Cmip03c PDF

Clay Mathematics Proceedings
This volume is the proceedings of

the 2002 Clay Mathematics Institute 3 Volume 3
School on Geometry and String
Theory. This month-long program
was held at the Isaac Newton
Strings and Geometry

Institute for Mathematical Sciences
in Cambridge, England, and was
organized by both mathematicians
and physicists: A. Corti, R. Dijkgraaf,
M. Douglas, J. Gauntlett, M. Gross,
Strings and
C. Hull, A. Jaffe and M. Reid. The
early part of the school had many
lectures that introduced various
Geometry
concepts of algebraic geometry
and string theory with a focus on
improving communication between
these two fields. During the latter Proceedings of the
part of the program there were also
a number of research level talks.
Clay Mathematics Institute
2002 Summer School
This volume contains a selection
of expository and research articles Isaac Newton Institute
by lecturers at the school, and Cambridge, United Kingdom
Douglas, Gauntlett and Gross, Editors

highlights some of the current
interests of researchers working March 24–April 20, 2002
at the interface between string
theory and algebraic geometry. The
topics covered include manifolds Michael Douglas
of special holonomy, supergravity,
supersymmetry, D-branes, the
Jerome Gauntlett
McKay correspondence and the Mark Gross
Fourier-Mukai transform. Editors
CMIP/3
www.ams.org American Mathematical Society

AMS
www.claymath.org CMI Clay Mathematics Institute
4-color process 392 pages • 3/4” spine

STRINGS AND
GEOMETRY
Volume 3
STRINGS AND
GEOMETRY
Proceedings of the
2002 Summer School
on Strings and Geometry
Isaac Newton Institute
Cambridge, United Kingdom
March 24–April 20, 2002
Michael Douglas
Jerome Gauntlett
Mark Gross
Editors
American Mathematical Society

2000 Mathematics Subject Classification. Primary 81T30, 83E30, 53C29, 32Q25, 14J32,
83E50, 14E15, 53C80, 32S45, 14D20.
ISBN 0-8218-3715-X
Copying and reprinting. Material in this book may be reproduced by any means for educa-
tional and scientific purposes without fee or permission with the exception of reproduction by ser-
vices that collect fees for delivery of documents and provided that the customary acknowledgment
of the source is given. This consent does not extend to other kinds of copying for general distribu-
tion, for advertising or promotional purposes, or for resale. Requests for permission for commercial
use of material should be addressed to the Clay Mathematics Institute, One Bow Street, Cam-
bridge, MA 02138, USA. Requests can also be made by e-mail to permissions@claymath.org.
Excluded from these provisions is material in articles for which the author holds copyright. In
such cases, requests for permission to use or reprint should be addressed directly to the author(s).
(Copyright ownership is indicated in the notice in the lower right-hand corner of the first page of
each article.)

c 2004 by the Clay Mathematics Institute. All rights reserved.
Published by the American Mathematical Society, Providence, RI,
for the Clay Mathematics Institute, Cambridge, MA.
Printed in the United States of America.
The Clay Mathematics Institute retains all rights
except those granted to the United States Government.
∞ The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
Visit the AMS home page at http://www.ams.org/
Visit the Clay Mathematics Institute home page at http://www.claymath.org/
10 9 8 7 6 5 4 3 2 1 09 08 07 06 05 04
Contents
The Geometry of String Theory 1

Michael R. Douglas
M theory, G2 -manifolds and Four Dimensional Physics 31
Bobby S. Acharya
Conjectures in Kähler geometry 71
Simon K. Donaldson
Branes, Calibrations and Supergravity 79
Jerome P. Gauntlett
M-theory on Manifolds with Exceptional Holonomy 127
Sergei Gukov
Special holonomy and beyond 159
Nigel Hitchin
Constructing compact manifolds with exceptional holonomy 177
Dominic Joyce
From Fano Threefolds to Compact G2 -Manifolds 193
Alexei Kovalev
An introduction to motivic integration 203
Alastair Craw
Representation moduli of the McKay quiver for finite Abelian subgroups of
SL(3, C) 227
Akira Ishii
Moduli spaces of bundles over Riemann surfaces and the Yang–Mills
stratification revisited 239
Frances Kirwan
On a classical correspondence between K3 surfaces II 285
Carlo Madonna and Viacheslav V. Nikulin
Contractions and monodromy in homological mirror symmetry 301
Balázs Szendrői
Lectures on Supersymmetric Gauge Theory 315
Nick Dorey
vii
viii CONTENTS
The Geometry of A-branes 337

Anton Kapustin
Low Energy D-brane Actions 349
Robert C. Myers
List of Participants 371
Preface
The 2002 Clay School on Geometry and String Theory was held at the Isaac
Newton Institute for Mathematical Sciences, Cambridge, UK from 24 March - 20
April 2002. It was organized jointly by the organizers of two concurrent workshops
at the Newton Institute, one on Higher Dimensional Complex Geometry organized
by Alessio Corti, Mark Gross and Miles Reid, and the other on M-theory orga-
nized by Robbert Dijkgraaf, Michael Douglas, Jerome Gauntlett and Chris Hull, in
collaboration with Arthur Jaffe, then president of the Clay Mathematics Institute.
This volume is one of two books which will provide the scientific record of the
school, and focuses on the topics of manifolds of special holonomy and supergravity.
Articles in algebraic geometry, Dirichlet branes and related topics are also included.
It begins with an article by Michael Douglas that provides an overview of the geom-
etry arising in string theory and sets the subsequent articles in context. A second
book, in the form of a monograph to appear later, will more systematically cover
mirror symmetry from the homological and SYZ points of view, derived categories,
Dirichlet branes, topological string theory, and the McKay correspondence.
On behalf of the Organizing Committee, we thank the directors of the Isaac
Newton Institute, H. Keith Moffatt and John Kingman, for their firm support.
We thank the Isaac Newton Institute staff, Wendy Abbott, Mustapha Amrani,
Tracey Andrew, Caroline Fallon, Jackie Gleeson, Louise Grainger, Robert Hunt,
Rebecca Speechley and Christine West, for their superlative job in bringing such
a large project to fruition, and providing the best possible environment for the
school. We thank the dining hall staff at Kings College, Magdelene College, Corpus
Christi College and Emmanuel College, and especially the Kings College singers, for
some memorable evenings. We thank the staff of the Clay Mathematics Institute,
and especially Barbara Drauschke, for their behind-the-scenes work, which made
the school possible. Finally we thank Arthur Greenspoon, Vida Salahi and Steve
Worcester for their efforts in helping to produce this volume.
Michael Douglas, Jerome Gauntlett and Mark Gross

September 2003
ix
Volume 3, 2004
The Geometry of String Theory
Michael R. Douglas
Abstract. An overview of the geometry of string theory, which sets the var-
ious contributions to this proceedings in this context.
1. Introduction
The story of interactions between mathematics and physics is very long and
very rich, too much so to summarize in a few pages. But from the beginning, a
central aspect of this interaction has been the evolution of the concept of geometry,
from the static conceptions of the Greeks, through the 17th century development
of descriptions of paths and motions through a fixed space, to Einstein’s vision of
space-time itself as dynamical, described using Riemannian geometry.
String/M theory, the unified framework subsuming superstring theory and su-
pergravity, is at present by far the best candidate for a unified quantum theory
of all matter and interactions, including gravity. One might expect that a worthy
successor to Einstein’s theory would be based on a fundamentally new concept of
geometry. At present, it would be fair to say that this remains a dream, but a very
live dream indeed, which is inspiring a remarkably fruitful period of interaction
between physicists and mathematicians.
Our school focused on the most recent trends in this area, such as compactifi-
cation on special holonomy manifolds, and approaches to mirror symmetry related
to Dirichlet branes. But before we discuss these, let us say a few words about
how these interactions began. To a large extent, this can be traced to before the
renaissance of string theory in 1984, back to informal exchanges and schools during
the mid-1970’s, at which physicists and mathematicians began to realize that they
had unexpected common interests.
Although not universally known, one of the most important of these encounters
came at a series of seminars at Stony Brook, in which C. N. Yang would invite
mathematicians to speak on topics of possible mutual interest. In 1975, Jim Simons
gave a lecture series on connections and curvature, and the group quickly realized
that this mathematics was the geometric foundation of Yang-Mills theory, and
could be used to understand the recently discovered non-Abelian instanton and
monopole solutions [23, 65]. These foundations are by now so familiar that it is
2000 Mathematics Subject Classification. 81T30, 83E30.
2004
c Clay Mathematics Institute
1
2 MICHAEL R. DOUGLAS
a bit surprising to realize that, for Yang-Mills theory, they date back only to this
time.
Among the participants in the Simons-Yang seminars was Is Singer, who carried
news of these developments to Atiyah at Oxford. Before long the mathematicians
were taking the lead in exploiting these solutions, culminating in the early 1980’s
with Donaldson’s use of instanton moduli spaces to formulate his celebrated invari-
ants [17], which revolutionized the study of four-dimensional topology.
While this case study in math-physics interaction might have ended there, with
the lesson being that mathematicians can find useful inspiration in physical devel-
opments but then must apply them to their own problems, of course it did not.
The deeper aspects of this interaction began with Witten’s 1988 reformulation of
the Donaldson invariants as observables in a topological field theory [58]. This set
the stage for the eventual application of deeper physical arguments, which led to
the 1994 Seiberg-Witten solution of N = 2 supersymmetric gauge theory [50]. Ap-
plying this to the topological field theory formulation led to the dramatic discovery
of the Seiberg-Witten invariants [60]. While the special role of four dimensions in
physics had led many to suspect physics would lead to new insights into special
features of four-dimensional space, this success went far beyond what anyone had
expected.
An extreme reading of these striking developments would take them as evidence
that fundamental theoretical physics and mathematics (or at least some subfields of
mathematics) have merged into a single unified field. However, few workers in either
field would agree with this claim, and consideration of the developments following
the Seiberg-Witten work bears this out: with some noteworthy exceptions such as
[42], the subsequent mathematical developments [43] and physical developments
(well discussed in Dorey’s lectures in this volume) have had little overlap so far. Of
course it would be overly pessimistic to rule out equally dramatic interactions in
the future, but the point we want to make is simply that the two fields are not the
same: they have different goals, different questions are raised, and even the sense
of when a question has been answered or an area understood is rather different in
the two fields. Thus the basic questions of why such a sustained interaction should
be possible at all, and whether we should expect this trend to continue indefinitely
or not, deserve serious consideration.
While we will not address these questions in depth here, the most basic point
to keep in mind is that a sustained interaction is only possible in an area which is
of fundamental interest on both sides, and in practice this interest is going to be for
very different reasons on the two sides. This was certainly the case for the study of
gauge theory, and it is equally true for string theory. Just because so many of the
recent physical developments start with interesting mathematics, say the existence
of manifolds of G2 holonomy, or the classification of Calabi-Yau threefolds, and
then suggest intriguing mathematical conjectures, does not in itself make these
interactions deep or significant. Rather, they have true significance only to the
extent that the ideas which cross over from one side to the other turn out to be
important in addressing the fundamental questions on the “other” side.
We will try to say a few words about this deeper significance in our conclusions,
but for now the main point we take from these comments is that one needs to
keep the fundamental questions from math and from physics clearly in mind in
evaluating the likely progress and significance of any given point of interaction.
THE GEOMETRY OF STRING THEORY 3
Thus I start in section 2 with a very brief overview for mathematicians of where
string theorists stand at present, and what for them are the fundamental physical
questions. While this will necessarily be a sketch with a lot of undefined terms,
besides serving as cultural background, it will outline the long road we have taken
from our experimental foundations to our present situation, in which we believe
interaction with mathematicians will provide us with essential insights. I will not
presume to do the same for the mathematical questions, instead referring to [47].
In the remainder of the article, we introduce the various frameworks used to
study string theory. Of course a basic problem in math-physics interaction is that
there is at present no precise definition of string theory or even most quantum field
theories, even to the satisfaction of physicists. On the other hand, the problem is
not as bad as one might think, in that these theories are highly overconstrained – it
is not hard to list axioms which almost all physicists would agree should characterize
them uniquely. Rather, the problem with treating the theories mathematically is
that along with these axioms, a very large number of supplementary assumptions
are used in physical arguments; one wants to reduce these down to a manageable
list which deserve the name “axioms.” Having even a redundant set of axioms
for a particular QFT or string theory, which sufficed to make interesting physical
arguments, would be a valuable advance, even before we reach the (probably still
distant) goal of proving the existence of the theories they characterize.
Most math-physics interaction is based on theories for which this problem is
not too serious, and we concentrate on these. In particular, classical field theory,
in which the basic mathematical framework is simply that of partial differential
equations, is far more generally applicable to these problems than one might think.
The only essential generalizations beyond what was needed to formulate general
relativity and Yang-Mills theory are fairly trivial mathematically, for example to
allow anticommuting variables.
While at first one might think classical field theory would be too simple to
address any of the questions of current interest, this is not true at all. Indeed, the
formulation of the Standard Model, which summarizes most existing observational
and experimental data, and most of the proposals which try to go beyond it, is
made using “effective field theory,” which summarizes quantum and string theoretic
results in classical terms.
The concept of effective field theory is also central in the connections to math-
ematics – in particular, the Seiberg-Witten solution is just the effective field theory
related to the original Donaldson theory. Thus we give an introduction to this idea
in section 3, emphasizing the sense in which effective field theory can be thought
of as a generalization of “moduli space.” As such, concepts developed in algebraic
geometry, such as formal deformation spaces and stacks, have simple effective field
theory analogs, and this is an important dictionary to learn.
In section 4, we discuss the physics of higher-dimensional field theories, Kaluza-
Klein reduction, and supersymmetry. For mathematician-friendly introductions to
supersymmetry, we recommend [24, 13]. Again, these are classical field theories,
but there is a precise sense in which a limit (the low energy limit) of string theory
is described by the ten- and eleven-dimensional supergravity theories, and many of
the recent developments involving duality were first discovered by working on this
level.
In section 5, we discuss results from the study of perturbative string theory, the
limit of string theory which can be defined in terms of a string world-sheet governed
by two-dimensional quantum field theory. This was the traditional definition of
string theory, and is still the basis of most of our understanding of string theory
beyond the low energy limit. As by far the most concrete formulation which goes
beyond conventional ideas of geometry, it has been a major focus of interaction
with algebraic geometers, leading to discoveries such as mirror symmetry, and its
more recent open string/Dirichlet brane analogs.
In section 6 we briefly discuss the more recent developments of superstring
duality and M theory. Most of the physics lectures in the volume address these
topics. While physicists have had a fairly clear picture of what M theory should
be for several years, it is still rather challenging to summarize this in a way which
mathematicians will find accessible. See [64] for another attempt at this.
In section 7 we conclude.
2. String theory and its physical goals

The stated goal of “fundamental physics” is to provide a single set of precise
laws which governs all observed physical phenomena. While ambitious, from many
points of view this goal has already been realized: it is difficult to find any hard
evidence for phenomena which are not very well described by the combination of
the “Standard Model,” a specific four-dimensional quantum field theory based on
Yang-Mills theory, and general relativity.
On the other hand, it is clear on theoretical grounds that this combination
of theories is incomplete. Most importantly, quantum effects in general relativity
predict the breakdown of the theory at a very high energy scale called the “Planck
scale,” Mpl ∼ 1019 GeV ≡ 1028 eV (electron volts). For comparison purposes, the
mass of the electron is about 5 × 105 eV.
To illustrate this breakdown, consider an experiment in which we collide an
electron and positron each moving with energy E. While the Standard Model
by itself makes definite predictions for any E, since the gravitational interaction
has strength roughly E 2 /Mpl 2
, as E ∼ Mpl this interaction is no longer a small
correction, and must be taken into account on an equal footing with the other
forces. Providing any candidate set of laws which does this remained an open
problem for many years. Over the course of the 1980’s, fairly convincing arguments
were developed to show that superstring theory could solve this problem.
Besides this basic problem, there are many features of the Standard Model
which strongly suggest that there are deeper layers of structure to be found. The two
best examples are known as “coupling unification” and the “hierarchy problem.”
To explain these, we start with an exceedingly brief overview of gauge symmetry
as it relates to the Standard Model.
2.1. The Standard Model and gauge symmetry. There are three fun-
damental forces of nature described by gauge theory, namely electromagnetism,
the weak interaction, and the strong interaction, These look very different in low
energy experiments: the electric and magnetic interactions are long range and rel-
atively weak; the weak interaction operates over nuclear distances and operates
between “currents” much like the electromagnetic current, while the strong inter-
action, while also operating only over nuclear distances, is exceedingly complicated
and on the surface bears little resemblance to Maxwell’s theory.
On the other hand, at the highest energies we attain in accelerator experiments

today, around 100 GeV, the three gauge forces look very similar, and one can
directly observe particles which form representations of the gauge groups U (1),
SU (2) and SU (3). Their very different behaviors at low energy are explained
through the existence of three phases of gauge theory, somewhat analogous to the
phases solid, liquid and gas of ordinary matter. One possibility, the “Coulomb
phase”, is the familiar behavior of electromagnetism as described by Maxwell’s
equations. The second is the “spontaneously broken” or “Higgs phase,” in which
the force is short range and weak; the weak interaction of the Standard Model
arises through breaking of SU (2) × U (1) gauge symmetry. Finally, the strong force
is in the “confined” phase, in which the gauge interactions between charged objects
(quarks) are not directly observable; indeed, the most basic physical effect is the
formation of a non-trivial vacuum “condensate” of quarks, so that the lightest
strongly interacting particles (the pions) are “spin waves” of this condensate.
While we will not get into the physical details of this, it is valuable to have
some mathematical analogies or counterparts for these effects, because they are
central in physicists’ thinking, and the actual study of supersymmetry and string
theory relies heavily on these ideas. Let us do this in the following framework: we
grant that the additional structure of string theory, M theory or whatever candidate
fundamental theory lies beneath the Standard Model can be described by a family
of “configurations”. Traditionally, in string theory a configuration is a choice of
structure for the compact manifold of “extra dimensions of space-time,” and might
include auxiliary structures such as vector bundles on this manifold, submanifolds
(branes), and so on. Eventually, we may want to describe string theory using
something other than differential geometry, and we might then use some more
abstract family of configurations. In any case, a possible physical history will be a
map from four-dimensional space-time into the family of configurations.
In this language, the gauge symmetry we observe arises because our vacuum
is a configuration with an isometry or endomorphism group. Since this group acts
at each point in observable space-time, it will lead to gauge symmetry, almost by
definition. Thus, physical experiment tells us that whatever may be the additional
structure which leads to the Standard Model, it must be preserved by an SU (3) ×
SU (2) × U (1) symmetry in some sense.
What about the three phases? The broken or Higgs phase is relatively easy to
explain in these terms: it corresponds to an “approximate symmetry.” More specif-
ically, it arises in the situation in which a family of configurations (say connections)
is naturally defined by a quotient construction. Reducible configurations then have
additional endomorphisms (by definition) and thus lead to enhanced gauge symme-
try. A configuration which is close to a reducible configuration (in terms of some
metric), will have broken gauge symmetry, at a scale proportional to the distance
to the reducibility locus. Thus, while our expectation of an SU (3) × SU (2) × U (1)
symmetry for the additional structure is correct, it must be that the effective po-
tential (to be explained below) has a minimum which is SU (3) symmetric, and near
but not precisely at the SU (2) × U (1) symmetric point.
The confined phase would seem the most difficult to explain in these terms.
In one way, it is not: from the point of view of families of configurations, if we
grant that the group G is confined, the consequence is that at low energies we must
identify all configurations related by G symmetry; in other words we can answer all
physical questions after performing the G quotient. Of course this begs the question
of whether or not a given symmetry G which appears at a particular reducibility
locus is confined; this is a central question we will return to below.
2.2. Coupling unification. At 100 GeV, the strengths of each of the gauge
interactions can be parameterized by a dimensionless “coupling constant” α1 ∼
1/60, α2 ∼ 1/30 and α3 ∼ 1/10. Now, due to renormalization effects, these cou-
pling constants depend on the energy at which they are measured. Going down to
everyday energies, a combination of α1 and α2 becomes the familiar “fine struc-
ture constant” ∼ 137 1
. On the other hand, if one extrapolates to high energies
one finds that all three become equal to αGUT ∼ 1/25 at a single energy scale
MGUT ∼ 1016 GeV called the “GUT scale,” where GUT stands for “grand unified
theory.”
Needless to say, while getting two functions (coupling as a function of energy)
to agree at some value would not be too impressive, the fact that three agree at
the same energy looks like clear evidence that something important happens at
that energy. The simplest hypothesis is that all three interactions arise from the
spontaneous breakdown of a larger gauge symmetry, say the group SU (5) which
contains SU (3)×SU (2)×U (1) as a (maximal) subgroup. This hypothesis of “grand
unification” implies a great deal of additional structure, and leads to very concrete
predictions, for example that the proton has a finite (although extremely long)
lifetime.
This picture was suggested in 1974 [27] and subsequently many attempts were
made to find a simple GUT from which all the observed structure of the Standard
Model would naturally arise. Although some progress was made, it is now generally
believed that there is no GUT which achieves this in a particularly simple and
natural way; the GUTs are almost as complicated as the Standard Model they
purport to explain. Rather, it appears that some other level of structure is required
to explain the origin of most of the features of the Standard Model, such as the
three generations of quarks and leptons.
2.3. The hierarchy problem. The Standard Model contains an explicit pa-
rameter setting the scale of electroweak SU (2)×U (1) symmetry breaking, the mass
squared of the Higgs boson. While not yet observed, if the Standard Model is right,
this must be between 110 GeV (the current experimental lower bound) and about
500 GeV. Either of these figures is far below the other two scales we mentioned, the
Planck scale and the GUT scale, and this already might lead one to deep scepticism
about any attempt to provide a complete fundamental theory – how can we hope
to extrapolate more than 13 orders of magnitude in energy, without experimental
input? On a more positive note, one can remark that no previously known physical
phenomenon relates such disparate scales, so if it is true here, there must be some
striking new physics which makes this possible, which we can hope to discover.
Of the various proposals in this direction, by far the most widely believed is
that this new physics is low energy supersymmetry, meaning not only that each
particle observed so far has a counterpart of the opposite statistics, but that these
“superpartners” must have mass around the same 100 − 500 GeV energy range.
Besides theoretical arguments, there is a powerful supporting piece of evidence in
the fact that the coupling unification we just mentioned has now been tested to
very high precision, and actually fails to work if one has only the Standard Model
particle content. On the other hand, it works extremely well if one assumes low
energy supersymmetry.
This is very good news for physicists as it would mean that the superpartners
should be observed in upcoming experiments at the Large Hadron Collider now
being constructed at CERN. These experiments should come into full swing around
the year 2008, giving physicists some time to try to sharpen our predictions in
preparation.
If we grant this conventional wisdom, the problem facing string theorists is fairly
clear. We must show that there is some solution of string theory which leads to
effectively four-dimensional physics at the energies of interest, contains low energy
supersymmetry, and agrees in detail with the tested predictions of the Standard
Model. Of course, we need not grant the conventional wisdom, and other scenarios
have been suggested, for example that some of the extra dimensions of string theory
will be observed in accelerator experiments. Other experimental surprises might
completely change this picture as well. Of course, any attempt to make physical
predictions is by definition speculative, and this is only a question of degree.
Leaving aside physical arguments as to where to place our bets, what is more
important for the present discussion is that the “conventional low energy super-
symmetry” scenario is the only one which has been developed physically to the
point where any of the important questions can be made mathematically precise,
and thus we continue our discussion in this framework.
3. Effective field theory

Once we establish the existence of solutions of string theory with low energy
supersymmetry, we will need to go on and derive their physical predictions. In
particular, we need a framework which is general enough to incorporate “stringy”
and quantum corrections, which will be essential for understanding effects such as
the breaking of supersymmetry at low energy.
In practice, the framework which is used is “effective field theory.” The physical
definition of an effective theory is as follows: one chooses an energy scale E, and asks
for the simplest theory which can reproduce all physical phenomena at energies up
to E, without knowing anything about the “degrees of freedom” present at energies
above E. The simplest example is quantum electrodynamics, which describes the
physics of electrons and photons. This theory is not complete, since of course there
are many other particles which can be created by the interaction of electrons and
photons. On the other hand, if one only does experiments at energies less than
105 MeV (the mass of the second lightest charged particle, the muon), the other
particles cannot be created directly. They still have “virtual” effects, but these can
be summarized by modifications (renormalizations) of the parameters entering the
Lagrangian, including those controlling “higher-dimensional operators.”
Thus, one can describe all physics at low energies solely in terms of the particles
we see, at the cost of needing to determine some additional parameters. The pro-
cess of finding the contributions of the more massive particles (equivalently, fields)
to these parameters is called “integrating out.” One can show that these effects
are controlled by powers of E/M , where M is the lightest mass of a particle not
explicitly present in the theory, so one only needs a finite (and small) number of
parameters to make real predictions.
This paradigm would appear quite appropriate for discussing the relation be-
tween string theory and the real world, at least until that happy day when we can
directly produce strings or higher-dimensional excitations, not described by conven-
tional field theory, in our experiments. For various reasons, at present most string
theorists do not consider this outcome very likely at LHC or any accelerator we can
imagine constructing, though in the end of course this is a question for experiment
to decide. We should also say that for some questions, especially those involving
gravitational effects, the applicability of effective field theory is controversial; see
for example [4] for arguments making this point.
In any case, the Standard Model is universally thought of as an effective field
theory, and thus it is a central problem to derive this particular effective field theory
from string theory.
For our purposes, an effective field theory is defined by the data (C, V, F, G):
• C is a manifold with metric called the “configuration space.” This is
not space-time; rather it is the space in which scalar fields take values.
Coordinates on this space are traditionally called “scalar fields” and are
often denoted φi .
• V is a real function on C, called the potential.
• F is a complex vector bundle over C, in which the fermions take values.
This also comes with a metric.
• G is a Lie group with an action on C, which preserves the other data.
Traditionally, all of this data is finite dimensional, and G is compact and semisimple.
It can be useful to consider infinite dimensional manifolds and groups as well.
The data would be used to define the effective field theory as follows (see
[24] for details). One takes as degrees of freedom a map M → C (the scalar
fields), a compatible section of Γ(M, F ⊗ Spin(M )) (the fermionic fields), etc. One
then uses the data to write an action for these fields. This requires postulating
additional data as well, most importantly the “Yukawa couplings,” a bilinear form
Y : F ⊗ F |φ → C at each φ ∈ C, but this will not appear in our discussion.
Finally, for present purposes, one treats this action classically; in other words
a possible dynamics for the system is given by a gauge equivalence class of solutions
of the classical equations of motion. Thus, a “configuration” as in section 2 is a
G-orbit of the form g(p, ξ) ∈ (C, F |p ) with g ∈ G.
The only solution we will discuss is the vacuum solution, in which the scalar
fields are constant,
∂φi (x) ∂V
= 0; = 0,
∂x ∂φi
and the fermions (the section of F ) are zero. We furthermore identify solutions un-
der gauge equivalence; in other words a solution is a G-orbit of critical points. Such
a solution preserves the maximal possible symmetry of four-dimensional space-time
and contains no particles anywhere in space; thus it is referred to as a “vacuum.”
For a stable vacuum, the solution should be stable to small perturbations. This
will be true if the critical point is a local minimum.
We say “a” and not “the” vacuum for the evident reason that a general func-
tion V can have many minima. A priori, any one of the minima is a candidate for
describing physics. All further physical predictions depend on the choice of mini-
mum φ: for example, the spectrum of scalar particle masses is the matrix of second
derivatives V (φ) evaluated in a local orthonormal basis at φ.
Our previous comments about gauge symmetry and its breaking can be sum-
marized in this language as follows: the unbroken gauge symmetry of a vacuum φ
is its stabilizer H ⊂ G, while an orbit at distance d from such a stabilizer has H
symmetry broken at energy E = d. Finally, whereas in general one can identify
G and distinguish points on a G-orbit by performing experiments at sufficiently
high energy, if an unbroken gauge group H confines at a point φH , then there is a
“confinement scale” Λ such that two points on the same H-orbit at distance d < Λ
from φH cannot be distinguished physically.
So far, we have not really distinguished classical field theory and effective field
theory. As we suggested earlier, the main physical point of the latter is that we
can incorporate all quantum mechanical effects, as well as all effects due to strings
or other objects which are not explicitly descibed by the effective field theory, in
the choice of the data of the effective Lagrangian. In particular, if we start with a
particular classical Lagrangian and quantize it, the resulting effective field theory
will obtain modifications or “renormalizations” of V and the metric on C.
In general, even data such as the choice of G or the topology of C can be dif-
ferent in the quantum theory from that one might have guessed at from a naive
interpretation of the classical Lagrangian, for example because confinement can
make the identification of G ambiguous, as we just discussed. While such possi-
bilities can be controlled using perturbative techniques at small , one needs more
subtle arguments to treat large ; much of the lectures of Acharya, Dorey and
Gauntlett are explicitly or implicitly devoted to this problem.
We can now summarize our problem as string theorists: it is to find the effective
field theories which can arise from string theory, with quantum corrections taken
into account, find their vacua, and find out which of these can agree with the
Standard Model.
3.1. The Standard Model and unification. In the Standard Model, G = ∼
SU (3) × SU (2) × U (1). G acts linearly on C and F , so the action is described by
a choice of linear representation. The irreps are usually denoted (N3 , N2 )Y , where
N3 and N2 are dimensions of SU (3) and SU (2) representations, and the notation
N̄3 is used to denote the complex conjugate representation. Finally, Y is the weight
for the U (1) action.
First, we specify C ∼
= (1, 2)1/2 , while V is an invariant quartic polynomial, for
which φ = 0 is a local maximum. The metric can be taken to be flat.
We then have
(3.1) F ∼ = 3[Q (3, 2)1/6 ⊕ D̄ (3̄, 1)1/3 ⊕ Ū (3̄, 1)−2/3 ⊕ L (1, 2)−1/2 ⊕ Ē (1, 1)1 ],
where we label the quark” fields Q, Ū and D̄, and the lepton fields L and Ē. In
other words, we have 3 copies of a specific 15-dimensional reducible representation
(a “generation” of quarks and leptons).
One must also specify (up to change of basis) three matrices m1 , m2 , m3 ∈
Mat3 (C), the Yukawa couplings. Together with the three gauge couplings, these
form 19 real parameters which must be determined by experiment.1
At first sight, a mathematician might say that while C and V seem reasonably
simple, F is a rather ugly and random looking collection of data. Needless to say,
physicists have had a lot of time to consider the matter, and to some extent have
1The discovery of neutrino oscillations is the first solid evidence for physics beyond the
Standard Model, but can be accommodated with a few more parameters and fields.
come to the opposite point of view: the G-representation which appears in (3.1)
can easily come out of a simpler starting point, but the presence of C and V is quite
difficult to understand.
Perhaps the simplest hypothesis which can lead to (3.1) is to postulate that
G = SO(10), F is 3 copies of an irreducible spinor representation, and C contains a
copy of the adjoint representation of G. It is then easy to find a φ whose stabilizer
is the Standard Model gauge group, and it is a simple exercise to check that the 16
dimensional spinor of SO(10) decomposes into the spectrum (3.1) plus an additional
trivial representation. Thus this is a “grand unified theory.” Other nice unification
groups, for which simple choices for F can decompose into (3.1), include SU (5) and
E6 .
Although this striking observation may look like direct evidence for grand uni-
fication, one can find other explanations for this matter content. For example, a
consistency condition called anomaly cancellation strongly constrains the allowed
matter representations in gauge theory, and determines the U (1) weight assign-
ments in (3.1). Still, this observation combined with coupling unification is very
suggestive.
Points which remain unexplained are the number 3 of generations and the
structure of the Yukawa couplings. Most significantly, as mentioned already, one
also has the hierarchy problem: the potential V must have a minimum of order
100 GeV, while other couplings are of order 1016 GeV. This is not easy to arrange,
and the only simple way to do it involves supersymmetry.
3.2. Supersymmetric effective field theory. An N = 1 supersymmetric
effective field theory can be defined by the data (C, G, µ, W ).
• C is a complex manifold with Kähler metric.
• G is a Lie group with a holomorphic action by isometries on C.
• µ are moment maps for the G action.
• W is a holomorphic G-invariant function on C, called the superpotential.
Note that there is less data than in the non-supersymmetric case, and that it obeys
the far more constraining conditions of complex geometry.
This data determines a specific effective theory which can also be described
in the general terms we gave earlier, but with very specific relations among the
particles and the couplings. In particular, each particle has a “superpartner” –
quarks come with scalar “squarks”, and gauge bosons come with partner “gaug-
inos.” While none of these have been observed yet, there is a precise hypothesis
which accounts for this: supersymmetry is spontaneously broken.
What this means for present purposes is the following. Given the data, the
potential of the effective theory is
(3.2) V = (∂W, ∂W ) + |µ|2 ,
where ∂ is the holomorphic gradient, (, ) is the Hermitian inner product derived
from the Kähler metric, and µ are the moment maps (the metric which appears
here is derived from the gauge couplings).
Such a potential has two types of critical points. One type, in which
(3.3) 0 = ∂W = µ,
is a “supersymmetric vacuum.” One can also have critical points V = 0 at which
(3.3) does not hold, called “non-supersymmetric” vacua.
In a supersymmetric vacuum, the naive consequences of supersymmetry hold –

for example, superpartners must have the same mass. This is clearly not true in our
universe, so if supersymmetry is relevant for real physics, it must be spontaneously
broken. In other words, we must be in a non-supersymmetric vacuum. Expanding
about such a vacuum, one can compute masses of all particles; now superpartners
have different masses, and it turns out to be natural for the presently observed
particles to be lighter than the others.
Thus, spontaneously broken supersymmetry is not easy to detect experimen-
tally, until one reaches the energy scales set by |∂W | and |µ|, which sets the mass
of the superpartners. On the other hand, whether or not the supersymmetry is
spontaneously broken, supersymmetric theories enjoy much better renormalization
properties, in particular the superpotential is not renormalized to all orders in .
This is an essential ingredient in most solutions to the hierarchy problem, as it
allows terms of very different order to coexist. It is not the entire story, as one
must somehow generate these terms in the first place. The field of “dynamical su-
persymmetry breaking” began with the realization that W could obtain corrections
from instanton effects, which are exponentially small in . One can find theories in
which supersymmetric vacua which would have been present without these effects
become nonsupersymmetric vacua in the exact theory, leading to a solution to the
hierarchy problem.
A lot of work has been done on supersymmetric extensions of the Standard
Model. Indeed, the simplest such extensions have the same G = SU (3) × SU (2) ×
U (1), while C is now the direct sum of F from (3.1), with the C as in the Standard
Model, with a “second Higgs” (1, 2)−1/2 (necessary because W is holomorphic,
so the two representations are different). The Yukawa couplings of the Standard
Model, are then derived from cubic terms in the superpotential,
(3.4) W = Ū m1 QH1 + D̄ m2 QH2 + Ē m3 LH2 ,
where Hi are the two Higgs fields, and m1 , m2 and m3 are 3 × 3 matrices encoding
the specific Yukawa couplings.
Thus, the Standard Model fits relatively easily into a supersymmetric extension,
and its grand unified extensions do as well. This is somewhat non-trivial as one
can easily write effective field theories which do not. Checking these constraints
from supersymmetry largely depends on discovering the Higgs boson, but it is also
generally believed that supersymmetry forces the Higgs bosons to be relatively
light, so this should also happen before long. Either way, the status of low energy
supersymmetry should become much clearer in a few years.
Having gotten this far, we can briefly mention the next important problem
in constructing a supersymmetric GUT, the problem of “fast proton decay,” and
the related “doublet-triplet splitting” problem. First, note the non-generic form of
(3.4), in which no terms appear which are odd in quarks or leptons separately. Such
terms would describe processes in which quarks decay (or transmute) into leptons,
but the observation that the proton lifetime is greater than 1033 years forces the
coefficient of any such term to be vanishingly small. The problem is then that this
is not easy to get from a GUT, which in the first instance treats quarks and leptons
on the same footing. It is not impossible, but we refer to [63] and references there
for further discussion of how this can be done.
3.3. Summary. We gave a brief sketch of the Standard Model, and a few of
the most promising ideas for what physicists believe lies beyond it, grand unification
and supersymmetry. All of these ideas can be described in the general framework
of four-dimensional effective field theory.
Attempts to derive these effective field theories from superstring theory and
understand their physics have been a major focus of the field for almost twenty
years. It is this line of work which has led to most of the mathematical “spin-offs,”
and explaining this will take up most of the rest of the lecture.
4. Supersymmetric Kaluza-Klein compactification

The field of superstring compactification started in 1985 with the work of Can-
delas, Horowitz, Strominger and Witten on compactification of the heterotic string
on Calabi-Yau manifolds, which led to the first quasi-realistic models with low
energy supersymmetry [6].
While this construction was discovered in the context of string theory, all of
the work was actually done using field theory, and in concept directly follows the
original picture of Kaluza and Klein, in which Maxwell’s equations were derived as
a consequence of compactification of five-dimensional general relativity. Thus we
review this picture very quickly before proceeding.
The modern version of the Kaluza-Klein idea can be stated succinctly as the
idea that all properties of the four-dimensional effective field theory of our universe
can be traced to geometric properties of extra dimensions of space. The central
problem in “the geometry of string theory,” is a definition of “geometry” which is
general enough to encompass all work on string and M theory compactification in
a reasonably unified way.
4.1. Kaluza-Klein compactification. Consider a solution of general rela-

tivity which is close to the product manifold M × K, where M is four-dimensional
Minkowski space-time with its flat metric, and K is a D-dimensional compact man-
ifold. The data of (D + 4)-dimensional general relativity is a D + 4 metric and its
associated Levi-Civita connection; we now try to interpret this as 4-dimensional
data.
Besides a 4-dimensional metric, we find two additional “fields.” First, there is
the D-dimensional metric on K. This is a choice which can be made independently
at each point in M , and thus this choice is a configuration in the terms of section
2. The vacua are direct product solutions of the Einstein equation, with a fixed
Ricci-flat metric on K. This can be described in effective field theory terms by
taking C = Met(K), the infinite dimensional space of metrics on K, and V to be
the Einstein-Hilbert action functional on these metrics.
Close to the direct product solutions are solutions in which the metric on K is
slowly varying as we move in M . These can be described to a good approximation
by a scalar field, a map from M into C.
This approximation is typically controlled by the ratio between the scale L of
this variation in M to the radius R of K. More precisely, it is controlled by the
ratio E/h, where E is the energy and h ∼ 1/R is the gap (the lowest non-zero
eigenvalue) for the scalar (and other) Laplacians) on K. If we want an effective
field theory for energies far below h, the description using an infinite dimensional
C is far more complicated than we need.
Rather, one can take for C the moduli space of Ricci-flat metrics on K. Since
all of these are solutions, one takes V = 0. This C carries a natural metric, the
Weil-Petersson metric, inherited from the natural metric on Met(K).
The remaining data contained in the (D+4)-dimensional metric can be regarded
as a connection on the fibration of K over M . While in general this connection need
not be linear, if we restrict attention to solutions which are slowly varying in M we
find that the connection must be trivial up to a possible action by an isometry of
K. In particular, the original result of Kaluza and Klein was that K = S 1 ∼ = U (1),
so in this case one recovers a U (1) connection. More generally, one obtains a G-
connection, where G is the isometry group of K. Furthermore, there is an induced
action of G on C; in physics terms one says that the scalar field is “charged” under
G.
While in pure (D + 4)-dimensional general relativity K must be Ricci-flat, by
considering other theories this can be generalized. The simplest possibility [25] is
to consider a (D + 4)-dimensional theory with an additional 3-form gauge potential
(generalizing Einstein-Maxwell theory). As discussed by Acharya and Gauntlett in
this volume, one can show that the resulting field equations can be solved by K
with constant positive scalar curvature. This greatly enlarges the possible isometry
group, at the cost of obtaining M with constant negative curvature (anti-de Sitter
space).
To end, we summarize the proposal of Kaluza and Klein as the idea that a
configuration is a manifold with metric, up to the natural equivalence by diffeomor-
phisms. Gauge symmetry arises at fixed points of the quotient by this equivalence.
4.2. String theory. For a long time, the Kaluza-Klein idea was completely
ignored, for the simple reason that it requires starting with a higher-dimensional
analog of general relativity. Even in four dimensions, it is very hard (most would
say impossible) to quantize general relativity by a direct approach, for the reason
mentioned earlier that the interaction becomes strong at short distances. This
problem gets worse in higher dimensions and, while mitigated in supergravity, is
not cured.
In superstring theory, point particles are replaced by strings. These come with
a natural length scale, their average size or “string length” ls , and it turns out that
all interactions become weak at distances shorter than ls , overcoming this problem.
On the other hand, at distances larger than ls , it is a good approximation to regard
the string as a convenient summary for a particular spectrum of particles, one for
each possible joint state of its modes of vibration. Only a small number of these
particles have mass small compared to 1/ls , and thus string theory will reduce to
a field theory with a finite number of fields at distances L >> ls and energies
E << 1/ls . This is the limit in which we want to think of string theory as reducing
to general relativity.
The fact that superstring theory includes general relativity is still one of the
most striking and mysterious facts about the subject, and we will not be able to
explain it satisfactorily here. Most of the explanations are indirect, developing
other properties of the theory and eventually seeing that these imply the existence
of a massless spin two particle, which by general arguments imply that the theory
must include general relativity.
Perhaps the simplest indirect explanation of this sort makes gravity a con-
sequence of supersymmetry. We will only describe this impressionistically: the
motion of a string can be described by a two-dimensional field theory of maps from

Σ, a two-dimensional manifold of topology S 1 × R parameterizing the space-time
history of a loop of string, into space-time M × K. There are also two-dimensional
field theories whose fields are fermionic, taking values in a space Λ with anticom-
muting coordinates, as well as more complicated possibilities. A theory of maps
into M × K × Λ will naturally have symmetries which mix all of the dimensions,
and these are the supersymmetries.
We now grant that string theory can realize supersymmetry (in fact string
theory came first, and was one of the clues which led to the discovery of supersym-
metry), and consider the consequences of supersymmetry.
4.3. Supersymmetry and supergravity. After the dimension of space-time,

the most basic characterization of a supersymmetric theory is the number of super-
charges (in Minkowski space-time), which we denote N s. The N = 1 theories in
four dimensions have N s = 4, but the upper bound (for the known unitary theories)
is N s = 32.
There are three basic supergravity theories with N s = 32. The simplest one,
discussed in Acharya and Gauntlett’s lectures, is formulated in 11 space-time di-
mensions; the supercharges form a real spinor representation of SO(1, 10). Besides
the metric, it contains a fermionic “gravitino,” and a three-form gauge potential.
The other two basic supergravities are formulated in 10 space-time dimensions, and
the supercharges form spinor representations of SO(1, 9) – depending on the choice
of representation, one obtains the IIa or the IIb theory.
The basic example of a supersymmetric theory with N s = 16 is super Yang-
Mills theory in ten dimensions; this theory contains a fermionic superpartner called
a “gaugino.” As with pure Yang-Mills theory, these theories are uniquely deter-
mined by the choice of “gauge group,” a finite dimensional compact semisimple Lie
group G. These theories can also be coupled to gravity, to obtain “type I super-
gravity.” In this case, the gauge coupling is determined by the expectation value
of a complex scalar field usually called the “dilaton-axion.”
The explicit Lagrangians and equations of motion for these theories are of course
known and discussed in other contributions in this volume. More important for us
here are the supersymmetry transformation laws, and the corresponding conditions
for unbroken supersymmetry,
(4.1) 0 = δψi = (Di + Γ · G)
(4.2) 0 = δχ = Γij Fij .
Here G denotes a direct sum of the various gauge field strengths of the particular
supergravity theory.
A background in which the supersymmetry variations of all fields are zero is
one which preserves supersymmetry. In classical backgrounds, fermionic fields have
zero expectation value, so this condition reduces to setting the right hand side of
(4.1) to zero. The analysis of these conditions, and variants of them which we will
come to, is the first step in the study of supersymmetric compactification.
4.4. Low energy supersymmetry from string theory. Let us start with
the simplest case of F = G = 0. In this case, the conditions (4.1) reduce to Di = 0;
in other words must be a covariantly constant spinor.
As is familiar, covariantly constant spinors are associated to a reduction of

the global holonomy group of K from SO(D) to a subgroup H, such that the
spinor representation contains an H-invariant subspace. The possibilities, listed in
Gauntlett’s lectures, include all of those on Berger’s list which support Ricci-flat
metrics. The simplest, used in the early works, is H = SU (n) with n = D/2.
Manifolds with SU (n) holonomy admit Ricci-flat metrics, and are thus solutions
of supergravity. In particular, compactification of the N s = 16 supergravity on a
Calabi-Yau threefold, leads to an N s = 4 supergravity in four dimensions.
On a Calabi-Yau, the second equation in (4.1) reduces to the Hermitian Yang-
Mills equations (in the special case c1 (V ) = 0), F (2,0) = F (1,1) ∧ J ∧ J = 0 (where J
is the Kähler form). Now one appeals to the theorems of Donaldson and Uhlenbeck-
Yau to argue that for a µ-stable holomorphic vector bundle V on K, these equations
have a unique solution.
This class of theories provides the simplest candidates for realistic physics, as
we discuss shortly. There are two more essential ingredients in the string theoretic
discussion. First, it turns out that to quantize a field theory in ten dimensions, the
spectrum of fermions must satisfy the very strong constraint of “anomaly cancella-
tion.” Without going into details, this arises because the determinant of the Dirac
operator acting on an irreducible spinor representation in ten (or any even number
of) dimensions is gauge invariant only up to a phase, and one must arrange for the
non gauge invariant phase to cancel between representations. In ten-dimensional
supergravity, this is true for the type II theories, and for type I theories only for
the gauge groups E8 × E8 and Spin(32)/Z2 . Both can be realized in the heterotic
string.
A full discussion of this point brings us to the second essential string theoretic
modification, the “Green-Schwarz term.” This imposes a topological constraint on
the gauge bundle,
c2 (K) = c2 (V ).
The resulting four-dimensional theory will have as a factor in its gauge group
the endomorphism group of V as an E8 × E8 or Spin(32)/Z2 bundle, while C will
include the formal deformation space of V , as we shortly discuss.
4.5. Deriving the supersymmetric effective field theory. Suppose we

grant the above, and find an interesting M and V ; how would we go on to derive
the effective field theory, to compare it with the supersymmetric Standard Model?
The first point to recognize is that the discussion above, based on Kaluza-Klein
reduction of a supergravity theory, does not lead to exact results in superstring
theory. Physically, one usually thinks of supergravity as the “weak coupling” and
“large volume” limit of superstring theory. These limits are characterized by two
parameters, controlling “stringy” and “quantum” corrections respectively.
The “stringy” corrections are functions of the length scales L of the internal
manifold K, measured in units of the string length ls ; in other words of ls /L, which
go to zero as L → ∞. These include “world-sheet instantons,” maps from two-
dimensional surfaces Σ into M , controlled by exp(−ls2 /L2 ) where L2 is the volume
of a two-cycle. In general they also include “perturbative” corrections, given by a
general Taylor series in ls /L. The “quantum” corrections are functions of the string
coupling gs (the stringy analog of Planck’s constant ) and can also be power-like
or exponentially suppressed.
In any case, physicists argue that as ls /L → 0 and gs → 0 the supergravity de-

scription becomes good. Thus the style of analysis described above can be extended
to derive the entire low energy effective Lagrangian in this limit.
Actually, gs is controlled by a scalar field, the dilaton-axion, so a better way
to think about the situation is that the effective Lagrangian data one derives in
the supergravity approximation is an approximation to the true data, which is
controlled in a subset of C, and progressively less reliable away from this subset.
The parameters ls /L are also controlled by fields (the Kähler class of the Ricci-
flat metric), so this statement of the situation is general: we only have direct
information about a limit of the effective field theory.
For quite a while, this realization, and general arguments that our universe
could not be described by the supergravity limit, turned most string theorists away
from detailed exploration of the problem we now discuss. On the other hand, it
does exhibit a lot of the structure one wants, and later considerations involving
duality make even limiting results much more interesting. So let us continue.
On some level, the result is not hard to describe, in the case G = 0. The
space C is a fibration, whose base is the product of the moduli space of Ricci-flat
metrics on M with its natural metric with a copy of the hyperbolic disc (in which
the dilaton-axion takes values).
The fiber F is harder to describe. One’s first attempt would be to take it
to be the moduli space of holomorphic E8 × E8 (or Spin(32)/Z2 ) bundles on K.
Within this moduli space, interesting bundles V lie on the reducibility locus with
endomorphism group SU (3) × SU (2) × U (1), while infinitesimal deformations away
from this locus correspond to the spectrum of quarks and leptons (3.1).
Of course, this is problematic: in general there is no good moduli space, and
even if we place restrictions to make it so (for example, consider µ-stable bundles), it
is not a manifold. In favorable cases it might be a variety with quotient singularities,
with some complicated obstruction theory. This is all relevant physical data; for
example the Massey products translate into Yukawa couplings in the effective field
theory.
One can postpone the problems of defining a moduli space by taking the fiber
to be the space of holomorphic connections on K, described as deformations of a
reference connection by an End G-valued (0, 1)-form. Since K is Kähler, this space
has a natural Kähler metric. The superpotential W is the holomorphic Chern-
Simons action,

2
(4.3) W [Ā] = Ω ∧ Tr(Ā∂¯Ā + Ā3 ).
M 3
This is a functional of Ā with the property that δW/δ Ā = F (0,2) = 0; in other
words its stationary points are connections on holomorphic bundles. Finally, the
G-action is the action of the group of holomorphic maps from K into GC , and
the corresponding moment maps µ are the Hermitian Yang-Mills equations, as
explained in [17]. Reducible connections lead to enhanced gauge symmetry, as in
our previous discussion, while the cubic coupling in (4.3) directly encodes the cubic
terms in W which become the Yukawa couplings.
While this is morally an effective field theory, it is infinite dimensional and
not concrete enough to address physicists’ questions. We need ways to reduce this
description to a finite number of variables. Now, many generalizations of “moduli
space” have been suggested by mathematicians to capture this information and get
a usable description of F . Personally, while one of my goals for the school was to
start a serious math-physics discussion of these ideas, I am not sure this effort was
a success; still, I hope some student will soon prove me wrong!
Perhaps more importantly, as we mentioned, the effective field theories one
obtains are only approximations valid in the large volume limit; as we move away
from this limit, the general result will obtain “quantum” and “stringy” corrections.
Rather than make this approximate effective field theory explicit, and then add
these corrections, one would rather find an equally general “geometric” characteri-
zation of the true effective field theory, and then find its explicit realizations.
5. Beyond differential geometry: the string world-sheet

So far we have been discussing applications of existing geometry – differential
and algebraic – to string theory. Why should we expect string theory to require,
or suggest, new concepts of geometry?
There are several intuitive reasons to expect this. Of course the most funda-
mental is that the primary definitions are not in terms of point-like particles but
instead extended objects, strings and the “branes” we discuss below. One thus feels
that concept of geometry based on points is somehow not “to the point,” and one
should start from new foundations.
These new foundations do not yet exist, though we will certainly come back to
this topic below. Now, even if we try to formulate string theory using the standard
language of field theory, we find not general relativity and Yang-Mills theory, but a
two-parameter deformation of these theories. The case of finite string length ls is
still far better understood than that of finite gs , and in this section we consider it.
In string theory terms, it corresponds to “perturbative” string theory (so, processes
involving one or a few strings), but moving through a background geometry in which
curvature and other geometric length scales can be arbitrarily small.
The propagation of a string moving through any background geometry, which
solves the stringy analog of Einstein’s equations, is described by a two-dimensional
conformal field theory, and thus we start here. This subject is on a somewhat
better mathematical footing than general field theory or string theory, as there are
many exactly solved cases (Wess-Zumino-Witten models, Gepner models, and so
on) which either have been or are in the process of being rigorously formulated. We
refer to [56] for a status report on this work, and continue our overview.
The most “geometric” definition of quantum field theory is the nonlinear sigma
model. This is defined as a functional integral over maps from Σ, here a two-
dimensional Riemannian manifold, to a target space K, a manifold with metric;
denote this g ∈ Met(K). One can also supersymmetrize this model, adding fermions
valued in the tangent bundle to K.
The foundational result here [26] is that the renormalization theory is geo-
metric, i.e. can be done in a generally covariant way. While the arguments are
perturbative, so is renormalization, so it is very plausible that an exact defini-
tion would share this property; of course more rigorous arguments would be very
valuable.
The primary result of the renormalization theory is the renormalization group
(RG) flow, a vector field on Met(K). This is a deformation of the Ricci flow, by
corrections in a Taylor series in ls and local tensors constructed from the metric.
The additional couplings of string theory, such as the “B-field” and dilaton, can
also be treated; adding the latter leads to the modified flow equations studied (for
example) in [46].
Fixed points of the RG flow are conformal field theories. Very few examples
are known for the original nonlinear sigma model, and these have either flat metric
or additional background fields (for example, the Wess-Zumino-Witten model has
a B-field) and at least hints at some underlying integrable structure. On the other
hand, it has been argued for the supersymmetric sigma model that any Ricci-flat
metric on K will correspond to a fixed point. Note that the fixed point metric
is typically not Ricci-flat, due to the ls corrections [31]; rather the obstruction to
solving the corrected fixed point equations vanishes in perturbation theory [45].
5.1. Strings as loops, and T-duality. The approach we just described,

based on constructing the stringy deformation order by order, is not too satisfying;
one would rather characterize the theory at finite ls , presumably in terms of some
radical reformulation of geometry.
While this goal is widely accepted, other satisfactory starting points are not
easy to find or work with. Perhaps the one which has been most studied is to
formulate stringy geometry as a geometry of loop spaces. So far, attempts to do
this, both physical and mathematical, rapidly get lost in the technicalities of infinite
dimensional manifolds, renormalization theory, and so on. Some sort of conceptual
breakthrough seems to be required to make useful progress in this direction.
Even without this, the intuitions coming from the picture of loops and extended
objects have motivated many important developments. Perhaps the most simple
intuitive ideas which came out of perturbative string theory are T-duality, of which
mirror symmetry is an example, and the noncommutativity of open string theory.
T-duality is a symmetry of perturbative string theory which appears in com-
pactification on tori, or torus fibrations. For example, a compactification of the
type IIa string on an S 1 of circumference 2πR is physically identical to a compact-
ification of the IIb string on an S 1 of circumference 2πls2 /R.
The picture which leads to this equivalence is the following. Compactification of
field theory on an S 1 leads to a spectrum of “Kaluza-Klein” or “momentum” modes,
whose masses are given by the square roots of the eigenvalues of the Laplacian, say
Mn = n/R for n ∈ Z. This is part of the string theory spectrum, obtained by
quantizing string loops with trivial topology (zero winding number).
By itself, this spectrum is obviously not invariant under R → ls2 /R. On the
other hand, string theory contains additional “winding modes,” defined as string
loops around the S 1 with fixed winding number, call this m. These have mass
equal to the string tension, 1/2πls2 , multiplied by the length of the geodesic with
this winding number, 2πRm, so Mm = Rm/ls2 . Now R → ls2 /R is a symmetry of
the union of these two mass spectra, exchanging momentum and winding modes.
This intuitive argument can be made precise using string world-sheet techniques.
To generalize this to higher-dimensional tori, one starts by noting that the
moduli space of string compactifications on T d with d > 1 is larger than the space
of metrics: one can also choose a harmonic two-form called the “Neveu-Schwarz
B-field.” The combined moduli space of metric and B-field can be shown to be the
homogeneous space SO(d, d, R)/SO(d) × SO(d) before T-duality, and T-duality
leads to identifications by any element of the discrete group SO(d, d, Z). The re-
sulting quotient space is a submanifold of C for type II string compactification on
tori, but not all of C – we will discuss this further under “duality” below.
What is most important for later developments is that there is much evidence
that T-duality applies not just to tori, but generalizes to any fibration with a torus
action which (at least approximately) preserves the metric.
5.2. Mirror symmetry and topological string theory. One can also take
the point of view that the framework of two-dimensional conformal field theory is
itself the starting point for “stringy geometry.” On the surface, this certainly
looks very different from conventional geometry. For example, some conformal field
theories can be defined in purely algebraic terms, without assuming that the fields
take values in any target space. One then might try to find geometric notions which
capture the structure of this algebraic problem, in the spirit of modern algebraic
geometry.
While attractive, this point of view has not been easy to implement in gener-
ality, because by themselves the axioms of conformal field theory are far too weak
to draw interesting conclusions. There is a special case of the problem, rational
conformal field theory [41], which is more accessible, and this led to one of the
most interesting early results, the construction of Gepner [28]. This was an ex-
plicit algebraic construction of certain (2, 2) theories, which Gepner proposed (and
others confirmed) were equivalent to sigma models with Calabi-Yau target space at
particular points in moduli space. On the other hand, the class of rational theories
is not preserved by deformation, and thus seems too special to form the foundation
of a new geometry.
The most interesting CFT results so far are for the special case of the (2, 2)
models. To make a long story short (see [39, 14, 29]), the supersymmetric nonlinear
sigma model with target space a complex Kähler manifold has additional world-
sheet supersymmetry, the (2, 2) superconformal algebra. This structure is very
central in all relations between string theory and algebraic geometry. It also leads
to the definition of “topological twisting” and the A and B twisted sigma models.
In early study of the (2, 2) models with Calabi-Yau target space [16, 35], it
was noted that the complex and Kähler moduli entered on a “symmetric” footing,
and that their roles could be interchanged by a simple automorphism of the (2, 2)
algebra. This observation was developed by Greene and Plesser [30], who used
Gepner’s construction and identifications to propose a specific equivalence between
a pair of sigma models with Calabi-Yau target. This proposal was the basis for the
famous work of Candelas et al [7], counting curves using mirror symmetry.
While the direct CFT point of view remains difficult, much progress has been
made in recent years by the use of the “linear sigma model” construction [59].
This essentially constructs Calabi-Yau target spaces as subvarieties of toric varieties
obtained by an explicit quotient of Cn by holomorphic isometries. We refer to [39]
for a detailed discussion of this topic, and the associated mathematical and physical
proofs of mirror symmetry.
5.3. Dirichlet branes. Although historically Dirichlet branes were not much
studied before the “duality revolution” we discuss below, they are also defined in
terms of the string world-sheet, and thus belong logically to the current discussion.
Strings can be closed, with topology S 1 , or open, with the topology of the
interval. Intuitively, a Dirichlet brane is an allowed endpoint for an open string.
Thus, in a geometrical framework, a Dirichlet brane could be specified by a choice
of a submanifold of the target space, on which a particular type of open string is

allowed to end.
There is an additional possible choice: since the end of an open string is a point,
it behaves physically like a particle, and can be coupled to a Yang-Mills field. Thus,
one can also specify a non-trivial Yang-Mills connection on the Dirichlet brane. In
fact, when one quantizes the string, one finds that this degree of freedom emerges
as a physical field, much in the same way that gravity emerged from the closed
string. Thus, in the physical discussion one cannot leave this out.
On a more general level, the Dirichlet brane turns out to be the simplest way
to embed Yang-Mills theory into the general framework of string theory, and has
led to many striking connections between the mathematical and physical study of
Yang-Mills theory. The simplest, which led to the original argument that Dirichlet
branes must be included in string theory [12], is that T-duality acts on a Dirichlet
brane as a sort of “Fourier-Mukai” or “spectral cover” transform. For example, a
brane embedded in T d carries a U (1) connection, which by the Yang-Mills equations
must be flat. Flat U (1) connections of course parameterize points in a dual T d ,
which turns out physically to be the moduli space of a T-dual Dirichlet brane which
sits at a point in the dual torus.
The next such relation to be discovered was the equivalence between “point-like
instantons” and Dirichlet branes [61, 18]. If one considers a Dirichlet brane with
spatial dimension p ≥ 4, the associated Yang-Mills field theory will have self-dual
instanton solutions. Among their moduli is a scale size, and one can ask how string
theory resolves the singularity which appears when the scale size goes to zero. In
fact, one finds that, in the limit, the instanton becomes equivalent to a second
Dirichlet brane of spatial dimension p − 4. Following this idea, one can find a
purely physical rederivation of the ADHM construction of instanton moduli space
[3], which admits many generalizations and has found many physical applications,
as discussed in Dorey’s lectures.
5.4. Noncommutative geometry. The physics of Dirichlet branes is rich

and complicated, and one soon feels the need for organizing principles. One of the
simplest of these starts with the elementary observation that an open string has
two ends. Furthermore, if we consider oriented strings, an interaction of two open
strings can take place when the end of the first string adjoins the beginning of the
second. This can be distinguished from another possibility – the end of the second
string could adjoin the beginning of the first – which leads to a different interaction.
These interactions can be formulated more precisely as the operator product
algebra on the two-dimensional world-sheet. Whereas for closed strings this prod-
uct is defined by taking a coincidence limit of two points on the sphere, a limit
without any natural ordering, for open strings one takes a coincidence limit of
points on the boundary of the disk, which are ordered. This distinction can be
encoded algebraically as noncommutativity – whereas closed string operator prod-
ucts form a commutative algebra, open string operator products naturally define a
noncommutative algebra.
One furthermore has a choice of boundary conditions – a choice of submanifold
and connection in the geometric description, or some more abstract choice in general
CFT. Since an open string has two ends, its spectrum depends on a choice of a pair
of boundary conditions, say A and B, and a string ending on (say) B can only
interact with another string beginning on B.
One way to encode these constraints is to say that Dirichlet branes (boundary
conditions) are objects in a category, and open strings are morphisms. This is not
the only way to think of this; one could instead say that open strings form a groupoid
algebra, as in [9], but given a general and diverse set of boundary conditions the
categorical language is perhaps more natural. In any case, the essential feature is the
noncommutativity – in this sense, the Dirichlet branes define a “noncommutative
geometry” associated to any CFT.
This idea has been made precise in several different ways. In principle, the
most general would be to work with open string field theory, along the lines of
[57], but so far this framework remains cumbersome. To avoid its difficulties, one
must consider a special case or limit in which the operator product algebra of CFT
reduces to an ordinary algebra. This essentially requires operator products to be
independent of the positions of operators on the world sheet; in other words one
must study topological string theory.
One special case in which the string reduces to a topological string is the study
of a background Neveu-Schwarz B-field, as discussed by Cattaneo and Felder [8]
following earlier work by Kontsevich [34]. In particular, in toroidal compactifica-
tion, this leads to the noncommutative torus algebra, as was argued in works of
Connes et al [10], Douglas and Hull [21], Schomerus [52] and Seiberg and Witten
[51].
5.5. Open string mirror symmetry. Another special case in which the
string can be made topological is the sigma model with Calabi-Yau target. Now
the topologically twisted open string theories provide categories of Dirichlet branes,
which can be defined using the data of the topological closed string theory.
For the B model, this data is a Calabi-Yau K considered as a complex variety.
The obvious category which arises is the category of coherent sheaves on K.
For the A model, the closed string data is a Calabi-Yau K considered as a sym-
plectic manifold. It is also clear from elementary considerations that world-sheet
instantons with disk topology contribute to the operator product algebra. The nat-
ural category which these considerations suggest is the Fukaya category of isotopy
classes of Lagrangian submanifolds, with morphisms given by Floer cohomology.
A naive conjecture of open string mirror symmetry would equate Coh M with
Fuk W on its mirror. However, Kontsevich, who was the first to get this far,
realized that this could not be – the two categories look too different in general.
This observation was subsequently made concrete in many different ways. For
example, mirror symmetry actually relates the B model on W , to a collection of A
models on various CY’s M , M , etc. which are birationally equivalent, but non-
homeomorphic. While Fuk W does not depend on the B model data (the complex
structure), Coh M = Coh M for non-homeomorphic Calabi-Yau’s, so one gets a
contradiction.
This contradiction can be fixed by instead postulating that the category arising
from the topological B model open string is the derived category D(Coh M ), which
(as first suggested by Bondal and Orlov [5]) is equivalent for birationally equivalent
CY’s. This is one way to arrive at Kontsevich’s homological mirror symmetry
conjecture, discussed in detail in the monograph [40].
Physical understanding of this idea came much later. A simpler physical picture
of open string mirror symmetry is to relate it to T-duality. As we discussed, this
naturally acts on torus fibrations, and this can fit with the known action of mirror
symmetry on homology only if these are T 3 fibrations. This picture leads to the
Strominger-Yau-Zaslow formulation of mirror symmetry [53], also discussed in [40],
in which Dirichlet branes wrapping the T 3 on W are mirror to point-like branes on
M.
As explained by Gauntlett, supersymmetry requires the relevant branes on W
to wrap special Lagrangian cycles. This suggests a mirror relation between the
complete set of special Lagrangian cycles on W and the coherent sheaves on M .
However, early pictures of this relation relied too much on properties of the large
volume approximation, and it seems fair to say that the full picture has not yet been
spelled out. Making this full picture will clearly require a good deal of math-physics
interaction, as indicated by a few results we now mention.
One of the simpler mathematical consequence of this symmetry, discussed in
Szendrői’s contribution, is a mirror relation between Fourier-Mukai transforms,
which realize the autoequivalences on D(Coh M ), and symplectomorphisms on W .
Both transformations are “generalized T-dualities” in physics language, but the
first is non-geometric, further demonstrating that D(Coh M ) must play a physical
role. This has been better understood, along lines discussed by Douglas in [22] and
to be explained in [40].
Another small volume subtlety is discussed in Kapustin’s contribution: to get
mirror symmetry in general, the Fukaya category must be enlarged to include
coisotropic submanifolds, objects which can also be shown to be supersymmetric
Dirichlet branes.
5.6. Resolution of orbifold singularities. The simplest non-trivial non-

commutative algebras of this sort arise as quotients of Cn by a discrete group
action. Taking infinite groups provides the rich set of examples discussed in [9],
but quotients by finite groups are very interesting as well, especially in the context
of algebraic geometry, where one has well understood methods to resolve and study
the resulting orbifold singularities.
Since quotients are natural, one would expect that all reasonable definitions of
the quotient will agree. This is certainly borne out by the Dirichlet brane example,
in which orbifolding leads directly to the concrete quiver theories which arose in the
study of the McKay and generalized McKay correspondence [48]. This connection
allowed immediately generalizing Witten’s small instanton work to gauge fields on
orbifold resolutions [19], and subsequent physical work has made fairly detailed
contact with the mathematical work on the McKay correspondence, explained in
contributions by Craw and Ishii to this volume. All of these topics will be discussed
at length in [40].
6. The duality revolution

So far, we discussed the new geometry associated with ls . The problem of
understanding quantum gs corrections is much more far-reaching, and much more
difficult. Clearly one should start by understanding the gs deformation in the
absence of ls corrections.
6.1. Quantum supersymmetric field theory. The typical problem here is

to start with a classical Lagrangian, not including gs corrections, and compute the
exact answer with gs corrections. Many such problems are discussed in the lectures
of Dorey and Acharya. Traditionally, one starts by showing that the quantity of
interest receives either no corrections in perturbation theory (say the superpotential

in d = 4, N = 1 gauge theory), or very specific corrections (say the metric on C
in N = 2 gauge theory, for which the only correction in the gs expansion is the
“one-loop” correction independent of gs ). This allows determining its behavior
in certain limits and, using an intricate combination of symmetry arguments and
global constraints on holomorphic functions, this can in simple cases be pushed to
get exact results.
Perhaps the most famous example is the Seiberg-Witten solution of four-dim-
ensional N = 2 (or N s = 8 in our conventions) gauge theory. In the language of
N = 1 effective field theory, this theory has C ∼ = G C (the complexified Lie algebra)
with the adjoint action of G, and the space of vacua is C //G ∼ = T C / Weyl(G) (the
complexified Cartan subalgebra modulo the Weyl group). In the “exact solution”
(meaning the exact low energy effective field theory; almost no other observables
are computable at present), one learns (essentially) two things.
First, one learns the exact metric on C, which (much as in mirror symmetry)
is given by an infinite series summing instanton contributions (in a tour de force
work [44], this was recently checked against the direct computation).
Second, one learns the locus in C on which new charged particles become mass-
less. These are the magnetic monopoles and dyons visible as solitonic solutions at
weak coupling, and this is the result which led to the formulation of the Seiberg-
Witten equations: they are the equations of motion for the effective field theory
of the monopole, and as such “must” (when suitably interpreted) reproduce the
Donaldson invariants. The argument is that since these invariants are “topologi-
cal” and do not depend on the metric, they cannot depend on the overall scale of
energy, therefore they must be the same for the effective field theory as they were
in the full quantum field theory.
Physically, this result determines the phase structure of the theory: in partic-
ular, the related N = 1 theory obtained by adding a quadratic superpotential is
in the confining phase. These ideas can be pushed further to compute the effective
superpotential as a functional of the original superpotential (taken to be a general
G-invariant function on C). Indeed, many more exact N = 1 superpotentials are
known for simple cases.
If we add a cubic or higher order superpotential, something even more unusual
happens: the result is a “non-trivial fixed point theory” [1], a theory with quantum
fluctuations at all length scales, analogous to the models of critical phenomena.
Another example can be obtained by coupling matter in sufficiently many copies
of the fundamental representation of G [49], and there is a whole zoo of further
examples. Such theories do not behave classically in the low energy limit and thus
are not well described by effective field theory; one needs more elaborate tools, more
analogous to those of two-dimensional conformal field theory (indeed, it is generally
believed that fixed point theories in any dimension are conformally invariant).
Shortly after the school, a general prescription for computing exact N = 1
superpotentials at finite gs was found by Dijkgraaf and Vafa [15]. It states that
the quantum superpotential can be obtained by a Legendre-type transform of the
result of an integral over C treated as a zero-dimensional (hence, matrix model)
integral.
6.2. The web of dualities. We turn to discuss the entire string theory at
finite gs . The remarkable and still amazing discovery of the “second superstring
revolution,” is that these theories are simple not just at small gs , but also in the
limit of very large gs .
The prototypical example is the relation between IIa superstring theory in
ten dimensions and eleven-dimensional supergravity. Naively, the IIa theory has a
dimensionless parameter, the string coupling gs . In fact, this parameter must be
thought of as the expectation value of a field, the dilaton. As it becomes large, the
theory actually becomes equivalent to eleven-dimensional supergravity compactified
on K ∼
3/2
= S 1 , of radius R ∼ gs .
Conversely, the spectrum of IIa supergravity can be obtained from eleven-
dimensional supergravity by Kaluza-Klein reduction on S 1 . Even the superstring
can be obtained by wrapping the membrane on S 1 , i.e. embedding one of the two
spatial dimensions with the S 1 , one obtains an object which looks like a string. In
the small radius limit, the claim is that this becomes the string.
Arguing in this vein, one concludes that the well-definedness of superstring
theory implies the existence of a “completion” of eleven-dimensional supergravity,
which makes sense at all energy scales, and contains the membrane and a (5 + 1)-
dimensional solution called the “fivebrane.” This theory is generally called “M
theory.” It turns out that similar arguments can be used to connect the IIb, type I
and heterotic string theories (suitably compactified), and this connected web of dual
theories is sometimes also called “M theory.” Evidently any starting point would
have this property, and thus the more democratic and inclusive term “string/M
theory” is also used.
While we do not have space to treat superstring duality in the depth it deserves,
the interested reader will find many overviews which do in [55] and references there.
6.3. Branes and singularities. A point in the preceding discussion worthy

of emphasis is the role of the spectrum of extended objects, or branes. This en-
tered into our discussion of IIa-M theory duality, and can be generalized into the
claim that, in a limit in which a wrapped brane becomes light, it will become a
fundamental object in the dual theory. So, for example, the wrapped fivebrane on
M theory compactified on the K3 manifold turns out to have the spectrum of the
heterotic string compactified on T 3 (both seven-dimensional theories); thus these
theories are dual.
Another variation on this argument is to consider a compactification with non-
trivial cycles of small volume. In this case, the wrapped branes will lead to new
localized degrees of freedom. This situation can be arranged by compactifying on
a singular manifold and then resolving or deforming it slightly. Depending on the
spectrum of branes, this can lead to instanton effects and a quantum deformation
of the effective field theory, or it can lead to new light fields in the effective field
theory.
To review the case of the fundamental string (which counts as a brane for this
argument), the first effect is seen in mirror symmetry (world-sheet instantons asso-
ciated with small two-cycles), while the second is seen in T-duality (new “winding
states”) associated with π1 .
A central ingredient in generalizing these results to higher-dimensional branes
is calibrated geometry. As explained in Gauntlett’s lectures, this provides the
conditions for a wrapped brane to preserve supersymmetry. It can be generalized
to compactifications with the non-metric fields (or “fluxes”) turned on as well.
Of the many other cases, a particularly interesting one is the resolution of a

canonical (or ADE) two-dimensional complex singularity, locally described by the
orbifold C2 /Γ for Γ ∈ SU (2). In a theory with a membrane, such as M theory, this
leads to an effective field theory with enhanced gauge symmetry, with the corre-
sponding ADE gauge group. In a bit more detail, the three-form gauge potential
leads to the Cartan subalgebra of the gauge symmetry, while the wrapped mem-
branes lead to the non-Abelian gauge bosons. This beautiful physical connection
between two interpretations of ADE has numerous generalizations, for example to
compactification on threefolds [32, 20].
Furthermore, it is perhaps the simplest example we know of for which the
basic questions of “the geometry of string theory” have not been satisfactorily
answered. A “correct stringy geometric” description of the singularity would make
this G-symmetry manifest, and predict the corresponding behaviors in families, for
higher-dimensional singularities, and so on. No fitting mathematicization or even
summary of the somewhat ad hoc physical arguments has yet emerged.
6.4. G2 and other supersymmetric compactifications. The basic anal-

ysis of supersymmetry we gave in section 4 of course has a role for most of the
special holonomies. For example, compactification on a manifold of G2 holonomy
preserves 1/8 of supersymmetry. This is particularly interesting starting from M
theory, as one then gets N = 1 supersymmetry in four dimensions, and candidate
physical models.
While we could have begun the discussion of G2 compactification in an earlier
section, it only becomes physically interesting when one considers compactification
on singular manifolds. This is because an easy index theorem argument shows that
on a smooth G2 manifold the matter must form real G-representations. On the
other hand, the Standard Model spectrum (3.1) involves a complex representation
(physically, one says the Standard Model has chiral fermions).
The arguments of the last section tell us how to get non-Abelian gauge sym-
metry; we need to compactify on a G2 manifold with an ADE singularity. This
will lead to non-Abelian gauge symmetry localized in real codimension four. As
discussed by Acharya, the simplest way to get chiral fermions is to have two such
singularities intersect at a point (the metric behavior near the singularity turns out
to allow this).
The novelty of these results has led to a lot of interest in G2 compactification,
but it must be admitted that the field is difficult as the appropriate mathematical
foundations barely exist. The work of Joyce, reviewed in this volume, is of course the
necessary starting point, and further discussion can be found in the contributions
of Hitchin and Kovalev as well as Acharya.
6.5. AdS/CFT. One of the most beautiful, yet startling, results of super-
string duality was the AdS/CFT correspondence, discovered by Maldacena [36].
To put the simplest example in a nutshell, N = 4 supersymmetric Yang-Mills the-
ory in four dimensions is dual to ten-dimensional IIb supergravity compactified on
S 5 . This duality between theories of different dimension will obviously require some
explanation; we refer to the review [2] and the lectures of Acharya and Gauntlett
in this volume.
As the name suggests, the central point is that (D + 1)-dimensional anti-de
Sitter space-time admits as symmetry group SO(D, 2), which is the same as the
group of conformal isometries of D-dimensional Minkowski space-time. Thus, the

simplest geometric questions which appear in this context involve string compact-
ifications which lead to anti-de Sitter space, and variations of these solutions of
supergravity.
In a real sense this duality goes beyond geometry, instead describing a sense
in which geometry can emerge from “pre-geometrical” ingredients. Indeed, for a
variety of reasons, many physicists have long suspected that a full theory of quantum
gravity would not have the same relation to geometry as classical general relativity.
After all, if the metric is a fluctuating quantum variable, it becomes difficult to
formulate the standard geometric observables such as geodesic or holonomy. On
the other hand, one might imagine that other, simpler observables might appear.
In AdS/CFT, these are boundary conditions on the fields in anti-de Sitter space,
dual to couplings in the gauge theory.
Some more mathematical papers inspired by these ideas include [62, 37].
7. Some reflections
As we have seen, string theory is a very rich subject, and the project of under-
standing it mathematically has barely begun. Most of us who attended the school,
and many readers, would happily agree with the claim that this project is destined
to be interesting for both physics and mathematics, and does not require further
justification. Let me nevertheless offer a few thoughts on where this interaction is
going, and what we can hope will come out of it.
We devoted most of our attention to the “classic” problem of string compactifi-
cation, to explain or at least reproduce the physics we have seen in experiments and
observations. While the original scenarios for this involved a fairly direct (though
ingenious) generalization of the Kaluza-Klein idea, during the 1980’s string theo-
rists came to realize that string theory was not adequately described by existing
notions of geometry, due to the ls and gs corrections. This led to the idea that
string theory would lead to completely new ideas of geometry.
On the other hand, the progress on this problem over the years has largely
come through a more modest agenda: one works on a subproblem which can be
formulated in geometric terms, but for which string theory provides a “twist.”
For example, both sides of closed string mirror symmetry can be formulated in
these terms, counting curves and varying Hodge structures; the twist is to connect
these seemingly very different problems. A related but more recent example is
the description of Dirichlet branes as objects in the derived category of coherent
sheaves, a notion formulated by mathematicians, but satisfying a modified stability
condition motivated by string theory, which incorporates the ls corrections [22].
Such developments seem very likely to continue, and this should be to the
benefit of both sides, but at present seem of more clear interest to mathematicians
than to physicists. Indeed, for those who feel string theory has something to say
about the real world, the following quote may be telling:
There is probably less difference between the positions of a math-
ematician and of a physicist than is generally supposed, and the
most important seems to me to be this, that the mathematician is
in much more direct contact with reality.
This was from G. H. Hardy’s A Mathematician’s Apology, but it seems to me
strangely appropriate as a description of the string theorists’ present situation.
Of course, Dirac is particularly well known for the point of view that the beauty
of equations is itself evidence for their correctness, but Hardy’s point is also well
taken: the mathematician knows what he (or she) is talking about, and in the act
of precisely formulating it makes it real, at least real as mathematics.2
String theory has far passed the threshold of convincing us that it is real in
this sense, despite our inability to precisely formulate it. On the other hand, it is
still hard to say how we will be convinced that it describes the physical world. It
could be that strings will be discovered in accelerators, but if not, it seems very
possible that we will have to live for some time with the argument that it is the
only candidate which is not “obviously wrong.”
One of the striking points which the school brought home to me is how different
the cultures of mathematics and physics are. It is interesting to ask whether or
not this interaction will ultimately lead to a merging of the two, or perhaps the
formation of some sort of intermediate culture.
One of the well-known differences in culture is the attitude towards generaliza-
tion. Physicists are always happier to consider examples, and feel that these bring
home the essential “point” of an idea or development much better than any abstract
treatment. On the other hand, while mathematicians regard examples as essential
as well, one of the defining characteristics of mathematics is the development of
general theory.
This is a very deep difference. It comes to the fore when one tries to prove
that some property or attribute of the objects under consideration is not possible,
as most such proofs require a general definition which does not presuppose the ex-
istence of an object with this property. In some cases, one can make reductio ad
absurdum proofs, but on reflection one realizes that this presupposes a far better
understanding of the objects under discussion than physicists usually can get. In-
deed, there is a saying among physicists that “no-go” theorems, which state that
some desired property or construction is impossible, would better be renamed “go-
go” theorems, as there are so many examples of loopholes or other ways around the
supposed impossibility. While this is more an illustration of selective memory than
anything else, it shows the deep distrust of physicists for general arguments.
If we ever attain a reasonably complete formulation of string/M theory, this
attitude will have to change. After all, by definition, we will have reached the point
where the axioms are not just obstacles to work one’s way around, but inevitable
constraints on the possibilities. Indeed, we must hope that these constraints are
strong enough in the end to enable us to make predictions. The most basic test of
whether a theory has content is whether it can be wrong or, more precisely, falsi-
fied by the observations. Assuming that physicists do discover supersymmetry and
additional phenomena, but only those which can be described by four-dimensional
effective field theory, the primary question which string theory will need to answer
is: can the effective field theory of our world arise from string theory? One can
answer this question by finding the “correct compactification,” but this is mean-
ingless if any effective theory can be obtained from some compactification. From
this point of view, the question of showing that certain possibilities cannot come
out of string theory is just as important as showing that the one which describes
2On the other hand, Jerome Gauntlett reminded me of another quote: “Mathematics may
be defined as the subject where we never know what we are talking about, nor whether what we
are saying is true.” (Bertrand Russell, Mysticism and Logic).
our universe can arise. This is at heart a mathematical question, in the sense I just
described, and it seems to me that physicists will have to adopt something of the
spirit of the mathematicians to answer it.
Comparing my colleagues’ attitudes with those I recall from my graduate stud-
ies in the early 80’s, I can only say that this process has come a long way already,
and probably has farther to go. As to whether mathematicians are being influenced
in equally deep ways, I leave for them, and especially the students in this school,
to tell me.
References
1. P. C. Argyres and M. R. Douglas, New phenomena in SU(3) supersymmetric gauge theory,
Nucl. Phys. B 448, 93 (1995) [arXiv:hep-th/9505062].
2. O. Aharony, S. S. Gubser, J. M. Maldacena, H. Ooguri, and Y. Oz, Large N field theories,
string theory and gravity, Phys. Rept. 323 (2000), 183–386.
3. M. Atiyah, V. Drinfeld, N. Hitchin and Y. Manin, Construction of instantons, Phys. Lett.
A65:185, 1978.
4. T. Banks, M. Dine and E. Gorbatov, Is there a string theory landscape?, arXiv:hep-
th/0309170.
5. A. Bondal and D. Orlov, Derived Categories of Coherent Sheaves, in Proceedings of ICM
2002, vol. 2, pp. 47–56, Higher Education Press, Beijing.
6. P. Candelas, G. T. Horowitz, A. Strominger and E. Witten, Vacuum configurations for
superstrings, Nucl. Phys. B. 258:46-74, 1985.
7. P. Candelas, X. C. De La Ossa, P. S. Green and L. Parkes, A Pair Of Calabi-Yau Manifolds
As An Exactly Soluble Superconformal Theory, Nucl. Phys. B 359, 21 (1991).
8. A. S. Cattaneo and G. Felder, A path integral approach to the Kontsevich quantization
formula, Commun. Math. Phys. 212, 591 (2000) [arXiv:math.qa/9902090].
9. A. Connes, Noncommutative Geometry, 1994, Academic Press, San Diego.
10. A. Connes, M. R. Douglas and A. Schwarz, Noncommutative Geometry and Matrix Theory:
Compactification on Tori, JHEP 9802 (1998) 003, [arXiv:hep-th/9711162].
11. E. Cremmer, B. Julia, and J. Scherk, Supergravity theory in 11 dimensions, Phys. Lett. B76
(1978), 409–412.
12. J. Dai, R. G. Leigh and J. Polchinski, New Connections Between String Theories, Mod.
Phys. Lett. A 4, 2073 (1989).
13. P. Deligne and J. W. Morgan, Notes on Supersymmetry (following J. Bernstein), in Quantum
Fields and Strings: A course for Mathematicians, eds. Deligne et al, pp. 41–97, 2 vols, AMS,
Providence, 1999.
14. R. Dijkgraaf, Fields, strings and duality, in Les Houches 1995, Quantum symmetries, eds.
A. Connes, K. Gawȩdzki and J. Zinn-Justin, pp. 3–148, North Holland, 1998 [arXiv:hep-
th/9703136].
15. R. Dijkgraaf and C. Vafa, A perturbative window into non-perturbative physics, [arXiv:hep-
th/0208048].
16. L. Dixon, in Superstrings, Unified Theories and Cosmology 1987, eds. G. Furlan et al, World
Scientific 1988.
17. S. K. Donaldson and P. B. Kronheimer, The Geometry of Four-Manifolds, Oxford Univ.
Press, 1990.
18. M. R. Douglas, Branes within branes, in Strings, Branes and Dualities, eds. L. Baulieu et
al., pp. 267–276, Kluwer 1999. [arXiv:hep-th/9512077]
19. M. R. Douglas and G. Moore, D-branes, Quivers, and ALE Instantons, [arXiv:hep-
th/9603167].
20. M. R. Douglas, S. Katz and C. Vafa, Small instantons, del Pezzo surfaces and type I theory,
21. M. R. Douglas and C. Hull, D-branes and the Noncommutative Torus, JHEP 9802 (1998)
008, [arXiv:hep-th/9711165].
22. M. R. Douglas, Dirichlet branes, homological mirror symmetry, and stability, in Proceedings
of ICM 2002, op. cit., vol. 3, pp. 395–408, [arXiv:math.AG/0207021].
23. R. G. Douglas, private communication.
24. D. S. Freed, Five Lectures on Supersymmetry, AMS, Providence, 1999.

25. P. G. Freund and M. A. Rubin, Dynamics Of Dimensional Reduction, Phys. Lett. B 97, 233
(1980).
26. D. H. Friedan, Nonlinear Models In Two + Dimensions, Annals Phys. 163, 318 (1985).
27. H. Georgi, H. R. Quinn and S. Weinberg, Hierarchy Of Interactions In Unified Gauge The-
ories, Phys. Rev. Lett. 33, 451 (1974).
28. D. Gepner, Space-time supersymmetry in compactified string theory and superconformal
models, Nucl. Phys. B296 (1988) 757.
29. B. R. Greene, Lectures on the Quantum Geometry of String Theory, in Les Houches 1995,
Quantum symmetries, op. cit.
30. B. R. Greene and M. R. Plesser, Duality In Calabi-Yau Moduli Space, Nucl. Phys. B 338,
15 (1990).
31. M. T. Grisaru, A. E. van de Ven and D. Zanon, Four-Loop Divergences For The N=1
Supersymmetric Nonlinear Sigma Model In Two-Dimensions, Nucl. Phys. B 277, 409 (1986).
32. K. A. Intriligator, D. R. Morrison and N. Seiberg, Five-dimensional supersymmetric gauge
theories and degenerations of Calabi-Yau spaces, Nucl. Phys. B 497, 56 (1997) [arXiv:hep-
th/9702198].
33. M. Kontsevich, Homological Algebra of Mirror Symmetry, Proceedings of the 1994 ICM,
[arXiv:alg-geom/9411018].
34. M. Kontsevich, Deformation quantization of Poisson manifolds, I, [arXiv:q-alg/9709040].
35. W. Lerche, C. Vafa and N. P. Warner, Chiral Rings In N=2 Superconformal Theories, Nucl.
Phys. B 324, 427 (1989).
36. J. M. Maldacena, The large N limit of superconformal field theories and supergravity, Adv.
Theor. Math. Phys. 2 (1998), 231–252.
37. Y. I. Manin and M. Marcolli, Holography principle and arithmetic of algebraic curves, Adv.
Theor. Math. Phys. 5, 617 (2002) [arXiv:hep-th/0201036].
38. M. Mariño, R. Minasian, G. W. Moore, and A. Strominger, Nonlinear instantons from
supersymmetric p-branes, JHEP 01 (2000), 005.
39. Mirror Symmetry, proceedings of the 2000 Clay School, eds. K. Hori et al, AMS 2003.
40. Mirror Symmetry II, proceedings of the 2002 Clay School, eds. P. Aspinwall et al, in prepa-
ration.
41. G. W. Moore and N. Seiberg, Classical And Quantum Conformal Field Theory, Commun.
Math. Phys. 123, 177 (1989).
42. G. W. Moore and E. Witten, Integration over the u-plane in Donaldson theory, Adv. Theor.
Math. Phys. 1, 298 (1998) [arXiv:hep-th/9709193].
43. J. W. Morgan, The Seiberg-Witten equations and applications to the topology of smooth
four-manifolds, Princeton Univ. Press, 1996.
44. N. A. Nekrasov, Seiberg-Witten prepotential from instanton counting, in Proceedings of ICM
2002, op. cit., vol. 3, pp. 477-496, [arXiv:hep-th/0306211].
45. D. Nemeschansky and A. Sen, Conformal Invariance Of Supersymmetric Sigma Models On
Calabi-Yau Manifolds, Phys. Lett. B 178, 365 (1986).
46. G. Perelman, The entropy formula for the Ricci flow and its geometric applications,
[arXiv:math.DG/0211159].
47. Proceedings of ICM 2002, ed. Li Tatsien, 3 vols., Higher Education Press, Beijing, 2002.
48. M. Reid, La correspondance de McKay, Séminaire Bourbaki (novembre 1999), no. 867,
[arXiv:math.AG/9911165].
49. N. Seiberg, Electric - magnetic duality in supersymmetric non-Abelian gauge theories, Nucl.
Phys. B 435, 129 (1995) [arXiv:hep-th/9411149].
50. N. Seiberg and E. Witten, E lectric-magnetic duality, monopole condensation, and confine-
ment in N=2 supersymmetric Yang-Mills theory, Nucl. Phys. B 426, 19 (1994) [Erratum-ibid.
B 430, 485 (1994)] [arXiv:hep-th/9407087].
51. N. Seiberg and E. Witten, String Theory and Noncommutative Geometry, JHEP 9909 (1999)
032, [arXiv:hep-th/9908142].
52. V. Schomerus, D-branes and Deformation Quantization, JHEP 9906 (1999) 030, [arXiv:hep-
th/9903205].
53. A. Strominger, S. T. Yau and E. Zaslow, Mirror symmetry is T-duality, Nucl. Phys. B 479,
243 (1996) [arXiv:hep-th/9606040].
54. R.P. Thomas, Examples of bundles on Calabi-Yau 3-folds for string theory compactifications,
Adv. Theor. Math. Phys. 4 (2000) 231-247.
55. Unity from Duality: Gravity, gauge theory and strings, eds. C. Bachas et al, proceedings of
the 2001 Les Houches Advanced Study Institute, Springer, 2002.
56. Vertex Operator Algebras in Mathematics and Physics, eds. Y.-Z. Huang et al, Fields Insti-
tute Communications, AMS, Providence, 2003.
57. E. Witten, Noncommutative geometry and string field theory, Nucl. Phys. B268 (1986) 253.
58. E. Witten, Topological Quantum Field Theory, Commun. Math. Phys. 117 (1988) 353.
59. E. Witten, Phases of N = 2 theories in two dimensions, Nucl. Phys. B 403, 159 (1993)
[arXiv:hep-th/9301042].
60. E. Witten, Monopoles and four-manifolds, Math. Res. Lett. 1, 769 (1994) [arXiv:hep-
th/9411102].
61. E. Witten, Small Instantons in String Theory, Nucl. Phys. B. 460, 541 (1996) [arXiv:hep-
th/9511030].
62. E. Witten and S. T. Yau, Connectedness of the boundary in the AdS/CFT correspondence,
Adv. Theor. Math. Phys. 3, 1635 (1999) [arXiv:hep-th/9910245].
63. E. Witten, Deconstruction, G2 Holonomy, and Doublet-Triplet Splitting, [arXiv:hep-
ph/0201018].
64. E. Witten, Singularities in String Theory, in Proceedings of ICM 2002, op. cit., vol. 1, pp.
495-506.
65. C. N. Yang, as discussed in Selected Papers 1945–1980, pp. 73–76, W. H. Freeman and Co.,
New York, 1983.
Department of Physics and Astronomy, Rutgers University, Piscataway, NJ 08855-

0849, USA, Isaac Newton Institute for Mathematical Sciences, Cambridge, CB3 0EH,
U.K., I.H.E.S., Le Bois-Marie, Bures-sur-Yvette, 91440 France
E-mail address: mrd@physics.rutgers.edu
Volume 3, 2004
M theory, G2 -manifolds and Four Dimensional Physics
Bobby S. Acharya
Abstract. M theory on a manifold X of G2 -holonomy is a natural framework

for obtaining vacua with four large spacetime dimensions and N = 1 supersym-
metry. The standard features of particle physics, namely non-Abelian gauge
symmetries and chiral fermions, emerge from singularities of X. The aim of
these lectures is to describe in detail how the above picture emerges. Along
the way we will see how interesting aspects of strongly coupled gauge theo-
ries, such as confinement and mass gap, receive relatively simple explanations
within the context of M theory. All of the singularities of G2 -manifolds we
discuss here are are constructed naturally from the familiar ADE-singularities.
For instance, codimension seven singular points which support chiral fermions
are obtained from a natural modification of the Kronheimer construction of
ALE-spaces.
1. Introdution
Supersymmetry is one of our best candidates for physics beyond the Standard
Model. M theory goes further in the sense that it is supersymmetric, contains
gravity and is quantum mechanically consistent. Since the theory is formulated
on spacetimes with eleven dimensions, a natural question to ask is whether there
are vacua of M theory with four macroscopic spacetime dimensions and a realistic
particle physics spectrum? Since supersymmetry is intrinsic to M theory, it is
perhaps more natural to look for vacua with supersymmetric particle physics in
four dimensions. Since non-minimal or extended supersymmetry in four dimensions
cannot accommodate chiral fermions, to answer this question we should really be
studying M theory vacua with N = 1 supersymmetry.
There are two (to date) natural looking ways to obtain four large spacetime
dimensions with N = 1 supersymmetry from M theory. Both of these require the
seven extra dimensions to form a manifold X whose metric obeys certain properties.
The first consists of taking X to be a manifold whose boundary ∂X is a Calabi-Yau
threefold [15]. The second possibility, which will be the subject of these lectures is
to take X to be a manifold of G2 -holonomy1. We will explain what this means in
section two.
2000 Mathematics Subject Classification. Primary 81T30, 83E30, 53C29, 81T60.

Key words and phrases. M theory, G2 -holonomy.
1We will often refer to X as a G -manifold.
2
2004
31
32 BOBBY S. ACHARYA
If X is large compared to the Planck scale (the only scale in M theory)

and smooth, then at low energies a good approximation is provided by eleven-
dimensional supergravity. Compactifications of the latter have been studied for
several decades. See [13] for a review. Unfortunately, Witten proved that none of
these could give rise to chiral fermions [22]. However, this does not mean that G2 -
manifolds are useless for obtaining models of particle physics from a fundamental
theory. This is because we have learned in recent years that additional light de-
grees of freedom can be “hidden” at singularities of X. These are typically branes
wrapped on submanifolds of X which have shrunk at the singularity2. In M theory
these are either the M 2-brane or the M 5-brane. This can provide a novel picture
of conventional field theory dynamics and can even lead to new theories. The su-
pergravity approximation breaks down at such singularities and the analysis of [22]
no longer applies.
Within the past couple of years there has been a tremendous amount of progress
in understanding M theory physics near singularities in manifolds of G2 -holonomy
[4 − 10]. In particular we now understand at which kinds of singularities in G2 -
manifolds the basic requisites of the standard model - non-Abelian gauge groups
and chiral fermions - are to be found. The purpose of these lectures will be to
explain how this picture was developed in detail. Along the way we will see how
important properties of strongly coupled gauge theories such as confinement and
the mass gap can receive a semi-classical description in M theory on G2 -manifolds.
The subject of these lectures may also be of interest to mathematicians. One
compelling mathematical theme is ADE-singularities. As we will see, all of the
singularities of interest here are built out of the basic ADE orbifold singularities.
Yang-Mills fields in four dimensions emerge by considering singularities of X param-
eterised by an associative 3-manifold W ⊂ X, near which X looks like C2 /ΓADE ×W .
Chiral fermions emerge from additional singular points in X through which W
passes and near which the description of X is given by a simple modification of
the Kronheimer construction of ADE-singularities: one simply picks a U (1) in the
Kronheimer gauge group and does not set its corresponding moment map to zero.
One then obtains a 7-manifold which fibers over R3 (given by the set of values of
this privileged moment map) and the fibers of this map are ADE-singularities. At
the origin in R3 the 7-manifold becomes more singular since some collection of S 2 ’s
in the fibers collapse at that point.
Another point which may also be of interest to mathematicians - especially
to those interested in mirror symmetry - will be a duality which we exploit here
between M theory on K3-fibered G2 -manifolds and heterotic string theory on T 3 -
fibered Calabi-Yau threefolds. On the string theory side the threefold is endowed
with a Hermitian-Yang-Mills connection A and chiral fermions emerge from zero
modes of the Dirac operator twisted by A. On the M theory side this gets mapped
to a statement about the singularities of X, which is how the aforementioned de-
scription of singularities is obtained.
By the end of these lectures, we should have a good and clear picture of the basic
properties that X should have in order that M theory on X produces something
like a realistic model of particle physics in four dimensions. An important problem
2A more conventional example of light states at a singularity is provided by string theory on

an orbifold. Typically one finds extra light states confined to the singularity. These arise in the
so-called twisted sectors.
M THEORY, G2 -MANIFOLDS AND FOUR DIMENSIONAL PHYSICS 33
- which we would like to pose to the mathematicians - is to construct compact

G2 -manifolds with these properties.
At the beginning of each main section we will offer a section summary. In
section two we derive the basic properties of M theory when X is large and smooth.
We also derive some basic properties of G2 -manifolds. Section three explains how
classical supersymmetric Yang-Mills theory can be obtained from M theory on a
singular G2 -manifold X. We describe these singularities in detail. Section four
describes how quantum properties of the Yang-Mills theory, confinement and a
mass gap, can be understood from M theory. The reason that this can be done
successfully is that M theory contains semi-classical limits which are not present
in the quantum gauge theory. Having understood how non-Abelian gauge groups
emerge, section five goes on to describe how additional singularities of X give rise
to chiral fermions.
2. Supersymmetry, G2 -holonomy and Kaluza Klein spectrum.

In this section we will describe in detail why G2 -holonomy manifolds natu-
rally emerge in the context of supersymmetric M theory compactification. We will
describe some of the basic properties of G2 -manifolds. We will then discuss the
Kaluza-Klein spectrum of M theory on a large and smooth G2 -manifold.
2.1. Supersymmetry and G2 -holonomy. At low energies M theory admits

a description in terms of eleven-dimensional supergravity. This description is valid
on smooth spacetimes whose smallest length scale is much larger than the eleven-
dimensional Planck length. The supergravity contains three fields, a metric g, a
three-form potential C and a gravitino Ψ. In addition to being generally covariant
and supersymmetric, the theory has a gauge invariance under which
(2.1) δC = dλ
with λ a 2-form, so the gauge invariant field is the derivative of C, denoted by G.
The action for the bosonic fields is of the form

√ 1 1
(2.2) S= gR − G ∧ ∗G − C ∧ G ∧ G.
2 6
The equations of motion for C and g are of the form,
(2.3) d∗G=G∧G
and
1
(2.4) RM N − gM N R = TM N (C)
2
where T is the energy-momentum tensor for the C-field.
Since the theory is supersymmetric, it is natural to look for supersymmetric
vacua. In the classical theory these are just the conditions that the supersymmetry
variations of the three fields vanish. In a Lorentz invariant background the expecta-
tion value of Ψ is zero, in which case the variations of g and C vanish automatically.
In order to find classically supersymmetric field configurations we must find values
of C and g for which the variation of Ψ is zero:
1
(2.5) δη ΨM ≡ ∇M η + (ΓP QRS GP QRS − 6ΓP QR GM P QR )η = 0.
288 M
34 BOBBY S. ACHARYA
The simplest way to solve these equations is to take G = 0 in which case we

are looking for 11-manifolds with metric g which admit a covariantly constant or
parallel spinor:
(2.6) ∇M η = 0.
We will re-write this equation in the more symbolic form,
(2.7) ∇g η = 0
where by ∇g we mean the Levi-Civita connection constructed from g. Solutions to

these conditions can be classified via the holonomy group of the connection ∇g .
The holonomy group of a connection acting on a field like η or a vector field
can be understood in terms of parallel transport. One takes a closed loop on the
manifold and literally transports the field around it. In the case of the Levi-Civita
connection, the field comes back to itself up to a rotation in SO(n) in the case of
Riemannian n-manifolds or SO(10, 1) in the case of M theory. The set of all such
rotations based at some point on the manifold generates a group, the holonomy
group of ∇g , Hol(g). If g is sufficiently generic then Hol(g) = SO(10, 1). However,
if we take special choices of g, then Hol(g) can be a proper subgroup of SO(10, 1).
For instance, if g is such that parallel spinors (supersymmetries) can be found, then
η is a field which undergoes no parallel transport at all and therefore Hol(g) must
be a subgroup of SO(10, 1) for which there are spinors in the trivial representation.
We will concern ourselves with compactifications of M theory to four dimen-
sions on a 7-manifold X. More precisely we will take the eleven-manifold to be a
product X×R3,1 with g a product metric of a metric g(X) and the Minkowski met-
ric on R3,1 . With this choice of eleven-metric, we have explicitly broken SO(10, 1)
to SO(7)×SO(3, 1). The second factor now plays the role of the Lorentz group
of the compactified theory. The conditions for supersymmetry can be satisfied by
taking g(X) to be such that it admits a spinor θ obeying
(2.8) ∇g(X) θ = 0
and choosing
(2.9) η =θ⊗
with a basis of constant spinors in Minkowski space.

The condition
(2.10) ∇g(X) θ = 0
implies that Hol(g(X)) is G2 or a subgroup. This is because G2 is the maximal

proper subgroup of SO(7) under which the spinor representation contains a singlet.
Specifically, a spinor of SO(7) can be regarded as a fundamental of G2 and a singlet:
(2.11) 8 → 7 + 1.
Therefore, if X is a compact manifold of precisely G2 -holonomy, the effective

theory in four dimensions will be minimally N = 1 supersymmetric. We get pre-
cisely N = 1 and no more because there is only one singlet spinor according to the
above group theory.
2.2. Properties of G2 -manifolds. From the parallel spinor θ we can con-

struct other covariantly constant fields on X. More precisely, any p-form with
components
(2.12) θ T Γi1 i2 ...ip θ
is obviously parallel with respect to ∇g(X) . In fact, since the antisymmetric rep-
resentations of SO(7), when decomposed as representations of G2 , contain singlets
only when p is 0, 3, 4, 7 the above p-forms are non-zero precisely for these values.
The 0-form is just a constant on X. The 7-form is the volume form. Locally the
three-form, which we will conventionally denote by ϕ, can be regarded as a set of
structure constants for the octonion algebra. This stems from the fact that G2
is the automorphism group of the octonion algebra, where we regard the tangent
space at a point on X as a copy of ImO, the imaginary octonions. A specific
representation of ϕ locally is
(2.13) ϕ0 = dx123 + dx145 + dx167 + dx246 − dx257 − dx347 − dx356
where the subscript refers to the fact that we are considering a local model. The
covariantly constant 4-form is the Hodge dual of ϕ, which in the local model is
given by
(2.14) ∗ϕ0 = dx4567 + dx2367 + dx2345 + dx1357 − dx1346 − dx1256 − dx1247 .
In addition to implying that there are other parallel fields on X, the existence
of a parallel spinor (or a G2 -holonomy metric) also has other implications. One of
these is that the metric of G2 -holonomy, g(X), is Ricci-flat. To see this, observe
that the commutator of the covariant derivative is the Riemann curvature. Acting
on θ this implies
1
(2.15) [∇g(X) , ∇g(X) ]mn θ = Rmnpq Γpq θ = 0.
4
Now, contract again with a Γ-matrix to obtain,
(2.16) Γn Γpq Rmnpq θ = 0
The Bianchi identity for Rmnpq which asserts that the components totally antisym-
metric in [npq] are zero then implies that
(2.17) Γj Rij θ = 0
which implies that the Ricci tensor vanishes. This shows that G2 -manifolds obey
the equations of motion of d = 11 supergravity when the 4-form field strength G and
the gravitino are zero: these equations are simply the vacuum Einstein equations.
A final implication - whose proof goes beyond the scope of these lectures but
which can be found in [16] - is that compact manifolds with Hol(g) = G2 have a
finite fundamental group. This implies that the first Betti number vanishes.
2.3. Kaluza-Klein Reduction. At low energies, the eleven-dimensional su-

pergravity approximation is valid when spacetime is smooth and large compared to
the eleven-dimensional Planck length. So, when X is smooth and large enough, we
can obtain an effective four-dimensional description by considering a Kaluza- Klein
analysis of the fields on X. This analysis was first carried out in [20].
In compactification of eleven-dimensional supergravity, massless scalars in four
dimensions can originate from either the metric or the C-field. If g(X) contains k
36 BOBBY S. ACHARYA
parameters, i.e. there is a k-dimensional family of G2 -holonomy metrics on X, then

there will be correspondingly k massless scalars in four dimensions.
The scalars in four dimensions which originate from C arise via the Kaluza-
Klein ansatz,
(2.18) C = ΣI ω I (x)φI (y) + ...
where ω I form a basis for the harmonic 3-forms on X. These are zero modes of
the Laplacian on X and are also closed. There are b3 (X) linearly independent such
forms. The dots refer to further terms in the Kaluza-Klein ansatz which will be
prescribed later. The φI (y) are scalar fields in four-dimensional Minkowski space
with coordinates y. With this ansatz, these scalars are classically massless in four
dimensions. To see this, note that,
(2.19) G = ΣI ω I ∧dφI
and d ∗ G is just
(2.20) d ∗ G = ∗ΣI ω I d ∗ dφI .
Since G∧G vanishes identically, the equations of motion actually assert that the
scalar fields φI are all massless in four dimensions. Thus, the C-field gives rise to
b3 (X) real massless scalars in four dimensions.
In fact it now follows from N = 1 supersymmetry in four dimensions that the
Kaluza-Klein analysis of g will yield an additional b3 (X) scalars in four dimensions.
This is because the superpartners of C should come from g as these fields are
superpartners in eleven dimensions. We should also add that (up to duality trans-
formations) all representations of the N = 1 supersymmetry algebra which contain
one massless real scalar actually contain two scalars in total which combine into
complex scalars. We will now describe how these scalars arise explicitly.
We began with a G2 -holonomy metric g(X) on X. g(X) obeys the vacuum
Einstein equations,
(2.21) Rij (g(X)) = 0.
To obtain the spectrum of modes originating from g we look for fluctuations in
g(X) which also satisfy the vacuum Einstein equations. We take the fluctuations in
g(X) to depend on the four-dimensional coordinates y in Minkowski space. Writing
the fluctuating metric as
(2.22) gij (x) + δgij (x, y)
and expanding to first order in the fluctuation yields the Lichnerowicz equation
(2.23) ∆L δgij ≡ −∇2M δgij − 2Rijmn δg mn + 2R(i
k
δgj)k = 0.
Next we make a Kaluza-Klein ansatz for the fluctuations as
(2.24) δgij = hij (x)ρ(y).
Note that the term ∇ is the square of the full d = 11 covariant derivative. If we
2
separate this term into two terms:

(2.25) ∇2M = ∇2µ + ∇2i ,
then we see that the fluctuations are scalar fields in four dimensions with squared
masses given by the eigenvalues of the Lichnerowicz operator acting on the hij :
(2.26) hij ∇2µ ρ(y) = −(∆L hij )ρ(y) = −λhij ρ(y).
Thus, zero modes of the Lichnerowicz operator give rise to massless scalar fields in
four dimensions. We will now show that we have precisely b3 (X) such zero modes.
On a 7-manifold of SO(7) holonomy, the hij - being symmetric 2-index tensors
- transform in the 27-dimensional representation. Under G2 this representation
remains irreducible. On the other hand, the 3-forms on a G2 -manifold, which are
usually in the 35 of SO(7), decompose under G2 as
(2.27) 35 −→ 1 + 7 + 27.
Thus, the hij can also be regarded as 3-forms on X. Explicitly,
(2.28) ϕn[pq hnr] = ωpqr .
The ω’s are 3-forms in the same representation as hij since ϕ is in the trivial
representation. The condition that h is a zero mode of ∆L is equivalent to ω being
a zero mode of the Laplacian:
(2.29) ∆L h = 0 ↔ ∆ω = 0.
This shows that there are precisely b3 (X) additional massless scalar fields coming
from the fluctuations of the G2 -holonomy metric on X.
As we mentioned above, these scalars combine with the φ’s to give b3 (X) mass-
less scalars, ΦI (y), which are the lowest components of massless chiral superfields
in four dimensions. There is a very natural formula for the complex scalars ΦI (y).
Introduce a basis αI for the third homology group of X, H3 (X, R). This is a basis
for the noncontractible holes in X of dimension three. We can choose the αI so
that

(2.30) ω J = δIJ .
αI
Since the fluctuating G2 -structure is

(2.31) ϕ = ϕ + δϕ = ϕ + ΣI ρI (y)ω I (x),
we learn that

(2.32) I
Φ (y) = ϕ + iC.
αI
The fluctuations of the four-dimensional Minkowski metric give us the usual

fluctuations of four-dimensional gravity, which due to supersymmetry implies that
the four-dimensional theory is locally supersymmetric.
In addition to the massless chiral multiplets, we also get massless vector multi-
plets. The bosonic component of such a mulitplet is a massless Abelian gauge field
which arises from the C-field through the Kaluza-Klein ansatz,
(2.33) C = Σα β α (x) ∧ Aα (y)
where the β’s are a basis for the harmonic 2-forms and the A’s are one-forms in
Minkowski space, i.e. Abelian gauge fields. Again, the equations of motion for C
imply that the A’s are massless in four dimensions. This gives b2 (X) such gauge
fields. As with the chiral multiplets above, the fermionic superpartners of the
gauge fields arise from the gravitino field. Note that we could have also included
an ansatz giving 2-forms in four dimensions by summing over harmonic 1-forms on
X. However, since b1 (X) = 0, this does not produce any new massless fields in four
dimensions.
38 BOBBY S. ACHARYA
We are now in a position to summarise the basic effective theory for the massless
fields. The low energy effective theory is an N = 1 supergravity theory coupled
to b2 (X) Abelian vector multiplets and b3 (X) massless, neutral chiral multiplets.
This theory is relatively uninteresting physically. In particular, the gauge group is
Abelian and there are no light charged particles. We will thus have to work harder
to obtain the basic requisites of the standard model - non-Abelian gauge fields and
chiral fermions - from G2 -compactifications. The basic point of these lectures is to
emphasise that these features emerge naturally from singularities in G2 -manifolds.
3. Super Yang-Mills from G2 -manifolds: Classical

In this section we will describe how to obtain non-Abelian gauge groups from
singular G2 -manifolds. We have known for some time now that non-Abelian gauge
groups emerge from M theory when space has a so-called ADE-singularity. We
learned this in the context of the duality between M theory on K3 and the heterotic
string on a flat three-torus, T3 [23]. So, our basic strategy will be to embed ADE-
singularities into G2 -manifolds. After reviewing the basic features of the duality
between M theory on K3 and heterotic string theory on T3 , we describe ADE-
singularities explicitly. We will then describe the M theory physics near such a
singularity. This is the physics of the McKay correspondence.
We then develop a picture of a G2 -manifold near an embedded ADE-singularity.
Based on this picture we analyse what kinds of four-dimensional gauge theories
these singularities give rise to. We then go on to describe local models for such
singular G2 -manifolds as finite quotients of smooth ones.
3.1. M theory - Heterotic Duality in Seven Dimensions. M theory
compactified on a K3 manifold is widely believed to be equivalent to the heterotic
string theory compactified on a 3-torus T3 . As with G2 compactification, both
of these are compactifications to flat Minkowski space. Up to diffeomorphisms,
K3 is the only simply connected, compact 4-manifold admitting metrics of SU (2)-
holonomy. SU (2) is the analog in four dimensions of G2 in seven dimensions.
Interestingly enough, in this case K3 is the only simply connected example, whereas
there are many G2 -manifolds.
There is a 58-dimensional moduli space of SU (2)-holonomy metrics on K3
manifolds of fixed volume. This space M(K3) is locally a coset space:
SO(3, 19)
(3.1) M(K3) = R+ × .
SO(3)×SO(19)
An SU (2) holonomy metric also admits two parallel spinors, which when ten-
sored with the 8 constant spinors of 7-dimensional Minkowski space give 16 global
supercharges. This corresponds to minimal supersymmetry in seven dimensions (in
the same way that G2 -holonomy corresponds to minimal supersymmetry in four
dimensions). If we work at a smooth point in M we can use Kaluza-Klein analysis
and we learn immediately that the effective d = 7 supergravity has 58 massless
scalar fields which parameterise M. These are the fluctuations of the metric on
K3. Additionally, since H 2 (K3, R) ≡ R22 there are twenty-two linearly indepen-
dent classes of harmonic 2-forms. These may be used a la equation (32) to give
a U (1)22 gauge group in seven dimensions. We now go on to describe how this
spectrum is the same as that of the heterotic string theory on T3 , at generic points
in M.
The heterotic string in ten dimensions has a low energy description in terms
of a supergravity theory whose massless bosonic fields are a metric, a 2-form B,
a dilaton φ and non-Abelian gauge fields of structure group SO(32) or E8 ×E8 .
There are sixteen global supersymetries. Compactification on a flat T3 preserves
all supersymmetries which are all products of constant spinors on both T3 and
Minkowski space. A flat metric on T3 involves six parameters so the metric gives rise
to six massless scalars, and since there are three independent harmonic two forms we
obtain from B three more. The condition for the gauge fields to be supersymmetric
on T3 is that their field strengths vanish: these are so called flat connections.
They are parameterised by Wilson lines around the three independent circles in
T3 . These are representations of the fundamental group of T3 in the gauge group.
Most of the flat connections actually arise from Wilson loops which are actually in
the maximal torus of the gauge group, which in this case is U (1)16 . Clearly, this
gives a 48-dimensional moduli space giving 58 scalars altogether. Narain showed
that this moduli space is actually also locally of the same form as M [19].
From the point of view of the heterotic string on T3 , the effective gauge group in
7 dimensions (for generic metric and B-field) is the subgroup of SO(32) or E8 ×E8
which commutes with the flat connection on T3 . At generic points in the moduli
space of flat connections, this gauge group will be U (1)16 . This is because the
generic flat connection defines three generic elements in U (1)16 ⊂ G. We can think
of these as diagonal 16 by 16 matrices with all elements on the diagonal non-zero.
Clearly, only the diagonal elements of G will commute with these. So, at a generic
point in moduli space the gauge group is Abelian.
Six more U (1) gauge fields arise as follows from the metric and B-field. T3 has
three harmonic one forms, so Kaluza-Klein reduction of B gives three gauge fields.
Additionally, since T3 has a U (1)3 group of isometries, the metric gives three more.
In fact, the local actions for supergravity theories in seven dimensions are actually
determined by the number of massless vectors. So, in summary, we have shown
that at generic points in M the low energy supergravity theories arising from M
theory on K3 or the heterotic string on T3 are the same.
At special points, some of the eigenvalues of the flat connections will vanish. At
these points the unbroken gauge group can get enhanced to a non-Abelian group.
This is none other than the Higgs mechanism: the Higgs fields are just the Wilson
lines. Additionally, because seven-dimensional gauge theories are infrared trivial
(the gauge coupling has dimension a positive power of length), the low energy
quantum theory actually has a non-Abelian gauge symmetry.
If M theory on K3 is actually equivalent to the heterotic string in seven di-
mensions, it too should therefore exihibit non-Abelian symmetry enhancement at
special points in the moduli space. These points are precisely the points in moduli
space where the K3 develops orbifold singularities. We will not provide a detailed
proof of this statement, but will instead look at the K3 moduli space in a neigh-
bourhood of this singularity, where all the interesting behaviour of the theory is
occurring. So, the first question is: what do these orbifold singularities look like?
3.1.1. ADE-singularities. An orbifold singularity in a Riemannian 4-manifold

can locally be described as R4 /Γ, where Γ is a finite subgroup of SO(4). For generic
enough Γ, the only singular point of this orbifold is the origin. These are the points
in R4 left invariant under Γ. A very crucial point is that on the heterotic side,
supersymmetry is completely unbroken all over the moduli space, so our orbifold
40 BOBBY S. ACHARYA
singularities in K3 should also preserve supersymmetry. This means that Γ is a

finite subgroup of SU (2) ⊂ SO(4). The particular SU (2) can easily be identified as
follows. Choose some set of complex coordinates so that C2 ≡ R4 . Then, a point in
C2 is labelled by a 2-component vector. The SU (2) in question acts on this vector
in the standard way:

u a b u
(3.2) −→ .
v c d v
The finite subgroups of SU (2) have a classification which may be described in
terms of the simply laced semi-simple Lie algebras: An , Dk , E6 , E7 and E8 . There
are two infinite series corresponding to SU (n + 1) = An and SO(2k) = Dk and
three exceptional subgroups corresponding to the three exceptional Lie groups of
E-type. The subgroups, which we will denote by ΓAn , ΓDk , ΓEi can be described
explicitly.
ΓAn−1 is isomorphic to Zn - the cyclic group of order n - and is generated by
2πi
e n 0
(3.3) −2πi .
0 e n
ΓDk is isomorphic to Dk−2 - the binary dihedral group of order 4k − 8 - and
has two generators α and β given by
πi
e k−2 0 0 i
(3.4) α= −πi β= .
0 e k−2 i 0
ΓE6 is isomorphic to T - the binary tetrahedral group of order 24 - and has two
generators given by
πi 2πi7 2πi7
e2 0 1 e 8 e 8
(3.5) −πi and √ 2πi5 2πi
0 e 2 2 e 8 e 8
ΓE7 is isomorphic to O - the binary octahedral group of order 48 - and has
three generators. Two of these are the generators of T and the third is
2πi
e 8 0
(3.6) 2πi7 .
0 e 8
Finally, ΓE8 is isomorphic to I - the binary icosahedral group of order 120 -
and has two generators given by
2πi3 2πi −2πi

e 5 0 1 e 5+e 5 1
(3.7) − 2πi2 and 2πi −2πi
−e 5 −e 5
2πi2 2πi3
0 e 5 e 5 −e 5 1
Since all the physics of interest is happening near the orbifold singularities of
K3, we can replace the K3 by C2 /ΓADE and study the physics of M theory on
C2 /ΓADE ×R6,1 near its singular set which is just 0×R6,1 . Since the K3 went from
smooth to singular as we varied its moduli we expect that the singular orbifolds
C2 /ΓADE are singular limits of non-compact smooth 4-manifolds X ADE . Because
of supersymmetry, these should have SU (2)-holonomy. This is indeed the case.
The metrics of SU (2)-holonomy on the X ADE are known as ALE-spaces, since
they asymptote to the locally Euclidean metric on C2 /ΓADE . Their existence was
proven by Kronheimer [18] - who constructed a gauge theory whose Higgs branch
is precisely the C2 /ΓADE with its SU (2)-holonomy (or hyper-Kähler) metric.
A physical description of this gauge theory arises in string theory. Consider
Type IIA or IIB string theory on C2 /ΓADE ×R5,1 . Take a flat Dp-brane (with p ≤ 5)
whose world-volume directions span Rp,1 ⊂ R5,1 , i.e. the D-brane is sitting at a point
on the orbifold. Then the world-volume gauge theory, which was first derived in
[12], is given by the Kronheimer gauge theory. This theory has eight supersym-
metries, which implies that its Higgs branch is a hyper-Kähler manifold. For one
D-brane this theory has a gauge group which is a product of unitary groups of
ranks given by the Dynkin indices (or dual Kac labels) of affine Dynkin diagram
of the corresponding ADE-group. So, for the An -case the gauge group is U (1)n+1 .
The matter content is also given by the affine Dynkin diagram - each link between
a pair of nodes represents a hypermultiplet transforming in the bifundamental rep-
resentation of the two unitary groups. This is an example of a quiver gauge theory
- a gauge theory determined by a quiver diagram.
We will make this explicit in the simplest case of ΓA1 . ΓA1 is isomorphic to Z2
and is in fact the center of SU (2). Its generator acts on C2 as

u −u
(3.8) −→ .
v −v
In this case, the Kronheimer gauge theory has a gauge group which is U (1) and
has two fields Φ1 and Φ2 . These are hypermultiplets in the string theory realisation
on a D-brane. The hypermultiplets each contain two complex scalars (ai , bi ) and
the a’s transform with charge +1 under U (1), whilst the b’s transform with charge
−1.
The potential energy of these scalar fields on the D-brane is
(3.9) 2 ≡ | µ|
V = |D|
where the three D-fields D (which are also known as the hyper-Kähler moment
maps µ associated with the U (1) action on the C4 parameterised by the fields) are
given by
(3.10) D1 = |a1 |2 + |b1 |2 − |a2 |2 − |b2 |2
and
(3.11) D2 + iD3 = a1 b1 − a2 b2 .
The space of zero energy minima of V is the space of supersymmetric ground
states S of the theory on the brane;
(3.12) S = {D
= 0}/U (1).
In supersymmetric field theories, instead of solving these equations directly, it is
equivalent to simply construct the space of gauge invariant holomorphic polynomials
of the fields and impose only the holomorphic equation above (this is the classical
F-term).
In the case at hand the gauge invariant polynomials are simply
(3.13) X = a1 b1 , Y = a2 b2 , Z = a1 b2 , W = a2 b1 .
These obviously parameterise C4 but are subject to the relation
(3.14) XY = W Z
42 BOBBY S. ACHARYA
However, the complex D-term equation asserts that

(3.15) X=Y
hence
(3.16) X 2 = W Z.
The space of solutions is precisely a copy of C2 /ΓA1 . To see this, we can
parameterise C2 /ΓA1 algebraically in terms of the ΓA1 invariant coordinates on
C2 . These are u2 , v 2 and uv. If we denote these three coordinates as w, z, x, then
obviously
(3.17) x2 = wz.
We prefer to re-write this equation by changing coordinates again. Defining x
= u2 − v 2 , y = iu2 + iv 2 and z = 2uv gives a map from C2 /ΓA1 to C3 . Clearly
however,
(3.18) x2 + y 2 + z 2 = 0
which means that C2 /ΓA1 is the hypersurface in C3 defined by this equation.
The orbifold can be deformed by adding a small constant to the right hand
side,
(3.19) x2 + y 2 + z 2 = r 2 .
If we take x, y and z to all be real and r to be real then it is clear that the deformed
4-manifold contains a 2-sphere of radius r. This 2-sphere contracts to zero size as
r goes to zero. The total space of the deformed 4-manifold is in fact the cotangent
bundle of the 2-sphere, T∗ S2 . To see this write the real parts of the x, y and z as
xi and their imaginary parts as pi . Then, since r is real, the xi are coordinates on
the sphere which obey the relation
(3.20) Σi xi pi = 0.
This means that the pi ’s parameterise tangential directions. The radius r sphere
in the center is then the zero section of the tangent bundle. Since the manifold
is actually complex it is natural to think of this as the cotangent bundle of the
Riemann sphere, T∗ CP1 . In the context of Euclidean quantum gravity, Eguchi
and Hanson constructed a metric of SU (2)-holonomy on this space, asymptotic to
the locally flat metric on C2 /ΓA1 .
In the Kronheimer gauge theory on the D-brane, deforming the singularity
corresponds to setting the D-terms or moment maps not to zero but to constants.
On the D-brane these constants represent the coupling of the background closed
string fields to the brane. These fields parameterise precisely the metric moduli of
the Eguchi-Hanson metric.
3.1.2. M theory Physics at the Singularity. This metric, whose precise

form we will not require, actually has three parameters which control the size and
shape of the two-sphere which desingularises the orbifold - the three possibilities
for setting the moment maps to constants . From a distance it looks as though
there is an orbifold singularity, but as one looks more closely one sees that the
singularity has been smoothed out by a two-sphere. The 2-sphere is dual to a
compactly supported harmonic 2-form, α. Thus, Kaluza-Klein reducing the C-field
using α gives a U (1) gauge field in seven dimensions. A vector multiplet in seven
dimensions contains precisely one gauge field and three scalars and the latter are
the parameters of the S2 . So, when T∗ CP1 is smooth the massless spectrum is an
Abelian vector multiplet.
From the duality with the heterotic string we expect to see an enhancement
in the gauge symmetry when we vary the scalars to zero i.e. when the sphere
shrinks to zero size. In order for this to occur, W ± -bosons must become massless
at the singularity. These are electrically charged under the U (1) gauge field which
originated from C. From the eleven-dimensional point of view the object which is
charged under C is the M 2-brane. If the M 2-brane wraps around the two-sphere,
it appears as a particle from the seven-dimensional point of view. This particle is
electrically charged under the U (1) and has a mass which is classically given by the
volume of the sphere. Since the M 2-brane has tension its dynamics will push it to
wrap the smallest volume two-sphere in the space. This least mass configuration is
in fact invariant under half of the supersymmetries 3 - a fact which means that it
lives in a short representation of the supersymmetry algebra. This in turn means
that its classical mass is in fact uncorrected quantum mechanically. The M 2-brane
wrapped around this cycle with the opposing orientation has the opposing U (1)
charge to the previous one.
Thus, when the two-sphere shrinks to zero size we find that two oppositely
charged BPS multiplets become massless. These have precisely the right quantum
numbers to enhance the gauge symmetry from U (1) to A1 = SU (2). Super Yang-
Mills theory in seven dimensions depends only on its gauge group. In this case we
are asserting that in the absence of gravity, the low energy physics of M theory on
C2 /ΓA1 ×R6,1 is described by super Yang-Mills theory on 0×R6,1 with gauge group
A1 .
The obvious generalisation also applies: in the absence of gravity, the low
energy physics of M theory on C2 /ΓADE ×R6,1 is described by super Yang-Mills
theory on 0×R6,1 with ADE gauge group. To see this, note that the smoothing
out of the orbifold singularity in C2 /ΓADE contains rank(ADE) two-spheres which
intersect according to the Cartan matrix of the ADE group. At smooth points
in the moduli space the gauge group is thus U (1)rank(ADE) . The corresponding
wrapped membranes give rise to massive BPS multiplets with precisely the masses
and quantum numbers required to enhance the gauge symmetry to the full ADE-
group at the origin of the moduli space.
Mathematically, the quantum numbers of wrapped branes are exactly the sim-
ple roots of the ADE-Lie algebra. This is our first encounter with the physics of the
McKay correspondence.
3.2. ADE-singularities in G2 -manifolds. We have thus far restricted our

attention to the ADE singularities in K3×R6,1 . However, the ADE singularity is a
much more local concept. We can consider more complicated spacetimes X 10,1 with
ADE singularities along more general seven-dimensional spacetimes, Y 6,1 . Then, if
X has a modulus which allows us to scale up the volume of Y , the large volume limit
is a semi-classical limit in which X approaches the previous maximally symmetric
situation discussed above. Thus, for large enough volumes we can assert that
3This is because the least volume two-sphere is an example of a calibrated or supersymmetric

cycle.
44 BOBBY S. ACHARYA
the description of the classical physics of M theory near Y is in terms of seven-

dimensional super Yang-Mills theory on Y - again with gauge group determined by
which ADE singularity lives along Y .
In the context of G2 -compactification on X×R3,1 , we want Y to be of the form
W ×R3,1 , with W the locus of ADE singularities inside X. Near W ×R3,1 , X×R3,1
looks like C2 /ΓADE ×W ×R3,1 . In order to study the gauge theory dynamics without
gravity, we can again focus on the physics near the singularity itself. So, we want
to focus on seven-dimensional super Yang-Mills theory on W ×R3,1 .
3.2.1. M theory Spectrum Near The Singularity. In flat space the super
Yang-Mills theory has a global symmetry group which is SO(3)×SO(6, 1). The
second factor is the Lorentz group, the first is the R-symmetry. The theory has
gauge fields transforming as (1, 7), scalars in the (3, 1) and fermions in the (2, 8) of
the universal cover. All fields transform in the adjoint representation of the gauge
group. Moreover the sixteen supersymmetries also transform as (2, 8).
On W ×R3,1 - with an arbitrary W - the symmetry group gets broken to
SO(3)×SO(3) ×SO(3, 1). Since SO(3) is the structure group of the tangent bun-
dle on W , covariance requires that the theory is coupled to a background SO(3)
gauge field - the spin connection on W . Similarly, though perhaps less intuitively,
SO(3) acts on the normal bundle to W inside X, hence there is a background SO(3)
gauge field also.
The supersymmetries transform as (2, 2, 2) + (2, 2, 2̄). For large enough W and
at energy scales below the inverse size of W , we can describe the physics in terms
of a four-dimensional gauge theory. But this theory as we have described it is not
supersymmetric as this requires that we have covariantly constant spinors on W .
Because W is curved, there are none. However, we actually want to consider the
case in which W is embedded inside a G2 -manifold X. In other words we require
that our local model - C2 /ΓADE ×W - admits a G2 -holonomy metric. When W is
curved this metric cannot be the product of the locally flat metric on C2 /ΓADE
and a metric on W . Instead the metric is warped and is more like the metric on
a fiber bundle in which the metric on C2 varies as we move around in W . Since
the space has G2 -holonomy we should expect the four-dimensional gauge theory
to be supersymmetric. We will now demonstrate that this is indeed the case by
examining the G2 -structure more closely. In order to do this however, we need to
examine the SU (2) structure on C2 /Γ as well.
4-dimensional spaces of SU (2)-holonomy are actually examples of hyperKähler
manifolds. They admit three parallel 2-forms ωi . These are analogous to the parallel
forms on G2 -manifolds. These three forms transform locally under SO(3) which
locally rotates the complex structures. On C2 these forms can be given explicitly
as
(3.21) ω1 + iω2 = du ∧ dv,

i
(3.22) ω3 = du ∧ dū + dv ∧ dv̄.
2
ΓADE is defined so that it preserves all three of these forms. The SO(3) which
rotates these three forms is identified with the SO(3) factor in our seven-dimensional
gauge theory picture. This is because the moduli space of SU (2)-holonomy metrics
is the moduli space of the gauge theory and this has an action of SO(3).
In a locally flat frame we can write down a formula for the G2 -structure on
C2 /ΓADE ×W ,
1
(3.23) ϕ= ωi ∧ ej δ ij + e1 ∧ e2 ∧ e3
6
where ei are a flat frame on W . Note that this formula is manifestly invariant under
the SO(3) which rotates the wi provided that it also acts on the ei in the same way.
The key point is that when the SO(3) of the gauge theory acts, in order for
the G2 -structure to be well defined the ei ’s must transform in precisely the same
way as the ωi . But SO(3) acts on the ei , because it is the structure group of the
tangent bundle to W . Therefore, if C2 /ΓADE ×W , admits a G2 -holonomy metric,
we must identify SO(3) with SO(3) . In other words, the connection on the tangent
bundle is identified with the connection on the normal SO(3) bundle. This breaks
the symmetries to the diagonal subgroup of the two SO(3)’s and implies that the
effective four-dimensional field theory is classically supersymmetric. Identifying the
two groups breaks the symmetry group down to SO(3) ×SO(3, 1) under which the
supercharges transform as (1, 2) + (3, 2) + cc. We now have supersymmetries since
the (1, 2) and its conjugate can be taken to be constants on W .
An important point, which we will not actually prove here, but will require in
the sequel is that the locus of ADE-singularities - namely the copy of W at the center
of C2 /Γ is actually a supersymmetric cycle (also known as a calibrated cycle). This
follows essentially from the fact that ΓADE fixes W and therefore the ϕ restricts to
be the volume form on W . This is the condition for W to be supersymmetric.
Supposing we could find a G2 -manifold of this type, what exactly is the four-
dimensional supersymmetric gauge theory it corresponds to? This we can answer
also by Kaluza-Klein analysis [1, 2], since W will be assumed to be smooth and
‘large’. Under SO(3) ×SO(3, 1), the seven-dimensional gauge fields transform as
(3, 1) + (1, 4), the three scalars give (3, 1) and the fermions give (1, 2) + (3, 2)+cc.
Thus the fields which are scalars under the four-dimensional Lorentz group are two
copies of the 3 of SO(3) . These may be interpreted as two one forms on W . These
will be massless if they are zero modes of the Laplacian on W (w.r.t. its induced
metric from the G2 -manifold). There will be precisely b1 (W ) of these i.e. one for
every harmonic one-form. Their superpartners are clearly the (3, 2) + cc fermions,
which will be massless by supersymmetry. This is precisely the field content of
b1 (W ) chiral supermultiplets of the supersymmetry algebra in four dimensions.
The (1, 4) field is massless if it is constant on W and this gives one gauge field
in four dimensions. The requisite superpartners are the remaining fermions which
transform as (1, 2) + cc.
All of these fields transform in the adjoint representation of the seven dimen-
sional gauge group. Thus the final answer for the massless fields is that they are
described by N = 1 super Yang-Mills theory with b1 (W ) massless adjoint chiral
supermultiplets. The case with pure “superglue” i.e. b1 (W ) = 0 is a particularly
interesting gauge theory at the quantum level: in the infrared the theory is believed
to confine colour, undergo chiral symmetry breaking and have a mass gap. We will
actually exhibit some of these very interesting properties semi-classically in M the-
ory! Much of the sequel will be devoted to explaining this. But before we can do
that we must first describe concrete examples of G2 -manifolds with the properties
we desire.
46 BOBBY S. ACHARYA
One idea is to simply look for smooth G2 -manifolds which are topologically
C2 ×W but admit an action of SU (2) which leaves W invariant but acts on C2 in
the natural way. Then we simply pick a ΓADE ⊂ SU (2) and form the quotient space
C2 /ΓADE ×W .
Luckily, such non-compact G2 -manifolds were constructed some time ago [9]!
Moreover, in these examples, W = S 3 , the simplest possible compact 3-manifold
with b1 (W ) = 0. Perfect.
3.2.2. Examples of G2 -manifolds with ADE-singularities. The G2 holo-

nomy metrics on C2 ×S 3 were constructed in [9]. These metrics are smooth and
complete and depend on one parameter a. There is a radial coordinate r and the
metric shows that there is a finite size S 3 (of size a) in the center which grows as we
move out along the r direction. They asymptote at infinity in a radial coordinate
r to a conical form
(3.24) ds2 = dr 2 + r 2 dΣ2
where dΣ2 is an Einstein metric on S 3 ×S 3 . Taking r large is equivalent to taking
a to zero, so the finite volume S 3 shrinks to zero size. This Einstein metric dΣ2
is not the standard product metric on the product of two spheres, although it is
the homogeneous metric on G/H with G = SU (2)3 and H = SU (2). H acts on G
as the ‘diagonal’ subgroup of the three SU (2)’s. G/H is obviously isomorphic as a
manifold to S 3 ×S 3 since S 3 is isomorphic to SU (2). This description of the conical
G2 -metric obviously has an asymptotic SU (2)3 ×Σ3 group of isometries with Σ3 the
group of permutations of the three SU (2) factors.
The conical metric is obviously incomplete, since the base of the cone goes to
zero size at r = 0. The complete G2 metrics can thus be regarded as completions of
the cones obtained by smoothing out the singularity at its apex. Topologically the
conical manifold is R+ ×S 3 ×S 3 , which gets smoothed out to R4 ×S 3 . Concretely
this amounts to choosing an SU (2) factor in G and ‘filling it in’ to form R4 . We
remind the reader that R4 − 0 is the same as R+ ×S 3 and filling in the origin gives
back R4 . In the case at hand we actually complete the cone by gluing in an S 3 .
This is the S 3 of size a.
Clearly there are three natural ways to carry out this procedure since G consists
of three copies of SU (2). Obviously each of these gives the same topological mani-
fold but it is very important for what follows that we realise that there are actually
three G2 manifolds that we can make in this way. The point is that the classical
moduli space of G2 -holonomy metrics consists of three real lines which intersect
at one point - the conical singularity. Moving off of the conical manifold in the
three different directions amounts to choosing an S 3 in G and filling it in. Along
these three directions three different S 3 ’s develop a finite volume. Another way to
say more or less the same thing in a perhaps more physical way is that there are
three smooth G2 -manifolds with the prescribed behaviour at infinity: the metric
on G/H. The transition between the three-manifolds via the conical singularity is
called a ‘flop’ - by analogy with what happens to spheres in complex manifolds.
We can give a simple algebraic description of the phenomenon of collapsing one
sphere and growing another with the following model taken from [6]. Consider the
hypersurface in R4 ×R4 cut out by the following equation
(3.25) Σi x2i − yi2 = a
where the x’s and y’ s are linear coordinates on the two R4 ’s. For a positive, we have
a radius a S 3 at the origin in the second R4 . This manifold is clearly topologically
S 3 ×R4 . For a negative, its again the same manifold topologically, but the roles of
x and y have been interchanged. Therefore as a passes from positive to negative
an S 3 shrinks to zero volume, the manifold becomes singular at 0 and another S 3
grows and the space remains smooth. This is obviously a much cruder description
of the space as a function of a, since as we have seen above there are actually three
directions in the moduli space and not two, but it has the advantage that it makes
the basic picture transparent.
3.2.3. M theory Physics on X = R4 ×S 3 . We saw earlier that the moduli of

the G2 -metric get complexified in M theory by the addition of the C-field. This is
necessary for supersymmetry. We observed that on a compact G2 -manifold the low
energy four-dimensional theory contains one massless scalar for every parameter
in the G2 -metric. The situation on a non-compact manifold X can in general
be quite different, since the metric fluctuations need not be localised on X. The
more delocalised these fields are, the more difficult it is to excite them. Indeed
to obtain a four-dimensional action we have to integrate over X, and this integral
will diverge if the fluctuations are not L2 -normalisable. If this is indeed the case
then we should not regard the corresponding four-dimensional fields as fluctuating:
rather they are background parameters, coupling constants, and we should study
the four-dimensional physics as a function of them.
In the case at hand, by examining the first order fluctation in the G2 -metric one
can readily see that a should indeed be treated as a coupling constant. It follows
from supersymmetry that its complex partner should also. We refer the reader to
[7] for the simple calculation which shows this explicitly. Our formula (32) can now
be applied to write this complex coupling constant as

3
(3.26) τ= ϕ + iC = V ol(S ) + i C
S3
where we integrate over the minimal volume three-sphere in the center of X. This
sphere generates the third homology. Note that there is no prime here, since the
field is not fluctuating.
So, we arrive at the conclusion that M theory on our G2 -manifold X is actually
a one complex dimensional family of theories parameterised by τ . There are three
semi-classical regimes corresponding to the three regions in which the spheres are
large, X is smooth and thus supergravity is valid. The four-dimensional spectrum
is massive in each of these semi-classical regimes since b2 (X) is zero and the zero
mode of the Lichnerowicz operator does not fluctuate. The physics in each of these
three regions is clearly the same since the metrics are the ‘same’. What about the
physics on the bulk of the τ -plane?
Our intuition asserts that there is only one further interesting point, τ = 0
where the manifold develops a conical singularity. In minimally supersymmetric
field theories in four dimensions which do have massless complex scalars which
parameterise a moduli space M , singularities in the physics typically only occur
at subloci in M which are also complex manifolds. In our case, we don’t have a
moduli space, but rather a parameter space, but we can think of τ as a background
superfield.
48 BOBBY S. ACHARYA
In any case, at τ = 0 we have zero size S 3 ’s and instanton effects can become
important here, since the action of an M 2-brane instanton is τ . These effects could
generate a non-zero quantum value of the C-field period and remove this potential
singularity, in which case we would be in the situation that there are no physical
phase transitions as a function of τ . Of course, physical quantities will depend on
τ , but the qualitative nature of the physics will remain the same for any value of τ .
This was first suggested in [6] and proven rigorously in [7].
The parameter space spanned by τ is thus a Riemann sphere with three spe-
cial points which correspond to the three large G2 -manifolds. This is the picture
Robbert Dijkgraaf alluded to in his lectures. To make contact with what he was
discussing we need to understand the S 1 quotients of our three G2 -manifolds. This
is because if the M theory 11-manifold Y is “fibered” by circles then the base 10-
manifold Q can be regarded as a spacetime in Type IIA string theory. The IIA
theory has a 1-form field A which is a connection on this circle bundle.
The quotient by S 1 of our three G2 -manifolds are basically the three geometries
Robbert was describing - the resolved conifold, its flop and the deformed conifold
with open strings!
To see this we define the S 1 action on the first big G2 -manifold to be the
standard action on the S 3 in R4 ×S 3 regarded as a Hopf fibration over CP1 . The
quotient is thus an R4 -bundle over CP1 . Because A is now topologically non-trivial
the metric one gets on this 6-manifold is not the Calabi-Yau metric on the resolved
conifold. However, it is a metric with the same symplectic structure! Thus, the
closed string Gromov-Witten invariants are the same as on the Calabi-Yau conifold.
The S 1 quotient of G2 -manifold number two gives the ‘same’ 6-manifold in Type
IIA - but now this is better regarded as the flop of the previous one.
Finally, S 1 -quotient number three acts on the R4 and not the S 3 . The quotient
by U (1) of R4 is R3 . The quotient of the G2 -manifold is thus an R3 -bundle over
S 3 . This action has a fixed point which is the S 3 of minimal volume in the middle
of the G2 space. The fact that there is a fixed point of the circle action means
in Type IIA language that A is singular along a codimension three submanifold in
the 10-dimensional spacetime. This is thus a 7-manifold which supports a Dirac
monopole. This is a D6-brane in string theory and is a place where open strings
can end. We thus learn that the third quotient gives Type IIA string theory on the
deformed conifold plus one D6-brane wrapping the S 3 .
Therefore M theory explains why the open string Gromov-Witten invariants
of the deformed conifold are exactly the same as the usual closed string Gromov-
Witten invariants of the resolved conifold. This is simply the statement that all
three Type IIA geometries are the same G2 -manifold in M theory metrically.
In M theory these invariants should be regarded as counting associative 3-cycles
in a G2 -manifold.
4. Super Yang-Mills from G2 -manifolds: Quantum.

We now move on to study the interplay between quantum super Yang-Mills
theory and M theory on G2 -manifolds. The results of this section are based upon
[2, 6, 7, 3]. We will be studying the physics of M theory on the G2 -manifolds
with ADE-singularities whose construction we described at the end of section 3.2.1.
We begin by reviewing the basic properties of super Yang-Mills theory. We then
go on to describe how these features are reflected in M theory. We first show
how membrane instantons can be seen to generate the superpotential of the theory.
Second, we go on to exhibit confinement and a mass gap semi-classically in M
theory. The result about confinement highlights the power of the physical McKay
correspondence.
The G2 -manifolds that actually interest us are obtained as quotients of R4 ×S 3
by ΓADE . We saw previously that, for large volume and low energies, four-dimen-
sional super Yang-Mills theory is a good description of the M theory physics. We
will thus begin this section with a review of the basic properties of the gauge theory.
4.1. Super Yang-Mills Theory. For completeness and in order to compare

easily with M theory results obtained later we briefly give a review of N = 1 pure
super Yang-Mills theory. We begin with gauge group SU (n). N = 1 SU (n) super
Yang-Mills theory in four dimensions is an extensively studied quantum field theory.
The classical Lagrangian for the theory is
1 1 θ
(4.1) L=− (F a )2 + 2 λ̄a iDλa + i F a F̃ aµν .
4g 2 µν g 32π 2 µν
F is the gauge field strength and λ is the gaugino field.
It is widely believed that this theory exhibits dynamics very similar to that of
ordinary QCD: confinement, chiral symmetry breaking, a mass gap. There are n
supersymmetric vacua. Supersymmetry constrains the dynamics of the theory so
strongly that the values of the low energy effective superpotential in the n vacua
are known. These are of the form
(4.2) Wef f = cµ3 e2πiτ /n .
Here τ is the complex coupling constant,
θ 4π
(4.3) τ= +i 2
2π g
and µ the mass scale. Shifting θ by 2π gives n different values for W .
In particular, the form of this potential suggests that it is generated by dynam-
ics associated with “fractional instantons”, i.e. instantonic objects in the theory
whose quantum numbers are formally of instanton number n1 . Such states are also
closely related to the spontaneously broken chiral symmetry of the theory. Let us
briefly also review some of these issues here.
Under the U (1)R-symmetry of the theory, the gauginos transform as
(4.4) λ → eiα λ
This is a symmetry of the classical action but not of the quantum theory (as can
easily be seen by considering the transformation of the fermion determinant in the
path integral). However, if the above transformation is combined with a shift in the
theta angle of the form
2n
(4.5) τ →τ+ α.
2π
then this cancels the change in the path integral measure. This shift symmetry
is a bona fide symmetry of the physics if α = 2π 2n , so that even in the quantum
theory a Z2n symmetry remains. Associated with this symmetry is the presence of
a non-zero value for the following correlation function,
(4.6) λλ(x1 )λλ(x2 )...λλ(xn )
50 BOBBY S. ACHARYA
which is clearly invariant under the Z2n symmetry. This correlation function is
generated in the 1-instanton sector and the fact that 2n gauginos enter is due to
the fact that an instanton of charge 1 generates 2n chiral fermion zero modes.
Cluster decomposition implies that the above correlation function decomposes
into ‘n constituents’ and therefore there exists a non-zero value for the gaugino
condensate:
(4.7) λλ =0.
Such a non-zero expectation value is only invariant under a Z2 subgroup of
Z2n implying that the discrete chiral symmetry has been spontaneously broken.
Consequently this implies the existence of n vacua in the theory.
In fact, it can be shown that
∂ 32π 2 3 2πiτ /n
(4.8) λλ = 16πi Wef f = − cµ e .
∂τ n
In view of the above facts it is certainly tempting to propose that ‘fractional
instantons’ generate the non-zero gaugino condensate directly. But this is difficult
to see directly in super Yang-Mills on R3,1 . We will return to this point later.
More generally, if we replace the SU (n) gauge group by some other gauge
group H, then the above statements are also correct but with n replaced every-
where with c2 (H), the dual Coxeter number of H. For ADE gauge groups c2 (H)
= Σr+1i=1 ai , where r is the rank of the gauge group and the ai are the Dynkin
indices of the affine Dynkin diagram associated to H. For An , all the ai = 1;
for Dn groups the four ‘outer’ nodes have index 1 whilst the rest have ai = 2.
E6 has indices (1, 1, 1, 2, 2, 2, 3), E7 has (1, 1, 2, 2, 2, 3, 3, 4) whilst E8 has indices
(1, 2, 2, 3, 3, 4, 4, 5, 6).
4.2. Theta angle and Coupling Constant in M theory. The physics of

M theory supported near the singularities of C2 /Γ×R6,1 is described by super
Yang-Mills theory on R6,1 . The gauge coupling constant of the theory is given by
1 1
(4.9) 2 ∼ l3
g7d p
where lp is the eleven-dimensional Planck length. In seven dimensions, one analog

of the theta angle in four dimensions is actually a three-form Θ. The reason for
this is the seven-dimensional interaction
(4.10) LI ∼ Θ∧F ∧F
(with F the Yang-Mills field strength). In M theory Θ is given by C, the three-form
potential for the theory.
If we now take M theory on our G2 -manifold R4 /Γ×W we have essentially
compactified the seven-dimensional theory on W and the four-dimensional gauge
coupling constant is roughly given by
1 VW
(4.11) 2 ∼ l3
g4d p
where VW is the volume of W . The four-dimensional theta angle can be identified

as

(4.12) θ= C.
W
The above equation is correct because under a global gauge transformation of C

which shifts the above period by 2π times an integer - a transformation which is a
symmetry of M theory - θ changes by 2π times an integer. Such shifts in the theta
angle are also global symmetries of the field theory.
Thus the complex gauge coupling constant of the effective four-dimensional
theory may be identified as the τ parameter of M theory

C ϕ
(4.13) τ= i + 3.
W 2π lp
This is of course entirely natural, since τ is the only parameter in M theory on this
space!
4.3. Superpotential in M theory. There is a very elegant way to calculate
the superpotential of super Yang-Mills theory on R3,1 by first compactifying it on
a circle to three dimensions [21]. The three-dimensional theory has a perturbative
expansion since the Wilson lines on the circle behave as Higgs fields whose vev’s
break the gauge symmetry to the maximal torus. The theory has a perturbative
expansion in the Higgs vevs, which can be used to compute the superpotential of
the compactified theory. One then takes the four-dimensional limit. In order to
compute the field theory superpotential we will mimic this idea in M theory [1].
Compactifying the theory on a small circle is equivalent to studying perturbative
Type IIA string theory on our G2 -manifold.
4.3.1. Type IIA theory on X = R4 /ΓADE ×S 3 . Consider Type IIA string

theory compactified to three dimensions on a seven-manifold X with holonomy G2 .
If X is smooth we can determine the massless spectrum of the effective supergravity
theory in three dimensions as follows. Compactification on X preserves four of
the 32 supersymmetries in ten dimensions, so the supergravity theory has three-
dimensional N = 2 local supersymmetry. The relevant bosonic fields of the ten-
dimensional supergravity theory are the metric, B -field, dilaton plus the Ramond-
Ramond one- and three-forms. These we will denote by g, B, φ, A1 , A3 respectively.
Upon Kaluza-Klein reduction the metric gives rise to a three-metric and b3 (X)
massless scalars. The latter parameterise the moduli space of G2 -holonomy metrics
on X. B gives rise to b2 (X) periodic scalars ϕi . φ gives a three-dimensional dilaton.
A1 reduces to a massless vector, while A3 gives b2 (X) vectors and b3 (X) massless
scalars. In three dimensions a vector is dual to a periodic scalar, so at a point
in moduli space where the vectors are free we can dualise them. The dual of the
vector field originating from A1 is the period of the RR 7-form on X, whereas the
duals of the vector fields coming from A3 are given by the periods of the RR 5-form
A5 over a basis of 5-cycles which span the fifth homology group of X. Denote
these scalars by σi . All in all, in the dualised theory we have, in addition to the
supergravity multiplet, b2 (X) + b3 (X) scalar multiplets. Notice that b2 (X) of the
scalar multiplets contain two real scalar fields, both of which are periodic.
Now we come to studying the Type IIA theory on X = R4 /ΓADE ×S 3 . Recall
that X = R4 /ΓADE ×S 3 is defined as an orbifold of the standard spin bundle of S 3 ,
denoted by S(S 3 ). To determine the massless spectrum of IIA string theory on X
we can use standard orbifold techniques. However, the answer can be phrased in a
simple way. X is topologically R4 /ΓADE ×S 3 . This manifold can be desingularised
to give a smooth seven-manifold M ΓADE which is topologically X ΓADE ×S 3 , where
X ΓADE is homeomorphic to an ALE space. The string theoretic cohomology groups
52 BOBBY S. ACHARYA
of X are isomorphic to the usual cohomology groups of M ΓADE . The reason for this
is simple: X is a global orbifold of S(S 3 ). The string theoretic cohomology groups
count massless string states in the orbifold CFT. The massless string states in the
twisted sectors are localised near the fixed points of the action of ΓADE on the spin
bundle. Near the fixed points we can work on the tangent space of S(S 3 ) near a
fixed point and the action of ΓADE there is just its natural action on R4 ×R3 .
Note that blowing up X to give M ΓADE cannot give a metric with G2 -holonomy
which is continuosly connected to the singular G2 -holonomy metric on X, since this
would require that the addition to homology in passing from X to M ΓADE receives
contributions from four-cycles. This is necessary since these are dual to elements
of H 3 (M ) which generate metric deformations preserving the G2 -structure. This
argument does not rule out the possibility that M ΓADE admits ‘disconnected’ G2 -
holonomy metrics, but is consistent with the fact that pure super Yang-Mills theory
in four dimensions does not have a Coulomb branch.
The important points to note are that the twisted sectors contain massless
states consisting of r scalars and r vectors where r is the rank of the corresponding
ADE group associated to Γ. The r scalars can intuitively be thought of as the periods
of the B field through r two-cycles. In fact, for a generic point in the moduli space of
the orbifold conformal field theory the spectrum contains massive particles charged
under the r twisted vectors. These can be interpreted as wrapped D2-branes whose
quantum numbers are precisely those of W -bosons associated with the breaking of
an ADE gauge group to U (1)r . This confirms our interpretation of the origin of this
model from M theory: the values of the r B -field scalars can be interpreted as the
expectation values of Wilson lines around the eleventh dimension associated with
this symmetry breaking. At weak string coupling and large S 3 volume these states
are very massive and the extreme low energy effective dynamics of the twisted sector
states is described by N =2 U (1)r super Yang-Mills in three dimensions. Clearly
however, the underlying conformal field theory is not valid when the W -bosons
become massless. The appropriate description is then the pure super Yang-Mills
theory on R2,1 ×S 1 which corresponds to a sector of M theory on X×S 1 . In this
section however, our strategy will be to work at a generic point in the CFT moduli
space which corresponds to being far out along the Coulomb branch of the gauge
theory. We will attempt to calculate the superpotential there and then continue the
result to four dimensions. This exactly mimics the strategy of [21] in field theory.
Note that we are implicitly ignoring gravity here. More precisely, we are assuming
that, in the absence of gravitational interactions with the twisted sector, the low
energy physics of the twisted sectors of the CFT is described by the Coulomb branch
of the gauge theory. This is natural since the twisted sector states are localised at
the singularities of X×R2,1 whereas the gravity propagates in bulk.
In this approximation, we can dualise the photons to obtain a theory of r
chiral multiplets, each of whose bosonic components (ϕ and σ ) is periodic. But
remembering that this theory arose from a non-Abelian one we learn that the
moduli space of classical vacua is
Cr
(4.14) Mcl =
ΛC
W Wg
where ΛC
W is the complexified weight lattice of the ADE group and Wg is the Weyl
group.
We can now ask about quantum effects. In particular, is there a non-trivial

superpotential for these chiral multiplets? In a theory with four supercharges BPS
instantons with only two chiral fermion zero modes can generate a superpotential.
Are there instantons in Type IIA theory on J? BPS instantons come from branes
wrapping supersymmetric cycles and Type IIA theory on a G2 -holonomy space
can have instantons corresponding to D6-branes wrapping the space itself or D2-
brane instantons which wrap supersymmetric 3-cycles. For smooth G2 -holonomy
manifolds these were studied in [14]. In the case at hand the D6-branes would
generate a superpotential for the dual of the graviphoton multiplet which lives in the
gravity multiplet but, since we wish to ignore gravitational physics for the moment,
we will ignore these. In any case, since X is non-compact, these configurations have
infinite action. The D2-branes on the other hand are much more interesting. They
can wrap the supersymmetric S 3 over which the singularities of X are fibered. We
can describe the dynamics of a wrapped D2-brane as follows. At large volume,
where the sphere becomes flatter and flatter the world-volume action is just the
Kronheimer gauge theory that we described previously in section 3.1.1! Here we
should describe this theory not just on S 3 but on a supersymmetric (or calibrated)
S 3 embedded in a space with a non-trivial G2 -holonomy structure. The upshot is
that the world-volume theory is in fact a cohomological field theory [8] so we can
trust it for any volume as long as the ambient space has G2 -holonomy. This is
because the supersymmetries on the world-volume are actually scalars on S 3 and
so must square to zero. This “topological version” of the Kronheimer theory can
be naturally obtained by “twisting” the usual gauge theory in the same way that
one obtains the topological A-model and B-model by twisting and was described by
Robbert Dijkgraaf in his lectures. In this case, the twisting is naturally incorporated
by the normal bundle of the calibrated three-sphere.
Note that, since we are ignoring gravity, we are implicitly ignoring higher de-
rivative corrections which could potentially also affect this claim. Another crucial
point is that the S 3 which sits at the origin in R4 in the covering space of X is the
supersymmetric cycle, and the spheres away from the origin are not supersymmet-
ric, so that the BPS wrapped D2-brane is constrained to live on the singularities of
X. In the quiver gauge theory, the origin is precisely the locus in moduli space at
which the single D2-brane can fractionate (according to the quiver diagram) and
this occurs by giving expectation values to the scalar fields which parameterise the
Coulomb branch which corresponds to the position of our D2-brane in the dimen-
sions normal to X.
What contribution to the superpotential do the fractional D2-branes make? To
answer this we need to identify the configurations which possess only two fermionic
zero modes. We will not give a precise string theory argument for this, but using
the correspondence between this string theory and field theory will identify exactly
which D-brane instantons we think are responsible for generating the superpoten-
tial. This may sound like a strong assumption, but, as we hope will become clear,
the fact that the fractional D2-branes are wrapped D4-branes is actually antici-
pated by the field theory! This makes this assumption, in our opinion, somewhat
weaker and adds credence to the overall picture being presented here.
In [11], it was shown that the fractionally charged D2-branes are actually D4-
branes which wrap the ‘vanishing’ 2-cycles at the origin in R4 /Γ. More precisely,
54 BOBBY S. ACHARYA
each individual fractional D2-brane which originates from a single D2-brane pos-
sesses D4-brane charge, but the total configuration, since it began life as a single
D2-brane, has zero D4-brane charge. The possible contributions to the superpoten-
tial are constrained by supersymmetry and must be given by a holomorphic function
of the r chiral superfields and also of the holomorphic gauge coupling constant τ
which corresponds to the complexified volume of the S 3 in eleven-dimensional M
theory. We have identified above the bosonic components of the chiral superfields
above. τ is given by

(4.15) τ = ϕ + iC
where ϕ is the G2 -structure defining 3-form on X. The period of the M theory

3-form through S 3 plays the role of the theta angle.
The world-volume action of a D4-brane contains the couplings
(4.16) L = B ∧ A3 + A5 .
Holomorphy dictates that there is also a term
(4.17) B∧ϕ
so that the combined terms are written as
(4.18) B ∧ τ + A5
Since the fourbranes wrap the ‘vanishing cycles’ and the S 3 we see that the
contribution of the D4-brane corresponding to the k-th fractional D2 takes the
form
(4.19) S = −β k .z
where we have defined
(4.20) z = τϕ + σ
and the βk are charge vectors. The r complex fields z are the natural holomorphic
functions upon which the superpotential will depend.
The wrapped D4-branes are the magnetic duals of the massive D2-branes which
we identified above as massive W -bosons. As such they are magnetic monopoles for
the original ADE gauge symmetry. Their charges are therefore given by an element
of the co-root lattice of the Lie algebra and thus each of the r + 1 β ’s is a rank
r vector in this space. Choosing a basis for this space corresponds to choosing a
basis for the massless states in the twisted sector Hilbert space which intuitively we
can think of as a basis for the cohomology groups Poincaré dual to the ‘vanishing’
2-cycles. A natural basis is provided by the simple co-roots of the Lie algebra of
ADE, which we denote by α∗k for k = 1, ..., r. This choice is natural, since these,
from the field theory point of view, are the fundamental monopole charges.
At this point it is useful to mention that the r wrapped D4-branes whose
magnetic charges are given by the simple co-roots of the Lie algebra correspond in
field theory to monopoles with charges α∗k and each of these is known to possess
precisely the right number of zero modes to contribute to the superpotential. Since
we have argued that, in a limit of the Type IIA theory on X, the dynamics at low
energies is governed by the field theory studied in [21] it is natural to expect that
these wrapped fourbranes also contribute to the superpotential. Another striking
feature of the field theory is that these monopoles also possess a fractional instanton
number - the second Chern number of the gauge field on R2,1 ×S 1 . These are
precisely in correspondence with the fractional D2-brane charges. Thus, in this
sense, the field theory anticipates that fractional branes are wrapped branes.
In the field theory on R2,1 ×S 1 it is also important to realise that there is
precisely one additional BPS state which contributes to the superpotential. The key
point is that this state, unlike the previously discussed monopoles, has dependence
on the periodic direction in spacetime. This state is associated with the affine node
of the Dynkin diagram. Its monopole charge is given by
(4.21) −Σrk=1 α∗k .
and it also carries one unit of instanton number.

The action for this state is
(4.22) S = Σrk=1 α∗k .z − 2πiτ
Together, these r + 1 BPS states can be regarded as fundamental in the sense

that all the other finite action BPS configurations can be thought of as bound states
of them.
Thus, in the correspondence with string theory it is also natural in the same
sense as alluded to above that a state with these corresponding quantum numbers
also contributes to the superpotential. It may be regarded as a bound state of
anti-D4-branes with a charge one D2-brane. In the case of SU (n) this is extremely
natural, since the total D4/D2-brane charge of the r +1 states is zero/one, and this
is precisely the charge of the D2-brane configuration on S 3 whose world-volume
action is the quiver gauge theory for the affine Dynkin diagram for SU (n). In
other words, the entire superpotential is generated by a single D2-brane which has
fractionated.
In summary, we have seen that the correspondence between the Type IIA string
theory on X and the super Yang-Mills theory on R2,1 ×S 1 is quite striking. Within
the context of this correspondence we considered a smooth point in the moduli space
of the perturbative Type IIA CFT, where the spectrum matches that of the Yang-
Mills theory along its Coulomb branch. On the string theory side we concluded
that the possible instanton contributions to the superpotential are from wrapped
D2-branes. Their world volume theory is essentially topological, from which we
concluded that they can fractionate. As is well known, the fractional D2-branes are
really wrapped fourbranes. In the correspondence with field theory, the wrapped
fourbranes are magnetic monopoles, whereas the D2-branes are instantons. Thus, if
these branes generate a superpotential they correspond in field theory to monopole-
instantons. This is exactly how the field theory potential is known to be generated.
We thus expect that the same occurs in the string theory on X.
Finally, the superpotential generated by these instantons is of affine-Toda type
and is known to possess c2 (ADE) minima corresponding to the vacua of the ADE
super Yang-Mills theory on R3,1 . The value of the superpotential in each of these
2πiτ
vacua is of the form e c2 . As such it formally looks as though it was generated
by fractional instantons, and in this context fractional M 2-brane instantons. This
result holds in the four-dimensional M theory limit because of holomorphy.
Let us demonstrate the vacuum structure in the simple case when the gauge
group is SU (2). Then there is only one scalar field, z. There are two fractional
56 BOBBY S. ACHARYA
D2-brane instantons whose actions are

(4.23) S1 = −z and S2 = z − 2πiτ
Both of these contribute to the superpotential as
(4.24) W = e−S1 + e−S2 .
Defining z = ln Y we have
e2πiτ
(4.25) W =Y + .
Y
The critical points of W are
2πiτ
(4.26) Y = ±e 2 .
This result about the superpotential suggests strongly that there is a limit
of M theory near an ADE singularity in a G2 -manifold which is precisely super
Yang-Mills theory. We will now go on to explore other limits of this M theory
background.
4.4. M theory Physics on ADE-singular G2 -manifolds. We saw previ-
ously that before taking the quotient by Γ, the M theory physics on R4 ×S 3 with
its G2 -metric was smoothly varying as a function of τ . In fact the same is true in
the case with ADE-singularities. One hint for this was that we explicitly saw just
now that the superpotential is non-zero in the various vacua and this implies that
the C-field is non-zero. This suggestion was concretely proven in [7].
Before orbifolding by Γ we saw there were three semiclassical limits of M theory
in the space parameterised by τ . These were described by M theory on three large
and smooth G2 -manifolds Xi , all three of which were of the form R4 ×S 3 . There
are also three semiclassical i.e. large volume G2 -manifolds when we orbifold by Γ.
These are simply the quotients by Γ of the Xi . One of these is the G2 -manifold
R4 /ΓADE ×S 3 . The other two are both of the form S 3 /ΓADE ×R4 . To see this, note
that the three S 3 ’s in the three G2 -manifolds Xi of the form R4 ×S 3 correspond to
the three S 3 factors in G = S 3 ×S 3 ×S 3 . ΓADE is a subgroup of one of these S 3 ’s.
If ΓADE acts on the R4 factor of X1 in the standard way, then it must act on S 3
in X2 - since X2 can be thought of as the same manifold but with the two S 3 ’s at
infinity exchanged. Then, because of the permutation symmetry it also acts on the
S 3 in X3 .
In the simple, crude, algebraic description in section 3.2.2, let X1 be the mani-
fold with a negative. Then define ΓADE to act on the R4 parameterised by xi . Then
xi = 0 is an S 3 of fixed points parameterised by yi . Thus X1 /ΓADE is isomorphic
to our G2 -manifold with ADE-singularities. Consider now what happens when a is
taken positive. This is our manifold X2 . Then, because xi = 0 is not a point on
X2 , there is no fixed point and ΓADE acts freely on the S 3 surrounding the origin
in the x-space. Thus, X2 /ΓADE is isomorphic to S 3 /ΓADE ×R4 , as is X3 .
On X1 /ΓADE in the large volume limit, we have a semi-classical description of
the four-dimensional physics in terms of perturbative super Yang-Mills theory. But,
at extremely low energies, this theory becomes strongly coupled, and is believed to
confine and get a mass gap. So, apart from calculating the superpotential in each
vacuum, as we did in section 4.3, we can’t actually calculate the spectrum here.
What about the physics in the other two semiclassical limits, namely large
X2,3 /ΓADE ? These G2 -manifolds are completely smooth. So supergravity is a good
approximation to the M theory physics. What do we learn about the M theory

physics in this approximation?
4.5. Confinement from G2 -manifolds. If it is true that the qualitative

physics of M theory on X2 /ΓADE and X3 /ΓADE is the same as that of M the-
ory on X1 /ΓADE , then some of the properties of super Yang-Mills theory at low
energies ought to be visible. The gauge theory is believed to confine ADE-charge
at low energies. If a gauge theory confines in four dimensions, electrically charged
confining flux tubes (confining strings) should be present. This is because the con-
fining potential is linear. If the classical fields of the gauge theory contain only
fields in the adjoint representation of the gauge group G, then these strings are
charged with respect to the center of G, Z(G). All other charges are screened by
quantum fluctuations. Can we see these strings in M theory on X2 /ΓADE ? Indeed,
the confining strings have a very simple description [3].
The natural candidates for such strings are M 2-branes which wrap around 1-
cycles in X2 /ΓADE or M 5-branes which wrap 4-cycles in X2 /ΓADE . Since X2 /ΓADE
is homeomorphic to S 3 /ΓADE ×R4 which is contractible to S 3 /ΓADE , the homology
groups of X2 /ΓADE are the same as those of the three-manifold S 3 /ΓADE . Thus, our
space has no four-cycles to speak of, so the confining strings can only come from
M 2-branes wrapping one-cycles in S 3 /ΓADE . The string charges are classified by the
first homology group H1 (S 3 /ΓADE , Z). For any manifold, the first homology group is
isomorphic to the Abelianisation of its fundamental group, Π1 . The Abelianisation
is obtained by setting all commutators in Π1 to be trivial, i.e.
Π1 (M )
(4.27) H1 (M, Z) ∼
= .
[Π1 (M ), Π1 (M )]
The fundamental group of S 3 /ΓADE is ΓADE . Hence, in order to calculate the
charges of our candidate confining strings we need to compute the Abelianisations
of all of the finite subgroups of SU (2).
ΓAn−1 ∼ = Zn . The gauge group is locally SU (n). Since Zn is Abelian, its
commutator subgroup is trivial and hence the charges of our strings take values in
Zn . Since this is isomorphic to the center of SU (n) this is the expected answer for
a confining SU (n) theory.
For Γ ∼= Dk−2 , the binary dihedral group of order 4k − 8, the local gauge group
of the Yang-Mills theory is SO(2k). The binary dihedral group is generated by two
elements α and β (see section 3.1.1) which obey the relations
(4.28) α2 = β k−2
(4.29) αβ = β −1 α
(4.30) α4 = β 2k−4 = 1.
To compute the Abelianisation of Dk−2 , we simply take these relations and
impose that the commutators are trivial. From the second relation this implies
that
(4.31) β = β −1
which in turn implies that
(4.32) α2 = 1 for k = 2p
58 BOBBY S. ACHARYA
and
(4.33) α2 = β for k = 2p + 1.
Thus, for k = 2p we learn that the Abelianisation of Dk−2 is isomorphic to Z2 ×Z2 ,
whereas for k = 2p + 1 it is isomorphic to Z4 . These groups are respectively the
centers of Spin(4p) and Spin(4p + 2). This is the expected answer for the confining
strings in SO(2k) super Yang-Mills which can be coupled to spinorial charges.
To compute the Abelianisations of the binary tetrahedral (denoted T), octahe-
dral (O) and icosahedral (I) groups which correspond respectively to E6 , E7 and
E8 super Yang-Mills theory, we utilise the fact that the order of F/[F, F ] - with
F a finite group - is the number of inequivalent one-dimensional representations of
F . The representation theory of the finite subgroups of SU (2) is described through
the McKay correspondence by the representation theory of the corresponding Lie
algebras. In particular the dimensions of the irreducible representations of T, O and
I are given by the coroot integers (or dual Kac labels) of the affine Lie algebras as-
sociated to E6 , E7 or E8 respectively. From this we learn that the respective orders
of T/[T, T], O/[O, O] and I/[I, I] are three, two and one. Moreover, one can easily
check that T/[T, T] and O/[O, O] are Z3 and Z2 respectively, by examining their
group relations. Thus we learn that T/[T, T], O/[O, O] and I/[I, I] are, respectively
isomorphic to the centers Z(E6 ), Z(E7 ) and Z(E8 ), in perfect agreement with the
expectation that the super Yang-Mills theory confines. Note that the E8 -theory
does not confine, since the strings are uncharged.
This result is also natural from the following point of view. In the singular
X1 /ΓADE (where the actual gauge theory dynamics is) the gauge bosons correspond
to M 2-branes wrapped around zero-size cycles. When we vary τ away from the ac-
tual gauge theory limit until we reach M theory on a large and smooth X2 /ΓADE the
confining strings are also wrapped M 2-branes. In the gauge theory we expect the
confining strings to be “built” from the excitations of the gauge fields themselves.
In M theory, the central role played by the gauge fields is actually played by the
M 2-brane.
4.6. Mass Gap from G2 -manifolds. We can also see the mass gap expected
of the gauge theory by studying the spectrum of M theory on the smooth G2 -
manifolds X2 /ΓADE and X3 /ΓADE . We already noted previously that the four-
dimensional spectrum of M theory on the Xi was massive. For precisely the same
reasons the spectrum of M theory on X2,3 /ΓADE is also massive.
5. Chiral Fermions from G2 -manifolds.

Thus far, we have seen that the simplest possible singularities of a G2 -manifold,
namely ADE orbifold singularities, produce a convincing picture of how non-Abelian
gauge groups emerge. However, for the purpose of obtaining a realistic model of
particle physics this is not enough. To this end, a basic requisite is the presence
of chiral fermions charged under these gauge symmetries. Chiral fermions are im-
portant in nature since they are massless as long as the gauge symmetries they are
charged under are unbroken. This enables us to understand the lightness of the
electron in terms of the Higgs vev.
What sort of singularity in a G2 -manifold X might we expect to give rise to
a chiral fermion in M theory? If the singularity is “bigger than a point” then we
don’t expect chiral fermions. This is because if the codimension of the singularity
is less than seven, the local structure of the singularity can actually be regarded as
a singularity in a Calabi-Yau threefold or K3 and these singularities give rise to a
spectrum of particles which form representations of N > 1 supersymmetry. Such
spectra are CP T self-conjugate. For instance, real codimension four singularities in
G2 -manifolds are the ADE-singularities we discussed above and the corresponding
four-dimensional spectra were not chiral. Similarly, if the singularity is codimension
six, i.e. is along a one-dimensional curve Σ in X, then everywhere near Σ the tangent
spaces of X naturally split into tangent and normal directions to X. Hence, the
holonomy of X near Σ actually reduces to SU (3) acting normally to Σ.
So we want to consider point-like singularities of X. The simplest such singu-
larities are conical, for which the metric looks locally like
(5.1) ds2 = dr 2 + r 2 g(Y )
for some six-dimensional metric g(Y ) on a 6-manifold Y . This has a singularity at
r = 0. We will argue that for many different choices of Y that chiral fermions are
part of the M theory spectrum.
5.1. Hints from Anomaly Inflow. The basic strategy of this subsection will
be to assume there is a G2 -manifold with a conical singularity of the above type
and consider the variation of bulk terms in the M theory effective action under
various gauge symmetries. These will be shown to be non-zero if Y obeys certain
conditions. If the theory is to be consistent, these anomalous variations must be
cancelled and this suggests the presence of chiral fermions in the spectrum. This is
based upon [24] who showed that when X is compact all these variations add up
to zero!
The gauge symmetries we will consider are the ones we have focussed on in
these lectures: the U (1) gauge symmetries from Kaluza-Klein reducing the C-field
and the ADE symmetries from the ADE-singularities.
We begin with the former case. We take M theory on X×R3,1 with X a cone
on Y so that X with the vertex removed is R×Y . The Kaluza-Klein ansatz for C
which gives gauge fields in four dimensions is
(5.2) C = Σα β α (x) ∧ Aα (y)
where the β’s are harmonic 2-forms on X. With this ansatz, consider the eleven-
dimensional Chern-Simons interaction

1
(5.3) S= C ∧ G ∧ G.
X×R 3,1 6
Under a gauge transformation of C under which
(5.4) C −→ C + d
S changes by something of the form4

(5.5) δS ∼ d( ∧ G ∧ G).
X×R3,1
We can regard X as a manifold with boundary ∂X = Y and hence

(5.6) δS ∼ ∧ G ∧ G.
Y ×R3,1
4We will not be too careful about factors in this section.

60 BOBBY S. ACHARYA
If we now make the Kaluza-Klein ansatz for the 2-form

(5.7) = Σi α β α
and use our ansatz for C, we find

(5.8) δS ∼ β ∧β ∧β
ρ σ δ
ρ dAσ ∧ dAδ .
Y R3,1
Thus if the integrals over Y (which are topological) are non-zero we obtain a non-
zero four-dimensional interaction characteristic of an anomaly in an Abelian gauge
theory. Thus, if the theory is to be consistent, it is natural to expect a spectrum
of chiral fermions at the conical singularity which exactly cancels δS.
We now turn to non-Abelian gauge anomalies. We have seen that ADE gauge
symmetries in M theory on a G2 -manifold X are supported along a three-manifold
W in X. If additional conical singularities of X are to support chiral fermions
charges under the ADE gauge group, then these singularities should surely also be
points Pi on W . So let us assume that near such a point the metric on X assumes
the conical form. In four-dimensional ADE gauge theories the triangle anomaly is
only non-trivial for An -gauge groups. So we restrict ourselves to this case. In this
situation, there is a seven-dimensional interaction of the form

(5.9) S= K ∧ Ω5 (A)
W ×R3,1
where A is the SU (n) gauge field and

(5.10) dΩ5 (A) = trF ∧ F ∧ F.
K is a two-form which is the field strength of a U (1) gauge field which is part of
the normal bundle to W . K measures how the An -singularity twists around W .
The U (1) gauge group is the subgroup of SU (2) which commutes with ΓAn .
Under a gauge transformation,
(5.11) A −→ A + DA λ
and

(5.12) δS ∼ (K ∧ dtrλF ∧ F )
W ×R3,1
so if K is closed, δS = 0. This will be the case if the An -singularity is no worse at the
conical singularity P than at any other point on W . If, however, the An -singularity
actually increases rank at P , then
(5.13) dK = 2πqδP
and we have locally a Dirac monopole of charge q at P . The charge is an integer
because of obvious quantisation conditions. In this situation we have that

(5.14) δS ∼ d(K ∧ trλF ∧ F ) = −q tr λF ∧ F
W ×R3,1 R3,1
which is precisely the triangle anomaly in an SU (n) gauge theory. Thus, if we have
this sort of situation in which the ADE-singularity along W degenerates further at
P we also expect chiral fermions to be present.
We now go on to utilise the M theory heterotic duality of subsection (3.1) to
construct explicitly conically singular manifolds for which we know the existence of
chiral fermions.
5.2. Chiral Fermions via Duality With The Heterotic String. In sec-
tion three we utilised duality with the heterotic string on T3 to learn about en-
hanced gauge symmetry in M theory. We applied this to G2 -manifolds quite suc-
cessfully. In this section we will take a similar approach. The following is based
upon [5].
We start by considering duality with the heterotic string. The heterotic string
compactified on a Calabi-Yau three fold Z can readily give chiral fermions. On the
other hand, most Calabi-Yau manifolds participate in mirror symmetry. For Z to
participate in mirror symmetry means, according to Strominger, Yau and Zaslow,
that, in a suitable limit of its moduli space, it is a T3 fibration (with singularities
and monodromies) over a base W . Then, taking the T3 ’s to be small and using on
each fiber the equivalence of the heterotic string on T3 with M theory on K3, it
follows that the heterotic string on Z is dual to M theory on a seven-manifold X
that is K3-fibered over W (again with singularities and monodromies). X depends
on the gauge bundle on Z. Since the heterotic string on Z is supersymmetric, M
theory on X is likewise supersymmetric, and hence X is a manifold of G2 holonomy.
The heterotic string on Z will typically have a four-dimensional spectrum of chi-
ral fermions. Since there are many Z’s that could be used in this construction (with
many possible classes of gauge bundles) it follows that there are many manifolds of
G2 holonomy with suitable singularities to give non-Abelian gauge symmetry with
chiral fermions. The same conclusion can be reached using duality with Type IIA,
as many six-dimensional Type IIA orientifolds that give chiral fermions are dual to
M theory on a G2 manifold [10]
Let us try to use this construction to determine what kind of singularity X
must have. (The reasoning and the result are very similar to that given in [17]
for engineering matter from Type II singularities. In [17] the Dirac equation is
derived directly rather than being motivated – as we will – by using duality with
the heterotic string.) Suppose that the heterotic string on Z has an unbroken
gauge symmetry G, which we will suppose to be simply-laced (in other words, an
A, D, or E group) and at level one. This means that each K3 fiber of X will have
a singularity of type G. As one moves around in X one will get a family of G-
singularities parameterized by W . If W is smooth and the normal space to W is a
smoothly varying family of G-singularities, the low energy theory will be G gauge
theory on W × R3,1 without chiral multiplets. This was the situation studied in
sections three and four. So chiral fermions will have to come from singularities of
W or points where W passes through a worse-than-orbifold singularity of X.
We can use the duality with the heterotic string to determine what kind of
singularities are required. The argument will probably be easier to follow if we
begin with a specific example, so we will consider the case of the E8 × E8 heterotic
string with G = SU (5) a subgroup of one of the E8 ’s. Such a model can very easily
get chiral 5’s and 10’s of SU (5); we want to see how this comes about, in the region
of moduli space in which Z is T3 -fibered over W with small fibers, and then we
will translate this description to M theory on X.
Let us consider, for example, the 5’s. The commutant of SU (5) in E8 is a
second copy of SU (5), which we will denote as SU (5) . Since SU (5) is unbroken,
the structure group of the gauge bundle E on Z reduces from E8 to SU (5) . Massless
fermions in the heterotic string transform in the adjoint representation of E8 . The
part of the adjoint representation of E8 that transforms as 5 under SU (5) transforms
62 BOBBY S. ACHARYA
as 10 under SU (5) . So to get massless chiral 5’s of SU (5), we must look at the
Dirac equation D on Z with values in the 10 of SU (5) ; the zero modes of that
Dirac equation will give us the massless 5’s of the unbroken SU (5).
We denote the generic radius of the T3 fibers as α, and we suppose that α is
much less than the characteristic radius of W . This is the regime of validity of the
argument for duality with M theory on X (and the analysis of mirror symmetry of
SYZ). For small α, we can solve the Dirac equation on Z by first solving it along
the fiber, and then along the base. In other words, we write D = DT + DW , where
DT is the Dirac operator along the fiber and DW is the Dirac operator along the
base. The eigenvalue of DT gives an effective “mass” term in the Dirac equation
on W . For generic fibers of Z → W , as we explain momentarily, the eigenvalues of
DT are all nonzero and of order 1/α. This is much too large to be canceled by the
behavior of DW . So zero modes of D are localized near points in W above which
DT has a zero mode.
When restricted to a T3 fiber, the SU (5) bundle E can be described as a flat
bundle with monodromies around the three directions in T3 . In other words, as in
section three, we have three Wilson lines on each fiber. For generic Wilson lines,
every vector in the 10 of SU (5) undergoes non-trivial “twists” in going around
some (or all) of the three directions in T3 . When this is the case, the minimum
eigenvalue of DT is of order 1/α. This is simply because for a generic flat gauge
field on the T3 -fiber there will be no zero mode.
A zero mode of DT above some point P ∈ W arises precisely if for some vector
in the 10, the monodromies in the fiber are all trivial.
This means that the monodromies lie in the subgroup of SU (5) that leaves fixed
that vector. If we represent the 10 by an antisymmetric 5 × 5 matrix B ij , i, j =
1, . . . , 5, then the monodromy-invariant vector corresponds to an antisymmetric
matrix B that has some nonzero matrix element, say B 12 ; the subgroup of SU (5)
that leaves B invariant is clearly then a subgroup of SU (2) × SU (3) (where in these
coordinates SU (2) acts on i, j = 1, 2 and SU (3) on i, j = 3, 4, 5). Let us consider
the case (which we will soon show to be generic) in which B 12 is the only nonzero
matrix element of B. If so, the subgroup of SU (5) that leaves B fixed is precisely
SU (2) × SU (3). There is actually a distinguished basis in this problem – the one
that diagonalizes the monodromies near P – and it is in this basis that B has only
one nonzero matrix element.
The commutant of SU (2) × SU (3) in E8 is SU (6). So, over the point P ,
the monodromies commute not just with SU (5) but with SU (6). Everything of
interest will happen inside this SU (6). The reason for this is that the monodromies
at P give large masses to all E8 modes except those in the adjoint of SU (6). So
we will formulate the rest of the discussion as if the heterotic string gauge group
were just SU (6), rather than E8 . Away from P , the monodromies break SU (6)
to SU (5) × U (1) (the global structure is actually U (5)). Restricting the discussion
from E8 to SU (6) will mean treating the vacuum gauge bundle as a U (1) bundle
(the U (1) being the second factor in SU (5) × U (1) ⊂ SU (6)) rather than an SU (5)
bundle.
The fact that, over P , the heterotic string has unbroken SU (6) means that,
in the M theory description, the fiber over P has an SU (6) singularity. Likewise,
the fact that away from P the heterotic string has only SU (5) × U (1) unbroken
means that the generic fiber, in the M theory description, must contain an SU (5)
singularity only, rather than an SU (6) singularity. As for the unbroken U (1), in
the M theory description it must be carried by the C-field. Indeed, over generic
points on W there is a non-zero size S 2 which shrinks to zero size at P in order
that the gauge symmetry at that point increases. Kaluza-Klein reducing C along
this S 2 gives a U (1).
If we move away from the point P in the base, the vector B in the 10 of SU (5)
is no longer invariant under the monodromies. Under parallel transport around
the three directions in T3 , it is transformed by phases e2πiθj , j = 1, 2, 3. Thus,
the three θj must all vanish to make B invariant. As W is three-dimensional,
we should expect generically that the point P above which the monodromies are
trivial is isolated. (Now we can see why it is natural to consider the case that,
in the basis given by the monodromies near P , only one matrix element of B is
nonzero. Otherwise, the monodromies could act separately on the different matrix
elements, and it would be necessary to adjust more than three parameters to make
B invariant. This would be a less generic situation.) We will only consider the
(presumably generic) case that P is disjoint from the singularities of the fibration
Z → W . Thus, the T3 fiber over P is smooth (as we have implicitly assumed in
introducing the monodromies on T3 ).
In [5] we explicitly solved the Dirac equation in a local model for this situation.
We found that the net number of chiral zero modes was one. We will not have time
to describe the details of the solution here.
In summary, before we translate into the M theory language, the chiral fermions
in the heterotic string theory on Z are localised at points on W over which the
Wilson lines in the T3 -fibers are trivial. In M theory this translates into the
statement that the chiral fermions are localised at points in W over which the ADE-
singularity “worsens”. This is also consistent with what we found in the previous
section.
5.3. M theory Description. So we have found a local structure in the het-

erotic string that gives a net chirality – the number of massless left-handed 5’s
minus right-handed 5’s – of one. Let us see in more detail what it corresponds to
in terms of M -theory on a manifold of G2 holonomy.
Here it may help to review the case considered in [17] where the goal was
geometric engineering of charged matter on a Calabi-Yau threefold in Type IIA.
What was considered there was a Calabi-Yau three fold R, fibered by K3’s with
a base W , such that over a distinguished point P ∈ W there is a singularity of
type Ĝ, and over the generic point in W this singularity is replaced by one of type
G – the rank of Ĝ being one greater than that of G. In our example, Ĝ = SU (6)
and G = SU (5). In the application to Type IIA, although R also has a Kähler
metric, the focus is on the complex structure. For Ĝ = SU (6), G = SU (5), let us
describe the complex structure of R near the singularities. The SU (6) singularity
is described by an equation xy = z 6 - cf. section three. Its “unfolding” depends
on five complex parameters and can be written zy = z 6 + P4 (z), where P4 (z) is a
quartic polynomial in z. If – as in the present problem – we want to deform the
SU (6) singularity while maintaining an SU (5) singularity, then we must pick P4 so
that the polynomial z 6 + P4 has a fifth order root. This determines the deformation
to be
(5.15) xy = (z + 5)(z − )5 ,
64 BOBBY S. ACHARYA
where we interpret as a complex parameter on the base W . Thus, the above

equation gives the complex structure of the total space R.
What is described above is the partial unfolding of the SU (6) singularity, keep-
ing an SU (5) singularity. In our G2 problem, we need a similar construction, but
we must view the SU (6) singularity as a hyper-Kähler manifold, not just a com-
plex manifold. In unfolding the SU (6) singularity as a hyper-Kähler manifold, each
complex parameter in P4 is accompanied by a real parameter that controls the area
of an exceptional divisor in the resolution/deformation of the singularity. The pa-
rameters are thus not five complex parameters but five triplets of real parameters.
(There is an SO(3) symmetry that rotates each triplet. This is the SO(3) rotating
the three Kähler forms in section three.)
To get a G2 -manifold, we must combine the complex parameter seen in (5.15)
with a corresponding real parameter. Altogether, this will give a three-parameter
family of deformations of the SU (6) singularity (understood as a hyper-Kähler man-
ifold) to a hyper-Kähler manifold with an SU (5) singularity. The parameter space
of this deformation is what we have called W , and the total space is a seven-manifold
that is our desired singular G2 -manifold X, with a singularity that produces the
chiral fermions that we analyzed above in the heterotic string language.
To find the hyper-Kähler unfolding of the SU (6) singularity that preserves an
SU (5) singularity is not difficult, using Kronheimer’s description of the general un-
folding via a hyper-Kähler quotient [18]. At this stage, we might as well generalize
to SU (N ), so we consider a hyper-Kähler unfolding of the SU (N + 1) singularity
to give an SU (N ) singularity. The unfolding of the SU (N + 1) singularity is ob-
tained by taking a system of N + 1 hypermultiplets Φ0 , Φ1 , . . . ΦN with an action of
K = U (1)N . Under the ith U (1) for i = 1, . . . , N , Φi has charge 1, Φi−1 has charge
−1, and the others are neutral. This configuration of hypermultiplets and gauge
fields is known as the quiver diagram of SU (N + 1) and appears in studying D-
branes near the SU (N + 1) singularity We let H denote R4 , so the hypermultiplets
parameterize HN +1 , the product of N + 1 copies of R4 . The hyper-Kähler quotient
of HN +1 by K is obtained by setting the D-field (or components of the hyper-
Kähler moment map) to zero and dividing by K. It is denoted HN +1 //K, and is
isomorphic to the SU (N + 1) singularity R4 /ZN +1 . Its unfolding is described by

setting the D-fields equal to arbitrary constants, not necessarily zero. In all, there
are 3N parameters in this unfolding – three times the dimension of K – since for
each U (1), D has three components, rotated by an SO(3) group of R-symmetries.
We want a partial unfolding keeping an SU (N ) singularity. To describe this,
we keep 3(N − 1) of the parameters equal to zero and let only the remaining
three vary; these three will be simply the values of D for one of the U (1)’s. The
seven-manifold which we propose admits a natural G2 -holonomy metric is easy to
describe. One picks a U (1) subgroup of K - the gauge group of the Kronheimer
construction. There are three D-terms D associated to this U (1). Then one simply
repeats Kronheimer’s construction, but one does not set D to zero. This gives a 7-
manifold which maps to R (parameterised by the space of values of D)
3 over which
the generic fiber is the ADE-singularity obtained from the Kronheimer construction
using K , the commutant of U (1) in K. However, at the origin, i.e. when D is zero,
the fiber degenerates further to an ADE-singularity of higher rank. This is exactly
the sort of picture we expected from the heterotic string.
To carry out this procedure, we first write K = K ×U (1) (where U (1) denotes
a chosen U (1) factor of K = U (1)N ). Then we take the hyper-Kähler quotient of
HN +1 by K to get a hyper-Kähler eight-manifold X̂ = HN +1 //K , after which
we take the ordinary quotient, not the hyper-Kähler quotient, by U (1) to get a
seven-manifold X = X̂/U (1) that should admit a metric of G2 -holonomy. X has
a natural map to W = R3 given by the value of the D-field of U (1) – which was
not set to zero – and this map gives the fibration of X by hyper-Kähler manifolds.
In the present example, we can easily make this explicit. We take U (1) to
be the “last” U (1) in K = U (1)N , so U (1) only acts on ΦN −1 and ΦN . K is
therefore the product of the first N − 1 U (1)’s; it acts trivially on ΦN , and acts on
Φ0 , . . . , ΦN −1 according to the standard quiver diagram of SU (N ). So the hyper-
Kähler quotient HN +1 //K is just (HN //K ) × H , where HN //K is the SU (N )
singularity, isomorphic to H/ZN and H is parameterized by ΦN . So finally, X will
be (H/ZN × H )/U (1) . To make this completely explicit, we just need to identify
the group actions on H and H . If we parameterize H and H respectively by pairs of
complex variables (a, b) and (a , b ) then the ZN action on H, such that the quotient
H/ZN is the SU (N ) singularity, is given by
2πik/N
a e a
(5.16) → ,
b e−2πik/N b
while the U (1) action that commutes with this (and preserves the hyper-Kähler
structure) is
iψ/N
a e a
(5.17) → .
b e−iψ/N b
The U (1) action on H is similarly
iψ/N
a e a
(5.18) → .
b e−iψ/N b
In all, if we set λ = eiψ/N , w1 = a , w2 = b , w3 = a, w4 = b̄, then the quotient
H/ZN × H /U (1) can be described with four complex variables w1 , . . . , w4 modulo
the equivalence
(5.19) (w1 , w2 , w3 , w4 ) → (λN w1 , λN w2 , λw3 , λw4 ), |λ| = 1
This quotient is a cone on a weighted projective space WCP3N,N,1,1 . In fact, if we
impose the above equivalence relation for all nonzero complex λ, we would get the
weighted projective space itself; by imposing this relation only for |λ| = 1, we get a
cone on the weighted projective space. Note that the conical metric of G2 -holonomy
on this space does not use usual Kähler metric on weighted projective space.
WCP3N,N,1,1 has a family of AN −1 -singularities at points (w1 , w2 , 0, 0). This is
easily seen by setting λ to e2πi/N . This set of points is a copy of CP1 = S2 . Our
proposed G2 -manifold is a cone over weighted projective space, so it has a family
of AN −1 -singularities which are a cone over this S2 . This is of course a copy of R3 .
Away from the origin in R3 the only singularities are these orbifold singularities.
At the origin however, the whole manifold develops a conical singularity. There,
the 2-sphere, which is noncontractible in the bulk of the manifold, shrinks to zero
size. This is in keeping with the anomaly inflow arguments of the previous section.
There we learned that an ADE-singularity which worsens over a point in W is a
good candidate for the appearance of chiral fermions. Here, via duality with the
66 BOBBY S. ACHARYA
heterotic string, we find that the conical singularity in this example supports one
chiral fermion in the N of the SU (N ) gauge symmetry coming from the AN −1 -
singularity. In fact, the U (1) gauge symmetry from the C-field in this example
combines with the SU (N ) to give a gauge group which is globally U (N ) and the
fermion is in the fundamental representation.
Some extensions of this can be worked out in a similar fashion. Consider the
case that away from P , the monodromies break SU (N +1) to SU (p)×SU (q)×U (1),
where p+q = N +1. Analysis of the Dirac equation along the above lines shows that
such a model will give chiral fermions transforming as (p, q̄) under SU (p) × SU (q)
(and charged under the U (1)). To describe a dual in M theory on a manifold of G2
holonomy, we let K = K × U (1) , where now K = K1 × K2 , K1 being the product
of the first p − 1 U (1)’s in K and K2 the product of the last q − 1, while U (1) is
the pth U (1). Now we must define X̂ = HN +1 //K , and the manifold admitting a
metric of G2 holonomy should be X̂/U (1) .
We can compute X̂ easily, since K1 acts only on Φ1 , . . . , Φp and K2 only on
Φp+1 , . . . , ΦN +1 . The hyper-Kähler quotients by K1 and K2 thus simply construct
the SU (p) and SU (q) singularities, and hence X̂ = H/Zp × H/Zq . X̂ has planes
of Zp and Zq singularities, which will persist in X = X̂/U (1) , which will also
have a more severe singularity at the origin. So the model describes a theory with
SU (p)×SU (q) gauge theory and chiral fermions supported at the origin. U (1) acts
on H/Zp and H/Zq as the familiar global symmetry that preserves the hyper-Kähler
structure of the SU (p) and SU (q) singularities. Representing those singularities by
pairs (a, b) and (a , b ) modulo the usual action of Zp and Zq , U (1) acts by
iψ/p −iψ/q
a e a a e a
(5.20) → and → .
b e−iψ/p b b eiψ/q b
Now if p and q are relatively prime, we set λ = eiψ/pq , and we find that the
U (1) action on the complex coordinates w1 , . . . , w4 (which are defined in terms of
a, b, a , b by the same formulas as before) is
(5.21) (w1 , w2 , w3 , w4 ) → (λp w1 , λp w2 , λq w3 , λq w4 ).
If p and q are relatively prime, then the U (1) action, upon taking λ to be a pth or
q th root of 1, generates the Zp ×Zq orbifolding that is part of the original definition
of X̂. Hence in forming the quotient X̂/U (1) , we need only to act on the w’s by
the equivalence relation. The quotient is therefore a cone on a weighted projective
space WCP3p,p,q,q . If p and q are not relatively prime, we let (p, q) = r(n, m) where
r is the greatest common divisor and n and m are relatively prime. Then we let
λ = exp(irψ/pq), so the equivalence relation above is replaced with
(5.22) (w1 , w2 , w3 , w4 ) → (λn w1 , λn w2 , λm w3 , λm w4 ).
To reproduce X̂/U (1) we must now also divide by Zr , acting by
(5.23) (w1 , w2 , w3 , w4 ) → (ζw1 , ζw2 , w3 , w4 ),
where ζ = 1. So X is a cone on WCP3n,n,m,m /Zr .
r
5.4. Other Gauge Groups and Matter Representations. We now ex-

plain how to generalise the above construction to obtain singularities with more
general gauge groups and chiral fermion representations. Suppose that we want to
get chiral fermions in the representation R of a simply-laced group G. This can be
achieved for certain representations. We find a simply-laced group Ĝ of rank one

more than the rank of G, such that Ĝ contains G × U (1) and the Lie algebra of Ĝ
decomposes as g ⊕ o ⊕ r ⊕ r̄, where g and o are the Lie algebras of G and U (1),
r transforms as R under G and of charge 1 under U (1), and r̄ transforms as the
complex conjugate. Such a Ĝ exists only for special R’s, and these are the R’s that
we will generate from G2 singularities.
Given Ĝ, we proceed as above on the heterotic string side. We consider a family
of T3 ’s, parameterized by W , with monodromy that at a special point P ∈ W leaves
unbroken Ĝ, and at a generic point breaks Ĝ to G×U (1). We moreover assume that
near P , the monodromies have the same sort of generic behavior assumed above.
Then the same computation as above will show that the heterotic string has, in this
situation, a single multiplet of fermion zero modes (the actual chirality depends on
solving the Dirac equation) in the representation R, with U (1) charge 1.
Dualizing this to an M theory description, over P we want a Ĝ singularity, while
over a generic point in W we should have a G singularity. Thus, we want to consider
the unfolding of the Ĝ singularity (as a hyper-Kähler manifold) that preserves a
G singularity. To do this is quite simple. We start with the Dynkin diagram of
Ĝ. The vertices are labeled with integers ni , the Dynkin indices. In Kronheimer’s
construction, the Ĝ singularity is obtained as the hyper-Kähler quotient of Hk (for
some k) by the action of a group K = i U (ni ). Its unfolding is obtained by

allowing the D-fields of the U (1) factors (the centers of the U (ni )) to vary.
The G Dynkin diagram is obtained from that of Ĝ by omitting one node,
corresponding to one of the U (ni ) groups; we write the center of this group as
U (1) . Then we write K (locally) as K = K × U (1) , where K is defined by
replacing the relevant U (ni ) by SU (ni ). We get a hyper-Kähler eight-manifold as
the hyper-Kähler quotient X̂ = Hk //K , and then we get a seven-manifold X by
taking the ordinary quotient X = X̂/U (1) . This maps to W = R3 by taking the
value of the U (1) D-field,
which was not set to zero. The fiber over the origin is

obtained by setting this D-field to zero after all, and gives the original Ĝ singularity,
while the generic fiber has a singularity of type G.
One can readily work out examples of pairs G, Ĝ. We will just consider the
cases most relevant for grand unification. For G = SU (N ), to get chiral fields in the
antisymmetric tensor representation, Ĝ should be SO(2N ). For G = SO(10), to
get chiral fields in the 16, Ĝ should be E6 . For G = SO(2k), to get chiral fields in
the 2k, Ĝ should be SO(2k + 2). (Note in this case that 2k is a real representation.
However, the monodromies in the above construction break SO(2k+2) to SO(2k)×
U (1), and the massless 2k’s obtained from the construction are charged under the
U (1); under SO(2k) × U (1) the representation is complex.) For 2k = 10, this
example might be used in constructing SO(10) GUT’s. For G = E6 , to get 27’s, Ĝ
should be E7 . A useful way to describe the topology of X in these examples is not
clear.
In this construction, we emphasized, on the heterotic string side, the most
generic special monodromies that give enhanced gauge symmetry, which corre-
sponds on the M theory side to omitting from the hyper-Kähler quotient a rather
special U (1) that is related to a single node of the Dynkin diagram. We could also
consider more general heterotic string monodromies; this would correspond in M
theory to omitting a more general linear combination of the U (1)’s.
68 BOBBY S. ACHARYA
6. Outlook.
Having gathered all the necessary ingredients we can now briefly describe how
one goes about building a model of particle physics from M theory on a G2 -
manifold, X. First it is natural that X admits a map to a three-manifold W .
The generic fibers of the map are all K3-surfaces which have an ADE singularity of
some fixed type. A4 = SU (5) is a promising possibility for particle physics. This
plays the role of the GUT gauge group.
At a finite number of points on X which are also on W , there are conical
singularities of the kind discussed in section five. These support chiral fermions in
various representations of the ADE gauge group. For instance, in the case of SU (5)
¯
we would like to obtain three 5’s and three 10’s. The singularities of X should be
of the required type.
We then take W to non-simply connected (e.g. W might be S 3 /Zn ). Wilson
lines (or flat connections) of the SU (5) gauge fields can then be used to break SU (5)
to SU (3)×SU (2)×U (1) - the gauge group of the standard model.
An analysis of some of the basic properties of such models (assuming a suitable
X exists) was carried out in [25]. It was found that one of the basic physical tests
of such a model - namely the stability of the proton - was not problematic. This is
because the various families of chiral fermions originate from different points on X
so it is natural for them to be charged under different discrete symmetries. These
symmetries prevent the existence of operators which would otherwise mediate the
decay of the proton too quickly.
On the mathematical side, we still do not have examples of compact G2 -
manifolds with these conical singularities. The physics suggests that they are nat-
ural spaces to construct and we hope that this will be done in the near future.
Of course, we are still a long way from having a realistic explanation of particle
physics through M theory, since we do not understand in detail why supersymmetry
is broken in nature and why the cosmological constant is so small.
References
1. B. S. Acharya, M theory, Joyce orbifolds and super Yang-Mills, Adv. Theor. Math. Phys. 3,
227 (1999) [arXiv:hep-th/9812205].
2. B. S. Acharya, On realising N = 1 super Yang-Mills in M theory, arXiv:hep-th/0011089.
3. B. S. Acharya, Confining strings from G2 -holonomy spacetimes, arXiv:hep-th/0101206.
4. B. S. Acharya and C. Vafa, On domain walls of N = 1 supersymmetric Yang-Mills in four
dimensions, arXiv:hep-th/0103011.
5. B. Acharya and E. Witten, Chiral fermions from manifolds of G2 holonomy, arXiv:hep-
th/0109152.
6. M. Atiyah, J. Maldacena and C. Vafa, An M-theory flop as a large N duality, J. Math. Phys.
42, 3209 (2001) [arXiv:hep-th/0011256].
7. M. Atiyah and E. Witten, M-theory dynamics on a manifold of G2 holonomy, Adv. Theor.
Math. Phys. 6 (2002) 1. arXiv:hep-th/0107177.
8. M. Bershadsky, C. Vafa and V. Sadov, D-Branes and Topological Field Theories, Nucl. Phys.
B. 463, 420 (1996) [arXiv:hep-th/9511222].
9. R. Bryant and S. Salamon, On the Construction of Complete Metrics wtih Exceptional Ho-
lonomy, Duke Math Journal,58 (1989) 829. G. W. Gibbons, D. N. Page and C. N. Pope,
Einstein Metrics On S 3 , R3 and R4 Bundles, Commun. Math. Phys. 127, 529 (1990).
10. M. Cvetic, G. Shiu and A. M. Uranga, Chiral four-dimensional N = 1 supersymmetric type IIA
orientifolds from intersecting D6-branes, Nucl. Phys. B. 615, 3 (2001) [arXiv:hep-th/0107166].
11. D. E. Diaconescu, M. R. Douglas and J. Gomis, Fractional branes and wrapped branes, JHEP
9802, 013 (1998) [arXiv:hep-th/9712230].
12. M. R. Douglas and G. W. Moore, D-branes, Quivers, and ALE Instantons, arXiv:hep-
th/9603167.
13. M. J. Duff, B. E. Nilsson and C. N. Pope, Kaluza-Klein Supergravity, Phys. Rept. 130, 1
(1986).
14. J. A. Harvey and G. W. Moore, Superpotentials and membrane instantons, arXiv:hep-
th/9907026.
15. P. Hořava and E. Witten, Eleven-Dimensional Supergravity on a Manifold with Boundary,
16. D. D. Joyce, Compact Manifolds of Special Holonomy, Oxford University Press, 2000.
17. S. Katz and C. Vafa, Matter from geometry, Nucl. Phys. B 497, 146 (1997) [arXiv:hep-
th/9606086].
18. P. Kronheimer, Construction of ALE-spaces as hyper-Kähler Quotients, J. Diff. Geom. 28
(1989) 665.
19. K. S. Narain, New Heterotic String Theories In Uncompactified Dimensions < 10, Phys. Lett.
B 169, 41 (1986).
20. G. Papadopoulos and P. K. Townsend, Compactification of D = 11 supergravity on spaces of
exceptional holonomy, Phys. Lett. B 357, 300 (1995) [arXiv:hep-th/9506150].
21. N. Seiberg and E. Witten, Gauge dynamics and compactification to three dimensions, The
mathematical beauty of physics (Saclay 1996) 333 Adv Ser. Math. Phys. 24 World Scientific.
arXiv:hep-th/9607163. N. M. Davies, T. J. Hollowood and V. V. Khoze, Monopoles, affine
algebras and the gluino condensate, arXiv:hep-th/0006011.
22. E. Witten, Fermion Quantum Numbers In Kaluza-Klein Theory, In Appelquist, T. (ED.) et
al: Modern Kaluza-Klein Theories, 438-511. Shelter Island 1983, Procedings, Quantum Field
Theory and the Fundamental Problems of Physics, 227-277.
23. E. Witten, String theory dynamics in various dimensions, Nucl. Phys. B 443, 85 (1995)
24. E. Witten, Anomaly cancellation on G2 -manifolds, arXiv:hep-th/0108165.
25. E. Witten, Deconstruction, G(2) holonomy, and doublet-triplet splitting, arXiv:hep-
ph/0201018.
Current address: Department of Physics, Rutgers University, 126 Frelinghuysen Road, Pis-
cataway, NJ 08854
E-mail address: bacharya@physics.rutgers.edu
Volume 3, 2004
Conjectures in Kähler geometry
Simon K. Donaldson
Abstract. We state a general conjecture about the existence of Kähler met-

rics of constant scalar curvature, and discuss the background to the conjecture.
1. The equations
In this article we discuss some well-known problems in Kähler geometry. The
general theme is to ask whether a complex manifold admits a preferred Kähler
metric, distinguished by some natural differential-geometric criterion. A paradigm
is the well-known fact that any Riemann surface admits a metric of constant Gauss
curvature. Much of the interest of the subject comes from the interplay between, on
the one hand, the differential geometry of metrics, curvature tensors etc. and, on
the other hand, the complex analytic or algebraic geometry of the manifold. This
is, of course, a very large field and we make no attempt at an exhaustive account,
but it seems proper to emphasise at the outset that many of these questions have
been instigated by seminal work of Calabi.
Let (V, ω0 ) be a compact Kähler manifold of complex dimension n. The Kähler
forms in the class [ω0 ] can be written in terms of a Kähler potential ωφ = ω0 +i∂∂φ.
In the case when 2π[ω0 ] is an integral class, eφ has a geometrical interpretation
as the change of metric on a holomorphic line bundle L → V . The Ricci form
ρ = ρφ is −i times the curvature form of KV−1 , with the metric induced by ωφ , so
[ρ] = 2πc1 (V ) ∈ H 2 (V ). In the 1950’s, Calabi [3] initiated the study of Kähler-
Einstein metrics, with
(1) ρφ = λωφ ,
for constant λ. For these to exist we need the topological condition 2πc1 (V ) = λ[ω].
When this condition holds we can write (by the ∂∂ Lemma)
ρ0 − λω0 = i∂∂f,
for some function f . The Kähler-Einstein equation becomes the second order, fully
nonlinear, equation
(2) (ω0 + i∂∂φ)n = ef −λφ ω0n .
2000 Mathematics Subject Classification. 32Q15, 53C55, 53C56.
2004
71
72 SIMON K. DONALDSON
More explicitly, in local coordinates zα and in the case when the metric ω0 is
Euclidean, the equation is

∂2φ
(3) det δαβ + = ef −λφ .
∂zα ∂z β
This is a complex Monge-Ampère equation and the analysis is very much related
to that of real Monge-Ampère equations of the general shape

∂2u
(4) det = F (x, u),
∂xi ∂xj
where u is a convex function on an open set in Rn . There is a tremendous body of
work on these real and complex Monge-Ampère equations. In the Kähler setting,
the decisive contributions, dating back to the 1970’s, are due to Yau [9] and Aubin
[2]. The conclusion is, roughly stated, that PDE techniques reduce the problem
of finding a solution to that of finding a priori bounds for φL∞ . In the case
when λ < 0 this bound is easily obtained from the maximum principle; in the case
when λ = 0 the bound follows from a more sophisticated argument of Yau. This
leads, of course, to the renowned Calabi-Yau metrics on manifolds with vanishing
first Chern class. When λ is positive, so in algebro-geometric language we are
considering a Fano manifold V , the bound on φL∞ , and hence the existence of a
Kähler-Einstein metric, may hold or may not, depending on more subtle properties
of the geometry of the manifold V and, in a long series of papers, Tian has made
enormous progress towards understanding precisely when a solution exists. Notably,
Tian made a general conjecture in [8], which we will return to in the next section.
In the early 1980’s, Calabi initiated another problem [4]. His starting point was
to consider the L2 norm of the curvature tensor as a functional on the metrics and
seek critical points, called extremal Kähler metrics. The Euler-Lagrange equations
involve the scalar curvature
S = (ρ ∧ ω n−1 )/ω n .
The extremal condition is the equation
(5) ∂(grad S) = 0.
On the face of it this is a very intractable partial differential equation, combining
the full nonlinearity of the Monge-Ampère operator, which is embedded in the
definition of the curvature tensor, with high order: the equation being of order six
in the derivatives of the Kähler potential φ. Things are not, however, quiet as bad
as they may seem. The extremal equation asserts that the vector field grad S on V
is holomorphic so if, for example, there are no non-trivial holomorphic vector fields
on V the equation reduces to the constant scalar curvature equation
(6) S = σ,
where the constant σ is determined by V through Chern-Weil theory. This reduction
still leaves us with an equation of order four and, from the point of view of partial
differential equations, the difficulty which permeates the theory is that one cannot
directly apply the maximum principle to equations of this order. From the point
of view of Riemannian geometry, the difficulty which permeates the theory is that
control of the scalar curvature—in contrast to the Ricci tensor—does not give much
control of the metric.
CONJECTURES IN KÄHLER GEOMETRY 73
In the case when [ρ] = λ[ω] the constant scalar curvature and Kähler-Einstein
conditions are equivalent. Of course this is a global phenomenon: locally the equa-
tions are quite different. Obviously Kähler-Einstein implies constant scalar curva-
ture. Conversely, one has an identity
∂S = ∂ ∗ ρ,
so if the scalar curvature is constant the Ricci form ρ is harmonic. But λω is also
a harmonic form so if ρ and λω are in the same cohomology class they must be
equal, by the uniqueness of harmonic representatives.
There are two parabolic evolution equations associated to these problems. The
Ricci flow
∂ω
(7) = ρ − λω.
∂t
and the Calabi flow
∂ω
(8) = i∂∂S.
∂t
Starting with an extremal metric, the Calabi flow evolves the metric by diffeomeor-
phims (the one-parameter group generated by the vector field grad S): the geometry
is essentially unchanged. The analogues of general extremal metrics (nonconstant
scalar curvature) for the Ricci flow are the “Ricci solitons”
(9) ρ − λω = Lv ω,
where Lv is the Lie derivative along a holomorphic vector field v.
2. Conjectural picture
We present a precise algebro-geometric condition which we expect to be equiva-
lent to the existence of a constant scalar curvature Kähler metric. This conjecture is
formulated in [6]; in the Kähler-Einstein/Fano case the conjecture is essentially the
same as that made by Tian in [8]. An essential ingredient is the notion of the “Fu-
taki invariant”. Suppose L → V is a holomorphic line bundle with c1 (L) = 2π[ω0 ]
and with a Hermitian metric whose induced connection has curvature −iω. Suppose
we have a C∗ -action α on the pair V, L. Then we get a complex-valued function H
on V by comparing the horizontal lift of the vector field generating the action on
V with that generating the action on L. In the case when S 1 ⊂ C∗ acts by isome-
tries H is real-valued and is just the Hamiltonian in the usual sense of symplectic
geometry. The Futaki invariant of the C∗ -action is

(S − σ)H,
V
where σ is the average value of the scalar curvature S (a topological invariant).
There is another, more algebro-geometric, way of describing this involving deter-
minant lines. For large k we consider the line
Λmax H 0 (V ; Lk ).
The C∗ -action on (V, L) induces an action on this line, with some integer weight wk .
Let dk be the dimension of H 0 (V ; Lk ) and F (k) = wk /kdk . By standard theory,
this has an expansion for large k:
F (k) = F0 + F1 k−1 + F2 k−2 + . . . .
The equivariant Riemann-Roch formula shows that the Futaki invariant is just the
coefficient F1 in this expansion. Turning things around, we can define the Futaki
invariant to be F1 , the advantage being that this algebro-geometric point of view
extends immediately to singular varieties, or indeed general schemes.
Given (V, L), we define a “test configuration” of exponent r to consist of
(1) a scheme V with a line bundle L → V;
(2) a map π : V → C with smooth fibres Vt = π −1 (t) for non-zero t, such that
Vt is isomorphic to V and the restriction of L is isomorphic to Lr ;
(3) a C∗ -action on L → V covering the standard action on C.
We define the Futaki invariant of such a configuration to be the invariant of
the action on the central fibre π −1 (0), (with the restriction of L) - noting that
this may not be smooth. We say that the configuration is “destabilising” if the
Futaki invariant is bigger than or equal to zero and, in the case of invariant zero the
configuration is not a product V ×C. Finally we say that (V, L) is “K-stable” (Tian’s
terminology) if there are no destabilising configurations. Then our conjecture is:
Conjecture 1. Suppose (V, ω0 ) is a compact Kähler manifold and [ω0 ] =
2πc1 (L). Then there is a metric of constant scalar curvature in the class [ω0 ] if and
only if (V, L) is K-stable.
The direct evidence for the truth of this conjecture is rather slim, but we will
attempt to explain briefly why one might hope that it is true.
The first point to make is that “K-stability”, as defined above, is related to
the standard notion of “Hilbert-Mumford stability” in algebraic geometry. That is,
we consider, for fixed large k, the embedding V → CPN defined by the sections
of Lk which gives a point [V, L]k in the appropriate Hilbert scheme of subschemes
of CPN . The group SL(N + 1, C) acts on this Hilbert scheme, with a natural
linearisation, so we have a standard notion of Geometric Invariant Theory stability
of [V, L]k . Then K-stability of (V, L) is closely related to the stability of [V, L]k for
all sufficiently large k. (The notions are not quite the same: the distinction between
them is analogous to the distinction between “Mumford stability” and “Gieseker
stability” of vector bundles.)
The second point to make is that there is a “moment map” interpretation of
the differential geometric set-up. This is explained in more detail in [5] , although
the main idea seems to be due originally to Fujiki [7]. For this, we change our
point of view and instead of considering different metrics (i.e. symplectic forms) on
a fixed complex manifold we fix a symplectic manifold (M, ω) and consider the set
J of compatible complex structures on M . Thus a point J in J gives the same
data–a complex manifold with a Kähler metric–which we denoted previously by
(V, ω). The group G of “exact” symplectomorphisms of (M, ω) acts on J and one
finds that the map J → S − σ is a moment map for the action. In this way, the
moduli space of constant scalar curvature Kähler metrics appears as the standard
symplectic quotient of J . In such situations, one anticipates that the symplectic
quotient will be identified with a complex quotient, involving the complexification
of the relevant group. In the case at hand, the group G does not have a bona fide
complexification but one can still identify infinite-dimensional submanifolds which
play the role of the orbits of the complexification: these are just the equivalence
classes under the equivalence relation J1 ∼ J2 if (M, J1 ) and (M, J2 ) are isomorphic
as complex manifolds. With this identification, and modulo the detailed notion of
stability, Conjecture 1 becomes the familiar statement that a stable orbit for the
complexified group contains a zero of the moment map. Of course all of this is
a formal picture and does not lead by itself to any kind of proof, in this infinite-
dimensional setting. We do however get a helpful and detailed analogy with the
better understood theory of Hermitian Yang-Mills connections. In this analogy
the constant scalar curvature equation corresponds to the Hermitian Yang-Mills
equation, for a connection A on a holomorphic bundle over a fixed Kähler manifold,
FA .ω = constant.
The extremal equation ∂(grad S) = 0 corresponds to the Yang-Mills equation
d∗A FA = 0,
whose solutions, in the framework of holomorphic bundles, are just direct sums of
Hermitian-Yang-Mills connections.
Leaving aside these larger conceptual pictures, let us explain in a down-to-earth
way why one might expect Conjecture 1 to be true. Let us imagine that we can
solve the Calabi flow equation (8) with some arbitrary initial metric ω0 . Then,
roughly, the conjecture asserts that one of four things should happen in the limit
as t → ∞. (We are discussing this flow, here, mainly for expository purposes.
One would expect similar phenomena to appear in other procedures, such as the
continuity method. But it should be stressed that, in reality, there are very few
rigorous results about this flow in complex dimension n > 1: even the long time
existence has not been proved.)
(1) The flow converges, as t → ∞, to the desired constant scalar curvature
metric on V .
(2) The flow is asymptotic to a one-parameter family of extremal metrics
on the same complex manifold V , evolving by diffeomeorphisms. Thus
in this case V admits an extremal metric. Transforming to the other
setting, of a fixed symplectic form, the flow converges to a point in the
equivalence class defined by V . In this case V cannot be K-stable, since
the diffeomorphisms arise from a C∗ -action on V with non-trivial Futaki
invariant and we get a destabilising configuration by taking V = V × C
with this action.
(3) The manifold V does not admit an extremal metric but the transformed
flow Jt on J converges. In this case the limit of the transformed flow lies in
another equivalence class, corresponding to another complex structure V
on the same underlying differentiable manifold. The manifold V admits
an extremal metric. The original manifold V is not K-stable because there
is a destabilising configuration where V is diffeomorphic to V × C but the
central fibre has a different complex structure V (“jumping” of complex
structure).
(4) The transformed flow Jt on J does not converge to any complex structure
on the given underlying manifold but some kind of singularities develop.
However, one can still make sufficient sense of the limit of Jt to extract a
scheme from it, and this scheme can be fitted in as the central fibre of a
destabilising configuration, similar to case (3).
We stress again that this is more of a programme of what one might hope
eventually to prove, rather than a summary of what is really known. In the Kähler-
Einstein/Fano situation one can develop a parallel programme (as sketched by Tian
in [8]) for the Ricci flow (about which much more is known), where Ricci solitons
take the place of extremal metrics. In any event we hope this brings out the point
that one can approach two kinds of geometric questions, which on the face of it
seem quite different.
(1) ALGEBRAIC GEOMETRY PROBLEM. Describe the possible destabil-
ising configurations and in particular the nature of the singularities of the
central fibre (e.g. does one need schemes as opposed to varieties?).
(2) PDE/DIFFERENTIAL GEOMETRY PROBLEM. Describe the possible
behaviour of the Calabi flow/Ricci flow (or other continuity methods),
and the nature of the singularities that can develop.
The essence of Conjecture 1 is that these different questions should have the same
answer.
3. Toric varieties and a toy model

One can make some progress towards the verification of Conjecture 1 in the case
when V is a toric variety [6]. Such a variety corresponds to an integral polytope P
in Rn and the metric can be encoded in a convex function u on P . The constant
scalar curvature condition becomes the equation (due to Abreu [1])
∂ 2 uij
(10) = −σ,
i,j
∂xi ∂xj
where (uij ) is the inverse of the Hessian matrix (uij ) of second derivatives of u.
This formulation displays very well the way in which the equation is an analogue,
of order 4, of the real Monge-Ampère equation (4). The equation is supplemented by
boundary conditions which can be summarised by saying that the desired solution
should be an absolute minimum of the functional

(11) F(u) = − log det(uij ) + L(u),
P
where

(12) L(u) = u dρ − σ u dµ.
∂P P
Here dµ is Lebesgue measure on P and dρ is a natural measure on ∂P . (Each

codimension-1 face of ∂P is defined by a linear form, which we can normalise to
have coprime integer coefficients. This linear form and the volume element on Rn
induce a volume element on the face.) We wish to draw attention to one interesting
point, which can be seen as a very small part of Conjecture 1. Suppose there is a
non-trivial convex function g on P such that L(g) ≤ 0. Then one can show that F
does not attain a minimum so there is no constant scalar curvature metric. Suppose
on the other hand that f is a piecewise linear, rational convex function (that is, the
maximum of a finite set of rational affine linear functions). Then one can associate
a canonical test configuration to f and show that this is destabilising if L(f ) ≤ 0.
Thus we have
Conjecture 2. If there is a non-trivial convex function g on P with L(g) ≤ 0
then there is a non-trivial piecewise-linear, rational convex function f with L(f ) ≤
0.
This is a problem of an elementary nature, which was solved in [6] in the case
when the dimension n is 2, but which seems quite difficult in higher dimensions.
(And one can also ask for a more conceptual proof than that in [6] for dimension 2.)
On the other hand if this Conjecture 2 is false then very likely the same is true for
Conjecture 1: in that event one probably has to move outside algebraic geometry
to capture the meaning of constant scalar curvature.
Even in dimension 2 the partial differential equation (10) is formidable. We

can still see some interesting things if we go right down to dimension 1. Thus is
this case V is the Riemann sphere and P is the interval [−1, 1] in R. The equation
(10) becomes
d2
(13) ((u )−1 ) = −σ,
dx2
which one can readily solve explicitly. This is no surprise since we just get a
description of the standard round metric on the 2-sphere. To make things more
interesting we can consider the equation
d2
(14) ((u )−1 ) = −A,
dx2
where A is a function on (−1, 1). This equation has some geometric meaning, cor-
responding to a rotationally invariant metric on the sphere whose scalar curvature
is a given function A(h) of the Hamiltonian h for the circle action. The boundary
conditions we want are, in this case, u ∼ (1 ± x)−1 as x → ±1. But if we have a
solution with
(15) u (x) → ∞
as x → ±1, these are equivalent to the normalisations
1 1
(16) A(x)dx = 1, xA(x)dx = 0,
−1 −1
which we suppose hold.

We now consider the linear functional
1
(17) LA (u) = u(1) + u(−1) − u(x)A(x)dx.
−1
and
1
(18) FA (u) = − log(u (x))dx + LA (u).
−1
Then we have
Theorem 1. There is a solution to equations (14), (15) if and only if LA (f ) >
0 for all (non-affine) convex functions g on [−1, 1]. In this case the solution is an
absolute minimum of the functional FA .
To prove this we consider the function φ(x) = 1/u (x). This should satisfy the
equation φ = −A with φ → 0 at ±1. Thus the function φ is given via the usual
Green’s function 1
φ(x) = gx (y)A(y)dy,
−1
where gx (y) is a linear function of y on the intervals (−1, x) and (x, 1), vanishing
on the endpoints ±1 and with a negative jump in its derivative at y = x. Thus
−gx (y) is a convex function on [−1, 1] and
1
LA (−gx ) = A(y)gx (y)dy = φ(x).
−1
Thus our hypothesis (LA (g) > 0 on convex g) implies that the solution φ is positive
throughout (−1, 1) so we can form φ−1 and integrate twice to solve the equation
u = φ−1
thus finding the desired solution u. The converse is similar. The fact that the
solution is an absolute minimum follows from the convexity of the functional FA .
References
1. M. Abreu Kähler geometry of toric varieties and extremal metrics International J. Math. 9
(1998) 641-51
2. T. Aubin Equations du type Monge-Ampère sur les variétés kähleriennes compactes Bul. Sc.
Math. 102 (1978) 63-95
3. E. Calabi The space of Kähler metrics Proc. Int. Congress Math. Amsterdam 2 (1954) 206-7
4. E. Calabi Extremal Kähler metrics In: Seminar in Differential geometry (Yau,ed.) Annals
of Math. Study 102 Princeton U.P 1982 259-290
5. S. Donaldson Remarks on gauge theory, complex geometry and 4-manifold topology Fields
Medallists’ Lectures (Atiyah, Iagolnitzer eds.) World Scientific 1997 13-33
6. S. Donaldson Scalar curvature and stability of toric varieties Jour. Differential Geometry,
To appear.
7. A. Fujiki The moduli spaces and Kähler metrics of polarised algebraic varieties Sugaku 42
(1990) 173-91
8. G. Tian Kähler-Einstein metrics with positive scalar curvature Inv. Math. 130 (1997) 1-37
9. S-T. Yau On the Ricci curvature of a compact Kähler manifold and the complex Monge-
Ampère equation. I., Comm. Pure Appl. Math. 31 (1978) 339-411
Department of Mathematics,Imperial College, London

E-mail address: s.donaldson@ic.ac.uk
Volume 3, 2004
Branes, Calibrations and Supergravity
Jerome P. Gauntlett
Abstract. We attempt to provide an elementary and somewhat self contained

discussion of the construction of supergravity solutions describing branes wrap-
ping calibrated cycles, emphasising the geometrical aspects and focusing on
D=11 supergravity. Following a discussion of the role of special holonomy
backgrounds in D=11 supergravity, the basic membrane and fivebrane solu-
tions are reviewed and the connection with the AdS/CFT correspondence is
made. The world-volume description of branes is introduced and used to argue
that branes wrapping calibrated cycles in special holonomy manifolds preserve
supersymmetry. The corresponding supergravity solutions are constructed first
in an auxiliary gauged supergravity theory which is obtained via Kaluza-Klein
reduction.
1. Introduction
Supergravity theories in D=10 and D=11 spacetime dimensions play an impor-
tant role in string/M-theory since they describe the low-energy dynamics. There
are five different supersymmetric string theories, all in D=10. At low energies the
type IIA and type IIB string theories give rise to type IIA and type IIB super-
gravity, respectively, while the type I, and the two heterotic string theories all give
rise to type I supergravities. The five string theories are all related to each other,
possibly after compactification, by various dualities. There are also dualities which
relate string theory to M-theory, which resides in D=11. M-theory is much less
understood than string theory, but one of the most important things that is known
about it is that its low-energy effective action is given by D=11 supergravity.
Solutions to the supergravity equations of motion, particularly those that pre-
serve some supersymmetry, are of interest for many reasons. One reason is that
they are useful in studying compactifications from D=10 or 11 down to a lower
dimensional spacetime. By compactifying down to four spacetime dimensions, for
example, one might hope to make contact with particle physics phenomenology.
In addition to strings, it is known that string theory has a rich spectrum of other
extended objects or “branes”. Indeed, supergravity solutions can be constructed
describing the geometries around such branes, and these provide a very important
description of the branes. Similarly, there are membrane and fivebrane solutions
2000 Mathematics Subject Classification. 83E30, 83E50, 53C38, 81T30.

The author thanks Bobby Acharya, Nakwoo Kim, Dario Martelli, Stathis Pakis and Daniel
Waldram for enjoyable collaborations upon which some of these notes are based.
2004
79
80 JEROME P. GAUNTLETT
of D=11 supergravity, which will be reviewed later, which implies that M-theory
contains such branes. An important application of brane and more general inter-
secting brane solutions is that they can be used to effectively study the quantum
properties of black holes.
Supergravity solutions also provide powerful tools to study quantum field the-
ories. The most significant example is Maldacena’s celebrated AdS/CFT corre-
spondence [108], which conjectures that string/M-theory on certain supergravity
geometries that include anti-de Sitter (AdS) space factors is equivalent to certain
conformally invariant quantum field theories (CFTs). The supergravity approxi-
mation to string/M-theory allows one to calculate highly non-trivial information
about the conformal field theories.
The AdS/CFT correspondence is truly remarkable. On the one hand it states
that certain quantum field theories, that a priori have nothing to do with gravity,
are actually described by theories of quantum gravity (string/M-theory). Similarly,
and equally surprising, it also states that quantum gravity on certain geometries is
actually quantum field theory. As a consequence much effort has been devoted to
further understanding and generalising the correspondence.
The basic AdS/CFT examples arise from studying the supergravity solutions
describing planar branes in flat space, in the “near horizon limit”. Roughly speak-
ing, this is the limit close to the location of the brane. Here we shall discuss more
general supergravity solutions that describe branes that are partially wrapped on
various calibrated cycles within special holonomy manifolds. We will construct ex-
plicit solutions in the near horizon limit, which is sufficient for applications to the
AdS/CFT correspondence.
To keep the presentation simple, we will mostly restrict our discussion to D=11
supergravity. In an attempt to make the lectures accessible to both maths and
physics students we will emphasise the geometrical aspects and de-emphasise the
quantum field theory aspects. To make the discussion somewhat self contained, we
begin with some basic material; it is hoped that the discussion is not too pedestrian
for the physics student and not too vague for the maths student!
We start with an introduction to D=11 supergravity, defining the notion of
a bosonic solution of D=11 supergravity that preserves supersymmetry. We then
describe why manifolds with covariantly constant spinors, and hence with special
holonomy, are important. Following this we present the geometries describing pla-
nar membranes and fivebranes. These geometries have horizons, and in the near
horizon limit we obtain geometries that are products of AdS spaces with spheres,
which leads to a discussion of the basic AdS/CFT examples.
To motivate the search for new AdS/CFT examples we first introduce the world-
volume description of branes. Essentially, this is an approximation that treats
the branes as “probes” propagating in a fixed background geometry. We define
calibrations and calibrated cycles, and explain why such probe-branes wrapping
calibrated cycles in special holonomy manifolds preserve supersymmetry. The aim
is then to construct supergravity solutions describing such wrapped branes after
including the back-reaction of the branes on the special holonomy geometry.
The construction of these supergravity solutions is a little subtle. In particular,
the solutions are first constructed in an auxiliary gauged supergravity theory. We
will focus, for illustration, on the geometries corresponding to wrapped fivebranes.
For this case the solutions are first found in SO(5) gauged supergravity in D=7.
BRANES, CALIBRATIONS AND SUPERGRAVITY 81
This theory arises from the consistent truncation of the dimensional reduction of
D=11 supergravity on a four-sphere, as we shall discuss. This means that any
solution of the D=7 supergravity theory automatically gives a solution of D=11
supergravity. We will present several details of the construction of the solutions
describing fivebranes wrapping SLAG 3-cycles, and summarise more briefly the
other cases. We also comment on some aspects of the construction of the solutions
for wrapped membranes and D3-branes of type IIB supergravity.
We conclude with a discussion section that outlines some open problems as
well as a brief discussion of other related work on the construction of wrapped
NS-fivebranes of type IIB supergravity.
2. D=11 supergravity
The bosonic field content of D=11 supergravity [36] consists of a metric, g, and
a three-form C with four-form field strength G = dC which live on a D=11 manifold
which we take to be spin. The signature is taken to be mostly plus, (−, +, . . . , +).
In addition the theory has a fermionic gravitino, ψµ . The action with ψµ = 0 is
given by

1 √ 1 1
(2.1) S = 2 d11 x −gR − G ∧ ∗G − C ∧ G ∧ G,
2κ 2 6
and thus the bosonic equations of motion, including the Bianchi identity for the
four-form, are
1 1
Rµν = (G2µν − gµν G2 )
12 12
1
d∗G+ G∧G = 0
2
(2.2) dG = 0 ,
where G2µν = Gµσ1 σ2 σ3 Gν σ1 σ2 σ3 and G2 = Gσ1 σ2 σ3 σ4 Gσ1 σ2 σ3 σ4 , with µ, ν, σ =
0, 1, . . . , 10. The theory is invariant under supersymmetry transformations whose
infinitesimal form is given schematically by
δg ∼ ψ
δC ∼ ψ
(2.3) δψ ∼ ∇
ˆ + ψψ ,
where the spinor parameterises the variation and the connection ∇ ˆ will be given
shortly.
Of primary interest are bosonic solutions to the equations of motion that pre-
serve at least one supersymmetry. These are solutions to the equations of motion
with ψ = 0 which are left inert under a supersymmetry variation. From (2.3) we
see that δg = δC = 0 trivially, and hence we seek solutions to the equations of
motion that admit non-trivial solutions to the equation ∇ ˆ = 0.
As somewhat of an aside we mention a potentially confusing point. For the
theory to be supersymmetric, it is necessary that all of the fermions are Grassmann
odd (anti-commuting) spinors. However, since the only place that fermions enter
into bosonic supersymmetric solutions is via ∇
ˆ = 0 and since this is linear in we
can, and will, take to be a commuting (i.e. ordinary) spinor from now on. The
Grassmann odd character of the fermions is certainly important in the quantum
theory, but this will not concern us here.
To be more precise about the connection ∇ ˆ let us introduce some further nota-
tion. We will use the convention that µ, ν, . . . are coordinate indices and α, β, . . .
are tangent space indices, i.e., indices with respect to an orthonormal frame. The
D=11 Clifford algebra, Clif f (10, 1), is generated by gamma-matrices Γα satisfying
the algebra
(2.4) Γα Γβ + Γβ Γα = 2η αβ ,
with η = diag(−1, 1, . . . , 1). We will work in a representation where the gamma-
matrices are real 32 × 32 matrices acting on real 32 component spinors, with
Γ0 Γ1 . . . Γ10 = +1. Recall that Spin(10, 1) is generated by
1 αβ 1 α β
(2.5) Γ ≡ Γ Γ − Γβ Γα ,
4 8
and here we have introduced the notation that Γα1 ...αp is an anti-symmetrised
product of p gamma-matrices. The charge conjugation matrix is defined to be Γ0
and ¯ ≡ T Γ0 .
We can now write the condition for a bosonic configuration to preserve super-
symmetry as
(2.6) ∇ˆ µ ≡ ∇µ + 1 [Γµ ν1 ν2 ν3 ν4 − 8δ νµ1 Γν2 ν3 ν4 ]Gν ν ν ν = 0 ,

1 2 3 4
288
where ∇µ is the usual covariant derivative on the spin bundle
1
(2.7) ∇µ = (∂µ + ωµαβ Γαβ ) .
4
Observe that the terms involving the four-form in (2.6) imply that ∇ ˆ takes values in
the Clifford algebra and not just the Spin subalgebra. This is the typical situation
in supergravity theories but there are exceptions, such as type I supergravity, where
the connection takes values in the spin subalgebra and has totally anti-symmetric
torsion [135].
Non-trivial solutions to (2.6) are called Killing spinors. The nomenclature is
appropriate since if i , j are Killing spinors then K ijµ ≡ ¯i Γµ j are Killing vectors.
To see this, first define Ωij µν = ¯i Γµν j and Σij µ1 ...µ5 = ¯i Γµ1 ...µ5 j . Then use (2.6)
to show that [76]
1 1
(2.8) ∇µ Kνij = Ωij σ1 σ2 Gσ1 σ2 µν + Σij σ1 σ2 σ3 σ4 σ5 ∗ Gσ1 σ2 σ3 σ4 σ5 µν ,
6 6!
ij
and hence in particular ∇(µ Kν) = 0. It can also be shown that the “diagonal”
Killing vectors K ii , for each Killing spinor i , are either time-like or null [27]. The
zeroth components of these vectors in an orthonormal frame are given by (i )T i ,
and are clearly non-vanishing if and only if i is, and hence so is K itself.
It is useful to know under what conditions a geometry admitting a Killing spinor
will also solve the equations of motion. In the case when there is a time-like Killing
spinor, i.e. a Killing spinor whose corresponding Killing vector is time-like, it was
proved in [76] that the geometry will solve all of the equations of motion providing
that G satisfies the Bianchi identity dG = 0 and the four-form equation of motion
d ∗ G + (1/2)G ∧ G = 0. If all of the Killing spinors are null, it is necessary, in
addition, to demand that just one component of the Einstein equations is satisfied
[76].
Note that given the value of a Killing spinor at a point, the connection defines
a Killing spinor everywhere, via parallel transport. Also, as the Killing spinor
equation is linear, the Killing spinors form a vector space whose dimension n can,
in principle, be from 1, . . . , 32. The fraction of preserved supersymmetry is then
n/32. Although solutions are known preserving many fractions of supersymmetry, it
is not yet known if all fractions can occur (for some recent speculations on this issue
see [51]). A general characterisation of the most general supersymmetric geometries
preserving one time-like Killing spinor is presented in [76]. It was shown that the
geometry is mostly determined by a ten-dimensional manifold orthogonal to the
orbits of the Killing vector that admits an SU (5)-structure with rather weakly
constrained intrinsic torsion. The analogous analysis for null Killing spinors has
not yet been carried out. A complete classification of maximally supersymmetric
solutions preserving all 32 supersymmetries is presented in [58].
Most of our considerations will be in the context of D=11 supergravity, but
it is worth commenting on some features of M-theory that embellish D=11 super-
gravity. Firstly, the flux G, which is unconstrained in D=11 supergravity, satisfies
a quantisation condition in M-theory. Introducing the Planck length lp , via
(2.9) 2κ2 ≡ (2π)8 lp9 ,
for M-theory on a D=11 spin manifold Y we have [138]
1 λ
(2.10) G − ∈ H 4 (Y, Z) ,
(2πlp )3 2
where λ(Y ) = p1 (Y )/2 with p1 (Y ) the first Pontryagin class of Y , given below.
Note that since Y is a spin manifold, p1 (Y ) is divisible by two. Actually, more
generally it is possible to consider M-theory on unorientable manifolds admitting
pinors and some discussion can be found in [138].
A second point is that the low-energy effective action of M-theory is given by
D=11 supergravity supplemented by an infinite number of higher order corrections.
It is not yet known how to determine almost all of these corrections, but there is
one important exception. Based on anomaly considerations it has been shown that
the equation of motion for the four-form G is modified, at next order, by [137, 52]
1 (2πlp )6 2
(2.11) d∗G+ G∧G=− p1 − 4p2 ,
2 192
where the first and second Pontryagin forms are given by
1 1 1
(2.12) p1 = − 2
tr R2 , p2 = − 4
trR4 + (trR2 )2 .
8π 64π 128π 4
At the same order there are other corrections to the equations of motion and also
to the supersymmetry variations, but these have not yet been determined. Thus it
is not yet known how to fully incorporate this correction consistently with super-
symmetry but nevertheless it does have important consequences (see e.g. [132]).
As this correction will not play a role in the subsequent discussion, we will ignore
it.
In the next sub-sections we will review two basic classes of supersymmetric
solutions to D=11 supergravity. The first class are special holonomy manifolds and
the second class are the membrane and fivebrane solutions.
2.1. Special Holonomy. First consider supersymmetric solutions that have

vanishing four-form flux G, where things simplify considerably. The equations of
motion and the Killing spinor equation then become

Rµν = 0
(2.13) ∇µ = 0.
That is, Ricci-flat manifolds with covariantly constant spinors. The second condi-
tion implies that the manifolds have special holonomy. To see this, observe that it
implies the integrability condition
1
(2.14) [∇µ , ∇ν ] = Rµναβ Γαβ = 0 .
4
The subgroup of Spin(10, 1) generated by Rµναβ Γαβ gives the restricted holonomy
group H. Thus (2.14) implies that a Killing spinor must be invariant under H.
i.e. it must be a singlet under the decomposition of the 32 spinor representation
of Spin(10, 1) into H representations, and this constrains the possible holonomy
groups H that can arise.
Of most interest to us here are geometries R1,10−d × Md , which are the direct
product of (11 − d)-dimensional Minkowski space, R1,10−d , with a d-dimensional
Riemannian manifold Md , which we mostly take to be simply connected. (For a
discussion of supersymmetric solutions with Lorentzian special holonomy, see [27,
57]). The possible holonomy groups of the Levi-Civita connection on manifolds Md
admitting covariantly constant spinors is well known, and we now briefly summarise
the different cases.
Spin(7)-Holonomy: In d = 8 there are Riemannian manifolds with Spin(7)
holonomy. These have a nowhere vanishing self-dual Cayley four-form Ψ whose
components in an orthonormal frame can be taken as
Ψ = e1234 + e1256 + e1278 + e3456 + e3478 + e5678
(2.15)
+ e1357 − e1368 − e1458 − e1467 − e2358 − e2367 − e2457 + e2468 ,
where e.g. e1234 = e1 ∧ e2 ∧ e3 ∧ e4 . The Cayley four-form is covariantly constant
for Spin(7) manifolds and this is equivalent to Ψ being closed:
(2.16) dΨ = 0 .
Spin(7) holonomy manifolds have a single covariantly constant chiral Spin(8) spinor,
which we denote by ρ. Moreover, the Cayley four-form can be constructed as a bi-
linear in ρ:
(2.17) Ψmnpq = −ρ̄γmnpq ρ ,
where here m, n, p, q = 1, . . . , 8. For more discussion on the spinor conventions for
this case and those below, see appendix B of [75].
G2 -Holonomy: In d = 7 there are Riemannian manifolds with G2 holonomy.
These have a nowhere vanishing associative three-form φ whose components in an
orthonormal frame can be taken as
(2.18) φ = e246 − e235 − e145 − e136 + e127 + e347 + e567 .
The three-form is covariantly constant and this is in fact equivalent to the conditions
(2.19) dφ = d ∗ φ = 0 .
These geometries possess a single covariantly constant minimal d = 7 spinor ρ. The
associative three-form can be constructed from ρ via
(2.20) φmnp = −iρ̄γmnp ρ .
SU (n)-Holonomy: In d = 2n there are Calabi-Yau n-folds (CYn ) with SU (n)

holonomy. The cases relevant for D=11 supergravity have n = 2, 3, 4, 5. Calabi-Yau
manifolds are complex manifolds, with complex structure J, and admit a nowhere
vanishing holomorphic (n, 0)-form Ω. The Kähler form, which we also denote by
J, is obtained by lowering an index on the complex structure. In an orthonormal
frame we can take
J = e12 + e34 + · · · + e(2n−1)(2n)

(2.21) Ω = (e1 + ie2 )(e3 + ie4 ) . . . (e2n−1 + ie2n ) .
Both J and Ω are covariantly constant and this is equivalent to the vanishing of
the exterior derivative of the Kähler form and the holomorphic (n, 0)-form:
(2.22) dJ = dΩ = 0 .
These manifolds have a covariantly constant complex chiral spinor ρ. The complex
conjugate of this spinor is also covariantly constant. For n = 2, 4 the conjugate
spinor has the same chirality, while for n = 3, 5 it has the opposite chirality. J and
Ω can be written in terms of the spinor ρ as
Jmn = iρ† γmn ρ

(2.23) Ωm1 ...m2n = ρT γm1 ...m2n ρ .
Sp(n)-Holonomy: In d = 4n there are hyper-Kähler n-manifolds (HKn ) with

Sp(n) holonomy. The cases relevant for D=11 supergravity have n = 1, 2. These
admit three covariantly constant complex structures J a satisfying the algebra of
the imaginary quaternions
(2.24) J a · J b = −δ ab + abc J c .
If we lower an index on the J a we obtain three Kähler forms and the condition for
Sp(n)-holonomy is equivalent to them being closed:
(2.25) dJ a = 0 .
Note that when n = 1, since Sp(1) ∼ = SU (2), four-dimensional hyper-Kähler man-

ifolds are equivalent to Calabi-Yau two-folds. From the CY2 side, the extra two
complex structures are obtained from the holomorphic two-form via Ω = J 2 + iJ 1 .
The remaining case of interest for D=11 supergravity is eight-dimensional hyper-
Kähler manifolds when n = 2. In this case, in an orthonormal frame we can take
the three Kähler forms to be given by
J 1 = e12 + e34 + e56 + e78
(2.26) J 2 = e14 + e23 + e58 + e67
J 3 = e13 + e42 + e57 + e86 .
Each complex structure J a has a corresponding holomorphic (4, 0)-form given by
Ω1 = 12 J 3 ∧ J 3 − 12 J 2 ∧ J 2 + iJ 2 ∧ J 3
(2.27) Ω2 = 12 J 1 ∧ J 1 − 12 J 3 ∧ J 3 + iJ 3 ∧ J 1
Ω3 = 12 J 2 ∧ J 2 − 12 J 1 ∧ J 1 + iJ 1 ∧ J 2 .
These manifolds have three covariantly constant Spin(8) spinors of the same chi-
rality ρa , a = 1, 2, 3. The three Kähler forms can be constructed as
1
Jmn = −ρ̄2 γmn ρ3
(2.28) 2
Jmn = −ρ̄3 γmn ρ1
3
Jmn = −ρ̄1 γmn ρ2 .
In addition to these basic irreducible examples we can also consider Md to be
the direct product of two manifolds. A rather trivial possibility is to consider the
product of one of the above manifolds with a number of flat directions. Two non-
trivial possibilities are to consider the product CY3 × CY2 with SU (3) × SU (2)
holonomy, or the product CY2 × CY2 with SU (2) × SU (2) holonomy.
We have summarised the possibilities in table 1. We have also recorded the
amount of D=11 supersymmetry preserved by geometries of the form R1,10−d ×Md .
As we noted, this corresponds to the total number of singlets in the decomposition
of 32 of Spin(10, 1) into representations of H. Let us illustrate the counting for
the d = 8 cases. The spinor representation 32 of Spin(10, 1) decomposes into
Spin(2, 1) × Spin(8) representations as
(2.29) 32 → (2, 8+ ) + (2, 8− ) ,
where the subscripts refer to the chirality of the two spinor representations of
Spin(8). When M8 is a Spin(7)-manifold we have the further decomposition under
Spin(7) ⊂ Spin(8)
(2.30) 8+ → 7 + 1, 8− → 8 .
The singlet corresponds to the single covariantly constant, Spin(7) invariant, spinor
on the Spin(7)-manifold discussed above. From (2.29) we see that this gives rise
to two preserved supersymmetries and that they transform as a minimal two real
component spinor of Spin(2, 1). This is also described as preserving N = 1 super-
symmetry in D=3 spacetime dimensions corresponding to the R1,2 factor. When
M8 is Calabi-Yau under SU (4) ⊂ Spin(8) we have
(2.31) 8+ → 6 + 1 + 1, 8− → 4 + 4̄ .
The two singlets combine to form the complex covariantly constant spinor on CY4
mentioned above. In this case four supersymmetries are preserved, transforming as
two minimal spinors of Spin(2, 1), or N = 2 supersymmetry in D=3. When M8 is
hyper-Kähler, under Sp(2) ⊂ Spin(8) we have
(2.32) 8+ → 5 + 1 + 1 + 1, 8− → 4 + 4 ,
and six supersymmetries are preserved, or N = 3 in D=3. Similarly, when M8 is
the product of two Calabi-Yau two-folds eight supersymmetries are preserved, or
N = 4 in D=3. If we also allow tori, then when M8 is the product of a Calabi-Yau
two-fold with T 4 , sixteen supersymmetries are preserved, or N = 8 in D=3, while
the simple case of T 8 preserves all thirty-two supersymmetries, or N = 16 in D=3.
An important way to make contact with four-dimensional physics is to con-
sider geometries of the form R1,3 × M7 with M7 compact. If we choose M7 to
be T 7 then it preserves all 32 supersymmetries or N = 8 supersymmetry in D=4
spacetime dimensions. T 3 × CY2 preserves 16 supersymmetries or N = 4 in D=4,
S 1 × CY3 preserves 8 supersymmetries or N = 2 in D=4 and G2 preserves four
supersymmetries or N = 1 in D=4.
dim(M ) Holonomy Supersymmetry

10 SU (5) 2
10 SU (3) × SU (2) 4
8 Spin(7) 2
8 SU (4) 4
8 Sp(2) 6
8 SU (2) × SU (2) 8
7 G2 4
6 SU (3) 8
4 SU (2) 16
Table 1. Manifolds of special holonomy and the corresponding

amount of preserved supersymmetry.
N = 1 supersymmetry in four spacetime dimensions has many attractive phe-

nomenological features and this is the key reason for the recent interest in manifolds
with G2 holonomy. As discussed in Acharya’s lectures at this school, it is important
to emphasise that the most interesting examples from the physics point of view are
not complete. In addition one can use non-compact G2 holonomy manifolds very
effectively to study various quantum field theories in four spacetime dimensions
(see e.g. [9]).
Another important class of examples is to consider d = 7 manifolds of the
form S 1 /Z2 × CY3 where the Z2 action has two fixed planes. It can be shown that
the orbifold breaks a further one-half of the supersymmetries and one is again left
with four supersymmetries in four spacetime dimensions. These configurations are
related to the the strongly coupled limit of heterotic string theory compactified on
CY3 [93, 92].
In summary, when G = 0, geometries of the form R1,10−d × Md preserve su-
persymmetry when Md admits covariantly constant spinors and hence has special
holonomy. For physical applications, Md need be neither compact nor complete.
In the next section we will consider the basic solutions with G
= 0, the fivebrane
and the membrane solutions.
2.2. Membranes and Fivebranes. The simplest, and arguably the most im-
portant supersymmetric solutions with non-vanishing four-form are the membrane
and fivebrane solutions. Further discussion can be found in e.g. [134].
The fivebrane geometry is given by

ds2 = H −1/3 dξ i dξ j ηij + H 2/3 dxI dxI
(2.33) GI1 I2 I3 I4 = −cI1 I2 I3 I4 J ∂J H , c = ±1 ,
where i, j = 0, 1, . . . 5, I = 1, . . . , 5 and H = H(xI ). This geometry preserves 1/2 of
the supersymmetry. In the obvious orthonormal frame {H −1/6 dξ i , H 1/3 dxI }, the
16 Killing spinors are given by
(2.34) = H −1/12 0 ,
where 0 is a constant spinor, and satisfy
(2.35) Γ012345 = c .
Since Γ012345 squares to unity and is traceless, we conclude that the geometry
admits 16 independent Killing spinors.
This geometry satisfies the equations of motion providing that we impose the
Bianchi identity for G which implies that H is harmonic. If we take H to have a
single centre
α5 N
(2.36) H =1+ 3 , r 2 = xI xI ,
r
with N positive and α5 = πlp3 , then the solution carries cN units of quantised
magnetic four-form flux

1
(2.37) G = cN ,
(2πlp )3 S 4
with N a positive integer, consistent with (2.10). When c = +1 the solution
describes N coincident fivebranes, that are oriented along the dξ 0 ∧ dξ 1 ∧ . . . dξ 5
plane. When c = −1 the solution describes N coincident anti-fivebranes. Roughly
speaking, the fivebranes can be thought of as being located at r = 0, where the
solution appears singular. However, this is in fact a regular horizon and moreover,
it is possible to analytically continue to obtain a completely non-singular geometry
[78]. Thus it is not possible to say exactly where the fivebranes are located.
In the directions transverse to the fivebrane the metric becomes asymptotically
flat. We can thus calculate the ADM mass per unit volume, or tension, and we find
1
(2.38) T ension = N T5 , T5 = ,
(2π)5 lp6
where T5 is the tension of a single fivebrane (for a careful discussion of numerical
coefficients appearing in T5 and the membrane tension T2 below, see [46]). It is
possible to show that the supersymmetry algebra actually implies that the tension
of the fivebranes is fixed by the magnetic charge. This “BPS” condition is equivalent
to the geometry preserving 1/2 of the supersymmetry. Note also that if H is taken
to be a multi-centred harmonic function then we obtain a solution with the N
coincident fivebranes separated.
It is interesting to examine the near horizon limit of the geometry of N coin-
cident fivebranes, when r ≈ 0. By dropping the one from the harmonic function in
(2.36) we get
r i j (α5 N )2/3 2
(2.39) ds2 = dξ dξ ηij + dr + r 2 dΩ4 .
(α5 N )1/3 r2
where dΩ4 is the metric on the round four-sphere. After a coordinate transformation
we can rewrite this as
i j
dξ dξ ηij + dρ2 R2
(2.40) ds2 = R2 + dΩ4 ,
ρ2 4
which is just AdS7 × S 4 , in Poincaré coordinates, with the radius of the AdS7 given
by
(2.41) R = 2(πN )1/3 lp .
There are still N units of flux on the four-sphere. This geometry is in fact a
solution to the equations of motion that preserves all 32 supersymmetries. A closely
related fact is that the Lorentz symmetry SO(5, 1) of the fivebrane solution has been
enhanced to the conformal group SO(6, 2). The interpretation of this fact and the
SO(5) isometries of the four-sphere will be discussed in the next section. Before
doing so, we introduce the membrane solution.
The membrane geometry is given by

ds2 = H −2/3 dξ i dξ j ηij + H 1/3 dxI dxI
(2.42) C = cH −1 dξ 0 ∧ dξ 1 ∧ dξ 3 , c = ±1 ,
with, here, i, j = 0, 1, 2, I = 1, . . . 8 and H = H(xI ). This geometry preserves one
half of the supersymmetry. Using the orthonormal frame {H −1/3 dξ i , H 1/6 dxI } we
find that the Killing spinors are given by
(2.43) = H −1/6 0 ,
where 0 is a constant spinor, and satisfy the constraint
(2.44) Γ012 = c .
Since Γ012 squares to unity and is traceless, we conclude that the geometry admits
16 independent Killing spinors.
This geometry solves all of the equations of motion providing that we impose
the four-form equation of motion. This implies that the function H is harmonic.
Now take H to be
α2 N
(2.45) H =1+ 6 , r 2 = xI xI ,
r
with N a positive integer and α2 = 32π 2 lp6 . The solution carries cN units of
quantised electric four-form charge:

1
(2.46) ∗G = cN .
(2πlp )6 S 7
When c = ±1, the solution describes N coincident (anti-)membranes oriented along
the dξ 0 ∧ dξ 1 ∧ dξ 3 plane. Transverse to the membrane the solution tends to flat
space, and we can thus determine the ADM tension of the membranes. We again
find that it is related to the charge as dictated by supersymmetry
1
(2.47) T ension = N T2 , T2 = ,
(2π)2 lp3
where T2 is the tension of a single membrane. If the harmonic function is replaced
with a multi-centre harmonic function we obtain a solution with the membranes
separated.
The solution describing N coincident membanes appears singular at r ≈ 0, but
one can in fact show that this is a horizon. The solution can be extended across
the horizon and one finds a timelike singularity inside the horizon (see e.g. [134]),
which can be mapped onto a membrane source with tension T2 . To obtain the near
horizon geometry, r ≈ 0, we drop the one in the harmonic function (2.45), to find,
after a coordinate transformation,
i j
dξ dξ ηij + dρ2
(2.48) ds2 = R2 + 4R2 dΩ7 ,
ρ2
which is simply the direct product AdS4 × S 7 with the radius of AdS4 given by
1/6
N π2
(2.49) R= lp .
2
There are still N units of flux on the seven-sphere. This configuration is itself a
supersymmetric solution preserving all 32 supersymmetries. The SO(2, 1) Lorentz
symmetry of the membrane solution has been enhanced to the conformal group
SO(3, 2) and the seven-sphere admits an SO(8) group of isometries.
This concludes our brief review of the basic planar membrane and fivebrane
solutions. There is a whole range of more general solutions describing the intersec-
tion of planar membranes and fivebranes, and we refer to the reviews [64, 133] for
further details.
3. AdS/CFT Correspondence
In the last section we saw that D=11 supergravity admits supersymmetric solu-
tions corresponding to N coincident membranes or coincident fivebranes, and that
in the near horizon limit the metrics become AdS4 × S 7 or AdS7 × S 4 , respectively.
The famous conjecture of Maldacena [108] states that M-theory on these back-
grounds is equivalent to certain conformal field theories in three or six spacetime
dimensions, respectively. For a comprehensive review of this topic, we refer to [6],
but we would like to make a few comments in order to motivate the construction
of the supersymmetric solutions of D=11 supergravity presented in later sections.
The best understood example of the AdS/CFT correspondence actually arises
in type IIB string theory, so we first pause to introduce it. The low-energy limit of
type IIB string theory is the chiral type IIB supergravity [131, 97]. The bosonic
field content of the supergravity theory consists of a metric, a complex scalar, a
complex three-form field strength and a self-dual five-form field strength. The
theory admits a 1/2 supersymmetric three-brane, called a D3-brane. The metric of
the corresponding supergravity solution is given by

(3.1) ds2 = H −1 dξ i dξ j ηij + H dxI dxI ,
where i, j = 0, 1, 2, 3, I, J = 1, . . . 6 and H = H(xI ) is a harmonic function. If we
choose
α3 N
(3.2) H =1+ 2 ,
r
with N a positive integer, and α3 some constant with dimensions of length squared,
then the solution corresponds to N coincident D3-branes. The only other non-
vanishing field is the self-dual five-form and the solution, for suitably chosen α3 ,
carries N units of flux when integrated around a five-sphere surrounding the D3-
branes. In the near horizon limit, r ≈ 0, we get AdS5 × S 5 , with equal radii.
The boundary of AdS5 is the conformal compactification of four-dimensional
Minkowski space, M4 . The AdS/CFT conjecture states that type IIB string theory
on AdS5 × S 5 is equivalent (dual) to N = 4 supersymmetric Yang-Mills theory
with gauge group SU (N ) on M4 . This quantum field theory is very special as it
has the maximal amount of supersymmetry that a quantum field theory can have.
Moreover, it is a conformal field theory (CFT), i.e. invariant under the conformal
group. The AdS/CFT correspondence relates parameters in the field theory with
those of the string theory on AdS5 × S 5 . It turns out that perturbative Yang-Mills
theory can be a good description only when the radius R of the AdS5 is small, while
supergravity is a good approximation only when R is large and N is large. The
fact that these different regimes don’t overlap is a key reason why such seemingly
different theories could be equivalent at all.
The natural objects to consider in a conformal field theory are correlation

functions of operators. A precise dictionary between operators in the conformal
field theory and fields (string modes) propagating in AdS5 is given in [84, 139].
Moreover, in the supergravity approximation, correlation functions of the operators
are determined by the dependence of the supergravity action on the asymptotic
behaviour of the fields on the boundary. For example, the conformal dimensions of
the operators is determined by the mass of the fields.
It remains very unclear how to prove the AdS/CFT conjecture. Neverthe-
less, it has now passed an enormous number of tests. Amongst the simplest is to
compare the symmetries on the two sides. N = 4 super Yang-Mills theory has
an internal SO(6) “R-symmetry” and is invariant under the conformal group in
four dimensions, SO(4, 2). But SO(4, 2) × SO(6) are precisely the isometries of
AdS5 × S 5 . Moreover, after including the supersymmetry, we find that both sides
are invariant under the action of the supergroup SU (2, 2|4) whose bosonic subgroup
is SO(4, 2) × SO(6).
Let us now return to the near horizon geometries of the membrane and five-
brane. For the membrane, it is conjectured that M-theory on AdS4 × S 7 with
N units of flux on the seven-sphere is equivalent to a maximally supersymmet-
ric conformal field theory on the conformal compactification of three-dimensional
Minkowski space, the boundary of AdS4 . More precisely this conformal field the-
ory is the infrared (low energy) limit of N = 8 super-Yang-Mills theory with gauge
group SU (N ) in three dimensions. It is known that this theory has an SO(8)
R-symmetry. For this case, the SO(3, 2) × SO(8) isometries of AdS4 × S 7 cor-
respond to the conformal invariance and the R-symmetry of the conformal field
theory. After including supersymmetry we find that both sides are invariant under
the supergroup OSp(8|4).
For the fivebrane, it is conjectured that M-theory on AdS7 × S 4 with N units
of flux on the four-sphere is equivalent to a maximally supersymmetric chiral con-
formal field theory on the conformal compactification of six-dimensional Minkowski
space, the boundary of AdS7 . This conformal field theory is still rather mysteri-
ous and the AdS/CFT correspondence actually provides a lot of useful information
about it (assuming the correspondence is valid!). The SO(6, 2) × SO(5) isometries
of AdS7 × S 4 correspond to the conformal invariance and the SO(5) R-symmetry
of the field theory. After including supersymmetry we find that both sides are
invariant under the supergroup OSp(6, 2|4). For the membrane and fivebrane ex-
amples, when N is large, the radius of the AdS spaces are large and M-theory is
well approximated by D=11 supergravity.
Much effort has been devoted to further understanding and generalising the
AdS/CFT correspondence. Let us briefly discuss some of the generalisations that
have been pursued, partly to put the solutions we will construct later into some
kind of context, and partly as a rough guide to some of the vast literature on the
subject.
The three basic examples of the AdS/CFT correspondence relate string/M-
theory on AdSd+1 × sphere geometries to conformally invariant quantum field the-
ories in d = 3, 4 and 6 with maximal supersymmetry. One direction is to find
new supersymmetric solutions of supergravity theories that are the products, pos-
sibly warped1, of AdSd+1 with other compact spaces, preserving less than maximal
supersymmetry. Since the isometry group of AdSd+1 is the conformal group, this in-
dicates that these would be dual to new superconformal field theories in d spacetime
dimensions. Non-supersymmetric solutions with AdS factors are also of interest,
as they could be dual to non-supersymmetric CFTs. However, one has to check
whether the solutions are stable at both the perturbative and non-perturbative
level, which is very difficult in general. By contrast, in the supersymmetric case,
stability is guaranteed from the supersymmetry algebra. The focus has thus been
on supersymmetric geometries with AdS factors.
One class of examples, discussed in [103, 2, 114], is to start with the fivebrane,
membrane or D3-brane geometry (2.33), (2.42) or (3.1), respectively, and observe
that if the flat space transverse to the brane is replaced by a manifold with special
holonomy as in table 1, and the associated flux left unchanged, then the resulting
solution will still preserve supersymmetry but will preserve, in general, a reduced
amount. Now let the special holonomy manifold be a cone over a base X, i.e. let
the metric of the transverse space be
(3.3) dr 2 + r 2 ds2 (X) ,
X must be Einstein and have additional well known properties to ensure that
the metric has special holonomy. For example a five-dimensional X should be
Einstein-Sasaki in order that the six-dimensional cone is Calabi-Yau. Apart from
the special case when X is the round sphere these spaces have a conical singularity
at r = 0. To illustrate this construction explicitly for the membrane, one replaces
the eight-dimensional flat space transverse to the membrane in (2.42) with an eight-
dimensional cone with special holonomy:

(3.4) ds2 = H −2/3 dξ i dξ j ηij + H 1/3 dr 2 + r 2 ds2 (X) ,
where H = 1 + α2 N/r 6 , as before. Clearly this can be interpreted as N coincident
membranes sitting at the conical singularity. By considering the near horizon limit
of (3.4), r ≈ 0, one finds that it is now the direct product AdS4 × X and this
provides a rich class of new AdS/CFT examples.
Another generalisation is to exploit the fact that the maximally supersymmetric
conformal field theories can be perturbed by certain operators. In some cases,
under renormalisation group flow, these quantum theories will flow in the infrared
(low energies) to new superconformal field theories, with less supersymmetry. It is
remarkable that corresponding dual supergravity solutions can be found. Given the
dictionary between operators in the conformal field theory in d dimensions and fields
in AdSd+1 mentioned above, the perturbation of the conformal field theory should
correspond to dual supergravity solutions that asymptotically tend to AdSd+1 in a
prescribed way. Now on rather general grounds it can be argued that this AdSd+1
boundary corresponds to the ultraviolet (UV) of the dual perturbed conformal
field theory, and that going away from the boundary into the interior corresponds
to going to the infrared (IR) in the dual quantum field theory [136]. This can be
seen, for example, by studying the action of the conformal group on AdSd+1 and
on the correlation functions in the d-dimensional conformal field theory. Thus if
the perturbed conformal field theory is flowing to another conformal field theory
1A warped product of two spaces with coordinates x and y corresponds to a metric of the
form f (y)ds2 (x) + ds2 (y), for some function f (y).
in the IR, we expect that there should be supergravity solutions that interpolate
from the perturbed AdS boundary to another AdS region in the interior. Indeed
such solutions can be found (see for example [102, 59, 126]).
The above examples concern gravity duals of superconformal field theories or
flows between superconformal field theories. Another way to generalise the cor-
respondence is to find supergravity solutions that are dual to non-conformal field
theories. For example, one might study perturbed superconformal field theories
that flow in the infrared to non-conformal phases, such as Coulomb, Higgs and
confining phases. The corresponding dual supergravity solutions should still have
an asymptotic AdSd+1 boundary, describing the perturbed conformal field theory,
but they will no longer interpolate to another AdSd+1 region but to different kinds
of geometry dual to the different phases. For an example of these kinds of solutions
see [125]. Often the geometries found in the IR are singular and further analysis
is required to determine the physical interpretation.
The generalisation we will discuss in the rest of the lectures was initiated by
Maldacena and Nuñez [110]. The idea is to construct supergravity solutions de-
scribing branes wrapping calibrated cycles in manifolds of special holonomy, in the
near horizon limit. The next section will explain the background for attempting
this, and in particular why such configurations preserve supersymmetry. The sub-
sequent section will describe the construction of the solutions using the technical
tool of gauged supergravity. The D=11 solutions describing wrapped membranes
and fivebranes and the D=10 solutions describing wrapped D3-branes provide a
large class of solutions with dual field theory interpretations. The simplest, and
perhaps most important, solutions are warped products of AdS spaces, cycles with
Einstein metrics and spheres. The presence of the AdS factor for these solutions
implies that they provide a large class of new AdS/CFT examples. In addition, as
we shall discuss, there are more complicated solutions that describe flows from a
perturbed AdS boundary, describing the UV, to both conformal and non-conformal
behaviour in the IR.
Type IIB string theory also contains NS fivebranes. In the discussion section we
will briefly discuss how supergravity solutions describing NS fivebranes wrapped on
calibrated cycles can be used to study non-conformal field theories. A particularly
interesting solution [111] (see also [33]) gives rise to a dual quantum field theory
that has many features of N = 1 supersymmetric Yang-Mills theory in four dimen-
sions. This is a very interesting theory as it has many features that are similar to
QCD. Two other interesting ways of studying N = 1 supersymmetric Yang-Mills
theory in four dimensions can be found in [104] and [128]. Related constructions
with N = 2 supersymmetry can be found in [69, 22, 20, 127] (for reviews see
[5, 19, 21]). The explicit regular solutions found in [104] are examples of a more
general construction discussed in [42] (see [40] for a review).
4. Brane worldvolumes and calibrations

The supergravity brane solutions that were presented in section 2 describe static
planar branes of finite tension and infinite extent. Physical intuition suggests that
these branes should become dynamical objects if they are perturbed. Moreover,
we also expect that branes with different topologies should exist. On the other
hand it is extremely difficult to study these aspects of branes purely from the
supergravity point of view. Luckily, there are alternative descriptions of branes
which can be used. In this section we will describe the low-energy world-volume
description of branes. Essentially, this is a probe-approximation in which the branes
are taken to be very light and hence propagate in a fixed background geometry
with no back-reaction. We will use this description to argue that branes can wrap
calibrated cycles in manifolds of special holonomy while preserving supersymmetry
[12, 13, 73, 79]. This then provides the motivation to seek D=11 supergravity
solutions that describe a large number of such wrapped branes, when the back-
reaction on the geometry will be very significant. We will construct the solutions in
the next section in the near horizon limit, which is the limit relevant for AdS/CFT
applications.
It will be useful to first review some background material concerning calibra-
tions on manifolds of special holonomy, before turning to the brane world-volume
theories.
4.1. Calibrations. A calibration [90] on a Riemannian manifold M is a p-
form ϕ satisfying two conditions:
dϕ = 0
(4.1) ϕ|ξp ≤ V ol|ξp , ∀ξ p ,
where ξp is any tangent p-plane, and V ol is the volume form on the cycle induced
from the metric on M . A p-cycle Σp is calibrated by ϕ if it satisfies
(4.2) ϕ|Σp = V ol|Σp .
A key feature of calibrated cycles is that they are minimal surfaces in their homology
class. The proof is very simple. Consider another cycle Σ such that Σ − Σ is the
boundary of a (p + 1)-dimensional manifold Ξ. We then have

(4.3) V ol(Σ) = ϕ= dϕ + ϕ= ϕ ≤ V ol(Σ ) .
Σ Ξ Σ Σ
The first equality is due to Σ being calibrated. The second equality uses Stokes’
theorem. The remaining steps use the closure of ϕ and the second part of the
definition of a calibration.
We will only be interested in calibrations that can be constructed as bilinears
of spinors, for reasons that will soon become clear. The general procedure for such
a construction was first discussed in [44, 89]. In fact all of the special holonomy
manifolds that we discussed earlier have such calibrations. We now summarise
the various cases, noting that the the spinorial construction and the closure of the
calibrations was already presented in section 2. That the calibrations also satisfy
the second condition in (4.1) was shown, for almost all cases, in [90]; it is also
straightforward to establish using the spinorial construction.
On Spin(7)-holonomy manifolds the Cayley four-form Ψ is a calibration and
the 4-cycles calibrated by Ψ are called Cayley 4-cycles.
G2 -holonomy manifolds have two types of calibrations, φ and ∗φ. The former
calibrates associative 3-cycles, while the latter calibrates co-associative 4-cycles.
Calabi-Yau n-folds generically have two classes of calibrations. The first class
1 n
is the Kähler calibrations given by n! J , where the wedge product is used. These
calibrate even 2n-dimensional cycles and this is equivalent to the cycles being holo-
morphic. The second type of calibration is the special Lagrangian (SLAG) cali-
bration given by the real part of the holomorphic n-form eiθ Ω, where the constant
θ ∈ S 1 , and these calibrate special Lagrangian n-cycles. Recall that for our purposes
n = 2, 3, 4, 5. When n = 2, there is no real distinction between SLAG and Kähler

2-cycles since the cycles that are Kähler with respect to one complex structure are
SLAG with respect to another (recall that CY2 = HK1 ). When n = 4 there are also
4-cycles that are calibrated by 12 J 2 + Re(eiθ Ω) – these are in fact Cayley 4-cycles
if we view the Calabi-Yau four fold as a special example of a Spin(7)-manifold.
Hyper-Kähler manifolds in eight dimensions are special cases of Calabi-Yau
four-folds. They thus admit Kähler and special Lagrangian calibrations with respect
to each complex structure. They also admit Cayley calibrations as just described. In
addition there are quaternionic calibrations [45] that calibrate quaternionic 4-cycles
which are Kähler with respect to all three complex structures: V ol = 12 (J 1 )2 =
1 2 2 1 3 2
2 (J ) = 2 (J ) , when restricted to the cycle. For example, with respect to the
hyper-Kähler structure (2.26), we see that e1234 is a quaternionic 4-cycle2. Of
more interest to us will be the complex-Lagrangian (C-Lag) calibrations [45] which
calibrate 4-cycles that are Kähler with respect to one complex structure and special
Lagrangian with respect to the other two: for example, V ol = 12 (J 1 )2 = Re(Ω2 ) =
−Re(Ω3 ) when restricted to the cycle. Referring to (2.26) and (2.27) we see that
e1256 is an example of such a C-Lag 4-cycle.
In constructing supergravity solutions describing branes wrapping calibrated
cycles in the next section, it will be very important to understand the structure of
the normal bundle of calibrated cycles. Let us summarise some results of Mclean
[113]. The tangent bundle of the special holonomy manifold restricted to the cycle
splits into the tangent bundle of the cycle plus the normal bundle
(4.4) T (M )|Σ = T (Σ) ⊕ N (Σ) .
In some, but not all cases, the normal bundle, N (Σ), is intrinsic to Σ. Given a
calibrated cycle, one can also ask which normal deformation, if any, is a normal
deformation through a family of calibrated cycles.
A simple case to describe are the special Lagrangian cycles, where N (Σ) is
intrinsic to Σ. It is not difficult to show that on a special Lagrangian cycle the
Kähler form, J, restricted to Σ vanishes. Thus, for any vector field V on Σ we
have that Jij V j are the components of a one-form on Σ which is orthogonal to all
vectors on Σ. In other words, J i j V j defines a normal vector field. Hence N (Σ) is
isomorphic to T (Σ).
In addition, the normal deformation described by the vector V is a normal
deformation through the space of special Lagrangian submanifolds if and only if the
one-form with components Jij V j is harmonic. Thus if Σ is compact, the dimension
of the moduli space of special Lagrangian manifolds near Σ is given by the first
Betti number, β 1 (Σ) = dim[H 1 (Σ, R)]. In particular if β 1 = 0, then Σ has no
harmonic one-forms and hence it is rigid as a special Lagrangian submanifold.
Next consider co-associative 4-cycles in manifolds of G2 holonomy, for which
N (Σ) is also intrinsic to Σ. In fact N (Σ) is isomorphic to the bundle of anti-self-
dual two-forms on Σ. A normal vector field is a deformation through a family of co-
associative 4-cycles if and only if the corresponding anti-self-dual two-form is closed
and hence harmonic. Thus if Σ is compact the dimension of the moduli space of co-
associative 4-cycles near Σ is given by the Betti number, β− 2
(Σ) = dim[H−2
(Σ, R)].
2
In particular if β− = 0, then Σ is rigid as a co-associative submanifold.
2Supergravity solutions describing fivebranes wrapping quaternionic 4-cycles in Ê 8 , which

are necessarily linear [45], were constructed in [120, 121].
The normal bundles of associative 3-cycles in manifolds of G2 holonomy are not

intrinsic to Σ in general. The normal bundle is given by S ⊗ V where S is the spin
bundle of Σ (oriented three-manifolds are always spin) and V is a rank two SU (2)
bundle. In other words the normal directions are specified by two-dimensional
spinors on Σ that carry an additional SU (2) index. A normal vector field gives a
deformation through a family of associative 3-cycles if and only if the corresponding
twisted spinor is harmonic, i.e. in the kernel of the twisted Dirac operator. In the
special case that the bundle V is trivial, the spinor must be harmonic. For example,
if Σ is an associative three-sphere and V is trivial, as in the G2 manifold constructed
in [28], then it is rigid.
The deformation theory of Cayley 4-cycles in manifolds of Spin(7) holonomy
has a similar flavour to the associative 3-cycles. The normal bundle is given by
S− ⊗ V where S− is the bundle of spinors of negative chirality on Σ and V is a rank
two SU (2) bundle. Although not all 4-cycles admit a spin structure, all Cayley
4-cycles admit such twisted spinors. A normal vector field gives a deformation
through a family of Cayley 4-cycles if and only if the corresponding twisted spinor
is harmonic, i.e. in the kernel of the twisted Dirac operator. In the special case that
the bundle V is trivial the spinor must be harmonic. For example, if Σ is a Cayley
four-sphere and V is trivial, as in the Spin(7) manifold constructed in [28], then it
is rigid.
Finally, let us make some comments concerning the Kähler cycles. These
cycles reside in Calabi-Yau manifolds M which have vanishing first Chern class,
c1 [T (M )] = 0. Since c1 [T (M )|Σ ] = c1 [T (Σ)] + c1 [N (Σ)] we conclude that in general
(4.5) c1 [N (Σ)] = −c1 [T (Σ)] .
If one considers the special case that Σ is a divisor i.e. a complex hypersurface
(i.e. real codimension two), then N (Σ) is intrinsic to Σ. Indeed one can show that
N (Σ) ∼ = K(Σ) where K(Σ) is the canonical bundle of Σ.
4.2. Membrane world-volume theory. Let us now turn to the world-vol-
ume theory of branes, beginning with membranes [15, 16]. We will consider the
membranes to be propagating in a fixed D=11 geometry which is taken to be a
bosonic solution to the equations of motion of D=11 supergravity, with metric g
and three-form C. The bosonic dynamical fields are maps xµ (σ) from the world-
volume of the membrane, W , to the D=11 target space geometry. If we let σ i
be coordinates on W with i = 0, 1, 2, and xµ be coordinates on the D=11 target
geometry with µ = 0, 1, . . . , 10, the reparametrisation invariant action is given by

1/2 1
(4.6) S = T2 d3 σ [−det ∂i xµ ∂j xν gµν (x)] + ijk ∂i xµ1 ∂j xµ2 ∂k xµ3 Cµ1 µ2 µ3 .
W 3!
The first term is just the volume element of the pullback of the metric to the world-
volume and is called the Nambu-Goto action. The second term arises because the
membrane carries electric four-form charge; it generalises the coupling of an electri-
cally charged particle to a vector potential. The full action also includes fermions
and is invariant under supersymmetry when the D=11 target admits Killing spinors.
The supersymmetry of brane-world volume theories is actually quite intricate, but
luckily we will not need many of the details. The reason is similar to the reason
that we didn’t need to discuss such details for D=11 supergravity. Once again,
our interest is bosonic solutions to the equations of motion that preserve some su-
persymmetry. For such configurations, the supersymmetry variation of the bosonic
fields automatically vanishes, and hence one only needs to know the supersymmetry
variation of the fermions, and this will be mentioned later.
To get some further insight, consider static bosonic D=11 backgrounds with
vanishing three-form, C = 0,and write the metric as
(4.7) ds2 = −dt2 + gM N dxM dxN ,
where M, N = 1, 2, . . . , 10. If we substitute this into (4.6) and partially fix the
reparametrisation invariance by choosing σ 0 = t the membrane action gives rise to
the energy functional

(4.8) E = T2 d2 σ [mab ]1/2 ,
W

where a, b = 1, 2, W is the spatial part of the world-volume, and mab is the spatial
part of the induced world-volume metric given by
(4.9) mab = ∂a xM ∂b xN gM N .
In other words, the energy is just given by the tension of the membrane times the
spatial area of the membrane. Now, static solutions to the equations of motion
minimise the energy functional. Thus static configurations minimise the area of
the membrane, which implies that the spatial part of the membrane is a minimal
surface. This is entirely in accord with expectations: the tension of the mem-
brane tends to make it shrink. It should be noted that the minimal surfaces can
be of infinite extent: the simplest example is an infinite flat membrane in D=11
Minkowski space. Of most interest to us will be membranes wrapping compact
minimal surfaces.
Let us further restrict to background geometries of the form R1,10−d × Md
with vanishing three-form that preserve supersymmetry. In other words Md has
special holonomy as discussed in section 2. The membrane world-volume theory is
supersymmetric with the number of supersymmetries determined by the number of
Killing spinors. Static membrane configurations that preserve supersymmetry wrap
cycles called supersymmetric cycles. We now argue that supersymmetric cycles are
equivalent to calibrated cycles with the associated calibration being constructed
from the Killing spinors.
In order that a bosonic world-volume configuration be supersymmetric the
supersymmetry variation of the fermions must vanish. Given the explicit super-
symmetry variations, it is simple to show this implies that [12]
(4.10) (1 − Γ) = 0 ,
where is a D=11 Killing spinor and the matrix Γ is given by
1
Γ = √ Γ0 γ
det m

1 ab
(4.11) γ = ∂a xM ∂b xN ΓM N ,
2!
where [ΓM , ΓN ]+ = 2gM N . The matrix Γ satisfies Γ2 = 1 and is Hermitian, Γ† = Γ.
We now calculate
(1 − Γ) (1 − Γ) (1 − Γ) (1 − Γ) 2
(4.12) † = † = || || ≥ 0 .
2 2 2 2
We thus conclude that † ≥ † Γ with equality if and only (1 − Γ) = 0, which

is equivalent to the configuration being supersymmetric. The inequality can be
rewritten
√
(4.13) det m ≥ † Γ0 γ = −¯
γ .
Thus the two-form defined by
1
(4.14) ϕ=− ¯ΓM N dxM ∧ dxN ,
2!
satisfies the second condition in (4.1) required for a calibration. One can argue
that the supersymmetry algebra [88] implies that it is closed and hence is in fact a
calibration (we will verify this directly in a moment). Moreover, the inequality is
saturated if and only if the membrane is wrapping a supersymmetric cycle, and we
see that this is equivalent to the cycle being calibrated by (4.14).
The only two-form calibrations on special holonomy backgrounds are Kähler
calibrations, and indeed ϕ is in fact equal to a Kähler two-form on the background.
To see this very explicitly and to see how much supersymmetry is preserved when
a membrane wraps a Kähler 2-cycle, first consider the D=11 background to be R ×
CY5 . We noted earlier that this background preserves two D=11 supersymmetries:
in a suitable orthonormal frame, the two covariantly constant D=11 spinors can be
taken to satisfy the projections (see, for example, the discussion in appendix B of
[75]):
(4.15) Γ1234 = Γ3456 = Γ5678 = −Γ78910 = − .
Note that these imply that Γ012 = . Substituting either of these spinors into
(4.14) we find that ϕ is precisely the Kähler calibration on CY5 :
(4.16) ϕ = J = e12 + e34 + e56 + e78 − e910 .
Consider now a membrane wrapping a Kähler 2-cycle in CY5 , i.e. its worldvolume
is R × Σ with Σ ⊂ CY5 . To be concrete, consider V ol(Σ) = e12 |Σ . We then
find that the supersymmetry condition (4.10) implies that Γ012 = , which is
precisely the projection on the spinors that we saw in the supergravity solution for
the membrane (2.44). For this case we see that this projection does not constrain
the two supersymmetries satisfying (4.15) further and thus a membrane can wrap
a Kähler 2-cycle in a CY5 “for free”. Clearly if we wrapped an anti-membrane,
satisfying Γ012 = −, there would be no surviving supersymmetry3. Let us now
consider the background to be R × CY4 × R2 . This preserves four supersymmetries
satisfying projections which we can take to be
(4.17) Γ1234 = Γ3456 = Γ5678 = ∓Γ78910 = − .
Two of these satisfy Γ012 = and two satisfy Γ012 = −. After substituting into
(4.14) they give rise to two Kähler forms on CY4 × R2 :
(4.18) ϕ = J = e12 + e34 + e56 + e78 ∓ e910 .
If we now wrap the membrane on a Kähler 2-cycle with V ol(Σ) = e12 |Σ , then
we see that the supersymmetry condition (4.10), Γ012 = , preserves two of the
supersymmetries. Similarly, if we wrapped an anti-membrane satisfying Γ012 = −
it would also preserve two supersymmetries.
3Note that if we change the orientation by switching e10 → −e10 , then (4.15) would assume
a more symmetric form and we would find that we could wrap an anti-membrane along Σ for free.
The amount of supersymmetry preserved by any brane wrapping a calibrated

cycle in a special holonomy background can be worked out in a similar way: one
considers a convenient set of projections for the background geometry and then
supplements them with those of the wrapped brane (or anti-brane). In almost all
cases, wrapping the brane breaks 1/2 of the supersymmetries preserved by the spe-
cial holonomy background. We have summarised the possibilities for the membrane
in table 2.
Calibration World-Volume Supersymmetry

Kähler R × (Σ2 ⊂ CY2 ) 8
R × (Σ2 ⊂ CY3 ) 4
R × (Σ2 ⊂ CY4 ) 2
R × (Σ2 ⊂ CY5 ) 2
Table 2. The different ways in which membranes can wrap cali-

brated cycles and the amount of supersymmetry preserved.
The action (4.6) describes the dynamics of a membrane propagating in a fixed

D=11 supergravity background. Such a membrane is often called a “probe mem-
brane”. Of course, the dynamics of the membrane will back-react on the geometry,
and so one should really supplement the D=11 supergravity action with the world-
volume action:
(4.19) S = SD=11 + SW V .
If there are many coincident membranes then this back-reaction could be large.
We have been emphasising the geometric aspects of the membrane world-
volume theory. The world-volume theory is also a quantum field theory. To
gain some insight into this aspect, let us restrict the target geometry to be D=11
Minkowski space and the world-volume to be R1,2 . Now fix the reparametrisation
invariance completely by setting σ 0 = t, σ 1 = x1 , σ 2 = x2 . We can then expand the
determinant to get

1
(4.20) S = T2 d3 σ(− ∂a xI ∂ a xI + f ermions + . . . ) ,
2
where we have dropped an infinite constant and the dots refer to higher derivative
terms. The eight scalar fields describe the eight transverse fluctuations of the
membrane. After quantisation, this action gives a three-dimensional quantum field
theory, with eight scalar fields plus fermions, that preserves 16 supersymmetries or
N = 8 supersymmetry in three dimensions. This quantum field theory is interacting
with gravity, via (4.19), but if we take the limit, lp → 0, it decouples from gravity.
In other words, in this decoupling limit we get a three-dimensional quantum field
theory living on the world-volume of the membrane. When there are N coincident
branes, the quantum field theory is much more complicated. There is a piece
describing the centre of mass dynamics of the branes given by (4.20) with T2 → N T2
and there is another piece describing the interactions between the membranes. This
latter theory is known to be a superconformal field theory that arises as the IR limit
of N = 8 supersymmetric Yang-Mills theory in three dimensions. Recall that this is
precisely the superconformal field theory that is conjectured to be dual to M-theory

on AdS4 × S 7 .
The important message here is that the supergravity solution describing the
membranes in the near horizon limit, AdS4 × S 7 , is conjectured to be equivalent to
the quantum field theory arising on the membrane world-volume theory, in a limit
which decouples gravity.
Now consider a more complicated example. Take the D=11 background to be
of the form R1,6 ×CY2 with a probe membrane wrapping a Kähler 2-cycle Σ ⊂ CY2 .
i.e. the world-volume of the membrane is R × (Σ ⊂ CY2 ). There is again a quantum
field theory living on the brane interacting with gravity. In the decoupling limit,
lp → 0 and keeping the volume of Σ fixed, we get a supersymmetric quantum field
theory on R×Σ. If Σ is compact, the low-energy infrared (IR) limit of this quantum
field theory corresponds to length scales much larger than the size of Σ. In this IR
limit the quantum field theory on R × Σ behaves like a quantum field theory on the
time direction R, which is just a quantum mechanical model.
If we could construct a supergravity solution describing membranes wrapping
such Kähler 2-cycles, in the near horizon limit, we would have an excellent candi-
date for an M-theory dual for this quantum field theory on R × Σ. Moreover, if
the supergravity solution has an AdS2 factor, it would strongly indicate that the
corresponding dual quantum mechanics, arising in the IR limit, is a superconformal
quantum mechanics. These kinds of supergravity solutions have been found [70]
and the construction will be described in the next section.
It is worth making some further comments about the quantum field theory on
R × Σ. For a single membrane the physical bosonic degrees of freedom describe
the transverse deformations of the membrane. In the case of a membrane with
world-volume R1,2 in R1,10 we saw above in (4.20) that there are eight scalar fields
describing these deformations. Geometrically, they are sections of the normal bun-
dle, which is trivial in this case. Now consider, for example, a membrane with
world-volume R × (Σ ⊂ CY2 ). There are six directions transverse to the membrane
that are also transverse to the CY2 and these lead to six scalar fields. There are
also two directions transverse to the membrane that are tangent to the CY2 : these
give rise to a section of the normal bundle. As we discussed earlier the normal
deformations of a Kähler 2-cycle Σ ⊂ CY2 (which are also SLAG 2-cycles with
respect to another complex structure) are specified by one-forms on Σ.
This “transition” from scalars to one-forms is intimately connected with the
way in which the field theory on R × Σ realises supersymmetry. In particular it
arises because the theory is coupled to external R-symmetry gauge fields. We will
discuss this issue again in the context of fivebranes wrapping SLAG 3-cycles, as
this is the example we will focus on when we construct the supergravity solutions
in the next section.
4.3. D3-brane and fivebrane world-volume theories. Let us now briefly

discuss the world-volume theories of the type IIB D3-brane and the M-theory five-
brane. The D3-brane action is given by a Dirac-Born-Infeld type action that in-
cludes a coupling to a four-form potential whose field strength is the self-dual
five-form (see Myers’ lectures for further discussion). If we consider, for simplicity,
a bosonic type IIB background with all fields vanishing except for the metric, the
world-volume action for a single D3-brane, with fermions set to zero, is given by

(4.21) S = T3 d4 σ [−det(∂i xµ ∂j xν gµν (x) + Fij )]1/2 .
W
The main new feature is that, in addition to the world-volume fields xµ , there is
now a U (1) gauge field with field strength F .
If we set F = 0, the action (4.21) reduces to the Nambu-Goto action. Following
a similar analysis to that of the membrane, we again find in static supersymmetric
backgrounds of the form R1,10−d × Md that supersymmetric cycles with F = 0 are
calibrated cycles. As the D3-brane has three spatial world-volume directions, there
are now more possibilities. A D3-brane can either wrap a calibrated 3-cycle in Md ,
with world-volume R × (Σ3 ⊂ Md ) or a Kähler 2-cycle in CYn with world-volume
R1,1 × (Σ2 ⊂ CYn ). The possibilities with the amount of supersymmetry preserved
are presented in table 3. For the Kähler cases, the world-volume has an R1,1
factor and we have also denoted by (n+ , n− ) the amount of d = 2 supersymmetry
preserved on the R1,1 factor, where n+ is the number of chiral supersymmetries
and n− the number of anti-chiral supersymmetries.

Kähler R × (Σ2 ⊂ CY2 ) 8, (4,4) d=2
1,1
R1,1 × (Σ2 ⊂ CY3 ) 4, (2,2) d=2

R1,1 × (Σ2 ⊂ CY4 ) 2, (1,1) d=2
SLAG R × (Σ3 ⊂ CY3 ) 4
Associative R × (Σ3 ⊂ G2 ) 2
Table 3. The different ways in which D3-branes can wrap cali-

In order to get some insight into the field theory living on D3-branes, consider
the target to be D=10 Minkowski space-time, the world-volume to be R1,3 and
fix the reparametrisation invariance by setting σ 0 = t, σ 1 = x1 , σ 2 = x2 , σ 3 = x3 .
After expanding the determinant and dropping a constant term, we get

1 1
(4.22) S = T3 d4 σ(− ∂i xI ∂ i xI − Fij F ij + f ermions + . . . ) .
2 4
In addition to F there are six scalar fields that describe the transverse displacement
of the D3-brane. This action is simply N = 4 super-Yang-Mills theory with gauge
group U (1). When there are N D3-branes, it is known that the DBI action should
be replaced by a non-Abelian generalisation but its precise form is not yet known.
However, it is known that after decoupling gravity, the leading terms give U (N )
N = 4 super-Yang-Mills theory. After dropping the U (1) centre of mass piece,
we find N = 4 SU (N ) Yang-Mills theory. Recall that this is the theory that is
conjectured to be dual to type IIB string theory on AdS5 × S 5 , which is the near
horizon limit of the type IIB supergravity solution describing a planar D3-brane.
Once again, the near horizon limit of the supergravity solution is dual to the field
theory arising on the brane, and the same should apply to the near horizon limits
of the supergravity solutions describing the wrapped D3-branes in table 3.
The world-volume theory of M-theory fivebranes is arguably the most intricate

of all branes [95, 96, 122, 10, 17]. The bosonic dynamical fields are maps xµ (σ)
along with a three-form field strength Hijk that satisfies a non-linear self-duality
condition. If we set H = 0 the dynamics is described by the Nambu-Goto action
and we again find in a static supersymmetric background of the form R1,10−d × Md
that supersymmetric cycles with H = 0 are calibrated cycles. As the fivebrane
has five spatial world-volume directions, there are many possibilities, which are
summarised in table 4. We have included the amount of supersymmetry preserved
including the number of supersymmetries counted with respect to the flat part of
the world-volume R1,q when q ≥ 1.

SLAG R × (Σ2 ⊂ CY2 )
1,3
8, N =2 d=4
R1,2 × (Σ3 ⊂ CY3 ) 4, N =2 d=3
R1,1 × (Σ4 ⊂ CY4 ) 2, (1,1) d=2
R × (Σ5 ⊂ CY5 ) 1
R1,1 × (Σ2 ⊂ CY2 ) × (Σ2 ⊂ CY2 ) 4, (2,2) d=2
R × (Σ2 ⊂ CY2 ) × (Σ3 ⊂ CY3 ) 2
Kähler R1,3 × (Σ2 ⊂ CY3 ) 4, N =1 d=4
R1,1 × (Σ4 ⊂ CY3 ) 4, (4,0) d=2
R1,1 × (Σ4 ⊂ CY4 ) 2, (2,0) d=2
C-Lag R1,1 × (Σ4 ⊂ HK2 ) 3, (2,1) d=2
Associative R1,2 × (Σ3 ⊂ G2 ) 2, N =1 d=3
Co-associative R1,1 × (Σ4 ⊂ G2 ) 2, (2,0) d=2
Cayley R × (Σ4 ⊂ Spin(7))
1,1
1, (1,0) d=2
Table 4. The different ways in which fivebranes can wrap cali-

The six-dimensional field theory living on a single planar fivebrane has five
scalar fields, describing the transverse fluctuations, the three-form H and fermions,
and has a chiral (2, 0) supersymmetry. The field theory when there are N coincident
fivebranes is not yet well understood. The AdS/CFT conjecture states that it is
dual to M-theory propagating on AdS7 × S 4 . Recall that the field theory has an
SO(5) R-symmetry.
In order to construct supergravity solutions describing wrapped branes, it is
very helpful to understand how supersymmetry is realised in the field theory living
on a probe-brane world-volume. The details depend on which calibrated cycle is
being wrapped and they are intimately connected to the structure of the normal
bundle of the calibrated cycle. Let us concentrate on the case of fivebranes wrapping
SLAG 3-cycles, as this will be the focus of the next section. The field theory on the
probe fivebrane world-volume lives on R1,2 × Σ3 . In order for this field theory to be
supersymmetric it is necessary that there be some notion of a constant spinor on Σ3 .
It is not immediately clear what this notion is, since, in general, Σ3 will not have a
covariantly constant spinor. However, the field theory living on the fivebrane with
world-volume R1,5 ⊂ R1,10 has an internal SO(5) R-symmetry, coming from the
five flat transverse directions, under which the fermions transform. The covariant
derivative of the spinors is schematically of the form

(4.23) (∂µ + ωµ − Aµ ) ,
where ω is the spin connection and A is the SO(5) gauge connection. Now consider
the fivebrane theory on R1,2 × Σ3 . If we decompose SO(5) → SO(3) × SO(2) and
choose the SO(3) gauge fields to be given by the SO(3) spin connection on Σ3 ,
A = ω, then clearly we can have constant spinors on Σ3 that could parameterise
the supersymmetry. This is exactly the way supersymmetry is realised for wrapped
branes [18]. It is sometimes said that the field theory is “twisted” because of the
similarities with the construction of topological field theories.
Geometrically, the identification of the SO(3) ⊂ SO(5) gauge fields with the
spin connection on Σ3 corresponds to the structure of the normal bundle of a SLAG
3-cycle. The five directions transverse to the fivebrane wrapping the SLAG 3-cycle
consist of three directions that are tangent to the CY3 and two flat directions that
are normal to the CY3 . This is responsible for breaking the SO(5) symmetry of the
flat fivebrane down to SO(3) × SO(2). We expect an SO(2) R-symmetry to survive
corresponding to the two flat directions, and thus the SO(2) gauge fields are zero in
the vacuum state. In section 4.1 we argued that for SLAG 3-cycles N (Σ3 ) ∼= T (Σ3 )
and this is responsible for the fact that there are non-zero SO(3) gauge fields in
the vacuum state and moreover, A = ω.
We can determine which external SO(5) gauge fields are excited for fivebranes
wrapping different supersymmetric cycles from our previous discussion of the nor-
mal bundles of calibrated cycles. We shall mention this again in the next section
in the context of constructing the corresponding supergravity solutions.
4.4. Generalised Calibrations. As somewhat of an aside, we comment that
there are more general supersymmetric cycles than those we have discussed above.
Firstly, we only considered background geometries when the background fluxes
(e.g. the four-form field strength G for D=11 supergravity) are all set to zero. By
analysing supersymmetric brane configurations when the fluxes are non-zero, one
is naturally lead to the notion of “generalised calibrations” [87, 88] (see also [11]).
The key new feature is that the exterior derivative of the generalised calibration is
no longer zero and is related to the flux. It is interesting to note that generalised
calibrations play an important role in characterising the most general classes of
supersymmetric supergravity solutions [67, 74, 76, 75]. They have also been
discussed in [86].
A second generalisation is to determine the conditions for supersymmetric
branes when non-trivial world-volume fluxes (F for D3-branes and H for the five-
branes) are switched on. This is related to the possibility of branes ending on
branes and is discussed in [14, 72, 65, 112].
5. Supergravity solutions for fivebranes wrapping calibrated cycles

At this point, we have established that D=11 supergravity has supersymmetric
membrane and fivebrane solutions. By considering the world-volume approxima-
tion to the dynamics of these branes we concluded that supersymmetric solutions
of D=11 supergravity describing branes wrapping calibrated cycles in special ho-
lonomy manifolds should also exist. In this section we will explain the explicit
construction of such solutions, in the near horizon limit, focusing on the richest
case of fivebranes.
At first sight it is not at all clear how to construct these solutions. One might
imagine that one should start with an explicit special holonomy metric, which are
rather rare, and then “switch on the brane”. In fact the procedure we adopt [110]
is more indirect and subtle. A key point is that we aim to find the solutions in
the near horizon limit, i.e. near to the brane wrapping the cycle, and this simplifies
things in two important ways. Firstly, we expect that only the local geometry of
the calibrated cycle in the special holonomy manifold, including the structure of its
normal bundle, enters into the construction. Secondly, we will be able to employ a
very useful technical procedure of first finding the solutions in D=7 SO(5) gauged
supergravity. This theory arises from the consistent truncation of the Kaluza-Klein
reduction on a four-sphere of D=11 supergravity, as we shall describe. In particular
any supersymmetric solution of the D=7 theory gives rise to a supersymmetric
solution of D=11 supergravity. Although the converse is certainly not true, the
D=7 gauged supergravity does include many interesting solutions corresponding to
the near horizon limit of wrapped fivebrane geometries.
We first discuss Kaluza-Klein reduction starting with the simplest case of re-
duction on a circle. We then describe the reduction on a four-sphere leading to D=7
gauged supergravity. Following this we will describe the construction of fivebranes
wrapping calibrated cycles. We focus on the case of fivebranes wrapping SLAG
3-cycles to illustrate some details and then summarise some aspects of the other
cases.
5.1. Consistency of Kaluza-Klein reduction. The basic example of Kal-

uza-Klein dimensional reduction is to start with pure gravity in five spacetime
dimensions and then reduce on a circle to get a theory of gravity in four spacetime
dimensions coupled to a U (1) gauge field and a scalar field. The procedure is to
first expand the five-dimensional metric in harmonics on the circle. One obtains
an infinite tower of modes whose four-dimensional mass is proportional to the in-
verse of the radius of the circle, as well as some massless modes consisting of the
four-dimensional metric, gauge field and scalar field just mentioned. Finally one
truncates the theory to the massless mode sector.
This truncation is said to be “consistent” in the sense that any solution of the
four-dimensional theory is automatically a solution to the five-dimensional theory.
The reason for this consistency is simply that the massless modes being kept are
independent of the coordinate on the circle, while the massive modes, which have
non-trivial dependence on the coordinate on the circle, are all set to zero. Note
that this is an exact statement that does not rely on the radius of the circle being
small, where one might argue that the massive modes are decoupling because they
are all getting very heavy. Similarly, as we shall shortly illustrate in more detail,
one can consistently truncate the reduction of theories with additional matter fields
on a circle, and more generally on tori.
The D=7 gauged supergravity that we shall be interested in arises from the
dimensional reduction of D=11 supergravity on a four-sphere. In general there
are no consistent truncations of the dimensional reductions of gravity theories on
spheres with dimension greater than one. The reason is that all of the harmonics
on the sphere, including those associated with the lowest mass modes, typically
depend on the coordinates of the sphere. Indeed, generically, if one reduces a
theory of gravity on a sphere and attempts to truncate to the lowest mass modes,
one will find that it is not consistent. That is, solutions of the truncated theory
will not correspond to exact solutions of the higher-dimensional theory. Of course if

the radius of the sphere were taken to be very small the truncated solutions could
provide very good approximations to solutions of the higher-dimensional theory.
However, in some special cases, including the reduction of D=11 supergravity on a
four-sphere, it has been shown that there is in fact a consistent truncation. We will
exploit this fact to construct exact solutions of D=11 supergravity by “uplifting”
solutions that we first find in the D=7 gauged supergravity.
Before we present the Kaluza-Klein reduction formulae for D=7 gauged super-
gravity, which are rather involved, let us first present them in the much simpler
setting of type IIA supergravity.
5.2. Reduction on S 1 to type IIA supergravity. Type IIA supergravity

in ten dimensions can be obtained from the Kaluza-Klein reduction of D=11 su-
pergravity on S 1 [77, 30, 100]. To see this, we construct an ansatz for the D=11
supergravity fields that just maintains the lowest massless modes. For the bosonic
fields we let
ds2 = e−2Φ/3 ds210 + e4Φ/3 (dy + C (1) )2
(5.1) C = C (3) + B ∧ dy .
Here the ten-dimensional line element ds210 , the scalar dilaton Φ, the “Ramond-
Ramond” one-form C (1) and three-form C (3) , and the the Neveu-Schwarz two-
form B are all independent of y. The field strengths of the forms will be denoted
F (2) = dC (1) , F (4) = dC (3) and H = dB. If we substitute this ansatz into the
D=11 equations of motion we find equations of motion for the ten-dimensional
fields which are derivable from the action
(5.2)
√ 1 1 2 1 2 1
S = d10 x −g e−2Φ [R + 4∂Φ2 − H 2 ] − F(4) − F(2) − B ∧ F4 ∧ F4 .
12 48 4 2
This is precisely the bosonic part of the action of type IIA supergravity. After
similarly including the fermions we find the full supersymmetric type IIA action,
which preserves 32 supersymmetries (two D=10 Majorana-Weyl spinors of opposite
chirality). Note that the isometries of S 1 give rise to the U (1) gauge field with field
strength F(2) .
The key point to emphasise is that, by construction, any solution of type II
supergravity automatically can be uplifted to give a solution of D=11 supergravity
that admits a U (1) isometry using the formulae in (5.1). Moreover, the D=11
supergravity solution will preserve at least the same amount of supersymmetry as
the type IIA solution4.
To illustrate with a simple example, consider the following supersymmetric
solution of type IIA supergravity:
ds2 = H −1 (dξ i dξ j ηij ) + (dxI dxI )
B = H −1 dξ 0 ∧ dξ 1
(5.3) e2Φ = H −1 ,
4Note that there are D=11 solutions with U (1) isometries that have supersymmetries that will
not survive the dimensional reduction because the Killing spinors have a non-trivial dependence
on the coordinate on the circle.
with i, j = 0, 1 and I, J = 1, . . . , 8. This is a solution to IIA supergravity providing

that H is harmonic in the transverse space. Choosing the simple single centre solu-
tion H = 1 + α2 N/r 6 we find that this solution carries N units of quantised electric
H-flux. In fact this solution describes the fields around N coincident fundamental
IIA strings of infinite extent. It preserves one-half of the supersymmetry. If we now
uplift this solution to get a solution of D=11 supergravity using (5.1) we obtain the
planar membrane solution (2.42).
5.3. Reduction on S 4 to D=7 SO(5) gauged supergravity. The dimen-
sional reduction of D=11 supergravity on a four-sphere can be consistently trun-
cated to give D=7 SO(5) gauged supergravity [116, 117]. The origin of the SO(5)
gauge symmetry is the SO(5) isometries of the four-sphere.
The explicit formulae for the D=11 bosonic fields is given by
∆4/5
ds2 = ∆−2/5 ds27 + DY A (T −1 )AB DY B
m2

1 (T · Y )A5
8G = A1 ...A5 − 3 DY A1 DY A2 DY A3 DY A4
3m Y ·T ·Y

4 (T · Y ) A4
+ 3 DY A1 DY A2 DY A3 D Y A5
3m Y ·T ·Y

2 A1 A2 (T · Y )A5 1
+ F DY A3 DY A4 + F A1 A2 F A3 A4 Y A5
m 2 Y ·T ·Y m
(5.4) +d(SB Y B ) ,
and the wedge product of forms is to be understood in the expression for G. Here
Y A , A, B = 1, . . . , 5, are constrained coordinates parametrising a four-sphere, sat-
isfying Y A Y A = 1. In this section xµ , µ, ν = 0, 1, . . . , 6 are D=7 coordinates and
ds27 is the D=7 line element associated with the D=7 metric g7 (x). The SO(5)
isometries of the round four-sphere lead to the introduction of SO(5) gauge fields
B A B (x), with field strength F A B (x), that appear in the covariant derivative DY A :
(5.5) DY A = dY A + 2mB A B Y B .
The matrix T is defined by
(5.6) T AB (x) = (Π−1 )i A (x)(Π−1 )j B (x)δ ij ,
where i, j = 1, . . . , 5 and ΠA i (x) are 14 scalar fields that parameterise the coset
SL(5, R)/SO(5). The warp factor ∆ is defined via
(5.7) ∆−6/5 = Y A T AB (x)Y B
and SA (x) are five three-forms.
If we substitute this into the D=11 supergravity equations of motion we get
equations of motion for g7 (x), B(x), Π(x) and S(x). The resulting equations of
motion can be derived from the D=7 action:

7 √ 1 1 i j AB 2
S = d x −g R + m2 T 2 − 2Tij T ij − Pµij P µij − ΠA ΠB Fµν
2 2

2
A
(5.8) − m2 Π−1 i Sµνρ,A − 6mSA ∧ FA
√ 1
+ 3ABCDE δ AG SG ∧ F BC ∧ F DE + (2Ω5 [B] − Ω3 [B]) ,
8m
where we have dropped the 7 subscript on g7 here and below, for clarity. This is
precisely the bosonic part of the action of D=7 SO(5) gauged supergravity [123].
In particular, the kinetic energy terms for the scalar fields are determined by Pµij
which is defined to be the part symmetric in i and j of
A

(5.9) (Π−1 )i δA B ∂µ + 2mBµ A B ΠB k δkj .
The potential terms for the scalar fields are defined in terms of
(5.10) Tij = (Π−1 )i A (Π−1 )j A , T = Tii .
Finally, Ω3 [B] and Ω5 [B] are Chern-Simons forms for the gauge fields, whose explicit
form will not be needed.
It is also possible to explicitly construct an ansatz for the D=11 fermions to
recover the fermions and supersymmetry of the D=7 gauged supergravity. In par-
ticular the construction implies that any bosonic (supersymmetric) solution of D=7
SO(5) gauged supergravity will uplift via (5.4) to a bosonic (supersymmetric) so-
lution of D=11 supergravity.
In order to find bosonic supersymmetric solutions of the D=7 gauged supergrav-
ity, we need the supersymmetry variations of the fermions. The fermions consist
of gravitini ψµ and dilatini λ and, in a bosonic background, their supersymmetry
variations are given by
1
δψµ = ∇µ − (γµ νρ
− 8δµν γ ρ )Γij ΠA i ΠB j Fνρ
AB
40
1 m 9 A
+ mT γµ + √ (γµ νρσ − δµ ν γ ρσ )Γi Π−1 i Sνρσ,A
20 10 3 2
1 µ j 1 1
δλi = γ Γ Pµij + γ µν (Γkl Γi − Γi Γkl ) ΠA k ΠB l Fµν
AB
2 16 5
1 1 m A
(5.11) + m(Tij − T δij )Γj + √ γ µνρ (Γi j − 4δij ) Π−1 j Sµνρ,A .
2 5 20 3
Here γ µ are the D=7 gamma matrices of Clif f (6, 1), while Γi are those of Clif f (5),
and these act on which is a spinor with respect to both Spin(6, 1) and Spin(5).
Note that γ µ and Γi commute. The covariant derivative appearing in the super-
symmetry variation for the gravitini is given by

1 ab 1
(5.12) ∇µ = ∂µ + ωµ γab + Qµij Γ ij
,
4 4
where Qµij is the part of (5.9) anti-symmetric in i and j. To obtain supersymmetric
configurations we set δψ = δλi = 0 and let be a commuting spinor. Note that
setting the variations of the dilatini to zero leads to algebraic constraints on .
Although still very complicated, D=7 gauged supergravity is a simpler theory
than D=11 supergravity as many degrees of freedom have been truncated. It is
clear from (5.4) that the “breathing mode”, corresponding to uniformly scaling
the round four-sphere, is one of the modes that has been truncated. This means
that there are no solutions of the D=7 gauged supergravity which give rise to the
full fivebrane solution of D=11 supergravity (2.33) with (2.36). However, the near
horizon limit of the fivebrane solution, AdS7 × S 4 , is easily found. Indeed it arises
as the simplest “vacuum” solution of the theory where the gauge fields, the three-
forms and the scalars are all set to zero: B = S = 0, Π = δ. In this case the
equations of motion reduce to solving

3
(5.13) Rµν = − m2 gµν ,
2
and the unique solution preserving all supersymmetry is AdS7 , which we can write
in Poincaré coordinates as

4 dξ i dξ j ηij + dρ2
(5.14) ds2 = 2 .
m ρ2
If we uplift this solution to D=11 using (5.4) we recover the AdS7 × S 4 solution
(2.40), with the radius of the AdS7 given by 2/m. Note that, in these coordinates,
the D=7 metric clearly displays the flat planar world-volume of the fivebrane.
5.4. Fivebranes wrapping SLAG 3-cycles. Supergravity solutions describ-
ing fivebranes wrapping different calibrated cycles have been constructed in D=7
gauged supergravity and then uplifted to D=11 in [110, 3, 71, 68]. We will il-
lustrate in some detail the construction of fivebranes wrapping SLAG 3-cycles [71]
and then comment more briefly on the other cases.
Consider the D=11 supersymmetric geometry R1,2 × CY3 × R2 and G = 0,
with a probe fivebrane wrapping a SLAG 3-cycle inside the Calabi-Yau three-fold.
i.e. the world-volume of the fivebrane is R1,2 × Σ3 with Σ3 ⊂ CY3 . The five
directions transverse to the fivebrane world-volume consist of three that are tangent
to the CY3 and two flat directions that are normal to the CY3 . If we wrap many
fivebranes, the back-reaction on the geometry will be significant and we aim to find
the corresponding supergravity solutions with G
= 0.
To construct these solutions, we seek a good ansatz for D=7 gauged super-
gravity. By analogy with the flat planar fivebrane solution, we think of the D=7
coordinates as being those of the world-volume of the fivebrane plus an additional
radial direction which, when the solution is uplifted to D=11, should correspond to
a kind of radial distance away from the fivebrane in the five transverse directions.
Thus an obvious ansatz for the D=7 metric is
(5.15) ds2 = e2f [ds2 (R1,2 ) + dr 2 ] + e2g ds̄2 (Σ3 ) ,
where f, g are functions of r only and ds̄2 (Σ3 ) is some metric on the 3-cycle. Now
before the back-reaction is taken into account, three of the directions transverse
to the fivebrane were tangent to CY3 and two were flat. We thus decompose the
SO(5)-symmetry of the D=7 theory into SO(3) × SO(2) and only switch on SO(3)
gauge fields. In order to preserve supersymmetry, as we shall elaborate on shortly,
the SO(3) gauge fields are chosen to be proportional to the spin connection of Σ3 :
(5.16) 2mB a b = ω̄ a b ,
for a, b = 1, 2, 3. This is a key part of the ansatz and it precisely corresponds to the
fact that the normal bundle of SLAG-cycles is isomorphic to the tangent bundle
of the cycle, as we discussed in section 4.1 and also at the end of section 4.3. An
ansatz for the scalar fields respecting SO(3) × SO(2) symmetry is given by
(5.17) ΠA i = diag(e2λ , e2λ , e2λ , e−3λ , e−3λ ) ,
where λ is a third function of r. It is consistent with the equations of motion to
set the three-forms S to zero, which we do to complete the ansatz.
For the configuration to preserve supersymmetry, we require that there exist
spinors such that the supersymmetry variations (5.11) vanish. Given the above
ansatz, the composite gauge fields Q appearing in (5.12) are given by the SO(3)
gauge fields Qab = 2mB ab . As a consequence, in demanding the vanishing of the
variation of the gravitini in the directions along Σ3 , we find that
1 m
(5.18) (∂ + ω̄ ab γab + B ab Γab ) = 0 ,
4 2
where ω̄ is the SO(3) spin connection on Σ3 . We now begin to see the significance
of the assumption (5.16). In particular, in the obvious orthonormal frame, (5.18)
can be satisfied by spinors independent of the coordinates along Σ3 , if we impose
the following projections on :
(5.19) γ ab = −Γab .
Note that this precisely parallels our discussion on the preservation of supersymme-
try in the context of the world-volume of the probe fivebrane at the end of section
4.3. To ensure that the variation of all components of the gravitini and dilatini
vanish we also need to impose
(5.20) γr =
and we find that the the only dependence of the Killing spinors on the coordinates is
radial: = ef /2 0 where 0 is a constant spinor. Note that only two of the conditions
(5.19) are independent and they break 1/4 of the supersymmetry. When combined
with (5.20) we see that 1/8 of the supersymmetry is preserved, in agreement with
table 4. Actually, if one follows through these preserved supersymmetries to D=11
one finds that they are precisely those that one expects when a fivebrane wraps a
SLAG 3-cycle: there are two projections corresponding to the CY3 and another for
the fivebrane.
A more detailed analysis of the conditions for supersymmetry implies that the
metric on Σ3 must in fact be Einstein. Given the factor e2g , we can always normalise
so that the Ricci tensor and the metric on Σ3 are related by
(5.21) R̄ic(Σ3 ) = lḡ(Σ3 ) ,
with l = 0, ±1. In three dimensions the Riemann curvature tensor is determined

by the Ricci tensor. When l = 0 (5.21) implies that Σ3 is flat: the resulting D=11
solution, after uplifting, corresponds to a fivebrane with a flat planar world-volume
and is thus not of primary interest. Indeed the solution turns out to be just a
special case of the fivebrane solution presented in (2.33) with a special harmonic
function with SO(3) × SO(2) symmetry.
The cases of most interest are thus when l = ±1. When l = 1 the Einstein
condition (5.21) implies that Σ3 is the three-sphere, S 3 , or a quotient by a discrete
subgroup of isometries of the isometry group SO(4), while for l = −1 it is hyper-
bolic three-space, H 3 , or a quotient by a discrete subgroup of SO(3, 1). Note in
particular that when l = −1 it is possible that the resulting geometry is compact.
We mentioned above that the Killing spinors are independent of the coordinates on
the cycle, and hence the quotients S 3 /Γ and H 3 /Γ are also supersymmetric.
Finally, in addition, supersymmetry implies the following first order BPS equa-
tions on the three radial functions:
m −4λ 3l 4λ−2g
e−f f = − 3e + 2e6λ + e
10 20m
m −4λ 7l 4λ−2g
e−f g = − 3e + 2e6λ − e
10 20m
m 6λ l 4λ−2g
(5.22) e−f λ = e − e−4λ + e .
5 10m
If these equations are satisfied then all of the D=7 equations of motion are satisfied
and we have found a supersymmetric solution.
Using (5.4) we can now uplift any solution of these BPS equations to obtain
supersymmetric solutions of D=11 supergravity. The metric is given by
1
(5.23) ds211 = ∆−2/5 ds27 + 2 ∆4/5 e4λ DY a DY a + e−6λ dY α dY α ,
m
where
DY a = dY a + ω̄ a b Y b
(5.24) ∆−6/5 = e−4λ Y a Y a + e6λ Y α Y α ,
with a, b = 1, 2, 3, α = 4, 5 and (Y a , Y α ) are constrained coordinates on S 4 satisfy-
ing Y a Y a + Y α Y α = 1. The expression for the four-form is given by substituting
into (5.4). Clearly, the four-sphere is no longer round and it is non-trivially fibred
over the three-dimensional Einstein space Σ3 .
It is illuminating5 to change coordinates from the constrained coordinates
(r, Y A ) to unconstrained coordinates (ρa , ρα ) via
1 f +g+2λ a
ρa = − e Y
m
1
(5.25) ρα = − e2f −3λ Y α .
m
The metric then takes the form:

ds2 = (∆−2/5 e2f ) ds2 (R1,2 ) + e2g−2f ds̄2 (Σ3 )

(5.26) +(∆4/5 e−4f ) e2f −2g (dρa + ω̄ a b ρb )2 + dρα dρα .
In these coordinates the warp factors have a similar form to the simple planar
fivebrane solution (2.33); in particular this form confirms the interpretation of the
solutions as describing fivebranes with world-volumes given by R1,2 ×Σ3 . In addition
the metric clearly displays the SO(2) symmetry corresponding to rotations in the
ρ4 , ρ5 plane.
Of course to find explicit solutions we need to solve the BPS equations. When
l = −1 it is easy to check that there is an exact solution given by
e10λ = 2
e8λ
e2g =
2m2
e8λ
(5.27) e2f = .
m2 r 2
5These types of coordinates were first noticed in the context of wrapped membranes in [70].
The D=7 metric is then the direct product AdS4 × (H 3 /Γ) (this D=7 solution
was first found in [124]). The uplifted D=11 solution is a warped product of
AdS4 × (H 3 /Γ) with a four-sphere which is non-trivially fibred over (H 3 /Γ). The
presence of the AdS4 factor in the D=11 solution indicates that M-theory on this
background is dual to a superconformal field theory in three spacetime dimensions.
We will return to this point after analysing the general BPS solution. Note that
when l = 1 there is no AdS4 × (S 3 /Γ) solution.
It seems plausible that the BPS equations could be solved exactly. However,
much of the physical content can be deduced from a simple numerical investigation.
If we introduce the new variables
a2 = e2g e−8λ
(5.28) eh = ef −4λ ,
then the BPS equations are given by
m 10λ l
e−h h = −2e −1 −
2 4ma2
a m 10λ 3l
e−h = − 2e −1 −
a 2 4ma2
m l
(5.29) e−h λ = e10λ − 1 + .
5 8ma2
We next define x = a2 and F = xe10λ to obtain the ODE
dF F [m2 x − 5α + 2β]
(5.30) = .
dx x[m2 (2F − x) + 2β]
FHxL
IR HBSL UV
1
HAdS 7 L
0.8
0.6
AdS âS
0.4
IR HGSL
0.2
BS
x
0.2 0.4 0.6 0.8 1 1.2
Figure 1. Behaviour of the orbits for five-branes wrapping SLAG

3-cycles with l = −1. Note the flow from the AdS7 -type region
when F, x are large to the IR fixed point and the flows to the good
and bad singularities in the IR, IR(GS) and IR(BS), respectively.
FHxL
UV HAdS 7 L
5
2
IR HGSL
1 IR HBSL
x
2 4 6 8
Figure 2. Behaviour of the orbits for fivebranes wrapping SLAG

3-cycles with l = 1.
The typical behaviour of F (x) is illustrated in figure 1 for l = −1 and figure 2

for l = 1. The region where both x and F are large is interesting. There we have
F ≈ x − l/m2 and using a as a radial variable we obtain the asymptotic behaviour
of the metric:
4
ds2 ≈ 2 2 da2 + a2 [ds2 (R1,2 ) + ds̄2 (Σ3 )] .
m a
This looks very similar to AdS7 in Poincaré coordinates except that the sections
with constant a are not R1,5 but R1,2 × Σ3 .
This clearly corresponds to the near horizon limit of the fivebrane wrapped
on the SLAG 3-cycle. By the general discussion on the AdS/CFT correspondence
earlier, this should be dual to the six-dimensional quantum field theory living on the
wrapped fivebrane worldvolume R1,2 × Σ3 , after decoupling gravity. More precisely,
the asymptotic behaviour of the solution (5.31), when lifted to D=11, is dual to the
UV behaviour of the quantum field theory. Following the flow of the solution as in
figures 1 and 2 correspond to flowing to the IR of the field theory. In the present
context, the IR corresponds to length scales large compared to the size of the cycle
Σ3 on which the fivebrane is wrapped. In other words, going to the IR corresponds
to taking Σ3 to be very small (assuming it is compact) and the six-dimensional
quantum field theory on R1,2 × Σ3 behaves more and more like a three-dimensional
quantum field theory on R1,2 .
Perhaps the most interesting solutions occur for l = −1. There is a solution
indicated by one of the dashed lines in figure 1 that flows from the UV AdS7
type region to the AdS4 × (H 3 /Γ) fixed point that was given in (5.27). This is a
supergravity solution that describes a kind of renormalisation group flow from a
theory on R1,2 × (H 3 /Γ) to a superconformal field theory on R1,2 . In particular, we
see that the natural interpretation of the AdS4 × (H 3 /Γ) solution is that, when it is
lifted to D=11, it is dual to a superconformal field theory on R1,2 that arises as the
IR limit of the fivebrane field theory on living on R1,2 × (H 3 /Γ). The interpretation
for non-compact H 3 /Γ is less clear.
The absence of an AdS4 × (S 3 /Γ) solution for l = 1 possibly indicates that the
quantum field theories arising on fivebranes wrapping SLAG 3-cycles with positive
curvature are not superconformal in the infrared. Alternatively, it could be that
there are more elaborate solutions lying outside of our ansatz that have AdS4
factors.
All other flows in figures 1 and 2 starting from the AdS7 type region flow
to singular solutions. Being singular does not exclude the possibility that they
might be interesting physically. Indeed, a criteria for time-like singularities in
static geometries to be “good singularities”, i.e. dual to some quantum field theory
behaviour, was presented in [110]. In particular, a good singularity is defined to
be one in which the norm of the time-like Killing vector with respect to the D=11
supergravity metric does not increase as one goes to the singularity (one can also
consider the weaker criterion that the norm is just bounded from above). It is not
difficult to determine whether the singularities that arise in the different asymptotic
limits are good or bad by this criterion and this has been presented in figures 1 and
2. It is likely that the good singularities describe some kind of Higgs branches of
the quantum field theory corresponding to the possibility of moving the coincident
wrapped fivebranes apart.
In summary, using D=7 gauged supergravity, we have been able to construct
D=11 supergravity solutions that describe fivebranes wrapping SLAG 3-cycles. The
cycle is Einstein and is either S 3 /Γ or H 3 /Γ where Γ is a discrete group of isometries.
Probably the most important solutions that have been found are the AdS4 ×(H 3 /Γ)
solutions which are dual to new superconformal field theories, after being uplifted
to D=11. More general flow solutions were also constructed numerically.
5.5. Fivebranes wrapping other cycles. The construction of supersym-

metric solutions corresponding to fivebranes wrapping other supersymmetric cycles
runs along similar lines. The ansatz for the D=7 metric is given by
(5.31) ds2 = e2f [ds2 (R1,5−d ) + dr 2 ] + e2g ds̄2 (Σd ) ,
where ds̄2d is the metric on the supersymmetric d-cycle, Σd , and the functions f
and g depend on the radial coordinate r only.
The SO(5)-gauge fields are specified by the spin connection of the metric on Σd
in a way determined by the structure of the normal bundle of the calibrated cycle
being wrapped. In general, we decompose the SO(5) symmetry into SO(p)×SO(q)
with p + q = 5, and only excite the gauge fields in the SO(p) subgroup. We will
denote these directions by a, b = 1, . . . , p. If we consider a probe fivebrane wrapping
the cycle inside a manifold of special holonomy M , this decomposition corresponds
to dividing the directions transverse to the brane into p directions within M and
q directions perpendicular to M . The precise ansatz for the SO(p) gauge fields is
determined by some part of the spin connection on the cycle and will be discussed
shortly.
In keeping with this decomposition, the solutions that we consider have a single
scalar field excited. More precisely we take
(5.32) ΠA i = diag(eqλ , . . . , eqλ , e−pλ , . . . , e−pλ ) ,

where we have p followed by q entries. Once again this implies that the composite
gauge-field Q is then determined by the gauge fields via Qab = 2mB ab .
It turns out that for the SLAG 5-cycle and most of the 4-cycle cases it is
necessary to have non-vanishing three-forms SA . The S-equation of motion is
C B 1
(5.33) m2 δAC Π−1 i Π−1 i SB = −m ∗ FA + √ ABCDE ∗ (F BC ∧ F DE ) .
4 3
The solutions have vanishing four-form field strength FA and hence SA are deter-
mined by the gauge fields.
Demanding that the configuration preserves supersymmetry, as for the SLAG
3-cycle case, we find that along the cycle directions:
1 m
(5.34) (∂ + ω̄ bc γbc + B ab Γab ) = 0 ,
4 2
bc
where ω̄ is the spin connection one-form of the cycle. For each case the specific
gauge fields, determined by the type of cycle being wrapped, go hand in hand with
a set of projections which allow (5.34) to be satisfied for spinors independent of the
coordinates along Σd . We can easily guess the appropriate ansatz for the gauge
fields from our discussion of the structure of the normal bundles of calibrated cy-
cles in section 4.1, and they are summarised below. The corresponding projections
are then easily determined and are given explicitly in [110, 3, 71, 68]. In the
uplifted D=11 solutions, these projections translate into a set of projections corre-
sponding to those of the special holonomy manifold and an additional projection
corresponding to the wrapped fivebrane.
By analysing the conditions for supersymmetry in more detail, one finds that
the metric on the cycle is necessarily Einstein, and we again normalise so that
(5.35) R̄ab = lḡab ,
with l = 0, ±1. When d = 2, 3 the Einstein condition implies that the cycles have
constant curvature and hence are either spheres for l = 1 or hyperbolic spaces for
l = −1, or quotients of these spaces by a discrete group of isometries. When d > 3
the Einstein condition implies that the Riemann tensor can be written
2l
(5.36) R̄abcd = W̄abcd + ḡa[c ḡd]b ,
d−1
where W̄ is the Weyl tensor, and there are more possibilities. By analysing the
D=7 Einstein equations one finds that for d = 4, 5 the part of the spin connection
that is identified with the gauge fields must have constant curvature.
We now summarise the ansatz for the SO(5) gauge fields for each case and
discuss the types of cycle that arise for d = 4, 5. We can always take a quotient of
the cycles listed by a discrete group of isometries.
SLAG n-cycles: Consider a probe fivebrane wrapping a SLAG n-cycle in a
CYn . The five directions transverse to the fivebrane consist of n directions tangent
to the CYn and 5−n normal to it. Thus, for the supergravity solution we decompose
SO(5) → SO(n) × SO(5 − n), and let the only non-vanishing gauge fields lie in the
SO(n) factor. Up to a factor of 2m these SO(n) gauge fields are identified with
the SO(n) spin connection on Σn , as in (5.16). This identification corresponds to
the fact that N (Σn ) ∼
= T (Σn ) for SLAG n-cycles. Since all of the spin connection
is identified with the gauge fields, the metric on Σn must have constant curvature
(W̄ = 0 for d = 4, 5) and hence Σn is S n for l = 1 and H n for l = −1.
Kähler 2-cycles: Kähler 2-cycles in CY2 are SLAG 2-cycles and have just been
discussed. For probe fivebranes wrapping Kähler 2-cycles in CY3 the five directions
transverse to the fivebrane consist of four directions tangent to the CY3 and one flat
direction normal to the CY3 . The normal bundle of the Kähler 2-cycle has structure
group U (2) ∼ = U (1) × SU (2), and now recall (4.5). Thus, for the supergravity
solution in D=7, we decompose SO(5) → SO(4) → U (2) ∼ = U (1) × SU (2) and
identify the U (1) spin connection of the cycle with the gauge fields in the U (1)
factor. We also set the SU (2) gauge fields to zero, which corresponds to considering
fivebranes wrapping Kähler 2-cycles with non-generic normal bundle. An example
of such a cycle is the two-sphere in the resolved conifold. It would be interesting to
find more general solutions with non-vanishing SU (2) gauge fields, preserving the
same amount of supersymmetry.
Kähler 4-cycles: We assume that Σ4 in the D=7 supergravity solution has
a Kähler metric with a U (2) ∼ = U (1) × SU (2) spin connection. When probe five-
branes wrap Kähler 4-cycles in CY3 the five transverse directions consist of two
directions tangent to the CY3 and three flat directions normal to the CY3 . Thus
we decompose SO(5) → SO(2) × SO(3), set the SO(3) gauge fields to zero and
identify the SO(2) ∼ = U (1) gauge fields with the U (1) part of the U (1) × SU (2) spin
connection on the 4-cycle. When probe fivebranes wrap Kähler 4-cycles in CY4 the
five transverse directions consist of four directions tangent to the CY4 and one flat
direction normal to the CY4 . Thus we now decompose SO(5) → SO(4) → U (2) ∼ =
U (1)×SU (2) and we set the SU (2) gauge fields to zero, which again corresponds to
considering non-generic normal bundles. We identify the U (1) gauge fields with the
U (1) part of the spin connection as dictated by (4.5). In both cases, the identifica-
tion of the gauge fields with part of the spin connection doesn’t place any further
constraints on Σ4 other than it is Kähler-Einstein. An example when l = 1 is CP 2 .
C-Lag 4-cycles: We again assume that Σ4 in the D=7 solution has a Kähler
metric with a U (2) ∼= U (1)×SU (2) spin connection. We again decompose SO(5) →
SO(4) → U (2) ∼ = U (1) × SU (2) but now we do not set the SU (2) gauge fields to
zero. Indeed, since the cycle is both SLAG and Kähler, with respect to different
complex structures, we must identify all of the U (2) gauge fields with the U (2) spin
connection. Einstein’s equations then imply that Σ4 must have constant holomor-
phic sectional curvature. This means that for l = 1 it is CP 2 while for l = −1 it is
the open disc in C2 with the Bergman metric. Note that the solutions corresponding
to fivebranes wrapping C-Lag CP 2 are different from the solutions corresponding
to fivebranes wrapping Kähler CP 2 , since they have more gauge fields excited and
preserve different amounts of supersymmetry.
Associative 3-cycles: When probe fivebranes wrap associative 3-cycles in G2
manifolds, there are four transverse directions that are tangent to the G2 manifold
and one flat direction normal to the G2 manifold. We thus decompose SO(5) →
SO(4) ∼= SU (2)+ × SU (2)− , where the superscripts indicate the self-dual and anti-
self-dual parts. Recall that the normal bundle of associative 3-cycles is given by
S ⊗ V where S was the SU (2) spin bundle on Σ3 and V is a rank SU (2) bundle. In
the non-generic case when V is trivial, for example for the G2 manifold in [28], then
the identification of the gauge fields is clear: we should identify the SO(3) ∼= SU (2)
spin connection on Σ3 with SU (2)+ gauge fields and set the SU (2)− gauge fields
to zero.
Co-associative 4-cycles: When probe fivebranes wrap co-associative 4-cycles

in G2 manifolds, there are three transverse directions that are tangent to the G2
manifold and two flat directions normal to the G2 manifold. We thus decompose
SO(5) → SO(3) × SO(2) and set the SO(2) gauge fields to zero. Recall that the
normal bundle of co-associative 4-cycles is isomorphic to the bundle of anti-self-dual
two-forms on the 4-cycle. This indicates that we should identify the SO(3) ∼ = SU (2)
gauge fields with the anti-self-dual part, SU (2)− , of the SO(4) ∼= SU (2)+ ×SU (2)−
spin connection on Σ4 . For the co-associative 4-cycles, Einstein’s equations imply
that the anti-self-dual part of the spin connection has constant curvature, or in
other words, the Weyl tensor is self-dual W̄ − = 0. These manifolds are sometimes
called conformally half-flat. If l = 1 the only compact examples are CP 2 and S 4 .
Cayley 4-cycles: When probe fivebranes wrap Cayley 4-cycles in Spin(7)
manifolds the five normal directions consist of four directions tangent to the Spin(7)
manifold and one flat direction normal to the Spin(7) manifold. We thus again
decompose SO(5) → SO(4) ∼ = SU (2)+ × SU (2)− , where the superscripts indicate
the self-dual and anti-self-dual parts. Recall that the normal bundle of Cayley
4-cycles is given by S− ⊗ V where S− is the SU (2) bundle of negative chirality
spinors on Σ3 and V is a rank SU (2) bundle. In the non-generic case when V
is trivial, for example for the Spin(7) manifolds in [28], then the identification of
the gauge fields is clear: if SU (2)± are the self-dual and anti-self-dual parts of the
SO(4) spin connection on Σ4 then we should identify the SU (2)− part of the spin
connection with SU (2)− gauge fields and set the SU (2)+ gauge fields to zero. As
for the co-associative 4-cycles, Einstein’s equations imply that the Weyl tensor is
self-dual, W̄ − = 0.
With this data the BPS equations can easily be derived and we refer to [110, 3,
71, 68] for the explicit equations. They have a similar appearance to those of the
SLAG 3-cycle case, with the addition of an extra term coming from the three-forms
S for most of the 4-cycle cases and the SLAG 5-cycles. When the curvature of the
cycle is negative, l = −1, in all cases except for Kähler 4-cycles in CY3 , we obtain
an AdS7−d × Σd fixed point. When l = 1, only for SLAG 5-cycles do we find such
a fixed point. This is summarised in table 5.
We note that using exactly the same ansatz for the D=7 supergravity fields,
some additional non-supersymmetric solutions of the form AdS7−d × Σd were found
[62]. In addition, by considering the possibility of extra scalar fields being ex-
cited, one more AdS3 solution was found. Such solutions could be dual to non-
supersymmetric conformal field theories, that are related to wrapped fivebranes
with supersymmetry broken. To develop this interpretation it is necessary that
the solutions be stable, which is difficult to determine. A preliminary perturba-
tive investigation revealed that some of these solutions are unstable. We have also
summarised these solutions in table 5.
5.6. Wrapped Membranes and D3-branes. D=11 supergravity solutions

describing membranes wrapping Kähler 2-cycles can be found in an analogous man-
ner [70]. The appropriate gauged supergravity for this case is maximal SO(8)
gauged supergravity in D=4 [47, 48] which can be obtained from a consistent
truncation of the dimensional reduction of D=11 supergravity on a seven-sphere
[50, 49]. The vacuum solution of this theory is AdS4 and this uplifts to AdS4 × S 7 ,
which is the near horizon limit of the planar membrane solution. More general
spacetime embedding cycle Σn supersymmetry

AdS5 × Σ2 Kähler 2-cycle in CY2 H2 yes
S2 no∗
Kähler 2-cycle in CY3 H2 yes
H2 no
AdS4 × Σ3 SLAG 3-cycle in CY3 H3 yes
H3 no
Associative 3-cycle H3 yes
H3 no
AdS3 × Σ4 Co-associative 4-cycle C− 4
yes
SLAG 4-cycle in CY4 H4 yes
H4 no
S4 no
4
Kähler 4-cycle in CY4 K− yes
4
K+ no
4
Cayley 4-cycle C− yes
4
C− no
CP , S 4
2
no
CLAG 4-cycle in HK8 B yes
B no
CP 2 no
SLAG 4-cycle in CY2 × CY2 H2 × H2 yes
H2 × H2 no∗
S2 × S2 no∗
S2 × H2 no∗
AdS2 × Σ5 SLAG 5-cycle in CY5 H5 yes
S5 yes
SLAG 5-cycle in CY2 × CY3 H × H2
3
yes
S2 × H3 no
S2 × H3 no
Table 5. AdS fixed point solutions for wrapped fivebranes: C−

and K± are conformally half-flat and Kähler–Einstein metrics with
the subscript denoting positive or negative scalar curvature and
B is the Bergman metric. Note that we can also take quotients
of all cycles by discrete groups of isometries and this preserves
supersymmetry. ∗ denotes a solution shown to be unstable.
solutions can be found that uplift to solutions describing the near horizon limit of
wrapped membranes.
Actually, the general formulae for obtaining SO(8) gauged supergravity from
the dimensional reduction of D=11 supergravity on the seven-sphere are rather
implicit and not in a form that is useful for uplifting general solutions. Luckily,
there is a further consistent truncation of the SO(8) gauged supergravity theory to
a U (1)4 gauged supergravity where the formulae are known explicitly (in the special
case that the axion fields are zero) [39] and this is sufficient for the construction of
D=11 wrapped membrane solutions.
To see why, let us describe the ansatz for the gauge fields for the D=4 solutions.
If we consider a probe membrane wrapping a Kähler 2-cycle in a CYn then the eight
directions transverse to the membrane consist of 2n − 2 directions that are tangent
to the CYn and 10 − 2n flat directions that are normal to the CYn . In addition
the normal bundle of the Kähler 2-cycle in CYn has structure group U (n − 1) ∼ =
U (1) × SU (n − 1), and recall (4.5). Thus, in the D=4 supergravity, we should
first decompose SO(8) → SO(2n − 2) × SO(10 − 2n) and only have non-vanishing
gauge fields in U (n − 1) ⊂ SO(2n − 2). The gauge fields in the U (1) factor of
U (n − 1) ∼= U (1) × SU (n − 1) are then identified with the U (1) spin connection on
the 2-cycle, corresponding to (4.5). In the solutions that have been constructed, the
remaining SU (n − 1) gauge fields are set to zero, which corresponds to the normal
bundle of the Kähler 2-cycle being non-generic when n ≥ 3. Thus, the ansatz for
the gauge fields is such that they always lie within the maximal Cartan subalgebra
U (1)4 of SO(8) and hence the truncation formulae of [39] can be used.
Once again the metric on the 2-cycle is Einstein and hence is either S 2 /Γ for l =
1 or H 2 /Γ for l = −1. In particular, the cycle can be an arbitrary Riemann surface.
General BPS equations have been found and analysed numerically. Interestingly,
AdS2 × Σ2 fixed points are found only for l = −1 and only for the cases of CY4
and CY5 . When uplifted to D=11 these solutions become a warped product with a
non-round seven-sphere that is non-trivially fibred over the cycle. The AdS2 fixed
point solutions should be dual to superconformal quantum mechanics living on the
wrapped membranes.
The appropriate gauged supergravity theory for finding D=10 type IIB so-
lutions describing D3-branes wrapping various calibrated cycles is the maximally
supersymmetric SO(6) gauged supergravity in D=5 [85]. This can be obtained
from the consistent truncation of the dimensional reduction of the type IIB super-
gravity on a five-sphere. In particular, the vacuum solution is AdS5 and this uplifts
to AdS5 × S 5 which is the near horizon limit of the planar D3-brane. Actually the
general formulae for this reduction are not yet known and one has to exploit further
consistent truncations that are known [129, 39, 107, 43].
All cases in table 3 have been investigated, and BPS equations have been
found and analysed. Once again, in the solutions, the 2- and 3-cycles that the D3-
branes wrap have Einstein metrics and hence have constant curvature. D3-branes
wrapping Kähler 2-cycles in CY2 and CY3 were studied in [110] while the CY4
case was analysed in [115]. AdS3 × H 2 /Γ fixed points were found for the CY3
and CY4 cases. D3-branes wrapping associative 3-cycles were analysed in [118]
and an AdS2 × H 3 /Γ fixed point was found. Finally, D3-branes wrapping SLAG
3-cycles were studied in [115] and no AdS2 fixed point was found. Note that non-
supersymmetric AdS solutions were sought in [115] for both wrapped membranes
and D3-branes, generalising the fivebrane solutions in [62], but none were found.
5.7. Other wrapped brane solutions. Let us briefly mention some other
supergravity solutions describing wrapped branes that have been constructed.
D6-branes of type IIA string theory carry charge under the U (1) gauge field
arising from the Kaluza-Klein reduction of D=11 supergravity on an S 1 (the field
C (1) in (5.1)). The planar D6-brane uplifts to pure geometry: R1,6 × M4 , where
M4 is Taub-NUT space with SU (2) holonomy. Similarly, when D6-branes wrap
calibrated cycles they uplift to other special holonomy manifolds in D=11 and this
has been studied in, e.g. [1, 8, 80, 53, 91, 82]. Other solutions related to wrapped
D6-branes that are dual to non-commutative field theories have been studied in
[25, 26].
There are solutions of massive type IIA supergravity with AdS6 factors which
are dual to the five-dimensional conformal field theory arising on the D4-D8-brane
system [56, 23]. Supersymmetric and non-supersymmetric solutions describing the
D4-D8 system wrapped on various calibrated cycles were found in [119, 115].
For some other supergravity solutions with possible applications to AdS/CFT,
that are somewhat related to those described here, see [7, 41, 54, 55, 24, 32,
105, 31, 106].
6. Discussion
We have explained in some detail the construction of supergravity solutions
describing branes wrapping calibrated cycles. There are a number of issues that
are worth further investigation.
It seems plausible that the BPS equations can be solved exactly. To date this
has only been achieved in a few cases. They were solved for the case of membranes
wrapping Kähler 2-cycles in CY5 [70], but this case is special in that all scalar
fields in the gauged supergravity are set to zero. When there is a non-vanishing
scalar field, the BPS equations were solved exactly for some cases in [110] and
they have been partially integrated for other cases. Of particular interest are the
exact solutions corresponding to the flows from an AdSD region to an AdSD−d ×Σd
fixed point (for example, one of the dashed lines in figure 1) as they are completely
regular solutions.
For the case of membranes wrapping Kähler 2-cycles in CY5 , the general flow
solution can be viewed as the “topological” AdS4 black holes discussed in [29].
When l = −1, there is a supersymmetric rotating generalisation of this black hole
[29]: when it is uplifted to D=11, it corresponds to waves on the wrapped mem-
brane [70]. The rotating solution is completely regular provided that the angular
momentum is bounded. It would be interesting to understand this bound from the
point of view of the dual field theory. In addition, the existence of this rotating
solution suggests, that for all of the regular flow solutions of wrapped branes start-
ing from an AdSD region and flowing to an AdSD−d × Σd region, there should be
rotating generalisations that are waiting to be found.
In all of the supergravity solutions describing wrapped branes that have been
constructed, the cycle has an Einstein metric on it. It would be interesting if a more
general ansatz could be found in which this condition is relaxed. While this seems
possible, it may not be possible to find explicit solutions. In some cases, such as
fivebranes wrapping Kähler 2-cycles in CY3 , we noted that the solutions constructed
correspond to fivebranes wrapping cycles with non-generic normal bundles. This
was because certain gauge fields were set to zero. We expect that more general
solutions can be found corresponding to generic normal bundles. Note that such a
solution, with an AdS factor, was found for D3-branes wrapping Kähler 2-cycles in
CY3 [110].
It would also be interesting to construct more general solutions that describe
the wrapped branes beyond the near horizon limit. Such solutions would asymptote
to a special holonomy manifold, which would necessarily be non-compact in order
that the solution can carry non-zero flux (a no-go theorem for D=11 supergravity
solutions with flux is presented in [76]). It seems likely that solutions can be
found that asymptote to the known cohomogeneity-one special holonomy manifolds.
For example, the deformed conifold is a cohomogeneity-one CY3 that is a regular
deformation of the conifold. It has a SLAG three-sphere and topologically the
manifold is the cotangent bundle of the three-sphere, T ∗ (S 3 ). It should be possible
to generalise the solutions describing fivebranes wrapping SLAG three-spheres in
the near horizon limit to solutions that include an asymptotic region far from the
branes that approaches the conifold metric. Of course, these solutions will still be
singular in the near horizon limit. It will be particularly interesting to construct
similar solutions for the l = −1 case. T ∗ (H 3 ) admits a CY3 metric, with a SLAG
H 3 , but there is a singularity at some finite distance from the SLAG 3-cycle. There
may be a solution with non-zero flux that interpolates from this singular behaviour
down to the regular near horizon solutions that we constructed. Alternatively, it
may be that the flux somehow “pushes off” the singularity to infinity and the entire
solution is regular. The construction of these more general solutions, when the four-
sphere transverse to the fivebrane is allowed to get large, will necessarily require
new techniques, as they cannot be found in the gauged supergravity.
The construction of the wrapped brane solutions using gauged supergravity
is rather indirect and it is desirable to characterise the D=11 geometries more
directly. For example, this may lead to new methods to generalise the solutions
along the lines mentioned above. One approach, is to guess general ansätze for
D=11 supergravity configurations that might describe wrapped brane solutions and
then impose the conditions to have supersymmetry. This approach has its origins in
the construction of the intersecting brane solutions, reviewed in [64, 133], and was
further extended in e.g. [54, 35, 98, 99]. Recently, it has been appreciated that it
is possible to systematically characterise supersymmetric solutions of supergravity
theories with non-zero fluxes using the notion of G-structures [67, 60, 101, 74, 76,
75] (see also [66, 63]). In particular, it was emphasised in some of these works that
generalised calibrations play a central role and this is intimately connected with the
fact that supergravity solutions with non-vanishing fluxes arise when branes wrap
calibrated cycles. It should be noted that while these techniques provide powerful
ways of characterising the D=11 geometries it is often difficult to obtain explicit
examples: indeed, even recovering the known explicit solutions found via gauged
supergravity can be non-trivial (see e.g. [74]).
The supergravity solutions can be used to learn a lot about the dual conformal
field theory, assuming the AdS/CFT correspondence is valid. For example, the
AdS fixed points can be used to determine the spectrum and correlation functions
of the operators in the dual field theory. For the case of wrapped D3-branes, since
the dual field theory is related to N = 4 super Yang-Mills theory, some detailed
comparisons can be made [110]. For the fivebrane case it will be more difficult
to do this since the conformal field theory living on the fivebrane is still poorly
understood. Perhaps some detailed comparisons can be made for the wrapped
membranes.
Recently some new supersymmetric solutions with AdS factors were constructed
in [38, 37]. It will be interesting to determine their dual CFT interpretation and
to see if they are related to wrapped branes. The non-supersymmetric solutions
containing AdS factors found in [62] might be dual to non-supersymmetric con-

formal field theories. A necessary requirement is that the solutions are stable: it
would be useful to complete the preliminary analysis of the perturbative stability
undertaken in [62].
Supergravity solutions that are dual to supersymmetric quantum field theories
that are not conformally invariant can be constructed using wrapped N S 5-branes
of type IIB string theory. The near horizon limit of the planar N S 5-brane is
dual to what is known as “little string theory” in six dimensions (for a review
see [4]). These still mysterious theories are not local quantum field theories but
at low energies they give rise to supersymmetric Yang-Mills (SYM) theory in six
dimensions. As a consequence, the geometries describing N S 5-branes wrapped on
various calibrated cycles encode information about various SYM theories in lower-
dimensions. The geometries describing N S 5-branes wrapped on Kähler 2-cycles
in CY3 were constructed in [34, 111] and are dual to N = 1 SYM theory in four
dimensions. If the 5-branes are wrapped on Kähler 2-cycles in CY2 the geometries
are dual to N = 2 SYM theory in four dimensions [69, 22] (see also [94]). By
wrapping on associative 3-cycles one finds geometries that encode information about
N = 1 SYM in D=3 [3, 34, 130, 109, 81], while wrapping on SLAG 3-cycles one
finds N = 2 SYM in D=3 [67, 83]. For the latter case, the solutions presented in
[67, 83] are singular and correspond to vanishing Chern-Simons form in the dual
SYM theory. There are strong physical arguments that suggest there are more
general regular solutions that are dual to SYM with non-vanishing Chern-Simons
form, and it would be very interesting to construct them. Note that supergravity
solutions describing N S 5-branes wrapping various 4-cycles were found in [115].
The D=10 geometry for wrapped N S 5-branes has been analysed in some detail in
[61, 67, 60, 101, 74, 75].
We hope to have given the impression that, while much is now known about
supergravity solutions describing branes wrapped on calibrated cycles, there is still
much to be understood.
References
1. B. S. Acharya, On realising N = 1 super Yang-Mills in M theory, hep-th/0011089.
2. B. S. Acharya, J. M. Figueroa-O’Farrill, C. M. Hull, and B. Spence, Branes at conical
singularities and holography, Adv. Theor. Math. Phys. 2 (1999), 1249–1286.
3. B. S. Acharya, J. P. Gauntlett, and N. Kim, Fivebranes wrapped on associative three-cycles,
Phys. Rev. D63 (2001), 106003.
4. O. Aharony, A brief review of ’little string theories’, Class. Quant. Grav. 17 (2000), 929–938.
5. , The non-AdS/non-CFT correspondence, or three different paths to QCD, hep-
th/0212193.
6. O. Aharony, S. S. Gubser, J. M. Maldacena, H. Ooguri, and Y. Oz, Large N field theories,
string theory and gravity, Phys. Rept. 323 (2000), 183–386.
7. M. Alishahiha and Y. Oz, AdS/CFT and BPS strings in four dimensions, Phys. Lett. B465
(1999), 136–141.
8. M. Atiyah, J. M. Maldacena, and C. Vafa, An M-theory flop as a large N duality, J. Math.
Phys. 42 (2001), 3209–3220.
9. M. Atiyah and E. Witten, M-theory dynamics on a manifold of G(2) holonomy, Adv. Theor.
Math. Phys. 6 (2003), 1–106.
10. I. Bandos, K. Lechner, A. Nurmagambetov, P. Pasti, D. P. Sorokin, and M. Tonin, Covariant
action for the super-five-brane of M-theory, Phys. Rev. Lett. 78 (1997), 4332–4334.
11. O. Barwald, N. D. Lambert, and P. C. West, A calibration bound for the M-theory fivebrane,
Phys. Lett. B463 (1999), 33–40.
12. K. Becker, M. Becker, and A. Strominger, Five-branes, membranes and nonperturbative

string theory, Nucl. Phys. B456 (1995), 130–152.
13. Katrin Becker, M. Becker, D. R. Morrison, H. Ooguri, Y. Oz, and Z. Yin, Supersymmetric
cycles in exceptional holonomy manifolds and Calabi-Yau 4-folds, Nucl. Phys. B480 (1996),
225–238.
14. E. Bergshoeff, R. Kallosh, T. Ortin, and G. Papadopoulos, kappa-symmetry, supersymmetry
and intersecting branes, Nucl. Phys. B502 (1997), 149–169.
15. E. Bergshoeff, E. Sezgin, and P. K. Townsend, Supermembranes and eleven-dimensional
supergravity, Phys. Lett. B189 (1987), 75–78.
16. , Properties of the eleven-dimensional super membrane theory, Ann. Phys. 185
(1988), 330.
17. E. Bergshoeff, D. P. Sorokin, and P.K. Townsend, The M5-brane hamiltonian, Nucl. Phys.
B533 (1998), 303–316.
18. M. Bershadsky, C. Vafa, and V. Sadov, D-branes and topological field theories, Nucl. Phys.
B463 (1996), 420–434.
19. M. Bertolini, Four lectures on the gauge-gravity correspondence, Internat. J. Modern Phys.
A. 18 (2003) 5647-5711.
20. M. Bertolini, P. Di Vecchia, M. Frau, A. Lerda, R. Marotta, and I. Pesando, Fractional
D-branes and their gauge duals, JHEP 02 (2001), 014.
21. F. Bigazzi, A. L. Cotrone, M. Petrini, and A. Zaffaroni, Supergravity duals of supersymmetric
four-dimensional gauge theories, hep-th/0303191.
22. F. Bigazzi, A. L. Cotrone, and A. Zaffaroni, N = 2 gauge theories from wrapped five-branes,
Phys. Lett. B519 (2001), 269–276.
23. A. Brandhuber and Y. Oz, The D4-D8 brane system and five dimensional fixed points, Phys.
Lett. B460 (1999), 307–312.
24. B. Brinne, A. Fayyazuddin, S. Mukhopadhyay, and D. J. Smith, Supergravity M5-branes
wrapped on Riemann surfaces and their QFT duals, JHEP 12 (2000), 013.
25. J. Brugués, J. Gomis, T. Mateos, and T. Ramirez, Commutative and noncommutative N =
2 SYM in 2+1 from wrapped D6-branes, Classical Quantum Gravity 20 (2003) S441-S448.
26. , Supergravity duals of noncommutative wrapped D6 branes and supersymmetry with-
out supersymmetry, JHEP 10 (2002), 016.
27. R. L. Bryant, Pseudo-Riemannian metrics with parallel spinor fields and vanishing Ricci ten-
sor, Global analysis and harmonic analysis (Marseille-Luminy, 1999), Sémin. Congr., vol. 4,
Soc. Math. France, Paris, 2000, pp. 53–94.
28. R. L. Bryant and S. M. Salamon, On the construction of some complete metrics with excep-
tional holonomy, Duke Math. J. 58 (1989), no. 3, 829–850.
29. M. M. Caldarelli and D. Klemm, Supersymmetry of anti-de sitter black holes, Nucl. Phys.
B545 (1999), 434–460.
30. I. C. G. Campbell and P. C. West, N=2 D = 10 nonchiral supergravity and its spontaneous
compactification, Nucl. Phys. B243 (1984), 112.
31. A. H. Chamseddine and W. A. Sabra, Magnetic and dyonic black holes in D = 4 gauged
supergravity, Phys. Lett. B485 (2000), 301–307.
32. , Magnetic strings in five dimensional gauged supergravity theories, Phys. Lett. B477
(2000), 329–334.
33. A..̋ Chamseddine and M. S. Volkov, Non-Abelian solitons in N = 4 gauged supergravity and
leading order string theory, Phys. Rev. D57 (1998), 6242–6254.
34. , Non-Abelian vacua in d = 5, N = 4 gauged supergravity, JHEP 04 (2001), 023.
35. H. Cho, M. Emam, D. Kastor and J. Traschen, Calibrations and Fayyazuddin-Smith space-
times, Phys. Rev. D63 (2001) 064003.
36. E. Cremmer, B. Julia, and J. Scherk, Supergravity theory in 11 dimensions, Phys. Lett. B76
(1978), 409–412.
37. S. Cucu, H. Lü, and J. F. Vázquez-Poritz, Interpolating from AdSD−2 × S2 to AdSD , Nucl.
Phys. B677 (2004) 181-222.
38. , A supersymmetric and smooth compactification of M-theory to AdS5 , Phys. Lett.
B568 (2003) 261-269.
39. M. Cvetic, M. J. Duff, P. Hoxha, J. T. Liu, H. Lu, J. X. Lu, R. Martinez-Acosta, C.N. Pope,
H. Sati, and T.A. Tran, Embedding AdS black holes in ten and eleven dimensions, Nucl.
Phys. B558 (1999), 96–126.
40. M. Cvetic, G. W. Gibbons, H. Lü, and C. N. Pope, Special holonomy spaces and M-theory,
Unity from Duality, Gravity, Gauge Theory and Strings (Les Houches 2001) 523-545. Nato
Adv. Study Inst.
41. M. Cvetic, H. Lu, C. N. Pope, and J. F. Vazquez-Poritz, Ad S in warped spacetimes, Phys.
Rev. D62 (2000), 122003.
42. M. Cvetic, H. Lu, and C. N. Pope, Brane resolution through transgression, Nucl. Phys. B600
(2001), 103–132.
43. M. Cvetic, H. Lu, C. N. Pope, A. Sadrzadeh, and T. A. Tran, Consistent SO(6) reduction
of type IIB supergravity on S5 , Nucl. Phys. B586 (2000), 275–286.
44. J. Dadok and R. Harvey, Calibrations and spinors, Acta Math. 170 (1993), no. 1, 83–120.
45. J. Dadok, R. Harvey, and F. Morgan, Calibrations on R8 , Trans. Amer. Math. Soc. 307
(1988), no. 1, 1–40.
46. S. P. de Alwis, A note on brane tension and M-theory, Phys. Lett. B388 (1996), 291–295.
47. B. de Wit and H. Nicolai, N=8 supergravity, Nucl. Phys. B208 (1982), 323.
48. , N=8 supergravity with local SO(8)×SU (8) invariance, Phys. Lett. B108 (1982),
285.
49. , The consistency of the S 7 truncation in d = 11 supergravity, Nucl. Phys. B281
(1987), 211.
50. B. de Wit, H. Nicolai, and N. P. Warner, The embedding of gauged N=8 supergravity into d
= 11 supergravity, Nucl. Phys. B255 (1985), 29.
51. M. J. Duff and J. T. Liu, Hidden spacetime symmetries and generalized holonomy in M-
theory, Nucl. Phys. B674 (2003) 217-230.
52. M. J. Duff, J. T. Liu, and R. Minasian, Eleven-dimensional origin of string / string duality:
A one-loop test, Nucl. Phys. B452 (1995), 261–282.
53. J. D. Edelstein and C. Nunez, D6 branes and M-theory geometrical transitions from gauged
supergravity, JHEP 04 (2001), 028.
54. A. Fayyazuddin and D. J. Smith, Localized intersections of M5-branes and four-dimensional
superconformal field theories, JHEP 04 (1999), 030.
55. , Warped AdS near-horizon geometry of completely localized intersections of M5-
branes, JHEP 10 (2000), 023.
56. S. Ferrara, A. Kehagias, H. Partouche, and A. Zaffaroni, Ad S6 interpretation of 5D super-
conformal field theories, Phys. Lett. B431 (1998), 57–62.
57. J. M. Figueroa-O’Farrill, Breaking the M-waves, Class. Quant. Grav. 17 (2000), 2925–2948.
58. J. M. Figueroa-O’Farrill and G. Papadopoulos, Maximally supersymmetric solutions of ten-
dimensional and eleven-dimensional supergravities, JHEP 03 (2003) 048.
59. D. Z. Freedman, S. S. Gubser, K. Pilch, and N. P. Warner, Renormalization group flows from
holography supersymmetry and a c-theorem, Adv. Theor. Math. Phys. 3 (1999), 363–417.
60. T. Friedrich and S. Ivanov, Killing spinor equations in dimension 7 and geometry of inte-
grable G2 -manifolds, J. Geom. Phys. 48 (2003) 1-11.
61. , Parallel spinors and connections with skew-symmetric torsion in string theory,
Asian J. Math. 6 (2001), no. 2, 303–335.
62. J. P. Gauntlett, N. Kim, S. Pakis, and D. Waldram, M-theory solutions with AdS factors,
Class. Quant. Grav. 19 (2002), 3927–3946.
63. J.P. Gauntlett and J.B. Gutowski, All supersymmetric solutions of minimal gauged super-
gravity in five dimensions, Phys. Rev. D68 (2003) 105009.
64. J. P. Gauntlett, Intersecting branes, hep-th/9705011.
65. , Membranes on fivebranes, Adv. Theor. Math. Phys. 3 (1999), 775–790.
66. J. P. Gauntlett, J. B. Gutowski, C. M. Hull, Stathis Pakis, and H. S. Reall, All supersym-
metric solutions of minimal supergravity in five dimensions, Classical Quantum Gravity 20
(2003) 4587-4634.
67. J. P. Gauntlett, Kim, D. N., Martelli, and D. Waldram, Fivebranes wrapped on SLAG three-
cycles and related geometry, JHEP 11 (2001), 018.
68. J. P. Gauntlett and N. Kim, M-fivebranes wrapped on supersymmetric cycles II, Phys. Rev.
D65 (2002), 086003.
69. J. P. Gauntlett, N. Kim, D. Martelli, and D. Waldram, Wrapped fivebranes and N = 2 super
Yang-Mills theory, Phys. Rev. D64 (2001), 106008.
70. J. P. Gauntlett, N. Kim, S. Pakis, and D. Waldram, Membranes wrapped on holomorphic
curves, Phys. Rev. D65 (2002), 026003.
71. J. P. Gauntlett, N. Kim, and D. Waldram, M-fivebranes wrapped on supersymmetric cycles,

Phys. Rev. D63 (2001), 126001.
72. J. P. Gauntlett, N. D. Lambert, and P. C. West, Supersymmetric fivebrane solitons, Adv.
Theor. Math. Phys. 3 (1999), 91–118.
73. J. P. Gauntlett, N. D. Lambert, and P. C. West, Branes and calibrated geometries, Commun.
Math. Phys. 202 (1999), 571–592.
74. J. P. Gauntlett, D. Martelli, S. Pakis, and D. Waldram, G-structures and wrapped NS5-
branes, hep-th/0205050.
75. J. P. Gauntlett, D. Martelli, and D. Waldram, Superstrings with intrinsic torsion, hep-
th/0302158.
76. J. P. Gauntlett and S. Pakis, The geometry of D= 11 Killing spinors, JHEP 0304 (2003)
039.
77. F. Giani and M. Pernici, N=2 supergravity in ten-dimensions, Phys. Rev. D30 (1984), 325–
333.
78. G. W. Gibbons, G. T. Horowitz, and P.K. Townsend, Higher dimensional resolution of
dilatonic black hole singularities, Class. Quant. Grav. 12 (1995), 297–318.
79. G. W. Gibbons and G. Papadopoulos, Calibrations and intersecting branes, Commun. Math.
Phys. 202 (1999), 593–619.
80. J. Gomis, D-branes, holonomy and M-theory, Nucl. Phys. B606 (2001), 3–17.
81. , On SUSY breaking and χSB from string duals, Nucl. Phys. B624 (2002), 181–199.
82. J. Gomis and T. Mateos, D6 branes wrapping Kaehler four-cycles, Phys. Lett. B524 (2002),
170–176.
83. J. Gomis and J. G. Russo, D = 2+1 N = 2 Yang-Mills theory from wrapped branes, JHEP
10 (2001), 028.
84. S. S. Gubser, I. R. Klebanov and A. M. Polyakov, Gauge theory correlators from non-critical
string theory, Phys. Lett. B 428 (1998), 105-114.
85. M. Gunaydin, L. J. Romans, and N. P. Warner, Compact and noncompact gauged supergrav-
ity theories in five-dimensions, Nucl. Phys. B272 (1986), 598.
86. J. Gutowski, S. Ivanov, and G. Papadopoulos, Deformations of generalized calibrations and
compact non-Kähler manifolds with vanishing first Chern class, Asian J. Math. 7 (2003)
39-79.
87. J. Gutowski and G. Papadopoulos, AdS calibrations, Phys. Lett. B462 (1999), 81–88.
88. J. Gutowski, G. Papadopoulos, and P. K. Townsend, Supersymmetry and generalized cali-
brations, Phys. Rev. D60 (1999), 106006.
89. R. Harvey, Spinors and calibrations, Perspectives in Mathematics, vol. 9, Academic Press
Inc., Boston, MA, 1990.
90. R. Harvey and H. B. Lawson, Jr., Calibrated geometries, Acta Math. 148 (1982), 47–157.
91. R. Hernandez, Branes wrapped on coassociative cycles, Phys. Lett. B521 (2001), 371–375.
92. P. Hořava and E. Witten, Eleven-dimensional supergravity on a manifold with boundary,
Nucl. Phys. B475 (1996), 94–114.
93. , Heterotic and type I string dynamics from eleven dimensions, Nucl. Phys. B460
(1996), 506–524.
94. K. Hori and A. Kapustin, Worldsheet descriptions of wrapped NS five-branes, JHEP 11
(2002), 038.
95. P. S. Howe and E. Sezgin, D = 11, p = 5, Phys. Lett. B394 (1997), 62–66.
96. P. S. Howe, E. Sezgin, and P. C. West, Covariant field equations of the M-theory five-brane,
Phys. Lett. B399 (1997), 49–59.
97. P. S. Howe and P. C. West, The complete N=2, d = 10 supergravity, Nucl. Phys. B238
(1984), 181.
98. T. Z. Husain, M2-branes wrapped on holomorphic curves, JHEP 12 (2003) 037.
99. , That’s a wrap!, JHEP 4 (2003) 053.
100. M. Huq and M. A. Namazie, Kaluza-Klein supergravity in ten-dimensions, Class. Quant.
Grav. 2 (1985), 293.
101. S. Ivanov, Connection with torsion, parallel spinors and geometry of Spin(7) manifolds,
math.dg/0111216.
102. A. Khavaev, K. Pilch, and N. P. Warner, New vacua of gauged N = 8 supergravity in five
dimensions, Phys. Lett. B487 (2000), 14–21.
103. I. R. Klebanov and E. Witten, Superconformal field theory on threebranes at a calabi-yau

singularity, Nucl. Phys. B536 (1998), 199–218.
104. I. R. Klebanov and M. J. Strassler, Supergravity and a confining gauge theory: Duality
cascades and χSB-resolution of naked singularities, JHEP 08 (2000), 052.
105. D. Klemm and W. A. Sabra, Supersymmetry of black strings in D = 5 gauged supergravities,
Phys. Rev. D62 (2000), 024003.
106. , Charged rotating black holes in 5d Einstein-Maxwell- (A)dS gravity, Phys. Lett.
B503 (2001), 147–153.
107. H. Lu, C. N. Pope, and T. A. Tran, Five-dimensional N = 4, SU(2) × U(1) gauged super-
gravity from type IIB, Phys. Lett. B475 (2000), 261–268.
108. J. M. Maldacena, The large N limit of superconformal field theories and supergravity, Adv.
Theor. Math. Phys. 2 (1998), 231–252.
109. J. M. Maldacena and H. Nastase, The supergravity dual of a theory with dynamical super-
symmetry breaking, JHEP 09 (2001), 024.
110. J. M. Maldacena and C. Nuñez, Supergravity description of field theories on curved manifolds
and a no go theorem, Int. J. Mod. Phys. A16 (2001), 822–855.
111. , Towards the large N limit of pure N = 1 super Yang Mills, Phys. Rev. Lett. 86
(2001), 588–591.
112. M. Marino, R. Minasian, G. W. Moore, and A. Strominger, Nonlinear instantons from
supersymmetric p-branes, JHEP 01 (2000), 005.
113. R. C. McLean, Deformations of calibrated submanifolds, Comm. Anal. Geom. 6 (1998), no. 4,
705–747.
114. David R. Morrison and M. Ronen Plesser, Non-spherical horizons. I, Adv. Theor. Math.
Phys. 3 (1999), 1–81.
115. M. Naka, Various wrapped branes from gauged supergravities, hep-th/0206141.
116. H. Nastase, D. Vaman, and P. van Nieuwenhuizen, Consistent nonlinear KK reduction of
11d supergravity on AdS7 × S4 and self-duality in odd dimensions, Phys. Lett. B469 (1999),
96–102.
117. , Consistency of the AdS7 × S4 reduction and the origin of self-duality in odd di-
mensions, Nucl. Phys. B581 (2000), 179–239.
118. H. Nieder and Y. Oz, Supergravity and D-branes wrapping special lagrangian cycles, JHEP
03 (2001), 008.
119. C. Nunez, I. Y. Park, M. Schvellinger, and T. A. Tran, Supergravity duals of gauge theories
from F(4) gauged supergravity in six dimensions, JHEP 04 (2001), 025.
120. G. Papadopoulos and A. Teschendorff, Multi-angle five-brane intersections, Phys. Lett. B443
(1998), 159–166.
121. , Grassmannians, calibrations and five-brane intersections, Class. Quant. Grav. 17
(2000), 2641–2662.
122. P. Pasti, D. P. Sorokin, and M. Tonin, Covariant action for a D = 11 five-brane with the
chiral field, Phys. Lett. B398 (1997), 41–46.
123. M. Pernici, K. Pilch, and P. van Nieuwenhuizen, Gauged maximally extended supergravity in
seven-dimensions, Phys. Lett. B143 (1984), 103.
124. M. Pernici and E. Sezgin, Spontaneous compactification of seven-dimensional supergravity
theories, Class. Quant. Grav. 2 (1985), 673.
125. K. Pilch and N. P. Warner, N = 2 supersymmetric RG flows and the IIB dilaton, Nucl.
Phys. B594 (2001), 209–228.
126. , N = 1 supersymmetric renormalization group flows from IIB supergravity, Adv.
Theor. Math. Phys. 4 (2002), 627–677.
127. J. Polchinski, N = 2 gauge-gravity duals, Int. J. Mod. Phys. A16 (2001), 707–718.
128. J. Polchinski and M. J. Strassler, The string dual of a confining four-dimensional gauge
theory, hep-th/0003136.
129. L. J. Romans, Gauged N=4 supergravities in five-dimensions and their magnetovac back-
grounds, Nucl. Phys. B267 (1986), 433.
130. M. Schvellinger and T. A. Tran, Supergravity duals of gauge field theories from SU(2)× U(1)
gauged supergravity in five dimensions, JHEP 06 (2001), 025.
131. J. H. Schwarz, Covariant field equations of chiral N=2 D = 10 supergravity, Nucl. Phys.
B226 (1983), 269.
132. S. Sethi, C. Vafa, and E. Witten, Constraints on low-dimensional string compactifications,

Nucl. Phys. B480 (1996), 213–224.
133. D. J. Smith, Intersecting brane solutions in string and M-theory, Class. Quant. Grav. 20
(2003), R233.
134. K. S. Stelle, BPS branes in supergravity, Quantum Field Theory: perspective and prospective
(Les Houches 1998) 257-350 Kluwer.
135. A. Strominger, Superstrings with torsion, Nucl. Phys. B274 (1986), 253.
136. L. Susskind and E. Witten, The holographic bound in anti-de Sitter space, hep-th/9805114.
137. C. Vafa and E. Witten, A one loop test of string duality, Nucl. Phys. B447 (1995), 261–270.
138. E. Witten, On flux quantization in M-theory and the effective action, J. Geom. Phys. 22
(1997), 1–13.
139. , Anti-de sitter space and holography, Adv. Theor. Math. Phys. 2 (1998), 253–291.
Department of Physics, Queen Mary, University of London, Mile End Rd, London,
E1 4NS, U.K.
E-mail address: J.P.Gauntlett@qmul.ac.uk
Volume 3, 2004
M-theory on Manifolds with Exceptional Holonomy
Sergei Gukov
Abstract. In these lectures I review recent progress on construction of man-

ifolds with exceptional holonomy and their application in string theory and
M-theory.
1. Introduction
Recently, M-theory compactifications on manifolds of exceptional holonomy
have attracted considerable attention. These models allow one to geometrically
engineer various minimally supersymmetric gauge theories, which typically have
a rich dynamical structure. A particularly interesting aspect of such models is
the behaviour near a classical singularity, where one might expect extra massless
degrees of freedom, enhancement of gauge symmetry, or a phase transition to a
different theory.
In these lectures I will try to explain two interesting problems in this subject
— namely, construction of manifolds with exceptional holonomy and the analysis
of the physics associated with singularities — and present general methods for their
solution. I will also try to make these lectures self-contained and pedagogical, so
that no special background is needed. In particular, below we start with an intro-
duction to special holonomy, then explain its relation to minimal supersymmetry
and proceed to the main questions.
2. Riemannian Manifolds of Special Holonomy

2.1. Holonomy Groups. Consider an oriented manifold X of real dimension
n and a vector v at some point on this manifold. One can explore the geometry
of X by doing a parallel transport of v along a closed contractible path in X; see
Figure1. Under such an operation the vector v may not come back to itself. In
fact, generically it will transform into a different vector that depends on the geom-
etry of X, on the path, and on the connection which was used to transport v . For
2000 Mathematics Subject Classification. 81T30, 53C29, 83E30 .

I would like to thank the organizers of the 2002 School on “Geometry and String Theory”
for a wonderful meeting. Various parts of these lectures are based on the work done together with
B. Acharya, A. Brandhuber, J. Gomis, S. Gubser, X. de la Ossa, D. Tong, J. Sparks, S.-T. Yau,
and E. Zaslow, whom I wish to thank for enjoyable collaboration. This research was conducted
during the period Sergei Gukov served as a Clay Mathematics Institute Research Fellow.
2004
127
128 SERGEI GUKOV
Figure 1. Parallel transport of a vector v along a closed path on

the manifold X.
a Riemannian manifold X, the natural connection is the Levi-Civita connection.

Furthermore, Riemannian geometry also tells us that the length of the vector co-
variantly transported along a closed path should be the same as the length of the
original vector. But the orientation may be different, and this is precisely what we
are going to discuss.
The relative orientation of the vector after parallel transport with respect to the
orientation of the original vector v is described by holonomy. On an n-dimensional
manifold holonomy is conveniently characterised by an element of the special or-
thogonal group, SO(n). It is not hard to see that the set of all holonomies them-
selves form a group, called the holonomy group, where the group structure is induced
by the composition of paths and an inverse corresponds to a path traversed in the
opposite direction. From the way we introduced the holonomy group, Hol(X),
it seems to depend on the choice of the point where we start and finish parallel
transport. However, for a generic choice of such point the holonomy group does
not depend on it, and therefore Hol(X) becomes a true geometric characteristic of
the manifold X. By definition, we have
(2.1) Hol(X) ⊆ SO(n)
where the equality holds for a generic Riemannian manifold X.
In some special instances, however, one finds that Hol(X) is a proper subgroup
of SO(n). In such cases, we say that X is a special holonomy manifold or a manifold
with restricted holonomy. These manifolds are in some sense distinguished, for
they exhibit nice geometric properties. As we explain later in this section, these
properties are typically associated with the existence of non-degenerate (in some
suitable sense) p-forms which are covariantly constant. Such p-forms also serve as
calibrations, and are related to the subject of minimal varieties.
The possible choices for Hol(X) ⊂ SO(n) are limited, and were completely
classified by M. Berger in 1955 [12]. Specifically, for X simply-connected and
neither locally a product nor symmetric, the only
possibilities
for Hol(X),
other
than the generic case of SO(n), are U n2 , SU n2 , Sp n4 × Sp(1), Sp n4 , G2 1,
Spin(7) or Spin(9), see Table 1. The first four of these correspond, respectively, to a
1The fourteen-dimensional simple Lie group G ⊂ Spin(7) is precisely the automorphism
2
group of the octonions, .
M-THEORY ON MANIFOLDS WITH EXCEPTIONAL HOLONOMY 129
Metric Holonomy Dimension

n
Kähler U 2 n = even
n
Calabi-Yau SU 2 n = even
n
HyperKähler Sp 4 n = multiple of 4
n
Quaternionic Sp 4 Sp(1) n = multiple of 4
Exceptional G2 7
Exceptional Spin(7) 8
Exceptional Spin(9) 16
Table 1. Berger’s list of holonomy groups.
Kähler, Calabi-Yau, quaternionic Kähler or hyper-Kähler manifold. The last three

possibilities are the so-called exceptional cases, which occur only in dimensions 7,
8 and 16, respectively. The case of 16-manifold with Spin(9) holonomy is in some
sense trivial since the Riemannian metric on any such manifold is always symmetric
[3].
2.2. Relation Between Holonomy and Supersymmetry. Roughly speak-

ing, one can think of the holonomy group as a geometric characteristic of the man-
ifold that tells us how much symmetry this manifold has. Namely, the smaller
the holonomy group, the larger the symmetry of the manifold X. Conversely, for
manifolds with larger holonomy groups the geometry is less restricted.
This philosophy becomes especially helpful in the physical context of super-
string or M-theory compactifications on X. There, the holonomy of X becomes
related to the degree of supersymmetry preserved in compactification: the mani-
folds with larger holonomy group typically preserve a smaller fraction of the super-
symmetry. This provides a nice link between the ‘geometric symmetry’ (holonomy)
and the ‘physical symmetry’ (supersymmetry). In Table 2 we illustrate this general
pattern with a few important examples, which will be used later.
The first example in Table 2 is a torus, T n , which we view as a quotient of an n-
dimensional real vector space, Rn , by a lattice. In this example, it is easy to deduce
directly from our definition that X = T n has trivial holonomy group, inherited
from the trivial holonomy of Rn . Indeed, no matter which path we choose on T n , a
parallel transport of a vector v along this path always brings it back to itself. Hence,
this example is the most symmetric one, in the sense of the previous paragraph,
Hol(X) = 1. Correspondingly, in string theory toroidal compactifications preserve
all of the original supersymmetries.
Our next example is Hol(X) = SU (3), which corresponds to Calabi-Yau mani-
folds of complex dimension 3 (real dimension 6). These manifolds exhibit a number
of remarkable properties, such as mirror symmetry, and are reasonably well studied
both in the mathematical and in the physical literature. We just mention here that
130 SERGEI GUKOV
Manifold X Tn CY3 XG2 XSpin(7)
dimR (X) n 6 7 8
Hol(X) 1 ⊂ SU (3) ⊂ G2 ⊂ Spin(7)
SUSY 1 > 1/4 > 1/8 > 1/16
Table 2. Relation between holonomy and supersymmetry for cer-

tain manifolds.
compactification on Calabi-Yau manifolds preserves 1/4 of the original supersym-

metry. In particular, compactification of heterotic string theory on X = CY3 yields
N = 1 effective field theory in 3 + 1 dimensions.
The last two examples in Table 2 are G2 and Spin(7) manifolds; that is, man-
ifolds with holonomy group G2 and Spin(7), respectively. They nicely fit into the
general pattern, so that as we read Table 2 from left to right the holonomy in-
creases, whereas the fraction of unbroken supersymmetry decreases. Specifically,
compactification of M-theory on a manifold with G2 holonomy leads to N = 1
four-dimensional theory. This is similar to the compactification of heterotic string
theory on Calabi-Yau three-folds. However, an advantage of M-theory on G2 man-
ifolds is that it is completely ‘geometric’. Compactification on Spin(7) manifolds
breaks supersymmetry even further, to an amount which is too small to be realised
in four-dimensional field theory.
Mathematically, the fact that all these manifolds preserve some supersymmetry
is related to the existence of covariantly constant spinors:
(2.2) ∇ξ = 0
For example, if Hol(X) = G2 the covariantly constant spinor is the singlet in the
decomposition of the spinor of SO(7) into representations of G2 :
8→7⊕1
Summarising, in Table 2 we listed some examples of special holonomy manifolds

that will be discussed below. All of these manifolds preserve a certain fraction of
supersymmetry, which depends on the holonomy group. Moreover, all of these
manifolds are Ricci-flat,
Rij = 0.
This useful property guarantees that all backgrounds of the form
R11−n × X
automatically solve eleven-dimensional Einstein equations with vanishing source

terms for matter fields.
Of particular interest are M-theory compactifications on manifolds with excep-
tional holonomy,
M − theory on M − theory on
G2 manifold Spin(7) manifold
(2.3)
⇓ ⇓
D = 3 + 1 QFT D = 2 + 1 QFT
since they lead to effective theories with minimal supersymmetry in four and three
dimensions, respectively. In such theories one can find many interesting phenomena,
e.g. confinement, various dualities, rich phase structure, non-perturbative effects,
etc. Moreover, minimal supersymmetry in three and four dimensions (partly) helps
with the following important problems:
• The Hierarchy Problem
• The Cosmological Constant Problem
• The Dark Matter Problem
All this makes minimal supersymmetry very attractive and, in particular, mo-
tivates the study of M-theory on manifolds with exceptional holonomy. In this
context, the spectrum of elementary particles in the effective low-energy theory
and their interactions are encoded in the geometry of the space X. Therefore,
understanding the latter may help us to learn more about dynamics of minimally
supersymmetric field theories, or even about M-theory itself!
2.3. Invariant Forms and Minimal Submanifolds. For a manifold X, we

have introduced the notion of special holonomy and related it to the existence of
covariantly constant spinors on X, cf. (2.2). However, special holonomy mani-
folds can be also characerised by the existence of certain invariant forms and the
corresponding minimal cycles.
Indeed, one can sandwich antisymmetric combinations of Γ-matrices with a
covariantly constant spinor ξ on X to obtain antisymmetric tensor forms of various
degree:
(2.4) ω (p) = ξ † Γi1 ...ip ξ
By construction, the p-form ω (p) is covariantly constant and invariant under Hol(X).
In order to find all possible invariant forms on a special holonomy manifold X, we
need to decompose the cohomology of X into representations2 of Hol(X) and iden-
tify all singlet components. For example, for a manifold with G2 holonomy such a
2This decomposition makes sense because the Laplacian on X commutes with Hol(X). In
a sense, for exceptional holonomy manifolds, decomposition of cohomology groups into represen-
tations of Hol(X) is a (poor) substitute for the Hodge decomposition in the realm of complex
geometry.
132 SERGEI GUKOV
decomposition reads [37]:

H 0 (X, R) = R
H 1 (X, R) = H71 (X, R)
H 2 (X, R) = H72 (X, R) ⊕ H14
2
(X, R)
H 3 (X, R) = H13 (X, R) ⊕ H73 (X, R) ⊕ H27
3
(X, R)
(2.5) H 4 (X, R) = H14 (X, R) ⊕ H74 (X, R) ⊕ H27
4
(X, R)
H 5 (X, R) = H75 (X, R) ⊕ H14
5
(X, R)
H 6 (X, R) = H76 (X, R)
H 7 (X, R) = R
Here, we used the G2 structure locally. The fact that the metric on X has irreducible
G2 -holonomy implies global constraints on X and this forces some of the above
groups to vanish when X is compact, viz.
H7k (X, R) = 0, k = 1, . . . , 6
Let us now return to the construction (2.4) of the invariant forms on X. From
the above decomposition we see that on a G2 manifold such forms can appear only
in degree p = 3 and p = 4. They are called associative and coassociative forms,
respectively. In fact, the coassociative 4-form is the Hodge dual of the associative
3-form. These forms, which we denote Φ and ∗Φ, enjoy a number of remarkable
properties.
For example, the existence of a G2 holonomy metric on X is equivalent to the
closure and co-closure of the associative form3,
(2.6) dΦ = 0
d∗Φ = 0.
This may look a little surprising, especially since the number of metric components
on a 7-manifold is different from the number of of components of a generic 3-form.
However, given a G2 holonomy metric,

7
ds2 = ei ⊗ ei ,
i=1
one can locally write the invariant 3-form Φ in terms of the vielbein ei ,
1
(2.7) Φ= ψijk ei ej ek
3!
Here, ψijk are totally antisymmetric structure constants of the imaginary octonions,
(2.8) σi σj = −δij + ψijk σk , i, j, k = 1, . . . , 7
and G2 is the automorphism group of the imaginary octonions. In a choice of basis
the non-zero structure constants are given by
(2.9) ψijk = +1 , (abc) = {(123), (147), (165), (246), (257), (354), (367)} .
3Another, equivalent, condition is to say that the G -structure (g, Φ) is torsion-free: ∇Φ = 0.

2
Hol(X) cycle S p = dim(S) Deformations dim(Def)
SU (3) SLAG 3 unobstructed b1 (S)
G2 associative 3 obstructed —
coassociative 4 unobstructed b+
2 (S)
Spin(7) Cayley 4 obstructed —
Table 3. Deformations of calibrated submanifolds.
It is, perhaps, less obvious that one can also locally reconstruct a G2 metric
from the associative 3-form:
(2.10) gij= det(B)−1/9 Bij
1
Bjk = − Φji i Φki3 i4 Φi5 i6 i7 i1 ...i7
144 1 2
This will be useful to us in the following sections.
Similarly, on a Spin(7) manifold X we find only one invariant form in degree
p = 4, called the Cayley form, Ω. Indeed, the decomposition of the cohomology
groups of X into Spin(7) representations looks as follows [37]:
H 0 (X, R) = R
H 1 (X, R) = H81 (X, R)
H 2 (X, R) = H72 (X, R) ⊕ H21
2
(X, R)
H 3 (X, R) = H83 (X, R) ⊕ H48
3
(X, R)
(2.11) H 4 (X, R) = H14+ (X, R) ⊕ H74+ (X, R) ⊕ H27
4
+ (X, R) ⊕ H35− (X, R)
4
H 5 (X, R) = H85 (X, R) ⊕ H48

5
(X, R)
H 6 (X, R) = H76 (X, R) ⊕ H21
6
(X, R)
H 7 (X, R) = H87 (X, R)
H 8 (X, R) = R
The additional label “±” denotes self-dual/anti-self-dual four-forms, respectively.
The cohomology class of the 4-form Ω generates H14+ (X, R),
H14+ (X, R) = [Ω]
Again, on a compact simply-connected Spin(7) manifold we have extra constraints,

H81 = H83 = H85 = H87 = 0, H72 = H74 = H76 = 0
Another remarkable property of the invariant forms is that they represent the
volume forms of the minimal submanifolds in X. The forms with these properties
are called calibrations, and the corresponding submanifolds are called calibrated
submanifolds [36]. More precisely, we say that Ψ is a calibration if it is less than or
equal to the volume on each oriented p-dimensional submanifold S ⊂ X. Namely,
combining the orientation of S with the restriction of the Riemann metric on X
to the subspace S, we can define a natural volume form vol(Tx S) on the tangent
134 SERGEI GUKOV
space Tx S for each point x ∈ S. Then, Ψ|Tx S = α · vol(Tx S) for some α ∈ R, and
we write:
Ψ|Tx S ≤ vol(Tx S)
if α ≤ 1. If equality holds for all points x ∈ S, then S is called a calibrated
submanifold with respect to the calibration Ψ. According to this definition, the
volume of a calibrated submanifold S can be expressed in terms of Ψ as:

(2.12) Vol(S) = Ψ|Tx S = Ψ
x∈S S
Since the right-hand side depends only on the cohomology class, we can write:

Vol(S) = Ψ= Ψ= Ψ|Tx S ≤ vol(Tx S ) = Vol(S )
S S x∈S x∈S

for any other submanifold S in the same homology class. Therefore, we have
just demonstrated that calibrated manifolds have minimal area in their homology
class. This important property of calibrated submanifolds allows us to identify
them with supersymmetric cycles. In particular, branes in string theory and M-
theory wrapped over calibrated submanifolds give rise to BPS states in the effective
theory.
Deformations of calibrated submanifolds have been studied by McLean [46],
and are briefly summarised in Table 3.
2.4. Why Exceptional Holonomy is Hard. Once we have introduced man-

ifolds with restricted (or special) holonomy, let us try to explain why until recently
so little was known about the exceptional cases, G2 and Spin(7). Indeed, on the
physics side, these manifolds are very natural candidates for constructing minimally
supersymmetric field theories from string/M-theory compactifications. Therefore,
one might expect exceptional holonomy manifolds to be at least as popular and
attractive as, say, Calabi-Yau manifolds. However, there are several reasons why
exceptional holonomy appeared to be a difficult subject; here we will stress two of
them:
• Existence
• Singularities
Let us now explain each of these problems in turn. The first problem refers to
the existence of an exceptional holonomy metric on a given manifold X. Namely,
it would be useful to have a general theorem which, under some favorable condi-
tions, would guarantee the existence of such a metric. Indeed, the original Berger’s
classification, described earlier in this section, only tells us which holonomy groups
can occur, but says nothing about examples of such manifolds or conditions un-
der which they exist. To illustrate this further, let us recall that when we deal
with Calabi-Yau manifolds we use such a theorem all the time — it is a theorem
due to Yau (originally, Calabi’s conjecture) which guarantees the existence of a
Ricci-flat metric on a compact, complex, Kähler manifold X with c1 (X) = 0 [58].
Unfortunately, no analogue of this theorem is known in the case of G2 and Spin(7)
holonomy (the local existence of such manifolds was first established in 1985 by
Bryant [14]). Therefore, until such a general theorem is found we are limited to
a case-by-case analysis of the specific examples. We will return to this problem in
the next section.
The second reason why exceptional holonomy manifolds are hard is associated
with singularities of these manifolds. As will be explained in the following sec-
tions, interesting physics occurs at the singularities. Moreover, the most interesting
physics is associated with the types of singularities of maximal codimension, which
exploit the geometry of the special holonomy manifold to the fullest. Indeed, singu-
larities with smaller codimension can typically arise in higher-dimensional compact-
ifications and, therefore, do not expose peculiar aspects of exceptional holonomy
manifolds related to the minimal amount of supersymmetry. Until recently, little
was known about these types of degenerations of manifolds with G2 and Spin(7)
holonomy. Moreover, even for known examples of isolated singularities, the dy-
namics of the M-theory in these backgrounds was unclear. Finally, it is important
to stress that the mathematical understanding of exceptional holonomy manifolds
would be incomplete too without proper account being taken of singular limits.
3. Construction of Manifolds With Exceptional Holonomy

In this section we review various methods of constructing compact and non-
compact manifolds with G2 and Spin(7) holonomy. In the absence of general ex-
istence theorems, akin to Yau’s theorem [58], these methods become especially
valuable. It is hard to give full justice to all the existing techniques in one section.
So we will try to explain only a few basic methods, focusing mainly on those which
have played an important role in recent developments in string theory. We also
illustrate these general techniques with several concrete examples that will appear
in the later sections.
3.1. Compact Manifolds. The first examples of metrics with G2 and Spin(7)
holonomy on compact manifolds were constructed by D. Joyce [37]. The basic idea
is to start with toroidal orbifolds of the form
(3.1) T 7 /Γ or T 8 /Γ
where Γ is a finite group, e.g. a product of Z2 cyclic groups. Notice that T 7 and
T 8 themselves can be regarded as G2 and Spin(7) manifolds, respectively. In fact,
they possess infinitely many G2 and Spin(7) structures. Therefore, if Γ preserves
one of these structures the quotient space automatically will be a manifold with
exceptional holonomy.
Example 3.1 (Joyce [37]). Consider a torus T 7 , parametrized by periodic
variables xi ∼ xi +1, i = 1, . . . , 7. As we pointed out, it admits many G2 structures.
Let us choose one of them:
Φ = e1 ∧e2 ∧e3 +e1 ∧e4 ∧e5 +e1 ∧e6 ∧e7 +e2 ∧e4 ∧e6 −e2 ∧e5 ∧e7 −e3 ∧e4 ∧e7 −e3 ∧e5 ∧e6
where ej = dxj . Furthermore, let us take
Γ = Z2 × Z2 × Z2
generated by three involutions
α : (x1 , . . . , x7 ) → (x1 , x2 , x3 , −x4 , −x5 , −x6 , −x7 )
β : (x1 , . . . , x7 ) → (x1 , −x2 , −x3 , x4 , x5 , 12 − x6 , −x7 )
γ : (x1 , . . . , x7 ) → (−x1 , x2 , −x3 , x4 , 12 − x5 , x6 , 12 − x7 )
It is easy to check that these generators indeed satisfy α2 = β 2 = γ 2 = 1 and
that the group Γ = α, β, γ
preserves the associative three-form Φ given above. It
136 SERGEI GUKOV
2
x C
n−4
R
Z2
Figure 2. A cartoon representing a Joyce orbifold T n /Γ with

C2 /Z2 orbifold points.
follows that the quotient space X = T 7 /Γ is a manifold with G2 holonomy. More

precisely, it is an orbifold since the group Γ has fixed points of the form T 3 ×C2 /Z2 .
The existence of orbifold fixed points is a general feature of the Joyce construction.
The quotient space (3.1) typically has bad (singular) points, as shown in Fig. 2.
In order to find a nice manifold X with G2 or Spin(7) holonomy one has to repair
these singularities. In practice, this means removing the local neighbourhood of
each singular point and replacing it with a smooth geometry, in a way which does
not affect the holonomy group. This may be difficult (or even impossible) for generic
orbifold singularities. However, if we have orbifold singularities that can also appear
as degenerations of Calabi-Yau manifolds, then things simplify dramatically.
Suppose we have a Z2 orbifold, as in the previous example:
Rn−4 × C2 /Z2
where Z2 acts only on the C2 factor (by reflecting all the coordiantes). This type
of orbifold singularity can be obtained as a singular limit of the A1 ALE space:
Rn−4 × ALEA1 → Rn−4 × C2 /Z2
Since both the ALE space and its singular limit have SU (2) holonomy group they
represent the local geometry of the K3 surface. This is an important point; we used
it implicitly to resolve the orbifold singularity with the usual tools from algebraic
geometry. Moreover, Joyce proved that resolving orbifold singularities in this way
does not change the holonomy group of the quotient space (3.1). Therefore, by the
end of the day, when all singularities are removed, we obtain a smooth, compact
manifold X with G2 or Spin(7) holonomy. In the above example, one finds a
smooth manifold X with G2 holonomy and Betti numbers:
b2 (X) = 12, b3 (X) = 43
There are many other examples of this construction, which are modelled not
only on K3 singularities, but also on orbifold singularities of Calabi-Yau three-folds
[37]. More examples can be found by replacing tori in (3.1) by products of the K3
surface or Calabi-Yau three-folds with lower-dimensional tori. In such models, finite
groups typically act as involutions on K3 or Calabi-Yau manifolds, to produce fixed
points of a familiar kind. Again, upon resolving the singularities one finds compact,
smooth manifolds with exceptional holonomy.
It may look a little disturbing that in Joyce’s construction one always finds
a compact manifold X with exceptional holonomy near a singular (orbifold) limit.
However, from the physics point of view, this is not a problem at all since interesting
phenomena usually occur when X develops a singularity. Indeed, compactification
on a smooth manifold X whose dimensions are very large (compared to the Planck
scale) leads to a very simple effective field theory; it is Abelian gauge theory with
some number of scalar fields coupled to gravity. To find more interesting physics,
such as non-Abelian gauge symmetry or chiral matter, one needs singularities.
Moreover, there is a close relation between various types of singularities and
the effective physics they produce. A simple, but very important aspect of this
relation is that a codimension d singularity of X can be associated with physics of
D ≥ 11 − d dimensional field theory. For example, there is no way one can obtain
four-dimensional chiral matter or parity symmetry breaking in D = 2+1 dimensions
from a C/Z2 singularity in X. Indeed, both of these phenomena are specific to their
dimension and can not be lifted to a higher-dimensional theory. Therefore, in order
to reproduce them from compactification on X one has to use the geometry of X
‘to the fullest’ and consider singularities of maximal codimension. This motivates
us to study isolated singular points in G2 and Spin(7) manifolds.
Unfortunately, even though Joyce manifolds naturally admit orbifold singu-
larities, none of them contains isolated G2 or Spin(7) singularities. Indeed, as we
explained earlier, it is crucial that orbifold singularities are modelled on Calabi-Yau
singularities, so that we can treat them using the familiar methods. Therefore, at
best such singularities can give us the same physics as one finds in the corresponding
Calabi-Yau manifolds.
Apart from a large class of Joyce manifolds, very few explicit constructions of
compact manifolds with exceptional holonomy are known. One nice approach was
recently suggested by A. Kovalev [39], where a smooth, compact 7-manifold X with
G2 holonomy is obtained by gluing ‘back-to-back’ two asymptotically cylindrical
Calabi-Yau manifolds W1 and W2 ,
X∼ = (W1 × S 1 ) ∪ (W2 × S 1 )
Although this construction is very elegant, so far it has been limited to very specific
types of G2 manifolds. In particular, it would be interesting to study deformations
of these spaces and to see if they can develop isolated singularities interesting in
physics. This leaves us with the following
Open Problem: Construct compact G2 and Spin(7) manifolds
with various types of isolated singularities
3.2. Non-compact Manifolds. As we explained, interesting physics occurs

at the singular points of the special holonomy manifold X. Depending on the
singularity, one may find, for example, extra gauge symmetry, global symmetry, or
massless states localized at the singularity. For each type of the singularity, the
corresponding physics may be different. However, usually it depends only on the
vicinity of the singularity, and not on the rest of the geometry of the space X.
Therefore, in order to study the physics associated with a given singularity, one
can imagine isolating the local neighbourhood of the singular point and studying it
separately. This gives the so-called ‘local model’ of a singular point. This procedure
is similar to considering one gauge factor in the standard model gauge group, rather
than studying the whole theory at once. In this sense, non-compact manifolds
138 SERGEI GUKOV
Figure 3. A cone over a compact space Y .
provide us with basic building blocks of low-energy energy physics that may appear
in compactifications on compact manifolds.
Here, we discuss a particular class of isolated singularities, namely conical sin-
gularities. They correspond to degenerations of the metric on the space X of the
form:
(3.2) ds2 (X) = dt2 + t2 ds2 (Y ),
where a compact space Y is the base of the cone; the dimension of Y is one less
than the dimension of X. It is clear that X has an isolated singular point at the
tip of the cone, except for the special case when Y is a sphere, Sn−1 , with a round
metric.
The conical singularities of the form (3.2) are among the simplest isolated
singularities one could study; see Figure3. In fact, the first examples of non-compact
manifolds with G2 and Spin(7) holonomy, obtained by Bryant and Salamon [15]
and, independently, by Gibbons, Page, and Pope [28], exhibit precisely this type
of degeneration. Specifically, the complete metrics constructed in [15, 28] are
smooth everywhere, and asymptotically look like (3.2), for various base manifolds
Y . Therefore, they can be considered as resolutions of conical singularities. In Table
4 we list known asymptotically conical (AC) complete metrics with G2 and Spin(7)
holonomy that were originally found in [15, 28] and in more recent literature [17,
31].
The method of constructing G2 and Spin(7) metrics originally used in [15, 28]
was essentially based on the direct analysis of the Ricci-flatness equations,
(3.3) Rij = 0,
for a particular metric ansatz. We will not go into details of this approach here
since it relies on finding the right form of the ansatz and, therefore, is not practical
for generalizations. Instead, following [32, 16], we will describe a very powerful
approach, recently developed by Hitchin [34], which allows us to construct all the
G2 and Spin(7) manifolds listed in Table 4 (and many more!) in a systematic
manner. Another advantage of this method is that it leads to first-order differential
equations, which are much easier than the second-order Einstein equations (3.3).
Before we explain the basic idea of Hitchin’s construction, notice that for all of
the AC manifolds in Table 4 the base manifold Y is a homogeneous quotient space
(3.4) Y = G/K,
Holonomy Topology of X Base Y
S4 × R3 CP3
G2 CP2 × R3 SU (3)/U (1)2
S3 × R4 SU (2) × SU (2)
T 1,1 × R2
S4 × R4 SO(5)/SO(3)
Spin(7) CP2 × R4 SU (3)/U (1)
S5 × R3
Table 4. Asymptotically conical manifolds with G2 and Spin(7) holonomy.
where G is some group and K ⊂ G is a subgroup. Therefore, we can think of X

as being foliated by principal orbits G/K over a positive real line, R+ , as shown
on Figure4. A real variable t ∈ R+ in this picture plays the role of the radial
coordinate; the best way to see this is from the singular limit, in which the metric
on X becomes exactly conical, cf. eq. (3.2).
As we move along R+ , the size and the shape of the principal orbit changes, but
only in a way consistent with the symmetries of the coset space G/K. In particular,
at some point the principal orbit G/K may collapse into a degenerate orbit,
(3.5) B = G/H
where symmetry requires
(3.6) G⊃H ⊃K
At this point (which we denote t = t0 ) the “radial evolution” stops, resulting in
a non-compact space X with a non-trivial topological cycle B, sometimes called a
bolt. In other words, the space X is contractible to a compact set B, and from the
relation (3.6) we can easily deduce that the normal space of B inside X is itself a
cone on H/K. Therefore, in general, the space X obtained in this way is a singular
space, with a conical singularity along the degenerate orbit B = G/H. However, if
H/K is a round sphere, then the space X is smooth,
H/K = Sk =⇒ X smooth
This simply follows from the fact that the normal space of B inside X in such a
case is non-singular, Rk+1 (= a cone over H/K). It is a good exercise to check that
for all manifolds listed in Table 4, one indeed has H/K = Sk , for some value of k.
To show this, one should first write down the groups G, H, and K, and then find
H/K.
The representation of the non-compact space X in terms of principal orbits
which are homogeneous coset spaces is very useful. In fact, as we just explained, the
140 SERGEI GUKOV
G
K
degenerate
orbitG H t
G
principal orbit K
Figure 4. A non-compact space X can be viewed as a foliation by

principal orbits Y = G/K. The non-trivial cycle in X correspond
to the degenerate orbit G/H, where G ⊃ H ⊃ K.
topology of X simply follows from the group data (3.6). For example, if H/K = Sk
so that X is smooth, we have
(3.7) X∼
= (G/H) × Rk+1
However, this structure can also be used to find a G-invariant metric on X. In
order to do this, all we need to know are the groups G and K.
First, let us sketch the basic idea of Hitchin’s construction [34], and then explain
the details in some specific examples. For more details and further applications we
refer the reader to [32, 16]. We start with a principal orbit Y = G/K which can
be, for instance, the base of the conical manifold that we want to construct. Let
P be the space of (stable4) G-invariant differential forms on Y . This space is finite
dimensional and, moreover, it turns out that there exists a symplectic structure on
P. This important result allows us to think of the space P as the phase space of
some dynamical system:
P = Phase Space

(3.8) ω= dxi ∧ dpi
where we parametrized P by some coordinate variables xi and the conjugate mo-

mentum variables pi .
Given a principal orbit G/K and a space of G-invariant forms on it, there
is a canonical construction of a Hamiltonian H(xi , pi ) for our dynamical system,
such that the Hamiltonian flow equations are equivalent to the special holonomy
4Stable forms are defined as follows [35]. Let X be a manifold of real dimension n, and
V = T X. Then, the form ρ ∈ Λp V ∗ is stable if it lies in an open orbit of the (natural) GL(V )
action on Λp V ∗ . In other words, this means that all forms in the neighborhood of ρ are GL(V )-
equivalent to ρ. This definition is useful because it allows one to define a volume. For example, a
symplectic form ω is stable if and only if ω n/2 = 0.
condition [34]:

dxi ∂H
= ∂p Special Holonomy Metric
(3.9) dt i
⇐⇒
dpi
dt = − ∂x
∂H on (t1 , t2 ) × (G/K)
i
where the ‘time’ in the Hamiltonian system is identified with the radial variable
t. Thus, solving the Hamiltonian flow equations from t = t1 to t = t2 with a
particular boundary condition leads to the special holonomy metric on (t1 , t2 ) ×
(G/K). Typically, one can extend the boundaries of the interval (t1 , t2 ) where the
solution is defined to infinity on one side, and to a point t = t0 , where the principal
orbit degenerates, on the other side. Then, this gives a complete metric with special
holonomy on a non-compact manifold X of the form (3.7). Let us now illustrate
these general ideas in more detail in a concrete example.
Example 3.2. Let us take G = SU (2)3 and K = SU (2). We can form the
following natural sequence of subgroups:
G H K
(3.10)
SU (2)3 ⊃ SU (2)2 ⊃ SU (2)
From the general formula (3.4) it follows that in this example we deal with a space
X, whose principal orbits are
Y = SU (2) × SU (2) ∼ = S3 × S3
Furthermore, G/H ∼ = H/K ∼ = S3 implies that X is a smooth manifold with topol-
ogy, cf. (3.7),
X∼ = S3 × R4
In fact, X is one of the asymptotically conical manifolds listed in Table 4.
In order to find a G2 metric on this manifold, we need to construct the “phase
space”, P, that is, the space of SU (2)3 –invariant 3-forms and 4-forms on Y = G/K:
P = Ω3G (G/K) × Ω4G (G/K)
In this example, it turns out that each of the factors is one-dimensional, generated
by a 3-form ρ and by a 4-form σ, respectively,

(3.11) ρ = σ1 σ2 σ3 − Σ1 Σ2 Σ3 + x d(σ1 Σ1 ) + d(σ2 Σ2 ) + d(σ3 Σ3 ) ,

(3.12) σ = p2/5 σ2 Σ2 σ3 Σ3 + σ3 Σ3 σ1 Σ1 + σ1 Σ1 σ2 Σ2 .
where we introduced two sets of left invariant one-forms (σa , Σa ) on Y
σ1 = cos ψdθ + sin ψ sin θdφ, Σ1 = cos ψ̃dθ̃ + sin ψ̃ sin θ̃dφ̃,
σ2 = − sin ψdθ + cos ψ sin θdφ, Σ2 = − sin ψ̃dθ̃ + cos ψ̃ sin θ̃dφ̃,
(3.13) σ3 = dψ + cos θdφ, Σ3 = dψ̃ + cos θ̃dφ̃
which satisfy the usual SU (2) algebra
1 1
(3.14) dσa = − abc σa ∧ σb , dΣa = − abc Σa ∧ Σb .
2 2
Therefore, we have only one “coordinate” x and its conjugate “momentum” p,
parametrizing the “phase space” P of our model. In order to see that there is a
natural symplectic structure on P, note that x and p multiply exact forms. For x
142 SERGEI GUKOV
this obviously follows from (3.11) and for p this can be easily checked using (3.12)
and (3.14). This observation can be used to define a non-degenerate symplectic
structure on P = Ω3exact (Y ) × Ω4exact (Y ). Explicitly, it can be written as
ω ((ρ1 , σ1 ), (ρ2 , σ2 )) = ρ1 , σ2
− ρ2 , σ1
,
where, in general, for ρ = dβ ∈ Ωkexact (Y ) and σ = dγ ∈ Ωn−k
exact (Y ) one has a
nondegenerate pairing

(3.15) ρ, σ
= dβ ∧ γ = (−1) k
β ∧ dγ.
Y Y
Once we have the phase space P, it remains to write down the Hamiltonian
flow equations (3.9). In Hitchin’s construction, the Hamiltonian H(x, p) is defined
as an invariant functional on the space of differential forms, P. Specifically, in the
context of G2 manifolds it is given by
(3.16) H = 2V (σ) − V (ρ),
where V (ρ) and V (σ) are suitably defined volume functionals5 [34]. Computing
(3.16) for the G-invariant forms (3.11) and (3.12) we obtain the Hamiltonian flow
equations:
ṗ = x(x − 1)2
ẋ = p2
These first-order equations can be easily solved, and the solution for x(t) and p(t)
determines the evolution of the forms ρ and σ, respectively. On the other hand,
these forms define the associative three-form on the 7-manifold Y × (t1 , t2 ),
(3.19) Φ = dt ∧ ω + ρ,
where ω is a 2-form on Y , such that σ = ω 2 /2. For x(t) and p(t) satisfying the
Hamiltonian flow equations the associative form Φ is automatically closed and co-
closed, cf. (2.6). Therefore, as we explained in section 2, it defines a G2 holonomy
metric. Specifically, one can use (2.10) to find the explicit form of the metric, which
after a simple change of variables becomes the G2 metric on the spin bundle over
S3 , originally found in [15, 28]:

3
r2 r02
3
dr 2 r2
(3.20) 2
ds = + (σa − Σa ) +
2
1− 2 (σa + Σa )2
1 − r02 /r 2 12 a=1 36 r a=1
The above example can be easily generalised in a number of directions. For

example, if instead of (3.10) we take G = SU (2)2 and K to be its trivial subgroup,
we end up with the same topology for X and Y , but with a larger space of G2
5The volume V (σ) for a 4-form σ ∈ Λ4 T ∗ Y ∼ Λ2 T Y ⊗ Λ6 T ∗ Y is very easy to define. Indeed,
=
we have σ 3 ∈ Λ6 T Y ⊗ (Λ6 T ∗ Y )3 ∼
= (Λ6 T ∗ Y )2 , and therefore we can take
1
(3.17) V (σ) = |σ 3 | 2 .
Y
to be the volume for σ. In order to define the volume V (ρ) for a 3-form ρ ∈ Λ3 T ∗ Y , one first
defines a map Kρ : T Y → T Y ⊗ Λ6 T ∗ Y, such that for a vector v ∈ T Y it gives K(v) = ı(v)ρ ∧ ρ ∈
Λ5 T ∗ Y ∼
= T Y ⊗ Λ6 T ∗ Y. Hence, one can define tr(K 2 ) ∈ (Λ6 T ∗ Y )2 . Since stable forms with
stabilizer SL(3, ) are characterised by tr(K)2 < 0, following [35], we define

(3.18) V (ρ) = | −trK 2 |.
Y
metrics on X. Indeed, for G = SU (2)2 the space of G-invariant forms on Y is much

larger. Therefore, the corresponding dynamical system is more complicated and
has a richer structure. Some specific solutions of this more general system have
been recently constructed in [11, 18, 13, 19], but the complete solution is still not
known.
There is a similar systematic method, also developed by Hitchin [34], of con-
structing complete non-compact manifolds with Spin(7) holonomy. Again, this
method can be used to obtain the asymptotically conical metrics listed in Table 4,
as well as other Spin(7) metrics recently found in [20, 29, 17, 42].
4. M-theory, Circle Fibrations, and Calibrated Submanifolds

Since compactification on smooth manifolds does not produce interesting physics
— in particular, does not lead to realistic quantum field theories — one has to study
the dynamics of string theory and M-theory on singular G2 and Spin(7) manifolds.
This is a very interesting problem which can provide us with many insights about
infrared behaviour of minimally supersymmetric gauge theories and even about
M-theory itself. From the experience with string theory and M-theory, one might
expect to find some new physics at the singularities of G2 and Spin(7) manifolds,
for example,
• New massless objects
• Extra gauge symmetry
• Restoration of continuous/discrete symmetry
• Topology changing transition
..
.
Before one talks about physics associated with G2 and Spin(7) singularities,
it would be nice to have a classification of all such degenerations. Unfortunately,
this problem is not completely solved even for Calabi-Yau manifolds (apart from
the ones in low dimension), and seems even less promising for real manifolds with
exceptional holonomy. Therefore, one starts with some simple examples.
One simple kind of singularities — which we already encountered in section 3.1
in the Joyce construction of compact manifolds with exceptional holonomy — is a
large class of orbifold singularities6. Locally, an orbifold singularity can be repre-
sented as a quotient of Rn by some discrete group Γ,
(4.1) Rn /Γ
Therefore, in string theory, the physics associated with such singularities can be
systematically extracted from the orbifold conformal field theory [21]. For G2
manifolds this was done in [51, 23, 52, 9, 24, 49], and for Spin(7) manifolds
in [51, 10]. Typically, one finds new massless degrees of freedom localized at the
orbifold singularity and other phenomena listed above. However, the CFT technique
is not applicable for studying M-theory on singular G2 and Spin(7) manifolds.
Moreover, as we mentioned in section 3.1, many interesting phenomena occur at
singularities which are not of the orbifold type, and to study the physics of those
we need some new methods.
6A large list of G orbifold singularities can be found in [33].

2
144 SERGEI GUKOV
S1
3
R
Figure 5. A cartoon representing Taub-NUT space as a circle

fibration over a 3-plane.
4.1. Low Energy Dynamics via IIA Duals. One particularly useful method
of analyzing M-theory on singular manifolds with special holonomy follows from
the duality between type IIA string theory and M-theory compactified on a circle
[53, 55]:
IIA Theory ⇐⇒ M-theory on S1
Among other things, this duality implies that any state in IIA theory can be
identified with the corresponding state in M-theory. In this identification, some of
the states acquire geometric origin when lifted to eleven dimensions. In order to
see this explicitly, let us write the eleven-dimensional metric in the form
ds2 = e− 3 φ gµν dxµ dxν + e 3 φ (dx11 + Aµ dxµ )2
2 4
(4.2)
Upon reduction to ten-dimensionsional space-time (locally parametrized by xµ ),
the field φ is identified with the dilaton, Aµ with the Ramond-Ramond 1-form,
and gµν with the ten-dimensional metric. Therefore, any IIA background that
involves excitations of these fields uplifts to a purely geometric background in eleven
dimensions. Moreover, from the explicit form of the metric (4.2) it follows that the
eleven-dimensional geometry is a circle fibration over the ten-dimensional space-
time, such that the topology of this fibration is determined by the configuration of
Ramond-Ramond 1-form field. This important observation will play a central role
in this section.
To be specific, let us consider a D6-brane in type IIA string theory. Since a
D6-brane is a source for the dilaton, for the Ramond-Ramond 1-form, and for the
metric, it is precisely the kind of state that uplifts to pure geometry. For example,
if both the D6-brane world-volume and the ambient space-time are flat, the dual
background in M-theory is given by the Taub-NUT space:
M-theory on IIA theory in flat space-time
⇐⇒
Taub-Nut space with a D6-brane
The Taub-NUT space is a non-compact four-manifold with SU (2) holonomy.
It can be viewed as a circle fibration over a 3-plane; see Figure5. The S1 fiber
degenerates at a single point — the origin of the R3 — which is identified with the
location of the D6-brane in type IIA string theory. On the other hand, at large
distance the size of the ‘M-theory circle’ stabilizes at some constant value (related
to the value of the string coupling constant in IIA theory). Explicitly, the metric
on the Taub-NUT space is given by [22]:
(4.3) ds2T N = Hdr 2 + H −1 (dx11 + Aµ dxµ )2 ,
∇
×A = −∇H,
1
H = 1+
2r
This form of the metric makes especially clear the structure of the circle fibration.
Indeed, if we fix a constant-r sphere inside the R3 , then it is easy to see that the
S1 fiber has ‘winding number one’ over this sphere. This indicates that there is a
topological defect — namely, a D6-brane — located at r = 0, where the S1 fiber
degenerates.
The relation between D6-branes and geometry can be extended to more general
manifolds that admit a smooth U (1) action. Indeed, if X is a space (not necessarily
smooth and compact) with a U (1) isometry, such that X/U (1) is smooth, then the
fixed point set, L, of the U (1) action must be of codimension 4 inside X [47, 43].
This is just the right codimension to identify L with the D6-brane locus7 in type
IIA theory on X/U (1):
M-theory on X ⇐⇒ IIA theory on X/U (1)

with D6-branes on L
For example, if X is the Taub-NUT space, then the U (1) action is generated by
the shift of the periodic variable x11 in (4.3). Dividing by this action one finds
X/U (1) ∼= R3 and that L = {pt} ∈ X indeed has codimension 4.
It may happen that a space X admits more than one U (1) action. In that
case, M-theory on X will have several IIA duals, which may look very different
but, of course, should exhibit the same physics. This idea was used by Atiyah,
Maldacena, and Vafa to realise geometric duality between certain IIA backgrounds
as a topology changing transition in M-theory [6]. We will come back to this in
section 5.
Now let us describe a particularly useful version of the duality between M-
theory on a non-compact space X and IIA theory with D6-branes on X/U (1) which
occurs when X admits a U (1) action, such that the quotient space is isomorphic to a
flat Euclidean space. Suppose, for example, that X is a non-compact G2 manifold,
such that
(4.4) X/U (1) ∼
= R6
Then, the duality statement reads:
IIA theory in flat space-
M-theory on non-compact
⇐⇒ time with D6-branes on
G2 manifold X
R4 × L
On the left-hand side of this equivalence the space-time is R4 × X. On the other
hand, the geometry of space-time in IIA theory is trivial (at least topologically)
and all the interesting information about X is mapped into the geometry of the
7The part of the D-brane world-volume that is transverse to X is flat and does not play an
important rôle in our discussion here.
146 SERGEI GUKOV
D6-brane locus L. For example, the Betti numbers of L are determined by the
corresponding Betti numbers of the space X [29]:
(4.5) bk (L) = bk+2 (X), k>0
b0 (L) = b2 (X) + 1
Notice the shift in degree by 2, and also that the number of the D6-branes (= the
number of connected components of L) is determined by the second Betti number
of X. We will not present the derivation of this formula here. However, it is a
useful exercise to check (e.g. using the Lefschetz fixed point theorem) that the
Euler numbers of X and L must be equal, in agreement with (4.5).
The duality between M-theory on a non-compact manifold X and a configura-
tion of D6-branes in a (topologically) flat space can be used to study singular limits
of X. Indeed, when X develops a singularity, so does L. Moreover, L must be a
supersymmetric (special Lagrangian) submanifold in R6 ∼ = C3 in order to preserve
8
the same amount of supersymmetry as the G2 space X. Therefore, the problem of
studying the dynamics of M-theory on G2 singularities can be restated as a problem
of studying D6-brane configurations on singular special Lagrangian submanifolds
in flat space [8].
Following Atiyah and Witten [8], let us see how this duality can help us to
analyze one of the conical G2 singularities listed in Table 4.
Example 4.1. Consider an asymptotically conical G2 manifold X with SU (3)
/ U (1)2 principal orbits and topology
(4.6) X∼ = CP2 × R3
Assuming that M-theory on this space X admits a circle reduction to IIA theory
with D6-branes in flat space, we can apply the general formula (4.5) to find the
topology of the D6-brane locus L. For the manifold (4.6) we find the following
non-zero Betti numbers:
Betti numbers of X Betti numbers of L
b4 = 1 −→ b2 = 1
b2 = 1 −→ b0 = 2
Therefore, we conclude that L should be a non-compact 3-manifold with two con-
nected components (since b0 = 2) and with a single topologically non-trivial 2-cycle
(since b2 = 1). A simple guess for a manifold that has these properties is

(4.7) L=∼ S2 × R ∪ R3
It turns out that there indeed exists a special Lagrangian submanifold in C3 with
the right symmetries and topology (4.7); see [8] for an explicit construction of the
circle action on X that has L as a fixed point set. If we choose to parametrize
Re(C3 ) = R3 and Im(C3 ) = R3 by 3-vectors φ 1 and φ 2 , respectively, then one can
explicitly describe L as the zero locus of the polynomial relations [38]:
√
(4.8) L = φ 2 = −|φ
1 · φ 1 ||φ
2 | ; |φ
1 |(3|φ
2 |2 − |φ
1 |2 ) = ρ ∪ |φ 1 − 3φ 2| = 0
8In general, in a reduction from M-theory down to Type IIA one does not obtain the stan-
dard flat metric on X/U (1) ∼ = 6 due to non-constant dilaton and other fields in the background.
However, one would expect that near the singularities of the D-brane locus L these fields exhibit
a regular behavior, and the metric on X/U (1) is approximately flat, cf. [8]. In this case the con-
dition for the Type IIA background to be supersymmetric can be expressed as a simple geometric
criterion: it says that the D-brane locus L should be a calibrated submanifold in X/U (1).
L
6
R
3
R
3
R3 R
Figure 6. Intersection of special Lagrangian D6-branes dual to a

G2 cone over the flag manifold SU (3)/U (1)2 , and its three non-
singular deformations.
It is easy to see that the first component of this manifold looks like a hyperboloid
in C3 = R3 × R3 . It has a ‘hole’ in the middle, resulting in the S2 × R topology,
required by (4.7). The second component in (4.8) is clearly a 3-plane, which goes
through this hole, as shown in Figure 6.
When X develops a conical singularity L degenerates into a collection of three
planes,
(4.9) Lsing ∼
= R3 ∪ R3 ∪ R3
intersecting at a single point, see Figure6. This can be seen explicitly by taking the
ρ → 0 limit in the geometry (4.8),
D61 : 1 = 0
φ
√
1 3
D62 : φ1 + φ2 = 0
2 2√
1 3
D63 : − φ 1+ φ2 = 0
2 2
which corresponds to the limit of a collapsing S2 cycle.
Therefore, in this example the physics of M-theory on a conical G2 singularity
can be reduced to the physics of three intersecting D6-branes in type IIA string
theory. In particular, since D6-branes appear symmetrically in this dual picture,
one can conclude that there must be three ways of resolving this singularity, de-
pending on which pair of D6-branes is deformed into a smooth special Lagrangian
submanifold (4.8). This is precisely what Atiyah and Witten found in a more careful
analysis [8]. We will come back to this example later, in section 5.
148 SERGEI GUKOV
There is a similar duality that relates M-theory on Spin(7) manifolds to D6-

brane configurations on coassociative cycles [25]. In particular, if X is a non-
compact Spin(7) manifold with a U (1) action, such that
(4.10) X/U (1) ∼
= R7
then one finds the following useful duality:
IIA theory in flat space-
M-theory on non-compact
⇐⇒ time with D6-branes on
Spin(7) manifold X
R3 × L
where L is a coassociative submanifold in R7 . Again, on the left-hand side the
geometry of space-time is R3 × X, whereas on the right-hand side space-time is
(topologically) flat and all the interesting information is encoded in the geometry
of the D6-brane locus L. The topology of the latter can be determined from the
general formula (4.5). Therefore, as in the G2 case, M-theory dynamics on singular
Spin(7) manifolds can be obtained by investigating D6-brane configurations on
singular coassociative submanifolds in flat space [29].
5. Topology Change in M-theory on Exceptional Holonomy Manifolds

5.1. Topology Change in M-theory. In this section we discuss topology
changing transitions, by which we mean a particular behavior of the manifold X
(and the associated physics) in the singular limit when one can go to a space with a
different topology. In Calabi-Yau manifolds many examples of such transitions are
known and can be understood using conformal field theory methods, see e.g. [4]
and references therein. Some of these transitions give rise to analogous topology
changing transitions in G2 and Spin(7) manifolds obtained from finite quotients
of Calabi-Yau spaces that we discussed in section 3.1. In the context of compact
manifolds with G2 holonomy this was discussed in [37, 48]. One typically finds
a transition between different topologies of X, such that the following sum of the
Betti numbers remains invariant:
(5.1) b2 + b3 = const
which is precisely what one would expect from the conformal field theory consid-
erations [51]. One can interpret the condition (5.1) as a feature of the mirror
phenomenon for G2 manifolds [2, 44, 32].
Here, we shall discuss topology change in non-compact asymptotically conical
G2 and Spin(7) manifolds of section 3.2, which are in a sense the most basic ex-
amples of singularities that reveal the peculiar aspects of exceptional holonomy.
Notice that all of these manifolds have the form
X∼
= B × (contractible)
where B is a non-trivial cycle (a bolt), e.g. B = S3 , S4 , CP2 , or something else.
Therefore, it is natural to ask:
“What happens if Vol(B) → 0 ?”
In this limit the geometry becomes singular and, as we discussed earlier, there are
many possibilities for M-theory dynamics associated with it. One possibility is a
topology change, which is indeed what we shall find in some of the cases below.
conifold
S3 singularity S2
TT
B@
B @ deformation Ac
Ac resolution
⇐= =⇒ T T
B @ A cc T T
B @ A c T T
B A T
B S2 A S2 T S2
3 3 3
S S S
mq = 0 mq = 0 q
= 0
Figure 7. Conifold transition in type IIB string theory.
Although we are mostly interested in exceptional holonomy manifolds, it is

instructive to start with topology changing transitions in Calabi-Yau manifolds,
where one finds two prototypical examples:
The Flop is a transition between two geometries, where one two-cycle shrinks
to a point and a (topologically) different two-cycle grows. This process can be
schematically described by the diagram:
S2(1) −→ · −→ S2(2)
This transition is smooth in string theory [5, 56].
The Conifold transition is another type of topology change, in which a three-
cycle shrinks and is replaced by a two-cycle:
S3 −→ · −→ S2
Unlike the flop, it is a real phase transition in the low-energy dynamics which can
be understood as the condensation of massless black holes [50, 26]. Let us briefly
recall the main arguments.
As the name indicates, the conifold is a cone over a five-dimensional base space
which has topology S2 × S3 (see Figure 7). Two different ways to desingularize
this space — called the deformation and the resolution — correspond to replacing
the singularity by a finite size S3 or S2 , respectively. Thus, we have two different
spaces, with topology S3 × R3 and S2 × R4 , which asymptotically look the same.
In type IIB string theory, the two phases of the conifold geometry correspond to
different branches in the four-dimensional N = 2 low-energy effective field theory.
In the deformed conifold phase, D3-branes wrapped around the 3-sphere give rise to
a low-energy field q, with mass determined by the size of the S3 . In the effective four-
dimensional supergravity theory these states appear as heavy, point-like, extremal
black holes. On the other hand, in the resolved conifold phase the field q acquires
an expectation value reflecting the condensation of these black holes. Of course, in
order to make the transition from one phase to the other, the field q must become
massless somewhere and this happens at the conifold singularity, as illustrated in
Figure7.
Now, let us proceed to topology change in G2 manifolds. Here, again, one finds
two kinds of topology changing transitions, which resemble the flop and the conifold
transitions in Calabi-Yau manifolds:
G2 Flop is a transition where a 3-cycle collapses and gets replaced by a (topo-
logically) different 3-cycle:
S3(1) −→ · −→ S3(2)
150 SERGEI GUKOV
3
Vol(S )
Figure 8. Quantum moduli space of M-theory on the G2 manifold

X with topology S3 × R4 . Green lines represent the ‘geometric
moduli space’ parametrized by the volume of the S3 cycle, which
is enlarged to a smooth complex curve by taking into account the
C-field and quantum effects. The resulting moduli space has three
classical limits, which can be connected without passing through
the point where the geometry becomes singular (represented by
the dot in this picture).
Note that this is indeed very similar to the flop transition in Calabi-Yau manifolds,
where instead of a 2-cycle we have a 3-cycle shrinking. The physics is also similar,
with membranes playing the role of string world-sheet instantons. Remember that
the latter were crucial for a flop transition to be smooth in string theory. For a
very similar reason, the G2 flop transition is smooth in M-theory. This was first
realised by Acharya [1], and by Atiyah, Maldacena, and Vafa [6], for a 7-manifold
with topology
X∼
= S3 × R4
and studied further by Atiyah and Witten [8]. In particular, they found that M-
theory on X has three classical branches, related by triality permutation symmetry,
so that the quantum moduli space looks as shown on Figure8. Once again, the
important point is that there is no singularity in quantum theory.
Let us proceed to another kind of topology changing transition in manifolds
with G2 holonomy.
A Phase Transition, somewhat similar to a hybrid of the conifold and the
G2 flop transition, can be found in M-theory on a G2 manifold with topology
X∼
= CP2 × R3
A singularity develops when the CP2 cycle shrinks. As in the conifold transition,
the physics of M-theory on this space also becomes singular at this point. Hence,
this is a genuine phase transition [8]. Note, however, that unlike the conifold
transition in type IIB string theory, this phase transition is not associated with
2
Vol(CP )
Figure 9. Quantum moduli space of M-theory on the G2 manifold

X with topology CP2 × R3 . Green lines represent the ‘geometric
moduli space’ parametrized by the volume of the CP2 cycle, which
is enlarged to a singular complex curve by taking into account
the C-field and quantum effects. The resulting moduli space has
three classical limits. In order to go from one branch to another
one necessarily has to pass through the point where the geometry
becomes singular (represented by the dot in this picture).
condensation of any particle-like states in M-theory9 on X. Indeed, there are no 4-

branes in M-theory, which could result in particle-like objects by wrapping around
the collapsing CP2 cycle.
Like the G2 flop transition, this phase transition has three classical branches,
which are related by triality symmetry, see Figure9. The important difference, of
course, is that now one can go from one branch to another only through the singular
point. In this transition one CP2 cycle shrinks and another (topologically different)
CP2 cycle grows:
CP2(1) −→ · −→ CP2(2)
One way to see that this is indeed the right physics of M-theory on X is to reduce
it to type IIA theory with D6-branes in flat space-time [8].
Finally, we come to the last and the hardest case of holonomy groups, namely
to Spin(7) holonomy.
The Spin(7) Conifold is the cone on SU (3)/U (1). It was conjectured in [31]
that the effective dynamics of M-theory on the Spin(7) conifold is analogous to
that of type IIB string theory on the usual conifold. Namely, the Spin(7) cone on
SU (3)/U (1) has two different desingularizations, obtained by replacing the conical
singularity either with a 5-sphere or with a CP2 ; see Figure10. As a result, we
obtain two different Spin(7) manifolds, with topology
CP2 × R4 and S5 × R3
which are connected via the topology changing transition
CP2 −→ · −→ S5
9However, such an interpretation can be given in type IIA string theory [31].
152 SERGEI GUKOV
S5 CP 2
first AA second
LL resolution A resolution
⇐= =⇒
LL
L A L
L A L
L A L
SU (3)/U (1)
mq = 0 mq = 0 q
= 0
Figure 10. Conifold transition in M-theory on a manifold with
Spin(7) holonomy.
As with the conifold transition [50, 26], the Spin(7) conifold has a nice interpreta-
tion in terms of the condensation of M5-branes. Namely, in the S5 phase we have
extra massive states obtained upon quantization of the M5-brane wrapped around
the five-sphere. The mass of these states is related to the volume of the S5 . At the
conifold point where the five-sphere shrinks, these M5-branes become massless as
suggested by the classical geometry. At this point, the theory may pass through
a phase transition into the Higgs phase, associated with the condensation of these
five-brane states; see Figure10.
To continue the analogy with the Calabi-Yau conifold, recall that the moduli
space of type II string theory on the Calabi-Yau conifold has three semi-classical
regimes. The deformed conifold provides one of these, while there are two large-
volume limits of the resolved conifold, related to each other by a flop transition. In
fact, the same picture emerges for the Spin(7) conifold. In this case, however, the
two backgrounds differ not in geometry, but in the G-flux. It was shown in [29] that,
due to the membrane anomaly of [57], M-theory on X ∼ = CP2 × R4 is consistent
only for half-integral units of G4 through the CP bolt. Namely, after the transition
2
from X ∼ = S5 × R3 , the G-flux may take the values ±1/2, with the two possibilities
related by a parity transformation [31]. Thus, the moduli space of M-theory on
the Spin(7) cone over SU (3)/U (1) also has three semi-classical limits: one with
the parity invariant background geometry S5 × R3 , and two with the background
geometry X ∼ = CP2 × R4 where parity is spontaneously broken; see Figure11. The
last two limits are mapped into each other under parity transformation.
This picture is reproduced in the effective low-energy theory if we include in
the spectrum light states corresponding to M5-branes wrapped over the five-sphere:
Effective Theory: N = 1, D = 3 Maxwell-Chern-Simons theory with one
charged complex scalar multiplet q
Here, it is the Higgs field q that appears due to quantizaion of the M5-branes.
In this theory, different topological phases correspond to the Coulomb and Higgs
branches:
(5.2) S5 × R3 ⇐⇒ Coulomb branch
(5.3) CP2 × R4 ⇐⇒ Higgs branch
Further agreement in favor of this identification arises from examining the various
extended objects that exist in M-theory on CP2 ×R4 , obtained from wrapped M5 or
M2-branes. For example, we can consider an M2-brane over CP1 inside CP2 × R4 .
vol(S5)
5 3
S R
k=−12 k=+ 1 2
2
CP R4
vol(CP2) vol(CP2)
Parity
Figure 11. The moduli space of M-theory on the Spin(7) cone

over SU (3)/U (1) can be compared to the vacuum structure of a
system with spontaneous symmetry
breaking. On this picture, the
G-flux is measured by k = CP2 G/2π.
This non-BPS state has a semi-classical mass proportional to the volume of CP1 ,
and is electrically charged under the global U (1)J symmetry of our gauge theory.
Therefore, this state can be naturally identified with a vortex. Note that this state
can be found only in the CP2 phase (i.e. in the Higgs phase), in complete agreement
with the low-energy physics.
In view of the interesting phenomena associated to branes in the conifold ge-
ometry, and their relationship to the conifold transition [41, 40], it would be in-
teresting to learn more about the Spin(7) transition using membrane probes in
this background, and also to study the corresponding holographic renormalization
group flows. For work in this area, see [30, 45, 27].
5.2. Relation to Geometric Transition. In the previous section we de-

scribed the basic examples of topology changing transitions in exceptional holo-
nomy manifolds, and commented on the important aspects of M-theory dynamics
in these transitions. As we explain in this section, some of these transitions also
have a nice interpretation in type IIA string theory, realizing dualities between
backgrounds involving D6-branes and Ramond-Ramond fluxes in manifolds with
more restricted holonomy. Specifically, we will consider two cases:
• SU(3) → G2 : We start with a relation between the conifold transition in the
presence of extra fluxes and branes and the G2 flop transition in M-theory [6]. Note
that these two transitions are associated with different holonomy groups10 and,
in particular, with different amounts of unbroken supersymmetry. The relation,
10That is why we refer to this case as SU (3) → G .

2
154 SERGEI GUKOV
RR flux
D6−brane
2
S 2 S
3 3
S S
Figure 12. Geometric transition in IIA string theory connecting
D6-branes wrapped around S3 in the deformed conifold geometry
and resolved conifold with Ramond-Ramond 2-form flux through
the S2 .
however, appears when we introduce extra matter fields, represented either by D-

branes or by fluxes. They break supersymmetry further, therefore, providing a
relation between two different holonomy groups.
In order to explain how this works in the case of the conifold, let us consider type
IIA theory on the deformed conifold geometry, T ∗ S3 ∼ = S3 ×R3 . This already breaks
supersymmetry down to N = 2 in four dimensions. One can break supersymmetry
further, to N = 1, by wrapping a space-filling D6-brane around the supersymmetric
(special Lagrangian) S3 cycle in this geometry. Then, a natural question to ask is:
“What happens if one tries to go through the conifold transition with the extra
D6-brane?”. One possibility could be that the other branch is no longer connected
and the transition is not possible. However, this is not what happens. Instead the
physics is somewhat more interesting. According to [54], the transition proceeds,
but now the two branches are smoothly connected, with the wrapped D6-brane
replaced by Ramond-Ramond 2-form flux through the S2 cycle of the resolved
conifold; see Figure12.
As we explained in section 4, both D6-branes and Ramond-Ramond 2-form
tensor fields lift to purely geometric backgrounds in M-theory. Therefore, the geo-
metric transition described above should lift to a transition between two purely
geometric backgrounds in M-theory (hence, the name). Since these geometries
must preserve the same amount of supersymmetry, namely N = 1 in four dimen-
sions, we conclude that we deal with a G2 transition. In fact, it is the familiar flop
transition in M-theory on a G2 manifold [1, 6]:
X∼
= S3 × R4
Indeed, if we start on one of the three branches of this manifold and choose the
‘M-theory circle’ to be the fiber of the Hopf bundle in the 3-sphere
S1 → S3 → S2
we obtain the resolved conifold as a quotient space, X/U (1) ∼= S2 × R4 . More

precisely, we obtain a resolved conifold with Ramond-Ramond 2-form flux and no
D6-branes because the circle action has no fixed points in this case. This gives
us one side of the brane/flux duality, namely the right-hand side on the diagram
below:
M-theory on G2 flop M-theory on G2
manifold S3(1) × R4 ←−−−−−−−−−−−−−−−→ manifold S3(2) × R4
↓ ↓
IIA on S3 ×R3 with geometric transition IIA on S2 ×R4 with

←−−−−−−−−−−−−−−−→ RR flux through S2
D6-brane on S3
Now, let us follow the G2 flop transition in M-theory on the manifold X ∼ =
S × R4 . As explained in the previous section, after the transition we obtain a
3
G2 manifold with similar topology, but the M-theory circle is now embedded in
R4 , rather than in S3 . Acting on each R4 fiber, it yields R3 = R4 /U (1) as a
quotient space with a single fixed point at the origin of the R4 (see the discussion
of M-theory on the Taub-NUT space in section 4). Applying this to each fiber
of the G2 manifold X, we obtain the deformed conifold as the quotient space,
X/U (1) ∼ = S3 × R3 , with the fixed point set L = S3 . Since the latter is identified
with the location of the space-filling D6-brane, we recover the other side of the
brane/flux duality, illustrated in the above diagram.
Thus, we explained that the geometric transition — which is a highly non-
trivial, non-perturbative phenomenon in string theory — can be understood as
a G2 flop transition in M-theory. Various aspects of this transition have been
discussed in [1, 6, 8, 11, 18, 13, 19, 7]. As we show next, there is a similar
relation between the phase transition in the G2 holonomy manifold X = CP2 × R3
and the Spin(7) conifold transition, discussed in the previous section.
• G2 → Spin(7) : Consider type IIA string theory on the G2 holonomy mani-
fold
(5.4) CP2 × R3
which is obtained by resolving the cone over SU (3)/U (1)2 . As was discussed in
the previous section, the corresponding moduli space has three classical branches
connected by a singular phase transition. Motivated by the geometric transition in
the conifold example, one could wrap an extra D6-brane over the CP2 cycle and ask
a similar question: “What happens if one tries to go through a phase transition?”.
Using arguments similar to [6], one finds that the transition is again possible,
via M-theory on a Spin(7) manifold [31]. More precisely, after the geometric tran-
sition one finds type IIA string theory in a different phase of the G2 manifold (5.4),
where the D6-brane is replaced by RR flux through CP1 ⊂ CP2 . This leads to a
fibration:
S1 → S5 → CP2
Hence the M-theory lift of this configuration gives a familiar Spin(7) conifold,
X∼ = S5 × R3 .
Similarly, one can identify the lift of CP2 × R3 with a D6-brane wrapped around
CP2 as the Spin(7) manifold CP2 × R4 , which is another topological phase of the
156 SERGEI GUKOV
RR flux
D6−brane
CP2
CP 2
SU(3)/ SU(3)/
U(1)2 U(1)2
Figure 13. Geometric transition in IIA string theory connecting
one branch of the G2 manifold CP2 × R3 , where D6-branes are
warpped around the CP2 cycle and another branch, where D6-
branes are replaced by Ramond-Ramond 2-form flux through the
CP1 ⊂ CP2 .
Spin(7) conifold. Summarizing, we find that the conifold transition in M-theory on

a Spin(7) manifold is related to a geometric transition in IIA string theory on the
G2 manifold (5.4) with branes/fluxes, as shown in the diagram below:
M-theory on Spin(7) conifold transition M-theory on Spin(7)
manifold CP2 × R4 ←−−−−−−−−−−−−−−−→ manifold S5 × R3
↓ ↓
IIA on CP2 ×R3 with geometric transition IIA on CP2 ×R3 with
←−−−−−−−−−−−−−−−→ RR flux through CP1
D6-brane on CP2
However, unlike its prototype with larger supersymmetry, this transition does
not proceed smoothly.
References
1. B.S. Acharya, On Realising N=1 Super Yang-Mills in M theory, hep-th/0011089.
2. B.S. Acharya, On Mirror Symmetry for Manifolds of Exceptional Holonomy, Nucl. Phys.
B524 (1998) 269; “Exceptional Mirror Symmetry 1-14,” in Winter School on Mirror Sym-
metry, Vector Bundles and Lagrangian Submanifolds, C. Vafa and S.-T. Yau, eds., AMS and
International Press, Boston, 2001.
3. D.V. Alekseevsky, Riemannian Spaces with Exceptional Holonomy, Funct. Anal. Appl. 2
(1968) 97.
R. Brown and A. Gray, Riemannian Manifolds with Holonomy Group Spin(9), Differential
Geometry (in honor of Kentaro Yano), Tokyo (1972) 41.
4. P. S. Aspinwall, B. R. Greene, D. R. Morrison, Spacetime Topology Change: The Physics of
Calabi-Yau Moduli Space, hep-th/9311186.
5. P. Aspinwall, B. Greene and D. Morrison, Multiple Mirror Manifolds and Topology Change
in String Theory, Phys. Lett. B303 (1993) 249, hep-th/9301043.
6. M. Atiyah, J. Maldacena and C. Vafa, An M-theory Flop as a Large N Duality, J. Math.

Phys. 42 (2001) 3209, hep-th/0011256.
7. M. Aganagic and C. Vafa, G2 manifolds, mirror symmetry and geometric engineering,
arXiv:hep-th/0110171.
8. M. Atiyah and E. Witten, M-Theory Dynamics On A Manifold Of G2 Holonomy, Adv.
Theor. Math. Phys. 6 (2002) 1.
9. R. Blumenhagen and V. Braun, Superconformal field theories for compact G2 manifolds,
JHEP 0112, 006 (2001) [arXiv:hep-th/0110232].
10. R. Blumenhagen and V. Braun, Superconformal field theories for compact manifolds with
Spin(7) holonomy, JHEP 0112, 013 (2001) [arXiv:hep-th/0111048].
11. A. Brandhuber, J. Gomis, S. S. Gubser and S. Gukov, Gauge theory at large N and new G2
holonomy metrics, Nucl. Phys. B. 611, 179 (2001), hep-th/0106034.
12. M. Berger, Sur les groupes d’holonomie des variétés á connexion affines et des variétés
riemanniennes, Bulletin de la Société Mathématique de France, 83 (1955) 279.
13. A. Brandhuber, G2 holonomy spaces from invariant three-forms, Nucl. Phys. B. 629, 393
(2002) [arXiv:hep-th/0112113].
14. R.L. Bryant, Metrics with Exceptional Holonomy, Ann. Math. 126 (1987) 525.
15. R. Bryant, S. Salamon, On the Construction of some Complete Metrics with Exceptional
Holonomy, Duke Math. J. 58 (1989) 829.
16. Z. W. Chong, M. Cvetic, G. W. Gibbons, H. Lu, C. N. Pope and P. Wagner, General
metrics of G2 holonomy and contraction limits, Nucl. Phys. B. 638 (2002) 459. arXiv:hep-
th/0204064.
17. M. Cvetic, G.W. Gibbons, H. Lu, C.N. Pope, Cohomogeneity One Manifolds of Spin(7) and
G2 Holonomy, Phys. Rev. D65 106004 (2002).
18. M. Cvetic, G. W. Gibbons, H. Lu and C. N. Pope, M-theory conifolds, Phys. Rev. Lett. 88,
121602 (2002) [arXiv:hep-th/0112098].
19. M. Cvetic, G. W. Gibbons, H. Lu and C. N. Pope, A G2 unification of the deformed and
resolved conifolds, Phys. Lett. B. 534, 172 (2002) [arXiv:hep-th/0112138].
20. M. Cvetic, G. W. Gibbons, H. Lu, C. N. Pope, New Complete Non-compact Spin(7) Mani-
folds, Nucl. Phys. B620 (2002) 29, hep-th/0103155.
21. L. J. Dixon, J. A. Harvey, C. Vafa and E. Witten, Strings On Orbifolds. 1 and 2, Nucl. Phys.
B 261 (1985) 678; Nucl. Phys. B. 274 (1986) 285.
22. T. Eguchi and A. J. Hanson, Asymptotically Flat Selfdual Solutions To Euclidean Gravity,
Phys. Lett. B 74 (1978) 249; T. Eguchi, P. B. Gilkey and A. J. Hanson, Gravitation, Gauge
Theories And Differential Geometry, Phys. Rept. 66 (1980) 213.
23. T. Eguchi and Y. Sugawara, CFT description of string theory compactified on non-compact
manifolds with G2 holonomy, Phys. Lett. B 519, 149 (2001) [arXiv:hep-th/0108091].
24. T. Eguchi and Y. Sugawara, String theory on G2 manifolds based on Gepner construction,
25. J. Gomis, D-Branes, Holonomy and M-Theory, Nucl. Phys. B606 (2001) 3, hep-th/0103115.
26. B. Greene, D. Morrison and A. Strominger, Black Hole Condensation and the unification of
String Vacua, Nucl. Phys. B451 (1995) 109, hep-th/9504145.
27. U. Gürsoy, C. Nuñez, M. Schvellinger, RG flows from Spin(7), CY 4-fold and HK manifolds
to AdS, Penrose limits and pp waves, JHEP 6, (2002) 015.
28. G. W. Gibbons, D. N. Page, C. N. Pope, Einstein Metrics on S 3 , 3 and 4 Bundles,
Commun.Math.Phys 127 (1990) 529-553.
29. S. Gukov and J. Sparks, M theory on Spin(7) manifolds. I, Nucl. Phys. B625 (2002) 3,
hep-th/0109025.
30. S. Gukov and D. Tong, D-brane probes of G2 holonomy manifolds, Phys. Rev. D66 (2002)
087901. hep-th/0202125.
S. Gukov and D. Tong, D-brane probes of special holonomy manifolds, and dynamics of
N = 1 three-dimensional gauge theories, JHEP 0204, 050 (2002), hep-th/0202126.
31. S. Gukov, J. Sparks and D. Tong, Conifold transitions and five-brane condensation in M-
theory on Spin(7) manifolds, Classical Quantum Gravity 20 (2003) 665. hep-th/0207244.
32. S. Gukov, S. T. Yau and E. Zaslow, Duality and fibrations on G2 manifolds, Turkish J.
Math. 7 (2003) 61. arXiv:hep-th/0203217.
33. Y.-H. He, G2 Quivers, JHEP 0302 (2003) 023.
158 SERGEI GUKOV
34. N. Hitchin, Stable forms and special metrics, in Global Differential Geometry: The math-
ematical legacy of Alfred Gray (Bilbao, Spain 2000) 70 Contemp. Math. 288, AMS 2001.
math.DG/0107101.
35. N. Hitchin, The geometry of three-forms in six and seven dimensions, math.DG/0010054.
36. R. Harvey and H.B. Lawson, Jr., Calibrated geometries, Acta Math. 148 (1982) 47.
37. D. Joyce, Compact Manifolds with Special Holonomy, Oxford University Press, 2000.
38. D. Joyce, Special Lagrangian m-folds in C m with symmetries, Duke Math. J. 115 (2002) 6.
math.DG/0008021.
39. A. Kovalev, Twisted connected sums and special Riemannian holonomy, J. Reine Ang. Math.
565 (2003) 125. math.dg/0012189.
40. I. R. Klebanov and M. J. Strassler, Supergravity and a confining gauge theory: Duality cas-
cades and chiSB-resolution of naked singularities, JHEP 0008, 052 (2000), hep-th/0007191.
41. I. Klebanov and E. Witten, Superconformal field theory on threebranes at a Calabi-Yau
singularity, Nucl. Phys. B536, 199 (1998), hep-th/9807080.
42. H. Kanno, Y. Yasui, On Spin(7) holonomy metric based on SU(3)/U(1) I and II, J. Geom.
Phys. 43 (2002) 293-209, 310-326. hep-th/0108226 and hep-th/0111198.
43. J. Levine, Semi-Free Circle Actions on Spheres, Invent. Math. 22 (1973) 161.
44. J.-H. Lee and N. C. Leung, Geometric Structures on G2 and Spin(7)-Manifolds,
math.DG/0202045.
45. A. Loewy, Y. Oz, Branes in Special Holonomy Backgrounds, Phys.Lett. B537 (2002) 147.
hep-th/0203092.
46. R. McLean, Deformations of Calibrated Submanifolds, Comm. Anal. Geom. 6 (1998) 705-
747.
47. D. Montgomery, L. Zippin, Topological transformation groups, New York: Interscience, 1955.
48. H. Partouche and B. Pioline, Rolling among G2 vacua, JHEP 0103 (2001) 005, [arXiv:hep-
th/0011130].
49. R. Roiban, C. Romelsberger and J. Walcher, Discrete torsion in singular G2 -manifolds and
real LG, Adv. Theor. Math. Phys. 6 (2003) 207. arXiv:hep-th/0203272.
50. A. Strominger, Massless black holes and conifolds in string theory Nucl. Phys. B. 451, 96
(1995), hep-th/9504090.
51. S. Shatashvili, C. Vafa, Superstrings and Manifolds of Exceptional Holonomy, Selecta Math
(N. S. ) 1 (1995) 347. hep-th/9407025.
52. K. Sugiyama and S. Yamaguchi, Cascade of special holonomy manifolds and heterotic string
theory, Nucl. Phys. B622, 3 (2002) [arXiv:hep-th/0108219].
53. P.K. Townsend, The eleven-dimensional supermembrane revisited, Phys. Lett. B350 (1995)
184.
54. C. Vafa, Superstrings and topological strings at large N, J. Math. Phys. 42 (2001) 2798,
hep-th/0008142.
55. E. Witten, String theory dynamics in various dimensions, Nucl. Phys. B443 (1995) 85,
56. E. Witten, ”Phases of N = 2 Models in Two Dimensions”, Nucl. Phys. B403 (1993) 159,
hep-th/9301042.
57. E. Witten, On Flux Quantisation in M-Theory and the Effective Action, J. Geom. Phys. 22,
1 (1997), hep-th/9609122.
58. S.-T. Yau, On the Ricci curvature of a compact Kähler manifold and the Monge-Ampére
equations. I, Commun. Pure Appl. Math. 31 (1978) 339.
Jefferson Physical Laboratory, Harvard University, Cambridge, MA 02138, U.S.A.

E-mail address: gukov@tomonaga.harvard.edu
Volume 3, 2004
Special holonomy and beyond
Nigel Hitchin
Abstract. Manifolds with special holonomy are traditionally associated with

the existence of parallel spinors, but Calabi-Yau threefolds and G2 manifolds
also arise naturally in the context of a nonlinear form of Hodge theory: finding
critical points of a natural functional defined on the differential p-forms in a
fixed cohomology class. The advantage of this point of view is that it gives
a natural approach to the moduli spaces of these structures, which appear as
open sets in the appropriate cohomology group, and also leads to other, less
familiar, geometric structures.
1. Invariant functionals and special holonomy

The list of special holonomy groups of Riemannian manifolds which emerged
from Berger’s work in the 1950’s was recognized later to contain all those manifolds
which possess covariant constant spinors, that is spinor fields ψ with ∇ψ = 0. They
are:
(1) SU (n): Calabi-Yau manifolds
(2) Sp(n): hyperkähler manifolds
(3) G2 : special manifolds in dimension 7
(4) Spin(7): special manifolds in dimension 8.
All of these are Einstein metrics with vanishing Ricci tensor. A second list of
special metrics consists of those with Killing spinors, spinor fields ψ satisfying
∇X ψ = λXψ. These are:
(1) spheres
(2) Einstein-Sasakian manifolds in dimension 2k + 1
(3) 3-Sasakian manifolds in dimension 4k + 3
(4) nearly Kähler manifolds in dimension 6
(5) weak holonomy G2 manifolds in dimension 7.
These are Einstein metrics with Rij = Λgij and Λ > 0. The link between them,
established by Bär in 1993 [1], is that the Killing spinor on a manifold M n extends
to become a covariant constant spinor on the (n + 1)-dimensional cone over M n .
In fact all of these geometries are defined by reference to a closed form or
collection of forms, and for some of them there is an alternative setting, independent
1991 Mathematics Subject Classification. Primary 53C29, 53C26, 32Q25,53C80, 58E11.

Key words and phrases. Special holonomy, Calabi-Yau.
2004
159
160 NIGEL HITCHIN
of spinors or Riemannian metrics, which we shall present here. The approach is

very natural, but leads on to some geometrical structures which appear not to have
been considered before.
We know what a non-degenerate symmetric bilinear form is, and it gives us the
notion of a metric, but what is a non-degenerate exterior form? One answer is to
ask that it should be an element of the vector space Λp V ∗ which lies in an open
orbit of the action of the general linear group GL(V ). For example, GL(2n, R)
acts on Λ2 R2n with an open orbit: the orbit consists of nondegenerate alternating
bilinear forms.
Now GL(V ) acts on Λp V ∗ , but dim Λp V ∗ = n!/(n − p)!p! and dim GL(V ) = n2
so in particular to get an open orbit we need
n!
≤ n2
(n − p)!p!
which puts serious constraints on when this occurs. We list the possibilities be-
low, but first we note that the word “nondegenerate” has specific connotations for
bilinear skew forms so instead of using this word we make the
Definition 1.1. If ρ ∈ Λp V ∗ lies on an open orbit under GL(V ) it is called
stable.
Note that an open orbit in Λp V ∗ implies the existence of an open orbit in Λn−p V ∗
by duality, so stability really occurs for pairs of forms in complementary dimensions.
There is a list, known in one form or another for many years, of when stable
forms arise:
(1) p = 1 all n: stable = non-zero
(2) p = 2 all n: if n = 2m + 1 or 2m, then stable forms are of rank 2m when
considered as bilinear forms. The stabilizer subgroup in dimension 2m is
Sp(2m, R).
(3) p = 3, n = 6: stabilizer SL(3, C) or SL(3, R) × SL(3, R)
(4) p = 3, n = 7: stabilizer G2 or its non-compact form
(5) p = 3, n = 8: stabilizer SU (3), SU (2, 1) or SL(3, R) in the adjoint repre-
sentation
and for the last three this means that we have normal forms at the linear algebra
level (for stabilizers SL(3, C), G2 and SU (3)) given by:
• dx1 dx2 dx3 − dy1 dy2 dx3 − dy2 dy3 dx1 − dy3 dy1 dx2
• (dx1 dx2 − dx3 dx4 )dx5 + (dx1 dx3 − dx4 dx2 )dx6 + (dx1 dx4 − dx2 dx3 )dx7 +
dx5 dx6 dx7
• dx1 dx2 dx3 −dx1 (dx7 dx4 +dx5 dx6 )−dx2 (dx7 dx5 +dx6 dx4 )+dx3 (dx7 dx6 +
dx4 dx5 ) + dx8 (dx4 dx5 + dx6 dx7 )
Now stability implies that there is an open set U = GL(V )/H ⊂ Λp V ∗ , and
the stabilizer H of a point in U is Sp(2m, R), G2 . . . etc. Each one of these groups
preserves also a volume form (metric or symplectic) so there is a GL(V )-invariant
map
φ : U → Λn V ∗ .
SPECIAL HOLONOMY AND BEYOND 161
Invariance under the scalar matrices implies that

φ(λp ρ) = λn φ(ρ)
so φ is homogeneous of degree n/p, and extends to a continuous map
φ : Λp V ∗ → Λn V ∗
which is smooth on the open set U ⊂ Λp V ∗ . The derivative of φ at ρ is a linear
map from Λp V ∗ to Λn V ∗ i.e.
Dφ ∈ (Λp V ∗ )∗ ⊗ Λn V ∗
But (Λp V ∗ )∗ ∼
= Λn−p V ∗ ⊗ Λn V so there is a unique
ρ̂ ∈ Λn−p V ∗
such that the derivative can be expressed as
Dφ(ρ̇) = ρ̂ ∧ ρ̇.
Note that if we take ρ̇ = ρ, then Euler’s formula for homogeneous functions implies
that
n
ρ̂ ∧ ρ = φ(ρ).
p
This nonlinear algebraic operation ρ → ρ̂ is familiar enough in the concrete cases:

(1) for n = 6, p = 3, ρ̂ is determined by the property that Ω = ρ + iρ̂ is a
complex (3, 0)-form preserved by SL(3, C),
(2) for n = 7, p = 3 or 4, ρ̂ = ∗ρ, where ∗ is the Hodge star operator for the
inner product on V ,
(3) for n = 8, p = 3 or p = 5, ρ̂ = − ∗ ρ.
Example 1.2. Take the symplectic case of ρ = ω ∈ Λ2 V ∗ . Then ω is stable

if it is nondegenerate, i.e. if ω lies in the open set U defined by det ωij = 0. The
volume form is the usual Liouville volume φ(ω) = ω m (up to a universal scale).
Differentiating, we get
ω̂ = mω m−1 ∈ Λ2m−2 V ∗
and ω ∧ mω m−1 = mω m verifying the homogeneity.
Suppose now that we have a compact manifold M n , and a p-form ρ ∈ Ωp (M )

which is stable at each point. This means that topologically we must have a re-
duction of the structure group of the tangent bundle to the stabilizer – one of the
groups in the list above. Then since ρ ∈ C ∞ (M, Λp T ∗ ) we obtain
φ(ρ) ∈ C ∞ (M, Λn T ∗ ).
Now define the diffeomorphism-invariant functional

Φ(ρ) = φ(ρ).
M

We now ask the question: what is a critical point of the functional Φ(ρ) = M
φ(ρ)
for ρ a closed form in a fixed cohomology class in H p (M, R)?
162 NIGEL HITCHIN
This is a nonlinear version of Hodge theory. The answer is obtained by calcu-

lating the first variation:

δΦ(ρ̇) = Dφ(ρ̇) = ρ̂ ∧ ρ̇.
M M
Because ρ is varying in a fixed cohomology class, ρ̇ = dα for some α and so

δΦ(ρ̇) = ρ̂ ∧ dα = ± dρ̂ ∧ α.
M M
The critical points which are stable forms therefore occur when
dρ = 0 = dρ̂.
Example 1.3. Take the symplectic case. If ω ∈ Ω2 (M ) lies in a fixed coho-

mology class in H 2 (M, R) then

Φ(ω) = ω m = [ω]m [M ] = const.
M
so any symplectic form is a critical point. On the other hand suppose we take
the stable form ρ = ω m−1 as the basic object and only suppose that this one is
closed and lies in a fixed cohomology class in H 2m−2 (M, R). This time Φ(ρ) is not
constant, and the critical points are where
dρ̂ = dω = 0
or in other words when ω is symplectic.
A critical point can never be an (isolated) Morse critical point since the func-
tional is invariant by the full group of diffeomorphisms of M . We can ask never-
theless if it is nondegenerate transverse to the action of Diff (M ) – an infinite di-
mensional version of a Morse-Bott critical point. In the symplectic case of ρ = ω m
above, we calculate the second variation to see this:

δ 2 Φ(ρ̇1 , ρ̇2 ) = m(m − 1) ω m−2 ω̇1 ω̇2
M
2
The degeneracy subspace is defined by δ Φ(ρ̇1 , ρ̇2 ) = 0 for all ρ̇2 = dσ, which gives

0 = (m − 1) ω m−2 ω̇1 ω̇2 = ω̇1 ρ̇2 = ω̇1 ∧ dσ.
M M M
So ρ̇1 is in the degeneracy space iff dω̇1 = 0.
Now recall the strong Lefschetz property:
∼ H 2m−2 (M, R)
∪[ω]m−2 : H 2 (M, R) =
for symplectic manifolds. We know that dω̇1 = 0 and since
ρ̇1 = (m − 1)ω m−2 ω̇1
if strong Lefschetz holds then ρ̇1 exact implies ω̇1 is exact, i.e. ω̇1 = dα. In this
case we define the vector field X by iX ω = α and then
ω̇1 = LX ω
which means that the variation is tangential to the action of Diff (M ). The features
we encounter here are:
(1) Φ is nondegenerate transverse to Diff (M ) orbits if strong Lefschetz holds,
(2) the local moduli space ∼

= open set in H 2m−2 (M, R),
(3) the Hessian of Φ defines an indefinite metric on the moduli space.
Note that traditionally the local moduli space of symplectic structures is given by
the de Rham cohomology class [ω] ∈ H 2 (M, R) but [ω] → [ω m−1 ] is a local diffeo-
morphism if strong Lefschetz holds, which brings the two points of view together.
A more interesting example than the symplectic one is the case of 3-forms in
six dimensions as in [6]. In this case ρ ∈ Λ3 T ∗ is the real part of a complex locally
decomposable form ν, and φ(ρ) = iν ∧ ν̄/2. Here ρ̂ = Im ν ∈ Λ3 T ∗ and so ν = ρ+iρ̂
is a (3, 0) form for an almost complex structure on M . The critical points of Φ on
a class in H 3 (M, R) give
d(ρ + iρ̂) = 0
and this implies that the complex structure is integrable and ν is a non-vanishing
holomorphic three-form. Thus M 6 is given the structure of a complex threefold
with trivial canonical bundle (for example a Calabi-Yau manifold). The essential
features in this case are:
¯
(1) Φ is nondegenerate transverse to Diff (M ) orbits if the ∂ ∂-lemma holds,
(2) the local moduli space is isomorphic to an open set in H 3 (M, R), which
gives another proof in this dimension of the unobstructedness of moduli,
(3) the Hessian of Φ defines an indefinite special Kähler metric on the moduli
space.
In dimension 7, ρ ∈ Λ3 T ∗ which is stable at each point defines an almost G2 -

structure and φ(ρ) is the Riemannian volume of the G2 -metric. We find ρ̂ = ∗ρ ∈
Λ4 T ∗ , with ∗ the Hodge star operator of this metric and the critical points of Φ are
where
dρ = 0 = d ∗ ρ.
This is known to be equivalent to the condition that ρ is covariant constant with
respect to the metric and so defines a metric with holonomy G2 .
Using Hodge theory, the features here are that transverse nondegeneracy holds
and, as a consequence, the moduli space is an open set in H 3 (M, R). The Hessian
of Φ again defines an indefinite metric on this space.
All of these structures clearly have a distinguished class of flat coordinates on

their moduli spaces. This would equally be true of the final structure thrown out
by the stability condition in dimension 8 if only we knew of a nontrivial example.
We study this geometry next nevertheless.
As noted above, ρ ∈ Λ3 T ∗ defines the structure of the Lie algebra of SU (3) on
T . This means that we have a metric defined by the Killing form tr(XY ) and the
3-form is then
ρ(X, Y, Z) = tr(X[Y, Z]).
The integrand in the functional is then the Riemannian volume form φ(ρ) and
ρ̂ = ∗ρ ∈ Λ5 T ∗ .
164 NIGEL HITCHIN
Critical points of the functional on a cohomology class in H 3 (M, R) are therefore

given by
dρ = 0 = d ∗ ρ.
This is directly analogous to the G2 situation but there are essential differences, in
particular we do not get a metric of special holonomy since SU (3) does not appear
in Berger’s list.
One approach to this geometry is through triality in 8 dimensions. Recall that
there are three real 8-dimensional representations of Spin(8): the vector represen-
tation T and the two half-spin representations S + , S − . These are permuted by an
outer automorphism ∆+ : Spin(8) → Spin(8) of order three. Now we can Clifford
multiply a spinor ψ ∈ S + by a vector x ∈ T to get x · ψ ∈ S − and since x is skew
adjoint in this action,
(x · ψ, x · ψ) = (−x2 · ψ, ψ)
But in the Clifford algebra x2 = −(x, x)1 so this expression is
x 2 ψ 2 .
If x, ψ and x · ψ all lived in the same space we would call this an orthogonal
multiplication. In fact, a simple weight calculation shows that, restricted to Ad :
SU (3) → Spin(8),
T ∼
= S+ ∼= S−
so we do get an orthogonal multiplication on the Lie algebra of SU (3), in fact it is:
i
A · B = ωAB − ω̄BA − √ tr(AB)1
3
√
where ω = (1 + i 3)/2.
Thus if M 8 has an adjoint SU (3) structure, as described above, the SU (3)-

invariant isomorphism S + ∼
= T defines a spin 3/2 field:
ψ ∈ C ∞ (M, S + ⊗ Λ1 ).
This ψ, it turns out, satisfies the Rarita-Schwinger equation (see [7])
Dψ = 0, d∗ ψ = 0
where for the first equation we think of ψ as a spinor with values in the bundle
Λ1 and apply the Dirac operator D and in the second we think of it as a one-form
with values in S + . Now take the second covariant derivative of ψ: this is a section
∇2 ψ ∈ C ∞ (S + ⊗ Λ1 ⊗ Λ1 ⊗ Λ1 ) or in more conventional terms
∇2 ψ = ψi;jk

Covariantly differentiate the equation Dψ = 0 to obtain j ej ψi;jk = 0 and then
contract to get

ej ψi;ji = 0.
i,j

Now differentiate d∗ ψ = 0 which gives i ψi;ij = 0 so that

ej ψi;ij = 0.
i,j
Subtract the two displayed formulae and re-express ψi;ji − ψi;ij using the curvature
tensor to get
Rij ei ψj = 0
i,j
which implies that 8 of the 36 components of the Ricci tensor vanish. These
manifolds are not quite Einstein manifolds, clearly, but yet have some common
features. Unfortunately the only known compact example is the group manifold
SU (3), though one can find nontrivial local examples.
2. Cohomogeneity one metrics

Variational characterizations can be useful in a practical way in situations
of symmetry but unfortunately compact manifolds with holonomy SU (n), G2 or
Spin(7) have no continuous symmetries because the Ricci tensor vanishes and so
any Killing vector is covariant constant. The formalism described above can still be
used, however, and we shall see that, as in [5], [3], it provides a practical approach
to deriving differential equations, if not solving them.
The first aspect of this involves duality of spaces of forms. If dβ ∈ Ωp is exact
and γ ∈ Ωn−p is closed then by Stokes’ theorem

dβ ∧ γ = 0.
M
This implies that the natural pairing of p-forms and (n − p)-forms gives a formal
isomorphism
(Ωpexact )∗ ∼ n−p
= Ωn−p /Ωclosed
and since d : Ωn−p → Ωn−p+1

exact has kernel the closed (n − p)-forms we formally have
(we are not concerned with distributions here)
(Ωpexact )∗ ∼ n−p+1
= Ωexact .
In the previous lecture we were restricting a functional to a cohomology class.

Now let’s consider the internal geometry of such a class
A = {α ∈ Ωp : dα = 0, [α] = a ∈ H p (M n , R)}.
This is an affine space modelled on Ωpexact , and so its tangent bundle is trivial and
TA =∼ A × Ωp
exact .
Using the duality above, its cotangent bundle is

T ∗A ∼= A × (Ω
p
)∗ = A × Ωn−p+1 .
exact exact
Bearing this in mind, let’s return to the 7-dimensional case, but now considering
4-forms instead of 3-forms. There are stable 4-forms and we can do the variational
approach of the last lecture and see G2 manifolds as critical points of a functional
on 4-forms in a given cohomology class, but the 4-form approach gives a bit more,
because we now have
(Ω4exact )∗ = Ω4exact
166 NIGEL HITCHIN
so A has a (flat indefinite) metric

(dη, dη) = η ∧ dη.
M
To make use of this, first restrict to the trivial cohomology class. Then we have
two functionals:
G2 volume
(1) Φ(dη) – the
(2) Q(dη) = M η ∧ dη – from the indefinite inner product.
We set up then a constrained variational problem – finding the critical points of Φ
subject to the constraint Q(dη) = 1. The first variation gives:

δΦ(dη̇) = ∗dη ∧ dη̇
M

δQ(dη̇) = 2 η̇ ∧ dη.
M
Using a Lagrange multiplier λ and Stokes’ theorem the equations become
d(∗ρ) = λρ
where ρ = dη.
A manifold with this structure is called a weak holonomy G2 manifold. It is

an Einstein manifold with positive scalar curvature and has a Killing spinor. As
such it defines an incomplete Spin(7) metric on the cone, but which also serves
as a model for asymptotically conical Spin(7) metrics. There are plenty of weak
holonomy G2 manifolds, since a 3-Sasakian manifold and its squashed deformation
provides an example and thanks to the work of Boyer, Galicki et al (see [2]) there
are infinitely many of these.
Example 2.1. We can find homogeneous examples by applying the variational

approach to invariant forms. For example, S 7 ⊂ H2 is acted on by Sp(2) × Sp(1)
(acting by the quaternionic matrix on the left and the quaternionic scalar on the
right). This makes S 7 → S 4 into a principal SU (2)-bundle with α1 , α2 , α3 the
components of a connection form (the 1-instanton bundle in fact) and ω1 , ω2 , ω3 the
components of the curvature form. The closed invariant 4-forms are then spanned
by dσ, dτ where
σ = α1 ∧ α2 ∧ α3 , τ = α1 ∧ ω1 + α2 ∧ ω2 + α3 ∧ ω3 .
This provides a two-dimensional vector space of invariant exact 4-forms xdσ + ydτ
on which we can define the two functionals and easily find the critical points, which
give the squashed 7-sphere.
In the case where the cohomology class A is non-trivial, Q defines an indefinite

metric and we take the gradient vector field X = dξ of Φ(ρ). Since

ξ ∧ dϕ = ∗ρ ∧ dϕ
M M
we have X = dξ = d ∗ ρ which yields the gradient flow equation:

∂ρ
= d ∗ ρ.
∂t
But dρ = 0, so
d(∗ρ ∧ dt + ρ) = 0
and then
∗ρ ∧ dt + ρ
satisfies precisely the algebraic condition to be the 4-form stabilized by Spin(7)
(not a “stable form” in the sense above.) Being closed, it defines a Riemannian
metric with holonomy Spin(7) on R × M 7 , or a subinterval thereof.
Example 2.2. The formalism above can be used to give cohomogeneity one
examples of non-compact Spin(7) manifolds. Again using S 7 as the 7-manifold with
symmetry group Sp(2) × Sp(1) one obtains the Bryant-Salamon Spin(7) manifold,
the first nontrivial complete example to be discovered. If we weaken the symmetry
to Sp(2) × U (1) then the invariant forms are spanned by
d(α1 ∧ α2 ∧ α3 ), d(α1 ∧ ω1 + α2 ∧ ω2 ), (α3 ∧ ω3 ).
With this, one can easily write the gradient flow which gives the system of ODE’s
solved by Cvetič et al [4]. Their solutions give in particular on M = R8 an analogue
of the Taub-NUT metric and one on the total space S + → S 4 of the spin bundle
over the sphere. They each have the form
(r + )2 dr 2 2 (r + 3)(r − ) 2 1 1
+ σ + (r + 3)(r − )(Dµi )2 + (r 2 − 2 )dΩ24
(r + 3)(r − ) (r + )2 4 2
for a different range of values.
If we move now to six dimensions then the duality of forms tells us that
(Ω3exact )∗ ∼
= Ω4exact .
Here we work with two objects – ρ ∈ Ω3 a stable form with stabilizer SL(3, C)
and ω 2 ∈ Ω4 with stabilizer Sp(6, R). Compatibility of these two forms leads to a
reduction of structure group to SU (3) and the compatibility conditions are:
ω ∧ ρ = 0, φ(ρ) = λω 3 .
The first says, from the complex point of view that ω is of type (1, 1), and from the
symplectic point of view that ρ is primitive. The second says that the norm of ρ is
constant.
We can consider here a constrained variational problem too, taking ρ = dα ∈ Ω3

an exact 3-form and
σ = ω 2 = dβ ∈ Ω4
an exact 4-form. The duality defines the bilinear pairing

B(ρ, ω 2 ) = dα ∧ β
M
and we have two functionals
(1) A(ρ, σ) = Φ(ρ) + Φ(σ) – the sum of volumes
168 NIGEL HITCHIN
(2) B(ρ, σ).

The critical points of A subject to B = 1 are given by the equations (after some
scale changes)
dρ̂ = ω 2 , dω = ρ.
Interestingly, the compatibility conditions to reduce to SU (3) follow from these
equations, for firstly
ω ∧ ρ = ω ∧ dω = d(ω 2 )/2 = 0.
It follows from this that ω is type (1, 1) so that ω ∧ ρ̂ = 0 too. Then
ω 3 = ω ∧ dρ̂ = d(ω ∧ ρ̂) − dω ∧ ρ̂ = ρ ∧ ρ̂.
The structure on M 6 defined by this pair of forms is called a weak holonomy SU (3)
structure or nearly Kähler structure. Here there are few known examples: S 6 , CP 3 ,
U (3)/T and S 3 × S 3 . The general class of manifolds of this type are Einstein with
positive scalar curvature and have Killing spinors.
To pursue the analogue of the gradient flow here we take two cohomology classes
A ∈ H 3 (M, R) and B ∈ H 4 (M, R). Then T (A × B) = A × B × Ω3exact × Ω4exact .
The pairing
B(ρ, ω 2 ) = dα ∧ β
M
defines an indefinite metric as before, but more usefully a symplectic structure
ω((ρ1 , σ1 ), (ρ2 , σ2 )) = ρ1 , σ2 − ρ2 , σ1
on A × B. We now take the function
H = Φ(ρ) − Φ(σ)
and derive the Hamiltonian flow equations which are:
∂ρ
= dω
∂t
∂σ ∂ω
= ω∧ = −dρ̂.
∂t ∂t
It happens that the compatibility conditions for SU (3) are preserved by the flow
and then
dt ∧ ω + ρ
defines a G2 metric on an interval of R × M 6 .
Example 2.3. In [3] the authors take M 6 = S 3 × S 3 and cohomology classes

in
H 3 (M ) ∼
= Z ⊕ Z, H 4 (M ) = 0.
Left-invariance of forms yields
ρ = n Σ1 Σ2 Σ3 − m σ1 σ2 σ3 + x1 (σ1 Σ2 Σ3 − σ2 σ3 Σ1 )
+ 2 cyclic terms
σ = y1 σ2 Σ2 σ3 Σ3 + y2 σ3 Σ3 σ1 Σ1 + y3 σ1 Σ1 σ2 Σ2
The equations are then:
y2 y3
ẋ1 =
y1
m n x1 + (m + n) x2 x3 + x1 (x22 + x23 − x21 )

ẏ1 = √
y1 y2 y3
Example 2.4. In [5] the authors take M 6 = S 3 × T 3 for which

H 3 (M ) ∼
= Z ⊕ Z, H 4 (M ) ∼= Z3 .
Left-invariance of forms gives:
ρ = n Σ1 Σ2 Σ3 − m σ1 σ2 σ3 + [x1 σ1 Σ2 Σ3
+ 2 cyclic terms]
σ = y1 σ2 Σ2 σ3 Σ3 + y2 σ3 Σ3 σ1 Σ1 + y3 σ1 Σ1 σ2 Σ2
The equations are then:
y2 y3
ẋ1 =
y1
m x2 x3
ẏ1 = √
y1 y2 y3
3. Generalized Calabi-Yau manifolds

We saw in the first Lecture that the Calabi-Yau condition could be derived
by a variational approach using 3-forms in six dimensions. In this lecture we shall
see how a variational approach using forms of arbitrary degree in six dimensions
produces a new differential geometric structure. It is perhaps easier to introduce
the structure first. This involves the Courant bracket on sections of T ⊕ Λp T ∗ . If
X + ξ, Y + η ∈ C ∞ (T ⊕ Λp T ∗ ) one defines
1
[X + ξ, Y + η] = [X, Y ] + LX η − LY ξ − d(iX η − iY ξ)
2
Unlike the Lie bracket on vector fields this has nontrivial automorphisms defined
by forms: if α ∈ Ωp+1 and dα = 0 then defining
A(X + ξ) = X + ξ + iX α
one obtains
A([X + ξ, Y + η]) = [A(X + ξ), A(Y + η)].
Example 3.1. In the case p = 0 we have X + f, Y + g ∈ C ∞ (T ) ⊕ C ∞ and one

finds
[X + f, Y + g] = [X, Y ] + Xg − Y f.
The automorphisms here are the closed 1-forms α ∈ Ω1 where
A(X + f ) = X + f + iX α
In fact this example has another interpretation as S 1 -invariant vector fields on
M × S 1 , for if
d
X +f
dθ
then the Lie bracket is equal to the Courant bracket. Moreover a gauge transfor-
mation g : M → S 1 defined by
(x, eiθ ) → (x, g(x)eiθ )
170 NIGEL HITCHIN
gives the automorphism of invariant vector fields

d 1 d
X +f → X + (f + iX g −1 dg)
dθ i dθ
so that α = g −1 dg/2πi is the closed 1-form defining it. Note however that α ∈ Ω1
in this case is closed with integral periods so that a general automorphism is a flat
connection and not just one gauge-equivalent to the trivial connection.
There is in fact a more general setting for this if we replace the product M × S 1
by a principal S 1 -bundle P over M . In that case we have an exact sequence of vector
bundles over M (the Atiyah sequence)
0 → 1 → T P/S 1 → T M → 0.
The sections of T P/S 1 are the invariant vector fields on P and so have their own
Lie bracket. A connection on P is a splitting of the exact sequence: T P/G ∼
= T ⊕ 1,
which therefore defines a bracket on sections of T ⊕ 1. which is
[X + f, Y + g] − 2F (X, Y )
where F is the curvature of the connection. Now we see that a non-flat connection
defines a deformation of the bracket by a closed 2-form with integral periods.
It is the case p = 1 which will interest us:

X + ξ, Y + η ∈ C ∞ (T ⊕ T ∗ ).
In this case the automorphisms are closed 2-forms
B ∈ Ω2 , dB = 0
which play a role very close to that of the B-field in string theory. Concretely, the
automorphism is
A(X + ξ) = X + ξ + iX B.
We can also twist the bracket by a closed 3-form H:
[X + ξ, Y + η] + 6iX iY H.
When H has integral periods it brings us close to the differential geometry of gerbes,
but this aspect will not be pursued here.
The vector bundle T ⊕ T ∗ has another structure defined on it apart from the
Courant bracket. The natural pairing between T and T ∗ defines an indefinite metric
on the bundle, with signature (n, n): for X + ξ ∈ T ⊕ T ∗ the inner product is
(X + ξ, X + ξ) = −X, ξ.
By naturality, any endomorphism of T preserves the inner product, but the full
orthogonal Lie algebra is
End T ⊕ Λ2 T ∗ ⊕ Λ2 T.
In particular a 2-form B, thought of as a map from T to T ∗ , lies in the Lie algebra.
The exponentiation of this action is
exp B(X + ξ) = X + ξ + iX B
which is just what we have seen in the context of the Courant bracket.
More interesting than the vector representation is the spin representation,

which has a concrete form in our situation, because of the decomposition T ⊕ T ∗
into maximal isotropic subspaces. We define the two spaces
S+ = Λev T ∗ ⊗ (Λn T )1/2
S− = Λod T ∗ ⊗ (Λn T )1/2
and the action of X + ξ by Clifford multiplication as:
(X + ξ) · ϕ = iX ϕ + ξ ∧ ϕ.
In this representation, the action of B ∈ Λ2 T ∗ is
1
exp B(ϕ) = (1 + B + B ∧ B + . . . ) ∧ ϕ.
2
We can also consider the exterior forms alone as spinors, removing the line bundle
twist by (Λn T )1/2 (like Spinc structures). The spin representations S + and S −
have an invariant symmetric or skew-symmetric form ϕ, ψ defined on them. On
exterior forms this takes values in the one-dimensional space Λn T ∗ . Concretely,
this is just the exterior product pairing, but with an alternating sign, sometimes
called in algebraic geometry the Mukai pairing.
Now we introduce a geometric structure compatible with this natural back-

ground. To motivate this recall one approach to the definition of a complex struc-
ture on a manifold of even dimension. An almost complex structure J : T → T ,
with J 2 = −1 has a +i eigenspace in T ⊗ C which is a subbundle E satisfying the
following properties:
(1) T ⊗ C = E ⊕ Ē
(2) the sections of E are closed under the Lie bracket.
The Newlander-Nirenberg Theorem tells us that any such subbundle actually de-
fines an integrable complex structure. We now make a more general definition:
Definition 3.2. A generalized complex structure on an even-dimensional
manifold is a subbundle E ⊂ (T ⊕ T ∗ ) ⊗ C such that
(1) (T ⊕ T ∗ ) ⊗ C = E ⊕ Ē
(2) sections of E are closed under the Courant bracket
(3) E is isotropic with respect to the natural inner product on T ⊕ T ∗ .
The final isotropy condition has two connotations. On the one hand it says that
the almost complex structure J : T ⊕ T ∗ → T ⊕ T ∗ is compatible with the indefinite
metric in the sense that the metric is pseudo-Hermitian. On the other, although it
is easy to let the words “closed under the Courant bracket” slip off the tongue, there
is a complication – the obstruction is only tensorial if E is isotropic. For example,
suppose that X + ξ, Y + η are sections of E and consider for some function f
1
[f (X + ξ), Y + η] = f [X + ξ, Y + η] + df ∧ (iX η + iY ξ) − (Y · f )(X + ξ).
2
Since
1
(iX η + iY ξ) = (X + ξ, Y + η)
2
if E is isotropic then iX η + iY ξ = 0 and the obstruction to being closed under the
bracket lies in
Λ2 E ∗ ⊗ E ∗ .
172 NIGEL HITCHIN
Without the isotropy condition, the embarrassing extra term df is involved.

A generalized complex structure is a reduction of the structure group of T ⊕ T ∗
from SO(2n, 2n) to U (n, n) together with an integrability condition. An ordinary
complex manifold is an example, where J = I on T and −I on T ∗ , but there are
more as we shall see.
Isotropic subspaces have a close connection with spinors. It is easy to see

that for any spinor ϕ the set of vectors X + ξ which annihilate ϕ is an isotropic
subspace, since Clifford multiplication has the characteristic property (X + ξ)2 =
−(X + ξ, X + ξ)1. A pure spinor ϕ ∈ S ± is defined by the condition
U = {X + ξ ∈ (T ⊕ T ∗ ) ⊗ C : (X + ξ) · ϕ = 0}
is maximally isotropic.
Example 3.3. We can build up examples from the simplest situation:
(1) take the pure spinor 1 ∈ Λ0 T ∗ ; its maximal isotropic subspace is U = T ⊂
T ⊕ T∗
(2) transform by B: one gets the pure spinor exp B(1) = 1 + B + 12 B 2 + . . . ;
its maximal isotropic subspace is U = exp B(T )
A generalized complex structure is thus defined in a neighbourhood of each point by
a complex pure spinor field, well defined up to scalar multiplication. In our set-up
such a spinor for T ⊕ T ∗ is a form of mixed degree on the manifold. In general there
is a complex line bundle, analogous to the canonical bundle, for which these locally
defined forms are local trivializations, but there is a situation where the pure spinor
is global:
Definition 3.4. A generalized Calabi-Yau manifold is an even-dimensional
manifold M with a closed form ρ ∈ Ωev/od ⊗ C which is a pure spinor and satisfies
ρ, ρ̄ = 0 at each point.
In the definition, the pure spinor ρ defines a maximal isotropic subbundle E ⊂
(T ⊕ T ∗ ) ⊗ C and the condition ρ, ρ̄ = 0 says that E ⊕ Ē = (T ⊕ T ∗ ) ⊗ C. One
can also prove the following:
Lemma 3.5. The condition dρ = 0 for a pure spinor ρ implies that sections of
the isotropic subbundle E ⊂ (T ⊕ T ∗ ) ⊗ C are closed under Courant bracket.
We see then that a generalized Calabi-Yau manifold is a particular case of a
generalized complex manifold. In practical terms it is often easier to check that a
form is closed rather than that a subbundle is closed under Courant bracket.
Surprisingly both complex and symplectic structures appear here, as in the

following examples:
(1) a Calabi-Yau manifold with non-vanishing holomorphic n-form ρ
(2) a symplectic manifold (M, ω) with ρ = exp iω
(3) the B-field transform of a symplectic manifold
ρ = exp B exp iω = exp(B + iω)
Example 3.6. Here is the situation in dimension 2, in the compact case:

(1) Odd type: ρ ∈ Ω1 (M ), dρ = 0, ρ ∧ ρ̄ = 0. There is no purity condition

here: the multiples of ρ define the (1, 0)-forms of a complex structure, and
then ρ is a non-vanishing holomorphic 1-form, so M 2 is an elliptic curve.
(2) Even type: here ρ = c + β (c constant, β a 2-form) such that
c + β, c̄ + β̄ = cβ̄ − c̄β = 0
This gives c = 0 and Im (β/c) = 0 implies that β/c = B + iω where ω is
a symplectic form, so
ρ = c exp(B + iω)
and M 2 is the B-field transform of a symplectic manifold.
In dimension 4, the purity condition is easy to state because of triality for SO(4, 4).
As we noted in Lecture 1, the vector representation and the two spin representa-
tions in eight dimensions are related by an outer automorphism. In the present
context this means that the internal structure of S + and S − is simply that of an
8-dimensional vector space with an inner product – the spin quadratic form ϕ, ϕ
– and a pure spinor is simply one that is null with respect to this.
Example 3.7. Here then is the situation in dimension 4 in the compact case:
(1) Odd type: ρ = β + γ ∈ Ω1 ⊕ Ω3 , where β, γ are closed and
β ∧ γ = 0, β ∧ γ̄ + β̄ ∧ γ = 0.
This implies that γ = β ∧ ν and β defines a foliation by surfaces with a
transverse holomorphic 1-form, while Im (ν) defines a symplectic form on
the leaves. An example is M 4 = M12 × M22 , where M1 is an elliptic curve,
and M2 a symplectic surface.
(2) Even type: ρ = c + β + γ ∈ Ω0 ⊕ Ω2 ⊕ Ω4 where the null (purity) condition
is
β 2 = 2cγ
If c = 0, then
1
ρ = c exp β = c + β + β 2
2c
and as before, this is the B-field transform of a symplectic structure, but
if c = 0, β 2 = 0 and is a closed, locally decomposable complex 2-form. It
defines the structure of a complex surface with nonvanishing holomorphic
2-form on M , which thus must be a torus or K3 surface.
Complex structures are not the only ones one can generalize in this way. Here
are some more obvious ones in 4 dimensions, determined by the number of spinors
fixed by a subgroup of Spin(4, 4):
(1) SU (2, 2) → SO(4, 2): generalized Calabi-Yau structure
(2) Sp(1, 1) → SO(4, 1): generalized hyperkähler structure
(3) Sp(1) × Sp(1) → SO(4): generalized hyperkähler metric
The last example applied to a K3 surface gives a geometrical structure whose moduli
space is a space of N = (4, 4) conformal field theories [9].
The case of dimension 6 brings us back to the idea of open orbits of forms and
invariant functionals, which is where we started. In fact, as described in [8], this is
the context in which this author found these objects.
174 NIGEL HITCHIN
Note that in general the dimension of the space of complex pure spinors for SO
(2m, 2m) is the dimension of C exp B which is 1 + m(2m − 1). When m = 3, the
real dimension is therefore
2 × 16 = 32 = dim S + = dim S − .
This numerical coincidence manifests itself in the more geometrical statement:
Lemma 3.8. For the spin representations of Spin(6, 6) in S ± there is an in-
variant open set of real spinors ϕ such that ϕ = ρ + ρ̄ where ρ is a pure spinor with
ρ, ρ̄ = 0.
As a consequence of this, ρ, ρ̄ defines SO(6, 6)-invariant maps
φ : Λev/od V ∗ → Λ6 V ∗
and on a compact manifold M 6 we get a functional

Φ(ϕ) = φ(ϕ)
M
defined on even forms or odd forms, invariant under diffeomorphisms and, this time,
the action of the B-field ϕ → exp Bϕ.
Example 3.9. If ρ = ρ0 + ρ2 + ρ4 + ρ6 , then

ρ, ρ̄ = ρ0 ρ̄6 − ρ̄0 ρ6 − ρ2 ρ̄4 + ρ̄2 ρ4 .
Take ϕ = 1 − ω /2, where ω is a symplectic form, then
2
1 1
ρ = exp(iω) = 1 + iω − ω 2 − i ω 3
2 6
and
4
φ(ϕ) = iω 3 .
3
So the functional Φ is essentially the Liouville volume of the symplectic form.
We now address our version of nonlinear Hodge theory and ask: what is a
critical point of the functional

Φ(ϕ) = φ(ϕ)
M
for ϕ a closed form in a fixed cohomology class in H ev/od (M, R)? Just as in Lecture
1, the critical points are forms where ϕ = ρ + ρ̄ and
dρ = 0
which means that the critical points are generalized Calabi-Yau manifolds.
In this situation, there is a non-degeneracy condition for the critical points

which is both a generalization of the ∂ ∂¯ lemma and of strong Lefschetz and implies
the existence of a unique nearby critical point in each nearby cohomology class,
so we deduce that the moduli space is an open set in H ev/od (M, R). The set-up
here however, is different from what we encountered before because we consider
structures modulo diffeomorphism and B-fields. In fact we have to take diffeomor-
phisms homotopic to the identity (which is normal as in Teichmüller theory) and
also cohomologically trivial B-fields – exact 2-forms.
To summarize, all the special geometries encountered above are in some way
associated to open orbits of groups. Such actions have in fact been classified and
in truth there are not many more, but to finish let me just point out that a real
form of the exceptional group E7 in its 56-dimensional representation shares many
of the properties of the groups considered above, but I have no idea what sort of
special geometry this generates or even in what dimension.
References
1. C. Bär, Real Killing spinors and holonomy, Comm. Math. Phys. 154 (1993), 509–521.
2. C. Boyer and K. Galicki, 3-Sasakian manifolds, Surveys in differential geometry: essays on
Einstein manifolds, Surv. Differ. Geom., VI, Int. Press, Boston, MA, 1999 pp. 123–184.
3. Z.W. Chong, M. Cvetič, G.W. Gibbons, H. Lü, C.N. Pope, P. Wagner, General Metrics of
G2 Holonomy and Contraction Limits, Nuclear Phys. B 620 (2002), 3–28.
4. M. Cvetič, G, Gibbons, H. Lü, C.N. Pope, New complete noncompact Spin(7) manifolds,
Nuclear Phys. B 620 (2002), 29–54
5. S. Gukov, S.-T. Yau, E. Zaslow, Duality and fibrations on G2 manifolds, Turkish J. Math.
27 (2003) 61-97
6. N. J. Hitchin, The geometry of three-forms in six dimensions., J. Differential Geom. 55
(2000), 547–576.
7. N. J. Hitchin, Stable forms and special metrics., Global differential geometry: the math-
ematical legacy of Alfred Gray (Bilbao, 2000), Contemp. Math., 288, Amer. Math. Soc.,
Providence, RI, 2001, pp. 70–89.
8. N. J. Hitchin, Generalized Calabi-Yau manifolds, Quarterly J. Math. 54 (2003) 281-308.
9. W. Nahm and K. Wendland A hiker’s guide to K3: aspects of N = (4, 4) superconformal
field theory with central charge c = 6, Comm. Math. Phys. 216 (2001) 85–138.
Mathematical Institute, 24-29 St Giles, Oxford, OX1 3LB, UK

E-mail address: hitchin@maths.ox.ac.uk
Volume 3, 2004
Constructing compact manifolds with exceptional holonomy
Dominic Joyce
Abstract. The exceptional cases in Berger’s classification of Riemannian ho-

lonomy groups are G2 in 7 dimensions, and Spin(7) in 8 dimensions. Metrics
with these holonomy groups are Ricci-flat, and have many interesting geomet-
rical properties. We survey the known constructions of examples of compact
7- and 8-manifolds with holonomy G2 and Spin(7).
We first construct a compact, singular Riemannian orbifold X with holo-
nomy a proper subgroup of G2 or Spin(7), such as X = T 7 /Γ for certain finite
groups Γ. Then we resolve the singularities of X using ideas from Calabi–
Yau geometry and analytic techniques, to get a compact, nonsingular 7- or
8-manifold M with holonomy G2 or Spin(7).
1. Introduction
In the theory of Riemannian holonomy groups, perhaps the most mysterious

are the two exceptional cases, the holonomy group G2 in 7 dimensions and the
holonomy group Spin(7) in 8 dimensions. This is a survey paper on the construction
of examples of compact 7- and 8-manifolds with holonomy G2 and Spin(7).
All of the material described can be found in the author’s book [9]. Some, but
not all, is also in the papers [5, 6, 7, 8, 10, 11]. In particular, the most compli-
cated and powerful form of the construction of compact manifolds with exceptional
holonomy by resolving orbifolds T n /Γ, and many of the examples, are given only
in [9] and not in any published paper.
The rest of this section introduces the holonomy groups G2 , Spin(7) and SU(m),
and the relations between them. Section 2 discusses constructions for compact 7-
manifolds with holonomy G2 . Most of the section explains how to do this by
resolving the singularities of orbifolds T 7 /Γ, but in § 2.5 we briefly discuss two
other methods starting from Calabi–Yau 3-folds.
Section 3 explains constructions of compact 8-manifolds with holonomy Spin(7).
One way to do this is to resolve orbifolds T 8 /Γ, but as this is very similar in outline
to the G2 material of § 2 we say little about it. Instead we describe a second
construction which begins with a Calabi–Yau 4-orbifold.
2000 Mathematics Subject Classification. 53C29.
2004
177
178 DOMINIC JOYCE
1.1. Riemannian holonomy groups.

Let M be a connected n-dimensional manifold, g a Riemannian metric on M ,
and ∇ the Levi-Civita connection of g. Let x, y be points in M joined by a smooth
path γ. Then parallel transport along γ using ∇ defines an isometry between the
tangent spaces Tx M , Ty M at x and y.
Definition 1.1. The holonomy group Hol(g) of g is the group of isometries of
Tx M generated by parallel transport around piecewise-smooth closed loops based
at x in M . We consider Hol(g) to be a subgroup of O(n), defined up to conjugation
by elements of O(n). Then Hol(g) is independent of the base point x in M .
The classification of holonomy groups was achieved by Berger [1] in 1955.
Theorem 1.2. Let M be a simply-connected, n-dimensional manifold, and g
an irreducible, nonsymmetric Riemannian metric on M . Then either
(i) Hol(g) = SO(n),
(ii) n = 2m and Hol(g) = SU(m) or U(m),
(iii) n = 4m and Hol(g) = Sp(m) or Sp(m) Sp(1),
(iv) n = 7 and Hol(g) = G2 , or
(v) n = 8 and Hol(g) = Spin(7).
Now G2 and Spin(7) are the exceptional cases in this classification, so they are
called the exceptional holonomy groups. For some time after Berger’s classification,
the exceptional holonomy groups remained a mystery. In 1987, Bryant [2] used the
theory of exterior differential systems to show that locally there exist many metrics
with these holonomy groups, and gave some explicit, incomplete examples. Then
in 1989, Bryant and Salamon [3] found explicit, complete metrics with holonomy
G2 and Spin(7) on noncompact manifolds.
In 1994-5 the author constructed the first examples of metrics with holonomy
G2 and Spin(7) on compact manifolds [5, 6, 7]. These, and the more complicated
constructions developed later by the author [8, 9] and by Kovalev [12], are the
subject of this article.
1.2. The holonomy group G2 .
Let (x1 , . . . , x7 ) be coordinates on R7 . Write dxij...l for the exterior form
dxi ∧ dxj ∧ · · · ∧ dxl on R7 . Define a metric g0 , a 3-form ϕ0 and a 4-form ∗ϕ0 on
R7 by g0 = dx21 + · · · + dx27 ,
ϕ0 = dx123 + dx145 + dx167 + dx246 − dx257 − dx347 − dx356 and
(1)
∗ϕ0 = dx4567 + dx2367 + dx2345 + dx1357 − dx1346 − dx1256 − dx1247 .
The subgroup of GL(7, R) preserving ϕ0 is the exceptional Lie group G2 . It also
preserves g0 , ∗ϕ0 and the orientation on R7 . It is a compact, semisimple, 14-
dimensional Lie group, a subgroup of SO(7).
A G2 -structure on a 7-manifold M is a principal subbundle of the frame bundle
of M , with structure group G2 . Each G2 -structure gives rise to a 3-form ϕ and a
metric g on M , such that every tangent space of M admits an isomorphism with
R7 identifying ϕ and g with ϕ0 and g0 respectively. By an abuse of notation, we
will refer to (ϕ, g) as a G2 -structure.
Proposition 1.3. Let M be a 7-manifold and (ϕ, g) a G2 -structure on M .
Then the following are equivalent:
CONSTRUCTING COMPACT MANIFOLDS WITH EXCEPTIONAL HOLONOMY 179
(i) Hol(g) ⊆ G2 , and ϕ is the induced 3-form,

(ii) ∇ϕ = 0 on M , where ∇ is the Levi-Civita connection of g, and
(iii) dϕ = d∗ ϕ = 0 on M .
We call ∇ϕ the torsion of the G2 -structure (ϕ, g), and when ∇ϕ = 0 the
G2 -structure is torsion-free. A triple (M, ϕ, g) is called a G2 -manifold if M is a 7-
manifold and (ϕ, g) a torsion-free G2 -structure on M . If g has holonomy Hol(g) ⊆
G2 , then g is Ricci-flat.
Theorem 1.4. Let M be a compact 7-manifold, and suppose that (ϕ, g) is
a torsion-free G2 -structure on M . Then Hol(g) = G2 if and only if π1 (M ) is
finite. In this case the moduli space of metrics with holonomy G2 on M , up to
diffeomorphisms isotopic to the identity, is a smooth manifold of dimension b3 (M ).
1.3. The holonomy group Spin(7).
Let R8 have coordinates (x1 , . . . , x8 ). Define a 4-form Ω0 on R8 by
Ω0 = dx1234 + dx1256 + dx1278 + dx1357 − dx1368 − dx1458 − dx1467
(2)
−dx2358 − dx2367 − dx2457 + dx2468 + dx3456 + dx3478 + dx5678 .
The subgroup of GL(8, R) preserving Ω0 is the holonomy group Spin(7). It also
preserves the orientation on R8 and the Euclidean metric g0 = dx21 + · · · + dx28 . It
is a compact, semisimple, 21-dimensional Lie group, a subgroup of SO(8).
A Spin(7)-structure on an 8-manifold M gives rise to a 4-form Ω and a metric
g on M , such that each tangent space of M admits an isomorphism with R8 iden-
tifying Ω and g with Ω0 and g0 respectively. By an abuse of notation we will refer
to the pair (Ω, g) as a Spin(7)-structure.
Proposition 1.5. Let M be an 8-manifold and (Ω, g) a Spin(7)-structure on
M . Then the following are equivalent:
(i) Hol(g) ⊆ Spin(7), and Ω is the induced 4-form,
(ii) ∇Ω = 0 on M , where ∇ is the Levi-Civita connection of g, and
(iii) dΩ = 0 on M .
We call ∇Ω the torsion of the Spin(7)-structure (Ω, g), and (Ω, g) torsion-free
if ∇Ω = 0. A triple (M, Ω, g) is called a Spin(7)-manifold if M is an 8-manifold and
(Ω, g) a torsion-free Spin(7)-structure on M . If g has holonomy Hol(g) ⊆ Spin(7),
then g is Ricci-flat.
Here is a result on compact 8-manifolds with holonomy Spin(7).
Theorem 1.6. Let (M, Ω, g) be a compact Spin(7)-manifold. Then Hol(g) =
Spin(7) if and only if M is simply-connected, and b3 (M ) + b4+ (M ) = b2 (M ) +
2b4− (M ) + 25. In this case the moduli space of metrics with holonomy Spin(7)
on M , up to diffeomorphisms isotopic to the identity, is a smooth manifold of
dimension 1 + b4− (M ).
1.4. The holonomy groups SU(m).
Let Cm ∼ = R2m have complex coordinates (z1 , . . . , zm ), and define the metric
g0 , Kähler form ω0 and complex volume form θ0 on Cm by
i
g0 = |dz1 |2 + · · · + |dzm |2 , ω0 = (dz1 ∧ dz̄1 + · · · + dzm ∧ dz̄m ),
(3) 2
and θ0 = dz1 ∧ · · · ∧ dzm .
180 DOMINIC JOYCE
The subgroup of GL(2m, R) preserving g0 , ω0 and θ0 is the special unitary group

SU(m). Manifolds with holonomy SU(m) are called Calabi–Yau manifolds.
Calabi–Yau manifolds are automatically Ricci-flat and Kähler, with trivial
canonical bundle. Conversely, any Ricci-flat Kähler manifold (M, J, g) with trivial
canonical bundle has Hol(g) ⊆ SU(m). By Yau’s proof of the Calabi conjecture
[16], we have:
Theorem 1.7. Let (M, J) be a compact complex m-manifold admitting Kähler
metrics, with trivial canonical bundle. Then there is a unique Ricci-flat Kähler
metric g in each Kähler class on M , and Hol(g) ⊆ SU(m).
Using this and complex algebraic geometry one can construct many examples of
compact Calabi–Yau manifolds. The theorem also applies in the orbifold category,
yielding examples of Calabi–Yau orbifolds.
1.5. Relations between G2 , Spin(7) and SU(m).
Here are the inclusions between the holonomy groups SU(m), G2 and Spin(7):
SU(2) −−−−→ SU(3) −−−−→ G2
  
  

SU(2) × SU(2) −−−−→ SU(4) −−−−→ Spin(7).
We shall illustrate what we mean by this using the inclusion SU(3) → G2 . As
SU(3) acts on C3 , it also acts on R ⊕ C3 ∼ = R7 , taking the SU(3)-action on R to be
trivial. Thus we embed SU(3) as a subgroup of GL(7, R). It turns out that SU(3)
is a subgroup of the subgroup G2 of GL(7, R) defined in § 1.2.
Here is a way to see this in terms of differential
forms. Identify R ⊕ C3 with

R7 in the obvious way in coordinates, so that x1 , (x2 + ix3 , x4 + ix5 , x6 + ix7 ) in
R ⊕ C3 is identified with (x1 , . . . , x7 ) in R7 . Then ϕ0 = dx1 ∧ ω0 + Re θ0 , where ϕ0
is defined in (1) and ω0 , θ0 in (3). Since SU(3) preserves ω0 and θ0 , the action of
SU(3) on R7 preserves ϕ0 , and so SU(3) ⊂ G2 .
It follows that if (M, J, h) is a Calabi–Yau 3-fold, then R × M and S 1 × M have
torsion-free G2 -structures, that is, are G2 -manifolds.
Proposition 1.8. Let (M, J, h) be a Calabi–Yau 3-fold, with Kähler form ω
and complex volume form θ. Let x be a coordinate on R or S 1 . Define a metric
g = dx2 + h and a 3-form ϕ = dx ∧ ω + Re θ on R × M or S 1 × M . Then (g, ϕ) is
a torsion-free G2 -structure on R × M or S 1 × M , and ∗ϕ = 12 ω ∧ ω − dx ∧ Im θ.
Similarly, the inclusions SU(2) → G2 and SU(4) → Spin(7) give:
Proposition 1.9. Let (M, J, h) be a Calabi–Yau 2-fold, with Kähler form ω
and complex volume form θ. Let (x1 , x2 , x3 ) be coordinates on R3 or T 3 . Define a
metric g = dx21 + dx22 + dx23 + h and a 3-form
(4) ϕ = dx1 ∧ dx2 ∧ dx3 + dx1 ∧ ω + dx2 ∧ Re θ − dx3 ∧ Im θ
on R3 × M or T 3 × M . Then (ϕ, g) is a torsion-free G2 -structure on R3 × M or
T 3 × M , and
(5) ∗ϕ = 12 ω ∧ ω + dx2 ∧ dx3 ∧ ω − dx1 ∧ dx3 ∧ Re θ − dx1 ∧ dx2 ∧ Im θ.
Proposition 1.10. Let (M, J, g) be a Calabi–Yau 4-fold, with Kähler form ω

and complex volume form θ. Define a 4-form Ω on M by Ω = 12 ω ∧ ω + Re θ. Then
(Ω, g) is a torsion-free Spin(7)-structure on M .
2. Constructing G2 -manifolds from orbifolds T 7 /Γ
We now explain the method used in [5, 6] and [9, §11–§12] to construct ex-
amples of compact 7-manifolds with holonomy G2 . It is based on the Kummer
construction for Calabi–Yau metrics on the K3 surface, and may be divided into
four steps.
Step 1. Let T 7 be the 7-torus and (ϕ0 , g0 ) a flat G2 -structure on T 7 . Choose a

finite group Γ of isometries of T 7 preserving (ϕ0 , g0 ). Then the quotient
T 7 /Γ is a singular, compact 7-manifold, an orbifold.
Step 2. For certain special groups Γ there is a method to resolve the singularities
of T 7 /Γ in a natural way, using complex geometry. We get a nonsingular,
compact 7-manifold M , together with a map π : M → T 7 /Γ, the resolving
map.
Step 3. On M , we explicitly write down a 1-parameter family of G2 -structures
(ϕt , gt ) depending on t ∈ (0, ). They are not torsion-free, but have small
torsion when t is small. As t → 0, the G2 -structure (ϕt , gt ) converges to
the singular G2 -structure π ∗ (ϕ0 , g0 ).
Step 4. We prove using analysis that for sufficiently small t, the G2 -structure (ϕt , gt )
on M , with small torsion, can be deformed to a G2 -structure (ϕ̃t , g̃t ), with
zero torsion. Finally, we show that g̃t is a metric with holonomy G2 on the
compact 7-manifold M .
We will now explain each step in greater detail.
2.1. Step 1: Choosing an orbifold.
Let (ϕ0 , g0 ) be the Euclidean G2 -structure on R7 defined in § 1.2. Suppose
Λ is a lattice in R7 , that is, a discrete additive subgroup isomorphic to Z7 . Then
R7 /Λ is the torus T 7 , and (ϕ0 , g0 ) pushes down to a torsion-free G2 -structure on
T 7 . We must choose a finite group Γ acting on T 7 preserving (ϕ0 , g0 ). That is, the
elements of Γ are the push-forwards to T 7 /Λ of affine transformations of R7 which
fix (ϕ0 , g0 ), and take Λ to itself under conjugation.
Here is an example of a suitable group Γ, taken from [9, §12.2].
Example 2.1. Let (x1 , . . . , x7 ) be coordinates on T 7 = R7 /Z7 , where xi ∈ R/Z.
Let (ϕ0 , g0 ) be the flat G2 -structure on T 7 defined by (1). Let α, β and γ be the
involutions of T 7 defined by
(6) α : (x1 , . . . , x7 )
−→ (x1 , x2 , x3 , −x4 , −x5 , −x6 , −x7 ),
(7) β : (x1 , . . . , x7 )
−→ (x1 , −x2 , −x3 , x4 , x5 , 12 − x6 , −x7 ),

(8) γ : (x1 , . . . , x7 )
−→ −x1 , x2 , −x3 , x4 , 12 − x5 , x6 , 12 − x7 ).
By inspection, α, β and γ preserve (ϕ0 , g0 ), because of the careful choice of exactly

which signs to change. Also, α2 = β 2 = γ 2 = 1, and α, β and γ commute. Thus
they generate a group Γ = α, β, γ ∼= Z32 of isometries of T 7 preserving the flat
G2 -structure (ϕ0 , g0 ).
182 DOMINIC JOYCE
Having chosen a lattice Λ and finite group Γ, the quotient T 7 /Γ is an orbifold,

a singular manifold with only quotient singularities. The singularities of T 7 /Γ
come from the fixed points of non-identity elements of Γ. We now describe the
singularities in our example.
Lemma 2.2. In Example 2.1, βγ, γα, αβ and αβγ have no fixed points on T 7 .
The fixed points of α, β, γ are each 16 copies of T 3 . The singular set S of T 7 /Γ is
a disjoint union of 12 copies of T 3 , 4 copies from each of α, β, γ. Each component
of S is a singularity modelled on that of T 3 × C2 /{±1}.
The most important consideration in choosing Γ is that we should be able to
resolve the singularities of T 7 /Γ within holonomy G2 . We will explain how to do
this next.
2.2. Step 2: Resolving the singularities.
Our goal is to resolve the singular set S of T 7 /Γ to get a compact 7-manifold
M with holonomy G2 . How can we do this? In general we cannot, because we have
no idea of how to resolve general orbifold singularities with holonomy G2 . However,
suppose we can arrange that every connected component of S is locally isomorphic
to either
(a) T 3 × C2 /G, for G a finite subgroup of SU(2), or
(b) S 1 × C3 /G, for G a finite subgroup of SU(3) acting freely on C3 \ {0}.
One can use complex algebraic geometry to find a crepant resolution X of C2 /G
or Y of C3 /G. Then T 3 × X or S 1 × Y gives a local model for how to resolve the
corresponding component of S in T 7 /Γ. Thus we construct a nonsingular, compact
7-manifold M by using the patches T 3 × X or S 1 × Y to repair the singularities
of T 7 /Γ. In the case of Example 2.1, this means gluing 12 copies of T 3 × X into
T 7 /Γ, where X is the blow-up of C2 /{±1} at its singular point.
Now the point of using crepant resolutions is this. In both case (a) and (b), there
exists a Calabi–Yau metric on X or Y which is asymptotic to the flat Euclidean
metric on C2 /G or C3 /G. Such metrics are called Asymptotically Locally Euclidean
(ALE). In case (a), the ALE Calabi–Yau metrics were classified by Kronheimer
[13, 14], and exist for all finite G ⊂ SU(2). In case (b), crepant resolutions of
C3 /G exist for all finite G ⊂ SU(3) by Roan [15], and the author [10], [9, §8]
proved that they carry ALE Calabi–Yau metrics, using a noncompact version of
the Calabi Conjecture.
By Propositions 1.8 and 1.9, we can use the Calabi–Yau metrics on X or Y
to construct a torsion-free G2 -structure on T 3 × X or S 1 × Y . This gives a local
model for how to resolve the singularity T 3 × C2 /G or S 1 × C3 /G with holonomy
G2 . So, this method gives not only a way to smooth out the singularities of T 7 /Γ as
a manifold, but also a family of torsion-free G2 -structures on the resolution which
shows how to smooth out the singularities of the G2 -structure.
The requirement above that S be divided into connected components of the
form (a) and (b) is in fact unnecessarily restrictive. There is a more complicated
and powerful method, described in [9, §11–§12], for resolving singularities of a more
general kind. We require only that the singularities should locally be of the form
R3 × C2 /G or R × C3 /G, for G a finite subgroup of SU(2) or SU(3), and when
G ⊂ SU(3) we do not require that G act freely on C3 \ {0}.
If X is a crepant resolution of C3 /G, where G does not act freely on C3 \ {0},
then the author shows [9, §9], [11] that X carries a family of Calabi–Yau metrics
satisfying a complicated asymptotic condition at infinity, called Quasi-ALE metrics.

These yield the local models necessary to resolve singularities locally of the form R×
C3 /G with holonomy G2 . Using this method we can resolve many orbifolds T 7 /Γ,
and prove the existence of large numbers of compact 7-manifolds with holonomy G2 .
2.3. Step 3: Finding G2 -structures with small torsion.
For each resolution X of C2 /G in case (a), and Y of C3 /G in case (b) above,
we can find a 1-parameter family {ht : t > 0} of metrics with the properties
(a) ht is a Kähler metric on X with Hol(ht ) = SU(2). Itsinjectivity
radius sat-
isfies δ(ht ) = O(t), its Riemann curvature satisfies R(ht )C 0 = O(t−2 ),
and ht = h + O(t4 r −4 ) for large r, where h is the Euclidean metric on
C2 /G, and r the distance from the origin.
(b) ht is Kähler on Y with Hol(ht ) = SU(3), where δ(ht ) = O(t), R(ht )C 0 =
O(t−2 ), and ht = h + O(t6 r −6 ) for large r.
In fact we can choose ht to be isometric to t2 h1 , and the properties above are easy
to prove.
Suppose one of the components of the singular set S of T 7 /Γ is locally modelled
on T 3 × C2 /G. Then T 3 has a natural flat metric hT 3 . Let X be the crepant
resolution of C2 /G and let {ht : t > 0} satisfy property (a). Then Proposition 1.9
gives a 1-parameter family of torsion-free G2 -structures (ϕ̂t , ĝt ) on T 3 × X with
ĝt = hT 3 + ht . Similarly, if a component of S is modelled on S 1 × C3 /G, using
Proposition 1.8 we get a family of torsion-free G2 -structures (ϕ̂t , ĝt ) on S 1 × Y .
The idea is to make a G2 -structure (ϕt , gt ) on M by gluing together the torsion-
free G2 -structures (ϕ̂t , ĝt ) on the patches T 3 × X and S 1 × Y , and (ϕ0 , g0 ) on T 7 /Γ.
The gluing is done using a partition of unity. Naturally, the first derivative of the
partition of unity introduces ‘errors’, so that (ϕt , gt ) is not torsion-free. The size of
the torsion ∇ϕt depends on the difference ϕ̂t − ϕ0 in the region where the partition
of unity changes. On the patches T 3 × X, since ht − h = O(t4 r −4 ) and the partition
of unity has nonzero derivative when r = O(1), we find that ∇ϕt = O(t4 ). Similarly
∇ϕt = O(t6 ) on the patches S 1 × Y , and so ∇ϕt = O(t4 ) on M .
For small t, the dominant contributions to the injectivity radius δ(gt ) and
Riemann curvature R(gt )are made by those of the metrics ht on X and Y , so we
expect δ(gt ) = O(t) and R(gt )C 0 = O(t−2 ) by properties (a) and (b) above. In
this way we prove the following result [9, Th. 11.5.7], which gives the estimates on
(ϕt , gt ) that we need.
Theorem 2.3. On the compact 7-manifold M described above, and on many

other 7-manifolds constructed in a similar fashion, one can write down the following
data explicitly in coordinates:
• Positive constants A1 , A2 , A3 and ,
• A G2 -structure (ϕt , gt ) on M with dϕt = 0 for each t ∈ (0, ), and
• A 3-form ψt on M with d∗ ψt = d∗ ϕt for each t ∈ (0, ).
These satisfy three conditions:
(i) ψt L2 A1 t4 , ψt C 0 A1 t3 and d∗ ψt L14 A1 t16/7 ,
(ii) the injectivity radius δ(gt ) satisfies δ(gt ) A2 t,
(iii) the Riemann curvature R(gt ) of gt satisfies R(gt )C 0 A3 t−2 .
Here the operator d∗ and the norms . L2 , . L14 and . C 0 depend on gt .
184 DOMINIC JOYCE
One should regard ψt as a first integral of the torsion ∇ϕt of (ϕt , gt ). Thus the
norms ψt L2 , ψt C 0 and d∗ ψt L14 are measures of ∇ϕt . So parts (i)–(iii) say
that the torsion ∇ϕt must be small compared to the injectivity radius and Riemann
curvature of (M, gt ).
2.4. Step 4: Deforming to a torsion-free G2 -structure.
We prove the following analysis result.
Theorem 2.4. Let A1 , A2 , A3 be positive constants. Then there exist positive
constants κ, K such that whenever 0 < t κ, the following is true.
Let M be a compact 7-manifold, and (ϕ, g) a G2 -structure on M with dϕ = 0.
Suppose ψ is a smooth 3-form on M with d∗ ψ = d∗ ϕ, and
(i) ψ L2 A1 t4 , ψ C 0 A1 t1/2 and d∗ ψ L14 A1 ,
(ii) the injectivity radius δ(g) satisfies δ(g) A2t, and
(iii) the Riemann curvature R(g) satisfies R(g)C 0 A3 t−2 .
Then there exists a smooth, torsion-free G2 -structure (ϕ̃, g̃) on M with ϕ̃−ϕ C 0
Kt1/2 .
Basically, this result says that if (ϕ, g) is a G2 -structure on M , and the torsion
∇ϕ is sufficiently small, then we can deform to a nearby G2 -structure (ϕ̃, g̃) that is
torsion-free. Here is a sketch of the proof of Theorem 2.4, ignoring several technical
points. The proof is that given in [9, §11.6–§11.8], which is an improved version of
the proof in [5].
We have a 3-form ϕ with dϕ = 0 and d∗ ϕ = d∗ ψ for small ψ, and we wish to
construct a nearby 3-form ϕ̃ with dϕ̃ = 0 and d∗ ϕ̃ = 0. Set ϕ̃ = ϕ + dη, where η is
a small 2-form. Then η must satisfy a nonlinear p.d.e., which we write as
(9) d∗ dη = −d∗ ψ + d∗ F (dη),

where F is nonlinear, satisfying F (dη) = O |dη|2 .
We solve (9) by iteration, introducing a sequence (ηj )∞
j=0 with η0 = 0, satisfying
the inductive equations
(10) d∗ dηj+1 = −d∗ ψ + d∗ F (dηj ), d∗ ηj+1 = 0.
If such a sequence exists and converges to η, then taking the limit in (10) shows
that η satisfies (9), giving us the solution we want.
The key to proving this is an inductive estimate on the sequence (ηj )∞
j=0 . The
inductive estimate we use has three ingredients, the equations
(11) dηj+1 L2 ψ L2 + C1 dηj L2 dηj C 0 ,

(12) ∇dηj+1 L14 C2 d∗ ψ L14 + ∇dηj L14 dηj C 0 + t−4 dηj+1 L2 ,

(13) dηj C 0 C3 t1/2 ∇dηj L14 + t−7/2 dηj L2 .
Here C1 , C2 , C3 are positive constants independent of t. Equation (11) is obtained
from (10) by taking the L2 -inner product with ηj+1 and integrating by parts. Using
the fact that d∗ ϕ = d∗ ψ and ψ L2 = O(t4 ), |ψ| = O(t1/2 ) we get a powerful
estimate of the L2 -norm of dηj+1 .
Equation (12) is derived from an elliptic regularity estimate for the operator
d + d∗ acting on 3-forms on M . Equation (13) follows from the Sobolev embedding
theorem, since L14 0
1 (M ) embeds in C (M ). Both (12) and (13) are proved on small
balls of radius O(t) in M , using parts (ii) and (iii) of Theorem 2.3, and this is where
the powers of t come from.
Using (11)-(13) and part (i) of Theorem 2.3 we show that if

(14) dηj L2 C4 t4 , ∇dηj L14 C5 , and dηj C 0 Kt1/2 ,
where C4 , C5 and K are positive constants depending on C1 , C2 , C3 and A1 , and if
t is sufficiently small, then the same inequalities (14) apply to dηj+1 . Since η0 = 0,
by induction (14) applies for all j and the sequence (dηj )∞ j=0 is bounded in the
3 ∗
Banach space L14 1 (Λ T M ). One can then use standard techniques in analysis to
prove that this sequence converges to a smooth limit dη. This concludes the proof
of Theorem 2.4.
Figure 1. Betti numbers (b2 , b3 ) of compact G2 -manifolds

2
b (M )
q
q
q q qq q
25
qqq q
20 q q qqq qq q q qq qq q
q qq qq qq q q qq q qq q q
15 q q q q
q q qq qq qq qq qq q qqq q q
qq q q q qq qq qq qq qq q q q q q q
q qqq qq q qq qq qq qq qq q q q q q qq qq
10
q qq qq q qq qq qq qq qq q qq q q q q q q q q qqq
q qqqqq
q q qqq q q
q qq q q q qq qq qq qq qq qq qq q q q q qq qq qq q q q q qq q q q q q q q q q q
q q q q q q q q q q q q q q
q qq qq q qqq
5
q q q q q q q q q q qq q qq qq qq qq q q q q q q b3 (M )
q q q
0
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210
From Theorems 2.3 and 2.4 we see that the compact 7-manifold M constructed
in Step 2 admits torsion-free G2 -structures (ϕ̃, g̃). Theorem 1.4 then shows that
Hol(g̃) = G2 if and only if π1 (M ) is finite. In the example above M is simply-
connected, and so π1 (M ) = {1} and M has metrics with holonomy G2 , as we
want.
By considering different groups Γ acting on T 7 , and also by finding topologically
distinct resolutions M1 , . . . , Mk of the same orbifold T 7 /Γ, we can construct many
compact Riemannian 7-manifolds with holonomy G2 . A good number of examples
are given in [9, §12]. Figure 1 displays the Betti numbers of compact, simply-
connected 7-manifolds with holonomy G2 constructed there. There are 252 different
sets of Betti numbers.
Examples are also known [9, §12.4] of compact 7-manifolds with holonomy G2
with finite, nontrivial fundamental group. It seems likely to the author that the
Betti numbers given in Figure 1 are only a small proportion of the Betti numbers
of all compact, simply-connected 7-manifolds with holonomy G2 .
2.5. Other constructions of compact G2 -manifolds.
Here are two other methods, taken from [9, §11.9], that may be used to con-
struct compact 7-manifolds with holonomy G2 . The first was outlined in [6, §4.3].
Method 1. Let (Y, J, h) be a Calabi–Yau 3-fold, with Kähler form ω and holo-
morphic volume form θ. Suppose σ : Y → Y is an involution, satisfying σ ∗ (h) = h,
σ ∗ (J) = −J and σ ∗ (θ) = θ̄. We call σ a real structure on Y . Let N be the fixed
186 DOMINIC JOYCE
point set of σ in Y . Then N is a real 3-dimensional submanifold of Y , and is in

fact a special Lagrangian 3-fold.
Let S 1 = R/Z, and define a torsion-free G2 -structure (ϕ, g) on S 1 × Y as in
Proposition 1.8. Then ϕ = dx ∧ ω + x ∈ R/Z
Re θ,where is the coordinate on S .
1
Define σ̂ : S × Y → S × Y by σ̂ (x, y) = −x, σ(y) . Then σ̂ preserves (ϕ, g)

1 1
and σ̂ 2 = 1. The fixed points of σ̂ in S 1 × Y are {Z, 12 + Z} × N . Thus (S 1 × Y )/σ̂

is an orbifold. Its singular set is 2 copies of N , and each singular point is modelled
on R3 × R4 /{±1}.
We aim to resolve (S 1 × Y )/σ̂ to get a compact 7-manifold M with holonomy
G2 . Locally, each singular point should be resolved like R3 × X, where X is an
ALE Calabi–Yau 2-fold asymptotic to C2 /{±1}. There is a 3-dimensional family
of such X, and we need to choose one member of this family for each singular point
in the singular set.
Calculations by the author indicate that the data needed to do this is a closed,
coclosed 1-form α on N that is nonzero at every point of N . The existence of a
suitable 1-form α depends on the metric on N , which is the restriction of the metric
g on Y . But g comes from the solution of the Calabi Conjecture, so we know little
about it. This may make the method difficult to apply in practice.
The second method is studied by Alexei Kovalev [12], and is based on an idea
due to Simon Donaldson.
Method 2. Let X be a projective complex 3-fold with canonical bundle KX , and
−1
s a holomorphic section of KX which vanishes to order 1 on a smooth divisor D
in X. Then D has trivial canonical bundle, so D is T 4 or K3. Suppose D is a K3
surface. Define Y = X \ D, and suppose Y is simply-connected.
Then Y is a noncompact complex 3-fold with KY trivial, and one infinite end
modelled on D × S 1 × [0, ∞). Using a version of the proof of the Calabi Conjecture
for noncompact manifolds one constructs a complete Calabi–Yau metric h on Y ,
which is asymptotic to the product on D × S 1 × [0, ∞) of a Calabi–Yau metric on
D, and Euclidean metrics on S 1 and [0, ∞). We call such metrics Asymptotically
Cylindrical.
Suppose we have such a metric on Y . Define a torsion-free G2 -structure (ϕ, g)
on S 1 × Y as in Proposition 1.8. Then S 1 × Y is a noncompact G2 -manifold with
one end modelled on D × T 2 × [0, ∞), whose metric is asymptotic to the product
on D × T 2 × [0, ∞) of a Calabi–Yau metric on D, and Euclidean metrics on T 2
and [0, ∞).
Donaldson and Kovalev’s idea is to take two such products S 1 × Y1 and S 1 × Y2
whose infinite ends are isomorphic in a suitable way, and glue them together to get
a compact 7-manifold M with holonomy G2 . The gluing process swaps round the
S 1 factors. That is, the S 1 factor in S 1 × Y1 is identified with the asymptotic S 1
factor in Y2 ∼ D2 × S 1 × [0, ∞), and vice versa.
3. Compact Spin(7)-manifolds from Calabi–Yau 4-orbifolds
In a very similar way to the G2 case, one can construct compact 8-manifolds
with holonomy Spin(7) by resolving the singularities of torus orbifolds T 8 /Γ. This
is done in [7] and [9, §13–§14]. In [9, §14], examples are constructed realizing 181
different sets of Betti numbers. Two compact 8-manifolds with holonomy Spin(7)
and the same Betti numbers may be distinguished by the cup products on their
cohomologies (examples of this are given in [7, §3.4]), so they probably represent
rather more than 181 topologically distinct 8-manifolds.
The main differences from the G2 case are, firstly, that the technical details of
the analysis are different and harder, and secondly, that the singularities that arise
are typically more complicated and more tricky to resolve. One reason for this is
that in the G2 case the singular set is made up of 1 and 3-dimensional pieces in a
7-dimensional space, so one can often arrange for the pieces to avoid each other,
and resolve them independently.
But in the Spin(7) case the singular set is typically made up of 4-dimensional
pieces in an 8-dimensional space, so they nearly always intersect. There are also
topological constraints arising from the Â-genus, which do not apply in the G2 case.
The moral appears to be that when you increase the dimension, things become more
difficult.
Anyway, we will not discuss this further, as the principles are very similar to
the G2 case above. Instead, we will discuss an entirely different construction of
compact 8-manifolds with holonomy Spin(7) developed by the author in [8] and [9,
§15], a little like Method 1 of § 2.5. In this we start from a Calabi–Yau 4-orbifold
rather than from T 8 . The construction can be divided into five steps.
Step 1. Find a compact, complex 4-orbifold (Y, J) satisfying the conditions:

(a) Y has only finitely many singular points p1 , . . . , pk , for k 1.
(b) Y is modelled on C4 /i near each pj , where i acts on C4 by complex
multiplication.
(c) There exists an antiholomorphic involution σ : Y → Y whose fixed
point set is {p1 , . . . , pk }.
(d) Y \ {p1 , . . . , pk } is simply-connected, and h2,0 (Y ) = 0.
Step 2. Choose a σ-invariant Kähler class on Y . Then by Theorem 1.7 there exists
a unique σ-invariant Ricci-flat Kähler metric g in this Kähler class. Let ω
be the Kähler form of g. Let θ be a holomorphic volume form for (Y, J, g).
By multiplying θ by eiφ if necessary, we can arrange that σ ∗ (θ) = θ̄.
Define Ω = 12 ω ∧ ω + Re θ. Then (Ω, g) is a torsion-free Spin(7)-
structure on Y , by Proposition 1.10. Also, (Ω, g) is σ-invariant, as σ ∗ (ω) =
−ω and σ ∗ (θ) = θ̄. Define Z = Y /σ. Then Z is a compact real 8-
orbifold with isolated singular points p1 , . . . , pk , and (Ω, g) pushes down
to a torsion-free Spin(7)-structure (Ω, g) on Z.
Step 3. Z is modelled on R8 /G near each pj , where G is a certain finite subgroup
of Spin(7) with |G| = 8. We can write down two explicit, topologically
distinct ALE Spin(7)-manifolds X1 , X2 asymptotic to R8 /G. Each car-
ries a 1-parameter family of homothetic ALE metrics ht for t > 0 with
Hol(ht ) = Z2 SU(4) ⊂ Spin(7).
For j = 1, . . . , k we choose ij = 1 or 2, and resolve the singularities
of Z by gluing in Xij at the singular point pj for j = 1, . . . , k, to get a
compact, nonsingular 8-manifold M , with projection π : M → Z.
Step 4. On M , we explicitly write down a 1-parameter family of Spin(7)-structures
(Ωt , gt ) depending on t ∈ (0, ). They are not torsion-free, but have small
torsion when t is small. As t → 0, the Spin(7)-structure (Ωt , gt ) converges
to the singular Spin(7)-structure π ∗ (Ω0 , g0 ).
188 DOMINIC JOYCE
Step 5. We prove using analysis that for sufficiently small t, the Spin(7)-structure
(Ωt , gt ) on M , with small torsion, can be deformed to a Spin(7)-structure
(Ω̃t , g̃t ), with zero torsion.
It turns out that if ij = 1 for j = 1, . . . , k we have π1 (M ) ∼ = Z2
and Hol(g̃t ) = Z2 SU(4), and for the other 2k − 1 choices of i1 , . . . , ik we
have π1 (M ) = {1} and Hol(g̃t ) = Spin(7). So g̃t is a metric with holonomy
Spin(7) on the compact 8-manifold M for (i1 , . . . , ik ) = (1, . . . , 1).
Once we have completed Step 1, Step 2 is immediate. Steps 4 and 5 are
analogous to Steps 3 and 4 of § 2, and can be done using the techniques and
analytic results developed by the author for the first T 8 /Γ construction of compact
Spin(7)-manifolds, [7], [9, §13]. So the really new material is in Steps 1 and 3, and
we will discuss only these.
3.1. Step 1: An example.
We do Step 1 using complex algebraic geometry. The problem is that conditions
(a)–(d) above are very restrictive, so it is not that easy to find any Y satisfying all
four conditions. All the examples Y the author has found are constructed using
weighted projective spaces, an important class of complex orbifolds.
Definition 3.1. Let m 1 be an integer, and a0 , a1 , . . . , am positive integers
with highest common factor 1. Let Cm+1 have complex coordinates (z0 , . . . , zm ),
and define an action of the complex Lie group C∗ on Cm+1 by
for u ∈ C∗ .
u
(z0 , . . . , zm )
−→(ua0 z0 , . . . , uam zm ),
m+1
The weighted projective space CPm a0 ,...,am is C \ {0} /C∗ . The C∗ -orbit of
(z0 , . . . , zm ) is written [z0 , . . . , zm ].
Here is the simplest example the author knows.
Example 3.2. Let Y be the hypersurface of degree 12 in CP51,1,1,1,4,4 given by

Y = [z0 , . . . , z5 ] ∈ CP51,1,1,1,4,4 : z012 + z112 + z212 + z312 + z43 + z53 = 0 .
Calculation shows that Y has trivial canonical bundle and three singular points p1 =
[0, 0, 0, 0, 1, −1], p2 = [0, 0, 0, 0, 1, eπi/3 ] and p3 = [0, 0, 0, 0, 1, e−πi/3 ], all modelled
on C4 /i.
Now define a map σ : Y → Y by
σ : [z0 , . . . , z5 ]
−→ [z̄1 , −z̄0 , z̄3 , −z̄2 , z̄5 , z̄4 ].
2
Note that σ = 1, though this is not immediately obvious, because of the geometry
of CP51,1,1,1,4,4 . It can be shown that conditions (a)–(d) of Step 1 above hold for Y
and σ.
More suitable 4-folds Y may be found by taking hypersurfaces or complete
intersections in other weighted projective spaces, possibly also dividing by a finite
group, and then doing a crepant resolution to get rid of any singularities that we
don’t want. Examples are given in [8], [9, §15].
3.2. Step 3: Resolving R8 /G.
Define α, β : R8 → R8 by
α : (x1 , . . . , x8 )
−→ (−x2 , x1 , −x4 , x3 , −x6 , x5 , −x8 , x7 ),
β : (x1 , . . . , x8 )
−→ (x3 , −x4 , −x1 , x2 , x7 , −x8 , −x5 , x6 ).
Then α, β preserve Ω0 given in (2), so they lie in Spin(7). Also α4 = β 4 = 1,

α2 = β 2 and αβ = βα3 . Let G = α, β. Then G is a finite non-Abelian subgroup
of Spin(7) of order 8, which acts freely on R8 \ {0}. One can show that if Z is
the compact Spin(7)-orbifold constructed in Step 2 above, then Tpj Z is isomorphic
to R8 /G for j = 1, . . . , k, with an isomorphism identifying the Spin(7)-structures
(Ω, g) on Z and (Ω0 , g0 ) on R8 /G, such that β corresponds to the σ-action on Y .
In the next two examples we construct two different ALE Spin(7)-manifolds
(X1 , Ω1 , g1 ) and (X2 , Ω2 , g2 ) asymptotic to R8 /G.
Example 3.3. Define complex coordinates (z1 , . . . , z4 ) on R8 by
(z1 , z2 , z3 , z4 ) = (x1 + ix2 , x3 + ix4 , x5 + ix6 , x7 + ix8 ),
Then g0 = |dz1 |2 + · · · + |dz4 |2 , and Ω0 = 12 ω0 ∧ ω0 + Re(θ0 ), where ω0 and θ0 are
the usual Kähler form and complex volume form on C4 . In these coordinates, α
and β are given by
α : (z1 , . . . , z4 )
−→ (iz1 , iz2 , iz3 , iz4 ),
(15)
β : (z1 , . . . , z4 )
−→ (z̄2 , −z̄1 , z̄4 , −z̄3 ).
Now C4 /α is a complex singularity, as α ∈ SU(4). Let (Y1 , π1 ) be the blow-up
of C4 /α at 0. Then Y1 is the unique crepant resolution of C4 /α. The action
of β on C4 /α lifts to a free antiholomorphic map β : Y1 → Y1 with β 2 = 1.
Define X1 = Y1 /β. Then X1 is a nonsingular 8-manifold, and the projection
π1 : Y1 → C4 /α pushes down to π1 : X1 → R8 /G.
There exist ALE Calabi–Yau metrics g1 on Y1 , which were written down ex-
plicitly by Calabi [4, p. 285], and are invariant under the action of β on Y1 . Let
ω1 be the Kähler form of g1 , and θ1 = π1∗ (θ0 ) the holomorphic volume form on Y1 .
Define Ω1 = 12 ω1 ∧ ω1 + Re(θ1 ). Then (Ω1 , g1 ) is a torsion-free Spin(7)-structure
on Y1 , as in Proposition 1.10.
As β ∗ (ω1 ) = −ω1 and β ∗ (θ1 ) = θ̄1 , we see that β preserves (Ω1 , g1 ). Thus
(Ω1 , g1 ) pushes down to a torsion-free Spin(7)-structure (Ω1 , g1 ) on X1 . Then
(X1 , Ω1 , g1 ) is an ALE Spin(7)-manifold asymptotic to R8 /G.
Example 3.4. Define new complex coordinates (w1 , . . . , w4 ) on R8 by
(w1 , w2 , w3 , w4 ) = (−x1 + ix3 , x2 + ix4 , −x5 + ix7 , x6 + ix8 ).
Again we find that g0 = |dw1 |2 + · · · + |dw4 |2 and Ω0 = 12 ω0 ∧ ω0 + Re(θ0 ). In these
coordinates, α and β are given by
α : (w1 , . . . , w4 )
−→ (w̄2 , −w̄1 , w̄4 , −w̄3 ),
(16)
β : (w1 , . . . , w4 )
−→ (iw1 , iw2 , iw3 , iw4 ).
Observe that (15) and (16) are the same, except that the rôles of α, β are reversed.
Therefore we can use the ideas of Example 3.3 again.
Let Y2 be the crepant resolution of C4 /β. The action of α on C4 /β lifts to
a free antiholomorphic involution of Y2 . Let X2 = Y2 /α. Then X2 is nonsingular,
and carries a torsion-free Spin(7)-structure (Ω2 , g2 ), making (X2 , Ω2 , g2 ) into an
ALE Spin(7)-manifold asymptotic to R8 /G.
We can now explain the remarks on holonomy groups at the end of Step 5. The
holonomy groups Hol(gi ) of the metrics g1 , g2 in Examples 3.3 and 3.4 are both
isomorphic to Z2 SU(4), a subgroup of Spin(7). However, they are two different
190 DOMINIC JOYCE
inclusions of Z2 SU(4) in Spin(7), as in the first case the complex structure is α

and in the second β.
The Spin(7)-structure (Ω, g) on Z also has holonomy Hol(g) = Z2 SU(4).
Under the natural identifications we have Hol(g1 ) = Hol(g) but Hol(g2 ) = Hol(g)
as subgroups of Spin(7). Therefore, if we choose ij = 1 for all j = 1, . . . , k, then
Z and Xij all have the same holonomy group Z2 SU(4), so they combine to give
metrics g̃t on M with Hol(g̃t ) = Z2 SU(4).
However, if ij = 2 for some j then the holonomy of g on Z and gij on Xij are
different Z2 SU(4) subgroups of Spin(7), which together generate the whole group
Spin(7). Thus they combine to give metrics g̃t on M with Hol(g̃t ) = Spin(7).
3.3. Conclusions.
The author was able in [8] and [9, Ch. 15] to construct compact 8-manifolds
with holonomy Spin(7) realizing 14 distinct sets of Betti numbers, which are given
in Table 1. Probably there are many other examples which can be produced by
similar methods.
Table 1. Betti numbers (b2 , b3 , b4 ) of compact Spin(7)-manifolds

(4, 33, 200) (3, 33, 202) (2, 33, 204) (1, 33, 206) (0, 33, 208)
(1, 0, 908) (0, 0, 910) (1, 0, 1292) (0, 0, 1294) (1, 0, 2444)
(0, 0, 2446) (0, 6, 3730) (0, 0, 4750) (0, 0, 11 662)
Comparing these Betti numbers with those of the compact 8-manifolds con-
structed in [9, Ch. 14] by resolving torus orbifolds T 8 /Γ, we see that in these
examples the middle Betti number b4 is much bigger, as much as 11 662 in one
case.
Given that the two constructions of compact 8-manifolds with holonomy Spin(7)
that we know appear to produce sets of 8-manifolds with rather different ‘geogra-
phy’, it is tempting to speculate that the set of all compact 8-manifolds with ho-
lonomy Spin(7) may be rather large, and that those constructed so far are a small
sample with atypical behaviour.
References
1. M. Berger, Sur les groupes d’holonomie homogène des variétés à connexion affines et des
variétés Riemanniennes, Bull. Soc. Math. France 83 (1955), 279–330.
2. R.L. Bryant, Metrics with exceptional holonomy, Ann. Math. 126 (1987), 525–576.
3. R.L. Bryant and S.M. Salamon, On the construction of some complete metrics with excep-
tional holonomy, Duke Math. J. 58 (1989), 829–850.
4. E. Calabi, Métriques kählériennes et fibrés holomorphes, Ann. Scient. Éc. Norm. Sup. 12
(1979), 269–294.
5. D.D. Joyce, Compact Riemannian 7-manifolds with holonomy G2 . I, J. Diff. Geom. 43
(1996), 291–328.
6. , Compact Riemannian 7-manifolds with holonomy G2 . II, J. Diff. Geom. 43 (1996),
329–375.
7. , Compact Riemannian 8-manifolds with holonomy Spin(7), Invent. Math. 123
(1996), 507–552.
8. , A new construction of compact 8-manifolds with holonomy Spin(7), J. Diff. Geom.
53 (1999), 89–130, math.DG/9910002.
9. , Compact manifolds with special holonomy, Oxford Mathematical Monographs, Ox-
ford University Press, Oxford, 2000, reprinted 2003.
10. , Asymptotically Locally Euclidean metrics with holonomy SU(m), Ann. Global Anal.
Geom. 19 (2001), 55–73, math.AG/9905041.
11. , Quasi-ALE metrics with holonomy SU(m) and Sp(m), Ann. Global Anal. Geom.
19 (2001), 103–132, math.AG/9905043.
12. A.G. Kovalev, Twisted connected sums and special Riemannian holonomy, J. Reine Angew.
Math 565 (2003) 125-160.
13. P.B. Kronheimer, The construction of ALE spaces as hyperkähler quotients, J. Diff. Geom.
29 (1989), 665–683.
14. , A Torelli-type theorem for gravitational instantons, J. Diff. Geom. 29 (1989), 685–
697.
15. S.-S. Roan, Minimal resolution of Gorenstein orbifolds, Topology 35 (1996), 489–508.
16. S.-T. Yau, On the Ricci curvature of a compact Kähler manifold and the complex Monge–
Ampère equations. I, Comm. Pure Appl. Math. 31 (1978), 339–411.
Lincoln College, Oxford, OX1 3DR

E-mail address: dominic.joyce@lincoln.ox.ac.uk
Volume 3, 2004
From Fano Threefolds to

Compact G2 -Manifolds
Alexei Kovalev
Abstract. G2 -manifolds are 7-dimensional Riemannian manifolds whose met-

rics have holonomy group G2 ; these are necessarily Ricci-flat. We explain a
systematic way to construct examples of compact G2 -manifolds by gluing a
pair of asymptotically cylindrical manifolds of holonomy SU (3) at their cylin-
drical ends. To obtain the latter SU (3)-manifolds one starts from complex
3-dimensional projective manifolds with c1 > 0 (Fano threefolds) endowed
with an appropriate choice of the anticanonical K3 divisor. The resulting
G2 -manifolds are topologically distinct from those previously constructed by
Joyce.
This article is an informal, introductory account of the ‘generalized connected

sum’ construction of compact Riemannian manifolds with holonomy G2 . Full de-
tails and proofs the results can be found in the author’s paper [6]. A good reference
on the Riemannian holonomy groups, including G2 and the previously known con-
struction of compact G2 -manifolds, is the book by Joyce [4].
We briefly review in Section 1 the background results on G2 holonomy. The
method of construction of manifolds of holonomy G2 is explained in Section 2.
Section 3 explains how to obtain examples of this construction using the theory of
Fano threefolds and K3 surfaces, and contains a discussion of the results.
1. Synopsis on the holonomy group G2

The holonomy group Hol(g) of a Riemannian manifold (M, g) is defined as the
group of isometries of the tangent space Tx M generated by parallel transport, using
the Levi–Civita connection of g, over closed loops based at x. Up to conjugation,
the holonomy group is well-defined as a subgroup of O(n), n = dim M . If M is
an oriented simply-connected Riemannian manifold, which is not locally isometric
to a Riemannian product or to a Riemannian symmetric space, then there are
very few groups which may occur as the holonomy of M , according to Berger’s
classification theorem. In fact, if in addition one assumes that the dimension of M
is odd then there are just two possibilities: either Hol(g) = SO(n) or dim M = 7
and Hol(g) = G2 .
2000 Mathematics Subject Classification. Primary 53C29, 14J45, 32Q25.
2004
193
194 ALEXEI KOVALEV
The group G2 may be defined as the group of automorphisms of the cross-

product algebra on R7 arising from the identification of R7 with the purely imagi-
nary octonions. It is a compact Lie group and a (proper) subgroup of SO(7). The
cross-product multiplication may be encoded by a 3-form ϕ0 ∈ Λ3 (R7 )∗ ,
ϕ0 (a, b, c) = a × b, c,
or, explicitly,
ϕ0 =e5 ∧ e6 ∧ e7 + (e1 ∧ e2 + e3 ∧ e4 ) ∧ e7
(1.1)
+ (e1 ∧ e3 − e2 ∧ e4 ) ∧ e6 + (e1 ∧ e4 + e2 ∧ e3 ) ∧ e5 ,
where ei denote an orthonormal basis of (R7 )∗ . Conversely, the formula
(1.2) 6a, b dvol7 = (aϕ0 ) ∧ (bϕ0 ) ∧ ϕ0 .
expresses the Euclidean inner product in terms of ϕ0 and the volume form of R7 .
The group G2 is thus identified as the stabilizer of ϕ0 in the natural action of
GL(7, R) on Λ3 (R7 )∗ . The form ϕ0 is stable, in the sense of Hitchin [2] —the
GL(7, R)-orbit of ϕ0 is open in Λ3 (R7 )∗ .
A G2 -structure on a 7-manifold, M say, may be given by a 3-form ϕ such that
at each point p ∈ M , ϕ(p) is the image of ϕ0 induced by a linear isomorphism
Tp M → R7 . Denote by Ω3+ (M ) the subset of 3-forms point-wise modelled on
ϕ0 in the latter sense; elements of Ω3+ (M ) will sometimes be referred to as the
G2 -structure 3-forms. Note that Ω3+ (M ) is an open subset of Ω3 (M ) in the sup-
norm topology, a direct consequence of the stablity property of ϕ0 .
Every 3-form ϕ ∈ Ω3+ (M ) defines an orientation and a Riemannian metric
g = g(ϕ) on M , as any G2 -structure is an instance of an SO(7)-structure. The
formula (1.2) determines g(ϕ) explicitly, up to a conformal factor. The holonomy
group of g(ϕ) will be a subgroup of G2 if and only if the form ϕ is parallel, ∇ϕ = 0,
with respect to the Levi–Civita connection of g. The latter condition is equivalent
to the system of partial differential equations on ϕ [9, Lemma 11.5],
(1.3) dϕ = 0 and d ∗ϕ ϕ = 0.
The second equation in (1.3) is non-linear as the Hodge star ∗ϕ is taken in the
metric g(ϕ) and depends on ϕ. The holonomy reduction Hol(g(ϕ)) ⊆ G2 implies
that g(ϕ) is Ricci-flat.
Proposition 1.1 ([4, pp.244–245]). Suppose that a 7-manifold M is compact

and let ϕ ∈ Ω3+ (M ). Then Hol(g(ϕ)) = G2 if and only if ϕ ∈ Ω3+ (M ) is a solution
to (1.3) and the fundamental group of M is finite.
We shall say that a Riemannian 7-manifold (M, g) is a G2 -manifold if

Hol(g) = G2 and shall use a similar terminology for other holonomy groups.
The first examples of compact G2 -manifolds were constructed in 1994-5 by
Joyce, using a generalized Kummer construction and resolution of singularities.
The most elaborate form of this construction can be found in [4]. Recently the
author obtained different examples of compact G2 -manifolds by a different method
[6] which we shall now describe.
FROM FANO THREEFOLDS TO COMPACT G2 -MANIFOLDS 195
2. The generalized connected sum construction

Our compact G2 -manifolds are constructed by forming a carefully chosen gen-
eralized connected sum of two non-compact Riemannian manifolds with asymptot-
ically cylindrical ends. The construction develops an idea due to Donaldson.
Firstly, we produce a class of complete Ricci-flat Kähler threefolds W of holo-
nomy SU (3) with an infinite cylindrical end asymptotic to the Riemannian product
D × S 1 × R>0 , where D is a K3 surface with a hyper-Kähler metric. This step re-
quires a proof of a non-compact version of the Calabi conjecture, which may be of
independent interest.
The Riemannian product W × S 1 carries a solution to (1.3). We consider a
pair of such 7-manifolds W1 × S 1 and W2 × S 1 . For certain pairs of hyper-Kähler
K3 surfaces Di ‘at the infinity of Wi ’ (i = 1, 2), there is a way to join the two
7-manifolds Wi × S 1 at their ends to obtain a compact 7-manifold M having finite
fundamental group and a 1-parameter family of G2 -structures ϕT compatible with
those on Wi × S 1 .
A G2 -structure ϕT on M is obtained using cut-off functions, which introduce
error terms in the equations (1.3). These error terms are exponentially small in T .
We use ‘stretching the neck’ analysis to prove a gluing theorem, obtaining a solution
to (1.3) on M from the solutions on Wi × S 1 .
The three parts of the construction are described in more detail below.
2.1. Asymptotically cylindrical Calabi–Yau manifolds. The equations

(1.3) define a metric g(ϕ) whose holonomy group is only contained in G2 . In
particular, the holonomy may be SU (3), a maximal subgroup of G2 .
We begin by introducing the holonomy SU (n) which will be needed in the cases
n = 3 and n = 2. The group SU (n) consists of all the complex linear isomorphisms
of Cn preserving the standard Hermitian inner product and the complex volume.
So SU (n) is the stabilizer of the pair of forms on Cn
ω0 = i
2 (dz1 ∧ dz̄1 + . . . + dzn ∧ dz̄n ) and Ω0 = dz1 ∧ · · · ∧ dzn
under the action of GL(n, C). Note that both ω0 and Ω0 are stable differential forms
(have open orbits under the action of GL(2n, R)). A metric g on a real 2n-manifold
Z will have holonomy contained in SU (n) if and only if Z has an SU (n)-structure
(I, ω, Ω) parallel with respect to g. Here I is an orthogonal complex structure with
respect to g, and ω and Ω are differential forms which are point-wise modelled on
ω0 and Ω0 , via a C-linear identification of tangent spaces to Z with Cn . That is
to say, a g-parallel SU (n)-structure makes Z into a Kähler complex n-fold with
the Kähler form ω ∈ Ω1,1 (Z), and Z has a nowhere vanishing holomorphic form
Ω ∈ Ωn,0 (Z) such that Ω ∧ Ω∗ is a constant multiple of ω n . Such a Ω is sometimes
called a holomorphic volume form. In particular, Z has trivial canonical bundle of
(n, 0)-forms and c1 (Z) = 0 and the Kähler metric is Ricci-flat.
Conversely, the following is a direct consequence of Yau’s proof of the Calabi
conjecture [12].
Theorem 2.1. Let Z be a Kähler complex n-fold with ωZ the Kähler form
on Z and suppose that c1 (Z) = 0. Then there exists on Z a unique Ricci-flat
¯ for some smooth
Kähler metric such that its Kähler form is given by ωZ + i∂ ∂u
real function u on Z. If Z is simply-connected then the holonomy of this Ricci-flat
Kähler metric is contained in SU (n).
196 ALEXEI KOVALEV
Kähler manifolds with holonomy in SU (n) are often called Calabi–Yau man-
ifolds. An important example is a K3 surface; recall that it may be defined as a
simply-connected complex surface with c1 = 0. By Yau’s theorem, a K3 surface
admits a unique Ricci-flat Kähler metric in every Kähler class.
Now let n = 3. The group SU (3) is a subgroup of G2 consisting of all those
elements of G2 which fix a particular one-dimensional subspace in R7 , thus it deter-
mines a decomposition R7 ∼ = C3 ⊕R. Consider a Kähler threefold W with holonomy
in SU (3) and let ω, Ω be respectively the Kähler form and a holomorphic volume
form on W . Then on the 7-manifold W × S 1 the 3-form
(2.1) ϕ = ω ∧ dθ + Im Ω
is in × S ) and defines a product metric, so W × S 1 has the same holonomy
Ω3+ (W 1
as W . In this metric, one has ∗ϕ = 12 ω ∧ ω − Re Ω ∧ dθ and ϕ is a solution to (1.3)

on W × S 1 .
We are now ready to introduce the class of complete SU (3)-manifolds that we
need. Let W be a compact simply-connected Kähler threefold, with ω ∈ Ω1,1 (W )
the Kähler form. Let D be a K3 surface in W such that there is a holomorphic
−1
section s of the anticanonical bundle KW vanishing to order 1 on D. It is easy to
see that the complement W = W \ D has trivial canonical bundle.
Assume further that the normal bundle of D in W is trivial. Then W can be
written as the union of two pieces,
(2.2) W Wcpt ∪ (D × S 1 × R+ )
a compact manifold Wcpt with boundary and a cylindrical end attached along the
boundary D × S 1 . Note that the relation (2.2) is only a diffeomorphism of the un-
derlying real manifolds. The complex structure on the end of W is not isomorphic,
but only asymptotic to the ‘obvious’ product complex structure on D × S 1 × R+ .
Let gD denote the Ricci-flat Kähler metric on D in the Kähler class [ω |D ]
determined by the embedding in W . We prove that the following non-compact
version of the Calabi conjecture is true.
Theorem 2.2. Let W and D be as above, so a K3 surface D is an anti-
canonical divisor and has trivial normal bundle in W . Suppose also that W is
simply-connected and the fundamental group of W = W \ D is finite.
Then W admits a complete Ricci-flat Kähler metric gW . The Kähler form and
holomorphic volume form of gW are exponentially asymptotic, along the cylindrical
end of W , to those of the product Ricci-flat Kähler structure on D ×S 1 ×R+ defined
using the metric gD on D. The holonomy of gW is SU (3).
There is nothing special to threefolds in the proof of Theorem 2.2 and the result
extends, with only minor modifications, to Kähler manifolds of arbitrary dimension.
We also remark at this point that previously a number of other non-compact
versions of the Calabi conjecture were proved by Tian and Yau, Bando and Koba-
yashi, and Joyce. These authors construct complete Ricci-flat Kähler metrics as-
ymptotic at infinity to the quotient Cn /Γ of Hermitian Cn by a finite subgroup Γ
of SU (n).
The main novelty of Theorem 2.2 is that it deals with the class of asymptot-
ically cylindrical manifolds. We build up on Theorem 5.2 in [11] using analysis
on exponentially weighted Sobolev spaces to work out the details of asymptotic
behaviour and provide control on the boundary data at infinity.
2.2. K3 surfaces and a hyper-Kähler rotation. A remarkable property

of a Ricci-flat Kähler metric on a complex surface D is that such a metric is hyper-
Kähler : the underlying real 4-manifold admits, in addition to the given complex
structure I, another complex structure J, such that IJ = −JI and the metric is
Kähler with respect to J too. Further, K = IJ is also a complex structure on
D, and I, J, K satisfy the quaternionic relations (and define an identification of
each tangent space of D with the quaternions). The three respective Kähler forms
κI ,κJ ,κK satisfy κ2I = κ2J = κ2K . There is a complete SO(3) symmetry between
I, J, K, in particular, they generate a 2-sphere of complex structures aI + bJ + cK
on D, where a2 + b2 + c2 = 1, and the metric is Kähler with respect to each of
these.
Let D be a Ricci-flat Kähler K3 surface and let κI be the Kähler form on D.
Then κJ + iκK defines a holomorphic volume form on D. Considering on D the
complex structure J we obtain in general a different Ricci-flat Kähler K3 surface DJ .
It has Kähler form κJ and holomorphic volume form κI − iκK and is sometimes
called a hyper-Kähler rotation of D. Note that there is an S 1 -ambiguity in choosing
J, as one may take any bJ + cK instead with b2 + c2 = 1.
Consider two asymptotically cylindrical SU (3)-manifolds W1 and W2 satisfy-
ing the assertions of Theorem 2.2 and, respectively, let D1 ,D2 be the Ricci-flat
Kähler K3 surfaces which determine the asymptotic model on the cylindrical ends
of W1 , W2 . For i = 1, 2, let ti ≥ 0 be the real parameter along the cylindrical end of
Wi , as defined by (2.2). Cut off at ti = T − 1 the Kähler and holomorphic volume
form on each Wi to their asymptotic model on the cylindrical end and consider
Wi (T ) Wcpt ∪ (Di × S 1 × [0, T ]). Then W1 (T ) × S 1 is a manifold with boundary
D1 × S 1 × S 1 and with a G2 -structure form which on a collar neighbourhood of the
boundary is given by
(2.3) ϕ(D1 ) = κI ∧ dθ1 + κJ ∧ dθ2 + κK ∧ dt + dθ1 ∧ dθ2 ∧ dt.
Here we used (2.1) and the cylindrical asymptotic model ω = κI + dθ2 ∧ dt, Ω =
(κJ + iκK ) ∧ (dθ2 + idt) of the Kähler and holomorphic volume forms on the end of
W1 . In particular, ϕ(D1 ) is a solution to (1.3) on the cylinder (D1 × S 1 × R) × S 1 .
Similar expressions hold for W2 × S 1 .
Now assume that the Ricci-flat Kähler K3 surface D2 is isomorphic to a hyper-
Kähler rotation of D1 . Let f : D1,J → D2 denote the isomorphism. Then the
pull-back action of f on the Kähler forms is given by
f ∗ : κI → κJ , κJ → κI , κK → (−κK ).
Define
(2.4) F : (y, θ1 , θ2 , t) ∈ D1 × S 1 × S 1 × [T − 1, T ] →
(f (y), θ2 , θ1 , 2T − 1 − t) ∈ D2 × S 1 × S 1 × [T − 1, T ]
and join the two manifolds with boundary to construct a closed oriented 7-manifold
M = (W1 (T ) × S 1 ) ∪F (W2 (T ) × S 1 ),
using the map F to identify collar neighbourhoods of the boundaries. The compact
7-manifold M is a generalized connected sum with the neck having the cross-section
D × S 1 × S 1 . We have F ∗ ϕ(D1 ) = ϕ(D2 ) , by the construction of F , therefore there
is a well-defined 1-parameter family of G2 -structures ϕT on M induced from those
198 ALEXEI KOVALEV
on Wi (T ) × S 1 , defined above using cut-off functions. Here the parameter T is

approximately half the length of the neck of M , measured by g(ϕT ).
The fundamental group of M is finite. This is because in the construction of M
the circle factor in W1 (T ) × S 1 is identified with a circle in W2 and the circle factor
in W2 (T ) × S 1 is identified with a circle in W1 , and we assumed that π1 (Wi ) are
finite. Therefore, by Proposition 1.1 any solution ϕ ∈ Ω3+ (M ) of the equations (1.3)
will define on M a metric g(ϕ) of holonomy G2 .
2.3. The gluing theorem. A G2 -structure form ϕT is constructed by patch-

ing the solutions of (1.3) when joining the two pieces of M . This uses cut-off
functions which introduce ‘error terms’ in the equations. In fact we can achieve
dϕT = 0 for all T and satisfy one of the two equations in (1.3), but the term d∗T ϕT
in general will not vanish. The error terms depend on the difference between the
SU (3)-structures on the end of Wi and on its cylindrical asymptotic model, and we
have an estimate
d ∗T ϕT Lpk < Cp,k e−λT ,
where 0 < λ < 1. Here ∗T denotes the Hodge star of the metric g(ϕT ).
We prove the following.
Theorem 2.3. There exists T0 ∈ R and for every T ≥ T0 a unique smooth

2-form ηT on M so that the following holds.
(1) ηT C 1 < const · e−µT , for some 0 < µ < 1, where the C 1 -norm is defined using
the metric g(ϕT ). In particular, ϕT + dηT is in Ω3+ (M ).
(2) The closed 3-form ϕT + dηT satisfies
(2.5) d ∗ϕT +dηT (ϕT + dηT ) = 0.
and so ϕT + dηT defines a metric of holonomy G2 on M .
The equation (2.5) can be rewritten, using the results of [4, §10.3] as a non-linear
elliptic PDE for η. For small η, this PDE has the form a(η) = a0 + Aη + Q(η) = 0,
where a0 = d ∗T ϕT , the linear elliptic operator A = AT is a compact perturbation
of the Hodge Laplacian of the form dd∗ + d∗ d + O(e−δT ), and Q(η) = O(|dη|2 ).
The central idea in the proof of Theorem 2.3 may be informally stated as
follows. For small η, the map a(η) is approximated by its linearization and so there
is a unique small solution η to the equation a(η) = 0, for every small a0 in the range
of A. This perturbative approach requires the invertibility of A and a suitable upper
bound on the operator norm A−1 T , as T → ∞. This bound determines what is
meant by ‘small’ a0 in this paragraph.
As we actually need the value of dη rather than η, there is no loss in restrict-
ing the equation (2.5) for η to the orthogonal complement of harmonic 2-forms
on M where the Laplacian is invertible. We use the technique of [5, §4.1] based on
Fredholm theory for the asymptotically cylindrical manifolds and weighted Sobolev
spaces to find an upper bound A−1 T < Ge . Here the constant G is independent
δT
of T and δ > 0 can be taken arbitrary small. So, for large T , the growth of A−1 T
is negligible compared to the decay of d ∗T ϕT and the ‘inverse function theorem’
strategy applies to give the required small solution ηT . Standard elliptic methods
show that this ηT is in fact smooth.
3. Examples arising from Fano threefolds

For applications of the construction given in Section 2 we need, as a start, to
find Kähler threefolds satisfying the hypotheses of Theorem 2.2.
3.1. Introduction to Fano threefolds. The following example is classical

in algebraic geometry.
The intersection of three generically chosen quadric hypersurfaces in CP 6 de-
fines a smooth Kähler threefold X8 . It is simply-connected and the characteristic
class c1 (X8 ) of its anticanonical bundle is the pull-back to X8 of the positive gen-
erator of the cohomology ring H ∗ (CP 6 ). That is to say, the anticanonical bundle
−1
KX 8
is the restriction to X8 of the tautological line bundle O(1) over CP 6 . It fol-
lows that any anticanonical divisor D on X8 is obtained by taking an intersection
D = X8 ∩ H with a hyperplane H in CP 6 . A generic such hyperplane section D is
a complex surface, isomorphic to a smooth complete intersection of three quadrics
in CP 5 . This is a well-known example of a K3 surface.
We next look at the normal bundle of X8 in D. An adjunction-type argument
shows that the normal bundle will be trivial if we can find another anticanonical
divisor D on X8 such that D does not meet D. But D = X8 ∩ H and the second
hyperplane section C = D ∩ D = X8 ∩ H ∩ H is never empty—it is an algebraic
curve (intersection of three quadrics) in CP 4 . Fortunately, a suitable threefold can
be obtained by blowing up the curve C. The K3 divisor D lifts via the blow-up
map X̃8 → X to an isomorphic K3 surface D̃ which is an anticanonical divisor in
X̃8 and has trivial normal bundle. Moreover, a Kähler metric on X̃8 may be chosen
so that D̃ and D are isometric Kähler manifolds.
Finally, as both D̃ and X8 are simply-connected we find that the only possibil-
ity for a nontrivial generator of π1 (X8 \ D) would be a circle around D̃. But this
circle contracts in an exceptional curve as this curve meets D̃ in exactly one point.
Hence X̃8 \ D̃ is simply connected. The pair X̃8 , D̃ now satisfies all the hypothe-
ses of Theorem 2.2, and so the quasiprojective threefold W = X̃8 \ D̃ admits an
asymptotically cylindrical Ricci-flat Kähler metric of holonomy SU (3).
The threefold X8 in the above example can be replaced by an arbitrary (smooth)
projective-algebraic threefold V with c1 (V ) > 0, i.e. a Fano threefold. Fano three-
folds have been extensively studied over the past few decades and a lot is known
about them. In particular, they are simply-connected and a generic anticanonical
divisor D on a Fano threefold is a K3 surface [10]. It can be shown that the three-
folds Ṽ \D̃ are again simply-connected and we obtain, by application of Theorem 2.2
the following.
Proposition 3.1. Let V be a Fano 3-fold, D ∈ |−KV | a K3 surface, and Ṽ the
blow-up of V along a self-intersection curve D·D, and D̃ the proper transform of D.
Then Ṽ \ D̃ has a complete Ricci-flat Kähler metric with holonomy SU (3). This
metric is asymptotic to the Riemannian product D × S 1 × R>0 , where the Ricci-flat
Kähler metric on D is in the Kähler class induced by the embedding in V .
3.2. Matching the K3 divisors. Let V1 , V2 be Fano threefolds and D1 , D2 ,
respectively, anticanonical K3 divisors on these. Recall that by Yau’s theorem each
of the Kähler K3 surfaces Di has a uniquely determined Ricci-flat Kähler metric
in its Kähler class. If the two Ricci-flat Kähler structures in the Kähler classes of
Di ⊂ Vi are hyper-Kähler rotations of each other then, in view of Proposition 3.1,
200 ALEXEI KOVALEV
we can proceed to the construction of a generalized connected sum M from Ṽi \ D̃i ,
as described in Section 2.2. The 7-manifold M admits metrics of holonomy G2
by Theorem 2.3. In this case, we shall say that a compact G2 -manifold M is
constructed from the pair of Fano threefolds V1 and V2 . Can we choose D1 and D2
so as to satisfy the required hyper-Kähler rotation condition?
We have the freedom to move a K3 surface in the anticanonical linear system
of V and to deform V in its algebraic family of Fano threefolds. Recall also from
Section 2.2 that there is an S 1 -family of choices for the second complex structure
J on each Di . It turns out that, with this freedom, a pair of ‘matching divisors’
D1 , D2 can always be found. We now briefly explain, ignoring some important
technical points, the ideas in the solution of the matching problem.
All the K3 surfaces are deformations of each other and are diffeomorphic as
real 4-manifolds. In particular, their second cohomology lattices are isomorphic
to the (unique) even unimodular lattice L of signature (3, 19), known as the K3
lattice. Respectively, L ⊗ C is isomorphic to the second cohomology with complex
coefficients and inherits the Hodge decomposition. A Kähler isometry between
two K3 surfaces induces a so-called effective Hodge isometry between their second
cohomology lattices, preserving the Hodge decomposition and mapping the Kähler
class of one K3 to the Kähler class of the other. Surprisingly, the global Torelli
theorem for K3 surfaces asserts that the converse is also true: any effective Hodge
isometry between second cohomology of two K3 surfaces arises as the pull-back of
a unique biholomorphic map between these K3 surfaces [1, Ch.VIII]. The latter
map will necessarily be an isometry between Ricci-flat Kähler K3 surfaces because
of the uniqueness of a Ricci-flat Kähler metric in a Kähler class.
We can identify, using a version of the Kodaira–Spencer–Kuranishi deforma-
tion theory, the data of Hodge decomposition and Kähler class which occurs in
the anticanonical K3 divisors in a given algebraic family of Fano threefolds. The
problem of choosing a matching pair of K3 divisors Di in Fano threefolds Vi then
reduces to a problem in the arithmetic of the K3 lattice.
The solution of this problem gives us the following general result.
Theorem 3.2. For any pair of algebraic families V1 ,V2 of Fano threefolds there
exists (smooth) V1 ∈ V1 , V2 ∈ V2 such that a compact G2 -manifold M can be
constructed from V1 ,V2 .
This G2 -manifold satisfies
0 ≤ b2 (M ) ≤ max{b2 (V1 ), b2 (V2 )} − 1
and
b2 (M ) + b3 (M ) = b3 (V1 ) − KV31 + b3 (V2 ) − KV32 + 27.
Example 3.3. A smooth complete intersection X8 of three quadrics in CP 6

has b2 = 1, b3 = 28, and −K 3 = 8. According to Theorem 3.2, an appropriate
(1) (2)
choice of two such complete intersections V8 , V8 ⊂ CP 6 and of a hyperplane
(i)
section Di in each of the V8 provides data for the construction of a compact G2 -
manifold M . We obtain b2 (M ) = 0 and b3 (M ) = 99. Also M is simply-connected.
This G2 -manifold is not homeomorphic to any of the examples constructed in [4].
3.3. Discussion of the results. There is a complete classification of smooth
Fano threefolds into 104 algebraic families [3, 8]. This provides 5,460 different
pairs to form the generalized connected sums, leading to examples of compact

G2 -manifolds. As any Fano threefold has 1 ≤ b2 ≤ 10, we have
b2 (M ) ≤ 9,
for any G2 -manifold M constructed from a pair of Fanos. Further inspecting the
classification list of Fanos, we find that
39 ≤ b2 (M ) + b3 (M ) ≤ 239
and, in particular, b3 (M ) ≥ 30. (Recall that b1 = 0 for any G2 -manifold, therefore
b2 , b3 determine all the Betti numbers of M .)
It is an interesting question to identify the most general class of Kähler three-
folds W for which the hypotheses of Theorem 2.2 hold. If the anticanonical linear
systems on a pair of such W are ‘large enough’ then the connected sum defined in
Section 2 can be formed and will admit G2 -metrics. It seems that the blow-ups Ṽ
of smooth Fano threefolds discussed in this section can be generalized to include
at least manifolds obtained by resolution of singularities in some singular Fano
varieties.
A pair of Fano threefolds in general yields several topologically distinct compact
G2 -manifolds. Example in [6, §8] shows two topologically distinct G2 -manifolds
constructed from a pair of CP 2 × CP 1 ’s, realizing both of the values b2 (M ) = 0
and b2 (M ) = 1 allowed in this case by Theorem 3.2. Of course, the counting of
pairs of Betti numbers (b2 , b3 ) only gives a lower estimate of the actual number of
topological types realized by our examples of compact G2 -manifolds.
In any event, the majority of smooth Fano threefolds have the Betti number
b2 ≤ 4 and respectively the G2 -manifolds constructed from these have b2 ≤ 3. On
the other hand, a majority of the compact G2 -manifolds constructed in [4] have
b2 > 3. Thus most of the compact G2 -manifolds constructed from smooth Fano
threefolds can be easily identified as new examples, topologically distinct from those
previously known.
Another interesting property of the connected sum construction is that it ex-
hibits a new type of boundary point in the moduli space of all G2 -metrics on the
given compact 7-manifold M . Any 1-parameter family ϕT + dηT of G2 -metrics
given by the gluing Theorem 2.3 defines a path in the moduli space. The boundary
point attained as T → ∞ corresponds to pulling apart a G2 -manifold at a cross-
section K3×(2-torus), obtaining a pair of asymptotically cylindrical pieces. The
approach to the boundary of the moduli space in this case involves no development
of singularities, nor a curvature growth. This becomes important in [7] where we
construct the first examples of fibrations of compact G2 -manifolds by certain min-
imal submanifolds called coassociative calibrated submanifolds. The fibrations are
an odd-dimensional non-holomorphic analogue of the well-known elliptic fibrations
of K3 surfaces.
References
1. W. Barth, C. Peters, and A. van de Ven, Compact complex surfaces. Springer-Verlag, 1984.
2. N. Hitchin, Stable forms and special metrics, Global differential geometry: the mathematical
legacy of Alfred Gray (Bilbao, 2000), Contemp. Math., vol. 288, Amer. Math. Soc., 2001,
pp. 70–89.
3. V.A. Iskovskih, Fano threefolds. I. II, Izv. Akad. Nauk SSSR Ser. Mat. 41 (1977), 516–
562,717, and 42 (1978), 506–549. English translation: Math. USSR Izvestia 11 (1977), 485–
527, and 12 (1978), 469–506.
202 ALEXEI KOVALEV
4. D.D. Joyce, Compact manifolds with special holonomy, OUP Mathematical Monographs
series, 2000.
5. A.G. Kovalev and M.A. Singer, Gluing theorems for complete anti-self-dual spaces, Geom.
Func. Anal. 11 (2001), 1229–1281.
6. A.G. Kovalev, Twisted connected sums and special Riemannian holonomy, J. Reine Angew.
Math. 565 (2003) 125-160.
7. A.G. Kovalev, Coassociative K3 fibrations of compact G2 -manifolds, in preparation.
8. S. Mukai and S. Mori, Classification of Fano threefolds with B2 ≥ 2, Manuscr. Math. 36
(1981/1982), 147–162.
9. S.M. Salamon, Riemannian geometry and holonomy groups, Pitman Res. Notes in Math.
201, Longman, Harlow, 1989.
10. V.V. Shokurov, Smoothness of a general anticanonical divisor on a Fano variety. Izv. Akad.
Nauk SSSR Ser. Mat. 43 (1979), 430–441. English translation: Math. USSR-Izv. 14 (1980),
395–405.
11. G. Tian and S.-T. Yau, Complete Kähler manifolds with zero Ricci curvature. I, J. Amer.
Math. Soc. 3 (1990), 579–609.
12. S.-T. Yau. On the Ricci curvature of a compact Kähler manifold and the complex Monge-
Ampère equation. I, Comm. Pure Appl. Math. 31 (1978), 339–411.
DPMMS, University of Cambridge, Centre for Mathematical Sciences, Wilberforce

Road, Cambridge CB3 0WB, U.K.
E-mail address: a.g.kovalev@dpmms.cam.ac.uk
Volume 3, 2004
An introduction to motivic integration
Alastair Craw
Abstract. By associating a motivic integral to every complex projective vari-

ety X with at worst Gorenstein canonical singularities, Kontsevich [20] proved
that, when a crepant resolution of singularities ϕ : Y → X exists, the Hodge
numbers of Y do not depend upon the choice of the crepant resolution. In
this article we provide an elementary introduction to the theory of motivic
integration, leading to a proof of the result described above. We calculate the
motivic integral of several quotient singularities and discuss these calculations
in the context of the cohomological McKay correspondence.
1. Introduction
The string-theoretic Mirror Symmetry Conjecture states that there exist mirror
pairs of Calabi–Yau varieties with certain compatibilities. For instance, if (X, X ∗ )
is a smooth, projective mirror pair of dimension n, then we expect the relation
(1.1) hp,q (X) = hn−p,q (X ∗ )
to hold between their Hodge numbers. Mirror pairs are not smooth in general
and the compatibility relation (1.1) can fail to hold if either X or its mirror have
singularities. In this case, if there exist crepant resolutions Y → X and Y ∗ → X ∗
then we expect the relation
(1.2) hp,q (Y ) = hn−p,q (Y ∗ )
to hold between the Hodge numbers of the smooth varieties Y and Y ∗ (recall that
a resolution ϕ : Y → X is said to be crepant if KY = ϕ∗ KX ). However, it is not
obvious that the revised relation (1.2) is well defined: even if a crepant resolution
exists it is not necessarily unique. In particular, given two crepant resolutions
Y1 → X and Y2 → X, it is not clear a priori that the Hodge numbers of Y1 and Y2
are equal.
Nevertheless, the consistency of string theory led Batyrev and Dais [5] to con-
jecture that for a variety X with only mild Gorenstein singularities, the Hodge
numbers of Y1 and Y2 are equal. In a subsequent paper [2], Batyrev used meth-
ods of p-adic integration to prove that the Betti numbers of Y1 and Y2 are equal.
2000 Mathematics Subject Classification. Primary 14E15, 32S45 Secondary 14E30, 14M25.
Key words and phrases. Motivic integration, Space of formal arcs, McKay correspondence.
2004
203
204 ALASTAIR CRAW
Kontsevich [20] later proved that the Hodge numbers are equal by introducing the
notion of motivic integration.
This article provides an elementary introduction to Kontsevich’s theory of mo-
tivic integration. The first step is to construct the motivic integral of a pair (Y, D),
for a complex manifold Y and an effective divisor D on Y with simple normal
crossings. We define the space of formal arcs J∞ (Y ) of Y and associate a function
FD defined on J∞ (Y ) to the divisor D. The motivic integral of the pair (Y, D) is
the integral of FD over J∞ (Y ) with respect to a certain measure µ on J∞ (Y ). This
measure is not real-valued. Indeed the subtlety in the construction is in defining
the ring R in which µ takes values: R is a completion of the polynomial ring in the
formal variable L−1 with values in the Grothendieck ring of algebraic varieties over
C (see Definition 2.11). We adopt the structure of the proof of Theorem 6.28 from
Batyrev [1] to establish the following user-friendly formula:
Theorem 1.1 (formula for the motivic integral). Let Y be a complex manifold
r
of dimension n and D = i=1 ai Di an effective divisor on Y with simple normal
crossings. The motivic integral of the pair (Y, D) is
 
L−1
(1.3) FD dµ = [DJ◦ ] ·   · L−n
J∞ (Y ) Laj +1 − 1
J⊆{1,...,r} j∈J
where we sum over all subsets J ⊆ {1, . . . , r} including J = ∅.

The motivic integral of a complex algebraic variety X with Gorenstein canonical
singularities is defined to be the motivic integral of a pair (Y, D), where Y → X is
a resolution of singularities for which the discrepancy divisor D has simple normal
crossings. Crucially, this is well defined independent of the choice of resolution.
The motivic integral induces a stringy E-function
 
uv − 1
(1.4) Est (X) := E(DJ◦ ) ·  
(uv)aj +1 − 1
J⊆{1,...,r} j∈J
which is also independent of the choice of resolution (see Warning 3.4). The E-
polynomials E(DJ◦ ) encode the Hodge–Deligne numbers of open strata DJ◦ ⊂ Y ,
and the stringy E-function records these numbers with certain ‘correction terms’
written in parentheses in formula (1.4). When Y → X is a crepant resolution the
correction terms disappear leaving simply the terms E(DJ◦ ) whose sum is the E-
polynomial of Y . As a result, when a crepant resolution Y → X exists, the function
Est (X) encodes the Hodge numbers of Y , thereby establishing Kontsevich’s result
on the equality of Hodge numbers. It is important to note however that crepant
resolutions do not exist in general. To get a better feeling for the stringy E-function
of varieties admitting no crepant resolution we calculate Est (X) for several 4- and
6-dimensional Gorenstein terminal cyclic quotient singularities.
We conclude this article with an application of motivic integration, namely
the proof by Batyrev [3, 4] of the cohomological McKay correspondence conjec-
ture which we now recall. A celebrated result of John McKay [22] states that the
graph of ADE type associated to a Kleinian singularity C2 /G can be constructed
using only the representation theory of the finite subgroup G ⊂ SL(2, C). This
establishes a one-to-one correspondence between a basis for the cohomology of the
AN INTRODUCTION TO MOTIVIC INTEGRATION 205
minimal resolution Y of C2 /G and the irreducible representations of G. In partic-

ular, the Euler number e(Y ) is equal to the number of irreducible representations
(or conjugacy classes) of G. Reid [27] proposed the following generalisation:
Conjecture 1.2 (The McKay correspondence). For G ⊂ SL(n, C) a finite
subgroup, suppose that the quotient variety X := Cn /G admits a crepant resolution
ϕ : Y → X. Then H ∗ (Y, Q) has a basis consisting of algebraic cycles corresponding
one-to-one with conjugacy classes of G. In particular, the Euler number of Y equals
the number of conjugacy classes of G.
Batyrev proved this result using the theory of motivic integration. The key step
in the proof is to show that the stringy E-function of the quotient Cn /G coincides
with the ‘orbifold E-function’ of the pair (Cn , G) (see §5.2 for the definition). We
present a simple, direct proof of the conjecture for a finite Abelian subgroup G ⊂
SL(n, C) to illustrate the simplicity of Batyrev’s approach in this case.
The original references on motivic integration are by Batyrev [1, §6] and Denef
and Loeser [14]. The more recent article by Looijenga [21] provides a detailed
survey of motivic integration.
Acknowledgements I’d like to thank the organisers of the 2002 Clay Mathematics
Institute School on Geometry and String Theory held at the Isaac Newton Institute
in Cambridge, especially Miles Reid who originally encouraged me to write this
article (which has existed in roughly the present form since November 1998). Also,
thanks to Willem Veys and to Victor Batyrev for useful comments.
2. Construction of the motivic integral

2.1. The space of formal arcs of a complex manifold.
Definition 2.1. Let Y be a complex manifold of dimension n. A k-jet over a
point y ∈ Y is a morphism
γy : Spec C[z]/z k+1 −→ Y
with γy (Spec C) = y. Once local coordinates are chosen, the space of k-jets over
y ∈ Y can be viewed as the space of n-tuples of polynomials of degree k whose
constant terms are zero. Let Jk (Y ) denote the bundle over Y whose fibre over
y ∈ Y is the space of k-jets over y. A formal arc over y ∈ Y is a morphism
γy : Spec C [[z]] −→ Y
with γy (Spec C) = y. Once local coordinates are chosen, the space of formal arcs
over y ∈ Y can be viewed as the space of n-tuples of power series whose constant
terms are zero. Let J∞ (Y ) denote the bundle whose fibre over y ∈ Y is the space
of formal arcs over y. For each k ∈ Z≥0 the inclusion C[z]/z k+1 → C [[z]] induces
a surjective map
πk : J∞ (Y ) −→ Jk (Y ),
where π0 : J∞ (Y ) → Y sends γy to the point y.
Recall that a subset of a variety is constructible if it is a finite, disjoint union
of (Zariski) locally closed subvarieties.
Definition 2.2. A subset C ⊆ J∞ (Y ) of the space of formal arcs is called a
cylinder set if C = πk−1 (Bk ) for k ∈ Z≥0 and Bk ⊆ Jk (Y ) a constructible subset.
206 ALASTAIR CRAW
It’s clear that the collection of cylinder sets forms an algebra of sets (see [28,
p. 10]), i.e., J∞ (Y ) = π0−1 (Y ) is a cylinder set, as are finite unions and complements
(and hence finite intersections) of cylinder sets.
2.2. The function FD associated to an effective divisor.

Definition 2.3. Let D be an effective divisor on Y and g a local defining
equation for D on a neighbourhood U of a point y ∈ Y . For an arc γu over a point
u ∈ U , define the intersection number γu · D to be the order of vanishing of the
formal power series g(γu (z)) at z = 0. Let
FD : J∞ (Y ) −→ Z≥0 ∪ ∞

be the function defined by FD (γu ) = γu · D. Write D = ri=1 ai Di
as a linear
combination of prime divisors. Then gdecomposes as a product g = ri=1 giai of
defining equations for Di , hence FD = ri=1 ai FDi . Furthermore
(2.1) FDi (γu ) = 0 ⇐⇒ u ∈
/ Di and FDi (γu ) = ∞ ⇐⇒ γu ⊆ Di .
Our ultimate goal is to integrate the function FD over J∞ (Y ), so we must
−1
understand the nature of the level set FD (s) ⊆ J∞ (Y ) for each s ∈ Z≥0 ∪ ∞.
−1
With this goal in mind, we introduce a partition of FD (s).
r
Definition 2.4. For D = i=1 ai Di and J ⊆ {1, . . . , r} any subset, define

j∈J Dj if J = ∅
DJ := and DJ◦ := DJ \ Di .
Y if J = ∅
i∈{1,...,r}\J
These subvarieties stratify Y and define a partition of the space of arcs into cylinder
sets:
Y = DJ◦ and J∞ (Y ) = π0−1 (DJ◦ ).
J⊆{1,...,r} J⊆{1,...,r}
For any s ∈ Z≥0 and any subset J ⊆ {1, . . . , r}, define

MJ,s := (m1 , . . . , mr ) ∈ Zr≥0 ai mi = s with mj > 0 ⇔ j ∈ J .
It now follows from (2.1) that

γu ∈ π0−1 (DJ◦ ) ∩ FD
−1
(s) ⇐⇒ (FD1 (γu ), . . . , FDr (γu )) ∈ MJ,s .
As a result we produce a finite partition of the level set

−1 −1
(2.2) FD (s) = i=1,...r FDi (mi ) .
J⊆{1,...,r} (m1 ,...,mr )∈MJ,s
Proposition 2.5. If D is an effective divisor with simple normal crossings

−1
then FD (s) is a cylinder set (see Definition 2.2) for each s ∈ Z≥0 .

Recall (see [19, p. 25]) that a divisor D = ri=1 ai Di on Y has only simple
normal crossings if at each point y ∈ Y there is a neighbourhood U of y with
coordinates z1 , . . . , zn for which a local defining equation for D is
aj
(2.3) g = z1a1 · · · zjy y for some jy ≤ n.
Proof of Proposition 2.5. A finite union of cylinder sets is cylinder and we

−1 −1
have a partition (2.2) of FD (s), so it is enough to prove that i=1,...r FD i
(m i ) is a
cylinder set1 for some J ⊆ {1, . . . , r} and (m1 , . . . , mr ) ∈ MJ,s . Cover Y by finitely
many charts
U on which D has a local equation of the form (2.3), and lift to cover
J∞ (Y ) = π0−1 (U ). We need only prove that the set
−1
Um1 ,...,mr := i=1,...r FD i
(mi ) ∩ π0−1 (U )
is cylinder. In the notation of (2.3), if J ⊆ {1, . . . , jy } then DJ◦ ∩ U = ∅ which

forces Um1 ,...,mr ⊂ π0−1 (DJ◦ ∩ U ) to be empty, and hence a cylinder set. We suppose
therefore that J ⊆ {1, . . . , jy }, thus |J| ≤ n holds by (2.3).
The key observation is that by regarding arcs γu as n-tuples (p1 (z), . . . , pn (z))
of formal power series with zero constant terms, the equality FDi (γu ) = mi is
equivalent to a condition on the truncation of the power series pi (z) to degree mi .
Indeed, since Di is cut out by zi = 0 on U , it follows that FDi (γu ) equals the order
−1
of vanishing of pi (z) at z = 0. Thus γu ∈ FD i
(mi ) if and only if the truncation
of pi (z) to degree mi is of the form cmi z , with cmi = 0. Truncating all n of the
mi
power series to degree t := max{mj j ∈ J} produces n − |J| polynomials of degree

t with zero constant term and, for each j ∈ J, a polynomial of the form
πt (pj (z)) = 0 + · · · + 0 + cmj z mj + c(mj +1) z mj +1 + · · · + ct z t ,

where cmj ∈ C∗ and ck ∈ C for all k > mj . The space of all such n-tuples is
isomorphic to Ct(n−|J|) × (C∗ )|J| × Ct|J|− j∈J mj , hence

(2.4) Um1 ,...,mr = πt−1 (U ∩ DJ◦ ) × Ctn− j∈J mj × (C∗ )|J| .

The set (U ∩ DJ◦ ) × Ctn− j∈J mj × (C∗ )|J| is constructible, so Um1 ,...,mr is a cylinder
set. This completes the proof of the proposition.
−1
It is worth noting that FD (∞) is not a cylinder set. Indeed, suppose otherwise,
−1
so there exists a constructible subset Bk ⊆ Jk (Y ) for which FD (∞) = πk−1 (Bk ).
−1
Each arc γy ∈ FD (∞) is an n-tuple of power series, at least one of which is
identically zero, whereas each γy ∈ πk−1 (Bk ) is an n-tuple of power series whose
terms of degree higher than k may take any complex value, a contradiction.
−1
Proposition 2.6. FD (∞) is a countable intersection of cylinder sets.
Proof. Observe that
−1
−1
(2.5) FD (∞) = πk−1 πk FD (∞)
k∈Z≥0
because a power series is identically zero if and only if its truncation to degree k is
−1
the zero polynomial, for all k ∈ Z≥0 . It is easy to see that the sets πk (FD (∞)) ⊂
Jk (Y ) are constructible.
2.3. A measure µ on the space of formal arcs. In this section we define

a measure µ on J∞ (Y ) with respect to which the function FD is measurable. The
measure is not real-valued, so we begin by constructing the ring in which µ takes
values.
1Finite intersections of cylinder sets are cylinder, so we could reduce to proving the result
−1
for FD (mi ). However we require (2.4) in §2.4.
i
208 ALASTAIR CRAW
Definition 2.7. Let VC denote the category of complex algebraic varieties.

The Grothendieck group of VC is the free Abelian group on the isomorphism classes
[V ] of complex algebraic varieties modulo the subgroup generated by elements of
the form [V ] − [V ] − [V \ V ] for a closed subset V ⊆ V . The product of varieties
induces a ring structure [V ] · [V ] = [V × V ], and the resulting ring denoted by
K0 (VC ) is the Grothendieck ring of complex algebraic varieties. Let
[ ] : Ob VC −→ K0 (VC )
denote the natural map sending V to its class [V ] in the Grothendieck ring. This
map is universal with respect to maps which are additive on disjoint unions of
constructible subsets, and which respect products.
Write 1 := [point] and L := [C] (see Appendix A: the class of C in K0 (VC )
corresponds to the Lefschetz motive L). Then
[C∗ ] = [C] − [{0}] = L − 1.
Also, if f : Y → X is a locally trivial fibration w.r.t. the Zariski topology and F is
the fibre over a (closed) point then [Y ] = [F × X].
Definition 2.8. Let K0 (VC )[L−1 ] := S −1 K0 (VC ) denote the ring of fractions
of K0 (VC ) with respect to the multiplicative set S := {1, L, L2 , . . . }.
Definition 2.9. Recall that cylinder sets in J∞ (Y ) are subsets πk−1 (Bk ) ⊂
J∞ (Y ) for k ∈ Z≥0 and for Bk ⊆ Jk (Y ) a constructible subset. The function

: cylinder sets in J∞ (Y ) −→ K0 (VC )[L−1 ]
µ
defined by
πk−1 (Bk ) → [Bk ] · L−n(k+1)
assigns a ‘measure’ to each cylinder set.
It is straightforward to show that
l
l
i=1 Ci =
µ (Ci )
µ for cylinder sets C1 , . . . , Cl
i=1
because the map [ ] introduced in Definition 2.7 is additive on disjoint unions of

a finitely additive measure.
constructible sets. For this reason we call µ
−1
Remark 2.10. Proposition 2.5 states that for s ∈ Z≥0 , the level set FD (s) is a
−1
-measurable, i.e., µ
cylinder set, and is therefore µ (FD (s)) is well defined. However,
−1
FD is not µ-measurable because FD (∞) is not cylinder. To proceed, we extend µ
−1
to a measure µ with respect to which FD (∞) is measurable.
The following discussion is intended to motivate the definition of µ (see Def-
−1
inition 2.12 to follow). The set J∞ (Y ) \ FD (∞) is a countable disjoint union of
cylinder sets

(2.6) J∞ (Y ) \ π0−1 π0 (FD
−1
(∞)) πk−1 πk (FD
−1 −1
(∞)) \ πk+1 −1
πk+1 (FD (∞)) .
k∈Z≥0
To see this, take complements in equation (2.5). Our goal is to extend µ to a

measure µ defined on the collection of countable disjoint unions of cylinder sets so
−1 −1
that the set J∞ (Y ) \ FD (∞), and hence its complement FD (∞), is µ-measurable.
One would like to define

(2.7) µ i∈N Ci := µ(Ci ) = (Ci ) for cylinder sets C1 , . . . , Cl .
µ
i∈N i∈N
However, countable sums are not defined in K0 (VC )[L−1 ]. Furthermore, given a
countable disjoint union C = i∈N Ci , it is not clear a priori that µ(C) defined by
formula (2.7) is independent of the choice of the Ci .
Kontsevich [20] solved both of these problems at once by completing the ring
K0 (VC )[L−1 ], thereby allowing
appropriate countable sums, in such a way that the
measure of the set C = i∈N Ci is independent of the choice of the Ci , assuming
that µ(Ci ) → 0 as i → ∞.
Definition 2.11. Let R denote the completion of the ring K0 (VC )[L−1 ] with
respect to the filtration
· · · ⊇ F −1 K0 (VC )[L−1 ] ⊇ F 0 K0 (VC )[L−1 ] ⊇ F 1 K0 (VC )[L−1 ] ⊇ · · ·
where for each m ∈ Z, F m K0 (VC )[L−1 ] is the subgroup of K0 (VC )[L−1 ] generated
by elements of the form
[V ] · L−i for i − dim V ≥ m.
The natural completion map is denoted φ : K0 (VC )[L−1 ] → R.
By composing µ with the natural completion map φ, we produce a finitely

additive measure with values in the ring R, namely φ ◦ µ (which we will also denote
) that sends
µ

πk−1 (Bk ) → φ [Bk ] · L−n(k+1) .
Given a sequence of cylinder sets {Ci } one may now ask whether or not µ
(Ci ) → 0
as i → ∞. We are finally in a position to define the measure µ on the space of
formal arcs.
Definition 2.12. Let C denote the collection of countable disjoint unions of

(Ci ) → 0 as i → ∞, together with the complements
cylinder sets i∈N Ci for which µ
of such sets. Extend µ to a measure
µ : C −→ R
by defining

Ci −→ (Ci ).
µ
i∈N i∈N
It is nontrivial to show (see [14, §3.2] or [1, §6.18]) that this definition is independent
of the choice of the Ci .
−1
Proposition 2.13. FD is µ-measurable, and µ(FD (∞)) = 0.
−1
Proof. We prove that FD (∞) (in fact its complement) lies in C. It’s clear
from (2.6) that we need only prove that µ(πk−1 πk (FD
−1
(∞))) → 0 as k → ∞.
−1 −1
Lemma 2.14 below reveals that µ(πk πk (FD (∞))) ∈ φ(F k+1 K0 (VC )[L−1 ]) which,
210 ALASTAIR CRAW
by the nature of the topology on R, tends to zero as k tends to infinity. This proves
the first statement. Using (2.6) we calculate
−1

(2.8) µ J∞ (Y ) \ FD (∞) = µ J∞ (Y ) \ π0−1 π0 (FD−1
(∞))

+ µ πk−1 πk (FD −1 −1
(∞)) \ πk+1 −1
πk+1 (FD (∞)) .
k∈Z≥0

This equals µ(J∞ (Y )) − limk→∞ µ πk−1 πk (FD−1
(∞)) . By the above remark, this
−1
is simply µ(J∞ (Y )), so µ(FD (∞)) = 0 as required.
(πk−1 πk (FD
Lemma 2.14. µ −1
(∞))) ∈ F k+1 K0 (VC )[L−1 ]
−1
Proof. It is enough to prove the result for a prime divisor D, since FD (∞) is
−1
the union of sets FDi (∞). Choose coordinates on a chart U in which D is (z1 = 0).
−1
Each γy ∈ FD (∞) ∩ π0−1 (U ) is an n-tuple (p1 (z), . . . , pn (z)) of power series over
y ∈ U ∩ D such that p1 (z) is identically zero. Truncating these power series to
degree k leaves n − 1 polynomials of degree k with zero constant term, and the zero
polynomial πk (p1 (z)). The space of all such polynomials is isomorphic to C(n−1)k , so
−1
that πk (FD (∞)∩π0−1 (U )) (U ∩D)×C(n−1)k . Thus [πk (FD −1
(∞)] = [D]·[C(n−1)k ]
and
(πk−1 πk (FD
µ −1 −1
(∞)) = [πk (FD (∞)] · L−n(k+1)
= [D] · L(n−1)k · L−n(k+1)
= [D] · L−(n+k)
This lies in F k+1 K0 (VC )[L−1 ] since D has dimension n − 1.
2.4. The motivic integral of a pair (Y, D).

Definition 2.15. Let rY be a complex manifold of dimension n, and choose
an effective divisor D = i=1 ai Di on Y with only simple normal crossings. The
motivic integral of the pair (Y, D) is
−1 −s
FD dµ := µ FD (s) · L .
J∞ (Y ) s∈Z≥0 ∪∞
−1
Since the set FD (∞) ⊂ J∞ (Y ) has measure zero (see Proposition 2.13), we need
−1
only integrate over J∞ (Y ) \ FD (∞), so we need only sum over s ∈ Z≥0 .
We now show that the motivic integral converges in the ring R introduced in
Definition 2.11. In doing so, we establish a user-friendly formula.
Theorem 2.16 (formula for the motivic integral). Let Y be a complex manifold
r
of dimension n and D = i=1 ai Di an effective divisor on Y with only simple
normal crossings. The motivic integral of the pair (Y, D) is
 
L−1
FD dµ = [DJ◦ ] ·   · L−n
J∞ (Y ) Laj +1 − 1
J⊆{1,...,r} j∈J

Proof. In the proof of Proposition 2.5 we cover Y by sets {U } and prove that
−1 −1
F (m i ) ∩ π0 (U ) is a cylinder set of the form

i=1,...r Di

πt−1 (U ∩ DJ◦ ) × Ctn− j∈J mj × (C∗ )|J| .
Since the map [ ] introduced in Definition 2.7 is additive on a disjoint union
of
constructible subsets, take the union over the cover {U } of Y to see that
−1 −1
F (m i = πt (Bt ) where
)

i=1,...r Di

|J|
[Bt ] = DJ◦ × Ctn− j∈J mj × (C∗ ) = [DJ◦ ] · Ltn− j∈J mj · (L − 1)|J| .

Since µ πt−1 (Bt ) = [Bt ] · L−(n+nt) , we have

−1

◦

− j∈J mj
µ i=1,...r FDi (mi ) = [DJ ] · L · (L − 1)|J| · L−n .
−1
Now use the partition (2.2) of FD (s) to compute the motivic integral:

−1 −s
µ FD (s) · L

s∈Z≥0

−1
= µ F
i=1,...r Di (m i ) · L− j∈J aj mj
s∈Z≥0 J⊂{1,...,r} (m1 ,...,mr )∈MJ,s

= [DJ◦ ] · (L − 1)|J| · L−n · L−(aj +1)mj
s∈Z≥0 J⊂{1,...,r} (m1 ,...,mr )∈MJ,s j∈J

= [DJ◦ ] · (L − 1) · mj >0 L−(aj +1)mj · L−n
J⊂{1,...,r} j∈J

1

= [DJ◦ ] · (L − 1) · −1 · L−n
j∈J
1 − L−(aj +1)
J⊂{1,...,r}
 
L − 1  −n
= [DJ◦ ] ·  ·L .
Laj +1 − 1
J⊂{1,...,r} j∈J

Warning 2.17. There is a small error in the proof of the corresponding result
in Batyrev [1, §6.28] which leads to the omission of the L−n term.
Corollary 2.18. The motivic integral of the pair (Y, D) lies in the subring

1
φ K0 (VC )[L−1 ]
Li − 1 i∈N
of the ring R introduced in Definition 2.11.
2.5. The transformation rule for the integral. The discrepancy divisor
W := KY − α∗ KY of a proper birational morphism α : Y → Y between smooth
varieties is the divisor of the Jacobian determinant of α. The next result may
therefore be viewed as the ‘change of variables formula’ for the motivic integral.
Theorem 2.19. Let α : Y −→ Y be a proper birational morphism between
smooth varieties and let W := KY − α∗ KY be the discrepancy divisor. Then

FD dµ = Fα∗ D+W dµ .
J∞ (Y ) J∞ (Y )
212 ALASTAIR CRAW
Proof. Composition defines maps αt : Jt (Y ) → Jt (Y ) for each t ∈ Z≥0 ∪ ∞.

An arc in Y which is not contained in the locus of indeterminacy of α−1 has a
birational transform as an arc in Y . In light of (2.1) and Proposition 2.13, α∞ is
bijective off a subset of measure zero.
−1 −1
The sets FW (k), for k ∈ Z≥0 , partition J∞ (Y ) \ FW (∞). Thus, for any
−1
s ∈ Z≥0 we have, modulo the set FW (∞) of measure zero, a partition

−1 −1
(2.9) FD (s) = α∞ (Ck,s ) where Ck,s := FW (k) ∩ Fα−1
∗ D (s).
k∈Z≥0
The set Ck,s is cylinder and, since the image of a constructible set is constructible
(
[24,
p. 72]), the set α∞ (C k,s ) is cylinder. Lemma 2.20 below states that µ C k,s =
µ α∞ (Ck,s ) · Lk . We use this identity and the partition (2.9) to calculate

FD dµ = µ α∞ (Ck,s ) · L−s = µ Ck,s · L−(s+k) .
J∞ (Y ) k,s∈Z≥0 k,s∈Z≥0

Set s := s + k. Clearly 0≤k≤s Ck,s −k = Fα−1
∗ D+W (s ).
Substituting this into the
above leaves

−1
−s
FD dµ = µ Fα∗ D+W (s ) · L = Fα∗ D+W dµ,
J∞ (Y ) s ∈Z≥0 J∞ (Y )
as required.

Lemma 2.20. µ Ck,s = µ α∞ (Ck,s ) · Lk .
Discussion of proof. Both Ck,s and α∞ (Ck,s ) are cylinder sets so there exists
t ∈ Z≥0 and constructible sets Bt and Bt in J∞ (Y ) and J∞ (Y ) respectively such
that the following diagram commutes:
α
Ck,s ⊂ J∞ (Y ) ∞
−→ α∞ (Ck,s) ⊂ J∞ (Y )
 
πt ! ! πt
α
Bt ⊂ Jt (Y ) −→
t
Bt ⊂ Jt (Y ).
We claim that the restriction of αt to Bt is a Ck -bundle over Bt . It follows that
[Bt ] = [Ck ] · [Bt ] and we have

µ Ck,s = [Bt ] · L−(n+nt) = [Bt ] · Lk · L−(n+nt) = µ α∞ (Ck,s ) · Lk
as required. The proof of the claim is a local calculation which is carried out in [14,
Lemma 3.4(b)]. The key observation is that the order of vanishing of the Jacobian
determinant of α at γy ∈ Ck,s is FW (γy ) = k.
Definition 2.21. Let X denote a complex algebraic variety with at worst
Gorenstein canonical singularities. The motivic integral of X is defined to be the
motivic integral of the pair (Y, D), where ϕ : Y → X is any resolution of singularities
for which the discrepancy divisor D = KY −ϕ∗ KX has only simple normal crossings.
Note first that the discrepancy divisor D is effective because X has at worst
Gorenstein canonical singularities. The crucial point however is that the motivic
integral of (Y, D) is independent of the choice of resolution:
Proposition 2.22. Let ϕ1 : Y1 −→ X and ϕ2 : Y2 −→ X be resolutions of X
with discrepancy divisors D1 and D2 respectively. Then the motivic integrals of the
pairs (Y1 , D1 ) and (Y2 , D2 ) are equal.
Proof. Form a ‘Hironaka hut’

ψ2
Y0 −→ Y2
 
ψ1 ! ! ϕ2
ϕ1
Y1 −→ X
and let D0 denote the discrepancy divisor of ϕ0 : Y0 −→ X. The discrepancy divisor
of ψi is D0 − ψi∗ Di . Indeed
KY0 = ϕ∗0 (KX ) + D0 = ψi∗ ◦ ϕ∗i (KX ) + D0 = ψi∗ (KYi − Di ) + D0
= ψi∗ (KYi ) + (D0 − ψi∗ Di ).
The maps ψi : Y0 −→ Yi are proper birational morphisms between smooth projec-
tive varieties so Theorem 2.19 applies:

FDi dµ = Fψi∗ Di +(D0 −ψi∗ Di ) dµ = FD0 dµ .
J∞ (Yi ) J∞ (Y0 ) J∞ (Y0 )
This proves the result.
3. Hodge numbers via motivic integration

This section describes how the motivic integral of X gives rise to the so-called
“stringy E-function” which encodes the Hodge–Deligne numbers of a resolution
Y → X.
3.1. Encoding Hodge–Deligne numbers. Deligne [8, 9] showed that the
cohomology groups H k (X, Q) of a complex algebraic variety X carry a natural
mixed Hodge structure. This consists of an increasing weight filtration
0 = W−1 ⊆ W0 ⊆ · · · ⊆ W2k = H k (X, Q)
on the rational cohomology of X and a decreasing Hodge filtration
H k (X, C) = F 0 ⊇ F 1 ⊇ · · · ⊇ F k ⊇ F k+1 = 0
on the complex cohomology of X such that the filtration induced by F • on the
graded quotient GrW k
l H (X) := Wl /Wl−1 is a pure Hodge structure of weight l.
Thus
Grl H k (X) ⊗ C = F p GrW
l H (X) ⊕ F
k l−p+1 GrW H k (X)
l
l H (X) denotes the complexified image of F ∩ Wl in the quotient

where F p GrW k p
Wl /Wl−1 ⊗ C. The integers

hp,q H k (X, C) := dimC F p GrW
p+q H (X) ∩ F Grp+q H (X)
k q W k
are called the Hodge–Deligne numbers of X. For a smooth projective variety X

over C, GrWl H (X, Q) = 0 unless l = k in which case the Hodge–Deligne numbers
k
are the classical Hodge numbers hp,q (X).

Danilov and Khovanskii [12] observed that cohomology with compact support
Hck (X, Q) also admits a mixed Hodge structure and they encode the corresponding
Hodge–Deligne numbers in a single polynomial:
Definition 3.1. The E-polynomial E(X) ∈ Z[u, v] of a complex algebraic
variety X of dimension n is defined to be

E(X) := (−1)k hp,q Hck (X, C) up v q .
0≤p,q≤n 0≤k≤2n
214 ALASTAIR CRAW
Evaluating E(X) at u = v = 1 produces the standard topological Euler number

ec (X) = e(X).
Theorem 3.2 ([12]). Let X, Y be complex algebraic varieties. Then

(i) if X = Xi is stratified by a disjoint union of locally
closed subvarieties
then the E-polynomial is additive, i.e., E(X) = E(Xi ).
(ii) the E-polynomial is multiplicative, i.e., E(X × Y ) = E(X) · E(Y ).
(iii) if f : Y → X is a locally trivial fibration w.r.t. the Zariski topology and F
is the fibre over a closed point then E(Y ) = E(F ) · E(X).
See Danilov and Khovanskii [12] for a proof.
3.2. Kontsevich’s theorem. Theorem 3.2 asserts that the map E : VC −→

Z[u, v] associating to each complex variety X its E-polynomial is additive on a
disjoint union of locally closed subvarieties, and satisfies E(X × Y ) = E(X) · E(Y ).
It follows from the universality of the map [ ] introduced in Definition 2.7 that
E factors through the Grothendieck ring of algebraic varieties, inducing a function
E : K0 (VC ) → Z[u, v]. By defining E(L−1 ) := (uv)−1 , this extends to2
E : K0 (VC )[L−1 ] → Z[u, v, (uv)−1 ].
Proposition 3.3. The map E can be extended uniquely to the subring

−1
1
φ K0 (VC )[L ]
Li − 1 i∈N
of the ring R introduced in Definition 2.11.
Proof. The kernel of the completion map φ : K0 (VC )[L−1 ] → R is

(3.1) F m K0 (VC )[L−1 ].
m∈Z

For [V ] · L−i ∈ F m K0 (VC )[L−1 ], the degree of the E-polynomial E [V ] · L−i is
2 dim V − 2i ≤ −2m. The E-polynomial of an element Z in the intersection (3.1)
must therefore be −∞; that is, E(Z) = 0. Thus E annihilates ker φ and hence
factors through φ K0 (VC )[L−1 ] . Defining E(1/(Li − 1)) := 1/((uv)i − 1) for i ∈ N
establishes the result.
By Corollary 2.18 the motivic integral of the pair (Y, D) lies in the subring of
Proposition 3.3. We now consider the image of the integral under E.
Warning 3.4. As Warning 2.17 states, the derivation of the motivic integral
in [1] contains a small error which leads to the omission of an L−n term. However,
in practise it is extremely convenient to omit this term (!). As a result, we define
the stringy E-function to be the image under E of the motivic integral times Ln . In
short, our stringy E-function agrees with that in [1], even though our calculation
of the motivic integral differs.
2One can use this function to define a finitely additive [u, v, (uv)−1 ]–valued measure µ :=
E
E ◦ µ on cylinder sets given by πk−1 (Bk ) → E(Bk ) · (uv)−n(k+1) , then construct the stringy
E-function directly. This is the approach adopted by Batyrev [1, §6].
Definition 3.5. Let X be a complex algebraic variety of dimension n with

at worst Gorenstein canonical singularities. Let ϕ : Y → X be a resolution of
r
singularities for which the discrepancy divisor D = i=1 ai Di has only simple
normal crossings. The stringy E-function of X is
" #  
uv − 1
Est (X) := E FD dµ · Ln = E(DJ◦ ) ·  ,
J∞ (Y ) (uv)aj +1 − 1
J⊆{1,...,r} j∈J

Theorem 3.6 ( [20]). Let X be a complex projective variety with at worst
Gorenstein canonical singularities. If X admits a crepant resolution ϕ : Y → X
then the Hodge numbers of Y are independent of the choice of crepant resolution.
r
Proof. The discrepancy divisor D = i=1 ai Di of the crepant resolution
ϕ : Y → X is by definition zero, so the motivic integral of X is the motivic integral
of the pair (Y, 0). Since each ai = 0 it’s clear that

Est (X) = J⊆{1,...,r} E(DJ◦ ) = E(Y ).
The stringy E-function is independent of the choice of the resolution ϕ. In par-
ticular, E(Y ) = Est (X) = E(Y2 ) for ϕ2 : Y2 → X another crepant resolution. It
remains to note that E(Y ) determines the Hodge–Deligne numbers of Y , and hence
the Hodge numbers since Y is smooth and projective.
4. Calculating the motivic integral

To perform nontrivial calculations of the stringy E-function we must choose
varieties which admit no crepant resolution. A nice family of examples is provided
by Gorenstein terminal cyclic quotient singularities.
4.1. Toric construction of cyclic quotient singularities. Consider the

action of the cyclic group G = Z/r ⊂ GL(n, C) generated by the diagonal matrix3

g = diag e2πiα1 /r , . . . , e2πiαn /r with 0 ≤ αj < r,
√
where i = −1. The quotient Cn /G is the cyclic quotient singularity of type
r (α1 , . . . , αn ). This fractional notation derives from the construction of C /G as
1 n
an affine toric variety as we now describe (see Reid [26, §4] for more details).
Write M ∼ = Zn for the lattice of Laurent monomials in x1 , . . . , xn , and N for
the dual lattice with basis e1 , . . . , en . Let σ = R≥0 e1 + · · · + R≥0 en denote the
positive orthant in N ⊗ R with dual cone σ ∨ ⊂ M ⊗ R. The overlattice
(4.1) N := N + Z · 1r (α1 , . . . , αn )
is dual to M := Hom(N, Z). A Laurent monomial in x1 , . . . , xn lies in the sublat-
tice M ⊂ M if and only if it is invariant under the action of the group G. Re-
stricting to Laurent monomials with only nonnegative powers leads to the equality
C[x1 , . . . , xn ]G = C[σ ∨ ∩ M ], and hence
Cn /G = Spec C[x1 , . . . , xn ]G = Spec C[σ ∨ ∩ M ] =: Uσ .
j , . . . , αn ) = 1 for all j = 1, . . . , n to ensure

3It is convenient to assume that gcd(r, α , . . . , α
1
that the group action is ‘small’. The notation αj means that αj is omitted.
216 ALASTAIR CRAW
In order to consider only Gorenstein terminal cyclic quotient singularities we impose

certain restrictions on the type 1r (α1 , . . . , αn ). Watanabe [31] showed that for a
small subgroup G ⊂ GL(n, C), the quotient Cn /G is Gorenstein if and only if
G ⊂ SL(n, C). Thus,
n
Uσ is Gorenstein ⇐⇒ j=1 αj ≡ 0 mod r.
To determine when Uσ is terminal we recall the discrepancy calculation for cyclic

quotients following Reid [25, 26] (see also Craw [7]). Write for the unit box in
N ⊗R ∼= Rn , i.e., the unit cell of the sublattice N ∼ = Zn . Each element g ∈ G ∼
= N/N
has a unique representative

1
vg = r(g) α1 (g), . . . , αn (g) ∈ N ∩ ;
here vg denotes both the vector in N ⊗ R and the lattice point in N . For each
primitive vector vg ∈ N ∩ , the simplicial subdivision of σ at vg determines a toric
blow-up ϕ : Y = XΣ → Uσ = X of the cyclic quotient. The exceptional divisor is
the closed toric stratum4 V (τ ) ⊂ Y , where τ is the ray with primitive generator vg .
Adjunction for the toric blow-up ϕ is
 
1 n
(4.2) KY = ϕ ∗ K X +  αj (g) − 1 V (τ ),
r(g) j=1
where the equality here denotes numerical equivalence. Therefore

n
1
Uσ is terminal ⇐⇒ αj (g) > 1 for each g ∈ G.
j=1
r(g)
That is, Uσ is terminal when every point vg ∈ N ∩ lies above the hyperplane

xi = 1.
Before proceeding to the examples we recall the simple formula that computes
the E-polynomial of the toric variety XΣ determined by a fan Σ in N ⊗ R.
Proposition 4.1. For a toric variety XΣ of dimension n we have
n
n−k
(4.3) E XΣ = dk · uv − 1 ,
k=0
where dk is the number of cones of dimension k in Σ.
Proof. The Hodge numbers of P1 are well known and, by Theorem 3.2, we
compute E(C∗ ) = E(P1 ) − E({0}) − E({∞}) = uv − 1. The E-polynomial is
n−k
multiplicative so E (C∗ )n−k = uv−1 . The action of the torus Tn (C∗ )n on
XΣ induces a stratification of XΣ into orbits of the torus action Oτ ∼= (C∗ )
n−dim τ
,
one for each cone τ ∈ Σ. The result follows from Theorem 3.2(i).
4The cones in Σ containing τ as a face define a fan denoted Star(τ ) in the vector space
N (τ ) ⊗ , where N (τ ) := N/(τ ∩ N ). The toric variety V (τ ) := XStar(τ ) is the closure of the
orbit Oτ . See Fulton [16, p. 52] for a nice picture.
4.2. The examples. For a finite subgroup G ⊂ SL(n, C) with n = 2 or 3, the

Gorenstein quotient Cn /G admits a crepant resolution5, so the stringy E-function
of Cn /G is simply the E-polynomial of the crepant resolution. We therefore begin
by considering 4-dimensional quotient singularities of type 1r (1, r − 1, a, r − a) with
gcd(r, a) = 1. Morrison and Stevens [23, Theorem 2.4(ii)] prove that these are the
only Gorenstein terminal 4-fold cyclic quotient singularities.
Remark 4.2. In each example below we calculate both E(Y ) and Est (Cn /G)
after resolving the singularity ϕ : Y → Cn /G. Note that E(Y ) is not equal to
the E-polynomial of the exceptional fibre D = ϕ−1 (π(0)), for π : Cn → Cn /G the
quotient map. Indeed,
E(Y ) = E(Y \ D) + E(D) = (uv)n − 1 + E(D).
The point is that the E-polynomial encodes the Hodge–Deligne numbers of com-
pactly supported cohomology, yet
H ∗ (D, C) = H ∗ (D, C) ∼
c = H ∗ (Y, C) = H ∗ (Y, C);
c
the first equality holds because D is compact, and the isomorphism is induced by
a deformation retraction of Y onto D ⊂ Y .
Example 4.3. Write X = Uσ for the quotient singularity of type 12 (1, 1, 1, 1),
i.e., for the quotient of C4 by the action of G = Z/2, where the nontrivial element
acts diaginally as −1. Add the ray τ generated by the vector v = 12 (1, 1, 1, 1) to the
cone σ, then take the simplicial subdivision of σ. This determines a toric resolution
ϕ : Y → X with a single exceptional divisor D = V (τ ) ∼= P3 . The discrepancy of D
is 1 by (4.2). Using Proposition 4.1 we calculate
E(Y ) = E(Y \ D) + E(P3 )

= (uv)4 − 1 + (uv)3 + (uv)2 + uv + 1
= (uv)4 + (uv)3 + (uv)2 + uv.
Compare this with the stringy E-function:
uv − 1
Est (X) = E(Y \ D) + E(P3 ) · = (uv)4 + (uv)2 .
(uv)2 − 1
Example 4.4. Write X = Uσ for the quotient singularity of type 13 (1, 2, 1, 2).
Add rays τ1 and τ2 generated by the vectors v1 = 13 (1, 2, 1, 2) and v2 = 13 (2, 1, 2, 1)
respectively to the cone σ, then take the simplicialsubdivision of σ. The resulting
fan Σ is determined by its cross-section ∆2 := σ ∩( xi = 2) illustrated in Figure 1.
There are eight 3-dimensional simplices in ∆2 (four contain a face of the tetra-
hedron and four contain the edge joining v1 to v2 ). Each of these simplices deter-
mines a 4-dimensional cone in Σ which is generated by a basis of the lattice N , so
Y = XΣ → Uσ is a resolution. The union of all eight 3-dimensional simplices in ∆2
contains eighteen faces, fifteen edges and six vertices. Write dk for the number of
cones of dimension k in Σ, so
d4 = 8; d3 = 18; d2 = 15; d1 = 6; d0 = 1 (the origin in N ⊗ R ).
5The moduli space Y = G -Hilb( n ) of G-clusters in n is a crepant resolution for n = 2, 3.
See Bridgeland, King and Reid [6] for definitions and details.
218 ALASTAIR CRAW
2e3
v2
2e4
2e1
v1
2e2
Figure 1. The simplex ∆2 for 13 (1, 2, 1, 2)
Apply Proposition 4.1 to compute

E(Y ) = (uv)4 + 2(uv)3 + 3(uv)2 + 2uv.
To compute Est (X) observe that for j = 1, 2 the exceptional divisor Dj := V (τj )
has discrepancy 1 by (4.2). Write dk (τj ) for the number of cones of dimension k in
the fan Star(τj ) defining V (τj ), so
d3 (τj ) = 6; d2 (τj ) = 9; d1 (τj ) = 5; d0 (τj ) = 1 (the origin in N (τj )).
Proposition 4.1 gives
E(Dj ) = (uv)3 + 2(uv)2 + 2(uv) + 1 for j = 1, 2.
Similarly, the fan Star(τ1 , τ2 ) contains four faces, four edges and one vertex so
Proposition 4.1 gives E(D1 ∩ D2 ) = (uv)2 + 2(uv) + 1. As a result
• E(D∅◦ ) = E(Y \ (D1 ∪ D2 )) = (uv)4 − 1.
◦ ◦
• E(D{1} ) = E(D{2} ) = E(Dj ) − E(D1 ∩ D2 ) = (uv)3 + (uv)2 .
◦
• E(D{1,2} ) = E(D1 ∩ D2 ) = (uv)2 + 2(uv) + 1.
Now compute the stringy E-function using formula (3.2):

◦ uv − 1 ◦ uv − 1
Est (X) = (uv)4 − 1 + E(D{1} )· + E(D{2} ) ·
(uv)2 − 1 (uv)2 − 1
2
◦ uv − 1
+ E(D{1,2} )·
(uv)2 − 1
= (uv)4 + 2(uv)2 .
Example 4.5. Let X = Uσ denote the cyclic quotient singularity of type
1
rays τ1 , τ2 and τ3 generated by the vectors v1 = 14 (1, 3, 1, 3) and
4 (1, 3, 1, 3). Add
v2 = 14 (2, 2, 2, 2)
and v3 = 14 (3, 1, 3, 1) respectively to the cone σ, then take the
simplicial subdivision of σ. The cross-section ∆2 of the resulting fan Σ has three
colinear points in the interior of the tetrahedron but is otherwise similar to that
shown in Figure 1. There are twelve 3-dimensional simplices in ∆2 containing 26
faces, 20 edges and 7 vertices. Proposition 4.1 calculates

E(Y ) = (uv)4 + 3(uv)3 + 5(uv)2 + 3uv.
For j = 1, 2, 3 the divisors Dj := V (τj ) have discrepancy 1 by (4.2). Following the
method of Example 4.4 we calculate
E(D1 ) = E(D3 ) = (uv)3 + 2(uv)2 + 2(uv) + 1
and E(D1 ∩ D2 ) = E(D2 ∩ D3 ) = (uv)2 + 2(uv) + 1. To compute the E-polynomial
of D2 observe that
d3 (τ2 ) = 8; d2 (τ2 ) = 12; d1 (τ2 ) = 6; d0 (τ2 ) = 1 (the origin in N (τ2 )),
where dk (τ2 ) denotes the number of cones of dimension k in Star(τ2 ). It follows
from Proposition 4.1 that
E(D2 ) = (uv)3 + 3(uv)2 + 3(uv) + 1.
Finally, since D1 ∩ D3 = ∅ we have E(D1 ∩ D3 ) = E(D1 ∩ D2 ∩ D3 ) = 0. As a
result
• E(D∅◦ ) = E(Y \ (D1 ∪ D2 ∪ D3 )) = (uv)4 − 1.
◦ ◦
• E(D{1} ) = E(D{3} ) = (uv)3 + (uv)2 .
◦
• E(D{2} ) = (uv)3 + (uv)2 − (uv) − 1.
◦ ◦
• E(D{1,2} ) = E(D{2,3} ) = (uv)2 + 2(uv) + 1.
◦ ◦
• E(D{1,3} ) = E(D{1,2,3} ) = 0.
Apply formula (3.2) to compute Est (X) = (uv)4 + 3(uv)2 .
Remark 4.6. The above examples feature only exceptional divisors with dis-
crepancy 1. To obtain examples of Gorenstein terminal cyclic quotient singularities
which admit resolutions containing divisors having discrepancy larger than one we
must work in dimension higher than four.
1
where n := dim X = kr for some k ∈ Z (by assuming that r
r (1, 1, 1, . . . , 1)
divides n we ensure that X is Gorenstein). Add a single ray τ generated by the
vector v1 = 1r (1, 1, 1, . . . , 1) to the cone σ, then take the simplicial subdivision of
σ. This determines a toric resolution ϕ : Y → X with a single exceptional divisor
D = V (τ ) ∼
= Pn−1 . The discrepancy of D is k − 1 by (4.2). Using Proposition 4.1
we calculate
E(Y ) = E(Y \ D) + E(Pn−1 )

= (uv)n − 1 + (uv)n−1 + (uv)n−2 + · · · + uv + 1
= (uv)n + (uv)n−1 + · · · + uv.
Compare this with the stringy E-function:
uv − 1
Est (X) = E(Y \ D) + E(Pn−1 ) ·
(uv)k − 1
= (uv)n + (uv)n−k + · · · + (uv)2k + (uv)k .
1
3 (1, 2, 1, 2, 1, 2) (compare Example 4.4). Add rays τ1 and τ2 generated by the
vectors v1 = 13 (1, 2, 1, 2, 1, 2) and v2 = 13 (2, 1, 2, 1, 2, 1) respectively to the cone
σ, then take the simplicial subdivision of σ. Both v1 and v2 lie in the simplex
220 ALASTAIR CRAW

∆3 := σ∩( xi = 0) of the resulting fan Σ so the corresponding exceptional divisors
D1 and D2 each have discrepancy 2 by (4.2). The cross-section ∆3 is difficult to
draw (it is 5-dimensional!) but, using Figure 1 as a guide, one can show that
d6 = 15; d5 = 48; d4 = 68; d3 = 56; d2 = 28; d1 = 8; d0 = 1,
where dk denotes the number of cones of dimension k in Σ. Hence
E(Y ) = (uv)6 + 2(uv)5 + 3(uv)4 + 4(uv)3 + 3(uv)2 + 2(uv).
As with Example 4.4, for j = 1, 2 write dk (τj ) for the number of cones of dimension
k in Star(τj ), so
d5 (τj ) = 12; d4 (τj ) = 30; d3 (τj ) = 34; d2 (τj ) = 21; d1 (τj ) = 7; d0 (τj ) = 1.
Proposition 4.1 gives
E(Dj ) = (uv)5 + 2(uv)4 + 3(uv)3 + 3(uv)2 + 2(uv) + 1 for j = 1, 2.
Similarly, counting simplices in the fan Star(τ1 , τ2 ) gives
E(D1 ∩ D2 ) = (uv)4 + 2(uv)3 + 3(uv)2 + 2(uv) + 1.
Now compute the stringy E-function using formula (3.2):

◦ uv − 1 ◦ uv − 1
Est (X) = (uv)6 − 1 + E(D{1} )· + E(D{2} ) ·
(uv)3 − 1 (uv)3 − 1
2
◦ uv − 1
+ E(D{1,2} )·
(uv)3 − 1
= (uv)6 + 2(uv)3 .
Remark 4.9. The stringy E-function of a Gorenstein canonical quotient sin-
gularity Cn /G can be calculated in terms of the representation theory of the finite
subgroup G ⊂ SL(n, C) using a simple formula due to Batyrev [3, 4] (see also
Denef and Loeser [15]). To state the formula, note that each g ∈ G is conjugate to
a diagonal matrix

(4.4) g = diag e2πiα1 (g)/r(g) , . . . , e2πiαn (g)/r(g) with 0 ≤ αj (g) < r(g),
√
where r(g) is the order of g and i = −1. To each conjugacy class [g] of the group
G we associate an integer in the range 0 ≤ age[g] ≤ n − 1 defined by
1
n
age[g] := αj (g).
r(g) j=1
In particular, for the diagonal action introduced in §4.1, the age grading on the
cyclic group G corresponds
to the slicing of the unit box ⊂ N R ∼ = Rn into
polytopes ∆k := σ ∩ ( xi = k) for k = 0, . . . , n − 1. The formula for the stringy
E-function of the quotient Cn /G is simply

(4.5) Est (Cn /G) = (uv)n−age[g] ,
[g]∈Conj(G)
where we sum over the conjugacy classes of G. For example, the nontrivial element
g of the group G = Z/2 acting on C4 in Example 4.3 has age two because vg =
2 (1, 1, 1, 1) ∈ ∆2 . Formula (4.5) gives Est (C /G) = (uv) + (uv) as shown in §4.2.
1 4 4 2
Observe that the same holds for the other examples of §4.2.
5. The McKay correspondence

In this section we discuss formula (4.5) in the wider context of the cohomological
McKay correspondence.
5.1. The McKay correspondence conjecture. Motivated by string theory,

Dixon et al. [10] introduced the orbifold Euler number for a finite group G acting
on a manifold M . This number can be written in the form

e(M, G) = e M g /C(g) ,
[g]∈Conj(G)
where the sum runs over the conjugacy classes of G, e denotes the topological Euler
number, M g is the fixed point set of g and C(g) is the centraliser of g (this version
of the formula is due to Hirzebruch and Höfer [18]). Observe that the orbifold
Euler number is the standard topological Euler number of the disjoint union

=
M/G M g /C(g).
[g]∈Conj(G)
Dixon et al. [11] formulated what became known as the “physicists’ Euler number
conjecture”:
Conjecture 5.1. If M/G is a Gorenstein Calabi–Yau variety which admits a

crepant resolution Y → M/G then e(Y ) = e(M, G).
Hirzebruch and Höfer [18] observed that for a finite subgroup G ⊂ U(n) acting
on M = Cn , the orbifold Euler number is equal to the number of conjugacy classes
of G because every fixed point set M g is contractible. For G ⊂ SU(2, C), the
classical McKay correspondence states that the second Betti number b2 (Y ) of the
minimal resolution Y → C2 /G equals the number of nontrivial conjugacy classes
of G. As a result, the equality

(5.1) e(Y ) = b2 (Y ) + 1 = # conjugacy classes of G = e(C2 , G)
can be viewed as a version of the McKay correspondence. Inspired by this obser-

vation, Reid [27] formulated Conjecture 1.2.
5.2. The orbifold E-function. Following the introduction of the orbifold

Euler number, Vafa [30] and Zaslow [32] considered the orbifold Hodge numbers

m(g)
p,q
(5.2) h (M, G) := dimC Hcp−age[g],q−age[g] (Mi (g)/C(g)),
[g]∈Conj(G) i=1
where M g = M1 (g) ∪ · · · ∪ Mm(g) (g) are the smooth connected components of the
fixed-point set. These numbers are the standard (compactly supported) Hodge
shifted according to the age6 of the appropriate conjugacy class.
numbers of M/G
It is natural to introduce the orbifold analogue of the E-polynomial.
6In this context, the age of a conjugacy class is called the Fermionic shift number.
222 ALASTAIR CRAW
Definition 5.2. The orbifold E-function of the pair (M, G) is

Eorb (M, G) := (−1)p+q hp,q (M, G) up v q
p,q

m(g)
= E(Mi (g)/C(g)) · (uv)age[g] ,
[g]∈Conj(G) i=1
where E denotes the standard E-polynomial. When Eorb (M, G) is evaluated at

u = v = 1 we produce the orbifold Euler number e(M, G).
Remark 5.3. Note that the second formula for Eorb (M, G) given in Defini-
tion 5.2 follows from a straightforward substitution. In the special case M = Cn
and G ⊂ GL(n, C), the fixed-point set M g is an affine subspace of dimension equal
to n − rank(g − id) = n − age[g] − age[g −1 ]. As a result,
−1
Eorb (Cn , G) = (uv)n−age[g ] = (uv)n−age[g] ,
[g]∈Conj(G) [g]∈Conj(G)
where the final equality follows from summing over conjugacy classes [g −1 ].
5.3. McKay correspondence via motivic integration. Batyrev proved

Conjecture 1.2 by first establishing that the stringy E-function coincides with the
orbifold E-function. That is, he proved formula (4.5):
Theorem 5.4 ([3]). Let G ⊂ SL(n, C) be a finite subgroup. Then

Est (Cn /G) = Eorb (Cn , G) = (uv)n−age[g] ,
[g]∈Conj(G)
where the sum runs over conjugacy classes of G.
Proof of the Abelian case. Choose coordinates on Cn so that every matrix

g ∈ G takes the form given in (4.4). The cyclic construction of §4.1 can be adapted
to show that Cn /G is the toric variety Uσ corresponding to the cone σ = R≥0 e1 +
· · · + R≥0 en and the lattice

(5.3) N := Ze1 + · · · + Zen + Z · vg ,
g∈G

1
for vg = r(g) α1 (g), . . . , αn (g) . Now, Uσ is Gorenstein because G ⊂ SL(n, C), so
there exists a continuous linear function ψK : N ⊗ R → R≥0 satisfying ψ(ei ) = 1
for i = 1, . . . , n. A straightforward computation due to Batyrev [1, Theorem 4.3]
gives

(5.4) Est (Uσ ) = (uv − 1)n (uv)−ψK (v) .
v∈N ∩σ
For vg ∈ , ψK (vg ) = age(g). In fact, for any lattice point v ∈ N ∩ σ we have

ψK (v) = k ⇐⇒ v ∈ ∆k = σ ∩ ( xi = k). Thus for each v ∈ N ∩ σ there
exist unique vg ∈ and (x1 , . . . , xn ) ∈ Zn≥0 such that v is the translate of vg by

(x1 , . . . , xn ), and ψK (v) = age(g) + ni=1 xi .
Covering the positive orthant σ by translations of gives

x
(uv)−ψK (v) = (uv)− age(g)− i
v∈N ∩σ vg ∈ (x1 ,...,xn )∈Zn

≥0

n
= (uv)− age(g) (uv)−xi
vg ∈ i=1 xi ∈Z≥0

n
1
= (uv)− age(g) .
i=1
1 − (uv)−1
g∈G
Substituting this into (5.4) gives

n
uv − 1
Est (Uσ ) = (uv)− age(g) −1
= (uv)n−age(g) .
i=1
1 − (uv)
g∈G g∈G
This proves the theorem for a finite Abelian subgroup G ⊂ SL(n, C).
Corollary 5.5 (strong McKay correspondence). Let G ⊂ SL(n, C) be a finite
subgroup and suppose that the quotient X = Cn /G admits a crepant resolution
ϕ : Y → X. The nonzero Betti numbers of Y are

dimC H 2k (Y, C) = # age kconjugacy classes of G
for k = 0, . . . , n − 1. In particular, the Euler number of Y equals the number of
conjugacy classes of G.
Proof. The Hodge structure in Hci (Y, Q) is pure for each i and Poincaré du-
ality Hc2n−i (Y, C) ⊗ H i (Y, C) → Hc2n (Y, C) respects the Hodge structure, so it is
enough to show that the only nonzero Hodge–Deligne numbers of the compactly
supported cohomology of Y are

hn−k,n−k Hc2n−2k (Y, C) = # age k conjugacy classes of G .

Now hn−k,n−k Hc2n−2k (Y, C) is the coefficient of (uv)n−k in the E-polynomial of
Y . Moreover, the resolution ϕ : Y → X is crepant so E(Y ) = Est (X) and the result
follows from Theorem 5.4.
Appendix A. The motivic nature of the integral

In this appendix we investigate the motivic nature of the integral and justify
the notation L for the class of the complex line C in the Grothendieck ring of
algebraic varieties.
The category MC of Chow motives over C is the category (see [29]) whose
objects are triples (X, p, m), where X is a smooth complex projective variety of
dimension d, p is an element of the Chow ring Ad (X × X) satisfying p2 = p and
m ∈ Z. If (X, p, m) and (Y, q, n) are motives then
HomMC ((X, p, m), (Y, q, n)) = qAd+n−m (X, Y )p
where composition of morphisms is given by composition of correspondences. The
category MC is additive, Q-linear and pseudo-Abelian. Tensor product of motives
is defined as (X, p, m) ⊗ (Y, q, n) = (X × Y, p ⊗ q, m + n). There is a functor
h : VC◦ → MC
224 ALASTAIR CRAW
from the (opposite) category of algbraic varieties over C to the category of Chow
motives over C, sending X to (X, ∆X , 0), the Chow motive of X, where the diagonal
∆X ⊂ X × X is the identity in A∗ (X × X). The motive of a point 1 = h(Spec C)
is the identity with respect to tensor product. The Lefschetz motive L is defined
implicitly via the relation h(P1C ) = 1 ⊕ L.
Definition A.1. The Grothendieck group of MC is the free Abelian group
generated by isomorphism classes of objects in MC modulo the subgroup generated
by elements of the form [(X, p, m)] − [(Y, q, n)] − [(Z, r, k)] whenever the relation
(X, p, m) (Y, q, n) ⊕ (Z, r, k) holds. Tensor product of motives induces a ring
structure and the resulting ring, denoted K0 (MC ), is the Grothendieck ring of
Chow Motives (over C).
Gillet and Soulé [17] exhibit a map M : VC −→ K0 (MC ) sending a smooth
projective variety X to the class [h(X)] of the motive of X. Furthermore the map
is additive on disjoint unions of locally closed subsets and satisfies M (X × Y ) =
M (X) · M (Y ).
We now play the same game as we did in §3. That is, M factors through
K0 (VC ) inducing
M : K0 (VC ) −→ K0 (MC ).
Observe that the image of [C] under M is the class of the Lefschetz motive L; this
explains why we use the notation L to denote the class of C in K0 (VC ) in § 2.
Sending L−1 ∈ K0 (VC )[L−1 ] to L−1 ∈ K0 (MC ) produces a map
M : K0 (VC )[L−1 ] −→ K0 (MC ).
It is unknown whether or not M annihilates the kernel of the natural completion
map φ : K0 (VC )[L−1 ] → R. Denef and Loeser conjecture that it does (see [13,
Remark 1.2.3]). If this is true, extend M to a ring homomorphism

−1
1 1
Mst : φ K0 (VC )[L ] −→ K0 (MC )
Li − 1 i∈N Li − 1 i∈N
such that the image of [DJ◦ ] under Mst is equal to M (DJ◦ ).
Definition A.2. Let X denote a complex algebraic variety with at worst
Gorenstein canonical singularities and let ϕ : Y →
X be any resolution of sin-
gularities for which the discrepancy divisor D = ai Di has only simple normal
crossings. The stringy motive of X is
" #  
L−1
Mst (X) := Mst FD dµ · Ln = M (DJ◦ ) ·  ,
J∞ (Y ) Laj +1 − 1
J⊆{1,...,r} j∈J
where we sum over all subsets J ⊆ {1, . . . , r} including J = ∅. As with the definition
of the stringy E-function (see Definition 3.5) we multiply by Ln for convenience.
References
1. V. Batyrev, Stringy Hodge numbers of varieties with Gorenstein canonical singularities,
Integrable systems and algebraic geometry (Kobe/Kyoto, 1997), (1998), pp. 1–32.
2. V. Batyrev, Birational Calabi-Yau n-folds have equal Betti numbers, New trends in algebraic
geometry, Klaus Hulek et al. (editors), CUP, pages 1–11, (1999).
3. V. Batyrev, Non-Archimedean integrals and stringy Euler numbers of log-terminal pairs, J.
Eur. Math. Soc. 1 (1999), 5–33.
4. V. Batyrev, Canonical Abelianization of finite group actions (correction to [Bat99b]),

Preprint math.AG/0009043, (2000).
5. V. Batyrev and D. Dais, Strong McKay correspondence, string-theoretic Hodge numbers and
mirror symmetry, Topology 35 (1996), 901–929.
6. T. Bridgeland, A. King, and M. Reid, The McKay correspondence as an equivalence of
derived categories, J. Amer. Math. Soc. 14 (2001), 535–554.
7. A. Craw, Notes from a course on Toric Geometry at the University of Utah, available from
www.math.utah.edu/∼craw/math7800, (2002).
8. P. Deligne, Théorie de Hodge II, Publ. Math. IHES 40 (1971), 5–57.
9. P. Deligne, Théorie de Hodge III, Publ. Math. IHES 44 (1974), 5–77.
10. L. Dixon, J. Harvey, C. Vafa, and E. Witten, Strings on Orbifolds I, Nuclear Physics B 261
(1985), 678–686.
11. L. Dixon, J. Harvey, C. Vafa, and E. Witten, Strings on Orbifolds II, Nuclear Physics B 274
(1986), 285–314.
12. V. Danilov and A. Khovanskii, Newton polyhedra and an algorithm for computing Hodge–
Deligne numbers, Math. USSR Izvestiya 29 (1987), 279–298.
13. J. Denef and F. Loeser, Motivic Igusa zeta functions, J. Alg. Geom. 7 (1998), 505–537.
14. J. Denef and F. Loeser, Germs of arcs on singular algebraic varieties and motivic integration,
Invent. Math. 135 (1999), 201–232.
15. J. Denef and F. Loeser, Motivic integration, quotient singularities and the McKay corre-
spondence, Compositio Math. 131 (2002), 267–290.
16. W. Fulton, Introduction to toric varieties, Annals of Math. Studies 131, Princeton University
Press, (1993).
17. H. Gillet and C. Soulé, Descent, motives and K-theory, J. Reine Angew. Math. 478 (1996),
127–176.
18. H. Hirzebruch and T. Höfer, On the Euler number of an orbifold, Math. Ann. 286 (1990),
255–260.
19. Y. Kawamata, K. Matsuda, and K. Matsuki, Introduction to the Minimal Model Problem,
Adv. Studies in Pure Math. 10 (1987), 283–360.
20. M. Kontsevich, Motivic integration, Lecture at Orsay, (1995).
21. E. Looijenga, Motivic measures, Séminaire Bourbaki, Vol. 1999/2000. Astérisque 276 (2002),
267–297.
22. J. McKay, Graphs, singularities and finite groups, Proceedings of Symposia in Pure Mathe-
matics 37, 1980, pp. 183–186.
23. D. Morrison and G. Stevens, Terminal quotient singularities in dimensions three and four,
Proceedings of the A.M.S. 90, (1984), pp. 15–20.
24. D. Mumford, The red book of varieties and schemes, Lecture Notes in Mathematics 1358,
Springer-Verlag, (1988).
25. M. Reid, Canonical 3-folds, Journées de Géométrie Algébrique d’Angers, A. Beauville (edi-
tor), Sijthoff and Noordhoff, pages 273–310, (1980).
26. M. Reid, Young person’s guide to canonical singularities, Proceedings of Symposia in Pure
Mathematics 46, (1987), pp. 345–414.
27. M. Reid, The McKay correspondence and the physicists’ Euler number conjecture, Lecture
notes given at the University of Utah (Sept.), and MSRI (Nov.), (1992).
28. W. Rudin, Real and complex analysis, McGraw–Hill Series in Higher Mathematics, (1987).
29. A. Scholl, Classical motives, Proceedings of Symposia in Pure Mathematics 55 Part 1, (1994),
pp. 163–187.
30. C. Vafa, String vacua and orbifoldized LG models, Modern Phys. Lett. A 4 (1989), 1169–
1185.
31. K. Watanabe, Certain invariant subrings are Gorenstein I and II, Osaka Journal 11 (1974),
1–8 and 379–388.
32. E. Zaslow, Topological orbifold models and quantum cohomology rings, Comm. Math. Phys.
156 (1993), 301–331.
Department of Mathematics, University of Utah, 155 South 1400 East, Salt Lake
City, UT 84112
E-mail address: craw@math.utah.edu
Volume 3, 2004
Representation moduli of the McKay quiver for finite

Abelian subgroups of SL(3, C)
Akira Ishii
Abstract. The representation moduli of the McKay quiver, or the moduli of

G-constellations, is known to be a crepant resolution of the quotient singularity
3 /G for a finite subgroup G of SL(3, ). The proof by Bridgeland, King
and Reid [1] uses derived category technique. In this note, we give local
(toric) coordinates of the moduli when G is Abelian, generalising Nakamura’s
coordinates on G-Hilb [9]. We also resume the chamber structure for stability
from [3] briefly.
1. Introduction
Let G be a finite subgroup of SL(3, C). Nakamura [9] showed that the moduli
space G-Hilb of G-clusters is a crepant resolution when G is Abelian. Bridgeland,
King and Reid [1] proved that the same holds for any finite subgroup of SL(3, C) by
∼
using the Fourier-Mukai transform Φ : D(Y ) → DG (C3 ), where D(Y ) and DG (C3 )
are the bounded derived categories of coherent sheaves on Y = G -Hilb and G-
equivariant coherent sheaves on C3 respectively. Therefore there is a special choice
G-Hilb among crepant resolutions of C3 /G.
We can consider more general moduli spaces Mθ introduced by Kronheimer
[7] and further studied by Sardo Infirri [11]. These spaces are again crepant reso-
lutions for generic parameters θ by the same arguments as in [1]. In [3], under the
assumption that G is Abelian, we studied variation of the moduli spaces Mθ (and
the associated Fourier-Mukai transform Φθ ) when the stability parameter θ moves.
In particular, we saw that every projective crepant resolution is isomorphic to Mθ
for some θ.
In this note, we introduce local coordinates on the moduli space Mθ when G
is Abelian. This gives an elementary proof that Mθ is a crepant resolution in the
Abelian case. To describe the local coordinates, we use the universal covering of
the McKay quiver and the G-igsaw puzzle as in [9] and [10]. Some of the ideas here
were also used in [3] §10.
In the rest of this section, we give basic definitions. We introduce the notion of
a G-constellation here which is regarded as a representation of the McKay quiver in
2000 Mathematics Subject Classification. 14E15, 14D20.
2004
227
228 AKIRA ISHII
§3. For a finite subgroup G of SL(3, C), we denote by R its regular representation
and by R(G) the representation ring.
Definition 1.1. A G-constellation is a G-equivariant coherent sheaf F on Cn
such that H 0 (F ) is isomorphic as a C[G]-module to R. Set

Θ := { θ ∈ HomZ (R(G), Q) θ(R) = 0 }.
For θ ∈ Θ, a G-constellation is said to be θ-stable if every proper G-equivariant
coherent subsheaf 0 ⊂ E ⊂ F satisfies θ(E) > 0, i.e., θ(H 0 (E)) > 0 = θ(H 0 (F )).
The notion of θ-semistable is the same with ≥ replacing >.
We denote by Mθ and Mθ the moduli spaces of θ-stable and θ-semistable G-
constellations respectively. The existence of these moduli spaces are ensured by
a work of King [6]. We recall the construction of these moduli spaces in §3. For
a particular choice of parameters θ, Mθ = Mθ is G-Hilb (See [5] §3 and also [3]
§9). Mθ and the quotient singularity X = Cn /G are related by the following
(set-theoretically) commutative diagram:
open immersion
Mθ −−−−−−−−−−→ Mθ
 
 projective

X −−−−−→ M0
injective
The first vertical arrow sends a G-constellation to its support and the second hori-
zontal arrow sends a G-orbit to the class of G-constellations supported by it.
Definition 1.2. We say a parameter θ ∈ Θ is generic if every θ-semistable
G-constellation is θ-stable.
Arguments in [1] show that if n = 3 and if the parameter θ is generic, then
Mθ is a crepant resolution of C3 /G and we have a Fourier–Mukai transform Φθ :
∼
D(Mθ ) −→ DG (C3 ). In §3, we give an elementary proof that Mθ is a crepant
resolution if θ is generic.
Acknowledgements The author wishes to thank A. King, H. Nakajima, I. Naka-

mura and M. Reid for discussions and encouragements. He is also grateful to
A. Craw for many discussions and useful suggestions. In particular, § 2 is a resumé
of the author’s joint work with Craw.
2. Chamber structure for stability

In this section, we briefly recall results from [3], a part of which the author
talked about in the conference. These results are not needed in the following sec-
tions. In [3], we studied the variation of pairs (Mθ , Φθ ) as the parameters θ vary,
assuming that G is Abelian. It is described by a chamber system in the parameter
space Θ; for a generic θ, (Mθ , Φθ ) depends only on the chamber C which contains
θ. Thus we can write (MC , ΦC ) instead of (Mθ , Φθ ). Our purpose in [3] was to
describe this chamber system together with the variation of (MC , ΦC ). One nat-
ural question is whether every projective crepant resolution is isomorphic to MC
for some chamber C. We gave an affirmative answer to this question in [3]:
REPRESENTATION MODULI OF THE MCKAY QUIVER 229
Theorem 2.1. For a finite Abelian subgroup G ⊂ SL(3, C), suppose Y → C3 /G

is a projective crepant resolution. Then there is a chamber C ⊂ Θ such that Y ∼
=
MC .
Another main result in [3] is the determination of the equivalences Φ−1 C ◦ΦC for
adjacent chambers C, C separated by a wall. Walls are classified into three types
0, I, III. (Types I and III correspond to the primitive birational contractions of types
I and III , and type 0 is the case where no contraction occurs; see [3].)
Theorem 2.2. Let C and C be adjacent chambers in Θ with W = C ∩ C .
Then the relation between the moduli spaces MC , MC and the explicit form of the
derived equivalence
∼
Φ−1
C ◦ ΦC : D(MC ) −→ D(MC )
is determined by the type of the wall W :
Type 0: MC is isomorphic to MC and the derived equivalence Φ−1 C ◦ ΦC
is a Seidel–Thomas twist [12] (up to a tensor product with a line bundle).
Type I: MC is a flop of MC and Φ−1 C ◦ ΦC is the induced derived equiva-
lence described by Bondal–Orlov [2, §3].
Type III: MC is isomorphic to MC and Φ−1 C ◦ΦC is an EZ-transformation
of Horja–Szendrői [4, 13] (up to a tensor product with a line bundle).
The key step to these results is to understand a chamber C in terms of the K-
theory of MC via the Fourier-Mukai transform ΦC . C is given by linear inequalities
that correspond to exceptional curves or line bundles on divisors via ΦC :
Theorem 2.3. Let C ⊂ Θ be a chamber. Then θ ∈ C if and only if
(1) for every exceptional curve we have θ([ΦC (O )]) > 0.
(2) for every compact reduced divisor D and irreducible representation ρ we
have
θ([ΦC (R−1
ρ ⊗ ωD )]) < 0 and θ([ΦC (R−1
ρ |D )]) > 0.
Here Rρ are the tautological bundles on Mθ associated to irreducible repre-

sentations ρ ∈ Irr(G). By identifying the Grothendieck group K0G (C3 ) of coherent
sheaves on C3 supported at the origin with the representation ring R(G) of G, we
can substitute the class [F] of F ∈ D0G (C3 ) into θ. For the details of these results,
see [3].
3. Local coordinates on Mθ
Hereafter, we assume that θ is generic and we give local coordinates of the
moduli space Mθ .
3.1. Representations of the McKay quiver. Let V be a three-dimensional
vector space over C and G ⊂ SL(V ) a finite subgroup. The moduli space Mθ of
G-constellations is constructed by regarding it as a moduli space of representations
of the McKay quiver. In this section, we recall the construction.
Consider the affine scheme

N = { B ∈ HomC[G] (R, V ⊗ R) B ∧ B = 0 },
where B ∧ B lies in HomC[G] (R, ∧2 V ⊗ R). Each map B ∈ N determines an action
of V ∗ on R and the condition B ∧B = 0 ensures that this action is commutative. In
this way, B endows R with a C[V ]-module structure, where C[V ] = ⊕∞ k ∗
k=0 Sym (V )
230 AKIRA ISHII
is the polynomial ring of functions on V . Thus B determines a G-constellation on

V.
Denote by AutC[G] (R) ⊂ GL(R) the group of G-equivariant automorphisms of
R. If we decompose the regular representation R = ⊕ρ∈Irr(G) Rρ ⊗ ρ as a G-module,
we have
AutC[G] (R) = GL(Rρ )
ρ∈Irr(G)
This group acts on N by conjugation and the diagonal scalar subgroup C∗ acts
trivially, leaving a faithful action of
PAutC[G] (R) := AutC[G] (R)/C∗
on N by conjugation. Let θ ∈ HomZ (R(G), Z) satisfy θ(R) = 0. Then θ deter-
mines a character χθ of PAutC[G] (R) mapping [(gρ )ρ ] to ρ det(gρ )θ(ρ) ∈ C∗ . This
character determines a linearisation of the trivial line bundle on N . We denote by
Mθ := Nθ / PAutC[G] (R) and Mθ := Nθ // PAutC[G] (R)
the geometric and the categorical quotients of the open subsets Nθ and Nθ of χθ -
stable and χθ -semistable points of N . King’s result (which holds in a more general
context) asserts that Mθ and Mθ are equivalently the moduli spaces of θ-stable and
θ-semistable G-constellations respectively. There is an isomorphism Mkθ ∼ = Mθ
for k ∈ Z>0 , so Mθ is well defined even for parameters θ ∈ HomZ (R(G), Q), i.e.,
for θ ∈ Θ.
To let R ⊗ ON and the universal homomorphism R ⊗ ON → V ⊗ R ⊗ ON
on N descend to Mθ , we have to determine an equivariant PAutC[G] (R)-action on
R ⊗ ON . Define a subgroup

AutC[G] (R) = { (gρ ) ∈ AutC[G] (R) gρ = 1 }
ρ∈Irr(G) 0
of AutC[G] (R), where ρ0 is the trivial representation and hence dim Rρ0 = 1. Then
the projection gives an isomorphism AutC[G] (R) → PAutC[G] (R). On the other
hand, AutC[G] (R) has a natural equivariant action on R ⊗ ON . Descent theory of
coherent sheaves for a faithfully flat morphism implies that R ⊗ ON and Rρ ⊗ ON
descend to locally free sheaves R := Rθ and Rρ := (Rθ )ρ respectively on Mθ such
that R = ⊕ρ∈Irr(G) Rρ ⊗ ρ. Moreover, we have a homomorphism R → V ⊗ R
determining the universal G-constellation Uθ on Mθ × C3 . Note by the definition
of AutC[G] (R) that the line bundle Rρ0 is trivial.
From now on, we consider the case where G is Abelian. We denote by G∗ =
Irr(G) the character group of G. Decompose V ∗ = ρ1 ⊕ ρ2 ⊕ ρ3 as a representation
of G. An element B ∈ HomC[G] (R, V ⊗ R) is decomposed as
B = (bρ1 , bρ2 , bρ3 )ρ∈G∗ = (bρi )i,ρ
where bρi is a linear map Rρ → Rρρi . Then the commutativity condition B ∧ B = 0
is expressed as
ρρj ρ
bρρ i ρ
j bi = bi bj .
The action of [(αρ )ρ∈G∗ ] ∈ PAutC[G] (R) is written as
[(αρ )ρ∈G∗ ] · (bρi )i,ρ = (αρρi αρ−1 bρi )i,ρ .
The three-dimensional torus T = AutC[G] (V ) acts naturally on N by
(t1 , t2 , t3 ) · (bρ1 , bρ2 , bρ3 )ρ = (t1 bρ1 , t2 bρ2 , t3 bρ3 )ρ .
We regard (bρi ) as a representation of the McKay quiver with dimension vector

(1, 1, . . . , 1) (see [11]). By definition, the McKay quiver is the quiver with vertex
set G∗ and arrows aρi from ρ to ρρi for i = 1, 2, 3 and ρ ∈ G∗ .
3.2. Universal covering of the McKay quiver. In the study of represen-
tations of the McKay quiver in our case, it is convenient to consider the universal
covering of it ( [9], [10]). By taking ρ1 , ρ2 , ρ3 as generators of G∗ , we have an exact
sequence
(3.1) 0 → M → Z3 → G∗ → 0.
Since G is in SL(V ), M contains (1, 1, 1). If we put M = M/Z(1, 1, 1) and H =
Z3 /Z(1, 1, 1) then we have an exact sequence
0 → M → H → G∗ → 0.
We denote by f1 , f2 , f3 the images of (1, 0, 0), (0, 1, 0), (0, 0, 1) respectively.
Definition 3.1. Consider the quiver with vertex set H and arrows from h ∈ H
to h + fi for i = 1, 2, 3. We call it the universal covering of the McKay quiver.
The McKay quiver is the quotient of the universal covering by the action of
M . For later purposes, we identify HR = H ⊗ R with the orthogonal complement
of (1, 1, 1) in R3 with respect to the Euclidean metric.
3.3. Torus fixed points and the G-igsaw puzzle. We look at torus fixed
points of Mθ . Let B ∈ Nθ give a T -fixed point of Mθ , where T = Spec(C[M ]) =
T/G. Define a graph Γ(B) whose vertex set is G∗ and whose edges are aρi with
bρi = 0, forgetting the orientation. We have a natural lift of Γ(B) in HR as follows.
[B] ∈ Nθ / AutC[G] (R) is a T-fixed point and AutC[G] (R) acts freely on N at B.
Therefore, we have a homomorphism α : T → AutC[G] (R) such that t · B = α(t) · B.
Write α as
−k −l −m

(3.2) (t1 , t2 , t3 ) → (t1 ρ t2 ρ t3 ρ )ρ ∈ C∗
ρ∈G∗
where (kρ , lρ , mρ ) are integers attached to ρ ∈ G∗ with kρ0 = lρ0 = mρ0 = 0. For
ρ ∈ G∗ , put
(3.3) h(ρ) = kρ f1 + lρ f2 + mρ f3 ∈ H.
Draw a line segment from h(ρ) to h(ρ) + fi if bρi = 0. In this case, we can see that
h(ρ) + fi = h(ρρi ) and hence we can embed Γ(B) in the plane HR so that the sets
of edges and vertices of Γ(B) are subsets of those of the universal covering of the
McKay quiver respectively. For h ∈ H we put
Hex(h) = { P ∈ HR | d(P, h) ≤ d(P, h ) for h ∈ H }
where d(∗, ∗) denotes the metric of HR . Then Hex(h) is a regular hexagon with
center h and these hexagons form a honeycombed tiling of HR . Define D(B) as

D(B) = Hex(h(ρ)).
ρ∈G∗
Then D(B) is a fundamental domain for the action of M on HR and we have a

tiling of HR by the fundamental domains parallel to D(B) as

(3.4) HR = (D(B) + m) .
m∈M
232 AKIRA ISHII
By the commutativity condition, Γ(B) has the following property: if aρi and
are edges of Γ(B), then so are aρj and ai j for ρ ∈ G∗ and i = j. Let C(ρ, i, j)
ρρ
aρρ
j
i
denote the cycle formed by arrows ai , aj , (ai j )−1 and (aρj )−1 which appear in
ρ ρρi ρρ
the commutativity condition. We denote by Cycle(B) the set of such cycles in

ρρ ρ
Γ(B) with i < j. Note that if C(ρ, i, j) ∈ Cycle(B), then the diagonal ak i j is not
contained in Γ(B), where {i, j, k} = {1, 2, 3}.
Lemma 3.2. There exists a unique representation σ ∈ G∗ such that bσ1 = bσρ 2
1
=
∗
bσρ
3
1 ρ2
= 0. Symmetrically, there also exists a unique representation τ ∈ G such
that bτ1 = bτ3 ρ1 = bτ2 ρ1 ρ3 = 0. Moreover, the cycles C(ρ, i, j) ∈ Cycle(B) form a
basis of H1 (Γ(B), Z).
Proof. Let p(resp. q) denote the number of vertices ρ ∈ G∗ such that exactly
two (resp. one) of bσ1 , bσρ 1
2 , b3
σρ1 ρ2
are non-zero. Note that at least one of these
three values is 0, since B gives a fixed point. Hence the assertion that σ exists
uniquely is equivalent to the equality n − (p + q) = 1.
Since Γ(B) has 2p + q edges and n vertices, the Euler characteristic of Γ(B) is
given by
χ(Γ(B)) = n − (2p + q).
On the other hand, we have exactly p cycles C(ρ, i, j) ∈ Cycle(B) which are minimal
loops of Γ(B) in the plane HR . It follows that the first Betti number satisfies
b1 (Γ(B)) ≥ p. The stability implies that Γ(B) is connected. Thus, χ(Γ(B)) ≤ 1 − p
and therefore n − (p + q) ≤ 1. So if we show there exists at least one σ or τ , then
every equality holds and all the assertions follow.
We embedded Γ(B) in HR and obtained a fundamental domain D(B) deter-
mined by it. There is a point P on the boundary of D(B) where D(B) and two
other fundamental domains of the tiling of HR meet. The three hexagons contain-
ing P belong to different domains and one of the central points of these hexagons
is the desired one.
Corollary 3.3. D(B) contacts 6 fundamental domains of the tiling (3.4).

The points where D(B) and two of these domains meet are the following:
h(σ) + 23 f1 + 13 f2 , h(σρ1 ) + 23 f2 + 13 f3 , h(σρ1 ρ2 ) + 23 f3 + 13 f1 ,
h(τ ) + 23 f1 + 13 f3 , h(τ ρ1 ) + 23 f3 + 13 f2 , h(τ ρ1 ρ3 ) + 23 f2 + 13 f1
Points in the first line and the second line above appear alternately on the boundary
∂D(B).
Proof. Only the last statement needs a proof. Assume Y and Y are points
on the first line adjacent in ∂D(B). We denote by Y Y the connected component
of ∂D(B) \ {Y, Y } on which there are no other points above. Since Y and Y
are congruent modulo M , there is m ∈ M such that Y = Y + m. Put L =

j∈Z (Y Y + jm). It is a connected infinite graph without trivalent points, dividing
HR into two parts. The points where three fundamental domains meet on L are
Y + jm (j ∈ Z). The opposite side of D(B) with respect to L must be contained
in a single domain parallel to D(B), a contradiction.
Let δ(B) denote the set of edges e of Hex(h(ρ)) for some ρ ∈ G∗ such that e is
not in the interior of any C(ρ, i, j) ∈ Cycle(B) and such that e ∩ Γ(B) = ∅.
Lemma 3.4. Any e ∈ δ is contained in the boundary ∂D(B) of D(B) and we

have

∂D(B) = e.
e∈δ(B)
Proof. The inclusion ∂D(B) ⊆ ∪e∈δ(B) e is obvious. If ∪e∈δ(B) e has a trivalent

point, it must be one of the 6 points in the above corollary. But at any of these
points three fundamental domains contact and it is impossible that the three edges
having this point as a vertex are contained in D(B). The commutativity condition
B ∧ B = 0 shows that ∪e∈δ(B) e does not have an end point. Therefore, ∪e∈δ(B) e
is homeomorphic to the disjoint union of finitely many copies of S 1 . Each S 1 must
surround an integral point and hence it follows from the simply-connectedness of
D(B) that ∪e∈δ(B) e is connected.
We fix the orientation of HR by taking f1 , f2 as a basis and say they are in the

anticlockwise order. Then we can determine the clockwise and the anticlockwise
orientations of ∂D(B). Denote the six points in Corollary 3.3 by
Y1 , A1 , Y2 , A2 , Y3 , A3
in the clockwise order. We assume the points in the first row in Corollary 3.3 are
Yi and the others are Ai (in appropriate orders).
Remark 3.5. The letter Y stands for the way in which the three domains
contact at a point. (To see the correct picture, we should assume that the vector
f1 faces to the right.) A denotes the way upside down, though this letter does not
show a correct picture.
Lemma 3.6. Consider the arrows of the universal covering of the McKay quiver
such that they intersect ∂D(B). Among them, assume ã is the nearest to Yi in the
clockwise direction of ∂D(B) and ã in the anti-clockwise direction. Then ã is
going out of D(B) and ã is going into D(B). For the points nearest to Ai , the
same assertion holds with directions of the arrows reversed.
Proof. The three arrows of the universal covering of the McKay quiver nearest
to Yi go around Yi in the anti-clockwise direction. ã and ã are two of these three
arrows and they have one point P in common, which must be contained in D(B).
Since we determined the direction of ∂D(B) to be clockwise, ã goes into P and ã
goes out of P .
3.4. Local coordinates. In this section, we introduce a local coordinate of

Mθ . Throughout this section, B ∈ Nθ denotes a representation of the McKay
quiver such that [B] ∈ Mθ is a T -fixed point.
Fix isomorphisms Rρ ∼ ρ
= C. Then we can regard the linear map bi : Rρ → Rρρi
ρ ρ
as a complex number. We call bi ∈ C the value of B at the arrow ai .
Lemma 3.7. There exists a unique element λ ∈ PAutC[G] (R) such that if we
put λ · B = (b i )i,ρ then b i is either 0 or 1.
ρ ρ
Proof. Fix ρ ∈ G∗ . Since Γ(B) is connected, we can take a sequence of arrows

γ = { α1ε1 , α2ε2 , . . . , αpεp }
234 AKIRA ISHII
such that αk is an arrow of the McKay quiver contained in Γ(B), εi = ±1 determines

the direction of arrows and such that γ gives a connected path from h(ρ0 ) to h(ρ).
Put
p
λρ = βk−εk
k=1
where if αk = aρi then βk = bρi is the value of B at this arrow. The last statement
of Lemma 3.2 and the commutativity condition imply that λρ is independent of the
choice of the path γ. Put
λ = (λρ )ρ ∈ AutC[G] (R).
Then λ·B has the desired property. The uniqueness follows from the connectedness
of Γ(B).
So we can assume bρi is 0 or 1 in the sequel. Put
(B) = { C = (cρ )i,ρ ∈ N | cρ = 0 if bρ = 0 }
U i i i
and
U (B) = { C = (cρi )i,ρ ∈ N | cρi = 1 if bρi = 0 }.
Corollary 3.8. The action of AutC[G] (R) induces an isomorphism
AutC[G] (R) × U (B) ∼
=U (B)
and U (B) ∼
=U (B)/ Aut (R) becomes an open neighborhood of [B] in Mθ .
C[G]
Proof. The first part is proved in the same way as in the above lemma.
(B), then every C-invariant subrepresentation of R is also B-invariant.
If C ∈ U
Therefore, C is also θ-stable and we obtain the second assertion.
Thus we can restrict the action of the torus T on Mθ to the open set U (B).
Corollary 3.9. For t = (t1 , t2 , t3 ) ∈ T and (cρi ) ∈ U (B), put t · (cρi ) = (dρi )
with respect to the above action. We put
(α1 , α2 , α3 ) = (kρ + δi1 , lρ + δi2 , mρ + δi3 ) − (kρρi , lρρi , mρρi ),
where kρ , lρ , mρ are the weights for ρ ∈ G∗ given in (3.2) and where δij denotes
Kronecker’s delta. Then dρi = t−α1
1 −α2 −α3 ρ
t2 t3 c i
Proof. Do the same argument for B = (ti bρi )ρi as in the lemma, assuming
bρiare either 0 or 1. By the definition of λρ in the proof of the lemma, we have
ti = λ−1 ρ
ρρi λρ if bi = 0. Then the definition of the weights kρ , lρ , mρ implies that
−kρ −lρ −mρ
λ ρ = t1 t2 t3 . Since dρi is given by dρi = λρρi λ−1 ρ
ρ ti ci , the assertion follows.
Now determine a coordinate on U (B). Divide ∂D(B) into 6 parts by the points
{Yi } and {Ai }. We denote them by Y ρ
1 A1 , A1 Y2 , . . . , A3 Y1 . For an arrow a = ai of
ρ
the McKay quiver, we denote by ã = ãi the arrow of the universal covering which
goes from h(ρ) to h(ρ) + fi . Note that if ãρi meets ∂D(B), then it is always going
out of D(B). Let C[ξ1 , ξ2 , ξ3 ] be the polynomial ring in three variables ξ1 , ξ2 , ξ3 .
We define a representation (uρi ) of the McKay quiver, with values uρi ∈ C[ξ1 , ξ2 , ξ3 ]
as follows.
(1) If ãρi is an edge of Γ(B), then uρi = 1.
(2) If ãρi meets Y ρ
p Ap , then ui = ξp .
(3) If ãρi meets A ρ

p Yp+1 , then ui = ξp ξp+1 (Set Y4 = Y1 etc.)
(4) If C(ρρi , j, k) ∈ Cycle(B) with {i, j, k} = {1, 2, 3} (i.e., ãρi is a diagonal of
the diamond), then uρi = ξ1 ξ2 ξ3 .
By virtue of Lemma 3.4, there are no other cases (note that edges of hexagons are
in one-to-one correspondence with arrows of the universal covering of the McKay
quiver).
Theorem 3.10. The above representation (uρi ) of the McKay quiver satisfies
the commutativity condition and gives rise to an isomorphism
Spec C[ξ1 , ξ2 , ξ3 ] ∼
= U (B).
Proof. We first prove that (uρi ) satisfies the commutativity condition. It fol-
lows from the equality uρ1 uρρ 1 ρρ1 ρ2
2 u3 = uρ1 uρρ 1 ρρ1 ρ3
3 u1 = ξ1 ξ2 ξ3 , because C[ξ1 , ξ2 , ξ3 ]
is an integral domain and each arrow gives a non-zero value in this ring. We
show uρ1 uρρ 1 ρρ1 ρ2
2 u3 = ξ1 ξ2 ξ3 , the other one is similar. First, if two of the arrows
ρ ρρ1 ρρ1 ρ2
a1 , a2 , a3 are edges of Γ(B), then they form half of a cycle in Cycle(B) by
virtue of the commutativity condition for B and the third arrow satisfies the as-
sumption of (4) above. Therefore, uρ1 uρρ 1 ρρ1 ρ2
2 u3 = 1 · 1 · ξ1 ξ2 ξ3 . Next assume exactly
one of the three arrows is in Γ(B). If we denote by a, a the other two arrows, then
Lemma 3.4 implies that the both of ã and ã meet ∂D(B). Moreover, we can see
that if one meets Y
p Ap , then the other meets Ap+1 Yp+2 . Therefore, we obtain
uρ1 uρρ
2
1 ρρ1 ρ2
u 3 = 1 · ξ p · ξ ξ
p+1 p+2 = ξ ξ ξ
1 2 3 . Finally, assume none of the three arrows
are in Γ(B). Then by Lemma 3.2, the arrows are aσ1 , aσρ 2
1
and aσρ
3
1 ρ2
. Assume that
Yp = h(σ) + (2/3)f1 + (1/3)f2 .
Since ãσ1 is going out of D(B), Lemma 3.6 implies that ãσi meets Y p Ap and hence
that uσ1 = ξp . Similar arguments show that uσρ 1
2 , uσρ
3
1 ρ2
are the other variables.
Thus we have proved that the commutativity relation holds for (uρi )i,ρ and we
obtain an injection Spec C[ξ1 , ξ2 , ξ3 ] → U .
Conversely, let R be a C-algebra and consider any R-valued point C = (cρi )i,ρ
of U . Let a and a be two arrows such that ã and ã meet Y p Ap . We show that the
values of C for a and a are the same. Y A
p p is a connected path formed by edges of

hexagons and the arrows of the universal covering meeting Yp Ap are in one-to-one
correspondence with these edges. Two adjacent edges belong to the same hexagon
and therefore one of two adjacent arrows(of the universal covering) meeting Y p Ap
is going out of D(B) and the other is going into D(B). Let a be an arrow of the
McKay quiver such that the lift ã meets Yp Ap , which is going out of D(B) by our
choice of the lift. Assume there is ã next to ã with the same properties. Then,
there exists an arrow β of the universal covering going into D(B) which meets both
ã and ã . Take arrows γ, γ of the universal covering such that ã, β, γ and ã , β, γ
form regular triangles respectively. Since both ã and ã meet Y p Ap , these triangles
surround neither Yp nor Ap . Therefore γ and γ are mapped to arrows in the McKay
quiver where B and hence C have value 1. Thus the commutativity relation implies
that the values of C for a and a are the same. Let ζp ∈ R be the common value.
We show that cρi are determined by ζ1 , ζ2 and ζ3 as in (1), (2), (3), (4) above. (1)
and (2) are obvious by the definitions of U and ζp .
Consider (3). The same argument as above shows that all the arrows a such
that ã meet A p YP +1 give the same value for C. Take such ã the nearest to Ap (if it
236 AKIRA ISHII
exists). Then do the similar arguments by using the triangle surrounding Ap . The
commutativity relation implies that these values are ζp ζp+1 .
Finally we show (4). The diagonals of adjacent diamonds give the same value
by the commutativity relation. Assume ã is the diagonal of a diamond, which
has an edge b̃ such that b̃ is not on any other diamond. Then the commutativity
relation for B implies that b̃ is the unique arrow in the triangle containing b̃ but not
containing ã such that it gives a non-zero value for B. Then (2) and (3) determine
the values of C at the other edges of the triangle. Again the commutativity relation
implies that C takes the value ζ1 ζ2 ζ3 at a.
3.5. Nowhere vanishing 3-form on Mθ . We now show that Mθ is a crepant

resolution by constructing a nowhere vanishing 3-form on Mθ . The following
construction is essentially due to Mukai [8]. Let E be a G-constellation which
gives rise to a point [E] ∈ Mθ . Then the Zariski tangent space of Mθ at [E] is
G-Ext1OC3 (E, E). We denote by ω the product (or composite) map
G-Ext1OC3 (E, E) × G-Ext1OC3 (E, E) × G-Ext1OC3 (E, E) −→ G-Ext3OC3 (E, E).
Serre duality and the stability of E imply
G-Ext3OC3 (E, E) ∼
= G-Ext0OC3 (E, E ⊗ KC3 )∨ ∼
= C.
We show that ω determines a nowhere vanishing 3-form on Mθ . Let B ∈ N

correspond to E. Then the Zariski tangent space is
{ C ∈ HomC[G] (R, V ⊗ R) | C ∧ B + B ∧ C = 0 }
G-Ext1OC3 (E, E) ∼
=
{ BA − (1V ⊗ A)B | A ∈ HomC[G] (R, R) }
and the above product is expressed as
ω(C1 , C2 , C3 ) = trace(C1 ∧ C1 ∧ C3 ).
Therefore, we have
ω(C1 , C2 , C3 ) = ω(C2 , C3 , C1 ).
On the other hand, the smoothness of Mθ implies that the obstruction class in
G-Ext2OC3 (E, E) vanishes. That is, we have ω(C, C, C3 ) = 0. Thus ω is skew-
symmetric.
We use the local coordinate ξ1 , ξ2 , ξ3 and the universal representation on U (B)
in Theorem 3.10. Then we obtain

∂ ∂ ∂
ω , , = ±1
∂ξ1 ∂ξ2 ∂ξ3
where the signature depends on the choice of a basis of ∧3 V . Thus ω does not
vanish on any affine open U (B) and is therefore nowhere vanishing on Mθ .
Remark 3.11. In contrast to the two-dimensional case [8], Serre duality doesn’t
imply that this is nowhere vanishing. But once we see that Mθ is smooth of
dimension three and that the (Fourier-Mukai) functor Φθ is fully faithful, then it
automatically follows that the above 3-form is nowhere vanishing.
References
1. T. Bridgeland, A. King, and M. Reid, The McKay correspondence as an equivalence of
derived categories, J. Amer. Math. Soc. 14 (2001), 535–554.
2. A. Bondal and D. Orlov, Semiorthogonal decomposition for algebraic varieties, Preprint
math.AG/9506012, (1995).
3. A. Craw and A. Ishii, Flops of G-Hilb and equvalences of derived categories by variaiton of
GIT quotient, To appear in Duke Math J., 43 pp.
4. R. P. Horja, Derived category automorphisms from mirror symmetry, Preprint
math.AG/0103231, (2001).
5. Y. Ito and H. Nakajima, The McKay correspondence and Hilbert schemes in dimension
three, Topology 39 (2000), 1155–1191.
6. A. King, Moduli of representations of finite dimensional algebras, Quart. J. Math. Oxford
45 (1994), 515–530.
7. P. Kronheimer, The construction of ALE spaces as hyper-Kähler quotients, J. Diff. Geom
29 (1989), 665–683.
8. S. Mukai, Symplectic structure of the moduli space of sheaves on an Abelian or k3 surface,
Invent. Math., 77 (1984), 101–116.
9. I. Nakamura, Hilbert schemes of Abelian group orbits, J. Alg. Geom. 10 (2000), 757–779.
10. M. Reid, McKay correspondence, Proc. of algebraic geometry symposium (Kinosaki, Nov
1996), T. Katsura (ed.), (1997), pp. 14–41.
11. A. Sardo-Infirri, Resolutions of orbifold singularities and the transportation problem on the
McKay quiver, Preprint math.AG/-9610005, (1996).
12. P. Seidel and R. Thomas, Braid group actions on derived categories of coherent sheaves,
Duke Math. J. 108 (2001), 37–108.
13. B. Szendrői, Diffeomorphisms and families of fourier–mukai transforms in mirror symmetry,
Applications of algebraic geometry to coding theory, physics and computation (Eilat, 2001),
NATO Sci. Ser. II Math. Phys. Chem., 36, Kluwer Acad. Publ., Dordrecht., (2001), pp. 317–
337.
Department of Urban and Environmental Engineering, Kyoto University, Kyoto

606-8501, Japan
E-mail address: akira@kusm.kyoto-u.ac.jp
Volume 3, 2004
Moduli Spaces of Bundles over Riemann Surfaces and the

Yang–Mills Stratification Revisited
Frances Kirwan
Abstract. Refinements of the Yang–Mills stratifications of spaces of connec-

tions over a compact Riemann surface Σ are investigated. The motivation
for this study is the search for a complete set of relations between the stan-
dard generators for the cohomology of the moduli spaces M(n, d) of stable
holomorphic bundles of rank n and degree d when n and d are coprime and
n > 2.
The moduli space M(n, d) of semistable holomorphic bundles of rank n and

degree d over a Riemann surface Σ of genus g ≥ 2 can be constructed as a quotient
of an infinite dimensional affine space of connections C(n, d) by a complexified gauge
group Gc (n, d), in an infinite-dimensional version of the construction of quotients in
Mumford’s geometric invariant theory [30]. When n and d are coprime, M(n, d) is
the topological quotient of the semistable subset C(n, d)ss of C(n, d) by the action of
Gc (n, d). Any nonsingular complex projective variety on which a complex reductive
group G acts linearly has a G-equivariantly perfect stratification by locally closed
nonsingular G-invariant subvarieties with its set of semistable points X ss as an
open stratum. This stratification can be obtained as the Morse stratification for
the normsquare of a moment map on X [22]; in the case of the moduli space
M(n, d) the rôle of the normsquare of the moment map is played by the Yang-Mills
functional. In [28] this Morse stratification of X is refined to obtain a stratification
of X by locally closed nonsingular G-invariant subvarieties with the set X s of stable
points of X as an open stratum. The other strata can be defined inductively in
terms of the sets of stable points of closed nonsingular subvarieties of X, acted on
by reductive subgroups of G, and their projectivised normal bundles.
In their fundamental paper [1], Atiyah and Bott studied a stratification of
C(n, d) defined using the Harder-Narasimhan type of a holomorphic bundle over
Σ, which they expected to be the Morse stratification of the Yang-Mills functional
(this was later shown to be the case [5]). The aim of this paper is to apply the
2000 Mathematics Subject Classification. Primary 14D20, 32G13, 14D21.

Key words and phrases. Moduli spaces of vector bundles, Yang-Mills stratification.
The author is a member of VBAC (Vector Bundles on Algebraic Curves), which is partially
supported by EAGER (EC FP5 Contract no. HPRN-CT-2000-00099), and acknowledges with
gratitude the hospitality of the University of Melbourne and the Australian National University
during the writing of part of this paper.
2004
239
240 FRANCES KIRWAN
methods of [28] to the Yang-Mills stratification to obtain refined stratifications of

C(n, d), and to relate these stratifications to natural refinements of the notion of
the Harder-Narasimhan type of a holomorphic bundle over Σ.
The motivation for this study was the search for a complete set of relations
among the standard generators for the cohomology of the moduli spaces M(n, d)
when the rank n and degree d are coprime and n > 2 [11]. The cohomology rings of
the moduli spaces M(n, d) have been the subject of much interest over many years;
see for example [1, 3, 7, 8, 14, 16, 17, 20, 21, 29, 31, 32, 38, 40, 41] among
many other pieces of work. In the case when n = 2 we now have a very thorough
understanding of the structure of the cohomology ring [2, 21, 37, 41]. For arbitrary
n it is known that the cohomology has no torsion and formulas for computing the
Betti numbers have been obtained, as well as a set of generators for the cohomology
ring [1, 6, 7, 16, 42]. When n = 2 the relations between these generators can be
explicitly described and in particular a conjecture of Mumford, that a certain set
of relations is a complete set, was proved some years ago [2, 21, 27, 37, 41].
However less is known about the relations between the generators when n > 2,
and the most obvious generalisation of Mumford’s conjecture to the cases when
n > 2 is false, although a modified version of the conjecture (using ‘dual Mumford
relations’ together with the original Mumford relations) is true for n = 3 [9]. There
is, however, a further generalisation of Mumford’s relations, and the attempt by
Richard Earl and the author to prove that this set is indeed complete was the
original stimulus for studying refinements of the Yang-Mills stratification, although
a different application has now appeared [11, 19].
The layout of this paper is as follows. §1 recalls background material on moduli
spaces of bundles and different versions of Mumford’s conjecture, while §2 reviews
the Morse stratification of the normsquare of a moment map and some refinements
of this stratification. §3 studies the structure of subbundles of semistable bundles
over Σ which are direct sums of stable bundles all of the same slope, and this is
used in §4 to define two canonical refinements of the Harder-Narasimhan filtration
of a holomorphic bundle over Σ, and thus to construct two refinements of the
Yang-Mills stratification. In the next two sections the stratification defined in §2
is applied to holomorphic bundles over Σ; its indexing set is studied in §5 and the
associated strata are investigated in §6. This stratification corresponds to a third
refinement of the Harder-Narasimhan filtration whose subquotients are all direct
sums of stable bundles of the same slope. The relationship between these three
filtrations is considered in §7, and §8 provides a brief conclusion.
1. Background material on moduli spaces of bundles

When n and d are coprime, the generators for the rational cohomology1 of
the moduli space M(n, d) given by Atiyah and Bott in [1] are obtained from a
(normalised) universal bundle V over M(n, d) × Σ. With respect to the Künneth
decomposition of H ∗ (M(n, d) × Σ) the rth Chern class cr (V ) of V can be written
as

2g
cr (V ) = ar ⊗ 1 + bjr ⊗ αj + fr ⊗ ω
j=1
1In this paper all cohomology will have rational coefficients.

THE YANG–MILLS STRATIFICATION REVISITED 241
where {1}, {αj : 1 ≤ j ≤ 2g}, and {ω} are standard bases for H 0 (Σ), H 1 (Σ) and
H 2 (Σ), and
(1.1) ar ∈ H 2r (M(n, d)), bjr ∈ H 2r−1 (M(n, d)), fr ∈ H 2r−2 (M(n, d)),
for 1 ≤ r ≤ n and 1 ≤ j ≤ 2g. It was shown by Atiyah and Bott [1, Prop. 2.20
and p.580] that the classes ar and fr (for 2 ≤ r ≤ n) and brj (for 1 ≤ r ≤ n and
1 ≤ j ≤ 2g) generate the rational cohomology ring of M(n, d).
Since tensoring by a fixed holomorphic line bundle of degree e gives an iso-
morphism between the moduli spaces M(n, d) and M(n, d + ne), we may assume
without loss of generality that
(2g − 2)n < d < (2g − 1)n.
This implies that H 1 (Σ, E) = 0 for any stable bundle of rank n and degree d [34,
Lemma 5.2], and hence that π! V is a bundle of rank d − n(g − 1) over M(n, d),
where
π : M(n, d) × Σ → M(n, d)
is the projection onto the first component and π! is the K-theoretic direct image
map. It follows that
cr (π! V ) = 0
for r > d − n(g − 1). Via the Grothendieck-Riemann-Roch theorem we can express
the Chern classes of π! V as polynomials in the generators ar , bjr , fr described above,
and hence their vanishing gives us relations between these generators. These are
Mumford’s relations, and they give us a complete set of relations when n = 2. We
can generalise them for n > 2 as follows.
Suppose that 0 < n̂ < n, and that dˆ is coprime to n̂. Then we have a universal
bundle V̂ over M(n̂, d)ˆ × Σ, and both V and V̂ can be pulled back to M(n̂, d) ˆ ×
M(n, d) × Σ. If
dˆ d
>
n̂ n
then there are no non-zero holomorphic bundle maps from a stable bundle of rank
n̂ and degree dˆ to a stable bundle of rank n and degree d, and hence, if
ˆ × M(n, d) × Σ → M(n̂, d)
π : M(n̂, d) ˆ × M(n, d),
is the projection onto the first two components, it follows that −π! (V̂ ∗ ⊗ V ) is a
bundle of rank nn̂(g − 1) − dn̂ + dn ˆ over M(n̂, d)
ˆ × M(n, d). Thus
0 = cr (−π! (V̂ ∗ ⊗ V )) ∈ H ∗ (M(n̂, d)

ˆ × M(n, d))
ˆ and hence the slant product
if r > nn̂(g − 1) − dn̂ + dn
cr (−π! (V̂ ∗ ⊗ V ))\γ ∈ H ∗ (M(n, d))
of cr (−π! (V̂ ∗ ⊗ V )) with any homology class γ ∈ H∗ (M(n̂, d))
ˆ vanishes when
ˆ
r > nn̂(g − 1) − dn̂ + dn.
The relations between the generators ar , bjr , fr obtained in this way for 0 < n̂ < n
and
d dˆ d
+1> >
n n̂ n
and
ˆ < r < nn̂(g + 1) − dn̂ + dn
nn̂(g − 1) − dn̂ + dn ˆ
242 FRANCES KIRWAN
(with a little more care taken when n̂ and dˆ are not coprime) are the ones we
consider. They are essentially Mumford’s relations when n = 2. To show that
these form a complete set of relations, a natural strategy is to consider the Yang-
Mills stratification which was used by Atiyah and Bott to obtain their generators
for the cohomology ring.
Recall that a holomorphic vector bundle E over Σ is called semistable (respec-
tively stable) if every holomorphic subbundle D of E satisfies
µ(D) ≤ µ(E), (respectively µ(D) < µ(E)),
where µ(D) = degree(D)/rank(D) is the slope of D. Bundles which are not
semistable are said to be unstable. Note that semistable bundles of coprime rank
and degree are stable.
Let E be a fixed C ∞ complex vector bundle of rank n and degree d over Σ. Let
C be the space of all holomorphic structures on E and let Gc denote the group of all
C ∞ complex automorphisms of E. Atiyah and Bott [1] identify the moduli space
M(n, d) with the quotient C ss /Gc where C ss is the open subset of C consisting of
all semistable holomorphic structures on E. The group Gc is the complexification
of the gauge group G which consists of all smooth automorphisms of E which are
unitary with respect to a fixed Hermitian structure on E. We shall write G for the
quotient of G by its U (1)-centre and G c for the quotient of Gc by its C∗ -centre.
There are natural isomorphisms
H ∗ (M(n, d)) ∼
= H ∗ (C ss /Gc ) = H ∗ (C ss /G c ) ∼
= H ∗ (C ss ) ∼
Gc = H ∗ (C ss )
G
between the cohomology of the moduli space and the G-equivariant cohomology of
C ss , since the C∗ -centre of Gc acts trivially on C ss , while G c acts freely on C ss and
G c is the complexification of G.
In order to show that the restriction map HG∗ (C) → HG∗ (C ss ) is surjective,
Atiyah and Bott consider the Yang–Mills (or Atiyah–Bott–Shatz) stratification of
C. This stratification {Cµ : µ ∈ M} is the Morse stratification for the Yang–
Mills functional on C, but it also has a more explicit description. It is indexed
by the partially ordered set M consisting of all the Harder–Narasimhan types of
holomorphic bundles of rank n and degree d, defined as follows. Any holomorphic
bundle E over M of rank n and degree d has a filtration
0 = E0 ⊂ E1 ⊂ · · · ⊂ EP = E
of subbundles such that the subquotients Qp = Ep /Ep−1 are semi-stable for 1 ≤
p ≤ P and satisfy
dp dp+1
µ(Qp ) = > = µ(Qp+1 )
np np+1
where dp and np are the degree and rank of Qp and µ(Qp ) is its slope. This filtration
is canonically associated to E and is called the Harder-Narasimhan filtration of E.
We define the type of E to be
µ = (µ(Q1 ), ..., µ(QP )) ∈ Qn
where the entry µ(Qp ) is repeated np times. The semistable bundles have type
µ0 = (d/n, ..., d/n) and form the unique open stratum. The set M of all possible
types of holomorphic vector bundles over Σ provides our indexing set, and if µ ∈ M
the subset Cµ ⊆ C is defined to be the set of all holomorphic vector bundles over Σ
of type µ. A partial order on M with the property that the closure of the stratum
indexed by µ is contained in the union of all strata indexed by µ ≥ µ is defined as

follows. Let σ = (σ1 , ..., σn ) and τ = (τ1 , ..., τn ) be two types; then

(1.2) σ ≥ τ if and only if σj ≥ τj for 1 ≤ i ≤ n − 1.
j≤i j≤i
The gauge group G acts on C preserving the stratification which is equivariantly

perfect with respect to this action, which means that its equivariant Thom-Gysin
sequences
j−2d
· · · → HG µ (Cµ ) → HGj (Uµ ) → HGj (Uµ − Cµ ) → · · ·
break up into short exact sequences
j−2dµ
0 → HG (Cµ ) → HGj (Uµ ) → HGj (Uµ − Cµ ) → 0.
Here

(1.3) dµ = (ni dj − nj di + ni nj (g − 1)),
i>j
is the complex codimension of Cµ in C and Uµ is the open subset of C which is the

union of all those strata labelled by µ ≤ µ; we also have

P
HG∗ (Cµ ) ∼
= ∗
HG(nj ,dj )
(C(nj , dj )ss ).
j=1
Atiyah and Bott show that the stratification is equivariantly perfect by considering
j−2d
the composition of the Thom-Gysin map HG µ (Cµ ) → HGj (Uµ ) with restriction to
Cµ , which is multiplication by the equivariant Euler class eµ of the normal bundle
to Cµ in C. They show that eµ is not a zero-divisor in HG∗ (Cµ ) and deduce that the
j−2d
Thom-Gysin maps HG µ (Cµ ) → HGj (Uµ ) are all injective.
So putting this all together Atiyah and Bott obtain inductive formulas for the
G-equivariant Betti numbers of C ss , and they also conclude that there is a natural
surjection
(1.4) H ∗ (BG) ∼= H ∗ (C) → H ∗ (C ss ) ∼
G G= H ∗ (M(n, d)).
Thus generators of the cohomology ring H ∗ (BG) give generators of the cohomology
ring M(n, d).
The classifying space BG can be identified with the space Mapd (Σ, BU (n)) of
all smooth maps f : Σ → BU (n) such that the pullback to Σ of the universal vector
bundle over BU (n) has degree d. If we pull back this universal bundle using the
evaluation map
Mapd (Σ, BU (n)) × Σ → BU (n) : (f, m) → f (m)
then we obtain a rank n vector bundle V over BG × Σ. If further we restrict the
pullback bundle induced by the maps

C ss × EG × Σ → C × EG × Σ → C ×G EG × Σ → BG × Σ
to C ss ×{e}×Σ for some e ∈ EG then we obtain a G-equivariant holomorphic bundle
on C ss × Σ. The C∗ -centre of G acts as scalar multiplication on the fibres, and
the associated projective bundle descends to a holomorphic projective bundle over
M(n, d) × Σ. In fact this is the projective bundle of a holomorphic vector bundle
V over M(n, d) × Σ which has the universal property that, for any [E] ∈ M(n, d)
representing a bundle E over Σ, the restriction of V to {[E]} × Σ is isomorphic to
244 FRANCES KIRWAN
E.
By a slight abuse of notation we define elements ar , bjr , fr in H ∗ (BG; Q) by
writing

2g
cr (V) = ar ⊗ 1 + bjr ⊗ αj + fr ⊗ ω 1 ≤ r ≤ n.
j=1
where, as before, ω is the standard generator of H 2 (Σ) and α1 , ..., α2g form a fixed
canonical cohomology basis for H 1 (Σ). In fact the ring H ∗ (BG) is freely generated
as a graded super-commutative algebra over Q by the elements
{ar : 1 ≤ r ≤ n} ∪ {bjr : 1 ≤ r ≤ n, 1 ≤ j ≤ 2g} ∪ {fr : 2 ≤ r ≤ n}
and if we omit a1 we get H ∗ (BG). These generators restrict to the generators
ar , bjr , fr given in (1.1) for H ∗ (M(n, d)) under the surjection (1.4).
The relations among these generators for H ∗ (M(n, d); Q) are then given by the
kernel of the restriction map (1.4) which is in turn determined by the map
(1.5) GH ∗ (C) ∼
= H ∗ (C) ⊗ H ∗ (BU (1)) → H ∗ (C ss ) ⊗ H ∗ (BU (1)) ∼
G G = H ∗ (C ss ),
G
and the proof that the Yang–Mills stratification is equivariantly perfect leads to
completeness criteria for a set of relations to be complete. Let R be a subset of the
kernel of the restriction map HG∗ (C) → HG∗ (C ss ). Suppose that for each unstable
type µ = µ0 there is a subset Rµ of the ideal generated by R in HG∗ (C) such that
the image of Rµ under the restriction map
HG∗ (C) → HG∗ (Cν )
is zero unless ν ≥ µ and when ν = µ contains the ideal of HG∗ (Cµ ) generated by the
equivariant Euler class eµ of the normal bundle to the stratum Cµ in C. Then R
generates the kernel of the restriction map HG∗ (C) → HG∗ (C ss ) as an ideal of HG∗ (C).
In fact Atiyah and Bott could have replaced the Yang-Mills stratification with
a coarser stratification of C and obtained equivalent results. For any integers n1
and d1 let Sn1 ,d1 be the subset of C consisting of all those holomorphic structures
with Harder-Narasimhan filtration 0 = E0 ⊂ E1 ⊂ · · · ⊂ Es = E where E1 has
rank n1 and degree d1 . We shall say that such a holomorphic structure has coarse
type (n1 , d1 ). Then Sn1 ,d1 is locally a submanifold of finite codimension
δn1 ,d1 = nd1 − n1 d + n1 (n − n1 )(g − 1)
in C and
(1.6) HG∗ (Sn1 ,d1 ) ∼ ∗
= HG(n1 ,d1 )
∗
(C(n1 , d1 )ss ) ⊗ HG(n−n1 ,d−d1 )
(U (n1 , d1 ))
where
U (n1 , d1 ) = S(n − n1 , d − d1 )n2 ,d2
d2 d
n2 < n1
1
is an open subset of C(n − n1 , d − d1 ). Moreover the equivariant Euler class en1 ,d1
of the normal to Sn1 ,d1 in C is not a zero divisor in HG∗ (Sn1 ,d1 ), so the stratification
of C by coarse type

d1 d
Sn1 ,d1 : 0 < n1 < n, > {Sn,d }
n1 n
is equivariantly perfect. This means that we can modify our completeness criteria,
ˆ with 0 < n̂ < n and dˆ > d we
so that for each pair of positive integers (n̂, d) n̂ n
require a set of relations whose restriction to HG∗ (Sn1 ,d1 ) is zero when d1 /n1 < d/n̂ ˆ
ˆ and n1 < n̂, and when (n1 , d1 ) = (n̂, d)
or d1 /n1 = d/n̂ ˆ equals the ideal of H ∗ (S ˆ)
G n̂,d
generated by the equivariant Euler class en̂,dˆ of the normal bundle to Sn̂,dˆ in C.
ˆ
G(n̂,d) ˆ s ) where d/n̂
ˆ > d/n and
It is easy enough to prove that if γ ∈ H∗ (C(n̂, d)
if r > nn̂(g − 1) + n̂d − ndˆ then the image of the slant product cr (−π! (V̂ ∗ ⊗ V))\γ
under the restriction map
H ∗ (C) → H ∗ (Sn ,d ) ∼
G G = H∗
1 1
(C(n1 , d1 )ss ) ⊗ H ∗
G(n1 ,d1 ) (U (n1 , d1 ))
G(n−n1 ,d−d1 )
ˆ
is zero when d1 /n1 < d/n̂, ˆ and n1 < n̂.
and also when d1 /n1 = d/n̂
ˆ ˆ ˆ is a manifold of dimen-
By Lefschetz duality, since C(n̂, d) /G(n̂, d) = Ms (n̂, d)
s
sion
ˆ = 2[(n̂2 − 1)(g − 1) + g] = 2(n̂2 g − n̂2 + 1)
D(n̂, d)
we have a natural map
LD : H∗
ˆ
G(n̂,d) ˆ s) ∼
(C(n̂, d) = H∗ (Ms (n̂, d))
ˆ
ˆ → H D(n̂,d)−∗ ˆ ss )
(C(n̂, d)
ˆ
G(n̂,d)
ˆ
G(n̂,d) ˆ
such that if γ ∈ H∗ ˆ s ) then LD(γ) lies in the dual of H G(n̂,d)
(C(n̂, d) ˆ ss )
(C(n̂, d)
ˆ
D(n̂,d)−∗
ˆ
and takes a G(n̂, d)-equivariant ˆ ss to its intersection, modulo G(n̂, d),
cycle on C(n̂, d) ˆ
with γ. When n̂ and dˆ are coprime then C(n̂, d) ˆ s and its quotient
ˆ ss equals C(n̂, d)
ˆ namely M(n̂, d),
by G(n̂, d), ˆ is a compact manifold. In this case Lefschetz duality
is essentially Poincaré duality and the map LD is an isomorphism.
We need to consider the restriction of a relation of the form cr (−π! (V̂ ∗ ⊗ V))\γ
ˆ
G(n̂,d)
to H ∗ (S ˆ). If γ ∈ H∗
G n̂,d
ˆ s ) where d/n̂
(C(n̂, d) ˆ > d/n and if r = nn̂(g − 1) + n̂d −
ndˆ+ 1 + j, then it turns out that the image of the slant product cr (−π! (V̂ ∗ ⊗ V))\γ
under the restriction map
HG∗ (C) → HG∗ (Sn̂,dˆ) ∼ ∗
= HG(n̂, ˆ ss ∗
ˆ (C(n̂, d) ) ⊗ HG(n−n̂,d−d)
d)
ˆ
ˆ (U (n̂, d))
(1)
is the product (−a1 )j LD(γ)en̂,dˆ of the equivariant Euler class en̂,dˆ of the normal
bundle to Sn̂,dˆ in C with the image of γ under the Lefschetz duality map LD
and j copies of minus the generator a
(1)
∈ H∗ ˆ ss ) ∼
(C(n̂, d) = H ∗ (BU (1)) ⊗
1 ˆ
G(n̂,d)
∗
HG(n̂, ˆs which comes from the copy of the polynomial ring H ∗ (BU (1)).
ˆ (C(n̂, d) )
d)
The proof of this is based on Porteous’s Formula (as in Beauville’s alternative
proof [2] of the theorem of Atiyah and Bott that the classes ar , bjr , fr generate
H ∗ (M(n, d)); cf. [39] and [12]), which allows us to deduce that the Poincaré dual
ˆ in
of ∆s × U (n̂, d)
∗ ˆs ∗ ˆ ss ∗ ˆ
HG(n̂, ˆ (C(n̂, d) ) ⊗ HG(n̂,d)
d) ˆ (C(n̂, d) ) ⊗ HG(n−n̂,d−d)
ˆ (U (n̂, d))
is cn̂2 (g−1)+1 (−π! (V̂ ∗ ⊗V1 ). In other words the restriction of cn̂2 (g−1)+1 (−π! (V̂ ∗ ⊗V1 ))
to C(n̂, d) ˆ ss × U (n̂, d)
ˆ s × C(n̂, d) ˆ is the image of 1 under the G(n̂, d)ˆ × G(n̂, d)
ˆ × G(n −
ˆ
n̂, d − d)-equivariant Thom-Gysin map associated to the inclusion of ∆s × U (n̂, d)) ˆ
ˆ ˆ ˆ
in C(n̂, d) × C(n̂, d) × U (n̂, d)). We can express the higher Chern classes of
s ss
−π! (V̂ ∗ ⊗ V1 ) in a similar way [11] by using Fulton’s Excess Porteous Formula [13].
∗ ˆ ss ∗ ˆ
Recall that given η ∈ HG(n̂, ˆ (C(n̂, d) ) ⊗ HG(n−n̂,d−d)
d) ˆ (C(n − n̂, d − d)) we
wish to find a relation whose restriction to C(n1 , d1 ) × U (n1 , d1 ) is zero when
ss
246 FRANCES KIRWAN
ˆ or d1 /n1 = d/n̂
d1 /n1 < d/n̂ ˆ and n1 < n, and when (n1 , d1 ) = (n̂, d) ˆ equals ηe ˆ.
n̂,d
We have found such a relation when η lies in the image of the Lefschetz duality
ˆ
G(n̂,d) ˆ s ) to H ∗ ˆ ss
map LD which maps H∗ (C(n̂, d) ˆ (C(n̂, d) ) and thus into
G(n̂,d)
∗ ˆ ss ∗ ˆ ss ∗
HG(n̂, ˆ (C(n̂, d) ) ⊗ H (BU (1)),
ˆ (C(n̂, d) ) = HG(n̂,d)
d)
(1)
and more generally when η has the form η = (−a1 )j LD(γ), for some element γ
ˆ
G(n̂,d) ˆ s ). When n̂ and dˆ are coprime this gives us all η ∈ H ∗
of H∗ (C(n̂, d) (C ss ).ˆ
G(n̂,d)
Moreover
∗ ˆ ss ∗ ˆ
HG(n̂, ˆ (C(n̂, d) ) ⊗ HG(n−n̂,d−d)
d) ˆ (C(n − n̂, d − d))
is generated as a module over HG∗ (C) by HG(n̂,

∗ ˆ ss ˆ
ˆ (C(n̂, d) ), so when n̂ and d are
d)
coprime, we have now obtained the relations we need from the slant products
ˆ
G(n̂,d)
{cr (−π! (V̂ ∗ ⊗ V))\γ : r ≥ nn̂(g − 1) + n̂d − ndˆ + 1, γ ∈ H∗ ˆ s )},
(C(n̂, d)
and a little more work reduces the range of r and dˆ (see [11] for more details).
This deals with the case when n̂ and dˆ are coprime, but the completeness
criteria have not yet been shown to hold for pairs n̂ and dˆ with common factors.
This was the original motivation for considering further modifications to the Yang–
Mills stratification. The difficulty with using the Yang–Mills stratification itself, or
the stratification of C by coarse type, is that in each case, although n and d are
chosen to be coprime so that semistability and stability coincide for n and d, in the
construction of the stratification other n̂ and dˆ appear for which semistability and
stability do not coincide.
2. Stratifying a set of semistable points

In this section we shall describe briefly how to stratify the set X ss of semistable
points of a complex projective variety X equipped with a linear action of a complex
reductive group G, so that the set X s of stable points of X is an open stratum (see
[15, 22, 30, 34] for background and [28] for more details).
We assume that X has some stable points but also has semistable points which
are not stable. In [23, 25] it is described how one can blow X up along a sequence
of nonsingular G-invariant subvarieties to obtain a G-invariant morphism X̃ → X
where X̃ is a complex projective variety acted on linearly by G such that X̃ ss = X̃ s .
The set X̃ ss can be obtained from X ss as follows. Let r > 0 be the maximal
dimension of a reductive subgroup of G fixing a point of X ss , and let R(r) be a
set of representatives of conjugacy classes of all connected reductive subgroups R
of dimension r in G such that
ss
ZR = {x ∈ X ss : R fixes x }
ss
is non-empty. Then R∈R(r) GZR is a disjoint union of nonsingular closed subva-
rieties of X . The action of G on X ss lifts to an action on the blow-up of X ss
ss
ss
along R∈R(r) GZR which can be linearised so that the complement of the set of
semistable points in the blow-up is the proper transform of the subset φ−1 (φ(GZRss
))
of X where φ : X → X//G is the quotient map (see [23] 7.17). Moreover no
ss ss
semistable point in the blow-up is fixed by a reductive subgroup of G of dimension

at least r, and a semistable point in the blow-up is fixed by a reductive subgroup
R of dimension less than r in G if and only if it belongs to the proper transform of

ss
the subvariety ZR of X ss .
If we repeat this process enough times, we obtain π : X̃ ss → X ss such that
X̃ = X̃ s . Equivalently we can construct a sequence
ss
ss
X(R 0)
= X ss , X(R
ss
1)
ss
, . . . , X(R τ)
= X̃ ss
where R1 , . . . , Rτ are connected reductive subgroups of G with
r = dim R1 ≥ dim R2 ≥ · · · dim Rτ ≥ 1,
and if 1 ≤ l ≤ τ then X(Rl ) is the blow up of X(R
ss
l−1 )
along its closed nonsingular
ss ∼
subvariety GZRl = G ×Nl ZRl , where Nl is the normaliser of Rl in G. Similarly,
ss
X̃//G = X̃ ss /G can be obtained from X//G by blowing up along the proper trans-
ss
forms of the images ZR //N in X//G of the subvarieties GZR of X ss in decreasing
order of dim R.
If 1 ≤ l ≤ τ then we have a G-equivariant stratification
{Sβ,l : (β, l) ∈ Bl × {l}}
of X(Rl ) by nonsingular G-invariant locally closed subvarieties such that one of the
strata, indexed by (0, l) ∈ Bl × {l}, coincides with the open subset X(Rss
l)
of X(Rl ) .
Here Bl is a finite subset of a fixed positive Weyl chamber t+ in the Lie algebra t
of a maximal compact torus T of G. In fact β ∈ t+ lies in Bl if and only if β is
the closest point to 0 in the convex hull in t of some nonempty subset of the set of
weights {α0 , . . . , αn } for the linear action of T on X(Rl ) .
There is a partial ordering on Bl given by γ > β if ||γ|| > ||β||, with 0 as its
minimal element, such that if β ∈ Bl then the closure in Xl of the stratum Sβ,l
satisfies

(2.1) Sβ,l ⊆ Sγ,l .
γ∈Bl ,γ≥β
If β ∈ Bl and β = 0 then the stratum Sβ,l retracts G-equivariantly onto its (trans-
verse) intersection with the exceptional divisor El for the blow-up X(Rl ) → X(R
ss
l−1 )
.
This exceptional divisor is isomorphic to the projective bundle P(Nl ) over GẐRss
l
,
ss
where ẐRl
is the proper transform of Z ss
Rl in X ss
(Rl−1 ) and Nl is the normal bundle
ss
to GẐRl
in XRss
l−1
. The stratification {Sβ,l : β ∈ Bl } is determined by the action of
Rl on the fibres of Nl over ZR ss
l
(see [23] §7).
There is thus a stratification {Σγ : γ ∈ Γ} of X ss indexed by
(2.2) Γ = {R1 } (B1 \{0}) × {1} · · · {Rτ } (Bτ \{0}) × {τ } {0}
defined as follows. We take as the highest stratum ΣR1 the nonsingular closed
ss
subvariety GZR whose complement in X ss can be naturally identified with the
ss ∼
1
complement X(R1 ) \E1 of the exceptional divisor E1 in X(R1 ) . We have GZR 1
=
G ×N1 ZR1 where N1 is the normaliser of R1 in G, and ZR1 is equal to the set
ss ss
of semistable points for the action of N1 , or equivalently for the induced action
of N1 /R1 , on ZR1 , which is a union of connected components of the fixed point
set of R1 in X. Moreover since R1 has maximal dimension among those reductive
subgroups of G with fixed points in X ss , we have ZR ss
1
= ZR s
1
s
where ZR l
denotes
the set of stable points for the action of Nl /Rl on ZRl for 1 ≤ l ≤ τ .
248 FRANCES KIRWAN
Next we take as strata the nonsingular locally closed subvarieties

Σβ,1 = Sβ,1 \E1 for β ∈ B1 with β = 0
of X(R1 ) \E1 = X ss \GZR ss
1
, whose complement in X(R1 ) \E1 is just X(R ss
1)
\E1 =
X(R1 ) \E1 where E1 = X(R1 ) ∩ E1 , and then we take the intersection of X(R
ss ss ss ss ss
1)
\E1
ss s s
with GZR 2
. This intersection is GZR2 where ZR2 is the set of stable points for the
action of N2 /R2 on ZR2 , and its complement in X(R1 ) \E1 can be naturally identified
ss
with the complement in X(R2 ) of the union of E2 and the proper transform Ê1 of
E1 .
The next strata are the nonsingular locally closed subvarieties
Σβ,2 = Sβ,2 \(E2 ∪ Ê1 ) for β ∈ B2 with β = 0
of X(R2 ) \(E2 ∪ Ê1 ), whose complement in X(R2 ) \(E2 ∪ Ê1 ) is X(R
ss
2)
\(E2 ∪ Ê1 ), and
the stratum after these is GZR3 . We repeat this process for 1 ≤ l ≤ τ and take X s
s
as our final stratum indexed by 0.

The given partial orderings on B1 , . . . , Bτ together with the ordering in the
expression (2.2) above for Γ induce a partial ordering on Γ, with R1 as the maximal
element and 0 as the minimal element, such that the closure in X ss of the stratum
Σγ indexed by γ ∈ Γ satisfies

(2.3) Σγ ⊆ Σγ̃ .
γ̃≥γ
s
It is possible to describe the strata Σγ in more detail. Either Σγ is GZR l
, for some
l, or else it is
Sβ,l \(El ∪ Êl−1 ∪ ... ∪ Ê1 )
for some l and β ∈ Bl \{0}, in which case by [22] §5 we have
ss ∼
(2.4) Sβ,l = GYβ,l = G ×Pβ Yβ,l
ss
ss
where Yβ,l ss
fibres over Zβ,l with fibre Cmβ,l for some mβ,l > 0, and Pβ is a parabolic
subgroup of G with the stabiliser Stab(β) of β under the adjoint action of G as its
maximal reductive subgroup. Here the fibration pβ : Yβ,lss
→ Zβ,l
ss
sends y to a limit
point of its orbit under the complex one-parameter subgroup of Rl generated by β.
Moreover
(2.5) Sβ,l ∩ El = G(Yβ,l
ss
∩ El ) ∼
= G ×Pβ (Yβ,l
ss
∩ El )
ss
where Yβ,l ∩ El fibres over Zβ,l
ss
with fibre Cmβ,l −1 (see [23] Lemmas 7.6 and 7.11).
Thus
(2.6) Sβ,l \El ∼
= G ×Pβ (Yβ,l
ss
\El )
ss
where Yβ,l \El fibres over Zβ,l
ss
with fibre Cmβ,l −1 ×(C\{0}). Furthermore if πl : El ∼
=
P(Nl ) → GẐRl is the projection then Lemma 7.9 of [23] tells us that when x ∈ ẐR
ss ss
l
the intersection of Sβ,l with the fibre πl−1 (x) = P(Nl,x ) of πl at x is the union
of those strata indexed by points in the adjoint orbit Ad(G)β in the stratification
of P(Nl,x ) induced by the representation ρl of Rl on the normal Nl,x to GẐR ss
l
at
x. Careful analysis [28] shows that we can, if we wish, replace the indexing set
Bl \{0}, whose elements correspond to the G-adjoint orbits Ad(G)β of elements of
the indexing set for the stratification of P(Nl,x ) induced by the representation ρl ,
by the set of their Nl -adjoint orbits Ad(Nl )β. Then we still have (2.4) – (2.6), but
now if qβ : Pβ → Stab(β) is the projection we have
Σγ = Σβ,l = Sβ,l \(El ∪ Êl−1 ∪ ... ∪ Ê1 ) ∼
\E
(2.7) = G ×Qβ,l Yβ,l
where Qβ,l = qβ−1 (Nl ∩ Stab(β)) and

\E ss
Yβ,l = Yβ,l \(El ∪ Êl−1 ∪ ... ∪ Ê1 ) ∩ p−1
β
ss
Zβ,l ∩ πl−1 (ẐR
ss
l
) ,
\E
and pβ : Yβ,l → Zβ,l
ss
∩ πl−1 (ẐR
ss
l
) is a fibration with fibre Cmβ,l −1 × (C\{0}).
This process gives us a stratification {Σγ : γ ∈ Γ} of X ss such that the stratum
indexed by the minimal element 0 of Γ coincides with the open subset X s of X ss . We
shall apply this construction to obtain a stratification of C ss , and thus inductively
to refine the Yang-Mills stratification {Cµ : µ ∈ M} of C by Harder–Narasimhan
type.
3. Direct sums of stable bundles of equal slope

In the good case when n and d are coprime, then C ss = C s and M(n, d) =
C /Gc is a nonsingular projective variety. When n and d are not coprime, we can
ss
use the description of M(n, d) as the geometric invariant theoretic quotient C//Gc
to construct a partial desingularisation M̃(n, d) of M(n, d). From this construction
we can use the methods described in §2 to obtain a stratification of C ss with C s as an
open stratum, and thus (by considering the subquotients of the Harder–Narasimhan
filtration) obtain a stratification of C refining the stratification {Cµ : µ ∈ M}. To
understand this refined stratification we need to use the description in [26] of the
partial desingularisation M̃(n, d).
In fact in [26] M̃(n, d) is not constructed using the representation of M(n, d)
as the geometric invariant theoretic quotient of C by Gc , although (as is noted at
[26, p.246]) this representation of M(n, d) would lead to the same partial desingu-
larisation. Instead in [26] M(n, d) is represented as a geometric invariant theoretic
quotient of a finite-dimensional nonsingular quasi-projective variety R(n̂, d) ˆ by a
ˆ ˆ
linear action of SL(p; C) where p = d + n̂(1 − g) with d >> 0. We may assume
that dˆ >> 0, since tensoring by a line bundle of degree l gives an isomorphism of
M(n, d) with M(n̂, dˆ + n̂l) for any l ∈ Z. By [34, Lemma 5.2] if E is a semistable
bundle over Σ of rank n̂ and degree dˆ > n̂(2g − 1) where g is the genus of Σ,
then E is generated by its sections and H 1 (Σ, E) = 0. If p = dˆ + n̂(1 − g), this
implies that dim H 0 (Σ, E) = p and that there is a holomorphic map h from Σ to
the Grassmannian G(n̂, p) of n̂-dimensional quotients of Cp such that the pullback
E(h) = h∗ T of the tautological bundle T on G(n̂, p) is isomorphic to E.
Let R(n̂, d)ˆ be the set of all holomorphic maps h : Σ → G(n̂, p) such that
∗
E(h) = h T has degree d and the map on sections Cp → H 0 (Σ, E(h)) induced
from the quotient bundle map Cp × Σ → E(h) is an isomorphism. For dˆ >> 0 this
ˆ has the structure of a nonsingular quasi-projective variety and there is
set R(n̂, d)
ˆ × Σ satisfying the following
a quotient E of the trivial bundle of rank p over R(n̂, d)
properties (see [34, §5]).
ˆ then the restriction of E to {h} × Σ is the pullback E(h) of
(i) If h ∈ R(n̂, d)
the tautological bundle T on G(n̂, p) along the map h : Σ → G(n̂, p).
ˆ then E(h1 ) and E(h2 ) are isomorphic as bundles
(ii) If h1 and h2 lie in R(n̂, d)
250 FRANCES KIRWAN
over Σ if and only if h1 and h2 lie in the same orbit of the natural action of GL(p; C)
ˆ
on R(n̂, d).
(iii) If h ∈ R(n̂, d) ˆ then the stabiliser of h in GL(p; C) is isomorphic to the
group Aut(E(h)) of complex analytic automorphisms of E(h).
If N >> 0 then R(n̂, d) ˆ can be embedded as a quasi-projective subvariety of the
product (G(n̂, p)) by a map of the form h → (h(x1 ), . . . . , h(xN )) where x1 , . . . , xN
N
are points of Σ. This embedding gives a linearisation of the action of GL(p; C) on

ˆ If N >> 0 and d >> 0 then we also have the following:
R(n̂, d).
(iv) The point h ∈ R(n̂, d) ˆ is semistable in the sense of geometric invari-
ant theory for the linear action of SL(p; C) on R(n̂, d) ˆ if and only if E(h) is a
ˆ ss
semistable bundle. If h1 and h2 lie in R(n̂, d) then they represent the same point
ˆ
of R(n̂, d)//SL(p; C) if and only if gr(E(h1 )) ∼ = gr(E(h2 )), and thus there is a
ˆ
natural identification of M(n, d) with R(n̂, d)//SL(p; C) (see for example [34, §5]).
It is shown in [24] that the Atiyah–Bott formulas for the equivariant Betti
numbers of C ss can be obtained by stratifying R(n, d) instead of C(n, d), and in fact
throughout this paper we could work with either R(n, d) or C(n, d). In particular
properties (i) to (iv) above imply that the analysis in [26] of the construction of
the partial desingularisation of M̃(n̂, d) ˆ as R̃(n̂, d)//SL(p;
ˆ C) applies equally well if
we work with C//Gc .
˜
To describe the construction of M̃(n, d), first of all we need to find a set
R of representatives of the conjugacy classes of reductive subgroups of SL(p; C)
which occur as the connected components of stabilisers of semistable points of
R(n, d). In fact it is slightly simpler to describe the corresponding subgroups of
G = GL(p; C), and since the central one-parameter subgroup of GL(p; C) consisting
of nonzero scalar multiples of the identity acts trivially on R(n, d), finding stabilis-
ers in GL(p; C) is essentially equivalent to finding stabilisers in SL(p; C). By [26,
pp. 248-9] such conjugacy classes in GL(p; C) correspond to unordered sequences
(m1 , n1 ), ..., (mq , nq ) of pairs of positive integers such that m1 n1 + ... + mq nq = n
and n divides ni d for each i. A representative R of the corresponding conjugacy
class is given by
R = GL(m1 ; C) × ... × GL(mq ; C)
q
embedded in GL(p; C) using a fixed isomorphism of Cp with i=1 (C
mi
⊗ Cpi ))
where di = ni d/n and pi = di + ni (1 − g) = ni p/n. Then GZR consists of all those
ss
holomorphic structures E with
(3.1) E∼
= (Cm1 ⊗ D1 ) ⊕ · · · ⊕ (Cmp ⊗ Dq )
s
where D1 , . . . , Dq are all semistable and Di has rank ni and degree di , while GZR
consists of all those holomorphic structures E with
(3.2) E∼
= (Cm1 ⊗ D1 ) ⊕ · · · ⊕ (Cmp ⊗ Dq )
where D1 , . . . , Dq are all stable and not isomorphic to one another, and Di has
rank ni and rank di . Moreover the normaliser N of R in GL(p; C) has connected
component

(3.3) N0 ∼= (GL(mi ; C) × GL(pi ; C))/C∗

1≤i≤q
where C∗ is the diagonal central one-parameter subgroup of GL(mi ; C)×GL(pi ; C),

and π0 (N ) = N/N0 is the product

(3.4) π0 (N ) = Sym(#{i : mi = j and ni = k})

j≥0,k≥0
where Sym(b) denotes the symmetric group of permutations of a set with b elements.
Furthermore in terms of the notation of §2, if R = Rl then a holomorphic structure
belongs to one of the strata Σβ,l with β ∈ Bl \{0} if and only if
it has a filtration
0 = E0 ⊂ E1 ⊂ ... ⊂ Es = E such that E is not isomorphic to 1≤k≤s Ek /Ek−1
but
Ek /Ek−1 ∼= (Cm1 ⊗ D1 ) ⊕ · · · ⊕ (Cmp ⊗ Dq )
1≤k≤s
where D1 , . . . , Dq are all stable and not isomorphic to one another, and Di has
rank ni and rank di [26, p. 248]. Thus to understand the refined Yang-Mills
stratification of C, we need to study refinements
0 = E0 ⊂ E1 ⊂ ... ⊂ Es = E
of the Harder-Narasimhan filtration of a holomorphic bundle E, such that each
subquotient Ej /Ej−1 is a direct sum of stable bundles all of the same slope. For
this recall the following standard result (cf. the proof of [26, Lemma 3.2] and [35]).
Proposition 3.1. Any semistable bundle E has a canonical subbundle of the
form
(Cm1 ⊗ D1 ) ⊕ · · · ⊕ (Cmq ⊗ Dq )
with D1 , . . . , Dq not isomorphic to each other and all stable of the same slope as E,
such that any other subbundle of the same form

(Cm1 ⊗ D1 ) ⊕ · · · ⊕ (Cmr ⊗ Dr ),
with D1 , ..., Dr not isomorphic to each other and all stable of the same slope as E,
satisfies r ≤ q and (after permuting the order of D1 , ..., Dq suitably) Dj ∼ = Dj for
all 1 ≤ j ≤ r and the inclusion

(Cm1 ⊗ D1 ) ⊕ · · · ⊕ (Cmr ⊗ Dr ) → E
factors through the inclusion (Cm1 ⊗ D1 ) ⊕ · · · ⊕ (Cmq ⊗ Dq ) → E via injections

Cmj → Cmj for 1 ≤ j ≤ r.
If we choose a subbundle of E of maximal rank among those of the required
form, this follows immediately from
Lemma 3.2. Let E, D1 , . . . , Dq , D1 , . . . , Dr be bundles over Σ all of the same
slope, with E semistable, D1 , ..., Dq and D1 , . . . , Dr all stable and Dj = Dj for 1 ≤
j ≤ k for some 0 ≤ k ≤ min{q, r}, but with no other isomorphisms between the
bundles D1 , ..., Dq and D1 , ..., Dr . If
α : (Cm1 ⊗ D1 ) ⊕ · · · ⊕ (Cmq ⊗ Dq ) → E
and
β : (Cm1 ⊗ D1 ) ⊕ · · · ⊕ (Cmr ⊗ Dr ) → E
are injective bundle homomorphisms, then there exist nonnegative integers n1 , ..., nk

and linear injections ij : Cmj → Cnj and ij : Cmj → Cnj for 1 ≤ j ≤ k and an
injective bundle homomorphism γ from
(Cn1 ⊗ D1 ) ⊕ · · · ⊕ (Cnk ⊗ Dk ) ⊕ (Cmk+1 ⊗ Dk+1 ) ⊕ · · · ⊕ (Cmq ⊗ Dq )
252 FRANCES KIRWAN

⊕(Cmk+1 ⊗ Dk+1 ) ⊕ · · · ⊕ (Cmr ⊗ Dr )
to E such that α and β both factorise through γ via the injections ij and ij for
1 ≤ j ≤ k in the obvious way.
Proof: Consider the kernel of

α ⊕ β : (Cm1 +m1 ⊗ D1 ) ⊕ · · · ⊕ (Cmk +mk ⊗ Dk ) ⊕
(Cmk+1 ⊗ Dk+1 ) ⊕ · · · ⊕ (Cmq ⊗ Dq ) ⊕

(Cmk+1 ⊗ Dk+1 ) ⊕ · · · ⊕ (Cmr ⊗ Dr ) → E.
The proof of [34, Lemma 5.1 (iii)] shows that this kernel is a subsheaf of the domain
of α ⊕ β which has the same slope as E, D1 , ..., Dq , and D1 , ..., Dr . Induction on
m1 + ... + mq + m1 + ... + mr using [34, Lemma 5.1] and the obvious projection
from Cm1 onto Cm1 −1 shows that such a subsheaf is in fact of the form

(U1 ⊗ D1 ) ⊕ · · · ⊕ (Uq ⊗ Dq ) ⊕ (Uq+1 ⊗ Dk+1 ) ⊕ · · · ⊕ (Uq−k+r ⊗ Dr )

for some linear subspaces Uj of Cm1 +m1 , ..., Cmk+1 , ..., Cmr . Then since α and β
are injective, we must have ker(α ⊕ β) = (U1 ⊗ D1 ) ⊕ · · · ⊕ (Uk ⊗ Dk ) for linear

subspaces Uj of Cmj ⊕ Cmj satisfying Uj ∩ Cmj = {0} = Uj ∩ Cmj for 1 ≤ j ≤ k.
The result follows easily.
A similar argument gives us

Corollary 3.3. Let E be semistable and D1 , . . . , Dr all stable of the same
slope as E and not isomorphic to each other. Then

k
H 0 (Σ, ((Cm1 ⊗ D1 ) ⊕ · · · ⊕ (Cmr ⊗ Dr ))∗ ⊗ E) ∼

= ((Cmj )∗ ⊗ Cmj )
j=1
where (Cm1 ⊗D1 )⊕· · ·⊕(Cmq ⊗Dq ) is the canonical subbundle of this form associated
to E as in Proposition 3.1, and without loss of generality we assume Dj ∼ = Dj for
1 ≤ j ≤ k for some 0 ≤ k ≤ min{q, r} and that there are no other isomorphisms
between the bundles D1 , ..., Dq and D1 , ..., Dr .
Definition 3.4. With the notation above we set

q
r
gr(E) = (Cmij ⊗ Di ) .
i=1 j=1
for any semistable bundle E.

Remark 3.5. Of course, by the Jordan–Hölder theorem, given any filtration
0 = D0 ⊂ D1 ⊂ ... ⊂ Dt = E of E such that Dj /Dj−1 is a direct sum of stable
t
bundles of the same slope as E, we have j=1 Dj /Dj−1 ∼
= gr(E).
4. Maximal and minimal Jordan–Hölder filtrations

Recall that the Harder-Narasimhan filtration of a holomorphic bundle E over
Σ is a canonical filtration
0 = F0 ⊂ F1 ⊂ ... ⊂ Fs = E
of E such that Fj /Fj−1 is semistable and slope(Fj /Fj−1 ) > slope(Fj+1 /Fj ) for
0 < j < s. In the last section we saw that any semistable bundle E has a canonical
maximal subbundle of the form
(Cm1 ⊗ D1 ) ⊕ · · · ⊕ (Cmq ⊗ Dq )
where D1 , . . . , Dq are not isomorphic to each other and are all stable of the same
slope as E. This subbundle is nonzero if E = 0, since any nonzero semistable
bundle is either itself stable or it has a proper stable subbundle of the same slope.
Therefore any semistable bundle E has a canonical filtration
(4.1) 0 = E0 ⊂ E1 ⊂ E2 ⊂ · · · ⊂ Er = E
whose subquotients are direct sums of stable bundles, which is defined inductively
so that
(4.2) Ej /Ej−1 ∼= (Cm1j ⊗ D1 ) ⊕ · · · ⊕ (Cmqj ⊗ Dq )
where D1 , ..., Dq are stable nonisomorphic bundles all of the same slope as E with
nonnegative integers mij for 1 ≤ i ≤ q and 1 ≤ j ≤ r, and Ej /Ej−1 is the maximal
subbundle of E/Ej−1 of this form. If, moreover, we assume that

r
mij > 0 for all 1 ≤ i ≤ q
j=1
then the filtration (4.1), the bundles Di and integers mij (for 1 ≤ i ≤ q and
1 ≤ j ≤ r) and the decompositions (4.2) are canonically associated to E up to
isomorphism of the bundles Di , the usual action of GL(mij ; C) on Cmij and the
obvious action of the permutation group Sym(q) on this data. We can generalise
the definition to the case when E is not necessarily semistable, by applying this
construction to the subquotients of the Harder-Narasimhan filtration of E. This
gives us a canonical refinement
0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E
of the Harder-Narasimhan filtration such that each subquotient Ej /Ej−1 is the
maximal subbundle of E/Ej−1 which is a direct sum of stable bundles all having
maximal slope among the nonzero subbundles of E/Ej−1 . We shall call this refine-
ment of the Harder-Narasimhan filtration the maximal Jordan–Hölder filtration of
E.
Definition 4.1. Let s, q1 , ..., qs and r1 , ..., rs be positive integers and let dk ,
nik and mijk (for 1 ≤ i ≤ qk , 1 ≤ j ≤ rk and 1 ≤ k ≤ s) be integers satisfying
nik > 0, mijk ≥ 0, nk > 0
and

s
s
dk nik d1 d2 ds
n= nk , d= dk , = dik ∈ Z, > > ... >
nk n1 n2 ns
k=1 k=1
where

qk
rk
nk = nik mijk .
i=1 j=1
qk ,s qk ,rk ,s
Denote by [d, n, m] = [(dk )sk=1 , (nik )i=1,k=1 , (mijk )i=1,j=1,k=1 ] the orbit of
qk ,s qk ,rk ,s
(d, n, m) = ((dk )sk=1 , (nik )i=1,k=1 , (mijk )i=1,j=1,k=1 )
254 FRANCES KIRWAN
under the action of the product of symmetric groups Σq1 × ... × Σqk on the set of
ˆ denote the set of all such orbits, for fixed n̂ and
such sequences, and let I = I(n̂, d)
ˆ ˆ
d. Given [d, n, m] ∈ I(n̂, d) let s([d, n, m] = s and let S[d,n,m]
maxJH
denote the subset
of C consisting of those holomorphic structures on our fixed smooth bundle of rank
n̂ and degree dˆ whose maximal Jordan–Hölder filtration
0 = E0,1 ⊂ E1,1 ⊂ · · · ⊂ Er1 ,1 = E0,2 ⊂ E1,2 ⊂ · · ·
· · · ⊂ Ers−1 ,s−1 = Ers ,0 ⊂ · · · ⊂ Ers ,s = E
satisfies
Ej,k /Ej−1,k ∼
= (Cm1jk ⊗ D1k ) ⊕ · · · ⊕ (Cmqk jk ⊗ Dqk k )
for 1 ≤ k ≤ s and 1 ≤ j ≤ qk , where D1k , . . . , Dqk k are nonisomorphic stable
bundles with
rank(Dik ) = nik and deg(Dik ) = dik
and Ej,k /Ej−1,k is the maximal subbundle of E/Ej−1,k isomorphic to a direct sum
maxJH
of stable bundles of slope dk /nk . To make the notation easier on the eye, S[d,n,m]
will often be denoted by S[d,n,m] .
Let I ss denote the subset of I consisting of all orbits [d, n, m] for which
s([d, n, m] = 1. For simplicity we shall write [n, m] for [d, n, m] when s([d, n, m]) =
1 (which means that d = (d)).
We have now proved
Lemma 4.2. C is the disjoint union of the subsets
{S[d,n,m] : [d, n, m] ∈ I}.
Remark 4.3. Note that when s([d, n, m]) = 1 and q1 = r1 = 1 and n11 = n̂
and m111 = 1 then we get S[(n),(1)] = C s , and moreover

C(n, d)ss = S[d,n,m] = S[n,m] .
[d,n,m]∈I,s([d,n,m])=1 [n,m]∈I ss
Remark 4.4. Let E be a semistable holomorphic structure on E. If E repre-

sents an element of the closure of S[n,m] in C(n, d)ss for some
[n, m] = [(ni )qi=1 , (mij )q,r
i=1,j=1 ] ∈ I ,
ss
then E has a filtration 0 = E0 ⊂ E1 ⊂ E2 ⊂ · · · ⊂ Er = E such that if 1 ≤ j ≤ r

then
Ej /Ej−1 ∼
= (Cm1j ⊗ D1 ) ⊕ · · · ⊕ (Cmqj ⊗ Dq )
where D1 , ..., Dq are semistable bundles all having the same slope as E, but this fil-
tration is not necessarily the maximal Jordan–Hölder filtration of E. If the bundles
D1 , ..., Dq are not all stable or if two of them are isomorphic to each other, then

q r
dim Aut(gr(E)) > ( mij )2
i=1 j=1

and so E lies in S[n ,m ] where m = (mij )qi=1,j=1
,r
satisfies

q
r
q r
( mij )2 > ( mij )2 .
i=1 j=1 i=1 j=1
If, on the other hand, D1 , ..., Dq are all stable and not isomorphic to each other,
then the maximal Jordan–Hölder filtration of E is of the form
0 = E0 ⊂ E1 ⊂ E2 ⊂ · · · ⊂ Er = E
with
Ej /Ej−1
∼
= (Cm1j ⊗ D1 ) ⊕ · · · ⊕ (Cmqj ⊗ Dq )
for 1 ≤ j ≤ r , where 1 ≤ r ≤ r and
mi1 + ... + mir = mi1 + ... + mir
and mi1 + ... + mij ≥ mi1 + ... + mij for 1 ≤ i ≤ q and 1 ≤ j ≤ r . Thus we can
define a partial order ≥ on I ss such that [n , m ] ≥ [n, m] if and only if either

q
r
q r
2
( mij ) > ( mij )2
i=1 j=1 i=1 j=1
or n = n and 1 ≤ r ≤ r and
mi1 + ... + mir = mi1 + ... + mir
and mi1 + ... + mij ≥ mi1 + ... + mij for 1 ≤ i ≤ q and 1 ≤ j ≤ r , and then the
closure of S[n,m] in C(n, d)ss is contained in

S[n ,m ] .
[n ,m ]≥[n,m]
Using (1.2), we can then extend this partial order to I so that

S[d,n,m] ⊆ S[d ,n ,m ] .
[d ,n ,m ]≥[d,n,m]
Proposition 4.5. Let [n, m] = [(ni )qi=1 , (mij )q,r

i=1,j=1 ] ∈ I . Then S[n,m] is a
ss
locally closed complex submanifold of C of finite codimension

ss
 
q
r−1
q q
r
mij mij+1 + (g − 1)  m i 1 j1 m i 2 j2 n i 1 n i 2 − (mij ni )2  .
i=1 j=1 i1 ,i2 =1 1≤j1 ≤j2 ≤r i=1 j=1
Proof: (cf. [1, §7]) The rank and degree are the only C ∞ invariants of a vector
bundle over Σ. Thus we may choose a C ∞ isomorphism of our fixed C ∞ bundle E
over Σ with a bundle of the form
q r
(Cmij ⊗ Di )
i=1 j=1
where Di is a fixed C ∞ bundle over Σ of rank ni and degree di = ni d/n̂

ˆ for 1 ≤ i ≤ q.
Let Y[n,m] be the subset of C consisting of all semistable holomorphic struc-
ss
tures E on E for which the subbundles

q j
Ej = (Cmik ⊗ Di )
i=1 k=1
are holomorphic for 1 ≤ j ≤ r, and for which there are nonisomorphic stable
∞
holomorphic structures D1 , . . . , Dq on the
q C mbundles D1 , . . . , Dq such that the
natural identification
of E j /E j−1 with i=1 C ij
⊗ Di becomes an isomorphism
of Ej /Ej−1 with qi=1 Cmij ⊗ Di for 1 ≤ j ≤ r, and finally for each 1 ≤ j ≤ r the
256 FRANCES KIRWAN
quotient Ej /Ej−1 is the maximal subbundle of E/Ej−1 isomorphic to a direct sum

of stable bundles of the same slope as E/Ej−1 . Let Gc [n, m] be the subgroup of
the complexified gauge group Gc consisting of all C ∞ complex automorphisms of E

by the subbundles qi=1 jk=1 (Cmik ⊗ Di ) and
which preserve the filtration of E
q
the decomposition of Ej /Ej−1 as i=1 (Cmij ⊗ Di ) up to the action of the general
linear groups GL(mij ; C) and the permutation groups
Sym(#{i : ni = k and mij = lj , j = 1, ..., r})
for all nonnegative integers k and l1 , ..., lr . Since the filtration (4.1) and decompo-
sitions of E1 /E0 ,...,Er /Er−1 are canonically associated to E up to the actions of
these general linear and permutation groups, we have
S[n,m] = Gc Y[n,m] ∼
= Gc ×Gc [n,m] Y[n,m] .
As in [1, §7] we have that Y[n,m] is an open subset of an affine subspace of the
infinite-dimensional affine space C and the injection
(4.3) Gc ×Gc [n,m] Y[n,m] → C
is holomorphic with image S[n,m] .
If E ∈ Y[n,m] , let End E be the subbundle of End E consisting of holomorphic
endomorphisms of E preserving the maximal Jordan–Hölder filtration (4.1) and
decomposition (4.2) up to isomorphism of the bundles Di and the vector spaces
Cmij . Let End E be the quotient of End E by End E. The normal to the Gc -
orbit of E in C can be canonically identified with H 1 (Σ, End E) (see [1, §7]) and
the image of TE Y[n,m] in this can be canonically identified with the image of the
natural map
H 1 (Σ, End E) → H 1 (Σ, End E)
which fits into the long exact sequence of cohomology induced by the short exact
sequence of bundles
0 → End E → End E → End E → 0.
Thus we get an isomorphism
TE C/(TE Y[n,m] + TE O) ∼
= H 1 (Σ, End E)
where O is the Gc -orbit of E in C.
We have short exact sequences
0 → (E/E1 )∗ ⊗ E → End E → E1∗ ⊗ E → 0
and
0 → E1∗ ⊗ E1 → E1∗ ⊗ E → E1∗ ⊗ (E/E1 ) → 0.
Let K be the kernel of the composition of surjections
End E → E1∗ ⊗ E → E1∗ ⊗ (E/E1 ).
Then we have short exact sequences
0 → K → End E → E1∗ ⊗ (E/E1 ) → 0
and
0 → (E/E1 )∗ ⊗ E → K → E1∗ ⊗ E1 → 0.
Now End E ⊆ K and the image of End E in E1∗ ⊗ E1 is

q
End(Cmi1 ) ⊗ End(Di ),
i=1

so since End E = End E/End E we get short exact sequences

K ∗ E
(4.4) 0→ → End E → E ⊗ →0
End E 1
E1
and
(4.5)
End E + ((E/E1 )∗ ⊗ E) K E1∗ ⊗ E1
0→ → → q mi1 ) ⊗ End(D )
→ 0.
End E End E i=1 End(C i
Since
End E + ((E/E1 )∗ ⊗ E) ∼ (E/E1 )∗ ⊗ E
=
End E
End E ∩ ((E/E1 )∗ ⊗ E)
∼ ((E/E1 )∗ ⊗ E)/((E/E1 )∗ ⊗ E1 )
=
(End E ∩ ((E/E1 )∗ ⊗ E))/((E/E1)∗ ⊗ E1 )
∼ (E/E1 )∗ ⊗ (E/E1 )
= = End (E/E1 ),
End (E/E1 )
the short exact sequence (4.5) becomes

E K E1∗ ⊗ E1
(4.6) 0 → End → → q m2i1 ⊗ D ∗ ⊗ D
→ 0.
i=1 C
E1 End E i i
From the sequences (4.4) and (4.6) it follows that the rank of End E is
rank(End E) = rank(K/End E) + rank(E1∗ ⊗ (E/E1 ))

q

= rank(End (E/E1 )) + rank(E1∗ ⊗ E1 ) − (mi1 )2 rank(Di∗ ⊗ Di )
i=1
+ rank(E1∗ ⊗ (E/E1 ))

q
= rank(End (E/E1 )) + rank(E1∗ ⊗ E) − (mi1 )2 (ni )2 .
i=1
Thus by induction on r we have
q
q
r
rank(End E) = m i 1 j1 m i 2 j2 n i 1 n i 2 − (ni )2 .
i1 ,i2 =1 1≤j1 ≤j2 ≤r i=1 j=1
ˆ as E we have deg(End E) = 0.
Since D1 , . . . , Dq all have the same slope d/n̂
Therefore by Riemann-Roch
dim H 1 (Σ, End E) = dim H 0 (Σ, End E)
 
q
q
r
+(g − 1)  m i 1 j1 m i 2 j2 n i 1 n i 2 − (ni )2  .
i1 ,i2 =1 1≤j1 ≤j2 ≤r i=1 j=1
Moreover the short exact sequences (4.4) and (4.6) give us long exact sequences of
cohomology
0 → H 0 (Σ, K/End E) → H 0 (Σ, End E) → H 0 (Σ, E1∗ ⊗ (E/E1 )) → · · ·
258 FRANCES KIRWAN
and
0 → H 0 (Σ, End (E/E1 )) → H 0 (Σ, K/End E)

q
→ H 0 (Σ, E1∗ ⊗ E1 / End(Cmij ) ⊗ End(Di )) → · · ·
i=1
q
Now E1 ∼ = i=1 Cmi1 ⊗ Di where D1 , ..., Dq are nonisomorphic stable bundles all
of the same slope as E1 , and

q
E2 /E1 ∼
= Cmi2 ⊗ Di
i=1
is the maximal subbundle of E/E1 which is a direct sum of stable bundles all of
the same slope as E/E1 . Since E1 and E/E1 have the same slope, it follows from
Corollary 3.3 that

q
H 0 (Σ, E1∗ ⊗ (E/E1 )) = H 0 (Σ, E1∗ ⊗ (E2 /E1 )) ∼
= (Cmi1 )∗ ⊗ Cmi2 .
i=1
By choosing an open cover U of Σ such that the filtration 0 = E0 ⊂ E1 ⊂ E2 ⊂
· · · ⊂ Er = E is trivial over each U ∈ U, and describing E in terms of upper
triangular transition functions on E1 ⊕ (E2 /E1 ) ⊕ ... ⊕ (E/Er−1 )|U∩V for U, V ∈ U
which induce the identity on Ej /Ej−1 for 1 ≤ j ≤ r, we see that the natural map
H 0 (Σ, End E) → H 0 (Σ, E1∗ ⊗ (E/E1 )) = H 0 (Σ, E1∗ ⊗ (E2 /E1 ))
is surjective. Also

q
H 0 (Σ, E1∗ ⊗E1 / End(Cmi1 ⊗End(Di )) = (Cmi1 )∗ ⊗Cmi1 ⊗H 0 (Σ, Di∗ ⊗Dj ) = 0,
i=1 i =j
so

q

0
dim H (Σ, End E) = mi1 mi2 + dim H 0 (Σ, End (E/E1 ))
i=1
and thus by induction on r we have

q
r−1
0
dim H (Σ, End E) = mij mij+1 .
i=1 j=1
Therefore dim H 1 (Σ, End E) is equal to

(4.7)  
q r−1 q
q
r
mij mij+1 + (g − 1)  m i 1 j1 m i 2 j2 n i 1 n i 2 − (mij ni )2  .
i=1 j=1 i1 ,i2 =1 1≤j1 ≤j2 ≤r i=1 j=1
In particular this tells us that the image in C of the derivative of the injection (4.3)
has constant codimension, and it follows as in [1, §7] (see also [1, §§14 and 15])
that the subset S[n,m] is locally a complex submanifold of C of finite codimension
given by (4.7).
Corollary 4.6. If [d, n, m] ∈ I then S[d,n,m] is a locally closed complex sub-
manifold of C of codimension

s
q
r−1
(nk1 dk2 − nk2 dk1 + nk1 nk2 (g − 1)) + mijk mij+1k
1≤k2 <k1 ≤s k=1 i=1 j=1
 

s
q
q
r
+(g − 1)  m i 1 j1 k m i 2 j2 k n i 1 k n i 2 k − (mijk nik )2  .
k=1 i1 ,i2 =1 1≤j1 ≤j2 ≤r i=1 j=1
Proof: This follows immediately from (1.3), Lemma 4.5 and the definition of
S[d,n,m] (Definition 4.1).
Remark 4.7. Let 0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E be the maximal Jordan–Hölder
filtration of a bundle E. Then the kernels of the duals of the inclusions Ej → E
give us a filtration
0 = F0 ⊂ F1 ⊂ ... ⊂ Ft = E
of the dual E of E, such that if 1 ≤ j ≤ t then
Fj /Fj−1 ∼
= (Et−j+1 /Et−j ) .
Thus Fj /Fj−1 is a direct sum of stable bundles all of the same slope, and moreover
Fj−1 is the minimal subbundle of Fj such that Fj /Fj−1 is a direct sum of stable
bundles all of which have minimal slope among quotients of Fj .
Applying this construction with E replaced by E , we find that every holomor-
phic bundle E over Σ has a canonical filtration
0 = F0 ⊂ F1 ⊂ ... ⊂ Ft = E,
which we will call the minimal Jordan–Hölder filtration of E, with the property
that if 1 ≤ j ≤ t then Fj−1 is the minimal subbundle of Fj such that Fj /Fj−1 is
a direct sum of stable bundles all of which have minimal slope among quotients of
Fj .
The minimal and maximal Jordan–Hölder filtrations of a bundle do not nec-
essarily coincide. For example, consider the direct sum E ⊕ F of two semistable
bundles with maximal Jordan–Hölder filtrations 0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E and
0 = F0 ⊂ F1 ⊂ ... ⊂ Fs = F where without loss of generality we may assume that
s ≤ t. If E and F have the same slope, then it is easy to check that the maximal
Jordan–Hölder filtration of E ⊕ F is
0 = E0 ⊕ F0 ⊂ E1 ⊕ F1 ⊂ ... ⊂ Es ⊕ Fs ⊂ Es+1 ⊕ Fs ⊂ ... ⊂ Et ⊕ Fs ;
that is, it is the direct sum of the maximal Jordan–Hölder filtrations of E and F with
the shorter one extended trivially at the top. Similarly the minimal Jordan–Hölder
filtration of E ⊕ F is the direct sum of the minimal Jordan–Hölder filtrations of E
and F with the shorter one extended trivially at the bottom. Thus if the minimal
and maximal Jordan–Hölder filtrations of E and F coincide (which will be the case
if, for example, each of the subquotients Ej /Ej−1 and Fj /Fj−1 are stable) but these
filtrations are not of the same length, then the minimal Jordan–Hölder filtration
0 = E0 ⊕ F0 ⊂ E1 ⊕ F0 ⊂ ... ⊂ Et−s ⊕ F0 ⊂ Et−s+1 ⊕ F1 ⊂ ... ⊂ Et ⊕ Fs
of E ⊕ F will be different from its maximal Jordan–Hölder filtration.
Definition 4.8. Given [d, n, m] ∈ I, let S[d,n,m]
minJH
denote the subset of C con-
sisting of those holomorphic structures on our fixed smooth bundle of rank n̂ and
degree dˆ whose minimal Jordan–Hölder filtration is of the form
0 = E0,1 ⊂ E1,1 ⊂ · · · ⊂ Er1 ,1 = E0,2 ⊂ E1,2 ⊂ · · ·
· · · ⊂ Ers−1 ,s−1 = Ers ,0 ⊂ · · · ⊂ Ers ,s = E
260 FRANCES KIRWAN
with
Ej,k /Ej−1,k ∼= (Cm1jk ⊗ D1k ) ⊕ · · · ⊕ (Cmqk jk ⊗ Dqk k )
for 1 ≤ k ≤ s and 1 ≤ j ≤ qk , where D1k , . . . , Dqk k are nonisomorphic stable
bundles with
rank(Dik ) = nik and deg(Dik ) = dik
and Ej−1,k is the minimal subbundle of Ej,k such that Ej,k /Ej−1,k is a direct sum
of stable bundles of slope dk /nk .
5. More indexing sets

In this section we will consider the indexing set Γ for the stratification {Σγ :
γ ∈ Γ} of C ss defined as in §2.
If γ ∈ Γ then by (2.2) either γ = 0 or γ = Rl or γ ∈ Bl \{0} × {l} for some
1 ≤ l ≤ τ . If γ = 0 then Σγ = C s , while by [26, pp.248-9] if γ = Rl then there
exists [n, m] = [(ni )qi=1 , (mij )q,r
i=1,j=1 ] ∈ I
ss
with r = 1 and q = q1 , such that

q
(5.1) Rl = GL(mi ; C)
i=1
where mi = mi1 , and ΣRl consists of all those holomorphic structures E with
(5.2) E∼= (Cm1 ⊗ D1 ) ⊕ · · · ⊕ (Cmq ⊗ Dq )
ˆ and not isomorphic to one another.
where D1 , . . . , Dq are all stable of slope d/n̂
In order to describe the strata {Σβ,l : β ∈ Bl \{0}} more explicitly, we need to
look at the action of Rl on the normal NRl to GZR ss
l
at a point represented by a
holomorphic structure E of the form (5.2), and to understand the stratification on
P(NRl ) induced by this action of Rl . If we choose a C ∞ isomorphism of our fixed
C ∞ bundle E with (Cm1 ⊗ D1 ) ⊕ · · · ⊕ (Cmq ⊗ Dq ), then we can identify C with the
infinite-dimensional vector space
Ω0,1 (End((Cm1 ⊗ D1 ) ⊕ · · · ⊕ (Cmq ⊗ Dq )))
and the normal to the Gc -orbit at E can be identified with H 1 (Σ, End E), where
End E is the bundle of holomorphic endomorphisms of E [1, §7]. If δij denotes the
ss
Kronecker delta then the normal to GZR can be identified with

q
i2
H 1 (Σ, End⊕ E) ∼
= Cmi1 mi2 −δi1 ⊗ H 1 (Σ, Di∗1 ⊗ Di2 )
i1 ,i2 =1
where End⊕ E
is the quotient of the bundle End E of holomorphic endomorphisms
of E by the subbundle End⊕ E consisting of those
q endomorphisms which preserve
the decomposition (5.2). The action of Rl = i=1 GL(mi ; C) on this is given by
i2
the natural action on Cmi1 mi2 −δi1 identified with the set of mi1 × mi2 matrices
if i1 = i2 and the set of trace-free mi1 × mi1 matrices if i1 = i2 ; its weights α
are therefore of the form α = ξ − ξ where ξ and ξ are weights of the standard
representation of Rl on ⊕qi=1 Cmi (see [26, pp.251-2] noting the error immediately
before (3.18)).
Any element β of the indexing set Bl is represented by the closest point to 0
of the convex hull of some nonempty set of these weights, and two such closest
points can be taken to represent the same element of Bl if and only if they lie in
the same Ad(Nl )-orbit, where Nl is the normaliser of Rl (see [22] or [28]). By (3.3)
the orbit of β under the adjoint action of the connected component of Nl is just its
Ad(Rl )-orbit, and so by (3.4) the Ad(Nl )-orbit of β is the union

w.Ad(Nl )(β)
w∈π0 (Nl )
where π0 (Nl ) is the product of permutation groups

π0 (Nl ) = Sym(#{i : mi = j and ni = k}).

j≥0,k≥0
We can describe this indexing set Bl more explicitly as follows. Let us take our
maximal compact torus Tl in Rl to be the product of the standard maximal tori of
the unitary groups U (m1 ),..., U (mq ) consisting of the diagonal matrices, and let tl
be its Lie algebra. Let
M = m1 + ... + mq
and let e1 , ..., eM be the weights of the standard representation of Tl on Cm1 ⊕ ... ⊕
Cmq . We take the usual invariant inner product on the Lie algebra u(p) of U (p)
given by A, B = −tr AB̄ t and restrict it to Tl . Since Rl is embedded in GL(p; C)
by identifying ⊕qi=1 (Cmi ⊗ Cpi ) with Cp , it follows that e1 , ..., eM are mutually
orthogonal and ||ej ||2 = 1/pi if m1 + ... + mi−1 < j ≤ m1 + ... + mi .
Proposition 5.1. Let β be any nonzero element of the Lie algebra tl of the
maximal compact torus Tl of Rl . Then β represents an element of Bl \{0} if and
only if there is a partition
{∆h,m : (h, m) ∈ J}
of {1, ..., M }, indexed by a subset J of Z × Z of the form
J = {(h, m) ∈ Z × Z : 1 ≤ h ≤ L, l1 (h) ≤ m ≤ l2 (h)}
for some positive integer L and functions l1 and l2 : {1, ..., L} → Z such that
l1 (h) ≤ l2 (h) for all h ∈ {1, ..., L}, with the following properties. If

rh,m = ||ej ||−2
j∈∆h,m
then the function : {1, ..., L} → Q defined by

  −1

l2 (h)

l2 (h)
(h) =  mrh,m   rh,m 
m=l1 (h) m=l1 (h)
satisfies −1/2 ≤ (h) < 1/2 and (1) > (2) > ... > (L), and
β
L
l2 (h)
ej
= ((h) − m) .
||β||2 ||ej ||2
h=1 m=l1 (h) j∈∆h,m
Remark 5.2. Note that because of the conditions on the function , the par-
tition {∆h,m : (h, m) ∈ J} and its indexing can be recovered from the coefficients
of β with respect to the basis e1 /||e1 ||2 , ..., eM /||eM ||2 of tl .
Proof of Proposition 5.1: β ∈ tl \{0} represents an element of Bl \{0} if and
only if it is the closest point to 0 of the convex hull of
{ei − ej : (i, j) ∈ S}
262 FRANCES KIRWAN
for some nonempty subset S of {(i, j) ∈ Z × Z : 1 ≤ i, j ≤ M }. Then β can be

expressed in the form β
β= λij (ei − ej )
(i,j)∈S

for some λβij ∈ R for (i, j) ∈ S such that λβij ≥ 0 and (i,j)∈S λβij = 1. Replacing S
with its subset {(i, j) ∈ S : λβij > 0} we may assume that λβij
> 0 for all (i, j) ∈ S.
Moreover clearly if S = {ei − ej } has just one element then β = ei − ej , and we
then take J = {(1, 0), (1, 1)} with ∆(1,0) = {j} and ∆(1,1) = {i}, so we can assume
without loss of generality that λβij < 1 for all (i, j) ∈ S. Since β = 0 we can also
assume that the convex hull of {ei − ej : (i, j) ∈ S} does not contain 0.
In order to find the closest point to 0 of the convex hull of {ei − ej : (i, j) ∈ S}
we minimise
|| λij (ei − ej )||2
(i,j)∈S

subject to the constraints that λij ≥ 0 for all (i, j) ∈ S and (i,j)∈S λij = 1. Since
the weights e1 , ..., eM are mutually orthogonal, we have

M
|| λij (ei − ej )||2 = || ( λij − λji )ei ||2
(i,j)∈S i=1 j:(i,j)∈S j:(j,i)∈S

M
= ( λij − λji )2 ||ei ||2 .
i=1 j:(i,j)∈S j:(j,i)∈S
Using the method of Lagrange multipliers, we consider

M
( λij − λji )2 ||ei ||2 − λ( λij − 1).
i=1 j:(i,j)∈S j:(j,i)∈S (i,j)∈S
If (i, j) ∈ S then i = j and (j, i) ∈ S since the convex hull of {ei − ej : (i, j) ∈ S}
does not contain 0, so
∂
M
( ( λij − λij )2 ||ei ||2 − λ( λij − 1))
∂λij i=1
j:(i,j)∈S j:(j,i)∈S (i,j)∈S
is equal to

2( λik − λki )||ei ||2 − 2( λjk − λkj )||ej ||2 − λ.
k:(i,k)∈S k:(k,i)∈S k:(j,k)∈S k:(k,j)∈S
β
Thus β = (i,j)∈S λij (ei − ej ) where for each (i, j) ∈ S we have either λβij = 0 or
λβij = 1 (both of which are ruled out by the assumptions on S) or

(5.3) ( λik − λki )||ei ||2 − ( λjk − λkj )||ej ||2 = λ/2
k:(i,k)∈S k:(k,i)∈S k:(j,k)∈S k:(k,j)∈S
where λ is independent of (i, j) ∈ S.

From S we can construct a directed graph G(S) with vertices 1, ..., M and
directed edges from i to j whenever (i, j) ∈ S. Let ∆1 ,..., ∆N be the connected
components of this graph. Then {ei − ej : (i, j) ∈ S} is the disjoint union of its
subsets {ei − ej : (i, j) ∈ S, i, j ∈ ∆h } for 1 ≤ h ≤ N , and {ei − ej : (i, j) ∈
S, i, j ∈ ∆h } is contained in the vector subspace of tl spanned by the basis vectors
{ek : k ∈ ∆h }. Since these subspaces are mutually orthogonal for 1 ≤ h ≤ N , it

follows that
L −1 L
1 βh
(5.4) β=
||βh ||2 ||βh ||2
h=1 h=1
where
βh = λβh
ij (ei − ej )
(i,j)∈S,i,j∈∆h
is the closest point to 0 of the convex hull of {ei − ej : (i, j) ∈ S, i, j ∈ ∆h } for

1 ≤ h ≤ N , and without loss of generality we assume that βh is nonzero when
1 ≤ h ≤ L and zero when L < h ≤ N . Note that then
L −2 L L −1
1 ||βh ||2 1
(5.5) ||β|| =
2
=
||βh ||2 ||βh ||4 ||βh ||2
h=1 h=1 h=1
so that
β βh L
(5.6) =
||β||2 ||βh ||2
h=1
and
λβh 2 β
ij = (||βh || /||β|| )λij
2
if i, j ∈ ∆h . For 1 ≤ h ≤ L let κh be defined by

κh = max{( λβik − λβki )||ei ||2 : i ∈ ∆h }
k:(i,k)∈S k:(k,i)∈S
where a sum over the empty set is interpreted as 0. Then by (5.3) for 1 ≤ h ≤ L
we can express ∆h as a disjoint union
∆h = ∆ ˆ h,0 ∆
ˆ h,1 . . . ∆
ˆ h,l
h
where

(5.7) ˆ h,m = {i ∈ ∆h : (
∆ λβij − λβji )||ei ||2 = κh − m|λ|/2}.
j:(i,j)∈S j:(j,i)∈S
Let us assume that λ ≤ 0; the argument is similar if λ ≥ 0. Then (5.3) tells us that
if (i, j) ∈ S then there exist h and m such that i ∈ ∆ ˆ h,m and j ∈ ∆ ˆ h,m+1 . Note
that if ∆ˆ h,m and ∆ˆ h,m are nonempty then so is ∆ˆ h,m whenever m1 < m < m2 , so
1 2
without loss of generality we may assume that ∆h,m is nonempty when 1 ≤ h ≤ L

ˆ
and 0 ≤ m ≤ lh . Thus if 1 ≤ h ≤ L we have

βh = (||βh ||2 /||β||2 ) ( λβij − λβji )ei
i∈∆h j∈∆h ,(i,j)∈S j∈∆h ,(j,i)∈S

lh ei
= (||βh ||2 /||β||2 ) (κh − m|λ|/2) .
m=1 i∈∆
||ei ||2
ˆ h,m
−2
For 1 ≤ h ≤ L and 0 ≤ m ≤ lh let rh,m = j∈∆ ˆ h,m ||ej || ; then by (5.7) we have

( λβij − λβji ) = rh,m (κh − m|λ|/2}).
ˆ h,m j:(i,j)∈S
i∈∆ j:(j,i)∈S
264 FRANCES KIRWAN
Recall that if (i, j) ∈ S then i ∈ ∆

ˆ h,m if and only if j ∈ ∆ ˆ h,m+1 , so we get

λβij = rh,0 κh ,
ˆ h,0 j:(i,j)∈S
i∈∆
and hence

λβij = ( λβij − λβji ) + λβji
ˆ h,1 j:(i,j)∈S
i∈∆ ˆ h,1 j:(i,j)∈S
i∈∆ j:(j,i)∈S ˆ h,0 j:(j,i)∈S
j∈∆
= rh,1 (κh − |λ|/2}) + rh,0 κh ,

and similarly
|λ|
λβij = (rh,0 + rh,1 + ... + rh,m )κh − (rh,1 + 2rh,2 + ... + mrh,m )
2
ˆ h,m j:(i,j)∈S
i∈∆

if 1 ≤ m ≤ lh . Since (i,j)∈S,i,j∈∆h λβh
ij = 1, it follows that
||β||2 /||βh ||2 = ((lh + 1)rh,0 + lh rh,1 + ... + rh,lh )κh

−(lh rh,1 + 2(lh − 1)rh,2 + 3(lh − 2)rh,3 + ... + lh rh,lh )|λ|/2,

and since (i,j)∈S,i,j∈∆h λβij − (j,i)∈S,i,j∈∆h λβji = 0 we have
0 = (rh,0 + rh,1 + ... + rh,lh )κh − (rh,1 + 2rh,2 + ... + lh rh,lh )|λ|/2.
Thus
|λ|/2 = (||β||2 /||βh ||2 )(rh,0 + rh,1 + ... + rh,lh )/µh
and
κh = (||β||2 /||βh ||2 )(rh,1 + 2rh,2 + ... + lh rh,lh )/µh
where

lh
µh = ((lh − i + 1)j − j(lh − j + 1))rh,i rh,j
i,j=0

= ((j − i)j + (i − j)i)rh,i rh,j = (j − i)2 rh,i rh,j .
0≤i<j≤lh 0≤i<j≤lh
Therefore

lh
lh
(j − m)rh,j ei
βh =
m=0 j=0
µh ||ei ||2
ˆ h,m
i∈∆
and
||βh ||2 = (rh,0 + rh,1 + ... + rh,lh ))/µh .
By defining ∆h,m = ∆ ˆ h,m−l (h) for an appropriate integer l1 (h), we can arrange
1
that the function defined in the statement of the proposition takes values in the
interval [−1/2, 1/2), and then by amalgamating those ∆h for which (h) takes the
same value and rearranging them so that is a strictly decreasing function, we can
assume that the required conditions on are satisfied, and we have
β
L
l2 (h)
ei
= ((h) − m)
.
||β||2 ||ei ||2
h=1 m=l1 (h) i∈∆h,m

This gives us all the required properties if L = N ; that is, if (h,m)∈J ∆h,m is equal

to {1, ..., M }. Otherwise we amalgamate {1, ..., M }\ (h,m)∈J ∆h,m with ∆h0 ,m0
where (h0 , m0 ) is the unique element of J such that (h0 ) = 0 = m0 if such an
element exists, and if there is no such element of J then we adjoin (L + 1, 0) to J

and define

L
l2 (h)
∆L+1,0 = {1, ..., M }\ ∆h,m .
h=1 m=l1 (h)
Conversely, suppose that we are given any partition {∆h,m : (h, m) ∈ J} of

{1, ..., M } indexed by
J = {(h, m) ∈ Z × Z : 1 ≤ h ≤ L, l1 (h) ≤ m ≤ l2 (h)}
for some positive integer L and functions l1 and l2 : {1, ..., L} → Z with l1 ≤ l2 ,
satisfying (h) ∈ [−1/2, 1/2) and (1) > (2) > ... > (L) where
  −1

l2 (h)

l2 (h)
(h) =  mrh,m   rh,m 
m=l1 (h) m=l1 (h)
−2
for rh,m = i∈∆h,m ||ei || . Suppose also that
β = β̂/||β̂||2
(or equivalently β̂ = β/|β|2 ) where

L
l2 (h)
ej
β̂ = ((h) − m) .
||ej ||2
h=1 m=l1 (h) j∈∆h,m
It suffices to show that β is the closest point to 0 of the convex hull of {ei − ej :
(i, j) ∈ S}, where S is the set of ordered pairs (i, j) with i, j ∈ {1, ..., M } such that
ei ∈ ∆h,m and ej ∈ ∆h,m+1 for some (h, m) ∈ J such that (h, m + 1) ∈ J. For this,
it is enough to prove firstly that β lies in the convex hull of {ei − ej : (i, j) ∈ S}
and secondly that (ei − ej ).β = ||β||2 (or equivalently that (ei − ej ).β̂ = 1) for all
(i, j) ∈ S. The latter follows easily from the choice of S: if (i, j) ∈ S then there
exists (h, m) ∈ J such that (h, m + 1) ∈ J and ei ∈ ∆h,m and ej ∈ ∆h,m+1 , so
(ei − ej ).β̂ = (h) − m − (h) + m + 1 = 1.
To show that β lies in the convex hull of {ei − ej : (i, j) ∈ S}, we note that

l2 (h)−1

m
(h)rh,k − krh,k
(ei − ej )
rh,m rh,m+1 ||ei ||2 ||ej ||2
m=l1 (h) i∈∆h,m j∈∆h,m+1 k=l1 (h)

l2 (h)
ej
= ((h) − m) .
||ej ||2
m=l1 (h) j∈∆h,m
This means that

β= λβij (ei − ej )
(i,j)∈S
where
λβij
m
(h)rh,k − krh,k
(5.8) =
||β|| 2 rh,m rh,m+1 ||ei ||2 ||ej ||2
k=l1 (h)
266 FRANCES KIRWAN
if i ∈ ∆h,m and j ∈ ∆h,m+1 for some (h, m) ∈ J such that (h, m + 1) ∈ J. Then
λβij
L l2 (h)−1
m
(h)rh,k − krh,k
=
||β|| 2 rh,m rh,m+1 ||ei ||2 ||ej ||2
(i,j)∈S h=1 m=l1 (h) i∈∆h,m j∈∆h,m+1 k=l1 (h)
l2 (h)−1
L l2 (h)−1
L l2 (h)−1
= ((h) − k)rh,k = (l2 (h) − k)((h) − k)rh,k .
h=1 k=l1 (h) m=k h=1 k=l1 (h)
Expanding the brackets, using the definition of and replacing the index k by m
shows that this equals

L
l2 (h)

L
l2 (h)
(m − (h) )rh,m =
2 2
((h) − m)2 rh,m
h=1 m=l1 (h) h=1 m=l1 (h)

L l2 (h)
((h) − m)2 1
= = ,
||ei ||2 ||β||2
h=1 m=l1 (h) i∈∆h,m
β
and so (i,j)∈S λij = 1. Finally note that

m
l2 (h)

m
l2 (h)
( ((h) − k1 )rh,k1 )( rh,k2 ) = (k2 rh,k2 − k1 rh,k2 )rh,k1 .
k1 =l1 (h) k2 =l1 (h) k1 =l1 (h) k2 =l1 (h)
This sum is positive because if k2 > m then the contribution of the pair (k1 , k2 ) to
the sum is (k2 − k1 )rh,k1 rh,k2 > 0, whereas if k2 ≤ m then the total contribution
of the pairs (k1 , k2 ) and (k2 , k1 ) is zero. Thus by (5.8) we have λβij ≥ 0 for all
(i, j) ∈ S, and hence β lies in the convex hull of {ei − ej : (i, j) ∈ S} as required.
Lemma 5.3. If β satisfies the conditions of Proposition 5.1 and if i ∈ ∆h,m
and j ∈ ∆h ,m , then
β.(ei − ej ) = ||β||2

if and only if h = h and m = m + 1, and
β.(ei − ej ) ≥ ||β||2
if and only if either m ≥ m + 2 or m = m + 1 and h ≥ h.
Proof: This follows immediately from the formula for β and conditions on the
function in the statement of Proposition 5.1.
Remark 5.4. Let β be as in Proposition 5.1. Then there is a unique bijection
φ : J → {1, ..., t} from the indexing set J of the partition {∆h,m : (h, m) ∈ J} of
{1, ..., M } to the set of positive integers {1, ..., t}, where t is the size of J, which
takes the Hebrew lexicographic ordering on J to the standard ordering on integers;
that is, φ(h, m) ≤ φ(h , m ) if and only if either m < m or m = m and h ≤ h .
We can define an increasing function
δ : {1, ..., t} → {1, ..., t}
such that δ(φ(h, m)) is the number of elements (h , m ) ∈ J such that either m <
m + 1 or m = m + 1 and h < h. Then δ(k) ≥ k for all k ∈ {1, ..., t}, and if
(h, m) and (h, m + 1) both belong to J then δ(φ(h, m)) = φ(h, m + 1) − 1 and
δ(φ(h, m)) < δ(φ(h, m) + 1). Conversely if
k2 − 1 = δ(k1 ) < δ(k1 + 1)
then there exists (h, m) ∈ J with (h, m + 1) ∈ J such that k1 = φ(h, m) and
k2 = φ(h, m + 1).
When it is helpful to make the dependence on β explicit, we shall write δβ :
{1, ..., tβ } → {1, ..., tβ } and {δh,m (β) : (h, m) ∈ Jβ }.
Lemma 5.3 tells us that if i ∈ ∆h,m and j ∈ ∆h ,m then β.(ei − ej ) ≥ ||β||2 if
and only if φ(h , m ) > δ(φ(h, m)).
Definition 5.5. Recall that if 1 ≤ i ≤ q then em1 +...+mi−1 +1 , ..., em1 +...+mi
are the weights
of the standard representation on Cmi of the component GL(mi ; C)
of Rl = qi=1 GL(mi ; C). If β and φ : J → {1, ..., t} are as in Remark 5.4 and
1 ≤ i ≤ q and 1 ≤ k ≤ t, then set
∆k = ∆φ−1 (k) , ∆ki = ∆φ−1 (k) ∩ {m1 + ... + mi−1 + 1, ..., m1 + ... + mi }
and let mki denote the size of ∆ki , so that m1i + ... + mti = mi .
Remark 5.6. By Remark 5.2 the partition {∆k (β) : 1 ≤ k ≤ tβ } of {1, ..., M }
and the function δβ : {1, ..., tβ } → {1, ..., tβ } are determined by β. Conversely, from
the partition {∆k (β) : 1 ≤ k ≤ tβ } of {1, ..., M } and the function δβ : {1, ..., tβ } →
{1, ..., tβ } we can recover β as the closest point to 0 of the convex hull of
{ei − ej : i ∈ ∆k1 (β) and j ∈ ∆k2 (β) where k2 > δβ (k1 )}.
Note, however, that although given any partition {∆k : 1 ≤ k ≤ t} of {1, ..., M }
and increasing function δ : {1, ..., t} → {1, ..., t} satisfying δ(k) ≥ k for 1 ≤ k ≤ t,
we can consider the closest point β to 0 of the convex hull of
{ei − ej : i ∈ ∆k1 and j ∈ ∆k2 where k2 > δ(k1 )},
it is not necessarily the case that the associated partition {∆k (β) : 1 ≤ k ≤ tβ }
of {1, ..., M } and function δβ : {1, ..., tβ } → {1, ..., tβ } coincide with the given
partition {∆k : 1 ≤ k ≤ t} of {1, ..., M } and function δ : {1, ..., t} → {1, ..., t}. For
example, some amalgamation and rearrangement may be needed as in the proof of
Proposition 9.1.
6. Balanced δ-filtrations
The last section studied the indexing set Γ for the stratification {Σγ : γ ∈ Γ} of
C ss defined as in §2. In this section we will consider what it means for a semistable
holomorphic bundle over the Riemann surface Σ to belong to a stratum Σγ = Σβ,l ,
where β is as in Proposition 5.1.
Definition 6.1. We shall say that a semistable bundle E has a δ-filtration
0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E
with associated function δ : {1, ..., t} → {1, ..., t} if δ is an increasing function such
that if 1 ≤ k ≤ t then δ(k) ≥ k and the induced filtration
Ek Ek+1 Eδ(k)
0⊂ ⊂ ⊂ ... ⊂
Ek−1 Ek−1 Ek−1
is trivial.
Let G(δ) be the graph with vertices 1, ..., t and edges joining i to j if j − 1 =
δ(i) < δ(i + 1). Then the connected components of G(δ) are of the form
{ihl1 (h) , ..., ihl2 (h) }
268 FRANCES KIRWAN
for 1 ≤ h ≤ L, where ihj − 1 = δ(ihj−1 ) < δ(ihj−1 + 1) if 1 < j ≤ sh , and ih1 − 1

is not in the image of δ and either δ(ihsh ) = u or δ(ihsh ) = δ(ihsh + 1). Moreover
l1 (h) ≤ l2 (h) can be chosen so that if
  −1

l2 (h)

l2 (h)
(h) =  mr̃h,m   r̃h,m  ,
m=l1 (h) m=l1 (h)
where r̃h,m = rank(Eihm /Eihm −1 ), then −1/2 ≤ (h) < 1/2, and the ordering of the
components of G(δ) can be chosen so that
(6.1) (1) ≥ (2) ≥ ... ≥ (L).
We shall say that the δ-filtration is balanced if the inequalities in (6.1) are all strict
and if
(6.2) ihm11 ≤ ihm22 if and only if m1 < m2 or m1 = m2 and h1 ≤ h2 ;
that is, if the usual ordering on {1, ..., t} is the same as the Hebrew lexicographic
ordering via the pairs (h, m).
Remark 6.2. If β is as in Proposition 5.1 then the proof of that proposition
shows that

1
L
l1 (h)≤i<j≤l2 (h) (j − i) rh,i rh,j
2
L
l2 (h)
(6.3) = = (m−(h))2 rh,m
||β||2
l1 (h)≤i≤l2 (h) r h,i
h=1 h=1 m=l1 (h)
where

q
−2
rh,m = ||ej || = ||ej ||−2
j∈∆h,m i=1 φ(h,m)
j∈∆i

q
q
φ(h,m)

q
φ(h,m)
= pi = mi pi = (1 − g + d/n) mi ni .
i=1 j∈∆φ(h,m) i=1 i=1
i
Thus

1 L
l1 (h)≤i<j≤l2 (h) (j − i)2 r̃h,i r̃h,j
(6.4) = (1 − g + d/n)
||β||2
l1 (h)≤i≤l2 (h) r̃h,i
h=1

L
l2 (h)
= (1 − g + d/n) (m − (h))2 r̃h,m
h=1 m=l1 (h)
where

q
φ(h,m)
r̃h,m = mi ni
i=1
and (h) is given by
  −1   −1

l2 (h)

l2 (h)

l2 (h)

l2 (h)
 mrh,m   rh,m  =  mr̃h,m   r̃h,m  .
m=l1 (h) m=l1 (h) m=l1 (h) m=l1 (h)
Note that if 0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E is a filtration such that

q
Ek /Ek−1 ∼
k
= Cmi ⊗ Di
i=1
if 1 ≤ k ≤ t, where t and mki for 1 ≤ i ≤ q and 1 ≤ k ≤ t are as in Definition 5.5

and D1 , ..., Dq are nonisomorphic stable bundles of ranks n1 , ..., nq and all of the
same slope d/n, then
r̃h,m = rank(Eφ(h,m) /Eφ(h,m)−1 ).
Proposition 6.3. Let β be as in Proposition 5.1, let δ be as in Remark 5.4
and let E be a semistable holomorphic structure on E.
(i) If E represents an element of the stratum Σβ,l then E has a unique balanced
δ-filtration 0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E such that

q
Ek /Ek−1 ∼
k
= Cmi ⊗ Di
i=1
and hence

q
Eδ(k) /Ek−1 ∼
k δ(k)
= Cmi +...+mi ⊗ Di
i=1
if 1 ≤ k ≤ t, where t and mki for 1 ≤ i ≤ q and 1 ≤ k ≤ t are as in Definition 5.5
and D1 , ..., Dq are nonisomorphic stable bundles of ranks n1 , ..., nq and all of the
same slope d/n.
(ii) Conversely, if E has a balanced δ-filtration 0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E
as in (i) then E represents an element of the stratum Σβ,l if and only if E has no
filtration with the corresponding properties for any β satisfying ||β || > ||β||.
Proof: Recall from (2.7) that
Σβ,l = Gc Yβ,l ∼
\E \E
(6.5) = Gc ×Qβ ,l Yβ,l
\E
and that if E represents an element of Yβ,l then its orbit under the complex one-
s
parameter subgroup of Rl generated by β has a limit point in ZR l
. This limit point
is represented by the bundle gr(E) which is of the form

q
gr(E) ∼
= Cmi ⊗ Di
i=1
where D1 , ..., Dq are nonisomorphic stable bundles of ranks n1 , ..., nq and all of the
same slope d/n. Recall also from [1, § 7] that C is an infinite dimensional affine
space, and if we fix a C ∞ identification of the fixed C ∞ Hermitian bundle E with
q
i=1 C ⊗ Di then we can identify C with the infinite dimensional vector space
mi

q
Ω0,1 (End( Cmi ⊗ Di ))
i=1
q
qof Ω mi(End( i=1 C ⊗Di )) corresponds to the
0,1 mi
in such a way that the zero element
given holomorphic structure
q on i=1 C ⊗ Di . With respect to this identification,
the action of R l = i=1 GL(m i ; C) on C is the action induced by the obvious
q
action of Rl on i=1 Cmi ⊗ Di . The one-parameter subgroup of Rl generated by
β acts diagonally on Cm1 ⊕ ... ⊕ Cmq with weights β.ej for j ∈ {1, ..., M } where
M = m1 + ... + mq , and so it acts on

q
q
Ω0,1 (End( Cmi ⊗ Di )) = Ω0,1 (Cmi1 ⊗ (Cmi2 )∗ ⊗ Di1 ⊗ Di∗2 )
i=1 i1 ,i2 =1
270 FRANCES KIRWAN
with weights β.(ei − ej ) for i, j ∈ {1, ..., M }. If E ∈ Σβ then E lies in the Gc -orbit
of an element of the sum of those weight spaces for which the weight β.(ei − ej )
satisfies β.(ei − ej ) ≥ ||β||2 . By Remark 5.4 and Definition 5.5 we have a partition
{∆1 , ..., ∆t } of {1, ..., M } such that β.(ei − ej ) ≥ ||β||2 if and only if i ∈ ∆k1 and
j ∈ ∆k2 where k2 > δ(k1 ). So if we make identifications

q
q
t
k
CM = Cmi = Cmi
i=1 i=1 k=1
using the induced partition {∆ki

: 1 ≤ i ≤ q, 1 ≤ k ≤ t} of {1, ..., M } as in Definition
5.5, then any E ∈ Σβ lies in the Gc -orbit of an element of

q
t
t
k1 k2
Ω0,1 (Cmi1 ⊗ (Cmi2 )∗ ⊗ Di1 ⊗ Di∗2 ).
i1 ,i2 =1 k1 =1 k2 =δ(k1 )+1

This completes the proof of (i), as such an element of Ω0,1 (End( qi=1 Cmi ⊗ Di ))
represents a holomorphic structure E on E with a filtration of the required form
(uniqueness follows from (6.5) and the fact that Qβ,l preserves the filtration), and
(ii) is now a consequence of (2.1).
Corollary 6.4. Let β be as in Proposition 5.1, let δ be as in Remark 5.4 and
let E be a semistable holomorphic structure on E with a balanced δ-filtration
(6.6) 0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E
whose subquotients satisfy

q
Ek /Ek−1 ∼
k
= Cmi ⊗ Di ,
i=1
where t and for 1 ≤ i ≤ q and 1 ≤ k ≤ t are as in Definition 5.5 and D1 , ..., Dq

mki
are nonisomorphic stable bundles of ranks n1 , ..., nq and all of the same slope d/n.
Then E represents an element of the stratum Σβ,l if and only if, in the notation of
Definition 6.1, there is no h ∈ {1, ..., L} and refinement
0 = E0 ⊂ ... ⊂ Eih −1 ⊂ Fl1 (h) ⊂ Eih ⊂ ... ⊂ Eihm −1 ⊂ Fm ⊂ Eihm ⊂ ...
l1 (h) l1 (h)
... ⊂ Eih −1 ⊂ Fl2 (h) ⊂ Eih ⊂ ... ⊂ Et = E

l2 (h) l2 (h)
of (6.6) with
l2 (h) l2 (h)
m=l1 (h) m rank(Fm /Eihm −1
) 1 m=l (h) m rank(Eihm /Eihm −1 )
l2 (h) < l (h)
2
= (h),
m=l1 (h) rank(Fm /Eih
m −1
) m=l1 (h) rank(Eihm /Eihm −1 )
such that the induced filtrations
Eihm −1 Fm
0⊂ ⊆
Eih1 −1 Eih1 −1
m1 m1
with l2 (h)

m=l (h) mrank(Fm /Eihm −1 )
m1 − (h1 ) ≤ l (h)
1
2
m=l1 (h) rank(Fm /Eihm −1 )
and
Eihm E ih 2
0⊂ ⊆
m2
Fm Fm
with
l2 (h)
m=l (h) m rank(Fm /Eihm −1 )
m2 − (h2 ) ≥ m − l (h)
1
2
m=l1 (h) rank(Fm /Eihm −1 )
are all trivial.
Remark 6.5. If 0 = E0 ⊂ E1 ⊂ ... ⊂ Ej−1 ⊂ F ⊂ Ej ⊂ ... ⊂ Et = E is a
refinement of the filtration (6.6) of E such that the induced filtration
Ej−1 F
0⊆ ⊂
Ei Ei
is trivial for some i < j − 1, then F/Ei is isomorphic to
Ej−1 F
⊕
Ei Ej−1
and so E can be given a filtration of the form

0 = E0 ⊂ E1 ⊂ ... ⊂ Ei ⊂ Ei+1 ⊂ Ei+2 ⊂ ... ⊂ Ej−1 ⊂ F ⊂ Ej ⊂ ... ⊂ Et = E

where Ei+1 /Ei ∼
= F/Ej−1 , Ek+1
/Ek ∼
= Ek /Ek−1 for i < k < j and F/Ej−1 ∼
=
Ej−1 /Ej−2 . A similar result is true if the induced filtration
F Ei
0⊂ ⊂
Ej−1 Ej−1
is trivial for some i > j.
Proof of Corollary 6.4: This follows from [28] and the proof of Proposition
5.1, which tells us that if β = β is the closest point to 0 of the convex hull
of {ei − ej : (i, j) ∈ S } where S is a subset of S, then S can be chosen so
that the connected components of the graph G(S ) give us a refinement of the
partition {∆h,m : (h, m) ∈ J} of {1, ..., M } associated to β, which in turn gives us
a refinement of the filtration (6.6) with the required properties.
Remark 6.6. Recall from Proposition 6.3 that a semistable bundle E repre-
sents an element of the stratum Σβ,l if and only if if has a balanced δ-filtration
(6.7) 0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E
k
such that if 1 ≤ k ≤ t then Ek /Ek−1 is of the form qi=1 Cmi ⊗Di where D1 , ..., Dq
are nonisomorphic stable bundles of ranks n1 , ..., nq and all of the same slope d/n,
and moreover E has no balanced δ-filtration with the corresponding properties
when β is replaced with β satisfying ||β || > ||β||. From (6.4) we have

1
L
l1 (h)≤i<j≤l2 (h) (j − i) r̃h,i r̃h,j
2
(6.8) = (1 − g + d/n) ,
||β||2 l1 (h)≤i≤l2 (h) r̃h,i
h=1
where
r̃h,m = rank(Eφ(h,m) /Eφ(h,m)−1 ).
This gives us some sort of measure of the triviality of the balanced δ-filtration (6.7);
very roughly speaking, the more trivial this filtration, the smaller the size of ||β||−2
and hence the larger ||β|| becomes.
272 FRANCES KIRWAN
Let us therefore define the triviality of the balanced δ-filtration (6.7) with as-
sociated function δ to be
L −1/2
l1 (h)≤i<j≤l2 (h) (j − i) r̃h,i r̃h,j
2

h=1 l1 (h)≤i≤l2 (h) r̃h,i
 −1/2

L
l2 (h)
(6.9) = (m − (h))2 r̃h,m  ;
h=1 m=l1 (h)
Remarks 5.2 and 5.6 tell us that this is well defined. Thus if E ∈ Σβ,l the balanced
δ-filtration (6.7) associated to E by Proposition 10.1 can be thought of as having
maximal triviality (according to this measure) among the balanced δ-filtrations of
E.
Remark 6.7. Let β be as in Proposition 5.1, let δ be as in Remark 5.4 and let
E be a semistable holomorphic structure on E with a δ-filtration
0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E
q
such that if 1 ≤ k ≤ t then Ek /Ek−1 ∼ = i=1 C
mk
i ⊗ D
i where D1 , ..., Dq are
nonisomorphic stable bundles of ranks n1 , ..., nq and all of the same slope d/n.
Then the proof of Proposition 6.3 and (2.3) shows that E represents an element of
Σγ̃ for some γ̃ ∈ Γ satisfying γ̃ ≥ γ = (β, l) with respect to the partial ordering on
Γ defined just before (2.3).
Remark
r 6.8. If [n, m] ∈ I ss then we can define β[n, m] ∈ Γ as follows. Let
mi = j=1 mij and suppose that Rl is the subgroup of GL(p; C) given by the

embedding of qi=1 GL(mi ; C) via a fixed identification of qi=1 Cni (1−g+d/n) ⊗Cmi
with Cp . If q = 1 let
β[n, m] = Rl ,
and if q > 1 let β[n, m] ∈ Bl \{0} be represented by the closest point to 0 of the
convex hull of the weights of the representation of the maximal compact torus Tl
of Rl on
q
Cmi1 j1 ⊗ (Cmi2 j2 )∗
i1 ,i2 =1 1≤j1 <j2 ≤r
r
given by identifying j=1 Cmij with Cmi for 1 ≤ i ≤ q. Then we have from Remark
6.7 that

(6.10) S[n,m] ⊆ Σγ ,
γ≥β[n,m]
and from Remark 4.4 and Proposition 6.3 that

(6.11) Σβ[n,m] ⊆ S[n ,m ]
[n ,m ]≥[n,m]
where ≥ denotes in (6.10) the partial order on Γ used in Remark 6.7, whereas in
(6.11) it denotes the partial order on I ss described in Remark 4.4.
Remark 6.9. It follows from Proposition 6.3 that if Rl is as at (5.1) then a

holomorphic structure belongs to

Σβ,l
β∈Bl \{0}
if and only if E ∼
= gr(E) ∼
= (Cm1 ⊗ D1 ) ⊕ · · · ⊕ (Cmq ⊗ Dq ) where D1 , . . . , Dq are
all stable of slope d/n and ranks n1 , . . . , nq and are not isomorphic to one another.
7. Pivotal filtrations
Let us now consider the relationship between the balanced δ-filtration associ-
ated to a semistable bundle E as in Proposition 6.3 and the maximal and minimal
Jordan–Hölder filtrations defined in §8.
Indeed, motivated by Proposition 6.3, we can try to carry our analysis of the
maximal Jordan–Hölder filtration
(7.1) 0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E
of a bundle E a bit further. Recall that if 1 ≤ j ≤ t then the subquotient Ej /Ej−1 is
the maximal subbundle of E/Ej−1 which is a direct sum of stable bundles all having
maximal slope among the nonzero subbundles of E/Ej−1 . We can ask whether it
is true for every subbundle F of E satisfying Ej−1 ⊂ F ⊂ Ej and slope(Ej /F ) =
slope(Ej /Ej−1 ) that Ej /F is the maximal subbundle of E/F which is a direct sum
of stable bundles all having maximal slope among the nonzero subbundles of E/F .
Of course if Ej /Ej−1 is itself stable there are no such intermediate subbundles F ,
so this is trivially true, but it is not always the case (as Example 7.1 below shows).
If there does exist such an intermediate subbundle F , then by Lemma 3.2 both
F/Ej−1 and Ej /F are of the form
(7.2) (Cm1 ⊗ D1 ) ⊕ · · · ⊕ (Cmq ⊗ Dq )
where D1 , . . . , Dq are all stable of slope d/n and ranks n1 , . . . , nq and are not
isomorphic to one another. So we can then ask whether it is possible to find a
canonical refinement
(7.3) 0 = F0 ⊂ F1 ⊂ ... ⊂ Fu = E
of the maximal Jordan–Hölder filtration (7.1) of E with an increasing function
δ : {1, ..., u} → {1, ..., u} such that δ(j) ≥ j if 1 ≤ j ≤ u, each subquotient
Fδ(j) /Fj−1 is of the form (7.2) and moreover for every subbundle F of E satisfying
Fj−1 ⊆ F ⊂ Fj and slope(Fj /F ) = slope(F/Fj−1 ), the quotient Fδ(j) /F is the
maximal subbundle of E/F which is a direct sum of stable bundles all having
maximal slope among the nonzero subbundles of E/F . The following example
shows that even this cannot be achieved in a canonical way.
Example 7.1. Let E1 and E2 be semistable bundles over Σ such that slope(E1 )
equals slope(E2 ), with maximal Jordan–Hölder filtrations
0 ⊂ D1 ⊂ E1 and 0 ⊂ D2 ⊂ E2
where D1 , D2 , E1 /D1 and E2 /D2 are nonisomorphic stable bundles all of the same
slope as E1 and E2 , and let
E = E1 ⊕ E2 .
We observed in Remark 4.7 that the maximal Jordan–Hölder filtration of a direct
sum of semistable bundles of the same slope is the direct sum of their maximal
274 FRANCES KIRWAN
Jordan–Hölder filtrations (with the shorter one extended trivially at the top if they
are not of the same length). Thus the maximal Jordan–Hölder filtration of E is
(7.4) 0 ⊂ D1 ⊕ D2 ⊂ E1 ⊕ E2 = E.
By Lemma 3.2 there are precisely two proper subbundles F of D1 ⊕ D2 with
slope(F ) = slope(D1 ⊕ D2 /F ) = slope(D1 ⊕ D2 ), namely D1 and D2 . The maximal
Jordan–Hölder filtration of E/D1 = (E1 /D1 ) ⊕ E2 is
0 ⊂ (E1 /D1 ) ⊕ D2 ⊂ (E1 /D1 ) ⊕ E2 ,
so we can refine the filtration (7.5) of E to get
(7.5) 0 = F0 ⊂ F1 ⊂ F2 ⊂ F3 ⊂ F4 = E
where F1 = D1 , F2 = D1 ⊕D2 and F3 = E1 ⊕D2 , and we can define δ : {1, 2, 3, 4} →
{1, 2, 3, 4} by δ(1) = 2, δ(2) = 3 and δ(3) = δ(4) = 4. If 1 ≤ j ≤ 4 then Fδ(j) /Fj−1
is the maximal subbundle of E/Fj−1 which is a direct sum of stable subbundles
all having maximal slope among the nonzero subbundles of E/Fj−1 . Moreover
this is trivially still true if we replace Fj−1 by any subbundle F of E satisfying
Fj−1 ⊆ F ⊂ Fj and slope(Fj /F ) = slope(Fj /Fj−1 ), since the only such F is Fj−1
itself. We can of course reverse the rôles of E1 and E2 in this construction, to get
another refinement
(7.6) 0 ⊂ D2 ⊂ D1 ⊕ D2 ⊂ D1 ⊕ E2 ⊂ E1 ⊕ E2 = E
of (7.4). Thus there are precisely two refinements of the maximal Jordan–Hölder
filtration of E1 ⊕ E2 with the required propertes, and if E1 has the same rank as
E2 and D1 has the same rank as D2 then by symmetry there can be no canonical
choice.
Notice that if rank(D1 )/rank(E1 ) = rank(D2 )/rank(E2 ) then neither of the δ-
filtrations (7.5) and (7.6) is balanced since the inequalities (6.1) are not strict; how-
ever the δ-filtration (7.4) is balanced and has maximal triviality, in the sense of (6.9),
among balanced δ-filtrations of E1 ⊕E2 . If on the other hand rank(D1 )/rank(E1 ) =
rank(D2 )/rank(E2 ) then precisely one of the δ-filtrations (7.5) and (7.6) is balanced
and it has maximal triviality, in the sense of (6.9), among balanced δ-filtrations of
E1 ⊕ E2 . This filtration then determines the stratum Σγ to which E belongs, and
in this case E represents an element of the open subset Σsβ,l of Σγ .
Lemma 7.2. Let E be a bundle over Σ with a filtration
0 = F0 ⊂ F1 ⊂ ... ⊂ Fu = E
and let {∆1 , ..., ∆L } be a partition of {1, ..., u} such that if 1 ≤ h ≤ L and ∆h =
{ih1 , ..., ihsh } where ih1 < ih2 < ... < ihsh , then the induced extension
(7.7) 0 → Fihj /Fihj −1 → Fihj+1 −1 /Fihj −1 → Fihj+1 −1 /Fihj → 0
is trivial. Then we can associate to this filtration of E, partition {∆1 , ..., ∆L } of
{1, ..., u} and trivialisations of the induced extensions (7.7) a sequence of elements
of
H 1 (Σ, (Fihj /Fihj −1 ) ⊗ (Fihj+1 /Fihj+1 −1 )∗ )
or equivalently of extensions
0 → Fihj /Fihj −1 → Ejh → Fihj+1 /Fihj+1 −1 → 0
for 1 ≤ h ≤ L and 1 ≤ j ≤ sh − 1.
Proof: This lemma follows immediately from the well known bijective correspon-
dence between holomorphic extensions of a holomorphic bundle D1 over Σ by an-
other holomorphic bundle D2 and elements of H 1 (Σ, D1 ⊗ D2 ). The extension
0 → Fihj+1 −1 /Fihj −1 → Fihj+1 /Fihj −1 → Fihj+1 /Fihj+1 −1 → 0
induced by the given filtration gives us an element of
H 1 (Σ, (Fihj+1 −1 /Fihj −1 ) ⊗ (Fihj+1 /Fihj+1 −1 )∗ )
and the given trivialisation of the extension (7.7) gives us a decomposition of this
as
H 1 (Σ, (Fihj /Fihj −1 ) ⊗ (Fihj+1 /Fihj+1 −1 )∗ ) ⊕ H 1 (Σ, (Fihj+1 −1 /Fihj ) ⊗ (Fihj+1 /Fihj+1 −1 )∗ ).
Projection onto the first summand gives us an extension
0 → Fihj /Fihj −1 → Ejh → Fihj+1 /Fihj+1 −1 → 0
as required.
Remark 7.3. Lemma 5.3 and Remark 5.4 tell us that if i ∈ ∆ki11 and j ∈ ∆ki22
where k2 > δ(k1 ) then (ei − ej ).β ≥ ||β||2 , and equality occurs if and only if there
exists (h, m) ∈ J with (h, m + 1) ∈ J such that k1 = φ(h, m) and k2 = φ(h, m + 1).
Thus, retaining the notation of the proof of Proposition 6.3, we observe that if E
is represented by an element of

q
t
t
k1 k2
Ω0,1 (Cmi1 ⊗ (Cmi2 )∗ ⊗ Di1 ⊗ Di∗2 )
i1 ,i2 =1 k1 =1 k2 =δ(k1 )+1
then the limit in C as t → ∞ of exp(−itβ)E is the bundle

q
gr(E) ∼
= Cmi ⊗ Di
i=1
which is represented by the zero vector in

q
t
t
k1 k2
Ω0,1 (Cmi1 ⊗ (Cmi2 )∗ ⊗ Di1 ⊗ Di∗2 ),
i1 ,i2 =1 k1 =1 k2 =δ(k1 )+1
and so the limit pβ (E) of exp(−itβ)E in the blow-up of C along Gc ZRss

l
is an element
of the fibre
q
i2
P(Nl,grE ) = P(H 1 (Σ, Cmi1 mi2 −δi1 ⊗ Di1 ⊗ Di∗2 )
i1 ,i2 =1
of the exceptional divisor over gr E. Indeed pβ (E) is the element of this fibre
represented by the sum in

L l2
(h)−1

q
φ(h,m)
q
φ(h,m+1)
1
H (Σ, ( C mi
1 ⊗ Di1 ) ⊗ ( Cmi2 ⊗ Di2 )∗ )
h=1 m=l1 (h) i1 i2
of the elements of

q
φ(h,m)
q
φ(h,m+1)
1
H (Σ, ( C mi
1 ⊗ Di1 ) ⊗ ( Cmi2 ⊗ Di2 )∗ )
i1 i2
276 FRANCES KIRWAN
for 1 ≤ h ≤ L and l1 (h) ≤ m ≤ l2 (h) corresponding to the extensions

q
φ(h,m)
mi

q
φ(h,m+1)
0→ C 1 ⊗ Di1 → h
Em → Cmi2 ⊗ Di2 → 0
i1 i2
associated as in Lemma 7.2 to the δ-filtration 0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E of

Proposition 6.3.
Proposition 7.4. Let β be as in Proposition 5.1 and let E be a semistable
bundle representing an element of Σβ,l with a balanced δ-filtration
0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E
q
mk
i ⊗ D
i as in Proposition 6.3.
Suppose also that k1 ∈ {1, ..., t} is such that
β.ej < 0 whenever j ∈ ∆k2 with k2 > δ(k1 ).
Then whenever F is a subbundle of E with slope(F ) = slope(E) and such that
Ek1 −1 ⊆ F ⊂ Ek1 , the subquotient Eδ(k1 ) /F is the maximal subbundle of E/F
which is a direct sum of stable bundles all having the same slope as E/F .
q
Proof: Since F/Ek1 −1 is a subbundle of Ek1 /Ek1 −1 ∼
k1
= i=1 Cmi ⊗ Di having the
same slope as D1 , ..., Dq , it follows from Lemma 3.2 that

q
F/Ek1 −1 = Ui ⊗ Di
i=1
k1
where Ui is a linear subspace of Cmi for 1 ≤ i ≤ q, and so

q
Eδ(k1 ) /F ∼
k1 k1 +1 δ(k1 )
= ((Cmi /Ui ) ⊕ Cmi +...+mi
) ⊗ Di
i=1
is a sum of stable bundles all having the same slope (which is equal to slope(E) and
slope(E/F )). Let us suppose for a contradiction that E/F has a subbundle E /F
which is not contained in Eδ(k1 ) /F and which is of the required form. Then we can
choose k2 > δ(k1 ) such that E ⊆ Ek2 but E is not contained in Ek2 −1 , and then
the inclusion of E in Ek2 induces a nonzero map

q
θ : E /F → Ek2 /Ek2 −1 ∼
k2
= Cmi ⊗ Di .
i=1
Since nonzero bundle maps between stable bundles of the same slope are always
isomorphisms, by replacing E by a suitable subbundle we can assume that E /F ∼
=
k2 k2
Di0 for some i0 ∈ {1, ..., q}, and that we can decompose Cmi0 as C⊕Cmi0 −1 in such
a way that the projection θ0 : E /F → Di0 of θ onto the corresponding component
Di0 of Ek2 /Ek2 −1 is an isomorphism. Then
θ0−1 : Di0 → E /F ⊆ Ek2 /F
gives us a trivialisation of the extension of Ek2 −1 /Ek1 by this component Di0 of
Ek2 /Ek2 −1 . By the definition of Σβ,l the limit pβ (E) ∈ P(Nl,grE ) of exp(−itβ)E
as t → ∞ is semistable for the induced action of Stab(β)/Tβc where Stab(β) is the
stabiliser of β under the coadjoint action of Rl and Tβc is the complex subtorus
generated by β (see [28]), and by Remark 7.3 pβ (E) is represented by the sum over
h ∈ {1, ..., L} and m ∈ {l1 (h), ..., l2 (h)} of the elements of

q
φ(h,m)
q
φ(h,m+1)
H 1 (Σ, ( Cmi1 ⊗ Di1 ) ⊗ ( Cmi2 ⊗ Di2 )∗ )
i1 i2
corresponding to the extensions

q
mi
φ(h,m)
q
φ(h,m+1)
0→ C 1 ⊗ Di1 → Em →
h
Cmi2 ⊗ Di2 → 0
i1 i2
associated by Lemma 7.2 to the filtration 0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E.

Let S0 be the set of ordered pairs (i, j) with i, j ∈ {1, ..., M } such that the
corresponding to the weight ei − ej for the
component of pβ (E) in the weight space
action of the maximal torus Tl of Rl = qi=1 GL(mi ; C) on

L l2
(h)−1

q
φ(h,m)
q
φ(h,m+1)
H 1 (Σ, ( Cmi1 ⊗ Di1 ) ⊗ ( Cmi2 ⊗ Di2 )∗ )
h=1 m=l1 (h) i1 i2
is nonzero. Since pβ (E) is semistable for the action of Stab(β)/Tβc , it follows that β
is the closest point to 0 in the convex hull of {ei − ej : (i, j) ∈ S0 }. We may assume
k2 k2
that Tl acts diagonally with respect to the decomposition of Cmi0 as C ⊕ Cmi0 −1 ;
k2
let ej0 be the weight of the action of Tl on the component C of Cmi0 with respect
to this decomposition. Since k2 > δ(k1 ) and the extension of Ek2 −1 /Ek1 by the
q
component of Ek2 /Ek2 −1 ∼
k
i=1 C ⊗ Di corresponding to the weight ej0 is

mi 2
=
trivial, it follows that if (i, j) ∈ S0 then j = j0 . Since β lies in the convex hull of
{ei − ej : (i, j) ∈ S0 } and e1 , ..., eM are mutually orthogonal, this means that
β.ej0 ≥ 0,
and as j0 ∈ ∆k2 and k2 > δ(k1 ), this gives us the required contradiction.
Remark 7.5. Dual to the definition of δ in Remark 5.4, we can define an
increasing function δ : {1, ..., t} → {1, ..., t} such that δ (φ(h, m)) − 1 is the number
of elements (h , m ) ∈ J such that either m < m − 1 or m = m − 1 and h ≤ h.
Then δ (k) ≤ k for all k ∈ {1, ..., t}, and if (h, m) and (h, m − 1) both belong to
J then δ (φ(h, m)) = φ(h, m − 1) + 1. Also k1 < δ (k2 ) if and only if k2 > δ(k1 ),
and Lemma 5.3 tells us that if i ∈ ∆k1 and j ∈ ∆k2 then β.(ei − ej ) ≥ ||β||2 if and
only if k1 < δ (k2 ). The dual version of Proposition 6.3 tells us that if 1 ≤ k ≤ t
then Ek /Eδ (k)−1 is a direct sum of stable bundles all of the same slope, and using
Remark 4.7 we obtain the following dual version of Proposition 7.4.
bundle representing an element of Σβ,l with δ-filtration
0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E
q
mk
i ⊗ D
Suppose also that k1 ∈ {1, ..., t} is such that
β.ej > 0 whenever j ∈ ∆k2 with k2 < δ (k1 ).
Then whenever F is a subbundle of E with slope(F ) = slope(E) and such that
Ek1 −1 ⊂ F ⊆ Ek1 , the subbundle Eδ (k1 )−1 is the minimal subbundle of F such that
F/Eδ (k1 )−1 is a direct sum of stable bundles all with the same slope as F .
278 FRANCES KIRWAN
Remark 7.7. It follows from the definition of ∆1 , ..., ∆t (Definition 5.5) that
if j1 ∈ ∆k1 and j2 ∈ ∆k2 then β.ej1 < β.ej2 if and only if k1 > k2 , so we can
choose k− and k+ such that β.ej < 0 (respectively β.ej > 0) if and only if j ∈ ∆k
with k > k− (respectively k < k+ ). Then k− = k+ or k− = k+ − 1, depending on
whether there exists j with β.ej = 0. Propositions 7.4 and 7.6 tell us that if E is a
semistable bundle representing an element of Σβ with filtration
0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E
as in Proposition 6.3, then
0 ⊂ Eδ(j)+1 /Ej−1 ⊂ Eδ(δ(j)+1)+1 /Ej−1 ⊂ ... ⊂ E/Ej−1
is the maximal Jordan–Hölder filtration of E/Ej−1 if j ≥ δ (k− ), and
0 ⊂ ... ⊂ Eδ (δ (j)−1)−1 ⊂ Eδ (j)−1 ⊂ Ej
is the minimal Jordan–Hölder filtration of Ej if j ≤ δ(k+ ). Note also that δ (k− ) ≤
k− ≤ k+ ≤ δ(k+ ), so there are values of j satisfying both j ≥ δ (k− ) and j ≤ δ(k+ ).
There is a converse to Propositions 7.4 and 7.6.
bundle with δ-filtration
0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E
q
mki ⊗ D
Suppose that every subbundle F of E with the same slope as E satisfies the following
two properties:
(i) if Ek1 −1 ⊆ F ⊂ Ek1 for some k1 ∈ {1, ..., t} such that β.ej < 0 whenever
j ∈ ∆k2 with k2 > δ(k1 ), then the subquotient Eδ(k1 )/F is the maximal subbundle
of E/F which is a direct sum of stable bundles all with the same slope as E/F ;
(ii) if Ek1 −1 ⊂ F ⊆ Ek1 for some k1 ∈ {1, ..., t} such that β.ej > 0 whenever
j ∈ ∆k2 with k2 < δ (k1 ), then the subbundle Eδ (k1 )−1 is the minimal subbundle of
F such that F/Eδ (k1 )−1 is a direct sum of stable bundles all with the same slope as
F.
Then E represents an element of the stratum Σβ,l .
Proof: Suppose for a contradiction that E does not represent an element of Σβ,l .
k
Then (cf. [28]) after applying a change of coordinates to Cmi for 1 ≤ i ≤ q and
1 ≤ k ≤ t, we can assume that β is not equal to the closest point to 0 in the convex
hull of {ei −ej : (i, j) ∈ S0 } where S0 is as in the proof of Proposition 8.4. Moreover
S0 ⊆ {(i, j) : β.(ei − ej ) ≥ ||β||2 }.
Thus β does not lie in the convex hull of
{ei − ej : (i, j) ∈ S0 ∩ S}
where S = {(i, j) : β.(ei − ej ) = ||β||2 }. ¿From Lemma 5.3 and Remark 5.4 we know
that if i ∈ ∆h,m and j ∈ ∆h ,m then β.(ei − ej ) ≥ ||β||2 if and only if m ≥ m + 2
or m = m + 1 and h ≥ h, and this happens if and only if φ(h , m ) > δ(φ(h, m)),
while β.(ei − ej ) = ||β||2 if and only if m = m + 1 and h = h. By Remark 7.7
the hypothesis (i) on subbundles F of E tells us that if Ek1 −1 ⊆ F ⊂ Ek1 where
k1 ≥ δ(k− ), then the subquotient Eδ(k1 )/F is the maximal subbundle of E/F which
is a direct sum of stable bundles all with the same slope as E/F . This implies that
if (h, m) and (h, m + 1) both lie in J and φ(h, m) ≥ k− then every pair (i, j) with
i ∈ ∆h,m and j ∈ ∆h,m+1 lies in S0 . Similarly the hypothesis (ii) tells us that if
(h, m) and (h, m − 1) both lie in J and φ(h, m) ≤ k+ then every pair (i, j) with
i ∈ ∆h,m and j ∈ ∆h,m−1 lies in S0 . Since k− ≤ k+ this means that
S0 ∩ S = S.
This contradicts the fact that β is the closest point to 0 in the convex hull of
{ei − ej : (i, j) ∈ S} but does not lie in the convex hull of {ei − ej : (i, j) ∈ S0 ∩ S},
and thus completes the proof.
Remark 7.9. Suppose that β corresponds to a partition {∆h,m : (h, m) ∈ J}
of {1, ..., M }, indexed by
J = {(h, m) ∈ Z × Z : 1 ≤ h ≤ L, l1 (h) ≤ m ≤ l2 (h)}
where l1 and l2 : {1, ..., L} → Z satisfy l1 (h) ≤ l2 (h) for all h ∈ {1, ..., L}, as in
Proposition 5.1. Let E be a semistable bundle representing an element of Σβ with
δ-filtration
0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E
as in Proposition 6.3. If
l1 (h1 ) − 1 (h1 ) < l1 (h2 ) − (h2 ) + 1
for all h1 , h2 ∈ {1, .., L}, or equivalently if
φ(h, l1 (h)) ≤ δ(1) + 1
for all h ∈ {1, ..., L}, then the proof of Proposition 7.8 shows that
(7.8) 0 ⊂ Eδ(1)+1 ⊂ Eδ(δ(1)+1)+1 ⊂ ... ⊂ E
is the maximal Jordan–Hölder filtration of E. Thus for such a β the stratum
Σβ,l is contained in the subset S[n maxJH
β ,mβ ]
of C ss defined at Definition 4.1, where nβ
and mβ are determined by the filtration (7.8). If, on the other hand, there exists
h0 ∈ {1, ..., L} with
φ(h0 , l1 (h0 )) > δ(1) + 1,
then Eδ(1)+1 may not be the maximal subbundle of E which is a direct sum
of stable bundles all with the same slope as E; there may be a subbundle of
Eφ(h0 ,l1 (h0 )) /Eφ(h0 ,l1 (h0 )−1) which provides a trivial extension of Eδ(1)+1 by a direct
sum of stable bundles all with the same slope as E (see Example 8.1 below). How-
ever, even in this case a careful analysis of the proof of Proposition 4.5 reveals that
it can be modified to show that the intersection Σβ ∩ S[n,m] maxJH
is a locally closed
complex submanifold of C of finite codimension for each β ∈ Γ and [n, m] ∈ I ss .
ss
Definition 7.10. We shall call a filtration 0 ⊂ P1 ⊂ ... ⊂ Pτ ⊂ E of a

semistable bundle E a pivot if each subbundle Pj has the same slope as E and P1
is the minimal subbundle of Pτ such that Pτ /P1 is a direct sum of stable bundles
of the same slope, while Pτ /P1 is the maximal subbundle of E/P1 which is a direct
sum of stable bundles of the same slope. Any pivot determines a filtration
0 ⊆ ... ⊆ Pτ−2 = P1−1 ⊆ ... ⊆ Pτ−1 −1
−1 ⊆ Pτ = P10 = P1 ⊆ ...
... ⊆ Pτ = Pτ0 = P1+1 ⊆ P2+1 ⊆ ... ⊆ Pτ+1 = P1+2 ⊆ ... ⊆ E
of E where
0 ⊂ Pj+1 /Pj ⊂ Pj+2 /Pj ⊂ ... ⊂ E/Pj
280 FRANCES KIRWAN
is the maximal Jordan–Hölder filtration of E/Pj , and

0 ⊂ ... ⊂ Pj−2 ⊂ Pj−1 ⊂ Pj
is the minimal Jordan–Hölder filtration of Pj . A filtration 0 = E0 ⊂ E1 ⊂ ... ⊂
Et = E of this form (once repetitions have been omitted) for some pivot 0 ⊂ P1 ⊂
... ⊂ Pτ ⊂ E will be called a pivotal filtration. It will be called a strongly pivotal
filtration if every subbundle F of E such that
m
Pj−1 ⊆ F ⊂ Pjm
for some m ≥ 0 has
0 ⊂ Pjm /F ⊂ Pjm+1 /F ⊂ ... ⊂ E/F
as the maximal Jordan–Hölder filtration of E/F , and every subbundle F of E such
that
Pjm ⊂ F ⊆ Pj+1
m
for some m ≤ 0 has

0 ⊂ ... ⊂ Pjm−1 ⊂ Pjm ⊂ F
as its minimal Jordan–Hölder filtration.
Note that a pivotal filtration is a δ-filtration where δ(k1 ) is the number of
k2 ∈ {1, ..., t} for which it is not the case that Ek1 = Pjm
1
1
and Ek2 = Pjm2
2
with
m1 ≥ m2 or m1 = m2 − 1 and j1 ≥ j2 ; if the associated δ-filtration is balanced then
we will call the pivotal filtration balanced.
Theorem 7.11. Let β be as in Proposition 5.1 and let E be a semistable bundle
representing an element of Σβ,l with δ-filtration
(7.9) 0 = E0 ⊂ E1 ⊂ ... ⊂ Et = E
q
such that if 1 ≤ k ≤ t then Ek /Ek−1 ∼
k
= i=1 Cmi ⊗ Di as in Proposition 6.3. Then
(7.9) is a balanced strongly pivotal δ-filtration with pivot
0 ⊂ Ek1 −1 ⊂ Ek1 ⊂ ... ⊂ Eδ(k1 ) ⊂ E
for any k1 satisfying δ (k− ) ≤ k1 ≤ k+ .
Proof: This is an immediate consequence of Propositions 7.4 and 7.6 as in Remark
7.7.
8. Refinements of the Yang-Mills stratification

We thus have three refinements of the Harder-Narasimhan filtration of a holo-
morphic bundle E over Σ: the maximal Jordan–Hölder filtration, the minimal
Jordan–Hölder filtration and the balanced δ-filtration of maximal triviality ob-
tained by applying Proposition 6.3 to the subquotients of the Harder-Narasimhan
filtration. Associated to these we have refinements of the Yang-Mills stratification
of C, each of which is a stratification of C by locally closed complex submanifolds
of finite codimension, and has the set C s of stable holomorphic structures on E as
its open stratum. The first of these refined stratifications is the stratification
(8.1) {S[d,n,m]
maxJH
: [d, n, m] ∈ I}
of C defined in Definition 8.1, and another is the stratification
(8.2) {S[d,n,m]
minJH
: [d, n, m] ∈ I}
defined dually using minimal Jordan–Hölder filtrations as in Remark 4.7 and Defi-
nition 4.8. A third refinement is the stratification obtained by applying the strat-
ification {Σγ : γ ∈ Γ} of C ss , whose indexing set was determined in §5 and whose
strata were described in terms of balanced δ-filtrations in §6, to the C(n , d )ss which
appear inductively in the description of the Yang–Mills stratification.
Example 8.1. Recall from Remark 4.7 that the maximal Jordan–Hölder fil-
tration of the direct sum E ⊕ F of two semistable bundles of the same slope is the
direct sum of the maximal Jordan–Hölder filtrations of E and F with the shorter
one extended trivially at the top, while the minimal Jordan–Hölder filtration of
E ⊕ F is the direct sum of their minimal Jordan–Hölder filtrations with the shorter
one extended trivially at the bottom. Suppose now that E and F have balanced
δ-filtrations of maximal triviality given by
0 ⊂ El1 (1) ⊂ El1 (1)+1 ⊂ ... ⊂ El2 (1) = E
and
0 ⊂ Fl1 (2) ⊂ Fl1 (2)+1 ⊂ ... ⊂ Fl2 (2) = F
where the indices l1 (1), ..., l2 (1) ∈ Z and l1 (2), ..., l2 (2) ∈ Z have been chosen so
that

l2 (1)
(1) = m rank(Em /Em−1 )
m=l1 (1)
and

l2 (2)
(2) = m rank(Fm /Fm−1 )
m=l1 (2)
lie in the interval [−1/2, 1/2). To simplify the notation let us assume that l1 (1) ≤
l1 (2) ≤ l2 (2) ≤ l2 (1). If (1) > (2) then E ⊕ F has a balanced δ-filtration given by
0 ⊂ El1 (1) ⊕ 0 ⊂ ... ⊂ El1 (2) ⊕ 0 ⊂ El1 (2) ⊕ Fl1 (2) ⊂ ...
... ⊂ Em−1 ⊕ Fm−1 ⊂ Em ⊕ Fm−1 ⊂ Em ⊕ Fm ⊂ ...
... ⊂ El2 (2) ⊕ Fl2 (2) ⊂ El2 (2)+1 ⊕ Fl2 (2) ⊂ ... ⊂ El2 (1) ⊕ Fl2 (2) = E ⊕ F.
If we assume that Ei /Ei−1 and Fj /Fj−1 are stable for l1 (1) ≤ i ≤ l2 (1) and
l1 (2) ≤ j ≤ l2 (2), then this filtration has no proper refinements with subquotients
of the same slope as E ⊕ F , so by Corollary 6.4 it is a balanced δ-filtration of E ⊕ F
with maximal triviality (and in fact it is not hard to check that this is still true
without the simplifying assumption). If (2) > (1) then we replace the filtration
above with the balanced δ-filtration
0 ⊂ El1 (1) ⊕ 0 ⊂ ... ⊂ El1 (2)−1 ⊕ 0 ⊂ El1 (2)−1 ⊕ Fl1 (2) ⊂ ...
... ⊂ Em−1 ⊕ Fm−1 ⊂ Em−1 ⊕ Fm ⊂ Em ⊕ Fm ⊂ ...
... ⊂ El2 (2) ⊕ Fl2 (2) ⊂ El2 (2)+1 ⊕ Fl2 (2) ⊂ ... ⊂ El2 (1) ⊕ Fl2 (2) = E ⊕ F.
Thus we see that the the maximal Jordan–Hölder filtration, the minimal Jordan–
Hölder filtration and the balanced δ-filtration of maximal triviality of a bundle
E can all be different from one another, and that none of them is necessarily a
refinement of the other two. Nonetheless, the concepts of maximal Jordan–Hölder
filtration, minimal Jordan–Hölder filtration and balanced δ-filtration of maximal
triviality on a bundle E are related by Theorem 7.11 via the notion of a pivotal
filtration (see also Remark 6.8, Propositions 7.4, 7.6 and 7.8 and Remark 7.9).
282 FRANCES KIRWAN
References
1. M.F. Atiyah and R. Bott, The Yang-Mills equations over Riemann surfaces Philos. Trans.
Roy. Soc. London Ser. A 308 (1982), 523-615.
2. V. Baranovsky, Cohomology ring of the moduli space of stable vector bundles with odd de-
terminant, Izv. Russ. Acad. Nauk. 58 n4 (1994), 204-210.
3. A. Beauville, Sur la cohomologie de certains espaces de modules de fibrés vectoriels, Geom-
etry and Physics, (Bombay 1992), Tata Inst. Fund. Res. (1995), 37-40.
4. S. Cappell, R. Lee and E. Miller, The action of the Torelli group on the homology of repre-
sentation spaces is non-trivial, Topology 39 (2000), 851-871.
5. G. Daskalopoulos, The topology of the space of stable bundles on a Riemann surface, J.
Differential Geometry 36 (1992), 699-746.
6. S. del Baño, On the Chow motive of some moduli spaces, J. Reine Angew. Math. 532 (2001),
105-132.
7. U.V. Desale and S. Ramanan, Poincaré polynomials of the variety of stable bundles, Math.
Ann. 216 (1975), 233-244.
8. S.K. Donaldson, Gluing techniques in the cohomology of moduli spaces, in Topological meth-
ods in modern mathematics (Proceedings of 1991 Stony Brook conference in honour of the
sixtieth birthday of J. Milnor), Publish or Perish (1993), 137-170.
9. R. Earl, The Mumford relations and the moduli of rank three stable bundles, Compositio
Math. 109 (1997), 13-48.
10. R. Earl and F. Kirwan, The Pontryagin rings of moduli spaces of arbitrary rank holomorphic
bundles over a Riemann surface, J. London Math. Soc. (2) 60 (1999), 835-846.
11. , Complete sets of relations in the cohomology rings of moduli spaces of arbitrary
rank holomorphic bundles over a Riemann surface, 2003 preprint.
12. G. Ellingsrud and S. Strømme, On the Chow ring of a geometric quotient, Ann. Math. 130
(1989), 130-159.
13. W. Fulton, Intersection Theory (Second Edition) Springer (1998).
14. D. Gieseker, A Degeneration of the Moduli Spaces of Stable Bundles, J. Differential Geom.
19 (1984), 173-206.
15. , Geometric Invariant Theory and applications to moduli problems, Proc. Int. Cong.
Math. (Helsinki, 1978) Academiae Scientarium Fennica (Helsinki, 1980).
16. G. Harder and M.S. Narasimhan, On the cohomology groups of moduli spaces of vector
bundles over curves, Math. Ann. 212 (1975) 215-248.
17. R. Herrera and S. Salamon, Intersection numbers on moduli spaces and symmetries of a
Verlinde formula, Comm. Math. Phys. 188 (1997), 521-534.
18. L. Jeffrey, Y-H. Kiem, F. Kirwan and J. Woolf, Cohomology pairings on singular quotients
in geometric invariant theory, Transformation Groups 8 217-259.
19. , Intersection pairings on singular moduli spaces of bundles over a Riemann surface,
in preparation.
20. L. Jeffrey and F. Kirwan, Intersection theory on moduli spaces of holomorphic bundles of
arbitrary rank on a Riemann surface, Ann. of Math. 148 (1998), 109-196.
21. A. King and P.E. Newstead, On the cohomology of the moduli space of rank 2 vector bundles
on a curve, Topology 37 (1998), 407-418.
22. F. Kirwan, Cohomology of quotients in symplectic and algebraic geometry, Math. Notes vol.
31 Princeton Univ. Press, Princeton, NJ 1985.
23. , Partial desingularisations of quotients of non-singular varieties and their Betti
numbers, Ann. Math. 122 (1985), 41-85.
24. , On spaces of maps from Riemann surfaces to Grassmannians and applications to
the cohomology of moduli of vector bundles Ark. Math. 24 (1986), 221-275.
25. , Rational intersection cohomology of quotient varieties, Invent. Math. 86 (1986),
471-505.
26. , On the homology of compactifications of moduli spaces of vector bundles over a
Riemann surface, Proc. London Math. Soc. 53 (1986), 237-266.
27. , Cohomology rings of moduli spaces of bundles over Riemann surfaces, J. Amer.
Math. Soc. 5 (1992), 853-906.
28. , Refinements of the Morse stratification of the normsquare of the moment map, 2003
preprint.
29. J. Le Potier, Lectures on Vector Bundles, Cambridge Studies in Advanced Mathematics 54

CUP (1997).
30. D. Mumford, J. Fogarty and F. Kirwan, Geometric Invariant Theory, Third Edition, Ser.
Modern Surveys Math. 34, Springer (1994).
31. M.S. Narasimhan and S. Ramanan, Deformations of the moduli space of vector bundles over
an algebraic curve, Ann. Math. 101 (1975), 391-417.
32. M.S. Narasimhan and C.S. Seshadri, Stable and unitary vector bundles over an algebraic
curve, Ann. Math. 82 (1965), 540-567.
33. P.E. Newstead, Characteristic classes of stable bundles of rank 2 over an algebraic curve,
Trans. Amer. Math. Soc. 169 (1972), 337-345.
34. , Introduction to moduli problems and orbit spaces, Tata Inst. Lect. 51 (1978).
35. C.S. Seshadri, Space of unitary vector bundles on a compact Riemann surface, Ann. Math.
85 (1967), 303-336.
36. S.S. Shatz, The decomposition and specialization of algebraic families of vector bundles,
Compositio Math. 35 (1977), 163-187.
37. B. Siebert and G. Tian, Recursive relations for the cohomology ring of moduli spaces of
stable bundles, Tr. J. of Math. 19 (1995), 131-144.
38. M. Thaddeus, Conformal field theory and the cohomology of the moduli space of stable
bundles, J. Differential Geom. 35 (1992), 131-149.
39. , An introduction to the topology of the moduli space of stable bundles on a Riemann
surface, in Geometry and Physics Lecture Notes in Pure and Applied Mathematics 184
Dekker (1997), 71-100.
40. E. Witten, Two dimensional gauge theories revisited, J. Geom. Phys. 9 (1992), 303-368.
41. D. Zagier, On the cohomology of moduli spaces of rank 2 vector bundles over curves, Progress
in Mathematics 129 The Moduli space of Curves (1995), 533-563.
42. , Elementary aspects of the Verlinde formula and of the Harder–Narasimhan–Atiyah–
Bott formula, Proceedings of the Hirzebruch 65 Conference on Algebraic Geometry (Ramat
Gan, 1993), Israel Math. Conf. Proc. 9 (1996), 445-462.
Mathematical Institute, Oxford University, Oxford OX1 3LB, UK

E-mail address: kirwan@maths.ox.ac.uk
Volume 3, 2004
On a classical correspondence between K3 surfaces II
Carlo Madonna and Viacheslav V. Nikulin
To memory of Andrei Nikolaevich Tyurin
Abstract. Let X be a K3 surface and H a primitive polarization of degree

H 2 = 2a2 , a > 1. The moduli space of sheaves over X with the isotropic Mukai
vector (a, H, a) is again a K3 surface Y which is endowed by a natural nef
element h with h2 = 2. We give necessary and sufficient conditions in terms
of Picard lattices N (X) and N (Y ) when Y ∼= X, generalising our results in [4]
for a = 2.
In particular, we show that Y ∼ = X if for at least one α = ±1 there
exists h1 ∈ N (X) such that h21 = 2αa, H · h1 ≡ 0 mod a, and the primitive
sublattice [H, h1 ]pr ⊂ N (X) contains x such that x · H = 1.
We also show that all divisorial conditions on moduli of (X, H) (i.e. for
Picard number 2) which imply Y ∼ = X and H · N (X) = are labelled by
pairs (d, ±µ) where d ∈ , ±µ ⊂ ( /2a2 )∗ such that d ≡ µ2 mod 4a2 and at
least for one of α = ±1 the equation p2 − dq 2 = 4a/α has an integral solution
(p, q) with p ≡ µq mod 2a. For each such ±µ and α, the number of d and the
corresponding divisorial conditions is infinite. Some of these conditions were
found (in different form) by A.N. Tyurin in 1987.
Introduction
Let X be a K3 surface with a primitive polarization H of degree H 2 = 2a2 . Let
Y be the moduli of sheaves over X with the isotropic Mukai vector v = (a, H, a)
(see [6], [7]). Then Y is a K3 surface which is endowed with a natural nef element
h with h2 = 2. It is isogenous to X in the sense of Mukai.
Question 1. When is Y isomorphic to X?
We want to answer this question in terms of the Picard lattices N (X) and
N (Y ) of X and Y . Then our question reads as follows:
Question 2. Assume that N is a hyperbolic lattice, H ∈ N a primitive element
2
with square 2a . What are conditions on N and H such that for any K3 surface X
with Picard lattice N (X) and a primitive polarization H ∈ N (X) of degree 2a2 the
corresponding K3 surface Y is isomorphic to X if the pairs of lattices (N (X), H)
and (N, H) are isomorphic as abstract lattices with fixed elements?
1991 Mathematics Subject Classification. 14J28.

Supported by Russian Fund of Fundamental Research (grant N 00-01-00170).
2004
285
286 CARLO MADONNA AND VIACHESLAV V. NIKULIN
In other words, what are conditions on (N (X), H) as an abstract lattice with

a primitive vector H with H 2 = 2a2 which are sufficient for Y to be isomorphic to
X and are necessary if X is a general K3 surface with Picard lattice N (X)?
In [4] we answered this question when a = 2. Here we give an answer for any
a ≥ 2. For odd a, we additionally assume that H · N (X) = Z. (For even a, this is
valid if Y ∼= X.) The answer is given in Theorem 2.2 and also Proposition 2.1.
In particular, if the Picard number ρ(X) = rk N (X) ≥ 12, the result is very
simple: Y ∼ = X if and only if either there exists x ∈ N (X) such that x · H = 1 or a
is odd and there exists x ∈ N (X) such that x · H = 2. This follows from results of
Mukai [7] and also [9], [10].
The polarized K3 surfaces (X, H) with ρ(X) = 2 are especially interesting.
Indeed, it is well-known that the moduli space of polarized K3 surfaces of any even
degree H 2 is 19-dimensional. If X is general, i.e. ρ(X) = 1, then the surface Y
cannot be isomorphic to X because N (X) = ZH where H 2 = 2a2 , a > 1, and
N (X) does not have elements with square 2, which is necessary if Y ∼ = X. Thus, if
Y ∼= X, then ρ(X) ≥ 2 and X belongs to a codimension ρ(X) − 1 submoduli space
of K3 surfaces. When ρ = 2 to describe connected components of the divisor it is
equivalent to describe Picard lattices N (X) of the surfaces X with fixed H ∈ N (X)
such that ρ(X) = rk N (X) = 2 and a general K3 surface X with Picard lattice
N (X) has Y ∼ = X.
The pair (N (X), H) with ρ = rk N (X) = 2 and H · N (X) = Z is defined up
to isomorphism by d = − det N (X) > 0 (it defines the Picard lattice N (X) up to
isomorphism, if Y ∼ = X) and by the invariant ±µ = {µ, −µ} ⊂ (Z/2a2 )∗ (this is
the invariant of the primitive vector H ∈ N (X)) such that µ2 ≡ d mod 4a2 . See
Proposition 3.1 concerning the definition of µ. We show that for a general X with
such N (X) we have Y ∼ = X and for odd a additionally H · N (X) = Z, if and only
if at least for one α = ±1 there exists integral (p, q) such that
(1) p2 − dq 2 = 4a/α and p ≡ µq mod 2a.
For each such ±µ and α the set Dαµof d having such a solution (p, q) and d ≡ µ2
2
mod 4a is infinite since it contains the infinite subset
(2) {(µ + 2ta/α)2 − 4a/α > 0 | tµ ≡ 1 mod a}
(set q = 1 in (1)). Thus, the set of possible divisorial conditions on moduli of
(X, H) which imply Y ∼ = X and H · N (X) = Z is labelled by the set of pairs
(d, ±µ) described above, and it is infinite.
Some infinite series of divisorial conditions on moduli of X which imply Y ∼ =X
were found by A.N. Tyurin in [17] — [19]. He found, in different form, infinite
series (2) for α = −1 and any ±µ.
Surprisingly, solutions (p, q) of (1) can be interpreted as elements of the Picard
lattices of X and Y . We get the following simple sufficient condition on (X, H)
which implies Y ∼ = X. It seems that many known examples when it happens that
Y ∼= X (e.g. see [2], [8], [17]) follow from this condition. This is one of the main
results of the paper, and we want to formulate it exactly (a similar statement can
also be formulated in terms of Y ).
Theorem 0.1. Let X be a K3 surface with a primitive polarization H of degree
2a2 , a ≥ 2. Let Y be the moduli space of sheaves on X with the Mukai vector
v = (a, H, a).
ON A CLASSICAL CORRESPONDENCE BETWEEN K3 SURFACES II 287
Then Y ∼
= X if at least for one α = ±1 there exists h1 ∈ N (X) such that
(3) (h1 )2 = 2αa, h1 · H ≡ 0 mod a,
and the primitive sublattice [H, h1 ]pr ⊂ N (X) generated by H, h1 contains x such
that x · H = 1.
These conditions are necessary to have Y ∼ = X and H · N (X) = Z for a odd, if
either ρ(X) = 1, or ρ(X) = 2 and X is a general K3 surface with its Picard lattice.
From our point of view, this statement is also very interesting because some
elements h1 of the Picard lattice N (X) with negative square (h1 )2 receive a very
clear geometrical meaning (when α < 0). For K3 surfaces this is well-known only
for elements δ of the Picard lattice N (X) with negative square δ 2 = −2: then δ or
−δ is effective.
As for the case a = 2, the fundamental tool to get the results above is the
Global Torelli Theorem for K3 surfaces [12] and results of Mukai [6], [7]. Using the
results of Mukai, we can calculate periods of Y using periods of X; by the Global
Torelli Theorem [12], we can find out if Y is isomorphic to X.
1. Preliminary notations and results about lattices and K3 surfaces

1.1. Some notations about lattices. We use notations and terminology
from [10] about lattices, their discriminant groups and forms. A lattice L is a non-
degenerate integral symmetric bilinear form, i.e. L is a free Z-module equipped with
a symmetric pairing x·y ∈ Z for x, y ∈ L, and this pairing should be non-degenerate.
We denote x2 = x · x. The signature of L is the signature of the corresponding real
form L ⊗ R. The lattice L is called even if x2 is even for any x ∈ L. Otherwise, L
is called odd. The determinant of L is defined to be det L = det(ei · ej ) where {ei }
is some basis of L. The lattice L is unimodular if det L = ±1. The dual lattice of
L is L∗ = Hom(L, Z) ⊂ L ⊗ Q. The discriminant group of L is AL = L∗ /L. It
has order | det L|. The group AL is equipped with the discriminant bilinear form
bL : AL × AL → Q/Z and the discriminant quadratic form qL : AL → Q/2Z if L is
even. To get these forms, one should extend the form of L to a form on the dual
lattice L∗ with values in Q.
For x ∈ L, we shall consider the invariant γ(x) ≥ 0, where
(4) x · L = γ(x)Z.
Clearly, γ(x)|x if x = 0.
2
We denote by L(k) the lattice obtained from a lattice L by multiplication of

the form of L by k ∈ Q. The orthogonal sum of lattices L1 and L2 is denoted by
L1 ⊕ L2 . For a symmetric integral matrix A, we denote by
A a lattice which is
given by the matrix A in some basis. We denote

0 1
(5) U= .
1 0
Any even unimodular lattice of signature (1, 1) is isomorphic to U .
An embedding L1 ⊂ L2 of lattices is called primitive if L2 /L1 has no torsion.
We denote by O(L), O(bL ) and O(qL ) the automorphism groups of the correspond-
ing forms. Any δ ∈ L with δ 2 = −2 defines a reflection sδ ∈ O(L) which is given
by the formula
x → x + (x · δ)δ,
x ∈ L. All such reflections generate the 2-reflection group W (−2) (L) ⊂ O(L).
1.2. Some notations about K3 surfaces. Here we recall some basic notions
and results about K3 surfaces, e.g. see [12], [13], [14]. A K3 surface S is a non-
singular projective algebraic surface over C such that its canonical class KS is zero
and the irregularity qS = 0. We denote by N (S) the Picard lattice of S, which is
a hyperbolic lattice with the intersection pairing x · y for x, y ∈ N (S). Since the
canonical class KS = 0, the space H 2,0 (S) of 2-dimensional holomorphic differential
forms on S has dimension one over C, and
(6) N (S) = {x ∈ H 2 (S, Z) | x · H 2,0 (S) = 0}
where H 2 (S, Z) with the intersection pairing is a 22-dimensional even unimodular
lattice of signature (3, 19). The orthogonal lattice T (S) to N (S) in H 2 (S, Z) is
called the transcendental lattice of S. We have H 2,0 (S) ⊂ T (S) ⊗ C. The pair
(T (S), H 2,0 (S)) is called the transcendental periods of S. The Picard number of S
is ρ(S) = rk N (S). A non-zero element x ∈ N (S) ⊗ R is called nef if x = 0 and
x · C ≥ 0 for any effective curve C ⊂ S. It is known that an element x ∈ N (S) is
ample if x2 > 0, x is nef , and the orthogonal complement x⊥ to x in N (S) has no
elements with square −2. For any element x ∈ N (S) with x2 ≥ 0, there exists a
reflection w ∈ W (−2) (N (S)) such that the element ±w(x) is nef; it then is ample
if x2 > 0 and x⊥ has no elements with square −2 in N (S).
We denote by V + (S) the light cone of S, which is the half-cone of
(7) V (S) = {x ∈ N (S) ⊗ R | x2 > 0 }
containing a polarization of S. In particular, all nef elements x of S belong to
V + (S): one has x · V + (S) > 0 for them.
The reflection group W (−2) (N (S)) acts in V + (S) discretely, and its fundamen-
tal chamber is the closure K(S) of the Kähler cone K(S) of S. It is the same as the
set of all nef elements of S. Its faces are orthogonal to the set Exc(S) of all excep-
tional curves r on S, which are non-singular rational curves r on S with r 2 = −2.
Thus, we have
(8) K(S) = {0 = x ∈ V + (S) | x · Exc(S) ≥ 0 }.
2. General results on the Mukai correspondence between K3 surfaces

with primitive polarizations of degrees 2a2 and 2 which gives
isomorphic K3’s
2.1. The correspondence. Let X be a K3 surface with a primitive polar-
ization H of degree 2a2 , a > 0. Let Y be the moduli space of (coherent) sheaves
E on X with the isotropic Mukai vector v = (a, H, a). This means that rk E = a,
χ(E) = 2a and c1 (E) = H. Let
(9) H ∗ (X, Z) = H 0 (X, Z) ⊕ H 2 (X, Z) ⊕ H 4 (X, Z)
be the full cohomology lattice of X with the Mukai product
(10) (u, v) = −(u0 · v2 + u2 · v0 ) + u1 · v1
for u0 , v0 ∈ H 0 (X, Z), u1 , v1 ∈ H 2 (X, Z), u2 , v2 ∈ H 4 (X, Z). We naturally identify
H 0 (X, Z) and H 4 (X, Z) with Z. Then the Mukai product is
(11) (u, v) = −(u0 v2 + u2 v0 ) + u1 · v1 .
The element
(12) v = (a, H, a) = (a, H, χ − a) ∈ H ∗ (X, Z)
is isotropic, i.e. v 2 = 0. In this case, Mukai showed [6], [7] that Y is a K3 surface,
and one has the natural identification
(13) H 2 (Y, Z) = (v ⊥ /Zv)
which also gives the isomorphism of the Hodge structures of X and Y . The element
h = (−1, 0, 1) ∈ v ⊥ has square h2 = 2, h mod Zv belongs to the Picard lattice
N (Y ) of Y and is nef . See [5], [13] and [15] concerning the geometry of (Y, h).
For a general X, the K3 surface Y is a double plane.
We want to answer Question 2 which we precisely formulated in the Introduc-
tion: Using N (X), say when Y ∼ = X.
2.2. Formulation of general results. We use notations from Sect. 2.1.

Thus, we assume that X is a K3 surface with a primitive polarization H with
H 2 = 2a2 where a > 1. The following statement follows from results of Mukai [7]
and some results from [10]. It is standard and well-known.
Proposition 2.1. If Y is isomorphic to X, then either γ(H) = 1 or a is odd
and γ(H) = 2 for H ∈ N (X) (see (4)).
Assume that either γ(H) = 1 or a is odd and γ(H) = 2 for H ∈ N (X).
Then the Mukai identification (13) canonically identifies the transcendental periods
(T (X), H 2,0 (X)) and (T (Y ), H 2,0 (Y )). It follows that the Picard lattices N (Y )
and N (X) have the same genus. In particular, N (Y ) is isomorphic to N (X) if
the genus of N (X) contains only one class. If the genus of N (X) contains only
one class, then Y is isomorphic to X, if additionally the canonical homomorphism
O(N (X)) → O(qN (X) ) is epimorphic. Both these conditions are valid (in particular,
Y ∼= X), if ρ(X) ≥ 12.
From now on we assume that γ(H) = 1 in N (X), which is automatically valid
for even a if Y ∼
= X.
The calculations below are valid for an arbitrary K3 surface X and a primitive
⊥
vector H ∈ N (X) with H 2 = 2a2 , a > 0 and γ(H) = 1. Let K(H) = HN (X) be
the orthogonal complement to H in N (X). Set H ∗ = H/2a2 . Then any element
x ∈ N (X) can be written as
(14) x = nH ∗ + k∗
where n ∈ Z and k∗ ∈ K(H)∗ , because
ZH ⊕ K(H) ⊂ N (X) ⊂ N (X)∗ ⊂ ZH ∗ ⊕ K(H)∗ .
Since γ(H) = 1, the map nH ∗ + [H] → k∗ + K(H) gives an isomorphism of the
groups Z/2a2 ∼
= [H ∗ ]/[H] ∼
= [u∗ + K(H)]/K(H) where u∗ + K(H) has order 2a2 in
∗
AK(H) = K(H) /K(H). It follows that
(15) N (X) = [ZH, K(H), H ∗ + u∗ ].
The element u∗ is defined canonically mod K(H). Since H ∗ + u∗ belongs to the
even lattice N (X), it follows that
1
(16) (H ∗ + u∗ )2 = + u∗ 2 ≡ 0 mod 2.
2a2
Let H ∗ = H ∗ mod [H] ∈ [H ∗ ]/[H] ∼

= Z/2a2 and k∗ = k∗ mod K(H) ∈ AK(H) =
∗
K(H) /K(H). We then have
(17) N (X)/[H, K(H)] = (Z/2a2 )(H ∗ + u∗ ) ⊂ (Z/2a2 )H ∗ + K(H)∗ /K(H).
Also N (X)∗ ⊂ ZH ∗ +K(H)∗ since H +K(H) ⊂ N (X), and for n ∈ Z, k∗ ∈ K(H)∗
we have x = nH ∗ + k∗ ∈ N (X)∗ if and only if
n
(nH ∗ + k∗ ) · (H ∗ + u∗ ) = 2 + k∗ · u∗ ∈ Z.
2a
It follows that
N (X)∗ = {nH ∗ + k∗ | n ∈ Z, k∗ ∈ K(H)∗ , n ≡ −2a2 (k∗ · u∗ ) mod 2a2 }
(18) ⊂ ZH ∗ + K(H)∗ ,
and
N (X)∗ /[H, K(H)] = {−2a2 (k∗ · u∗ ) H ∗ + k∗ } | k∗ ∈ AK(H) }
(19)
⊂ (Z/2a2 )H ∗ + AK(H) .
We introduce the characteristic map of the polarization H
(20) κ(H) : K(H)∗ → AK(H) /(Z/2a2 )(u∗ + K(H)) → AN (X)
where for k∗ ∈ K(H)∗ we have
(21) κ(H)(k∗ ) = −2a2 (k∗ · u∗ )H ∗ + k∗ + N (X) ∈ AN (X) .
It is epimorphic, its kernel is (Z/2a2 )(u∗ + K(H)), and it gives the canonical iso-
morphism
(22) κ(H) : AK(H) /(Z/2a2 )(u∗ + K(H)) ∼ = AN (X) .
For the corresponding discriminant forms we have
(23) κ(k∗ )2 mod 2 = (k∗ )2 + 2a2 (k∗ · u∗ )2 mod 2.
Now we can formulate our main result:
Theorem 2.2. The surface Y is isomorphic to X if the following conditions
(a), (b), (c) are valid:
(a) γ(H) = 1 for H ∈ N (X);
(b) there exists h ∈ N (X) with h2 = 2, γ(
h) = 1 and such that there exists an
embedding
f : K(H) → K( h)
of negative definite lattices such that
h) = [f (K(H)), 2af (u∗ )], w∗ + K(
K( h) = af (u∗ ) + K(
h);
(c) the dual to f embedding f ∗ : K(
h)∗ → K(H)∗ commutes (up to multiplica-
tion by ±1) with the characteristic maps κ(H) and κ(
h), i.e.
(24) κ(
h)(k∗ ) = ±κ(H)(f ∗ (k∗ ))
for any k∗ ∈ K( h)∗ .
The conditions (a), (b) and (c) are necessary if for odd a additionally γ(H) = 1
for H ∈ N (X), and rk N (X) ≤ 19, and X is a general K3 surface with Picard lattice
N (X) in the following sense: the automorphism group of the transcendental periods
(T (X), H 2,0 (X)) is ±1. (Recall that Y ∼
= X if rk N (X) = 20.)
2.3. Proofs. Let us denote by e1 the canonical generator of H 0 (X, Z) and

by e2 the canonical generator of H 4 (X, Z). They generate the sublattice U in
H ∗ (X, Z). Consider the Mukai vector v = ae1 + H + ae2 = (a, H, a). We have
(25) N (Y ) = v ⊥
U⊕N (X) /Zv.
⊥
Let us calculate N (Y ). Let K(H) = HN (X) . Then we have the embedding of
lattices of finite index
(26) ZH ⊕ K(H) ⊂ N (X) ⊂ N (X)∗ ⊂ ZH ∗ ⊕ K(H)∗
where H ∗ = H/2a2 . We have the orthogonal decomposition up to finite index
(27) U ⊕ ZH ⊕ K(H) ⊂ U ⊕ N (X) ⊂ U ⊕ ZH ∗ ⊕ K(H)∗ .
Let s = x1 e1 + x2 e2 + yH ∗ + z ∗ ∈ vU⊕N
⊥ ∗ ∗
(X) , z ∈ K(H) . Then −ax1 − ax2 + y = 0
since s ∈ v ⊥ and hence (s, v) = 0. Thus, y = ax1 + ax2 and
(28) s = x1 e1 + x2 e2 + a(x1 + x2 )H ∗ + z ∗ .
Here s ∈ U ⊕ N (X) if and only if x1 , x2 ∈ Z and a(x1 + x2 )H ∗ + z ∗ ∈ N (X). This
orthogonal complement contains
(29) [Zv, K(H), Zh]
where h = −e1 + e2 , and this is a sublattice of finite index in (v ⊥ )U⊕N (X) . The
generators v, generators of K(H) and h are free, and we can rewrite s above using
these generators with rational coefficients as follows:
−x1 + x2 x1 + x2
(30) s= h+ v + z∗ ,
2 2a
where a(x1 + x2 )H ∗ + z ∗ ∈ N (X). Equivalently, for h∗ = h/2,
v
(31) s = x1 h∗ + x2 + z∗ ,
2a
where x1 , x2 ∈ Z, z ∗ ∈ K(H)∗ , x1 ≡ x2 mod 2, and ax2 H ∗ + z ∗ ∈ N (X).
From these calculations, we get
Claim. Assume that γ(H) = 1. Then
H
(32) N (X) = [H, K(H), + u∗ ],
2a2
h h
(33) N (Y ) = [h, K(h) = [K(H), 2au∗ ], + au∗ ] = [h, K(h), + w∗ ],
2 2
where u∗ + K(H) has order 2a2 in AK(H) , w∗ = au∗ , K(h) = [K(H), 2w∗ = 2au∗ ].
Here we match our notations with Sect. 2.2. We have det N (X) = det K(H)/2a2
and det N (Y ) = det K(h)/2 (in particular, γ(h) = 1). Thus, det N (X) = det N (Y )
for this case, since det K(h) = det K(H)/a2 . We can formally put here h = H
2 a
since h2 = H a = 2.
From the claim, we get
Lemma 2.3. For the Mukai identification (13), the sublattice T (X) ⊂ T (Y )
has index 1 if γ(H) = 1 for H ∈ N (X).
Proof. Indeed, since H ∈ N (X), T (X) ⊥ N (X) and T (X) ∩ Zv = {0},

the Mukai identification (13) gives an embedding T (X) ⊂ T (Y ). We then have
det T (Y ) = det T (X)/[T (Y ) : T (X)]2 . Moreover, | det T (X)| = | det N (X)| and
| det T (Y )| = | det N (Y )| because the transcendental and the Picard lattice are or-
thogonal complements to each other in the unimodular lattice H 2 (X, Z) ∼ = H 2 (Y, Z).
By (32) and (33) we get the statement.
The statement of Lemma 2.3 is a particular case of the general statement by

Mukai [7] that
(34) [T (Y ) : T (X)] = q = min |v · x|
for all x ∈ H 0 (X, Z) ⊕ N (X) ⊕ H 4 (X, Z) such that v · x = 0. For our Mukai vector
v = (a, H, a) we obviously get that q = 1 if and only if for H ∈ N (X) either
γ(H) = 1 or γ(H) = 2 and a is odd. Thus, for a even we have the only case
γ(H) = 1.
Proof of Proposition 2.1. By (34), if Y ∼ = X, we have either γ(H) = 1 or
γ(H) = 2 and a is odd.
Assume that γ(H) = 1 or γ(H) = 2 and a is odd. Then T (X) = T (Y ) for
the Mukai identification (13). By the discriminant forms technique (see [10]), the
discriminant quadratic forms qN (X) = −qT (X) and qN (Y ) = −qT (Y ) are isomorphic.
Thus, the lattices N (X) and N (Y ) have the same signatures and discriminant
quadratic forms. It follows (see [10]) that they have the same genus: N (X) ⊗ Zp ∼ =
N (Y ) ⊗ Zp for any prime p and the ring of p-adic integers Zp . Additionally, assume
that either the genus of N (X) or the genus of N (Y ) contains only one class. Then
N (X) and N (Y ) are isomorphic.
If additionally the canonical homomorphism O(N (X)) → O(qN (X) ) (equiva-
lently, O(N (Y )) → O(qN (Y ) )) is epimorphic, then the Mukai identification T (X) =
T (Y ) can be extended to give an isomorphism φ : H 2 (X, Z) → H 2 (Y, Z) of cohomol-
ogy lattices. The Mukai identification is an identification on H 2,0 (X) = H 2,0 (Y ).
Multiplying φ by ±1 and by elements of the reflection group W (−2) (N (X)), if nec-
essary, we can assume that φ(H 2,0 (X)) = H 2,0 (Y ) and φ maps the Kähler cone of
X to the Kähler cone of Y . By the global Torelli Theorem for K3 surfaces [12], φ
is then defined by an isomorphism of K3 surfaces X and Y .
If ρ(X) ≥ 12, by [10] Theorem 1.14.4, the primitive embedding of T (X) = T (Y )
into the cohomology lattice H 2 (X, Z) of K3 surfaces is unique up to automorphisms
of the lattice H 2 (X, Z). As above, it then follows that X is isomorphic to Y . The
given proof of Proposition 2.1 is standard and well-known.
Proof of Theorem 2.2. Assume that γ(H) = 1. The Mukai identification then
gives the canonical identification
(35) T (X) = T (Y ).
Thus, it gives the canonical identifications
AN (X) = N (X)∗ /N (X) = (U ⊕ N (X))∗ /(U ⊕ N (X)) = T (X)∗ /T (X)
(36)
= AT (X) = AT (Y ) = T (Y )∗ /T (Y ) = N (Y )∗ /N (Y ) = AN (Y ) .
Here AN (X) = N (X)∗ /N (X) = (U ⊕N (X))∗ /(U ⊕N (X)) because U is unimodular,
(U ⊕ N (X))∗ /(U ⊕ N (X)) = T (X)∗ /T (X) = AT (X) because U ⊕ N (X) and T (X)
are orthogonal complements to each other in the unimodular lattice H ∗ (X, Z). Here
AT (Y ) = T (Y )∗ /T (Y ) = N (Y )∗ /N (Y ) = AN (Y ) because T (Y ) and N (Y ) are
orthogonal complements to each other in the unimodular lattice H 2 (Y, Z). E. g.

the identification (U ⊕ N (X))∗ /(U ⊕ N (X)) = T (X)∗ /T (X) = AT (X) is given by
the canonical correspondence
(37) x∗ + (U ⊕ N (X)) → t∗ + T (X)
if x∗ ∈ (U ⊕ N (X))∗ , t∗ ∈ T (X)∗ and x∗ + t∗ ∈ H ∗ (X, Z).
By (33), we also have the canonical embedding of lattices
(38) K(H) ⊂ K(h) = [K(H), 2au∗ ].
We have the key statement:
Lemma 2.4. Assume that γ(H) = 1. The canonical embedding (38) (it is given
by (33)) K(H) ⊂ K(h) of lattices, and the canonical identification AN (X) = AN (Y )
(given by (36)) agree with the characteristic homomorphisms κ(H) : K(H)∗ →
AN (X) and κ(h) : K(h)∗ → AN (Y ) , i.e. κ(h)(k∗ ) = κ(H)(k∗ ) for any κ∗ ∈ K(h)∗ ⊂
K(H)∗ (this embedding is dual to (38)).
Proof. As the proof of Lemma 2.3.2 in [4].
Let us finish the proof of Theorem 2.2. We have the Mukai identification (it is
defined by (13)) of the transcendental periods
(39) (T (X), H 2,0 (X)) = (T (Y ), H 2,0 (Y )).
For general X with the Picard lattice N (X), it is the unique isomorphism of the
transcendental periods up to multiplication by ±1. If X ∼ = Y , this (up to ±1)
isomorphism can be extended to φ : H 2 (X, Z) ∼= H 2 (Y, Z). The restriction of φ on
N (X) gives then isomorphism φ1 : N (X) ∼ = N (Y ) which is ±1 on AN (X) = AN (Y )
under the identification (36). The element h = (φ1 )−1 (h) and f = φ−1 satisfy
Theorem 2.2 by Lemma 2.4.
The other way around, under conditions of Theorem 2.2, by Lemma 2.4, one
can construct an isomorphism φ1 : N (X) ∼ = N (Y ) which is ±1 on AN (X) = AN (Y ) .
It can be extended to be ±1 on the transcendental periods under the Mukai iden-
tification (39). Then it is defined by the isomorphism φ : H 2 (X, Z) → H 2 (Y, Z).
Multiplying φ by ±1 and by reflections from W (−2) (N (X)), if necessary (the group
W (−2) (N (X)) acts identically on the discriminant group N (X)∗ /N (X)), we can
assume that φ maps the Kähler cone of X to the Kähler cone of Y . By the global
Torelli Theorem for K3 surfaces [12], it is then defined by an isomorphism of X
and Y .
3. The case of Picard number 2

3.1. The case when ρ(X) = 2 and γ(H) = 1 for a odd. Here we apply
results of Sect. 2 to X and Y with Picard number 2.
We start with some preliminary considerations on K3 surfaces X with Picard
number 2 and a primitive polarization H of degree H 2 = 2a2 , a ≥ 1. Thus, we
assume that rk N (X) = 2. Additionally, we assume that γ(H) = 1 for H ∈ N (X)
(we have this condition if a is even and Y ∼
= X). Let
⊥
K(H) = HN (X) = Zδ
and δ 2 = −t where t > 0 is even. Then δ ∈ N (X) is defined uniquely up to ±δ. It

then follows that
δ
N (X) = [ZH, Zδ, µH ∗ + 2 ]
2a
where H ∗ = H/2a2 and g.c.d.(µ, 2a2 ) = 1. The element
±µ mod 2a2 ∈ (Z/2a2 )∗
is the invariant of the pair (N (X), H) up to isomorphisms of lattices with the
primitive vector H of H 2 = 2a2 . If δ changes to −δ, then µ mod 2a2 changes to
−µ mod 2a2 . We have
δ 2 1 t
(µH ∗ + 2
) = 2 (µ2 − 2 ) ≡ 0 mod 2.
2a 2a 2a
Then t = 2a2 d, for some d ∈ N and µ2 ≡ d mod 4a2 . Thus, d mod 4a2 ∈
(Z/4a2 )∗ 2 . Obviously, −d = det(N (X)).
Any element z ∈ N (X) can be written as z = (xH + yδ)/2a2 where x ≡ µy
mod 2a2 . In these considerations, one can replace H by any primitive element of
N (X) with square 2a2 . Thus, we have:
Proposition 3.1. Let X be a K3 surface with Picard number ρ = 2 and a
primitive polarization H of degree H 2 = 2a2 , a > 0, and γ(H) = 1 for H ∈ N (X).
The pair (N (X), H) has the invariants d ∈ N and ±µ mod 2a2 ∈ (Z/2a2 )∗
such that µ2 ≡ d mod 4a2 . (It follows that d ≡ 1 mod 4.)
⊥
For the invariants d, µ we have: det N (X) = −d, and K(H) = HN (X) = Zδ
where δ = −2a d. Moreover,
2 2
(40) N (X) = [H, δ, (µH + δ)/2a2 ],
(41) N (X) = {z = (xH + yδ)/2a2 | x, y ∈ Z and x ≡ µy mod 2a2 },

and z 2 = (x2 − dy 2 )/2a2 .
For any primitive element H ∈ N (X) with (H )2 = H 2 = 2a2 and the same
invariant ±µ, there exists an automorphism φ ∈ O(N (X)) such that φ(H) = H .
Applying Proposition 3.1 to a = 1, we get that the pair (N (Y ), h) with h2 = 2
and γ(h) = 1 is defined by its determinant det N (Y ) = −d where d ≡ 1 mod 4.
Thus, from Proposition 3.1, we get
Proposition 3.2. Under conditions and notations of Propositions 3.1, all el-
ements h = (xH + yδ)/(2a2 ) ∈ N (X) with (h )2 = 2 are in one to one correspon-
dence with integral solutions (x, y) of the equation
(42) x2 − dy 2 = 4a2
such that x ≡ µy mod 2a2 .
The Picard lattices of X and Y are isomorphic, N (X) ∼
= N (Y ), if and only if
there exists such a solution.
Proof. It follows from the fact that γ(h) = 1 for h ∈ N (Y ) because of (33).

The crucial statement is

Theorem 3.3. Let X be a K3 surface, ρ(X) = 2 and H a primitive polarization

of X of degree H 2 = 2a2 , a > 1. Let Y be the moduli space of sheaves on X with the
Mukai vector v = (a, H, a) and the canonical nef element h = (−1, 0, 1) mod Zv.
Assume that γ(H) = 1 (for even a only for this case can we have Y ∼ = X). Then
we can introduce the invariants ±µ mod 2a2 ∈ (Z/2a2 )∗ and d ∈ N of (N (X), H)
as in Proposition 3.1. Thus, we have
(43) γ(H) = 1, det N (X) = −d where µ2 ≡ d mod 4a2 .
With the notations of Propositions 3.1, all elements
h = (xH + yδ)/(2a2 ) ∈ N (X)
2
with square h = 2 satisfying Theorem 2.2 are in one to one correspondence with
integral solutions (x, y) of the equation
(44) x2 − dy 2 = 4a2
with x ≡ µy mod 2a2 and x ≡ ±2a mod d.
In particular (by Theorem 2.2), for a general X with ρ(X) = 2 and γ(H) = 1
for odd a, we have Y ∼ = X if and only if the equation x2 − dy 2 = 4a2 has an
integral solution (x, y) with x ≡ µy mod 2a2 and x ≡ ±2a mod d. Moreover, a
nef element h = (xH + yδ)/2a2 with h2 = 2 defines the structure of a double plane
on X which is isomorphic to the double plane Y if and only if x ≡ ±2a mod d.
Proof. Let h ∈ N (X) satisfy conditions of Theorem 2.2.
By Proposition 3.2, all primitive
(45)
h = (xH + yδ)/(2a2 ) ∈ N (X)
with
h2 = 2 are in one to one correspondence with integral (x, y) which satisfy the
equation x2 − dy 2 = 4a2 and x ≡ µy mod 2a2 , and any integral solution of the
equation x2 − dy 2 = 4a2 with x ≡ µy mod 4a2 gives such an h.
Let k = aH + bδ ∈ h⊥ = Zα. Then (k, h) = ax − byd = 0 and (a, b) = λ(yd, x).
Hence, we have
(46) (λ(ydH + xδ))2 = λ2 (2a2 y 2 d2 − 2a2 dx2 ) = 2a2 λ2 d(y 2 d − x2 ) = −2(2a2 )2 λ2 d.
Since α2 = −2d, we get λ = 1/2a2 and α = (ydH + xδ)/2a2 . There exists a unique
(up to ±1) embedding f : K(H) = Zδ → K(h) = Zα of one-dimensional lattices.
It is given by f (δ) = aα up to ±1. Thus, its dual is defined by f ∗ (α∗ ) = aδ ∗ where
α∗ = α/2d and δ ∗ = δ/2a2 d. To satisfy the conditions of Theorem 2.2, we should
have
(47) κ(
h)(α∗ ) = ±aκ(H)(δ ∗ ).
Further we denote ν = µ−1 mod 2a2 . We have u∗ = νdδ ∗ , w∗ = af (u∗ ) = ν α2 ,
and
(48) h)(α∗ ) = (−2α∗ · w∗ )
κ( h∗ + α∗ + N (X)
by (21). Here
h∗ =
h/2. We then have α∗ · w∗ = − ν2 , and
νx + y ∗ νyd + x ∗
(49) κ(h)(α∗ ) = ν
h∗ + α∗ + N (X) = ( )H + ( )δ
2 2
where H ∗ = H/2a2 .
We have u∗ = νdδ ∗ = νδ/2a2 . By (21), κ(H)(δ ∗ ) = (−2a2 δ ∗ · u∗ )H ∗ + δ ∗ + N (X).
We have δ ∗ · u∗ = −ν/2a2 . It follows that
(50) κ(H)(δ ∗ ) = νH ∗ + δ ∗ .
By (49) and (50), we then obtain that κ(H)(aδ ∗ ) = ±κ( h)(α∗ ) is equivalent to
(νyd + x)/2 ≡ ±a mod d and hence x + νyd ≡ ±2a mod d since the group
N (X)∗ /N (X) is cyclic of order d and it is generated by νH ∗ + δ ∗ + N (X). Thus,
finally we get x ≡ ±2a mod d.
This finishes the proof.
By Theorem 3.3, for given a > 1, ±µ ∈ (Z/2a2 )∗ and d ∈ N such that d ≡ µ2

mod 4a2 , we should look for all integral (x, y) such that.
(51) x2 − dy 2 = 4a2 , x ≡ µy mod 2a2 , x ≡ ±2a mod d.
We first describe the set of integral (x, y) such that
(52) x2 − dy 2 = 4a2 , x ≡ ±2a mod d.
Considering ±(x, y), we can assume that x ≡ 2a mod d. Then x = 2a − kd where
k ∈ Z. We have 4a2 − 4akd + k2 d2 − dy 2 = 4a2 . Thus,
y 2 + 4ak
d= .
k2
Let l be prime. As in [4], it is easy to see that if l2t+1 |k and l2t+2 does not divide
k, then l2t+2 |4ak. It follows that k = −αq 2 where q ∈ Z, α ∈ Z is square-free and
α|2a. Then
y 2 − 4aαq 2
d= .
α2 q 4
It follows that αq|y and y = αqp where p ∈ Z. We then get
p2 − 4a/α
(53) d= .
q2
Equivalently,
4a
(54) α|2a is square-free, p2 − dq 2 = .
α
We recall that α can be negative.
Thus, solutions α, (p, q) of (54) give all solutions
(55) (x, y) = ±(2a + αdq 2 , αpq)
of (52). We call them associated solutions. Thus, all solutions (x, y) of (52) are
associated solutions (55) to all solutions α, (p, q) of (54). If one additionally assumes
that q > 0, then (x, y) and α, (p, q) are in one to one correspondence (by our
construction).
Now let us consider associated solutions (55) which satisfy the additional con-
dition x ≡ µy mod 2a2 . We have
(56) 2a + αdq 2 ≡ µαpq mod 2a2 .
Using d ≡ µ2 mod 4a2 , we get
2a
(57) ≡ µq(p − µq) mod 2a2 /α.
α
Since p2 − dq 2 = 4a/α, we get from (56)
2a
(58) − ≡ −p(p − µq) mod 2a2 /α.
α
Taking the sum of (57) and (58), we get

(59) (p − µq)2 ≡ 0 mod 2a2 /α.
Since α|2a2 and α is square-free, it follows easily that a|(p − µq) and (p − µq)/a is
an integer. From (58), we then get

p − µq
(60) a|p − µq and 2 ≡ αp mod 2a.
a
The condition x ≡ µy mod 2a2 is equivalent to (60).
From (60), we get α|2. Thus, α = ±1 or α = ±2. Let us consider both cases.
Assume α = ±1. By (59), then 2a|(p − µq), and we can rewrite (60) as

p − µq
(61) 2a|p − µq and ± 1 ≡ p mod a.
2a
For α = ±1 we have p2 − dq 2 = ±4a. It follows that (p − µq)(p + µq) ≡ ±4a
mod 4a2 . If 2a|(p − µq), then ((p − µq)/2a)(p + µq) ≡ ((p − µq)/2a)2p ≡ ±2
mod 2a which is equivalent to (61).
Thus, for α = ±1, associated solutions (x, y) to α = ±1, (p, q) satisfy the
additional condition x ≡ µy mod 2a2 if and only if p ≡ µq mod 2a. Equivalently,
ap ≡ µaq mod 2a2 which is equivalent to h1 = (apH + aqδ)/2a2 ∈ N (X). The
equation p2 −dq 2 = ±4a is equivalent to (h1 )2 = ±2a. We also have h1 ·H = ap ≡ 0
mod a. Conversely, assume h1 = (uH + vδ)/2a2 ∈ N (X), and (h1 )2 = ±2a,
h1 · H ≡ 0 mod a. We then have h1 · H = u ≡ 0 mod a. Thus, u = ap and
h1 = (ap + vδ)/2a2 . Since h1 ∈ N (X), then ap ≡ µv mod 2a2 . Thus, v = aq and
p ≡ µq mod 2a. Since (h1 )2 = ±2a, we get p2 − dq 2 = ±4a. We also remark that
from the conditions (60) and p2 − dq 2 = ±4a, it follows that
(62) g.c.d(αa, p) = g.c.d(αa, q) = 1; g.c.d(p, q)|(2/α).
Thus, (p, q) is an “almost primitive” solution of the equation p2 − dq 2 = ±4a. It is
primitive if a is even, and g.c.d(p, q)|2 if a is odd.
Now assume that α = ±2. Then (60) is equivalent to

p − µq
(63) a|p − µq and ± 1 ≡ p mod a.
a
Assume that a|p − µq and p2 − dq 2 = ±2a. Then (p − µq)(p + µq) ≡ ±2a mod 4a2
and ((p − µq)/a)(p + µq) ≡ ((p − µq)/a)2p ≡ ±2 mod a. If a is odd, this is
equivalent to (63).
Assume that a is even. If (p − µq)/a is even, then p + µq is also even. From
p2 −dq 2 = ±2a, we then get ((p−µq)/2a)(p+µq) ≡ ±1 mod 2a which is impossible
for even p + µq. Thus, p − µq ≡ a mod 2a and µq ≡ p + a mod 2a. From
((p − µq)/a)(p + µq) ≡ ±2 mod 4a, we then get ((p − µq)/a)(2p + a) ≡ ±2 mod 2a
and ((p − µq)/a)p + ((p − µq)/a)(a/2) ≡ ±1 mod a. Since (p − µq)/a is odd, it
follows that (63) is never satisfied for even a.
Thus, we get that for α = ±2 the number a is odd and the condition x ≡ µy
mod 2a2 is equivalent to p ≡ µq mod a. Let us consider this case. From p2 −dq 2 =
±2a and d odd, we get that p ≡ q mod 2. Since a is odd, we then get p ≡ µq
mod 2a and p2 ≡ µ2 q 2 mod 4a. This contradicts p2 − dq 2 = ±2a because d ≡ µ2
mod 4a2 . Thus, α = ±2 is impossible for odd a too.
Finally we get the main results.
Theorem 3.4. Under the conditions of Theorem 3.3, for a general X with
ρ(X) = 2 and γ(H) = 1 for odd a, we have Y ∼ = X if and only if at least for one
α = ±1 there exists integral (p, q) such that
4a
(64) p2 − dq 2 = and p ≡ µq mod 2a.
α
Solutions (p, q) of (64) are “almost primitive”; they satisfy (62).
Solutions (p, q) of (64) give all solutions (44) of Theorem 3.3 as associated
solutions
(x, y) = ±(2a + αdq 2 , αpq).
Interpreting, as above, solutions (p, q) of (64) as elements of N (X), we also get
Theorem 3.5. Under the conditions of Theorem 3.3, for a general X with
ρ(X) = 2 and γ(H) = 1 for odd a, we have Y ∼
= X if and only if at least for one
α = ±1 there exists h1 ∈ N (X) such that
h21 = 2αa and h1 · H ≡ 0 mod a.
Applying additionally Theorem 2.2, we get the following simple sufficient con-
dition when Y ∼ = X which is valid for X with any ρ(X). This is one of the main
results of the paper.
Theorem 3.6. Let X be a K3 surface and H a primitive polarization of degree
2a2 , a ≥ 2. Let Y be the moduli space of sheaves on X with the Mukai vector
v = (a, H, a).
Then Y ∼= X if at least for one α = ±1 there exists h1 ∈ N (X) such that
(65) (h1 )2 = 2αa, h1 · H ≡ 0 mod a,
and γ(H) = 1 for H ∈ [H, h1 ]pr where [H, h1 ]pr is the primitive sublattice of N (X)
generated by H, h1 .
This condition is necessary to have Y ∼= X and γ(H) = 1 for a odd if either
ρ(X) = 1, or ρ(X) = 2, and X is a general K3 surface with its Picard lattice (i.e.
the automorphism group of the transcendental periods (T (X), H 2,0 (X)) is ±1).
Proof. The cases ρ(X) ≤ 2 have been considered. We can assume that
ρ(X) > 2. Let N = [H, h1 ]pr . All considerations above for N (X) of rk N (X) = 2
will be valid for N . We can construct an associated with h1 solution h ∈ N with

h2 = 2 such that H and h satisfy conditions of Theorem 2.2 for N (X) replaced by
N . It is easy to see that the conditions (b) and (c) will still be satisfied if we extend
⊥
f in (b) by ± the identity on the orthogonal complement NN (X) .
It seems that many known examples of Y ∼ = X (e. g. see [2], [8], [17]) follow
from Theorem 3.6. Theorems 3.5 and 3.6 are also interesting because they give
a very clear geometric interpretation of some elements h1 ∈ N (X) with negative
square h21 (for negative α).
Below we consider an application of Theorem 3.4.
3.2. Divisorial conditions on moduli (X, H) when γ(H) = 1 for odd
a. Further we use the following notations. We fix a ∈ N, α ∈ {1, −1} and µ =
{µ, −µ} ⊂ (Z/2a2 )∗ . We denote by D(a)α µ
the set of all d ∈ N such that d ≡ µ2
mod 4a and there exists an integral (p, q) such that p2 − dq 2 = 4a/α and p ≡ µq
2
mod 2a. We denote by D(a)µ the union of D(a)α µ

for all α ∈ {1, −1}, by D(a)α the
union of D(a)µα for all µ = {µ, −µ} ⊂ (Z/2a2 )∗ , and by D(a) the union of all D(a)α
for all α ∈ {1, −1}.
Assume that X is a K3 surface with a primitive polarization H of degree H 2 =
2
2a where a > 1. The moduli space Y of sheaves on X with Mukai vector v =
(a, H, a) has the canonical nef element h = (−1, 0, 1) mod Zv with h2 = 2. It
follows that Y is never isomorphic to X if ρ(X) = 1. Since the dimension of
moduli of (X, H) is equal to 20 − ρ(X), it follows that describing general (X, H)
with ρ(X) = 2 and Y ∼ = X, we at the same time describe all possible divisorial
conditions on moduli of (X, H) when Y ∼ = X. See [9], [10] and also [3]. They are
described by invariants of the pairs (N (X), H) where rk N (X) = 2. By Theorem
3.4, we get
Theorem 3.7. All possible divisorial conditions on moduli of polarized K3 sur-

faces (X, H) with a primitive polarization H with H 2 = 2a2 , a > 1, which imply
Y ∼= X and γ(H) = 1 for a odd, are labelled by the set Div(a) of all pairs
(d, µ)

where µ = {µ, −µ} ⊂ (Z/2a2 )∗ , d ∈ D(a)µ = α D(a)µα . Here α ∈ {1, −1}.
For any µ = {µ, −µ} ⊂ (Z/2a2 )∗ and any α ∈ {1, −1} the set
p2 − 4a/α
D(a)µα = {d = ∈ N | q ∈ N, p ≡ µq mod 2a, d ≡ µ2 mod 4a2 }
q2
⊃ {(µ + t(2a/α))2 − 4a/α ∈ N | tµ ≡ 1 mod a}
is infinite (put q = 1 to get the last infinite subset).
In particular, for any a > 1, the set of possible divisorial conditions on moduli
of (X, H) which imply Y ∼ = X is infinite.
To enumerate the sets D(a)α µ
, α ∈ {1, −1}, it is most important to enumerate
the sets D(a)α . This is almost equivalent to finding all possible d ∈ N such that
d mod 4a2 ∈ (Z/4a2 )∗2 and the equation p2 − dq 2 = 4a/α has a solution (p, q)
satisfying (62) (it is “almost primitive”). Each such solution (p, q) defines a unique
(if it exists) µ mod 2a2 such that µ2 ≡ d mod 4a2 and p ≡ µq mod 2a. The pair
(d, µ) then gives an element of the set Div(a). Thus, to find all possible µ (for
the given α, d), it is enough to find all “almost primitive” solutions (p, q) of the
equation p2 − dq 2 = 4a/α and ν mod 2a such that p ≡ νq mod 2a.
Indeed, let (p, q) be a solution of the equation p2 − dq 2 = 4a/α. For example,
assume that it is primitive. Let ν ≡ p/q mod 2a. Then ν 2 ≡ d mod 4a. We
have µ = ν + 2ka mod 2a2 , k ∈ Z, and µ2 ≡ ν 2 + 4kνa + 4k2 a2 ≡ d mod 4a2 .
Equivalently,
4νka ≡ d − ν 2 mod 4a2 .
Finally, we get
d − ν2
νk ≡ mod a
4a
which determines µ mod 2a2 uniquely. If g.c.d(p, q) = 2, then a is odd. For this
case, one should again start with ν mod 2a such that p ≡ νq mod 2a and ν 2 ≡ d
mod 4a. Then there exists a unique lifting µ mod 2a2 such that µ ≡ ν mod 2a
and µ2 ≡ d mod 4a2 .
References
1. Borevich, Z. I. and Shafarevich, I. R., Number Theory (3d edition) Nauka, Moscow, 1985,
English transl., Academic Press, 1966
2. Cossec, F. R., Reye Congruences, Trans. Amer. Math. Soc., 280, (1983) 737–751.
3. James, D. G., On Witt’s theorem for unimodular quadratic forms, Pacific J. Math., 26,
(1968), 303–316.
4. Madonna C., and Nikulin, V. V., On a classical correspondence between K3 surfaces, Proc.
Steklov Math. Inst., 241 (2003) 132-168 (Russian).
5. Mayer, A., Families of K-3 surfaces, Nagoya Math. J., 48 (1982), 1–17.
6. Mukai, S., Symplectic structure of the moduli space of sheaves on an Abelian or K3 surface,
Invent. math., 77, (1984) 101–116.
7. Mukai, S., On the muduli space of bundles on K3 surfaces, in Vector bundles on algebraic
varieties (Bombay, 1984), Tata Inst. Fund. Res. Studies in Math. no. 11, Bombay, 1987,
341–413.
8. Mukai, S., Duality of polarized K3 surfaces, in New trends in algebraic geometry. Selected
papers presented at the Euro conference, K. Hulek (ed.), Warwick, UK, July 1996, Cambridge
University Press. London Math. Soc. Lect. Notes Ser. 264, Cambridge, 1999, 311–326.
9. Nikulin, V. V., Finite automorphism groups of Kählerian surfaces of type K3, Trudy Mosk.
Matem. Ob-va, 38, (1979), 75–137. English transl. in Trans. Moscow Math. Soc., 38, (1980)
71–135.
10. Nikulin, V. V., Integral symmetric bilinear forms and some of their geometric applications,
Izv. Akad. Nauk SSSR Ser. Mat., 43, (1979) 111–177. English transl. in Math. USSR Izv.,
14, (1980).
11. Nikulin, V. V., On correspondences between K3 surfaces, Izv. Akad. Nauk SSSR Ser. Mat.,
51, (1987) 402–411. English transl. in Math. USSR Izv., 30, (1988).
12. Pjatetski-Shapiro, I. I., and Shafarevich, I. U., A Torelli theorem for algebraic surfaces of
type K3, Izv. Akad. Nauk SSSR Ser. Mat., 35, (1971) 530–572. English transl. in Math.
USSR Izv., 5, (1971) 547–588.
13. Saint-Donat, B., Projective models of K–3 surfaces, Amer. J. of Mathem., 96, (1974) 602–
639.
14. Shafarevich, I. R. (ed.) Algebraic surfaces, Trudy Matem. Inst. Steklov, T. 75, 1965. English
transl. in Proc. Stekov Inst. Math., 75 (1965).
15. Shokurov, V. V., The Noether–Enriques theorem on canonical curves, Mat. Sbornik, 86,
(1971) 367–408. English transl. in Math. USSR-Sb. 15, (1971) 361–404.
16. Tyurin, A. N., The intersection of quadrics, Uspekhi Mat. Nauk, 30, (1975) 51–99. English
transl. in Russian Math. Surveys, 30 (1975) 51–105.
17. Tyurin, A. N., Special 0-cycles on a polarized K3 surface Izv. Akad. Nauk SSSR Ser. Mat.,
51, (1987) 131–151. English transl. in Math. USSR Izv., 30, (1988) 123–143.
18. Tyurin, A. N., Cycles, curves and vector bundles on algebraic surfaces, Duke Math. J., 54,
(1987) 1–26.
19. Tyurin, A. N., Symplectic structures on the varieties of moduli of vector bundles on algebraic
surfaces with pg > 0, Izv. Akad. Nauk SSSR Ser. Mat., 52, (1988) 139–178. English transl.
in Math. USSR Izv., 33, (1989) 139–177.
Dipartimento di Matematica, Università degli studi di Ferrara, Italia

E-mail address: madonna@mat.uniroma3.it
Dept. of Pure Math., The University of Liverpool, Liverpool L69 3BX, UK
Steklov Mathematical Institute, ul. Gubkina 8, Moscow 117966, GSP-1, Russia

E-mail address: vnikulin@liv.ac.uk slava@nikulin.mian.su
Volume 3, 2004
Contractions and monodromy in homological mirror

symmetry
Balázs Szendrői
Abstract. This paper discusses the mirror correspondence between contrac-

tions and degenerations of Calabi–Yau varieties, originally due to Morrison,
in the light of recent developments. In homological mirror symmetry, de-
generations lead to symplectomorphisms, whereas contractions give rise to
Fourier–Mukai functors. Several explicit examples are treated, many of them
conjectural.
Introduction
Kontsevich’s Homological Mirror Symmetry conjecture [14] connects holomor-
phic and symplectic geometry in a deep and surprising way. It relates a pair of
Calabi–Yau varieties, one with fixed holomorphic (complex) structure, one with
fixed symplectic structure, and predicts an equivalence of two very different kinds
of categories. On the symplectic side, “the (derived) Fukaya category” has gone
through many transfigurations over the years, and there is still no unique defi-
nition. On the holomorphic side, the category is well known: it is the derived
category of coherent sheaves. The conjectured equivalence between these two cat-
egories implies that there should be a correspondence between their symmetries.
The obvious symmetries are symplectic, respectively holomorphic automorphisms;
however, typically there are many more of the former than the latter. The idea
of the Fourier–Mukai functor [19] comes to the rescue: the derived category of a
Calabi–Yau variety possesses more symmetries than just the obvious ones, just by
virtue of the fact that its canonical bundle is globally trivial.
The purpose of this note is to review the dictionary between symplectic iso-
morphisms and Fourier–Mukai functors in a pedagogical way. Versions of a global
correspondence (at least on the cohomological level) are discussed for example in
[3, 10, 30]; a comprehensive account will presumably be attempted in [7]. My
aim here is rather more limited, summarizing and slightly extending the range of
constructions that are local to singularities of Calabi–Yau varieties. I will discuss
surface double points, threefold nodes, isolated threefold singularities and finally
curves of singularities on threefolds, proving no new results but posing several open
2000 Mathematics Subject Classification. 14J32, 14E15, 18E30.
2004
301
302 BALÁZS SZENDRŐI
problems. The presentation is at the level of suggestive analogies, relying on the

shape of cohomology actions, relations, toric examples and the like.
The real question this note completely disregards is of course what exactly is
mirror symmetry for singularities. Recent ideas of Kontsevich, Kapustin–Li [13]
and Orlov [20] on the one hand, related on the physics side to D-branes in Landau–
Ginzburg models, and Seidel [24, 25] on categories defined by vanishing cycles on
the other, will provide the tools to ask and eventually answer this question in much
more precise detail than attempted here.
I thank Mark Gross for pointing out the relation between toric Gorenstein
singularities, polytopes and degenerations.
1. Homological mirror symmetry in a nutshell

1.1. I shall make no attempt to give a general overview of mirror symmetry.
There are several good sources available; most relevant for this note is Kontsevich’s
1994 ICM address [14] and its 2000 ECM update by Manin [16], both with ex-
tensive bibliography. As originally formulated in string theory, mirror symmetry
relates two Calabi–Yau manifolds (X, ωX ) and (Y, ωY ), both equipped with com-
plex structures and compatible symplectic (Kähler) forms, and certain structures
defined on (families of) them. In fact, at least deep inside the moduli space, mirror
symmetry “decouples” the symplectic and the complex structures, and interchanges
one with the other. Kontsevich conjectured that eventually all mirror symmetry
constructions should be understood as an equivalence of categories depending on
symplectic, respectively holomorphic data:
m : Db Fuk(Y, ω) −→ Db (X).
Here Db Fuk(Y, ω) is the “derived Fukaya category”. It should be constructed purely
and functorially in symplectic terms, using Lagrangian submanifolds of (Y, ω) (now
simply thought of as a symplectic manifold, with no complex structure), and their
Floer homology. The analytic details are highly non-trivial, and in fact to date not
completely settled. On the right hand side, Db (X) is the bounded derived category
of coherent sheaves on the smooth Calabi–Yau variety X. This is a perfectly well-
defined and relatively well-known (triangulated) category; see [9] for a detailed
introduction.
Irrespective of the details of its definition, functoriality of the Fukaya cate-
gory should imply that it carries an action of the group of symplectomorphisms
Symp(Y, ω). Symplectomorphisms symplectically homotopic to the identity (in the
C ∞ -topology) are expected to act trivially. The mirror symmetry equivalence m
should then give rise to a map
(1) µ : π0 (Symp(Y, ω)) → AutEq(Db (X))/[1],
which can be studied independently of the intricacies of the Fukaya category. The
group of self-equivalences AutEq(Db (X)) of the derived category (together with its
triangulated structure) contains [1], the translation functor, which is expected to
correspond to the translation functor in the derived Fukaya category. This corre-
sponds to the difference between “ordinary” and “graded” symplectic geometry, as
explained in [26, Introduction].
1.2. The previous discussion implies that we need a handy supply of self-
equivalences of the derived category. As I already discussed in the introduction,
CONTRACTIONS AND MONODROMY IN HOMOLOGICAL MIRROR SYMMETRY 303
there are two obvious sources: automorphisms of the complex manifold X, as well
as line bundles acting by tensor product. However, one soon realizes that this is
not enough. For example, a generic quintic has no automorphisms other than the
identity, and only a Z worth of line bundles. Its mirror in turn has many more
symplectomorphisms.
A fundamental construction of Mukai [19] comes to the rescue. By analogy
with the theory of classical correspondences, an object F ∈ Db (X × X) defines a
functor
ΦF : Db (X) → Db (X)
by
L
ΦF (−) = Rp1∗ (F ⊗ p∗2 (−)),
L
where pi : X × X → X are the two projections, and ⊗, Rp1∗ denote operations
in the derived category. It often happens that, under appropriate conditions, ΦF
is a self-equivalence of the triangulated category Db (X), and then it is called a
Fourier–Mukai functor [19].
As a special case, this definition includes the example of a self-equivalence
defined by an automorphism; here F is (the structure sheaf of) the graph of the
automorphism. The self-equivalence defined by tensoring with a line bundle L is
realized by F = O∆ (L) where ∆ ∈ X × X is the diagonal. However, the main
virtue of this framework is that in the Calabi–Yau context, there are many more
Fourier–Mukai functors. Several examples will be given below.
1.3. The mirror symmetry map m is expected to be compatible with a coho-
mology isomorphism
m̄ : H middle (Y, Q) −→ H even (X, Q)
under, on the symplectic side, the map taking a Lagrangian submanifold to its
cohomology class, and on the complex side, the map taking a sheaf or complex
to its K-theory class and then via the Chern character to cohomology. For tech-
nical reasons,
√ one uses a slight modification of the Chern character, the Mukai
map ch(−) TdX , to pass from K-theory to cohomology, which makes the map
compatible with natural bilinear pairings. This plays no role in this note.
In fact, here I would like to invoke the “symmetry” of mirror symmetry, which
says that actually the complex geometry of Y is also expected to be equivalent to
the symplectic geometry of (X, ωX ) in the sense of a categorical equivalence. For
the examples studied here (K3s, Calabi–Yau threefolds, but also for elliptic curves)
this means that one actually expects a full isomorphism
(2) m̄ : H ∗ (Y, Q) −→ H ∗ (X, Q)
between Q-vector spaces.
A symplectomorphism of (Y, ω) acts on homology by pushforward. In the
context of cohomology actions, Y will always be a compact manifold; hence I can
use Poincaré duality to think of cohomology as covariant for symplectomorphisms.
A Fourier–Mukai functor ΦF : Db (X) → Db (X) acts on the cohomology of X via
the cohomological correspondence induced by the Chern class of the complex F.
One expects the correspondence µ of (1) between symplectomorphisms and Fourier–
Mukai functors to be compatible with cohomology actions via the isomorphism m̄
of (2).
One specific issue in the Calabi–Yau context is that cupping with the holomor-
phic n-form induces an isomorphism, well-defined up to constant, H 1 (X, TX ) ∼ =
H 1 (X, Ωn−1
X ). Via Hodge theory, the latter is a direct summand of cohomology.
The map ϕF preserves Hodge structures, so one obtains an action of a Fourier–
Mukai functor on H 1 (X, TX ). This is the tangent space to deformations of X,
and the interpretation of this action is that the Fourier–Mukai self-equivalence only
deforms as a self-equivalence along deformation directions fixed by its cohomology
action. This issue is spelled out in [31, Theorem 2.1].
2. Degenerations and contractions

2.1. The specific issue considered here is the relation of symplectomorphisms
and derived equivalences arising from mirror symmetry between degenerations and
contractions [17]. Fix some projective ambient space P, and consider a family
Y ⊂ T × P with projection π : Y → T over a smooth base T , having smooth total
space Y and smooth Calabi–Yau fibres over an open set T 0 ⊂ T whose complement
is a divisor with normal crossings. Assume also that P is equipped with a symplectic
form Ω, compatible with its complex structure, whose restriction ωs to every smooth
fibre Ys of π is still non-degenerate. The map π then becomes a Lefschetz fibration
[23]. In particular, there is a notion of symplectic parallel transport over T 0 . Fixing
a base point t ∈ T 0 with fibre (Y, ω) = (Yt , ωt ), there is a map
σ : π1 (T 0 ) → π0 (Symp(Y, ω)),
the symplectic monodromy of the family π.
As always, if the fibre (Ys , ωs ) is smooth for all s ∈ T , then the symplectic
monodromy is trivial. The interesting case is when Ys is singular for s ∈ T \ T 0 ,
in other words when π represents a degeneration of the complex structure on the
Calabi–Yau manifold Yt , and T \ T 0 lies in the boundary of the space of complex
structures. There is a vast literature on degenerations of Calabi–Yau manifolds;
I will describe some specific situations below.
As mirror symmetry interchanges complex and Kähler moduli, the construc-
tion mirror to degenerations of the complex structure should be a degeneration of
the Kähler structure of the mirror X of Y . The way this can be thought of as
monodromy in an actual moduli space around some boundary should not concern
us here; compare [18] and the much more advanced approach [5] for details. I will
take the old-fashioned view [17] which says that in many cases, Kähler degener-
ation means that a collection of Kähler classes [ωt ] on the (complex) Calabi–Yau
manifold X tends to some limit class [ω0 ] on the boundary of the Kähler cone [33]
of the complex Calabi–Yau manifold X, the cone of Kähler classes on X in the
vector space H 2 (X, R). Such boundary points, in favorable cases, correspond to
contractions on the complex manifold X: algebraic morphisms X → X̄ satisfying
certain conditions [33].
We then get our basic correspondence: the map µ arising from mirror symmetry
considerations should in certain situations relate symplectic monodromy around
boundary points of the complex moduli space of Y to Fourier–Mukai transforms
defined from contractions on the complex Calabi–Yau manifold X. I will further
restrict attention to cases where the contraction X → X̄ is birational, contracting
a locus E ⊂ X to a locus C ⊂ X̄ and inducing an isomorphism between non-empty
open subsets X \ E ∼ = X̄ \ C. As such contractions are more easily classified,
at least into broad classes, I will always start from the contraction and discuss
the corresponding degeneration. I will attempt to be rigorous in notation: X will
be a Calabi–Yau threefold (with fixed complex structure) and X → X̄ a birational
contraction. On the mirror side, Y0 will always be the degenerate complex manifold
(possibly local) with a symplectic smoothing (Y, ω) and, if needed, a resolution Y0 →
Y0 .
2.2. Note that the above discussion assumed a number of details about the
specific form of the degeneration of the complex structure of Y . Just because the
Calabi–Yau variety Y is embedded in some ambient space P (such as a weighted
projective space or a toric variety), it is by no means certain that all deformations
of Y can also be embedded in P or that all symplectic forms on Y arise as restrictions
of some form Ω from P. Hence for a general discussion of the relation between
diffeomorphisms and Fourier–Mukai functors, one needs to treat both symplectic
structure and complex structure in families; compare [30]. However, for my present
purposes the above considerations suffice.
3. Surface double points

In dimension two, there is only one class of birational contractions on Calabi–
Yau varieties: the contraction of a (−2)-curve on a K3 surface, or more generally
the contraction of a tree of such curves. I begin by discussing the case of a single
curve.
3.1. Let X be a smooth K3 surface containing a rational curve P1 ∼ = E ⊂ X of
square (−2). There is a contraction X → X̄ contracting this curve to a point P ∈ X̄,
which is locally analytically isomorphic to the surface node (simple double point)
{x21 + x22 + x23 = 0} ⊂ A3 .
The mirror degeneration to this contraction is known to be a degeneration to
the same type of singularity: this means a one-dimensional family of K3 surfaces
: Y → ∆ with smooth fibres over the punctured disc ∆∗ and a singular fibre Y0
π
containing a single double point. Locally analytically the family Y is isomorphic to
the family
{x21 + x22 + x23 = t} ⊂ A4
where t is the coordinate in the base ∆ and the symplectic form ωt on an open set
of Yt is given by the restriction of the standard form on C3 .
Fix t = 0 as the base point and assume for simplicity that t is positive real.
Then Y = Yt contains a two-sphere S 2
S ⊂ Yt given by
S = {x21 + x22 + x23 = t | xi ∈ R}.
In fact this sphere is Lagrangian in (Y, ω), and it represents the vanishing cycle:
its homology class generates the subspace of homology disappearing under passage
from Y to Y0 . The symplectic monodromy of the family π : Y → ∆∗ was constructed
explicitly with the help of the vanishing cycle S by Arnold [2] as follows.
Consider first a model situation: the cotangent bundle T ∗ S 2 with its canonical
symplectic form η. By means of the standard metric, identify T ∗ S 2 with the tangent
bundle T S 2 . The latter has a circle action σ, defined by the normalized geodesic
flow, transporting a tangent vector ξ with unit speed along the geodesic emanating
from it, irrespective of its length; regard this as a circle action on T ∗ S 2 . Consider
also an auxiliary smooth real function ψ such that ψ(t) + ψ(−t) = 2π for all t ∈ R,
and ψ(t) = 0 for t >> 0. Now define τ : T ∗ S 2 → T ∗ S 2 by

σ(eiψ(|ξ|) )(ξ) ξ ∈ T ∗ S 2 \ S 2
τ (ξ) =
A(ξ) ξ ∈ S2
where A is the antipodal map on the sphere. It is easy to check that τ is continuous,
and acts trivially away from a small neighbourhood of the zero-section. A short
argument also shows that it is a symplectomorphism of (T ∗ S 2 , η) which, up to
symplectic isotopy, is independent of the choice of ψ. τ is the model Dehn twist
of T ∗ S 2 with respect to its zero section.
If now (Y, ω) is a symplectic manifold containing a Lagrangian two-sphere S,
then by a theorem of Weinstein a neighbourhood of this two-sphere can be identified
with a neighbourhood of the zero-section in (T ∗ S 2 , η). Since the model Dehn twist
of T ∗ S 2 acts trivially outside a neighbourhood of the zero-section which can be
made arbitrarily small, there is a symplectomorphism τS of Y defined by the model
Dehn twist in a suitable neighbourhood of S.
Proposition 3.1. (Arnold) The symplectic monodromy of the family Y → ∆
deforming the surface node is generated by the Dehn twist [τS ] ∈ π0 (Symp(Y, ω)) in
the vanishing cycle S. The cohomology action of [τS ] is the map (τS )∗ : H ∗ (Y, Q) →
H ∗ (Y, Q) given by
(τS )∗ (α) = α + ([S] · α)[S]
which is a reflection since [S]2 = −2.
What is then the mirror of this symplectic monodromy transformation under
the map µ of (1)? According to our basic principle, it should be a Fourier–Mukai
functor associated to the contraction of E in X.
Proposition 3.2. (after Seidel–Thomas) Consider the structure sheaf F =
OX×X̄ X of the correspondence X ×X̄ X ⊂ X × X. This sheaf defines a Fourier–
Mukai equivalence
ΦF : Db (X) → Db (X)
with cohomology action ϕF : H ∗ (X, Q) → H ∗ (X, Q) given by
ϕF (α) = α + ([E] · α)[E]
which is a reflection since [E]2 = −2.
A plausible guess, which under a suitable choice of mirror map can indeed
be made precise, is that the mirror µ([τS ]) of the symplectic monodromy [τS ] is
the functor ΦF . Their cohomology actions are indeed easy to match. I wrote the
sheaf F in the above form to show its explicit dependence on the contraction, but in
fact it can be shown that the above definition is equivalent to the original definition
of [26] as a twist functor. Without going into the details, which can be found in
[26], I note that this particular transform is the (inverse) twist on X defined by the
spherical sheaf OE (−1).
3.2. The story for simple nodes can be generalized to other double point singu-
larities. In the local situation, mirror symmetry relates deformations and smooth-
ings of the arbitrary An surface singularity
{x21 + x22 + xn+1
3 = 0} ⊂ A3 .
As shown in [22], a smoothing Y of this singularity contains a collection S1 , . . . , Sn

of Lagrangian spheres, with a single transversal intersection point between Si
and Si+1 and no other intersections. There are Dehn twists in all these spheres,
which satisfy the relations of the braid group on n + 1 strings [22]. On the contrac-
tion side, X contains holomorphic spheres (rational curves) E1 , . . . , En , and has a
corresponding collection of derived self-equivalences which act by the braid group.
The actions on cohomology are given on both sides by reflections generating the
symmetric (Weyl) group. For details, consult [26].
4. Isolated threefold singularities

I now turn to birational contractions of three-dimensional Calabi–Yau varieties.
There are now three classes of possibilities, depending on the dimension of the
exceptional locus E ⊂ X and its image. In this section I consider cases where this
image is a point.
4.1. The first possibility to consider is when E is one-dimensional, and hence
necessarily rational. The simplest case is when E ∼
= P1 with normal bundle NE/X ∼ =
OP1 (−1, −1), the (−1, −1)-curve. The contraction of such a curve leads to a three-
fold X̄ with an ordinary double point singularity
{x21 + x22 + x23 + x24 = 0} ⊂ A4 .
This contraction is the first half of the conifold transition in physics, which is
considered to be self-mirror [17]. Hence its degeneration mirror should be the
other half of the conifold transition, the smoothing Y → ∆∗ of the node locally
given by
{x21 + x22 + x23 + x24 = t} ⊂ A5 .
For fixed t positive real, this local geometry contains a Lagrangian three-sphere
given by xi ∈ R, which is again the vanishing cycle and is Lagrangian. Globally
we have a symplectic Calabi–Yau threefold (Y, ω) containing a Lagrangian S 3 , and
the situation is analogous to the surface case:
Proposition 4.1. (Arnold) The symplectic monodromy of the smoothing of the
threefold node is given by a Dehn twist [τS ] in the vanishing cycle S
S 3 defined
exactly as in the surface case. Its cohomology action is
(τS )∗ (α) = α + ([S] · α)[S];
as [S]2 = 0, this is a map of infinite order.
One plausible mirror of this symplectic monodromy map is given by
Proposition 4.2. (Seidel–Thomas) Consider a (−1, −1)-curve E in a Calabi–
Yau manifold X. Then its structure sheaf OE defines a spherical sheaf, and there
is a corresponding twist functor
TOE ∈ AutEq(Db (X))
with cohomology action given by
α → α + (ch(OE ) · α)ch(OE ).
One notable feature of this pair of symmetries is their cohomology action. As

the formulae show, (τS )∗ acts trivially on even cohomology as [S] is in degree three;
mirror to this, TOE acts trivially on odd cohomology as ch(OE ) lives in even degree.
As mentioned in Section 1, this implies that TOE acts trivially on the deformation
space of X. Thus this Fourier–Mukai functor must exist on all (small) deformations
of X; indeed, contractions of (−1, −1)-curves are known to have this property [33].
I leave the world of the simple node, though of course there would be much
more to say regarding the issue of multiple nodes and flops. I leave these to more
able hands; [17, 26] and [27] respectively have all the details.
4.2. The mirror pairs considered so far can be encoded in toric geometry, which
suggests generalizations. Namely, consider a convex polytope Π with vertices in a
lattice Zn−1 . Embed Zn−1 ⊂ Zn as the affine hyperplane with last coordinate 1.
Let N be the sublattice of Zn spanned by the origin and the vertices of Π, and let τ
be the cone over Π. Then the toric variety X̄ = XN,τ (in the covariant description
[8]) is an affine n-fold with a canonical Gorenstein singularity at the origin, which
in dimension at most three has toric crepant resolution(s) X → X̄.
On the other hand, the polytope Π can be thought of as the Newton polytope
of a polynomial f in variables xi , x−1
i for i = 1, . . . , n − 1. Adding two auxiliary
variables u, v, one obtains a family of affine n-folds
Y = {u2 + v 2 + f (xi , x−1
i ) = 0} ⊂ (A \ {0})
n−1
× A2 × T
over the base T defined by the moduli of the polynomial f . The complex structure
of this variety will degenerate for particular f , giving rise to a boundary locus T \T 0
and symplectic monodromy. Toric geometry suggests that this family is the mirror
of the contraction X → X̄ = XN,τ .
It is a simple matter to see that the surface An singularity is the case when
P = [0, n + 1] is 1-dimensional, whereas the threefold node arises from the two-
dimensional polytope P = [0, 1] × [0, 1].
4.3. Examples of birational maps contracting a surface E ⊂ X to a point
also arise in this way. For example, consider the polytope in R2 with vertices at
(−1, −1), (1, 0) and (0, 1). The contraction X → X̄ = XN,τ contracts a projective
plane P2 ⊂ X to a Gorenstein singularity P ∈ X̄. The mirror of this singularity is
the family of threefolds
Y = {a0 + a1 x1 + a2 x2 + a3 x−1 −1
1 x2 + u + v = 0} ⊂ (A \ {0}) × A × T
2 2 2 2
over a base T ⊂ A4 with coordinates ai . The symplectic geometry of this family is

studied in detail in [25, Proposition 3.2].
4.4. The list can be continued with various two-dimensional polytopes. One
interesting example is the following: consider a Calabi–Yau threefold Y0 containing
a contractible surface E ⊂ Y0 abstractly isomorphic to P2 blown up in three points.
Then E 3 = (KE )2 = 6 and after contraction, Y0 contains a toric Gorenstein singu-
larity (P ∈ Y0 ) corresponding to the polytope Π with vertices at (−1, −1), (0, −1),
(1, 0), (1, 1), (0, 1) and (−1, 0). The mirror to this singularity is the family
X = {u2 + v 2 + f (xi , x−1
i ) = 0} ⊂ A × (A \ {0}) × T
2 2
defined by polynomials f with Newton polytope Π. It might be interesting to study

the symplectic geometry of degenerations in this family in more detail.
On the other hand, the deformation theory of (P ∈ Y0 ) is quite interesting: by

[1, 2.1 and 8.4], the local first-order deformation space is three-dimensional, but
only a one- and a two-dimensional subspace can be realized as actual deformations,
the two components corresponding to two essentially different ways in which Π de-
composes into the Minkowski sum of two polytopes Π1 and Π2 . There are therefore
(j)
two different (not even diffeomorphic) symplectic smoothings Ys here. On the
other side, as Mark Gross points out, for every decomposition Π = Π1 + Π2 into
Minkowski summands, there are degenerate members
X̄ (j) = {u2 + v 2 + f1 (xi , x−1 −1
i )f2 (xi , xi ) = 0} ⊂ A × (A \ {0}) ,
2 2
in the family X , where fi has Newton polytope Πi . These varieties are singular,
with nodes at the points where u = v = f1 = f2 = 0, and have small resolutions
(j)
X (j) → X̄ (j) mirroring the two symplectic smoothings Ys .
4.5. Beyond toric cases, one example discussed in the literature [28] concerns,
on the symplectic side, the degeneration to the simple threefold triple point
{x31 + x32 + x33 + x34 = 0} ⊂ A4 .
Denote as usual by (Y, ω) a symplectic smoothing.
Proposition 4.3. (Smith–Thomas) The symplectic manifold (Y, ω) contains a
collection of 16 Lagrangian three-spheres S1 , . . . , S16 meeting in an intricate config-
uration depicted on [28, Figure 2]. In particular, there are 16 corresponding Dehn
twists
τSi ∈ π0 (Symp(Y, ω)).
As far as I know, the mirror of this singularity has not been discussed in the
literature. It should have 16 Fourier–Mukai transforms corresponding to the Dehn
twists of Proposition 4.3.
5. Non-isolated threefold singularities

The final case left out so far is that of a contraction (E ⊂ X) → (C ⊂ X̄)
with dim C = 1. Assuming that X is smooth, it follows that the curve C is also
smooth [33].
5.1. Start with the simplest case: a projective Calabi–Yau variety X̄, smooth
outside of a smooth curve C ⊂ X̄ of genus g, locally analytically isomorphic to
{x21 + x22 + x23 = 0} ⊂ A4
along C (which is locally the line A1x4 ). Blowing up the ideal of C gives a Calabi–
Yau resolution X → X̄ containing an exceptional divisor E geometrically ruled
over C.
Proposition 5.1. (Horja, Szendrői) The structure sheaf F = OX×X̄ X on the
product X × X gives a Fourier–Mukai self-equivalence
ΦF : Db (X) → Db (X).
The corresponding cohomology action maps
i. a class α ∈ H 2 (X, Q) as
ϕF (α) = α + ([l] · α)[E] mod H 4 (X, Q)
with [l] ∈ H 4 (X, Q) the class of the ruling of E (this is a reflection since
[l] · [E] = −2);
ii. on third cohomology as an involution with a codimension 2g fixed locus
and (−1)-eigenspace given by the image of the cylinder homomorphism
H 1 (C, Q) → H 3 (X, Q) (note that there is no odd cohomology in different
degrees; see [29, Proposition 4.6] for more details).
The problem is then clear: find the symplectic mirror of this contraction, to-
gether with its symplectic monodromy corresponding to ΦF .
5.2. Before I move on, I discuss a slight generalization, which will be just as easy
(or difficult) to study: suppose that X̄ contains a curve of An -singularities C ⊂ X̄
locally of the form
{x21 + x22 + xn+13 = 0} ⊂ A4 .
Assume also that repeated blowup leads eventually to a smooth Calabi–Yau reso-
lution X containing a collection E1 , . . . , En of smooth ruled surfaces. Then there is
a Fourier–Mukai equivalence ΦFi ∈ AutEq(Db (X)) for each of the surfaces, which
as a matter of fact satisfy the relations of the braid group [32] just as in the surface
case.
5.3. To approach the problem of finding the mirror, I recall the discussion of the
above setup in the toric context from [15, Section 3.2]. Assume therefore that X̄ is
in fact a hypersurface in a toric variety P∆ given by a reflexive polytope ∆ spanned
by some lattice points in a lattice M ∼ = Z4 ; compare [4]. The singularities of X̄
are most easily studied in terms of the normal fan Σ consisting of cones spanned
by vertices of the dual polytope ∆◦ in the dual N = HomZ (M, Z). The specific
singularity along C then arises from an edge τ = v0 , vn+1 of ∆◦ containing n
interior lattice points v1 , . . . , vn . The genus g can also be read off from the toric
data: it is the number of interior lattice points in the dual two-dimensional face τ ◦ ⊂
∆ ⊂ M.
The mirror Y of X is given, by Batyrev’s mirror duality [4], as a hypersurface
in a partial toric resolution P ∆◦ of the toric variety P∆◦ . The parameter space of
the hypersurface Y is the vector space

H0 P ∆◦ , −K ∼
= C xm .
P∆◦
m∈∆◦ ∩N

m
Here x represents a monomial xm
i ,
i
where xi are to be thought of as affine
∗ 4
variables in the affine torus (C ) ⊂ P∆◦ , and Y is the closure in P ∆◦ of the subva-
riety of the torus given by the chosen Laurent polynomial. By [6], this vector space
can also be viewed dually as the appropriately graded piece of the homogeneous
polynomial ring S of the abstract toric variety P ∆◦ , with one generating variable yη
for every lattice point η ∈ N on the boundary of the polytope ∆.
Now return to our concrete situation, restricting attention to g = 1, the reasons
for which will appear presently; let η1 ∈ M be the unique lattice point in the
interior of the face τ ◦ ⊂ ∆ with corresponding homogeneous coordinate y1 . The
hypersurface Y is given by a homogeneous equation
{y1 f (yj ) + g(yj ) = 0} ⊂ P∆◦
where I have separated out the monomials not involving the variable y1 . It can
however be checked, using the correspondence between Laurent and homogeneous
polynomials [6], that the only terms not involving the monomial y1 correspond to
Laurent monomials xm with m ∈ τ , one of the n + 2 lattice points responsible
for the singularity of X̄. The linear relations between these lattice points translate
to multiplicative relations between the Laurent monomials, which implies that the
equation of Y can be written on a suitable affine piece of P ∆◦ as

n+1
y1 f + a j xj = 0
j=0
where x is an auxiliary affine variable. Moreover, the contraction of X back to X̄

corresponds to the degeneration to the hypersurface with equation
y1 f + (b1 x + b0 )n+1 = 0
which (assuming appropriate regularity of f ) is a threefold exactly of the studied
type: singular along the curve {y1 = f = b1 x + b0 = 0}, with a curve of An+1
singularities.
This is then the correspondence suggested by toric geometry: the mirror to a
contraction of a collection of ruled surfaces to an elliptic curve of An singularities
should be a degeneration to a single curve of An singularities.
5.4. Before moving on to more theory, it might be illustrative to give an ex-
ample, appearing as [12, Example II]. Let
= {m ∈ M ∼
∆ = Z4 : mi ≥ −1, 1 ≥ m1 + 2m2 + 2m3 + 2m4 }
with dual polytope ∆ ◦ spanned by e1 , e2 , e3 , e4 and v0 = −e1 − 2e2 − 2e3 − 2e4
= ⊕Zei . The toric variety P is simply weighted projective space P4 [12 , 23 ],
in N ∆
containing the family of octic Calabi–Yau hypersurfaces. The edge v0 , e1 contains
one interior lattice point, giving rise to a curve of A1 singularities in the general
octic. This curve has genus 3.
Now consider a quotient of this family; let

N =N + 1 , 1 , 0, 1 Z + 0, 0, 1 , 3 Z
4 4 2 4 4
with ∆◦ = ∆ ◦ now thought of as a polytope in the lattice N . It can be checked
◦
that ∆ is still reflexive, and now v0 , e1 contains seven interior lattice points, indi-
cating singularities of type A7 along a curve. The dual to (N, ∆◦ ) is the pair (M, ∆),
where
M =M + 0, 1 , 1 , 1 Z
2 4 4

and ∆ = ∆ as a polytope with vertices in M . The mirror two-dimensional face
to v0 , e1 looks like the toric diagram for a 14 (1, 1, 2) singularity, with one interior
lattice point. Hence the generic threefold X̄ ⊂ P∆ has a genus-1 curve of A7
singularities, so falls under the rubric of the above discussion.
5.5. I now consider the symplectic geometry of the mirror (Y, ω) smoothing
the hypersurface Y0 . The monodromy of the degeneration found above has in fact
been studied already by Seidel [22].
Definition/Proposition 5.2. (Seidel) Let (M 2n , ω) be a symplectic manifold.

A set of construction data for a generalized Dehn twist on M consists of a compact
symplectic manifold (M , ω ) of dimension 2n − 2r, a fibre bundle p : S → M with
fibre S r and structure group O(r + 1), and an embedding i : S
→ M such that
i∗ (ω) = p∗ (ω ). Given a set of such data, there is a well-defined generalized Dehn
twist
[τS ] ∈ π0 (Symp(M, ω)).
Dehn twists as discussed in previous sections correspond to the case M = point.
The general idea should now be clear:
Conjecture 5.3. Let X → X̄ be a contraction of ruled surfaces E1 , . . . , En to
an elliptic curve C ⊂ X̄ of An -singularities on a Calabi–Yau threefold. Then the
symplectic mirror (Y, ω) of X contains a collection pi : Si → Σ of S 2 -bundles over
a symplectic Riemann surface (Σ, ω ), providing a set of data for generalized Dehn
twists
[τSi ] ∈ π0 (Symp(Y, ω)).
These symplectomorphisms satisfy the relations of the braid group on n + 1 strings
in the obvious way. Their cohomology actions are given for α ∈ H 2 (Y, Q) by
(τSi )∗ (α) = α + ([li ] · α)[Si ]
with [li ] ∈ H 4 (Y, Q) the class of the fibre of pi ; (τSi )∗ is a reflection since
[li ] · [Si ] = −2. These symplectic automorphisms mirror the action of the Fourier–
Mukai transforms of Proposition 5.1.
Note that the cohomology action is formally identical to that of Proposi-
tion 5.1.i. However, remembering that we are in the threefold case, the more
important point is the relation to the action in Proposition 5.1.ii, which brings
me back to the g = 1 issue. Note that in Proposition 5.1.i, the fixed locus is of
codimension 2g. Hence for a simple formula like that in Conjecture 5.3 to hold,
with a codimension-one fixed locus, we need to be in the case g = 1 (the other half
of the fixed locus in the symplectic case is in H 4 by Poincaré duality).
Geometrically, it is abundantly clear where the fibred manifolds Si come from,
if the sketched degeneration picture is correct. The base Σ of the fibration is the
singular locus of the degenerate variety Y0 . The S 2 sphere fibres are the vanishing
cycles in the local An degeneration transverse to that curve.
5.6. In the toric argument, the g = 1 assumption manifested itself in the
presence of a single coordinate y1 which can be pulled out of the equation of Y .
In the case of higher genus, there are several such coordinates, and the picture
suggested by [15] is that of a degeneration to a reducible curve of An singularities.
These curves will however begin to intersect, necessarily producing singularities
worse than An , and the geometric picture is not so clear any more. The discussion of
the cohomology action also suggests a more complicated symplectic automorphism.
It might be worthwhile to study this case in more detail.
Incidentally, the g = 0 case is also of some interest. In that case the cohomology
action of the Fourier–Mukai functors of Proposition 5.1 in odd degree is trivial.
Mirror to that, I expect a symplectomorphism induced by a submanifold S ⊂ (Y, ω)
generically fibred in spheres by p : S → Σ, so that the spheres collapse over special
points P ∈ Σ making their cohomology class trivial.
5.7. Return to the example of 5.4. This is discussed in [12] as Example II,
where a pair of (special) Lagrangian cycles S 1 × S 2
Ni ⊂ (Y, ω) is constructed as
the fixed locus of a real involution. If Conjecture 5.3 holds, the natural guess is that
in fact Ni = p−1
4 (Bi ) for a pair of Lagrangian circles S
Bi ⊂ (Σ, ω ); remember
1
that the singularity is of type A7 and the construction using real variables in the
complex A7 equation gives the middle vanishing 2-cycle. The one-dimensional local
moduli space of Ni as a special Lagrangian cycle is geometrically realized then as
coming simply from moving the circle Bi locally in Σ. Conjecture 5.3 would imply
that there is a host of other Lagrangian cycles around, though their realization
as special Lagrangians is bound to run into the usual problem of finding sLag
representatives of vanishing cycles.
As for the other two examples of [12], Example I involves the original octic
as X, with a contraction to a g = 3 curve. Its mirror contains a complicated
(special) Lagrangian cycle N with b2 (N ) = 5. As I discussed above, I expect this
case to be quite complicated. [12, Example III] involves on the complex side a
contraction to a g = 0 curve, and again S 1 × S 2 in the mirror; this would arise
naturally from the g = 0 speculation at the end of 5.6.
5.8. To conclude, I want to return to one point which was swept under the
carpet above. Namely, just because a threefold has a curve of An singularities, it
does not follow that in its resolution one finds n irreducible surfaces all ruled over
the same curve. This is an issue of monodromy (in a different sense now, over the
curve C), which is discussed in detail in [31, 32]. I only want to point out that [31,
Example 4.3] constructs an example of a threefold X̄ with an elliptic curve of A3
singularities, where in the resolution there are only two irreducible surfaces E1 , E2 ⊂
X. There are two corresponding Fourier–Mukai functors Φ1 , Φ2 ∈ AutEq(Db (X)),
which satisfy the braid relation
Φ1 ◦ Φ 2 ◦ Φ 1 ◦ Φ 2 = Φ2 ◦ Φ 1 ◦ Φ 2 ◦ Φ 1
of the C3 braid group. The analogue of Conjecture 5.3 would suggest that the
mirror (Y, ω) of X should contain a pair of S 2 -fibred submanifolds together with the
necessary symplectic data, giving rise to a pair of Dehn twists [τi ] ∈ π0 (Symp(Y, ω))
which satisfy the same relation
τ1 ◦ τ2 ◦ τ1 ◦ τ2 = τ2 ◦ τ1 ◦ τ2 ◦ τ1
up to symplectic isotopy. I leave the problem of filling in details as a final challenge
for you, my Dear Reader.
References
1. K. Altmann, The versal deformation of an isolated toric Gorenstein singularity, Inv. Math.
128 (1997), 443–479.
2. V. Arnold, Some remarks on symplectic monodromy of Milnor fibrations, The Floer memorial
volume (ed. Hofer et al), 99–103, Progr. Math., 133, Birkhauser, Basel, 1995.
3. P. Aspinwall and M.R. Douglas, D-brane stability and monodromy, JHEP 0205 (2002) 031.
4. V.V. Batyrev, Dual polyhedra and mirror symmetry for Calabi-Yau hypersurfaces in toric
varieties, J. Alg. Geom. 3 (1994) 493–535.
5. T. Bridgeland, Stability conditions on triangulated categories, math.AG/0212237.
6. D.A. Cox, The homogeneous coordinate ring of a toric variety, J. Alg. Geom. 4 (1995) 17-50.
7. R. Dijkgraaf et al (eds), Mirror Symmetry at the Clay School in Cambridge, in preparation
to be in preparation.
8. W. Fulton, Introduction to toric varieties, Princeton University Press, Princeton, 1993.
9. S.I. Gelfand and Y.I. Manin, Methods of homological algebra, Springer, Berlin, 1996.
10. P. Horja, Hypergeometric functions and mirror symmetry in toric varieties,
math.AG/9912109.
11. P. Horja, Derived category automorphisms from mirror symmetry, math.AG/0103231.
12. S. Kachru, S. Katz, A. Lawrence and J. McGreevy, Mirror symmetry for open strings,
Phys. Rev. D (3) 62 (2000).
13. A. Kapustin and Y. Li, D-Branes in Landau-Ginzburg models and algebraic geometry, hep-
th/0210296.
14. M. Kontsevich, Homological algebra of mirror symmetry, in: Proceedings of the International
Congress of Mathematicians, (Zurich, 1994, ed. S.D. Chatterji), 120–139, Birkhauser, Basel,
1995.
15. S. Katz, D.R. Morrison and R. Plesser, Enhanced gauge symmetry in type II string theory,
Nucl. Phys. B 477 (1996), 105–140.
16. Y.I. Manin, Moduli, motives, mirrors, in: European Congress of Mathematics, Vol. I
(Barcelona, 2000), 53–73, Progr. Math., 201, Birkhuser, Basel, 2001.
17. D.R. Morrison, Through the looking glass, Mirror Symmetry III (D. H. Phong, L. Vinet, and
S.-T. Yau, eds.) 263-277, American Mathematical Society and International Press 1999.
18. D.R. Morrison, Compactifications of moduli spaces inspired by mirror symmetry, Journées
de Géométrie Algébrique d’Orsay (Orsay, 1992), Astérisque 218 (1993), 243–271.
19. S. Mukai, Duality between D(X) and D(X̂) with its application to Picard sheaves, Nagoya
Math. J. 81 (1981), 153–175.
20. D. Orlov, Triangulated categories of singularities and D-branes in Landau–Ginzburg models,
math.AG/0302304.
21. M. Reid, Canonical 3-folds, Journées de Géométrie Algébrique d’Angers (Angers, 1979, ed.
A. Beauville), 273–310, Sijthoff and Noordhoff, Alphen aan den Rijn—Germantown, Md.,
1980.
22. P. Seidel, Lagrangian two-spheres can be symplectically knotted, J. Diff. Geom. 52 (1999),
145–171.
23. P. Seidel, Symplectic monodromy, manuscript 1996.
24. P. Seidel, Vanishing cycles and mutations, European Congress of Mathematics, Vol. II
(Barcelona, 2000), 65–85, Progr. Math., 202, Birkhuser, Basel, 2001.
25. P. Seidel, More about vanishing cycles and mutation, Symplectic geometry and mirror sym-
metry (Seoul, 2000), 429–465, World Sci. Publishing, River Edge, NJ, 2001.
26. P. Seidel and R. Thomas, Braid group actions on derived categories of coherent sheaves,
Duke Math. J. 108 (2001), 37–108.
27. I. Smith, R. Thomas and S.-T.Yau, Symplectic conifold transitions, J. Diff. Geom. 62 (2002)
209-242.
28. I. Smith and R. Thomas, Symplectic surgeries from singularities, Turkish J. Math. 27 (2003)
231-250.
29. B. Szendrői, Calabi–Yau threefolds with a curve of singularities and counterexamples to the
Torelli problem, Int. J. Math. 11 (2000), 449–459.
30. B. Szendrői, Diffeomorphisms and families of Fourier–Mukai transforms in mirror symme-
try, in: Applications of algebraic geometry to coding theory, physics and computation (Eilat,
2001), 317–337, NATO Sci. Ser. II Math. Phys. Chem. 36, Kluwer, Dordrecht, 2001.
31. B. Szendrői, Enhanced gauge symmetry and generalized braid group actions, Comm. Math.
Phys. 238 (2003) 35-51.
32. B. Szendrői, Artin group actions on derived categories of threefolds, math.AG/0210121,
submitted.
33. P.M.H. Wilson, The Kähler cone on Calabi–Yau threefolds, Inv. Math. 107 (1992), 561-583.
Erratum, Inv. Math. 114 (1993), 231–233.
Department of Mathematics, Utrecht University, PO. Box 80010, NL-3508 TA Utrecht,

The Netherlands
E-mail address: szendroi@math.uu.nl
Volume 3, 2004
Lectures on Supersymmetric Gauge Theory
Nick Dorey
Abstract. These lectures give an informal introduction to recent develop-

ments in supersymmetric (SUSY) gauge theories. In particular, we describe
how heuristic ideas about the strongly-coupled vacuum of QCD such as electric-
magnetic duality and large N string theory have found a precise realization in
the context of SUSY gauge theory.
1. Introduction
Since its discovery almost 30 years ago, supersymmetry (SUSY) has become
one of the key theoretical ideas in contemporary high-energy physics. Although
there is, as yet, no direct experimental evidence for SUSY there are a number of
related reasons why it remains close to theorists’ hearts. One general motivation
is based on the central role that symmetry principles have played so far in the
development of the subject. Given this historical perspective, looking for theories
with the largest possible symmetry groups seems like a sensible strategy. Under
certain fairly general assumptions, supersymmetry is the only possible extension of
the known symmetries of particle physics and this fact alone makes it worthy of
study. There are also (at least) three more specific motivations for SUSY. Two of
these we will only mention in passing as they will not concern us directly in these
lectures:
• Supersymmetry offers a potential solution to the hierarchy problem of the
Standard Model. In particular, it provides a natural way of keeping scalar particles
light by pairing them with fermions whose masses in turn are protected by (ap-
proximate) chiral symmetries. For this reason SUSY is an important ingredient in
many possible extensions of the Standard Model.
• Supersymmetry plays a key role in string theory. Indeed, spacetime super-
symmetry, in the form of the GSO projection, seems to be necessary for removing
the tachyon from the spectrum and ensuring a stable ground state for the string.
Either of these subjects would require a lecture course of its own. And so we
come to the main subject of these lectures, which is lessons that supersymmetry
can teach us about the dynamics of non-Abelian gauge theories in four dimen-
sions. These gauge theories are, of course, the cornerstone of the standard model.
However, despite their long familiarity, we still understand very little about them
2000 Mathematics Subject Classification. 81T60, 81T13.
2004
315
316 NICK DOREY
outside the regime of perturbation theory. The problem is particularly pressing in

QCD-like theories with asymptotic freedom, which flow to strong coupling in the
infrared (IR). In particular, the strongly-coupled dynamics in the IR are responsible
for the enduring mystery of quark confinement. After many years of effort, we have
yet to find analytic methods which can provide quantitative understanding of the
confining phase. In the absence of such methods, progress has been made only by
resorting to lattice simulations.
Although QCD itself remains as intractable as ever, theorists have often at-
tempted to study the key physical phenomena such as quark confinement and chiral
symmetry breaking in the context of simpler models. For example, in the 1970’s
there were many theoretical studies of field theories in two spacetime dimensions
such as the Schwinger model and the Gross-Neveu model, which have some QCD-
like features. Recent progress in this direction has come from the realization that
supersymmetric non-Abelian gauge theories in four dimensions provide a much more
realistic theoretical laboratory. Supersymmetry has the effect of constraining the
form of quantum corrections while still allowing pheneomena of physical interest
including confinement. As we will review below, interesting qualitative ideas about
QCD and confinement such as electric-magnetic duality and the large N string,
which remain heuristic in the physical case, can be implemented in a very precise
way in the context of SUSY theories. The reader may justifiably ask what super-
symmetry has to do with QCD and the strong interactions. At present there is no
satisfactory answer to this question. There is instead the hope that some of the
lessons we have learnt, particularly about the physics of the large N limit, will be
applicable also in the non-supersymmetric case.
In this introdution, we will briefly review two old ideas about confinement in
QCD which have recently taken on a much more quantitative form in the super-
symmetric context. The first idea is the possibility of a duality between electric and
magnetic charges. This idea has its genesis in the obvious symmetry of Maxwell’s
equations in the vacuum: E → B, B → −E. In the real world this symmetry
is of course broken by the asymmetry between electric charges which are common
place and magnetic charges which are yet to be seen in Nature. Despite this, the
existence of magnetic charges remains an important theoretical possibility. As we
will discuss below, non-Abelian gauge theories often have a Coulomb phase where
the unbroken gauge group includes an Abelian factor. This is true, for example,
of many candidate GUT theories which embed the U (1) of electromagnetism in-
side a simple group. These theories often contain classical solitons which carry the
corresponding magnetic charge. At weak coupling, these states are very massive
(possibly explaining why we have never seen them). On the other hand, the clas-
sical mass formula suggests that they may become light at strong coupling, and
many authors have speculated on their potential importance for strong coupling
phenomena like confinement.
In any case, non-Abelian gauge theories in their Coulomb phase often have
both electric and magnetically charged particles in their spectrum and, in some ex-
amples, they appear symmetrically in a way which suggests that the trivial electric-
magnetic symmetry of Maxwell’s equations in vacuum can be extended to include
charges. This ‘symmetry’ exchanges electrically charged elementary particles with
magnetically charged solitons. Because of the Dirac quantization condition (see
Section 3 below), it also necessarily exchanges weak coupling, where the electric
LECTURES ON SUPERSYMMETRIC GAUGE THEORY 317
degrees of freedom are light, with strong coupling, where the magnetic degrees of
freedom are light. An equivalence between theories with different values of the
coupling constant is known as a duality. In some special cases, of which the best
understood is N = 4 supersymmetric Yang-Mills theory, it is now believed that
this electric-magnetic duality is an exact property of the theory. If true, this gives
us a precise description of the strong coupling regime which was previously out of
reach. Of course, to test such a duality one needs an independent calculation of
some quantity in the strongly coupled theory.
To understand the possible relevance of electric-magnetic duality to confine-
ment, we must extend our discussion of gauge theory from the Coulomb phase to the
Higgs phase. Specifically, we can think about what happens to magnetic monopoles
when we spontaeously break the U (1) of electromagnetism. The Higgs mechanism
occurs when a massless electrically-charged scalar field acquires a vacuum expec-
tation value. In other words light electrically-charged particles form a condensate
in the vacuum. When this occurs, magnetic fields are screened and cannot pen-
etrate the vacuum. This is a precise relativistic analog of the Meissner effect in
superconductors, where the Cooper pairs of electrons form an electrically-charged
condensate and magnetic fields are excluded from the superconducting sample. In
this case also, magnetic flux lines cannot spread out and it is energetically preferred
for them to form flux tubes or strings of finite tension. The only configurations of
finite energy are those where a monopole and anti-monopole are joined by such a
flux tube. Thus magnetic charges are confined by a linear potential in the Higgs
phase.
So far we have described the consequences of the Higgs mechanism for our
conjectural magnetic charges. Of course we are really interested in the confinement
of elementary particles which carry ordinary electric charges or their non-Abelian
analog. However, in a theory with electric-magnetic duality, the two things should
be related. Duality indicates that magnetic charges can become light at strong
coupling. In this regime, it is plausible that they can be encouraged (by the intro-
duction of a suitable potential) to condense in the vacuum. In this dual Higgs phase
where the vacuum contains a magnetic condensate, electric charges are confined.
Thus the elusive confining phase of gauge theory is mapped onto the Higgs phase,
which we can understand easily within conventional perturbation theory.
There are several obvious objections to the idea described above: as stated it
really only applies to confinement of Abelian electric charge. One may ask how it is
expected to work in the context of QCD where there are no stable monopole solu-
tions and the coupling constant inverted by duality is replaced by the dimensionful
parameter ΛQCD . Despite this, there has been considerable progress in realizing
these ideas quantitatively in supersymmetric gauge theory. First, there is now
substantial evidence that electric-magnetic duality is an exact property of N = 4
supersymmetric Yang-Mills theory. The famous Seiberg-Witten solution [19] of
N = 2 supersymmetric Yang-Mills theory shows explicitly that monopoles can be-
come massless at strong coupling in an asymptotically free gauge theory. Further
soft breaking to N = 1 explicitly exhibits confinement due to monopole conden-
sation. Further, a quantitative duality between the Higgs and confining phases of
N = 1 theories is now well established. We will review some of these developments
in the the following.
318 NICK DOREY
The second theoretical idea for the 1970’s which has gained renewed impor-
tance in the supersymmetric context is that of the large N expansion due to Gerard
’t Hooft [1]. As mentioned above, because of dimensional transmutation, QCD lacks
a dimensionless coupling in which to expand. However, a parameter can be intro-
duced by replacing the three colours of QCD with an arbitrary number N or, in
other words, working with gauge group SU (N ). Following a strategy used earlier in
statistical mechanics, one may then look for simplifications in the N → ∞ limit and
attempt to expand around this limit in powers of 1/N . To illustrate how this idea
is implemented in practice, we consider vacuum diagrams in pure SU (N ) gauge
theory with bare coupling constant g. The correct limit to think about for any
non-Abelian gauge theory is one in which we take N → ∞ and g 2 → 0 holding
the combination λ = g 2 N (known as the ’t Hooft coupling) fixed. In this limit a
vacuum diagram with E propagators, V vertices and F closed loops scales like N χ
where χ = V − E + F . If we consider the Feynman diagram as a triangulation
of the two-dimensional surface on which it can be drawn without internal lines
crossing, then χ = V − E + F is precisely the Euler characteristic of that surface.
All compact two-dimensional surfaces are characterized up to homeomorphism by
their number of handles, the genus G, which is related to the Euler characteristic
by χ = 2 − 2G. The 1/N expansion is therefore an expansion in the number of
handles of a two-dimensional surface.
This beautiful result means that the leading behavior at large N is governed
by planar diagrams which can be drawn on a sphere without lines crossing. Planar
diagrams can have an arbitrary number of loops and calculating their sum is a
formidable problem which remains unsolved, except in certain lower-dimensional
models. However the complexity of the leading order reflects the fact that large N
QCD retains many of the features which are hard to analyse directly in the N = 3
case. In particular, large N QCD has asymptotic freedom in the UV and quark
confinement in the IR1. Despite the complexity there are tantalizing indications
that the problem of summing the planar diagrams should, ultimately, be tractable.
In particular, Witten has argued that the leading order at large N is essentially
semi-classical in the sense that the path integral is dominated by a single field
configuration known as the ‘master field’. A useful review of this and other aspects
of the large N limit is [2]. Another indication lies in the topological nature of the
1/N expansion described above. In particular, as an expansion over 2D surfaces of
different topology, it is similar to string perturbation expansion with string coupling
constant proportional to 1/N . A long-standing conjecture is that large N QCD is
dual to a weakly-coupled string theory: the QCD string. If this were true, the
large N hadron spectrum could be descibed as the (tree-level) spectrum of a string,
confirming other indicators of stringy behaviour in the strong interactions, such as
Regge behaviour.
The conjecture described above has been around for over 20 years. The key
question, “which string theory describes QCD?” has so far remained unanswered.
Obvious problems include the fact that the only weakly-coupled string theories
we know about are the critical string theories which live in ten dimensions rather
then four. These theories inevitably contain a massless spin two particle, identified
with the graviton in the context of fundamental strings. Despite these objections
1Or at least, the resons why we believe ordinary QCD is confining, such as the area law for
the Wilson loop in lattice QCD, remain valid at large N .
significant progress has been made in the more controlled context of supersymmetric
gauge theories. For the first time a precise proposal for the large N string dual to
a four-dimensional non-Abelian gauge theory has emerged. As in our discussion of
electric-magnetic duality above, the maximally supersymmetric case is understood
best. Maldacena [8] has conjectured that the large N dual of N = 4 SUSY Yang-
Mills theory with gauge group SU (N ) is the IIB superstring on the spacetime
manifold AdS5 × S 5 . Note that this is a theory which lives in ten dimensions and
has a massless graviton. As we will describe in the following, this reflects some
rather special features of the N = 4 theory. Cases with less supersymmetry have
also been analysed successfully, although the QCD string itself remains elusive.
We will see below that, like electric-magnetic duality, Maldacena’s conjecture
relates the strong coupling behaviour of one theory to the weak coupling behaviour
of another. Testing any such conjecture clearly requires us to calculate some quan-
tity in both theories and compare. In particular this requires us to perform an
independent calculation on the strongly-coupled side of the correspondence. For-
tunately, this is where supersymmetry can help. In supersymmetric theories some
quantities are protected against quantum corrections and can be reliably extrap-
olated to strong coupling. Such quantities can often provide quantitative tests of
strong-weak coupling duality. In the next section we will give an overview of this
aspect of SUSY.
The remainder of these notes are organized as follows. In Section 2 we give
a brief overview of protected quantities in SUSY gauge theory. In Section 3 we
consider N = 4 supersymmetric Yang-Mills theory at the classical level. In the
following two sections we consider the quantum theory in its Coulomb and confor-
mal phases respectively. This naturally includes a discussion of S-duality and the
AdS/CFT correspondence. Section 5 is devoted to a brief sketch of the Seiberg-
Witten solution of N = 2 SUSY Yang-Mills theory. Finally, we give our conclusions
and a brief bibliography.
2. Supersymmetry and Exact Results

In this section, we will briefly review of some of the special properties of super-
symmetric theories which allow exact statements to be made about their dynamics.
Some of these features are present even in the simplest possible model: supersym-
metric quantum mechanics with a single (complex) supercharge Q. In this case the
SUSY algebra takes the simple form
(2.1) {Q , Q† } = H
where H is the Hamiltonian. We also introduce a conserved (modulo-two) fermion
number operator (−1)F which anti-commutes with Q. Bose and Fermi eigenstates
have eigenvalues ±1 respectively. One immediate consequence of the SUSY algebra
is that the energy of all states is bounded below by zero: E ≥ 0. The bound is only
saturated for states which are annihilated by Q. It follows that the groundstates of
the system have zero energy and are invariant under supersymmetry. In contrast,
Q acts non-trivially on states of non-zero energy and it follows immediately that
these states necessarily come in degenerate Bose-Fermi pairs.
States in the Hilbert space therefore come in two different ‘multiplets’ of su-
persymmetry depending on whether their energies saturate the bound E ≥ 0. This
is a characteristic feature of supersymmetry which we will meet again. Another
320 NICK DOREY
such feature is the existence of quantities which are protected from quantum cor-
rections. So far we have met the groundstate energy which is strictly zero. A more
sophisticated example is the Witten Index, defined by,
(2.2) I = Tr (−1)F exp (−βH) .
Because states of non-zero energy come in Bose-Fermi pairs their contribution to
the trace cancels. Thus we have I = nB − nF , where nB and nF are the numbers of
Bose and Fermi groundstates respectively. As we vary the parameters of the model
groundstates can only be lifted to non-zero energy in Bose-Fermi pairs. Similarly
new groundstates can only appear if a Bose-Fermi pair descends to zero energy.
Neither of these processes change the value of I, which therefore remains invariant
under all smooth variations of the parameters which preserve SUSY2.
Now we move on to the more interesting case of supersymmetric quantum field
theory. The global supersymmetry algebra in four dimensions has left and right
handed Weyl spinor supercharges QA α and Q̄α̇A respectively. Here we are adopting
the standard notation of Wess and Bagger [5]: α and α̇ are left and right-handed
Weyl spinor indices and A = 1, 2, . . . , N labels distinct families of supercharges.
In four-dimensional renormalizable quantum field theory we are only interested
in algebras which lead to multiplets of particles with spins ≤ 1: this limits the
possiblities to N = 1, 2 and 4. The so-called extended SUSY algebras with N = 2
and 4 have an SU (N ) internal symmetry which rotates the supercharges. The left-
and right-handed supercharges are in the N and N̄ of this group, denoted by the
raised and lowered index A = 1, 2, . . . , N . The basic anti-commutator is
µ
(2.3) α , Q̄β̇B } = 2σαβ̇ Pµ δB .
{QA A
For theories with extended SUSY, supercharges of the same chirality also have a
non-trivial anti-commutator,
(2.4) {QA
α , Qβ } = αβ Z
B AB
.
Here Z AB = −Z BA is a central charge which commutes with all the other elements
of the algebra.
We will now briefly list some of the special features of these theories which
constrain the form of quantum corrections:
Non-renormalization: As in the example of SUSY quantum mechanics con-
sidered above, the SUSY algebra implies that the energy of a supersymmetric vac-
uum of the theory is exactly equal to zero. Typically in quantum field theory the
vacuum energy may be set to zero at the classical level, but it will certainly re-
cieve quantum corrections from loop diagrams. Unbroken supersymmetry implies
that these corrections vanish to all orders. This vanishing is due to a cancella-
tion between bosons and fermions, which reflects the additional factor of −1 in the
Feynman rules for a fermionic loop.
BPS states: Just as in SUSY quantum mechanics the supersymmetry algebra
implies that the energy of all states is bounded below by zero: E ≥ 0. In the case of
extended SUSY, one can actually derive a stronger bound for the masses of particles
of the schematic form M ≥ |Z| where Z is the central charge. This is known as the
Bogomol’nyi bound. Another similarity to the quantum mechanical case is the fact
2There are several caveats here: the logic described above only applies when the Hamiltonian
has a discrete spectrum. In theories with non-compact directions we must also limit ourselves to
variations which do not change the asymptotic behaviour of the potential in these directions.
that states which saturate this bound live in smaller representations of the SUSY
algebra. Specifically, states which saturate the bound are invariant under exactly
half the SUSY generators and typically fill out ‘short’ or BPS representations of
dimension 2N . In contrast, generic states transform non-trivially under all the
SUSY generators and form ‘long’ representations of dimension 22N . Considerations
like those described above leading to the Witten index in the case of quantum
mechanics mean that BPS states cannot be lifted above the Bogomol’nyi bound
except in multiples of 2N . Typically, the necessary additional BPS states are not
present and we conclude that the BPS mass formula M = |Z| is exact.
Holomorphy: Certain special quantities in SUSY gauge theories are con-
strained to be holomorphic functions of the fields and parameters. Much recent
progress can be traced to the fact that holomorphic functions can be reconstructed
exactly from knowledge of their singularities.
Flat directions: In theories with extended SUSY the potential energy typ-
ically has flat directions along which scalar fields can acquire expectation values
without energetic cost. Fluctuations along these directions are massless scalars. It
is important to emphasize that these scalars are not necessarily Goldstone modes
for any global symmetry. Their masslessness is instead a consequence of unbro-
ken supersymmetry. This situation is often described by saying that the theory
has a moduli space, M, of gauge-inequivalent vacua. It is often useful to think
of the moduli space as a Riemannian manifold with the scalar fields {Xi }, with
i = 1, 2, . . . dimR M as real coordinates. The natural metric Gij on M is provided
by the lowest term in the derivative expansion of the Wilsonian effective action for
the massless scalars,
1
(2.5) Leff = Gij (X) ∂µ X i ∂ µ X j + . . . .
2
Of course we must really consider a supersymmetric version of (2.5) with kinetic
terms for the superparteners of the massive fields. It turns out that the existence
of such an extension provides stringent constraints on the geometry of M. In
particular, for N = 1 SUSY in four dimensions, it is constrained to be a Kähler
manifold. In various situations with extended supersymmetry the requirement is
for hyper-Kähler or special Kähler geometry.
In the next sections we will review what we know about theories with N = 4
and N = 2 supersymmetry.
3. N = 4 SUSY Yang-Mills: the Classical Theory

N = 4 supersymmetric Yang-Mills theory in four dimensions can be obtained
by dimensional reduction of the minimal supersymmetric gauge theory in ten di-
mensions. In the following we will consider the theory with gauge group SU (N )
(or sometimes U (N )). The field content is a single massless vector multiplet. In
addition to the gauge field Aµ , the multiplet includes four adjoint-valued left-
handed Weyl fermions λA α with A = 1, 2, 3, 4 which transform in the fundamen-
tal representation of the SU (4) R-symmetry mentioned in the introduction. The
charge-conjugate degrees of freedom are right-handed Weyl spinors, λ̄α̇A in the anti-
fundamental of SU (4)R . The multiplet is completed by six adjoint scalar fields φa ,
with a = 1, 2, . . . , 6 which form a 6 of SU (4)R . As SU (4) is locally the same as
SO(6), this can also be thought of as the vector representation of the latter group.
322 NICK DOREY
The Lagrangian consists of the Maxwell term for the gauge fields together
with minimal coupling for the adjoint scalars and fermions. The Lagrangian also
includes a non-trivial potential for the scalar fields,

(3.1) V= TrN [φa ,φb ]2 .
a>b
Finally the supersymmetric completion of these terms include Yukawa couplings be-
tween the fermions and scalars. All these terms have mass dimension four and the
resulting Lagrangian is therefore scale-invariant. In fact the classical Lagrangian
is invariant under the group of conformal transformations in four dimensions,
SO(4, 2). This group includes the Poincaré generators together with dilatations
and special conformal transformations. The coefficients of each term in the La-
grangian are uniquely determined by N = 4 supersymmetry up to a single overall
coefficient, 1/g 2 . Here g is the dimensionless gauge coupling constant. One may
also add to the Lagrangian the topological term TrN (F ∧ F ) with coefficient the
vacuum angle θ. It is customary to combine g 2 and θ in a single complex coupling,
4πi θ
(3.2) τ= + .
g2 2π
Apart from the choice of gauge group, this is the only free parameter in the theory.
As mentioned above, the theory is invariant under the conformal group and
also under N = 4 supersymmetry transformations. The closure of these generators
results in the largest possible global spacetime symmetry group in four dimensions.
This is the N = 4 superconformal group, denoted SU (2,2|4). Its bosonic part
is SO(4, 2) × SU (4)R . In addition to the sixteen real components of the SUSY
generators QA α and Q̄α̇A , there are sixteen independent superconformal generators,
SαA and S̄α̇A .
The N = 4 theory exhibits a characteristic feature of theories with sixteen
supercharges mentioned in the previous section: the potential V has flat directions.
Specifically, if we choose each of the six scalar fields to have vacuum expectation
values (VEVs) in the Cartan subalgebra of U (N ), then the potential vanishes and
we have a supersymmetric vacuum. The corresponding manifold of vacuum states
is known as the Coulomb branch. The behaviour of the theory depends crucially
on whether these VEVs are non-zero. The two cases correspond to different phases
of the system which we describe in turn:
Coulomb Phase: If the scalar VEVs are non-zero then the gauge group
G = U (N ) is spontaneously broken down to its Cartan subgroup U (1)N . The
theory has massless photons corresponding to unbroken gauge generators which give
rise to long range Abelian gauge fields (hence the term ‘Coulomb’). As the scalar
fields have mass dimension, choice of a VEV also spontaneosly breaks the conformal
symmetry down to its Poincaré subgroup and also spontaneously breaks part of the
SU (4)R symmetry. Some of the massless scalars which appear as superpartners of
the photons can be thought of as Goldstone bosons in this context. In addition to
massless neutral fields, the theory also has massive particles which carry electric
and magnetic charges with respect to the Cartan subgroup.
Conformal Phase: If the scalar VEVs are zero, then the U (N ) gauge sym-
metry remains unbroken and so does the classical superconformal symmetry. The
theory is then in an interacting conformal phase. There is no mass gap and states
in the Hilbert space form multiplets of the superconformal group.
Before moving on to describe the quantum theory, we will take a closer look
at the spectrum of the theory with gauge group SU (2) on the Coulomb branch. In
this theory we can expand the bosonic fields in terms of the three Pauli matrices
τi , i = 1, 2, 3, in the obvious way as Aµ = Aiµ τi /2, φia τi /2. Up to a gauge rotation,
we can choose scalar VEVs of the form φa = v a τ3 /2. The Coulomb branch is
therefore parameterized by the complex six-component vector v a with an SU (4)R -
6
invariant length defined by |v|2 = a=1 v a v̄ a . The Weyl group of SU (2) acts as
v a → −v a and, modding this out, the manifold of gauge inequivalent vacua is just
R6 /Z2 .
As above, when |v| > 0, the gauge group is spontaneously broken down to the
Cartan subalgebra, which, in this case, is the U (1) subgroup of SU (2) generated
by the third Pauli matrix τ3 . In principle, particle states may carry electric and/or
magnetic charges, denoted QE and QM respectively, with respect to this U (1).
On general grounds, the allowed charges are constrained by the Dirac quantization
condition QE QM ∈ 2π Z. A semiclassical analysis of the theory reveals the following
states:
• A massless N = 4 vector multiplet with QE = QM = 0 containing the photon
and its superpartners. These are the elementary quanta which remain massless after
the Higgs mechanism. They correspond to fluctuations of the fields living in the
unbroken U (1).
• An N = 4 vector multiplet containing massive gauge bosons and their su-
perpartners. Each of these states has QE = ±g, QM = 0 and mass M = |v|. The
spin one members of the multiplet are fluctuations of the fields Wµ± = A1µ ± iA2µ
and are referred to as ‘W-bosons’.
In addition to these elementary quanta, the classical theory also contains mag-
netically charged solitons or magnetic monopoles. These are classical solutions of
the field equations for the Yang-Mills field minimally coupled to an adjoint scalar.
The configurations have finite mass M = 4π|v|/g 2 and carry magnetic charge
QM = 4π/g. Somewhat miraculously, the units of electric and magnetic charge
carried by the W-bosons and magnetic monopoles respectively turn out to satisfy
the Dirac quantization.3 At weak coupling the monopole can be quantized in a
semiclassical approximation and leads to massive one-particle states in the Hilbert
space. These include,
• An N = 4 supermultiplet of states with QE = 0 and QM = 4π/g.
• An infinite tower of dyonic states which carry integer multiples of the quan-
tum of electric charge in addition to one unit of magnetic charge: QE = ng,
QM = 4π/g with integer n.
The mass of each of these states is described by the universal mass formula:

|v|
(3.3) M= Q2E + Q2M .
g
3This appears miraculous because the Dirac quantization condition is derived from quantum
mechanical considerations, while the quantization of magnetic charge observed in this theory
happens already at the classical level. A further miracle is the fact that the monopole actually
carries twice the minimum allowed quantum of magnetic charge. This means that the theory
remains consistent when particles in the fundamental representation of G = SU (2) are introduced
which have half the electric charge of the W-boson.
324 NICK DOREY
4. The Quantum Theory of the Coulomb Phase

In this section and the next we describe some of the remarkable results for
the quantum N = 4 theory which have emerged over the last few years. As men-
tioned in the introduction the large space-time symmetry group of the theory places
strong restrictions on the form of quantum corrections to the classical results of the
last section. Of course an important general issue in any theory is whether the
symmetries of the classical theory survive quantization or, in other words, whether
the theory has anomalies. For example, the ordinary Maxwell action for any non-
Abelian gauge theory is scale invariant at the classical level, manifesting itself in
the presence of a dimensionless coupling constant g 2 . However, this scale invariance
is usually broken at one loop by the running of the coupling with energy scale µ:
g 2 → g 2 (µ). The breaking of scale invariance is measured by the Renormalization
∂
Group β-function, β(g) = µ( ∂µ )g 2 (µ). In asymptotically free gauge theories such
as QCD we have β < 0, which means that the running coupling becomes large in
the IR.
A remarkable fact about the N = 4 theory is that the β-function is exactly
zero. The cancellation occurs order by order in perturbation theory due to a can-
cellation between bosonic and fermionic loops. This is a striking example of the
non-renormalization properties of SUSY gauge theories mentioned in the introduc-
tion. It reflects the fact that the N = 4 theory is exactly scale-invariant at the
quantum level. A related fact is that the theory does not have a chiral anomaly. In
theories with massless fermions and a chiral anomaly the vacuum angle can be set to
zero by a chiral rotation of the fermions. In contrast, in the N = 4 theory, θ remains
as a genuine parameter of the theory. More precisely the physics depends periodi-
cally on θ via the Yang-Mills instanton factor q = exp(2πiτ ) = exp(−8π 2 /g 2 + iθ)
which is invariant under the transformation T : θ → θ + 2π.
As in the previous section, the physics depends critically on whether the adjoint
scalars acquire a vacuum expectation value. As before we have two phases: in the
conformal phase where the gauge group is unbroken the theory is invariant under
the full N = 4 superconformal algebra. In the language of the Renormalization
Group, the dimensionless parameter τ = 4πi/g 2 + θ/2π is an exactly marginal
coupling which parametrizes a fixed line of interacting superconformal theories. The
situation in the Coulomb phase is slightly different. As before, the superconformal
symmetry is spontaneously broken by a choice of scalar VEV. As there is no explicit
breaking of the symmetry the classical Goldstone bosons remain massless in the full
quantum theory. There are remarkable properties of the physics in each of these
phases which we will now discuss in turn.
In the classical Coulomb phase, the theory has a spectrum of massive particles
carrying magnetic and electric charges as described in the previous section. In a
generic QFT, we would be able to say little about the spectrum of the quantum
theory apart from computing the radiative corrections to the classical masses in
a weak coupling expansion. In the N = 4 theory however, we are able to do
much better than this, and we can actually make some exact statements about the
spectrum. The underlying reason is the presence of a central charge in the N = 4
supersymmetry algebra. As above we have the anti-commutator,
(4.1) α , Qβ } = αβ Z
{QA B AB
,
where Z AB is an anti-symmetric 4 × 4 matrix. By a unitary rotation we can put

this matrix in the canonical form,
 
0 +1 0 0
 −1 0 0 0 
Z = 2|Z|   0
.
0 0 +1 
0 0 −1 0
By working in the restframe of a massive particle, the algebra can be used to prove
a Bogomol’nyi bound for the mass of any state: M ≥ |Z|. Generic states in the
spectrum whose masses lie above the bound live in ‘long’ multiplets of N = 4 SUSY
of dimension 22N = 256. In contrast, states with M = |Z|, which saturate the
bound, are invariant under half the generators and live in ‘short’ or BPS multiplets
of dimension 2N = 16. As mentioned in the introduction, the key fact is that, as we
vary the parameters of the theory, BPS-saturated states cannot be lifted above the
Bogomol’nyi bound except by combining with fifteen other short multiplets with
the same quantum numbers to make a long multiplet. Normally the additional
states are simply not available and we can safely conclude that BPS states which
are present in one regime of parameters will be present throughout the parameter
space. Further, the relation M = |Z| is the exact mass formula for these states.
To proceed further we must identify the central charge. In the SU (2) theory
it is given by
v
(4.2) Z = (QE + iQM )
g
where QE and QM are electric and magnetic charges with respect to the unbroken
U (1) subgroup of SU (2) on the Coulomb branch. The mass formula for BPS states
is then precisely (3.3) above and we can conclude that the entire spectrum of electric
and magnetically charged states discussed in the previous section are actually BPS
states. The fact that the ’t Hooft-Polyakov monopoles in this theory yield BPS
states is closely connected with the fact that the corresponding classical soliton
solutions obey first order Bogomol’nyi equations of the form Bi = ±Di φ. This
equation is precisely the condition that the classical field configuration is invariant
under half the supersymmetry generators. It is also easy to find hints as to why the
W-bosons should be BPS saturated. These states are obtained from the spectrum
of massless states via the Higgs mechanism and massless states are automatically
BPS saturated.
In the classical analysis of the previous section, we saw that the basic unit of
electric charge is that of the W-boson, which is g, while the basic unit of magnetic
charge is that carried by the ’t Hooft-Polyakov monopole, which is 4πv/g. We can
express the charges of an aibitrary state in terms of integer multiples of these unit
charges by setting QE = nE g and QM = nM (4π/g) where nE and nM are integers.
In these units the central charge of the SU (2) theory can be written as
(4.3) Z = v (nE + nM τ ) .
Actually this formula is a slight extension of the previous one as we have incorpo-
rated the effects of a non-zero θ angle. It reflects a phenomenon first discovered
by Witten in [15]. In the presence of non-zero θ, monopoles with nM units of
magnetic charge acquire an electric charge ∆QE = nM gθ/2π. The Dirac quanti-
zation condition is also modified by the non-zero vacuum angle to accommodate
this shifted charge. Thus, in the presence of nM units of magnetic charge, the shift
326 NICK DOREY
T : θ → θ + 2π is no longer a trivial invariance of the theory. It shifts the integer-

valued electric charge nE of every state by exactly nM units. However T will still
leave the spectrum of the theory unchanged provided the theory already contains
all the possible image states under the repeated application of this transformation
(and its inverse).
In fact the BPS spectrum described above has another much more profound
invariance. As we have a theory with both electric and magnetic charges, we ask
to what extent they appear symmetrically in the theory. One may look for a
transformation which interchanges electric and magnetic charges and fields in a
way which leaves Maxwell’s equations invariant. The action on the charges is
QE → QM and QM → −QE . As our units of electric and magnetic charge (for
θ = 0) are determined by the Dirac quantization condition to be QE = nE g and
QM = nM 4π/g, this requires the transformations nE → nM , nM → −nE together
with g → 4π/g. Olive and Montonen [13] were the first to observe that this is
indeed an invariance of the BPS spectrum described above, and they were also the
first to conjecture that this could be an exact duality of a non-Abelian gauge theory
in its Coulomb phase.
Several comments are in order about the conjectured duality. First of all, as it
inverts the coupling constant, it relates the strong coupling behaviour of the theory
to its weak coupling behaviour. As discussed in the introduction, this makes the
proposal extremely interesting but hard to test without independent knowledge
about the strongly-coupled regime. The transformation also interchanges the roles
of the electric charges which appear as elementary particles at weak coupling with
the magnetic charges which appear as classical solitons. Thus Olive-Montonen
duality shares many features with the duality between the sine-Gordon and Thirring
models in two dimensions [4], where the kink of the former model is identified with
the elementary fermion of the latter. The conjecture of Olive and Montonen refers
in the first instance to an ordinary bosonic gauge theory without supersymmetry,
and in this context there are several obvious objections. First the invariance is only
demonstrated at the level of the classical mass formula, which in bosonic gauge
theory will certainly be corrected by quantum effects. Quantum effects also lead to
the running of the coupling constant and the generation of a dynamical mass scale
which is the only true parameter of the theory. Strong and weak coupling both occur
in the same theory at different energy scales and it is unclear how to implement
the tranformation g → 4π/g. Finally the symmetry between the monopole and the
electrically charged gauge boson is broken by the fact that the latter states have
spin one. In the absence of fermions, the monopole field configuration is essentially
spherically symmetric and therefore does not carry angular momentum.
It is a remarkable fact that each of the objections raised above is without force
in the context of N = 4 supersymmetric Yang-Mills theory. As mentioned above
the BPS mass formula M = |Z| is exact and, together with the stability properties
of BPS states described above, it can be used to determine the BPS spectrum at
strong coupling. As the β-function of the N = 4 theory vanishes exactly, there
are none of the problems associated with the running of the coupling. Perhaps
most remarkably, in the N = 4 theory, the W-boson and the BPS monopole lie in
supermultiplets with the same spin content [14]. Although the classical monopole
itself does not carry spin, it can form threshold bound-states with the elementary
fermions of the theory which do. The fact that these states fill out a short multiplet
is intimately related to the fact mentioned above that the BPS monopole is invariant
under half the supersymmetry generators.
It is instructive to consider the extension of electric-magnetic duality occurring
when the vacuum angle is non-zero. In this case, duality acts on the complexified
coupling τ = 4πi/g 2 + θ/2π as the transformation S : τ → −1/τ . Obviously S 2 is
the identity and repeated application of this transformation does not generate any
thing new. However, we should also remember the transformation T introduced
above which shifts the vacuum angle by 2π. This acts on the complexified coupling
as T : τ → τ + 1. Importantly though, S and T do not commute with each
other and, together, they generate an infinite group of transformations known as
SL(2, Z). The action of a general element of this group on the coupling is
aτ + b
τ →
cτ + d
(4.4)
where a, b, c and d are integers with ad − bc = 1. The corresponding action on the

integer-valued electric and magnetic charges is

nE a b nE
→ .
nM c d nM
This group is well known in complex analysis as the modular group of the torus.
If complex tori are parameterized by the ratio of their periods τ = ω2 /ω1 , then
values of τ related by an SL(2, Z) transformation as in (4.4) describe tori which are
equivalent as complex manifolds. This geometric realization of the duality group
is central to how electric-magnetic duality appears on branes living in toroidal
compactifications of M-theory.
The extended electric-magnetic duality described above is known as S-duality.
There is now much evidence that S-duality is actually an exact invariance of the
N = 4 theory. The simplest evidence in favour of this comes from the spectrum
of BPS states. Although the BPS mass formula is certainly invariant under S-
duality, invariance of the spectrum clearly requires the existence in the theory
of the SL(2, Z) images of all the known BPS states. In particular, it is easy to
prove that the SL(2, Z) orbit of the W-boson with electric charge nE = 1 and
zero magnetic charge includes states with all coprime values of nE and nM . For
nM = 1, this amounts to the existence of the dyonic states mentioned in the
previous section. However, for nM > 1, we also have a prediction for the existence
of many new states which can be thought of as bound-states of monopoles and
dyons. By the usual reasoning, it suffices to check the existence of these states at
weak coupling where we may use a semiclassical quantization of the multi-monopole
system. The existence of the required bound-states translates into the existence of
certain normalizable harmonic forms on the moduli space of BPS monopoles. This
has been checked explicitly by Sen [16] in the case of magnetic charge two, where
an explicit metric on the monopole moduli space is known. Progress has also been
made in demonstrating the existence of the necessary forms in the case of arbitrary
magnetic charge [17].
328 NICK DOREY
5. The Quantum Theory of the Conformal Phase

We now return to the conformally invariant phase where the scalar VEVs are
zero and the SU (N ) gauge symmetry is unbroken. As mentioned in Section 2, in this
phase the theory is characterized by invariance under the supergroup SU (2, 2|4),
which contains the four-dimensional conformal group SO(4, 2) and the R-symmetry
group SU (4)R together with 32 real supercharges. The Hilbert space of the theory
is made up of states transforming in irreducible representations of this supergroup.
States are labeled by their eigenvalues under the appropriate Casimir operators.
These include the left and right moving spins (jL , jR ), the scaling dimension D,
which is the eigenvalue of the dilatation operator, and the Casimirs of SU (4)R . As
in any conformal field theory, there is a natural map between local operators and
states provided by radial quantization, so we can equivalently think in terms of
local, gauge-invariant operators labeled by the same quantum numbers. Roughly
speaking, the state corresponding to a given operator is that obtained by acting on
a canonically chosen vacuum state with the operator in question.
In the previous sections, we encountered the phenomenon of short or BPS-
saturated multiplets of supersymmetry. It turns out that an exactly analogous
phenomenon occurs in the representation theory of the superconformal algebra. The
easiest case to describe is that of N = 1 superconformal symmetry, the conformal
extension of the minimal supersymmetry algebra in four dimensions. The N = 1
superconformal group contains a U (1) global R-symmetry. States or operators
are labeled by a definite scaling dimension D and a definite U (1) R-charge R.
There is an absolute lower-bound on these scaling dimensions, D ≥ |R|. This is
analogous to the Bogomol’nyi bound in ordinary supersymmetry and, as in that
case, states which saturate the bound are invariant under half of the algebra and
consequently live in short representations. The corresponding operators are known
as chiral primaries and, by a now familiar argument, they cannot be lifted above
the bound except in very special circumstances. In cases where the theory has a
weakly-coupled limit, we will typically find a set of these chiral primary operators in
the classical theory with classical dimensions which saturate the bound. Their BPS
stability property then ensures that the corresponding states remain in the spectrum
for all values of the coupling. In the usual language of the Renormalization Group,
the exact formula D = |R| ensures that these operators do not acquire anomalous
dimensions.
The N = 4 superconformal algebra also has chiral primary representations.
Roughly speaking, the states are chiral primary with respect to the N = 4 super-
conformal algebra if they are chiral primary with respect to any N = 1 subalgebra.
The condition on their scaling dimensions reads D = |R|, where R is the charge
under any U (1) subgroup of SU (4)R . One infinite class of such operators is formed
by taking symmetric traceless combinations of the six adjoint scalar fields φa . For
example, the operators
(5.1) Oa1 a2 ...an = TrN [φa1 φa2 . . . φan ]
are chiral primary provided we symmetrize the SU (4)R indices and subtract off
traces.
The quantum behaviour of the N = 4 theory in its conformal phase is highly
constrained by superconformal symmetry. Apart from the spectrum of states, the
natural observables are the correlation functions of local gauge-invariant operators.
These obey superconformal Ward identities which are sufficient to determine all
two- and three- point functions of chiral primary operators exactly. However, scaling
dimensions and correlation functions of other operators in the theory are much less
constrained and, in general, they will receive complicated perturbative and non-
perturbative corrections. Despite this, two remarkable properties of the quantum
theory have emerged in recent years. The first is the electric-magnetic duality
discussed in the previous section. Unlike the Coulomb phase, the SU (N ) gauge
group is unbroken and we cannot define Abelian electric and magnetic charges.
Nevertheless there is still considerable evidence that the conformal theory has an
exact SL(2, Z) duality acting on the complexified coupling constant as in (4.4).
A precise formulation is that the operators in the theory and their correlation
functions transform as modular forms of definite weight4. In [27], Vafa and Witten
evaluated the partition function of a topologically-twisted version of the N = 4
theory on various four-manifolds and verified its modular properties explicitly.
The second major development is one we will touch on only briefly. As men-
tioned in the introduction, there is a long-standing conjecture that the large N limit
of non-Abelian gauge theory should be dual to a weakly-coupled string theory. Un-
til recently, the identity of the dual string theory was essentially unknown. In late
1997, Maldacena [8] made a remarkable proposal concerning the large N dual of
the N = 4 theory at its superconformal point. Specifically it concerns the N = 4
theory with gauge group SU (N ) in the ’t Hooft limit N → ∞, g 2 → 0 with the ’t
Hooft coupling λ = g 2 N held fixed. Maldacena conjectured that the N = 4 theory
in this limit (and perhaps beyond it) is equivalent to Type IIB superstring the-
ory propagating on the ten-dimensional spacetime, AdS5 × S 5 . Here AdS5 denotes
five-dimensional anti-de-Sitter space and S 5 is the five-sphere. This proposal is also
known as the AdS/CFT correspondence. The identification of the parameters of the
two theories is as follows: the complexified coupling τ = 4πi/g 2 +θ/2π of the N = 4
theory is identified with a corresponding complex parameter τIIB = i/gstr + 2πC (0)
of the IIB theory. Here gstr is the string coupling and C (0) is the VEV of the scalar
field arising from the Ramond-Ramond sector of the IIB theory. The radius, L,
of both√the AdS5 and S 5 factors of the spacetime, measured in units of the string√
length α , scales like a positive power of the ’t Hooft coupling: L2 /α ∼ λ.
Additionally there are N units of R-R fiveform flux through the five-sphere.
Several comments are in order. The first point of agreement between the two
sides of the correspondence is the global symmetries apprearing on both sides. The
SO(4, 2) group of conformal transformations on the N = 4 side emerges as the
isometry group of AdS5 . Similarly, the SU (4) R-symmetry of the N = 4 theory
corresponds to the isometry group of the five-sphere. The number of supercharges
of the IIB theory is thirty-two and all of these remain unbroken in the AdS5 ×
S 5 background. In fact they enlarge the bosonic symmetry group to form the
supergroup SU (2, 2|4) which is precisely the N = 4 superconformal group in four
dimensions. In retrospect, Maldacena’s proposal is strongly suggested by the fact
that IIB on AdS5 × S 5 is the only known string theory with this symmetry group.
More complicated dynamical aspects of the N = 4 theory are also reproduced on
the IIB side. In particular the IIB theory has an SL(2, Z) duality group acting on
τIIB , including transformations which interchange the NS-NS and R-R sectors of
4f (τ, τ̄ ) is a modular form of weight (w, w̄) if it transforms as f → (cτ + d)w (cτ̄ + d)w̄ f under
the SL(2, ) transformation (4.4)
330 NICK DOREY
the theory. This corresponds to the S-duality group of the N = 4 theory acting on
τ.
The fact that the string coupling is identified with the Yang-Mills coupling
constant (squared), which scales like 1/N , means that the dual string theory is
weakly coupled at large N as originally envisaged by ’t Hooft. In principle the
sum of planar diagrams in the N = 4 theory can be evaluated by solving the
tree-level string theory of AdS5 × S 5 . Unfortunately, the latter task has proved
to be extremely difficult. The main problem is that the background includes non-
zero R-R flux, something which is hard to incorporate in the theory on the string
world-sheet. Rather than working with the full string theory on AdS5 ×S 5 , one may
instead consider only the low-energy approximation to the theory, which amounts to
studying IIB supergravity on the same background. This approximation is reliable
as long as the stringy excitations are more massive than the Kaluza-Klein modes
of the SUGRA fields on the S 5 . This is true provided the radius of the √ sphere
is much larger that the string length. Given the identification L2 /α = λ, this
corresponds to an interesting strongly-coupled regime of the N = 4 theory. In
contrast, Feynman perturbation theory is only valid when the ’t Hooft coupling is
small. Hence the regimes in which we may calculate reliably on each side of the
correspondence are non-overlapping.
In this sense, AdS/CFT is a weak-coupling/strong-coupling duality like the
electric-magnetic duality discussed above. In particular, we are confronted with the
same problem we encountered in that context: in order to test Maldacena’s proposal
we need to perform an independent calculation when one of the two theories in
question is strongly coupled. As in the case of S-duality, the special properties of
supersymmetric gauge theory come to the rescue. In particular, as described above,
we know the theory contains a large set of operators whose scaling dimensions are
independent of the coupling. One may safely extrapolate the classical spectrum
of N = 4 chiral primaries to the regime of large ’t Hooft coupling where they
can be compared with the supergravity spectrum. In particular, the IIB theory
on AdS5 × S 5 contains an infinite tower of states living in small representations
of the SU (2, 2|4) superalgebra. These states correspond to the expansion of the
massless ten-dimensional fields of IIB supergravity in spherical harmonics on the
five-sphere. This comparison was first performed by Witten in [9], and yields
impressive agreement.
6. Theories with less supersymmetry

So far we have only discussed the N = 4 theory. We have seen that the large
degree of supersymmetry gives us enough control over quantum effects to formulate
and test highly non-trivial dualities like SL(2, Z) and AdS/CFT. However, the
extended supersymmetry also leads to features, like the vanishing β-function, which
are rather unrealistic from the point of view of QCD. In fact, powerful exact results
also exist for theories with less supersymmetry, most notably the celebrated solution
of N = 2 SUSY Yang-Mills due to Seiberg and Witten [19]. In the remainder of
this article we will provide a brief sketch of the physics of these theories emphasizing
the role of electric-magnetic duality.
As in the original Seiberg-Witten paper we will focus on the minimal N = 2
theory with gauge group SU (2). This consists of a single vector multiplet of N = 2
SUSY which contains the SU (2) gauge field, two Weyl fermions and a single complex
scalar field Φ. All fields are in the adjoint representation of the gauge group and
the theory has a potential term,

(6.1) V = Tr [Φ,Φ† ]2 .
As in the N = 4 theory the potential has a flat direction along which the complex
scalar acquires a VEV breaking the SU (2) gauge symmetry down to its Cartan
subalgebra. By gauge rotation we can choose the generator of the unbroken U (1)
to be the third Pauli matrix. The resulting Coulomb branch can be parameterized
by a single complex number a, up to Weyl reflection a → −a, as Φ = aτ3 /2. Al-
ternatively we may introduce the gauge-invariant modulus u = Tr Φ2 . Classically
we have u = a2 /2 although, as u is a composite operator, this relation can receive
quantum corrections. The classical theory on the Coulomb branch is very similar
to that of the N = 4 theory. Apart from the massless photon multiplet, there are
also massive N = 2 multiplets corresponding to the massive W-bosons, as well as
a spectrum of monopoles and dyons. All of these states live in short multiplets of
N = 2 SUSY. The classical Bogomol’nyi bound is M ≥ |Z|, with central charge,
(6.2) Z = a (nE + nM τ )
where nE and nM are integer-valued electric and magnetic charges.
The differences between the N = 4 and N = 2 theories emerge as soon as we
compute quantum corrections. In the minimal N = 2 theory the β-function is non-
zero and negative at one-loop. This means that the N = 2 theory is asymptotically
free in the UV. The coupling constant runs and is replaced by a dynamical scale
Λ. In the absence of scalar VEVs, the theory runs to strong coupling at the scale
Λ and, a priori, it is very hard to say what the IR physics is like. However, if the
scalar VEV a is non-zero, the gauge group is broken down to U (1) at the scale
|a|. Below this energy scale the effective theory is Abelian and the coupling does
not run. It follows that the theory has two different regimes. When the VEVs are
large, |a| >> Λ, the running coupling is frozen before it has a chance to get large
and the theory is weakly coupled at all length scales and semiclassical methods are
valid. On the other hand when |a| ∼ Λ the IR coupling is strong and perturbation
theory breaks down. If we think of the Coulomb branch as the complex u-plane,
the first regime is the asymptotic region near the point at infinity and the second
is the region near the origin.
As in the N = 4 theory, to make progress we must concentrate on quantities
which are highly constrained by supersymmetry. As in that case these include
the spectrum of BPS states. However, although the BPS bound M = |Z| is an
exact mass formula in this case also, the N = 2 story is complicated by the fact
that the central charge can recieve quantum corrections. Indeed we have already
commented on the fact that the bare coupling τ appearing in the central charge
gets renormalized. The most general possibility for the exact central charge is,
(6.3) Z = nE a(u) + nM aD (u)
where a and aD are holomorphic in u and λ. This is an example of the general
feature of SUSY gauge theories described earlier, namely that SUSY constrains
some observables to be holomorphic functions of the fields and parameters.
The low-energy theory on the Coulomb branch is the N = 2 supersymmetric
extension of QED. It is characterzed by an effective complexified coupling τef f
which is also constrained to be holomorphic in u and Λ. In the absence of charged
332 NICK DOREY
matter it has an exact SL(2, Z) duality which acts on the effective coupling τef f
rather than the bare one as in (4.4). This is essentially an extension of the Hodge
duality of Maxwell theory to N = 2 superspace and it can be implemented by an
exact transformation of the path integral. The renormalized BPS mass formula
given above is also SL(2, Z) invariant provided that aD /a transforms in the same
way as τef f under modular transfomations. Although the duality is only a property
of the low-energy theory, it plays an important role in the exact solution of the
model.
Previously we indicated that a, aD and τef f depended holomorphically on u
and Λ. However it is not quite precise to say that they are holomorphic functions
defined on the u-plane. In particular we should allow for the possiblity that the
spectrum changes by an SL(2, Z) transformation as we traverse a closed curve
on the u-plane. Thus we should really require that (a, aD ) defines a holomorphic
section of a flat SL(2, Z) bundle over the u-plane. In fact, a one-loop perturbative
analysis is sufficient to show that this bundle is non-trivial. The classical theory
has a U (1) R-symmetry under which the adjoint scalar has charge two. Acting with
this generator on a point far from the origin of the Coulomb branch we traverse
a large circle on the u-plane. However, at one loop the symmetry suffers from an
anomaly. The effect of this is that a U (1) rotation through an angle α is no longer
exactly a symmetry of the theory but rather leads to a shift in the effective vacuum
angle θef f → θef f + 2α. Via the Witten effect described in the previous section,
this shifts the electric charges of (nM = 1) magnetic monopoles in the theory by an
amount α/π. Rotation through 2π, accomplished by traversing a large circle in the
u-plane, therefore implements the non-trivial SL(2, Z) transformation5 denoted T 2
in the notation of the previous section. Such a transformation associated with a
closed path is known as a monodromy.
In summary, the low energy effective action and the BPS spectrum of the the-
ory are determined by the holomorphic quantities τef f (u) and (a(u), aD (u)), which
together define a holomorphic section of an SL(2, Z) bundle over the u-plane. Like
a holomorphic function, this bundle is essentially determined by its behaviour at its
singular points in the u-plane. In particular, the allowed singularities are logarith-
mic cuts which induce SL(2, Z) monodromies associated with closed paths in the
u-plane which encircle the singular point. The non-trivial monodromy described in
the previous paragraph is associated with a logarithmic branch point at u = ∞.
The logarithmic divergence as u → ∞ in turn reflects the logarithmic running of
the coupling in the UV. In principle the SL(2, Z) bundle can be reconstructed from
a knowledge of the monodromies of the BPS spectrum around each singular point.
As the branch cut coming from the point at infinity must terminate somewhere
there must be at least one singular point in the interior of the u-plane. To make
further progress it is necessary to consider the possible physical significance of such
points.
The singular points are points at which the couplings in the low-energy effec-
tive Lagrangian diverge. This breakdown of the low-energy description is a classic
symptom of extra massless degrees of freedom which need to be taken into account.
It is reasonable therefore to assume that the singular points in the interior of the
u-plane are points at which new massless particles appear. In the classical theory
5The real story is slightly more complicated than I have indicated as the transformation also
changes the sign of the magnetic charge.
there is only one such point, the point u = 0 where all the classical BPS states
become massless and the full SU (2) gauge symmetry is restored. One might think
that the simplest possibility is that this situation persists in the quantum theory.
However, a basic constraint on the solution is that the effective gauge coupling
constant Im τef f should be positive everywhere to ensure unitarity of the effective
action. Using an elementary theorem from complex analysis, Seiberg and Witten
were able to rule out scenarios involving only a single singularity at the point u = 0.
The next simplest possibility is that of a pair of singularities away from the
origin interchanged by the global symmetry u → −u. For various reasons discussed
in [19], it is hard to associate these points with the restoration of non-Abelian
gauge symmetry. However, Seiberg and Witten found an interesting alternative
explanation. They considered the possibility of points at which a single BPS state
becomes massless. Given the presence of a new light state in the vicinity of the
singular point the effective Lagrangian can then be improved by including a charged
N = 2 matter multiplet and explicitly evaluating the one-loop running of the effec-
tive coupling due to the charged particle. Just as at weak coupling, the perturbative
logarithm dictates the SL(2, Z) monodromy associated with contours encircling the
singularity. Remarkably, they found a single consistent solution with two singular
points at u = ±Λ2 . The corresponding massless states at these two singularities
are the BPS monopole and the dyon of electric-magnetic charge (nE , nM ) = (1, 1).
The resulting exact formulae for the BPS mass spectrum and effective Lagrangian
have been tested precisely in many different ways.
The Seiberg-Witten solution realizes electric-magnetic duality in a quite dif-
ferent way than the N = 4 theory. The magnetic degrees of freedom which are very
massive in the semiclassical regime become massless in the vicinity of the singular
points. At low energies we have a dual description in terms of an Abelian N = 2
gauge theory with charged matter. In the N = 4 theory, electric-magnetic duality
holds at all length scales. The monopoles become massless only in the strong cou-
pling limit and the dual theory in this regime is non-Abelian: it is just the same
N = 4 theory again with but with dual coupling τD = −1/τ .
Finally we can return to the issue we hoped to address all along, that of
confinement. In particular we can study the effect of soft-breaking of the N = 2
supersymmetry down to N = 1. In terms of N = 1 supersymmetry, the N = 2
vector multiplet splits into an N = 1 vector multiplet containing the gauge field
and a single Weyl fermion and a chiral multiplet containing the complex adjoint
scalar Φ and its N = 1 superpartner which is another Weyl fermion. Soft breaking
of N = 2 supersymmetry down to N = 1 means giving a non-zero mass µ to the
adjoint chiral multiplet. At energy scales far below µ, the massive scalar and its
superpartner decouple leaving the theory of a single N = 1 vector multiplet. This
theory is N = 1 supersymmetric Yang-Mills theory, a theory which is believed to
have many features in common with QCD. In particular, it is an asymptotically
free theory which runs to strong coupling in the IR. Strong coupling dynamics lead
to the dynamical breaking of chiral symmetry by a fermion condensate just as they
do in QCD. We can say this with confidence as the so-called gluino condensate
can be calculated exactly using the fact that it is a holomorphic function of the
parameters [22]. As in QCD, we interpret this as the signal of a gauge theory
realized in the confining phase. One important difference from QCD is that the
non-anomalous chiral symmetry is Z2N and, as always, spontaneous breaking of
334 NICK DOREY
a discrete symmetry leads to degenerate vacua. For the pure N = 1 theory with
gauge group SU (N ), the condensate breaks Z2N down to Z2 and we expect to find
exactly N degenerate supersymmetric vacuum states.
As soon as we introduce a small non-zero mass µ for the adjoint chiral multi-
plet, the IR dynamics should be qualitatively similar to the N = 1 super Yang-Mills
theory with gauge group SU (2)6. Hence we expect to find a confining theory with
two supersymmetric vacua. However, when µ is small compared to the dynamical
scale Λ, we can actually study the dynamics quantitatively as a small perturbation
of the Seiberg-Witten solution. In the classical theory, the introduction of a mass
term for Φ means that this field can no longer acquire a VEV and the only surviving
classical vacuum is the one in which the VEV is zero and the non-Abelian symme-
try is restored. However, the Seiberg-Witten solution shows that such a vacuum
simply does not exist in the full quantum theory. Naively this might suggest that
the perturbed theory has no SUSY vacua.
To understand what really happens one must remember to include the effects
of the extra light degrees of freedom which appear near the singular points on
the u-plane. For example, in the vicinity of the point u = Λ2 the low energy
theory actually contains three scalar fields. Apart from the modulus field u, we
have complex scalar fields M and M̃ whose quanta are the light BPS monopoles.
Including these degrees of freedom one finds a supersymmetric vacuum at u√= Λ
with a non-zero value for the VEVs of the monopole fields M = M̃ ∼ µΛ.
Thus the light monopoles have condensed in the vacuum. This immediately leads
to the confinement of electric charges due to the dual Meissner effect! As the dual
theory is weakly coupled, one can study the formation of electric flux tubes very
explicitly. This is the first explicit demonstration of confinement in a non-Abelian
gauge theory in four dimensions.
Another supersymmetric vacuum appears at the other singular point u =
−Λ2 . By similar arguments, one finds that the light dyons condense in the vacuum
realizing a phase with oblique confinement. These two SUSY vacua are believed to
be continuously connected to the two SUSY vacua of the pure N = 1 theory which
emerges as the mass parameter µ is taken to infinity.
7. Conclusion
In these notes we have seen how two old ideas about QCD have been real-
ized very explicitly in the more controlled context of supersymmetric gauge the-
ory. Seiberg-Witten theory conclusively demonstrates that confinement due to the
condensation of magnetic monopoles really does occur in four-dimensional field
theory. Maldacena’s conjecture identifies the large N dual string theory of a four-
dimensional non-Abelian gauge theory for the first time. In both cases, we are still
a long way from being able to apply these ideas in their original context. However,
the author believes that the developments described above represent important
steps in the right direction.
The author would like to thank the organisers of the TMR Winter school
in Utrecht and the Clay Institute/INI School on “Complex Geometry and String
Theory” where these lectures were given.
6Quantitative agreement with the N = 1 theory should only be obtained when we fully
decouple the extra matter and this requires taking the limit µ → ∞ holding the dynamical scale
on the N = 1 theory fixed.
8. Further Reading
A (very) partial list of references for my lectures is as follows. The classic
textbook on N = 1 SUSY is the book of Wess and Bagger [5]. The basic facts about
the representations of extended SUSY, including reduction from higher dimensions
are reviewed in Lykken’s 1996 TASI lectures [6].
Different aspects of N = 4 SUSY Yang-Mills are reviewed by different authors.
The superconformal phase and the AdS/CFT correspondence are extensively dis-
cussed in the review of Aharony et al [7]. Another very useful reference for this
subject is [10]. The relation between theories with 16 supercharges in different
dimensions is covered in [11].
The original conjecture of electric-magnetic duality in non-Abelian gauge the-
ory was made by Montonen and Olive in [13]. Its particular relevance to the N = 4
theory was suggested by Osborn [14]. Other important ingredients for the S-duality
conjecture were supplied in [12] (BPS states) and [15] (the “Witten effect”). An
impressive semiclassical test of S-duality is described by Sen in [16]. These devel-
opments are reviewed in Harvey’s lecture notes [18].
The original paper by Seiberg and Witten on N = 2 SUSY Yang-Mills is [19].
The generalization to include matter is described in their second paper [20]. An
important earlier work is by Seiberg [21], where the general form of the prepotential
is derived. Direct semiclassical tests of the Seiberg-Witten solution are given in
[23] and [24]. There are many reviews of Seiberg-Witten theory, but the one which
is most relevant to these lectures is Peskin’s 1996 TASI lectures [22]. This also
includes an excellent review of N = 1 theories. A more stringy perspective on SW
theory is given by Lerche [25].
References
1. G. ’t Hooft, A planar diagram theory for the strong interactions, Nucl. Phys. B72 (1974)
461.
2. S. R. Coleman, 1/N, in C79-07-31.7 SLAC-PUB-2484 Presented at 1979 Int. School of Sub-
nuclear Physics, Pointlike Structures Inside and Outside Hadrons, Erice, Italy, Jul 31-Aug
10, 1979.
3. S. Coleman and J. Mandula, All Possible Symmetries of the S-matrix, Phys. Rev. 159 (1967)
1251.
4. S. R. Coleman, Quantum Sine-Gordon Equation As The Massive Thirring Model, Phys.
Rev. D 11 (1975) 2088.
5. J. Wess and J. Bagger, Supersymmetry and Supergravity, Princeton University Press (1992).
6. J. D. Lykken, Introduction to supersymmetry, Fields, Strings and duality (1997) 85-153
World Sci. [arXiv:hep-th/9612114].
7. O. Aharony, S. S. Gubser, J. Maldacena, H. Ooguri and Y. Oz, Phys. Rept. 323 (2000) 183
8. J. M. Maldacena, The large N limit of superconformal field theories and supergravity,
Adv. Theor. Math. Phys. 2 (1998) 231 [Int. J. Theor. Phys. 38 (1999) 1113] [arXiv:hep-
th/9711200].
9. E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253
10. K. A. Intriligator, Bonus symmetries of N = 4 super-Yang-Mills correlation functions via
AdS duality, Nucl. Phys. B 551 (1999) 575 [arXiv:hep-th/9811047].
11. N. Seiberg, Notes on theories with 16 supercharges, Nucl. Phys. Proc. Suppl. 67 (1998) 158
12. E. Witten and D. I. Olive, Supersymmetry Algebras That Include Topological Charges, Phys.
Lett. B 78 (1978) 97.
336 NICK DOREY
13. C. Montonen and D. I. Olive, Magnetic Monopoles As Gauge Particles?, Phys. Lett. B 72
(1977) 117.
14. H. Osborn, Topological Charges For N=4 Supersymmetric Gauge Theories And Monopoles
Of Spin 1, Phys. Lett. B 83 (1979) 321.
15. E. Witten, Dyons Of Charge eθ/2π, Phys. Lett. B 86 (1979) 283.
16. A. Sen, Dyon - monopole bound states, selfdual harmonic forms on the multi-monopole mod-
uli space, and SL(2, ) invariance in string theory, Phys. Lett. B 329 (1994) 217 [arXiv:hep-
th/9402032].
17. G. Segal and A. Selby, The Cohomology Of The Space Of Magnetic Monopoles, Commun.
Math. Phys. 177 (1996) 775.
18. J. A. Harvey, Magnetic monopoles, duality, and supersymmetry, arXiv:hep-th/9603086.
Fields, Strings and duality. World Sci. (1997) 157-216.
19. N. Seiberg and E. Witten, Electric - magnetic duality, monopole condensation, and confine-
ment in N=2 supersymmetric Yang-Mills theory, Nucl. Phys. B 426 (1994) 19 [Erratum-ibid.
B 430 (1994) 485] [arXiv:hep-th/9407087].
20. N. Seiberg and E. Witten, Monopoles, duality and chiral symmetry breaking in N=2 super-
symmetric QCD, Nucl. Phys. B 431 (1994) 484 [arXiv:hep-th/9408099].
21. N. Seiberg, Supersymmetry And Nonperturbative Beta Functions, Phys. Lett. B 206 (1988)
75.
22. M. E. Peskin, Duality in supersymmetric Yang-Mills theory, Fields, Strings and duality.
World Sci. (1997) 729-809.
23. D. Finnell and P. Pouliot, Instanton calculations versus exact results in four-dimensional
SUSY gauge theories, Nucl. Phys. B 453 (1995) 225 [arXiv:hep-th/9503115].
24. N. Dorey, V. V. Khoze and M. P. Mattis, Multi-Instanton Calculus in N=2 Supersymmetric
Gauge Theory, Phys. Rev. D 54 (1996) 2921 [arXiv:hep-th/9603136].
25. W. Lerche, Introduction to Seiberg-Witten theory and its stringy origin, Nucl. Phys. Proc.
Suppl. 55B (1997) 83 [Fortsch. Phys. 45 (1997) 293] [arXiv:hep-th/9611190].
26. R. Donagi and E. Witten, Supersymmetric Yang-Mills Theory And Integrable Systems, Nucl.
Phys. B 460 (1996) 299 [arXiv:hep-th/9510101].
27. C. Vafa and E. Witten, A Strong coupling test of S duality, Nucl. Phys. B 431 (1994) 3
Department of Physics, University of Wales, Swansea, Singleton Park, Swansea,

SA2 8PP, UK
E-mail address: n.dorey@swansea.ac.uk
Volume 3, 2004
The Geometry of A-branes
Anton Kapustin
Abstract. I discuss the geometry of A-branes in the context of the Homolog-

ical Mirror Symmetry conjecture. I argue that there exist A-branes which are
coisotropic, rather than Lagrangian, submanifolds of a Calabi-Yau. I propose
a formal analogue of Floer homology for coisotropic A-branes.
Mirror Symmetry
The goal of my talk is to describe the current state of knowledge about topologi-
cal D-branes of type A, also known as A-branes. The most important motivation for
studying topological D-branes is the Homological Mirror Symmetry conjecture put
forward by Maxim Kontsevich. Therefore I will begin by reviewing this conjecture.
A weak Calabi-Yau manifold X is a compact complex manifold of Kähler type
whose canonical line bundle is trivial. In physics, weak Calabi-Yau manifolds are
usually equipped with additional structure: a Kähler form ω and a B-field B ∈
H 2 (X, R)/H 2 (X, Z). The triviality of the canonical line bundle is equivalent to the
requirement that the holonomy group of the Kähler metric belong to SU (n), where
n = dimC X. If the holonomy group is exactly SU (n), then according to standard
terminology X is called a Calabi-Yau manifold. But in this talk, for reasons of
brevity, I will use the term “Calabi-Yau” to mean “weak Calabi-Yau.”
It is believed that to any weak Calabi-Yau equipped with a Kähler form and
a B-field one can naturally associate an N = 2 superconformal field theory. An
N = 2 superconformal field theory is a special kind of quantum field theory in two
dimensions which describes propagation of superstrings on X. Its mathematical
definition is complicated (see e.g. [23, 8]). The most important feature of an
N = 2 SCFT is that its symmetry algebra contains the direct sum of two copies of
the N = 2 super-Virasoro algebra with equal central charges. These two copies are
referred to as left-moving and right-moving super-Virasoro algebras. In the case of
an N = 2 SCFT associated to a Calabi-Yau manifold the central charge is equal
to 3n. The N = 2 super-Virasoro algebra has a canonical N = 1 super-Virasoro
subalgebra, and a certain involution, called the mirror automorphism, which acts
as the identity on the N = 1 subalgebra. If two Calabi-Yaus are complex conjugate,
2000 Mathematics Subject Classification. Primary 14J32, 81T45; Secondary 53D40.

Key words and phrases. Mirror symmetry, topological field theory, D-branes.
The author was supported in part by the DOE grant DE-FG03-92-ER40701.
2004
337
338 ANTON KAPUSTIN
then the corresponding N = 2 SCFTs are “almost isomorphic”: they are isomorphic
as N = 1 SCFTs, but the isomorphism induces the mirror automorphism on both
copies of the N = 2 super-Virasoro algebra. The idea of mirror symmetry is “to
take a square root” of the operation of complex conjugation. That is, one says
that two Calabi-Yau manifolds are mirror if there exists an isomorphism of the
corresponding N = 1 SCFTs which acts by the mirror automorphism on one copy
of the N = 2 super-Virasoro algebra, and by the identity on the other one.
An explicit construction of the N = 2 SCFT corresponding to X is currently
possible only when X is a complex torus with a flat metric. The reason for this
is that N = 2 SCFTs are very complicated objects depending on the Ricci-flat
metric on X, whose explicit form is unknown. To facilitate the discussion of the
mirror relation, E. Witten introduced two “truncated” versions of N = 2 SCFT
called the A-model and the B-model [24, 25]. The truncation procedure used by
Witten is known as topological twisting. The main idea of topological twisting
is to regard a certain element in the super-Virasoro algebra (supercharge) as a
BRST operator, and focus on its cohomology. There are two essentially different
choices of the BRST operator, which are exchanged by the mirror automorphism,
so one gets two topologically twisted versions of the N = 2 SCFT. As indicated
by the name, topologically twisted theories are independent of the metric on the
2d world-sheet, i.e. they are topological field theories (TFT). 2d TFTs are much
simpler than 2d SCFTs; for a definition and discussion, see e.g. [2, 6]. From
the mathematical viewpoint, the notion of a 2d TFT is equivalent to the notion
of a super-commutative Frobenius algebra, i.e. a super-commutative algebra with
an invariant inner product. For example, the algebra corresponding to the A-
model of a Calabi-Yau X is the quantum cohomology ring of X, while the algebra
corresponding to the B-model is
⊕np,q=0 H p (Λq T 1,0 X).
The advantage of the “topologically twisted” viewpoint is that it allows one

to treat complex and symplectic aspects of a Calabi-Yau separately. That is, the
A-model is independent of the complex structure of X and the (0, 2) and (2, 0)
parts of the B-field, while the B-model is independent of the Kähler form ω and
the (1, 1) part of the B-field. Since the mirror automorphism exchanges the A and
B-twists, mirror symmetry exchanges the A and B-models. This observation is the
starting point for relating the counting of holomorphic curves in X with periods of
the mirror X .
In view of the above, mirror symmetry should be regarded as some kind of dual-
ity between complex and symplectic manifolds, which implies the equivalence of the
B and A-models of the respective manifolds. The origin of this duality was greatly
clarified by the Homological Mirror Symmetry Conjecture of M. Kontsevich [16].
Kontsevich proposed that the equivalence of B and A-models reflects an equivalence
between certain triangulated categories attached to complex and symplectic mani-
folds. On the complex side, the relevant category is the bounded derived category
of coherent sheaves on X, which we will denote Db (X).1 On the symplectic side,
1This statement is strictly true only if the (0, 2) part of the B-field is zero. Otherwise, one
has to consider the bounded derived category of “twisted” sheaves on X [12, 3]. It is denoted
D b (X, B).
THE GEOMETRY OF A-BRANES 339
the relevant category is the so-called derived Fukaya category DF (X). Its defini-
tion is due to K. Fukaya and M. Kontsevich and will be sketched below. One can
regard these categories as enriched versions of the B and A-models, respectively. It
is believed that the Homological Mirror Symmetry Conjecture completely captures
the mathematical meaning of the mirror relation. That is, Calabi-Yau manifolds X
and X are mirror if and only if Db (X) (more precisely, Db (X, B), see the preceding
footnote) is equivalent to the derived Fukaya category DF (X ), and vice versa.
Topological D-branes
Now let us discuss the physical interpretation of the Homological Mirror Sym-
metry Conjecture. From the physical viewpoint, a natural way to enrich the A
and B-models is to consider 2d TFTs with boundaries. Recall [2, 6] that a 2d
TFT assigns to any closed connected one-dimensional manifold (i.e. to a circle) a
vector space V , and to any oriented 2d bordism from a collection of k circles to a
collection of m circles a linear map from V ⊗k to V ⊗m . These maps must satisfy
certain compatibility conditions [2, 6]. To define a 2d TFT with boundaries, we
replace closed connected one-dimensional manifolds with compact connected one-
dimensional manifolds with boundaries. There are two such manifolds: a circle
and an interval. Thus we have two vector spaces, V and W . Instead of bordisms
between collections of circles, we now have bordisms between collections of cir-
cles and intervals. Whereas for usual 2d TFTs bordisms are compact oriented 2d
manifolds with boundaries, for 2d TFTs with boundaries bordisms are compact
oriented 2d manifolds with boundaries and “corners.” To any such bordism a 2d
TFT assigns a linear map from a tensor product of several copies of V and W to
another such tensor product. Whereas ordinary 2d TFT describes propagation of
closed strings (circles), a 2d TFT with boundaries describes propagation of open
and closed strings (intervals and circles). For a detailed discussion of 2d TFTs with
boundaries, see [21, 17].
In physics, 2d TFTs are defined by means of a path integral over spaces of fields
defined on the 2d world-sheet. In the case of a 2d TFT with boundaries, one must
specify boundary conditions on the fields. In most cases there are many different
choices of boundary conditions, and this suggests that the two endpoints of the
open string should not be treated symmetrically: the boundary conditions may be
different on the two endpoints. If we denote the set of boundary conditions for a
2d TFT by I, then we will have a vector space analogous to W for any ordered
pair (i, j) of elements of I. This strongly suggests that a 2d TFT with boundaries
produces a category, instead of an algebra. Objects of this category are boundary
conditions, while morphisms are vector spaces Wij . Using the axioms of the 2d
TFT with boundaries, one can show that this is indeed the case [17]. One usually
refers to boundary conditions for a 2d TFT as topological D-branes. There are also
non-topological D-branes, which are boundary conditions for a 2d SCFT which are
compatible with the super-Virasoro algebra.
Applying these observations to the A and B-models, we get two categories
of topological D-branes, which are called, naturally enough, the categories of A-
branes and B-branes. Alternatively, one can view A-branes and B-branes as non-
topological D-branes for the underlying N = 1 SCFT which are compatible with the
A and B-twist, respectively. Topological D-branes were introduced by E. Witten
[26]. He showed that the category of A-branes does not depend on the complex
340 ANTON KAPUSTIN
structure of X, while the category of B-branes is independent of the symplectic

structure. One can think of these categories as the enriched versions of the A and
B-models: while A and B-models describe closed strings, A and B-models with
boundaries describe closed and open strings with all possible boundary conditions.
Since the mirror automorphism exchanges the A and B-twists, it is clear that the
category of A-branes for X is equivalent to the category of B-branes for its mirror
X , and vice versa.
Now we are ready to explain Kontsevich’s conjecture in physical terms. Ac-
cording to Witten [26], holomorphic vector bundles are examples of B-branes,
and spaces of morphisms between vector bundles E and F are global Ext groups
Extk (E, F ) = H k (F ⊗E ∗ ) [26]. Further, Witten showed that examples of A-branes
are provided by Lagrangian submanifolds equipped with vector bundles with flat
connections, and spaces of morphisms between them are Floer homology groups
(the definition of the Floer homology groups is sketched below). Now, Ext groups
are spaces of morphisms in the derived category of coherent sheaves, therefore if
is natural to conjecture that arbitrary complexes of coherent sheaves are also ex-
amples of B-branes, and morphisms between them are morphisms in the derived
category. Similarly, it is reasonable to assume that arbitrary “complexes” of La-
grangian submanifolds with flat vector bundles (in the sense explained in [16]) are
examples of A-branes. The Homological Mirror Symmetry Conjecture is basically
the statement that all topological B-branes and A-branes arise in this way (as com-
plexes of coherent sheaves or complexes of Lagrangian submanifolds with flat vector
bundles). In other words, the Homological Mirror Symmetry Conjecture would fol-
low if we could prove that the category of A-branes (resp. B-branes) is equivalent
to DF (X) (resp. Db (X)).
Category of A-branes versus the Fukaya category

What is the current status of the Homological Mirror Symmetry Conjecture?
On the mathematical side, a certain modification of the conjecture has been proved
in the case when X and X are elliptic curves [22]. Unfortunately the methods
of [22] do not extend to higher-dimensional Calabi-Yaus. On the physical side,
there is now a good understanding of why complexes of coherent sheaves can be
regarded as B-branes and why morphisms between B-branes coincide with mor-
phisms in the derived category [7, 18, 1, 5, 14]. Similar arguments can be made
on the symplectic side to show that objects of the derived Fukaya category are valid
A-branes [19, 20]. There is also some evidence that the category of B-branes is
equivalent to the derived category of (twisted) coherent sheaves [12]. On the other
hand, it has been shown recently that the Fukaya category is only a full subcate-
gory of the category of A-branes [13], i.e. there are A-branes which are not related
to Lagrangian submanifolds. This implies that the symplectic side of Kontsevich’s
conjecture needs substantial modification. In this section we outline the argument
of [13] showing that in the case of flat tori there are not enough Lagrangian sub-
manifolds for the Homological Mirror Symmetry Conjecture to be true, and that
there are non-Lagrangian A-branes.
The idea is to work on the level of K-theory, and to show that for certain mirror
pairs X and X the K-theory of Db (X) is strictly bigger than the K-theory of the
Fukaya category DF (X ). In fact, to simplify life, we will tensor K-theory with
Q and use the Chern character to map the rational K-theory to an appropriate
cohomology group. In the case of Db (X), the Chern character takes values in the
intersection of H ∗ (X, Q) and ⊕np=0 H p,p (X), which are both subgroups of H ∗ (X, C).
(The Hodge conjecture says that the image of the Chern character map coincides
with this intersection.) In the case of the Fukaya category, the situation is somewhat
less clear. Mirror symmetry maps H ∗ (X, C) into H ∗ (X , C), therefore the Chern
character for the Fukaya category should take values in some subgroup of H ∗ (X, C).
The only obvious candidate for the Chern character of an object of the Fukaya
category is the Poincaré dual of the corresponding Lagrangian submanifold (taken
over Q). This guess can be physically motivated, and we will assume it in what
follows.
Now we will exhibit an example showing that the Fukaya category is not big
enough for the Homological Mirror Symmetry Conjecture to hold for all known
mirror pairs. Let E be an elliptic curve, e be an arbitrary point of E, and Ende (E)
be the ring of endomorphisms of E which preserve e. For a generic E we have
Ende (E) = Z, but for certain special E Ende (E) is strictly larger than Z. Such
special E’s are called elliptic curves with complex multiplication. One can show
that E has complex multiplication if and only if its Teichmüller parameter τ is a
root of a quadratic polynomial with integral coefficients. Let E be an elliptic curve
with complex multiplication. Consider the Abelian variety X = E n , n ≥ 2. One
can show that for such a variety the dimension of the image of the map

ch : K(Db (X)) ⊗ Q −→ H ∗ (X, Q) ⊕np=0 H p,p (X)
is
2n
dimQ Im(ch) = .
n
On the other hand, X is related by mirror symmetry to a symplectic torus X of
real dimension 2n. Cohomology classes Poincaré-dual to Lagrangian submanifolds
in X lie in the kernel of the map
∧ω
(1) H n (X , R) −−−−→ H n+2 (X , R).
This
map isan epimorphism,
and therefore the dimension of the kernel is equal
2n 2n
to − . Thus the image of the Chern character map for the Fukaya
n n+2
2n 2n
category of X has dimension less than or equal than − . Therefore
n n+2
K(DF (X )) cannot be isomorphic to K(Db (X)).
This leaves us with a question: if not all A-branes are Lagrangian submanifolds,
what are they? On the level of cohomology, if the Chern character of A-branes does
not take values in the kernel of the map Eq. (1), where does it take values? In the
case of flat tori, we can answer the second question. In this case we know that a
mirror torus is obtained by dualizing a Lagrangian sub-torus, and can infer how
the cohomology classes transform under this operation. The answer is the following
[10]. Suppose the original torus is of the form X = A × B, where A and B are
Lagrangian sub-tori, and the mirror torus is X = Â × B, where Â is the dual of
A. Consider a torus Z = A × Â × B. It has two obvious projections π and π to X
and X . On A × Â we also have the Poincaré line bundle whose Chern character
will be denoted P . Given a cohomology class α ∈ H ∗ (X, Q), we pull it back to
Z using π, multiply by P , and then push forward to X using π . This gives a
342 ANTON KAPUSTIN
cohomology class α ∈ H ∗ (X , Q) which is mirror to α. The requirement that α be

in the intersection of H ∗ (X, Q) and ⊕p H p,p (X) implies that α is orthogonal to the
class exp(iω) [10].2 Cohomology classes dual to Lagrangian submanifolds satisfy
this condition, but one can also give examples of coherent sheaves on X such that
α is not even a middle-dimensional cohomology class. For example, in [13] we have
constructed a holomorphic line bundle on X such that α has the form α = ea ,
where a ∈ H 2 (X , Z). This suggests that in this case the mirror of the holomorphic
line bundle on X is a line bundle on a symplectic torus X whose Chern character
ea is orthogonal to eiω . We will see in the next section that this guess is correct.
Coisotropic A-branes
To make further progress in understanding A-branes, we need to rely on phys-
ical arguments. As explained above, an A-brane is a boundary condition for an
N = 2 SCFT which is compatible with the A-twist. In [13] we analyzed this con-
dition assuming that an A-brane is a submanifold Y of a Kähler manifold X, and
carries a Hermitian line bundle E equipped with a unitary connection dE . (Our
analysis was on the classical level; some comments on possible quantum effects are
made below.) We showed that in order for a triple (Y, E, dE ) to be an A-brane, the
following three conditions are necessary and sufficient.
(i) Y must be a coisotropic submanifold of X. This means that the restriction of
the symplectic form ω to Y must have constant rank, and its kernel is an integrable
distribution LY ⊂ T Y . We will denote by N Y the quotient bundle T Y /LY . Note
that ω|Y is only a pre-symplectic form on the vector bundle T Y (i.e. it has non-
trivial kernel), but it descends to a symplectic form σ on the vector bundle N Y .
(ii) The curvature 2-form F = (2πi)−1 d2E , regarded as a bundle map from T Y
to T Y ∗ , annihilates LY . (The factor (2πi)−1 is included to make F a real 2-form
with integral periods). This implies that F descends to a section f of Λ2 (N Y ∗ ).
(iii) The symplectic form σ and the skew-symmetric bilinear form f, regarded
as bundle maps from N Y to N Y ∗ , satisfy (σ −1 f )2 = idN Y . This means that
J = σ −1 f is a complex structure on the bundle N Y .
Let us make some comments on these three conditions. The condition (i)
implies the existence of a foliation of Y whose dimension is equal to the codimension
of Y in X. It is known as the characteristic foliation of Y . The form σ is a basic
2-form with respect to it. If the characteristic foliation happens to be a fiber bundle
with a smooth base Z, then N Y is simply the pull-back of T Z to Y . In general,
N Y is a foliated vector bundle over the foliated manifold Y , and it makes sense to
talk about local sections of N Y locally constant along the leaves of the foliation.
It is useful to think of such sections as vector fields on the generally non-existent
quotient manifold Z. In the same spirit, the 2-form σ should be interpreted as
a symplectic form on Z. One can summarize the situation by saying that Y is a
foliated manifold with a transverse symplectic structure σ. N Y will be called the
transverse tangent bundle of Y .
The condition (ii) says that for any section v of LY we have iv F = 0. Since
dF = 0, this implies that the Lie derivative of F along such v vanishes. In other
words, F is also a basic 2-form. In the case when the characteristic foliation is a
fibration, this is equivalent to saying that f is a pull-back of a closed 2-form on the
base Z. In general, it is useful to think of f as a pull-back from a non-existent Z.
2We assume that the B-field vanishes, for simplicity.
The condition (iii) implies, first of all, that f is non-degenerate. Thus f is

another transverse symplectic structure on Y . Second, the condition (iii) says that
the ratio of the two transverse symplectic structures f and σ is a complex structure
on the transverse tangent bundle N Y . In other words, the ratio is a transverse
almost complex structure. If the characteristic foliation is a fibration with base Z,
then J = σ −1 f is simply an almost complex structure on Z.
An easy consequence of these conditions is that the dimension of Y must be
n + 2k, where n = 12 dimR X, and k is a non-negative integer. k has the meaning
of “transverse complex dimension.” If k = 0, then Y is a Lagrangian submanifold,
and the second condition forces F to vanish. (The third condition is vacuous in
this case). Another extreme case occurs when Y = X (this is possible only if n is
even). In this case the leaves of the characteristic foliation are simply points, the
second condition is vacuous, and the third condition says that ω −1 F is an almost
complex structure on X.
A less obvious property is that the transverse almost complex structure J is
integrable [13]. This follows easily from the well-known Gelfand-Dorfman theorem
[9] which plays an important role in the theory of integrable systems. Thus Y
is a transverse complex manifold, i.e. it has complex structure in the directions
transverse to the leaves of the characteristic foliation. It is also easy to see that
both f and σ have type (0, 2)+(2, 0) with respect to J. In fact, f +iσ is a transverse
holomorphic symplectic form on the transverse complex manifold Y .
The somewhat mysterious condition (iii) can be rewritten in several equivalent
forms. For example, an equivalent set of conditions is
∧r (f + iσ) = 0, r < k, ∧k (f + iσ) = 0.
Here k is related to the dimension of Y as above. This form is convenient for com-
parison with the conditions on the Chern character of A-branes on tori explained
above. For example, let us set n = 2 and let Y = X. In this case the above
conditions are equivalent to
F ∧ ω = 0, F ∧ F = ω ∧ ω.
On the level of cohomology, this is equivalent to the condition that the Chern
character eF be orthogonal to eiω .
It is not yet known how to generalize these considerations to the case when A-
branes carry a vector bundle of rank higher than one (except in the case when Y is
a Lagrangian submanifold, and E is flat). The difficult part is to find a meaningful
generalization of (iii). For general E the curvature 2-form is a form with values in
End(E), and an (almost) complex structure on N Y does not seem to arise. For
example, it is clear that if both (Y, E, dE ) and (Y, E , dE ) are rank-one A-branes
satisfying the conditions above, then their sum is also a valid A-brane. However,
instead of one transverse almost complex structure on Y, we now get two! If a
rank-two bundle E on Y does not decompose into a direct sum of two line bundles,
it is not clear what conditions on the curvature of dE ensure that it is an A-brane.
In principle, the physical definition of an A-brane as a boundary condition for a
sigma-model should enable one to find these conditions, but there are some technical
difficulties when the gauge group is non-Abelian.
So far our discussion of A-branes was classical. The main source of worry is the
possibility that N = 2 super-Virasoro used to define the A and B-twist is broken
by quantum anomalies. One can argue that any possible anomaly must arise from
344 ANTON KAPUSTIN
non-perturbative effects on the world-sheet, i.e. from the contributions to the path
integral from Riemann surfaces in X whose boundaries lie in Y and which cannot be
continuously deformed to a point. In the case when Y is Lagrangian, the conditions
for the absence of anomalies have been analyzed by K. Hori [11]. The result is that
there are no anomalies if and only if the so-called Maslov class of Y vanishes. Let
us recall how the Maslov class is defined. Let us choose a holomorphic section Ω of
the canonical line bundle (which is trivial for Calabi-Yau manifolds). Restricting
it to Y , we obtain a nowhere vanishing n-form. On the other hand, we also have a
volume form vol on Y , which comes from the Kähler metric on X. This is also a
nowhere vanishing n-form on Y , and therefore Ω|Y = h · vol, where h is a nowhere
vanishing complex function on Y . h can be thought of as an element of H 0 (C∗Y ),
where C∗Y is the sheaf of C∗ -valued functions on Y . The standard exponential
exact sequence gives a homomorphism from H 0 (C∗Y ) to H 1 (Y, Z), and the Maslov
class of Y is defined as the image of h under this homomorphism. (Explicitly, the
Čech cocycle representing the Maslov class is constructed as follows: choose a good
cover of Y , take the logarithm of h on each set of the cover, divide by 2πi, and
compare the results on double overlaps). Although the definition of the Maslov
class seems to depend both on the complex and symplectic structures on X, in fact
it is independent of the choice of complex structure. Note also that if the Maslov
class vanishes, the logarithm of h exists as a function, and is unique up to addition
of 2πim, m ∈ Z. A Lagrangian submanifold Y together with a choice of the branch
of log h is called a graded Lagrangian submanifold [16]. In the Fukaya category, all
Lagrangian submanifolds are graded.
For coisotropic Y the condition for anomaly cancellation is not yet known, but
there is an obvious guess. Let F be the curvature 2-form of the line bundle on Y ,
and let the dimension of Y be n + 2k, as before. It is easy to see that the (n + 2k)-
form Ω|Y ∧ F k is nowhere vanishing, and therefore we have Ω|Y ∧ F k = h · vol.
where vol is the volume form, and h is an element of H 0 (C∗Y ). We propose that the
vanishing of the image of h in H 1 (Y, Z) is the condition of anomaly cancellation.
It would be interesting to prove or disprove this guess.
In the case when X is a torus with a constant symplectic form, Y is an affine
sub-torus, and the curvature 2-form F is constant, one can quantize the sigma-
model and verify directly that the N = 2 super-Virasoro algebra is preserved on
the quantum level. This shows that non-Lagrangian A-branes exist on the quantum
level.
Now let us suppose that X is a non-toroidal Calabi-Yau manifold. The analysis
of [13] applies to such cases just as well, but it is harder to find explicit examples of
non-Lagrangian A-branes. If X has even complex dimension, we expect that there
exist non-Lagrangian A-branes with Y = X, i.e. vector bundles on X. For example,
let X be a hyperkähler manifold with three symplectic forms ω1 , ω2 , ω3 . If we set
ω = ω1 , we can solve the condition (iii) by setting F = ω1 cos θ + ω2 sin θ with
θ ∈ R. If one can find θ such that F has integral periods, then a line bundle with
curvature F is a non-Lagrangian A-brane on X. If X has odd complex dimension,
all non-Lagrangian A-branes must have non-zero codimension. As a particularly
simple example, let us take X to be a product of a K3 surface and an elliptic curve.
If K3 admits a non-Lagrangian A-brane as above (a line bundle), we may consider
a product of this A-brane on K3 with a 1-cycle on the elliptic curve. This brane has
real codimension 1, and one can easily check that all our conditions are satisfied.
In the most-often discussed case when X is a simply-connected Calabi-Yau

3-fold, there seem to be no non-Lagrangian A-branes of rank one. Indeed, our con-
ditions imply that such an A-brane would have real codimension one, and therefore
would be homologically trivial. We suspect that the situation remains the same for
A-branes of rank higher than one.
Morphisms between coisotropic A-branes

We hope that we have given convincing arguments that non-Lagrangian A-
branes exist, and must be included in order for the Homological Mirror Symmetry
Conjecture to be true. Unfortunately, we do not have a proposal for what should
replace the Fukaya category. In the remainder of this lecture we will describe ideas
which could help to solve this problem..
As we saw above, we know two kinds of A-branes. First, we have objects of the
Fukaya category, i.e. triples (Y, E, dE ) where Y is a graded Lagrangian submanifold,
E is a trivial vector bundle on Y with Hermitian metric, and dE is a flat unitary
connection of E.3 Second, we have triples (Y, E, dE ), where (E, dE ) is a Hermitian
line bundle on Y with a unitary connection, and the conditions (i)-(iii) are satisfied.
For objects of the Fukaya category we also know how to compute morphisms and
their compositions. Let us try to guess what the recipe should be for objects of the
second kind.
To begin with, let us recall how morphisms in the Fukaya category are defined.
Suppose we are given two objects (Y1 , E1 , d1 ) and (Y2 , E2 , d2 ). We will assume that
Y1 and Y2 intersect transversally at a finite number of points; if this is not the case,
we should deform one of the objects by flowing along a Hamiltonian vector field,
until the transversality condition is satisfied. Let I be the set of intersection points
of Y1 and Y2 . Now we consider the Floer complex. As a vector space, it is a direct
sum of vector spaces
Vi = Hom(E1 (ei ), E2 (ei )), i ∈ I.
The grading is defined as follows. At any point p ∈ Y the space Tp Y defines a
point q in the Grassmannian of Lagrangian planes in Tp X. Let us denote by Lag p
the universal cover of the Lagrangian Grassmannian of Tp X. On a Calabi-Yau X,
these spaces fit into a fiber bundle over X denoted by Lag [16]. Grading of Y

provides a canonical lift of q to Lagp for all p; these lifts assemble into a section
to Y [16]. Thus for each intersection point ei we have
of the restriction of Lag
e . The grade of the component of the Floer complex
a pair of points q1 , q2 ∈ Lag i
corresponding to ei is the Maslov index of q1 , q2 (see [4] for a definition of the
Maslov index.) Finally, we need to define the differential. Let ei and ej be a pair of
points whose grades differ by one. The component of the Floer differential which
maps Vi to Vj is defined by counting holomorphic disks in X with two marked
points, so that the two marked points are ei and ej (the Maslov index of ej is the
Maslov index of ei plus one), and the two intervals which make up the boundary of
the disks are mapped to Y1 and Y2 . Note that in order to compute the differential
one has to choose an (almost) complex structure J on X such that the form ω(·, J·)
is a Hermitian form on the tangent bundle of X. For a precise definition of the
3We assume that the B-field is trivial. The modification needed when B = 0 is explained in
[12].
346 ANTON KAPUSTIN
Floer differential, see [16]. The space of morphisms in the Fukaya category is
defined to be the Floer complex. The composition of morphisms can be defined
using holomorphic disks with three marked points and boundaries lying on three
Lagrangian submanifolds. It is associative only up to homotopy, given by a triple
product of morphisms. Actually, there is an infinite sequence of higher products in
the Fukaya category, which are believed to satisfy the identities of an A∞ category
(see [15] for a review of A∞ categories). It is also believed that changing the almost
complex structure J gives an equivalent A∞ category, so that the equivalence class
of the Fukaya category is a symplectic invariant.
What we need is a generalization of the Floer complex to non-Lagrangian A-
branes. To guess the right construction, the following heuristic viewpoint on the
Floer complex (due to A. Floer himself) is very useful. Consider the space of smooth
paths in X, which we will denote P X. This space is infinite-dimensional, but let
us treat it as if it were a finite-dimensional manifold. Since X is symplectic, we
have a natural 1-form α obtained by integrating ω along the path. More precisely,
if γ : I → X is a path, and β is a tangent vector to P X at the point γ (i.e. a vector
field along γ(t)), then the value of α on β is defined to be

·
ω(γ (t), β(t))dt.
I
Note that the space P X has two natural projections to X, which we denote π1 and
π2 . It is easy to see that dα = π1∗ ω − π2∗ ω. Thus α is closed if we restrict it to the
subspace of P X consisting of paths beginning and ending on isotropic submanifolds
of X.
In particular, let us consider the subspace P X(Y1 , Y2 ) consisting of paths which
begin at Y1 and end at Y2 . Since Y1 and Y2 are Lagrangian, the restriction of α
to P X(Y1 , Y2 ) is closed. Thus the operator d + 2πα on the space of differential
forms on P X(Y1 , Y2 ) squares to zero, and we may try to compute the corresponding
cohomology groups (in the finite-dimensional case this complex is called the twisted
de Rham complex.) Since P X(Y1 , Y2 ) is infinite-dimensional, it is not easy to make
sense of this. Floer solved this problem by a formal application of Morse theory
to this complex. Namely, the Morse-Smale-Witten-Novikov complex in this case is
precisely the Floer complex for the pair Y1 , Y2 .
This construction ignores the bundles E1 and E2 , but it is easy to take them into
account. We can pull back the bundles E1∗ and E2 , together with their connections,
to P X(Y1 , Y2 ) using π1 and π2 , respectively. This gives us a pair of Hermitian
vector bundles on P X(Y1 , Y2 ) with unitary flat connections. We tensor them, and
then add the 1-form 2πα to the connection on the tensor product. The resulting
connection is still flat, but no longer Hermitian. Finally, we formally apply Morse
theory to compute the cohomology of the resulting twisted de Rham complex on
P X(Y1 , Y2 ). This gives the Floer complex for a pair of objects of the Fukaya
category.
Now consider a pair of coisotropic A-branes, instead of a pair of Lagrangian
A-branes. We assume that the bundles E1 and E2 are line bundles, for reasons
discussed above. By P X(Y1 , Y2 ) we still denote the space of smooth curves in
X beginning at Y1 and ending on Y2 . The first difficulty, as compared to the
Lagrangian case, is that the restriction of α to P X(Y1 , Y2 ) is not closed, so we
cannot use it to define a complex. The second difficulty is that connections on E1
and E2 are not flat, and neither are their pull-backs to P X(Y1 , Y2 ). However, these
two difficulties cancel each other, as we will see in a moment. So let us proceed as
in the Lagrangian case: pull back (E2 , d2 ) by π2 , pull back the dual of (E1 , d1 ) by
π1 , tensor them, and add 2πα to the connection. The resulting connection on the
bundle on P X(Y1 , Y2 ) is not flat, but it has the following interesting property. Note
that since Y1 and Y2 are foliated manifolds with transverse complex structures, so
is P X(Y1 , Y2 ). A leaf of this foliation consists of all smooth paths in X which begin
on a fixed leaf in Y1 and end on a fixed leaf of Y2 . The codimension of the foliation is
finite and equal to the sum of the codimensions of the characteristic foliations of Y1
and Y2 . The connection on the bundle on P X(Y1 , Y2 ) has the following properties:
(a) it is flat along the leaves of the foliation;
(b) its curvature π2∗ (F2 + iω) − π1∗ (F1 + iω) has type (2, 0) in the transverse
directions.
According to property (b), if we consider sections of our bundle which are
covariantly constant along the leaves of the foliation, tensored with basic differential
forms, then the anti-holomorphic part of the transverse covariant derivative squares
to zero. Thus we have a natural substitute for the cohomology of the twisted de
Rham complex: the cohomology of the sheaf of sections which are covariantly
constant along the leaves and holomorphic in the transverse directions. This is our
formal proposal for the space of morphisms between a pair coisotropic A-branes.
This formal proposal must be properly interpreted, before one gets a concrete
recipe for computing spaces of morphisms. In the case when Y1 and Y2 are La-
grangian submanifolds, the above sheaf becomes the sheaf of covariantly constant
sections of a flat line bundle on P X(Y1 , Y2 ), and one can interpret its cohomology
using Morse-Smale-Witten-Novikov theory. We do not know how to make sense of
our formal proposal in general.
The difficulty of defining A-branes and morphisms between them in geometric
terms suggests that perhaps the approach based on Floer homology and its gen-
eralizations is not the right way to proceed. An analogy which comes to mind is
the notion of a holomorphic line bundle, as compared to the general notion of a
holomorphic vector bundle. One can study line bundles in terms of their divisors,
but this approach does not extend easily to higher rank bundles. Perhaps objects
of the Fukaya category, as well as coisotropic A-branes of rank one, are symplectic
analogues of divisors, and in order to make progress one has to find a symplectic
analogue of the notion of a holomorphic vector bundle (or a coherent sheaf). This
analogy is strengthened by the fact that both divisors and geometric representatives
of A-branes provide a highly redundant description of objects in the respective cate-
gories: line bundles correspond to divisors modulo linear equivalence, while objects
of the Fukaya category are unchanged by flows along Hamiltonian vector fields.
We believe that a proper understanding of the mirror phenomenon will require a
profound change in our viewpoint on the Fukaya category and its generalizations,
and that this new viewpoint may be important for symplectic geometry in general.
References
1. P. S. Aspinwall and A. E. Lawrence, Derived categories and zero-brane stability, JHEP 0108
(2001), 004 [arXiv:hep-th/0104147].
2. M. F. Atiyah, Topological quantum field theories, Publ. Math. IHES 68 (1988).
3. A. Caldararu, S. Katz, E. Sharpe, D-branes, B-fields, and Ext groups, arXiv:hep-th/0302099.
4. S. Cappell, R. Lee, and E. Y. Miller, On the Maslov index, Comm. Pure and Appl. Math.
47 (1994), 121.
348 ANTON KAPUSTIN
5. D. Diaconescu, Enhanced D-brane categories from string field theory, JHEP 0106 (2001),
016 [arXiv:hep-th/0104200].
6. R. Dijkgraaf, Intersection theory, integrable hierarchies and topological field theory,
arXiv:hep-th/9201003.
7. M. R. Douglas, D-branes, categories and N = 1 supersymmetry, J. Math. Phys. 42 (2001),
2818 [arXiv:hep-th/0011017].
8. K. Gawȩdzki, Lectures On Conformal Field Theory, preprint IHES-P-97-2.
9. I. M. Gelfand and I. Ya. Dorfman, The Schouten bracket and Hamiltonian operators, Funkts.
Anal. Prilozh. 14 (1980), 71.
10. V. Golyshev, V. Lunts, and D. Orlov, Mirror symmetry for Abelian varieties, J. Alg. Geom.
10 (2001), 433 [arXiv:math.ag/9812003].
11. K. Hori et al., Mirror Symmetry, Lectures at the Clay Mathematical School on Mirror
Symmetry, Harvard, Spring 2000, Clay Mathematics Monographs, 1 AMS (2003).
12. A. Kapustin and D. Orlov, Vertex algebras, mirror symmetry, and D-branes: The case of
complex tori, Commun. Math. Phys. 233 (2003), 79 [arXiv:hep-th/0010293].
13. A. Kapustin and D. Orlov, Remarks on A-branes, mirror symmetry, and the Fukaya cate-
gory, to appear in J. Geom. Phys., arXiv:hep-th/0109098.
14. S. Katz and E. Sharpe, D-branes, open string vertex operators, and Ext groups, arXiv:hep-
th/0208104.
15. B. Keller, Introduction to A∞ algebras and modules, arXiv:math.ra/9910179.
16. M.Kontsevich, Homological Algebra of Mirror Symmetry, talk at the International Congress
of Mathematicians, Zürich 1994, arXiv:alg-geom/9411018.
17. C. I. Lazaroiu, On the structure of open-closed topological field theory in two dimensions,
Nucl. Phys. B 603 (2001), 497 [arXiv:hep-th/0010269].
18. C. I. Lazaroiu, Unitarity, D-brane dynamics and D-brane categories, JHEP 0112 (2001),
031 [arXiv:hep-th/0102183].
19. C. I. Lazaroiu, Generalized complexes and string field theory, JHEP 0106 (2001), 052
20. C. I. Lazaroiu, Graded Lagrangians, exotic topological D-branes, and enhanced triangulated
categories, JHEP 0106 (2001), 064 [arXiv:hep-th/0105063].
21. G. Moore and G. B. Segal, unpublished; see also lectures by G. Moore at
http://online.itp.ucsb.edu/online/mp01.
22. A. Polishchuk and E. Zaslow, Categorical mirror symmetry: The Elliptic curve, Adv. Theor.
Math. Phys. 2 (1998), 443 [arXiv:math.ag/9801119].
23. G. B. Segal, The Definition Of Conformal Field Theory, in Differential Geometrical Methods
in Theoretical Physics, Como, 1987, pp. 165-171.
24. E. Witten, Topological Sigma Models, Commun. Math. Phys. 118 (1988), 411.
25. E. Witten, Mirror manifolds and topological field theory, in Essays on Mirror Manifolds, ed.
S. T. Yau, International Press, Hong Kong, 1991, pp. 120-158. [arXiv:hep-th/9112056]
26. E. Witten, Chern-Simons gauge theory as a string theory, Prog. Math. 133 (1995), 637
California Institute of Technology, Pasadena, CA 91125, USA

E-mail address: kapustin@theory.caltech.edu
Volume 3, 2004
Low Energy D-brane Actions
Robert C. Myers
Abstract. The low energy action describing the dynamics of D-branes con-
sists of two parts: the Born-Infeld action and the Wess-Zumino action. This
nonlinear action reliably captures the physics of D-branes with great accuracy.
One remarkable feature is the appearance of a non-Abelian gauge theory in the
description of several (nearly) coincident branes. This non-Abelian structure
plays an important role in realizing various geometric effects with D-branes.
In particular, the branes’ transverse displacements are described by matrix-
valued scalar fields and so noncommutative geometry naturally appears in this
framework.
1. Introduction
Dirichlet branes have played a central role in all of the major advances in
string theory in the past seven years [78]. A primary reason for this is that there
are many diverse perspectives from which these objects can be studied. Initially
D-branes were discovered within perturbative string theory as surfaces which sup-
ported open string excitations [33, 69, 70, 56, 107]. It was later found that the
same systems had a description in terms of gravitational solutions within low en-
ergy supergravity [71] — see also, e.g., [105, 92] for a more extensive review. More
recently, however, D-branes have been described in terms of boundary conformal
field theory, e.g., [50, 109, 111], K theory, e.g., [132, 96, 133], derived cate-
gories, e.g., [86, 42, 117], tachyon condensation, e.g., [115, 97, 62] and various
approaches from noncommutative geometry, e.g., [11, 27, 114]. This paper will
discuss the worldvolume action describing the low energy dynamics of D-branes.
This approach emerges naturally from the perturbative description of D-branes,
but it provides an exceptionally reliable framework to study D-branes and allows
one to develop useful physical intuition for these remarkable systems.
One of the most interesting aspects of the physics of D-branes is the appearance
of a non-Abelian gauge symmetry when several D-branes are brought together.
Of course, we understand that this symmetry emerges through the appearance
of new massless states corresponding to open strings stretching between the D-
branes [131]. Thus, while the number of light degrees of freedom is proportional
2000 Mathematics Subject Classification. 81T30, 81T75.

The author thanks David Winters for a thorough proofreading of this manuscript. This
research was supported in part by NSERC of Canada and Fonds FCAR du Québec.
2004
349
350 ROBERT C. MYERS
to N for N widely separated D-branes, this number grows like N2 for N coincident
D-branes. The non-Abelian symmetry and the rapid growth in massless states
are crucial elements in the statistical mechanical entropy counting for black holes
[105, 34, 35, 92], in Maldacena’s conjectured duality between type IIB superstrings
in AdS5 × S5 and four-dimensional N =4 super-Yang-Mills theory [2], and in the
development of M(atrix)-theory as a nonperturbative description of M-theory [120].
Referring to the worldvolume theory for N (nearly) coincident D-branes as a
non-Abelian U(N) gauge theory emphasizes the massless vector states, which one
might regard as internal excitations of the branes. Much of the present discussion
will focus, however, on the scalar fields describing the transverse displacements of
the branes. For coincident branes, these coordinate fields become matrix-valued,
appearing in the adjoint representation of the U(N) gauge group. These matrix-
valued coordinates then provide a natural framework in which to consider noncom-
mutative geometry. What has become evident, after a detailed study of the world-
volume action governing the dynamics of the non-Abelian U(N) theory [98, 123], is
that noncommutative geometries appear dynamically in many physical situations
in string theory. One finds that D-branes of one dimension metamorphose into
branes of a higher dimension through noncommutative configurations. These con-
figurations provide hints of dualities relating gauge theories in different dimensions,
and point to a symbiotic relationship between D-branes of all dimensions. The
emerging picture is reminiscent of the ‘brane democracy’ speculations in very early
investigations of D-branes [124, 79].
One important example of this brane transmogrification is the ‘dielectric effect’
in which a set of D-branes are polarized into a higher dimensional noncommutative
geometry by nontrivial background fields [98]. String theory seems to employ this
brane expansion mechanism in a variety of circumstances to regulate spacetime
singularities [106, 99, 10, 87]. So not only do string theory and D-branes provide
a natural physical framework for noncommutative geometry, but it also seems that
they may provide a surprising realization of the old speculation that noncommuta-
tive geometry should play a role in resolving the chaotic short-distance structure
of spacetime in quantum gravity.
An outline of this paper is as follows: First, section 2 presents a preliminary
discussion of the worldvolume actions governing the low energy physics of a single D-
brane (the Abelian case). Then section 3 discusses the extension to the non-Abelian
action, relevant for a system of several (nearly) coincident D-branes. Section 4
provides a brief outline of the dielectric effect for D-branes. In appendix A, we
give a short discussion of various aspects of noncommutative geometry and fuzzy
spheres, which in particular are relevant for the physical system described in the
last section.
2. Worldvolume D-brane actions

Within the framework of perturbative string theory, a Dp-brane is a (p + 1)-
dimensional extended surface in spacetime which supports the endpoints of open
strings [78]. The massless modes of this open string theory form a supersymmetric
U(1) gauge theory with a vector Aa , 9 − p real scalars Φi and their superpartner
fermions — for the most part, the latter are ignored throughout the following
discussion. At leading order, the low-energy action corresponds to the dimensional
reduction of that for ten-dimensional U(1) super-Yang-Mills theory. However, as
LOW ENERGY D-BRANE ACTIONS 351
is usual in string theory, there are higher order α = 2s corrections — s is the
string length scale. For constant field strengths, these stringy corrections can be
resummed to all orders, and the resulting action takes the Born-Infeld form [88]

(2.1) SBI = −Tp dp+1 σ e−φ −det(P [G + B]ab + λ Fab ) ,
where Tp is the Dp-brane tension and λ denotes the inverse of the (fundamental)
string tension, i.e., λ = 2π2s . This Born-Infeld action describes the couplings of
the Dp-brane to the massless Neveu-Schwarz fields of the bulk closed string theory,
i.e., the (string-frame) metric Gµν , dilaton φ and Kalb-Ramond two-form Bµν . The
symbol P [. . .] denotes the pull-back of the bulk spacetime tensors to the D-brane
worldvolume.
The interactions of the Dp-brane with the massless Ramond-Ramond (RR)
fields are incorporated in a second part of the action, the Wess-Zumino term [45,
89, 57]

(2.2) SW Z = µp P C (n) eB eλ F .
Here C (n) denote the n-form RR potentials. Eq. (2.2) shows that a Dp-brane is
naturally charged under the (p+1)-form RR potential, with charge µp , and super-
symmetry dictates that µp = ±Tp . If we consider the special case of the D0-brane
(a point particle), the Born-Infeld action reduces to the familiar worldline action of
a point particle, where the action is proportional to the proper length of the particle
trajectory. Actually, the string theoretic D0-brane action is not quite this simple
geometric action, rather it is slightly embellished with the additional coupling to
the dilaton which appears as a prefactor to the standard Lagrangian density. (Note,
however, that the tensors B and F drop out of the action since the determinant
is implicitly over a one-dimensional matrix.) Turning to the Wess-Zumino action,
we see that a D0-brane couples to C (1) (a vector). Then eq. (2.2) reduces to the
familiar coupling of a Maxwell field to the worldline of a point particle, i.e.,

dxµ
(2.3) µ0 P C (1) q Aµ dτ .
dτ
Higher dimensional Dp-branes can also support a flux of B + F , which compli-
cates the worldvolume actions above. From eq. (2.2), we see that such a flux allows
a Dp-brane to act as a charge source for RR potentials with a lower form degree
than p+1 [45]. Such configurations represent bound states of D-branes of different
dimensions [131]. To illustrate this point, let us assume that B vanishes and then
we may expand the Wess-Zumino action (2.2) for a D4-brane as

λ2 (1)
(2.4) µ4 C + λC ∧F + C ∧ F ∧ F .
(5) (3)
2
Hence, the D4-brane is naturally a source for the five-form potential C (5) . However,
by introducing a worldvolume gauge field with a nontrivial first Chern class, i.e.,
exciting a nontrivial magnetic flux on the worldvolume, the D4-brane also sources
C (3) , which is the potential associated with D2-branes. Hence a D4-brane with
magnetic flux is naturally interpreted as a D4-D2 bound state. Similarly a D4-
brane that supports a gauge field with a nontrivial second Chern class will source
the vector potential C (1) and is interpreted as a D4-D0 bound state.
352 ROBERT C. MYERS
Hence, as already alluded to above, the Born-Infeld action (2.1) has a geometric
interpretation, i.e., it is essentially the proper volume swept out by the Dp-brane,
which is indicative of the fact that D-branes are actually dynamical objects. This
dynamics becomes more evident with an explanation of the static gauge choice
implicit in constructing the above action. To begin, we employ spacetime diffeo-
morphisms to position the worldvolume on a fiducial surface defined as xi = 0 with
i = p + 1, . . . , 9. With worldvolume diffeomorphisms, we then match the worldvol-
ume coordinates with the remaining spacetime coordinates on this surface, σ a = xa
with a = 0, 1, . . . , p. Now the worldvolume scalars Φi play the role of describing
the transverse displacements of the D-brane, through the identification
(2.5) xi (σ) = 2π2s Φi (σ) with i = p + 1, . . . , 9.
With this identification the general formula for the pull-back reduces to
∂xµ ∂xν
(2.6) P [E]ab = Eµν
∂σ a ∂σ b
= Eab + λ Eai ∂b Φi + λ Eib ∂a Φi + λ2 Eij ∂a Φi ∂b Φj .
In this way, the expected kinetic terms for the scalars emerge to leading order in an
expansion of the Born-Infeld action (2.1). Note that our conventions are such that
both the gauge fields and worldvolume scalars have the dimensions of length−1 —
hence the appearance of the string scale in eq. (2.5).
Although it was mentioned above, we want to stress that these worldvolume
actions are low energy effective actions for the massless states of the open and
closed strings, which incorporate interactions from all disk amplitudes (all orders
of tree level for the open strings). The Born-Infeld action was originally derived [88]
using standard beta function techniques applied to worldsheets with a boundary
[1, 21, 22]. In principle, they could also be derived from a study of open and closed
string scattering amplitudes and it has been verified that this approach yields the
same interactions to leading order [52, 51, 64]. As a low energy effective action
then, eqs. (2.1) and (2.2) include an infinite number of stringy corrections, which
essentially arise through integrating out the massive modes of the string — see the
discussion in [59]. For example, consider the Born-Infeld action evaluated for a flat
Dp-brane in empty Minkowski space

SBI −Tp dp+1 x −det(ηab + 2π2s Fab )

(2π2s )2 2
(2.7) −Tp dp+1 x 1 + F + (2π2s )4 F 4 + (2π2s )6 F 6 + · · ·
4
where the precise structure of the F 4 and F 6 terms may be found in [126, 60] and
[85], respectively. Hence, as well as the standard kinetic term for the worldvolume
gauge field, eq. (2.7) includes an infinite series of higher dimension interactions,
which are suppressed as long as the typical components 2s Fab (referred to an or-
thonormal frame) are small. The square-root expression in the first line resums this
infinite series and so one may consider arbitrary values of 2s Fab in working with
this action. However, the full effective action would also include stringy correction
terms involving derivatives of the field strength, e.g., ∂a Fbc or ∂a ∂b Fcd — see, for
example, [85] or [134]. None of these have been incorporated in the Born-Infeld
action and so one must still demand that the variations in the field strength are
relatively small, e.g., components of s ∂a Fbc are much smaller than those of Fab .
Of course, this discussion extends in the obvious way to derivatives of the scalar
fields Φi .
At this point, we should also note that the bulk supergravity fields appearing in
eqs. (2.1) and (2.2) are in general functions of all of the spacetime coordinates, and
so they are implicitly functionals of the worldvolume scalars. In static gauge, the
bulk fields are evaluated in terms of a Taylor series expansion around the fiducial
surface xi = 0. For example, the metric functional appearing in the D-brane action
would be given by

(2.8) Gµν = exp λΦi ∂xi G0µν (σ a , xi )|xi =0
∞
λ n i1
= Φ · · · Φin (∂xi1 · · · ∂xin )G0µν (σ a , xi )|xi =0 .
n=0
n!
Hence the worldvolume action implicitly incorporates an infinite class of higher
dimension interactions involving derivatives of the bulk fields as well. However,
beyond this class of interactions incorporated in eqs. (2.1) and (2.2), once again the
full effective action includes other higher derivative bulk field corrections [57, 24,
8, 31, 32, 113]. It is probably fair to say that the precise domain of validity of
the D-brane action from the point of view of the bulk fields is poorly understood.
3. Non-Abelian D-brane action

As N parallel D-branes approach each other, the ground state modes of strings
stretching between the different D-branes become massless. These extra mass-
less states carry the appropriate charges to fill out representations under a U(N)
symmetry. Hence the U(1)N of the individual D-branes is enhanced to the non-
Abelian group U(N) for the coincident D-branes [131]. The vector Aa becomes a
non-Abelian gauge field
(3.1) Aa = A(n)
a Tn , Fab = ∂a Ab − ∂b Aa + i[Aa , Ab ] ,
where Tn are N2 hermitian generators with Tr(Tn Tm ) = N δnm . Central to the
following is that the scalars Φi are also matrix-valued, transforming in the adjoint
of U(N). The covariant derivative of the scalar fields is given by
(3.2) Da Φi = ∂a Φi + i[Aa , Φi ] .
Understanding how to accommodate this U(N) gauge symmetry in the world-
volume action is an interesting puzzle. For example, the geometric meaning (or
even the validity) of eq. (2.5) or (2.8) seems uncertain when the scalars on the right
hand side are matrix-valued. In fact, the identification of the scalars as transverse
displacements of the branes does remain roughly correct. Some intuition comes
from the case where the scalars are commuting matrices and the gauge symmetry
can be used to simultaneously diagonalize all of them. In this case, one interprets
the N eigenvalues of the diagonal Φi as representing the displacements of the N
constituent D-branes — see, e.g., [120]. Further the gauge symmetry may be used
to simultaneously interchange any pair of eigenvalues in each of the scalars and so
to ensure that the branes are indistinguishable. Of course, to describe noncommu-
tative geometries, we will be more interested in the case where the scalars do not
commute and so cannot be simultaneously diagonlized.
In refs. [98] and [123], progress was made in constructing the worldvolume
action describing the dynamics of non-Abelian D-branes. The essential strategy
354 ROBERT C. MYERS
in both of these papers was to construct an action which was consistent with the
familiar string theory symmetry of T-duality [55]. Acting on D-branes, T-duality
acts to change the dimension of the worldvolume [78]. The two possibilities are:
(i) if a coordinate transverse to the Dp-brane, e.g., y = xp+1 , is T-dualized, it
becomes a D(p + 1)-brane where y is now the extra worldvolume direction; and
(ii) if a worldvolume coordinate on the Dp-brane, e.g., y = xp , is T-dualized, it
becomes a D(p−1)-brane where y is now an extra transverse direction. Under these
transformations, the roles of the corresponding worldvolume fields change according
to
(3.3) (i) Φp+1 → Ap+1 , (ii) Ap → Φp ,
while the other scalars, Φi , and the remaining components of A are left unchanged.
Hence, in constructing the non-Abelian action, one can begin with the D9-brane
theory, which contains no scalars since the worldvolume fills the entire spacetime.
In this case, the non-Abelian extension of eqs. (2.1) and (2.2) is given by simply
introducing an overall trace over gauge indices of the non-Abelian field strengths
appearing in the action [41, 40]. Then applying T-duality transformations on
9 − p directions yields the non-Abelian action for a Dp-brane. Of course, in this
construction, one also T-dualizes the background supergravity fields according to
the known transformation rules [55, 18, 19, 20, 16, 94]. As in the Abelian
theory, the result for the non-Abelian action has two distinct pieces [98, 123]: the
Born-Infeld term

SBI = −Tp d σ STr e−φ det(Qi j )
p+1

(3.4) −1
× − det (P [Eab + Eai (Q − δ) Ejb ] + λ Fab ) ,
ij
with Eµν = Gµν + Bµν and Qi j ≡ δ i j + iλ [Φi , Φk ] Ekj ; and the Wess-Zumino term

(3.5) SW Z = µp STr P eiλ iΦ iΦ ( C (n) eB ) eλ F .
Let us enumerate the non-Abelian features of this action:

1. Non-Abelian field strength: The Fab appearing explicitly in both terms is now
non-Abelian, of course.
2. Non-Abelian Taylor expansion: The bulk supergravity fields are again functions
of all of the spacetime coordinates, and so they are implicitly functionals of the
non-Abelian scalars. In the action given by eqs. (3.4) and (3.5), these bulk fields
are again interpreted in terms of a Taylor expansion as in eq. (2.8); however the
transverse displacements are now matrix-valued.
3. Non-Abelian Pullback : As was noted in refs. [72, 39], the pullbacks of various
spacetime tensors to the worldvolume must now involve covariant derivatives of
the non-Abelian scalars in order to be consistent with the U(N) gauge symmetry.
Hence eq. (2.6) is replaced by
(3.6) P [E]ab = Eab + λ Eai Db Φi + λ Eib Da Φi + λ2 Eij Da Φi Db Φj .
4. Non-Abelian Interior Product: In the Wess-Zumino term (3.5), iΦ denotes the
interior product with Φi regarded as a vector in the transverse space, e.g., acting
(n)
on an n-form C (n) = 1
n! Cµ1 ···µn dx
µ1
· · · dxµn , we have
1 (n)
(3.7) iΦ iΦ C (n) = [Φi , Φj ] Cjiµ3 ···µn dxµ3 · · · dxµn .
2(n − 2)!
Note that when acting on forms, the interior product is an anticommuting operator
and hence for an ordinary vector (i.e., a vector v i with values in R9−p ): iv iv C (n) = 0.
It is only because the scalars Φ are matrix-valued that eq. (3.7) yields a nontrivial
result.
5. Non-Abelian Gauge Trace: As is evident above, both parts of the action are highly
nonlinear functionals of the non-Abelian fields, and so eqs. (3.4) and (3.5) would
be incomplete without a precise definition for the ordering of these fields under
the gauge trace. Above, STr denotes the maximally symmetric trace [127]. To
be precise, the trace includes a symmetric average over all orderings of Fab , Da Φi ,
[Φi , Φj ] and the individual Φk appearing in the non-Abelian Taylor expansions of
the background fields. This choice matches that inferred from Matrix theory [121],
and a similar symmetrization arises in the leading order analysis of the boundary
beta functions [39]. However, we should note that with this definition an expansion
of the Born-Infeld term (2.1) does agree with the string theory to fourth order in
F [127, 128], but it does not seem to capture the full physics of the non-Abelian
fields in the infrared limit at higher orders [65] — we expand on this point below.
Some general comments on the non-Abelian action are as follows: In the Born-
Infeld term (3.4), there are now two determinant factors as compared to one in the
Abelian action (2.1). The second determinant in eq. (3.4) is a slightly modified
version of that in eq. (2.1). One might think of this as the kinetic factor, since to
leading order in the low energy expansion it yields the familiar kinetic terms for the
gauge field and scalars. In the same way, one can think of the new first factor as the
potential factor, since to leading order in the low energy expansion it reproduces
the non-Abelian scalar potential expected for the super-Yang-Mills theory — see
eq. (4.1) below. Further, note that the first factor reduces to simply one when the
scalar fields are commuting, even for general background fields.
The form of the action and in particular the functional dependence of the
bulk fields on the adjoint scalars can be verified in a number of independent
ways. Douglas [41, 40] observed on general grounds that, whatever their form,
the non-Abelian worldvolume actions should contain a single gauge trace, as do
both eqs. (3.4) and (3.5). This observation stems from the fact that the action
should encode only the low energy interactions derivable from disk amplitudes in
superstring theory. Since the disk has a single boundary, the single gauge trace
arises from the standard open string prescription of tracing over Chan-Paton fac-
tors on each worldsheet boundary. Further then, one may note [41, 40] that the
only difference in the superstring amplitudes between the U(1) and the U(N) the-
ories is that the amplitudes in the latter case are multiplied by an additional trace
of Chan-Paton factors. Hence, up to commutator ‘corrections’, the low energy in-
teractions should be the same in both cases. Hence, since the background fields
are functionals of the neutral U(1) scalars in the Abelian theory, they must be
precisely the same functionals of the adjoint scalars in the non-Abelian theory, up
to commutator corrections. The interactions involving the non-Abelian inner prod-
uct in the Wess-Zumino action (3.5) provide one class of commutator corrections.
The functional dependence on the adjoint scalars also agrees with the linearized
356 ROBERT C. MYERS
couplings for the bulk fields derived from Matrix theory [121]. As an aside, we
would add that the technology developed in Matrix theory remains useful in gain-
ing intuition and manipulating these non-Abelian functionals [80, 122, 100, 101].
Finally we add that by the direct examination of string scattering amplitudes using
the methods of refs. [52] and [64], one can verify at low orders the form of the
non-Abelian interactions in eqs. (3.4) and (3.5), including the appearance of the
new commutator interactions in the non-Abelian Wess-Zumino action [53, 54].
As noted above, the symmetric trace perscription is known not to agree with the
full effective string action [65]. Rather, at sixth order and higher in the worldvolume
field strength, additional terms involving commutators of field strengths must be
added to the action [9]. The source of this shortcoming is clear. Recall from
the discussion following eq. (2.7) that in the Abelian action we have disgarded all
interactions involving derivatives of the field strength. However, this prescription
is ambiguous in the non-Abelian theory since
(3.8) [Da , Db ]Fcd = i[Fab , Fcd ] .
It is clear that with the symmetric trace we have eliminated all derivative terms,
including those antisymmetric combinations that might contribute commutators
of field strengths. One might choose to improve the action by reinstating these
commutators. This problem has been extensively studied and the commutator cor-
rections at order F 6 are known [9]. Considering the supersymmetric extension
of the non-Abelian action [38, 116, 83, 15, 85] has lead to a powerful iterative
technique relying on stable holomorphic bundles [84, 36, 25] which seems to pro-
vide a constructive approach to determine the entire effective open string action,
including all higher derivative terms and fermion contributions as well. A similar
iterative procedure seems to emerge from studying the Seiberg-Witten map [114]
in the context of a noncommutative worldvolume theory [30].
As described below eq. (2.2), an individual Dp-brane couples not only to the
RR potential with form degree n = p + 1, but also to the RR potentials with
n = p − 1, p − 3, . . . through the exponentials of B and F appearing in the Wess-
Zumino action (2.2). Above in eq. (3.5), iΦ iΦ is an operator of form degree –2, and
so worldvolume interactions appear in the non-Abelian action (3.5) involving the
higher RR forms. Hence in the non-Abelian theory, a Dp-brane can also couple
to the RR potentials with n = p + 3, p + 5, . . . through the additional commutator
interactions. To make these couplings more explicit, consider the D0-brane action
(for which F vanishes):

SCS = µ0 STr P C (1) + iλ iΦ iΦ C (3) + C (1) B

λ2 1
− (iΦ iΦ )2 C (5) + C (3) B + C (1) B 2
2 2
3

λ 1 (3) 2 1 (1) 3
(3.9) −i (iΦ iΦ ) C + C B + C B + C B
3 (7) (5)
6 2 6
4

λ 4 (9) (7) 1 (5) 2 1 (3) 3 1 (1) 4
+ (iΦ iΦ ) C + C B + C B + C B + C B .
24 2 6 24
Of course, these interactions are reminiscent of those appearing in Matrix theory
[11, 12]. For example, eq. (3.9) includes a linear coupling to C (3) , which is the
potential corresponding to D2-brane charge,

iλ µ0 Tr P iΦ iΦ C (3)

λ (3) (3)
(3.10) = i µ0 dt Tr Ctjk (Φ, t) [Φk , Φj ] + λCijk (Φ, t) Dt Φi [Φk , Φj ] ,
2
where we assume that σ 0 = t in static gauge. Note that the first term on the
right hand side has the form of a source for D2-brane charge. This is essentially
the interaction central to the construction of D2-branes in Matrix theory with the
large N limit [11, 12]. Here, however, with finite N, this term would vanish upon
(3)
taking the trace if Ctjk was simply a function of the worldvolume coordinate t
(since [Φ , Φ ] ∈ SU(N) ). However, in general these three-form components are
k j
functionals of Φi . Hence, while there would be no ‘monopole’ coupling to D2-brane

charge, nontrivial expectation values of the scalars can give rise to couplings to an
infinite series of higher ‘multipole’ moments [80, 122].
4. Dielectric Branes
In this section, we wish to consider certain physical effects arising from the new
non-Abelian interactions in the worldvolume action, given by eqs. (3.4) and (3.5).
To begin, consider the scalar potential for Dp-branes in flat space, i.e., Gµν = ηµν
with all other fields vanishing. In this case, the entire scalar potential originates in
the Born-Infeld term (3.4) as

Tp λ 2
(4.1) V = Tp Tr det(Qi j ) = NTp − Tr([Φi , Φj ] [Φi , Φj ]) + . . .
4
The commutator-squared term corresponds to the potential for ten-dimensional
U(N) super-Yang-Mills theory reduced to p + 1 dimensions. A nontrivial set of
extrema of this potential is given by taking the 9 − p scalars as constant commuting
matrices, i.e.,
(4.2) [Φi , Φj ] = 0
for all i and j. Since they are commuting, the Φi may be simultaneously diago-
nalized and, as discussed above, the eigenvalues are interpreted as the separated
positions in the transverse space of N fundamental Dp-branes. This solution reflects
the fact that a system of N parallel Dp-branes is supersymmetric, and so they can
sit in static equilibrium with arbitrary separations in the transverse space [78].
From the results described in the previous section, it is clear that in going
from flat space to general background fields, the scalar potential is modified by new
interactions and so one should reconsider the analysis of the extrema. It turns out
that this yields an interesting physical effect that is a precise analog for D-branes
of the dielectric effect in ordinary electromagnetism. That is, when Dp-branes are
placed in a nontrivial background field for which the Dp-branes would normally be
regarded as neutral, e.g., nontrivial F (n) with n > p + 2, new terms will be induced
in the scalar potential, and generically one should expect that there will be new
extrema beyond those found in flat space, i.e., eq. (4.2). In particular, there can
be nontrivial extrema with noncommuting expectation values of the Φi , e.g., with
Tr Φi = 0 but Tr(Φi )2 = 0. This would correspond to the external fields ‘polarizing’
the Dp-branes to expand into a (higher dimensional) noncommutative worldvolume
geometry. This is the analog of the familiar electromagnetic process where an
358 ROBERT C. MYERS
external field may induce a separation of charges in neutral materials. In this

case, the polarized material will then carry an electric dipole (and possibly higher
multipoles). The latter is also seen in the D-brane analog. When the worldvolume
theory is at a noncommutative extremum, the gauge traces of products of scalars
will be nonvanishing in various interactions involving the supergravity fields. Hence,
at such an extremum, the Dp-branes act as sources for the latter bulk fields.
To make these ideas explicit, we will now illustrate the process with a simple
example. We consider N D0-branes in a constant background RR field F (4) , i.e., the
field strength associated with D2-brane charge. We find that the D0-branes expand
into a noncommutative two-sphere which represents a spherical bound state of a
D2-brane and N D0-branes.
Consider a background where only the RR four-form field strength is nonvan-
ishing, with
(4)
(4.3) Ftijk = −2f εijk for i, j, k ∈ {1, 2, 3} ,
where f is a constant (of dimensions length−1 ). Since F (4) = dC (3) , we must con-
sider the coupling of the D0-branes to the RR three-form potential, which is given
above in eq. (3.10). If one explicitly introduces the non-Abelian Taylor expansion
(2.8), one finds that the leading order interaction may be written as

i 2 (4)
(4.4) λ µ0 dt Tr Φi Φj Φk Ftijk (t) .
3
This final form might have been anticipated since one should expect that the world-
volume potential can only depend on gauge invariant expressions of the background
field. Given that we are considering a constant background F (4) , the higher order
terms implicit in eq. (3.10) will vanish, as they can only involve spacetime deriva-
tives of the four-form field strength. Combining eq. (4.4) with the leading order
Born-Infeld potential (4.1) yields the scalar potential of interest for the present
problem
λ 2 T0 i (4)
(4.5) V (Φ) = NT0 − Tr([Φi , Φj ]2 ) − λ2 µ0 Tr Φi Φj Φk Ftijk (t) .
4 3
Substituting in the (static) background field (4.3) and µ0 = T0 , the extremisa-
tion condition δV (Φ)/δΦi = 0 yields
(4.6) 0 = [[Φi , Φj ], Φj ] + i f εijk [Φj , Φk ] .
Note that commuting matrices (4.2) describing separated D0-branes still solve this
equation. The value of the potential for these solutions is simply V0 = NT0 , the
mass of N D0-branes. Another interesting solution of eq. (4.6) is
f i
(4.7) Φi = α
2
where the αi are any N×N matrix representation of the SU(2) algebra
(4.8) [αi , αj ] = 2i εijk αk .
For the moment, let us focus on the irreducible representation for which one finds
N 2
(4.9) Tr[(αi )2 ] = (N − 1) for i = 1, 2, 3.
3
Now evaluating the value of the potential (4.5) for this new solution yields

T0 λ 2 f 2
3
π 2 3s f 4 3 1
(4.10) VN = NT0 − Tr[(Φi )2 ] = NT0 − N 1− 2
6 i=1
6g N
using T0 = 1/(gs ). Hence the noncommutative solution (4.7) has lower energy than
a solution of commuting matrices, and so the latter configuration of separated D0-
branes is unstable towards condensing out into this noncommutative solution. One
can also consider reducible representations of the SU(2) algebra (4.8); however, one
finds that the corresponding energy is always larger than that in eq. (4.10). Hence it
seems that the irreducible representation describes the ground state of the system.
Geometrically, one can recognize the SU(2) algebra as that corresponding to
the noncommutative or fuzzy two-sphere [68, 37, 90]. The physical size of the
fuzzy two-sphere is given by
3 1/2 1/2
1
(4.11) R=λ i 2
Tr[(Φ ) ]/N = πs f N 1 − 2
2
i=1
N
in the ground state solution. From the Matrix theory construction of Kabat and
Taylor [23], one can infer that this ground state is not simply a spherical arrange-
ment of D0-branes, rather the noncommutative solution actually represents a spher-
ical D2-brane with N D0-branes bound to it. In the present context, the latter can
be verified by seeing that this configuration has a ‘dipole’ coupling to the RR four-
form field strength. The precise form of this coupling is calculated by substituting
the noncommutative scalar solution (4.7) into the worldvolume interaction (4.4),
which yields
−1/2
R3 1 (4)
(4.12) − 1 − dt Ft123
3πg3s N2
for the ground state solution. Physically this F (4) -dipole moment arises because
antipodal surface elements on the sphere have the opposite orientation and so form
small pairs of separated membranes and anti-membranes. Of course, the spherical
configuration carries no net D2-brane charge.
Given that the noncommutative ground state solution corresponds to a bound
state of a spherical D2-brane and N D0-branes, one might attempt to match the
above results using the dual formulation. That is, this system can be analyzed
from the point of view of the (Abelian) worldvolume theory of a D2-brane. In this
case, one would consider a spherical D2-brane carrying a flux of the U(1) gauge
field strength representing the N bound D0-branes and, at the same time, sitting
in the background of the constant RR four-form field strength (4.3). In fact, one
does find stable static solutions, but what is more surprising is how well the results
match those calculated in the framework of the D0-branes. The results for the
energy, radius and dipole coupling are the same as in eqs. (4.10), (4.11) and (4.12),
respectively, except that the factors of (1 − 1/N2 ) are absent [98]. Hence, for large
N, the two calculations agree up to 1/N2 corrections.
One expects that the D2-brane calculations would be valid when R s while
naively the D0-brane calculations would be valid when R s . Hence it appears
there is no common domain where the two pictures can both produce reliable re-
sults. However, a more careful consideration
√ of range of validity of the D0-brane
calculations only requires that R Ns . This estimate is found by requiring that
360 ROBERT C. MYERS
the scalar field commutators appearing in the full non-Abelian potential (4.1) are
small so that the Taylor expansion of the square root converges rapidly. Hence, for
large N, there is a large domain of overlap where both of the dual pictures are reli-
able. Note that the density of D0-branes on the two-sphere
√ is N/(4πR2 ). However,
even if R is macroscopic it is still bounded by R Ns and so this density must
be large compared to the string scale, i.e., the density is much larger than 1/2s .
With such large densities, one can imagine the discreteness of the fuzzy sphere is
essentially lost and so there is good agreement with the continuum sphere of the
D2-brane picture. More discussion on the noncommutative geometry appears in
Appendix A.
Finally note that the Born-Infeld action contains couplings to the Neveu-
Schwarz
two-form which are similar to that in eq. (4.4). From the expansion of
det(Q), one finds a cubic interaction

i 2
(4.13) λ T0 dt Tr Φi Φj Φk Hijk (t) .
3

Hence the noncommutative ground state, which has Tr Φi Φj Φk = 0, also acts as
a source of the B field through the worldvolume coupling

R03 1
(4.14) − 1 − dt H123 .
3πg3s N2
This coupling is perhaps not so surprising given that the noncommutative ground
state represents the bound state of a spherical D2-brane and N D0-branes. Explicit
supergravity solutions describing D2-D0 bound states with a planar geometry have
been found [17, 110], and are known to carry a long-range H field with the same
profile as the RR field strength F (4) . One can also derive this coupling from the dual
D2-brane formulation. Furthermore, we observe that the presence of this coupling
(4.13) means that we would find an analogous dielectric effect if the N D0-branes
were placed in a constant background H field. This mechanism plays a role in
describing D-branes in the spacetime background corresponding to a WZW model
[7, 102, 91, 5, 3, 4, 47]. It seems that quantum group symmetries may be useful
in understanding these noncommutative configurations [103, 104].
The example considered above must be considered simply a toy calculation
demonstrating the essential features of the dielectric effect for D-branes. A more
complete calculation would require analyzing the D0-branes in a consistent super-
gravity background. For example, the present case could be extended to consider
the asymptotic supergravity fields of a D2-brane, where the RR four-form would
be slowly varying but the metric and dilaton fields would also be nontrivial. Al-
ternatively, one can find solutions with a constant background F (4) in M-theory,
namely the AdS4 ×S7 and AdS7 ×S4 backgrounds — see, e.g., [46]. In lifting the
D0-branes to M-theory, they become gravitons carrying momentum in the inter-
nal space. Hence the expanded D2-D0 system considered here corresponds to the
‘giant gravitons’ of ref. [93]. The analog of the D2-D0 bound state in a constant
background F (4) corresponds to M2-branes with internal momentum expanding
into AdS4 [58, 63], while that in a constant H field corresponds to the M2-branes
expanding on S4 [93]. The original analysis of these M-theory configurations was
made in terms of the Abelian world-volume theory of the M2-brane [93] — a Ma-
trix theory description of such states in terms of noncommutative geometry was
only developed recently [77, 6]. Further, one finds that M5-branes will expand
in a similar way for these backgrounds, and that expanded D3-branes arise in the
type IIB supergravity background AdS5 × S5 . Remarkably, a detailed analysis
[58, 93, 63] shows that these expanded branes are BPS states with the quantum
numbers of a graviton. Ref. [95] extends this discussion to more general expanded
configurations.
Alternatively, the dielectric effect has been found to play a role in other string
theory contexts, for example, in the resolution of certain singularities in the AdS/-
CFT correspondence [106]. Further, one can consider more sophisticated back-
ground field configurations, which, through the dielectric effect, generate more
complicated noncommutative geometries [125, 48]. There is also an interesting
generalization to open dielectric branes, in which the extended brane emerging from
the dielectric effect ends on another D-brane [129]. Other interesting applications
of the dielectric effect for D-branes can be found in refs. [73, 74, 75].
Appendix A. Noncommutative geometry

The idea that noncommutative geometry should play a role in physical theories
is an old one [119, 118, 26]. Suggestions have been made that such noncommuta-
tive structure may resolve the ultraviolet divergences of quantum field theories, or
appear in the description of spacetime geometry at the Planck scale. In the past
few years, it has also become a topic of increasing interest to string theorists. From
one point of view, the essential step in realizing a noncommutative geometry is
replacing the spacetime coordinates by noncommuting operators: xµ → x̂µ . In this
replacement, however, there remains a great deal of freedom in defining the non-
trivial commutation relations which the operators x̂µ must satisfy. Some explicit
choices that have appeared in physical problems are as follows:
(i) Canonical commutation relations:
[x̂µ , x̂ν ] = iθ µν , θ µν ∈ C .
Such algebras have appeared in the Matrix theory description of planar D-branes
[11] — for a review, see [120]. This work also stimulated an ongoing investigation
by string theorists of noncommutative field theories, which arise in the low energy
limit of a planar D-brane with a constant B-field flux — see, e.g., [114, 27, 43, 44].
(ii) Quantum space relations:
x̂µ x̂ν = q −1 Rµν ρτ x̂ρ x̂τ , Rµν ρτ ∈ C .
These algebras received some attention from physicists in the early 1990’s — see,
e.g., [130, 112] — and have appeared more recently in the geometry of the moduli
space of N =4 super-Yang-Mills theory [13, 14].
(iii) Lie algebra relations:
[x̂µ , x̂ν ] = if µν ρ x̂ρ , f µν ρ ∈ C .
Such algebras naturally arise in the description of fuzzy spheres, as was discovered
in early attempts to quantize the supermembrane [68, 37]. These noncommutative
geometries have also been applied in Matrix theory to describe spherical D-branes
[23, 81]. As discussed in the main text, noncommutative geometries with a Lie-
algebra structure and these noncommutative spheres, in particular, arise very nat-
urally in various D-brane systems since the transverse scalars are matrix-valued in
the adjoint representation of U(N).
362 ROBERT C. MYERS
Beginning with a prescribed set of commutation relations for the coordinates

on a given manifold, the bulk of the problem in noncommutative geometry is to
understand the algebra of functions in this framework. Of course, mathematicians
are typically careful in defining the specific class of functions with which they
wish to work; however, these details are usually glossed over in physical models.
That is, as physicists, we are usually confident that the physics will guide the
choice of functions. For a fuzzy sphere, one finds that not only is the product
structure modified but that the space of functions is naturally truncated to be
finite dimensional. The remainder of the discussion in the appendix will focus on
these noncommutative spheres [68, 37, 90], in part because they have been found
to play an interesting physical role in string theory — see, e.g., [98, 28, 29, 81, 23].
We also elaborate on these examples because, in contrast to the above discussion,
the natural presentation given below de-emphasizes the role of the commutation
relations. In particular, the fuzzy four-sphere will provide an intriguing example
below.
To begin the construction of a fuzzy sphere, we begin with the standard defi-
nition of a k-sphere using the embedding in (k+1)-dimensional Cartesian space

k+1
i 2
(A.1) x = R2 , xi ∈ Rk+1 .
i=1
Now, functions on Sk can be expanded in terms of spherical harmonics as

∞

(A.2) f (xi ) = fi1 ···i xi1 · · · xi ,
=0
where fi1 ···i are completely symmetric and traceless tensors. We could be more
precise in defining a basis of these tensors, but here we will be satisfied with noting
that each term in the sum is a linear combination of spherical harmonics with prin-
cipal quantum number . Showing this is straightforward: Denoting the individual
terms in the sum as f and setting aside the constraint (A.1), it is clear that the
Laplacian on the Cartesian space annihilates each of these terms, i.e., ∇2 f = 0.
Now, in spherical polar coordinates on Rk+1 , the Laplacian may be written as

(A.3) ∇2 = R−k ∂R Rk ∂R + R−2 ∇2Ω ,
where ∇2Ω is the angular Laplacian on the unit k-sphere. Hence it follows that
∇2Ω f = ( + k − 1)f .
In general, to produce a fuzzy sphere, one might proceed by replacing the
k+1 continuum coordinates above by finite dimensional matrices, xi → x̂i , whose
commutation relations we leave aside for the moment. The matrices are chosen to
satisfy a constraint analogous to eq. (A.1):

k+1
i 2
(A.4) x̂ = R2 1N .
i=1
Similarly the continuum functions are replaced by

max
(A.5) fˆ(x̂i ) = fi1 ···i x̂i1 · · · x̂i ,

=0
where fi1 ···i are the same symmetric and traceless tensors considered in (A.2).
Notice that the ‘noncommutative’ sum is truncated at some max because for fi-
nite dimensional matrices, such products will only yield a finite number of linearly
independent matrices. Thus this matrix construction truncates the full algebra of
functions on the sphere to those with ≤ max , and the star product on the fuzzy
sphere differs from that obtained by the deformation quantization of the Poisson
structure on the embedding space, i.e., the latter acts on the space of all square
integrable functions on the sphere [49, 66].
The simplest example of this construction is the fuzzy two-sphere, which was
already encountered in section 4. In this case, one chooses x̂i = λ αi with i = 1, 2, 3,
where the αi are the generators of the irreducible N×N representation of SU(2)
satisfying the commutation relations given in eq. (4.8). These generators satisfy
the Casimir relation
2
(A.6) αi = (N2 − 1) 1N
and so, in order to satisfy the constraint (A.4), the normalization constant should
be chosen as
R
(A.7) λ= √ .
N2 − 1
With these N×N matrices, one finds the cutoff in eq. (A.5) is max = N − 1. So
heuristically, we might say that with this construction we can only resolve distances
on the noncommutative sphere for ∆d > ∼ R/N . Hence, in the limit N → ∞, one
expects to get agreement with the continuum theory, as was illustrated with the
physical models in the main text.
We should mention that the entire space of functions (A.5) plays a role in the
stringy constructions. To illustrate this point, we consider the example of the fuzzy
two-sphere appearing in the example of the dielectric effect in section 4. In the static
ground state configuration, three of the transverse scalars have a noncommutative
expectation value: Φi = (f /2) αi for i = 1, 2, 3. Now one might consider excitations
of this system. In particular, it is natural to expand fluctuations of the scalars in
terms of the noncommutative spherical harmonics
(A.8) δΦm (t) = ψim1 ...i (t) αi1 · · · αi
where as above the coefficients ψim1 ···i are completely symmetric and traceless. In
the case of overall transverse scalars, i.e., m = 0, 1, 2, 3, the linearized equations of
motion reduce to
∂t2 δΦm (t) = −[Φi , [Φi , δΦm (t)]]
(A.9) = −( + 1)f 2 δΦm (t) .
Hence inserting the ansatz δΦm (t) ∝ e−iωt , we find excitations with frequencies
ω 2 = ( + 1)f 2 . Of course, these fluctuations inherit the cutoff ≤ max from
the noncommutative framework. The analysis of the fluctuations in the i = 1, 2, 3
directions is more involved but a nice description is given in [76]. In the interesting
regime√where the dielectric calculations
√ are expected to be valid, i.e., large N and
R Ns , we have f 1/ Ns . Hence the low energy excitations are well
below the string scale, giving further corroboration that the low energy effective
action provides an adequate description of the physics [98]. We might add that
these frequencies match the results found from calculations in the dual continuum
364 ROBERT C. MYERS
framework of the expanded D2-brane, again up to 1/N2 corrections. Of course,

the latter description gives no upper cutoff on the angular momentum. A similar
discussion [28] applies for the excitations of the D3⊥D1-system.
We now turn to a discussion of the fuzzy four-sphere, which is relevant for the
construction of the D-string description of the D5⊥D1 system [29]. This construc-
tion also played a role in ref. [23] for the Matrix theory description of spherical
D4-branes (longitudinal M5-branes). At first sight, the construction of the fuzzy
four-sphere appears very similar to the fuzzy two-sphere above, but in fact it yields
a very different object. One begins by choosing x̂i = λ Gi where the Gi are an
appropriate set of N×N matrices. These matrices were first constructed in ref. [61]
— see also ref. [23]. The Gi are given by the totally symmetric n-fold tensor
product of 4 × 4 gamma matrices:

(A.10) G i = Γi ⊗ 1 ⊗ · · · ⊗ 1 + 1 ⊗ Γi ⊗ 1 ⊗ · · · ⊗ 1

+ · · · + 1 ⊗ · · · ⊗ 1 ⊗ Γi Sym ,
where Γi , i = 1, . . . , 5, are 4×4 Euclidean gamma matrices, and 1 is the 4×4 identity
matrix. The subscript ‘Sym’ means the matrices are restricted to the completely
symmetric tensor product space. With the latter restriction, the dimension of the
matrices becomes
(n + 1)(n + 2)(n + 3)
(A.11) N= ,
6
where n is the integer denoting the size of the tensor product in eq. (A.10). The
‘Casimir’ associated with the Gi matrices, i.e., Gi Gi = c 1N , is given by
(A.12) c = n(n + 4) .
Hence, to satisfy eq. (A.4), we choose the normalization constant in x̂i = λGi as
λ = R/ n(n + 4).
These matrices were presented as a representation of the fuzzy four-sphere on
the basis of a discussion of representations of SO(5) in ref. [61]. Working within
Matrix theory, ref. [23] provided a series of physical arguments towards the same
end. That is, the Gi produce a spherical locus, are rotationally invariant under the
action of SO(5) and give an appropriate spectrum of eigenvalues.
With these Gi , one can construct matrix harmonics (A.5) with ≤ max = n,
as before [61]. However, a key difference between the fuzzy two-sphere and the
fuzzy four-sphere is that the Gi do not form a Lie algebra (in contrast to the αi
used to construct the fuzzy two-sphere). As a result the algebra of these matrix
harmonics does not close! [23, 61] In particular, one finds that the commutators
Gij ≡ [Gi , Gj ]/2 define linearly independent matrices. The commutators of the Gi
and Gjk are easily obtained:1
[Gij , Gk ] = 2(δ jk Gi − δ ik Gj ),
[Gij , Gkl ] = 2(δ jk Gil + δ il Gjk − δ ik Gjl − δ jl Gik ).
Note then that the Gik are the generators of SO(5) rotations. Hence, combined,
the Gi and Gij give a representation of the algebra SO(1, 5), as can be seen from
the definition of the Gij and the commutators in eq. (A.13). Hence a closed algebra
of matrix functions would given by
(A.13) a1 G
ãa a ···a G a2 · · · G
a ,
1 2
1We refer the interested reader to refs. [29, 23] for more details.
where the G a are generators of SO(1, 5) with a = 1, . . . , 15, and the ã are naturally
symmetric in the SO(1, 5) indices. Identifying G a = Ga for a = 1 . . . 5, the desired
matrix harmonics would correspond to the subset of ã with nonvanishing entries
only for indices ai ≤ 5. Thus, while the fuzzy four-sphere construction introduces an
algebra that contains a truncated set of the spherical harmonics on S 4 , the algebra
also contains a large number of elements transforming under other representations
of the SO(5) symmetry group that acts on the four-sphere. The reader may find a
precise description of the complete algebra in ref. [61], in terms of representations
of SO(5) (or rather Spin(5) = Sp(4)).
Given this extended algebra, or alternatively the appearance of ‘spurious’ modes,
one might question whether the above construction provides a suitable noncommu-
tative description of the four-sphere. Nonetheless, in the context of non-Abelian
D-branes (or Matrix theory), the Gi certainly form a basis for the description of
physically interesting systems such as a spherical D4-D0 bound state [23] or a
D5⊥D1 system [29]. In the latter case, N D-strings open up into a collection of n
perpendicular D5-branes and the fuzzy four-sphere forms the cross-section of the
funnel describing this geometry. From the point of view of the D5-brane theory,
the four-sphere is endowed with a nontrivial SU(n) bundle. In this context, one
can show that the relation between N and n in eq. (A.11) essentially arises from
demanding that the gauge field configuration is homogeneous on the four-sphere
[29]. Given that there are extra instantonic or gauge field degrees of freedom in the
full physical system, it is natural that the ‘spurious’ modes above should be related
to non-Abelian excitations of the D5-brane theory. Evidence for this identification
can be found by studying the linearized excitations of the fuzzy funnel describing
the D5⊥D1 system and their couplings to the bulk RR fields [29]. A thorough
analysis of the noncommutative geometry [108, 67, 82] also indicates that this
identification is correct. Hence this fuzzy four-sphere construction has a natural
physical interpretation within string theory.
References
1. A. Abouelsaood, C. G. Callan, C. R. Nappi, and S. A. Yost, Open strings in background
gauge fields, Nuclear Phys. B 280 (1987), no. 4, 599–624.
2. Ofer Aharony, Steven S. Gubser, Juan Maldacena, Hirosi Ooguri, and Yaron Oz, Large N
field theories, string theory and gravity, Phys. Rep. 323 (2000), no. 3-4, 183–386.
3. Anton Yu. Alekseev, Andreas Recknagel, and Volker Schomerus, Non-commutative world-
volume geometries: branes on SU(2) and fuzzy spheres, J. High Energy Phys. (1999), no. 9,
Paper 23, 20 pp. (electronic).
4. , Brane dynamics in background fluxes and non-commutative geometry, J. High En-
ergy Phys. (2000), no. 5, Paper 10, 25.
5. Anton Yu. Alekseev and Volker Schomerus, D-branes in the WZW model, Phys. Rev. D (3)
60 (1999), no. 6, 061901, 2.
6. B. Janssen, Y. Lozano, and D. Rodriguez-Gomez, A microscopical description of giant gravi-
tons. II: The AdS5 × S 5 background, Nucl. Phys. B 669 (2003) 363-378.
7. Constantin Bachas, Michael R. Douglas, and Christoph Schweigert, Flux stabilization of
D-branes, J. High Energy Phys. (2000), no. 5, Paper 48, 18.
8. Constantin P. Bachas, Pascal Bain, and Michael B. Green, Curvature terms in D-brane
actions and their M-theory origin, J. High Energy Phys. (1999), no. 5, Paper 11, 30 pp.
(electronic).
9. Pascal Bain, On the non-Abelian Born-Infeld action, Progress in string theory and M-
theory (Dordrecht, The Netherlands) (Marco Picco, Laurent Baulieu, Michael Green and
Paul Windey, eds.), Kluwer, 2001.
366 ROBERT C. MYERS
10. Vijay Balasubramanian and Asad Naqvi, Giant gravitons and a correspondence principle,
Phys. Lett. B 528 (2002), no. 1-2, 111–120.
11. T. Banks, W. Fischler, S. H. Shenker, and L. Susskind, M theory as a matrix model: a
conjecture, Phys. Rev. D (3) 55 (1997), no. 8, 5112–5128.
12. Tom Banks and Nathan Seiberg, Strings from matrices, Nuclear Phys. B 497 (1997), no. 1-2,
41–55.
13. David Berenstein, Vishnu Jejjala, and Robert G. Leigh, Marginal and relevant deformations
of N = 4 field theories and non-commutative moduli spaces of vacua, Nuclear Phys. B 589
(2000), no. 1-2, 196–248.
14. , Non-commutative moduli spaces, dielectric tori and T -duality, Phys. Lett. B 493
(2000), no. 1-2, 162–168.
15. E. A. Bergshoeff, M. de Roo, and A. Sevrin, Non-Abelian Born- Infeld and kappa-symmetry,
J. Math. Phys. 42 (2001), no. 7, 2872–2888.
16. Eric Bergshoeff, Chris Hull, and Tomás Ortı́n, Duality in the type-II superstring effective
action, Nuclear Phys. B 451 (1995), no. 3, 547–575.
17. J. C. Breckenridge, G. Michaud, and R. C. Myers, More D-brane bound states, Phys. Rev.
D (3) 55 (1997), no. 10, 6438–6446.
18. T. H. Buscher, Quantum Corrections and Extended Supersymmetry in New Sigma Models,
Phys. Lett. B 159 (1985), 127.
19. , A symmetry of the string background field equations, Phys. Lett. B 194 (1987),
no. 1, 59–62.
20. , Path-integral derivation of quantum duality in nonlinear sigma-models, Phys. Lett.
B 201 (1988), no. 4, 466–472.
21. C. G. Callan, C. Lovelace, C. R. Nappi, and S. A. Yost, Adding holes and crosscaps to the
superstring, Nuclear Phys. B 293 (1987), no. 1, 83–113.
22. , Loop corrections to superstring equations of motion, Nuclear Phys. B 308 (1988),
no. 2-3, 221–284.
23. Judith Castelino, Sangmin Lee, and Washington Taylor, Longitudinal 5-branes as 4-spheres
in Matrix theory, Nuclear Phys. B 526 (1998), no. 1-3, 334–350.
24. Yeuk-Kwan E. Cheung and Zheng Yin, Anomalies, branes, and currents, Nuclear Phys. B
517 (1998), no. 1-3, 69–91.
25. Andres Collinucci, Mees de Roo, and Martijn G. C. Eenink, Supersymmetric Yang-Mills
theory at order α3 , J. High Energy Phys. (2002), no. 6, no. 24, 16.
26. Alain Connes, Noncommutative geometry, Academic Press Inc., San Diego, CA, 1994.
27. Alain Connes, Michael R. Douglas, and Albert Schwarz, Noncommutative geometry and
matrix theory: compactification on tori, J. High Energy Phys. (1998), no. 2, Paper 3, 35 pp.
(electronic).
28. Neil R. Constable, Robert C. Myers, and Øyvind Tafjord, Noncommutative bion core, Phys.
Rev. D (3) 61 (2000), no. 10, 106009, 14.
29. , Non-Abelian brane intersections, J. High Energy Phys. (2001), no. 6, Paper 23, 37.
30. L. Cornalba, On the general structure of the non-Abelian Born- Infeld action, Adv. Theor.
Math. Phys. 4 (2000), no. 6, 1259–1318.
31. Ben Craps and Frederik Roose, Anomalous D-brane and orientifold couplings from the
boundary state, Phys. Lett. B 445 (1998), no. 1-2, 150–159.
32. , (Non-)anomalous D-brane and O-plane couplings: the normal bundle, Phys. Lett.
B 450 (1999), no. 4, 358–362.
33. Jin Dai, R. G. Leigh, and Joseph Polchinski, New connections between string theories, Mod-
ern Phys. Lett. A 4 (1989), no. 21, 2073–2083.
34. S.R. Das and S.D. Mathur, The quantum physics of black holes: Results from string theory,
Ann. Rev. Nucl. Part. Sci. 50 (2000), no. 6, 153–206.
35. Justin R. David, Gautam Mandal, and Spenta R. Wadia, Microscopic formulation of black
holes in string theory, Phys. Rep. 369 (2002), no. 6, 549–686.
36. Mees de Roo, Martijn G. C. Eenink, Paul Koerber, and Alexander Sevrin, Testing the
fermionic terms in the non-Abelian D-brane effective action through order α3 , J. High
Energy Phys. (2002), no. 8, no. 11, 9.
37. B. de Wit, J. Hoppe, and H. Nicolai, On the quantum mechanics of supermembranes, Nuclear
Phys. B 305 (1988), no. 4, FS23, 545–581.
38. Frederik Denef, Alexander Sevrin, and Jan Troost, Non-Abelian Born-Infeld versus string
theory, Nuclear Phys. B 581 (2000), no. 1-2, 135–155.
39. H. Dorn, Non-Abelian gauge field dynamics on matrix D-branes, Nuclear Phys. B 494 (1997),
no. 1-2, 105–118.
40. Michael R. Douglas, D-branes in curved space, Adv. Theor. Math. Phys. 1 (1997), no. 2,
198–209.
41. , D-branes and Matrix theory in curved space, Nuclear Phys. B Proc. Suppl. 68
(1998), 381–393.
42. , D-branes, categories and N = 1 supersymmetry, J. Math. Phys. 42 (2001), no. 7,
2818–2843.
43. Michael R. Douglas and Chris Hull, D-branes and the noncommutative torus, J. High Energy
Phys. (1998), no. 2, Paper 8, 5 pp. (electronic).
44. Michael R. Douglas and Nikita A. Nekrasov, Noncommutative field theory, Rev. Modern
Phys. 73 (2001), no. 4, 977–1029.
45. M.R. Douglas, Branes within Branes, hep-th/9512077, in Strings, Branes and Dualities
(Cargèse 1997) 267-275.
46. M. J. Duff, TASI lectures on branes, black holes and anti-de Sitter space, Strings, branes
and gravity, TASI 99 (Boulder, CO), World Sci. Publishing, River Edge, NJ, 2001, pp. 3–125.
47. Stefan Fredenhagen and Volker Schomerus, Branes on group manifolds, gluon condensates,
and twisted K-theory, J. High Energy Phys. (2001), no. 4, Paper 7, 30.
48. , D-branes in coset models, J. High Energy Phys. (2002), no. 2, No. 5, 27.
49. Laurent Freidel and Kirill Krasnov, The fuzzy sphere -product and spin networks, J. Math.
Phys. 43 (2002), no. 4, 1737–1754.
50. Jürgen Fuchs and Christoph Schweigert, Branes: from free fields to general backgrounds,
Nuclear Phys. B 530 (1998), no. 1-2, 99–136.
51. Mohammad R. Garousi, Superstring scattering from D-branes bound states, J. High Energy
Phys. (1998), no. 12, Paper 8, 28 pp. (electronic).
52. Mohammad R. Garousi and Robert C. Myers, Superstring scattering from D-branes, Nuclear
Phys. B 475 (1996), no. 1-2, 193–224.
53. , World-volume interactions on D-branes, Nuclear Phys. B 542 (1999), no. 1-2, 73–88.
54. , World-volume potentials on D-branes, J. High Energy Phys. (2000), no. 11, Paper
32, 28.
55. A. Giveon, M. Porrati, and E. Rabinovici, Target space duality in string theory, Phys. Rep.
244 (1994), no. 2-3, 77–202.
56. Michael B. Green, Space-time duality and Dirichlet string theory, Phys. Lett. B 266 (1991),
no. 3-4, 325–336.
57. Michael B. Green, Jeffrey A. Harvey, and Gregory Moore, I-brane inflow and anomalous
couplings on D-branes, Classical Quantum Gravity 14 (1997), no. 1, 47–52.
58. Marcus T. Grisaru, Robert C. Myers, and Øyvind Tafjord, SUSY and Goliath, J. High
Energy Phys. (2000), no. 8, Paper 40, 28.
59. David J. Gross and John H. Sloan, The quartic effective action for the heterotic string,
Nuclear Phys. B 291 (1987), no. 1, 41–89.
60. David J. Gross and Edward Witten, Superstring modifications of Einstein’s equations, Nu-
clear Phys. B 277 (1986), no. 1, 1–10.
61. H. Grosse, C. Klimčı́k, and P. Prešnajder, On finite 4D quantum field theory in non-
commutative geometry, Comm. Math. Phys. 180 (1996), no. 2, 429–438.
62. Jeffrey A. Harvey and Per Kraus, D-branes as unstable lumps in bosonic open string field
theory, J. High Energy Phys. (2000), no. 4, Paper 12, 14.
63. Akikazu Hashimoto, Shinji Hirano, and N. Itzhaki, Large branes in AdS and their field theory
dual, J. High Energy Phys. (2000), no. 8, Paper 51, 20.
64. Akikazu Hashimoto and Igor R. Klebanov, Scattering of strings from D-branes, Nuclear
Phys. B Proc. Suppl. 55B (1997), 118–133.
65. Akikazu Hashimoto and Washington Taylor, Fluctuation spectra of tilted and intersecting
D-branes from the Born-Infeld action, Nuclear Phys. B 503 (1997), no. 1-2, 193–219.
66. E. Hawkins, Noncommutative regularization for the practical man, Tech. Report hep-
th/9908052, August 1999.
67. Pei-Ming Ho and Sanjaye Ramgoolam, Higher-dimensional geometries from matrix brane
constructions, Nuclear Phys. B 627 (2002), no. 1-2, 266–288.
368 ROBERT C. MYERS
68. Jens Hoppe, Quantum theory of a massless relativistic surface and a two-dimensional
bound state problem, Ph.D. thesis, MIT, July 1982 (available at http://www.aei-
potsdam.mpg.de/ hoppe).
69. Petr Hořava, Background duality of open-string models, Phys. Lett. B 231 (1989), no. 3,
251–257.
70. , Strings on world-sheet orbifolds, Nuclear Phys. B 327 (1989), no. 2, 461–484.
71. Gary T. Horowitz and Andrew Strominger, Black strings and p-branes, Nuclear Phys. B 360
(1991), no. 1, 197–209.
72. Chris M. Hull, Matrix theory, U -duality and toroidal compactifications of M-theory, J. High
Energy Phys. (1998), no. 10, Paper 11, 22 pp. (electronic).
73. Yoshifumi Hyakutake, Torus-like dielectric D2-brane, J. High Energy Phys. (2001), no. 5,
Paper 13, 22.
74. , Expanded strings in the background of NS5-branes via a M2-brane, a D2-brane and
D0-branes, J. High Energy Phys. (2002), no. 1, No. 21, 31.
75. , Notes on the construction of the D2-brane from multiple D0-branes, Nucl. Phys. B
675 (2003) 241-269.
76. Satoshi Iso, Yusuke Kimura, Kanji Tanaka, and Kazunori Wakatsuki, Noncommutative gauge
theory on fuzzy sphere from matrix model, Nuclear Phys. B 604 (2001), no. 1-2, 121–147.
77. B. Janssen and Y. Lozano, A microscopical description of giant gravitons, Nuclear Phys. B
658 (2003), 281.
78. Clifford Johnson, D-branes, Cambridge University Press, Cambridge, 2002.
79. Clifford V. Johnson, Nemanja Kaloper, Ramzi R. Khuri, and Robert C. Myers, Is string
theory a theory of strings?, Phys. Lett. B 368 (1996), no. 1-2, 71–77.
80. Daniel Kabat and Washington Taylor, Linearized supergravity from Matrix theory, Phys.
Lett. B 426 (1998), no. 3-4, 297–305.
81. , Spherical membranes in M(atrix) theory, Adv. Theor. Math. Phys. 2 (1998), no. 1,
181–206.
82. Yusuke Kimura, Noncommutative gauge theory on fuzzy four-sphere and matrix model, Nu-
clear Phys. B 637 (2002), no. 1-3, 177–198.
83. Paul Koerber and Alexander Sevrin, The non-Abelian open superstring effective action
through order α3 , J. High Energy Phys. (2001), no. 10, Paper 3, 30.
84. , Testing the α3 term in the non-Abelian open superstring effective action, J. High
Energy Phys. (2001), no. 9, Paper 9, 13.
85. , The non-Abelian D-brane effective action through order α4 , J. High Energy Phys.
(2002), no. 10, 46.
86. Maxim Kontsevich, Homological algebra of mirror symmetry, Proceedings of the Interna-
tional Congress of Mathematicians, Vol. 1, 2 (Zürich, 1994) (Basel), Birkhäuser, 1995,
pp. 120–139.
87. Frédéric Leblond, Robert C. Myers, and David C. Page, Superstars and giant gravitons in
M-theory, J. High Energy Phys. (2002), no. 1, No. 26, 16.
88. R. G. Leigh, Dirac-Born-Infeld Action from Dirichlet Sigma Model, Modern Phys. Lett. A
4 (1989), no. 21, 2767.
89. Miao Li, Boundary states of D-branes and dy-strings, Nuclear Phys. B 460 (1996), no. 2,
351–361.
90. J. Madore, An introduction to noncommutative differential geometry and its physical appli-
cations, London Mathematical Society Lecture Note Series, vol. 206, Cambridge University
Press, Cambridge, 1995.
91. Juan Maldacena, Nathan Seiberg, and Gregory Moore, Geometrical interpretation of D-
branes in gauged WZW models, J. High Energy Phys. (2001), no. 7, Paper 46, 63.
92. Juan M. Maldacena, Black holes in string theory, Ph.D. thesis, Princeton University, July
1996.
93. John McGreevy, Leonard Susskind, and Nicolaos Toumbas, Invasion of the giant gravitons
from anti-de Sitter space, J. High Energy Phys. (2000), no. 6, Paper 8, 14.
94. Patrick Meessen and Tomás Ortı́n, An Sl(2, Z) multiplet of nine-dimensional type II super-
gravity theories, Nuclear Phys. B 541 (1999), no. 1-2, 195–245.
95. Andrei Mikhailov, Giant gravitons from holomorphic surfaces, J. High Energy Phys. (2000),
no. 11, Paper 27, 23.
96. Ruben Minasian and Gregory Moore, K-theory and Ramond- Ramond charge, J. High En-
ergy Phys. (1997), no. 11, Paper 2, 7 pp. (electronic).
97. Nicolas Moeller, Barton Zwiebach, and Ashoke Sen, D-branes as tachyon lumps in string
field theory, J. High Energy Phys. (2000), no. 8, Paper 39, 31.
98. Robert C. Myers, Dielectric-branes, J. High Energy Phys. (1999), no. 12, Paper 22, 41.
99. Robert C. Myers and Øyvind Tafjord, Superstars and giant gravitons, J. High Energy Phys.
(2001), no. 11, Paper 9, 19.
100. Yuji Okawa and Hirosi Ooguri, How noncommutative gauge theories couple to gravity, Nu-
clear Phys. B 599 (2001), no. 1-2, 55–82.
101. , Energy-momentum tensors in matrix theory and in noncommutative gauge theories,
Strings, 2001 (Mumbai), Clay Math. Proc., vol. 1, Amer. Math. Soc., Providence, RI, 2002,
pp. 263–277.
102. Jacek Pawelczyk, SU(2) WZW D-branes and their noncommutative geometry from DBI
action, J. High Energy Phys. (2000), no. 8, Paper 6, 7.
103. Jacek Pawelczyk and Harold Steinacker, Matrix description of D-branes on 3-spheres, J.
High Energy Phys. (2001), no. 12, Paper 18, 14.
104. , A quantum algebraic description of D-branes on group manifolds, Nuclear Phys. B
638 (2002), no. 3, 433–458.
105. Amanda W. Peet, Black holes in string theory, TASI 99: Strings, Branes, And Gravity
(S. Kachru J. Harvey and E. Silverstein, eds.), World Scientific, 2001.
106. J. Polchinski and M.J. Strassler, The string dual of a confining four-dimensional gauge
theory, Tech. Report hep-th/0003136, March 2000.
107. Joseph Polchinski, Dirichlet branes and Ramond-Ramond charges, Phys. Rev. Lett. 75
(1995), no. 26, 4724–4727.
108. Sanjaye Ramgoolam, On spherical harmonics for fuzzy spheres in diverse dimensions, Nu-
clear Phys. B 610 (2001), no. 3, 461–488.
109. A. Recknagel and V. Schomerus, D-branes in Gepner models, Nuclear Phys. B 531 (1998),
no. 1-3, 185–225.
110. J. G. Russo and A. A. Tseytlin, Waves, boosted branes and BPS states in M -theory, Nuclear
Phys. B 490 (1997), no. 1-2, 121–144.
111. Volker Schomerus, Lectures on branes in curved backgrounds, Classical Quantum Gravity 19
(2002), 5781–5847.
112. Peter Schupp, Paul Watts, and Bruno Zumino, Differential geometry on linear quantum
groups, Lett. Math. Phys. 25 (1992), no. 2, 139–147.
113. Claudio A. Scrucca and Marco Serone, A note on the torsion dependence of D-brane RR
couplings, Phys. Lett. B 504 (2001), no. 1-2, 47–54.
114. Nathan Seiberg and Edward Witten, String theory and noncommutative geometry, J. High
Energy Phys. (1999), no. 9, Paper 32, 93 pp. (electronic).
115. Ashoke Sen, Non-BPS states and branes in string theory, Progress in string theory and
M-theory (Dordrecht, The Netherlands) (Marco Picco, Laurent Baulieu, Michael Green and
Paul Windey, eds.), Kluwer, 2001 187-234.
116. Alexander Sevrin, Jan Troost, and Walter Troost, The non-Abelian Born-Infeld action at
order F 6 , Nuclear Phys. B 603 (2001), no. 1-2, 389–412.
117. Eric Sharpe, D-branes, derived categories, and Grothendieck groups, Nuclear Phys. B 561
(1999), no. 3, 433–450.
118. H.S. Snyder, The Electromagnetic Field In Quantized Space-Time, Phys. Rev. 72 (1947),
68.
119. , Quantized Space-Time, Phys. Rev. 71 (1947), 38.
120. Washington Taylor, M(atrix) theory: matrix quantum mechanics as a fundamental theory,
Rev. Modern Phys. 73 (2001), no. 2, 419–461.
121. Washington Taylor and Mark Van Raamsdonk, Multiple D0-branes in weakly curved back-
grounds, Nuclear Phys. B 558 (1999), no. 1-2, 63–95.
122. , Supergravity currents and linearized interactions for Matrix theory configurations
with fermionic backgrounds, J. High Energy Phys. (1999), no. 4, Paper 13, 76 pp. (electronic).
123. , Multiple Dp-branes in weak background fields, Nuclear Phys. B 573 (2000), no. 3,
703–734.
124. Paul Townsend, P-brane democracy, Particles, strings and cosmology (J. Bagger, ed.), World
Scientific, 1996.
370 ROBERT C. MYERS
125. Sandip P. Trivedi and Sachindeo Vaidya, Fuzzy cosets and their gravity duals, J. High Energy
Phys. (2000), no. 9, Paper 41, 33.
126. A. A. Tseytlin, Vector field effective action in the open superstring theory, Nuclear Phys. B
276 (1986), no. 2, 391–428.
127. , On non-Abelian generalisation of the Born-Infeld action in string theory, Nuclear
Phys. B 501 (1997), no. 1, 41–52.
128. , Born-Infeld action, supersymmetry and string theory, The many faces of the super-
world, World Sci. Publishing, River Edge, NJ, 2000, pp. 417–452.
129. Mark Van Raamsdonk, Open dielectric branes, J. High Energy Phys. (2002), no. 2, No. 1,
55.
130. Julius Wess and Bruno Zumino, Covariant differential calculus on the quantum hyperplane,
Nuclear Phys. B Proc. Suppl. 18B (1990), 302–312 (1991).
131. Edward Witten, Bound states of strings and p-branes, Nuclear Phys. B 460 (1996), no. 2,
335–350.
132. , D-branes and K-theory, J. High Energy Phys. (1998), no. 12, Paper 19, 41 pp.
(electronic).
133. , Overview of K-theory applied to strings, Internat. J. Modern Phys. A 16 (2001),
no. 5, 693–706.
134. Niclas Wyllard, Derivative corrections to D-brane actions with constant background fields,
Nuclear Phys. B 598 (2001), no. 1-2, 247–275.
Perimeter Institute for Theoretical Physics, Waterloo, Ontario N2J 2W9 Canada
Department of Physics, McGill University, Montréal, Québec H3A 2T8 Canada

E-mail address: rmyers@perimeterinstitute.ca
List of Participants
Bobby Acharya Nero Budur

Rutgers University University of Illinois
Piscataway, NJ, USA Chicago, USA
bacharya@physics.rutgers.edu nero@math.uic.edu
Cecilia Albertsson Igor Burban

Stockholm Centre for Physics Universitat Kaiserlautern
Stockholm University, Sweden Kaiserlautern, Germany
cecilia@physto.se burban@mathematik.uni-kl.de
Luis Alvarez-Consul Fabrizio Catanese

University of Bath Universtaet Bayreuth
Bath, U.K. Bayreuth, Germany
lalvarez@math.uiuc.edu fabrizio.catanese@uni-bayreuth.de
Costas Bachas Richard Cleyton

Ecole Normale Supérieure University of Southern Denmark
Paris, France Odense, Denmark
bachas@physique.ens.fr cleyton@imada.sdu.dk
Serguei Barannikov Alessio Corti

DMA Ecole Normale Superieure University of Cambridge
Paris, France Cambridge, UK
serguei.barannikov@ens.fr a.corti@dpmms.cam.ac.uk
Victor Batyrev Alastair Craw

Universitaet Tuebingen University of Utah
Tuebingen, Germany Salt Lake City, USA
victor.batyrev@uni-tuebingen.de craw@math.utah.edu
Ingrid Bauer Andrew Dancer

Universitaet Bayreuth University of Oxford
Bayreuth, Germany Oxford, UK
ingrid.bauer@uni-bayreuth.de dancer@maths.ox.ac.uk
Marcin Bobienski Sarah Dean

Warsaw University Harvard University
Warsaw, Poland Cambridge, USA
bobi@fuw.edu.pl sarah@math.harvard.edu
Volker Braun Nihat Sadik Deger

Humboldt University Feza Gursey Institute Rasathane Yolu
Berlin, Germany Istanbul, Turkey
volker.braun@physik.hu-berlin.de deger@gursey.gov.tr
Gavin Brown Robert Henricus Dijkgraaf

University of Sydney University of Amsterdam
Sydney, Australia Amsterdam, Netherlands
gavin@maths.usyd.edu.au rhd@wins.uva.nl
371
372 Participants
Igor Dolgachev Daniel Grunberg

University of Michigan University of Amsterdam
Ann Arbor, USA Amsterdam, Netherlands
idolga@math.lsa.umich.edu grunberg@science.uva.nl
Ron Donagi Marco Gualtieri
University of Pennsylvania University of Oxford
Philadelphia, USA Oxford, UK
donagi@math.upenn.edu marco.gualtieri@sjc.ox.ac.uk
Simon Donaldson Stefano Guerra
Imperial College University of Pennsylvania
London, UK Philadelphia, USA
s.donaldson@ic.ac.uk guerras@math.upenn.edu
Nick Dorey Sergei Gukov
University of Wales Swansea Harvard University
Swansea, UK Cambridge, USA
n.dorey@swansea.ac.uk gukov@gauss.harvard.edu
Michael Douglas Sean Hartnoll
Rutgers University University of Cambridge
Piscataway, USA Cambridge, UK
mrd@physics.rutgers.edu s.a.hartnoll@damtp.cam.ac.uk
Christian Van Enckevort Yang-Hui He
Johannes Gutenberg-Universitaet MIT
Mainz, Germany Cambridge, USA
enckevor@mathematik.uni-mainz.de yhe@ctp.mit.edu
Has-Georg Freiermuth Nigel Hitchin
Columbia University University of Oxford
New York, USA Oxford, UK
freiermuth@math.columbia.edu hitchin@maths.ox.ac.uk
Jerome Gauntlett Petr Horava
Queen Mary, University of London University of California
London, UK Berkeley, USA
j.p.gauntlett@qmw.ac.uk horava@socrates.berkeley.edu
Jos Gheerardyn Chris Hull
University of Leuven Queen Mary, University of London
Leuven, Belgium London, UK
jos.gheerardyn@fys.kuleuven.ac.be c.m.hull@qmw.ac.uk
Gary Gibbons Tasneem Zehra Husain
University of Cambridge University of Texas
Cambridge, UK Austin, USA
g.w.gibbons@damtp.cam.ac.uk tasneem@zippy.ph.utexas.edu
Mark Gross Akira Ishii
University of Warwick Kyoto University
Coventry, UK Kyoto, Japan
mgross@maths.warwick.ac.uk akira@kusm.kyoto-u.ac.jp
Participants 373
Yukari Ito Frances Kirwan

Tokyo Metropolitan University University of Oxford
Tokyo, Japan Oxford, UK
yukari@comp.metro-u.ac.jp kirwan@maths.ox.ac.uk
Arthur Jaffe Masanori Kobayashi
Clay Mathematics Institute Tokyo Metropolitan University
Cambridge, USA Tokyo, Japan
USA masanori@math.metro-u.ac.jp
Clifford Johnson Alexey Kokotov
University of Durham 197101 Kronverkskaya 8, apt. 26
Durham, UK St Petersburgh, Russia
c.v.johnson@durham.ac.uk alexey kokotov@hotmail.com
Dominic Joyce Boris Kors
Oxford University Utrecht University
Oxford, UK Utrecht, Netherlands
dominic.joyce@lincoln.ox.ac.uk kors@phys.uu.nl
Bernard Julia Alexei Kovalev
Ecole Normale Superieure University of Cambridge
Paris, France Cambridge, UK
bernard.julia@ens.fr a.g.kovalev@dpmms.cam.ac.uk
Dmitry Kaledin S. Prem Kumar
Independent University of Moscow University of Wales
Moscow, Russia Swansea, UK
kaledin@mccme.ru s.p.kumar@swansea.ac.uk
Anton Kapustin Adrian Langer
California Institute of Technology University of Warwick
Pasadena, USA Coventry, UK
kapustin@theory.caltech.edu langer@maths.warwick.ac.uk
Yujiro Kawamata Frederic Leblond
University of Tokyo Perimeter Institute
Tokyo, Japan Waterloo, Canada
kawamata@ms.u-tokyo.ac.jp fleblond@hep.physics.mcgill.ca
Masayuki Kawakita Felipe Leitner
University of Cambridge Humboldt University
Cambridge, UK Berlin, Germany
kawakita@ms.u-tokyo.ac.jp leitner@mathematik.hu-berlin.de
Oskar Kedzierski Wolfgang Lerche
UJ Krakow CERN
Krakow, Poland Geneva, Switzerland
kedziers@in.uj.edu.pl lerche@nxth04.cern.ch
Alastair King Tim Logvinenko
University of Bath University of Bath
Bath, UK Bath, UK
a.d.king@maths.bath.ac.uk maptl@maths.bath.ac.uk
374 Participants
Juha Petteri Loikkanen Rob Myers

Stockholm University Perimeter Institute
Stockholm, Sweden Waterloo, Canada
juha@theophys.kth.se rcm@hep.physics.mcgill.ca
Jan Louis Yasunari Nagai
Martin-Luther-Universitat University of Tokyo
Halle, Germany Tokyo, Japan
j.louis@physik.uni-halle.de nagai@ms.u-tokyo.ac.jp
Fernando Marchesano Viacheslav Nikulin
Universidad Autonoma de Madrid University of Liverpool
Madrid, Spain Liverpool, UK
marchesa@delta.ft.uam.es vnikulin@liv.ac.uk
Antonio Mateos Carlos Nunez
University of Barcelona University of Cambridge
Barcelona, Spain Cambridge, UK
tonim@ecm.ub.es cn239@damtp.cam.ac.uk
Tomoo Matsumura Aybike Ozer
Boston University Queen Mary, University of London
Boston, USA London, UK
mushmt@math.bu.edu a.c.ozer@qmul.ac.uk
James McKernan David Page
University of California University of Durham
Santa Barbara, USA Durham, UK
mckernan@math.ucsb.edu d.c.page@durham.ac.uk
Ilarion Victor Melnikov Stathis Pakis
Duke University Queen Mary, University of London
Durham, USA London, UK
lmel@cgtp.duke.edu s.pakis@qmw.ac.uk
Ruben Minasian Ari Pankiewicz
Ecole Polytechnique Max-Planck-Institut
Palaiseau, France Golm, Germany
ruben@cpht.polytechnique.fr apankie@aei.mpg.de
Gabriele Mondello George Papadopoulos
Scuola Normale Superiore King’s College London
Pisa, Italy London, UK
mondello@cibs.sns.it gpapas@mth.kcl.ac.uk
Gregory Moore Louis Paulot
Rutgers University Ecole Normale Superieure
Piscataway, USA Paris, France
gmoore@physics.rutgers.edu paulot@lpt.ens.fr
Mircea Mustata Simon Philip
University of California University of Edinburgh
Berkeley, USA Edinburgh, UK
mustata@math.berkeley.edu simonp@maths.ed.ac.uk
Participants 375
Sanjay Ramgoolam Emmanuel Scheidegger

Brown University Institut fuer Theoretische Physick
Providence, USA Wien, Austria
ramgosk@het.brown.edu esche@hep.itp.tuwien.ac.at
Miles Reid Savdeep Sethi
University of Warwick University of Chicago
Coventry, UK Chicago, USA
miles@maths.warwick.ac.uk sethi@theory.uchicago.edu
Soo-Jong Rey Annamaria Sinkovics
Seoul National University University of Wales
Seoul, Korea Swansea, UK
sjrey@phya.snu.ac.kr a.sinkovics@swansea.ac.uk
Sven Rinke Aninda Sinha
Duke University University of Cambridge
Durham, USA Cambridge, UK
rinke@phy.duke.edu a.sinha@damtp.cam.ac.uk
Daniel Roggenkamp Daniel Stevenson
Universitaet Bonn University of Adelaide
Bonn, Germany Adelaide, Australia
roggenka@th.physik.uni-bonn.de dstevens@maths.adelaide.edu.au
Gregory Sankaran Dr Balázs Szendrői
University of Bath University of Warwick
Bath, UK Coventry, UK
G.K.Sankaran@bath.ac.uk balazs@maths.warwick.ac.uk
Hisham Sati Hiromichi Takagi
University of Michigan Kyoto University
Ann Arbor, USA Kyoto, Japan
hsati@umich.edu takagi@kurims.kyoto-u.ac.jp
Natalia Saulina Richard Thomas
Princeton University Imperial College, London
Princeton, USA London, UK
saulina@princeton.edu richard.thomas@ic.ac.uk
Justin Sawon Andrea Tomatis
University of Oxford Imperial College, London
Oxford, UK London, UK
sawon@maths.ox.ac.uk andrea.tomatis@ic.ac.uk
Sakura Schafer-Nameki Alessandro Tomasiello
University of Cambridge Ecole Polytechnique
Cambridge, UK Palaiseau, France
s.schafer-nameki@damtp.cam.ac.uk tomasiel@cpht.polytechnique.fr
Chad(mark) Schoen Burt Totaro
Max-Planck-Institut University of Cambridge
Bonn, Germany Cambridge, UK
schoen@mpim-bonn.mpg.de b.totaro@dpmms.cam.ac.uk
376 Participants
Dimitrios Tsimpis Chen-Gang Zhou

Chalmers Institute of Technology Rutgers University
Goteborg, Sweden Piscataway, USA
tsimpis@fy.chalmers.se czhou@physics.rutgers.edu
Andrei Tyurin Francesco Zucconi
Steklov Math Institute University of Cambridge
Moscow, Russia Cambridge, UK
Tyurin@newton.kias.re.kr f.zucconi@dpmms.cam.ac.uk
Nikolaos Tziolas
University of Warwick
Coventry, UK
tziolas@maths.warwick.ac.uk
Misha Verbitsky
11-1-83 Orekhovy Blvrd
Moscow, Russia
verbit@mccme.ru
Dan Waldram
Queen Mary, University of London
London, UK
d.j.waldram@qmw.ac.uk
Brian Wecht
Univerity of California
San Diego, USA
bwecht@physics.ucsd.edu
Robert Wendt
University of Vienna
Vienna, Austria
robert.wendt@univie.ac.at
Professor Pelham Wilson
University of Cambridge
Cambridge, UK
pmhw@dpmms.cam.ac.uk
Brookie Williams
University of California
Santa Barbara, USA
brook@physics.ucsb.edu
Takehiko Yasuda
University of Tokyo
Tokyo, Japan
takehiko@tkg.att.ne.jp
Alberto Zaffaroni
Universita di Milano-Bicocca
Milano, Italy
alberto.zaffaroni@mib.infn.it
This volume is the proceedings of
the 2002 Clay Mathematics Institute 3 Volume 3
School on Geometry and String
Theory. This month-long program
was held at the Isaac Newton
Strings and Geometry

Institute for Mathematical Sciences
in Cambridge, England, and was
organized by both mathematicians
and physicists: A. Corti, R. Dijkgraaf,
M. Douglas, J. Gauntlett, M. Gross,
Strings and
C. Hull, A. Jaffe and M. Reid. The
early part of the school had many
lectures that introduced various
Geometry
concepts of algebraic geometry
and string theory with a focus on
improving communication between
these two fields. During the latter Proceedings of the
part of the program there were also
a number of research level talks.
2002 Summer School
This volume contains a selection
of expository and research articles Isaac Newton Institute
by lecturers at the school, and Cambridge, United Kingdom
Douglas, Gauntlett and Gross, Editors

highlights some of the current
interests of researchers working March 24–April 20, 2002
at the interface between string
theory and algebraic geometry. The
topics covered include manifolds Michael Douglas
of special holonomy, supergravity,
supersymmetry, D-branes, the
Jerome Gauntlett
McKay correspondence and the Mark Gross
Fourier-Mukai transform. Editors
CMIP/3
www.ams.org American Mathematical Society

AMS
www.claymath.org CMI Clay Mathematics Institute
4-color process 392 pages • 3/4” spine

Cmip03c PDF

Uploaded by

Copyright:

Available Formats

Cmip03c PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cmip03c PDF

Uploaded by

Copyright:

Available Formats

Clay Mathematics Proceedings

This volume is the proceedings of

Strings and Geometry

Douglas, Gauntlett and Gross, Editors

www.ams.org American Mathematical Society

4-color process 392 pages • 3/4” spine

American Mathematical Society

The Geometry of String Theory 1

The Geometry of A-branes 337

Michael Douglas, Jerome Gauntlett and Mark Gross

The Geometry of String Theory

2000 Mathematics Subject Classiﬁcation. 81T30, 83E30.

2. String theory and its physical goals

On the other hand, at the highest energies we attain in accelerator experiments

3. Eﬀective ﬁeld theory

In a supersymmetric vacuum, the naive consequences of supersymmetry hold –

(3.4) W = Ū m1 QH1 + D̄ m2 QH2 + Ē m3 LH2 ,

4. Supersymmetric Kaluza-Klein compactiﬁcation

4.1. Kaluza-Klein compactiﬁcation. Consider a solution of general rela-

motion of a string can be described by a two-dimensional ﬁeld theory of maps from

4.3. Supersymmetry and supergravity. After the dimension of space-time,

As is familiar, covariantly constant spinors are associated to a reduction of

4.5. Deriving the supersymmetric eﬀective ﬁeld theory. Suppose we

In any case, physicists argue that as ls /L → 0 and gs → 0 the supergravity de-

5. Beyond diﬀerential geometry: the string world-sheet

5.1. Strings as loops, and T-duality. The approach we just described,

of a submanifold of the target space, on which a particular type of open string is

5.4. Noncommutative geometry. The physics of Dirichlet branes is rich

5.6. Resolution of orbifold singularities. The simplest non-trivial non-

6. The duality revolution

6.1. Quantum supersymmetric ﬁeld theory. The typical problem here is

interest receives either no corrections in perturbation theory (say the superpotential

6.3. Branes and singularities. A point in the preceding discussion worthy

Of the many other cases, a particularly interesting one is the resolution of a

6.4. G2 and other supersymmetric compactiﬁcations. The basic anal-

group of conformal isometries of D-dimensional Minkowski space-time. Thus, the

24. D. S. Freed, Five Lectures on Supersymmetry, AMS, Providence, 1999.

Department of Physics and Astronomy, Rutgers University, Piscataway, NJ 08855-

M theory, G2 -manifolds and Four Dimensional Physics

Abstract. M theory on a manifold X of G2 -holonomy is a natural framework

2000 Mathematics Subject Classiﬁcation. Primary 81T30, 83E30, 53C29, 81T60.

If X is large compared to the Planck scale (the only scale in M theory)

2A more conventional example of light states at a singularity is provided by string theory on

- which we would like to pose to the mathematicians - is to construct compact

2. Supersymmetry, G2 -holonomy and Kaluza Klein spectrum.

2.1. Supersymmetry and G2 -holonomy. At low energies M theory admits

The simplest way to solve these equations is to take G = 0 in which case we

We will re-write this equation in the more symbolic form,

where by ∇g we mean the Levi-Civita connection constructed from g. Solutions to

with  a basis of constant spinors in Minkowski space.

implies that Hol(g(X)) is G2 or a subgroup. This is because G2 is the maximal

Therefore, if X is a compact manifold of precisely G2 -holonomy, the eﬀective

2.2. Properties of G2 -manifolds. From the parallel spinor θ we can con-

2.3. Kaluza-Klein Reduction. At low energies, the eleven-dimensional su-

parameters, i.e. there is a k-dimensional family of G2 -holonomy metrics on X, then

separate this term into two terms:

Since the ﬂuctuating G2 -structure is

The ﬂuctuations of the four-dimensional Minkowski metric give us the usual

3. Super Yang-Mills from G2 -manifolds: Classical

with a basis of constant spinors in Minkowski space.

If we now make the Kaluza-Klein ansatz for the 2-form

where we interpret as a complex parameter on the base W . Thus, the above

(2.6) ∇ˆ µ ≡ ∇µ + 1 [Γµ ν1 ν2 ν3 ν4 − 8δ νµ1 Γν2 ν3 ν4 ]Gν ν ν ν = 0 ,

We thus conclude that † ≥ † Γ with equality if and only (1 − Γ) = 0, which