You are on page 1of 142

D-branes and String Topology

Classifying D-brane RR-Charge at Low-Energies using K-theory

Sheer El-Showk

Supervisor: Prof.dr. R.H. Dijkgraaf

University of Amsterdam Institute for Theoretical Physics Valckenierstraat 65 1018 XE Amsterdam The Netherlands

August 2005

Abstract In this masters thesis we review recent developments in our understanding of the topological character of D-branes. As suggested in the title this is to be viewed as a starting point for the study of more sophisticated aspects of what has become referred to as String Topology. Starting from the notion that D-branes act as sources for the RR-elds in Supergravity we argue, using the anomaly canceling argument of [CY] [GHM] [MM], that their charges take values in the K-groups of spacetime. This argument is rened and reformulated, following [Wit2], by incorporating Sens work on tachyon condensation and brane/anti-brane annihilation to argue that all lower dimensional branes can be seen as the decay product of a stack of N D9 and N D9-branes. As a consequence we arrive at a direct geometric interpretation of the notion that the RR-charge of a D-brane, and hence its topological stability, is given by a class in the relative group K(X, Y ) where X is the spacetime manifold and Y is the complement, in X, of a tubular neighborhood of the D-brane world-volume. These arguments only hold in the case when the cohomology class of the three-form eld-strength, H, is trivial. In the nal section of this thesis we review modications of this argument required in order to incorporate non-trivial eld-strengths, H, and we will say something about their consequences. The focus will mostly be on the torsion case following [Kap]. There are several other review papers available in this area [OS][Wit4]. This review diers from these mostly in its level and its scope. It is intended to be relatively pedagogical (as its length will attest to) aiming at graduate students with a basic knowledge of string theory. It also attempts to cover the major developments in the early part of this eld which are often neglected in other reviews.

CONTENTS

CONTENTS

Contents
1 Introduction 1.1 How to Read this Paper . . . . . 1.2 Overview and Introduction . . . 1.3 Physical Motivation . . . . . . . 1.4 Further Reading (the literature) 1.5 Updates and Corrections . . . . . 1.6 Acknowledgments . . . . . . . . . 1.7 Notation and Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 4 4 6 7 8 9 10 11 11 12 12 14 16 16 18 19 20 22 23 25 28 29 32 35 35 37 40 42 43 44 45 48 51 51 53 56 57 61 62 62 63 64 65 68 71 73 74 75 76 78 81

2 RR-Charge and K-theory 2.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 The Physical Theory . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Perturbative String Theory . . . . . . . . . . . . . . . . 2.2.2 Low-Energy Eective Field Theories . . . . . . . . . . . 2.3 RR Field Strengths . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 Generalized Electromagnetism . . . . . . . . . . . . . . 2.3.2 Poincar Duality and the (Co)homological Classication e 2.4 The D-brane World-volume Theory . . . . . . . . . . . . . . . . 2.4.1 D-branes as Supersymmetric Excitations . . . . . . . . 2.4.2 D-brane Actions . . . . . . . . . . . . . . . . . . . . . . 2.4.3 Chan-Paton Factors and Adjoint Bundles . . . . . . . . 2.4.4 The Spin-Bundle and Chiral R-Symmetry . . . . . . . . 2.4.5 The World-Volume Dirac Operator . . . . . . . . . . . . 2.4.6 I-Branes and Chiral Fermions . . . . . . . . . . . . . . . 2.4.7 I-brane Spin Bundle . . . . . . . . . . . . . . . . . . . . 2.5 The D-brane Anomaly . . . . . . . . . . . . . . . . . . . . . . . 2.5.1 Gauge Anomalies and the Descent Procedure . . . . . . 2.5.2 Index of the D-brane Dirac Operator . . . . . . . . . . . 2.5.3 Index of the I-brane Dirac Operator . . . . . . . . . . . 2.5.4 Anomaly Factorization and the Euler Class . . . . . . . 2.6 Canceling the D-brane Anomalies: Anomaly Inow . . . . . . . 2.6.1 IIB with D-brane Sources . . . . . . . . . . . . . . . . . 2.6.2 RR Equations of Motion . . . . . . . . . . . . . . . . . . 2.6.3 The Thom Isomorphism and the Euler Class . . . . . . 2.7 Normal Bundles with Spinc Structures . . . . . . . . . . . . . . 2.7.1 Fermions and Spin(c )-structures . . . . . . . . . . . . . 2.7.2 I-branes and Spinc -structures . . . . . . . . . . . . . . . 2.7.3 Spinc and Anomaly Factorization . . . . . . . . . . . . . 2.8 The RR-Charge of a D-brane . . . . . . . . . . . . . . . . . . . 2.9 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . of Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3 D-branes and K-theory 3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Topological Defects, Chern Classes, andLower Dimensional Brane Charge 3.2.1 RR Equations of Motion and Lower-Dimensional Brane Charge . . 3.2.2 The First Chern Classes of a Line Bundle . . . . . . . . . . . . . . 3.2.3 Poincar Dual of Zero Loci . . . . . . . . . . . . . . . . . . . . . . e 3.2.4 Higher Chern Classes and Generalized Winding . . . . . . . . . . . 3.3 Tachyon Condensation: Brane-Anti-brane Annihilation . . . . . . . . . . . 3.3.1 RR-Charge and the Brane-Anti-Brane System . . . . . . . . . . . . 3.3.2 SYM on a D9-D9 System . . . . . . . . . . . . . . . . . . . . . . . 3.3.3 Spontaneous Symmetry Breaking and the Closed String Vacuum . 3.3.4 Topologically Stable Vortices in Flat Spacetime . . . . . . . . . . . 3.3.5 Topologically Stable Vortices in Non-trivial Backgrounds . . . . . 2

CONTENTS

CONTENTS

3.4

3.5

3.6

3.3.6 Topological Defects in Higher Rank Bundles . . . . . . . . . . . . D-branes as K-theory Classes . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 K-theory Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.2 Some Aspects of K-theory . . . . . . . . . . . . . . . . . . . . . . 3.4.3 Some Concrete Examples . . . . . . . . . . . . . . . . . . . . . . 3.4.4 Higher K-groups and Bott Periodicity . . . . . . . . . . . . . . . 3.4.5 D-branes in Flat Spacetimes . . . . . . . . . . . . . . . . . . . . D-branes in General Space-Times . . . . . . . . . . . . . . . . . . . . . . 3.5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.2 Line Bundles and Codimension Two . . . . . . . . . . . . . . . . 3.5.3 Incorporating CP-Bundles and Lower-Dimensional Brane Charge 3.5.4 The Thom Isomorphism in K-Theory . . . . . . . . . . . . . . . 3.5.5 Examples and Remarks . . . . . . . . . . . . . . . . . . . . . . . 3.5.6 RR Field Equations and the Thom Isomorphism . . . . . . . . . 3.5.7 Lower Dimensional Branes and the Thom Isomorphism . . . . . 3.5.8 Normal Bundles with Spinc Structure . . . . . . . . . . . . . . . Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . .

82 85 85 87 89 90 92 93 93 94 96 97 99 100 101 103 104 106 106 108 108 109 112 114 116 120 120 121 122 122

4 The B-eld and Twisted Vector Bundles 4.1 Overview . . . . . . . . . . . . . . . . . . . . . . . 4.2 The Freed-Witten Anomaly and Holonomy . . . . 4.2.1 The B-eld . . . . . . . . . . . . . . . . . . 4.2.2 Geometry of A and B for Torsion H . . . . 4.2.3 The Global Freed-Witten Anomaly . . . . . 4.2.4 Holonomies and Line Bundles on M . . . . 4.2.5 Surface Holonomy and Line Bundles on M 4.3 Twisted CP Bundles . . . . . . . . . . . . . . . . . 4.3.1 Obstruction to Dening U (N )-bundles . . . 4.3.2 Spinc and pfa(iD) . . . . . . . . . . . . . . 4.3.3 Some Basic Notions of Twisted K-Theory . 4.4 Outlook and Current Research . . . . . . . . . . .

A Spin- and Spinc -Structures 123 A.1 Spin-lifts of Principle SO(n)-bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 A.2 Spinc -lifts of Principle SO(n)-bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 B Some K-theory Technicalities 129 B.1 Constructing the K-theoretic Thom Class . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 B.2 The Thom Isomorphism in Codimension Two . . . . . . . . . . . . . . . . . . . . . . . . . 130 C Anomalies and the Index Theorem 132 C.1 Abelian Anomaly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

1 INTRODUCTION

1
1.1

Introduction
How to Read this Paper

The level of exposition in this paper is perhaps best suited for an entry/intermediate level graduate student (i.e. someone who has completed a masters degree) who has had at least a basic course on string theory and is also comfortable with quantum eld theory. A knowledge of dierential geometry is also assumed as well as some basic algebraic topology. For the latter a working knowledge of de Rham cohomology is essential as well as a general idea of how Cech cohomology is dened. No familiarity with K-theory is required. A reader with this background should benet from the level of detail presented in the text and the background material introduced along the way. Several sections, such as Section 2.7 and parts of Section 4, require some more advanced knowledge of some concepts and the reader should feel free to skip these (this will be discussed in the introduction to these sections). A more advanced reader may still benet from this text but might wish to skip some of the background material, either on string theory (Sec. 2.2 to Sec. 2.4.3), or on characteristic classes and K-theory (Sec. 3.2 and Sec 3.4.1-3.4.4).

1.2

Overview and Introduction

This thesis is intended as a pedagogical introduction to a central topic in string topology namely the topological character of D-branes. The term string topology is being used here (somewhat loosely) to denote a large body of work concerned with interesting topological properties of spaces that occur in string theory. This idea is discussed nicely in [Seg] and it is this usage of the term that we wish to imply (rather than the more technical usage found in [CV]). The basic K-theoretic classication is, in some sense, not particularly interesting, either from a mathematical or a physical perspective, because it does not introduce any new mathematical structures and, on the physical side, is a technicality which can often times be ignored. However, in a general setting, when various background elds are turned on the K-theoretic classication has to be modied or may even break down. It is in this setting that both new mathematics and new physics starts to emerge. We will have occasion to discuss this somewhat in Section 4 but we should forewarn the reader that the majority of the text is concerned with the rather more trivial setting when K-theory can be applied and results in a straightforward classication. Even though this is well-understood it involves many dierent aspects of string theory and mathematics and the purpose of this text is to bring these together and present them in a manner that is accessible to entry level graduate students. Along the way we will attempt to bolster the arguments found in the extant reviews [OS] [CY] [MM] [Wit2] and original papers to make the exposition as rigorous and transparent as possible. The last section should also serve as an introduction to the more interesting and open questions of how to understand D-branes in a more general background. Having said all this let us briey discuss the results reviewed in this thesis while being as non-technical as possible. At low-energies (compared to the string scale) string theory can be approximated by a eld-theory in ten dimensions. If only closed strings are considered then this eld theory is a supersymmetric theory containing gravity and is referred to as supergravity. It contains, amongst other things, a series of even or odd rank (depending on the particular string theory being approximated) antisymmetric tensors which act as higher-rank analogs of electromagnetism. These are referred to as the RR eld-strengths. Open strings can be incorporated by dening D-branes, submanifolds of the full spacetime on which the end-points of the open strings are stuck. If the open strings are allowed to propagate anywhere in spacetime then it is assumed that there is a spacetime lling D-brane, referred to as a D9-brane. It can be shown that the D-branes carry charge with respect to the RR elds and, as such, act as sources for the latter. This coincides well with the fact that the RR eld strengths are higher rank tensors so their potentials, rather than being one-forms, will also be of higher rank and so must couple minimally to higher dimensional objects, namely the D-branes. So far this has the character of a higher rank generalization of electromagnetism with the RR eld strengths playing the role of the electromagnetic two-form eld strength and the D-branes acting as generalized higher dimensional sources. As with regular electromagnetism, this theory admits a description via cohomology with the generalized current

1.2 Overview and Introduction

1 INTRODUCTION

dening an element of compact cohomology. Thus, naively, one might imagine that D-brane charge can be classied by cohomology alone (as is the case with say magnetic charge). A complication arises, however, if one recalls that the D-branes are dened as the spaces on which open string endpoints may propagate. This is because the open strings also can be approximated, at low energy, by a eld theory which will be dened only on the D-brane where the strings can propagate. This will be a supersymmetric Yang-Mills theory with gauge-elds and charged fermions. In some particular instances these fermions will either be chiral or will couple chirally to the gauge elds and this has the potential to generate gauge anomalies which are well known to make a theory inconsistent. In particular, such anomalies lead to a lack of charge conservation. In order to cancel these anomalies and make the overall theory consistent we must couple topological gauge defects of the gauge eld on the D-brane, which are the source of such anomalies, to the RR elds in the bulk, thus making the former charged under the latter. This cancels the anomaly by allowing charge to ow from the bulk theory (the supergravity theory) into or out of the defects on the D-brane, thus countering the loss of charge due to the anomaly. Implicit in this cancellation argument, however, is that the D-brane charge or current is a more complex mathematical object than that of an electron. The charge of a D-brane is not only related to the support of the D-brane as a submanifold of spacetime but also to the topological character of its gauge bundles. The exact form of this relationship is very suggestive of the fact that the charge, rather than being viewed as an element of cohomology, should be viewed as an element of K-theory. At this level, however, the only motivation we have is that the actual cohomological form of the D-brane charge, after incorporating anomaly canceling terms, looks like the image of a K-theory class under a well-known map, the Chern homomorphism, from K-theory to cohomology. To make this argument more intuitive and also to establish it more rmly we can take another approach to studying D-branes. Namely, there is a body of work suggesting that D-branes can, themselves, be dened as gauge defects on D9-branes. Recall, we have already suggested that D-branes are higher dimensional analogs of charged particles so, in particular, there are corresponding anti-branes. Thus to a D9-brane (that lls spacetime) there is a corresponding anti-brane which we denote D9. Since they coincide (because they both ll all of space time) and they have opposite charge they can annihilate (because they have a higher energy density than the vacuum and the net charge of the two is zero so their annihilation would not violate charge conservation). However, we have already mentioned that the charge of a D-brane has a complex dependence, not only on the support of the brane (which is identical for D9 and D9), but also on the topology of the gauge bundles on the brane. If the bundles on the D9 and the D9 have dierent topology then there is a net charge in the system so the pair cannot uniformly annihilate everywhere in spacetime. Rather, on the support of the gauge defects, there will remain RR charge and this can be associated with the existence of lower-dimensional branes (the full argument is more involved and will be presented in detail in the thesis). Thus lower-dimensional branes can be described by an annihilation process of pairs of D9 and D9 and, in particular, by the gauge bundles associated with such pairs. This turns out to lead very naturally to K-theory since the latter is dened in terms of pairs vector bundles (related to the aforementioned gauge bundles) on a space. Hence it is possible to identify congurations of D9/D9 pairs with classes in the K-theory of the spacetime manifold. Since such congurations decay to dene stable congurations of lower-dimensional branes one can identify the latter with the K-theory class of the former. This identication leads to precisely the correct RR charge for the lower-dimensional branes if their K-theory class is mapped into cohomology via the Chern homomorphism mentioned above. This description also correctly reproduces the energy density associated with the lowerdimensional branes. This argument suggests that the K-theoretic classication is in fact the correct one and the cohomological form of the charge is simply an approximation viewed through the Chern homomorphism. The dierence between the two is generally due to torsion subgroups in K-theory which are eliminated when mapped into cohomology. All the discussion above has taken place against a particular backdrop where certain background elds in the supergravity theory described above have been turned o. This is because when one of these elds, the anti-symmetric rank two tensor known as the B-eld, is turned on and has a topologically non-trivial eld strength associated with it certain parts of the previous analysis break down. The B-eld couples minimally to the fundamental string in string theory and, in the presence of open strings, this coupling leads to an anomaly in the two-dimensional world-sheet theory on the string itself. This anomaly would imply an inconsistency in the formulation of the string theory and so must be canceled. This can be

1.3 Physical Motivation

1 INTRODUCTION

done by modifying the gauge elds on the D-branes, which couple to the boundary of the open-strings (recall that D-branes are dened as submanifolds on which open strings can end), in such as way as to cancel the anomaly induced by the coupling of the B-eld. In so doing, however, the gauge elds become connections on ill-dened twisted vector bundles. This requires a reformulation of the previous arguments in terms of twisted K-theory but we will not have time to develop this in this thesis. The thesis is organized as follows. We rst introduce some background material on string theory and its low-energy approximations, which furnish the backdrop for the rest of the discussion, in Section 2.2. We then discuss some general notions of how electromagnetism can be described in the language of cohomology in Section 2.3. We then introduce D-branes, the eld theories living on them and the causes of potential anomalies for these theories in Section 2.4. The actually anomaly calculation and its cancellation are the subject of Sections 2.5 and 2.6. A technical subtlety as well as a minor new result regarding Spinc structures on D-brane intersections is discussed in Section 2.7. Finally, the connection between D-brane charges and K-theory via the Chern homomorphism is made in Section 2.8. In Section 3 we start by introducing characteristic classes and their relationship to topological defects in Section 3.2. We then discuss Sens construction for brane/anti-brane annihilation in Section 3.3 and relate it to the previous discussion on topological defects. This leads nally to a more direct classication of D-brane charges as K-theory classes in Section 3.4 where the discussion is limited to a topologically simple setting. This analysis is generalized in Section 3.5 with the use of the Thom isomorphism in K-theory. The last Section, 4, is rather shorter than the rst two and is provided primarily to suggest why the arguments given in Sections 2 and 3 must be modied if the B-eld has non-trivial topology. A discussion of some further reading is provided in the conclusions to the various sections as well as in Section 1.4 below. Two technical appendices on Spinc and on some technical K-theoretic constructions (App. A and B, respectively) are provided for readers who wish to understand some of these subtleties. Finally, a pedagogical introduction to anomalies and their relationship to index theorems is provided in Appendix C.

1.3

Physical Motivation

In this thesis we will review the earliest part of a growing body of literature pertaining to the topological properties of D-branes. These objects emerge naturally as higher-dimensional degrees of freedom in the various incarnations of string theory and have played an important role in the continued development and coherence of the theory. They admit many descriptions ranging from a characterization as boundary states in a world-sheet theory to a acting as extended generalized electromagnetic sources in an eective spacetime eld theory approximating string theory at low energies. Our focus here will be predominantly on the spacetime perspective as this was the approach taken in much of the earlier literature (for notable departures see [Moo2] [Moo1] and references therein). The main aim of this review will be to demonstrate how, in certain spacetime backgrounds, X, the charges of a D-brane will take values in a certain Abelian group, K 0 (X), determined by the exotic cohomology theory K-theory. One may wonder at the relevance, particularly the physical relevance, of this undertaking and if it warrants a text as long as this thesis. A further concern might be that this work, even if valid, is rather unrewarding as it is often dicult to calculate the K-groups of a space and they often may not dier signicantly from the ordinary cohomology groups. Let us address these concerns in turn by rst motivating an interest in studying purely topological aspects of a theory and then discussing the relevance of K-theory (or some more exotic variant) over the approximation provided by (Cech) cohomology. K-theory, as any other cohomology theory, concerns topological invariants of spaces and is, as such, a source of rather coarse information. Physics has traditionally been far more concerned with more dicult questions such as the exact geometry of a space or the dynamics of certain elds on it. All these questions, however, have always been posed, implicitly, against a backdrop of topology. That is to say that, although it is not always appreciated, a proper understanding of dynamics or geometry is often sensitive to topological issues. In many physical scenarios, however, the relevant topologies were uninteresting so have been ignored or dealt with rather simplistically. There are several areas, however, where it is clear that global topological issues can be very important. For instance, in the study of anomalies in gauge eld theories or topological defects in condensed matter systems. In string theory, moreover, where the topology of spacetime itself may be highly non-trivial, it is clear that one cannot

1.4 Further Reading (the literature)

1 INTRODUCTION

always neglect topological considerations in favor of local physics alone. It is in this vain that the author rst wishes to suggest the relevance of K-theory in classifying Dbranes. The K-theoretic invariant associated with a D-brane is a topological invariant meaning it should not be altered by any reversible adiabatic process. This alone makes it interesting to study but, of even more relevance, this invariant is related to the charge of the D-brane under a generalized electromagnetic eld much as the charge of a magnetic monopole is related to a topological invariant of the gauge bundle it is associated to. In the case of D-branes, in particular, the K-theoretic characterization provides an elegant explanation for certain facts such as the stability of only certain types of branes in certain theories. Moreover, this description is very closely tied to a description of D-branes in terms of topological defects in higher dimensional branes and hence provides an elegant synthesis of two dierent lines of research. Of course these arguments, based on elegance, may still not seem to warrant the attention that has been focused on this area in the literature or the rather laborious extent of this document. To address the second issue let us note that the bulk of this thesis is not immediately related to the Ktheoretic classication of D-branes. Rather, large parts of the paper are spent developing the necessary background material. This includes deriving the exact form of the coupling of D-branes to the bulk elds in supergravity. While this is of central importance to the argument for K-theory it is also obviously of great importance in any study of D-branes in low-energy string theory. A lot of time is also spent discussing the theory of tachyon condensation and brane/anti-brane annihilation which has been a subject of considerable interest in string theory. Finally, as a masters thesis that is intended to be mostly self-contained and pedagogical, a large amount of this presentation has been dedicated to furnishing the appropriate background knowledge in string theory and basic algebraic topology. In fact, as a consequence of this pedagogical approach, this document only reviews the earliest and most basic developments in this eld. This brings us to the nal point, namely, that there is still a continued and perhaps growing interest in this subject or its spin-os. Although the notion that D-branes are classied by K-theory has been mostly settled in a certain case, where the background B-eld is topologically trivial or has a torsion cohomology class, there is still a body of research on the topic. This research is not motivated purely by the desire to catalog the possible D-brane charges in various backgrounds or theories since such a task is not likely to generate the kind of interest that does seem to exist in the eld. Rather, there are a number of problems raised by the basic classication of D-branes as K-theory classes and it is hoped that their resolution might provide a deeper insight into the nature of string theory. For instance, when the background B-eld has a non-torsion class it would seem like one must introduce an innite number of D-branes and the physical interpretation in this case is not entirely clear. Witten [Wit2] [Wit4] has suggested that this might be related to going o-shell and, as such, a clearer physical understanding of this might suggest something about the o-shell nature of string theory. Moreover, understanding and properly formulating the self-duality of the RR-elds, for which the D-branes are a source and which are themselves classied by a K-theory, has motivated a deeper study of quantum theories with self-dual eld strengths which has beneted from a formulation using K-theory [FH] [MW]. Understanding how the K-theoretic description of D-branes lifts to objects in M -theory is also an area of active interest as it is possible that this might provide a source of insight into the latter. Other interesting avenues of research initiated by the K-theoretic classication of D-branes are mentioned through-out the thesis (see Sections 2.9, 3.6, and 4.4) and also in Section 1.4 below.

1.4

Further Reading (the literature)

The reader wishing to learn more about the basic classication of D-branes can consult the original references, [CY] [GHM] [MM] [Wit2], or some reviews [Wit4] [OS] [Man]. Wittens analysis in IIB was extended to IIA in [Hor]. There is also some work extending the anomaly cancellation arguments to the brane/anti-brane case [SW] and to non-BPS systems [Sza2]. The original paper on extending the classication to non-trivial B-elds was done, for the torsion case, in [Kap]. Work on the non-torsion case has been pursued in a large body of work [BM] [BCM+ ] [GR] [CJM] and is still ongoing. There are many other papers by the same authors on this subject that have not been listed here. One of the most recent is [CW]. A related line of research is the existence of higher gauge theories in string/M-theory and the exact topological character of these theories. A general theme occurring in the more mathematical descriptions is that string theory requires lifting standard geometric

1.5 Updates and Corrections

1 INTRODUCTION

constructions to the loop space of a manifold and this has developed into an industry of its own. See [Seg] [CV] [TS] [Sch]. Another interesting line of research that has developed out of the earliest analysis reviewed in this paper is that of quantizing theories with self-dual eld strengths [Wit3] [MW] [FH]. Mathematically this has led to attempting to dene cohomology theories that can adequately incorporate both magnetic and electric sources [Fre1] [HS]. Finally the study of D-branes and K-theory has been conducted in the controlled setting of WZW models in a series of papers [GGR] [FS]. See also other papers by the same authors. As the reader can see there are many directions in which the work reviewed in this thesis has been extended and neither the list provided above nor the bibliography should be considered comprehensive.

1.5

Updates and Corrections

The most recent version of this thesis can be found at: http://www.netherrealm.net/sheer/ The author would be grateful for any corrections or suggestions. These can be emailed to him at the address found on the website above.

1.6 Acknowledgments

1 INTRODUCTION

1.6

Acknowledgments

I would like to take this opportunity to thank the various individuals to whom I am indebted, not only for their help in preparing this thesis, but also for their support and guidance through-out my masters studies. The particular choice of topic for this masters thesis proved to be very rewarding as it allowed me to explore many interrelated areas of physics and mathematics that are of great interest to me. Thus I would rst like to extend my thanks to my supervisor, Robbert Dijkgraaf, for encouraging me to pursue this topic and for providing me with very friendly and insightful guidance when it was needed. I would also like to thank Robbert for clarifying many mathematical subtleties that I missed as I slowly tried to absorb the background material necessary to tackle this subject. I am also very grateful to N.P. Landsman for initially suggesting this thesis topic and for spawning within me an appreciation for the beauty and elegance of formal mathematics. He has been a constant source of support as well as a truly inspirational lecturer. I have also been fortunate enough to benet from the general atmosphere of camaraderie and intellectual stimulation fostered at the Institute for Theoretical Physics (ITFA) in Amsterdam and wish to extend my gratitude to all the people who have nurtured this environment. Several names, in particular, come to mind as people have been very helpful, both in guiding me and providing me with support (often in the form of numerous of letters of recommendation). For this I would like to once again thank Robbert Dijkgraaf and N.P. Landsman as well as Jan de Boer and Jan Smit. I am also very grateful to Jan Smit for introducing me to eld theory and for inciting my interest in the subject. To Jan de Boer I would like to oer my thanks, not only for various letters of recommendation, but also for elding innumerable queries both technical and bureaucratic. It is very nice to have someone who is as intimidatingly knowledgeable as Jan but who is, at the same time, very friendly and personable and I look forward to his supervision of my PhD thesis. An inestimable debt of gratitude is owed by myself to Vyacheslav Rychkov (Slava) for being both an excellent mentor and a good friend. I can only hope that I have been able to absorb something of the physical intuition you tried to convey to us in your AQFT course. With your support this became one of the best courses I have ever had. I would also like to thank Ioannis Papadimitriou for our many long and challenging discussions over so many excellent Greek dinners and for clarifying many aspects of string theory for me. In him, also, I have found that most wonderful combination of a friend and a colleague. In addition to Ioannis and Slava I would like to thank the other postdocs and PhD students at the institute who have often been available for useful and interesting discussions, both academic and otherwise. I would, in particular, like to mention Ben Craps for several illuminating discussions during the early phase of writing this thesis. To the other masters students who have spent this year with me writing our thesis and commiserating on shared hardships I would like to extend fond thanks. It has been very pleasant spending my time with you and Im sure well be seeing more of each other. I would also like to particularly thank Charles Mathy, Marc Emanuel, and Balt Rees for our often irresponsible but always educational discussions. You know how much of a pleasure it has been. A special debt of gratitude is owed to Maxim Henry Lagrilli`re e who, as well as being a constant and amusing distraction through-out the writing of this thesis, played a vital role in its nal moments. Many thanks Maxim. I would also like to thank Hendrik van Eerten for being such a fun oce mate. Finally, I would like to thank those who have not directly been part of my academic life yet have been of central importance in the rest of my life. First, and foremost, my parents, Nabil and Luna El-Showk, for their unrelenting support and for teaching me the the value of things (other than brevity). Also, to my brothers and sister, Sedeer, Hedeer and Shiraz El-Showk, who have had to deal with my unending rants and other oddities. I also oer my thanks and apologies to my girlfriend, Hanna Valdis Thorsteinsdottir, for putting up with weekends of me working on my thesis and with my endless, uninteresting updates on the latest traumatic turn of events. You are as kind as you are beautiful. I would like to dedicate this thesis to my parents.

1.7 Notation and Conventions

1 INTRODUCTION

1.7

Notation and Conventions

Through-out the text both coordinate and index-free (where dierential geometric objects will be represented abstractly) notation will be used. In general the latter will be favored as it is less cluttered, tends to be more convenient and high-lights the mathematical nature of the object under discussion. When calculations are undertaken, however, coordinate notation will often be favored. Through-out the text an attempt will be made to use certain symbols consistently to enhance readability. The following table is provided for reference. Symbol X M , p T , N or N N () S(N ), S (N ) C, C(i) G, G(i) B, B(2) , or B H, H(3) , or H A, A , Aa F , F , , a, b, c m, n, o , 1 , 2
H (X), HdR (X)

H (X, G)

Meaning The spacetime manifold. This may be at R10 in some cases but not always. More often it will be a general 10-dimensional spin manifold. A Riemann surface giving the string worldsheet. The manifold representing a Dp-brane world-volume that is immersed in spacetime. The tangent bundle and normal bundle of a Dp-brane. A tubular neighborhood of the Dp-brane world-volume dieomorphic to N or N when is a submanifold. The spin-bundle dened as a lift from the (usually even real dimensional) bundle N and its decomposition into positive and negative chirality components (when applicable). The formal sum of Ramond-Ramond potentials or a particular degree i RR potential The formal sum of Ramond-Ramond eld-strengths or a particular degree i RR eld strength The NS-NS antisymmetric tensor potential which is a 2-form. The 3-form eld strength associated with the potential B as H = dB. The open string NS-sector gauge eld potential (a 1-form) in geometric, spacetime coordinate and world-volume coordinate form respectively. The 2-form eld strength associated with the potential A. Greek indices will denote full spacetime indices ranging from 0 to 9 unless specically noted otherwise. Lower case letters from the beginning of the alphabet denote D-brane indices unless otherwise noted (range 0, . . . , p). Lower case letters from the later part of the alphabet denote indices transverse to the D-brane unless otherwise noted (range p + 1, . . . , 9) Unless otherwise specied these maps will be consistently for the following sequence of inclusions: 1 : N () X, 2 : N , and : X. de Rham cohomology. Subscript c or cv indicates de Rham cohomology with compact support, or compact support in the vertical direction (for a vector bundle). Cech cohomology with coecients in the group G.

10

2 RR-CHARGE AND K-THEORY

RR-Charge and K-theory

In this section we will present the arguments that initially led to the modern understanding of D-brane charges as being classied by the K-theory of the ambient spacetime in which they exist. The relevant original literature is [CY], [GHM], and [MM] wherein the coupling of the D-branes to the RR-form potentials is determined on the basis of an anomaly cancellation argument. As a consequence, it is found that the associated RR-eld strength source is an element of the even cohomology ring that is very suggestive of K-theory. When it was later suggested that D-branes can be understood as topological defects on the world-volumes of higher-dimensional branes ([Sen3], [Sen4], [Sen5]) a more geometric understanding of the K-theoretic classication of D-branes emerged [Wit2] which is developed in Section 3. For simplicity and consistency we will generally restrict the discussion here to Type IIB string theory in 10 uncompactied spacetime dimensions though all the arguments given in this section can be modied for the type I and type IIA theories [Wit2] [Hor]. Moreover through-out Sections 2 and 3 the topological class of the eld-strength, H3 , of the B-eld will be trivial; non-trivial cases will be discussed in Section 4.

2.1

Overview

In the closed string sector of superstring theory the massless excitations of the Ramond-Ramond sector can be decomposed into antisymmetric spacetime elds (n-form elds). Type IIA and IIB string theories are characterized by dierent relative signs of the GSO projection on left-moving and right-moving closed string modes which results in a dierent decomposition of the RR-spectrum into spacetime n-forms. In the IIA theories the RR-sector decomposes into even dimensional elds strengths with corresponding odd-dimensional RR potentials while for IIB the situation is reversed. It is natural, therefore, to search for degrees of freedom which couple to the RR-potentials and provide a source for the associated eld strengths. Fundamental strings do not directly couple to the RR-potentials but rather to the eld strengths and therefore do not provide an appropriate candidate. D-branes, special subspaces on which open-strings can end, are a natural candidate and in fact their introduction can be motivated as sources for the RR-eld strengths (alternately they can be seen as required by T-duality since open strings with Neumann boundary conditions in one direction will have Dirichlet boundary conditions when T-dualizing that direction). Naively, it would appear that D-brane charges are classied by the homology class (or the Poincar e dual cohomology class) of the cycle they wrap in a manner analogous to electromagnetism. In particular we can add a term to the low-energy eective action of the type II string theories of the form SRR =

C=
X

(1)

Where we integrate the formal sum of all the RR potentials C = i C(i) (i, the degree of the potential, is odd for type IIA and even for type IIB) over the D-brane world-volume, , with the integration selecting out only the appropriate degree form. In the last term we have used a current localized on the D-brane to represent integration over the worldvolume by integration over the spacetime manifold X (if C were a closed form, would be the Poincar dual of ; as it is it can be seen an the analog of a delta-function for e a point source). This latter form makes manifest the analogy with electromagnetism with C a generalized electromagnetic potential and representing a current source localized on . Naively then would be a cohomology class representing the D-brane RR-charge. A more careful analysis ([MM], [CY], [GHM]) reveals that there are additional terms in the low-energy action arising from the need to cancel anomalies generated on the intersection of D-brane worldvolumes. The open string spectrum on the D-brane worldvolume contains fermions from the Ramond sector. These are sections of the spacetime spin bundle lifted from the space-time tangent bundle. The GSO projection eliminates one chirality of the spinor representation of the structure group of the space-time tangent, SO(1, 9). Although the remaining spinors are chiral with respect to the 10-dimensional chiral grading matrix, 11 , their decomposition into spinors lifted from the D-brane tangent and normal bundles is no longer generically chiral. None-the-less, it may contain chiral gauge couplings due to an R-symmetry associated with normal bundle rotations which couple dierently to left and right moving spinors (when

11

2.2 The Physical Theory

2 RR-CHARGE AND K-THEORY

the normal bundle is topologically non-trivial). Moreover, when two D-branes intersect, the spectrum changes and the result is that open strings with one end on either brane can become chiral. As in any quantum eld theory the presence of charged chiral fermions generates an anomaly in the action that can be calculated from the index of the appropriate Dirac operator (via the descent procedure). The anomaly on the worldvolume implies that the charge associated with the anomalous current is not conserved. This can be rectied using a standard mechanism by allowing charge to ow in or out from a higher dimensional theory within which the anomalous theory is embedded and thereby resulting in overall charge conservation [CH]. In order to do this it is necessary to add additional terms to the bulk action which are localized on the D-brane worldvolume. These terms take the form of additional RRpotential couplings to the brane with the interpretation that the anomaly on the D-brane is canceled by an anomalous variation of the RR-potentials. As a consequence (1) is modied so that the charge takes values in the even cohomology ring in a very particular way1 SRR =

C ch(W) e 2 d

A(T ) A(N )

(2)

It is not hard to see that the values the charges are in the image of the group K 0 () under a modied version of the Chern homomorphism: ch : K 0 (X) H 2n (X, Q). This suggests that K-theory might more adequately classify the charge of D-branes than cohomology. This can also be seen in other ways and is suggested by the fact that K-theory, due to Bott periodicity, is cyclic with period two (i.e. the higher K-groups which can be dened iteratively, are cyclic) as is the D-brane spectrum of IIA and IIB string theories. The form of the bulk RR-coupling (2) implies that RR-elds also couple to topological defects on the brane world-volume theory. This is because the various terms in the integrand in (2) are cohomology classes whose integral denes winding numbers of gauge congurations in various co-dimensions (more precisely they are characteristic classes measuring non-triviality of vector bundles). Witten, in [Wit2], combined this observation with earlier work by Sen ([Sen3], [Sen4], [Sen5]) to generate a more explicit (topological) construction of D-brane charge via K-theory. This construction will be the subject of Section 3.

2.2

The Physical Theory

In order to orient the unfamiliar reader this section, and the next, outline the structure of the physical theory under discussion. This theory is a non-renormalizable quantum eld theory that is none-the-less presumed to be well dened because it can be shown to be the low-energy limit of certain string theories (as will be made more precise shortly). The exposition will not be particularly detailed as the exact form of the theory (in a Lagrangian formulation) is rather complicated and is mostly irrelevant for the rest of the thesis. Rather, this introduction is intended to outline, schematically, how the eld theories discussed in much of the rest of the thesis emerge from a perturbative description of string theory as a low-energy approximation and how string theoretic calculations can be used to x the nature of this approximation. The presentation will summarize discussions in [Pol2, ch. 10, 12-13], [Sza1], [Cra], and [Bac]. These references can be consulted for more particulars. In general we will avoid constant normalization factors unless they become relevant to the discussion at some point (such as the Regge slope, , below). 2.2.1 Perturbative String Theory

As mentioned above the setting of much of this thesis will be the low-energy eective theories derived from superstring theory. Perturbative string theory is described in terms of a map from a 2-dimensional worldsheet into a target space, x : M X, which maps the worldsheet coordinates (, ) to spacetime coordinates, x (, ).2 This map is constrained to be independent of the parameterization of the manifold M . This has the interpretation of describing the 2-dimensional worldsheet of a string propagating through the spacetime manifold X. Classically, the dynamics of this theory are given by extremizing the proper area of the image of M in X. This is analogous to the statement, familiar from general relativity, that
1 In (2) it has been assumed that the B-eld has a topologically trivial eld strength; this assumption is maintained throughout Sections 2 and 3. 2 One can also formulate this in a more geometric manner without reference to coordinates.

12

2.2 The Physical Theory

2 RR-CHARGE AND K-THEORY

particles proceed along trajectories that extremize the proper length of their worldline. Quantization of this theory is given by integration over inequivalent maps, x, with a weight factor proportional to the proper area of the embedded worldsheet (actually the Polyakov action given below is only classically equivalent to the area of the embedded worldsheet but one can use this as an ansatz and simply dene the string action to be given by (3) below). Thus the quantized theory is given by the Polyakov path integral which is essentially a path integral for the 2-dimensional theory on the manifold M . From the worldsheet, M , perspective this is a 2-dimensional quantum eld theory with elds given by x and action (the Polyakov action) given by the area of the image: Spolyakov = 1 4 d2 g gab a x b x (3)

From the spacetime, X, perspective this is a rst quantized theory in the sense that it provides a description of a single string propagating rather than a eld of string excitations (as required for a full string eld theory). Perturbative amplitudes in this theory can be calculated by summing up contributions from dierent Riemann surfaces, M . Since the Riemann surfaces correspond to the path of the string, higher genus surfaces calculate loop amplitudes (and one must sum over all possible inequivalent loops given by Riemann surfaces with dierent moduli). The elds in this world-sheet theory, x , satisfy certain boundary conditions depending on whether the string is closed or open. Their energy eigenstates correspond to oscillatory modes of a (classical) string and hence can be partitioned into right- or leftmoving components (these are also referred to as holomorphic and anti-holomorphic for reasons that will not be discussed here) with a given frequency. More details about the basic formulation of string theory can be found in the appendices and the references mentioned above. Phenomenological constraints, as well as certain pathologies in the purely bosonic theory, require the presence of fermions in the spectrum. The basic Polyakov string action above can be augmented to give a supersymmetric worldsheet theory by adding worldsheet superpartners, (, ), for each x in the theory above and dening a supersymmetric version of the Polyakov action. This introduces several new choices of boundary conditions for the fermionic elds referred to as Ramond (R) and Neveu-Schwarz (NS) which correspond to the freedom to make the fermions periodic or anti-periodic, respectively. For closed strings two such choices must be made resulting in four sectors of the theory: RR, R-NS, NS-R, and NS-NS. The two choices correspond to separate choices of boundary conditions for left and right moving oscillators and the various sectors are given by dierent combinations of choices. In the open string only one such choice is possible resulting in two sectors: R and NS. This is because the open string end-points interchange left and right moving oscillators so the two must have the same fermionic boundary conditions. The spectrum in all these sectors will look dierent and gives the wealth of particle states associated with the massless modes. In particular these particle states will include spacetime fermions and can be arranged into spacetime supersymmetric multiplets. In the closed string, for instance, the rst excited state in the NS-NS sector implies that a single string in this sector is a massless object transforming as a spacetime two-tensor. Its symmetric, anti-symmetric and trace components are associated with the graviton, G , the anti-symmetric tensor, B , and the Dilaton, . The modes from the R-NS and the NS-R sector, on the other hand, transform as spacetime fermions and provide the fermionic components in the supersymmetric multiplets. An NS-sector open string in its rst excited state is also massless but transforms under a vector representation of the spacetime Lorentz group. It can thus be associated with a spacetime gauge eld, A .3 It is non-trivial, but it can be shown, that the spectrum thus generated implies that the resulting spacetime theories, which will be discussed below, are also supersymmetric, assuming that certain topological conditions 4 on X are satised. Perturbative worldsheet anomalies imply that the quantized, supersymmetric worldsheet theory is only a consistent quantum theory when the target spacetime is 10 dimensional. Hence, from now on, we shall always assume the spacetime manifold, X, to be 10 dimensional.
3 In Minkoswki spacetime it can be argued, on quite general grounds, that a massless vector particle must always have gauge degrees of freedom in order to eliminate its temporal component. If this were not the case then the kinetic term in a Lagrangian formulation would be unbounded from below (because of the opposite sign of the temporal derivatives). In a hamiltonian formulation this pathology is manifest as states with negative norms that must be eliminated. 4 Spectrum calculations and equations of motion provide only local information but global supersymmetry and even the presence of certain elds, such as fermions, provides certain topological constrains such as the existence of a spin-structure on X. Some of these will be discussed in later sections.

13

2.2 The Physical Theory

2 RR-CHARGE AND K-THEORY

It is also possible to model a string embedded in a gravitationally non-trivial background by turning on a coherent state of gravitons. That is, we dene a state which corresponds to creating an innite number of gravitons associated to a gravitational eld, and then include it as an additional insertion in every correlation function we wish to calculate. Assume the target space metric is approximately at and can hence be decomposed as G (x) = + h (x). Then, in the path integral on the string worldsheet, we can add an insertion5 exp d10 k d2 h (k)eikx x x = exp d2 h (x) x x (4)

to any correlation function calculated implying that all such functions must allow for interactions with this background. The LHS of this expression, interpreting in a correlation function, looks like a vertex operator corresponding to a coherent state of gravitons while on the RHS it is evident that it can be incorporated as a simple modication of the original Polyakov action, (3). Specically, the form of the coherent state allows us to incorporated it into the Polyakov action by replacing x x with G (x) x x . Similarly we can introduce backgrounds for any of the other NS-NS elds such as the anti-symmetric tensor eld, B , and the Dilaton, , resulting in a modied Polyakov action [Pol2] (omitting the fermionic components) S= 1 4 d2 g g ab G (x) + i
ab

B (x) a x b x + R(x)

(5)

Here, G , B , and correspond to the decomposition of the massless NS-NS sector modes into a symmetric, anti-symmetric and trace component (R is the Ricci curvature of X). ab is the two-dimensional anti-symmetric tensor. The RR sector modes are not so easily incorporated in this formulation. Note that the action given above applies for closed strings only since we have not allowed for interactions with open string modes such as A . More details can be found in the references. 2.2.2 Low-Energy Eective Field Theories

The essential point about (5) is that it can be used to constrain the nature of the low-energy quantum eld theory that corresponds to the limit of superstring theory when massive, stringy modes are integrated out. As described the references [Pol1] the string scale is set by ( )1 which appears as a prefactor in the energy scale associated with string excitations. In the limit 0 only the lowest energy string modes, corresponding to massless excitations, are expected to survive (since all other modes become innitely massive and their contribution to the path integral is correspondingly damped). is related to the string length so this is essentially a statement about an eective eld theory probed at a distance much larger than the string scale. For generic values of the coupling constants, G and B , the quantum theory corresponding to (5) is no longer Weyl-invariant. This is undesirable since a Weyl rescaling of the worldsheet metric g ab e(, )gab corresponds to a change of worldsheet scales rather than a change in the string embedding 6 and should have no observable consequences. As such, (5) should be constrained to be Weyl-invariant which can be done by calculating the beta-function for the various coupling constants in the 2-dimensional eld theory and requiring them to disappear [For]. The resulting equations relating the various coupling constants can be interpreted as space-time equations of motions for massless elds transforming under the same Lorentz representation as the coupling constant. This constrains the form of the background elds interacting with the string and hence provides a way to derive an associated low-energy eective action. This is perhaps somewhat unfamiliar so we will re-phrase it another way. The constraint of scale-invariance (or, in fact, conformal invariance) in the 2-dimensional world-sheet theory implies that its coupling constants obey certain equations. These coupling constants can be interpreted as elds (or more properly the pull-backs of elds) on the spacetime X and the equations they must satisfy can be thought of as equations of motions for these elds. It is then possible to derive a Lagrangian associated
5 Without delving greatly into the details let us remind the reader that in the string world-sheet theory there is a map between operators and states under which the operator x x creates a massless state with two spacetime indices corresponding to a massless two-tensor in spacetime. 6 The string map, x , is not necessarily an embedding but we will often be sloppy and refer to it this way for want of a better term.

14

2.2 The Physical Theory

2 RR-CHARGE AND K-THEORY

with these equations of motion and consequently derive a (generally non-renormalizable) quantum eld theory. It should be noted that the constraints on the elds so derived take the form of an expansion in , of which, generally, on the lowest order terms are known. The fact that superstring theory results in a spacetime theory with such a high-degree of supersymmetry strongly constrains the associated spacetime action in 10-dimensions (up to two derivative couplings [Cra]). The elds in these theories will depend on the coupling constants that appear in (5) and hence will be sensitive to the choices made in dening the string theory itself such as (for closed strings) the relative sign of the GSO projection in the left- and right-moving sector, the orientability of the fundamental string, and the character of worldsheet supersymmetry. The rst eld theories of relevance in this thesis arise from the closed string sector. Both these theories have two gravitinos from the R-NS sector and the NS-R sectors which implies that they posses N = 2 spacetime supersymmetry and are hence labeled type II theories. In fact, they correspond to the two possible 10-dimensional N = 2 supersymmetric eld theories incorporating gravity, otherwise known as supergravity theories. The two supergravity theories in 10-dimensions are type IIA and IIB and they are the low-energy limits of type IIA and IIB superstring theory. The dierence between them arises (in the string theory context) from the relative choice of sign between the GSO projections in left- and right-moving excitations. The GSO projection is a consistent truncation of the spectrum that, from the spacetime point of view, is required to make the spectrum supersymmetric and to eliminate tachyonic states with negative mass squared. It is also required from a world-sheet perspective to make the spectrum invariant under modular transformations which are symmetries of the worldsheet theory [Sza1]. In IIA the opposite sign projection is taken in the two directions and results in a non-chiral spectrum while in IIB the inverse is true. It is non-trivial to generate an action corresponding to the equations of motion described above (which arise from requiring the beta-functions of the 2-dimensional eld theory to disappear) in the case of type IIA and IIB string theory, particularly as the latter has a self-dual eld strength (as will be discussed below) which does not admit Lorentz-invariant descriptions. None-the-less a form of the action does exist and is presented below for the bosonic elds only [Pol2] Type IIA: SIIA = SN S + SR + SCS 1 SN S = 2 d10 x g e2 R + 4 |H3 |2 2 SR = SCS = Type IIB: SIIB = SN S + SR + SCS 1 SN S = 2 d10 x g e2 R + 4 |H3 |2 2 1 SR = d10 x G1 G1 + G3 G3 + G5 G5 2 SCS =
X

(6) (7) (8) (9)

d10 x G2 G2 + G4 G4
X

B2 G 4 G 4

(10) (11) (12) (13)

C4 H3 G 3

As only a few terms in the above actions will play any role in this thesis the others will not be discussed extensively. They are provided mostly to give a schematic sense of the background to the discussion that will follow. As these actions are generated by coherent excitations of massless modes (as seen above) the elds are all roughly in correspondence with the various massless excitations of closed type IIA or IIB string theory. The actions above have both been decomposed into terms containing NS-NS elds only, RR elds only, and terms containing a mixture of both (the Chern-Simons terms). The R-NS and NS-R modes are fermionic and have not been included. The integral measure has been explicitly denoted in 15

2.3 RR Field Strengths

2 RR-CHARGE AND K-THEORY

the above to distinguish between cases where a metric is required and where it is not (the Chern-Simons terms are top-forms and hence can be integrated without the need for a metric). In the future we will generally avoid denoting the measure explicitly. Both actions contain a spacetime metric7 , g , the 3-form eld strength of the anti-symmetric tensor, H3 H = [ B] = dB2 , the Dilaton, , and the Ricci curvature R. They also contain a host of RR-eld strengths, Gi , which are i-form eld-strengths (anti-symmetric tensors) arising from strings in the Ramond-Ramond sector. They will be the subject of the next section.

2.3

RR Field Strengths

As this thesis concerns sources for RR elds strengths, Gi we will restrict our attention to these. The RR eld strengths in the supergravity equations above are bispinors. The zero modes of a Ramond oscillator satisfy the dening relation of a Cliord algebra v v = Q(v) where v V for some vector space V and quadratic form Q. In this case the vector space is Tp X R1,9 , the tangent space at a point p X, = the vectors are the world-sheet spinors r whose spacetime indices identify them as elements of (the pull-back of) T X and the quadratic form is . The world-sheet anticommutator relations for r are
{r , s } = r,s

(14)

Only for r = s = 0 does a single set of modes dene a Cliord Algebra associated with the quadratic form . One can check that it is possible to construct a representation of this algebra with basis states 8 [Pol2, App. B]
4

|, , , , =

k=0

2k+1 2k (0 i0 ) |p , 0

R(R)

(15)

Where the action of 0 on this basis is just given by Cliord multiplication. Here |p , 0 R(R) is the ground-state of the Ramond (R) or Ramond-Ramond (RR) Hilbert space with momentum p . The states |, , , , form the 25 basis elements for a 10-dimensional spacetime spinor representation given by the 5 choices of (though many of these are eliminated by a world-sheet theory constraint and by the GSO projection). These states are also eigenstates of the 10-dimensional chirality operator, (11) , with eigenvalues +1 or 1 according to whether the number of pluses in (15) are even or odd, respectively. Since each Ramond oscillator denes a Cliord algebra as above closed Ramond-Ramond string states, with both a left- and a right-moving oscillator, are a tensor product of two such representations. A general result of Cliord algebra representation theory tells us that such a product decomposes into antisymmetric tensors of various ranks. In the IIA theory the rank of the tensors is always even while in IIB they are odd.

IIA : 16 16 = [0] [2] [4] IIB : 16 16 = [1] [3] [5] Here [n] denotes a rank n anti-symmetric tensor or an n-form and 16 or 16 are the irreducible positive and negative chirality representations of Spin(1, 9). See [Pol2, App. B], [Arg, Lect. 3] and [LM, Cor. 5.19] for more details. Note the for a general X with a general metric on its tangent bundle this construction must be undertaken using orthonormal frames [Nak], [LM]. 2.3.1 Generalized Electromagnetism

In [Pol2, Ch. 10] it is shown that the physical state condition on the supersymmetric worldsheet current is equivalent to a massless Dirac equation on the ground state spinors of the R sector. When this is applied to the above construction for producing anti-symmetric tensors from products of two copies of this sector it turns out that the antisymmetric tensors are not merely n-forms but are closed n-forms as
have switched to a lower-case g to avoid confusion with the RR-eld strengths, G i , which we are about to introduce. 1 0 1 0 spinors actually have to be formed somewhat dierently and have 0 + 0 rather than 0 i0 but this is inconsequential for the discussion and complicates the notation so in this section and in general later, when explicitly writing out spinor states, SO(p + 1) spinors will normally be described.
8 SO(1,p) 7 We

16

2.3 RR Field Strengths

2 RR-CHARGE AND K-THEORY

they adhere to dGi = 0 and d Gi = 0 [Pol2, Ch. 12]. In general the decomposition of bispinors into antisymmetric tensors would generate tensors of each degree up to the dimension of the space but the fact that the right and left moving spinors in the RR ground states are always chiral implies that in the decomposition of their product tensors of rank [n] and [10-n] are Hodge duals so that G i = G10i . This is evident in (9) and (12) since only forms of rank 5 or less appear. Moreover, this produces a subtle problem for degree 5 eld strengths since they are self-dual, G5 = G5 . This self-duality cannot be enforced at the level of the action (12) since then it would imply G5 G5 = G5 G5 = 0 (16)

This would generate no equations of motion for G5 . It is not possible to construct a Lorentz-invariant action for a self-dual eld so instead an action such as (12) can be used and self-duality enforced on the equations of motion rather than the action [Pol2]. While this prescription suces for the classical theory it does not work for the full quantized theory (the reader interested in learning more about this may consult [MW] [Wit3] and references therein). Since the Gi are closed they can be expressed locally as an exact form, Gi = dCi1 . More precisely, we employ a good open cover, U = {U }, of X for which all open sets, U , and their n-fold intersections are contractable for all n; such a cover is available for any manifold. Then, on each U there are closed i-forms, G(i) , and, because U is contractable and hence has trivial cohomology, they are also exact so G(i) = dC(i1) . On intersections of two open sets, U U , the Gi must match, G(i) = G(i) , to give a globally well-dened i-form but this need not hold for the C(i1) . This is a higher degree analog of electromagnetism with the C(i1) acting as potentials for the eld strengths. As with electromagnetism, C(i1) is only a local choice of representative of a class in i1 (U )/di2 (U ), the degree i1 forms on U modulo the image of the degree i 2 under the exterior derivative, since a dierent choice, C (i1) = C(i1) + d(i2) , of representatives still gives the same eld strength G(i) = dC(i1) = dC(i1) . In the supergravity actions above, however, there are additional subtleties. The relation dG i = 0 was derived for a string embedded in a trivial background (and without magnetic sources). In the case that the target space has a background H3 ux or non-trivial Dilaton there can be additional complications and the physical, invariant eld-strength might have non-standard Bianchi identities.9 This is the reason why in some cases Gi instead of Gi appear in equations (9) and (12). In the IIB theory these are [Pol2] G 3 = G 3 C 0 H3 1 1 G 5 = G 5 C2 H3 + B 2 G 3 2 2 (17) (18)

For the IIB theory we can neglect this complication by setting H3 = B2 = 0. As mentioned in the introduction this will be assumed to hold throughout this section. The complications that arise when this is not the case will be the subject of later sections. Thus far the RR elds have been seen to satisfy a generalized version (for higher degree forms) of the sourceless Maxwell equations. A natural extension is to introduce an electric source for these equations which would be of the form d Gi = j

dGi = 0

(19)

Here j is an electric source for the eld Gi (eqn. (19) can easily be seen to be the analog of maxwells equations with an electric source if written out in index notation in four dimensions and if j is replaced by j) however making the substitution Gi = G10i and using the fact that Hodge duality is an isomorphism so = 1 (depending on the degree of the form and the dimension) shows d (G10i ) = dG10i = j d G10i = 0
9

(20) (21)

See [Pol2, ch. 12] for an example with a non-trivial Dilaton background.

17

2.3 RR Field Strengths

2 RR-CHARGE AND K-THEORY

Thus an electric source for Gi is also a magnetic source for G10i implying that the latter no longer satises the Bianchi identity (dG10i = 0) and hence is not closed (near the magnetic source). The RR zero-mode spectrum has an obvious electric-magnetic duality given by interchanging an [n] form with a [10-n] form. Since the two are isomorphic and both appear in the spectrum it often suces (as above) to deal with only one of them but it then becomes inconsistent to introduce only electric sources since this would break the duality. One way to deal with magnetic sources is to consider the eldstrength only on the complement of the source where it is closed; this is the approach generally taken with magnetic monopoles, for instances. Alternatively one may resort to the language of gerbes. For a discussion of the subtleties involved in dening the eld-strength globally in the presence of a magnetic source see [MW] [FH]. 2.3.2 Poincar Duality and the (Co)homological Classication of Sources e

If one considers only electric sources then the discussion simplies somewhat. To generate a eld equation like (19) requires coupling the potential Ci1 to some closed form in the action10 Sj =
X

j Ci1

(23)

From eqn. (19) it follows that j = d Gi is clearly an (11 i)-degree form and is, moreover, closed since dj = d2 Gi = 0, hence an element of H 11i (X). Since j is exact it is trivial in H 11i (X) but it can still have topological signicance. Recall that, in general, a source has support only in some compact region whereas the eld strength has support that extends well beyond this region. This is true, for instance, for an electron or an electric charge distribution, whose charge can be measured by integrating the electric ux through a sphere enclosing the source (where, on the surface of the sphere, there is no 11i charge density). Thus if we consider the class of j in compact cohomology, Hc (X), it need not be zero 11 since G is not compactly supported. This is also relevant for magnetic sources where, as described above, the eld-strength is well-dened on the complement of the support of the magnetic charge and can be integrated along a sphere surrounding the magnetic source to calculate its charge. A nice discussion of this can be found in [MW, 2]. 11i The topological signicance of the class [j] Hc (X) is that it is the Poincar dual of a compact e submanifold, , which is the support of the charge density generating Gi . Let us develop this notion briey. Given a compact (and hence closed), oriented d-dimensional submanifold X , its compactness allows us to integrate any closed form, H d (X), over it

(24)

where denotes the pullback of the inclusion : X (see [BT, 5]). , as a compact submanifold of X, is an element of (Hc )d (X), the compact homology of X, and the integration can be shown to dene a duality (in the sense of vector spaces) H d (X) ((Hc )d (X)) . Another duality exists between H d (X) = 10d 10d and Hc (X) given by taking the wedge product of representative forms, (, ) H d (X) Hc (X) and then integrating over X, since the resultant form is a top form with compact support, and can be integrated to give a number (25)

X
10

Note that in standard electromagnetism the coupling is Z dd x Z p (|g|) j A = j A


X

(22)

but this is somewhat misleading since it is not necessary to use a metric to formulate this coupling. Rather, as will be discussed later, there is a closed form, , called the Poincar dual (or a current that represents it; see end of section), e which can be used to integrate A over X without using a metric. In the standard physics literature the hodge dual of this form rather than the form itself is used so j in (22) is equal to so j = = . For the purpose of our exposition it does not make sense to do this so we will always use j = as in eqn. (19) and (23) rather than j = as in (22). 11 The notion of compact support here depends on the particular situation. In general one might consider cohomology with compact support in the spatial direction.

18

2.4 The D-brane World-volume Theory

2 RR-CHARGE AND K-THEORY

In [BT, 5] it is shown that these dualities imply (with some other hypothesis) that (Hc )d (X) Hc = 10d (X) 10d so, for any Hd (X), there is a xed class, j = Hc (X), referred to as the Poincar dual of . e The class, , has the property that, for any H d (X)
X

(26)

In general we will denote the Poincar dual of via . Poincar duality will be use frequently through-out e e this thesis to shift integration from X to a submanifold and to classify such submanifolds. Although we have stated it here only for a submanifold the actual duality is between homology classes and cohomology classes (for various kinds of homologies and cohomologies) with arbitrary coecients. Thus even if is not a proper submanifold but is a cohomology cycle then it will have a Poincar dual. More e properties of this class will be developed in later sections as they become relevant. A reader unfamiliar with Poincar duality can think of it as a non-singular version of a delta-function (with support centered e on in X). From the discussion above it is easy to see that the dimension of in the case of eqn. (23) must be i1 (this also follows easily since it is only meaningful to integrate a top form on a manifold). This means that the candidate sources should be odd dimensional in the type IIA theories and even dimensional in type IIB (recall the potentials are one degree lower than the eld strengths). The natural choice for these sources are D-branes which can already be seen to exist as objects in open string theory in other ways. The notion that D-branes are classied, naively, by cohomology is essentially the statement that they are classied by the cohomology class of the Poincar dual of their world-volume. This is because this class e is a topological invariant and hence presumed invariant under reversible continuous deformations of the world-volume. In this case Ci is not a closed form so eqn. (26) does not apply directly. It is however possible to regard j as current in the mathematical sense. Such an object is an element of de Rham cohomology with distribution valued coecients. A distribution is essentially a linear functional T : Cc (M ) R which maps (innitely dierentiable) compactly supported functions to real numbers continuously (in a particular topology dened on Cc (M )). As an example, for any element L1 (R) the functional taking Cc (R) to dx (x)(x) is a distribution. Distributions also include delta-functions which take (x) to (0). If we consider the space of n-forms on M with coecients in Cc (M ) its topological dual (analogous to the distributions just described) is the space of currents. Constructing a cochain from these spaces (for dierent values of n) and dening its cohomology as the cohomology with distribution valued coecients it is possible to show that this space is isomorphic to the normal de Rham cohomology of a manifold. Hence we can replace j by the appropriate current given under this isomorphism. The reason for this construction was to establish that even for a regular form C i (as opposed to a closed form satisfying d = 0) j Ci = (Ci )

(27)

for some cycle . This means that potential coupling to a current term can be interpreted as integrating the potential over a submanifold which is the world-volume of the source for the associated eld strength. In the case of regular electromagnetism this submanifold would be the worldline of an electron or another charged particle. More details of the formalism of currents can be found in [GH, ch. 3][dR] and their application in this situation is discussed somewhat in [CY].

2.4

The D-brane World-volume Theory

In the last section we saw that in order to incorporate sources into the low-energy eective theory it is necessary to include subspaces, (cycles in the language of homology), over which the RR-potentials are integrated. Although these may often be submanifolds and in general we will restrict the actual world-volume, , to being a manifold, we do not wish to restrict to be a submanifold of X. This is too restrictive as it would not allow for self-intersections of in X while the latter is not forbidden by any physical consideration. In general the map : X will be an immersion meaning that, any point p has a neighborhood V such that (V ) is a submanifold of X [Lan3] (i.e. the self-intersections are not pathological). 19

2.4 The D-brane World-volume Theory

2 RR-CHARGE AND K-THEORY

From a string-theory perspective these are subspaces dened by open string boundary conditions which break some of the translational and rotation invariance of space-time. Normal (covariant) open string boundary conditions do not constrain the location of the string ends. However, it can be shown that a theory with covariant boundary conditions is actually T-dual (see below) to another theory where the boundaries are constrained to a specic coordinate value in a one direction. This denes hyperplanes which open strings are restricted to propagate on. Their properties are developed more in [Pol1, Ch. 8] but for the current exposition it suces to know that they are in fact dynamical objects as can be seen by various dualities relating D-branes in dierent backgrounds (of other open and closed string excitations). They also couple to the elds in the NS-NS sector and to the RR-potentials. Moreover the open string modes propagating on a D-brane admit an eective, low-energy description as a eld theory (which may be non-renormalizable if the brane is high dimensional) restricted to the worldvolume of the D-brane. The nature of this theory will be the topic of this section and will have important consequences for the rest of this thesis. It is this, from a low-energy description, that will invalidate the naive reasoning above which described RR-sources as nothing more than cycles, such as , in X. Here we provide a brief introduction to the worldvolume theory of a D-brane. In general, such a theory will correspond to a dimensionally reduced Super Yang-Mills theory with the NS-sector producing the bosonic part of the supersymmetric multiplet and the R-sector zero-modes acting as the corresponding superpartners. Once again, the exposition here will be schematic with more detail available in the references. 2.4.1 D-branes as Supersymmetric Excitations

The rst issue to note is that D-branes are in some sense excitations of the theory and thus are not expected to preserve the full symmetry of the ground state. If they are considered as subspaces (in the case of a at R1,9 spacetime) they break translational invariance and also reduce the rotational symmetry from SO(1, 9) to SO(1, p) SO(p) (this is the simplest case when a D-brane lies on a p + 1 dimensional hyperplane with one direction along the time axis). Recall from the previous section that the two string theories under consideration have N = 2 supersymmetry corresponding to the two spacetime supersymmetry generators which will be denoted Q and Q ( is a spinor index). These two generators derive, in the same sense that other low-energy currents and elds do, from specic string states. In particular, from the worldsheet currents associated to two gravitinos [Pol2]
1/2 |0; s; k 1/2 |0; s; k

N SR

u,s u,e s

(28) (29)

,s

RN S

,s

The objects above are simply states in the worldsheet theory quantum Hilbert space corresponding to a linear superposition of basis states, each of which has one NS creation operator ( 1/2 or 1/2 ) and a set of R creation operators decomposed in a basis of eigenvectors of a spin representation. Here s enumerates the spin states and is of the form (15) and u,s is a coecient. As these states contain R and NS oscillators they are from the NS-R or R-NS sector. The tildes above are used to dierentiate the right moving oscillations from the left moving ones. The important point is to note that the spacetime chiralities of the two supersymmetry charges are given by the spacetime chiralities of the Ramond ground states in the left and right moving sectors. These are xed by the GSO projection (or the relative sign of the GSO projection on the left and right moving oscillators) which in type IIA is taken to have the opposite sign in the two directions while in type IIB it has the same sign. As a consequence the supersymmetry generators have opposite or equal chiralities in IIA or IIB respectively. As mentioned above, one way to introduce D-branes is as a subspace that open strings can propagate on. In open string theory, however, the boundary conditions intermix the left and right moving oscillators so they are no longer independent. Whereas in the closed string theories above we have two separate (spacetime) supersymmetries generated by the right and left moving worldsheet Ramond zero modes, in open string theory it is clear that there can be at most one supersymmetry since there is only one set of Ramond oscillators. In this sense D-branes also break spacetime supersymmetry from N = 2 to N = 1 (or in some cases N = 0). This can be understood intuitively by noting that, in the presence of a D-brane, 20

2.4 The D-brane World-volume Theory

2 RR-CHARGE AND K-THEORY

a closed string can break into an open one (on the brane) which would mix the two supersymmetries on it into one. This point is relevant as it will allow us to determine the possible dimensionality of the D-branes in a theory. Type IIA and IIB string theory, as dened above, involve only closed strings but there is a consistent supersymmetric string theory containing open and closed strings. These strings, however, must be unoriented meaning that their worldsheet is an unoriented manifold. This is type I string theory, so-called because it has only one spacetime supersymmetry generator. Type I theory without D-brane sources is related to type IIA and IIB theories with D-brane sources via T-duality. T-duality is a ubiquitous tool in determining properties of D-branes and will be used frequently in the sequel. Essentially, a given string theory can considered with one dimension, x9 for deniteness, compactied to a circle of radius R and shown to be related to a T-dual theory compactied along the same dimension but with radius 1/R. The two theories can be shown to be equivalent in that, after proper identications have been made, they have the same spectrum and correlation functions and hence represent the same theory described in terms of dierent variables.12 The necessary identication (i.e. change of variables needed to switch between descriptions) is given by splitting the string modes associated with x 9 into their right- and left-moving parts (including the zero modes which include spacetime momenta) and then performing a spacetime parity transformation on only one side. In the open string case this results in changing the normal, covariant open string boundary conditions on x9 , which do not x the string endpoints but conserve momenta on the end of the string, in such a manner that the open string endpoints in the dual theory are xed at a particular value of x9 . The standard, covariant boundary conditions are referred to as Neumann boundary conditions while the boundary conditions which imply the strings are at xed coordinate values are Dirichlet. Thus T-duality interchanges Neumann and Dirichlet boundary conditions. A reader unfamiliar with this duality should consult [Pol1, Ch. 8]. Starting with type I string theory with freely propagating open strings will result in a T-dual theory where the open strings are restricted to lie on a xed hyperplane (with a xed coordinate value in the compact direction). For the closed unoriented strings of type I, T-duality combines with the orientation projection13 to restrict the closed string spectrum but in a non-local way. Rather than relating the rightand left-moving modes of a single string, the orientation projection instead relates the modes of strings at opposite spacetime points (e.g. x9 and x9 ). Hence the local physics in the T-dual theory (away from a D-brane) is that of oriented, closed strings. Since type I theory has equal chirality in both left- and right-movers its T-dual theory has opposite chirality in both directions. This follows because T-duality is given by taking a spacetime parity transformation on only one-half of the worldsheet oscillators. For 9 the Ramond modes this means that, if the T-duality acts in the x9 direction that the oscillator 0 is 9 9 related to 0 in the T-dual theory while the operator 0 is unaected. Recall, from the Section 2.3, 2k+1 2k ) and that the chirality operator has eigenvalues that the spinor states are of the form k (0 i0 1 according to whether the number of pluses is even or odd. A parity transformation on just on one coordinate would change the number of pluses by one so results in a change of chirality (on whichever set of left or right moving oscillators T-duality acts on). Hence the T-dual of type I string theory is locally (away from a brane) type II with opposite chiralities on both sides; this is just type IIA. From T-duality on the open string sector, which xes all open string endpoints to a particular coordinate value of x9 , we see that this theory contains D8-branes, 9 dimensional hyperplanes on which open strings can propagate. If two directions rather than one are T-dualized on then the relative chirality is not changed and the result is type IIB theory in the bulk with D7-branes. The terminology Dp-brane implies a p + 1-dimensional subspace of spacetime (the time direction always being included implicitly because the D-branes propagate in time).14 T-dualizing more dimensions shows that the bulk physics is IIA for an odd number of (T-dualized) dimensions and IIB for an even number implying that IIA has only odd-dimensional branes (like the D8-brane given above which is 8+1 dimensional) while IIB has only even-dimensional branes. This shows that the dimensionality of the branes corresponds to the degree of the RR-form potentials which is necessary for the D-branes to be sources for the associated
12 In string theory the world-sheet theory is considered the fundamental description of the theory so if two target spaces, X and Y , generate the same world-sheet theory then they are assumed to describe the same string theory and are thought of as being dual to one another; T-duality is an instance of this. 13 This is an operator on the closed string Hilbert space which allows one to describe unoriented strings starting from oriented ones. 14 We are ignoring the obvious exceptions of D-instantons for pedagogical reasons.

21

2.4 The D-brane World-volume Theory

2 RR-CHARGE AND K-THEORY

eld strengths. 15 It should be noted that this approach is somewhat incomplete; we generate theories with D-branes and N = 1 supersymmetry by starting with a string theory without D-branes but with N = 1 and then T-dualizing. Because T-duality is an exact symmetry in this case it cannot change the amount of supersymmetry in the theory so the dual theory must also be N = 1. A more complete approach would be to actually calculate the spectrum of open strings ending on D-branes and then see if this spectrum is supersymmetric. 2.4.2 D-brane Actions

Having xed the dimension of the D-branes, the next issue is to determine the kind of theories that live on the branes. As open strings are restricted to propagate on D-branes (theories with open strings propagating throughout spacetime can be described as having spacetime-lling D9-branes) there is a low-energy theory localized on the D-branes corresponding to integrating out the massive modes. This is analogous to the supergravity theories given in Section 2.2 as the low-energy descriptions of closed type IIA and IIB string theory. In the present discussion the detailed form of the world-volume theory will not be relevant. Rather it will be important to understand the kinds of eld that exist on the D-brane and their geometric character. This is because we are trying to derive the coupling of the D-brane to the bulk string action and this will be constrained by the existence of an anomaly in the D-brane world-volume theory (which is given from the topology of certain bundles dened on the D-brane). The bulk coupling will have to be determined in such a way as to cancel any anomaly emerging on the D-brane theory or else the theory will not be consistent. The reader unfamiliar with gauge or chiral anomalies is urged to consult Appendix C and references cited therein. As shown in Appendix C, in a theory with chiral fermions that couple to a gauge eld the classical conservation of the current coupling to the gauge eld may become anomalous. That is, the classical equation dJ = 0 where J is the current coupling to a particular gauge eld may no longer hold as an operator equation indicating that the associated gauge symmetry is no longer present in the quantum theory. As this is a gauge symmetry its absence implies the introduction of extra local degrees of freedom in the quantum theory that were considered unphysical in the classical theory. If the quantum theory was renormalizable to begin with then, as discussed in the appendix, this would render the theory nonrenormalizable and inconsistent. As the theories under discussion are often already non-renormalizable (for dimensional reasons) the argument here against anomalies in the low-energy eective action is that they would introduce additional local degrees of freedom that we know are unphysical in the original high-energy description (string theory). Thus it will be important to consider what kind of fermions exist in the world-volume theory and how they couple to various gauge elds in that theory. Note that it is important to separate, conceptually, fermions as seen by the D-brane world-volume theory from fermions transforming in a Spin representation of the bulk spacetime Lorentz group. This is because the above mentioned anomaly is given in terms of the Dirac operator on the world-volume (see below) and this operator only contains world-volume Lorentz indices. We will make this more precise momentarily. The eld content of the world-volume theory can be deduced, to some extent, merely by examining the massless excitations of the open string spectrum as these are the only modes that will be present at low-energies (relative to the string scale). The bosonic part can be calculated, much as for the closed strings by a beta-function calculation [For] or by consider a tree-level open string amplitude (a disk) coupled to a background open string gauge eld (i.e. a background generated by a coherent state of open string massless modes which are massless gauge particles, A , as mentioned above). The latter is done in [Sza1, Sec. 7, 8] and can be used to derive the D-brane worldvolume action SDBI = dp+1 e det (g (x) + B + 2 F ) x x a b (30)

where the are worldvolume coordinates and the open strings have also been coupled to closed string
the discussion here actually applies if we require the branes to preserve some of the vacuum supersymmetry. In that case the branes are BPS states and hence stable. This is directly related to their being charged under the RR-eld because the RR-repulsion between such branes cancels the gravitational attraction and implies that parallel congurations of such branes are stable. It also implies that they carry conserved quantum numbers associated with their RR-charge and hence must be stable.
15 Actually

22

2.4 The D-brane World-volume Theory

2 RR-CHARGE AND K-THEORY

background elds (g , B , and ). Here F is the worldvolume eldstrength associated with the gauge eld coming from the open-string NS zero mode, A (this should not be confused with Gi which is a eld-strength in the bulk theory coming from the closed string RR sector). This is the Dirac-Born-Infeld action (DBI). The rst thing to note is that this action contains a factor of so it can be expanded and considered in the regime where 0 as a low-energy approximation. In the case where the D-brane is 10 dimensional (only p < 9 dimensional branes were described above but D9-branes with 9+1 dimensions also exist) and the background is trivial the zeroth order action in is N = 1 supersymmetric Yang-Mills SDBI = d10 x T r(F F ) + 2iT r(iD) (31)

The full version of the above action, in a more general setting, where the dimension of the brane can be less than 10 and where multiple branes are allowed to overlap, is harder to derive as it must be generalized to accommodate A becoming a U (N ) gauge eld for N the number of D-branes and also includes fermionic modes and non-commutative scalar modes. The trace in the action above is to accommodate A becoming a U (N ) gauge eld as discussed below. 2.4.3 Chan-Paton Factors and Adjoint Bundles

Without worrying about the action let us try to understand what happens to the eld content of the theory if we allow N D-branes to wrap the same submanifold of spacetime, X. If each D-brane is labeled by an integer from 1 to N then any string state must be labeled by a pair |m |n where 1 n, m N corresponding to the labels of the D-branes on which each end of the string is xed (recall D-branes are dened as subspaces on which open string endpoints are forced to reside). These labels are referred to as Chan-Paton (CP) factors [Pol1, Ch. 6,8]. If the N D-branes all wrap the same submanifold, , then the string states |m |n are all degenerate; if however the D-branes wrap dierent submanifolds then strings between spatially separated D-branes would have a nite minimum length which would raise the energy associated with their oscillator modes and lift the degeneracy. Let us assume that the branes all wrap the same submanifold. A general open string state will be given by a linear superposition of these degenerate N N states which can be arranged into a matrix, which will be referred to as a Chan-Paton matrix, (Amn ) dx c11 |1 |1 c12 |1 |2 c13 |1 |3 1/2 |0; k N S c21 |2 |1 c22 |2 |2 c23 |2 |3 (mn ) = (32) |0; s; k R c31 |3 |1 c32 |3 |2 c33 |3 |3 ()(s) Tmn |0; k

where the cmn are normalized phase factors associated to each state. The CP matrices also carry subscripts or s depending on whether the state is in the NS or R sector. If each cmn is assumed to come from the product of two phase factors associated to the two states |m and |n so c mn = ei(m n ) then it is clear that the above matrix is anti-hermitian16 and hence takes values in the Lie algebra of U (N ), which we will denote u(N ). The column on the LHS indicates a eld in the low-energy eective action while the column on the RHS is the corresponding (massless) string state that generates it. Generally the low-energy actions of such elds must be determined using the calculations outlined above (e.g. by taking 0 limit of string amplitudes) but the tensor structure, as well as the mass of each state, makes it clear what kind of elds such states correspond to. The rst eld, A , is locally, at each point, mn a massless vector particle taking values in anti-hermitian N N matrices. Constraints on the string spectrum imply that each component of the CP matrix A is invariant under A A + mn for mn mn mn any local, Lie-algebra valued function . The second eld, mn , comes from open strings in the Ramond sector. Recall that the ground oscillators in this sector form a representation for the Cliord algebra associated to the tangent bundle of spacetime and the quadratic form and so this eld is locally a massless spinor taking values u(N ). The third eld, T , has no spacetime indices and so is, locally, a spacetime scalar taking values in u(N ). This eld has negative mass squared and is normally projected
16 The conventions in the physics literature dier from those in the mathematical literature by using e.g. iT a as a Lie-algebra element for a unitary group so Lie-algebra elements for such groups are considered hermitian rather than antihermitian in the physics literature. In this thesis we adopt the mathematical convention (T a ) = T a as it simplies the notation.

23

2.4 The D-brane World-volume Theory

2 RR-CHARGE AND K-THEORY

out of the spectrum by the GSO projection but there are instances when this will not occur and they will be very important later in this thesis so it is presented here for completeness. The local character of the elds above was determined by examining their tensor structure and invariances. Before using this information to determine the global geometric character of these elds there is an additional local symmetry that must be noted. Namely, that the low-energy elds dened by states such as (32) are invariant under conjugation of their Chan-Paton matrices by an element of U (N ). The reason for this is that in any string amplitude involving Chan-Paton matrices there must always be a trace over the product of these matrices. The states |m are invariant under the string hamiltonian and so do not change in the course of an interaction. So, in a string interaction, such as Figure 1 connected end-points must always carry the same CP label (another way to say this is that the lines connecting the end-points carry kroeneker delta factors). Summing over all degenerate in- and out-going CP factors on each end-point ultimately results in a trace over the product of the matrices associated with each state. For more details we refer the reader to [Pol1, ch. 6.5]. Recall that terms in the low-energy eective action will arise as an 0 limit of such amplitudes (in Figure 1 one can imagine, in the eld theory limit, that the incoming string is a gauge eld A while the outgoing ones are spinors and so that this would be a standard photon decay vertex, A ) so they will also contain a trace over the product of all the relevant CP matrices. Letting (1 )ij , (2 )kl and (3 )mn label the CP matrices associated with the strings at each end of the interaction in Figure 1 the associated eld theory vertex will contain a factor T r(1 2 3 ). Thus the eld theory will be invariant under conjugation of each of the i by the same U (N ) valued function i (p) g(p)i (p)g (p) where g(p) U (N ) and p . The fact that this symmetry is local, as suggest by the U (N )-valued functions g(p), rather than global can be deduced by studying string amplitudes [Pol1, Ch. 6].

k
ik

i
lm

j
PSfrag replacements jn

n
Figure 1: A string interaction with one in-going open string with CP-label |i |j and two out-going open strings with CP-labels |k |j and |m |n . Because CP-labels do not evolve the lines between the endpoints come with factors ik , lm and jn . Let us use this information to determine the geometric nature of these elds. Consider a good open cover of given by contractable open sets U (whose intersections are all also contractable) and note that on each U the considerations outlined above indicate that there is a Lie-algebra valued eld, A . On the intersections of two elements of this cover U U U these local elds are related as
1 A = g A g + d

(33)

The last factor, d , is to account for the fact that, on each U , the eld A is dened only up to a local choice of a gauge. Here g is a U (N ) valued function on U and is valued in u(N ). Likewise,

24

2.4 The D-brane World-volume Theory

2 RR-CHARGE AND K-THEORY

the elds and T will be related between elements of the cover by conjugation by U (N )
1 = g g 1 T = g T g

(34)

Note that the origin of this U (N ) symmetry, namely the trace in the string amplitude, implies that the matrices g must be the same in all three cases. Consistency on triple overlaps, U U U U , of (34) implies that on U the g satisfy the cocycle condition in the adjoint representation Ad(g )Ad(g )Ad(g ) = 1 (35)

and since the center, U (1), of U (N ) is in the kernel of the adjoint representation this implies that the actual matrices g (in the dening representation) only satisfy g g g = IN N (36)

for some U (1)-valued function, . From (34) it follows that and T are sections of a vector bundle with ber u(N ) and structure group Aut(u(N )). The relation (33) for A suggests that it is a local representative 1 for a connection on a principle bundle associated with the CP-bundle so long as = g dg . We will proceed under this assumption and describe the principle bundle below. Thus, in the presence of N branes wrapping a submanifold, , there is a vector bundle V adj with base-space in an adjoint representation of U (N ), Ad(U (N )) Aut(u(N )), and the elds in the lowenergy eective action are sections of this bundle. Given such a bundle, it may be possible to dene it as the tensor product of two vector bundles with structure group U (N ) in the fundamental and the conjugate representations. That is, we may dene W to be the vector bundle whose transition functions are given by the same U (N )-valued matrices, {g } above, but which act on the ber CN of W via the fundamental representation. We can then try to show that V adj W W where W is the bundle with = conjugate transition functions {g }. However, for W to be a well-dened vector bundle, the twisted cocycle condition (36) must actually be g g g = 1
17

(37)

It will be shown in Section 4 that this condition is satised when the eldstrength, H3 , has trivial topology. Since we are assuming B = 0 this condition is satised so W is well dened and V adj = W W. It should be pointed out that when one considers D-branes in a spacetime, X, where the bulk 18 antisymmetric tensor eld, B , is non-zero then the gauge transformation rules for B couple to those of A so the latter can no longer be interpreted as a connection on a principle bundle. This will be the starting point for Section 4. When B is xed at zero, however, the eld A is a well dened connection on a principle U (N )-bundle, PU (N ) , which is the principle bundle to which W is associated.19 Throughout the text we will use the superscript adj to explicitly denote adjoint vector bundles while bundles in the fundamental representation will carry no superscript. The elds A and also have additional local symmetries, such as Lorentz symmetries in the fundamental representation of the Lorentz group, for the former, and in a spin-lift of the same representation, for the latter. Thus these elds are also sections of the space-time cotangent bundle T X and the spin-lift of the spacetime tangent bundle S(T X) (see below) respectively. 2.4.4 The Spin-Bundle and Chiral R-Symmetry

The actions of relevance throughout this part of the thesis will be variations of (31) dimensionally reduced (to the dimension of the relevant brane) and with added couplings to the closed string sector. The exact
17 More precisely, it will be shown that a consistent choice of phase can be made on each double intersection, U , such that (37) is satised. Readers familiar with Cech cohomology can see that this relates to the cohomology class of in H 2 (, Cont(U (1))). See Section 4. 18 Bulk refers to the bulk of the spacetime manifold X rather than just the world-volume . This language is standard and will be used throughout the text. 19 P U (N ) is xed up to isomorphism by the set of transition function {g } dened on a xed cover.

25

2.4 The D-brane World-volume Theory

2 RR-CHARGE AND K-THEORY

form of such an action is not needed as the focus will be on perturbative gauge anomalies which will be xed by the form of the Dirac operator, iD. This operator acts on the fermions of the D-brane world-volume theory which come from the Ramond sector. As described above, the massless excitations of the Ramond sector are the zero modes which form a Cliord Algebra associated to the quadratic form and hence can be used to build a representation of Spin(1, 9). There are some subtleties here related to certain constraints on the worldsheet that reduce the dimensionality of this representation (see [Pol2, Ch. 10]) but these are not relevant here. It is relevant that the GSO projection acts on this reducible representation by eliminating one chirality so, as a Spin(1, 9), representation the associated fermions are chiral. For spacetime manifold X these fermions will be sections of a spinor bundle S(T X) which is the spin-bundle associated to the tangent space T X of X. This can be done precisely by taking a frame bundle over X such that the coordinates given with respect to this bundle are orthonormal and the transition functions are then SO(N ) valued rather than GL(R, N ) (assuming that X is orientable). The associated spin bundle S(T X) can be constructed assuming X is a spin-manifold (the second StiefelWhitney class vanishes [LM] [Nak]) which has already been assumed since we assume spacetime admits fermions. A more detailed discussion of the topological obstruction to dening spinors (fermions) is given in Appendix A.1. Thus the fermionic elds on a 10-dimensional brane (a D9-brane) will be sections of the tensor bundle S(T X) V adj . Consider now the case when this theory is restricted to a Dp-brane with p < 9 and let us restrict the discussion to IIB so p will be odd (and the p + 1 dimensional branes will always be even dimensional). The open string fermions are now restricted to lie on a Dp-brane and hence must transform under the decomposed representation Spin(1, 9) = Spin(1, p) Spin(p). This is not hard to see directly from the states corresponding to the fermion zero modes. Consider the spinor basis dened in eqn. (15)
4 2k+1 2k (0 i0 ) |p , 0 R

(38)

k=0

In 10-dimensions the above state carries a spinor representation of the Lorentz 20 algebra with generators S [ , ]. Under the full algebra where 0 , 9 the above state forms a single (albeit reducible via the chirality grading) representation. On a Dp-brane the oscillators in tangential directions have dierent boundary conditions than those in normal directions so the local symmetry group is reduced to SO(1, p) SO(9 p). If we consider only elements of the reduced algebra so(1, p) so(9 p) (so(N ) denotes the lie algebra of SO(N )) then either 0 , p or p + 1 , 9 but the indices cannot mix these dierent values which implies that we can split (38) into
(p+1)/21 2k+1 2k (0 i0 ) 4 2l+1 2l (0 i0 ) |p, 0 R

(39)

k=0

l=(p+1)/2

It is not hard to see that the two parts of this product transform independently as separate spinors of Spin(1, p) and Spin(p) and so form a (graded) tensor product. A reader familiar with Cliord Algebras may recognize this as an instance of a general theorem[ABS], [LM, Prop. 1.5] in the representation theory of Cliord Algebras (and hence of Spin-groups which are subgroups of the latter and have closely related representation theory). Namely, that if V , W , and U are vector spaces and V = W U , where the decomposition is with respect to the quadratic form Q, then the Cliord algebra associated to V and Q is given by C(V ) C(W )C(U ). Here is a graded tensor product [LM] and C(W ) and C(U ) are = constructed with the restriction of Q to these spaces. As a consequence, a representation of C(V ) (or its spin subgroup) gives a representation of C(W )C(U ). The result above is an application of this theorem 1,9 1,p 9p to the decomposition R = R R of the bers of the tangent bundle T X into the transverse and normal components. Denoting the tangent bundle of the D-brane, , as T and its normal bundle as N , this decomposition can be see as S(T X) = S(T N ) = S(T ) S(N ) in the notation xed above where S(T X) denotes the spin-bundle lifted from the frame-bundle of the tangent space of X. Letting S + (T X) and S (T X) denote the positive and negative chirality bundles21 one nds
20 As mentioned before the spinor basis shown above is actually for a Euclidean SO(10) spinor simply because it is notationally simpler. This will not be relevant for the discussion here. 21 Recall we have restricted to IIB and even dimensional branes here so T and N are both even dimensional.

26

2.4 The D-brane World-volume Theory

2 RR-CHARGE AND K-THEORY

S(T X) =S + (T X) S (T X) = [S + (T ) S (T )] [S + (N ) S (N )] =[S (T ) S (N ) S (T ) S (N )]
+ + + +

(40) (41) (42)

[S (T ) S (N ) S (T ) S (N )] S (T X) = [S + (T ) S (N )] [S (T ) S (N )]

The positive and negative chirality components of S(T X) decompose as (43)

Note that the chirality grading on the various spaces is with respect to their own, dierent, chirality matrices. It is not hard to see that acting with 11 = in 9 0 (here d = 2n + 2) decomposes into =0 the correct chirality grading on the two separate representations.
(p+1)/21 0 =0 p (p1)/2 =0 k=0 0 k=0 2k+1 2k (0 i0 ) (p+1)/21 2k (0

4 2l+1 2l (0 i0 ) |p, 0 9 R

in

=
4

l=(p+1)/2

2k+1 i0 )

(9p)/2 =p+1

0 l=(p+1)/2

2l+1 2l (0 i0 ) |p, 0

(44)

The equations above imply the decomposition (43) since the eigenvalue of 11 will be the product of the eigenvalue of the two other chirality matrices. We have assumed even co-dimension here so that p + 1 is always even dimensional (this is necessary for the equality (44) to hold). The chirality grading is only dened on even dimension. A spin representation of SO(N ) with N odd includes the chiral matrix in its algebra and hence is irreducible and has no chirality grading so we need only consider even codimension which is why we have restricted the discussion to type IIB theory. The anomalies we will consider in the next sections are chiral anomalies that only occur when the fermions can be decomposed into dierent chiralities and so will not occur in odd dimensional theories (there are anomalies in IIA string theory but only on the intersection of two D-branes which can be even dimensional). Thus under dimensional reduction the spinors of the worldvolume theory are sections of a tensor bundle S(T ) S(N ). They must also be in the adjoint of U (N ) with dimensionally reduced gauge eld, Aa , where a = 0, . . . , p is a world-volume index. The original gauge eld A decomposes under dimensional reduction to world-volume gauge eld, Aa , and scalars, m , where m = p + 1, . . . , 9. The scalars m are simply the components of the original gauge eld A with indices in directions transverse to the brane. For this reason m transforms as a vector under the SO(9 p) symmetry group of the D-brane normal bundle. On the eld theory dened on the brane, however, this looks like an internal symmetry group. Supersymmetry transforms both Aa and m into a fermion with elements from the aforementioned tensor product (this theory has the same number of states as the non-compactied one so this is not hard to see). Hence the fermions are actually sections of S(T ) S(N ) V adj where once more V adj is an associated vector bundle (in the adjoint representation) for the principle bundle for which Aa is a connection. Moreover S(N ), being the spin-lift of the normal bundle, is charged under the same internal SO(9 p) symmetry as the scalars. We briey mention that these scalar can be interpreted as D-brane coordinates in the transverse directions [Pol1]. When there are N D-branes wrapping a submanifold then these coordinates become matrix valued. The spectrum must still be truncated using the GSO projection22 so what remains is (if we project onto the positive chirality modes) S + (T X) = [S + (T ) S + (N )] [S (T ) S (N )] (45)

Note these spinors, seen as world-volume fermions, do not have denite chirality with respect to the world-volume grading. Hence they do not couple chirally to the gauge elds Aa . However they have an additional, internal, symmetry given by Lorentz transformation in the directions transverse to the brane
22 This is always dened on the full 10 dimensional theory so acts as a 10-dimensional chirality projection on the spinors irrespective of the D-brane dimension.

27

2.4 The D-brane World-volume Theory

2 RR-CHARGE AND K-THEORY

(this is evident from the coupling of the normal bundle above). Moreover this symmetry is chiral since the positive chirality world-volume bundle, S + (T ), is charged dierently under these transformation than the negative chirality bundle, S (T ). This is because S + (T ) is charged under a positive chirality spin representation of SO(9 p) while S (T ) is charged under a negative chirality representation and these may dier if the SO(9 p)-bundle (the normal bundle N ) has non-trivial topology. This is a potential source of gauge anomalies and will be explored in the next section. The additional internal Lorentz symmetry is an R-symmetry meaning it does not commute with the supersymmetry since the fermion and scalar elds are charged under this symmetry (recall that the m transform as components of a vector under transverse rotations; this is obvious from the decomposition of A and also follows because they represent the location of the D-brane in the transverse coordinates) but the gauge eld is not even though it is in the same supersymmetric multiplet. Another way to think of this is that the single ten dimensional spinor (a section of S (T X)) looks, under dimensional reduction to n dimensions, like 2(10n)/21 dierent n-dimensional spinors (sections of S (T )) that transform into each other under a Spin(10n) R-symmetry, where the latter is the spin-lift of SO(10n). This is essentially the statement that S(T X) S(T ) S(N ). = 2.4.5 The World-Volume Dirac Operator

For completeness, and to make contact with more familiar anomaly arguments, we present below the Dirac operator for the worldvolume theory which can be xed by noting the gauge symmetries of the fermions. As shown above, these are the internal adjoint U (N ) gauge symmetry of the gauge elds A a as well as the internal symmetry associated with transverse rotations; the latter is simply rotational symmetry in an internal coordinate. Moreover the fermions are charged under gravity on the world-volume which is also a gauge symmetry. The Dirac operator is then the standard operator for a theory with gauge symmetries (no attempt will be made to justify this form though it follows as the only gauge invariant kinetic term for fermions elds coupled to gauge elds which, as suggested above, is expected to be the nature of the worldvolume theory)
bc mn iD = if ef a a + [Aa , ] + a Sbc + a Smn bc mn = if ef a a + [Aa , ] + a [d ed b , e ee c ] + a [p ep m , q eq n ]

(46)

Note that the gamma matrices (which are represented as part of the world-volume theory not as actual string oscillator modes 0 ) must be dened with respect to an orthonormal frame bundle dened eb as mentioned previously (see [Nak] for more on this) resulting in the rather complex expression above. Also, the gauge elds are non-Abelian and act in the adjoint representation on the spinors from the right and the left which is denoted in (46) via [Aa , ]. Recall the conventions given in Section 1.7: letters from the beginning of the alphabet are world-volume indices, those from the later part are transverse indices, and a bar over the index indicates it represents a frame-index. The Sab and Smn are generators of world-volume and transverse Lorentz transformations, respectively. Viewed as a dierential operator on the space of sections, (46) acts as follows iD : S + (T ) S + (N ) V adj S (T ) S + (N ) V adj

iD : S (T ) S (N ) V adj S + (T ) S (N ) V adj

(47)

This follows most easily by noting that it acts on the spinors as a world-volume gamma matrix which anti commutes with the worldvolume chirality operator {a , (p+2) } = 0 (where (p+2) is the world-volume chirality matrix in p + 1 dimensions). Equation (46) will not be used to calculate any anomalies explicitly (though this can be done [SS]). It is provided only to give a concrete sense, from a physical perspective, of what the theory looks like. Potential anomalies associated to fermion zero modes on the world-volume theory due to chiral coupling of the fermions to the R-symmetry of transverse rotations (generated by Smn ) will be calculated in a later section by appealing to a more geometric formulation.

28


(D, N, N)
X1

PSfrag replacements

It is possible to see that such strings can be chiral directly by considering the string mode expansion with mixed DN or ND boundary conditions but a simpler approach will be taken here. Recall that Dp-branes can be obtained from type I string theory by T-dualizing 9 p directions. The type I theory possess only N = 1 supersymmetry. To see this most clearly represent the two generators of the type IIA/B N = 2 supersymmetry, Q and Q , in a new basis, Q + Q and Q Q . In standard IIA or IIB these would be linearly independent, 10-dimensional chiral spinors however, in type I string theory or its T-dual, type II A/B with D-branes, the open string and unoriented close string boundary conditions imply Q = Q (as these generators derive from left- and right-moving worldsheet oscillators respectively) so only one linearly independent generator, Q + Q , actually remains giving an N = 1 theory [Pol2]. If the supersymmetry generators in the theory are chiral then the resultant fermion spectrum must itself be Figure 2: A depiction of strings (in R3 ) with ND, DN, NN, and DD endpoints. The gure shows two D2-branes (2+1 dimensional but only spatial dimensions are shown above) in R3 (or R1,3 but the gure only depicts the spatial dimensions x1 , x2 , x3 ) one wrapping the x1 x3 plane and the other on the x2 x3 plane. The coordinates of the string endpoints are labeled by a triple (x1 , x2 , x3 ) where, in the place of each coordinate, an N indicates Neumann boundary conditions along that direction and a D indicates Dirichlet. Note that the string with ends on dierent D-branes has mixed boundary conditions in the x 1 and x2 coordinates (for its dierent ends). Before turning to the calculation of the R-symmetry anomaly it is important to investigate another source of potential anomalies, namely D-brane intersections. When two D-branes intersect, their intersection, which will be referred to as an I-brane, may have a dierent spectrum than the two branes themselves (the spectrum will not be equivalent to the spectrum of a D-brane of the same dimension as the I-brane). The reason for this is that near, or on, the intersection open strings can have one end-point on each brane. Strings not restricted to the I-brane but near it will have some directions in which their boundary conditions are Dirichlet (D) on one end and Neumann (N) on the other. This could not happen for strings xed to a single brane. Some illustrative examples are shown in the simplied 3-dimensional setting in Figure 2 below. 2.4.6 2.4 The D-brane World-volume Theory X2 (X1 , X2 , X3 ) I-Branes and Chiral Fermions

(D, N, N)

(N, D, N)

29

(D, N, N)

2 RR-CHARGE AND K-THEORY

(N, D, N)

(N, D, N)

D-brane 2 D-brane 1 X3

2.4 The D-brane World-volume Theory

2 RR-CHARGE AND K-THEORY

chiral (since its fermion spinorial indices come from a supersymmetric variation of a bosonic eld). In the 10-dimensional open string theory the generators are chiral because of the GSO projection. The question that will be addressed below is whether these generators, which are chiral with respect to (11) , are chiral with respect to the grading matrix of a particular D-brane worldvolume theory (or the theory on the intersection of two worldvolumes). If so, then the resultant fermions are chiral and might be anomalous. To study this we will, once more, resort to T-duality to understand the supersymmetry on the D-branes. 9 9 Recall that T-duality acts as a one-sided parity transformation on the worldsheet taking 0 to 0 9 9 and leaving 0 invariant (for T-duality in x ). This can be implemented on the spectrum by acting with 9 (11) which commutes with all the oscillator modes except 9 with which it anti-commutes.23 Let (d+1) denote the grading matrix of Spin(d); we will adopt this convention consistently in the sequel. If we T-dualize on multiple coordinates we must modify this operator to incorporate each dualized (transverse) coordinate resulting in the product m m (11) . From the explicit Hilbert-space states corresponding to Q and Q given in (28) and (29) it follows that the two spacetime generators have spinor indices coming from dierent sides of the string worldsheet and hence will be transformed dierently by T-duality. For deniteness assume Q is unaected while Q must be transformed by m m (11) . Let us start by considering what the N = 1 supersymmetry in the type I theory, given by the linear state Q + Q , implies for D-branes in the T-dual theory. We have already suggested that Dbranes break the bulk N = 2 to N = 1 but let us see how this emerges. Consider a Dp-brane with worldvolume coordinates S1 = {0, . . . , p} and let S1 = {p + 1, . . . , 9} denote its transverse coordinates m (11) be the associated transformation accounting for the T-duality. Then a and let 1 = mS 1 Dp-brane dual to type I string theory has one N = 1 supersymmetry given by
Q + (1 ) Q

(48)

In words, eqn. (48) means that the single supersymmetry generator of type I string theory, Q + Q , gets translated, under T-duality, which relates type I string theory to type II A/B string theory with D-branes, into (48) on the D-brane world-volume. The pair, Q and Q , look, locally, away from the D-brane, like the two independent generators of type II string theories (which is why we have distinguished them rather than using into a single variable for the N = 1 generator, (48)). This local independence is illusory however (as discussed earlier) because these generators are related to each other, even in the dual theory, but in a non-local way; that is Q (x9 ) is related to Q (x9 ) in the dual theory (assuming T-duality in the x9 coordinate as discussed previously). At the D-brane this relationship becomes local (in this simple setup the D-bane is xed at x9 = 0). Alternatively one can phrase this condition24 as Q = 1 Q, which implies that strings ending on the D-brane require this matching between their right- and left-moving Ramond modes and hence break one of the bulk supersymmetries (i.e. the normal condition that right and left-moving modes must be equal is twisted by the T-duality). For more background consult [Pol2, Ch. 13]. If one now considers intersections of two branes and in particular strings with endpoints on both branes then they must satisfy this condition for both branes [Pol2][PCJ]. This implies
Q = (1 ) Q Q = (2 ) Q

(49) (50) (51)

Q =

(2 )1 (1 ) Q

If (51) is not satised then the spectrum of strings stretching between the two branes will not be supersymmetric. This is because the rst brane is only supersymmetric under components of the supersymmetry generators satisfying (49) while the second is only supersymmetric under components satisfying (50) so the intersection will only have a supersymmetric spectrum if these components overlap (recall Q and
23 The gamma-matrices are being used abstractly here as operators on the zero-mode spinors. They normally are just given by the 0 themselves but there may be normalization factors. In the case of product representations the e sometimes represent tensor products of 0 and 0 so for simplicity we just denote them abstractly. 24 This discussion can also be used to show that supersymmetric branes (preserving at least have the supersymmetries) e exist in odd/even dimension for IIA/B theories as Q = 1 Q implies opposite/equal chirality of the two sides in odd/even codimension.

30

2.4 The D-brane World-volume Theory

2 RR-CHARGE AND K-THEORY

Q are generators with many components). For any two branes it is possible (assuming the spacetime geometry admits T-duality in the appropriate directions) to T-dualize enough directions that one of them becomes a spacetime lling D9-brane in the dual theory and this is the case that will be considered rst (it will then be shown that the result is independent of T-duality). Thus the transverse coordinates are S1 = {p + 1, . . . , 9} and S2 = so the condition for strings on the intersection to be supersymmetric, eqn. (51), reduces to
Q = (1 ) Q

(52)

For (52) to hold it is necessary that the matrix by squaring the matrix
2 (1 )2 =
mS1

(1 )

have some +1 eigenvalues. This can be checked

m (11)
p+1

=
mS1

m . . . 8 9 ) = (p+1 . . . 7 )(p+1 . . . 8 9 8 9 )
|S1 | 2

= (

. . . )(

8 9

p+1

= (p+1 . . . 7 )(p+1 . . . 8 8 9 9 ) = (1)

It has been assumed that |S1 |, the number of transverse dimensions, is even which indeed must be the case if there is a spacetime lling D9-brane (since the theory is then IIB).25 If (1 )2 = 1 then its eigenvalues will all be i and (52) will have no solutions. Hence |S1 | must be a multiple of 4 which implies that the possible values of p are 1, 5, and 9. It is shown in [Pol2, Ch. 12] that the general requirement for chiral fermions is that the number of ND (or DN) directions be a multiple of 4. The case p = 9 is somewhat uninteresting at this point so the other two will be considered instead. Recall that the GSO projection implies that the supersymmetry generators are each chiral with respect to (11) so, in addition to (52), the condition ((11) ) Q = Q holds. Together these imply Q = (1 )Q =
mS1

m (11) Q = (p+1 . . . 9 )Q

(53)

and Q = (11) Q = (0 . . . 9 )Q which gives Q = (0 . . . 9 )Q = (0 . . . p )(p+1 . . . 9 )Q = (0 . . . p )Q = (p+1) Q (54)

So the supersymmetry on the Dp-brane (which, in this case, is the intersection because the other brane is a D9-brane) is chiral and there are chiral fermions in the spectrum. Note that (53) only holds for some of the Q , namely those in eigenspaces of 1 with a +1 rather than 1 eigenvalue. This has no eect on the discussion above since it is only those spinors with +1 eigenvalue that continue to generate supersymmetry and hence fermions on the intersection but it is worthwhile to note that it is not necessarily all the components of the original 10-dimensional N = 1 supersymmetry which are preserved here. The discussion has so far been restricted to the case where one brane is a spacetime lling D9-brane. To see that this eect is more generic one must consider the eect of T-duality on the relations above (to return to the original conguration of a Dp-Dp -brane that existed before T-dualizing). Consider the Dp-D9 brane system discussed above. There are two ways of applying T-duality to this system. If the T-dual is considered with respect to a common dimension of the D-branes (i.e. an element of S 1 ) then the result is that both branes lose a dimension to give a D(p 1)-D8 brane. Alternatively, a dimension transverse to the Dp-brane can be dualized to give a D(p + 1)-D8 system. Consider the latter case and take, for deniteness, p = 5 (the discussion will be independent of this value but the exposition will
25 We

will see how this carries through to IIA by T-duality momentarily.

31

2.4 The D-brane World-volume Theory

2 RR-CHARGE AND K-THEORY

be simplied by the use of a denite value). Then, after T-dualizing (on x9 ) to a D6-D8 system the transverse coordinates are S1 = {6, 7, 8} and S2 = {9}. As a consequence (52) becomes Q = ((2 )1 ) (1 ) Q By noting that, for any , ( ) = 1 it is easy to see and S2 sum to S1 so (55) becomes (note, this depends
2

(55) = = 0) 2 . Also note that the sets S1 (56)

that (2 )1 on S1 S2

Q = (2 ) (1 ) Q = (1 ) Q

Thus (52) still holds up to a plus or minus which is irrelevant since the analysis only required the square of 1 to have eigenvalues +1 (so is insensitive to the dierence between 1 ). Also note that the 1 S2 = {0, . . . , 5} so the rest of the analysis intersection of the two branes has not changed and is still S applies. Continuing to T-dualize on elements of S1 gives the same results so leads to D7-D7 systems. Likewise it is possible to start from p = 1, which is also chiral (as noted above), and apply the same analysis resulting in a D2-D8, D3-D7, D4-D6, and D5-D5 system, all intersecting on a 1-brane with a chiral world-volume spectrum. Thus in all these cases there will (potentially) be an anomaly on the branes, localized on their intersection, resulting from the chiral gauge coupling (because there are only chiral fermions, see Appendix C). This is in addition to the other source of anomalies described above on individual branes in IIB. In order for the whole theory to be consistent the bulk theory must also be anomalous in such a way as to cancel the two anomalies discussed so far to give a consistent theory. To make this more precise rst requires a calculation of the value of the anomaly. T-duality on tangential coordinates (common dimensions of the branes) will not generate chiral fermions on the intersection as can easily be checked using the methods above. The fermions generated, however, will have chiral coupling, similar to the R-symmetry coupling, as will be discussed in Section 2.4.6 below. Note that we have taken the D-branes to lie on coordinate axis so that they always intersect at right angles. More generally, this could be done using orthonormal frames (which are, in any case, required to introduce spinors). On the intersection the separate D-branes are then each assumed to coincide with local coordinate directions in the veilbein basis. It is possible to consider D-branes that intersect at non-orthogonal angles and still preserve some supersymmetry but this will not be done here [Pol2]. The use of T-duality in the discussion above limits its applicability as it requires a suitable set of isometries26 in a compact direction and requires the branes to be oriented along these. The discussion can be made more general by calculating the local string spectrum (as will be done below) near the brane intersection and then checking if any local spinors patch to form a global spinor (using topological arguments similar to those found in Appendix A.1). The T-duality argument given above has been favored, pedagogically, for its simplicity. See also [Pol2, Ch. 13.4-5]. 2.4.7 I-brane Spin Bundle

In order to calculate the anomaly, which is done in the next section, it will be necessary to have an understanding of the geometrical nature of the spinors of the I-brane world-volume theory. The anomaly is related to the analytic index of the Dirac operator on the I-brane worldvolume which can be calculated using an expression similar to (47). The latter was derived by noting that the fermions on a D-brane worldvolume are sections of the spin bundle lifted from the space-time tangent bundle decomposed into tangential and normal components: S(T ) S(N ). As already mentioned, the spectrum of an I-brane diers from that of a D-brane of equal dimension and so the analysis given for a single D-brane cannot simply be applied. Rather, it will be found that the fermions on the I-brane world-volume are spinors lifted from the bundle T 1 T 2 N 1 N 2 (57)

To see where (57) comes from requires a closer examination of the string spectrum for strings on the I-brane. This is discussed in [Pol2, ch. 13.4-5] and [PCJ]. Consider a local coordinate chart near the intersection of two branes such that the branes are dened to extend over particular coordinate
26 To T-dualize on a particular coordinate requires a killing vector in a compact direction that will dene that coordinate and along which all the relevant background elds, not just the metric, are invariant.

32

2.4 The D-brane World-volume Theory

2 RR-CHARGE AND K-THEORY

(or veilbein) directions (recall that we are considering branes intersecting at right angles). Let S 1 = {0, 1 , . . . , p } be the coordinates of one brane and S2 = {0, 1 , . . . , p } the coordinates of the other. Let S12 = S1 S2 and S1 = S1 S12 and S2 = S2 S12 and T be the complement of S1 + S2 (in 0, . . . , 9). Consider the bosonic elds, x , on the string worldsheet corresponding to its spacetime coordinates and consider a string with one end on either brane. Both ends of such a string can propagate freely along coordinate axis tangential to both branes (i.e. S12 ) while for axis tangential to one brane but perpendicular to the other only one end of the string is free to move. Both ends of the string are at xed locations for axis perpendicular to both branes. This is depicted pictorially in Figure 3.
Z

Dp -brane

N
X

D N

Dp-brane

N N

PSfrag replacements

Figure 3: An I-brane string stretched between a Dp-brane and a Dp -brane. The label N or D on the arrow in each direction indicates the type of boundary condition on x in that direction. Here S1 = {x, y}, S2 = {x, z}, S12 = {x}, S1 = {z}, S2 = {y}, and T = . Recall that Neumann boundary conditions in x imply that the string ends ( = 0, ) are free to propagate along the coordinate-axis while Dirichlet boundary conditions imply that the ends are xed along that coordinate axis (i.e. they assume some constant value). x (, )|=0, = 0 x (, )|=0, = 0

(Neuman) (Direchlet)

(58) (59)

From the description of string propagation in the last paragraph it follows that if S 12 then Neumann boundary conditions should be applied to both ends of the x (, ) world-sheet eld (such a eld will be referred to as NN). If however, S1 then one string end, = 0 for deniteness, is xed while the other can propagate so x (, ) should have mixed boundary conditions: Dirichlet at = 0 and Neumann at = (such a eld will be referred to as DN). Likewise it follows that S2 gives an ND eld (where the rst boundary condition applies to = 0 and the second to = ) and if T then that eld has DD boundary conditions.

33

2.4 The D-brane World-volume Theory

2 RR-CHARGE AND K-THEORY

If one considers the decomposition of x into oscillator modes (and ignores the zero-mode oscillators) x (, ) = i
r

r ir( +) e eir( ) r

(60)

it is easy to see (by checking the boundary conditions) that if x is NN or DD then r Z and the must be a + and , respectively. If x is ND or DN then r Z + 1 and is + and , respectively. Because of 2 the way the worldsheet supercurrents are dened and the associated denition of the Ramond and NeveuSchwarz sectors the Ramond sector must always have the same modding (e.g. integer or half-integer) as the bosonic elds while the NS sector has the opposite modding (or is o-set by a half-integer). Recall that worldsheet modes only transform as spacetime fermions if they form a Cliord algebra associated with the quadratic form and this happens when, in the worldsheet anticommutator re lationship, {r , s } = r,s , the kronecker delta reduces to unity (i.e. when r = s = 0). To see when this is the case we consider rst the Ramond sector and restrict the discussion to massless states. Because the worldsheet fermions in the Ramond sector have the same periodicity as the bosons the ground state energy in this sector is always zero. Any bosonic oscillators would raise the energy to some positive value and are hence not considered. Consider a single (worldsheet) fermionic oscillator, r . If S12 or T then r Z so r = 0 modes exist that do not raise the energy of the ground state and, 1 1 moreover, form a Cliord algebra. If S1 or S2 however then r Z + 2 so r = 2 is the lowest energy excitation but still raises the ground state energy and hence, in the massless spectrum, should be neglected. The massless spectrum therefore consists of spacetime fermions with indices from S 12 + T . Note that S12 or equivalently x being NN implies that is a direction tangential to both branes and consequently associated with T 1 T 2 . Likewise T implies is perpendicular to both branes so it is a coordinate in N 1 N 2 . The same reasoning implies that ND and DN modes corresponds to coordinates in T 1 N 2 and N 1 T 2 . With this correspondence it is easy to see that the fermions coming from the Ramond sector with indices in S12 + T must be sections of the spinor bundle lifted from T 1 T 2 N 1 N 2 which is just (57). As to the NS sector, because they are half-integer modded in the NN directions they cannot be worldvolume fermions since their (worldsheet) fermion components in S12 do not form a Cliord algebra. Even though their DN and ND excitations make them internal spinors they do not contribute to the fermionic eld-content of the I-brane worldvolume theory and, hence, can be neglected for the purpose of anomaly calculation. A complete analysis of the spectrum along the lines already began can also be used as an alternative method to detect when the spectrum on the D-brane is chiral but this will not be pursued here [Pol2]. One nal point is worth mentioning here. The fermions on a single D-brane were shown above to be lifted from the spin-bundle associated to T N . This spin-bundle must exist because T X = T N is assumed to admit a spin-lift. Hence, even though S(T ) and S(N ) may not be well dened bundles their tensor product, S(T ) S(N ) S(T X), must be. The same, however, cannot be said of = S(T 1 T 2 ) S(N 1 N 2 ) since the relevant bundles do not sum up to T X. Rather T X = T X T X = [T 1 N 1 ] [T 2 N 2 ] (61)

= [T 1 T 2 ] [N 1 N 2 ] [T 1 N 2 ] [N 1 T 2 ]

This issue was pointed out in [FW] and its resolution is related to a constraint placed on the D-brane worldvolume by a global27 or non-perturbative anomaly in the string worldsheet theory. In order to cancel the Freed-Witten anomaly, as it is known, the manifold on which a D-brane is wrapped must be a Spinc -manifold28 which means that, even if S(T ) is not a well-dened bundle, there exists some line bundle over whose transition functions are undened in the same way as S(T ) so that the tensor product of both is a well-dened bundle (see Appendix A.2 for more details). This will be of importance later and will be discussed in more depth in Section 2.7.
27 The term global anomaly here does not refer to an anomaly in a global symmetry but rather an anomaly associated with disconnected components of a symmetry group (such as a gauge group). This is discussed in [Nas]. 28 When H is topologically trivial; see Section 4.

34

2.5 The D-brane Anomaly

2 RR-CHARGE AND K-THEORY

2.5

The D-brane Anomaly

The results of the previous section will now be used to calculate the anomalous gauge variation of the Dbrane world-volume theory. It is shown in Appendix C that when gauge elds are coupled to fermions in chiral manner in an even dimensional eld theory the associated fermion determinant, which is a factor in the quantum eective action, is no longer gauge invariant. As gauge degrees of freedom are unmeasurable and hence unphysical29 this implies that the theory is inconsistent. Such anomalies are generally given in terms of topological invariants which are integrals of geometric quantities such as curvature forms; this relates to the fact that anomalies emerge from the non-triviality of certain bundles dened either on spacetime or on some conguration space [AGG1], [Ber], [dAI]. Specically, the anomalies will be given by polynomials in the gauge eld-strength and Riemannian curvature tensor on the D-branes; such polynomials are known as characteristic classes. For anomalies on intersecting D-branes it will be important that the form of these polynomials allows them to be factored. That is to say, the anomaly polynomial associated with intersecting D-branes can be split in such a way that the two separate factors depend only on one of the two branes each. This will allow the anomalies to be canceled by modifying the bulk coupling to the individual D-branes. Otherwise, if the cancellation were to depend on the intersection world-volume, then the action would have to be modied any time two D-branes intersected and this would be very unnatural. The anomalies below were rst calculated in [GHM] and then rened in [MM] and [CY]. The exposition follows [CY] most closely though more detailed calculations are presented and some additional points deriving from [MM] and [FW] are brought up in the calculation of the anomaly on the intersection. 2.5.1 Gauge Anomalies and the Descent Procedure

Recall from the previous section that there are two sources of anomalies associated with D-brane worldvolume theories: chirally coupled R-symmetry on individual branes and fully chiral theories on the intersection of two D-branes. The former will only be an issue in the IIB theory where the brane worldvolumes are even dimensional since chiral anomalies only occur in even dimensional theories. The latter occurs in both avors (IIA and IIB) since I-brane volumes can be even dimensional even if the individual branes are not. The anomalies can be calculated explicitly using Feynman diagrams but we eschew this technique as the high-dimensionality of the theory makes it cumbersome. Moreover, we are ultimately going to be interested in the geometric character of the anomaly polynomials and these are somewhat obscured in the explicit diagrammatic calculation. Rather we will use a technique described in Appendix C wherein the gauge anomaly in dimension d can be calculated using the Dirac operator on a d+2 dimensional manifold. We review and motivate this very briey here. The un-familiar reader is urged to consult the references given below as it is unlikely that this exposition will provide a sucient introduction. Start by considering the quantum eective action which is assumed, for simplicity, to be the functional of a single gauge eld, A. eiW [A] Z[, ] DD eiS[,,A] (62)

W [A] is a formal quantity called the quantum eective action which denes the quantum theory in terms of something that resembles a classical action with an innite number of terms. It is useful in instances such as this when one is interested in the presence of a symmetry in the quantum theory. Note that in expression above we are not integrating over A but rather treating it as a xed background eld. W [A] can be seen as a functional on the innite-dimensional space, U, of gauge connections. We will not attempt to make this mathematically precise. If W [A] is gauge invariant it should really be considered a functional not on U itself, since many of the individual points in U are related by a gauge transformation, but on U/G which is the quotient of U under the action of all gauge transformation, G. If W [A] is not gauge invariant then it looks more like a section of a non-trivial bundle on U/G because its value at a xed reference connection, A1 U/G, will depend on a choice of gauge (so, in some imprecise sense it is associated to the bundle U U/G). If this is the case then it is not possible to dene the full path
29 This is similar to phase in quantum mechanics. Phase dierence can be measured but the actual value of a phase is not a physical quantity.

35

2.5 The D-brane Anomaly

2 RR-CHARGE AND K-THEORY

integral DD DA eiS[,,A] = eiW [A]


U/G

(63)

because the integrand in the last expression is not a well-dened functional on the space, U/G. A reader unfamiliar with this perspective on anomalies can consult Section 4 for a similar discussion of the Freed-Witten anomaly or see [Fre2]. The purpose of introducing these spaces is to formally dene a dierential, S, on the total space U. This dierential is dened to adhere to S 2 = 0 and can be used to dene a cohomology on this space. If we consider the space U U/G as a bundle with base space, U/G, given by gauge inequivalent connections and typical ber G corresponding to gauge transformations then we can dene = S| f ibre which gives the component of the dierential acting along the ber. This is an innitesimal gauge transformation and, moreover, is nil-potent 2 = 0. The details of this construction can be found in [AGG1], [AGG2], [Ber], [Nak], [dAI] and [HT]. On the physical manifold on which the theory is dened there is already a dierential operator d, dened on dierential forms on this space; commutes with this form since they act on dierent spaces [dAI]. The dierential can now be used to calculate the anomaly as follows. The anomaly is given by the non-triviality of the quantum eective action under a gauge transformation (I + )W [A] = W [A] + W [A] (64)

As the anomaly is given by W [A] it is clear that it must be in the kernel of the dierential generating such innitesimal transformations since (W [A]) = 2 W [A] = 0 which is, it turns out, a sucient constraint to x it up to a normalization. The condition G[A] = 0, where G[A] = W [A] is the anomaly polynomial, is referred to as the Wess-Zumino consistency condition. There is a procedure to nd such a G[A] starting with P(d+2) (A), the degree d + 2 part of the curvature polynomial associated with an Abelian anomaly in d + 2 dimensions (see Appendix C). This is a somewhat subtle point and is reviewed more in Appendix C and the references but we are essentially formally calculating the Abelian, global anomaly associated with the same Dirac operator (extended to d + 2 dimensions) assuming it was not chirally coupled (so given iDP+ we take just iD) but we are considering the terms of degree d + 2 which are only dened formally in dimension d (since they normally vanish). It is not obvious why this is the correct procedure but there is a geometrical interpretation of this given in [AGG1], [AGG2] and [dAI]. The exact form of P(d+2) (A) will be discussed below but it is essentially a polynomial in the various relevant curvatures. For instance in the case when there is only a gauge coupling (such as A here) then it is a polynomial in the associated curvature, F = dA + A A, which is, moreover, gauge invariant so P (gF g 1 ) = P (F ) (where g is a gauge transformation). Such invariant polynomials can be used to build characteristic classes which are topological invariants of a manifold associated to vector bundles [MS] [LM] [Nak]. Their geometric signicance is discussed in Section 3.2. These are closed forms, dP = 0, (dening elements of cohomology) and hence locally exact (0) so they can be represented, locally, as the dierential of a lower degree form: P (d+2) = dP(d+1) + N . There is then a procedure to generate a chain of descent on this equation determining secondary and other characteristics P(d+1) = dP(d)
(1) (2) (0) (1)

(65) (66) (67)

P(d) = dP(d1) P(d1) = dP(d2) ...


(3)

(2)

The upper indices give an enumeration of this chain (they also have another meaning which will not be (1) discussed here). It is through this chain that a candidate for G[A] can be found. Note that P (d) in (66) is a top-form (on a d-dimensional manifold) whose gauge-variation is a closed form. Assuming that the

36

2.5 The D-brane Anomaly

2 RR-CHARGE AND K-THEORY

theory is dened on a space X without boundary one can dene G[A] =


X

P(d)

(1)

(68)

which implies G[A] =


X (1)

P(d) =

(1)

dP(d1) =

(2)

P(d1) = 0

(2)

(69)

The procedure for calculating P(d) from P(d+2) is referred to as the descent procedure and will be used to calculate the gauge anomalies on D-branes and I-branes starting from the index of the Dirac operator dened on their world-volume theories. 2.5.2 Index of the D-brane Dirac Operator

Consider rst the anomaly associated with the R-symmetry on a single d-dimensional D-brane. The Dirac operator on the brane, iD, can be represented in the basis where (11) is diagonal as iD = iDP+ + iDP = 0 i /+ i / 0 (70)

The above form follows easily if one recalls that iD always acts on a chiral spinor to change its chirality since { , (11) } = 0. In a non-chiral theory iD would be hermitian (see footnote 78), (i /+ ) = i / . The Dirac operator on the D-brane world-volume theory is not however since the latter involves gauge elds that couple chirally to the fermions which means that the operators iDP are not equal and iD is not actually hermitian. However, as mentioned above, the correct procedure to calculate an anomaly for such a theory is to consider a version of this Dirac operator with the gauge elds extended, in a particular way, over d + 2 dimensions and calculate the index of this operator to determine a degree d + 2 product of characteristic classes on which we can perform the descent procedure. To calculate this index requires the Atiyah-Singer index formula [LM][Nak] [AS1], a version of which is given in Theorem C.1. The index is usually calculated with respect to an elliptic complex, the denition of which can be found in [Nak] and Appendix C, but is simply a cyclic (in this case) sequence of maps S+ S S+ where S are the spin-bundles on which the Dirac operator acts and the index is dened as ind iD dim ker i /+ dim ker i / (72) The sequence (71) is schematic, suggesting that the operator maps the positive chirality bundle to the negative one and the index of such an operator over a manifold, X, is then given by the Atiyah-Singer index formula as ind iD =
X i/+ i/

(71)

(ch(S + ) ch(S ))

T d(T X C ) e(T X)

(73)

The integrand in the above expression is a sum of invariant polynomials, the characteristic classes, of the form described in Section 2.5.1. The d + 2 degree part of the integrand in expression above will provide the Pd+2 (A) used as the starting point in the descent procedure. The characteristic classes that appear above are the Chern character, ch(), the Todd class, T d(), and the Euler class, e(). They all have topological signicance and are cohomology classes associated to vector bundles on a manifold. They are given above with respect to particular bundles which means they are polynomials in some curvature form or cohomology class associated with that bundle. Furthermore, the Chern character and the Todd class are essential in dening the relationship between K-theory and ordinary cohomology. The former denes a homomorphism between the K-groups and the cohomology ring while the latter emerges in dening the relationship between the K-theoretic and the cohomological Thom class. All of these notions will be discussed in greater detail later. These topological invariants will 37

2.5 The D-brane Anomaly

2 RR-CHARGE AND K-THEORY

occur frequently through-out this thesis and will be essential in understanding the topological character of D-branes. The explicit form of the Chern character in terms of a two-form curvature will be given later and some geometric intuition for the Chern character (and the related Chern class) will be developed. The reader unfamiliar with these objects, however, is urged to consult the literature [BT], [Kar], [MS], [LM],[Nak], or the appendices since we will often use these classes without a detailed exposition. In the specic case of the D-brane worldvolume theory in type IIB string theory the complex is given by (47) iD : [S + (T ) S + (N )] [S (T ) S (N )] V adj

[S (T ) S + (N )] [S + (T ) S (N )] V adj

(74)

If one considers what this map actually does one nds that iD maps S + (T X) to S (T X) even if the physical theory does not contain fermions in the latter bundle. It is clear from the derivation in Appendix C that the relationship between the anomaly and Dirac index zero modes is simply a consequence of calculating a formal Jacobian. At one stage this requires expanding in a complete basis of eigenstates of the Dirac operator and hence should include modes of both chirality regardless of whether certain modes are projected out of the physical theory. The index of (74) is given by ind iD =

ch([S + (T ) S + (N )] [S (T ) S (N )] V adj )
+ + adj

ch([S (T ) S (N )] [S (T ) S (N )] V

T d(T C) ) e(T )

(75)

Several identities that are easy calculations simplify this expression. First ch(W W ) = ch(W)+ch(W ) and ch(W W ) = ch(W) ch(W ) so the expression above factors to ind iD =

[ch(S + (T )) ch(S (T ))] [ch(S + (N )] ch(S (N ))] ch(V adj )

T d(T C ) e(T )

(76)

The wedge product of forms is somewhat unnecessary as the relevant forms are all even degree and hence commute. Furthermore, part of this expression reduces to the Dirac or roof genus A(T ) given by T d(T ) A(T ) = [ch(S + (T )) ch(S (T ))] e(T ) and also (since N is orientable and assumed, for now, to be a spin-bundle [CY]) ch(S + (N )) ch(S (N )) = so nally (75) simplies to ch(V adj ) A(T ) e(N ) A(N ) e(N ) A(N ) (78)
C

(77)

ind iD =

(79)

All of the relations above are a simple consequence of the denitions of the characteristic classes and can be derived using the explicit form of these classes given in [Nak] in terms of curvature 2-forms. Recall from Section 2.4.3 that when the B-eld is xed at zero on the D-brane then the ChanPaton bundle in the adjoint, V adj , can be factored into the tensor product of two bundles bundles in the fundamental and the conjugate of U (N ), V adj W W. Thus the term in (79) associated with the gauge = coupling, ch(V adj ), decomposes into two terms, ch(V adj ) = ch(W) ch(W) and (79) can be re-expressed ind iD =

e(N )

ch(W)

A(T ) ) A(N

38

ch(W)

A(T ) ) A(N

e(N ) Y Y

(80)

2.5 The D-brane Anomaly

2 RR-CHARGE AND K-THEORY

Where for convenience the following notation has been introduced A(T ) A(N ) A(T ) A(N )

Y = ch(W)

Y = ch(W)

(81)

For complex vector bundles over a manifolds with a connection there is a particularly convenient representation of the Chern character in terms of the local curvature 2-form of the associated vector bundle, namely ch(W) = T r(exp( iF )). Here F = dA+AA is the local form of the curvature associated with the 2 connection A (locally) on W and the trace is over the Lie-algebra indices (recall A is a Lie-algebra valued one-form and F is a Lie-algebra valued two-form [Nak]). Using this formula it is easy to see how ch(W) a a and ch(W) are related. The conjugate representation is given by the right action of g = (e T ) where a is a real-parameter and T a is an anti-hermitian generator of the associated unitary group U (N ). It is not hard to see that, as a consequence, the curvature of W is given by F = (F a T a ) = F a T a = F where F is the curvature associated with W. Hence ch(W) = T r(exp( F )) which can be expanded in a 2 series of even-degree forms. Denoting the kth term in this series ch2k (W) we nd that ch2k (W) = (1)k ch2k (W). The square root given in (81) is the formal square root of a series, in a curvature two-form, which denes the roof genus A; the curvature, in this case, is the Riemannian curvature of T or N . Likewise the denominator is dened formally as the inverse of the series. The series giving A has only four-form terms which means (k+2j+2l) k that, since (1) = (1) for all j, l Z, the sign in front of (1)k ch2k (W)A4j (T )A4l (N ) = (1)(k+2j+2l) ch2k (W)A4j (T )A4l (N ) (82) can be written to depend only on the total degree 2k + 4j + 4l (rather than on the degree of ch 2k (W) only) so Y and Y can be compared degree by degree Y(2k+4j+4l) = (1)k Y(2k+4j+4l) = (1)(2k+4j+4l)/2 Y(2k+4j+4l) so Y(j) = (1)j/2 Y(j) (84) (83)

In (84) an overall relative sign between Y and Y has been introduced and left undetermined as it will depend on the choice of orientation [CY] (see equation (122)) . That is to say that the sign of the anomaly depends on the choice of orientation of the brane and, as such, is a convention. The proper choice of convention will emerge from the requirement that the anomaly be canceled and will be xed in the next section. Eqn. (80) gives the index of the Dirac operator associated to the manifold or equivalently the Abelian anomaly associated with this theory but, as discussed in Section 2.5.1, what is actually needed in order to calculate the non-Abelian gauge anomaly is the secondary characteristic of the integrand of (80) which is an invariant polynomial. Recall that what is needed is the d + 2-form part of this expression, (1) P(d+2) , from which we can determine a d-form, P(d) , whose integral calculates the non-Abelian anomaly. In this case, we can begin with the d + 2-form part of e(N ) Y Y and calculate its descendent, (e(N ) Y Y)(1) , where the subscripts indicating the degree of the form have been omitted. The gauge anomaly resulting from chiral R-symmetry coupling on a single even-dimensional brane, i , is then given by (e(N i ) Yi Yi )(1) (85)

where Yi and Yi are dened above (i is an index indicating the particular brane in question) and the ambiguity regarding choice of orientation has been absorbed into the denition of Y in (84). It may not be clear, at this stage, why the expression for the anomaly has been factored into the two components Y i and Yi in (85) or in (80). The reason has to do with cancellation of the anomaly which will be addressed in the next section. Moreover, the need for this factorization is most obvious in the case of the I-brane anomaly associated with intersecting branes (that will be treated presently) rather than the R-symmetry 39

2.5 The D-brane Anomaly

2 RR-CHARGE AND K-THEORY

anomaly for a single brane. For now it suces to note that the anomaly can be written in the form (85). This expression is not actually the most convenient form to deal with in order to cancel the anomaly and so it will be modied one more time before the end of this section. In order to understand why this is necessary however it is useful to rst consider the second type of anomaly that must be canceled, namely the anomaly on intersecting branes. 2.5.3 Index of the I-brane Dirac Operator

To calculate the anomaly on an intersection recall from Section 2.4 that the appropriate spin complex is the spinor bundle lifted from (57) tensored with a vector bundle, which we will denote V, which carries a representation of the Chan-Paton gauge groups associated to both D-branes ( 1 and 2 ). This is because the strings which generate chiral fermions on the I-brane world-volume are those with one end on either D-brane so they are charged under both gauge groups. Unlike the case of a single brane this vector bundle is not in the adjoint of U (N ). Rather a CP-matrix associated with an open string stretched between two world-volumes, 1 and 2 , supporting N and M branes, respectively, is an N M matrix, . From the considerations discussed in Section 2.4.3 this matrix has a U (N ) U (M ) symmetry which translates into the following transition functions on intersections of an open cover, U U U , = g h (86)

with (g , h ) U (N )U (M ) (consider Figure 1 with the end-points of the incoming strings on dierent world-volumes, not just dierent branes). The two factors, g and h , are the same as those that occur in (34) but from two dierent brane world-volumes (recall, e.g., that g is not associated with a given brane but with N branes wrapping 1 ). If B = 0 so that there are well-dened Chan-Paton bundles, W1 and W2 , in the fundamental represtation on the world-volumes 1 and 2 , respectively, then V W1 W 2 W 1 W 2 = (87) The direct sum above reects the fact that there are two types of strings with dierent ends on dierent branes. The anomaly associated with the fermions on the I-brane can be calculated using the same methods as above (for the R-symmetry anomaly) by starting with the d + 2-form (here d is the dimension of the I-brane) of the integrand of the abelian anomaly associated with the I-brane world-volume theory. Note that the analysis of Section 2.4.6 implies that the I-brane fermions are chiral with respect to the I-brane grading, (p+2) , and the ten-dimensional grading, (11) , so long as N 1 N 2 = 0. If N 1 N 2 = 0 then one checks that the GSO projection, combined with the analysis of Section 2.4.6, implies that the fermions on the I-brane are chiral with respect to the grading of the full bundle T 1 T 2 N 1 N 2 (88)

so they are not chiral with respect to the I-brane grading matrix. However, as with the R-symmetry anomaly on individual branes, the two I-brane chiralities couple dierently to the chiralities 30 of N 1 N 2 and, as such, generate an anomaly if the latter is not trivial. Consider the case N 1 N 2 = 0 so that the spin complex is (assuming the GSO projection is onto positive chiralities, for deniteness) iD : [S + (T 1 T 2 ) S + (N 1 N 2 )] [S (T 1 T 2 ) S (N 1 N 2 )] V

[S (T 1 T 2 ) S + (N 1 N 2 )] [S + (T 1 T 2 ) S (N 1 N 2 )] V

(89)

Recall that iD has only I-brane indexed (i.e. T 1 T 2 ) gamma-matrices so it will only change the chirality of the spinor bundle lifted from the I-brane tangent bundle. To unburden the notation somewhat
30 N 1

N 2 is even dimensional when T 1 T 2 is because of the ND=4 condition for supersymmetry on the inter-

sections.

40

2.5 The D-brane Anomaly

2 RR-CHARGE AND K-THEORY

let T = T 1 T 2 and N = N 1 N 2 . The analytic index of the complex, given by the Atiyah-Singer index formula applied to the complex (89), is ch([S + (T ) S + (N )] [S (T ) S (N )] V)
+ +

ind iD =
1 2

T d(T C ) ch([S (T ) S (N )] [S (T ) S (N )] V) e(T ) This has the same form as (75) and can be simplied in a similar way [ch(S + (T )) ch(S (T ))]
A(T )

(90)

indiD =
1 2

T d(T C ) e(T )

[ch(S + (N )) ch(S (N ))] ch(V)


e(N ) A(N )

(91)

which gives indiD =


1 2

e(N 1 N 2 )

Note that although we derived this for N 1 N 2 = 0 it also applies when N 1 N 2 = 0 since then the I-brane fermions are fully chiral (not just chirally coupled) and the anomaly can be calculated by just neglecting the S (N ) terms in eqn. (91); thus, although these are very dierent types of anomalies they can be canceled the same way. We now expand ch(V) using (87) A(T 1 T 2 ) [ch(W1 ) ch(W2 ) + ch(W1 ) ch(W2 )] A(N 1 N 2 ) A(T 1 T 2 ) A(N 1 N 2 )
2

A(T 1 T 2 ) ch(V) A(N 1 N 2 )

(92)

indiD =
1 2

e(N 1 N 2 )

(93)

Using the Whitney sum rule for the roof-genus [Nak], A(E F ) = A(E) A(F ), it is easy to show that31 A(T 1 )A(T 2 ) = A(N 1 )A(N 2 ) so (93) can be re-written (94)

indiD =
1 2

e(N 1 N 2 )

A(T 1 )A(T 2 ) ch(W1 ) ch(W2 )+ A(N 1 )A(N 2 ) A(T 1 )A(T 2 ) ch(W1 ) ch(W2 ) A(N 1 )A(N 2 )

(95)

Recalling the denitions of Yi and Yj it follows that (95) can be written in a semi-factored form indiD =
1 2

e(N 1 N 2 ) Y1 Y2 + Y1 Y2

(96)

As with the R-symmetry anomaly the index of the Dirac operator provides a means of calculating the gauge anomaly via the descent procedure so the anomaly is actually given by the secondary characteristic of the integrand of (96)
(1) 1 2

e(N 1 N 2 ) Y1 Y2 + Y1 Y2

(97)

31 Write A(T ) = A(T T T N ) = A(T T ) A(T N ) and likewise for N and the result i i j i j i j i j i will follow.

41

2.5 The D-brane Anomaly

2 RR-CHARGE AND K-THEORY

2.5.4

Anomaly Factorization and the Euler Class

It is perhaps now more evident why the single D-brane anomaly associated to chiral R-symmetry couplings was written in the factored form (85). In the introduction to this section it was noted that the anomaly associated with an intersection must be canceled by contributions from both branes separately and cannot depend on the intersection itself. This observation was made in [CY] and much of the exposition in this and the next section is based on it. In the next section it will be shown that by modifying the RRpotential, C, coupling to the D-brane to include a coupling to the anomaly terms, C Yi , it will be possible to cancel the anomalies due both to R-symmetry on a given brane and to chiral fermions on the I-branes. For this to work however the anomalies need to be in a factored form. That is, the associated invariant polynomials must be factorizable into separate factors, each of which depends only on bundles associated to one particular brane. At this point eqn. (85) and (97) are not properly factored in this way as the Euler class in each case cannot be separated into two contributions (in the case of (85) where it is the Euler class of the normal bundle of a single brane this may seem like an unnecessary constraint but since the anomaly cancellation mechanism for a single brane and for brane intersections will be the same it is important that (85) can also be factored into the same factors that occur in (97); this will be made more clear in the next section). The solution to this problem is pointed out in [CY] and has to do with a certain topological properties of the Euler class. It also provides more insight as to the meaning of the anomaly term for the case of a single D-brane. This will be discussed once the actual mechanism for anomaly cancellation has been introduced but in the meantime an important modication to (85) and (97) will be proposed. Consider the integrand of (85) (e(N i ) Yi Yi )(1) (98)

and note that it is composed of a product of invariant polynomials. We will show that this actually equivalent to e(N i ) (Yi Yi )(1) (99)

and similar reasoning can be applied to the two terms in the integrand of (97). This reformulation will be relevant in the next section as it is possible to generate a term such as (99) by modifying the coupling to the D-brane to be C Yi . The factor Y will be seen to follow from the bulk equations of motion and the Euler factor will be related to the topology of the brane (but will not occur as an explicit coupling term). To see how (99) is equivalent to (98) one starts by noting that Y Y (dropping subscripts for convenience) is also an invariant polynomial which we dene as Z = Y Y. Recalling, from Section 2.5.1, that for such polynomials there is a local expression as an exact form, Y = dY (0) , shows that and for the purposes of the descent procedure Z can replace Y Y with its secondary characteristic Z (1) given by Z (0) = (Y (0) dY (0) ) = Y (0) dY (0) + Y (0) d Y (0) = dY (1) dY (0) + Y (0) d2 Y (1) = dY (1) dY (0) = d(Y (1) dY (0) ) = dZ (1) Z = Y Y = dY (0) dY (0) = d(Y (0) dY (0) ) = dZ (0) (100)

(101)

where use has been made of eqns (65) and (66) and the commutativity of the and d operators. Note also that is dened to be a derivative and hence obey a Leibnitz identity [HT][Nak]. Hence (98) can be written (with e = e(N )) as (e Z)(1) This can be calculated using the descent procedure once again e Z = de(0) dZ (0) = d(de(0) Z (0) ) = d(e Z)(0) (103) (102)

42

2.6 Canceling the D-brane Anomalies: Anomaly Inow

2 RR-CHARGE AND K-THEORY

where (e Z)(0) is just a label for the rst descendent of e Z so to calculate the second we take its gauge variation (e Z)(0) = (de(0) Z (0) ) = de(0) Z (0) + de(0) Z (0) = d2 e(1) Z (0) + de(0) dZ (1) = de(0) dZ (1) = d(de(0) Z (1) ) = d(e Z)(1)

(104)

Hence (e Z)(1) is given by de(0) Z (1) = e Z (1) showing that (98) is indeed equivalent to (99). This simply reects a general ambiguity in dening descendents which we have exploited in order to keep the Euler class unaltered in the nal expression. The same argument shows that the integrand of (97) can be rewritten as e(N 1 N 2 ) Y1 Y2 + Y1 Y2
(1)

(105)

In the next section it is shown how this form (derived in [CY]) allows the anomaly to be canceled by modifying the bulk coupling of D-branes to the RR-form potentials.

2.6

Canceling the D-brane Anomalies: Anomaly Inow

In this section it will be shown how a modication of the naive D-brane coupling term in the bulk theory i C = (C)

(106)

can be used to cancel the anomalies on the D-brane given by integrating (105) and (99) over the D-brane and I-brane worldvolume, respectively. To do so requires considering the classical equations of motion associated with the bulk theory (which is supergravity) and introducing an anomalous gauge variation for the RR-form potential, C. This anomalous variation is seen as a consequence of the equations of motions and Bianchi identity (which are essentially xed by each other since the total RR-form eld strength, G, is self-dual). This anomalous gauge variation will be localized on the D-brane and will cancel the variation from the quantum anomaly on the brane and render the overall theory consistent. The exposition below is essentially inspired by [CY] but diers from it in the use of dierent bulk equations. In [CY] a non-covariant form of the RR kinetic terms is introduced in order to address the self-dual nature of the RR-form eld-strengths whereas here the standard covariant kinetic term will be used and the selfduality will be imposed on the equations of motions rather than the action (as suggested in Section 2.3). The calculations that follow are also considerably more detailed than [CY] and we attempt a rigorous treatment of some of the cohomological subtleties involved. The calculations in this section will involve many subtle signs arising from commuting forms of various degrees and these will be sensitive to the signature of the metric and the degree of the forms. To simplify the exposition the discussion is limited to type IIB theory (in which things are generally more transparent in any case) and a Minkoswki metric. Thus the eld strengths G(i) will have odd degree while the potentials C(i1) will have even degree and the D-brane currents, , will also be of even degree. G will denote the sum of all eld strengths and C the sum of potentials which will be respectively of odd and even degrees. It will be important to recall that the exterior derivative on dierential forms obeys a graded Leibnitz identity d( ) = d + (1)|| d (107)

where || is the degree of . This means that even degree forms can be partially integrated the same way as functions whereas for odd-degree forms an additional minus sign is picked up. It is also important to note that the self-duality of G(i) is given by G(i) = (1)(i1)/2 G(10i) (108)

43

2.6 Canceling the D-brane Anomalies: Anomaly Inow

2 RR-CHARGE AND K-THEORY

2.6.1

IIB with D-brane Sources

Consider the modied D-brane RR-coupling [CY] [MM]


src SRR =

Ni (C) (G) Yi

(0)

(109)

This modies the RR-sector supergravity action, (12), in the presence of D-brane sources much as the naive equation (27) originally did. Note that the bulk forms C and G have been pulled back to the D-brane by the inclusion : i X. Ni is an unspecied integer which will be seen to give the number src of branes wrapping the manifold i . In what follows it will be shown that SRR is not invariant under a src gauge transformation of the D-brane worldvolume gauge elds given by the dierential . In fact, S RR will be shown to be exactly cancel the anomalous gauge variation (of the D-brane world-volume action) that was described schematically as W [A] in the last section (given by (85) and (97)). To see how this works we must rst derive the equations of motion associated with (12) in the presence of sources coupling via (109). To this end (12) will be modied to include all the RR-form eld strengths and the source term while the self-duality constraint will only be applied once the equations of motion have been derived. The full RR-form action is then
f ull SRR =

1 2

G G

Ni (C) (G) Yi

(0)

(110)

where the factor of 1 accounts for the fact that weve formulated (12) in terms of elds and their dual. 2 This can be written more uniformly using D-brane currents discussed in Section 2.3.2
f ull SRR =

1 2

G G +

i Ni C G (Yi )

(0)

(111)

where i is the Poincar dual to i . In (111) a standard cohomological construction has been used to e (0) (0) push Yi forward (even though Yi is not closed and hence not an element of cohomology). For clarity, let us quote the necessary result [BT, Prop. 6.15] Proposition 2.1 (Projection Formula). Let : W by an oriented rank n (real) vector bundle, a form on and a form on W with compact support along the ber. Then (( ) ) = Proposition 2.2. With the same hypothesis as Prop. 2.1 suppose that is oriented and of dimension d+nq d, q (W), and c (). Then with the local product orientation on W cv
W

( ) =

The map is given, with respect to some cover, = U , that trivializes W, so W|U = U Rn , by integrating the components of the form along the bers, Rn . This local map from (W)|U to cv n ()|U can be shown, [BT, 6], to patch to a globally well-dened map. Here (W) is the space of forms of arbitrary degree and cv indicates compact-vertical support; this notation follows [BT]. The change of degree n, where n is the dimension of the ber, is due to the integration. (0) Applying this result to the normal bundle, : N i i and Yi gives32
N

( Yi ) =

(0)

Yi

(0)

(112)

where is a form on N with compact support in the vertical direction. Essentially because has (0) compact support so does the product (Yi ) and hence it can be integrated.
compactness of Yi this is the case.
32 The (0)

as in Proposition 2.2 is required for its integration over to be well-dened. We will assume

44

2.6 Canceling the D-brane Anomalies: Anomaly Inow

2 RR-CHARGE AND K-THEORY

The normal-bundle : N i i can now be identied, via a dieomorphism, with an innitesimal tubular neighborhood33 of , N (). This identication is only possible when is a submanifold in X so : X is an embedding, not just an immersion. In such cases we will often use N and N () interchangeably, as convenience dictates. This dieomorphism is described in [BT, 6] and references mentioned therein. Note that, up until now, it was sucient to work with an an immersion. As mentioned in the beginning of Section 2.4 there is no physical reason to restrict to be an embedding but mathematically it will make things simpler. We will not deal with the more general case in cohomology but in K-theory we will nd there are some results [Kar] which generalize the constructions we need (the Gysin map and its relation to the Thom isomorphism; see Section 3.5.7) to more general . Thus for now we require to be an embedding under the assumption that there are results which can be used to extend our constructions to a more general setting. (0) As N is dieomorphic to a tubular neighborhood of and (Yi ) tends to zero near the boundary of this neighborhood it can be extended to a form on all of X and34 ( Yi ) =
(0)

X (0)

( Yi ) =

(0)

Yi

(0)

(113)

since the support of (Yi ) is contained in N . In this case = (G) where 1 : N X 1 is the inclusion of the tubular neighborhood into X and let 2 : N be the inclusion of in N as the zero section so : X is given by = 1 2 and = . Applying this to (113) gives 2 1 G ( Yi ) = =
(0)

( (G)) Yi 1

(0)

(114)
(0)

( (G)) Yi

( ) (G) Yi

(0)

where Proposition 2.1 ( (G)) = ( ) (G)


(115)

has been used as well as the fact that = ( Throughout the argument it has been assumed that = has compact vertical support. The reason for this is that is a representative of the Poincar dual of which is also the Thom class of its e normal bundle [BT, 6, 12] [Kar], [OS, Sec. 7.2]. The Thom class and the Thom isomorphism, as well as their relationship to Poincar duality, will be discussed below in Section 2.6.3 but one important property e that relates (114) to (110) is that ( ) = 1. This means that the class is a bump function with compact support in the vertical direction that integrates to 1 along the bers of the normal bundle. This is the standard denition of the Thom class. Once again, the reader unfamiliar with Poincar duality e can think of as a -function with support on . 2.6.2 RR Equations of Motion

) 2

= .35 1 (G) 1

Although the equations of motion in the presence of a source (which will be electric with respect to some RR-elds and magnetic with respect to their duals) and Bianchi identities will imply G = dC but rather G = dC + . . . this does eect the derivation of the equations themselves. The equations of motion are
33 This is an open neighborhood in X consisting of all points less than > 0 distance from with respect to some metric on X. 34 The notation here is somewhat imprecise as (Y (0) ) , which is in (N ), is being identied with its image in cv (X) under the map which simply extends it by zero o the tubular neighborhood dened by N (). 35 The argument for = ( ) = 1 comes from the homotopy equivalence 1 and the homotopy invariance 2 2 2 of cohomology. In subsequent sections, however, we will see that G is not always closed and hence not an element of cohomology. In the absence of closed string backgrounds, such as the B-eld and the Dilaton, G is closed in the bulk, away from the brane and we can then try and use these cohomology arguments to pull its class back to a neighborhood of the brane.

45

2.6 Canceling the D-brane Anomalies: Anomaly Inow

2 RR-CHARGE AND K-THEORY

derived by considering a variation of C by C + where is necessarily an even degree form with an innitesimal coecient.
f ull SRR + S =

1 2

(dC + d + . . . ) G +

i Ni (C + ) (dC + d + . . . ) (Yi ) (116)

(0)

The variation is given by (note the exterior derivative commutes with the pullback) 1 2
(0)

S =

dG dG

i Ni + (dYi ) i (Ni + dYi )


(0)

1 = 2

1 2

dG

i (Yi ) (117)

which gives (when required to vanish for all ) equations of motion dG= i (Yi ) (118)

The Bianchi identities for the eld can now be determined by imposing the self-duality relation (108) which shows, for a xed term in the expansion, of degree 11 j, d G(j) = (1)(j1)/2 dG(10j) = so the Bianchi identity becomes, in degree j + 1, dG(j) = (1)(9j)/2
i i

i (Yi )

(119)
(11j)

i (Yi )

(120)
(j+1)

where the notation indicates that only the (j + 1) degree terms in the sum appear on the RHS (recall Y i is a polynomial with terms of varying degree and has degree of codim()). Consider an expansion of Y in terms of various degrees Y = k Y(k) and recall that is always of even degree as is dG(j) so k will always be even. Since i is of degree 10 dim(i ) the relationship between k and j can be determined by 10 dim(i ) + k = j + 1 so j = 9 dim(i ) + k which implies (using that k is even) dG(j) =
i

(1)(dim(i )k)/2 (Yi )(k)


k (j+1)

=
i

i (1)(dim(i )/2) i (Yi )

((1)k/2 Yi )(k)

(121)
(j+1)

(j+1)

In the nal equality above the factor (1)(dim(i )/2) , which is independent of the degree of Y(j) is used to x the sign ambiguity in (84) which also xes a convention for the orientation of the branes (Yi )(j) = (1)(dim(i )/2) (1)j/2 (Yi )(j) (122)

From the modied Bianchi identity (121) it is clear that the relationship G = dC no longer holds. A suitable modication of this consistent with (121) is G = dC i (Yi )
(0)

(123)

46

2.6 Canceling the D-brane Anomalies: Anomaly Inow

2 RR-CHARGE AND K-THEORY

which works because i is closed. However, from this equation it would appear that G is no longer gauge invariant under a gauge transformation of the world-volume theory generated by the operator (0) (1) (described in the last section) since Yi = dYi . As eld-strengths are physical observables they must be gauge invariant. To rectify this it is necessary to modify the RR-form potential C so that it transforms under gauge transformation on the D-brane but in such a way that the variation is localized on the brane. To see how this works consider the consequences of requiring G = 0 0 = G = (dC = dC (0) i (Yi )) = dC
(1)

(0) i ( Yi )) i (Yi ))
(1)

i (dYi )) = dC d

(124)

which, when considered locally (for which cohomology is trivial so closed forms must be exact), gives C = d +
i

(1) i (Yi )

(125)

where d is a closed, exact form that can be absorbed into C by a gauge transformation and is of no consequence. Thus the modied Bianchi identity, arising from the coupling of the RR-form elds to gauge elds on the brane (and, in particular, topological invariants of gauge elds on the branes) require that the bulk RR-form potential, C, transforms non-trivially under a gauge transformation of the D-brane worldvolume elds. This will be referred to as an anomalous gauge transformation though this is not intended to suggest a quantum anomaly. It is also important to note that this non-trivial transformation is localized to the part of C on the D-brane since i is a codim(i )-form with a support localized on i . Although it may seem strange that a bulk eld transforms under a gauge group localized on the D-brane (at least from the point of view of the low-energy action) this might well be expected from string-scale considerations. The RR-form eld strengths are closed strings in the RR-sector. It was already mentioned that D-branes can emit and absorb closed strings (which can be seen as a closed string splitting open on a D-brane or open string endpoints meeting to form a closed string that can leave the D-brane) so it is not surprising that closed strings near the brane or on the brane become charged under the Chan-Paton groups of the brane. This is the sense in which the anomaly cancellation, which will be described presently, can be considered an inow from the bulk. To see exactly how the anomalous gauge transformation rule of the bulk RR-form potentials cancel the quantum anomaly on the D-brane and I-brane world volumes consider the gauge variation of the D-brane to bulk couplings, (111), and note that the latter can also be seen as a term on the D-brane worldvolume theory (the kinetic terms have been omitted as they are, in any case, gauge invariant) 2 i Ni C G (Yi ) i Ni i Ni i Ni j (Yj ) G (dYi )
(1) (1) j (Yj ) dG (Yi ) (1) (1) (0)

SRR =

= 2 = 2

= 2 = 2 = 2
2

(1) j (Yj ) +

(1) j (Yj ) (Yi )

(126)

i,j

(1) (1) i j Ni (Yj ) + (Yj ) (Yi )

2 i,j X

i j

Yi Yj

(1)

47

2.6 Canceling the D-brane Anomalies: Anomaly Inow

2 RR-CHARGE AND K-THEORY

The last equality follows from considering Y Y = N + dY (0) ) (N + dY (0) = N N + N dY (0) + NdY (0) + dY (0) dY (0) = N N + d N Y (0) + N Y (0) + Y (0) dY (0)

(127)

so the rst descendent, (Y Y)(0) , is given by N Y (0) + N Y (0) + Y (0) dY (0) and the gauge variation of this denes the exterior derivative of the second descendent, d(Y Y)(1) (see (65)), as (Y Y)(0) = N Y (0) + NY (0) + Y (0) dY (0) = N dY (1) + N dY (1) + Y (0) dY (0) + Y (0) d Y (0)

= N dY (1) + N dY (1) + dY (1) dY (0) + Y (0) d2 Y (1) = d N Y (1) + N Y (1) + Y (1) dY (0) = d N Y (1) + Y (1) Y d(Y Y)(1)

(128)

which shows that that last two lines of (126) are in fact equal. 2.6.3 The Thom Isomorphism and the Euler Class

Eqn. (126) almost has the same form as (85) and (97) when the modications coming from (99) and (105) are taken into account. There remain normalization factors to be xed but note that the relative sign is xed since the orientation of the world-volumes has been absorbed into the denition of Y (so (85) and (97) will only have a positive coecient and it is clear that (126) will have a negative one). In fact, given the normalization for (85) and (97) it is then possible to calculate the D-brane coupling constant, (or charge in the sense of fundamental charge like e in electromagnetism), by requiring the normalizations to match up. Before discussing this it remains to show that the integrals in (126) and those from (85) and (97) actually have exactly the same form. To do so both sets of integrals must rst be expressed as integrals over the same manifold. This will involve the introduction of some cohomological machinery that has only been briey mentioned until now. In Section 2.3.2 the Poincar dual, , of a d-dimensional submanifold : X was introduced and e related to integrating an element of H d (X) over via
X

(129)

At the end of Section 2.6.1 the Projection formula, Proposition 2.1, was used with (G) by claiming 1 that could be seen as an element of q (N ), the q-forms of the normal bundle with compact support cv along the bers. The relationship between the Poincar dual and the normal bundle cohomology arises e from the Thom isomorphism which we discuss presently. Let : W be a rank n vector-bundle and : W the dieomorphic embedding of into the zero section. Then [BT, 6, 12], [Kar], [OS, Sec. 7.2] Theorem 2.3 (Thom Isomorphism). For any p d there is an isomorphism
p+n ! : Hcv (W) H p ()

(130) (131)

! : H ()

p+k Hcv (W)

Here ! is just the restriction to cohomology of the map that was previously described as integration over the bers in the vertical direction. There is a particular cohomology class, [W], the Thom Class which is uniquely characterized by the fact that it restricts, on each ber of W, to the generator of the ber cohomology [BT, Prop. 6.18]. This means that it integrates to one along the vertical direction ! ([W]) = ([W]) = 1 48 (132)

2.6 Canceling the D-brane Anomalies: Anomaly Inow

2 RR-CHARGE AND K-THEORY

The inverse of ! , ! , is given by the pull-back by following by multiplication by [W]. This follows easily from Proposition 2.1; let H p () ! () [W] ! ( [W]) = ([W]) = 1 (133) (134)

From this it is easy to see that [W] = ! (1). To understand the relationship between the Thom isomorphism and Poincar duality, recall the defe initions of 2 : N , 1 : N X and : X from Section 2.6.1. Let : N be the projection where N is seen as the normal bundle (dieomorphic to a tubular neighborhood N ()). Then there are some useful propositions [BT, Prop. 6.24-6.25] Proposition 2.4. The Poincar dual, , of a closed, oriented submanifold in an oriented manifold e X and the Thom class of the normal bundle, [N ], of can be represented by the same form. Proposition 2.5 (Localization Principle). The support of the Poincar dual of a submanifold can e be shrunk to any given tubular neighborhood of .
p+n By construction elements of Hcv (N ) have compact support in the vertical direction (meaning that within each ber of N the form is compactly supported) so they can be extended by 0 to elements of H p+n (X). Recall that the cohomology class [N ] was described as a bump function integrating to 1 in the vertical direction. To see, roughly, why Proposition 2.4 holds consider H d (X) so H d (),

=
N

[N ] =

[N ]

(135)

where the rst equality follows from Proposition 2.2 and the fact that ([N ]) = 1 and the second follows because the support of [N ] is, by denition, restricted to N X. A comparison between eqn. (135) and eqn. (26) should motivate Proposition 2.4. Proposition 2.5 follows because the normal bundle is dieomorphic to an arbitrary tubular neighborhood of . Two results related to the Thom class that we will need are given below [BT, Prop. 6.41, 6.19] Theorem 2.6. The pull-back of the Thom class to by the zero section, 2 : N , is the Euler class e(N ) = ((2 )! (1)) 2 Since the Thom class of the normal bundle is also the Poincar dual this implies that e e(N ) = ( ) (136)

Theorem 2.7. If V and W are two oriented vector bundles over a manifold , and 1 and 2 are the projections V 1 V W 2 W

then the Thom class of V W is [V W] = 1 [V] 2 [W].

In the discussion above, and also that in Section 2.6.1, it is important to consider when certain identities apply only at the level of cohomology (such as Poincar duality and the Thom isomorphism) e or when they hold more generally (such as the Projection Formula, prop. 2.1). In particular, in this discussion , Y, and Y are closed but the secondary characteristics such as Y (0) are not. All these results can now be applied to represent (126) as an integral over . First however it is necessary to note that even though = ( 2 ) is not equal to the identity (on forms in general) it 2 is an isomorphism on the level of cohomology because 2 splits (i.e. 2 is homotopic to the identity map [BT]). On cohomology with compact vertical support (in the ber direction) the proper isomorphism is given by ! (2 )! = id. Now consider the image of e(N ) = ( ) under this isomorphism 2 e(N ) = ! (2 )! e(N ) = ! ( ) = ! 2 49 (137)

2.6 Canceling the D-brane Anomalies: Anomaly Inow

2 RR-CHARGE AND K-THEORY

The last equality follows as an equality in cohomology with compact vertical support because the total class already has compact support due to the wedge product with (i.e. even though ( ) itself 2 may not have compact support). Now consider the integral in (126) schematically for the case when i = j (the integral has been shifted from X to N since all the terms in the integrand have support only in N ) and recall that ! is given by integration along the ber directions i i Yi Yi
(1)

=
i

N i

[e(N i )] Yi Yi

(1)

(138)

where the dependence on the Euler class has been made explicit. Now (138) is of the same form as (99) but (bearing in mind that (138) has a prefactor 2 /2) of the opposite sign and so cancels the anomalies arising on a single brane (when the value of is set to match any normalization terms of (99) which have generally been neglected in this exposition). To show that a similar form holds for i = j requires using Theorem 2.7. In this case the relevant bundles will be T i and N i restricted to the I-brane world-volume i j . Then we calculate [CY] (recall the intersections are assumed to be perpendicular) i j = [N i ] [N j ] = [N i T j N i N j ] [T i N j N i N j ] = [N i T j T i N j N i N j ] [N i N j ] = i j [N i N j ] = i j ! (1)

(139)

In the calculation above all bundles and cohomology classes have been pulled back to i j via the inclusions into i or j . The second-to-last equality follows because the normal bundle to i j is the complement of T i T j in T X which is just N i T j T i N j N i N j and the last equality follows by dening : i j N i N j (140)

as the zero section of the bundle : N i N j i j and from the remarks in the last few paragraphs on the Thom class. The same reasoning as (138) now shows for i = j Yi Yj
(1)

N (i j )

i j
i j

+ Yj Yi

(1)

=
(1)

(141)

[e(N i N j )] Yi Yj + Yj Yi

which is of the exact same form as the I-brane anomaly and hence cancels it (once normalization factors have been matched). A few remarks are in order before proceeding. The rst regards the exact method of anomaly cancellation above. Throughout this discussion there have been two low-energy theories: the classical supergravity in the bulk (given in Section 2.2) and the Super Yang-Mills theories living on the D-branes, which is anomalous. Even though, in higher dimension, neither theory can be consistently quantized we can study the anomalous gauge variation of the quantum eective action, W [A], associated to the Super YangMills theory. In a standard perturbative calculation the anomaly would occur at the one-loop level and so can be calculated even though the full theory cannot be quantized (without going to a string theory). Hence, in order to see how the gauge anomaly cancels, one can consider the supergravity RR-action with src only the eld-strength kinetic terms, (12), and add to it the brane-bulk coupling, SRR , given in (109) as well as the quantum eective action of the brane SYM theory, denoted schematically as W [A] above to dene a total Lagrangian for the full theory with D-brane sources (enumerated by i)
src Stotal = SN S + SCS + SR + SRR + i D-brane terms

Wi [A]

(142)

50

2.7 Normal Bundles with Spinc Structures

2 RR-CHARGE AND K-THEORY

Under a gauge variation of the above, given by the action of the operator , all terms but the last two src are invariant. The variation of SRR is given by (138) and (141) while the variation of the Wi [A] is given by (99) and (105). These two come with opposite relative signs and cancel each other once the src D-brane charge, , that appears in SRR is tuned appropriately. The tuned value of is shown, in [CY], to correspond to the value for D-brane charge originally calculated in [Pol3].

2.7

Normal Bundles with Spinc Structures

In the next section we will study the consequences of the modied D-brane coupling (109) but before doing so it is necessary to address an issue raised at the end of Section 2.4, namely the spin-structure of the D-brane. Throughout the exposition we have made the tacit assumption that the D-brane admits spinors transforming under a spinorial representation of both its tangent and its normal bundle. This is not true for a general manifold, , and a general embedding : X even if X itself admits a spin-structure (which we will assume is the case). In the literature this was rst discussed in [FW] where it was shown that a single D-brane can only wrap a submanifold, , if the latter admits a Spin c -structure (again only when B is topologically trivial). This was later incorporated in [MM] to justify a necessary factor in the anomaly polynomials. Specically, we will argue in this section that it is necessary to modify eqn. (81) as follows A(T ) A(N ) A(T ) A(N )

Y = ch(W) ed/2

Y = ch(W) ed/2

(143)

Where d = c1 (Z) is a Chern class with odd integral value corresponding to a particular (choice of) line bundle, Z, associated to the normal bundle of the D-brane world-volume, , embedded in X. The discussion in [FW] was extended to multiple branes in [Kap] and will be reviewed in Section 4 when non-trivial B-eld backgrounds are incorporated. Although the issue of Spinc structures has been discussed frequently in the literature it does not seem to have ever been treated carefully in the context of I-brane anomaly cancellation36 so we will attempt to do so here. Specically we will use the result of [FW] to show that I-brane world-volumes must also admit Spinc structures and that these structure are related to that of the branes they are part of in such a way that (143) results in the proper anomaly cancellation. The discussion in this section will be rather technical. It is also somewhat outside the main stream of the exposition so a reader lacking the appropriate background should feel free to skip this section and simply accept that eqn. (143) is a necessary modication of (81) when some topological criteria are not satised (when the worldvolumes, , are not Spin-manifolds). For this reason we will assume, in the exposition below, that the reader is already familiar with the notion of Spin- and Spinc -structures and some other cohomological constructions. A reader wishing to learn about these may consult Appendix A and [LM] for some background on Spin- and Spinc -lifts of SO(n) bundles and [BT] for an introduction to some of the cohomological arguments that occur below. 2.7.1 Fermions and Spin(c )-structures

The notion of a Spin- or a Spinc -lift is relevant to the physics of fermions for the following reasons. Locally, in any eld theory, fermions are dened to be elds that transform under a spinorial representation of the Lorentz group. More precisely, let be a Lorentzian manifold on which the theory is dened and let T be its p + 1 dimensional tangent bundle which has structure group SO(1, p) (recall that it is always possible to work in a frame such that the local basis vectors are orthonormal with respect to the metric; see [Nak] for details). Under a local Lorentz transformation fermions must, by denition, transform under a spinorial representation so that a rotation by 2 about a xed axis is not equivalent to the identity but rather to 1. Mathematically what this means is that fermions must locally be vectors in
36 In Section 3 we will see that the direct use of K-theory to classify D-branes incorporates the results of this section quite naturally.

51

2.7 Normal Bundles with Spinc Structures

2 RR-CHARGE AND K-THEORY

a representation, S, of Spin(1, p) which is compatible with the action of SO(1, p) on the tangent bundle. By virtue of the sequence 0 Z2 Spin(1, p) SO(1, p) 0

(144)

any representation of SO(1, p) is also a representation of its double cover, Spin(1, p). The compatibility alluded to above implies that a local Lorentz rotation by an element g SO(1, p) must lift to a rotation of the spinors by g Spin(1, p) for some g such that (g) = g. Since is a double cover there is a freedom to lift either to g or g. As a consequence not every principle SO(1, p)-bundle, P SO(1,p) , lifts to (denes) a principle Spin(1, p)-bundle, PSpin(1,n) . There is a topological obstruction generated by the need to lift consistently on dierent elements of a cover in order to satisfy the cocycle condition. For the vector bundle, T , associated with a principle SO(1, p)-bundle via the fundamental representation this obstruction is measured by the second Stiefel-Whitney class of T , w2 (T ), which is discussed in depth in Appendix A.1. Thus to determine whether it is possible to have fermions on a manifold, , it is necessary to determine whether the bundle T is a Spin-bundle in the sense dened above. If this is not the case then there is no bundle of which the spinors can be a section and so the manifold will not admit fermions (even if local considerations, such as a spectrum calculation, suggest that they exist). As discussed in previous sections, in the case of D-branes, we will often also need to know whether it is possible to dene spinors transforming under the local Lorentz symmetry associated with the normal bundle; that is, whether N admits a spin-lift. Although it is true that for S(T ) and S(N ) to be well-dened bundles requires that w2 (T ) = 0 or w2 (N ) = 0, it turns out that this is not so relevant for the physics of fermions on a single D-brane. The reason is that it has been shown, in Section 2.4.4, that the spinors on are sections of S(T ) S(N ) which is always a well-dened bundle even if the individual bundles making it up are not well dened.37 This follows since w2 (T X) = 0 by assumption and T X = T N so S(T X) S(T ) S(N ) is a well-dened bundle. None-the-less, it will be useful for us to continue and = address the issue of whether it is possible to dene some kind of global spinor related to any given real vector bundle, such as T or N , on a general (sub)manifold, . Even if T or N do not admit Spin-lifts it is still possible to dene spinors associated to these bundles if they are Spinc . In this case the spinors must be charged under some gauge group (which is U (1) in the simplest case) but with half-integer charges (in the appropriate units) in such a way that the line-bundle associated with the gauge group is an ill-dened square root, L1/2 , of an actual line-bundle, L, as described in Appendix A.2. There it is shown that such a bundle, with the appropriate properties, exists if and only if the third Stiefel-Whitney class, w3 (T ), vanishes. In this case the fermions will be sections of the Spinc -bundle S(T ) L1/2 , which is well-dened. Locally, in a trivialization U = {U }, the sections will look like functions : U S C which is exactly what charged fermions are expected to look like locally. The use of a Spinc -bundle rather than a Spin-bundle does not conict with local constraints such as spectrum calculations of open-strings ending on a D-brane. Most of the constructs and theorems that are in place for Spin-bundles also work for Spinc -bundles so dening the fermion eld, , as a section of S(T ) L1/2 will not introduce signicant complications. Thus if either w2 (T ) = 0 or w3 (T ) = 0 it is possible to dene a bundle with Spin(1, p) structure group lifting T (and likewise for the normal bundle). It is not obvious that these topological constraints are satised for any D-brane worldvolume, , but Freed and Witten [FW] have shown that for any on which open-string can end (i.e. a D-brane) the tangent and normal bundle must satisfy w3 (T ) = 0 and w3 (N ) = 0 so long as the pull-back of the eld-strength, H = dB, onto is topologically trivial. That is, so long as ([H]) = 0 where : X. More generally [FW] showed that38 w3 (T ) = w3 (N ) = ([H]) (145) This condition was derived in [FW] by considering a global (non-perturbative) anomaly, known as the Freed-Witten anomaly (see footnote 27), in the string worldsheet theory. Equation (145) is to be under stood as a relation between Z or Z2 valued Cech cohomology classes.
37 In Appendix A.2 we explain how two putative bundles may not be well dened while their tensor product is a well-dened bundle. 38 In the case where more than one D-brane wraps the same submanifold this condition is modied [Kap] as will be discussed in Section 4.

52

2.7 Normal Bundles with Spinc Structures

2 RR-CHARGE AND K-THEORY

The rigorous derivation of this condition is quite involved mathematically and will not be presented here. As it is essential for understanding how the B-eld modies the geometric character of the worldvolume elds (and the bundles they are sections of) we will discuss it in some detail in Section 4. Even there, however, we will restrict our attention mostly to those aspects of the argument that are relevant to the exposition. Readers interested in the full derivation of (145) are referred to the original paper [FW]. For now the reader may either simply accept (145) or may turn to Section 4 to learn something more of its origin. 2.7.2 I-branes and Spinc -structures

Above we have argued that the existence of fermions on a general manifold, , associated to some real bundle (usually T ) requires that this bundle admits either a Spin- or a Spinc -lift. In the case of Dbranes, cancellation of the Freed-Witten anomaly implies that this condition is always fullled for the tangent and normal bundle of a world-volume supporting a single D-brane and a topologically trivial B-eld (though we should emphasize again that, for a single D-brane, this is irrelevant since the fermions are sections of S(T ) S(N ) which is always well-dened). In Section 2.5, however, anomalies were also considered deriving from spinors on the world-volume theory of an I-brane, the intersection of two D-branes. In this case a Spin- or Spinc -structure is necessary for the existence of fermions in the lowenergy action since, unlike the D-brane world-volume theory, the spinors of the I-brane derive from (are a lift of) a bundle that is not equivalent to the spacetime tangent bundle, T X (see the end of Section 2.4.7). In this section we will show that the relevant bundle on the I-brane (see (57)), : (T i T j ) (N i N j ) i j (146)

is a Spin -bundle under some mild hypothesis which will be outlined below. Recall that spacetime, X, is assumed to be an oriented spin-manifold so that w1 (T X) = w2 (T X) = 0 (we have not discussed this but the rst Stiefel-Whitney class is the obstruction to orientability so necessarily vanishes for orientable bundles [LM] [Nak]). Moreover, the D-brane world-volumes are assumed to be orientable (because they are sources of RR-ux and string worldsheets are orientable in the IIB theory) so w1 (T 1 ) = w1 (T 2 ) = 0. Although the Stiefel-Whitney classes, as they have been introduced so far, seem to be independent they can, in fact, be dened as a single characteristic class (the total Stiefel-Whitney class) associated to a real vector bundles : W X so that [Nak], [LM] where wk (W) H k (X, Z2 ). This class is multiplicative so W = W1 W2 implies that w(W) = w(W1 )w(W2 ). Thus, in degree k
k

w(W) 1 + w1 (W) + w2 (W) + w3 (W) + . . .

(147)

wk (W) =

wi (W1 )wki (W2 )


i=0

(148)

Note that, on 1 2 , T 1 can be decomposed into T1 T12 where T12 T (1 2 ) = T 1 T 2 and T1 T12 is its orthogonal complement in T 1 . Likewise T 2 = T2 T12 so it is possible to decompose T X into T X = T1 T2 T12 N12 where N12 N 1 N 2 . Although we are being somewhat careless notationally we intend that all these bundles should be pulled back to 1 2 . Recall that the bundle intersections shown above (and also used below) are well-dened because we take the D-branes to intersect at right angles (relative to the metric) and so their tangent bundles are spanned by a choice of veilbeins on an orthonormal frame for the spacetime tangent bundle. This constitutes one of our mild hypothesis. Let us now state a lemma we will need. Lemma 2.8. Let 1 and 2 be orientable submanifolds of an oriented, spin-manifold X (as above) and let 1 2 be their oriented perpendicular intersection (see above). The second Stiefel-Whitney class of the restricted bundle : (T 1 T 2 ) (N 1 N 2 ) 1 2 53 (149)

2.7 Normal Bundles with Spinc Structures

2 RR-CHARGE AND K-THEORY

satises w2 (T 1 T 2 N 1 N 2 ) = w2 (T 1 ) w2 (T 2 ) where all bundles are implicitly pulled back to 1 2 . Proof. The multiplicative nature of the Stiefel-Whitney class implies (in the notation introduced above and with all bundles pulled back to 1 2 ) that w(T X) = w(T1 T2 T12 N12 ) = w(T1 ) w(T2 ) w(T12 ) w(N12 ) In degree one the fact that T X is orientable implies 0 = w1 (T X) = w1 (T1 ) + w1 (T12 ) +w1 (T2 ) + w1 (N12 )
w1 (T 1 )=0

(150)

(151)

(152)

Combining this with the degree one part of w(T 2 ) = w(T2 )w(T12 ) 0 = w1 (T 2 ) = w1 (T2 ) + w1 (T12 ) (153)

and the orientability of T 1 , T 2 and T X (the tangent bundles of oriented manifolds are necessary oriented [BT, 6]) gives 0 = w1 (T12 ) + w1 (N12 ) (154) which implies w1 (N12 ) = 0 since 1 2 is assumed to be orientable. This simplies the degree two expansion of (151) (because all products of degree one contributions vanish) so that it becomes 0 = w2 (T X) = w2 (T1 ) + w2 (T2 ) + w2 (T12 ) + w2 (N12 ) w2 (T12 ) + w2 (N12 ) = w2 (T1 ) w2 (T2 ) This result can now be applied to show w2 (T 1 ) + w2 (T 2 ) = w2 (T1 ) + w2 (T2 ) +2w2 (T12 ) =
w2 (T12 )w2 (N12 )

(155) (156)

(157) (158)

w2 (T12 ) w2 (N12 ) = w2 (T12 ) + w2 (N12 ) = w2 (T 1 T 2 N 1 N 2 )

In the second line we have used the fact that, since the Stiefel-Whitney classes are Z 2 -valued, addition and subtraction are equivalent implying that wk (W) = wk (W). Applying this reasoning to w2 (T 2 ) = w2 (T 2 ) completes the proof. It is now easy to calculate w3 (T 1 T 2 N 1 N 2 ) by applying the connecting homomorphism (see Appendix A.2) w3 (T 1 T 2 N 1 N 2 ) (w2 (T 1 T 2 N 1 N 2 )) (159)

This can be shown to be given by (w2 (T 1 )) (w2 (T 2 )) using a Mayer-Vietoris argument [BT] applied to the exact sequence of topological spaces. In the following sequence 1 and 2 are two maps of 1 2 into each factor of the disjoint union and p is a projection (see [BT] for more details) 1 0 1 2 1
2

2 1 2 0

Because cohomology theories are dened via contravariant functors, H k , this sequence induces the fol lowing long exact sequence in Cech cohomology with coecients in F
2 p H k (1 2 , F) H k (1 , F) H k (2 , F) 1 H k (1 2 , F)

54

2.7 Normal Bundles with Spinc Structures

2 RR-CHARGE AND K-THEORY

On each side the sequence is connected to the next Cech group via the connecting homomorphism . The sequence is exact because p maps an element H k (1 2 , F) to its restrictions (|1 , |2 ) k (1 , F) H k (2 , F) while maps an element (, ) H k (1 , F) H k (2 , F) to the dierence H 2 1 k (1 2 , F). Here we have let 12 = 1 2 . The deniof their restrictions |12 |12 H tion of and a demonstration of its exactness are given in [BT] and are standard constructions in cohomology/homology theories. Combining this sequence with the sequence 0 Z Z Z2 0 and using F = Z or F = Z2 gives the double complex
mod 2 H 2 (1 , Z) H 2 (2 , Z) H 2 (1 , Z2 ) H 2 (2 , Z2 ) H 3 (1 , Z) H 3 (2 , Z)
1 2 1 2 1 2

mod 2

(160)

H 2 (1 2 , Z)

mod 2

H 2 (1 2 , Z2 )

H 3 (1 2 , Z)

We now demonstrate the two main results Proposition 2.9. Under the same hypothesis as Lemma 2.8 the following two identities hold. a. w3 (T 1 T 2 N 1 N 2 ) = 0 b. There exist two classes, d1 H 2 (1 , Z) and d2 H 2 (2 , Z), such that w2 (T 1 ) = d1 mod 2 w2 (T 2 ) = d2 mod 2 w2 (T 1 T 2 N 1 N 2 ) = [ (d1 ) (d2 )] mod 2 1 2 (161) (162) (163)

Proof. The proof of Proposition 2.9 is a direct application of the commutativity of (160) to the results of Lemma 2.8. From the double complex (160) it follows that if a class, w2 (T12 + N12 ) H 2 (1 2 , Z2 ), is 2 (1 , Z2 ) H 2 (2 , Z2 ), given by the dierence of (the restriction of) two classes (w2 (T 1 ), w2 (T 2 )) H as was in Lemma 2.8, then the commutativity of the right-hand square in (160) implies that w3 (T12 + N12 ) (w2 (T12 + N12 )) = (w2 (T 1 )|12 w2 (T 2 )|12 ) = ( )(w2 (T 1 ), w2 (T 2 )) = ( ) (w2 (T 1 ), w2 (T 2 )) 1 2 1 2 = ( 1 )(w3 (T 1 ), w3 (T 2 )) 2 = w3 (T 1 ) w3 (T 2 ) = 0 since w3 (T 1 ) = w3 (T 2 ) = 0. This completes the demonstration of (a). The fact that ((w2 (T 1 ), w2 (T 2 ))) = w3 (T 1 ) w3 (T 2 ) = 0 (w2 (T 1 ), w2 (T 2 )) = (d1 mod 2, d2 mod 2) which implies (161) and (162). Now Lemma 2.8, along with the commutativity of the left-hand square of (160), implies that [ (d1 ) (d2 )] mod 2 = [( )(d1 , d2 )] mod 2 = ( )[(d1 mod 2, d2 mod 2)] 1 2 1 2 1 2 = ( )(w2 (T 1 ), w2 (T 2 )) = w3 (T12 + N12 ) 1 2 This completes the proof of (163). (167) (168) (165)

(164)

implies that there is a class, (d1 , d2 ) H 2 (1 , Z) H 2 (2 , Z) such that

(166)

55

2.7 Normal Bundles with Spinc Structures

2 RR-CHARGE AND K-THEORY

Recall from the results in Appendix A.2 that a real vector bundle for which the third Stiefel-Whitney class vanishes admits a Spinc -structure. Thus Proposition 2.9 implies that for any D-brane intersection the bundle (146) on the I-brane world-volume is Spinc . Recall also that the data involved in specifying a Spinc -structure associated to a principle SO(n)-bundle is a (non-unique) choice of line bundle, L12 . In Appendix A.2 it is shown that the Chern class of this line bundle, c1 (L12 ) H 2 (, Z), which species it up to isomorphism, must be in the preimage of the second Stiefel-Whitney class of the bundle c1 (L12 ) mod 2 = w2 (T12 + N12 ) Thus Proposition 2.9 (b) implies that w2 (T12 + N12 ) = (c1 (L1 )) mod 2 (c1 (L2 )) mod 2 1 2 (170) (169)

where L1 and L2 are the choice of line bundle associated with the Spinc -structures on T 1 and T 2 . As discussed in previous sections, factorization of characteristic classes involved in anomalies on Dbrane intersections is important. The spinors on the I-brane are expected to be sections of a Spin-bundle lifted from T 1 T 2 N 1 N 2 (171)

However, as T 1 T 2 N 1 N 2 = T X there is no guarantee that S(T 1 T 2 N 1 N 2 ) = S(T 1 T 2 ) S(N 1 N 2 ) is a well-dened bundle. Because w3 (T 1 T 2 N 1 N 2 ) = 0 implying that T 1 T 2 N 1 N 2 is a Spinc -bundle the line bundle L12 on 12 has an ill-dened square root bundle that makes the following bundle well-dened S(T 1 T 2 N 1 N 2 ) L12
1/2

(172)

Moreover, from the discussion above, it is evident that this bundle can be factorized into the product of two bundles dened on 1 and 2 so that (172) is equivalent to S(T 1 T 2 N 1 N 2 ) L1
1/2

This will be used below to modify the anomaly polynomials Y and Y in the case where T 1 T 2 N 1 N 2 is Spinc but not Spin. 2.7.3 Spinc and Anomaly Factorization

L2

1/2

(173)

It is possible to modify the anomaly calculation given in Section 2.5 to account for (173) relatively easily. Recall that, in that section, the anomaly was factorized using the following polynomials A(T ) ) A(N A(T ) ) A(N

Y = ch(W)

Y = ch(W)

(174)

To understand how (173) eects the denition of these polynomials requires rst interpreting the physical 1/2 1/2 meaning of L1 and L2 . It was argued in Section 2.7.1 that these ill-dened bundles are the gauge bundles, with connections, under which the fermions in the theory have half-integral charge so they should replace the CP bundle W in (174). However, many of the results that will be applied to the polynomials in (174) will assume that W is a well-dened bundle. For that reason it makes more sense 1/2 1/2 1/2 to split Li (for i = 1, 2) into two separate bundles Zi and Zi so that Li Zi Zi and c1 (Li ) = = c1 (Zi ) + c1 (Zi ) in such a way that c1 (Zi ) is an integral class while c1 (Zi ) is half-integral. This is possible 1/2 since c1 (Li ) itself is a half-integral class. Thus Zi will be a well-dened line bundle while Zi will be a square-root bundle that can serve to make (173) well-dened. Of course there is a great ambiguity in how this is done corresponding to the freedom in splitting any half-integral quantity into an integral and a half-integral component. It will turn out, in later sections, that this ambiguity is not meaningful in any way since the K-theoretic interpretation will be insensitive to this choice. With this splitting one can dene anomaly polynomials for (173) via

56

2.8 The RR-Charge of a D-brane

2 RR-CHARGE AND K-THEORY

Yi = ch(Zi ) edi /2

A(T i ) A(N i )

Yj = ch(Zj ) edj /2

A(T j ) A(N j )

(175)

where di = c1 (Zi2 ) is the well-dened Chern class of the proper line bundle Zi2 . It is not hard to see that this does not eect the analysis of Section 2.5 at all since it involves nothing more than splitting 1/2 the contribution of the Chern character of Li , treated as the gauge bundle on i , into two parts in the anomaly polynomial (recall that for a line bundle ch(L) = ec1 (L) ). This splitting is there only to make 1/2 manifest the dierence between the well dened parts of Li and the parts that does not correspond to a proper line bundle. The reason that this has been emphasized is that it will play an important role in the relation to K-theory in the next section and also in Section 3. To see that the modications above do in fact yield the correct anomaly structure recall from (85) that the R-symmetry anomaly is (e(N i ) Yi Yi )(1) (176)

and from (97) that the I-brane anomaly is


(1) 1 2

e(N 1 N 2 ) Y1 Y2 + Y1 Y2

(177)

Plugging in (175) gives integrand (neglecting the Euler class) for the D-brane anomaly ch(Zi ) ch(Zi ) and for the I-brane ch(Zi ) ch(Zj ) edi /2dj /2

A(T i ) A(N i )

(178)

It is not hard to check that these are the correct polynomial expressions expected for the anomaly (see (79) and (92)) with additional contribution of a Spinc -factor only in the case of I-brane intersections.39 Before proceeding let us clarify an issue that has been glossed over. The treatment in this section only applies strictly to Abelian bundles corresponding to single branes wrapping a world-volume, . The non-Abelian generalization of these arguments is more involved and will be discussed in conjunction with the case of a non-trivial B-eld in Section 4. Essentially, comparing the analysis of Section 4 on the Freed-Witten anomaly and its eect on W and the arguments of Appendix A.2 on transition functions for Spinc bundles suggests that the correct generalization is to make the rank N CP bundle, W, be a twisted vector bundle (see Section 4) that fails to satisfy the cocycle condition by a Z2 factor; that is, its twisting is related to the Spinc structure of N . See also [DK] for a discussion of K-theory and twistings.

A(T i T j ) A(N i N j )

(179)

2.8

The RR-Charge of a D-brane

The requirement for overall consistency of the low-energy eective action has generated a very specic form of the coupling of the D-brane to the bulk theory 2 Ni (C) (G) Yi
(0)

(180)

These coupling could have been derived in other ways but the method treated in this section is rather ecient in generating the many dierent terms implicit in (180) and also highlighting their geometrical
39 As noted earlier this is not really an additional contribution so much as a re-formulation of the same expression. The Chern class c1 (L1/2 ) is not the class of a real bundle whereas all the terms that appear (175) are proper characteristic e classes associated with well-dened bundles, though in the case of Z this involved squaring it to get a proper bundle.

57

2.8 The RR-Charge of a D-brane

2 RR-CHARGE AND K-THEORY

signicance. In this thesis, however, we are concerned not so much with the exact form of the low-energy actions but rather the geometric character of the objects that occur in that action, namely the RR-form elds, the NS-NS-form elds, and the associated charged objects. In this section, by considering the eld-equations for the RR-form eld strengths, which are n-form generalizations of Maxwells equations with a source, it will be possible to characterize D-branes topologically. It will turn out [MM] that although the D-brane, which plays the role of a source in the equations of motion, seem, naively, to be characterized by even cohomology classes the exact form of the latter will be very suggestive of Ktheory. In particular the source term in the RR equation of motion can be seen as the image, under a homomorphism between K-theory and even cohomology, of a K-theory class of the Chan-Paton bundle of the D-brane pushed forward into the K-theory of spacetime using a K-theoretic construction analogous to the Thom isomorphism already encountered in cohomology. This observation was rst made in [MM] and provides quantitative evidence for the interpretation of D-brane charges as K-theory classes. This will be the starting point, in Section 3, for a construction which more naturally incorporates K-theory and within which many of the strange cohomological constructions through-out this section have a natural interpretation. From the expression (180) it is easy to extract the actual coupling of the brane to the RR-form potential 2 (C) Yi (181)

which gives the associated equation of motion dG= i (Yi ) (182)

This equation can be expressed another way by noting that Yi is a closed form on i and i is a representative of the cohomology class of the Poincar dual to i in its normal bundle. Recall from e Section 2.6.3 that, for 2 : N the inclusion of the zero section, the Thom isomorphism gives [(2 )! (1)] = [i ] and for any closed H (i ) (2 )! () = () (2 )! (1) = () [i ] Hence, on the level of cohomology (i.e. up to an exact form), dG= i (Yi ) = ! (Yi )
i

(183)

(184)

To simplify the exposition consider the presence of only a single world-volume, (possibly with multiple branes wrapping it), and apply (175) to expand Y in terms of characteristic classes of a bundle on d G = ! ch(W) ed/2 A(T ) ) A(N (185)

Recall that the normal bundle N is dened using the following exact sequence 0 T (T X) N 0 (186)

X) so that N (T or equivalently (T X) = T N . Here : X is the embedding of into T X. Thus the factors of the roof genus in (185) can be re-written as

A(T ) = ) A(N

A(T ) ) A(N

A(N ) = A(N )

A(T N ) = (A(N ))2

A(T X) A( (T X)) = A(N ) A(N )

(187)

58

2.8 The RR-Charge of a D-brane

2 RR-CHARGE AND K-THEORY

Note that the last equality follows from naturality of characteristic classes. Also, recall that = 1 2 where 2 : N and 1 : N X then = and so (185) becomes 2 1 d G = (2 )! ch(W) ed/2 1 2 ) A(N (A(T X)) 1 (188)

Consider now the expression in the argument of the Thom isomorphism ! in the equation above and recall (183) 1 A(N )

(2 )! ch(W) ed/2

(A(T X)) 1 (A(T X)) 1 [ ]

= ch(W) ed/2 = ch(W) ed/2 = ch(W) ed/2 = ch(W) ed/2

1 2 ) A(N 1 A(N ) 1 A(N ) 1 ) A(N 1 ) A(N

(A(T X)) [ ] 1 (189)

(A(T X)) [ ] 1 (A(T X)) 1

[ ]

= (2 )! ch(W) ed/2

(A(T X)) 1

The reason for manipulating (185) to get it in this particular is because the rst factor in (189), while seeming rather involved and unnatural in cohomology, has a particularly natural interpretation in K-theory. This provides the rst quantitative connection between D-brane charges and K-theory. Specically we will use a theorem of Atiyah-Hirzebruch in [Kar, Theorom V.4.11]. Let : V be the zero-section embedding of a (real) 2n-dimensional vector bundle that has a Spinc -structure with d the Chern class of the associated line bundle. The statement of the theorem is 40 Theorem 2.10 (Atiyah-Hirzebruch). For each element x in K 0 () we have the relation ch(K (x)) = ! ch(x) ed/2 A1 (V) ! (190)

Here x is an element of the complex K-theory of the base manifold given by the formal dierence of two complex vector bundles, [W1 ] [W2 ]. The homomorphism K : K 0 () K 0 (V) is the K-theoretic ! Thom isomorphism that is in many ways analogous to its cohomological counterpart (it will be developed in detail in Section 3.5.4). The use of x in (190) is somewhat ambiguous (as an element of K() normally entails two vector bundles). In the formula above it can be interpreted as x V ect() K 0 () for some embedding of V ect(), the abelian semi-group of vector bundles over , into K 0 (), its Grothendieck group. These denitions will be introduced properly in Section 3; for now it suces to think of x as a K-theory class associated with a given vector bundle. In this case the real, 2n-dimensional vector bundle is the normal bundle N (which is even dimensional because the discussion has been restricted to type IIB string theory) and x = [W] is a K-theory class associated with the complex Chan-Paton bundle, W. Thus (190) becomes ch((2 )K ([W])) = (2 )! ch(W) ed/2 A1 (N ) ! (191) A proof of the identity (190) can be found in [Kar] but it is not hard to see why it follows once an explicit form of the map K is known; see Section 3.5.4. !
40 Warning: the theorem in [Kar] seems to have a factor of A(N ) rather than A1 (N ) but this is because Karoubis denition of the roof genus is the inverse of ours (see [Kar, p. 288] versus [Nak, p.442]).

59

2.8 The RR-Charge of a D-brane

2 RR-CHARGE AND K-THEORY

As a consequence of (189) and (191) the RR-eld equations with D-brane source terms can be written in terms of D-branes as K-theory classes associated with their Chan-Paton bundles [MM]. d G = ch (2 )K ([W]) ! (A(T X)) 1 (192)

Equation (192) is suggestive because it implies that the RR-forms are sourced by the D-branes but with the latter appearing as elements of a K-theory. Of course (192) relates cohomology classes in X since the anomaly cancellation argument was phrased in this language. In Section 3 an alternative argument will be presented with a greater emphasis on the topology of the Chan-Paton bundles and, in this context, it will be clear that (192) is a cohomological approximation to a K-theoretic identity. To make this more precise, and suggestive, it is possible to note that (192) is in fact the image of the isometric isomorphism between rational K-theory (K 0 (X) Q, which eliminates torsion subgroups in K 0 (X)) and rational Cech cohomology ch : K 0 (X) Q H even (X, Q) ch([W, V]) (ch(W) ch(V)) A(T X)

(193)

A discussion of the exact way in which (193) is an isometry and of why (191) is very natural in K-theory will be postponed until Section 3.5.6 when enough machinery will have been developed to understand these formulas. Jumping somewhat ahead of ourselves we will briey suggest the origin of (191); the reader should not, at this point, expect to understand this explanation in its entirety but should, none-the-less, use it to try and get a sense for how things t together. At a heuristic level these constructions can be understood as follows. In arriving at (184) we are attempting to push-forward characteristic cohomology classes in HdR (), associated with a complex vector bundle 1 : W , to elements of HdR (X) where X is the ambient spacetime containing , the D-brane world-volume. In cohomology there is a well-dened pull-back of the map : X but to push-forward it is necessary to resort to the Thom isomorphism or, its generalization, the Gysin map. The latter uses Poincar duality to map a class in H dR () to e an element of the dual homology, H (), then use the push-forward (which is natural in homology) to map this to a class in H (X), and nally map this back to HdR (X) via Poincar duality once more. e This is known as the Gysin homomorphism and is equivalent to using the Thom isomorphism to map HdR () HdR (N ) HdR (X). The need for this construction is because RR-charges should be interpreted within the context of some cohomology theory on X such as HdR (X) whereas the charge formula derived via anomaly cancellation is in terms of cohomology class in HdR (). More fundamentally, however, the arguments of this section have suggested that the D-brane should be regarded as a vector bundle rather than just a submanifold. If this is the case then the cohomological arguments given above amount to measuring the class of a vector bundle using its characteristic cohomology classes. A more natural approach would be to classify D-brane charges using the Chan-Paton bundles directly rather than their cohomology classes. Translating the construction of the previous paragraph into the language of vector bundles one can ask how to represent an element of V ect() as an element of V ect(X) where V ect(X) is essentially the semi-group of vector bundles with base-space X (under the operator of Whitney sums). Like cohomology classes vector bundles admit natural pull-backs but not push-forwards. Thus a construction similar to the Thom isomorphism must be used. In this case the vector bundle is rst pulled back to a tubular neighborhood N () of . Recall that this neighborhood is constructed to be dieomorphic to the normal bundle, 2 : N . If this vector bundle, 2 (W) for some bundle 1 : W , were trivial on the boundary of N () it would then be possible to extend it to a trivial bundle all over X by just extending trivially. This is not generally the case; a rather more elaborate construction ensures that the bundle constructed on N () has a trivial K-theory class at the boundary of N () and so can be extended to an element of K(X). The details will be presented in the Section 3.5.4 but this essentially involves tensoring 2 (W) with spinors lifted from the normal bundle. These spinor bundles are used to trivialize 2 (W) at the boundary of N (). The reason we mention this here is that it already suggests somewhat the origin of (191) if one recalls that the factor ed/2 is associated to a Spinc -structure on the normal bundle and also that for positive and negative chirality

60

2.9 Remarks

2 RR-CHARGE AND K-THEORY

spin-bundles, S (N ), lifted from N the following relation holds [ch(S + (N )] ch(S (N ))] = e(N ) A(N ) (194)

This is the origin of the factor A1 (N ) in (191) (the Euler class would also appear in (191) if it were pulled back to as the pull-back of the Thom class, , that appears in the image of (2 )! , is just the Euler class).

2.9

Remarks

Let us pause and make some remarks before moving on to the more geometrical derivation in the next section. We have, rather laboriously, derived the D-brane bulk couplings (often referred to as WessZumino coupling) by resorting to an anomaly cancellation argument. This is an important result that has applications well beyond the K-theoretic nature of D-branes and is hence well worth the eort. In fact, the anomaly canceling arguments above do not correctly determine all the D-brane bulk couplings which is evident in the fact that the terms given in (181) are not T-duality invariant. A discussion of the additional terms required to make the overall coupling T-duality invariant can be found in [HM] and references therein. An interesting application of the chiral anomalies associated with D-brane intersection and their relation to D-brane production can be found in [BDG]. Although we have developed a lot of mathematical and physical machinery to calculate the D-brane coupling there are heuristic arguments that can be used to x several terms in the coupling (181) such as the dependence on the Chern class of the CP-bundles. It is also possible to calculate, term by term, the roof genus contribution by considering string amplitudes. The derivation presented above is somewhere between these two extremes of eort and, moreover, high-lights the geometric origin of the terms which will be essential in establishing the fact that D-brane charges are valued in K-theory. In fact, the normal bundle contribution will be a very important part of the argument in favor of K-theory given in Section 3.5.4. As mentioned in the introduction, Section 1.3, our motivation is not merely to catalog the possible values that D-brane charges can assume but to better understand the topological character of D-branes. A deeper knowledge of the latter can, perhaps, provide a (non-perturbative?) probe into string or M-theory.

61

3 D-BRANES AND K-THEORY

D-branes and K-theory

In this section we will re-examine the results of Section 2 from a dierent perspective, namely that of D-branes as topological gauge defects on the world-volume theory of higher-dimensional branes [Sen1], [Sen3], [Sen4], [Sen5]. This will provide a geometrical interpretation [Wit2] of D-brane charges as elements of K 0 (X) which allows the discussion to be extended to more general settings such as orbifold/orientifold spacetimes and also the case of a non-zero B-eld. To fully develop these ideas it will rst be necessary to say something about characteristic classes and their relationship to vector bundle topology. We will also spend a considerable amount of time introducing K-theory and some of its important properties. Finally, in order to relate these ideas to physics we will review the arguments of Sen regarding brane/anti-brane annihilation and topological defects.

3.1

Overview

In the last section it was shown that the equations of motion for the generalized Maxwell action of the RR-form eld strengths is dG= i (Yi ) = i ch(W) ed/2 A(T ) A(N ) (195)

and it was suggested that this expression implies that, in fact, the current on the RHS of (195) should be regarding as an element of the K-theory of spacetime, K 0 (X). The arguments leading to this are rather technical, however, and obscure the natural geometric interpretation of (195). In retrospect, it is not surprising that the RR-charge should be sensitive to more than the homology (or the dual cohomology) class of the world-volume, , since can carry arbitrary Chan-Paton bundles. As the Chan-Paton bundles derive from open-string end-points on and carry a connection given by massless NS-sector modes one can argue, at a purely heuristic level, that the RR-form potential coupling to should be sensitive to the CP-bundles.41 While consistency requirements, in the form of anomaly cancellation, proved to be an ecient way to determine this coupling they do not shed much light on its geometric character. In this section a dierent approach, due mostly to Witten [Wit2] building on earlier work by Sen [Sen1][Sen3] will be presented. In essence, Witten combined the observation that (195) implies that RR-elds couple, not only to D-branes, but to topological defects on the Chan-Paton bundles with a conjecture of Sens which suggested that D-branes could be constructed from topological defects on the unstable world-volume theory of pairs of higher-dimensional branes. This leads to a characterization of all D-branes as elements of K 0 (X), the generalized cohomology group classifying the CP-bundles of spacetime-lling D9-branes. Wittens construction realizes all lower-dimensional branes as topological defect on these CP-bundles.42 In the simplest case, with at spacetime and world-volume topology, the Dp-brane is constructed from a brane/antibrane pair of p + 2 dimensional branes. The brane and the anti-brane wrap a R 1,p+2 submanifold of spacetime, R1,9 , in such a way that their CP-bundles are conjugate and carry no conserved RR (p + 4)-form charge43 but rather only G(p+2) charge. This implies that their conserved quantum numbers are associated with a Dp-brane, R1,p R1,p+2 , rather than D(p + 2)-branes. To understand why the brane and the anti-brane carry no G(p+4) charge we consider both the eld theory on the branes and string spectrum, both of which contain a tachyonic, Higgs-like scalar eld of negative mass squared. The presence of a tachyon signals an instability in both theories which is conjectured to result in the annihilation of the brane-anti-brane pair. This is analogous to annihilation of particles and anti-particles because, in both cases, the lack of conserved charge implies the total state is not protected from decaying to a lower-energy state by any symmetry.
41 Consider the standard string cylinder diagram which, by a modular transform, can be interpreted either as open strings stretched between two branes or closed strings exchanged between two branes; in this amplitude the existence of a topologically non-trivial A-eld background on one of the branes would eect the closed string amplitude as well. 42 Here the term topological defect is being used in a very general sense to suggest zero-loci of generic sections of vector bundle. 43 Recall that a (p + 3)-dimensional D(p + 2)-brane is normally a source for the (p + 4)-form eld strength, G (p+4) .

62

3.2 Topological Defects, Chern Classes, and Lower Dimensional Brane Charge

3 D-BRANES AND K-THEORY

In the eld theory on the brane-antibrane world-volume, R1,p+2 , this is explicitly realized by allowing the tachyon eld, T , which is a section of the CP-bundle, to assume its vacuum expectation value (VEV) everywhere except on the R1,p submanifold, on which it vanishes. Moreover, the tachyon is a complex scalar eld that supports winding congurations about this codimension two zero-locus. This winding is related to the topology of the Chan-Paton bundle44 and is encoded in the characteristic classes appearing on the RHS of (195). It implies that the zero-locus of T , which is a generic section of the CP-bundle, is Poincar dual to certain characteristic classes appearing on the RHS of (195) and suggests that this e topological vortex on the CP-bundle should somehow be thought of as a Dp-brane in its own right. In order to make this correspondence precise we can examine the local energy density in the region R1,p+2 R1,p , where the tachyon has assumed its VEV, and note, using some results in [Sen1] and [Sen3], that the local energy density in this region is that of the closed string vacuum rather than of a pair of D(p + 2)-branes. Combining these two observations it is plausible to suppose that the brane and antibrane on R1,p+2 have annihilation but the solitonic conguration of their CP-bundles, which supported a codimension two vortex centered on R1,p , has resulted in the latter becoming a genuine Dp-brane. The evidence for this supposition lies in the fact the nal state has the correct conserved charges (both in terms of RR-charge and energy-density) to correspond to a Dp-brane wrapping R 1,p . This construction will be developed in more detail below. This approach recties several shortcomings in the anomaly calculation given in Section 2. First, it provides an intuitive picture of why K-theory (and the particular kind of K-theory) is the right framework for classifying D-branes and specically how K-theory with dierent compactness requirements correspond to dierent branes. In combining the notion of D-brane charge with the new perspective of D-branes as generalized topological defects (instantons, solitons, monopoles, . . . ) it also provides a very natural interpretation of the non-constant45 terms in the expansion (195) as lower-dimensional branes spread-out over the higher-dimensional brane. Mathematically this treatment shows that the natural framework is actually K-theory with cohomology making an appearance only via its relation to the latter. Finally, this approach will make it possible to generalize to the setting with non-trivial B-eld which will be discussed in Section 4. Note that, throughout, the discussion will generally be restricted to type IIB string theory where the constructions will be most transparent. Similar constructions exist in type IIA (where the appropriate K-theory is actually K 1 (X)) and the other string theories such as type I (where real KO-theory must be used) but the focus in this section will be on the type IIB case. The relevant literature is primarily [Wit2] with a brief review in [Wit4] and a much more detailed one given in [OS]. It was already noticed earlier, in [Dou] and [Wit1], that RR-forms couple to topological defects on D-branes and that this might somehow be lifted to a lower-dimensional brane. The notion of tachyon condensation was developed by Sen in a series of papers [Sen1], [Sen3], [Sen4], [Sen5]. This built on observations in [BS] and other references in [Sen1], [Sen3], [Sen4], [Sen5]. Wittens paper [Wit2], which is the focus of this section, is already reviewed in some depth in [OS]. Rather than regurgitate [OS] we attempt to give give a complementary perspective here by focusing more on the relationship with (195), which is not very prominent in [OS], and attempting to formulate Wittens construction in slightly more technical detail.

3.2

Topological Defects, Chern Classes, and Lower Dimensional Brane Charge

In order to understand topics such as brane-anti-brane annihilation and conserved charge associated with topological defects it will be useful to rst introduce the relevant mathematical formalism and show how it is related to some simple examples. Below, it will be shown that the equations of motion derived in the rst part of this thesis imply that RR-forms are sensitive to the isomorphism classes of various vector bundles on a D-brane as well as its actual world-volume. Then, a simple example of a line bundle on a two-sphere will be used to illustrate the mathematical theory of characteristic classes and their relation to zero-loci of vector bundles. While rather specic (and highly overused) this example does demonstrate the qualitatively relevant points (and their relation to homotopy theory) without invoking too much sophisticated machinery. We will suggest that some of these characteristic classes (elements of even cohomology) are Poincar dual to the zero-locus of generic global section of a non-trivial vector bundle e
44 To 45 Recall

support non-trivial bundles we actually consider the compactication of the base-space R n . that locally Y = N + dY (0) where N is constant.

63

3.2 Topological Defects, Chern Classes, and Lower Dimensional Brane Charge

3 D-BRANES AND K-THEORY

(from which it can be seen that they measure the twisting of the vector bundle since a trivial bundle admits one or more independent global non-zero section). This will be important in the next section when, for instance, Dp-branes will be realized as the zero-locus of a particular section corresponding to a charged, tachyonic scalar particle in the world-volume theory of a higher dimensional D(p + 2)-brane. 3.2.1 RR Equations of Motion and Lower-Dimensional Brane Charge

Recall from Section 2 that the source, or current density, associated to the RR-form eld strength, G, is given by the eld equation (195). Let us consider this equation in the presence of N Dp-brane wrapping the world-volume, . If the simplifying assumptions are made that the normal and tangent bundles of have trivial topology (so that the roof genera, A(N ) and A(T ), are both equal to one) and that the c normal bundle is a Spin-bundle (so the Spin -factor, d/2, is trivial) then (195) can be expanded as d G = i (ch(W)) = i (T r e 2 )
n odd
iF

d G(n) = i (N + T r

iF + Tr 2

iF 2

+...)

(196)

Where we have used the dierential geometric representation of the Chern character in terms the curva2 ture, F , of a connection on the bundle seen as an element of HdR () with integer periods. The trace is over the Lie-algebra indices; recall that F is a Lie-algebra valued two-forms so T r[F ] is just a two-form. This is more familiar to physicists but another representation can be given in terms of the Chern classes, ck (W) H 2k (, Z), of W and this will sometimes be more convenient (see Section 3.2.2 or [BT], [LM] and references cited therein) 1 d G(n) = i (N + c1 (W) + [c1 (W)2 2c2 (W)] + . . . ) 2

(197)

n odd

The relation between these two representations comes from the isomorphism between de Rham coho mology and Cech cohomology with real (constant) coecients, HdR () H (, R). The Chern classes = are actually elements of H (, Z) which can be mapped into H (, R) and the curvature forms, F , are assumed to be normalized to have integer periods. We will have occasion to discuss this further below but a reader wishing to learn more should consult [BT] and [Alv]. Recall that is a k-form Poincar dual to a Dp-brane with p = 9 k and that T r[F n ] or cn (W) are e 2n-form so we can re-write (197) degree by degree d G(p+2) = N d G(p) = c1 (W) (198) (199) (200) (201)

1 d G(p2) = ch2 (W) = [c1 (W)2 2c2 (W)] 2 ...

Eqn. (198) is nothing more than the standard Maxwell equation for the RR-form eld strength G (p+2) with a source extended over a submanifold, , with a Poincar dual . The factor of N comes from e the rank of the Chan-Paton bundle on and is consistent with the notion that, when N D-branes wrap a submanifold , the Chan-Paton bundle changes from U (1) to U (N ). Thus, as indicated in earlier sections, the constant part of Y corresponds to the number of branes wrapping . Favoring a naive analogy with electromagnetism the current associated with G (p+2) (i.e. which couples to C(p+1) ) will generally be referred to as Dp-brane current (or charge) even though, as we will see, the current may not be associated with an actual Dp-brane. For example, with respect to this terminology (199) implies that the Dp-brane, , actually carries D(p 2)-brane charge (i.e. is a source for the eldstrength G(p) ). This is because it couples to C(p1) which can be seen by returning to (181), expanding Yi , and recalling that the integral over i picks out forms whose total degree is p + 1 = dim(i ). Yi contains a constant (degree 0) part, N , which will couple to C(p+1) but it also contains other forms of 64

3.2 Topological Defects, Chern Classes, and Lower Dimensional Brane Charge

3 D-BRANES AND K-THEORY

even degree, such as c1 (W) or c2 (W) (of degree 2 and 4 respectively), and these will couple to the lower degree RR potentials via terms such as C(p1) c1 (W) and C(p3) c2 (W). Thus terms like (199) and (200) imply that Dp-branes also carry D(p 2) and D(p 4)-brane charge but that this charge is not given simply by the topology of (i.e. by the homology class of as a closed submanifold of spacetime X) but also by characteristic classes associated with the Chan-Paton bundle, W, on . It is this statement that leads to the K-theoretic interpretation of D-brane charge. For example, it will be suggested shortly that, when W is a line bundle, terms such as c1 (W) should be interpreted as the Poincar dual (in ) to a D(p 2)-brane world-volume, p2 , smeared out on the world-volume e of . This interpretation rests, largely, on the fact that p2 can be seen to carry the correct RR-charge because of equation (199). Note that, by the localization principle, Proposition 2.5, the support of (199) is localized to a neighborhood of p2 . Then, if there is no conserved higher-dimensional brane charge, we can interpret p2 as an independent brane. This notion was suggested as early as [Dou] but it was only with Sens conjecture relating to tachyon condensation that a good mechanism for annihilating higher brane RR charge, as well as the energy density on , associated with its tension, emerged. Although, through-out this section the discussion will be limited to the case where the roof-genus disappears this merely simplies the exposition but does not alter its qualitative features. The roofgenus is related to the topological twisting of the tangent or normal bundles which are real bundles. As a consequence, the associated characteristic classes come in degrees that are integer multiples of four rather than two. They also can be seen to represent vortex-like, or other topologically non-trivial congurations, of sections of the normal or tangent bundle in the same way the the Chern classes of the CP bundles do for the latter (see Section 3.2.3 below). As these bundles are associated with gravity these can be seen as solitonic congurations in some gravitational tensor or spinor elds (such as spinors on the D-brane world-volume). As with the case of CP bundles these vortices are sensitive only to the cohomology class of the gravitational curvature rather than to its actual value as a 2-form. 3.2.2 The First Chern Classes of a Line Bundle

Before proceeding it will be useful to extract some intuition for the couplings (199) and (200) and others like them. To do so requires a better understanding of the meaning of the classes c i (W). This is most tractable when W is a line bundle in which case only c1 (W) may be non-trivial and, in fact, completely determines W up to isomorphism. The case where W is not a line-bundle can be related to the simpler case when it is and, using some results from cohomology, one can extract a similar understanding. If W is not a line-bundle then the various classes ci (W) need to all be considered and it is much harder to do things explicitly. In all cases, the non-vanishing of these classes can interpreted as some topological non-triviality of the bundle which is stable in the sense that it is a homotopy invariant of the base-space X. The arguments in this section will require a level of prociency with Cech cohomology and long-exact sequences in cohomology. Rather than develop this here the reader is referred to [BT] or to the more concise exposition in [Alv]. Fixing a good open cover, U = {U }, of , a line bundle, : L , is completely determined by specifying its U (1)-valued transition functions, {g }, where This is actually a Cech co-chain with values in the sheaf, Cont(U (1)), of continuous U (1)-valued functions, 1 (U, Cont(U (1))). This cochain must be closed because transition functions of vector bundles must satC isfy the cocycle condition (g) = id; hence, it denes an element in H 1 (, Cont(U (1))).46 It is not hard to see that the isomorphism class of L actually depends only on the class of {g } in H 1 (, Cont(U (1))) (this is discussed for SO(N )-bundles in Appendix A.1 but a similar argument holds for complex bundles). The rst Chern class of the line bundle L is the image of this cocycle under the isomorphism H 1 (, Cont(U (1))) H 1 (, Z). Consider the following exact sequence of Abelian groups = 0 Z R U (1) 0
i ei()

g : U U U (1)

(202)

(203)

46 We have dropped the explicit dependence on the choice of open cover because U is a good open cover and hence denes the same Cech cohomology group as any other good open cover.

65

3.2 Topological Defects, Chern Classes, and Lower Dimensional Brane Charge

3 D-BRANES AND K-THEORY

which induces the following long-exact sequence in Cech cohomology (204) ... The construction of the boundary map, for a representative, g = {g } C 1 (U, Cont(U (1))), of the class, [g], is as follows. Select an f C 1 (U, Cont(R)) in the preimage of g (so g = eif ) then take its image in C 2 (U, Cont(R)) under the dierential and dene ([g]) = [(i )1 ((f ))] where (i )1 : C 2 (U, Cont(R)) C 2 (U, Z) is the inverse of the injective map i (dened only on its image). This is a standard construction in dening long-exact sequences in cohomology ([BT, 1]) and can be shown to be independent of the choices made (such as the element f ). The map is an isomorphism because the groups H k (, Cont(R)) are trivial47 for k > 0 and the sequence is exact. Using the construction above we can give the rst Chern class of a line bundle, L, explicitly. Consider a section, s (L), which is given on U by specifying local functions s : U C such that
e . . . H 1 (X, Cont(R)) H 1 (X, Cont(U (1))))
i()

H 2 (X, Z)

H 2 (X, Cont(R))

s = g s = eif s Where f are Cont(R) valued cochains. Let us normalize s to |s | = 1 so it is given by s = e C 0 (U, Cont(R)). These must satisfy exp(i ) = exp(i + if )
i

(205) where

(206)

and, in the case where there are multiple charts, this relation can be applied multiple times to derive something similar to the cocycle condition on f exp(i ) = exp(i + if + if if ) (207)

However this does not imply the cocycle condition f f +f f = 0 but rather, since the equation above only needs to holds in an exponential, f need only be an integral multiple of 2 1 (f + f f ) Z (208) 2 It follows that , so dened, as (f ) where is the dierential map in Cech cohomology, is closed under because 2 = 0. Although it is exact as an element of the trivial group H 2 (, Cont(R)), if 1 considered as an element of H 2 (X, Z) it is not exact unless 2 f is integer valued. That is to say 1 is trivial only when L is a trivial bundle since if 2 f Z then g = id for all intersections. Hence H 2 (, Z) measures the twisting of the vector bundle and is indeed the rst Chern class, 2 (X, Z) (or equivalently the Euler class of the line bundle [BT]). c1 (W) H It can easily be checked that , dened this way, coincides with the construction above so that (g) = H 2 (X, Z). Thus (208) gives an explicit representation of c1 (W) in terms of the transition functions of L and by the sequence (204) can be seen to depend only on the cohomology class of the transition functions g . Let us now briey relate this to the dierential geometric representation of the Chern class. Let F be the curvature of any connection on L. Then, because each U is contractable, the connection has a local representative, A , such that = F = dA (A) A A = i df (f ) f + f f = 2 (209) (210) (211)

where the f are the same cochains as given in (205). Eqn. (210) is the standard relation between the local form of a connection on dierent elements of a cover (see [Nak]). That is, the local forms of A are related by a gauge transformation. Now consider the following double exact sequence, known as the Cech-de Rham complex,
47 This follows because any cocycle in this group is also a coboundary. It is easy to show this explicitly using a partition of unity subordinate to U .

66

3.2 Topological Defects, Chern Classes, and Lower Dimensional Brane Charge

3 D-BRANES AND K-THEORY

r ker C 0 (U, 2 ) C 1 (U, 2 ) C 2 (U, 2 ) d d d

r ker C 0 (U, 1 ) C 1 (U, 1 ) C 2 (U, 1 ) d d d

The Cech-de Rham complex given above is a double complex where the object in each cell is an element of C p (U, q ), a p-cochain with values in the q-forms. Note that r = 0 and d = 0. Thus, the bottom row correspond to constant, R-valued functions and hence are elements of C p (U, R) while the left-most column corresponds to globally well-dened forms and hence to q (). The map r is the restriction map taking a global two-form, , to its restriction |U and is the embedding map : C (U, R) C (U, Cont(R)) C (U, 0 ). It can be shown [BT] that the cohomology of the bottom = most row and the left-most columns of (3.2.2) are isomorphic and this is essentially the statement that HdR () H (, R). = The Cech-de Rham complex can be used to represented eqns. (209)-(211) in order to show that is the image of F under the aforementioned isomorphism 2 () F
r

r ker C 0 (U, 0 ) C 1 (U, 0 ) C 2 (U, 0 )

ker d

ker d

ker d

1 ()

{A }

{F } d

0 ()

{(A) } i{df } = 0 d i{f }

{(f ) } = { } C 2 (U, R)

C 0 (U, R)

C 1 (U, R)

The explicit form of the isomorphism, HdR () H (, R), is given by working from de Rham cohomology = to Cech cohomology by zig-zagging down the complex as done for F above. This does not depend on 2 the fact that the class F is the curvature of a connection. For any HdR (), one can nd local 0 (U, 1 ) such that |U = d because each U is contractable and hence all one-forms, { } C closed forms are also exact. Because is a global closed two-form it satises = on overlaps which implies = + d for some C 1 (U, 0 ). Because the operators d and commute (which is easy to check explicitly) d() = (d) = () = 0 so () is constant and hence an element of C 2 (U, R). Also, = 2 = 0 so H 2 (, R) and is the image of under the isomorphism (one 2 (, R) to H 2 () using this sort of logic). What is special in the case when F is can also go from H dR the curvature of a connection is that the { } H 2 (, R) must actually be integer valued as shown in (208). This implies that F has to have integer periods (must integrate to an integer on any element in H2 ()). Please note that through-out this section the term integer is being used loosely to refer either to Z or to an isomorphic group like 2Z. The dierence between these two is obviously just a question of normalization. See [Alv] for more details.

67

3.2 Topological Defects, Chern Classes, and Lower Dimensional Brane Charge

3 D-BRANES AND K-THEORY

3.2.3

Poincar Dual of Zero Loci e

An essential characteristic of topologically non-trivial vector bundles, : W , (i.e. vector bundles which are not isomorphic to the direct product Cn) is that they may not admit global, non-vanishing sections. Rather, generic sections, seen as an embedding s : W, may necessarily vanish on a certain submanifold, Z , which we will refer to as a zero locus. More precisely Z s() s 0 () where s0 : W is the inclusion of in W as the zero section (the origin in each ber, Cn ). We will refer to the section, s, as transverse if its intersection with s0 () is transverse in W, meaning that, at each point in Z the tangent space of s() and s0 () span the tangent space of W (i.e. the total space of W). For such sections we have the following theorem [BT, Prop. 12.8] Proposition 3.1. For an oriented vector bundle : W over an oriented manifold the Euler class e(W) is Poincar dual to the zero locus, Z , of a transversal section. e Note also that for complex vector bundles the Euler class is equivalent to the top Chern class, c n (W), where n is the rank of the bundle [BT, 21]. What this means is that, when this class is non-vanishing, generic sections must vanish on a codimension 2n submanifold of given by Z which is Poincar dual e to cn (W) = Z . Of course for this to be the case the section must be transversal. Note that s() and s0 () will have the same dimension as and so if n, the rank of W, is less than dim() then a non-transversal intersection will necessary have tangent vectors parallel to and this will simply increase the dimensionality of Z. We will not prove Proposition 3.1 but refer the reader to [BT]; rather we will consider some explicit examples when n = 1 so W is a line bundle and we can work things out explicitly. We rst consider the rather well known example of a line-bundle, L, on S 2 . Ordinarily all our covers, U, above have been good open covers meaning that all of their intersections are contractable but in this example we will start by considering a cover of S 2 that does not satisfy this requirement. The reason for this is that it is more straight-forward; we will switch to a good open cover afterwords to connect with the discussion in the previous section. Let U consist of two contractable charts, one to cover the northern hemisphere, one to cover the southern. For deniteness let the charts be (see Fig. 4 (a)) UN = {0 /2 + , 0 2} US = {/2 , 0 2} These charts are contractable and so the restriction of L to either chart must be a trivial bundle. Hence, any topological non-triviality of the bundle L will derive entirely from its transition function on U N US . A single transition function is dened on this intersection which is an innitesimal neighborhood of the equator gN S : UN US U (1) (212)

As UN US is homotopic to S 1 and U (1) is topologically S 1 the function gN S denes an element of 1 (S 1 ). Such elements are classied by their winding number m Z. If L is the trivial bundle, S 2 C, then gN S = id is a representative of a the trivial element in 1 (S 1 ) with winding number 0. If, however, gN S represents a non-trivial element in 1 (S 1 ) with winding number m > 0 then it clearly cannot be continuously deformed to the trivial element with m = 0 and so corresponds to a non-trivial line bundle as we will see in more detail below. It is not hard to show that equivalence classes of g N S in 1 (S 1 ) correspond to isomorphism classes of bundles so line bundles on S 2 are classied by 1 (S 1 ). Let us now consider whether it is possible to dene a non-vanishing global section of L. To dene L we must specify gN S as a function from UN US to U (1) gN S () = eifN S () (213) where is the angular coordinate along the equator, = /2; it is the angle parameterizing the S 1 UN US . We will dene a section by specifying its local representatives, s = |s |ei on each element of the cover. We rst attempt to construct a non-vanishing section by extending one of the charts, U N or US , as far as possible. Consider the new charts (see Fig. 4 (b))

68

3.2 Topological Defects, Chern Classes, and Lower Dimensional Brane Charge

3 D-BRANES AND K-THEORY UN UN U S

UN

US

UN U S PSfrag replacements US

(a)

(b)

Figure 4: Two charts, UN and US covering S 2 (a). These can be contracted to an arbitrarily small neighborhood of the north pole (b).

UN = {0 , 0 2}

US = { , 0 2}

so the northern hemisphere has been contracted to an innitesimal neighborhood of the north pole. On US , which is contractable and hence topologically trivial, this allows us to dene non-vanishing section sS = 1 with its U (1) phase xed at S = 0. Specifying sS completely xes sN on the overlap, UN US , as sN = gN S sS = eifN S
1

(214)

Now if gN S represents a non-trivial element in 1 (S ) then it must have some winding number, m, and we can let fN S = m. Hence sN has winding at the boundary of UN . On the circle, UN US , bounding UN , given by = , the section is given by sN ( , ) |sN |eiN (
1 ,)

= eifN S ()

(215)

so its phase e denes a map from S to U (1) that winds m times since N ( , ) = fN S () = m. The winding number of this map is invariant under continuous deformations. To dene N (, ) in the interior of UN we must extend the map N ( , ) continuously over the region 0 . This denes a continuous, one-parameter family of maps. () N (, ) : S 1 U (1) (216)

iN ( ,)

All the members of this family must be homotopic as it is parameterized continuously by but this is problematic at = 0 which is topologically a point. Maps from a point to S 1 can never have winding so it is impossible to continuously vary from with winding number m to 0 with winding number 0. The resolution to this problem lies in the fact that N , and hence , are only well-dened if |sN | = 0. Hence at some point (which has been labeled = 0 here) |sN | = 0 implying that the section s vanishes at at least one point. One might try to redene sS in such a way as to avoid this but it would not be possible. Intuitively this follows because the transition function fN S introduces winding in the phase of any section which 69

3.2 Topological Defects, Chern Classes, and Lower Dimensional Brane Charge

3 D-BRANES AND K-THEORY

can be absorbed either by the northern or the southern hemisphere or by both (in the case of m > 1) but in any case it will result in a vortex conguration on one of the two hemispheres which implies the section must vanish at some point (another way to see this is that the hemispheres are homotopic to a point so cannot support winding maps to S 1 unless a point is removed somewhere to make them non-contractable). This example is intended to illustrate Proposition 3.1. Although we have not specied the Chern class of L (and indeed it is not evident how to do so since we are not working on a good open cover) we expect it will be non-trivial since the bundle L cannot be isomorphic to the trivial bundle (due to the non-triviality of gN S in 1 (S 1 )). If c1 (L) is a non-trivial element of H 2 (S 2 , Z) then its inclusion in 2 H 2 (S 2 , R) HdR (S 2 ) denes a closed two-form (with integer periods) which, by Proposition 3.1, will be = Poincar dual to the zero locus of a generic transversal section. In the example the base space, S 2 , is e two-dimensional so the relevant zero-locus must be zero-dimensional (a point) which was shown to be the case for the section s above. The construction of the previous paragraph already provided an intuitive sense of how, in this particular example, the Euler class was localized around the north pole. This essentially means contracting all the curvature of the connection to an innitesimal neighborhood of the north pole. We have presented the above using a cover of S 2 which contains non-contractable intersections because it allowed us to arrive at the desired results quite quickly. We will not spend much time discussing how to rephrase this using a good open cover wherein the role of the Chern class will be more transparent. We will introduce this briey and leave it to the motivated reader to work it out for themselves. To cover S 2 with a good open cover let us divide US down the middle into UR and UL so now all charts and their intersections are contractable (see Fig 5). On UL UR the transition function gLR is just dened to be the identity; even if we started with some other transition function we could always choose coordinates on UL UR so that gLR is trivial since the union UL UR is contractable. Once again all the winding is about the equator in the double intersections UN UL and UN UR (as well as the triple intersections UN UL UR ).

UN p2 PSfrag replacements UN U L p1

UN U L U R

UN U R

UN U L U R

UL

UR

UL U R Figure 5: A good open cover of S 2 given by three charts, UN , UL , and UR . Note that all intersections are contractable. The new transition functions gN R and gN L are simply given by restricting the old transition function gN S to these regions. We can determine the Chern class = f by comparing the winding of a section s ei about the equator. This can be done rst entirely in UN using sN (/2, ) = eiN (/2,) which

70

3.2 Topological Defects, Chern Classes, and Lower Dimensional Brane Charge

3 D-BRANES AND K-THEORY

gives (for some m Z) N (/2, 2) N (/2, 0) = dN (/2, ) = 2m


UN

(217)

To derive a condition on = f let us re-write the above. Let SL UN UL and SR UN UR be two semi-circles partitioning the equator and meeting at the points = p1 , p2 UN UL UR (see Figure 5). Eqn. (217) should be equivalent to 2m =
UN p2

dN (/2, ) =
SR

dN (/2, ) +
SL p1

dN (/2, )

=
p1

d R (/2, ) + fN R () +
p2

d L (/2, ) + fN L ()

(218)

= R (p2 ) + fN R (p2 ) L (p2 ) fN L (p2 ) + L (p1 ) + fN L (p1 ) R (p1 ) fN R (p1 ) = fRL (p2 ) + fN R (p2 ) + fLN (p2 ) + fLR (p1 ) + fN L (p1 ) + fRN (p1 ) = RLN (p2 ) + LRN (p1 ) So the rst Chern class, , summed over S 2 (recall it is only dened on UN UL UR ) gives the winding number of the bundle. One also checks that the curvature, F , of any connection on L must satisfy F =
S2 UN

FN +
US

FS =
UN

AN +
US

AS (219)

=
UN

AN A S = i

dfN S = i 2m
UN

where eqns. (209)-(211) have been used as well as the dierent relative orientations of UN and US . This is what is meant by the statement that F has integer periods. Recalling that S 2 is a deformation retraction of R3 {0} one can glean some intuition as to how this works in the higher dimensional setting. As the radial coordinate r in R3 {0} can be used to 2 parameterize the retraction, one can imagine a one-parameter family of two-spheres S r , each described as above with the sections sr varying continuously between the spheres. Hence the zero-locus at = 0 in the example above becomes a continuous line through the one-parameter family of spheres. This illustrates the fact that the Poincare dual of e(L) in R3 {0} is actually a 1-cycle rather than a 0-cycle. 3.2.4 Higher Chern Classes and Generalized Winding

Having spent so much time on the simplest case of a line bundle one might wonder how this extends to bundles of higher-rank, W. For a bundle of arbitrary rank the Euler class coincides with the top Chern class, cn (W) for n = rank(W) and is Poincar dual to the zero-locus of any global transversal section. e For rank n > 1 bundles, however, there are many more Chern classes that come into play and the simple description that existed for line bundles is (generally) no longer present. There are instances, though they are not at all generic, when the simple logic applies to a line bundle can be extended to a general rank n bundle and this is when the bundle W splits. That is, suppose the bundle W is of the form where the Li are line bundles. In this case the following relations hold c(W) = c(L1 L2 . . . Ln ) = c(L1 )c(L2 ) . . . c(Ln ) = (1 + c1 (L1 ))(1 + c1 (L2 )) . . . (1 + c1 (Ln )) (221) ch(W) = ch(L1 L2 . . . Ln ) = ch(L1 ) + ch(L2 ) + + ch(Ln ) = ec1 (L1 ) + ec1 (L2 ) + + ec1 (Ln ) 71 W = L 1 L2 . . . L n (220)

3.2 Topological Defects, Chern Classes, and Lower Dimensional Brane Charge

3 D-BRANES AND K-THEORY

In this case we can actually work with the separate line bundles that make up W and just apply the analysis of the previous section. However, it must be emphasize that this is generally not true. Rather, in the more general case we must consider the higher Chern classes. It is possible to relate these Chern classes directly to the topology of the bundle (in terms of transition functions) but it is much less trivial than for line bundles and we will only discuss it briey below. Let us rst consider another case where we have some explicit control of the situation and which will be relevant later on. This is the case of rank n > 1 complex vector bundles on a 2k sphere, S 2k generalizing the discussion in the last section. Let W S 2k be a rank n vector bundle. We cover S 2k with two charts, UN and US , as in Section 3.2.3 and apply the same analysis to show that they cannot support any non-trivial bundle topology so W|UN UN Cn = W|US US Cn = (222)

The analysis proceeds in direct analogy with the S 2 case. The equator of S 2k is homotopic to S 2k1 and the structure group of W is U (n) so isomorphism classes of rank n vector bundles on S 2k are given by 2k1 (U (n)). If the transition function gN S : S 2k1 U (n) (223)

is not homotopic to the identity in U (n) then the same analysis as in Section 3.2.3 implies that any global section must vanish on one or the charts since we can x one of the charts to have a locally constant function and then the section on the boundary of the other chart will have a U (n) phase associated with gN S which continuously extended to the origin (via a foliation of the chart by copies of S 2k1 ). We will not attempt to construct the relevant Chern class here as it is much more dicult. Thus in the cases of rank n > 1 it is much more dicult to work explicitly in terms of cohomology, excepting the two rather special cases mentioned above (both of which will occur again below when discussing the physics of Chan-Paton bundles). When we switch to the language of K-theory we will nd a rather explicit construction, the Thom isomorphism in K-theory, which allows us to construct vector bundles whose sections vanish on any given submanifold of spacetime X. Although we will not have to work with higher Chern classes explicitly in the rest of this thesis we would like to at least suggest to the reader how the intuition gleaned from the rst Chern class of a line bundle on a two-sphere can be generalized to understand what the Chern classes measure for higher rank bundles on arbitrary spaces. We will briey summarize a nice discussion found in [Hat1, Ch. 3] without making any pretense at rigor or formality. The point of view we will take is that the argument given in Section 3.2.2 for the vanishing of a section of a line bundle in some chart on a two-sphere can be generalized by approximating spaces by something called a CW-complex. The main idea behind this is that, for the purposes of homotopy or even cohomology, it is possible to approximate a large class of spaces as being constructed by gluing together higher dimensional discs or balls along their boundaries. Recall that Dn = {|x| 1|x Rn } and Dn = S n1 . We can imagine starting from a number of points (recall S 0 is the disjoint union of two points) and gluing them together by connecting them via copies of D1 (a line segment) by mapping D 1 into two points. This is dened a 1-skeleton. We can now attach copies of D 2 by mapping their boundary D 2 = S 1 into any circles in the 1-skeleton. This procedure can be used inductively to build a CW-complex and there are theorems to the eect that many spaces we are familiar with, such as smooth nite dimensional manifolds, can be approximated (homeomorphically or homotopically) by CW-complexes. This is somewhat similar to triangulating a manifold by mapping simplices into it. A full denition of CW-complexes can be found in [Hat2]. Using the approximation of a space by its CW-complex (and recalling the homotopy invariance of vector bundles) one can study if it is possible to dene vector bundles on a manifold by considering the question on its CW complex. One can imagine how this will allow us to extend the notions already developed to study vector bundles on spheres to study vector bundles on more general spaces. For instance, if a vector bundle, W X, restricted to the one-skeleton, X (1) , of a manifold X, is trivial then one can ask if it is possible to extend this vector bundle as a trivial bundle over the two skeleton, X (2) . Assume a rank N for the vector bundle and since it is trivial on the restriction to the one-skeleton we can dene N orthogonal global (non-vanishing) sections, si , over this space. We can now pull W and si back from X to D2 via the map D2 X (2) which maps the boundary into the one-skeleton, D 2 = S 1 X (1) . The pull-backs of si on the boundary S 1 dene a map from S 1 to U (N ) and hence an element of 1 (U (N )). If this

72

3.3 Tachyon Condensation: Brane-Anti-brane Annihilation

3 D-BRANES AND K-THEORY

map is not homotopic to the trivial map then, as discussed for line bundles in Section 3.2.2, we will nd an obstruction to extending the N sections over D 2 (in Section 3.2.2 we had only one section and so considered elements of 1 (U (1))). This obstruction can be related to an element of H 2 (X, 1 (U (N ))) though indirectly by relating it to cellular cohomology [Hat1]. Since k (U (N )) = Z for high enough N and k this suggests the origin of the Chern classes as elements of H (X, Z) that measure the obstruction to extending N orthogonal sections onto higher dimensional cells. It is also possible to measure the obstruction to extending N < N orthogonal sections though this is slightly more complicated. Although we have skirted many details we would like to high-light the essential point that the analysis of bundles on spheres (which, for line bundles on two-spheres, we were able to relate explicitly to Chern classes via winding numbers) can be generalized to any manifold by using CW-complexes. A reader interested in more of the details of this argument should consult [Hat1] and [Hat2]. In the next section the various constructions given above will be related to the specic case of gaugeelds on D-branes. It will then become clear why so much time was spent developing an intuitive sense for what the Chern classes are and how they relate to topological defects and zero-locus of sections. Namely, when a D(p + 2)-brane and a D(p + 2)-anti-brane wrap the same p + 3 dimensional worldvolume it can be seen, in various ways, that there is no D(p + 2)-brane charge. Anti-branes will be discussed presently; intuitively they can be thought of as the brane analog of anti-particles (i.e. identical to regular branes but with opposite charges). If the Chan-Paton bundles on the D(p + 2)-brane diers from those of the D(p + 2)-antibrane (henceforth referred to as a D(p + 2)-brane) then there may still be conserved quantities in the system associated with generalized solitons on either brane (if their CPbundles are identical then the defects would cancel). Terms in the equations of motion such as (199) and (200) imply that these defects are sources of RR-charge and hence must be conserved. Hence, even though the D(p + 2)-D(p + 2) system might annihilate something must remain to carry this RR-charge and, from the arguments suggested above and elaborated on below, this must be the world-volume of a lower-dimensional Dp-brane which coincides with the defects. This process is somewhat similar to electron-positron annihilation. In that process when the world-lines of an electron and an anti-electron intersect they can potentially annihilate as the total system has no conserved electric charge. However, the electron and positron world-volumes also have additional charges associated with their spin. This is charged with respect with local Lorentz symmetry and the electron-positron system carries a net charge with respect to this symmetry. As a consequence, even after the annihilation of the electron and the positron, there must remain a photon to carry the net charge associated with this symmetry.

3.3

Tachyon Condensation: Brane-Anti-brane Annihilation

To develop a more complete understanding of how K-theory can be applied to classify brane charges it will rst be important to realize Dp-branes for p < 9 as topological defects in the world-volume of higher dimensional branes. To understand how this works it is necessary to consider the notion of an anti-brane, analogous to that of an anti-particle such as a positron. An anti-brane diers from a brane only in the sign of its RR-charge (corresponding to a dierent choice of orientation) and as a consequence the combined system has no conserved RR-charge and hence can decay. This is signaled, at the string scale, by the emergence of a negative mass-squared tachyon in the spectrum which can be interpreted, in the eective world-volume eld theory, as a Higgs-like potential for a charged scalar eld implying that perturbation theory about the zero-point of the tachyon eld is unreliable. The spontaneously-broken theory in this situation will be identied with the system state after brane and anti-brane have annihilated. As has been emphasized before, however, a D-brane carries additional charge associated with its Chan-Paton bundles and these may be conserved even after annihilation. This can be understood from the eld theory on the brane as a consequence of topological defects that prevent the tachyonic eld from realizing its vacuum expectation value (VEV) everywhere. The subspace on which the tachyon does not realize its VEV is to be identied with the worldvolume of the lower-dimensional brane that is the source of the lower-dimensional brane charge carried by the CP bundles of the original brane/anti-brane pair. This is an explicit realization of the idea that somehow lower dimensional brane charge should be interpreted as a lower-dimensional brane spread out on the world-volume of a higher-dimensional one [Dou] [Wit1]. Once this notion has been developed it will be possible to gain a more direct understanding of the role of K-theory in classifying D-brane charge. The ideas discussed here are due primarily to Sen ([Sen1], [Sen3]) but some discussion can also be found in [Wit2] and [OS]. Further discussion can be

73

3.3 Tachyon Condensation: Brane-Anti-brane Annihilation

3 D-BRANES AND K-THEORY

found in [Sen3] [Sen1] [Sen4] [Sen5] [Ues] and references therein. It should be emphasized that the central idea of this section, namely the identication of the tachyon condensate with the closed string vacuum, is still a conjecture. Although this idea has been studied extensively and there is considerable evidence to support the conjecture it is still not entirely understood. Much of the more credible evidence involves rather sophisticated machinery [Ues] which we make no attempt to develop here. As a consequence the introduction below will be rather heuristic; we will simply derive certain assumptions which we will use later to show that D-branes can be classied by K-theory classes. The fact that the tachyon condensation arguments lead us to K-theory provides, in some sense, re-enforcement for both conjectures (by way of a non-trivial consistency check for both ideas). 3.3.1 RR-Charge and the Brane-Anti-Brane System

In Section 2.6.2 it was noted that the sign of the charge,48 , that multiplies the D-brane current is xed by a choice of orientation of the D-brane world-volume and is in this sense arbitrary (much as the sign of the charge of an electron is an arbitrary convention). However, one can consider two Dp-branes wrapping a q = p + 1-dimensional cycle, q X, and select dierent combinations of relative orientations on the branes. A single brane wrapping q generates a line bundle L q with an associated charge (again, we assume the tangent and normal bundle are trivial merely to simplify the exposition) d G = q (ch(L)) (224)

Adding a second D-brane with the same orientation expands the gauge group from U (1) to U (2) as massless NS modes with ends on dierent branes are now possible. If, however, this second brane is chosen with the opposite relative orientation then the analysis changes somewhat. To understand fully what happens it is necessary to return to the original calculation of the string spectrum and note that the strings with ends on dierent branes are no longer indistinguishable from those with ends on the same brane. Hence, rather than changing from a U (1)- to U (2)-bundle, there are actually two separate U (1) bundles on q , one associated to the brane and one to the anti-brane. Likewise if there are N pairs of branes and N pairs of anti-branes (we assume the numbers to be equal for now though we need not in general) there will be two Chan-Paton bundles on q , W1 and W2 , one from the brane and one from the anti-brane. Strings with both ends on the brane or anti-brane will generate elds that are sections of Wi Wi (or its conjugate) while those with one end on the brane and the other end on the anti-brane will generate elds which are sections of Wi Wj (or its conjugate) with i = j. Note that here, as in Section 2.4.3, we have relied on the fact that there are well dened vector bundles, W, in the fundamental of U (N ) on the branes because B = 0. Deferring for now a more detailed discussion of the string spectrum let us consider what kind of fermions there are in this theory and how they might eect the anomaly cancellation argument of Section 2. As in Section 2, the situation here involves intersecting D-branes and it is necessary to account for new fermion species in the eld theory deriving from Ramond strings stretched between the branes. As in the previous analysis, anomaly cancellation will only be possible if these contributions factor to components from the separate branes. In this case, as the branes wrap the same world-volume, the analysis is much easier. Ramond strings both between the brane and the anti-brane and on only one of them will generate fermions but, as mentioned above, these will be sections of dierent CP-bundles. Thus ch(V adj ) or ch(V) in the previous section will be replaced by (1)(1ij ) ch(Wi ) ch(Wj ) (225)

i,j

where the factors of 1 are necessary to account for the relative orientations of the branes when integrated over q . It is not hard to see that this factors gives Y = [ch(W1 ) ch(W2 )]
48

(226)
H 0 ()

can be seen as a constant function on , the D-brane worldvolume hence an element of that gets pushed, via k k the Thom isomorphism, to in Hcv (N ) which can be extended to an element to Hc (X) where k is the codimension of . This justies the terminology of or e (for an electron) as charge.

74

3.3 Tachyon Condensation: Brane-Anti-brane Annihilation

3 D-BRANES AND K-THEORY

as the anomalous RR-coupling (we have neglected the normal and tangent bundle contribution since these do not change). We have been very brief because this argument is not signicantly dierent from that of Section 2. A somewhat more detailed discussion can be found in [SW]. Let us return to the case N = 1 so we replace Wi with Li to indicate a line bundle. The charge formula associated with eqn. (226) is d G = q (ch(L1 )) q (ch(L2 )) = q 1 + c1 (L1 ) + 1 c1 (L1 ) 2
2

+...

1 + c1 (L2 ) +

1 c1 (L2 ) 2

+...

(227)

Now considering the dierent kind of D-brane charge associated with this conguration one nds d G(p+2) = 0 (228) (229)

d G(p) = (c1 (L1 ) c1 (L2 ))

So this system carries no Dp-brane charge (associated with the world-volume, q ) but does carry D(p2)brane charge. If D-branes are considered as higher dimensional generalizations of particles in the bulk supergravity action (12) then (229) is an indication that there is no conserved charge associated with the Abelian gauge eld G(p+2) and hence the decay of the brane/anti-brane system into a system with no G(p+2) sources is not prohibited by any symmetry of the system. However, as with particles, there may still be other conserved charges, associated to dierent symmetries, that must be preserved in the decay. By analyzing the relevant conserved charges it should be possible to determine what the possible output of the decay process is. It should be understood that this argument is merely a useful analogy as this situation is dierent in several fundamental ways from particle scattering. 3.3.2 SYM on a D9-D9 System

One essential dierence is that the D-branes support a eld theory described by a Super Yang-Mills action similar to (31) and one must understand what the brane/anti-brane annihilation means in terms of this theory. To do so requires rst determining what the open string-spectrum looks like when strings are able to stretch between a brane and an anti-brane and what the resultant low-energy action will be (it will not be given completely by (31)). We will not attempt to give a derivation of the tachyonic mode that emerges in the open string spectrum of strings with one end on the brane and one on the anti-brane because this would involve developing more details of string perturbation theory than we have thus far in this thesis. For such open strings what happens is essentially that the NS sector GSO projection is inverted and no longer annihilates the tachyonic ground state but rather the excited state that would normally generate the gauge eld (see [Wit2] and references mentioned there). There are several ways that one can see this. It is possible to study a closed string exchange between D-branes and consider the eects of reversing the sign of the RR-sector contribution (so the relative RR-charge of the D-branes are dierent) when this exchange is mapped, via a modular transformation, to an open string vacuum diagram. One can also study the potential associated with a brane/anti-brane pair using such a diagram and note that it suggests the lowest energy string mode has negative mass squared at small separations [BS]. Another approach is to study the spectrum explicitly by repeating the analysis of the string mode expansion in Section 2.4.7 and extending it to branes intersecting at arbitrary angles [BDL]. One can then vary the intersection angle continuously from 0 to (so the branes are anti-aligned) and see how the spectrum is eected. In all cases one would nd that the NS modes are tachyonic scalars with negative mass squared while the Ramond modes are fermions. Using this information on the spectrum of the open strings in the brane/anti-brane system it is possible to consider the corresponding low-energy eective action. As in Section 2 the detailed structure of the action will not be as important as the geometrical character of the elds that occur in it. In order to provide a concrete working example, however, we will rst consider a specic action for a single pair of D9-D9-branes wrapping R1,9 . As this example uses D9-branes it obviously applies only to type IIB string

75

3.3 Tachyon Condensation: Brane-Anti-brane Annihilation

3 D-BRANES AND K-THEORY

theory wherein stable branes must be even dimensional. For a D9-D9-brane system in at spacetime the action has a relatively simple form (this is a modication of an action given in [Pes]) d10 x F1 F1 + F2 F2 + 2i1 iD1 1 + 2i2 iD2 2 + 2i12 iD12 12 + 2i21 iD21 21 + D12 T D12 T m T T + V (T T ) The subscripts above indicate the nature of the eld: subscript 1 or 2 indicate elds generated by open strings with both on the brane or anti-brane, respectively, while 12 indicates elds from open strings stretched between the branes. Note, as mentioned above, that there are fermions in all these sectors. The strings on a given brane transform under the adjoint of the branes gauge-group (which would be trivial for a single brane with a U (1) group) whereas strings stretched between two branes transform in a mixed representation. Thus the covariant derivative has the form Dij = d + Ai () + () (Aj ) though in (230) above, where only a pair of branes is considered, the Lie algebra action on itself is simply given by multiplication. The eld T is the eld associated to the tachyonic open string mode stretched between the brane and anti-brane so T (L1 L2 ). In the non-Abelian generalization of (230) all the elds become matrix valued and the action will involve traces over the terms (recall the discussion in Section 2.4.3). V (T T ) is a polynomial expression in T T representing the tachyon potential. More generally there will be couplings to other elds in this expression (for instance, on lower dimensional branes, there will be scalar elds associated with transverse components of the gauge eld under dimensional reduction which will still couple to T T because they are T-dual to the gauge eld kinetic terms). These will be neglected in this discussion but are discussed in more detail in [Pes]. The non-derivative terms in V (T T ) depend on T only via T T . In the Abelian case (i.e. the case when T and T have one component each) this follows from the requirement that all terms in the Lagrangian be real and gauge invariant. When there are more than one pair of D9 and D9 branes, even though the exact form of V (T T ) is not known, several qualitative features can be determined. Let there be M branes and N anti-branes. In this case T is the matrix-valued scalar eld generated by all strings with one end on the brane and the other end of the anti-brane; likewise T is the complex conjugate of T corresponding to strings with the opposite orientation (recall that the CP-matrix (32) is anti-hermitian). Consider the string amplitude in Fig 1 and note that each T string takes a brane CP label to an anti-brane CP label and each T string takes an anti-brane CP label to a brane CP label. Thus to have an amplitude with a non-vanishing CP trace requires an equal number of T and T . This can also be seen directly by looking at the CP-matrices of T and T T = 0N N TM N 0N M 0M M T = 0N N 0M N T N M 0M M (231)
2

SDBI =

(230)

Note that neither matrix can have diagonal entries since strings with both ends on the same brane are never tachyonic. Since all string amplitudes will come with a trace over the CP factors and the tachyon potential, V , is a 0 limit of such amplitudes we argue that it must be of the form T r(V (T T )). 3.3.3 Spontaneous Symmetry Breaking and the Closed String Vacuum

From the string spectrum we know that the mass of the tachyon is imaginary so its quadratic term will have the wrong sign. It is assumed that some higher order term in V (T T ) will provide a corrective contribution so that the potential for this eld is bounded from below. V (T T ) is not as easily determined as the terms given in (230) which are mostly constrained by symmetry. It is possible to calculate its form by calculating the relevant string amplitudes and then taking the point-like limit ( 0); this can also determine the dependence of V on other elds (such as the transverse scalars in the case of Dp-branes for p < 9). Such calculations are undertaken in [Pes] but will not be reviewed here. Not only does this require a considerable amount of work but it is not very rewarding as the entire eld-theoretic discussion is, at best, somewhat naive. The reason for this is that we have calculated the string spectrum in a xed background and used it to determine a low-energy eective theory. It does not then follow that modifying this theory by giving one of its elds a VEV corresponds to the same string theory. It is not clear, from this perspective, what it even means to modify the background theory. Rather, this situation should be 76

3.3 Tachyon Condensation: Brane-Anti-brane Annihilation

3 D-BRANES AND K-THEORY

treated using considerably more sophisticated machinery such as boundary conformal string eld theory [Ues] or by considering deformations of the CFT dened by the original background corresponding to the tachyon rolling down its potential [Sen1]. A discussion of the dierent approaches to interpreting the tachyon VEV is given in [Ues] where the validity of these approaches is also reviewed. It was shown in [Pes] that, in the Abelian case, the potential V (T T ) gives a Higgs-like Mexican hat function (see Fig. 6 below) for the tachyon potential so that there is some Tvev = 0 for which V (Tvev ) is a minimum. The quantum theory associated with (230) is given, perturbatively, by considering quantum uctuations around some a classical conguration which is a solution to the equations of motion associated with (230). These equations of motion must hold at extremum of the action (230); that is, congurations of T for which S/T = 0 (and likewise for other elds). Such congurations however need not correspond to minima of the action or the energy but may correspond to maxima. In the latter case the classical solutions to the equations of motion for T will be plane-waves with space-like momenta. This pathology can be associated to the fact that the minimum of the potential is not at T = 0. Although the conguration T = 0 does extremize the action, any excited conguration automatically has a lower energy because the energy of a plane wave is given by k 2 where k is the associated Fourier mode and, because of its space-like momenta, any excited state will have negative energy k 2 = m2 < 0 (here the potential energy has been normalized so that V (0) = 0). V (|T |)

|T | = Tvev

|T | = Tvev
T

PSfrag replacements

Figure 6: Mexican hat potential for the tachyon eld on a single D9-D9-pair. The z-axis is potential and x-y plane is tachyons complex value as T = |T |ei C. In passing to a quantum theory associated with a given action one normally solves the free theory (i.e. the quadratic parts) and then derives the full interacting quantum theory by treating it as a perturbation of the free theory given by a small parameter. However, in the case where the action is not at a minimum but a maximum this perturbative approach can no longer be considered reliable. This theory will have pathologies like negative mass states (such as the tachyon). Rather, the quantum theory should be expanded around the stable classical solutions corresponding, in the diagram above, to congurations with |T | = Tvev . Of course, one must check that the quantum eective potential has the same form as the classical one. This potential is given in terms of expectation values of the elds T rather than T itself. In the absence of anomalies the symmetries of the classical ground state will also be present in the quantum theory so the quantum eective potential will have the same form as the classical one and it will suce to work with the latter; this is discussed in [PS, ch. 11]. Sen [Sen1], [Sen3] has argued that the value of V (T T ) for T = Tvev is equivalent to twice the tension of a D-brane so that one can associate the elevated energy of the T = 0 conguration with the energy of

77

3.3 Tachyon Condensation: Brane-Anti-brane Annihilation

3 D-BRANES AND K-THEORY

the brane/anti-brane system, which is unstable. g 1 V (Tvev ) + 29 = 0 (232)

Here g is the string coupling constant and 9 is the tension associated with a D9-brane (it is the prefactor in the D9-brane action). This equation does not follow from a direct examination of the eld theory dened by (230) but rather, as mentioned earlier, requires a more careful analysis that will not be reviewed here. This suggests an interpretation of the tachyon rolling down its potential as somehow modeling brane/anti-brane annihilation so that, wherever |T | = Tvev , there are no branes and hence no open strings and only the closed string vacuum remains. Essentially the overall system has no conserved charges that can dierentiate it from the closed string vacuum: the net D9-brane charge is zero and the energy associated with the D9-D9-brane pair is canceled by the tachyon VEV [Sen3]. Later we will argue that this reasoning applies locally as well so that, in any region of spacetime where (232) holds we will assume that the brane and anti-brane have annihilated locally there (assuming the region does not support any net D9-brane RR charge). For the discussion that is to follow it will only be necessary to extract certain qualitative features from the works cited above and to proceed on relatively naive assumptions. Certain problems will result but they will not be impede the analysis and, in some cases, their resolution is still an open problem. The most pressing example of this, that will occur below, is that proceeding under the naive assumption that the spontaneous-breaking of symmetry in the eld theory models brane/anti-brane annihilation will suggest that there is always an additional U (N ) symmetry remaining in the closed string vacuum that emerges after the annihilation. How to address this issue is still an open problem. 3.3.4 Topologically Stable Vortices in Flat Spacetime

Consider the tachyonic eld theory given by (230) for a single pair of D9-D9-branes with a U (1) U (1) gauge group. From the form of V (T T ) it is clear that, in the U (1) case, V (T T ) will only depend on the modulus of T rather than its phase (since T T = |T |2 ). This implies that even if the tachyon eld has rolled-down its potential to assume its minimal value it may still have winding. The tachyon, at its VEV, is of the form T (x) = |Tvev |ei(x) (233)

Note that, assuming the tachyon has attained its VEV everywhere, the modulus must be constant and can have no spatial dependence. The phase, however, is some (smooth) function of spacetime which can be selected arbitrarily but which, once xed, will break some of the initial U (1) U (1) symmetry (since rotations will only be allowed which preserve the choice of VEV). Normally one would then quantize this theory by expanding the path integral around this stable minimal conguration. Implicit in the form of (233) is the assumption that the modulus of the tachyon eld can be made to assume a constant value everywhere. Following the arguments in Section 3.3.3, this has the interpretation that the brane and anti-brane have annihilated everywhere leaving only the closed string vacuum. Recall, however that T (L1 L2 ). It is not always the case that such bundles admit global non-vanishing sections (as was seen in Section 3.2); rather this is only the case when c1 (L1 L2 ) = 0. As a tensor product of line bundles L L1 L2 is, itself, a line bundle with rst Chern class When this class is non-trivial T must vanish on some co-dimension two subspace by Proposition 3.1. Let us work out an explicit example on the worldvolume of a single D9-D9 pair. The world-volume here is R1,9 which we will consider as a normal bundle, R1,7 R2 , for the embedding R1,7 R1,9 . Consider a particular conguration of the tachyon eld where the phase in (233) has a dependence only on the ber coordinates of the R2 . In general, for the action or the energy to be well dened in a theory it is necessary for the various terms in the action to have compact support (for a non-compact manifold) otherwise they cannot be integrated. As such, the various terms in (230) must vanish outside of some compact subspace Y R1,9 . Restricting to the ber, R2 , and switching to radial coordinates r, this implies that for some positive real number, R, (here the i and j are once again brane or anti-brane labels)
ij Fi (r, ) = D T (r, ) = V (|T (r, )|) = 0 for r > R

c1 (L) = c1 (L1 L2 ) = c1 (L1 ) + c1 (L2 ) = c1 (L1 ) c1 (L2 )

(234)

(235)

78

3.3 Tachyon Condensation: Brane-Anti-brane Annihilation

3 D-BRANES AND K-THEORY

The same must hold true for the fermion kinetic terms and all other terms in the action but these are irrelevant to the discussion. In fact what is really meant by (235) is V (|T (r)|) Vmin = 0 since potential energy is only dened up to an additive constant. Hence what is really required is that, in the region r > R, the potential is minimized so that T (r, ) = Tvev (r, ) = |Tvev |ei(r,) . This compactness requirement will allow us to compactify the R2 associated with the ber to a sphere S 2 . More precisely, such compact congurations will be physically equivalent to congurations dened on S 2 . This is because the various constraints (235) already imply that, for r > R, the eld-strength F ij must be zero and only the phase of T may vary. Moreover this phase must satisfy D T = 0 (for i = j) and hence
ij D T = ( + Ai Aj )|Tvev |ei(r,) = (i (r, ) + Ai Aj )|Tvev |ei(r,) = 0

(236)

The anti-hermiticity of unitary Lie-algebras has already been used to give A = Aj . Thus the phase for j r > R must satisfy i (r, ) + Ai Aj = 0. Using Fi (r, ) = 0 which implies that Ai = f for some function f it is possible to make a physical choice of gauge such that Ai = 0 (since, for r > R, A is pure gauge). This implies that for r > R the phase satises (r, ) = 0 so the tachyons conguration is physically equivalent to a conguration with a xed value T = |Tvev |ei where has no spatial dependence for r > R. As all elds are now constant on r > R on each ber, R2 , the physical theory is unaected by changing to the ber S 2 by compactifying the region r > R to a point. This is important because the compactness requirement allows for the construction of non-trivial vector bundles on this space (as shown in Section 3.2 the contractable spaces Rn cannot support non-trivializable vector bundles). We have only discussed compactness in the ber directions but we also need to require it along the R1,7 directions. The relevant point is that, in these directions, there is no winding of the phase since only depends on the ber coordinates. On the R1,7 we simply require that all elds vanish at innity. The compactness requirement (in all directions) makes the base space topologically non-trivial (a better way to say this would be that we consider the compact cohomology of the base space and that may be non-trivial even for a contractable space). Hence the setting is topologically S 2 (or a continuous eight-parameter family of such) and the tachyon is a section of the bundle L1 L2 on this space. In such a setting it possible to construct a nite energy vortex conguration of the tachyon eld that is none-the-less stable due to its topology [Sen1]. That is to say it is possible to dene congurations of the tachyon eld which are not uniformly T (x) = Tvev for all x R1,9 yet which are still associated to a stable eld theory because they are nite energy solutions to the equations of motion (and hence extremize the action) and are also stable for topological reasons. Such congurations are often referred to as solitons or vortices. In such a conguration it will be shown that there must be some space of codimension two for which T = 0 implying that open strings can still propagate in this space. We can show this explicitly for the conguration dened above on R1,7 R2 . Let the two charts covering the compactied ber, S 2 , correspond to the disc in R2 with r < R (so on this chart the elds need not vanish) and the complement of the closed disc r > R compactied with a point at innity (where the compact support condition on the elds discussed above implies they satisfy (235) on this patch). Denote these UN and US respectively and note that UN US is homotopic to S 1 . Then it is possible to consider congurations with winding at the boundary of UN so that for r R the tachyon VEV phase is given by (r, ) = m for some m Z. In Section 3.2.3 we showed that a section such as T with TS constant (= Tvev ) and TN having winding at the boundary (TN (R, ) = Tvev em ) correspond to a non-trivial line bundles, L, with non-vanishing rst Chern class. In this case T is playing the same role as s in Section 3.2 so its phase is correlated with the transition functions on UN US via an equation like (206). Thus the sum of the c1 (L) H 2 (X, Z) will be given by the winding number m (see eqn. (218)). The phase associated with such a conguration is depicted below in Fig. 7. The importance of the above construction is that, as T is a section of such a non-trivial line bundle it cannot have a globally non-vanishing section. Irrespective of its phase, the requirement that |T | = |T vev | everywhere would constitute a global non-vanishing section so somewhere this must be relaxed. As in the case of the section s (in Section 3.2.3), it is possible to minimize the number of points where T vanishes to a single point at the center of the northern chart UN which corresponds to the origin of R2 in the initial conguration. Physically this would mean that the tachyon has not assumed its vacuum expectation value at this point. This is the zero section, R1,7 , of the bundle R1,7 R2 and corresponds 79

3.3 Tachyon Condensation: Brane-Anti-brane Annihilation

3 D-BRANES AND K-THEORY

Brane

Brane

PSfrag replacements x9 x10

x9 x10

Figure 7: Left: winding of tachyon phase with m = 1; Right: winding with m = 2. Arrows represent complex phase of tachyon eld T at xed radius. In the diagram Y has been shrunken to a smaller radius but the winding can be considered at the boundary of the compact support of Fi (namely r < R). to a codimension two subspace of R1,9 . Although, for simplicity, we have restricted the dependence of (r, ) to the ber coordinates only, we can consider an 8-parameter family of such functions, x (r, ) where x is an 8-vector, so long as we require they all have the same winding about the equator in S 2 . Let us now say something about the physics of this conguration. Previously it had been argued that a system with a classical eld theory with a eld of imaginary mass would not be stable at the naive solution T (x) = 0 for all x R1,9 . This is because such a conguration is at a much higher energy than a conguration with T (x) = Tvev . In the case we discussed above, however, even though the conguration with T (x) = Tvev everywhere is energetically favored over the conguration with T (x, r, ) = T vev only for r > R, the latter is stable for topological reasons. To decay from the latter to the former would require continuously unwinding the phase of the tachyon eld at the circle r = R and, as has been suggested in examples in the previous section, this is not possible. It is topologically possible to reduce the value of R to some innitesimal value and, moreover, it is energetically favorable to do so, hence, one would 2 imagine that such a conguration would decay until T is approximated by T (r) (1 e br )Tvev for a very large, positive b, so that the tachyon assumes its VEV everywhere except near the origin. The localization of the zeros of T to the origin in the ber S 2 is related, via the Localization Principle and Proposition 3.1, to the localization of the Euler class of L (or rst Chern class, since L is a line bundle) to an innitesimal neighborhood of the zero section of the bundle, namely R1,7 . Sens conjecture would then suggest that outside of this core, where the tachyon eld has attained its VEV, the D9 and D9-branes have annihilated to leave only the closed string vacuum. The energy density and the RR-form charge in this region are equivalent to the closed string vacuum so there is nothing to distinguish it from the latter. Presumably another supersymmetry is also restored locally in this region 49 so here only closed IIB strings propagate. One of the unresolved problems50 with the conjecture is that in the region r > R, where we expect to nd only the closed string vacuum, there is a left-over U (1)-gauge symmetry. The original conguration of D9-D9-branes had a U (1) U (1) gauge symmetry which we can see is broken, in the region outside the vortex, to a U (1) symmetry. What is meant, exactly, by this is that xing a particular a non-zero classical eld conguration to expand around, T (x) = Tvev ei(x) , still leaves an invariance under a diagonal symmetry subgroup of U (1) U (1). Any pair (eif1 (x) , eif2 (x) ) act trivially on such a conguration so long as f1 (x) = f2 (x). Recall, from the discussion below (236) that in the region r > R the value of Ai Aj is xed by i (r, ); this corresponds to breaking the anti-diagonal subgroup of U (1) U (1) but the diagonal subgroup remains as suggested by the fact that Ai + Aj is unconstrained. As remarked
49 See 50 As

footnote 3 in [Sen2] and the references cited there. far as the author is aware.

80

3.3 Tachyon Condensation: Brane-Anti-brane Annihilation

3 D-BRANES AND K-THEORY

in [Pes] no open string modes can couple to this linear combination of gauge connections since any such modes must be oppositely charged with respect to the gauge groups of either ends of the strings. This is problematic outside the vortex as there is no U (1) symmetry associated with the closed string vacuum. At the vortex core the tachyon has not attained its VEV so (232) cannot be applied to suggest that the energy density is the same as the closed string vacuum. Moreover, the fact that F is non-zero in some tubular neighborhood of this region and that it has non-trivial cohomology implies, via (229), that the vortex has D7-brane charge. The vortex is a conguration of the tachyon eld, T (L 1 L2 ) and by construction the latter bundle has a non-trivial rst Chern class associated with the winding number of (r, ) along the equator. The RR equations of motion in the presence of such a vortex are given by (note that = 1 for a spacetime lling D9-brane worldvolume) d G = ch(L1 ) ch(L2 ) [F1 ] [F2 ] 2 2 (237) (238) (239)

d G(p+2) = = 0

d G(p) = (c1 (L1 ) c1 (L2 )) =

The most natural interpretation is then that the vortex is somehow an actual D7-brane as it has both the energy density and the RR-charge of a D7-brane. Note that the vortex is not a vortex of the Chan-Paton bundle on either brane but rather a relative vortex associated with their dierence. That is, it is given by ch(L1 ) ch(L2 ). Thus the charge of the resultant D7-brane will depend on the sign of this dierence. 3.3.5 Topologically Stable Vortices in Non-trivial Backgrounds

Let us now re-phrase this construction more mathematically so that it can be generalized to less trivial spaces. To do this we will abandon our explicit construction and simply apply Proposition 3.1 to the Chern class of L. Consider a general cycle q-cycle q X and wrap a brane and an anti-brane (at this stage the discussion is still limited to a single pair). This denes a pair, L1 and L2 , of vector bundles on n . The tachyon eld is a section of L1 L2 and its conjugate is a section of L1 L2 . Proposition 3.1 implies that the Euler class of this complex line bundle (given by the Euler class of the underlying real bundle) is Poincar dual to the zero locus of a generic transversal section. For a line e bundle the Euler class coincides with the rst Chern class and, as they are both of degree 2, they are Poincar dual to a codimension two cycle embedded in q . Hence there is some (q 2)-cycle, q2 n , e on which a generic section of L1 L2 must vanish. Of course only the homology class of q2 is xed so it is possible to move it around by selecting another homologous representative of the same class (this corresponds in the concrete example above to the freedom to move the core of the vortex around in a continuous manner in every element of the eight-parameter family of congurations). The same physical analysis as above holds: this sub-cycle looks like the world-volume of a Dp-brane for p = q 3. Specically, the Dp-brane charge of this conguration is given by the formula (this holds even in the case when the normal or tangent bundle of q are non-trivial because all terms in the Roof-genus are of degree four or eight) d G(p) = q (c1 (L1 ) c1 (L2 )) (240)

e Since q2 c1 (L1 ) c1 (L2 ) is Poincar dual to q2 in q , eqn. (240) can be recast in a form reminiscent of (198), namely d G(p) = q q2 = q2 (241)

where q2 q q2 is obviously the Poincar dual of q2 in X since q2 is a proper subspace e of q . The construction of the previous paragraph showed how to generate Dp-branes from a D(p + 2)D(p + 2)-brane pair. In doing so, it is necessary to dene a pair of vector bundles on the cycle, p+3 , 81

3.3 Tachyon Condensation: Brane-Anti-brane Annihilation

3 D-BRANES AND K-THEORY

corresponding to a single brane/anti-brane pair. We would like to modify this so that the vector bundles need only be dened on all of spacetime, X, rather than on some specic cycle. This would reduce the problem of classifying Dp-branes to that of classifying bundles on X (though this has not been made precise yet). One might imagine doing this iteratively using the construction above by embedding a chain of cycles in X via q q+2 . . . 10 X where q = p + 1. Thus a Dp-brane can be dened as some generalized vortex in on multiple pairs of D9-D9-branes. One immediate problem with this construction is that it is not evident that such a chain of embeddings exists (though in topologically trivial settings they do). Also, a CP-bundle on X will not, in general, factor into a sum of line bundles that can be treated separately as such a construction would require. This is only the case for bundles that split as discussed in Section 3.2.4. Another approach would be to consider q as a closed submanifold of X and determine its Poincar e dual, q . Inspired by the construction above one may try to construct a pair of vector bundles on X with Euler class q . Proposition 3.1 applies for a bundle of any rank but only for rank one (complex) bundles does the Euler class correspond to the rst Chern class. For a general rank n bundle the Euler class coincides with the (top) nth Chern class, e(W) = cn (W). Since q has codimension 10q it follows that this construction would require a rank n = (10 q)/2 vector bundle which would have a top Chern class of degree 10 q. However, as remarked in Section 3.2.4, the case of rank n > 1 bundles is considerably more complicated than rank one bundles. There are several reasons for this. First, it is not clear that it will be possible to construct a pair of rank n bundles in such a way that all the lower-degree Chern classes will cancel (so there will be no higher dimensional branes charge associated to branes of lower codimension). That is to say, it is not clear that there exist two rank n bundles, W1 and W2 , which can be used to dene n pairs of D9-D9-branes wrapping X in such a way that all the (lower-degree) classes that appear in . . . below will vanish d G = (ch(W1 ) ch(W2 ))

= + cn (W1 ) cn (W2 )

(242)

This is because there are relations amongst the Chern classes and the lower Chern classes cannot always be arbitrarily chosen so that (242) holds. If such bundles could be constructed then they would carry no Dp-brane charge for p = 9 (2n 2), 9 (2n 4), . . . , 9 but would only carry Dp-brane charge for p = 9 2n and also would have a tachyon eld that vanished on a (p + 1)-cycle Poincar dual to e cn (W1 )cn (W2 ); thus a Dp-brane could be realized as a conguration of multiple pairs of D9-D9-branes. For a general (p + 1)-cycle in a general spacetime manifold X it is not clear, however, how to dene such congurations. Moreover, the relation cn (W1 W2 ) = cn (W1 ) cn (W2 ) no longer holds for n > 1 so, even if we could achieve (242), the LHS of this equation would not be Poincar dual to a generic section e of W1 W2 . A full topological invariant of rank n > 1 bundles is given by H 2 (X, G) where G = U (n) is the structure group of the bundle51 but for non-Abelian G this does not carry a group structure so is not a well-dened element of a cohomology theory (see [Bry] and Appendix A). The more natural (generalized) cohomology theory appropriate for classifying vector bundles is K-theory and it is within this framework that the construction of higher-codimension D-branes will be addressed in the next section. One immediate point is that K-theory classies vector bundles of all ranks on a given space rather than classifying them rank by rank (as was attempted with cohomology above); it will be shown in the next section that this is in fact a useful and necessary feature. Before proceeding to this, however, some time will be spent generalizing the discussion above to higher rank bundles which, at least for topologically simple enough spaces, can be done explicitly. 3.3.6 Topological Defects in Higher Rank Bundles

For more than one brane/anti-brane pair we know, from the discussion at the end of Section 3.3.2, that the the tachyon potential will be of the form T r(V (T T )); it will depend on the trace of a polynomial in T T . We restrict to an equal number, N , of branes and anti-branes and work on R1,9 (or R10 ). The
51 For

a line bundle this is just H 2 (X, U (1)) as was already shown in the previous section.

82

3.3 Tachyon Condensation: Brane-Anti-brane Annihilation

3 D-BRANES AND K-THEORY

where Tvev is just some constant numerical pre-factor. The VEV in eqn. (243) would only break a small subgroup of UL (N ) UR (N ), the structure group of W1 W2 . Condensing a single eigenvalue in a matrix-valued scalar theory is quite standard in spontaneous symmetry breaking and is often used as an illustrative example [PS, Ch. 11]. In this case, however, it can be argued, heuristically, that the potential V should be minimized by some value of T with eigenvalues of equal modulus [Wit2]. To argue this imagine separating the N pairs of D9-D9-branes into N spatially separated pairs which would reduce the structure group from UL (N ) UR (N ) to [UL (1) UR (1)]N (at least in the region where the bundle is trivial there should be no obstruction to doing this). The tachyonic elds on each of these pairs will have identical potentials and so will condense to a VEV that has the same modulus but possibly dierent phases (the phase will be selected due to some quantum uctuation or other external selection factor but will not respect the full UL (1) UR (1) symmetry of the Lagrangian). Thus the separate tachyon elds on each of these branes would condense to a VEV of the form (233). When the brane pairs are brought back together additional modes enter the spectrum corresponding to o diagonal CP factors (i.e. strings stretching between a brane from one pair and an anti-brane from another). However, because the tachyons on a single pair have assumed a VEV they are no longer degenerate with the o-diagonal strings and the resulting tachyon eld does not have the full UL (N ) UR (N ) symmetry but rather only the symmetry associated with the stabilizer of (the traceless matrix) i (x) e 1 0 ... 0 0 ei2 (x) . . . 0 (2) . . . . . . . . . . . . . . . . . . . . . . . T (x) = Tvev (244) . . . . . . . . . . . . . . . . . . . . . . . 0 . . . . . . . . . ein (x) Thus we will assume that the tachyon eld, where it assumes its VEV, has this form, though it need not be globally diagonalizable. Unfortunately, it is not possible to make this very heuristic argument more rigorous because the relation between the theory before and after the tachyon has assumed its VEV is hard to capture purely in eld theoretic terms; none-the-less qualitatively one imagines that all the separate pairs must annihilate separately in order to result in the stable closed string vacuum. 52 To see that this stabilizer of (244) is a U (N ) subgroup53 of UL (N ) UR (N ) note that the requirement that U T V = T for (U, V ) UL (N ) UR (N ) xes U with respect to V as U = T V T so the subgroup

coupling to the quadratic term, T r(T T ), in the non-Abelian generalization of (230) will have a negative sign as it will correspond to the negative mass squared of the tachyonic states state (this value is the same for all CP-states with on end on either brane) [BS] [Sre]. We expect that, as in the case with a single pair of branes, there will be a corrective contribution from higher order terms in T r(V (T T )), otherwise the theory will have no ground state. Thus the stability condition on the tachyon VEV would be of the form T r(V (T T )) = C for some constant C. The solutions to this equation would yield a possible vacuum expectation values for the tachyon. Let us consider the diagonalized value of this solution. Recall that, outside of some compact region (i.e. towards innity) all eld congurations must become trivial, the tachyon must assume its VEV, and even the CP bundles are supposed to be trivial when restricted to this region. Hence it is possible to globally diagonalize the tachyon section in this region (where it assumes its VEV). One might thus imagine possible VEVs with non-equal eigenvalues such as the following traceless matrix (this is obviously not in the basis indexed by individual branes since then there could be no diagonal entries) i(x) e 0 ... 0 0 ei(x) . . . 0 (1) . . . . . . . . . . . . . . . . . . . T (x) = Tvev (243) . . . . . . . . . . . . . . . . . . . 0 .......... 0

52 Of course it is possible for individual pairs to annihilate separately via a conguration like (243) where the prefactor Tvev is associated, via (232), not to the tension of all the combined D-branes but only of a single pair. This would still leave behind an unstable conguration of N 1-brane pairs however. This will be exploited later as the basis of the step-wise construction of lower dimensional branes. 53 Ref. [Wit2] seems to suggest that not only the modulus of the eigenvalues of T vev but also the phase must also be equal but it is not clear to the author how this can be argued. In either case the remaining subgroup is U (N ) though in [Wit2] the subgroup ends up being the diagonal subgroup of UL (N ) UR (N ) as Tvev is then a central element.

83

3.3 Tachyon Condensation: Brane-Anti-brane Annihilation

3 D-BRANES AND K-THEORY

stabilizing (244) at a point is isomorphic to one copy of U (N ). Thus, even after tachyon condensation, there is an unexplained U (N ) gauge symmetry in what we presume to be the closed string vacuum (analogous to the U (1) symmetry found in the case of a single pair of branes). The rest of the construction in the non-Abelian case is qualitatively similar to the Abelian one. Rather than R2 one xes a higher dimensional normal space, R2k , and denes a conguration of the tachyon eld which only depends on these 2k ber coordinates. The same compactness arguments follow as before but now the kinetic term for the tachyon must account for the non-Abelian action of the gauge group so has the form (where i = j) Dij Tac = Tac + Ai Tbc + Tab (Aj ) ab bc (245)

Outside of a compact region the requirement that F i = F j = 0 allows us to make a gauge choice Ai = Aj = 0 so the phases of all the diagonal elements of (244) must be constant and, in the complement of the compact region in R2k , all the elds have a xed value so that region can be compactied to a point giving a ber topology of S 2k . Even though generally T is a section of W1 W2 (implying its transition functions are valued in UL (N ) UR (N )) the requirement that T assumes it VEV with all its eigenvalues having equal modulus implies that T is an element of U (N ) (see (244)), after a trivial rescaling of Tvev to 1. Hence T , at its VEV, is just a map from the base space to U (N ). Topologically non-trivial tachyon congurations, corresponding to sections of the non-trivial vector bundle W1 W2 can be dened by covering S 2k with two charts which overlap on the equator which is homotopic to S 2k1 . If there are maps from S 2k1 into U (N ) with winding then it is possible for T to assume its VEV at the boundary of the northern hemisphere in such a way that its has winding at the equator. The same arguments as given for the S 2 case then imply that at some point in the northern hemisphere the tachyon eld has to have modulus zero. These arguments will hold so long as the tachyon eld at the equator, seen as a map T : S 2k1 U (N ), represents an element of the group 2k1 (U (N )) which is non-trivial. Note that this construction is slightly dierent than the one used in Section 3.2.4 to classify rank N bundles on S 2k . There we consider vector bundles in the fundamental representation of U (N ) and classied them by the winding number of their U (N ) valued transition functions on the equator. In this case, T is a section of W1 W2 , hence is not in the fundamental representation. However, we have restricted T to be of the form (244) which means T U (N ) and we restrict to elements in the structure group, UL (N ) UR (N ), that preserve this form. Thus we have, in eect, a principle bundle with U (N ) valued transition functions. This results in the same classication even though the bundles look very dierent. In general, for large enough N , the group 2k1 (U (N )) = Z but there is a natural value of N that is suggested by the fact that this same construction could be undertaken stepwise. Recall from an earlier part of this section that one can attempt to construct a Dp-brane from successive annihilations of brane pairs in such a way that the Dp-brane is seen as the decay product of N D9-D9-brane pairs. As mentioned before this construction may be subject to topological obstructions but if these are neglected one can nd a suggested value for the number of D9-D9-pairs. Namely, starting from N pairs of D9-D9-branes it would have been possible to condense them pairwise by allowing the tachyons between each brane pair to condense separately (by separating the pairs). It is possible to arrange a conguration such that the decay products are N/2 pairs of D7-D7-branes. These could then be arranged to decay pairwise once more to give N/4 pairs of D5-D5-brane pairs and so on until the nal pair decays into a single Dp-brane. This suggests a value of N = 2k1 where p = 9 2k. Although this construction has demonstrated how to generate topologically stable vortex solutions of the tachyon eld which one would like to identify with lower dimensional branes, it is not clear, from the point of view of cohomology exactly how to calculate the charges of these solutions. Moreover, the elegant construction of the tachyon elds zero locus as Poincar dual to the Euler class or rst Chern class of a e line bundle does not hold for the non-Abelian case. Although the top Chern class is Poincar dual to the e zero locus, even for a higher-rank bundle, it does not appear naturally in the RR eld equations, (195), as a source term (i.e. it only appears via its non-trivial relationship to the Chern character). As such it is not clear, from this perspective, that the vortex really carries localized RR-charge. These vortex like solutions can be studied in the context of K-theory where, moreover, they will be completely xed by the K-theory class of the relative CP-bundle congurations. In this setting it will be evident that the conguration carries the correct charge and it will also be possible to localize it via a K-theoretic version 84

3.4 D-branes as K-theory Classes

3 D-BRANES AND K-THEORY

of the Thom isomorphism. An important consideration in guiding us towards the K-theory picture is that (195) suggests that D-brane charges respect the algebraic structure on the category, V ect(X), of all vector bundles on X. This follows because ch(W V) = ch(W) + ch(V) and ch(W V) = ch(W) ch(V). K-theory naturally incorporates this algebraic structure whereas studying vector bundles via their Chern classes, as we have been doing, does not.

3.4

D-branes as K-theory Classes

In this section it will be argued that D-branes are most naturally regarded as classes in the relative Ktheory of the spacetime manifold with compact support. This argument, rst presented in [Wit2], hinges on the observation that Dp-branes can be realized, even in a topologically non-trivial settings, as unstable congurations of D9-D9-brane pairs. Moreover a particular congurations of such pairs corresponding to a given Dp-brane is not unique but is rather a representative of a large number of physical congurations which will still ultimately decay to leave behind the same Dp-brane. Subsuming all such congurations into a single equivalence class and studying the set of such classes precisely reproduces the K-theory of the spacetime manifold. Moreover, the image of this equivalence class under the Chern homomorphism 54 from K-theory to even cohomology gives the Dp-brane RR-charge as calculated in Section 2. From this perspective K-theory is much more natural than cohomology where these equivalence classes do not appear naturally and which does not naturally55 respect the semi-group structure on V ect(X), the category of vector bundles with base-space X. The explicit construction of a Dp-brane in terms of a relative K-theory of X will be given rst for some topologically simple cases and then it will be generalized to X an arbitrary manifold. 3.4.1 K-theory Basics

We start by considering a D7-brane and assuming that it can be generated by a pair of D9-D9-branes (as in some of the examples given in the last section). That is, there exist Chan-Paton bundles L 1 and K1 over X corresponding to the D9 and the D9 such that they carry only net D7-brane charge (recall eqn. (240)). This conguration is not unique however. Another conguration, given by n pairs of D9D9-branes can also give the same D7-brane charge. Specically, consider adding to the original pair n 1 pairs of D9-D9 branes but with the brane and anti-brane in each pair carrying isomorphic Chan-Paton bundles. As each of these new pairs carry no net charge the total net charge of the conguration of n pairs should be the same as that of the original. That is, the RR eld equations with D-brane sources should be unaected because ch(L1 ) ch(K2 ) = ch(W1 ) ch(V1 ) (246)

where W1 and V1 are the new CP-bundles on X associated with the combined branes and anti-branes, respectively. This is obvious physically and follows mathematically from the fact that bundles W 1 and V1 are constructed as the direct sums of the separate line bundles associated with each additional D9-brane and D9-brane, respectively. Note that here we have dened W1 and V1 to split by construction. Thus, ch(W1 ) ch(V1 ) = ch(L1 L2 . . . Ln ) ch(K1 L2 . . . Ln ) = ch(L1 ) ch(K2 ) (247)

where each Li , i > 1, is the bundle associated with the ith D9-D9-pair (the same abstract bundle is used in both cases because the bundles of both are isomorphic by construction). Thus the same D7-brane can be dened by the conguration (L1 , K1 ) but also by any other conguration of the form (L1 W, K1 W) (for W an arbitrary rank bundle) and thus in, classifying D-branes, one would like to identify these congurations.
q fact one must consider not the Chern homomorphism but a modied version, ch() A(T X). 55 What is intended by this somewhat loose statement is that the Chern classes, which to some degree classify the vector bundle do not respect the multiplicative and additive structure on V ect(X); rather one has to pass to the Chern character which is what appears in the RR equations of motion.
54 In

85

3.4 D-branes as K-theory Classes

3 D-BRANES AND K-THEORY

Consider the category56 V ect(X) of complex vector bundles on X and dene an equivalence relation on ordered pairs of objects from this category (vector bundles) via (L1 , K1 ) (L2 , K2 ) if and only if there exists another bundle, W, such that L1 K 2 W L2 K 1 W = (248) Note that this equivalence implies (L1 , K1 ) (L1 V, K1 V) for any bundle V. Let [, ] denote equivalence classes under this relation so the equivalence class [L1 , K1 ] corresponds both to (L1 , K1 ) and to (L1 V, K1 V) for any V. Denition 3.2 (K-theory). Let X be a compact topological space (in fact we will generally take X to be a manifold). Equivalence classes of pairs of complex vector bundles (W, V), under the equivalence relation dened above, dene an Abelian group, K 0 (X), known as the K 0 -group of the space X. This is a group under the operation of addition [W1 , V1 ] + [W2 , V2 ] [W1 W2 , V1 V2 ] (249)

The zero element in K (X) is given by [W, W] for any W V ect(X) and the inverse of an element is given by changing the order [W, V] + [V, W] = [W V, V W] = 0 (250)

The fact that [W, W] 0 follows from (L1 , K1 ) (L1 W, K1 W) because adding the same bundle to both elements of an ordered pair does not change its equivalence class. This construction of a group, K 0 (X), from a semi-group V ect(X) has certain universal properties, which we need not develop here, and K 0 (X) is known as the Grothendieck group of V ect(X). A suggestive notation for the equivalence classes in this group is [W, V] = [W] [V] implying that these ordered pairs represent formal dierences. A more detailed discussion of the Grothendieck construction can be found in [Kar]. The denition above normally requires that X be a compact space as some of the relevant properties for K-theory can only be demonstrated if this is the case. It is possible, however, to weaken this requirement to local compactness and then use the one point compactication, X + , of X to dene the K-theory of a locally compact space X. This will be discussed below. Note that we have restricted the discussion to complex vector bundles. There are a variety of K-theories for dierent kinds of vector bundles (real, complex, etc. . . ) but here we are only concerned with the complex case. Hence, the group K 0 (X) (and the higher K-groups dened below) should be understood to always refer to the complex K-groups, 0 KC (X). Likewise, unless otherwise stated, all vector bundles should be assumed to be complex. A pair such as [W1 , V1 ] is called a virtual bundle because it encodes the dierence between two bundles rather than anything about them (this is evident from the denition of the equivalence relation above). Clearly this is important when considering lower-dimensional branes as relative vortices on the world-volume of higher-dimensional branes and follows quite naturally from eqn. (242) for the charge of such lower dimensional branes where it is evident that only the dierence between these two bundles is relevant. It would even be possible to consider dening a D9-brane with Chan-Paton bundle W as a pair [W V, V] (for an arbitrary V) but it turns out that this is not physically relevant because tadpole cancellation in IIB requires that the number of D9- and D9-branes be equal so it is possible to restrict the construction above to pairs of equal rank bundles [Wit2]. The construction has thus far been motivated by an example of a D7-brane but it was already suggested in the last section that lower-dimensional branes can also be dened as congurations of higher-rank bundles associated to multiple pairs of D9-D9-branes. This will be developed below but it certainly ts well into the construction so far described. K 0 (X) was dened using ordered pairs such as (L1 , K1 ) but these were not restricted to be of rank one and indeed such a restriction would not work since then one could not dene the equivalence relation given in the previous paragraph.
56 The language of Category theory is natural in many areas of mathematics and so will be used when convenient. A deep understanding of category theory will not be necessary in any part of this thesis. The unfamiliar reader can consult [Lan1, 6] or references therein.

86

3.4 D-branes as K-theory Classes

3 D-BRANES AND K-THEORY

3.4.2

Some Aspects of K-theory

Denition 3.3 (Reduced K-theory). The reduced K-group, K 0 (X), of a compact topological space, X, with base-point, is dened as the kernel of the map, ,
0 K 0 (X) K 0 (X) K 0 ({pt}) 0

It will be relevant, for applications to D-branes, to rene the denition of K-theory given above. The exposition here will be very brief with more details available in references such as [Kar], [Ati], and [LM]. K-theory, as dened above, is an un-reduced generalized cohomology theory because it does not satisfy all of the axioms of Eilenberg and Steenrod [Lan2]; in particular K 0 ({pt}) = Z = 0 which follows since vector bundles over a point are easily seen to be classied by their dimension (the possibility for negative integers comes from the Grothendieck group construction of a group K 0 ({pt}) from the semi-group V ect({pt}) given above). To rectify this one can dene the reduced K-theory of a space once we have selected a base-point {pt} X

(251)

where : {pt} X is the inclusion of the base-point in X and is the induced map in K-theory. Thus the reduced K-theory is a generalized cohomology theory and for a connected space, X, we will show below that it is actually the correct K-theory to consider when classifying D-branes. The physical reasons for this will be developed below but it is not hard to understand why this is the case. Recall, in the Denition 3.2, that K-theory is dened via pairs of vector bundles (W, V) with no restrictions on the rank of the vector bundle. When considering the physics of our situation we will nd, however, that this is too general. Loosely speaking, the condition that the tachyon eld assume its VEV at innity (or outside of some compact region) can be translated into an isomorphism of W and V in the neighborhood of innity (this will be made more precise below). Since X is a connected space two vector bundles isomorphic in any region must be of equal rank on the entire space hence we are interested only in congurations where rk(W) = rk(V). This can be encoded mathematically by dening the rank function rk : K 0 (X) H 0 (X, Z) given by That the image of this map is in H (X, Z) follows quite easily from the fact that the dierence in rank of two vector bundles is a constant Z valued function on X (which are exactly classied by H 0 (X, Z)). Thus the K-theory of relevance in physical cases is the kernel of the function rk dened above which we can see ts into the exact sequence
rk 0 ker(rk) K 0 (X) H 0 (X, Z) 0 (253) 0 Surjectivity in this case is not hard to see since for any k H (X, Z) with the ki denoting the value on a connected component, Xi , we can always dene elements of K 0 (X) via pairs of rank ni and rank mi trivial bundles on each connected component, Xi , in such a way as to satisfy ki = ni mi . For a connected space H 0 (X, Z) = Z so (253) is exactly the same as (251) after recalling that K 0 ({pt}) = Z. Both these sequences also split since it is possible to dene right inverses for the maps rk and (which is also a necessary condition to get exactness of the two sequences (253) and (251)) which implies that

rk([W, V]) = rk(W) rk(V)

(252)

(254) ker(rk) and so we Thus for a connected space, X, which is always the physically relevant case, K (X) = are interested in the reduced K-group. Denition 3.2 only holds when the base space, X, is compact as this condition is necessary in several essential results in dening K-theory. For a space that is not compact but is locally compact it is possible to extend the denition to give K-theory with compact support, Kcpt . 0 Denition 3.4 (Compact K-theory). Let X be a locally compact space and dene X + to the the one-point compactication of X. This is the space X {pt} with open sets given by the open sets of X plus subsets in X {pt} whose complement is compact in X (see [Lan1, 5]). The compact K-theory of X is dened as
0 Kcpt (X) K 0 (X + )

K 0 (X) ker(rk) Z K 0 (X) Z = =

(255)

87

3.4 D-branes as K-theory Classes

3 D-BRANES AND K-THEORY

The point {pt} is then the base point of X + . Intuitively, the one point compactication means adding a boundary to the space to make it compact and then identifying the boundary to a point (this is just an intuitive description not an exact formulation). This construction will occur frequently because we will often be working with non-compact spaces such 0 as Rn . For compact spaces this denition coincides with the regular denition of K-theory so K cpt (X) = K 0 (X) for X compact. Another construction that will be useful is the relative K-theory of two spaces, denoted K 0 (X, Y ) for Y a closed subspace of the compact space X. This can be dened in various ways. One useful description is in terms of triples (W, V; ) where W and V are vector bundles over X which are isomorphic over Y via : W|Y V|Y . Two such triples, (W1 , V1 ; 1 ) and (W2 , V2 ; 2 ), are considered isomorphic if there are isomorphisms : W1 W2 and : V1 V2 which commute with the isomorphisms over Y
1 W 1 |Y V 1 |Y | | Y Y

(256)

The space of such triples denes a semi-group, (X, Y ), under the action (W1 , V1 ; 1 ) + (W2 , V2 ; 2 ) (W1 W2 , V1 V2 ; 1 2 ) (257)

2 W 2 |Y V 2 |Y

In the same spirit as the Grothendieck construction it is possible to dene an equivalence relation on this semi-group that will make it a group. First dene the notion of an elementary triple, (W, V; ), to be a triple where W V everywhere and is homotopic to the identity over Y . = Denition 3.5 (Relative K-theory). Let be an equivalence on (X, Y ) dened by setting (W 1 , V1 ; 1 ) (W2 , V2 ; 2 ) if there exist elementary triples, (Z1 , Z2 ; ) and (Z1 , Z2 ; ), such that (W1 , V1 ; 1 ) + (Z1 , Z2 ; ) (W2 , V2 ; 2 ) + (Z1 , Z2 ; ) = The quotient (X, Y )/ denes the reduced K-group, K (X, Y ). The inverse in this reduced K-group is given by [Kar] [W, V; ] = [V, W; 1 ] A result in [Kar] provides another formulation of relative K-theory K 0 (X, Y ) K 0 (X/Y ) =
0 0 0

(258)

(259)

(260)

So the group K (X, Y ) is the reduced K -group of the quotient space X/Y . This space is dened as the space X with the closed subspace Y reduced to a point, {y}, and treated as the base point of X/Y . This is essentially the one-point compactication of X Y [Kar]. To understand how this connects with the earlier description note that the image of Y in this space is the distinguished base point of X/Y . By denition, the reduced K-theory of X/Y is given by K-theory classes, [W, V] in K 0 (X/Y ), which are isomorphic when restricted to the base point, {y} X/Y . Since this point corresponds to the image of Y under the projection, W|{pt} V|{pt} implies that, if such bundles came from bundles on X then they = must have been isomorphic on the restriction to Y which is the pre-image of {y} under the quotient. The group K 0 (X, Y ) ts into the exact sequence [Kar] K 0 (X, Y ) K 0 (X) K 0 (Y )
r

(261)

The map r : K 0 (X) K 0 (Y ) is dened via the natural restriction map r : V ect(X) V ect(Y ) that takes the bundle W X to its restriction over Y denoted W|Y Y . This map is clearly well dened with respect to formal pairs of bundles (W, V). Since trivial elements of K 0 (Y ) correspond to pairs (W, V) which are isomorphic (recall this from the discussion of the additive zero in the K-groups) it is clear that the kernel of this map is given by bundles on X that restrict to isomorphic bundles on Y . That is, pairs (W, V) such that W|Y V|Y . The reason that the group K 0 (X, Y ) will be relevant is that we = 88

3.4 D-branes as K-theory Classes

3 D-BRANES AND K-THEORY

will often want to consider bundles that are equivalent away from a certain cycle X corresponding to the world-volume of a D-brane as we often have done in the previous section. In such cases we will dene Y = X N (), where N () is some tubular neighborhood of (dieomorphic to its normal bundle) and then consider K 0 (X, Y ). 3.4.3 Some Concrete Examples

At this point it is useful to consider a concrete example of K 0 (X, Y ) that might help elucidate exactly what these groups are and also show their relevance in some of the physical cases considered below. Specically, consider X = B n and Y = S n1 = B n so we are considering vector bundles over the closed unit ball, B n = {x Rn | |x| 1}, which must be isomorphic on their restriction to the boundary, S n1 . Since B n is contractable all vector bundles on it are trivializable so there is only one isomorphism class of bundles in any given dimension. However, in the group K 0 (B n , S n1 ) we identify two vector bundles, W and V, over B n according to the homotopy class of the isomorphism that relates them over S n1 . Hence even though W V on B n the fact that this isomorphism, restricted to S n1 , may not be homotopic to = the identity means that the pairs (W, V; ) and (W, W; id) can dene dierent classes in K 0 (X, Y ). As a specic example consider two trivial line bundles B 2 C and dene an bundle morphism between them as follows. Let : B 2 C B 2 C take the form (x, z) = (x, r eim z) where we have passed to radial coordinates, r and , to parameterize B 2 and xed some arbitrary m Z (see Fig. 8 below). Although, over B 2 , these vector bundles are isomorphic and hence dene a trivial element in K-theory, their image under the quotient X X/Y will not be trivial because the isomorphism eim : C C is not homotopic to the identity isomorphism over S 1 . Note also that B 2 /S 1 is topologically S 2 and more generally we have the homeomorphism B n /S n1 S n so it is not surprising that there are suddenly = non-trivial elements in the reduced K-group K 0 (S n ) = K 0 (B n , S n1 ) since S n support non-trivial vector bundles. In fact, as has been shown before, such vector bundles are classied by homotopy classes of maps from the equator, S n1 , to the group of isometries of the bers, namely U (k) (for k the rank of the bundle). From this description it should be clear why the relative K-groups are important since they have exactly captured the arguments given in Section 3.3.4 for the treatment of bundles on Rn with compact support as bundles over S n (here B n denes the compact support of the bundles in Rn ).

r= PSfrag replacements

1 2

r=1

r=

1 2

r=1

(a)

(b)

Figure 8: The map (x, z) = (x, r eim z) sends a at section on B 2 , (a) above, to the section with winding at S 1 = B 2 and which vanishes at the origin of B 2 , (b) above. Note that in (b) the modulus of the section is reduced at the radius r = 1/2.

89

3.4 D-branes as K-theory Classes

3 D-BRANES AND K-THEORY

3.4.4

Higher K-groups and Bott Periodicity

Besides K 0 it is possible to construct additional K-groups, labelled57 K n (X). The higher K-groups can also be introduced via the notion of reduced suspension which is useful since it relates Bott Periodicity to the periodicity of the K-theory of spheres. This is dened in terms of the smash product of two topological spaces X Y X Y /X Y (262)

where X Y = (X {ptY }) ({ptX } Y ) and {ptX } and {ptY } are distinguished base-points of the pointed spaces X and Y . This operation is best illustrated by the case of S 1 S 1 . Here each S 1 is modeled by a line with the end-points identied and its distinguished point is given by the (identied) endpoints of the line (see Fig. 9). From the gure one can see that, after quotienting by X {ptY }, the left and right sides of the square must be identied with the end-points of the top and bottom sides (Fig 9 (b)). If one then quotients out {ptX } Y then the top and bottom of the square are likewise identied with the endpoints of the right and left sides (which have now become points). What is left is a two-sphere, S 2 , so S 1 S 1 = S 2 and more generally S n S m = S n+m . ptY ptX Y ptY ptX Y

X PSfrag replacements ptX ptY Y ptY

ptX ptY

ptX ptY

ptX

(b) (a) Figure 9: Let X = Y = S 1 be two copies of S 1 with base points given by ptX and ptY respectively. Each S 1 is modeled by a line with the base-point at each of its ends which must be identied. Figure (a) is the product X Y . In Figure (b) the space X {ptY } has been quotiented out. A further quotienting by {ptX } Y will result in an identication of all the boundaries in (b) resulting in S 2 . Denition 3.6 (K n -groups). The higher K-groups for a compact space can be dened as K n (X) K 0 (S n X) (263)

The smash product with S 1 is referred to as the reduced suspension. For non-compact, but locally compact spaces there is a denition of the higher compact K-groups
n 0 Kcpt (X) = Kcpt (X Rn ) K 0 ((X Rn )+ )

(264)

We have dened this using compact K-theory but it is also possible to do directly using the K-theory dened earlier for compact spaces as is done in the last equality above. Recall that the compactication represented by the last equality is a one point compactication so the compactication of the space (for example by taking the closure) must be augmented by identifying all the additional compactication points, including those from both factors of in the topological product. This means that the product
57 The reason for the negative integer is that K-theory is a cohomology theory so is contravariant with respect to maps (i.e. inverts directions of arrows) which becomes relevant when constructing long-exact sequences. This is, in any case, a convention.

90

3.4 D-branes as K-theory Classes

3 D-BRANES AND K-THEORY

above is not a trivial product of spaces because the compactication combines the two spaces. An illustrative example would be R1 R1 which has as its one point compactication S 2 rather than T2 , the two torus. This is consistent with the earlier denition because the one point compactication of X R is equivalent to the reduced suspension X S 1 since in both cases it is necessary to identify base-points in the two spaces. All the constructions dened above for K 0 can be extended to K n . An essential property of these groups, however, is that there are only two unique K-groups, which we will label K 0 (X) and K 1 (X). That all higher groups are isomorphic to one of these follows from the property known as Bott Periodicity [Ati] [Kar]. Theorem 3.7 (Bott Periodicity). For any compact space X the following periodicity holds K n (X) K n2 (X) = (265)

This theorem is a very dicult and important result in K-theory. It connects nicely with the periodicity of the D-brane spectrum seen in R10 . At this point the necessary understanding to make this identication has not been developed but one may note that both in IIA and IIB string theories on trivial spaces Dpbranes only exist with p having periodic values with period 2. Bott periodicity is intricately connected with the periodicity of the K-theory of spheres (this is also linked with the periodicity of the representation ring of Cliord modules but we do not develop this here) and (265) can be read K n (S 2 X) K n (X) = which for n-spheres implies K m (S n+2 ) K m (S n ) =
2n

(266)

(267) 0
2n

For S enough machinery has already been developed to determine K (S ). One can start with S 0 {pt} {pt} and use the fact that K-theory is additive under disjoint unions to calculate This also gives, using eqn. (254), K 0 (S 2n ) = Z. Another way to calculate this that is not quite so trivial 0 (S 2 ). The arguments for this are somewhat more involved but it has already been is to calculate K shown in previous sections that non-trivial line bundles on S 2 are entirely classied by winding numbers around the equator, S 1 , and some more technical arguments can be used to show that one needs only consider line bundles so the semi-group V ect(X) is given by Z+ which implies that the full K-group is the associated Grothendieck group, namely Z. To calculate the group K 0 (S 2n+1 ) or equivalently K 1 (S 2n ) requires some machinery from the representation theory of Cliord modules that we do not wish to develop here so we will simply cite the result K 0 (S 2n+1 ) = K 0 (S 2n+1 ) = 0 (269) K 0 (S 2n ) K 0 (S 0 ) = K 0 ({pt}) K 0 ({pt}) Z Z = = (268)

and leave the details to the references [OS], [ABS], [LM], [Kar]. It is worth noting that the group K 1 (X) = K 1 (X) for any X because K 1 (X) K 1 (X) K 1 ({pt}) = K 1 (X) K 0 (S 1 ) = K 1 (X) (270)

At this point enough formalism has been developed to allow us to consider the problem of classifying Dbranes within the framework of K-theory, at least in the topologically simple cases. After addressing these cases it will be necessary to extend some other notions from cohomology, such as the Thom isomorphism and Chern character, to K-theory in order to treat the problem in greater generality.

91

3.4 D-branes as K-theory Classes

3 D-BRANES AND K-THEORY

3.4.5

D-branes in Flat Spacetimes

It remains to see how Dp-branes can be realized as congurations of D9-D9-brane pairs in the context of K-theory. This will rst be discussed for D7-branes in trivial (Euclidean) spacetime, R10 , with compact support and then extended to lower-dimensional brane congurations. In the next section, these constructions will be generalized to topologically non-trivial settings where some complications will arise. For the case of a D7-brane in R10 most of the construction has already been introduced. One considers a pair of D9-D9-branes with line bundles L and K. The compactness requirement on the eld theory means that the space transverse to a point on the D7-brane is eectively S 2 rather than R2 and since the D7-brane has world-volume R8 the tachyon eld is eectively dened on the space R8 S 2 . More exactly, from the discussion of tachyon condensation in Section 3.3 it is expected that a D7-brane can be regarded as the decay product of a D9-D9-pair with Chan-Paton bundles that have a relative vortex localized on the putative world-volume of the D7-brane, . The tachyon eld is a section of L K and, for reasons explained in the previous section, it is assumed to have a constant value outside the closure of some compact tubular neighborhood of which will be denoted N ().58 Physical arguments were already given indicating why this neighborhood should be minimized in size (to minimize the overall energy of the conguration) to some innitesimal tubular neighborhood of . As the tachyon eld, T , is a section of L K it is also59 a section of L K . Any non-vanishing section of such a bundle can be used to dene an isomorphism from K to L simply by contracting the second factor in the tensor product with an element of K (on each ber). Hence in the region Y = X N () where a constant non-zero section of T exists (because the tachyon assumes its VEV in this region) these bundles are isomorphic L|Y K|Y . = In sum, it is expected that D7-branes with eight-dimensional worldvolumes, X, are going to be classied by pairs of vector bundles, (L, K), such that L|Y K|Y for Y = X N (). Combining this with = earlier observations regarding equivalence classes of these congurations (by supplementing both vector bundles with additional isomorphic bundles) leads to the conclusion that this classication is given by 0 Kcpt (X, Y ) where the compactness requirement (in the R8 directions) follows from standard nite-energy arguments. Below we will see that it is actually not necessary to consider X = R 10 but rather to shrink it to R8 B 2 and to shrink Y to the boundary of N () given by R8 S 1 . Consider the case of X = R10 and = R8 so that N () = R8 B 2 where B 2 is as dened above. We will be interested in bundles on this space which are isomorphic at the boundary R8 S 2 . Thus the relevant K-theory to classify D7 branes in R10 is given by
0 Kcpt (R8 B 2 , R8 S 1 ) = K 0 ((R8 B 2 )/(R8 S 1 )) = K 0 (S 2 ) = Z (271) Where we have applied the denition of relative K-theory and noted that the quotient space is homotopic to S 2 . This result has already been derived several times in previous sections so it should come as no surprise. Although, in this case, considerably more machinery has been introduced to arrive at the same result it is none-the-less satisfactory for several reasons. First, these results can be extended to cases where the relevant normal bundle of the Dp-brane is non-trivial (i.e. not B m ) and where the spacetime itself is non-trivial so it is not obvious that one can work purely in the normal bundle of the Dp-brane and extend things to all of spacetime. Extending the constructions introduced thus far to the non-trivial setting will be the subject of the next section but let us rst treat the higher-codimensional branes in R 10 . It is not very dicult to generalize the trivial normal bundle case to R10n R10 for arbitrary codimension n = 9 p 0 Kcpt (R10n B n , R10n S n1 ) = K 0 ((R10n B n )/(R10n S n1 )) = K 0 (S n ) =

Z n even (272) 0 n odd

58 This notation follows as this neighborhood is, by construction, dieomorphic to the normal bundle of [BT, 6] and N () is its closure. 59 Because, for a unitary group, an associated vector bundle via the conjugate representation is isomorphic to the dual of a bundle associated via the fundamental representation. To see this note that for a Lie group G and a representation V acting as g v = gv where g G and v V there is an induced representation on V given, for f V , by (g f )(v) f (g 1 v). The need to use g 1 instead of g follows from the denition of a representation and the non-commutativity of G (consider the action of g followed by h versus the action of gh and it will be clear why g 1 is used). For a unitary group this is given by f (g v).

92

3.5 D-branes in General Space-Times

3 D-BRANES AND K-THEORY

In this form the spectrum of Dp-branes in IIB string theory emerges as a consequence of Bott periodicity as has been suggested earlier. Several points should be addressed here to relate this description to the more explicit description in Section 3.3 of D-branes as relative vortices in pairs (or multiple pairs) of associated U L (n) UR (n)bundles. Clearly when the reduced K-group is non-zero the space admits non-trivial bundle congurations. Recall, for instance, the case of a D7-brane dened as a K-theory class [L, K]. Although these bundles where dened on R8 B 2 R10 as bundles that are isomorphic on Y = R8 S 1 they actually should be considered bundles on the whole space X = R10 . It is, in fact, always possible to dene elements of K 0 (R10 ) that correspond to these bundles. A simple prescription is given in [Wit2] that applies in cases of both trivial and non-trivial topology. Since L|Y K|Y , if L extends from the boundary, Y , = of R8 B 2 to a bundle on all of R10 then we can just extend K in the same way and hence dene an element [L, K] in K 0 (X). This element is necessarily trivial since R10 is contractable but, as an element of K 0 (R8 B 2 , R8 S 1 ), it is non-trivial. Hence its not sucient to simply consider K 0 (X) to classify Dp-branes in X but the necessary compactness conditions (corresponding to nite energy requirements) must be enforced; otherwise, for instance, K-theory would predict no stable Dp-branes in R 10 for any p which is clearly incorrect. The reader should compare this with the discussion of compact cohomology and sources for generalized electromagnetism in Section 2.3.2. In the above formulation it may occur that the bundle L cannot be extended to all of R 10 . It is possible to overcome this obstruction, however, by applying the following very useful theorem [LM, Corollary 9.9]. Theorem 3.8. For any vector bundle W on a compact space, X, there exists a complementary vector bundle W such that W W is a trivializable bundle. It follows that L L is trivial and can be extended over all of R10 (as any space admits a trivial bundle) and furthermore the class, [L L , K L ] = [K, L] by construction (and so represents the same D-brane). This shows that such Dp-branes can always be dened using bundles on all of X and also highlights the relationship between the compact K-theory of the Dp-brane normal bundle 0 Kcpt (N ()) K 0 (R8 B 2 , R8 S 1 ) and the K-theory of the spacetime manifold K 0 (X).

3.5

D-branes in General Space-Times

In the previous section the necessary K-theoretic machinery was introduced to classify D-branes in trivial spacetimes. In this setting the classication does not add anything new to our understanding of D-branes though it does provide a much more natural framework that readily incorporates the freedom to add additional net-charge zero brane pairs. In the previous section it was argued that D-branes in a spacetime X could be classied using a relative K-theory of X which corresponds to a compactness requirement for the gauge bundles on a D9-D9-system. This obscures somewhat the relationship between the world-volume of the D-brane and the K-theory of the spacetime manifold. In this section a slightly dierent perspective will be taken that the D-branes wrapping a submanifold X can be classied by K 0 () and that this K-theory can then be embedded in the relative K-theory (of X) described above. The relative K-theory is intended to be a generalization of compact K-theory where the compactness requirement is being implemented dierently in dierent directions. The appearance of K 0 () is required to allow for more general congurations where the Dp-brane itself carries lower dimensional p 2, p 4, . . . charges. This construction is also closer in spirit to that of Section 2 where cohomology classes on the Dp-brane worldvolume were extended, via the Thom isomorphism, to cohomology classes in the compact cohomology of spacetime. In this setting it will be possible to deal with non-trivial manifolds, and X, and also with a non-trivial normal bundle N (). 3.5.1 Overview

Some of the forthcoming constructions may seem technical so it is useful to rst review the overall logic before introducing too many technical details. The physical reasoning is as follows. D-branes are essentially classied by certain quantum numbers or conserved charges associated with them; namely their RR-charges and their energy density. Both these quantities have topological interpretations: the former is given by characteristic classes of gauge, tangent and normal bundles on the brane worldvolume and the latter can be thought of as a constant function on the brane worldvolume (hence an element

93

3.5 D-branes in General Space-Times

3 D-BRANES AND K-THEORY

of H 0 (, R)). Another, more useful, topological description of the energy density is as a classical eld conguration of the tachyon eld seen as a generic section of the bundle W V over X associated with D9-D9-brane pairs. This follows because wherever such a section can assume a constant, non-zero value, then it can be set to the tachyon VEV which implies, via eqn. (232), that the local energy density there is equivalent to the closed string vacuum (because the tachyon potential at the eld value |T vev | exactly cancels the D-brane tension of the D9-D9-pairs). Thus there will be a non-zero energy density (suggesting a possible D-brane worldvolume) only on the zeros of such a generic section. The zero-locus can be determined via characteristic classes of W V and is, in any case, a topological invariants of the bundle of which the tachyon eld is a section. Note that the two conserved charges we have discussed are related to topological invariants of dierent bundles; the RR-charge is given by characteristic classes of the Chan-Paton bundle on the Dp-brane worldvolume, , while the energy density is given by the zero-locus of a generic section of the Chan-Paton bundles of the D9-D9-branes wrapping spacetime, X. The D-brane is thus best dened as the some vector bundle whose base space, , is embedded in X. Such objects are classied, by construction, using K 0 () but to address the more general question of what kinds of D-branes can be embedded in X requires understanding how elements of K 0 () can be 0 extended to elements of Krel (X) where the subscript suggests some particular relative K-theory of X.60 This requires understanding how to extend a bundle, W , over to a bundle W X in such a way as to map the K-theories into each other. An analogous question has already been posed in cohomology; namely, given the class H k () of degree k = dim(), what is the extension of to a class on X, H dim(X) (X), that satises =
X

(273)

In Section 2.3.2 the class was given as = () where is the Poincar dual of and : N e is the normal bundle of in X viewed as a tubular neighborhood, N (), of . This construction provides m+codim() a way to extend classes in H m () to globally dened, compactly supported classes in Hc (X). An analogous construction is desired for vector bundles (or K-theory classes) where here what is to be preserved is the topological structure of the bundle on . In fact we will see that such a construction is possible and that the D-brane charge formula, eqn. (195), has a very natural interpretation in terms of this construction. 3.5.2 Line Bundles and Codimension Two

The program outlined above will rst be undertaken in the codimension two case (D7-branes) as presented in [Wit2] and then extended to the more general case. The rational is as before; in this case cohomology classes (namely the rst Chern class) provide a complete topological invariant for line bundles whose zero-loci is thus always of codimension two. Higher rank (and hence higher codimension) bundles require considering higher Chern classes which are more complicated to deal with so K-theory provides a more natural formulation. The reason that we are concerned with zero-loci and their cohomological classication is that Dp-branes are to be constructed as topological solitons in systems with multiple D9-D9-pairs. The Poincar dual form, , to the Dp-brane world-volume must coincide with the Euler e class of some bundle, W V , of which the tachyon is a section which vanishes on the brane world-volume. In codimension two the direct cohomological construction of line bundles from the Poincar dual of the e D7-brane works and provides a nice example where things can be dened explicitly. The construction in the codimension two case is equivalent to the more general one using K-theory but here we will formulate it using cohomology to provide some intuition for the more technical construction to follow. 2 Let X be of codimension two and let HdR (X) be its Poincar dual. A D7-brane on can e now be constructed by dening two line bundles, L and K, over X corresponding to the Chan-Paton bundles of a D9- and a D9-brane respectively. For deniteness, x c1 (K) = 0 so K is the trivial bundle over X and then dene L via c1 (L) = . The tachyon eld T is a section of L K which can be dened as follows. Recall Proposition 3.1, that the Euler class, e(L) = c1 (L) = , is Poincar dual to the zeroe locus of a generic section of L. Thus a section s : X L must be zero-valued on (or some homologous
60 In fact, K 0 () will not, in any way, involve the normal bundle of since this is only dened with respect to X so somehow the embedding of K 0 () into a relative K-theory of X needs to provide this dependence.

94

3.5 D-branes in General Space-Times

3 D-BRANES AND K-THEORY

cycle) and we can normalize it to |s| = |Tvev | everywhere outside of some tubular neighborhood of . Now dene the tachyon elds ground state conguration to be T = s 1 where 1 is the unit section of the trivial bundle K. Since s vanishes on so does T . Here T denes a classical eld conguration around which the quantum eld theory will be expanded. Thus the zero-locus of T at implies that the eld has not assumed its VEV there and, as has been argued above, this can be interpreted, speculatively, as the existence of a D7-brane wrapping (see Fig. 10). To establish this more rmly recall the arguments given in Section 3.3 that, on X N (), where |T | = |Tvev | the net energy density is equivalent to the closed string vacuum so there are no D-branes. Recall, also, that the RR charge of the system is given by 1 ch(L) ch(K) = ec1 (L) ec1 (K) = c1 (L)(1 + c1 (L) + . . . ) (274) 2 The Localization Principle can be used to restrict the support of c1 (L) (seen as the Poincar dual of e ), and hence of the entire expression above, to N (). Here N () is the tubular neighborhood of dieomorphic to its normal bundle in X (this is the neighborhood over which s smoothly increases from 0 to its value away from ). Thus the net RR-charge of the system is localized to the world-volume of the D7-brane as is the energy density. Applying (274) to the general charge formula eqn. (195) or, more specically, to eqn. (199) shows that this system has only D7- and lower brane charge (i.e. it is only charged electrically with respect to the elds G(i) for i < 9) so in all respects this looks like a single D7-brane. Eqn. (274) also clearly supports lower-dimensional brane charge from terms such as 1 c1 (L) c1 (L), 2 which is a 4-form, but this can be recast as which is Poincar dual [BT, 6] [CY] to the transversal e self-intersection of . The anomaly canceling argument in Section 2 has already suggested that D-branes carry lower dimensional RR-charge on their intersections (including self-intersections which correspond to a non-trivial normal bundle and hence an R-symmetry anomaly that must be canceled) but it does not seem possible to ever interpret these as lower dimensional branes (unlike lower-dimensional charge associated with the Chan-Paton bundle). This is because such terms are intimately tied to the inclusion of in X and so would be identical for a corresponding anti-brane wrapping . Hence it is not possible to wrap an anti-brane on with a dierent set of lower-dimensional charges associated to non-trivial normal bundle topology in the hope of having some net lower-dimensional brane charge (since the normal bundle of the brane and the anti-brane are always isomorphic). Fibre of L K

Tachyon Section

PSfrag replacements

Figure 10: The modulus of a section, T = s 1, of L K with a co-dimension two vortex providing the classical ground state conguration about which the quantum theory is dened. The phase of T will have winding about this zero locus. In the construction above it is easy to see that a non-zero section T (L K ) denes an element of Hom(K, L) and that in fact it is invertible so provides an isomorphism of vector bundles. Restricting these bundles to Y = X N () where T is constant and non-zero shows that L|Y K|Y . The K-theory =

95

3.5 D-branes in General Space-Times

3 D-BRANES AND K-THEORY

class associated with this conguration, [L, K], is thus an element of K 0 (X, Y ) since the bundles L and K are isomorphic on Y . 3.5.3 Incorporating CP-Bundles and Lower-Dimensional Brane Charge

More generally might support a a possibly non-trivial, rank n gauge bundle W (corresponding to multiple D7-branes wrapping ). Such a conguration can carry lower-dimensional brane charge associated with topological non-triviality of W and it should be possible to construct an element of K 0 (X, Y ) corresponding to these congurations. Assuming, for the moment, that there exists a bundle W X which extends W over X, in a way that will be made more precise below, and such that ch(W) = ch(W), then such a D7-brane system can be dened as the decay product of n pairs of D9D9-branes with gauge bundles L W and K W (respectively) where L and K are as dened above. The extended tachyon eld in such a conguration would be a section of L W K W so all the arguments given above still apply, suggesting that T condenses away from N (). 61 The RR-charge of the system is given by 1 ch(L W) ch(K W) = ch(L) ch(K) ch(W) = c1 (L) (1 + c1 (L) + . . . )ch(W) 2

(275)

Once more, localization (of the support of c1 (L) = ) will restrict the support of this charge to a neighborhood of and so the system looks like a D7-brane wrapped on with RR charge ch( W) = ch(W). The rst term (in the last expression) in (275), c1 (L) ch(W), is equivalent to ch(W) which is the charge, from (195) or (199), for a D7-brane with world-volume and CP-bundle W. Thus, in this more general case, the K-theory class dening the D7-brane is [L W, K W] K 0 (X, Y ). The explicit construction of W from W is simple in this case (in the more general setting with higher codimension a construction based on the Thom isomorphism for K-theory must be used). First, W is pulled back to a bundle over N (), W (W) N (), via the projection, : N () . This gives two bundles L W and K W (where it is understood that L and K here have been restricted to N ()) dened on N () which might be extendable to X. Even if this is not the case it is possible to consider a modication that will still result in the same nal D7-brane conguration. Specically, invoking Theorem 3.8, consider the bundle Z = (L W) dened so that (L W) Z is isomorphic to a trivial bundle over N (). Then, since Y includes the boundary of the closure, N (), and since L|Y K|Y , the isomorphism restricts to the boundary N () = N () N () so = (L W Z)|N () (K W Z)|N () = (276)

Although W has only been dened on N () it extends uniquely to N () (because any continuous function, including transition functions, extend uniquely onto closures). This isomorphism implies that both bundles are trivial on N () so both can be extended to all of X (simply by dening the extension to be the trivial bundle of the same rank on Y , which obviously agrees on N ()). Thus the bundles L W Z and K W Z are well dened bundles on all of X and can be used to dene a conguration of D9-D9-pairs which will decay into the D7-brane conguration W . This follows because, once more, the pairs are isomorphic on Y so the bundle [L W Z] [K W Z] , of which T is a section, admits a constant non-vanishing section on Y (indicating tachyon condensation in this region) and the net charge of the D9-D9 system is given by 1 ch(L W Z) ch(K W Z) = c1 (L) (1 + c1 (L) + . . . )ch(W) (277) 2 So the nal decay product looks exactly the same as that of the simpler case where W extended directly to a bundle on all of X. The only dierence is that here additional pairs of isomorphic yet topologically non-trivial pairs of D9-D9-branes (corresponding to Z) had to be added to dene the conguration; because the net charge of these additional branes is zero this addition of virtual branes violates no
f construction W is only dened on N () and is extended trivially o it (see discussion later). Hence away from f f N () the factors W and W in L W K W are trivial.
61 By

96

3.5 D-branes in General Space-Times

3 D-BRANES AND K-THEORY

conservation law. Mathematically this statement follows from the relation [L W, K W] = [L W Z, K W Z] and the fact that RR-charge is measured by the Chern homomorphism image of K-theory classes. The relationship ch(W) = ch(W) and similar equalities for any characteristic classes follows from the construction of W via the pull-back, , which is an isomorphism on cohomology. Thus, for any codimension two brane in any spacetime and with any embedding topology, this gives a well dened prescription to associate to it a unique element, [L W, K W] in this case, of K 0 (X, Y ) (where Y is dened relative to the D-brane worldvolume). Note that in (275), L and W, play very dierent roles. The former denes the world-volume of the D7-brane via its zero locus and its rst Chern class acts as the Poincar dual of the brane. The e latter, on the other hand, can only have topological defects on (since it is trivial o and must be isomorphic to a bundle dened on ) and these defects can conceivably develop into lower-dimensional branes (for instance by placing a sucient number of anti-branes on with trivial CP bundles). If all lower-dimensional branes occurred like this it would be possible to simply use this construction iteratively. To dene a Dp-brane with p + 1 dimensional worldvolume p we could apply the above prescription to generate a set of D(p + 2)-D(p + 2)-pairs wrapping some cycle p+2 (nothing in the above construction constrained it to D7-branes dened via D9-D9-pairs; any codimension two conguration will work) such that p p+2 and repeat iteratively until p+2k = X for some iteration k. This is not possible, however, for a general p since the chain of embeddings p p+2 . . . X is not always guaranteed to exist. Thus a more general construction must be invoked that works for any codimension. It will turn out that the codimension two case is a special case of this more general construction. 3.5.4 The Thom Isomorphism in K-Theory

In codimension two we used the Poincar dual to to construct, explicitly, a line bundle, L, with e sections vanishing on which could be tensored with an arbitrary bundle, W, so that the product could be extended all over X to dene a conguration of D9-D9-branes with the same RR-charges and energy density as a D7-brane on with CP bundle W. For higher codimension it is not generally possible to dene the analog of L using only the Poincar dual of . Rather, a construction from K-theory, the e K-theoretic Thom isomorphism [Kar], [OS], [LM], can be used to directly construct a K-theory class (corresponding to a pair of bundles) that plays the same role as the Thom class in cohomology (recall, from Proposition 2.4, that the Poincar dual is the Thom class of the normal bundle of a submanifold). e That is, a pair of bundles, dened on the normal bundle, N , whose K-theory class has compact support in the vertical direction (i.e. the pair are isomorphic outside of some compact subspace of the bundle) 0 0 and which is also a generator of the compact K-theory of the bundle, Kcpt (N ). By Kcpt (N ) we 0 essentially mean the relative K-group K (B(N ), S(N )) where B(N ) and S(N ) are the ball and sphere bundles associated to N (see below). The fact that this class is a generator of the K-group is analogous to the fact that the cohomological Thom class is the generator of the compactly supported (in the vertical direction) cohomology of a vector bundle since it is a bump function which can be integrated vertically (along the bers) to give the value one (see Section 2.6.3). The construction of the K-theoretic Thom class, as this class is known, is rather technical and has been deferred to Appendix B.1 (also see [LM], [Kar] for more details and proofs). We highlight the essential denitions here. Consider a real vector bundle : N (which will correspond to the normal bundle in our construction) where N is of dimension 2k and admits a spin-structure (this, and the generalization to a Spinc -structure, are discussed in the Appendices). Associated to N are two bundles, B(N ) N and S(N ) N , which are the ball and sphere bundle, respectively, whose bers are the 2k-dimensional ball and and the 2k 1-dimensional sphere, B 2k , and S 2k1 = B 2k . Note that the former is homotopic to the base space (and to N itself) since each ber is contractable so there is an induced bijection V ect() V ect(B(N )) ( V ect(N )) dened by the pull-back via : B(N ) ( is the restriction of = = : N to B(N ) N ; we will be sloppy and denote them both using the same symbol). In Appendix B.1 the Thom class in K-theory for the bundle N is constructed in terms of the complex Spin-bundles, S (N ), associated to N . These are pulled back to B(N ) via : B(N ) and a bundle morphism, : (S + (N )) (S (N )), is dened that restricts on the boundary, S(N ) = B(N ), to an isomorphism which is homotopically non-trivial. This denes a class, U = [ (S + (N )), (S (N )); ] K 0 (B(N ), S(N )) 97 (278)

3.5 D-branes in General Space-Times

3 D-BRANES AND K-THEORY

which is trivial outside of a compact subspace of each ber of N . Assuming is, itself, compact then U 0 corresponds to a class in Kcpt (N ).62 If is not compact the class U is not well-dened but can still be used to give a well-dened product (281) below (see [Kar, IV.5] or [LM, App. C]). We denote this class 0 UN = [ (S + (N )), (S (N ))] Kcpt (N ) (279)

The notation here is somewhat sloppy since we have dropped the dependence of the class on the the bundle morphism (because classes in compact K-theory are given by pairs of bundles, not a pair and a morphism). In relative K-theory there might be several dierent classes associated to the pair (S + (N )) and (S (N )) but since we know we are referring to the Thom class which is dened with a particular morphism (dened to have winding number one on the boundary, S(N )) we will just be lax about this. More details can be found in the Appendix B.1. 0 In the sequel we will discuss only Kcpt (N ) and K 0 () whether is compact or not assuming, in all cases, that compactly supported K-theory is used ( is always locally compact). The relevance of U N is that it can be shown to be a generator of the group K 0 (N ) [Kar, IV]. Theorem 3.9 (Thom Isomorphism (in K-theory)). For a real 2k-dimensional bundle N , with 0 a Spinc -structure, the product with UN Kcpt (N ) induces an isomorphism K 0 () Kcpt (N ) = 0 Moreover, the group
0 Kcpt (N )

(280)

is a free K ()-module

63

with generator UN .

More concretely this means that any element, K 0 (N ), can be written as = ()


0 0

UN
0

(281)

for some K (). Since K (N ) is a free module (of rank 1) over K () the two groups are isomorphic K 0 (N ) K 0 (). The isomorphism maps any element of K 0 () to an element of K 0 (N ) by the pull-back = of the projection, , followed by multiplication with UN as given above. To make sense of this statement it is rst necessary to understand what it means that K 0 (N ) is a 0 K ()-module. Recall that a module is an Abelian group admitting an action given by a homomorphism from K 0 () K 0 (N ) to K 0 (N ). For example, for a Lie group, G, with a representation, V , there is a map (g, v) g v V . In this case the homomorphism is given by composing the maps K 0 () K 0 (N ) K 0 ( N ) K 0 ( N ) K 0 (N ) The second map is simply the pull-back of the proper map id : N N while the rst map, , is much more involved. Given = [W, V; ] K 0 () where is an isomorphism outside a compact neighborhood (if is not compact) and UN as dened above64 the product, , is given by [LM] = UN [v1 , v2 ; ] v1 = (W v2 = (V (S + )) (V (S )) (S )) 1 (283)
(id)

(282)

(S + ) (W = 1

1 1

is topologically the one-point compactication of N . module is essentially a representation of a ring, R. A free R-module, F , is a module that is isomorphic to a number L of copies of the ring so F N R. See also discussion below. = i=0 64 The product, , is a well dened product for any two spaces, and N , and any of their K-theory classes, not only the Thom class, UN , of a bundle, N . We use this particular example here because it will be useful later in the discussion. Hopefully this will not cause confusion.
63 A

62 B(N )/S(N )

98

3.5 D-branes in General Space-Times

3 D-BRANES AND K-THEORY

Note here the pretense of (S ) instead of just S is an artifact of how UN was dened and is not related to the denition of the product . Above and in the sequel we will be somewhat sloppy in dierentiating between the bundle S (N ) and its bers, S , since the context will generally make it clear which is intended. The symbol denotes the outer tensor product W (S + ) (1 (W)) (2 ( (S + ))) where 1 : N and 2 : N N are projections onto factors. Thus the statement of the Thom isomorphism for K-theory is that, under the action of K 0 () on K 0 (N ) just dened, K 0 (N ) is a free module with its K-theoretic Thom class, UN , as the generator. 3.5.5 Examples and Remarks

To gain an intuitive sense for this rather technical construction let us see exactly what this means for a vector bundle W . This vector bundle can be used to dene an element of K 0 (), namely [W, I n ] where I n is the trivial rank n bundle over and n is the rank of W. Now consider the image of this bundle under the isomorphism, K 0 () Kcpt (N ) described above = 0 [W, I n ] ( (W) (S + )) ( (I n ) (S )),
0 (S + )) Kcpt (N )

( (W)

(S )) ( (I n )

The image of [W, I n ] is more cleanly represented as (W) (S + ), (W) (S ) + (I n ) (S ), (I n ) (S + ) (284) So the relative bundle W has been extended to a bundle over N with compact support by tensoring it with both (S + ) and (S ) which are isomorphic to each other outside of a compact neighborhood of N homotopic to (i.e. the origin in each ber). That is to say, both classes above are trivial outside of some compact region corresponding to the region where : (S + ) (S ) is an isomorphism. This is exactly the point of the Thom isomorphism. Namely to extend a bundle W on represented by a K-theory class [W, I n ] to a bundle on N which has compact support in the vertical direction. Notice that one might also consider a more trivial isomorphism given by the map (W, I n ) ( (W), (I n )) via the pull-back of the projection (that this is a bijection follows from the fact that the total space of N is homotopic to ) but these bundles no longer have compact support in N so will not extend to global elements in X. It is also instructive to consider the relationship between the K-theoretic Thom isomorphism and the cohomological one. In both cases an object dened on a manifold, , is lifted to an object dened on a bundle N over that manifold; since, in both cases, the objects are sensitive only to the homotopy type of the base space and since any vector bundle, N , is homotopic to this lifting is an isomorphism. In cohomology the Thom class, (N ), of the bundle is a form with compact support in the vertical direction and is the image of 1 H 0 () under the cohomological Thom isomorphism. Thus, under the isomorphism, (N ) restricts to the generator of compactly supported cohomology on the ber. Likewise, the K-theoretic Thom class dened above, UN , is shown [LM] to be the generator of the compactly supported K-theory of the bers of N .65 Let us consider the image of the Thom class, UN , under the 0 Chern homomorphism between Kcpt (N ) and H even (N, Q) ch(UN ) = ch([ (S + (N )), (S (N ))]) ch( (S + (N ))) ch( (S (N ))) = (ch(S + (N )) ch(S (N ))) = e(N ) A(N ) (285)

= (A(N ))1 (s ((N ))) = (A(N ))1 (N ) 0 Where we have recalled that the Euler class is the pullback of the Thom class via the zero section s0 : N and have applied eqn. (78). A slight subtlety that is being neglected in this equation is that is only an isomorphism onto normal cohomology not cohomology with compact vertical support;
65 A quick way to see this is that the compactied bers are topologically S n and the Thom class in K-theory is dened via a pair of bundles, (S (N )) which are isomorphic on the equator, S n1 , via an isomorphism of winding number one. Recall that all vector bundles on S n are classied by their winding number along the equator.

99

3.5 D-branes in General Space-Times

3 D-BRANES AND K-THEORY

thus the class (N ) should be seen as a class in H (N, Q). Thus the image of the K-theoretic Thom class diers from the Cohomological Thom class by the inverse of the roof-genus of the normal bundle. We will show in the next section that this contribution gives a very natural interpretation to the charge formula for a D-brane, (195). Namely, instead of being the push-forward of a cohomology class on associated to W, eqn. (195) should be interpreted as the K-theoretic push forward of the K-theory class associated with W (as in our construction above). Before going on to use the K-theoretic Thom isomorphism to dene D9-D9 congurations, as promised, let us rst make a few more observations about this construction that relate it to things we have already seen. Although the construction of the Thom isomorphism might seem somewhat abstract we have, in fact, already undertaken it by hand for some special cases. The rst is the codimension two construction in the general topological setting discussed in the Section 3.5.2. To understand to relation between the Thom isomorphism and the cohomological construction of a line bundle requires using some results from the representation theory of Cliord Algebras related to the construction of the Thom class (see Appendix B.1). The interested reader should consult Appendix B.2. The construction presented here can also be seen as a general case of the construction used for trivial normal bundles N R102k B 2k in Section 3.4. = There, pairs of bundles where considered that were isomorphic on the boundary N = R 102k S 2k1 . These bundles were xed by their winding number from S 2k1 to U (n) for n the rank of the bundle. Although it is not immediately evident, this does, in fact, coincide with the construction above since is required to be an isomorphism on S(N ) which looks locally like S 2k1 . By construction the isomorphism restricts to the generator of 2k1 (U (n)) for n = 2k1 . 3.5.6 RR Field Equations and the Thom Isomorphism

Although we have already spent considerable eort developing the Thom isomorphism for K-theory, and have connected it to some earlier constructions, all we have, by way of motivation, is the geometric intuition that it can be used to extend K-theory classes from to X. Let us now see if this construction can be motivated by appealing directly to the RR eld equations. Starting with the charge formula in Hc (X) dG = i ch(W) A(T ) A(N ) (286)

and pulling it back to using ( ) ((N )) = e(N ) and A( (T X)) = A(T )A(N ) where : X is the embedding, this can be re-written as (d G) = =
i

ch(W)

e(T ) A(N )
+

A( (T X)) (287)

ch(W) (ch(S (N )) ch(S (N )))

A( (T X))

The second line follows from the identity, used in Section 2, that holds for any spin, orientable vector bundle E ch(S + (E)) ch(S (E)) = e(E) A(E) (288)

Recall that the Chern character denes a homomorphism between K-theory and rational ( Cech) cohomology ch : K 0 (X) H even (X, Q) ch([W, V]) ch(W) ch(V)

(289)

100

3.5 D-branes in General Space-Times

3 D-BRANES AND K-THEORY

Using the denitions of the Chern character and K-theory it is easy to see that this is indeed a homomorphism. Moreover, if K 0 (X) is tensored with Q which eectively removes any torsion66 subgroups, then this can be shown to be an isomorphism (see [Kar, Thrm V.3.25] and references therein) ch : K 0 (X) Q H even (X, Q) It is also possible to dene a natural bilinear pairing on each space given by [W, V], [W , V ] , indDWV indDW
X =

(290)

K dR

(291)

Where the index of a dierence bundle, [W][V], is dened to be the dierence of the indices indDWV indDW indDV . indDW is the index of the Dirac operator on X twisted by a bundle W (see [Nak] for this terminology). The Chern character, as given, above does not preserve this bilinear form but it is easy to see that a modied version does [MM] ch : K 0 (X) Q H even (X, Q) ch([W, V]) (ch(W) ch(V)) A(T X)

(292)

Under this modied isomorphism the last line of (287) appears as the image of [ (W) (S + ), (W) (S )] pulled down to . We observe, therefore, that the RR eld equations are themselves very suggestive of the K-theoretic Thom isomorphism. This argument is based primarily on the presence of the term e(N )/A(N ) in eqn (287) and is not so reliant on the (perhaps arbitrary) choice of bilinear pairing dened above (i.e. the modied Chern homomorphism is required to get an exact match with equation (287) but even without it, the presence of the term e(N )/A(N ) is suggestive of the Ktheoretic Thom isomorphism). 3.5.7 Lower Dimensional Branes and the Thom Isomorphism

The techniques introduced in Section 3.5.4 can now be used to generalize the construction of a D7-brane conguration to congurations of lower dimensional branes. It is important in dening this construction to recall that the equivalence between Dp-branes and congurations of D9-D9-brane pairs is based on the fact that the two systems have the same conserved charges, both in terms of the RR-eld strengths and energy density. Thus in specifying a conguration of D9-D9-pairs which we expect to decay into a Dp-brane we are constrained only by the total charges in both systems. Let us now consider a Dp-brane for p odd (since we are in IIB) on a world-volume with a ChanPaton bundle W . This information encodes all the charges of the system since the RR-charge is given by the formula (195) and the energy density is a constant function (whose value is xed by the value of p via the D-brane tension, p ) on the world-volume . Recall, also, that on a D9-D9 system the energy density can be related to the tachyon, seen as a section of tensor product of Chan-Paton bundles, via equation (232) which implies that wherever this section has a value coinciding with the tachyon VEV the local energy density is equivalent to the closed string vacuum. Thus there is a non-vacuum energy density, coinciding with a putative Dp-brane worldvolume, on the zero-locus of the tachyon section. With this introduction we hope to have motivated the following construction. The general program will be, following [Wit2], to dene a Dp-brane wrapping with a CP-bundle W as a K-theory class in K 0 (X) with some compactness condition (in transverse directions) by rst pushing the class of W into a class in K 0 (N ) via the Thom isomorphism and then extending that to a class in K 0 (X) with compact support. This general theme has already been pursued several times, rst for cohomology classes in Section 2, and then for line bundles in the last few subsections. The bundle associated to W in K 0 (X) will be the pullback of W onto the normal bundle tensored with spinor bundles (S ) associated to the
66 A torsion group is a group with a nilpotent generator such as Z . The tensor product, over Z, of a torsion group with 2 Q will always be zero since any element of the tensor product will be equivalent, via the tensor relations, to zero. Consider p p 1 p Zn Z Q for any p, q Z. Then, under the tensor product, this is equal to 1 np = n nq = 0 nq = 0. q nq

101

3.5 D-branes in General Space-Times

3 D-BRANES AND K-THEORY

normal bundle as described in Section 3.5.4. The physical interpretation of this will be that the extended spinor bundles (S ) dened on all of X and viewed as complex, rank n, vector bundles (but also carrying additional algebraic structure) are actually the CP bundles of the D9-D9-pairs. This inverts the usual logic where we dene Chan-Paton bundles associated to D9-D9-pairs and then declare that the zero-locus of a generic section of such a bundle is the world-volume of a lower-dimensional brane. Here the world-volume is already given and, via the Thom isomorphism, it is possible to construct bundles with the correct generic zero-loci. In addition, lower-dimensional brane charge is encoded in W which then multiplies (S ) to dene spacetime lling brane congurations that encode, not only the embedding of , but also the lower dimensional brane charge of its CP-bundles. Let us work this out in detail. Let now be the world-volume of a Dp-brane (p odd), : N its even dimensional normal bundle, and W its Chan-Paton bundle (which may be the trivial rank 1 bundle, I 1 , for a single brane without any lower dimensional brane charge). We assume that is connected (see below) but this is not a serious limitation since the K-theory of disjoint unions is easily calculated from the K-theory of each connected component (one can dene a Dp-brane to be a connected component and then treat a disconnected as multiple Dp-branes). We also assume is compact or, if it is locally compact, we work on its (one-point) compactication so its K-theory is well-dened. Now we want to dene a map V ect() K 0 () by xing a bundle V and dening the map W [W, V]. It is necessary to x a V in order to dene an inclusion of V ect() into K 0 (), its Grothendieck group, and this will have the physical interpretation of dening the zero-charge of Dp-branes. We want to push this K-theory class in K 0 () to a class in K 0 (X, Y ), where the use of relative Ktheory ensures that the image of the class is trivial on the complement of N (), namely Y = X N (). This is done using the K-theoretic Gysin map (see [Kar]) which is essentially the Thom isomorphism, 0 0 used to map K 0 () into Kcpt (N ), followed by an extension of the class Kcpt (N ) to a class in K 0 (X, Y ). This is analogous to the cohomological Gysin map which was used in Section 2 to push forward elements of cohomology from H () to H (X). Note that, to apply the Thom isomorphism, we must assume that N is a spin-bundle. If this is not true a slight modication is required that will be addressed in the next section. Another point to note is that the Gysin map, as dened above, requires that be embedded into X so that there exists tubular neighborhood, N (), dieomorphic to the normal bundle, N . In [Kar, Ch. IV, 5] it is shown, however, that any proper, continuous map : X for which the induced normal bundle has a Spinc -structure can be used to dene a Gysin map, : K 0 () K 0 (X). We will restrict the discussion here to the case of an embedding but using the above mentioned theorem it is clear that this can be generalized to the case where is not a submanifold of X. 0 Explicitly then, the class [W, V] K 0 () is mapped to the class in Kcpt (N ) given by (284) (W) (S + ), (W) (S ) + (V) (S ), (V) (S + ) = (293)

(W) (S + ), (W) (S ) (V) (S + ), (V) (S ) =

By construction, we know that these classes are trivial on the boundary of the tubular neighborhood, N () (since there (S ) are isomorphic), which we will denote N () (W) (S + ) |N () (W) (S ) |N () = (V) (S ) |N () (V) (S + ) |N () =

(294)

We would like to extend these bundles, dened on N () X, to all of X but this may not always be possible for topological reasons. If not, we proceed as in the case of a line bundle. First, we apply Theorem 3.8 to demonstrate the existence of a bundle Z = (W) (S + ) such that for some m (and likewise nd a Z for (V) (S )) and then use (W) (S + ), (W) (S ) = (W) (S + ) Z, (W) (S ) Z (296) ( (W) (S + )) Z I m = (295)

102

3.5 D-branes in General Space-Times

3 D-BRANES AND K-THEORY

because they dene the same K-theory class. Since (W) (S + ) Z is trivial, and it is isomorphic to (W) (S ) Z on the boundary, N (), both bundles can be extended trivially over all of X (by just gluing to the trivial bundle dened on Y = X N ()) giving a class in K 0 (X, Y ). This is exactly the same procedure as undertaken in Section 3.5.2 with respect to line bundles. We now associate to this conguration, W , only the rst class in this sum, (W) (S + ) Z, (W) (S ) Z (297)

This is a consistent choice because we dene the class of the conguration to be its image under the Gysin map minus the class (V) (S + ), (V) (S ) . The choice is consistent because we dened the map from V ect() to K 0 () using the bundle V and it is easy to check that, after subtracting this contribution, the nal class in K 0 (X) corresponding the the Dp-brane conguration no longer depends on the choice of V. Physically one can think of V as a zero-point background charge against which other charges are measured. One also easily checks that, with respect to the Dp-brane charge formula, eqn. (287), this is the correct prescription. In order to complete the construction it remains to dene a conguration of D9-D9-brane pairs from the data we have so far. To do this we will have to use some simple results from the representation theory of Cliord Algebras (see [LM, I.5] and Appendix B). First, note that the bundles S have bers Sx k1 (x ) which are complex vector spaces of complex dimension 2 where the dimension of N is 2k. The transition functions for these bundle are SpinC (2k) GL(2k1 , C) valued (in fact they correspond to the two irreducible representations of SpinC (2k)) and we would like to use these to treat S as 2k1 dimensional complex vector bundles that we can declare to be (part of) the Chan-Paton bundles of the D9-D9-brane pair. Recall that such a bundle must be of the form W V where W and V are equal rank complex vector bundles and here the minimum rank suggested is 2k1 corresponding to the complex dimension of the bers Sx . Observe that such a bundle requires the same number of D9-D9-brane pairs as does the step-wise iterative construction, described at the end of Section 3.3.6, which can be applied when a suitable embedding of cycles exists. To complete the construction above we want to show that the bundles S have unitary transition functions (i.e. valued in U (2k1 )). Normally any complex vector bundle admits a reduction to a unitary structure group but this assumes the freedom to reduce the structure group using any elements of GL(2k1 , C) not only elements of the subgroup SpinC (2k). Thus, in this case, it is necessary to show that the SpinC action is, itself, unitary. This follows from a proposition in [LM, Prop. 5.16] showing that the complex representation of SpinC are unitary (or can be made unitary). Starting with a xed inner product on S it is possible to dene a new one as follows. First, dene a subgroup of the Cliord Algebra C(R2k ) that is nite and then dene a new inner product by averaging over the action of this subgroup. This is a standard construction which can be used to construct an inner product for which a given representation of a compact (or nite) group is unitary. Under this new inner product the representation of the Spin subgroup will also be unitary. Thus the action of the SpinC valued transition functions of S on these bundles can be made unitary which means they are complex vector bundles in the fundamental of U (2k1 ). 3.5.8 Normal Bundles with Spinc Structure

We have thus far assumed that the bundle N admits a spin-structure, meaning that its second StiefelWhitney class vanishes. This was necessary in the Thom isomorphism because it allowed for the construction of Spin-bundles lifted from N which dene the Thom class of the normal bundle. The notion of Spinc has already been mentioned in Section 2.7 and is discussed in some depth in Appendix A so we will be brief here. We will assume the reader is familiar with the content of these sections so the reader who has skipped them should also feel free to skip this section as well as it mostly concerns technical details that do not signicantly alter the argument. Recall that in [FW] it was shown, by analyzing a global anomaly on the string worldsheet theory, that the normal bundle of a D-brane must always admit a Spinc structure if the B-eld has a topologically trivial eld strength (which has been the assumption so far). Thus even if N does not lift to a Spin bundle, meaning that the bundles S do not satisfy the cocycle condition and are not well dened, there exists a line bundle, L, on whose square-root bundle L1/2 fails to satisfy the cocycle condition in the 103

3.6 Conclusions

3 D-BRANES AND K-THEORY

same way as S so that the tensor product bundles, S L1/2 , are well-dened. This can be used in the construction of the Thom class to dene it as [ (S + L1/2 ), (S L1/2 )]. As demonstrated in [FW] and discussed in Section 2.7, the line bundle L1/2 is actually the ChanPaton bundle of a single D-brane and, if N is Spinc but not Spin (i.e. w3 (N ) = 0 but w2 (N ) = 0), then this bundle is not well dened as it does not satisfy the cocycle condition. Even for the case of an Abelian Chan-Paton bundle (a single brane) there is an ambiguity regarding exactly what the denition of the Spinc component is. This was discussed under eqn. (174) and comes from the ability to factor L1/2 into two parts, Z and Z, the rst of which enters the D-brane charge formula, eqn. (185), as the actual gauge bundle (W in that formula) of the D-brane and the second of which gives the Spin c -factor (ed/2 in that formula). It is obvious, however, from the construction of the Thom isomorphism and its relation to Theorem 2.10 (see Section 2.8) that this splitting has no eect on the resultant D-brane charge or K-theory class. Consider the image of the CP-bundle L1/2 under the Thom isomorphism where L1/2 Z Z and Z has a half-integer Chern class and so denes a Spinc bundle. Then the Thom class = would be given by [ (S + Z), (S Z)] and the Chan-Paton bundle is considered the well-dened factor, Z (i.e. the factor with an integer Chern class). The image of Z (or [Z, V]) under the Thom isomorphism is [ (Z S + Z), (Z S Z)] which is just [ (L1/2 S + ), (L1/2 S )] (299) (298)

and so is entirely independent of the factorization, L1/2 Z Z, of the Chan-Paton bundle into a = gauge and a Spinc -bundle. The use of Spinc -bundles instead of Spin bundles does not aect the rest of the construction of the Thom isomorphism or its use in dening D9-D9-congurations corresponding to a D-brane on a worldvolume with Chan-Paton bundle W. The arguments of [FW] only strictly apply in the Abelian case where the Chan-Paton bundle is a line bundle but we will not attempt to deal rigorously with the more general case until the next section. As noted in Section 2.7 the more general analysis of Spinc suggests that the bundle W is a twisted vector bundle that does not satisfy the cocycle condition by a Z 2 factor (which is also true of L1/2 above). Thus one might imagine a simple extension of the arguments above to the more general case where the twisted line bundle, L1/2 , is given by the determinant bundle L1/2 det(W).

3.6

Conclusions

Our analysis has provided an elegant geometric understanding of exactly why D-branes are classied by K-theory, at least in the simplied setting to which weve restricted ourself (no background B eld). Of course the fact that D-brane charges are classied by K-theory was, in some sense, already evident at the end of Section 2. It followed from the identication of D-brane charge as an element of rational K-theory via the modied Chern isomorphism (292). In geometrising the analysis we have gained more than just elegance. The identication of D-brane charge with rational K-theory is insensitive to torsion classes since K 0 (X) must be tensored with Q. By making a direct connection between D-branes and K-theory using Sens construction it is clear that the actual D-brane charge is given by the full K-theory class, not just its image in cohomology. This of course brings up the issue of when the two dier and of how much relevance the dierence between rational cohomology and K-theory is for our physical understanding. There is a nice discussion of the sorts of situations in which the exact K-theoretic understanding is important in [MMS]. Specically, they note there that it is possible to have a brane wrapping a homologically stable cycle, , yet be unstable because of K-theoretic considerations. The most straightforward consequence of this analysis is simply that, to determine the stability of a congurations of D-branes, one must consider the class of the branes in the correct K-groups. More interestingly it turns out that the RR eld-strengths are, themselves, also classied by K-theory [MW]. This is not so surprising since they are generated by sources classied by K-theory but it also turns out to be very natural in dealing with the fact that they are self-dual. This has led to ongoing work

104

3.6 Conclusions

3 D-BRANES AND K-THEORY

on quantizing theories with such objects, a problem which is sensitive to their geometric nature. This problem is also being studied in M-theory [DMW] where it is being used to better understand the theory. In the next section we will nally lift the constraint B = 0 and suggest how this eects the analysis and introduces complications whose resolution is not entirely clear (as far as the author is aware). It is these open problems that make this eld tantalizing as they suggest that their resolution might provide more insights into a non-perturbative formulation of string theory or might provide some more insight into the low-energy approximation to M-theory.

105

4 THE B-FIELD AND TWISTED VECTOR BUNDLES

The B-eld and Twisted Vector Bundles

So far the discussion in this thesis has been limited to a very special regime in string theory where we have taken the limit going to zero and set the value of the background antisymmetric tensor eld, B , to zero. Ultimately, we would like to extend our results to a more general setting. In the particular case where B = 0 and the cohomology class of its eld-strength, [H], is a torsion class this can be done [Kap] [Wit2] and suggests that D-branes should, more generally, be classied using twisted K-theory [DK]. The case of a non-torsion [H] is more speculative [BM] [BCM+ ] and suggests a possible generalization to (C -)algebraic or other kinds of K-theory. Even more generally, the idea that there is a generalization of K-theory that applies when the point-limit ( 0) of strings is not taken is an interesting one that has only been recently studied [TS] [Sch]. Unfortunately, constraints of time and space do not allow us to develop the notions of twisted Ktheory (or its variants) in this thesis. Rather we will focus on the physical derivation of the twisted cocycle condition of the CP bundles when H is non-trivial. This twisting implies the analysis of Sections 2 and 3 must be modied and ultimately leads to some of the open problems in the eld. We oer this section as a brief enticements and encourage the interested reader to further pursue the references starting with [BM] [BCM+ ] [Kap] [AS2] [CJM].

4.1

Overview

In this section we will determine how the analysis of Sections 2 and 3 must be modied in the presence of a non-trivial background B-eld with a torsion class [H] Tor(H 3 (X, Z)). Recall, from Section 2.4.3, that open string modes for strings ending on a D-brane with world-volume, , were shown to be sections of a vector bundle V adj in the adjoint representation67 of U (N ). In that section, we claimed that, when B = 0 or when it is topologically trivial, the bundle V adj was equal to W W where W is a vector bundle in a fundamental representation of U (N ). We also claimed that on the D-brane world-volume there is a eld, A , generated by the open string modes, that acts as a connection on the Chan-Paton bundle, W dened in Section 2.4.3. To understand how B and A are related and when V adj factors to well-dened bundles, W and W, requires analyzing the original string world-sheet theory given by the action (5). This action is modied, for an open string, by including the holonomy of a background A eld [Pol1]. We write the path integral measure schematically as W [] = F [] pfa(iD ) exp i (A) exp i (B)
M

(300)

where : M X is map of the string world-sheet, M , into spacetime, X. W [] is the exponentiated form of the action so the path integral would be given by integrating W [] over the space, M = M ap(M, X), of inequivalent maps.68 The factors in its denition will be explained in due course. In general, holonomies of gauge elds, such as A and B, are well-dened (gauge-invariant) quantities so long as they are taken over spaces with no boundaries [Alv]. For closed strings the world-sheet, M , has no boundary so the term HolM ( A) vanishes and the term HolM ( B) is well-dened. For open strings, however, HolM ( A) does not vanish but is a well dened quantity while HolM ( B) is no longer gauge invariant [GR][Gaw]. Since HolM ( B) is a factor in W [] the latter also becomes illdened. We will suggest, below, how to make this vague notion more rigorous. Essentially, if we attempt to consider HolM ( B) as a function on the space of maps, M, we nd that it depends upon a choice of coordinates in this space and, as such, is not really a well-dened function but rather a section of a line bundle, LB M. This implies that, in this case, W [] also becomes a section of a line bundle on M and hence can only be integrated if that line bundle is trivial. It is this last observation that introduced a dependence of the gauge eld, A, on the background Beld. To make W [] a well-dened functional we must modify A in such a way that the factor Hol M ( A), rather than being a well-dene functional on the space M, becomes a section of a non-trivial line bundle, LA , on M. This second line bundle is chosen in such a way that the tensor product LA LB , of which
some variation of V adj in the presence of other branes or anti-branes. for now the need to include the world-sheet metric in the moduli space. The factors that will concern us are, in any case, independent of the world-sheet metric as they are holonomies of forms.
68 Neglecting 67 Or

106

4.1 Overview

4 THE B-FIELD AND TWISTED VECTOR BUNDLES

W [] is a section, is a trivial bundle. This will require modifying the cocycle condition on W , the Chan-Paton bundle on the D-brane, on which A is a connection to be g g g = eiK (301)

where eiK is an element of H 2 (, Cont(U (1))) related to the (restriction to of the) class of [H] 3 (, Z) in such a way that, as remarked in Section 2.4.3, it will only be possible to set eiK = 1 when H the class of [H] is trivial. The argument given above only holds when the class [H], restricted to , is a torsion class, [H] 3 Tor(H 3 (, Z)) meaning that its image in de Rham cohomology, HdR () H 3 (, R), vanishes. We will = show that the torsion degree of [H] is related to the rank of W, the putative vector bundle, on which we attempted to dene the connection, A. If this bundle is of rank n then n[H] = 0 so [H] must be of degree m where m is a divisor of n. The failure of W [] to be a well-dened function on M would generate a global or non-perturbative anomaly in the string world-sheet theory known as the Freed-Witten anomaly [FW]. A similar problem, resulting from the failure of the factor pfa(iD ) to be a well-dened function was studied in [FW] and led to the conclusion that, if B is topologically trivial, must be a Spinc manifold in order to cancel the anomaly (see Section 2.7 and Appendix A.2). We will discuss this briey below. The physical input, in terms of the Freed-Witten anomaly, implies that the bundles W, that formed the basis of our discussion in Sections 2 and 3, are no longer well-dened vector bundles and thus cannot be classied using the techniques of K-theory so far introduced. Several mathematical alternatives have been introduced 1. Azumaya algebras, their modules, and twisted K-theory [Wit2] [Kap]. 2. Bundle gerbes, bundle gerbe modules and their K-theory [BCM+ ]. 3. P U (H)-bundles and their K-theory [AS2], In all these cases the relevant twisted K-theory can be shown to coincide when [H] is a torsion class. It is shown in [AS2] that there is a bijection between P U (H)-bundles (bundles with structure group, P U (H), the projective unitary group on some Hilbert space, H) and H 3 (, Z) which provides a very natural generalization, in mathematical terms, of the torsion construction to the non-torsion case. It is not clear physically, however, what such a generalization would correspond to. Let us briey recapitulate the physically relevant points 1. If we attempt to factor the vector bundle bundle, V adj on , in the adjoint of U (N ) U (N ) into a product of vector bundles, W W, in the fundamental of U (N ) the bundle W turns out not to be well-dened when the class of H = dB restricted to H 3 (, Z) is non-trivial. 2. This is not a problem in terms of a global denition of the string modes or the low-energy elds they generate because these are all sections of V adj not of the ill-dened bundles W. 3. There is a restriction, however, coming from the global anomaly, on the rank of W. This rank, n, must be a multiple of the degree, m, of [H] in Tor(H 3 (, Z)). Physically, this means we can only wrap D-branes on in multiples of m (if we want them to be stable). As has been mentioned before, what happens in the limit when m, the degree of [H], is sent to innity (corresponding to the non-torsion case) is not clear. Naively, it would seem like we would need an innite number of D-branes wrapping in which case it is non-trivial to extend many parts of the analysis below (for instance the holonomy of the gauge eld, A, on the CP bundle). It should be clear that certain parts of the argument to be presented below depend crucially either on having a nite m or on having [H]dR = 0 (both of which imply torsion). A possible formulation of the m going to innity limit is the subject of on-going research [BM] [BCM+ ][CJM].

107

4.2 The Freed-Witten Anomaly and HolonomyTHE B-FIELD AND TWISTED VECTOR BUNDLES 4

4.2

The Freed-Witten Anomaly and Holonomy

The main physical input into the relation between D-branes and twisted K-theory is the Freed-Witten anomaly which relates the global geometric structure of the gauge-eld on the D-brane with the restriction of the anti-symmetric bulk tensor to the brane world-volume. To study this we will rst introduce this relationship naively, as it was rst discussed in the literature [Pol1] [Pol2], so that the reader can have some understanding of its nature before too much detail is developed. We will then more carefully review how the surface holonomy of B on the string world-sheet requires, in order for the world-sheet path integral to be well-dened, a modication of the gauge transformations of A and what this implies for the geometry of the CP-bundles. Most of the discussion in this section is a review of [Kap]. Rather than work in a general setting, as done there, here we will use a specic example to make the presentation more accessible. 4.2.1 The B-eld

Let us rst focus on the local physics of the B-eld without introducing too many topological complications. The string state generating the B-eld is the anti-symmetric component of the massless state in the closed NS-NS sector. This excitation is generated by two world-sheet oscillators and takes the form M (k) 1/2 1/2 |0, k
N SN S

(302)

M is a general matrix whose symmetric components were identied with the graviton, g , and whose anti-symmetric components dene B . A discussion of how a coherent state generated by such modes can be related to a spacetime background eld with the same tensor structure and how the equations of motion69 for this eld can subsequently be determined can be found in Section 2.2 and [Pol1]. Note that, as it is a closed string state, it lives in the bulk of spacetime, X, and is not restricted to the world-volume of any D-brane. As an anti-symmetric rank two tensor, B is a two-form and from the string action, eqn. (5), in the presence of a background B-eld one can see that the B-eld is coupled minimally to the string worldsheet, exactly like a one-form gauge eld in electromagnetism couples to the world-line of an electron Spolyakov B
M

(303)

For a closed world-sheet, M , without boundary Stokes theorem implies an invariance of the theory under the transformation B B + d for any one-form, . Hence, in this theory, the background B-eld is not a physically well-dened eld, but is only dened up to a gauge-transformation (a total derivative). The three-form eld, H [ B] , which shows up as the eld-strength associated to B in the low-energy string action, is a globally well-dened three-form. It is closed because dH = d 2 B = 0 but it need not be exact because B is not a well-dened global two-form. This means that H is an element of 3 the de Rham cohomology, HdR (X), of spacetime. A standard argument for eld-strengths in quantized theories [Alv], indicates that it can be normal ized to have integer periods and hence is actually an element of H 3 (X, Z) which can be mapped into 3 (X, R) H 3 (X) (see also Section 3.2.2). The essence of the argument is that, when calculating the H = dR holonomy of B on a closed surface, M , there is an ambiguity given by the class of H in H 3 (X, R). As this ambiguity occurs in an exponential (in the holonomy) it has no eect so long as the class is an integer multiple of 2 so we must enforce this condition on H. Recall that the holonomy of B occurs in the string world-sheet path integral so it must be well-dened. So far we have described only closed world-sheets, M . If we now generalize to open world-sheets so M has a boundary then we must include a boundary term in the action allowing the one-form gauge eld, A , generated by the open string modes, to couple minimally to the boundary, M , of M . The origin of this coupling is similar to the other terms in the action (5). It can be seen to be generated by including, in all correlation functions, a coherent state of massless open-string modes that generate A , modeling
69 The

Einstein equation in the case of the graviton and a generalized Maxwell equation in the case of B .

108

4.2 The Freed-Witten Anomaly and HolonomyTHE B-FIELD AND TWISTED VECTOR BUNDLES 4

a background eld that the string can interact with [Pol1]. For simplicity, let us rst restrict to A an Abelian gauge eld. Then eqn. (303) becomes Spolyakov B+
M M

(304)

Note that this action has an invariance under A A + d for any function because M is a closed loop (or disjoint union thereof). The action (304) is not invariant, however, under B B + d because M has a boundary and this will generate a boundary term. This lack of invariance is inconsistent. All open string theories contain closed strings and in the closed string theories the string modes generating B have unphysical degrees of freedom associated to this gauge invariance which would suddenly become physically meaningful in the presence of open strings. Thus, to restore gauge invariance, even in the presence of open strings, we postulate the following, coupled, gauge transformation rules B B + d A A + d (305) (306)

It is easy to check that eqn. (304) is invariant under these combined transformations. We will revisit this argument shortly and show that if we posit such a gauge transformation we can use it to cancel an anomaly on the string worldsheet theory. If one neglects the B-eld then there is a gauge-invariant two-form, F = dA, that is the eld-strength associated with the gauge eld A. The modied gauge transformation rule (306) implies that this is no longer the case in the presence of a B-eld since then, under a gauge transformation of B by , F must now transform as F F d. This has the interesting consequence that now there is a globally well-dened two-form given by B + F since their respective gauge transformations cancel. One might imagine that, as it admits a local gauge invariance, B can be seen as a connection on some bundle on X. This is not the case, however, since its eld strength, H H 3 (X, Z), implies a constraint on quadruple intersections whereas bundles only require constraints on triple intersections [Hit]. This can be overcome in some sense by lifting B to the loop space, LX, of spacetime, X. Each point in this space corresponds to a loop in X so, heuristically, by integrating B over a loop, , in X we lower its degree by one and so dene a one form over the point, LX. This turns out to dene B as a connection on a line bundle of the space LX. On X, B does have a simple geometric meaning. Rather it is associated to a Gerbe [Hit] [Bry] [Moe] but we will not make an attempt to develop this notion here though it is central to many of the arguments found in [BM] [BCM+ ] [CJM]. 4.2.2 Geometry of A and B for Torsion H

To be able to work with A and B more exactly let us attempt to determine their geometric nature taking global topology into account. Let us start by considering the B-eld and the A-eld independently from each other. We will start by assuming that A is a connection on a well-dened bundle in the fundamental representation of U (N ). That is to say, we assume that the bundle V adj of Section 2.4.3 factors as well dened bundles, W W, and A is a connection on W. This is implicit, for instance, in eqn. (309) and eqn. (315) below. Even though, as suggested in Section 2.4.3, these bundles are ill-dened for non-trivial [H], we will be able to start with this ansatz and modify it to dene connections on V adj later on. On a good open cover of X (which restricts to a cover on ) one checks that the A and B elds must satisfy the following set of equations F = dA (A) A A = i df (f ) f + f f = 2 and (307) (308) (309)

109

4.2 The Freed-Witten Anomaly and HolonomyTHE B-FIELD AND TWISTED VECTOR BUNDLES 4

() + = dK (K) K + K K + K = 2 m

H = dB (B) B B = d

(310) (311) (312) (313)

Eqns. (312) and (313) can be derived by considering multiple intersections and recalling that, on a good open cover, all such intersections are contractable and hence all closed forms are exact. Recall, from Section 3.2.2, that eqns. (307)-(309) can be organized into a Cech-de Rham complex as in 3.2.2 showing 2 (, R) is the Cech representative of F . Likewise it is possible to t eqns (310)-(313) to that H 3 show that m is the Cech representative of H HdR (X). By considering a line bundle with connection F we showed, in that section, that F has normalized integer periods so H 2 (, Z); a related analysis 3 (X, Z). in [Alv] (using holonomies rather than line bundles) shows that m H So far we have only discussed Abelian gauge elds, corresponding to a single D-brane. Below, we will be working in the more general setting where the gauge eld, A , will become non-Abelian. Let us recast eqns. (307)-(309) in a form that will be more convenient in this case F = dA + A A + (314) (315) (316)

A =

1 g A g

1 ig dg

g g g = 1

In the non-Abelian case it is more convenient to work directly with the transition functions, g = eif . Recall that these are U (N ) valued transition functions for the CP-bundle, W, in the fundamental representation (as opposed to V adj ). See Section 2.4.3 for a discussion of this. Let us see what the consequences are of setting [H]dR | = 0. The eld-strength is an element of de 3 Rham cohomology, H HdR (X) H 3 (X, R), but, because it must have integer periods, it is in fact in = 3 the image of the group H (X, Z) into H 3 (X, R). Recall that the sequence 0 Z R U (1) 0 induces, in cohomology,
exp i H 2 (X, U (1)) H 3 (X, Z) H 3 (X, R) H 3 (X, U (1)) (318) n Because the sheaf used in H (X, R) is R these Cech cohomology groups, which are isomorphic to de Rham cohomology, will not have any torsion (R can easily be seen to eliminate torsion; see footnote 66). Requiring that [H]dR | = 0 implies that [H] H 3 (, Z) is a class that is in the kernel of the map i above (restricted from X to ). We will see eventually that this class is in a torsion subgroup of H 3 (X, Z). Since [H]dR | = 0 there must be a global two form, P , on such that H| = dP . Let us note several things. First, as the discussion is to be restricted to from hence forth we will stop denoting the restriction in the notation and leave it implicit. Second, P need not be the B-eld which is given by a set of local two-forms {B }. P is simply some global two-form on . If fact, B may still not be well dened as a global two-form. Rather, on a good open cover, {U }, of , the following set of equations must hold

exp

(317)

H = dP = dB P = P B = B + d B = P + d = d

(319) (320) (321) (322) (323) (324)

( + + ) + K = 110

4.2 The Freed-Witten Anomaly and HolonomyTHE B-FIELD AND TWISTED VECTOR BUNDLES 4

Eqns. (319) and (320) together suce to imply HdR = 0. Eqn. (319) implies eqn. (322) for some one-forms, , dened locally on each chart. The transition for the B-eld on overlaps, eqn. (321), is unaltered from its earlier form. Substituting (322) into (321) and using the fact that {U } is a good cover so all intersections are contractable (with trivial cohomology) implies there exist zero-forms (functions), , on each double intersection satisfying eqn. (323). Finally, substituting the value for in (323) into (312) and using the contractability of the triple intersection implies that there is a constant function, satisfying (324). Note that eqn. (324) is equivalent to saying that K is a constant up to a coboundary which follows from m = 0 in H 2 (X, R) since then m = for some H 2 (X, R) and = K which follows from (324). Thus plays the same role for K as P does for B. The relations above derived in [Kap]. Let us also modify the transition functions on A to account for the necessary coupling to the gauge transformations of B . In Section 4.2.1 a naive analysis suggested that B B + d should imply A A + d so we will begin with this ansatz. Thus eqn. (315) should be modied to
1 1 A = g A g + ig dg

(325)

where g are the U (N )-valued transition functions satisfying (316). With this modication A no longer transforms as the local representative of a connection on some line bundle. In this forms its geometric meaning is no longer clear and it is not obvious how to calculate its holonomy around a closed loop (such as M ) as we will eventually need to do. In [Kap] a geometrical meaning is given to a connection satisfying (325) as a connection on an Azumaya algebra but we will not develop this notion here (this description does not really have even a mathematical generalization to non-torsion H). Before proceeding to resolve this issue let us make a slight generalization. Recall that A is a connection on the bundle W which we have taken to be a well-dened bundle in the fundamental of U (N ). We have already mentioned that this will no longer be the case when [H] has a non-trivial cohomology class (even if it is just a torsion class). The reason is that, if we continue to allow g to satisfy the cocycle condition (316) then it will not be possible to cancel the anomaly discussed in the next section. Thus, let us make a, for now, unmotivated modication of (316) which will be shown below to result in the appropriate anomaly cancellation. Thus we posit g g g = eiK (326)

where K is dened in eqn. (312). From this modication it is now clear how, when [H] is a trivial class, so K can be set to zero, the bundle W is a well-dened bundle in the fundamental of U (N ). Otherwise, the twisted cocycle condition implies W is a twisted vector bundle. Note, however, that this modication does not resolve the issue above with A not being a well-dened connection. Let us see if we can now nd an object related to A that does transform like a connection on some vector bundle. Substituting (323) into (325) gives
1 1 A + = g (A + )g + ig dg + d

(327)

If we now dene new local connections A := A + and dene modied transition functions g = g ei it is easy to see that they satisfy
1 1 A = g A g + ig dg

(328)

g g g = g g g exp i( + + ) = ei =: (329)

Recall that is a constant R-valued function on U which implies that is a constant U (1)valued function on the same domain. Note that is actually a cocycle as can be checked explicitly 1 using the fact that g = g . Hence H 2 (X, U (1)). Let us interpret what we have done. We started by trying to identify A as a a connection on a vector bundle in the fundamental representation of U (N ). It was associated to transition functions, g U (N ), that satised the cocycle condition in the fundamental representation. In anticipation of the fact that W 111

4.2 The Freed-Witten Anomaly and HolonomyTHE B-FIELD AND TWISTED VECTOR BUNDLES 4

would later no longer be able to be a well-dened vector-bundle in the fundamental of U (N ) we modied g such that they satised the cocycle condition only up to a U (1) phase, exp(iK ). This particular form for the cocycle condition was used because it will later allow us to cancel an anomaly coming from the ill-dened holonomy of B around a open-string world-sheet, M , with a boundary. We then tried to couple A to the gauge transformation of the B-eld and found that the additional factor of in the gauge transformation rule for A no longer allowed it to be interpreted as a connection on any bundle. However, using the fact that H is torsion so satised eqn. (323) we were able to dene a new connection, A, and new transition functions, g. This new connection does have the correct transformation properties for a connection associated with a bundle with transition functions g. From (329) it is clear that g only correspond to well-dened transition functions for a vector bundle in the adjoint of U (N ), not in the fundamental. Hence, A is a connection on a bundle in the adjoint of U (N ). To reconnect with our discussion in Section 2.4.3 recall that there was a bundle V adj in the adjoint of U (N ) which we wanted to factor into the product of two U (N ) bundles, V adj W W. The connection = A would have been the connection on the U (N )-bundle, W, if that bundle was well-dened (when H is zero). The modication (326) however implied that W is no longer a well-dened vector bundle in the fundamental. The connection A, however, is a connection on the vector bundle V adj and the transition functions (329) ensure that V adj is a well-dened vector bundle in the adjoint of U (N ). 4.2.3 The Global Freed-Witten Anomaly

Let us now examine the relationship between A and B more carefully and relate it to a potential nonperturbative anomaly on the string world-sheet theory. Canceling this anomaly will justify the modication (326) of the transition functions of W, making the latter no longer a well-dened vector bundle. Recall that (rst) quantized string theory is dened by evaluating the Polyakov path integral over a xed world-sheet topology and then summing over topologies. Consider the following manipulations that relate the original Polyakov path integral, eqn. (5), to eqn. (300) Z= = =
M

DDgDD exp iSbos [, g] + iD + i

(B) Tr exp
M M

(A)

DDg eiSbos [,g] pfa(iD ) HolM ( B) Tr HolM ( A) F [] pfa(iD ) HolM ( B) Tr HolM ( A) W []


M

(330)

Here g is the world-sheet metric (we are neglecting subtleties such as gauge xing since this factor will be irrelevant in our analysis), : M X is the world-sheet embedding and , are the world-sheet fermions. Sbos [] is the bosonic part of the action including any factors that are not explicitly shown above. In the second line we have integrated over the fermionic world-sheet elds to generate a fermionic determinant, pfa(iD ), known as a Pfaan. In the third line we have incorporated all the terms except the Pfaan and the two holonomies into F [] (including the integral over gauge-inequivalent world-sheet metrics) which we will generally ignore in the rest of the exposition. The trace of the holonomy of A is evaluated in the fundamental representation and not in the adjoint. The argument for this follows from the arguments given in Section 2.4.3 for why the trace shows up in all low-energy eective actions. Recall that M is the space of all maps : M X and does not include any integral over metrics. This denes the functional W [] over the space M given in eqn. (300). Note, we have said nothing about the topology of the world-sheet so far. Here, we will deal only with the sphere and the annulus for simplicity but it should be clear from the arguments below that higher genus generalizations are straight-forward. If W [] were a well dened functional of then we would have a well-dened path integral (in so far as any path integral can be). If M has no boundary (i.e. closed string theory) this is generally true. In open string theory however, M has a boundary, M , whose image must lie on a D-brane, (M ) . In this case, which we will restrict ourselves to from now on, the three factors pfa(iD ) HolM ( B) Tr HolM ( A) 112 (331)

4.2 The Freed-Witten Anomaly and HolonomyTHE B-FIELD AND TWISTED VECTOR BUNDLES 4

are all subject to ambiguities in their denition that imply that W [] is not actually a well-dened functional. That is to say there is no well-dened number we can ascribe to W [] for a given . Rather W [] has the character of a section of a line bundle, L, on M, its value xed only up to a choice of phase. This is obviously problematic because it is not clear how to integrate a section of a line bundle. One way to resolve this problem is to divide the section by another section of unit norm to dene a number at every point on M Z
M

W [] 1

(332)

where 1 is a global section of the line bundle L M with xed length 1. Of course such a section only exists if L is trivial. The topology of L can be determined by the transition functions of the line bundles LP f , LA and LB over M dened by the factors in (331). In fact it is not hard to see that L = LP f LA LB . Hence, for the path integral to be well-dened, it is necessary for these bundles to trivialize each other in the sense that their tensor product is trivial. If this is not the case then there is a quantum anomaly, referred to as a global or non-perturbative anomaly, because, as we will see below, its derivation requires a global PSfrag replacementsof the topology of the elds A and B on X (or ). analysis Before actually dening the bundles LA and LB (we will not discuss LP f in much detail, leaving this to [FW]) let us explore the notion of a bundle on the (innite dimensional) space M. Such spaces can be treated with some mathematical rigor but such a delicate treatment will not be attempted here. Rather we will be somewhat informal and generally work by analogy with the nite dimensional case. The space M M ap(M, X) can be covered with charts as follows. Fix a good open cover, {U }I , of X indexed by a set, I. Recall that M has a boundary and let us for now restrict M to be homeomorphic to the two-dimensional disc (or ball) so M D2 and M S 1 . This represents the rst term in the world-sheet = = topology expansion. The open sets in M are given by {V(,) } which are dened as follows. Here is a triangulation of D2 (which is identied, via a xed homeomorphism, with M ). Such a triangular includes a set of two-simplices (triangles) or faces, {sa }aA , a set of one-simplices (lines) or edges, {lb }bB , and a set of zero-simplices (points) or vertices, {vc }cC . On a closed surface (which D 2 is not) every one-simplex is associated to two faces as their common edge whereas every vertex is associated to two or more faces. Figure 11 illustrates two dierent triangulation of D 2 that we will be using shortly. v41 s1 l41 l12 l4 s4 v1234 l23 l34 s3 v34 l3 (a) v23 lk vjk (b) s2 l2 sk vijk ljk sj lki lj l1 v12 vki si lij li vij

Figure 11: Two dierent triangulations of a disc, D 2 . We have used dierent index sets (1, . . . , 4 vs. i, j, k) for the faces to distinguish the triangulations. For convenience we have indexed the edges and vertices in such a way as to be able to immediately tell which faces they are associated to. Note that the order of the indices for the edges and vertices relates to a choice of orientation of D 2 .

113

4.2 The Freed-Witten Anomaly and HolonomyTHE B-FIELD AND TWISTED VECTOR BUNDLES 4

The map maps the index set of the faces, A, into the index set, I, of the xed open cover by associating, to each face sa on D2 , a chart. A choice of and denes a set of maps V(,) as follows. is dened to be in V(,) if (sa ) U(a) . This implies that maps an edge, lb , joining two faces, sa and sa , into the intersection Ua Ua and likewise for vertices. This is described in more detail in Section 5 of [Kap]. This denes a topology on M though we will not bother trying to show this. Note that this prescription works whether or not M has a boundary, though, in the former case the number of edges and vertices associated to a face may change. Also, note that if M has a boundary, M , then the prescription given above restricts to an analogous prescription, in one dimension lower, for S 1 . Namely, restricts to a triangulation (via zero- and one-simples) of S 1 and maps these into the index set of the cover, {U }. A map, , is in the intersection of two sets V(,) and V( , ) if it satises the conditions specied above with respect to both sets of triangulations and maps, (, ) and ( , ). It is via these intersections that we will dene line bundles on M. For later convenience let us agree to denote the triangulations in Figure 11 (a) and (b) as (a , a ) and (b , b ). They each generate open sets on M which we will denote Va V(a ,a ) and Vb V(b ,b ) . The indices a and b here refer to left and right triangulations of Figure 11 and should not be confused with elements of the index sets A and B of two-simplices and one-simplices respectively. 4.2.4 Holonomies and Line Bundles on M

Let us consider the terms Tr HolM ( A) and HolM ( B) to see how they dene line bundles on M. We will not attempt to work this out in full rigor because doing so would involve an inordinate amount of detail. We will go over the basic idea and invite the reader so motivated to work out the details themselves. First consider the holonomy of a general one-form connection A but, for simplicity, let us restrict A to being Abelian once more HolM ( A) = exp i
M ?

(A)

(333)

The denition given above does not work if A is not a globally well-dened one-form since then the pull back has no intrinsic geometric meaning. Thus, in the general case, we must rst x a good open cover {U }I and apply (307) to dened local one-forms A . We still may not be able to pull these one forms back to M because the image of M may span several charts. If this is the case then we nd a pair (, ) as above70 so that each one-simplex, lb , is mapped into U(b) and tentatively dene the holonomy HolM ( A) =
bB ?

exp i
lb

(A(b) )

(334)

It is not hard to check, however, that this naive denition is dependent on our choice of triangulation, (, ). The reader can check that the two triangulations of S 1 given, in a notation similar to Figure 11, by {l1 , l2 , v12 , v21 } and {li , lj , vij , vji }, where the vertices are not at the same points, results in dierent values of the holonomy. Specically the dierence will be given by the (exponentiated) sum or dierence of the functions f , from eqn. (308), evaluated at the vertices. See [Alv] or check it explicitly using eqn. (308). The correct ansatz turns out to be [Alv] [Kap] [GR] HolM ( A) :=
bB

exp i
lb

(A(b) )
cC

g(c1 )(c2 ) ((vc1 c2 ))

(335)

where g = eif are the transition functions of the bundle on which A is a connection and we have formalized our naming convention that vij is a common vertex of li and lj . Note, that the ordering 1 of the indices is important as g = g . In [Alv] it is shown that, when S 1 is triangulated by three one-simplices, the denition (335) is only independent of the triangulation if the cocycle dened in (309) is integer valued.
70 Previously we required to map two-simplexes into charts but since we are considering one-form holonomy we work with one-simplices instead.

114

4.2 The Freed-Witten Anomaly and HolonomyTHE B-FIELD AND TWISTED VECTOR BUNDLES 4

Eqn. (335) suces for triangulating S 1 where every vertex is associated with exactly two one-simplices but a more general form of the holonomy, allowing for vertices joining three or more one-simplices, can be found in [GR]. For our purposes eqn. (335) will suce. Moreover, we can generalize this denition to incorporate non-Abelian connections Tr HolM ( A) := Tr
bB

exp i
lb

(A(b) )
cC

gc1 c2 ((vc1 c2 ))

(336)

Recall once more that the trace in the holonomy (336) must be taken in the fundamental, not in the adjoint representation. This will be very relevant in calculating the holonomy [Kap]. This is because we are going to be calculating the holonomy of A, which is a connection on the adjoint bundle V adj . Although the holonomy of A is well-dened (i.e. independent of the choice of triangulation) if the traces are taken in the adjoint representation this will no longer be the case in the fundamental representation. This follows from the fact that the cocycle condition in the adjoint representation implies that the actual U (n)-valued matrices only satisfy the cocycle condition up to a U (1) phase g g g = (337)

where is a constant U (1)-valued function given in (329) and g are the transition functions of V adj (seen as elements of U (N ) under the representation Ad : U (N ) Aut(u(N ))). If g is U (N ) valued for N > 1 then the RHS should be read as IN N where the latter is the N N identity matrix. We can now apply (335) to calculate the holonomy of A around M for the two triangulations given in Figure 11 (a) and (b). We show only the nal result but, as this is not overly arduous, the reader is encouraged to attempt the calculation themselves. The reader can now check that the dierence between Tr HolM ( A) when calculated using triangulation a and b is given by Tr HolM ( A) Tr HolM ( A)

a b

= 4i1 (v41 )1ij (vij )1j2 (v12 )2j3 (v23 )3jk (vjk )3k4 (v34 )4ki (vki ) =: ga b ()

(338)

To make this manageable we have abused the notation to identify the indices of the open cover, , , . . . , by the simplex that is mapped into them so, for instance, if (i) = and (1) = then we have denoted g as gi1 . In comparing the two triangulations, a and b , above we have implicitly assumed that Va Vb . Eqn. (338) implies that, on such double intersections, the holonomy of A satises Tr HolM ( A)
a

= ga b () Tr HolM ( A)

(339)

where ga b is a U (1)-valued functional on Va Vb . This looks like the relationship between local representatives of sections of a line bundle when compared on a double intersections. Selecting a third triangulation, (c , c ), which denes a third set Vc M such that Va Vb Vc it should be possible to show, using the fact that is a cocycle, that g a b g b c g c a = 1 (340)

though this is somewhat involved. Hence, in this case, Tr HolM ( A) is actually a section of a line bundle, which we denote LA , over the space M. e If A where a connection on a proper U (N ) vector or principle bundle then = 1 and so all the transition functions for LA would be trivial and it would correspond to a trivial line bundle. We do not e want this to be the case, however, since, as mentioned in the introduction we need LA to trivialize LB . e Thus we will be forced to postulate that H 2 (X, Cont(U (1))) is a non-trivial class meaning that g do not satisfy the cocycle condition and hence are not transition functions for a well-dene bundle on . We will return to this presently.

115

4.2 The Freed-Witten Anomaly and HolonomyTHE B-FIELD AND TWISTED VECTOR BUNDLES 4

4.2.5

Surface Holonomy and Line Bundles on M

Let us now turn to calculating the holonomy of B on the surface, M , of the string world-sheet. As with the connection A, B is not a globally well-dened eld so we must calculate the surface holonomy by pulling back local two-forms B , dened with respect to an open cover, {U }, and using a triangulation, , of M . Let us for a moment consider M closed by simply taking two D 2 and attaching them at the boundary to form an S 2 . It is then possible to dene the holonomy of B over M in such a manner that it is independent of the choice of open cover of X. A detailed description of this can be found in [Alv] so we will not repeat it here. Rather, we will quote the prescription for calculating the holonomy given in [Kap] HolM ( B) :=
aA

exp i

(B(a) )
sa bB

exp i

lb

((ab1 )(ab2 ))
cC

Cvc ((vc ))

(341)

where, for each edge, lb , we dene ab1 and ab2 are the two faces containing it. The factors Cv are given as follows. If Cv = 1 if v connects two or less faces. If v is a common vertex of three faces, s a , sa , and sa then Cv = eiK(a)(a )(a ) where K is given in eqn. (312). If v is shared by n > 3 faces, {a1 , . . . , an } then we x one of these, ak , and dene Cv = exp i(K(a1 )(a2 )(ak ) + K(a2 )(a3 )(ak ) + + K(an )(a1 )(ak ) ) (342)

This is essentially a more complicated version of eqn. (335) that works for calculating the holonomy of B on a closed surface like S 2 . Once more, a generalization to more complicated surfaces, where, for instance, each edge can connect more than two faces, is given in [GR]. Note that we have included minus signs (in the exponentials) in the denition to facilitate the eventual cancellation with the holonomy of A (this is a matter of convention). Eqn. (341) is somewhat cumbersome but can be checked to be gauge invariant for dierent triangulations of S 2 . If we now consider M D2 then this is no longer the case. In fact, is is not clear how = to apply (341) to a surface with a boundary because there will now be edges contained in only one face (namely the edges on the boundary). Let us describe an extended prescription for Hol M ( B) that will apply when M = 0 [Kap]. This prescription can only be used, however, when H is a torsion class as will be obvious from the denition. For a surface, D 2 , separate the edges and vertices into two sets, internal and boundary, in an obvious manner. Let us continue to use the indexes B and C for internal edges and vertices and dene the new index sets B and C for external edges and vertices. The holonomy of B on D 2 will have an internal and a boundary contribution dened as follows. The internal contribution is simply given by (341) applied to the internal edges and vertices only (but all faces). This will be multiplied by a boundary factor exp i ((ab ) )
cC

bB

lb

Cvc ((vc ))

(343)

Since each boundary edge, lb , is contained in only one face, sa , we can associate a unique ab to each b. The vertex contributions are dened as follows. If vc is contained in only one face then Cvc = 1. If vc belongs to two-faces, sa and sa , ordered with this orientation relative to the boundary then Cvc = ei(a)(a ) . Finally, if vc connects to n faces, {sa1 , . . . , san }, with sa1 and san having boundary edges then Cvc is dened as Cvc = exp i(K(a3 )(a2 )(a1 ) + K(a4 )(a3 )(a1 ) + + K(an )(an1 )(a1 ) ) ei(a1 )(an ) (344)

This denition of the holonomy can be understood as follows. If the holonomy of D 2 is computed this way for two dierent charts and then the two copies of D 2 are glued together on the boundary to form an S 2 one can check that the product of the two holonomies gives the holonomy of S 2 calculated using the formula (341) (which is independent of the triangulation of S 2 ). Note that the use of and dened in (322) and (323) imply that H must be torsion. As with the holonomy of A on a closed loop, the independence of HolM ( B) (on a closed surface) with respect to a choice of covers will follow from applying the cocycle condition (313). However, when 116

4.2 The Freed-Witten Anomaly and HolonomyTHE B-FIELD AND TWISTED VECTOR BUNDLES 4

M has a boundary a complete cocycle will not appear in the calculation because, for instance, some edges will be contained in only one face. From the denition above it is clear that the factor of HolM ( B) coming from the interior will be independent of the triangulation while the external factor will not. In order to understand how to cancel this contribution using the holonomy of A calculated along the boundary, as in the last section, let us explicitly calculate the dierence between the holonomies of B over D2 triangulated using Figure 11 (a) and (b). This can be calculated using the intersections of the two triangulations depicted in Figure 12 below. l1 l4 vki l2 lki l4 s4 s k l34 v41 s4 s i s1 s i l41 l2 vij v12 l1

v1234

v34 l3 vjk l3

v23

Figure 12: When the two triangulations are superimposed we get several regions which have been denoted by the two-simplex (face) that corresponded to them, such as s1 si . Edges that have been split by an intersection are labeled using accents so that, e.g., lki = lki + lki . On the boundary where the lines li,j,k and l1,2,3,4 were used in Figure 11 on overlapping line segments we have just used the l1,2,3,4 and split these lines into multiple segments where they intersect vertices. For practical reasons we have only labeled the components that will be used in the analysis below. Let us rst give the holonomy for Figure 11 (a) and (b) separately. HolM ( B)(a) = exp i exp i
s1

B1 1

l12

21
l3

s2

B2
l4

l23

32

s3

B3

l34

43

s4

B4

14
l41

l1

l2

exp i(12 (v12 ) + 23 (v23 ) + 34 (v34 ) + 41 (v41 )) (345)

exp i(K121 (v1234 ) + K231 (v1234 ) + K341 (v1234 ) + K411 (v1234 ))

HolM ( B)(b) = exp i exp i


si

Bi i

lij

ji
lk

sj

Bj

ljk

kj

sk

Bk

ik
lki

(346)

li

lj

exp i(ij (vij ) + jk (vjk ) + ki (vki ))

exp i(Kijk (vijk ))

117

4.2 The Freed-Witten Anomaly and HolonomyTHE B-FIELD AND TWISTED VECTOR BUNDLES 4

Let us now consider HolM ( B)(a) HolM ( B)(b) (347)

In the interest of brevity let us focus on the region given by the overlap of the faces s 4 and si as depicted in Figure 12. We will denote this region s4 si . The contribution to the quotient (347) coming from this region is given by (the exponential of) (B4 Bi ) = d4i = 4i 4i 4i 4i
l4

(348)

s4 si

s4 si

l41

l34

lki

The notation l4 and l4 denote two dierent parts of the original edge l4 when it is split by an intersection with another edge. See Figure 12 for some examples of the notation. Something similar to eqn. (348) will happen to all the faces. Let us now consider the sum of all contributions on a particular internal edge. Let us focus on l41 , one of the edges of s4 si . There are three terms in (347) relating to this edge. One term from (348), another similar contribution from the adjacent region s1 sj , and nally a contribution from the numerator, (345), of (347). Summing these gives

l41

(4i 1i + 14 ) =

l41

(4i + i1 + 14 ) =

l41

dK4i1 = K4i1 (v41 ) + K4i1 (v1234 )

(349)

Once more something similar to (349) will happen to all the internal edges. So we have reduced the problem, at least on the interior, to considering vertices only. Note that the external edges (with only single digit subscripts) will not have a contribution from a second face to provide a full cocycle. We will come back to these momentarily. Let us now focus on the internal vertex v1234 . It will get a contribution similar to (349) from the other three edges that it connects (see Figure 12). After some one work can check that this contribution is K4i1 (v1234 ) + K3i4 (v1234 ) + K1i2 (v1234 ) + K2i3 (v1234 ) It also receives contributions from the numerator, (345), of (347) given as K121 (v1234 ) + K231 (v1234 ) + K341 (v1234 ) + K411 (v1234 ) = K231 (v1234 ) + K341 (v1234 ) (351) (350)

where we have applied the antisymmetry of K in all its indices to get rid of the rst and last term. Let us combine all of these, dropping the dependence on v1234 in the notation,
Ki13 +2 mi134

K231 + K341 + K4i1 + k3i4 +K1i2 + K2i3 = K231 + Ki13 + K1i2 + K2i3 +2 mi134
2 mi123

(352)

So the total contribution at v1234 is given by the exponential of 2 mi134 + 2 mi123 which vanishes if m H 3 (, Z) which we have already assumed. Something similar will happen at the other internal vertices and so the contribution to the quotient (347) from internal vertices will be a factor of unity. Let us now turn our attention to the more problematic boundary. From eqn. (348) we see that the surface integral of B4 Bi on the region marked s4 si in Figure 12 will add a contribution to the boundary edge l4 and the boundary vertex v4 of K4i1 (v41 ) 4i
l4

(353)

118

4.2 The Freed-Witten Anomaly and HolonomyTHE B-FIELD AND TWISTED VECTOR BUNDLES 4

Another contribution comes from the dierence between the boundary terms in (345) and the terms in (346) along the overlap l4 li 41 (v41 ) +
l4

4 i

(354)

The other faces with boundary edges will contribute similar factors and the total factor around the boundary will be ta b = K4i1 (v41 ) K1ij (vij ) K1j2 (v12 ) K2j3 (v23 ) K3jk (vjk ) K3k4 (v34 ) K4ki (vki ) + 41 (v41 ) ij (vij ) + 12 (v12 ) + 23 (v23 ) jk (vjk ) + 34 (v34 ) ki (vki ) +
l4

(4 i 4i ) + (3 j 3j ) +

l1

(1 i 1i ) + (3 k 3k ) +

l1

(1 j ij ) + (4 k 4k )

l2

(2 j 2j )

(355)

+
l3

l3

l4

The quotient (347) will be given by the exponential of this sum. By applying eqns. (323) and (324) the exponential of (355) can be calculated as
1 1 1 1 1 1 1 ha b () := exp(ita b ) = 4i1 (v41 )1ij (vij )1j2 (v12 )2j3 (v23 )3jk (vjk )3k4 (v34 )4ki (vki )

(356)

so that the relation between HolM ( B) for V(a ,a ) V(b ,b ) is given by HolM ( B)(a) = ha b () HolM ( B)(b) (357) Again, this looks suggestive of the transition function of a line bundle, LB , on M. In fact it is easy to check that ha b = g1 b so ha b obviously denes a line bundle and is, moreover, isomorphic to L , the a e A dual bundle to LA . Thus the tensor product of the two bundles, LA LB , will be trivial. e e To recapitulate briey we have calculated the quotient of the holonomies of B over D 2 using triangulations a and b . This was reduced to considering integrals over external edges and sums over internal vertices. All internal vertices where shown to give a total contribution of unity to this quotient. Only edges and vertices on the boundary gave non-trivial contributions and the total contribution is given by (356). This has the interpretation of a transition function, ha b , for a line bundle, LB M, of which the holonomy, HolM ( B), is a section. It can be checked that ha b satisfy the cocycle condition. Because the transition functions associated with the line bundles LA and LB are inverse to each other the tensor e product of these two bundles is trivial. Let us repeat the overall logic, since it is somewhat convoluted, to understand what we have accomplished. Recall, from Section 4.2.3, that the integration measure, W [], is not a functional of M but a section of a bundle LA LB LP f . For now we are ignoring the factor LP f associated with the pfa(iD) e term in the path integral; we essentially assume this latter bundle is trivial. To ensure that a global section of unit norm, 1 (LA LB ), exists on this bundle the bundle must be trivial. Above we have e demonstrated that this is so. The careful reader will note, however, that this was only possible at the expense of modifying the transition functions, g , of W, so that they satisfy g g g = eiK =: (358) This modication was introduced in Section 4.2.2 without motivation. The essential reason for this was to provide the necessary factors that would allow the transition functions, g a b , calculated in Section 4.2.4 to be equal to the inverse of the transition functions ha b . We will now consider, very briey, how the modication (358), which implies W is no longer a proper vector bundle, will eect the analysis of Sections 2 and 3 in terms of classifying D-branes via vector bundles and K-theory.

119

4.3 Twisted CP Bundles

4 THE B-FIELD AND TWISTED VECTOR BUNDLES

4.3

Twisted CP Bundles

In the previous section we showed that if the transition functions, g , of the Chan-Paton bundle V adj , seen as elements of U (N ) satisfy the twisted cocycle condition g g g = IN N (359)

where H 2 (, U (1)) is related to H via (329) then the Freed-Witten global world-sheet anomaly cancels. Even though g do not satisfy the cocycle condition as U (N ) matrices, the fact that they fail by a U (1) valued factor implies that they do dene a proper vector bundle, V adj , in the adjoint representation. We will show below that there is a class, y H 2 (, ZN ) and a map 1 such that 1 (y) = [H] H 3 (, Z). The fact that y has N torsion so N y = 0 implies the same for H. This implies that if H is a torsion class with mH = 0 then the number of bundles wrapping must always be a multiple of m. While these arguments already use the fact that H is a torsion class they do also suggest, heuristically, that if H is non-torsion then we could only wrap an innite number of D-branes on . Of course in the latter case it is not clear how to extend the arguments of the previous few sections (for instance, as N goes to innity A becomes an innite dimensional matrix that may no longer have nite trace) though some attempt is made to deal with this in [CJM]. 4.3.1 Obstruction to Dening U (N )-bundles

Although we have claimed the g are the transition functions for V adj , as U (N )-valued transition functions they are not unique. Any U (N )-valued two-cochain, h = g q , denes the same adjoint bundle, V adj , so long as q Cont(U (1)). We use this freedom to dene SU (N )-valued transition functions by letting q = (det g )1/N IN N (360)

Note that these are not uniquely dened but are xed only up to an N th root of unity. The cocycle condition for h now becomes h h h = q q q IN N =: y IN N (361)

This denes a cocycle, y H 2 (, Cont(U (1))) as can easily be checked. Since the determinant of both sides of this equation is one y ZN U (1) (it is an N th root of unity on any triple intersection) which implies that it is actually in H 2 (, ZN ). There exact sequence 0 Z Z ZN 0 induces the long exact sequence in cohomology
1 H 2 (, Z) H 2 (, ZN ) H 3 (, Z) N exp(2i()/N )

(362)

(363)

where the map 1 (y) is dened as follows. Let p Z be a two-cochain such that y = exp(i2p /N ). The fact that y = 0 implies that p N Z so denes a three-cochain in N Z. 1 (y) H 3 (, Z) is dened as the image of p under the inverse image of N , namely 1/N . Thus 1 (y) = 1/N p. This construction of 1 follows the standard construction of the connecting homomorphism given in [BT]. Let us work this out explicitly for y . Let l be R-valued two-cochains dened by q = exp(il ). Then q q q = exp i( + l + l + l ) = y = exp(i2p /N ) which implies that 2p = + (l) + 2n N (365) (364)

120

4.3 Twisted CP Bundles

4 THE B-FIELD AND TWISTED VECTOR BUNDLES

Where n is some Z-valued three-cocycle. is dened in eqn. (324) as = K + so the above becomes 2p = K + () + (l) + 2n N Applying the coboundary operator to this gives 2 (p) = (K) + 2(n) N = 2m + 2(n) (366)

(367) (368)

Since 1 (y) = 1/N p = m + n and m is the Cech representative of H in H 3 (, Z) this implies that 2 (, ZN ) has N -torsion, N y = 0, so 1 (y) = [H]. Since the class of y in H N (y) = N [H] = 0 (369)

and H is in a torsion subgroup of H 3 (, Z). The degree of this subgroup need not be N but it must be a divisor of N . Thus we arrive at the conclusion that the rank of the Chan-Paton bundle is constrained by the torsion degree of the restriction of H to . Let us also note, via this analysis, that when H = 0 as a class in H 3 (, Z) then it is possible to choose adj U (N )-valued transition functions, h , of V that also lift to transition functions of the bundle W in the fundamental of U (N ). Referring once more to the exact sequence (363) note that if H = 1 (y) = 0 then y must be in the image of an element in H 2 (, Z). This means, for instance, that we can regard p as an element of H 2 (, Z). These dene, via the isomorphism H 1 (, Cont(U (1))) H 2 (, Z) discussed in = Section 3.2.2, cocycles, t = eif , with values in U (1) such that f + f + f = 2p We use these to dene h = h t . One now checks that, as U (N ) matrices, h h h = 1 (371)
1/N

(370)

It is also clear, since Ad(h ) = Ad(h ) = Ad(g ), that these also act, in an adjoint representation, as transition functions for V adj . Hence, in this case, when 1 (y) = 0 it is possible to dene V adj as a tensor product of two well-dened vector bundles, W and W, in the fundamental representation. 4.3.2 Spinc and pfa(iD)

Before moving on let us touch briey on the relation of this analysis to the requirement that the D-brane tangent and normal bundles be Spinc . The original work on this was done in [FW] where it was shown that the factor pfa(iD) in the path integral is also ill-dened. The exact arguments for this are given in [FW] and are too involved to discuss here. What is shown there, that can be related to the analysis above, is that this factor is ill-dened in a manner that is similar to way the factor HolM ( B) is ill-dened. They both dene sections of line bundles on M and the rst Chern class of these line bundles is, in both cases, related to the transgression of an element of H 3 (, Z) [Kap]. We have already mentioned this briey and will not develop it in any detail but essentially it is possible to integrate a class in H 3 (, Z) 1 2 (L, Z) where L is the loop space of , around a loop, : S , in in order to dene a class in H the space whose points each correspond to a loop in . The classes in H 2 (L, Z) so dened correspond to the Chern classes of line bundles and it can be shown that the transgression of H denes the Chern class of LB while the transgressions of 1 (y) and w3 (N ) H 3 (, Z2 ) dene the Chern classes of LA e and LP f , respectively. Thus the condition that LB LA LP f be trivial implies that e w3 (N ) + 1 (y) = H| (372)

Thus if H is trivial and the bundle W is well-dened (so y is a trivial class) then this condition becomes w3 (N ) = 0 which is the condition that the D-brane is Spinc which we have enforced through-out Sections 2 and 3. 121

4.4 Outlook and Current Research

4 THE B-FIELD AND TWISTED VECTOR BUNDLES

4.3.3

Some Basic Notions of Twisted K-Theory

In order to extend the discussion of Sections 2 and 3 to the cases when H is non-trivial and hence W is not well-dened as a vector bundle requires the introduction of twisted K-theory, a generalization of the topological K-theory introduced in Section 3. There are many dierent formulations of twisted Ktheory, however they all require the use of algebraic rather than topological K-theory. In Section 3.4.1, K-theory was introduced as a contravariant cohomology theory to study topological spaces by studying pairs of vector bundles on those spaces. In fact, this is only one formulation of topological K-theory. An equivalent formulation is given, in a more algebraic language, by studying projective modules of the ring Cont(X) of continuous functions on the topological space X. In this thesis we have focused on the more topological construction of topological K-theory as it requires less machinery and also connects more naturally to Cech and de Rham cohomology. None-the-less the two formulations are equivalent. In order to classify twisted vector bundles and thereby dene twisted K-theory one should, in fact, consider the algebra of sections of the twisted analog of Cont(X). This is given by the algebra of sections of a particular, innite dimensional, bundle [BM]. Developing these notions, however, would require introducing algebraic K-theory and a lot of other mathematical machinery which we do not have time to do. We leave it to the motivated reader to pursue the references mentioned earlier and peruse the literature.

4.4

Outlook and Current Research

We have already said something of the open problems in this eld in the introduction to the thesis and to this section. There, and previously, we have mentioned that there are still several open problems in the eld and that their resolution might lead to greater insights into regimes of string/M-theory that are hard to probe in the standard formulation. The basis for this statement is Wittens remark [Wit2] that the notion that an innite number of branes is required when H is non-torsion can be related to going o-shell. Besides this, there are several interesting paradoxes suggested by the K-theoretic classication and its extension to twisted K-theory. D-branes are related by dualities to other objects, such as the fundamental string or the N S5-brane, which are sources for H and its dual yet the latter seem to be classied by cohomology rather than K-theory [KS]. If the relevant dualities are exact symmetries of the theory then this is inconsistent and a more complete understanding clearly needs to be reached. D-branes are also related by dualities to branes in M-theory and so the K-theoretic classication of the former should imply some similar classication for the latter but this is still not entirely well-understood 71 [DMW]. The physical analysis in this section serves as the basis for a lot of the later generalizations such as the (torsion degree) m limit and it is hoped that the reader will be able to use it to more productively approach the literature. Attempts to take this limit are hampered by the fact that the holonomy becomes undened since A becomes an operator on a Hilbert space whose trace may not be nite. In [BCM+ ] and [CJM] some solutions are suggested for this but this requires a description of B as a gerbe and a lot of additional mathematical machinery. It does not seem, to the author, that conclusive physical progress has been made in this area yet.

71 As

far as the author is aware.

122

A SPIN- AND SPINC -STRUCTURES

Spin- and Spinc-Structures

In this appendix we will develop some results on Spin- and Spinc -lifts of real vector bundles. Although these results will be used in several points in the main text they are sometimes rather technical and so have been placed in an Appendix. Moreover, the details of many of the constructions below are not necessary for an understanding of the main text and will only be of interest to readers who wish to study the subtleties that come into play when a bundle admits only a Spinc rather than a Spin-structure. The reader is warned that we will need to use some results from the representation theory of Cliord algebras which we will assume the reader to be familiar with. A detailed exposition of the latter can be found in [LM]. Although the groups Spin and Spinc will be introduced below with respect to SO(n) the derivation applies just as well to SO(p, q) for general p and q since, for all of these, Spin(p, q) is the double cover of the group [LM].

A.1

Spin-lifts of Principle SO(n)-bundles

Consider an oriented, real rank n vector bundle W . This bundle is completely specied by its SO(n)-valued transition functions on intersections g : U U SO(n) As these functions must, by denition, satisfy the cocycle condition g g g = 1 1 g = 1 (374) (373)

they dene an element of H (, C(SO(n))). Here, and through-out the appendix, we will let C(G) denote the sheaf of continuous G-valued functions for a group G. Since SO(n) is non-Abelian for n > 2 this is no longer a cohomology group (or even a group at all) but it can still be studied as a set and there are results that demonstrate that it is still possible to partially form a long exact sequence using such objects (see [LM, App. A] for the relationship to principle bundles or [Bry, Ch. 4.1] for a more general discussion). Working with such objects as if they were cohomology groups we can see that there is a one-to-one correspondence between elements of H 1 (, C(SO(n))) and (isomorphism classes) of bundles with SO(n)valued transition functions. This is because two bundles, W and W , with transition functions g and g (dened with respect to the same cover U = {U } by taking common renements if necessary) dene isomorphic bundles if there are ber-wise isomorphisms, h , (i.e. that commute with the projections) dened on each element of a trivialization, U , which commute with the transition functions so g = h1 g h (375)

Thus h can be thought of as a coboundary for a non-Abelian cohomology group and eqn. (375) is the statement that two sets of transition functions give rise to the same isomorphism class of bundle so long as they dier by a coboundary (with the coboundary map given by (h) = h1 ()h ). Combined with the cocycle condition, eqn. (374), showing that g is closed under the boundary map, this result implies that each isomorphism class of rank n vector bundles corresponds uniquely to an element of H 1 (, SO(n)). The discussion above referred to an SO(n) vector bundle, W, but in fact the cocycle [g] H 1 (, SO(n)) denes a ber bundle for any space (i.e. typical ber) that has SO(n) as a subgroup of its automorphism group. For instance, using the cover U and g dened above one denes a principle SO(n) bundle, PSO(n) , which is the associated principle bundle to W. This is done by taking the disjoint union

U SO(n)

(376)

for U U and then quotienting by the equivalence which identies pairs (x1 , h) U SO(n) and (x2 , g (x2 ) h) U SO(n) if x1 = x2 is the same point in U U (that is, we glue together trivial topological products using the transition functions). The vector bundle W can now be seen as an associated bundle to the principle SO(n)-bundle PSO(n) associated via the fundamental representation. That 123

A.1 Spin-lifts of Principle SO(n)-bundles

A SPIN- AND SPINC -STRUCTURES

is, we can dene W = PSO(n) Rn where : SO(n) Hom(Rn , Rn ) is the fundamental representation and the product given associates points (p, x) PSO(n) Rn via (p, x) (pg 1 , (g)x) for all g SO(n). So far we have discussed bundles with SO(n) structure group but SO(n) is not simply connected (for n > 2) and in physics it is often relevant to consider bundles with structure group given by the double cover, Spin(n), of SO(n), which is simply connected. Moreover, such bundles are generally expected to be a lift of a given SO(n)-bundle in the following sense. There is an exact sequence 0 Z2 Spin(n) SO(n) 0

(377)

where is the inclusion of Z2 as a subgroup of Spin(n) given by the two elements 1 and 1 and is the projection dened by quotienting out this group (this is why Spin(n) is the double cover of SO(n)). Locally, of course, one can always embed SO(n) in Spin(n) by selecting one element, either 1 or 1, in the kernel of and using it to make a consistent choice of pre-image. However, it is not always true that given an SO(n)-bundle there is a corresponding, well-dened, Spin(n)-bundle. Any given transition function g SO(n), dened on U U can be lifted to an element g Spin(n) by a choice of an element in Z2 (i.e. 1 in this case) but it is not true that such choices can always be made in a consistent manner. There is a global topological obstruction that can be studied using the following exact sequence. Recalling the previous argument about the existence of (part of) the long exact sequence, even in the case with coecients in a non-Abelian group, the sequence (377) induces H 1 (, Z2 ) H 1 (, C(Spin(n))) H 1 (, C(SO(n))) 1 H 2 (, Z2 )

(378)

If the class [g] H 1 (, C(SO(n))) corresponding to PSO(n) (given, for instance, on U U by g ) is in the image of a class [g] H 1 (, Spin(n)) so that [g] = ([g]) such a lift can be shown to exist. In this case there is a principle Spin(n)-bundle, PSpin(n) , corresponding to [g] in the same way as [g] corresponds to PSO(n) . The fact that [g] = ([g]) implies that, over every intersection U U U , the Spin(n)-valued transition functions, g of PSO(n) are a lift (via an element of Z2 ) of the SO(n)-valued transition functions g . Exactness of sequence (378) implies that for this to be the case requires that w2 (W) 1 ([g]) = 0 as an element of H 2 (, Z2 ). This class, w2 (W), is the second Stiefel-Whitney class of the vector bundle W and its vanishing implies that W admits a lift to a Spin(n)-valued bundle. To see more concretely what the topological obstruction is let us work it out in detail. Consider selecting, on each intersection, U , an element, g , in the pre-image of 1 (g ) so that (g ) = g (the sequence (377) implies that for any such g , another valid choice would have been g ). This can be used to dene new transition functions, {g }, for a putative Spin(n)-bundle. The cocycle condition (374) for the SO(n) bundle implies (g ) (g ) (g ) = 1 (g g g ) = 1 g g g =
1

(379)

(1) = 1

This means that the Spin(n)-valued transition functions g may not satisfy the cocycle condition and so do not dene a proper Spin(n)-bundle. However, it may be possible to modify the choice of lifts, g , in such a way as to ensure that the 1 in the above expression is always +1. First, note that the last line in (379) denes an element in H 2 (, Z2 ) (1)
1 1

g g g = 1

(380)

because ( ) = = 1 (this can be seen by simply working it out in terms of g) so is a cocycle. This cocycle, [ ], is actually the Stiefel-Whitney class, w2 (W). If this element is trivial so [ ] = 0 then there exists an element [f ] H 1 (, Z2 ) such that = f . This means that

= f f f

(381)

If this is the case then it is possible to select a dierent lift, h , of the g so that the cocycle condition will be satised for these for these Spin(n)-valued transition functions. Specically, let h = g f 124

A.2 Spinc -lifts of Principle SO(n)-bundles

A SPIN- AND SPINC -STRUCTURES

and then note that (using the fact that f Z2 Spin(n) is a central element) h h h = g (1)f g (1)f g (1)f = (g g g )(1)(f f f ) = (1)
2(
)

(382)

=1

Thus we see explicitly that if [ ] is a trivial class we can use it to construct a well-dened Spin(n)-lift of the SO(n)-bundle. That this condition is not only sucient but is necessary follows from the arguments of the previous paragraphs via the long-exact sequence.

A.2

Spinc -lifts of Principle SO(n)-bundles

Having determined the topological obstruction to dening a Spin-lift of an SO(n)-bundle we now investigate the possibilities when no such lifts exists. We will show that there is a closely related structure, the Spinc -structure, which can often be used to replace the Spin-structure. We will also determine the topological obstruction to dening a Spinc -structure. Suppose that [ ] = 0 so it is not possible to lift PSO(n) to a well dened Spin-bundle, PSpin(n) . It may still be possible to construct a bundle which transforms under a Spin(n)-representation but in a non-trivial twisted way. Namely, in some circumstances, it is possible to construct a principle Spinc(n) Spin(n) Z2 U (1) bundle; the representations of Spinc(n) restrict to representations of Spin(n) via the morphism Spin(n) Spinc (n) (383)

The group Spinc (n) is dened as the quotient of the group Spin(n) U (1) by the subgroup Z2 = {(1, 1), (1, 1)}. That is, any two elements (v, ) and (v, ) are identied. Let C(Rn ), CC (Rn ) denote the Cliord algebra (and its complexication) associated with Rn and the standard inner product (this notation follows [LM]; see there for more details and for the relation to Spin(n)). Since Spin(n) C(Rn ) C(Rn ) C CC (Rn ), its irreducible complex representations can all = be derived by restricting a representation (on a vector space, V ) given by the complex homomorphism : CC (Rn ) HomC (V, V ) to the subgroup Spin(n) CC (Rn ). Likewise, Spinc (n) Spin(n)Z2 U (1) C(Rn ) C CC (Rn ) but here the inclusion in C(Rn ) C is not merely an extension of one in C(Rn ) = because of the U (1) factor in the denition of Spinc(n) (which is mapped in a non-trivial way into the C in the product C(Rn ) C). Thus representations of Spinc (n) can also be derived by restricting complex representations of CC (Rn ) and, as in the case of Spin(n), the irreducible representations are actually given by representations of 0 the subgroup CC (Rn ) CC (Rn ) which acts irreducibly on each component of the decomposition V = V + V of the irreducible CC (Rn )-module, V . That is, each irreducible CC (Rn )-module, V , decomposes 0 into two, inequivalent CC (Rn )-modules, V , which are also inequivalent as Spinc (n)-modules (via the inclusion Spinc(n) Spin(n) Z2 U (1) C 0 (Rn ) C CC (Rn ) and the restriction of the representation = 0 homomorphism). Thus representations of Spinc (n) are given by representations of Spin(n) but where the action is twisted by the equivalence relation dening Spinc(n). As Spin(n) is the double cover of SO(n) so Spinc (n) is the double cover of SO(n) U (1) and the exact sequence (377) becomes 0 Z2 Spinc (n) SO(n) U (1) 0

(384)

Where now the map takes both (1, 1) and (1, 1) in Spinc(n) to the identity in SO(n) U (1). As in the case of SO(n) and Spin(n) it is possible to extend this analysis of groups to principle ber bundles and try and determine if a principle SO(n) U (1) bundle lifts to a principle Spinc (n) bundle. This leads to considering the following long exact sequence (analogous to (378)) H 1 (, Z2 ) H 1 (, C(Spinc (n))) H 1 (, C(SO(n))) H 1 (, C(U (1))) 2 H 2 (, Z2 )) (385)

125

A.2 Spinc -lifts of Principle SO(n)-bundles

A SPIN- AND SPINC -STRUCTURES

For any manifold, , there is an isomorphism72 H 1 (, C(U (1))) H 2 (X, Z) which we use to rewrite = (385)
H 1 (, Z2 ) H 1 (, C(Spinc (n))) H 1 (, C(SO(n))) H 2 (, Z) 2 H 2 (, Z2 ))

(386)

We will see below, when explicitly considering the isomorphism, H 1 (, U (1)) H 2 (X, Z), that the = 2 (X, Z) by taking its mod 2 reduction. connecting homomorphism 2 acts on an element of H The image of [g] H 1 (, C(SO(n))) under 2 is given, as in the Spin(n)-case, by the class [ ] w2 (PSO(n) ) as demonstrated in Appendix A.1. Thus on an element, ([g], [d]) H 1 (, C(SO(n))) H 2 (, Z) the map acts as 2 ([g], [d]) = 1 ([g]) + ([d] mod 2) = w2 (PSO(n) ) + ([d] mod 2) (387)

Thus, if w2 (PSO(n) ) = ([d] mod 2) then, as an element of H 2 (, Z2 )) the sum w2 (PSO(n) )+([d] mod 2) = 1 (, C(SO(n)))) H 2 (, Z) must be in the image of by exactness of the 0. In this case ([g], [d]) H sequence so there must be some class [g] H 1 (, C(Spinc (n))) such that ([g]) = ([g], [d]). This class would correspond to a well-dened principle Spinc -bundle, PSpinc (n) , whose transitions functions are lifts of the principle SO(n) U (1)-bundle, PSO(n)U (1) , given by the element ([g], [d]) H 1 (, C(SO(n))) 2 H (, Z). In this way one can start with W L, the product of an SO(n)-bundle and a complex line bundle, L, as dened by ([g], [d]) where [d] = c1 (L) and then there is a well dened bundle S(W)L1/2 transforming under Spin(n) Z2 U (1) where S(W) is a spin-lift of W and L1/2 is the square-root of the bundle L (the complex bundle whose transition functions are given by the square root of the transition functions of L). If w2 (PSO(n) ) = 0 or c1 (L) mod 2 = 0 then these two objects are not well-dened individually, but the fact that w2 (PSO(n) ) + ([d] mod 2) = 0 implies that the tensor product S(W) L1/2 is. We will return to this presently but rst let us try to reproduce this construction more concretely (as was done for the Spin(n)-case). Consider, once again, an SO(n)-bundle, W, whose transition functions, lifted to Spin(n)-valued functions, g , dene a non-trivial element H 2 (X, Z2 ) so that (1)[w2 (W)] (1)

= g g g = 1

(388)

and so that [ ] = 0. We want to nd a line bundle, L, whose rst Chern class, [d] = c1 (L), has a reduction mod 2 equal to [w2 (W)]. This was suggested by the analysis in the previous paragraph and we will attempt to show what the point of this is. The rst Chern class of a line bundle was discussed in great detail in Section 3.2.2 where it was given by [d] = 1 ( + ) Z 2 (389)

where the ei are the U (1)-valued transition function on U and the fact that [d] Z follows from the cocycle condition ei ei ei = 1 (390)

since L is a proper line bundle. This condition, however, does not constrain [d] to be even. The analysis of the long-exact sequence (386) suggests that we want a bundle, L, such that [d] mod 2 = [ ] in H 2 (, Z2 ); such an element will only be trivial if [d] is always even (or can be made even by an appropriate choice of coboundary). Let us assume this is not the case and let us attempt to dene the square root bundle, L 1/2 , as the complex line bundle whose transition functions are given by ei /2 . It is immediately obvious that this
72 See

Section 3.2.2 and the sequence (204) in particular.

126

A.2 Spinc -lifts of Principle SO(n)-bundles

A SPIN- AND SPINC -STRUCTURES

is not a well dened bundle because [d] can be odd so [d/2] H 1 (, Z). More concretely, the cocycle / condition ei /2 ei /2 ei /2 = ei( + + )/2 =e
i[ ]

= ei[d] = ei2([d] /2) ei([d] = (1)


[ ]

mod 2)

(391)

is not always satised. In the second to last line the term ([d] /2) is always rounded down to be an integer so that, even if [d] is odd, [d] = ([d] /2) + ([d] mod 2). Eqn. (391) provides a map from H 2 (, Z) to H 2 (, Z2 ) that denes the action of 2 on H 2 (, Z). Note that the cocycle condition for L1/2 fails in exactly the same way as the Spin(n)-valued transition functions, g , of S(W), the ill-dened Spin-lift of W. This was actually the whole purpose of constructing L1/2 because we can now consider the cocycle condition for the tensor product bundle S(W) L 1/2 g ei /2 g ei /2 g ei /2 = (g g g )(ei /2 ei /2 ei /2 ) = (1)2[
]

=1

(392)

which is satised. Thus S(W) L1/2 is a well-dened vector bundle even though the separate bundles, S(W) and L1/2 , are not. It is not hard to see that S(W) L1/2 is a bundle with structure group Spinc(n). Clearly Spin(n) acts on the factor S(W) while U (1) acts on L1/2 but the element (1, 1) Spin(n) U (1) acts trivially which induces an action of Spinc (n). Thus we have a concrete realization of the notion that to have a well-dened Spin c -structure associated with an SO(n)-bundle, W, it is sucient that there exists a line bundle, L, such that w2 (W) = c1 (L) mod 2 (393) in H 2 (, Z2 ). Let us examine this condition in more detail. To see how c1 (L) H 2 (, Z) gives an element in 2 (, Z2 ) consider the following exact sequence of abelian groups H 0 Z Z Z2 0 which induces the following long exact sequence in cohomology H 2 (, Z) H 2 (, Z) H 2 (, Z2 ) 3 H 3 (, Z)
2 mod 2 2 mod 2

(394)

(395)

and then set w2 (W) H 2 (, Z2 ) to be the image of a class [d] = c1 (L) H 2 (, Z). The map mod 2 is not at all injective so for any [d] in the pre-image of w2 (W) it is possible to select another [d ] such that [d ] [d] is in the kernel of mod 2. Specically, for any line bundle K, let M = K K, then if w2 (W) = c1 (L) mod 2 then c1 (L M) mod 2 = c1 (L K K) mod 2 = c1 (L) + c1 (K) + c1 (K) mod 2 = c1 (L) mod 2 (396)

Hence, rather than selecting L, it would also have been possible to select L M and the tensor product bundle S(W) (L M)1/2 would still be well dened. Thus, there is not a unique choice of line bundle for the Spinc -lift of a given vector bundle, W. In the context of D-branes, which is discussed in Sections 2.7 and 3.5.8, this at rst seems like a problematic ambiguity but it is not hard to see that it actually is not at all. To close the discussion of Spinc(n) lifts of a SO(n)-bundle, W, note that the topological constraint on W for it to be Spinc (i.e. to admit a Spinc-lift) is compactly represented by the sequence (395). Namely, for w2 (W) = c1 (L) mod 2 for some line bundle, L, it is necessary and sucient that 3 (w2 (W)) = 0. The class w3 (W) 3 (w2 (W)) is the third Stiefel-Whitney class and it encodes the topological obstruction to having a Spinc structure (the rst Stiefel-Whitney class, which we have not discussed, is the topological obstruction to lifting an O(n)-valued bundle to an SO(n)-valued bundle; it encodes the obstruction to 127

A.2 Spinc -lifts of Principle SO(n)-bundles

A SPIN- AND SPINC -STRUCTURES

orientability of the bundle).73 . Thus a vector bundle is Spinc if and only if its third Stiefel-Whitney class vanishes.

73 [LM] has a nice discussion about how these various obstructions relate to the homotopy groups of the structure groups of a bundle. Briey, O(n) has non-trivial 0th homotopy (as it is disconnected) and w 1 (W) measures the obstruction to lifting the structure group to a connected group, namely SO(n). Likewise SO(n) is not simply connected so has non-trivial 1st homotopy and w2 (W) measures the topological obstruction to lifting to the simply connected group Spin(n).

128

B SOME K-THEORY TECHNICALITIES

Some K-theory Technicalities

This is a technical appendix in which we will discuss the explicit construction of the Thom class, in complex K-theory, for real, even dimensional Spinc vector bundles (Appendix B.1). We will also describe how this relates to the cohomological construction of D9-D9-brane congurations in the codimension two case (Appendix B.2). These results have been deferred to this appendix because they rely on some details of Cliord algebra representation theory that would have been cumbersome to develop. Thus the reader of this appendix is assumed to have some familiarity with this subject (see [LM]). The reader should also note that the contents of this Appendix are not required to understand the main body of the text though they do add some depth to the exposition.

B.1

Constructing the K-theoretic Thom Class

We consider a compact base manifold, , and a real, even dimensional bundle N that is spin. If N is Spinc but not Spin then the construction dened here can be modied in a suitable way (see Section 3.5.8). Associated to N are two bundles, B(N ) N and S(N ) N , which are the ball and sphere bundle, respectively, whose bers are the 2k-dimensional ball and and the 2k 1-dimensional sphere, B 2k , and S 2k1 = B 2k . Note that the former is homotopic to the base space (and to N itself) since each ber is contractable so there is an induced bijection V ect() V ect(B(N )) ( V ect(N )) dened = = by the pull-back via : B(N ) ( is the restriction of : N to B(N ) N ; we will be sloppy and denote them both using the same symbol). If N is a spin-bundle, meaning that its second Stiefel-Whitney class vanishes (i.e. the SO(2k) valued transition functions lift to Spin(2k) valued functions in a consistent way; see [LM], [Nak], and Appendix A.1), it is possible to dene a bundle transforming under a spin representation of SO(2k) which we will denote S(N ) . More precisely, the SO(2k) valued transition functions of N are lifted to the double cover of SO(2k), Spin(2k), by a consistent choice of 1 in the lift on every element of the cover (that such a choice can be made is guaranteed because w2 (N ) = 0; see Appendix A.1 for the details of this construction). These Spin(2k) valued transition functions can now be used to dene a new bundle on as follows. Let {U } be a cover of trivializing N , let g be the SO(2k) valued transition functions on the overlaps U U , and let g Spin(2k) be the spin-lifts of these functions. For any module (representation), W , of the group Spin(2k) it is possible to dene an associated bundle by gluing the spaces U W together using the transition functions g (x) where x U U and g acts on W via the homomorphism Spin(2k) Hom(W, W ). We dene the topological space ((U W ))/ where is the equivalence dened by (x , v) (x , g (x ) v) for (x , v) U W and x = x . In this case, we will be interested in complex representations, even though N is a real bundle. This is because we will ultimately treat these spinors as U (n) bundles for some appropriate n so we want them to be complex spin-bundles (as the latter can also be shown to be a representation of U (n)). In particular, we will use the two inequivalent irreducible complex representations of Spin C (2k), which we will denote S + and S . The derives from the fact that these two are, in fact, components of a single irreducible representation, S = S + S , of the complexied Cliord algebra CC (R2k ) which restrict to two inequivalent representations of the subgroup Spin(2k) C(R2k ) CC (R2k ).74 Here the refer to the grading of the CC (R2k ) representation by the action of the grading75 element, CC (R2k ), so S = [(1 /2)] S. The construction of such Cliord modules is well understood and, as mentioned in the thesis, is intimately related to the K-theory of spheres and Bott periodicity. A slightly more detailed exposition is provided in Appendix B.2 below. These two bundles can be pulled back via : B(N ) to dene two bundles over B(N ) (S + (N )) B(N )

(S (N )) B(N )

74 The notation C(R2k ) indicates the Cliord algebra associated to R2k using the standard metric as a quadratic form and CC (R2k ) is the complexication of this algebra; see [LM]. This is the Cliord algebra familiar to physicists; cf [Pol2, App. B]. 75 In the physics literature this corresponds to the chirality grading and the grading element is = (2k+1) in dimension 2k.

129

B.2 The Thom Isomorphism in Codimension Two

B SOME K-THEORY TECHNICALITIES

where S (N ) refers to two dierent SpinC (2k) bundles dened on in the manner described above with bers given by the two inequivalent SpinC (2k) modules, S . On the compliment of the zero section of B(N ), namely the image of the embedding s0 : B(N ), an isomorphism between these bundles can be dened as follows. Let (S + (N ))|(x,v) and (S (N ))|(x,v) denote bers of the two bundles, isomorphic to S + and S , respectively. The coordinates (x, v) U B 2k are local coordinates on the total space B(N ) that correspond to a trivializing cover, {U } of B(N ) as a bundle over . The two chiral spinor representations admit an isomorphism via spinorial multiplication by a homogeneous element76 v R2k C(R2k ). That is, the local coordinates, (x, v) of B(N ), can be used to dene an isomorphism between the bers (S + )|(x,v) S + and (S )|(x,v) S given by treating = = v B 2k R2k as an automorphism of S via the inclusion R2k CC (R2k ). This is shown in [ABS], [LM, Prop. 3.6], and [Kar]; the details will not be presented here but it is not hard to gain some sense for why this is true. The grading isomorphism, (2k+1) , anticommutes with any odd degree homogeneous element (e.g. v or ) so the action of such an element, like v, on the representation will change the grading (recall that S + and S are dened as positive and negative eigenspaces of the chirality operator, respectively). That is, (v S ) = v ( S ) = v (S ) = v S so v S S . That v acting on the CC (R2k )-module S is an automorphism follows from the denition of CC (R2k ) since this algebra is dened by enforcing the relationship v v = Q(v) 1
2k

(397)

where v R , Q(v) is a quadratic form dening CC (R ) and 1 is the abstract unit in the algebra CC (R2k ). Thus, as long as Q(v) = 0 it is possible to invert v using v 1 = (Q(v))1 v. In CC (R2k ) the form Q corresponds to the standard metric on R2k (complexied in the complexied Cliord algebra CC (R2k )) so for any non-zero v the induced morphism on S is an isomorphism and for |v| = 1 it is isometric. Hence the map dened above is a ber-wise isomorphism between the bundles (S + ) and (S ) over B(N ) except at the point (x, 0) corresponding to the zero section of B(N ), namely . This morphism is actually a bundle morphism because it is dened continuously over all of B(N ) and it restricts to an isomorphism on S(N ) = B(N ) because the bers of S(N ) are given by S 2k1 = B 2k over which the morphism is invertible. Let us denote this morphism as : (S + (N )) (S (N )). Consider the K-theory class U = [ (S + (N )), (S (N )); ] K 0 (B(N ), S(N )) (398)

2k

77

Note that this class has compact support in the bers since, on the boundary of the unit ball, the two bundles (S + (N )) and (S (N )) are isomorphic. This isomorphism has winding number one since the map from the bers, S 2k1 , of S(N ) to Aut(S) is simply the inclusion S 2k1 R2k and it is clear that all the elements except the origin in R2k act invertibly on S (hence the homogeneous elements in Aut(S) have the topology of a sphere). Clearly this inclusion map has winding number one. 0 Since is compact, U can be extended to dene an element of Kcpt (N ) since B(N )/S(N ) is the 2k one point compactication of N . Since N has bers R which are homotopic to the bers, B 2k , of B(N ) the bundles (S ) can be extended to all of N and the morphism extends naturally since its original denition via multiplication by an element (x, v) Ui B 2k extends easily to multiplication by an element of U R2k . The resulting class (recall the abuse of notation for ),
0 is the image of U under the isomorphism between Kcpt (N ) and K 0 (B(N ), S(N )), is the Thom class of the bundle N .

0 UN = [ (S + (N )), (S (N ))] Kcpt (N )

(399)

B.2

The Thom Isomorphism in Codimension Two

In Section 3.5.2 we demonstrated how to explicitly construct a pair of line bundles over spacetime, X, 0 dening a class in Kcpt (X) that corresponded to the K-theory class of a codimension two brane (a D7brane in this case). This was done using only the Poincar dual cohomology class of the world-volume, e
76 In 77 Its

standard physics notation v = for = 0, . . . , 2k. action on a general Q(v, w) can be derived from bi-linearity and symmetry.

130

B.2 The Thom Isomorphism in Codimension Two

B SOME K-THEORY TECHNICALITIES

, of the D7-brane and dening the line bundles in terms of this class. In this Appendix we will show that this construction is actually a special instance of the Thom isomorphism. Recall that the Cliord algebra CC (R2k ) is isomorphic to C(2k ), the 2k 2k complex matrices [LM, k Tbl. III], which has a single irreducible representation given by C2 . This is a graded space of complex dimension 2k and can be given by the exterior algebra of Ck , namely Ck (which has dimension 2k ). The (chirality) grading of the representation is then given by the grading of the exterior algebra into even, even Ck , and odd, oddCk , parts. To make the connection with construction of spinor representations of SO(2k) in the physics literature [Pol2, App. B] let be an orthonormal basis for R2k and dene 1 z i 2 ( 2i + i 2i+1 ) for i = 0, . . . , k as a map from R2k to Ck . Note now that the standard anticommutation relations { , } = 2 now imply {z i , z j } = 0 so that z i z j = z j z i which is the expected algebraic structure for elements in Ck . Representations of the spin-subgroup of CC (R2k ) are built from states given by formal products of z i and z i (i.e. elements of Ck ). Recall that the grading matrix, (2k+1) , has eigenvalues 1 depending on whether there are an odd or even number of z i in an element of Ck . This explains the grading into even Ck and oddCk . The spin subgroup SpinC (2k) C 0 (R2k ) CC (R2k ) acts irreducibly on each of these gradings so even C and odd Ck provide two inequivalent representations of SpinC (2k). Specializing to the codimension two case, which we have already developed explicitly, gives C C (R2 ) which has as its irreducible representation C = 0 C 1 C. This space has two complex dimensions with basis 1 (here seen as the abstract unit of 0 C) and e = 1 C (here seen as the basis element for the complex dimension 1 C). Each of these complex dimensions is an irreducible representation of SpinC (2) so in this case S are each a copy of C (with S + coinciding with elements of degree zero, 0 C = {C 1}, and S coinciding with elements of degree one, 1 C = {C e}, in the above decomposition of C). Thus, the bundles S (N ) are just line bundles lifted from the SO(2) valued transition functions via the map ei (recall each element of SO(2) is just a two dimensional rotation matrix with a single angle ). In our codimension two construction discussed in the previous sections the line bundles L and K, restricted to the tubular neighborhood dieomorphic to N , coincide, in this construction, to (S + (N )) and (S (N )) dened on N . This follows because the relative Chern character of L and K is Poincar e dual to in X and hence is the Thom class of its normal bundle, N . This class restricts to the Euler class of the normal bundle on . Likewise one can see (here we have already restricted to and used s = id) 0 ch(S + (N )) ch(S (N )) = e(N ) = e(N ) A(N ) (400)

where the last equation holds because the roof genus (containing only forms of degree 4 or higher) is trivial for a line bundle. Note also that the section s 1 : X L K , restricted to N and considered as a morphism from K to L, has the same character as the isomorphism : S + (N ) S (N ) in that it is an isomorphism away from the zero section, . So, in fact, the earlier codimension two construction was a specic instance of the Thom isomorphism which admitted, in that dimension, a purely cohomological formulation.

131

C ANOMALIES AND THE INDEX THEOREM

Anomalies and the Index Theorem

An anomaly is said to occur if a symmetry in the classical action is no longer present in the associated quantum action. This can sometimes be understood in terms of a failure to lift a classical symmetry from its representation on the projective quantum Hilbert space to the underlying linear Hilbert space (see [Wei, ch. 2] or [vdB] for a discussion of this). In the present context the anomalies that will be of relevance are anomalies in gauge symmetries associated with gauge theories coupled chirally to fermions. These anomalies are so called local anomalies as they are anomalies in a local symmetry. They are closely connected however with global anomalies arising from the anomalous breaking of a global symmetry of the classical action. The latter is more easily understood and thus will provide a more transparent introduction to the subject. Anomalies will occupy an important place in this thesis because they are related to topological invariants of principle ber bundles associated with the classical eld theory. It is this connection which ultimately allows of the interpretation of D-brane charges in terms of K-theory classes. When a local symmetry is anomalous the associated quantum eective action is no longer gauge-invariant which makes the theory inconsistent (since gauge degrees of freedom are unphysical). Thus the requirement that gauge anomalies must somehow vanish provides a strong constraint on generating sensible quantum theories. This will be used, in the exposition in the main text, to x certain terms of the low-energy action associated to D-branes. In this appendix anomalies will be introduced with an emphasis on a eld theoretic calculation of the Abelian anomaly. Once the topological underpinnings of the anomaly have emerged however additional tools become available that are discussed briey in Sections 2.5.1 and 2.5.2. We will not develop the descent procedure or its mathematical underpinnings in this appendix as this is a rather complex and involved subject and hence best left to the extensive literature [AGG1] [AGG2] [Ber] [dAI] [HT] [Nak]. In this appendix our main goal will be to show that the Abelian anomaly, which is used as the starting point for the descent procedure in Section 2.5.1, is given by the index of the relevant Dirac operator. This will demonstrate the connection between gauge anomalies and vector bundle topology (and ultimately K-theory). The introduction to anomalies below follows [Nak, ch. 13] with some additional input from [AGG1] and [AGG2]. A more eld-theoretic perspective can be found in [PS, ch. 19].

C.1

Abelian Anomaly

A classical eld theory with a single fermion species coupled to a background electromagnetic eld has action S=i dd x ( A ) = i dd x + ( A )+ + ( A ) (401)

By a background electromagnetic gauge eld it is intended that, in the quantum theory, this eld will, for now, not be integrated over and moreover will have no corresponding kinetic terms. Note also that the normal gauge coupling iA has been replaced by A , absorbing the i in the denition of A so the latter is now anti-hermitian (complex). This is counter to standard physics conventions but simplies the overall 1 treatment. In the last equality we have used the decomposition I = P+ + P where P = 2 (1 d ) are the positive and negative chirality projectors and d is the d-dimensional gamma matrix. + = P+ and + = P and likewise for the negative chirality components. It is not hard to see that with these denitions the last equality above holds. The action (401) is invariant under the global transformations ei ei + ei + ei
d

ei ei
d

It is easy to see that the second set of transformations correspond to chiral rotations + + ei ei

132

C.1 Abelian Anomaly

C ANOMALIES AND THE INDEX THEOREM

Because this is a symmetry of the classical action Noethers procedure can be used to derive an associated d current that must be conserved. To do this the elds are transformed to = ei(x) and d ei(x) . Because the transformations are now spacetime dependent via (x) this is no longer a real symmetry of the classical action but an associated variation can be derived to rst order. This rst order term must vanish when the equations of motion hold because they extremize the action. Letting = 0 + 0 with 0 = i(x) d 0 (i.e. considering only an innitesimal transformation gives S S 0 + O( 2 ) dd x ( A )(i(x) d ) dd x i (x)J d

S[ , ] = S[, ] + 0 = S[, ] + = S[, ] + = S[, ]

+
0

(402) (403) (404) (405)

dd x i(x) d ( A ) +

dd x i (x) d S[, ] + dd x i(x) J d

Thus classically the second term on the left hand side is expected to vanish for all (x) when the equation of motions hold which implies that the current J d = 0. The current J d is the noether current associated with the continuous symmetry given by a chiral rotation. When the theory is quantized the identity J d = 0 would normally be expected to hold as an operator identity J d = 0. This will not be the case in an anomalous theory as will be shown below. The question of whether this symmetry lifts to the quantum theory can be studied using path integral techniques. Consider the path integral that denes this theory Z[, ] DD eiS[,] (406)

Since the transformation given above is nothing more than a change of variable it should not eect the path integral dened above. Z[, ] = DD eiS[,] = D D eiS[ , ] = Z[ , ] (407)

Writing S[ , ] = S[, ] + S with S given in (405) the above can be re-written DD eiS[,] = = D D eiS[ , ] = D D eiS[,] eiS

(408) (409)

DD (J ) eiS[,] (1 + iS + O((S)2 ))

J is the Jacobian associated to the change of integration variables; this will ultimately prove to be the source of the anomalous non-conservation of J d . J is most easily calculated by enacting another change of integration variable. To see this note that can be expanded in an orthonormal eigenbasis of the hermitian operator iD i A as = n an n . 78 Likewise we can expand = n bn . The coecients an and bn are necessarily grassmanian since the eigenfunctions are merely functions and have normal statistics. This transformation is unitary since it corresponds to change of basis (the measure in the path integral is normally dened with respect to either a momentum or position eigenbasis of ) and hence does not generate a Jacobian. To see how the measure DD = dan dbn transforms under the chiral rotation note = n an n so, by the orthonormality of the eigenbasis
78 To see that this operator is hermitian note that A, as an element of a unitary Lie algebra, is anti-hermitian while is anti-hermitian as an operator on functions since (schematically) Z Z dx f g = (f )g

check this

133

C.1 Abelian Anomaly

C ANOMALIES AND THE INDEX THEOREM

an = n | = =
m

dd x
m

am n (x)ei(x) m (x)

(410) (mn + Cmn )am = (I + C)a (411)


m

am nm +

dd x
k=1

i(x) k!

k n (x)( d )k m (x) =

The Jacobian associated with this transformation is Ja = [det(I + C)]1 = exp(tr(ln(I + C))) (412)
2

exp(tr(C)) = (I tr(C) + O((tr(C)) )

(413)

In the second line terms of higher order than rst in C have been consistently ignored (the log has been expanded and then the exponential). This is because, from (410), it can be seen that C only contains terms rst and higher order in (x) so higher powers of C will contain terms of second and higher order in (x). The expectation value of J d will determined by matching terms of the same order in (409) and since J d is coupled to only a rst power of (x) only such terms will be needed. The Jacobian for bn will be exactly the same and so the combined Jacobian will have an additional factor of 2 in front of the leading order term J = Ja Jb = (I 2tr(C) + O((tr(C))2 ) Combining the above with (409) gives DD eiS[,] = = = DD (I 2tr(C) + O((tr(C))2 ) eiS[,] (1 + iS + O((S)2 )) DD eiS[,] (I 2tr(C) + iS + O((x)2 )) DD eiS[,] (I 2tr(C) i dd x i(x) J d + O((x)2 ))

(414) (415) (416)

Matching terms of rst order in (x) and taking the trace using the denition of Cmn given in (411) yields the following identity which is nothing but the restriction on terms of order (x) imposed by the invariance of the path integral under a change of variables DD eiS[,] i dd x i(x) J d = 2 = 2 DD eiS[,] tr(C) DD eiS[,]
dd x (i(x))n (x) d n (x) n

(417) (418)

Dividing both sides by the partition function which is a standard normalization in quantum eld theory and noting that the modes n and n are not actually elds but merely solutions to an eigenvalue problem so they factor out of the path integral gives (here denotes a quantum correlation function) dd x (x) J d = 2i J d = 2i
dd x ((x))n (x) d n (x) n n (x) d n (x) n

(419) (420)

The second line is a consequence of the arbitrariness of (x). The right-hand side is divergent and must be regulated if it is to be calculated. This can be done and will give a non-zero value [Nak, ch. 19]. For the purpose of this thesis, however, it is more interesting to determine what the equation above 134

C.1 Abelian Anomaly

C ANOMALIES AND THE INDEX THEOREM

means rather than what its exact value is. To gain some deeper understanding of the equation above it is interesting to integrate it (i.e. calculate the integrated anomaly) dd x J d = 2i = 2i
dd x n (x) d n (x) n

(421) (422)

n | d n

The n were dened to be eigenvectors of iD with eigenvalues n . Since { d, iD} = 0 the eigenvalue of d n = d n can easily be seen to be n from iD( d n ) = d iDn = d n n = n ( d n ). Since iD is hermitian so (iD) = iD it is not hard to see that eigenvectors with dierent eigenvalues must be orthogonal. Let n and m have unequal non-zero eigenvalues, n and m . It is important below that the eigenvalues of iD are real as it is a hermitian operator. Then iD m |n = m |iDn (423) (424) (425)

m m |n = n m |n

Since m = n , m |n = 0. This implies that for n with non-zero eigenvalue n | d n = 0 so the sum in (420) reduces to a sum over only the zero modes. The zero modes are invariant under d as seen 0 by the arguments above so we can diagonalize the set of n (n with n = 0) relative to d and arrive d d at a set of zero modes that are eigenvalues of . Since has eigenvalues 1 this will give the following decomposition for (420) J d = 2i 0, n, +
n,0

0, n
n,0,

= 2i(v+ v ) 2i(ind (iD))

(426)

Where the sum is only over the zero modes and the modes themselves have been separated into positive and negative eigenvectors of d . The last equality uses the orthonormality of modes to show that the sum is actually equal to the number of zero modes with positive chirality minus the number of zero modes with negative chirality. The latter quantity is dened to be the analytic index of the Dirac operator. This is, remarkably, a topological invariant of the principle ber bundle associated with the spinors and is given by the famous Atiyah-Singer index theorem [LM] [AS1] [Nak] Theorem C.1 (Atiyah-Singer Index Formula (for twisted spin bundle)). The index of the Dirac operator, iD, coupled to a vector bundle E on a compact manifold M is given by ind(iD) =
M

A(T M )ch(E)|vol

(427)

The content of this theorem, which is a very specic case of a much more general theorem, is that to calculate the index of the Dirac operator, which is nothing more than the dierence between the dimension of its positive and negative chirality zero eigenspaces, it is necessary to integrate certain characteristic classes over the entire manifold. These classes are associated to the tangent space of the manifold and also to the gauge-elds which couple to the spinors. The later is manifest in the formula above in the form of E. The spinor eld transforms under the spin representation of the space-time tangent bundle (and hence is a spinor) but also under the gauge-group of the gauge eld it couples to. There is a subtlety here: although the gauge eld is dened as a background eld and given no dynamic content it still represents, classically, a topological object dened by a principle ber bundle and an associated vector bundle the latter of which appears above as E. As such, the eld, , is classical a section of a tensor product between the spin-bundle, p : S M carrying a representation of Spin(d) for d the dimension of space-time and : E M , carrying a representation of the gauge group. Although the calculation undertaken above could be completed to yield a specic case of the formula above, the latter is far more powerful as it incorporates possibly non-trivial topology of the base manifold 135

C.1 Abelian Anomaly

C ANOMALIES AND THE INDEX THEOREM

and also couplings to additional elds such as gravity. That is to say, it is valid even if the Dirac operator were to include a spin-connection so that it would be of the form iD = i e ( A ) where is a spin-connection. Although the derivation above was not rigorous it should be clear that it was centered almost entirely about the fact that the action was of the form |iD and the particular properties of the operator iD and hence can be generalized to the non-trivial setting just mentioned. This means that even in this settings it is possible to calculate the anomaly using the index theorem and that the anomaly is always related to the index of the Dirac operator. In four-dimensional Euclidean spacetime the the index formula reduces to ind(iD) = d4 x ch(F ) = 1 2 d4 x T r iF 2
2

(428)

Here F = dA is the eld strength associated to the connection A. It is not hard to check that the value given above coincides with that calculated directly by regulating the integral in (420) in a gauge-invariant way [Nak], [PS]. The anomaly can be calculated in several ways which are discussed at length in [PS, ch. 19]. The relevance to this thesis lies in the topological origin of the anomaly, its relation to K-theory, and its use in calculating the non-Abelian anomaly. The latter is discussed briey in Sections 2.5.1 and 2.5.2. The reader who wishes to pursue the non-Abelian anomaly in more detail should consult the references given above.

136

REFERENCES

REFERENCES

References
[ABS] M.F. Atiyah, R. Bott, and A. Shapiro. Cliord modules. Topology, 3:338, 1964. [AGG1] Luis Alvarez-Gaume and Paul H. Ginsparg. The topological meaning of nonabelian anomalies. Nucl. Phys., B243:449, 1984. [AGG2] Luis Alvarez-Gaume and Paul H. Ginsparg. The structure of gauge and gravitational anomalies. Ann. Phys., 161:423, 1985. [Alv] Orlando Alvarez. Topological quantization and cohomology. Commun. Math. Phys., 100:279, 1985. Available from: http://projecteuclid.org/Dienst/UI/1.0/Summarize/euclid.cmp/1103943448. Philip Argyres. Introduction to global supersymmetry [online]. http://www.physics.uc.edu/%7Eargyres/661/index.html. Available from:

[Arg] [AS1] [AS2] [Ati] [Bac]

M. F. Atiyah and I. M. Singer. The index of elliptic operators. 3. Annals Math., 87:546604, 1968. Michael Atiyah and Graeme Segal. Twisted k-theory. 2004. arXiv:math.KT/0407054. M. F. Atiyah. K-theory. Benjamin, New York, USA, 1967. Constantin P. Bachas. Lectures on d-branes. 1998. arXiv:hep-th/9806199.

[BCM+ ] Peter Bouwknegt, Alan L. Carey, Varghese Mathai, Michael K. Murray, and Danny Stevenson. Twisted k-theory and k-theory of bundle gerbes. Commun. Math. Phys., 228:1749, 2002. arXiv:hep-th/0106194. [BDG] [BDL] [Ber] [BM] [Bry] [BS] [BT] [CH] [CJM] [Cra] [CV] [CW] Constantin P. Bachas, Michael R. Douglas, and Michael B. Green. Anomalous creation of branes. JHEP, 07:002, 1997. arXiv:hep-th/9705074. Micha Berkooz, Michael R. Douglas, and Robert G. Leigh. Branes intersecting at angles. Nucl. Phys., B480:265278, 1996. arXiv:hep-th/9606139. R. A. Bertlmann. Anomalies in quantum eld theory. International series of monographs on physics: 91. Oxford, UK: Clarendon, 1996. P. Bouwknegt and V. Mathai. D-branes, b-elds and twisted k-theory. JHEP, 03:007, 2000. arXiv:hep-th/0002023. J. L. Brylinski. Loop spaces, characteristic classes and geometric quantization. Progress in mathematics, 107. Boston, USA: Birkhaeuser, 1993. Tom Banks and Leonard Susskind. Brane - antibrane forces. 1995. arXiv:hep-th/9511194. Raoul Bott and Loring W. Tu. Dierential Forms in Algebraic Topology. Springer-Verlag, New York, USA, 1982. Jr. Callan, Curtis G. and Jerey A. Harvey. Anomalies and fermion zero modes on strings and domain walls. Nucl. Phys., B250:427, 1985. Alan L. Carey, Stuart Johnson, and Michael K. Murray. arXiv:hep-th/0204199. Holonomy on d-branes. 2002.

Ben Craps. D-branes and boundary states in closed string theories. 2000. arXiv:hep-th/0004198. Ralph L. Cohen and Alexander A. Voronov. arXiv:math.gt/0503625. Notes on string topology. 2005.

Alan L. Carey and Bai-Ling Wang. Thom isomorphism and push-forward map in twisted ktheory. 2005. arXiv:math.kt/0507414. 137

REFERENCES

REFERENCES

[CY] [dAI]

Yeuk-Kwan E. Cheung and Zheng Yin. Anomalies, branes, and currents. Nucl. Phys., B517:69 91, 1998. arXiv:hep-th/9710206. Jose A. de Azcarraga and Jose M. Izquierdo. Lie groups, Lie algebras, cohomology and some applications in physics. Cambridge Monographs on Mathematical Physics. Cambridge University Press, 1995. Peter Donavan and Max Karoubi. Graded brauer groups and k-theory with local coecients. Publications Mathmatiques de lIHS, 38:525, 1970. Available from: http://archive.numdam.org/article/PMIHES_1970__38__5_0.pdf.

[DK]

[DMW] Duiliu-Emanuel Diaconescu, Gregory W. Moore, and Edward Witten. E(8) gauge theory, and a derivation of k-theory from m- theory. Adv. Theor. Math. Phys., 6:10311134, 2003. arXiv:hep-th/0005090. [Dou] [dR] [FH] [For] [Fre1] [Fre2] [FS] [FW] [Gaw] [GGR] [GH] [GHM] [GR] [Hat1] [Hat2] [Hit] [HM] Michael R. Douglas. Branes within branes. 1995. arXiv:hep-th/9512077. Georges de Rham. Dierentiable Manifolds. Springer-Verlag, 1984. Forms, Currents, Harmonic Forms. Daniel S. Freed and Michael J. Hopkins. On ramond-ramond elds and k-theory. JHEP, 05:044, 2000. arXiv:hep-th/0002027. Stefan Forste. Strings, branes and extra dimensions. arXiv:hep-th/0110055. Fortsch. Phys., 50:221403, 2002.

Daniel S. Freed. Dirac charge quantization and generalized dierential cohomology. 2000. arXiv:hep-th/0011220. Daniel S. Freed. K-theory in quantum eld theory. 2002. arXiv:math-ph/0206031. Stefan Fredenhagen and Volker Schomerus. Branes on group manifolds, gluon condensates, and twisted k-theory. JHEP, 04:007, 2001. arXiv:hep-th/0012164. Daniel S. Freed and Edward Witten. arXiv:hep-th/9907189. Anomalies in string theory with d-branes. 1999.

K. Gawedzki. Topological actions in two-dimensional quantum eld theories. in *Cargese 1987, PROCEEDINGS, Nonperturbative Quantum Field Theory* 101-141. Matthias R. Gaberdiel, Terry Gannon, and Daniel Roggenkamp. The d-branes of su(n). JHEP, 07:015, 2004. arXiv:hep-th/0403271. Phillip Griths and Joseph Harris. Principles of Algebraic Geometry. John Wiley and Sons, Inc., New York, USA, 1978. Michael B. Green, Jerey A. Harvey, and Gregory W. Moore. I-brane inow and anomalous couplings on d-branes. Class. Quant. Grav., 14:4752, 1997. arXiv:hep-th/9605033. Krzysztof Gawedzki and Nuno Reis. Wzw branes and gerbes. Rev. Math. Phys., 14:12811334, 2002. arXiv:hep-th/0205233. Allen Hatcher. Vector bundles and k-theory. http://www.math.cornell.edu/~hatcher/VBKT/VB.pdf. Allen Hatcher. Algebraic Topology. Cambridge University Press, 2001. http://www.math.cornell.edu/~hatcher/AT/ATpage.html. Available from:

Available from:

Nigel J. Hitchin. Lectures on special lagrangian submanifolds. 1999. arXiv:math.dg/9907034. S. F. Hassan and Ruben Minasian. D-brane couplings, rr elds and cliord multiplication. 2000. arXiv:hep-th/0008149. 138

REFERENCES

REFERENCES

[Hor] [HS] [HT] [Kap] [Kar] [KS] [Lan1] [Lan2] [Lan3] [LM] [Man] [MM] [MMS] [Moe] [Moo1] [Moo2] [MS] [MW] [Nak] [Nas] [OS] [PCJ]

Petr Horava. Type iia d-branes, k-theory, and matrix theory. Adv. Theor. Math. Phys., 2:1373 1404, 1999. arXiv:hep-th/9812135. M. J. Hopkins and I. M. Singer. Quadratic functions in geometry, topology, and m-theory. 2002. arXiv:math.at/0211216. Marc Henneaux and Claudio Teitelboim. Quantization of Gauge Systems. Princeton University Press, 1994. Anton Kapustin. D-branes in a topologically nontrivial b-eld. Adv. Theor. Math. Phys., 4:127 154, 2000. arXiv:hep-th/9909089. Max Karoubi. K-theory: An introduction. Springer-Verlag, New York, USA, 1978. Igor Kriz and Hisham Sati. Type iib string theory, s-duality, and generalized cohomology. 2004. arXiv:hep-th/0410293. N.P. Landsman. Lecture notes on c*-algebras and k-theory, part i [online]. Available from: http://staff.science.uva.nl/~npl/CK.pdf. N.P. Landsman. Lecture notes on c*-algebras and k-theory, part ii [online]. Available from: http://www.science.uva.nl/~npl/CKmaster.pdf. Serge Lang. Dierential Manifolds. Addison-Wesley Series in Mathematics. Addison-Wesley Publishing Company, Reading, Massachusetts, 1972. H. Blaine Lawson and Marie-Louise Michelsohn. Spin Geometry. Princeton University Press, Princeton, New Jersey, USA, 1989. Juan Jose Manjarin. Topics on d-brane charges with b-elds. Int. J. Geom. Meth. Mod. Phys., 1:N4, 2004. arXiv:hep-th/0405074. Ruben Minasian and Gregory W. Moore. K-theory and ramond-ramond charge. JHEP, 11:002, 1997. arXiv:hep-th/9710230. Juan M. Maldacena, Gregory W. Moore, and Nathan Seiberg. D-brane instantons and k-theory charges. JHEP, 11:062, 2001. arXiv:hep-th/0108100. Ieke Moerdijk. Introduction to the language of stacks and gerbes. 2002. arXiv:math.AT/0212266. Gregory W. Moore. Kitp talk notes: D-branes, rr-elds and k-theory. http://online.itp.ucsb.edu/online/mp01/moore1/. Available from:

Gregory W. Moore. K-theory from a physical perspective. 2003. arXiv:hep-th/0304018. John Milnor and James D. Stashe. Characteristic Classes. Princeton University Press, 1974. Gregory W. Moore and Edward Witten. Self-duality, ramond-ramond elds, and k-theory. JHEP, 05:032, 2000. arXiv:hep-th/9912279. M. Nakahara. Geometry, topology and physics. Graduate student series in physics. Hilger, Bristol, UK, 1990. C. Nash. Dierential topology and quantum eld theory. Academic Press Limited, London, UK, 1991. Kasper Olsen and Richard J. Szabo. Constructing d-branes from k-theory. Adv. Theor. Math. Phys., 3:8891025, 1999. arXiv:hep-th/9907140. Joseph Polchinski, Shyamoli Chaudhuri, and Cliord V. Johnson. Notes on d-branes. 1996. arXiv:hep-th/9602052.

139

REFERENCES

REFERENCES

[Pes] [Pol1] [Pol2] [Pol3] [PS] [Sch] [Seg] [Sen1] [Sen2] [Sen3] [Sen4] [Sen5] [Sre] [SS] [SW] [Sza1] [Sza2] [TS]

I. Pesando. On the eective potential of the dp dp-bar system in type ii theories. Mod. Phys. Lett., A14:15451564, 1999. arXiv:hep-th/9902181. J. Polchinski. String theory. Vol. 1: An introduction to the bosonic string. Univ. Pr., Cambridge, UK, 1998. J. Polchinski. String theory. Vol. 2: Superstring theory and beyond. Univ. Pr., Cambridge, UK, 1998. Joseph Polchinski. Dirichlet-branes and ramond-ramond charges. Phys. Rev. Lett., 75:4724 4727, 1995. arXiv:hep-th/9510017. Michael E. Peskin and D. V. Schroeder. An Introduction to quantum eld theory. AddisonWesley, Reading, USA, 1995. Urs Schreiber. From loop space mechanics to nonabelian strings [online]. Available from: http://www-stud.uni-essen.de/~sb0264/SchreiberDiss.pdf. PhD Thesis. G. Segal. Topological structures in string theory. Phil. Trans. Roy. Soc. Lond., A359:13891398, 2001. Ashoke Sen. So(32) spinors of type i and other solitons on brane- antibrane pair. JHEP, 09:023, 1998. arXiv:hep-th/9808141. Ashoke Sen. Stable non-bps bound states of bps d-branes. arXiv:hep-th/9805019. JHEP, 08:010, 1998.

Ashoke Sen. Tachyon condensation on the brane antibrane system. JHEP, 08:012, 1998. arXiv:hep-th/9805170. Ashoke Sen. Descent relations among bosonic d-branes. Int. J. Mod. Phys., A14:40614078, 1999. arXiv:hep-th/9902105. Ashoke Sen. Non-bps states and branes in string theory. 1999. arXiv:hep-th/9904207. Mark Srednicki. Iib or not iib. JHEP, 08:005, 1998. arXiv:hep-th/9807138. Claudio A. Scrucca and Marco Serone. Anomalies and inow on d-branes and o-planes. Nucl. Phys., B556:197221, 1999. arXiv:hep-th/9903145. John H. Schwarz and Edward Witten. Anomaly analysis of brane-antibrane systems. JHEP, 03:032, 2001. arXiv:hep-th/0103099. Richard J. Szabo. Busstepp lectures on string theory: An introduction to string theory and d-brane dynamics. 2002. arXiv:hep-th/0207142. Richard J. Szabo. Superconnections, anomalies and non-bps brane charges. J. Geom. Phys., 43:241292, 2002. arXiv:hep-th/0108043. Peter Teichner and Stephan Stolz. What is an elliptic object? In U. Tillmann, editor, Proceedings of the 2002 Oxford Symposium in Honour of the 60th Birthday of Graeme Segal. Cambridge University Press, 2004. Available from: http://math.ucsd.edu/~teichner/Papers/Oxford.pdf. Tadaoki Uesugi. Worldsheet description of tachyon condensation in open string theory. 2002. arXiv:hep-th/0302125. Erik van den Ban. Representation theory and applications in classical quantum mechanics [online]. Available from: http://www.math.uu.nl/people/ban/lecnotes/repq.pdf. Steven Weinberg. The Quantum theory of elds. Vol. 1: Foundations. Univ. Pr., Cambridge, UK, 1995.

[Ues] [vdB] [Wei]

140

REFERENCES

REFERENCES

[Wit1] [Wit2] [Wit3] [Wit4]

Edward Witten. Small instantons in string theory. arXiv:hep-th/9511030.

Nucl. Phys., B460:541559, 1996.

Edward Witten. D-branes and k-theory. JHEP, 12:019, 1998. arXiv:hep-th/9810188. Edward Witten. Duality relations among topological eects in string theory. JHEP, 05:031, 2000. arXiv:hep-th/9912086. Edward Witten. Overview of k-theory applied to strings. Int. J. Mod. Phys., A16:693706, 2001. arXiv:hep-th/0007175.

141

You might also like