A Window into Zeta and Modular Physics (Mathematical Sciences Research Institute Publications 57)

This book provides an introduction, with applications, to three interconnected mathematical topics: • zeta functions in ...

Author: Klaus Kirsten | Floyd L. Williams | Editors

18 downloads 633 Views 5MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

This book provides an introduction, with applications, to three interconnected mathematical topics: • zeta functions in their rich variety: those of Riemann, Hurwitz, Barnes, Epstein, Selberg, and Ruelle, plus graph zeta functions; • modular forms: Eisenstein series, Hecke and Dirichlet L-functions, Ramanujan’s tau function, and cusp forms; • vertex operator algebras: correlation functions, quasimodular forms, modular invariance, rationality, and some current research topics including higher-genus conformal field theory. Applications of the material to physics are presented, including Kaluza–Klein extra-dimensional gravity, bosonic string calculations, a Cardy formula for black hole entropy, Patterson–Selberg zeta function expressions of one-loop quantum field and gravity partition functions, Casimir energy calculations, atomic Schr¨ odinger operators, Bose–Einstein condensation, heat kernel asymptotics, random matrices, quantum chaos, elliptic and theta function solutions of Einstein’s equations, a soliton–black hole connection in two-dimensional gravity, and conformal field theory.

Mathematical Sciences Research Institute Publications

57

A Window into Zeta and Modular Physics

Mathematical Sciences Research Institute Publications 1 2 3 4 5 6 7 8 9 10–11 12–13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57

Freed/Uhlenbeck: Instantons and Four-Manifolds, second edition Chern (ed.): Seminar on Nonlinear Partial Differential Equations Lepowsky/Mandelstam/Singer (eds.): Vertex Operators in Mathematics and Physics Kac (ed.): Infinite Dimensional Groups with Applications Blackadar: K-Theory for Operator Algebras, second edition Moore (ed.): Group Representations, Ergodic Theory, Operator Algebras, and Mathematical Physics Chorin/Majda (eds.): Wave Motion: Theory, Modelling, and Computation Gersten (ed.): Essays in Group Theory Moore/Schochet: Global Analysis on Foliated Spaces, second edition Drasin/Earle/Gehring/Kra/Marden (eds.): Holomorphic Functions and Moduli Ni/Peletier/Serrin (eds.): Nonlinear Diffusion Equations and Their Equilibrium States Goodman/de la Harpe/Jones: Coxeter Graphs and Towers of Algebras Hochster/Huneke/Sally (eds.): Commutative Algebra Ihara/Ribet/Serre (eds.): Galois Groups over Q Concus/Finn/Hoffman (eds.): Geometric Analysis and Computer Graphics Bryant/Chern/Gardner/Goldschmidt/Griffiths: Exterior Differential Systems Alperin (ed.): Arboreal Group Theory Dazord/Weinstein (eds.): Symplectic Geometry, Groupoids, and Integrable Systems Moschovakis (ed.): Logic from Computer Science Ratiu (ed.): The Geometry of Hamiltonian Systems Baumslag/Miller (eds.): Algorithms and Classification in Combinatorial Group Theory Montgomery/Small (eds.): Noncommutative Rings Akbulut/King: Topology of Real Algebraic Sets Judah/Just/Woodin (eds.): Set Theory of the Continuum Carlsson/Cohen/Hsiang/Jones (eds.): Algebraic Topology and Its Applications Clemens/Koll´ ar (eds.): Current Topics in Complex Algebraic Geometry Nowakowski (ed.): Games of No Chance Grove/Petersen (eds.): Comparison Geometry Levy (ed.): Flavors of Geometry Cecil/Chern (eds.): Tight and Taut Submanifolds Axler/McCarthy/Sarason (eds.): Holomorphic Spaces Ball/Milman (eds.): Convex Geometric Analysis Levy (ed.): The Eightfold Way Gavosto/Krantz/McCallum (eds.): Contemporary Issues in Mathematics Education Schneider/Siu (eds.): Several Complex Variables Billera/Bj¨ orner/Green/Simion/Stanley (eds.): New Perspectives in Geometric Combinatorics Haskell/Pillay/Steinhorn (eds.): Model Theory, Algebra, and Geometry Bleher/Its (eds.): Random Matrix Models and Their Applications Schneps (ed.): Galois Groups and Fundamental Groups Nowakowski (ed.): More Games of No Chance Montgomery/Schneider (eds.): New Directions in Hopf Algebras Buhler/Stevenhagen (eds.): Algorithmic Number Theory: Lattices, Number Fields, Curves and Cryptography Jensen/Ledet/Yui: Generic Polynomials: Constructive Aspects of the Inverse Galois Problem Rockmore/Healy (eds.): Modern Signal Processing Uhlmann (ed.): Inside Out: Inverse Problems and Applications Gross/Kotiuga: Electromagnetic Theory and Computation: A Topological Approach Darmon/Zhang (eds.): Heegner Points and Rankin L-Series Bao/Bryant/Chern/Shen (eds.): A Sampler of Riemann–Finsler Geometry Avramov/Green/Huneke/Smith/Sturmfels (eds.): Trends in Commutative Algebra Goodman/Pach/Welzl (eds.): Combinatorial and Computational Geometry Schoenfeld (ed.): Assessing Mathematical Proficiency Hasselblatt (ed.): Dynamics, Ergodic Theory, and Geometry Pinsky/Birnir (eds.): Probability, Geometry and Integrable Systems Albert/Nowakowski (eds.): Games of No Chance 3 Kirsten/Williams (eds.): A Window into Zeta and Modular Physics

Volumes 1–4, 6–8, and 10–27 are published by Springer-Verlag

A Window into Zeta and Modular Physics

Edited by

Klaus Kirsten Baylor University

Floyd L. Williams University of Massachusetts, Amherst

Klaus Kirsten Department of Mathematics Baylor University Waco, TX 76798 United States Klaus [email protected]

Floyd L. Williams Department of Mathematics and Statistics University of Massachusetts Amherst, MA 01003-9305 United States [email protected]

Silvio Levy (Series Editor ) Mathematical Sciences Research Institute Berkeley, CA 94720 [email protected] The Mathematical Sciences Research Institute wishes to acknowledge support by the National Science Foundation and the Pacific Journal of Mathematics for the publication of this series.

cambridge university press Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, S˜ ao Paulo, Delhi, Dubai, Tokyo Cambridge University Press 32 Avenue of the Americas, New York, NY 10013-2473, USA www.cambridge.org Information on this title: www.cambridge.org/9780521199308 c Mathematical Sciences Research Institute 2010

This publication is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press.

First published 2010 Printed in the United States of America A catalog record for this publication is available from the British Library.

ISBN 978-0-521-19930-8 hardback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party Internet Web sites referred to in this publication and does not guarantee that any content on such Web sites is, or will remain, accurate or appropriate.

A Window Into Zeta and Modular Physics MSRI Publications Volume 57, 2010

Contents Introduction K LAUS K IRSTEN AND F LOYD L. W ILLIAMS

1

Introductory Lectures Lectures on zeta functions, L-functions and modular forms with some physical applications F LOYD L. W ILLIAMS

7

Basic zeta functions and some applications in physics K LAUS K IRSTEN

101

Zeta functions and chaos AUDREY T ERRAS

145

Vertex operators and modular forms G EOFFREY M ASON AND M ICHAEL T UITE

183

Research Lectures Applications of elliptic and theta functions to FRLW cosmology with cosmological constant J ENNIE D’A MBROISE

279

Integrable systems and 2D gravitation: How a soliton illuminates a black hole S HABNAM B EHESHTI

295

Functional determinants in higher dimensions using contour integrals K LAUS K IRSTEN

307

The role of the Patterson–Selberg zeta function of a hyperbolic cylinder in three-dimensional gravity with a negative cosmological constant F LOYD L. W ILLIAMS

vii

329


Introduction Some exciting, bold new cooperative explorations of various interconnections between traditional domains of “pure” mathematics and exotic new developments in theoretical physics have continued to emerge in recent years. The beautiful interlacing of theory and application, and cross-discipline interaction has led, as is usual, to notable, fruitful, and bonus outcomes. These interconnections range from topology, algebraic geometry, modular forms, Eisenstein series, zeta functions, vertex operators, and knot theory to gauge theory, strings and branes, quantum fields, cosmology, general relativity, and Bose–Einstein condensation. They are broad enough in scope to present the average reader with not only a measure of enchantment but with some mild bewilderment as well. A new journal, Communications in Number Theory and Physics, has recently been launched to follow and facilitate interactions and dynamics between these two disciplines, for example. Various books that are now available, in addition to an array of conference and workshop activity, accent this fortunate merger of mathematics and physical theory and assist greatly in bridging the divide, although in some cases the themes are pitched at a level more suitable for advanced readers and researchers. In an attempt to further bridge the divide, at least in some modest way for students and non-experts, and to provide a window into this adventurous arena of intertwining ideas, a graduate workshop entitled “A Window into Zeta and Modular Physics” was presented at MSRI during the period from 16 to 27 June 2008. The workshop consisted of daily expository lectures, speakers’ seminar lectures (where the material was more technical and represented their research, but which at the same time did connect to and enlarge on the daily lectures), and four special student lectures. Given the excellent preparation and presentation of the lectures, it was proposed that it would be of further benefit to the students (especially given their enthusiastic response), and to the mathematical physics community in general, if the lectures were eventually molded in some form as a book. Thus the present volume, with the workshop title and similar style, evolved. Besides the editors, there were three workshop speakers: Geoffrey Mason, Audrey Terras, and Michael Tuite. The student speakers were Jennie 1

2

INTRODUCTION

D’Ambroise, Shabnam Beheshti, Savan Kharel, and Paul Nelson. Among these four, Jennie and Shabnam (who have since earned their PhDs) accepted the invitation to have their lectures appear here. The volume consists of two parts. Part I contains basic, expository lectures, except that it was found convenient to also include some seminar material in the combined presentations of Geoff and Michael, and thus to make those one long set of lectures. Part II consists of speakers’ seminar lectures, where we have included the lectures of Jennie and Shabnam. A more specific description of Part I is as follows. Lectures of Floyd Williams cover topics such as the Riemann zeta function (proof of its meromorphic continuation and functional equation), Euler products (in particular, of Hecke and Dirichlet L-functions), holomorphic modular forms, Dedekind’s eta function, the quasimodular form G2 , the Rademacher–Zuckerman formula for the Fourier coefficients of forms of negative weight, and non-holomorphic Eisenstein series, with some physical applications including gravity in extra dimensions, finite temperature zeta functions, the zeta regularization of Casimir energy, a determinant (or path integral) computation in Bosonic string theory, and an abstract Cardy formula for black hole entropy, based on the asymptotic behavior of Fourier coefficients of modular forms of zero weight. Several appendices are included. These provide a proof of the Poisson summation formula, the Jacobi inversion formula, the Fourier expansion of a holomorphic periodic function, etc., and thus they render further completeness of the material. The lectures, in general, provide some background material for some of the other lectures, with some overlap. Further applications of zeta functions — those of Hurwitz, Barnes, and Epstein, and also spectral zeta functions — are presented in the lectures of Klaus Kirsten. Integral representations of these functions are established and their association with eigenvalue problems for partial differential operators is discussed by way of concrete computational examples, where various types of boundary conditions are imposed. For the atomic Schrödinger operator corresponding to a harmonic oscillator potential in three dimensions, for example, the spectral zeta function is shown to be given in terms of the Barnes zeta function. Applications of zeta functions to other areas such as the Casimir effect and Bose–Einstein condensation are also discussed in the Kirsten lectures, along with some provocative, motivating questions: Can one hear the shape of a drum? What does the Casimir effect know about a boundary? What does a Bose gas know about its container? Regarding the first of these questions — which was posed in 1966 by M. Kac, but may even go back to H. Weyl — some heat kernels and their small-time asymptotics and Minakshisundaram–Pleijel coefficients are presented by way of examples, and by way of a general result

INTRODUCTION

3

for Laplace-type differential operators acting on smooth sections of a vector bundle over a smooth compact Riemannian manifold with or without boundary. Bose–Einstein condensation is the final topic treated, where zeta functions continue to play a relevant role. A Bose–Einstein condensate is the ground state (lowest energy state) assumed by interacting bosons under the influence of some external trapping potential. This condensation phenomenon, which was predicted in 1924 by S. Bose and A. Einstein, is an example of phase transition at the quantum level. It took some 70 years, however, for its first experimental verification, in some 1995 vapor studies of rubidium and sodium. The subject matter of zeta functions (of Riemann, Selberg, Ruelle, and Ihara) and quantum chaos is taken up in the lectures of Audrey Terras, where emphasis is laid on the Ihara zeta function of a finite graph, and its connections to chaos and random matrix theory. The Selberg zeta function is a zeta function attached to a compact Riemann surface (say of genus at least two), and gives rise to a duality between lengths of closed geodesics and the spectrum of the Laplace– Beltrami operator — closed geodesics being regarded as Selberg “primes”. This zeta function can be attached, more generally, to compact quotients of hyperbolic spaces, or in fact to the quotient of a rank 1, noncompact symmetric space modulo a cofinite volume discrete group of isometries. The Ruelle zeta function is a dynamical systems zeta function, whose motivation goes back to work of M. Artin and B. Mazur on the zeta function of a projective algebraic variety over a finite field, where the Frobenius map is replaced by a suitable diffeomorphism of a smooth compact manifold. It is shown in the lectures that the Ihara zeta function is generalized by the Ruelle zeta function, the former function being the graph version of the Selberg zeta function where the duality is replaced by that between closed paths and the spectrum of an adjacency matrix. Some particular topics include edge and path zeta functions, Ihara determinant formulas, a graph prime number theorem, a Riemann hypothesis regarding the poles of the Ihara zeta function (which for a regular connected graph is true precisely when the graph is Ramanujan — a condition on the eigenvalues of the adjacency matrix, roughly speaking), the Alon conjectures for regular and irregular graphs (which refer to Riemann hypotheses for these graphs), and some closing material with a focus on the quantum chaos question that concerns a connection between the poles of Ihara zeta and eigenvalues of a random matrix. Here a comparison is made between the spacing of poles and the spacing of the eigenvalues of a certain (non-symmetric) edge adjacency matrix. Quantum chaos, which is not always precisely defined, is described sometimes as the statistics of eigenvalues (or energy levels) of particular non-classical systems. For example, E. Wigner, in the 1950s, first introduced the notion of the statistical distribution of energy levels of heavy atomic nuclei. From this

4

INTRODUCTION

work there evolved the Wigner surmise (which Audrey also discusses) for the probability density for finding two adjacent eigenvalues with a given spacing — a density that differs markedly from a Poisson density. Part I concludes with an extended set of lectures by Geoff Mason and Michael Tuite that cover basic, introductory material on vertex operator algebras (VOAs) and modular forms, as well as some research material indicative of their presentations during the workshop speakers’ seminar. VOAs, also known as chiral algebras to physicists, special cases of which include W-algebras, were formally defined by R. Borcherds. They figure prominently in many areas such as finite group theory (in particular in regards to “monstrous moonshine”), representations of affine Kac–Moody Lie algebras, knot theory, quantum groups, geometric Langlands theory, etc., and in fact they provide for a mathematical, axiomatic formulation of two-dimensional conformal field theory (CFT). Various CFT manipulations, that sometimes are in want of rigor, are neatly handled, algebraically, in the VOA world. In the so-called operator product expansion for primary fields, for example (as postulated by A. Belavin, A. Polyakov, and A. Zamolodchikov), the fundamental requirement of associativity encodes neatly in VOA structure — which in the lectures amounts to the Jacobi identity (page 188). This identity also embodies the key notion of locality. Locality relaxes the requirement of commutativity of fields, which if imposed would be too strong a physical condition. String theory for bosons can be regarded as a CFT, and it therefore has natural ties to VOAs. For example, to an even, positive definite lattice L is attached a VOA (pages 221–224), which corresponds to a bosonic string compactified on the torus defined by L. Other naturally constructed VOAs are Heisenberg VOAs (that correspond to a free boson), Virasoro VOAs, and VOAs attached to a Lie algebra g equipped with a symmetric, non-degenerate, invariant bilinear form, from whence is attached to g an affine Kac–Moody Lie algebra. The moonshine module also has a VOA structure (due to I. Frenkel, J. Lepowsky, and J. Meurman), which is discussed on page 233. Modular forms and elliptic functions arise naturally in VOA theory by way of VOA characters (or partition functions) and, more generally, correlation functions. For nonrational VOAs, however, these functions might only be quasimodular. But the character of the (nonrational) Heisenberg VOA, for example, is a modular form (of weight 12 ) as it is the reciprocal of the Dedekind eta function. A characteristic property of rational VOAs is that they have only finitely many inequivalent, irreducible representations. In particular, a distinguished class of rational VOAs (called Virasoro minimal models) is provided by certain Virasoro quotient modules of zero conformal weight and a central charge c D 1 6.p q/2 =pq parametrized by a pair of coprime integers p; q greater

INTRODUCTION

5

than 1. The (finite) number of irreducible representations here is .p 1/.q 1/. p D 4 and q D 3, for example, give rise to a central charge 21 , which at the physical level corresponds to the Ising model in statistical mechanics, whereas p D 5, 7 , and p D 6, q D 4 corresponds to the tricritical Ising model with central charge 10 q D 5 corresponds to the three-states Potts model with central charge 54 . In addition to one-point correlation functions, Geoff and Michael also consider two-point correlation functions that have relevance to their research on higher genus CFT. These functions are shown to be elliptic. Zhu recursion formulas (for correlation functions), Zhu’s finiteness condition (referred to as C2 -cofiniteness in the lectures), the important issue of modular invariance, and other important matters are given careful attention in the latter part of the paper, with a discussion of current research areas. A deep, initial result of Y. Zhu is that for rational VOAs subject to his finiteness condition, the complete (finite) set of inequivalent, irreducible characters enjoys the beautiful modular property of being holomorphic functions on the upper half-plane that span a subspace invariant under the action of the modular group SL.2; Z /. At the Lie-algebra representation theory level, apart from VOA theory, modular properties of characters have profound implications — which is the case for Weyl–Kac characters of affine SU.2/, for example, where modularity is exploited in the study of BTZ black hole entropy corrections. Seminar lectures comprise Part II of the volume (with the exception of those of Geoff and Michael, which, as mentioned, appear in Part I). We describe them briefly. Jennie’s lecture deals with elliptic and theta function solutions of Einstein’s gravitational field equations in the FLRW model. A soliton–black hole connection in two-dimensional gravity is considered in Shabnam’s lecture. Floyd discusses the role of the Patterson–Selberg zeta function in threedimensional gravity. Klaus points out how contour integration methods lead to closed formulas for functional determinants in low and high dimensions. The editors express sincere thanks and appreciation to their workshop colleagues for their valued contributions, and to MSRI Director Dr. Robert Bryant and Associate Director Dr. Kathleen O’Hara for the invitation to present the workshop and the kind hospitality we were accorded. We are also appreciative of the gifted workshop students that we were fortunate to interact with. Our sincere thanks extend moreover to the MSRI Book Series Editor, Dr. Silvio Levy, for his helpful guidance during the preparation of the volume. Klaus Kirsten Floyd L. Williams August 2009


Lectures on zeta functions, L-functions and modular forms with some physical applications FLOYD L. WILLIAMS

Introduction We present nine lectures that are introductory and foundational in nature. The basic inspiration comes from the Riemann zeta function, which is the starting point. Along the way there are sprinkled some connections of the material to physics. The asymptotics of Fourier coefficients of zero weight modular forms, for example, are considered in regards to black hole entropy. Thus we have some interests also connected with Einstein’s general relativity. References are listed that cover much more material, of course, than what is attempted here. Although his papers were few in number during his brief life, which was cut short by tuberculosis, Georg Friedrich Bernhard Riemann (1826–1866) ranks prominently among the most outstanding mathematicians of the nineteenth century. In particular, Riemann published only one paper on number theory [32]: ¨ “Uber die Anzahl der Primzahlen unter einer gegebenen Grösse”, that is, “On the number of primes less than a given magnitude”. In this short paper prepared for Riemann’s election to the Berlin Academy of Sciences, he presented a study of the distribution of primes based on complex variables methods. There the now famous Riemann zeta function 1 X 1 .s/ D ; ns def

(0.1)

nD1

defined for Re s > 1, appears along with its analytic continuation to the full complex plane C , and a proof of a functional equation (FE) that relates the values .s/ and .1 s/. The FE in fact was conjectured by Leonhard Euler, who also obtained in 1737 (over 120 years before Riemann) an Euler product 7

8

FLOYD L. WILLIAMS

representation .s/ D

Y

p>0

1 1 p

s

.Re s > 1/

(0.2)

of .s/ where the product is taken over the primes p. Moreover, Riemann introduced in that seminal paper a query, now called the Riemann Hypothesis (RH), which to date has defied resolution by the best mathematical minds. Namely, as we shall see, .s/ vanishes at the values s D 2n, where n D 1; 2; 3; : : : ; these are called the trivial zeros of .s/. The RH is the (yet unproved) statement that if s is a zero of that is not trivial, the real part of s must have the value 21 ! Regarding Riemann’s analytic approach to the study of the distribution of primes, we mention that his main goal was to set up a framework to facilitate a proof of the prime number theorem (which was also conjectured by Gauss) which states that if .x/ is the number of primes x, for x 2 R a real number, then .x/ behaves asymptotically (as x ! 1) as x= log x. That is, one has (precisely) that .x/ D 1; x!1 x= log x lim

(0.3)

which was independently proved by Jacques Hadamard and Charles de la ValléePoussin in 1896. A key role in the proof of the monumental result (0.3) is the fact that at least all nontrivial zeros of .s/ reside in the interior of the critical strip 0 Re s 1. Riemann’s deep contributions extend to the realm of physics as well - Riemannian geometry, for example, being the perfect vehicle for the formulation of Einstein’s gravitational field equations of general relativity. Inspired by the definition (0.1), or by the Euler product in (0.2), one can construct various other zeta functions (as is done in this volume) with a range of applications to physics. A particular zeta function that we shall consider later will bear a particular relation to a particular solution of the Einstein field equations — namely a black hole solution; see my Speaker’s Lecture. There are quite many ways nowadays to find the analytic continuation and FE of .s/. We shall basically follow Riemann’s method. For the reader’s benefit, we collect some standard background material in various appendices. Thus, to a large extent, we shall attempt to provide details and completeness of the material, although at some points (later for example, in the lecture on modular forms) the goal will be to present a general picture of results, with some (but not all) proofs. Special thanks are extended to Jennie D’Ambroise for her competent and thoughtful preparation of all my lectures presented in this volume.

LECTURES ON ZETA FUNCTIONS, L-FUNCTIONS AND MODULAR FORMS

9

C ONTENTS Introduction 1. Analytic continuation and functional equation of the Riemann zeta function 2. Special values of zeta 3. An Euler product expansion 4. Modular forms: the movie 5. Dirichlet L-functions 6. Radiation density integral, free energy, and a finite-temperature zeta function 7. Zeta regularization, spectral zeta functions, Eisenstein series, and Casimir energy 8. Epstein zeta meets gravity in extra dimensions 9. Modular forms of nonpositive weight, the entropy of a zero weight form, and an abstract Cardy formula Appendix References

7 9 17 21 30 46 50 57 66 70 78 98

Lecture 1. Analytic continuation and functional equation of the Riemann zeta function Since j1=ns j D 1=nRe s , the series in (0.1) converges absolutely for Re s > 1. Moreover, by the Weierstrass M-test, for any ı > 0 one has uniform convergence of that series on the strip def

Sı D fs 2 C j Re s > 1 C ıg; since j1=ns j D 1=nRe s < 1=n1Cı on Sı , with 1 X

nD1

1 n1Cı

< 1: def

Since any compact subset of the domain S0 D fs 2 C j Re s > 1g is contained in some Sı , the series, in particular, converges absolutely and uniformly on compact subsets of S0 . By Weierstrass’s general theorem we can conclude that the Riemann zeta function .s/ in (0.1) is holomorphic on S0 (since the terms 1=ns are holomorphic in s) and that termwise differentiation is permitted: for Re s > 1 1 X log n 0 .s/ D : (1.1) ns nD1

We wish to analytically continue .s/ to the full complex plane. For that purpose, we begin by considering the world’s simplest theta function .t /, defined

10

FLOYD L. WILLIAMS

for t > 0: def

.t / D

X

e

n2 t

n2Z

D 1C 2

1 X

e

n2 t

(1.2)

nD1

where Z denotes the ring of integers. It enjoys the remarkable property that its values at t and t inverse (i.e. 1=t ) are related: .t / D

.1=t / p : t

(1.3)

The very simple formula (1.3), which however requires some work to prove, is called the Jacobi inversion formula. We set up a proof of it in Appendix C, based on the Poisson Summation Formula proved in Appendix C. One can of course define more complicated theta functions, even in the context of higherdimensional spaces, and prove analogous Jacobi inversion formulas. For s 2 C define Z 1 .t / 1 s def J.s/ D t dt: (1.4) 2 1

By Appendix A, J.s/ is an entire function of s, whose derivative can be obtained, in fact, by differentiation under the integral sign. One can obtain both the analytic continuation and the functional equation of .s/ by introducing the sum 1 Z 1 X def . n2 / s e t t s 1 dt; (1.5) I.s/ D nD1 0

which we will see is well-defined for Re s > 21 , and by computing it in different ways, based on the inversion formula (1.3). Recalling that the gamma function .s/ is given for Re s > 0 by Z 1 def .s/ D e t t s 1 dt (1.6) 0

we clearly have def

I.s/ D

s

X 1

nD1

1 n2s

.s/ D

s

.2s/ .s/;

(1.7)

so that I.s/ is well-defined for Re 2s > 1: Re s > 12 . On the other hand, by the change of variables u D t = n2 we transform the integral in (1.5) to obtain I.s/ D

1 Z X

nD1 0

1

e

n2 t s 1

t

dt:


11

We can interchange the summation and integration here by noting that 1 Z X

nD1

0

1ˇ

ˇe

n2 t s 1 ˇ

ˇ

t

dt D

1 Z X

nD1

1

n2 t Re s 1

e

t

0

dt D I.Re s/ < 1

for Re s > 21 ; thus

I.s/ D

Z

D

Z

0

1 1X

n2 t s 1

e

t

nD1

1

0

.t / 1 s t 2

1

dt D

dt C

Z

1

Z

1

1

.t / 1 s t 2

0

.t / 1 s t 2

1

1

dt

dt;

(1.8)

by (1.2). Here Z

1

t

s 1

0

dt D lim

"!0C

Z

1

ts

1

"

dt D

1 s

(1.9)

for Re s > 0. In particular (1.9) holds for Re s > 12 , and we have Z

1 0

.t / 1 s t 2

1

1 dt D 2

Z

1

.t /t s

1

dt

0

1 : 2s

(1.10)

By the change of variables u D 1=t , coupled with the Jacobi inversion formula (1.3), we get Z 1 Z 1 Z 1 1 1 s 1 t 1 s dt D .t /t 2 t 1 s dt .t /t dt D t 1 0 1 Z 1 Z 1 1 1 D ..t / 1/ t 2 s dt C t 2 s dt 1

D D

Z

Z

1

1

..t /

1/ t

1 2

s

1 1 1

..t /

1/ t

1 2

s

dt C dt C

Z

1

3 2 CsD.s

u

0

s

1 1 2

1 2/

1

du

;

where we have used (1.9) again for Re s > 12 . Together with equations (1.8) and (1.10), this gives Z Z 1 1 1 .t / 1 s 1 1 1 1 s 2 dt C C t dt ..t / 1/ t I.s/ D 1 2 1 2 2.s 2 / 2s 1 Z 1 1 1 1 .t / 1 s 1 t C t 2 s dt C ; D 2 2s 1 2s 1

12

FLOYD L. WILLIAMS

which with equation (1.7) gives Z 1 .t / 1 s s .2s/ .s/ D t 2 1

1

Ct

1 2

s

dt C

1 2s

1

1 ; 2s

(1.11)

for Re s > 12 . Finally, in (1.11) replace s by s=2, to obtain s Z 1 .t / 1 s s 1 1 1 s=2 D t 2 1 C t 2 2 dt C (1.12) .s/ 2 2 s 1 s 1 s s for Re s > 1. Since z .z/ D .z C 1/, we have sD2 C 1 , which 2 2 proves: T HEOREM 1.13. For Re s > 1 we can write s s s Z 1 s 2 .t / 1 s 1 2 2 1 : dt C t2 Ct 2 .s/ D s s s 2 .s 1/ 2 C1 1 2 2 2 R1 The integral 1 in this equality is an entire function of s, since, by (1.4), it s 1 . Also, since 1= .s/ is an entire function of s, equals J 2s 1 C J 2 it follows that the right-hand side of the equality in Theorem 1.13 provides for the analytic continuation of .s/ to the full complex plane, where it is observed that .s/ has only one s D 1 is a simple pole. singularity: 1 1=2 The fact that allows one to compute the corresponding residue: 2 D 1

s

lim .s

s!1

1/.s/ D lim

s!1

2

s 2

D

2

1 2

D 1:

An equation that relates the values .s/ and .1 s/, called a functional equation, easily follows from the preceding discussion. In fact define s def XR .s/ D s=2 .s/ (1.14) 2

for Re s > 1 and note that the right-hand side of equation (1.12) (which provides for the analytic continuation of XR .s/ as a meromorphic function whose simple poles are at s D 0 and s D 1) is unchanged if s there is replaced by 1 s: T HEOREM 1.15 (T HE FUNCTIONAL EQUATION FOR .s/). Let XR .s/ be given by (1.14) and analytically continued by the right-hand side of the (1.12). Then XR .s/ D XR .1 s/ for s ¤ 0; 1. One can write the functional equation as s s .s/ 2 2 D .1 s/ D 1 s 1 s 2 2

sC 12

s .s/ 2 1 s 2

(1.16)


13

1 s , 2

use the

s 1 = for s ¤ 0; 1, multiply the right-hand side here by 1 D 2 1 s 1 s 3 s identity 2 D , and thus also write 2 2 1 s 1/.s/ sC 2 2 .s .1 s/ D ; 2 32 s

(1.17)

an equation that will be useful later when we compute 0 .0/. For the computation of 0 .0/ we make use of the following result, which is of independent interest. Œx denotes the largest integer that does not exceed x 2 R . T HEOREM 1.18. For Re s > 1, Z 1 Œx x C 12 dx 1 C Cs .s/ D s 1 2 x sC1 1 Z 1 Œx x 1 C 1C s dx: D s 1 x sC1 1 1

(1.19)

R1 That these two expressions for .s/ are equal follows from the equality 1 xdx sC1 D 1s for Re s > 0; this with the inequalities 0 x Œx < 1 allows one to deduce that the improper integrals there converge absolutely for Re s > 0. We base the proof of Theorem 1.18 on a general observation: L EMMA 1.20. Let .x/ be continuously differentiable on a closed interval Œa; b. Then, for c 2 R , Z b Z b x c 12 0 .x/ dx D b c 12 .b/ a c 12 .a/ .x/ dx: a

a

In particular for Œa; b D Œn; n C 1, with n 2 Z one gets Z nC1 .n C 1/ C .n/ x Œx 12 0 .x/ dx D 2 n

Z

nC1

.x/ dx: n

P ROOF. The first assertion is a direct consequence of integration Using R nC1by parts. 0 .x/ dx D it, one obtains for the choice c D n the second assertion: Œx n R nC1 0 n .x/ dx (since Œx D n for n x < n C 1); hence n

Z

nC1

x

1 2

Œx

n

D

Z

0 .x/ dx

nC1

x

1 2

n

n 1 2

0 .x/ dx

D nC1 n .n C 1/ n n 12 .n/ Z nC1 D 21 .n C 1/ C 21 .n/ .x/ dx; )

n

Z

nC1

.x/ dx n

˜

14

FLOYD L. WILLIAMS

As a first application of the lemma, note that for integers m2 ; m1 with m2 > m1 , m2 X

nDm1

Œ.nC1/ C .n/ m2 X

D

nDm1

.nC1/ C

m2 X

.n/

nDm1

D .m1 C1/C.m1 C2/C C.m2 C1/C.m1 /C.m1 C1/C C.m2 / D .m2 C1/ C .m1 /C2 Also

R nC1 nDm1 n

Pm2

D

m1

. Therefore

m2 X

nDm1C1

m2 X

nDm1

Z

nDm1C1

.n/ D

.n/ D

m2 1 X Œ.n C 1/ C .n/ 2 nDm 1

nC1

.x

0 1 2 / .x/ dx C

Œx

n

(by Lemma 1.20), which equals Thus m2 X

.n/:

nDm1 C1

R m2 C1

.m2 C 1/ C .m1 / C 2 D

m2 X

R m2 C1 m1

.x Œx

.m2 C 1/ .m1 / 2 Z m2 C1 Z C .x/ dx C m1

m2 C1 m1

m2 Z X

nDm1

nC1

.x/ dx

n

R m2 C1 1 0 .x/ dx. 2 / .x/ dx C m1

.x

Œx

1 0 / .x/ dx 2

(1.21)

for .x/ continuously differentiable on Œm1 ; m2 C 1. Now choose m1 D 1 and def

.x/ D x

s

R1 for x > 0, Re s > 1. Then 1 dx D s 1 1 . Also .m2 C 1/ D .m2 C 1/ xs as m2 ! 1, since Re s > 0. Thus in (1.21) let m2 ! 1: Z 1 1 X 1 1 1 D 2C C .x Œx 12 /. sx s 1 / dx: ns s 1 1 nD2

That is, for Re s > 1 we have Z 1 1 X .Œx x C 21 / 1 1 1 dx; D C C s .s/ D 1 C ns 2 s 1 x sC1 1 nD2

s

!0


15

which proves Theorem 1.18. We turn to the second integral in equation (1.19), which we denote by Z 1 .Œx x/ def dx f .s/ D x sC1 1 Rn x/ for Re s > 0. We can write f .s/ D limn!1 1 .Œx dx, where x sC1 Z

n

1

n 1 Z j C1

X .Œx x/ dx D x sC1

j D1

j

n 1 Z j C1

X .Œx x/ dx D x sC1

j D1

since Œx D j for j x < j C 1. That is, f .s/ D def

aj .s/ D

Z

j

j C1

j x j dx D sC1 s x

1 js

1 .j C1/s

j

P1

j x dx; x sC1

j D1 aj .s/

1

s

1

where

1 js

(1.22)

1

1 .j C1/s

1

for s ¤ 0; 1, and where for the second term here s D 1 is a removable singularity: 1 1 1 lim .s 1/ D 0: s 1 j s 1 .j C 1/s 1 s!1 Similarly, for the first term s D 0 is a removable singularity. That is, the aj .s/ are entire functions. In particular each aj .s/ is holomorphic on the domain def D C D fs 2 C j Re s > 0g. At the same time, for WD Re s > 0 we have Z j C1 1 1 dx 1 D jaj .s/j j .j C1/ x C1 j (where the inequality comes from jj xj D x j 1 for j x j C1); moreover 1 n X X 1 1 1 1 1 D1 D1 ) j .j C 1/ .n C 1/ j .j C 1/ j D1

j D1

P (i.e. 1=.nC1/ ! 0 as n ! 1 for > 0). Hence, by the M-test, j1D1 aj .s/ converges absolutely and uniformly on D C (and in particular on compact subsets of D C ). f .s/ is therefore holomorphic on D C , by the Weierstrass theorem. Of course, in equation (1.19), Z 1 Œx x C 12 dx D sf .s/ C 12 s sC1 x 1 is also a holomorphic function of s on D C . We have deduced:

16

FLOYD L. WILLIAMS

C OROLLARY 1.23. Let Z

def

f .s/ D

1

1 2/ dx: x sC1

.Œx

1

Then f .s/ is well-defined for Re s > 0 and is a holomorphic function on the def domain D C D fs 2 C j Re s > 0g. For Re s > 1 one has (by Theorem 1.18) .s/ D

1

s

1

C 1 C sf .s/:

(1.24)

From this we see that .s/ admits an analytic continuation to D C . Its only singularity there is a simple pole at s D 1 with residue lims!1 .s 1/.s/ D 1, as before. This result is obviously weaker than Theorem 1.13. However, as a further application we show that 1 lim .s/ D (1.25) s 1 s!1 where

1 1 1 def

D lim 1 C C C C (1.26) log n n!1 2 3 n is the Euler–Mascheroni constant; ' 0:577215665. By the continuity (in particular) of f .s/ at s D 1, f .1/ D lims!1 f .s/. That is, by (1.24), we have 1 D lim 1 C sf .s/ lim .s/ s 1 s!1 s!1 n X1 Z j C1 j x .1:22/ D 1 C f .1/ D 1 C lim dx n!1 x2 j j D1 n X1 1 log .j C 1/ log j D 1 C lim n!1 j C1 j D1

D 1 C lim

n!1

D 1 C lim

n!1

D 1 C lim

n!1

D ; as desired.

n X1

nX1

j D1

1 j C1

j D1

j D1

1 j C1

log n

nX1

log.j C 1/

1 1 1 1C 1C C C C 2 3 n

log j

log n


17

Since s D 1 is a simple pole with residue 1, .s/ has a Laurent expansion .s/ D

1 s

1

C 0 C

1 X

1/k

k .s

(1.27)

kD1

on a deleted neighborhood of 1. By equation (1.25), 0 D . One can show that, in fact, for k D 0; 1; 2; 3; : : : X n .log l/k .log n/kC1 . 1/k ; (1.28) lim

k D k! n!1 l k C1 lD1

a result we will not need (except for the case k D 0 already proved) and thus which we will not bother to prove. The inversion formula (1.3), which was instrumental in the approach above to the analytic continuation and FE of .s/, provides for a function F.t /, t > 0, def that is invariant under the transformation t ! 1=t . Namely, let F.t / D t 1=4 .t /. Then (1.3) is equivalent to statement that F.1=t / D F.t /, for t > 0.

Lecture 2. Special values of zeta In 1736, L. Euler discovered the celebrated special values result .2n/ D

. 1/nC1 .2/2n B2n 2.2n/!

(2.1)

for n D 1; 2; 3; : : : , where Bj is the j-th Bernoulli number, defined by z ez

1

D

1 X Bj j z ; j!

j D0

for jzj < 2, which is the Taylor expansion about z D 0 of the holomorphic def function h.z/ D z=.e z 1/, which is defined to be 1 at z D 0. Since e z 1 vanishes if and only if z D 2 i n, for n 2 Z , the restriction jzj < 2 means that the denominator e z 1 vanishes only for z D 0. The Bj were computed by Euler up to j D 30. Here are the first few values: B0 B1 B2 B3 B4 B5 B6 B7 B8 B9 B10 B11 B12 B13 1

1 2

1 6

0

1 30

0

1 42

0

1 30 def

0

5 66

0

691 2730

0

(2.2)

In general, Bodd>1 D 0: To see this let H .z/ D h.z/ C z=2 for jzj < 2, which we claim is an even function. Namely, for z ¤ 0 the sum z=.e z 1/Cz=.e z 1/ equals D z by simplification: z z z z D z Cz D H .z/: H . z/ D z e 1 2 e 1 2

18

FLOYD L. WILLIAMS

Then 1

1

j D1

j D1

X B2j X B2j C1 z C B0 C B1 z C z 2j C z 2j C1 2 .2j /! .2j C 1/! D

z C 2

D

1 X

j D0

Bj j z D H .z/ D H . z/ j! 1

1

j D1

j D1

X B2j X B2j C1 z C B0 C B1 . z/ C . z/2j C . z/2j C1 ; 2 .2j /! .2j C 1/!

which implies 1 X B2j C1 2j C1 z ; 0 D .1 C 2B1 /z C 2 .2j C 1/! j D1

and consequently B1 D (2.1) (in particular)

1 2

and B2j C1 D 0 for j 1, as claimed. By formula

1 1 1 X X X 2 4 6 1 1 1 D D D .2/ D ; .4/ D ; .6/ D ; (2.3) 6 90 945 n2 n4 n6 nD1

nD1

nD1

P 2 2 the first formula, 1 nD1 1=n D =6, being well-known apart from knowledge of the zeta function .s/. We provide a proof of (2.1) based on the summation formula 1 X 1 1 coth a D (2.4) 2 2 2a n Ca 2a2 nD1

for a > 0; see Appendix E on page 92. Before doing so, however, we note some other special values of zeta. As we have noted, 1= .s/ is an entire function of s. It has zeros at the points s D 0; 1; 2; 3; 4; : : : . By Theorem 1.13 and the remarks that follow its statement we therefore see that for n D 1; 2; 3; 4; : : : , . 2n/ D

n D 0; 2 . n C 1/

.0/ D

1 D 2 .1/

1 : 2

(2.5)

Thus, as mentioned in the Introduction, .s/ vanishes at the real points s D 2; 4; 6; 8; : : :, called the trivial zeros of .s/. The value .0/ is nonzero — it equals 21 by (2.5). Later we shall check that 0 .0/ D .0/ log 2 D

1 2

log 2:

(2.6)


19

t in (2.4), Turning to the proof of (2.1), we take 0 < t < 2 and choose a D 2 obtaining successively 1

2 t coth t 2 1 t coth 2 2

X 1 2 2 2 D 4 ; 2 2 t t C 4 2 n2 nD1

1 X

1 ; t 2 C 4 2 n2 nD1 e t=2 C e t=2 1 1 2 C e t 1 e t=2 D C D et 1 2 2.e t 1/ e t=2 2.e t=2 e t=2 / 1 D 2t t

1 X 1 cosh.t =2/ 1 1 t 1 D ; D coth D C 2t 2 2 sinh.t =2/ 2 2 t t C 4 2 n2 nD1

t et

1

C

t D 1 C 2t 2 2

1 X

nD1

1 t 2 C 4 2 n2

:

(2.7)

1 2

(see (2.2)), and since B2kC1 D 0 for k 1, we can Since B0 D 1 and B1 D write 1 1 X B2k 2k t t def X Bk k D t D1 C t ; t e 1 k! 2 .2k/! kD0

kD1

and (2.7) becomes

1 1 X X B2k 2k 1 2 : t D 2t 2 .2k/! t C 4 2 n2

(2.8)

nD1

kD1

For 0 < t < 2, we can use the convergent geometric series 1 X 4 2 n2 t2 k 1 D D ; t2 4 2 n2 t 2 C 4 2 n2 1 C 2 2 kD0 4 n

(2.9)

to rewrite (2.8) as 1 1 X 1 X X B2k 2k 1 t2 k 2 t D 2t .2k/! 4 2 n2 4 2 n2 nD1 kD0 kD1 1 X 1 X t2 k 1 D 2t 2 4 2 n2 4 2 n2

(2.10) 1

:

nD1 kD1

The point is to commute the summations on n and k in this equation. Now k 1 ˇ X 1 ˇ 1 X 1 2.k 1/ X 1 2.k 1/ X 1 1 X X ˇ 1 ˇ t2 t t 1 1 ˇ ˇD ˇ 4 2 n2 ˇ 2 2 2 k 2k 2 k 4 n n2 .4 / n .4 /

kD1 nD1

kD1

nD1

nD1

nD1

20

FLOYD L. WILLIAMS

P P1 2.k 1/ 2 which is finite since 1 =.4 2 /k < 1, nD1 1=n D .2/ < 1 and nD1 t by the ratio test (again for 0 < t < 2). Commutation of the summation is therefore justified: 1 1 X X . t 2 /k 1 B2k t 2k D 2t 2 .2k/! 4 2 .4 2 /k

kD1

kD1

1

1 1 X X 2. 1/k 1 1 D .2k/t 2k n2k .4 2 /k

nD1

kD1

on .0; 2/. By equating coefficients, we obtain B2k 2. 1/k 1 .2k/ D .2k/! .4 2 /k

for k 1;

which proves Euler’s formula (2.1). Next we turn to a proof of equation (2.6). We start with an easy consequence of the quotient and product rules for differentiation. L EMMA 2.11 (L OGARITHMIC DIFFERENTIATION WITHOUT LOGS ). If F.s/ D

1 .s/2 .s/3 .s/ ; 4 .s/

on some neighborhood of s0 2 C , where the j .s/ are nonvanishing holomorphic functions there, then F 0 .s0 / 10 .s0 / 20 .s0 / 30 .s0 / 40 .s0 / D C C : F.s0 / 1 .s0 / 2 .s0 / 3 .s0 / 4 .s0 / def 1 def 3 s s Now choose 1 .s/ D 2 s , 2 .s/ D , .s/ D 2 4 2 2 , say on a small neighborhood of s D 1. For the choice of 3 .s/, we write .s/ D g.s/=.s 1/ on a neighborhood N of s D 1, for s ¤ 1, where g.s/ is holomorphic on N and g.1/ D 1. This can be done since s D 1 is a simple pole of .s/ with residue def D 1; for example, see equation (1.27). Assume 0 … N and take 3 .s/ D g.s/ on N . By equation (1.17), .1 s/ D 1 .s/2 .s/3 .s/=4 .s/ near s D 1, so that def by Lemma 2.11 and introducing the function .s/ D 0 .s/= .s/, we obtain ˇ 0 .1 s/ ˇˇ .1 s/ ˇsD1 ˇ ˇ 1 ˇˇ ˇˇ 2 s . log / ˇˇ g 0 .s/ ˇˇ s 1ˇ 1 ˇ 3 s D : (2.12) 1 ˇ C 2 2 ˇ C g.s/ ˇ s 2 2 ˇsD1 2 sD1 sD1 sD1 If is the Euler–Mascheroni constant of (1.26), the facts .1/ D and 1 2 log 2 are known to prevail, which reduces equation (2.12) to 2 D

0 .0/ D .0/ log C C log 2 g 0 .1/ C 2 2 0 1 D 2 log C C log 2 g .1/ ;


21

since g.1/ D 1 and .0/ D 21 ; see (2.5). But g 0 .1/ D , as we will see in a minute; hence we have reached the conclusion that 0 .0/ D 21 log 2, which is (2.6). There remains to check that g 0 .1/ D . We have g.s/ 1 ; s 1 again since g.1/ D 1; this in turn equals lims!1 .s/ (1.25). def

g 0 .1/ D lim

s!1

1 s 1

D , by equation

To obtain further special values of zeta we appeal to the special values formula p . 1/n 22n n! 1 n D (2.13) 2 .2n/!

for the gamma function, where n D 1; 2; 3; 4; : : : . This we couple with (2.1) and the functional equation (1.16) to show that . 1/ D

1 12

and

Namely, .1 2n/ D

.1

2nC 21

B2n for n D 1; 2; 3; 4; : : : : (2.14) 2n .n/.2n/= 12 n , by (1.16); this in turn equals 2n/ D

2nC 12

.n 1/!.2n/.2n/! p ; . 1/n 22n n!

by (2.13); whence (2.1) gives .1

2n/ D

. 1/n .2/

Taking n D 1 gives . 1/ D

2n .2n/.2n/!

n B2 2

D

B2n : 2n

1 , by (2.2), which confirms (2.14). 12

Lecture 3. An Euler product expansion For a function f .n/ defined on the set Z C D f1; 2; 3; : : :g of positive integers one has a corresponding zeta function or Dirichlet series 1 X f .n/ ; f .s/ D ns def

nD1

defined generically for Re s sufficiently large. If f .n/ D 1 for all n 2 Z C , for example, then for Re s > 1, f .s/ is of course just the Riemann zeta function .s/, which according Q to equation (0.2) of the Introduction has an Euler product expansion .s/ D p2P 1 1p s over the primes P in Z C . It is natural to inquire whether, more generally, there are conditions that permit an analogous Euler product expansion of a given Dirichlet series f .s/. Very pleasantly, there is

22

FLOYD L. WILLIAMS

an affirmative result when, for example, the f .n/ are Fourier coefficients (see Theorem 4.32, where the n-th Fourier coefficient there is denoted by an ) of certain types of modular forms, due to a beautiful theory of E. Hecke. Also see equations (3.20), (3.21) below. Rather than delving directly into that theory at this point we shall instead set up an abstract condition for a product expansion. The goal is to show that under suitable conditions on f .n/ of course the desired expansion assumes the form 1 X f .1/ f .n/ DQ f .s/ D s 2s n p2P 1 C ˛.p/p def

nD1

f .p/p

s

(3.1)

for some function ˛.p/ on P ; see Theorem 3.17 below. Here we would want to have, in particular, that f .1/ ¤ 0. Before proceeding toward a precise statement and proof of equation (3.1), we note that (again) if f .n/ D 1 for all n 2 Z C , for example, then for the choice ˛.p/ D 0 for all p 2 P , equation (3.1) reduces to the classical Euler product expansion of equation (0.2). Given f W Z C ! R or C , and ˛ W P ! R or C , we assume the following abstract multiplicative condition: f .np/ if p - n, (3.2) f .n/f .p/ D n if p j n, f .np/ C ˛.p/f p

for .n; p/ 2 Z C P ; here p j n means that p divides n and p - n means the opposite. Given condition (3.2) we observe first that if f .1/ D 0 then f vanishes identically, the proof being as follows. For a prime p 2 P , (3.2) requires that f .1/f .p/ D f .p/, since p - 1; that is, f .p/ D 0. If n 2 Z C with n 2, there exists p 2 P such that p j n, say ap D n, a 2 Z C . Proceed inductively. If p - a, f .a/f .p/ D f .ap/ D f .n/, by (3.2), so f .n/ D 0, as f .p/ D 0. If p j a, we have 0 D f .a/f .p/ (again as f .p/ D 0), and this equals f .ap/ C ˛.p/f .a=p/ D f .n/ C ˛.p/f .a=p/, where 1 < p a (so 1 a=p < a D n=p < n/. Thus f .a=p/ D 0, by induction, so f .n/ D 0, which completes the induction. Thus we see that if f 6 0 then f .1/ ¤ 0. As in Appendix D (page 88) we set, m; n 2 Z C , def 1 if m j n, d.m; n/ D 0 if m - n. Fix a finite set of distinct primes S D fp1 ; p2 ; : : : ; pl g P and define g.n/ D gS .n/ on Z C by g.n/ D f .n/

l Y

j D1

1

d.pj ; n/ :

(3.3)


Fix p 2 P

23

fp1 ; p2 ; : : : ; pl g. Then the next observation is that, for n 2 Z C , ( g.np/ if p - n, ; (3.4) g.n/f .p/ D n if p j n g.np/ C ˛.p/g p

which compares with equation (3.2).

P ROOF. If pj j n then of course pj j pn. If pj - n then pj - pn; for if pj jpn then p j n since pj ; p are relatively prime, given that p ¤ each pj . Thus d.pj ; n/ D d.pj ; pn/

for n 2 Z C , 1 j l.

(3.5)

Similarly suppose p j n, say bp D n, with b 2 Z C . If pj - n=p then pj - n; for otherwise pj j n D bp again with pj ; p relatively prime, implying that pj jb D n=p. Thus we similarly have d.pj ; n=p/ D d.pj ; n/ for n 2 Z C , 1 j l, such that p j n.

(3.6)

Now if n 2 Z C is such that p - n, then .3:3/

g.n/f .p/ D f .n/f .p/

l Q

1

d.pj ; n/

j D1

l .3:2/ Q 1 D f .np/ .3:5/

j D1

d.pj ; pn/ .3:3/

D g.np/:

On the other hand, if p j n, then .3:3/

g.n/f .p/ D f .n/f .p/ .3:2/

D

.3:5/

.3:3/

1

d.pj ; n/

j D1

f .np/ C ˛.p/f . pn /

D f .np/

.3:6/

l Q

l Q

j D1

1

l Q

j D1

1

d.pj ; n/

l Q 1 d.pj ; pn/ C ˛.p/f . pn /

D g.np/ C ˛.p/g

j D1

n

p

d.pj ; pn /

;

which proves (3.4). ˜ P1 Let h .s/ D nD1 h.n/=ns be a Dirichlet series that converges absolutely, say at some fixed point s0 2 C . Fix p 2 P and some complex number .p/ corresponding to p such that h.np/ if p - n, (3.7) h.n/.p/ D n h.np/ C ˛.p/h p if p j n,

24

FLOYD L. WILLIAMS

for n 2 Z C . Then h .s0 / 1 C ˛.p/p P ROOF. Define

2s0

s0

.p/p

D

8 n < ˛.p/h p def s0 an D : .pn/ 0

C

1 X 1

nD1

d.p; n/ h.n/ : ns0

(3.8)

if p j n, if p - n,

for n 2 Z . Since p j pn, we have apn D

pn p s .ppn/ 0

˛.p/h

D ˛.p/p

2s0 h.n/ ; ns0

P which shows that 1 The Scholium of Appendix D (page nD1 apn converges. P 91) then implies that the series 1 d.p; n/an converges, and one has nD1 1 X

nD1

apn D

1 X

d.p; n/an ;

(3.9)

nD1

where the left-hand side here is ˛.p/p 2s0 h .s0 /. Since both d.p; n/; an D 0 if p - n, d.p; n/an D an , which is also clear if p j n. On the other hand, if p j n, def an D ˛.p/h pn .pn/ s0 D h.n/.p/ h.np/ .pn/ s0 (3.10)

by equation (3.7). Equation (3.10) also holds by (3.7) in case p - n, for then both sides are zero. That is, (3.10) holds for all n 1 and equation (3.9) reduces to the statement 1 X 2s0 Œh.n/.p/ h.np/.pn/ s0 : (3.11) ˛.p/p h .s0 / D nD1

def

s0 We apply the Scholium a second time, P1 where this time we define an D h.n/=n ; since nD1 d.p; P1 jd.p; n/an j jan j, the Psum P1 n/an converges. By the Scholium, 1 a converges and a D pn pn nD1 nD1 nD1 d.p; n/an ; that is, 1 P

nD1

h.pn/.pn/

s0

D

1 P

d.p; n/h.n/n

s0

;

nD1

which one plugs into (3.11), to obtain ˛.p/p 2s0 h .s0 / D .p/p s0 h .s0 / P 1 s0 . This proves equation (3.8). nD1 d.p; n/h.n/n The proof of the main result does involve various moving parts, and it is a bit lengthy as we have chosen to supply full details. We see, however, that the proof is elementary. One further basic ingredient is needed. Again let fp1 ; : : : ; pl g by a fixed, finite set of distinct primes in P . With f; ˛ subject to the multiplicative


25

condition (3.2), we assume that f .s0 / converges absolutely where s0 2 C is Ql some fixedPnumber. 0 sj D1 1 d.pj ; n/ 1; therefore QlFor n 1, we have .1 d.p ; n// f .n/=n 0 converges absolutely. We now the series 1 j j D1 nD1 show by induction on l that l Q

j D1

1 C ˛.pj /pj

2s0

f .pj /pj

s0

f .s0 / D

l 1 Q P

1

d.pj ; n/

nD1 j D1

f .n/ ns0

: (3.12)

For l D 1, the claim follows by (3.8) with p D p1 , h.n/ D f .n/, .p/ D f .p/. Proceeding inductively, we consider a set fp1 ; p2 ; : : : ; pl ; plC1 g of l C1 distinct primes in P . Then lC1 Q

j D1

1C˛.pj /pj

2s0

f .pj /pj

s0

f .s0 /

2s0 s0 D 1C˛.plC1 /plC1 f .plC1 /plC1

l Q

2s0 s0 f .plC1 /plC1 D 1C˛.plC1 /plC1

2s0 s0 D 1C˛.plC1 /plC1 f .plC1 /plC1

j D1

1C˛.pj /pj

l 1 Q P

2s0

f .pj /pj

1 d.pj ; n/

nD1 j D1

1 g.n/ P ; s0 nD1 n

f .n/

s0

f .s0 /

ns0

(3.13)

where the second equality follows by induction and the last one by definition Ql P d.pj ; n// f .n/=ns0 (3.3). We noted, just above (3.12), that 1 j D1 .1 nD1 P1 converges absolutely; that is, nD1 g.n/=ns0 converges absolutely. Thus we choose h.n/ D g.n/, p D plC1 , .p/ D f .p/. Condition (3.7) is then a consequence of equation (3.4), since plC1 2 P fp1 ; p2 ; : : : ; pl g, and one is therefore able to apply formula (3.8) again: 2s0 s0 1 C ˛.plC1 /plC1 f .plC1 /plC1 g .s0 / 1 1 d.p P lC1 ; n/ g.n/ D ns0 nD1 l 1 1 d.p .3:3/ P lC1 ; n/ Q 1 d.p ; n/ f .n/ D j ns0 j D1 nD1 D

1 lC1 P Q

nD1 j D1

1

d.pj ; n/

f .n/ ns0

(3.14)

which, together with equation (3.13), allows one to complete the induction, and thus the proof of the claim (3.12). ˜

26

FLOYD L. WILLIAMS

Now let pl be the l-th positive prime, so p1 D 2, p2 D 3, p3 D 5, p4 D 7, . . . . We can write the right-hand side of (3.12) as f .1/ C

l 1 Q P

1

d.pj ; n/

nD2 j D1

f .n/ ns0

;

since no pj divides 1. We show that if 2 n l then l Q

1

j D1

d.pj ; n/ D 0:

(3.15)

Namely, for n 2 choose q 2 P such that q j n. If no pj divides n, 1 j l, then q ¤ p1 ; : : : ; pl (since q j n); hence q plC1 (since pl is the l-th prime) and so q l C 1. But this is impossible since n l (by hypothesis) and q n (since q j n). This contradiction proves that some pj divides n, that is, 1 D d.pj ; n/, which gives (3.15). It follows that l 1 Q 1 Q l f .n/ f .n/ P P 1 d.pj ; n/ f .1/ D ; 1 d.p ; n/ j ns0 ns0 nD1 j D1 nDlC1 j D1 Q where (using again that 0 jl D1 1 d.pj ; n/ 1) we have ˇ ˇ ˇ ˇ 1 ˇ ˇ ˇ P Q l 1 ˇ f .n/ ˇ f .n/ ˇˇ P ˇˇ f .n/ ˇˇ P ˇ 1 ˇ ˇ 1 d.pj ; n/ D ˇ ˇ ˇnDlC1 j D1 ns0 ˇ nDlC1 ˇ ns0 ˇ nD1 ˇ ns0 ˇ

ˇ ˇ l ˇ f .n/ ˇ P ˇ ˇ ˇ ns0 ˇ :

nD1

But this difference tends to 0 as l ! 1. That is, by equation (3.12), the limit Q 1 C ˛.p/p 2s0 f .p/p s0 f .s0 /

p2P

def

D lim

l Q

l!1 j D1

1 C ˛.pj /pj

2s0

f .pj /p

s0

exists, where pj D the j-th positive prime, and it equals f .1/.

f .s0 / (3.16)

We have therefore finally reached the main theorem. def

T HEOREM 3.17. (As before, Z C D f1; 2; 3; : : :g and P denotes the set of positive primes.) Let f W Z ! R or C and ˛ W P ! R or C be functions where f is not identically zero and where f is subject to the multiplicative condition (3.2). Then f .1/ ¤ 0. Let C be some subset on which the corresponding PD 1 Dirichlet series f .s/ D nD1 f .n/=ns converges absolutely. Then on D, Y 1 C ˛.p/p 2s f .p/p s f .s/ D f .1/ p2P


(see equation (3.16)). In particular both f .s/ are nonzero on D and f .s/ D Q

p2P

Q

p2P

1C ˛.p/p

f .1/ 1 C ˛.p/p 2s

on D (which is equation (3.1)).

f .p/p

2s

s

f .p/p

27 s

and

(3.18)

We remark (again) that the proof of Theorem 3.17 is entirely elementary, if a bit long-winded; it only requires a few basic facts about primes, and a weak version of the fundamental theorem of arithmetic — namely that an integer n 2 is divisible by a prime. As a simple example of Theorem 3.17, suppose f W Z C ! R or C is not identically zero, and is completely multiplicative: f .nm/ P D f .n/fs .m/ for all n; m 2 Z C . Assume also that s 2 C is such that 1 nD1 f .n/=n converges absolutely. Then 1 X 1 f .n/ D Q (3.19) : s f .p/ n 1 nD1 ps p2P

To see this, first we note directly that f .1/ ¤ 0. In fact since f 6 0, choose n 2 Z C such that f .n/ ¤ 0. Then f .n/ D f .n 1/ D f .n/f .1/, so f .1/ D 1. Also f satisfies condition (3.2) for the choice ˛ D 0 (that is, ˛.p/ D 0 for all p 2 P ). Equation (3.19) therefore follows by (3.18). In Lecture 5, we apply (3.19) to Dirichlet L-functions. Before concluding this lecture, we feel some obligation to explain the pivotal, abstract multiplicative condition (3.2). This will involve, however, some facts regarding modular forms that will be discussed in the next lecture, Lecture 4. Thus suppose that f .z/ is a holomorphic modular form of weight k D 4; 6; 8; 10; : : : , with Fourier expansion 1 X an e 2 inz (3.20) f .z/ D nD0

on the upper half-plane series

C.

Then there is naturally attached to f .z/ a Dirichlet f .s/ D

1 X an ; ns

(3.21)

nD1

called a Hecke L-function, which is known to converge absolutely for Re s > k, and which is holomorphic on this domain. Actually, if f .z/ is a cusp form (i.e., a0 D 0) then f .s/ is holomorphic on the domain Re s > 1 C k2 . As in the case of the Riemann zeta function, Hecke theory provides for the meromorphic

28

FLOYD L. WILLIAMS

continuation of f .s/ to the full complex plane, and for an appropriate functional equation for f .s/. For each positive integer n D 1; 2; 3; : : : , there is an operator T .n/ (called a Hecke operator) on the space of modular forms of weight k given by X X nz C da k 1 .T .n/f /.z/ D n d k: (3.22) f 2 d d>0 a2Z =d Z d jn

Here the inner sum is over a complete set of representatives a in Z for the cosets Z =d Z . We shall be interested in the case when f .z/ is an eigenfunction of all Hecke operators: T .n/f D .n/f for all n 1, where f ¤ 0 and .n/ 2 C . We assume also that a1 D 1, in which case f .z/ is called a normalized simultaneous eigenform. For such an eigenform it is known from the theory of Hecke operators that the Fourier coefficients and eigenvalues coincide for n 1 W an D .n/ for n 1. Moreover the Fourier coefficients satisfy the “multiplicative” condition X d k 1 an1 n2 =d 2 (3.23) a n1 a n2 D d>0 d j n1 ;d j n2

for n1 ; n2 1. In particular for a prime p 2 P and an integer n 1, condition (3.23) clearly reduces to the simpler condition anp if p - n, (3.24) an ap D anp C p k 1 anp=p 2 if p j n, which is the origin of condition (3.2), where we see that in the present context we have f .n/ D an and ˛.p/ D p k 1 ; f .n/ here is the function f W Z ! C of condition (3.2), of course, and is not the eigenform f .z/. By Theorem 3.17, therefore, the following strong result is obtained. T HEOREM 3.25 (E ULER PRODUCT FOR H ECKE L- FUNCTIONS ). Let f .z/ be a normalized simultaneous eigenform of weight k (as desribed above), and let f .s/ be its corresponding Hecke L-function given by definition (3.21) for Re s > k. Then, for Re s > k, f .s/ D Q

p2P

1 1 C pk

1 2s

ap p

s

;

(3.26)

where ap is the p-th Fourier coefficient f .z/; see equation (3.20). If , moreover, f .z/ is a cusp form (i.e., the 0-th Fourier coefficient a0 of f .z/ vanishes), then for Re s > 1C k2 , f .s/ converges (in fact absolutely) and formula (3.26) holds. The following example is important, though no proofs (which are quite involved)


29

are supplied. If def

.z/ D e iz=12 is the Dedekind eta function on

C,

1 Y

.1

e 2 inz /

(3.27)

nD1

then the Ramanujan tau function .n/ on

Z C is defined by the Fourier expansion

.z/

24

D

1 X

.n/e 2 inz ;

(3.28)

nD1

which is an example of equation (3.20), where in fact .z/24 is a normalized simultaneous eigenform of weight k D 12. It turns out, remarkably, that every .n/ is real and is in fact an integer. For example, .1/ D 1, .2/ D 24, .3/ D 252, .4/ D 1472, .5/ D 4830. Note also that, since the sum in (3.27) starts at n D 1, .z/24 is a cusp form. By Theorem 3.25 we get: C OROLLARY 3.29. For Re s > 1 C k2 D 7,

1 X .n/ 1 DQ s 11 2s n p2P 1 C p

.p/p

nD1

s

:

(3.30)

The Euler product formula (3.30) was actually proved first by L. Mordell (in 1917, before E. Hecke) although it was claimed earlier to be true by S. Ramanujan. Since we have introduced the Dedekind eta function .z/ in (3.27), we check, as a final point, that it is indeed holomorphic on C . For def

an .z/ D e 2 inz

and

def

G.z/ D

1 Q

.1 C an .z//;

nD1

write .z/ D e iz=12 The product G.z/ converges absolutely on C since P P1G.z/. 2 1 ny (for z D x Ciy, x; y 2 R ; y > 0) is a convergent nD1 jan .z/j D nD1 e 2y geometric series as e < 1. We note also that an .z/ ¤ 1 since (again) 2 ny jan .z/j D e < 1 for n 1. If K C is any compact subset, then the continuous function Im z on K has a positive lower bound B: Im z B > 0 for every z 2 K. Hence P1

jan .z/j D e

2 n Im z

e

2 nB

on K;

where nD1 e 2 nBP is a convergent geometric series as e 2B < 1 for B > 0. Therefore the series 1 nD1 an .z/ converges uniformly on compact subsets of C (by the M-test), which means that the product G.z/ converges uniformly on compact subsets of C . That is, G.z/ is holomorphic on C (as the an .z/ are holomorphic on C ), and therefore .z/ is holomorphic on C .

30

FLOYD L. WILLIAMS

Lecture 4. Modular forms: the movie In the previous lecture we proved an Euler product formula for Hecke Lfunctions, in Theorem 3.25 which followed as a concrete application of Theorem 3.17. That involved, in part, some notions/results deferred to the present lecture for further discussion. Here the attempt is to provide a brief, kaleidoscopic tour of the modular universe, whose space is Lobatchevsky–Poincáre hyperbolic space, the upper half-plane C , and whose galaxies of stars are modular forms. As no universe would be complete without zeta functions, Hecke L-functions play that role. In particular we gain, in transit, an enhanced appreciation of Theorem 3.25. There are many fine texts and expositions on modular forms. These obviously venture much further than our modest attempt here which is designed to serve more or less as a limited introduction and reader’s guide. We recommend, for example, the books of Audrey Terras [35], portions of chapter three, (also note her lectures in this volume) and Tom Apostol [2], as supplements. We begin the story by considering a holomorphic function f .z/ on C that satisfies the periodicity condition f .z C 1/ D f .z/. By the remarks following Theorem B.7 of the Appendix (page 84), f .z/ admits a Fourier expansion (or q-expansion) X X f .z/ D an q.z/n D an e 2 inz (4.1) n2Z

n2Z

def 2 iz De ,

on C , where q.z/ and where the an are given by formula (B.6). We say that f .z/ is holomorphic at infinity if an D 0 for every n 1: f .z/ D

1 X

an e 2 inz

(4.2)

nD0

on C . Let G D SL.2; R / denote the group of 22 real matrices g D ac db with determinant D 1, and let D SL.2; Z / G denote the subgroup of elements

D ac db with a; b; c; d 2 Z . The standard linear fractional action of G on C , given by def az C b gz D 2 C for .g; z/ 2 G C (4.3) cz C d restricts to any subgroup of G, and in particular it restricts to . A (holomorphic) modular form of weight k 2 Z , k 0, with respect to , is a holomorphic function f .z/ on C that satisfies the following two conditions: (M1) f . z/ D .cz C d/k f .z/ for D ac db 2 , z 2 C .

(M2) f .z/ is holomorphic at infinity.


31

def Here we note that for the case of D T D 10 11 2 , we have z D z C 1 by (4.3). Then f .z C 1/ D f .z/ by (M1), which means that condition (M2) is well-defined, and therefore f .z/ satisfies equation (4.2), which justifies the statement of equation (3.20) of Lecture 3. One can also consider weak modular forms, where the assumption that an D 0 for every n 1 is relaxed to allow finitely many negative Fourier coefficients to be nonzero. One can consider, moreover, modular forms with respect to various subgroups of . There are def two other quick notes to make. First, if D 1 D 10 01 2 , then by (4.3) and (M1) we must have f .z/ D . 1/k f .z/ which means that f .z/ 0 if k is odd. For this reason we always assume that k is even. Secondly, condition (M1) is equivalent to the following two conditions: (M1)0 f .z C 1/ D f .z/, and

(M1)00 f . 1=z/ D z k f .z/ for z 2 C .

For we have already noted that (M1) ) (M1)0 , by the choice D T . Also def 0 1 choose D S D 1 0 2 . Then by (4.3) and (M1), condition (M1)00 follows. Conversely, the conditions (M1)0 and (M1)00 together, for k 0 even, imply condition (M1) since the two elements T; S 2 generate ; a proof of this is provided in Appendix F (page 96). Basic examples of modular forms are provided by the holomorphic Eisenstein series Gk .z/, which serve in fact as building blocks for other modular forms: def

Gk .z/ D

X

.m;n/2Z Z f.0;0/g

1 .m C nz/k

(4.4)

for z 2 C , k D 4; 6; 8; 10; 12; : : : . The issue of absolute or uniform convergence of these series rests mainly on the next observation, whose proof goes back to Chris Henley [2]. Given A; ı > 0 let def

SA;ı D f.x; y/ 2 R 2 j jxj A; y ıg be the region C , as illustrated:

y

ı A

(4.5)

0

x A

32

FLOYD L. WILLIAMS

L EMMA 4.6. There is a constant K D K.A; ı/ > 0, depending only on A and ı, such that for any .x; y/ 2 SA;ı and .a; b/ 2 R 2 with b ¤ 0 the inequality

holds. In fact one can take

.a C bx/2 C b 2 y 2 K a2 C b 2 def

KD

ı2 : 1 C .A C ı/2

(4.7)

(4.8) def

P ROOF. Given .x; y/ 2 SA;ı and .a; b/ 2 R 2 with b ¤ 0, let q D a=b. Then (4.7) amounts to .q C x/2 C y 2 K: 1 C q2 Two cases are considered. First, if jqj ACı, then 1Cq 2 1C.ACı/2 , so 1 1 : 2 1C q 1 C .A C ı/2

Also .q C x/2 C y 2 y 2 ı 2 (since y ı for .x; y/ 2 SA;ı ). Therefore ı2 .q C x/2 C y 2 D K; 1 C q2 1 C .A C ı/2

with K as in (4.8). If instead jqj > ACı, we have 1=jqj < 1=.ACı/, so jxj=q jxj=.ACı/. Use the triangular inequality and the fact that jxj A for .x; y/ 2 SA;ı to write ˇ ˇ ˇ ˇ ˇ ˇ ˇ ˇ jxj A ı ˇ1 C x ˇ 1 ˇ x ˇ 1 1 D : ˇqˇ ˇ ˇ q ACı ACı ACı

That is, jq C xj jqj

ı q2ı2 , which implies .q C x/2 , or again ACı .ACı/2

.q C x/2 ı2 q2 .q C x/2 C y 2 : 1 C q2 1 C q2 .ACı/2 1Cq 2 def

(4.9)

On the other hand, f .x/ D x 2 =.1 C x 2 / is a strictly increasing function on .0; 1/ since f 0 .x/ D 2x=..1 C x 2 /2 / is positive for x > 0. Thus, since jqj > A C ı, we have

which leads to

q2 .A C ı/2 D f .jqj/ > f .A C ı/ D ; 1 C q2 1 C .A C ı/2 ı2 .ACı/2 .q C x/2 C y 2 1 C q2 .ACı/2 1C.ACı/2


33

by the second inequality in (4.9). But the right-hand side is again the constant K of (4.8). This concludes the proof. ˜ 2 2 2 2 .aCbx/ Cb y a Now suppose b D 0, but a ¤ 0. Then D 2 D 1 > 12 ; which a2 Cb 2 a by Lemma 4.6 says that .a C bx/2 C b 2 y 2 def K1 D min 2 2 a Cb

1 2; K

for .x; y/ 2 SA;ı , .a; b/ 2 R R f.0; 0/g. Hence if z 2 SA;ı , say z D x Ciy, and .m; n/ 2 Z Z f.0; 0/g, we get jmCnzj2 D .mCnx/2 Cn2 y 2 .m2 Cn2 /K1 D ˛=2 K1 jm C nij2 . This implies, for ˛ 0, that jm C nzj˛ K1 jm C nij˛ , or 1 1 : ˛=2 ˛ jm C nzj K1 jm C nij˛

(4.10)

def

Moreover — setting for convenience Z 2 D Z Z f.0; 0/g — we know from P results in Appendix G that .m;n/2Z 2 1=jm C nij˛ converges for ˛ > 2. This shows that Gk .z/ converges absolutely and uniformly on every SA;ı for k > 2, which (since k is even) is why we take k D 4; 6; 8; 10; 12; : : : in (4.4). In particular, since any compact subset of C is contained in some SA;ı , the holomorphicity of Gk .z/ on C is established. Since the map .m; n/ ‘ .m n; n/ is a bijection of Z 2 , we have P 1 D k .m;n/2Z 2 .m C n C nz/ .m;n/2Z 2 .m

Gk .z C 1/ D

P

1 D Gk .z/: n C n C nz/k

Similarly, Gk

X 1 D z

.m;n/2Z 2

1 .m n=z/k

D zk

X

.m;n/2Z 2

1 .mz n/k

D zk

X

1 ; .nzCm/k 2

.m;n/2Z

since the map .m; n/ ‘ .n; m/ is a bijection of Z 2 . This shows that Gk .z/ satisfies the conditions (M1)0 and (M1)00 . To complete the argument that the Gk .z/, for k even 4, are modular forms of weight k, we must check condition (M2). Although this could be done more directly, we take the route whereby the Fourier coefficients of Gk .z/ are actually computed explicitly. For this, consider the function def

k .z/ D

X

m2Z

1 .z C m/k

(4.11)

34

FLOYD L. WILLIAMS

on C for k 2 Z , k 2. The inequality (4.10) gives 1 1 1 k=2 D k=2 k jm C zj K1 jm C ijk K1 .m2 C 1/k=2

(4.12)

for z 2 SA;ı . Since X

m2Z

1 1 X X 1 1 1 D 1 C 2 1 C 2 1, we see that k .z/ converges absolutely and uniformly on every SA;ı , and is therefore a holomorphic function on C such that k .z C 1/ D

X

m2Z

X 1 1 D D k .z/: k .z C m C 1/ .z C m/k m2Z

Thus (again) there is a Fourier expansion X k .z/ D an .k/e 2 inz

(4.13)

n2Z

on C , where by formula (B.6) of page 84 (with the choice b1 D 0, b2 D 1) an .k/ D

Z

0

1

k .t C ib/e

2 in.tCib/

dt

(4.14)

for n 2 Z , b > 0. P ROPOSITION 4.15. In the Fourier expansion (4.13), an .k/ D 0 for n 0 and an .k/ D . 2 i/k nk 1 =.k 1/! for n 1. Therefore 1 . 2 i/k X k k .z/ D n .k 1/!

1 2 inz

e

nD1

is the Fourier expansion of the function k .z/ of (4.11) on C . Here k 2 Z , k 2 as in (4.11). def

P ROOF. For fixed n 2 Z and b > 0, define hm .t / D e 2 int=.t CibCm/k on Œ0; 1, for m 2 Z . Since .t; b/ 2 S1;b for t 2 Œ0; 1 according to (4.5), the inequality in (4.12) gives 1 jhm .t /j k=2 : K1 .m2 C 1/k=2


35

P P As we have seen, m2Z 1=.m2 C 1/k=2 < 1 for k > 1, so m2Z hm .t / converges uniformly on Œ0; 1. By (4.11) and (4.14), therefore, we see that Z 1X XZ 1 2 nb 2 nb an .k/ D hm .t /e dt D e hm .t / dt 0 m2Z

D e 2 nb

m2Z

XZ

m2Z

0

1

0

2 int

e dt: .t C ib C m/k

(4.16)

By the change of variables x D t C m, we get Z 1 Z mC1 2 in.x m/ Z mC1 e 2 int dt e e 2 inx D dx D dx; k .x C ib/k .x C ib/k 0 .t C ib C m/ m m

so

XZ

m2Z

1 0

e 2 int dt .t C ib C m/k X Z mC1 e 2 inx dx D .x C ib/k m2Z m 1 Z mC1 1 Z mC1 X X e 2 inx e 2 inx dx C dx D .x C ib/k .x C ib/k m m mD1 mD0 Z 1 Z 0 Z 1 2 inx e e 2 inx e 2 inx D dx C D dx: (4.17) k k k 0 .x C ib/ 1 .x C ib/ 1 .x C ib/

(Note that the integrals on the last line are finite for k > 1, since ˇ ˇ ˇ e 2 inx ˇ 1 ˇ ˇ ˇD 2 ˇ k ˇ .x C ib/ ˇ .x C b 2 /k=2

(4.18)

and the map x ‘ 1=.x 2 C b 2 / lies in L1 .R ; dx/ for 2 > 1.) From (4.16) and (4.17), we have Z 1 e 2 inx an .k/ D e 2 nb dx: (4.19) k 1 .x C ib/ In particular, Z dx 2 nb jan .k/j e 2 k=2 ; x R bk C1 b by (4.18). Setting t D x=b, we can rewrite the right-hand side as Z b dt e 2 nb ck e 2 nb D ; b k R .t 2 C 1/k=2 bk 1

(4.20)

36

FLOYD L. WILLIAMS

def R where ck D R dt =.t 2 C 1/k=2 < 1 for k > 1. Since b > 0 is arbitrary, we let b ! 1. For n D 0 or for n < 0, we see by the inequality (4.20) that an .k/ D 0. Also for n D 1; 2; 3; 4; : : : and k > 1, it is known that Z k e 2 inx 2 nb . 2 i/ dx D e nk 1 I (4.21) k .k 1/! R .x C ib/

see the Remark below. The proof of Proposition 4.15 is therefore completed by way of equation (4.19). ˜ R EMARK . Equation (4.21) follows from a contour integral evaluation: Z 1Cib .2/k k 1 e k i=2 e 2 iz dz D .k/ zk 1Cib

(4.22)

where ; b > 0, k > 1. The left-hand side here is Z 1 Z 1 e 2 i.xCib/ e 2 ix 2b dx D e dx: k k 1 .x C ib/ 1 .x C ib/ R1 Thus we can write 1 e 2 ix dx=.x C ib/k D e 2b . 2 i/k k 1=.k 1/! for ; b > 0 and k > 1 an integer. The choice D n (n D 1; 2; 3; 4; : : : ) gives (4.21). For n 2 Z , define n .z/

def

D k .nz/ D

X

m2Z

1 .nz C m/k

(see (4.11) for the lastPequality) on C . In n D 0 contriPdefinition k(4.4), Pthe 1 k bution to the sum is m2Z f0g 1=mk D 1 1=m C mD1 mD1 1=. m/ D P1 2 mD1 1=mk (since k is even) D 2.k/: Thus we can write Gk .z/ D 2.k/ C P P P1 P1 k n .z/. But n2Z f0g m2Z 1=.m C nz/ D 2.k/ C nD1 n .z/ C nD1 n .z/ D n .z/, again because k is even (the easy verification is left to the reader). Therefore 1 1 X X k .nz/ .z/ D 2.k/ C 2 Gk .z/ D 2.k/ C 2 n nD1

nD1

D 2.k/ C

2.2 i/k .k

1/!

1 1 X X

mk

1 2 imnz

e

nD1 mD1

(by Proposition 4.15, for k even), leading to Gk .z/ D 2.k/ C

1 2.2 i/k X k .k 1/! nD1

by formula (D.8) of Appendix D. This proves:

1 .n/e

2 inz

;


37

T HEOREM 4.23. The holomorphic Eisenstein series Gk .z/, k D 4; 6; 8; 10; 12; : : : , defined in (4.4) satisfy conditions (M1)0 , (M1)00 , and are holomorphic at infinity. In fact, Gk .z/ has Fourier expansion (4.2), where a0 D 2.k/ and an D

2.2 i/k k .k 1/!

def

1 .n/ D

2.2 i/k X k d .k 1/!

1

d>0 d jn

for n 1. The Gk .z/ are therefore modular forms of weight k. Since k is even in Theorem 4.23, formula (2.1) applies to .k/. As mentioned, a modular form is a cusp form if its initial Fourier coefficient a0 in equation (4.2) vanishes. By Theorem 4.23 the Gk .z/, for example, are not cusp forms since .k/ ¤ 0 for k even, k 4. In fact we know (by Theorem 3.17) that since .s/ is given by an Euler product, it is nonvanishing for Re s > 1. We return now to the discussion of Hecke L-functions, where we begin with results on estimates of Fourier coefficients of modular forms. The Gk .z/ already provide the example of how the general estimate looks. This involves only an estimate of the divisor function .n/ for > 1, where we first note that d > 0 runs through the divisors of n 2 Z , n 1, as does n=d: def

.n/ D

X

0 0, depending only on f and k, such that jf .z/j < M.f; k/.Im z/

k=2

(4.27)

on C . Once (4.25) is established for a cusp form, the weaker result jan j < C.f; k/nk 1 , n 1, for an arbitrary modular form f .z/ of weight k follows from the fact that it holds for Gk .z/ (as shown above), and the fact that f .z/ differs from a cusp form (where one can apply the inequality (4.25)) by P 2 inz , a constant multiple of Gk .z/. In fact, write f .z/ D a0 C 1 nD1 an e P1 Gk .z/ D b0 C nD1 bn e 2 inz (by equation (4.2)) where b0 D 2.k/, for examdef ple (by Theorem 4.23). Then f0 .z/ D f .z/ .a0 =b0 / Gk .z/ is a modular form of weight k, with Fourier expansion f0 .z/ D a0 C D

1 X

nD1

1 X

nD1

an e 2 inz

an

a0 b0 b0

1 X a0 bn e 2 inz b0

nD1

a0 bn 2 inz e ; b0

which shows that f0 .z/ is a cusp form such that f .z/ D f0 .z/ C

a0 Gk .z/: b0

(4.28)

To complete our sketch of the proof of Theorem 4.24, details of which can be found in section 6.15 of [2], for example, we should add further remarks regarding F . By definition, a fundamental domain for the action of on C is an open set F C such that (F1) no two distinct points of F lie in the same -orbit: if z1 ; z2 2 F with z1 ¤ z2 , then there is no 2 such that z1 D z2 . We also require condition: (F2) given z 2 C , there exists some 2 such that z 2 FN (= the closure of F ). The standard fundamental domain, as is well-known, is given by def ˚ (4.29) F D z 2 C j jzj > 1; j Re zj < 12 ; and is shown at the the top of the next page.


39

F is this interior region

1 2

0

x

1 2

If Mk . / and Sk . / denote the space of modular forms and cusp forms of weight k D 4; 6; 8; 10; : : : , respectively, there is the C -vector space direct sum decomposition Mk . / D Sk . / ˚ C Gk ;

(4.30)

by (4.28). The sum in (4.30) is indeed direct since (as we have seen) Gk .z/ is not a cusp form. If f 2 Mk . / with Fourier expansion (4.2), the corresponding Hecke Lfunction L.sI f / is given by definition (3.21): 1 X an : L.sI f / D f .s/ D ns def

(4.31)

nD1

By Theorem 4.24 this series converges absolutely for Re s > k, and for Re s > 1C k2 if f 2 Sk . /. On these respective domains L.sI f / is holomorphic in s (by an argument similar to that for the Riemann zeta function), as we have asserted in Lecture 3. Since Theorem 3.25 is based on equation (3.24), which is based on equation (3.23), our proof of it actually shows the following reformulation: T HEOREM 4.32. Suppose the Fourier coefficients an of f 2 Mk . / satisfy the multiplicative condition (3.23), with at least one an nonzero: a n1 a n2 D

X

dk

d>0 d j n1 ; d j n2

1 a n1 n2 d2

(4.33)

40

FLOYD L. WILLIAMS

for n1 ; n2 1. Then L.sI f / has the Euler product representation Y 1 L.sI f / D k 1 2s 1C p ap p s

(4.34)

p2P

for Re s > k. If f 2 Sk . /, equation (4.34) holds for Re s > 1 C k=2. Here, we only need to note that a1 D 1. For if some an ¤ 0, then by (4.33) an a1 D an1 =12 D an and so a1 D 1. Theorem 4.32 raises the question of finding modular forms whose Fourier coefficients satisfy the multiplicative condition (3.23) = condition (4.33). This question was answered by Hecke (in 1937), who found, in fact, all such forms. As was observed in Lecture 3, the multiplicative condition is satisfied by normalized simultaneous eigenforms: nonzero forms f .z/ with a1 D 1, that are simultaneous eigenfunctions of all the Hecke operators T .n/; n 1; see definition (3.22). More concretely, among the non-cusp forms the normalized simultaneous eigenforms turn out to be the forms f .z/ D

.k 1/! Gk .z/; 2.2 i/k

where indeed, by Theorem 4.23, a1 D k 1 .1/ D 1. We mention that the Hecke operators fT .n/gn1 map the space Mk . / to itself, and also mapPthe space Sk . / to itself. For f 2 Mk . / with Fourier 2 inz on C , as in (4.2), .T .n/f /.z/ has Fourier expansion f .z/ D 1 nD0 an e expansion 1 X 2 imz a.n/ .T .n/f /.z/ D m e mD0

.n/ a0

P .n/ on C , where D a0 k 1 .n/ and am D 0 k. Here the integral Jf .s/ D

1 1

f .i t /

a0 t s

1

dt

(4.36)


41

is an entire function of s. Notice that equation (4.35) is similar in form to (1.12). Again since 1= .s/ is an entire function, and since .1=s/ .1= .s// D 1= .s C 1/, we can write equation (4.35) as L.sI f / D

.2/s Jf .s/ C i k Jf .k .s/

.2/s a0 i k s/ C .s/ .s k/

.2/s a0 .sC1/

(4.37)

for Re s > k. If f .z/ is a cusp form, we see that L.sI f / extends to an entire function by way of the first term in (4.37). In general, we see that L.sI f / extends meromorphically to C , with a single (simple) pole at s D k with residue .2/k a0 i k = .k/ D .2/k a0 i k =.k 1/!. Also by equation (4.37), for k s ¤ k (i.e., s ¤ 0), we have i k .2/s k

k

D i .2/

.k

s/L.k

sI f /

.2/k s a0 i k .k s/ Jf .k s/ C i k Jf .s/ C .k s/ s k a0 i a0 s/ C Jf .s/ C s k s

s k

D i k Jf .k

a0 .k s/ .k sC1/

(again since .w C 1/ D w .w/, and since i 2k D 1 for k even); the right-hand side in turn equals .2/ s .s/L.sI f /. That is, .2/

s

.s/L.sI f / D i k .2/s

k

.k

s/L.k

sI f /

(4.38)

for s ¤ 0, which is the functional equation for L.sI f /, which compares with the functional equation for the Riemann zeta function; see [17; 18]. The Eisenstein series Gk .z/ can be used as building blocks to construct other modular forms. It is P known that any modular form f .z/ is, in fact, a finite sum of the form f .z/ D n;m0 cnm G4 .z/n G6 .z/m for suitable complex numbers cnm . Of particular interest are the discriminant form def

.z/ D .60G4 .z//3

27 .140G6 .z//2

(4.39)

and the modular invariant def

J.z/ D .60G4 .z//3 =.z/

(4.40)

which is well-defined since it is true that .z/ never vanishes on C . .z/ is a modular form of weight 12, since if f1 .z/; f2 .z/ are modular forms of weight k1 ; k2 , then f1 .z/f2 .z/ is a modular form of weight k1 C k2 . Similarly J.z/ is a weak modular form of weight k D 0: J. ; z/ D J.z/ for 2 , z 2 C . The form J.z/ was initially constructed by R. Dedekind in 1877, and by F. Klein in 1878. Associated with it is the equally important modular j-invariant def

j .z/ D 1728J.z/:

(4.41)

42

FLOYD L. WILLIAMS

.z/ is connected with the Dedekind eta function .z/ (see definition (3.27)) by the Jacobi identity .z/ D .2/12 .z/24 ; (4.42)

which with equation (3.28) shows that .z/ has Fourier expansion .z/ D .2/12

1 P

.n/e 2 inz ;

(4.43)

nD1

where .n/ is the Ramanujan tau function, and which in particular shows that .z/ is a cusp form: .z/ 2 S12 . /, which can also be proved directly by definition (4.39) and Theorem 4.23, for k D 4; 6. There are no nonzero cusp forms of weight < 12. Note that by Theorem 4.24, there is a constant C > 0 such that j.n/j < C n6 for n 1. However P. Deligne proved the Ramanujan conjecture j.n/j 0 .n/n11=2 for n 1, where 0 .n/ is the number of positive divisors of n. As we remarked in Lecture 3, the .n/ (remarkably) are all integers. This can be proved using definition (4.39) and Theorem 4.23, for k D 4; 6. It is also true that, thanks to the factor 1728 in definition (4.41), all of the Fourier coefficients of the modular j-invariant are integers: j .z/ D 1e with each an 2 Z :

2 iz

C

1 P

an e 2 inz

(4.44)

nD0

a0 D 744; a1 D 196;884; a2 D 21;493;760; a3 D 864;299;970;

a4 D 20;245;856;256; a5 D 333;202;640;600; : : :

(4.45)

An application of the modular invariant j .z/, and of the values in (4.45), to three-dimensional gravity with a negative cosmological constant will be given in my Speaker’s Lecture; see especially equation (5-8) on page 343 and the subsequent discussion. In the definition (4.4) of the holomorphic Eisenstein series Gk .z/, one cannot take k D 2 for convergence reasons. However, Theorem 4.23 provides a suggestion of how one might proceed to construct a series G2 .z/. Namely, take k D 2 there and thus define def

G2 .z/ D 2.2/C2.2 i/2

nD1

.n/e 2 inz D

1 P 2 .n/e 2 inz (4.46) 8 2 3 nD1

def P on where .n/ D 1 .n/ D 0 0 on K, so j .n/e 2 inz j 21 n.n C 1/e P P 1 1 2 nB < 1 for B > 0. By the M -test, 2 inz connD1 n.nC1/e nD1 .n/e verges uniformly on K, which (by the Weierstrass theorem) means that G2 .z/ is a holomorphic function on C . Another expression for G2 .z/ is X X 1 : (4.47) G2 .z/ D 2.2/ C 2 .m C nz/ m2Z n2Z f0g

To check this, start by taking k D 2 in definition (4.11) and in Proposition 4.15: X 1 C z2

1

m2Z f0g

X 1 2 ke 2 ikz : D .z/ D . 2 i/ 2 .z C m/2

(4.48)

kD1

Replace z by nz in (4.48) and sum on n from 1 to 1: 1

X 1 .2/ C z2

1 1 X X 1 2 ke 2 ik nz : D . 2 i/ .nz C m/2

X

nD1 m2Z f0g

(4.49)

nD1 kD1

By (D.6) (see page 90), we obtain .n/ D

1 X

d.k; n/k:

(4.50)

kD1

P def For an D e 2 inz , n 1, z 2 C , and for k 1 fixed the series 1 nD1 d.k; n/an clearlyP converges absolutely, since Im z P > 0 and 0 d.k; n/ 1. Then the 1 series 1 a converges and equals nD1 k n nD1 d.k; n/an , by the Scholium of Appendix D (page 91): 1 X

e

2 ik nz

nD1

D

1 X

d.k; n/e 2 inz

(4.51)

nD1

which gives, by equation (4.50) 1 X

nD1

.n/e 2 inz D D

1 X 1 X

d.k; n/e 2 inz D

nD1 kD1 1 1 X X

k

kD1

nD1

e 2 ik nz D

1 X

kD1

1 1 X X

k

1 X

d.k; n/e 2 inz

nD1

ke 2 ik nz ;

(4.52)

nD1 kD1

which in turn allows for the expression 1

X 2 .2/ C 2 z2

X

nD1 m2Z f0g

1

X 1 2 .n/e 2 inz D 2.2 i/ .nzCm/2 nD1

D G2 .z/

2.2/

(4.53)

44

FLOYD L. WILLIAMS

by equation (4.49), provided the commutations of the summations over k; n in (4.52) are legal. But for y D Im z, 1 1 P P

nD1 kD1

jd.k; n/ke 2 inz j D

1 1 P P

d.k; n/ke

nD1 kD1

P1

2 ny

D

j .n/e 2 inz j,

1 P

.n/e

2 ny

nD1

by (4.50), and this equals nD1 which is finite, as we have seen. This justifies the first commutation. Similarly, 1 1 1 1 1 1 P P P P P P k e 2ky D k k .e 2ky /n D jke 2 ik nz j D 2ky 2ky 1 e 1 kD1 e kD1 kD1 nD1 kD1 nD1 is finite by the integral test: Z 1 Z 1 t dt 1 u du < 1; D u 2yt 2 e 1 e 1 .2y/ 1 2y

(4.54)

as we shall see later by Theorem 6.1, for example. This justifies the second commutation in equation (4.52). Since for n 1 X X X 1 1 1 D D ; (4.55) .m nz/2 . m nz/2 .m C nz/2 m2Z f0g

m2Z f0g

m2Z f0g

the double sum on the right-hand side of (4.47) can be written as 1 X X 1 1 C .nz/2 .m C nz/2

n2Z f0g

m2Z f0g

D2 D2

1 X

nD1

1

X 1 C n2 z 2

X

nD1 m2Z f0g

1

X .2/ C 2 z2

X

nD1 m2Z f0g

1

X 1 C .m C nz/2

1 ; .m C nz/2

X

nD1 m2Z f0g

1 .m nz/2 (4.56)

which is the left-hand side of (4.53). That is, G2 .z/ 2.2/ equals the double sum on the right-hand side of (4.47), proving (4.47). Using equation (4.47) one can eventually show that G2 .z/ satisfies the rule 1 D z 2 G2 .z/ 2 iz: (4.57) G2 z Because of the term 2 iz in (4.57), G2 .z/ is not a modular form of weight 2. That is, condition (M1)00 above is not satisfied, although condition (M1)0 is: G2 .z C 1/ D G2 .z/ by definition (4.46). Equation (4.57) also follows by a transformation property of the Dedekind eta function, whose logarithmic derivative turns out to be a constant multiple of G2 .z/. Thus we indicate now an alternative derivation of the rule (4.57).


45

def

On the domain D D fw 2 C j jwj < 1g, the holomorphic function 1 C w is nonvanishing and it therefore has a holomorphic logarithm g.w/ that can def beP chosen so as to vanish at w D 0: e g.w/ D 1 C w on D. In fact g.w/ D def 1 n =n. Again for q.z/ D e 2 iz , z 2 C , consider the n-th partial nD1 . w/P def sum sn .z/ D nkD1 g. q.z/k /, which is well-defined, because q.z/ 2 D for z 2 C implies q.z/k 2 D for k > 0. We claim that the series 1 X

def

.z/ D

kD1

g. q.z/k / D

lim sn .z/

n!1

(4.58)

P P1 k n converges. We have .z/ D 1 kD1 nD1 .q.z/ / =n, where for y D Im z (again) ˇ ˇ 1 1 X 1 ˇ 1 k nˇ X ˇ .q.z/ / ˇ X 1 X .e 2 ny /k ˇ ˇD ˇ ˇ n n nD1 kD1 nD1 kD1 X 1 1 X 1 e 2 ny 1 1 D ; D n 1 e 2 ny n e 2 ny 1 nD1

nD1

which is finite test; comparePwith (4.54), This allows Pby thePintegral P1 for example. 1 1 k /n =n D k /n =n, and shows us to write 1 .q.z/ .q.z/ nD1 kD1 kD1 nD1 Q1 the finiteness of these series. Hence .z/ is finite. Now q.z/n / D nD1 .1 Qn Qn k/ k g. q.z/ limn!1 kD1 .1 q.z/ / D limn!1 kD1 e , by the definition of g.w/, and this equals e .z/ by (4.58). That is, for the Dedekind eta function .z/ D e

iz=12

1 Y

q.z/n /

.1

(4.59)

nD1

on C defined in (3.27) we see that .z/ D e iz=12

.z/

:

(4.60)

def

Differentiation of the equation e g.w/ D 1Cw gives g 0 .w/ D 1=e g.w/ D 1=.1Cw/ (of course), which with termwise differentiation of (4.58) (whose justification we skip) gives 0

.z/ D

1 X

0

g . q.z/ /. kq.z/

kD1 1 X

D 2 i

k

kD1

1 X kq.z/k q .z// D 2 i 1 q.z/k

k 1 0

kD1

k

1 X

.q.z/k /n D 2 i

nD1

1 X

nD1

.n/e 2 inz ;

46

FLOYD L. WILLIAMS

by equation (4.52). Therefore, by (4.60), we obtain i 1 P 2 i .n/e 2 inz ; 0 .z/ D .z/ 12 nD1

or

0 .z/ D .z/

G2 .z/ 4 i

(4.61)

by definition (4.46). Now .z/ satisfies the known transformation rule p 1 D e i=4 z.z/; z

where we take arg z 2 . ; /. Differentiation gives p p 0 z i=4 1 1 .z/ z .z/ C e 0 0 .z/ 1 2z z z2 D D C (4.62) 1 1 .z/ 2z z z 1 which by equation (4.61) says that G2 1z =4 iz 2 D G2 .z/=4 i C . This 2z is immediately seen to imply the transformation rule (4.57). In the lectures of Geoff Mason and Michael Tuite the particular normalization Gk .z/=.2 i/k of the Eisenstein series is considered, which they denote by Ek .z/. In particular, by definition (4.46), E2 .z/ D

1

X 1 .n/e 2 inz : C2 12 nD1

However, other normalized Eisenstein series appear in the literature that also might be denoted by Ek .z/. For example, in [20] there is the normalization def (and notation) Ek .z/ D Gk .z/=2.k/.

Lecture 5. Dirichlet L-functions Equation (3.19) has an application to Dirichlet L-functions, which we now consider. To construct such a function, we need first a character modulo m, where m > 0 is a fixed integer. This is defined as follows. Let Um denote the def group of units in the commutative ring Z .m/ D Z =mZ . Thus if nN D n C mZ denotes the coset of n 2 Z in Z .m/ , we have nN 2 Um ( ) 9aN 2 Z .m/ such N that aN nN D 1. One knows of course that nN 2 Um ( ) .n; m/ D 1 (i.e. n and m are relatively prime). A character modulo m is then (by definition) a group def homomorphism W Um ! C D C f0g. For our purpose, however, there is an


47

equivalent way of thinking about characters modulo m. In fact, given , define Z W Z ! C by N if .n; m/ D 1, def .n/ Z .n/ D (5.1) 0 otherwise, for n 2 Z . For n; n1 ; n2 2 Z , Z satisfies: (D1) Z .n/ D 0 ( ) .n; m/ ¤ 1;

(D2) Z .n1 / D Z .n2 / when nN 1 D nN 2 in Z .m/ ; (D3) Z .n1 n2 / D Z .n1 /Z .n2 / when .n1 ; m/ D 1 and .n2 ; m/ D 1.

Conversely, suppose 0 W Z ! C is a function that satisfies the three conditions (D1), (D2), and (D3). Define W Um ! C by .n/ N D 0 .n/ for n 2 Z such that .n; m/ D 1. The character is well-defined by (D2), and W Um ! C by (D1). By (D3), .ab/ D .a/.b/ for a; b 2 Um , so we see that is a character modulo m. Moreover the induced map .0 /Z W Z ! C given by definition (5.1) coincides with 0 . Note that Z is completely multiplicative: (D4) Z .n1 n2 / D Z .n1 /Z .n2 / for all n1 ; n2 2 Z .

For if either .n1 ; m/ ¤ 1 or .n2 ; m/ ¤ 1, then .n1 n2 ; m/ ¤ 1, so that by (D1) both Z .n1 /Z .n2 / and Z .n1 n2 / are zero. If both .n1 ; m/ D 1 and .n2 ; m/ D 1, then already Z .n1 n2 / D Z .n1 /Z .n2 / by (D3). Note also that since .a/ 2 C for every a 2 Um (that is, .a/ ¤ 0), we have N D .1N 1/ N D .1/ N .1/ N by (D4), so .1/ N D 1. Moreover since .1; m/ D 1, 0 ¤ .1/ N by (5.1), which in turn equals 1. Z .1/ D .1/, One final property of Z that we need is: (D5) j.a/j D 1 for all a 2 Um ; hence jZ .n/j 1 for all n 2 Z . The proof of (D5) makes use of a little theorem in group theory which says that if G is a finite group with jGj elements, then ajGj D 1 for every a 2 G. Now, N D .ajUm j / D .a/jUm j (since given a 2 Um , we can write (as just seen) 1 D .1/ is a group homomorphism), which shows that .a/ is a jUm j-th root of unity: j.a/j D 1 for all a 2 Um . Hence jZ .n/j 1 for all n 2 Z , by definition (5.1). Given a character modulo m, it follows that we can form the zeta function, or Dirichlet series def

L.s; / D

1 X Z .n/ D ns

nD1

X

.n;m/D1

.n/ N ns

;

(5.2)

called a Dirichlet L-function, which converges for Re s > 1, by (D5). L.s; / is holomorphic on the domain Re s > 1, by the same argument given for the Riemann zeta function .s/. Since Z 6 0 (Z .1/ D 1), and since Z is completely multiplicative, formula (3.19) implies:

48

FLOYD L. WILLIAMS

T HEOREM 5.3 (E ULER Re s > 1. Then

PRODUCT FOR

L.s; / D Q

1

1 Z .p/p

D IRICHLET L- FUNCTIONS ). Assume

s

p2P

D Q

1 Z .p/p

1

p2P p-m

s

:

(5.4)

The second statement of equality follows by (D1), since for a prime p 2 P , saying that .p; m/ ¤ 1 is the same as saying that p j m. As an example, define 0 W Z ! C by 0 .n/ D 1 if .n; m/ D 1, 0 otherwise, for n2 Z . Then 0 satisfies (D1). If n1 ; n2 ; l 2 Z such that n1 D n2 C lm (i.e., nN 1 D nN 2 /, then .n1 ; m/ D 1 ( ) .n2 ; m/ D 1, so 0 satisfies (D2). If .n1 ; m/ D 1 and .n2 ; m/ D 1, then .n1 n2 ; m/ D 1, so 0 also satisfies (D3), and 0 therefore defines a Dirichlet character modulo m. We call 0 (or the induced character Um ! C ) the principal character modulo m. Again since p is a prime, we see by equation (5.4) that for Re s > 1 L.s; 0 / D Q

1 .1 p

s/

:

(5.5)

p2P p-m

Then for Re s > 1 L.s; 0 / Q

p2P pjm

1 .1 p

s/

D Q

1 .1

s/

p

D .s/

(5.6)

p2P

by formula (0.2). That is, L.s; 0 / D .s/

Y

p2P pjm

.1

p

s

/

(5.7)

for Re s > 1. If 0 W Um ! C also denotes the character modulo m induced by 0 W Z ! C , then 0 .n/ N D 1 for every nN 2 Um (by (5.1)) since .n; m/ D 1. N 1; N 2; N 3; N 4g, N and it is As another simple example, take m D 5: Z.5/ D f0; def N 2; N 3; N 4g. N Moreover the equations .1/ N D .4/ N def easily checked that U5 D f1; D 1, def def N D .3/ N D 1 define a character .5/ D W U5 ! C modulo 5. The .2/ induced map Z W Z ! C in definition (5.1) is given by Z .1/ D 1, Z .2/ D 1, Z .3/ D 1, Z .4/ D 1, Z .5/ D 0 (since .5; 5/ ¤ 1), Z .6/ D 1, Z .7/ D 1, N 7N D 2, N 8N D 3, N 9N D 4, N with .n; 5/ D 1 Z .8/ D 1, Z .9/ D 1, . . . (since 6N D 1,


49

for n D 6; 7; 8; 9). The corresponding Dirichlet L-function is therefore given, for Re s > 1, by 1 1 1 1 1 1 C sC s C s ˙; s s s 3 4 6 7 8 9 N N N N N N N N N N 5; N 7g: N The map Next take m D 8 W Z.8/ D f0; 1; 2; 3; 4; 5; 6; 7g, U8 D f1; 3; def def def def N D .7/ N D 1, .3/ N D .5/ N D 1 is a character .8/ D W U8 ! C given by .1/ modulo 8, with Z W Z ! C in definition (5.1) given by Z .1/ D 1, Z .2/ D 0, Z .3/ D 1, Z .4/ D 0, Z .5/ D 1, Z .6/ D 0, Z .7/ D 1, Z .8/ D 0, Z .9/ D 1, Z .10/ D 0, Z .11/ D 1, Z .12/ D 0, Z .13/ D 1, Z .14/ D 0, Z .15/ D 1, Z .16/ D 0, Z .17/ D 1, . . . . Then L.s; .5/ / D 1

1 2s

1 1 1 1 1 1 1 1 C C C C ˙: 3s 5s 7s 9s 11s 13s 15s 17s From formula (5.7) it follows that the L-function L.s; 0 / admits a meromorphic continuation to the full s D 1 as its only singularity — Qcomplex plane,1 with a simple pole with residue p2P; p j m 1 p . If ¤ 0 it is known that L.s; / at least extends to Re s > 0 and, moreover, that L.1; / ¤ 0. For example, for the characters .5/ , .8/ modulo 5 and 8, respectively, constructed in the previous examples, one has p p 1 1 3C 5 .5/ L.1; / D p log ; L.1; .8/ / D p log.3 C 2 2/: 2 5 8 If ¤ 0 is a primitive character modulo m, a notion that we shall define presently, then L.s; / does continue meromorphically to C , and it has a decent functional equation. L.s; .8/ / D 1

First we define the notion of an imprimitive character. Suppose k > 0 is a divisor of m. Then there is a natural (well-defined) map q W Z =mZ ! Z =k Z , given by def q.n C mZ / D n C k Z for n 2 Z . If .n; m/ D 1 then .n; k/ D 1 since k j m. def Therefore the restriction q D qjUm maps Um to Uk , and is a homomorphism between these two groups. Let .k/ W Uk ! C be a character modulo k. By definition, .k/ is a homomorphism and hence so is .k/ ı q W Um ! C . That def is, given a positive divisor k of m we have an induced character D .k/ ı q modulo m. Characters modulo m that are induced this way, say for k ¤ m, are called imprimitive. is called a primitive character if it is not imprimitive, in which case m is also called the conductor of . Thus for a primitive character modulo m, the L-function L.s; / satisfies a theory similar to (but a bit more complicated than) that of the Riemann zeta function .s/. In the Introduction we referred to the prime number theorem, expressed in equation (0.3), as a monumental result, and we noted quite briefly the role of .s/ in its proof. Similarly, the study of the L-functions L.s; / leads to a

50

FLOYD L. WILLIAMS

monumental result regarding primes in an arithmetic progression. Namely, in 1837 Dirichlet proved that there are infinitely many primes in any arithmetic progression n, n C m, n C 2m, n C 3m, : : : , where n; m are positive, relatively prime integers — a key aspect of the proof being the fact (pointed out earlier) that L.1; / ¤ 0 if ¤ 0 . Dirichlet’s proof relates, moreover, L.1; / to a Gaussian class number — an invariant in the study of binary quadratic forms. One can obtain, also, a prime number theorem for arithmetic progressions (from the Siegel–Walfisz theorem), where the counting function .x/ in (0.3) is replaced def by the function .xI m; n/ D the number of primes p x, with p n.mod m/, for n; m relatively prime. One can also formulate and prove a prime number theorem for graphs. This is discussed in section 3.3 of the lectures of Audrey Terras.

Lecture 6. Radiation density integral, free energy, and a finite-temperature zeta function Theorems 1.13 and 1.18 provide for integral representations of .s/, for Re s > 1, that serve as starting points for its analytic continuation. The following, nice integral representation also serves as a starting point. We apply it to compute Planck’s radiation density integral. We also consider a free energy – zeta function connection. T HEOREM 6.1. For Re s > 1 .s/ D

1 .s/

Z

1 s 1 t dt

0

et

1

:

(6.2)

We can regard the integral on the right as the sum Z 1 s 1 Z 1 s 1 t dt t dt C ; t 1 et 1 1 0 e where the second integral converges absolutely s 2 C , and the first inteR 1 s 1 for all t gral, understood as the limit lim˛!0C ˛ t dt =.e 1/, exists for Re s > 1.

The proof of Theorem 6.1 is developed ˛; ˇ; a 2 R and R ˇ in two stages. First,Rfor aˇ s 2 C , with ˛ < ˇ and a > 0, write ˛ e at t s 1 dt D a s a˛ e v v s 1 dv, by the change of variables v D at . In particular, Z ˇ Z aˇ at s 1 s e t dt D a e t t s 1 dt for ˇ > 1; (6.3) 1

Z

a

1

e ˛

at s 1

t

dt D a

s

Z

a

a˛

e tts

1

dt

for ˛ < 1.

(6.4)


For s 2 C ; m > 0, consider the integral Z 1 s 1 mt t e e def Im .s/ D t 1 e 0

t

dt D

Z

1 s 1 t e mt

et

0

1

dt

51

(6.5)

which we check does converge for Re s > 1. For t > 0, we have e t > 1 C t , so 1=.e t 1/ < 1=t , and hence ˇ s 1 mt ˇ ˇt ˇ t 1 e mt e ˇ ˇ D t 2 e mt (6.6) ˇ et 1 ˇ < t Rˇ R mˇ for D Re s. By (6.3), 1 t 2 e mt dt D m . 1/ m e t t 2 dt for ˇ > 1. R1 Let ˇ ! 1: then 1 t 2 e mt dt exists and Z 1 Z 1 1 t 2 e mt dt D 1 e t t 2 dt: m 1 m R 1 t s 1 e mt dt converges absolutely for every In view of (6.6), therefore, 1 et 1 s 2 C and m > 0, and ˇ Z 1 s 1 mt ˇ Z 1 ˇ s 1 mt ˇ Z 1 ˇ ˇ ˇ ˇt t e 1 e ˇ ˇ ˇ ˇ dt ˇ e t t 2 dt: (6.7) ˇ ˇ e t 1 ˇ dt m 1 et 1 1 1 m By the change of variables v D 1=t for t > 0, s 1 Z 1=˛ 1 Z 1 s 1 mt e m.1=v/ t e v dv dt D et 1 .e 1=v 1/v 2 1 ˛

(6.8)

for 0 < ˛ < 1. Here, by the inequality in (6.6), we can write ˇ ˇ ˇ s 1 e m.1=v/ ˇ 2 e m=v ˇ ˇ 1 1 D v e m=v v ˇ< ˇ 1=v 2 v ˇ v v2 .e 1/v ˇ

But

Z

1=˛

v

1

dv D

.1=˛/1 1 Z

1

1

1

, so lim

Z

˛!0C 1

1=˛

v

dv D

1

1

:

(6.9)

for > 1, i.e.,

s 1 1 e m.1=v/ v dv .e 1=v 1/v 2

converges absolutely for Re s > 1 and s 1 ˇ 1 s 1 m.1=v/ ˇ ˇZ ˇ Z 1 ˇ 1 1 ˇ e m.1=v/ ˇ Z 1 ˇ e ˇ ˇ ˇ ˇ v v dv v dv ˇ ˇ ˇ ˇ 1=v ˇ 1 ˇ .e 1=v 1/v 2 1/v 2 ˇ 1 ˇ .e 1

e

m=v

dv;

52

FLOYD L. WILLIAMS

Rˇ by the inequality in (6.9). The right-hand side equals limˇ!1 1 v e m=v dv, or, by the change of variables t D 1=v, Z 1 Z m lim t 2 e mt dt D lim m . 1/ e t t 2 dt ˇ!1 1=ˇ ˇ!1 m=ˇ Z m 1 e t t 2 dt: D 1 m 0 Z 1 s 1 mt t e That is, by (6.8), lim dt exists for Re s > 1, and, with D Re s, t e 1 ˛!0C ˛ ˇ Z m Z 1 s 1 mt ˇ ˇ ˇ t e 1 ˇ lim ˇ e t t 2 dt: (6.10) dt ˇ 1 ˇ ˛!0C t e 1 m 0 ˛

We have therefore checked that theR integral Im .s/ defined by (6.5) converges 1 for Re s > 1 (and in fact the portion 1 t s 1 e mt =.e t 1/ dt converges absolutely for all s 2 C ), and that, moreover, for D Re s, we have ˇ Z 1 s 1 mt ˇ Z m ˇ ˇ t e 1 ˇ ˇ e t t 2 dt; dt ˇ 1 ˇ et 1 m 0 0 ˇ Z 1 s 1 mt ˇ Z 1 ˇ ˇ t e ˇ ˇ 1 dt e t t 2 dt; ˇ ˇ m 1 t e 1 1 m by the inequalities (6.7) and (6.10). This says that ˇ ˇ Z m Z 1 ˇ ˇ t 2 t 2 ˇIm .s/ˇ 1 e t dt C e t dt ˇ m 1 ˇ 0 m Z 1 1 1 D 1 e t t 2 dt D 1 . 1/; m m 0 by definition (1.6). Since m

1

! 0 as m ! 1 for > 1, we see that lim Im .s/ D 0

m!1

(6.11)

for Re s > 1 ! We move now to the second stage of the proof of Theorem 6.1, which is quite brief. Fix integers 1 n m and Re s > 1, and set v D nt . R 1n; mntwith R 1 s 2vC swith s 1 s 1 Again by (1.6), 0 e t dt D n v dv D .s/=ns , so 0 e X Z 1 Z 1 m m t mt / X 1 s 1 nt s 1 e .1 e .s/ D dt t dt D e t ns 1 e t 0 0 nD1 nD1 Z 1 s 1 t t e dt C Im .s/: D t 1 e 0


53

(Here the the last equality comes from (6.5) and the last but one from the formula for the partial sum of the geometric series.) Therefore Z 1 s 1 m X 1 t D dt C Im .s/: (6.12) .s/ s n et 1 0 nD1

We check the existence of the integral on the right-hand side for Re s > 1: write ts 1 D ts et 1

1

e

t

C

t s 1e t et 1

R1 for t > 0; the integral 0 t s 1 e t dt D .s/ converges for Re s > 0 and the integral of the last term, which equals I1 .s/ by (6.5), converges for Re s > 1, as established in the first stage of the proof. Z 1 s 1 t dt by (6.11) for Let m ! 1 in equation (6.12): then .s/.s/ D t e 1 0 Re s > 1, concluding the proof of Theorem 6.1. C OROLLARY 6.13. For a > 0, Re s > 1, Z 1 s 1 t dt D at e 1 0

.s/.s/ : as

In particular, for a > 0, n D 1; 2; 3; 4; : : : Z 1 2n 1 . 1/nC1 .2=a/2n B2n t dt D : e at 1 4n 0

(6.14)

(6.15)

P ROOF. This follows from (6.2) once the obvious change of variables v D at is executed. By formula (6.14), Z 1 2n 1 .2n/.2n/ t dt D e at 1 a2n 0

for n 2 Z , n 1. Since .2n/ D .2n 1/! (because .m/ D .m 1/! for m 2 Z ; m 1), one can now appeal to formula (2.1) to conclude the proof of equation (6.15). ˜ As an application of formula (6.15) we shall compute Planck’s radiation density integral. But first we provide some background. On 14 December 1900, a paper written by Max Karl Ernst Ludwig Planck and entitled “On the theory of the energy distribution law of the normal spectrum” was presented to the German Physical Society. That date is considered to be the birthday of quantum mechanics, as that paper set forth for the first time the hypothesis that the energy of emitted radiation is quantized. Namely, the energy cannot assume arbitrary values but only integral multiples 0; h; 2h; 3h; : : : of the basic energy value E D h, where is the frequency of the radiation and h is what is now called Planck’s constant. We borrow a quotation from Hermann

54

FLOYD L. WILLIAMS

Weyl’s notable book [37]: “The magic formula E D h from which the whole of quantum theory is developed, establishes a universal relation between the frequency of an oscillatory process and the energy E associated with such a process”. The quantization of energy has profound consequences regarding the structure of matter. Planck was led to his startling hypothesis while searching for a theoretical justification for his newly proposed formula for the energy density of thermal (or “blackbody”) radiation. Lord Rayleigh had proposed earlier that year a theoretical explanation for the experimental observation that the rate of energy emission f .I T / by a body at temperature T in the form of electromagnetic radiation of frequency grows, under certain conditions, with the square of , and the total energy emitted grows with the fourth power of T . In the quantitative form derived by James Jeans a few years later, Rayleigh’s formula reads f .I T / D

8 2 kT; c3

(6.16)

where c is the speed of light and k is Boltzmann’s constant. As grows, however, this formula was known to fail. Wilhelm Wien had already proposed, in 1896, the empirically more accurate formula f .I T / D a 3 e

b=T

:

(6.17)

Unlike the Rayleigh–Jeans formula (6.16), Wien’s avoids the “ultraviolet catastrophe”. (This colorful name was coined later by Paul Ehrenfest for the notion that aRfunctional form for f .I T / might yield an infinite value for the 1 total energy, 0 f .I T /d D 1 — “ultraviolet” because the divergence sets in at high frequencies.) However, the lack of a theoretical explanation for Wien’s law, and its wrong prediction for the asymptotic limit at low frequencies — proportional to 3 rather than 2 — made it unsatisfactory as well. By October 1900 Planck had come up with a formula that had the right asymptotic behavior in both directions and was soon found to be very accurate: f .I T / D

8 2 h 3 h=kT c e

1

;

(6.18)

where the new constant h was introduced. In his December paper, already mentioned, he provides a justification for this formula using the earlier notions of electromagnetic oscillators and statistical-mechanical entropy, but invoking the additional assumption that the energy of the oscillators is restricted to multiples of E D h. This then is the genesis of quantization. Note that given the approximation e x ' 1Cx for a very small value of x, one has the low frequency approximation e h=kT 1 ' h=kT , which when used


55

in formula (6.18) gives f .I T / ' 8 2 kT =c 3 — the Rayleigh–Jeans result (6.16), as expected by our previous remarks. We now check that in contrast to an infinite total energy value implied by formula (6.16), integration over the full frequency spectrum via Planck’s law (6.18) does yield a finite value. The result is: P ROPOSITION 6.19. Planck’s radiation density integral Z Z 1 8h 1 3 d def I.T / D f .I T /d D 3 h=kT c 1 0 e 0 (see (6.18)) has the finite value 8 5 k 4 T 4 =15c 3 h3 . The proof is quite immediate. In formula (6.15) choose n D 2, a D h=kT ; then 8h 2kT 4 B4 I.T / D 3 . 1/3 : h 8 c

Since B4 D 1=30 by (2.2), the desired value of I.T / is achieved. The radiation energy density f .I T / in (6.18) is related to the thermodynamics of the quantized harmonic oscillator; namely, it is related to the thermodynamic internal energy U.T /. We mention this because U.T /, in turn, is related to the Helmholtz free energy F.T / which has, in fact, a zeta function connection. A quick sketch of this mix of ideas is as follows, where proofs and details can be found in my book on quantum mechanics [42] (along with some historic remarks). One of the most basic, elementary facts of quantum mechanics is that the quantized harmonic oscillator of frequency has the sequence ˚ def 1 En D n C 21 h nD0

as its energy levels. The corresponding partition function Z.T / is given by def

Z.T / D

1 X

exp

nD0

En ; kT

(6.20)

where again T denotes temperature and k denotes Boltzmann’s constant. This sum is easily computed: Z.T / D i.e.,

1 X

nD0

e

h=2kT

e

h=kT n

Z.T / D

De

1 2 sinh

h 2kT

1

h=2kT

1

:

e

h=kT

I

(6.21)

56

FLOYD L. WILLIAMS

The importance of the partition function Z.T / is that from it one derives basic thermodynamic quantities such as the Hemholtz free energy

def

F.T / D kT log Z.T /; def

the entropy the internal energy

S.T / D @F=@T ; def

U.T / D F.T / C T S.T / D F.T /

Using (6.21), one computes that by these definitions h h ; C kT log 1 exp F.T / D 2 kT h=kT h S.T / D k ; log 1 exp kT exp.h=kT / 1 U.T / D

T

@F : @T

(6.22)

h h C ; 2 exp.h=kT / 1

which means that the factor h=.e h=kT 1/ of f .I T / in equation (6.18) differs from the internal energy U.T / exactly by the quantity E0 D 12 h, which is the ground state energy (also called the zero-point energy) of the quantized 1 harmonic oscillator since En D .n C 2 /h . However, our main interest is in setting up a free energy – zeta function connection. Here’s how it goes. def For convenience let ˇ D 1=.kT / denote the inverse temperature. Form the finite temperature zeta function X 1 def ; (6.23) .sI T / D 2 n2 C h2 2 ˇ 2 /s .4 n2Z which turns out to be well-defined and holomorphic for Re s > 21 . In [42] we show that .sI T / has a meromorphic continuation to Re s < 1 given by Z p 1 2a sin s 1 .x 2 1/ s .s 21 / p dx (6.24) C 2. a/ .sI T / D p 4 .s/as 1=2 1 exp. ax/ 1 def

for a D h2 2 ˇ 2 . Moreover .sI T / is holomorphic at s D 0 and p p 0 .0I T / D a 2 log 1 exp. a/ I

(6.25)

p see Theorem 14.4 and Corollary 14.2 of [42]. By definition, a D hˇ D h=kT . Therefore by formula (6.25) (which does require some work to derive from (6.24)), and by the first formula in (6.22), one discovers that kT 0 .0I T /; 2 which is the free energy – zeta function connection. F.T / D

(6.26)


57

We will meet the zero-point energy E0 D 21 h again in the next lecture regarding the discussion of Casimir energy. In chapter 16 of [42] another finite temperature zeta function is set up in the context of Kaluza–Klein space-times with spatial sector R m nG=K, where is a discrete group of isometries of the rank 1 symmetric space G=K; here K is a maximal compact subgroup of the semisimple Lie group G. In this broad context a partition function Z.T / and free energy-zeta function connection still exist.

Lecture 7. Zeta regularization, spectral zeta functions, Eisenstein series, and Casimir energy Zeta regularization is a powerful, elegant procedure that allows one to assign to a manifestly infinite quantitiy a finite value by providing it a special value zeta interpretation. Such a procedure is therefore of enormous importance in physics, for example, where infinities are prolific. P As a simple example, we consider the sum S D 1 C 2 C 3 C 4 C D 1 nD1 n, which is obviously infinite. This sum arises naturally in string theory — in the discussion of transverse Virasoro operators, for example. A string (which replaces the notion of a particle in quantum theory, at the Plancktian scale 10 33 cm) sweeps out a surface called a world-sheet as it moves in d-dimensional space-time R d D R 1 R d 1 - in contrast to a world-line of a point-particle. For Bosonic string theory (where there are no fermions, but only bosons) certain Virasoro constraints force d to assume a specific value. Namely, the condition d 2 S (7.1) 1D 2

arises, which as we shall see forces the critical dimension d D 26,P 1 being the s value of a certion normal ordering constant. In fact, we write S D 1 nD1 1=n , where s D 1, which means that it is natural to reinterpret S as the special zeta value . 1/. Thus we zeta regularize the infinite quantity S by assigning to it 1 , according to (2.14). Then by condition (7.1), indeed we must the value 12 have d D 26. Interestingly enough, the “strange” equation 1C 2 C 3 C 4 C D

1 12 ;

(7.2)

which we now understand to be perfectly meaningful, appears in a paper of Ramanujan — though he had no knowledge of the zeta function. It was initially dismissed, of course, as ridiculous and meaningless. As another simple example we consider “1! ”, that is, the product P D 1 2 3 4P , which is also infinite. To zeta regularize P , we consider first 0 log P D 1 nD1 log n (which is still infinite), and we note that since .s/ D

58

FLOYD L. WILLIAMS

P1

s

for s > 1, by equation (1.1), if we illegally take s D 0 the PRe 1 D nD1 log n D log P follows. However the left-hand side here is well-defined and in fact it has the value 21 log 2 by equation (2.6). The finite value 12 log 2, therefore, is naturally assigned to log P and, consequently, we define 1 p Q 1 0 def (7.3) PD n D 1! D e .0/ D e 2 log 2 D 2: nD1 .log n/n false result 0 .0/

nD1

More complicated products can be regularized in a somewhat similar manner. A typical set-up for this is as follows. One has a compact smooth manifold M with a Riemannian metric g, and therefore a corresponding Laplace–Beltrami operator D .g/ where has a discrete spectrum 0 D 0 < 1 < 2 < 3 < ;

lim j D 1:

j !1

(7.4)

If nj denotes the (finite) multiplicity of the j-th eigenvalue j of , then one can form the corresponding spectral zeta function (cf. definition (0.1)) 1 X nj M .s/ D ; js

(7.5)

j D1

which is well-defined for Re s > 21 dim M , due to the discovery by H. Weyl of the asymptotic result j j 2=dim M , as j ! 1. S. Minakshisundaram and A. Pleijel [26] showed that M .s/ admits a meromorphic continuation to the complex plane and that, in particular, M .s/ is holomorphic at s D 0. Thus e M .0/ is well-defined and, as in definition (7.3), we set 0

det D

1 Y

j D1

n def

j j D e

0 .0/ M

;

(7.6)

where the prime 0 here indicates that the product of eigenvalues (which in finite dimensions corresponds to the determinant of an operator) is taken over the nonzero Q1ones. Indeed, similar to the preceding example with the infinite product P D j D1 j , the formal, illegal computation ˇ ˇ X X 1 1 1 ˇ nj d X nj ˇˇ ˇ nj log j log j ˇ D exp D exp exp ds js ˇsD0 js sD0 j D1

D

1 Y

j D1

j D1

e nj log j D

1 Y

j D1

n

j j

(7.7)

j D1

serves as the motivation for definition (7.6). Clearly this definition of determinant makes sense for more general operators (with a discrete spectrum) on other


59

infinite-dimensional spaces. It is useful, moreover, for Laplace-type operators on smooth sections of a vector bundle over M . The following example is more involved, where M is a complex torus (in fact M is assumed to be the world-sheet of a bosonic string; see Appendix C of [42], for example). For a fixed complex number D 1 C i2 in the upper half-plane (2 > 0) and for the corresponding integral lattice, def

def

L D fa C b j a; b 2 Z g;

M D C nL :

(7.8)

In this case it is known that has a multiplicity free spectrum (i.e., every nj D 1) given by o n def 4 2 ; mn D 2 jm C nj2 m;n2Z 2 and consequently the corresponding spectral zeta function of (7.5) is given by M .s/ D

2s .4 2 /s

E .s; /

(7.9)

for Re s > 1, where def

E .s; / D

2s

X

.m;n/2Z 2

(7.10)

jm C nj2s

(with Z 2 D Z Z f.0; 0/g as before) is a standard nonholomorphic Eisenstein series. That is, in contrast to the series Gk ./ in definition (4.4), E .s; / is not a holomorphic function of . As a function of s, it is a standard fact that E .s; /, which is holomorphic for Re s > 1, admits a meromorphic continuation to the full complex plane, with a simple pole at s D 1 as its only singularity. Hence, by equation (7.9), the same assertion holds for M .s/. By [7; 14; 35; 38], for example, the continuation of E .s; / is given by p .s 1/

E .s; / D 2.2s/2s C 2.2s C

1 1 4 s 1=2 X X 2 e .s/

C

2 imn1

1 2/

.s/ n s 1 2

m

mD1 nD1 1 X 1 X 1=2 2 e 2 imn1 .s/ mD1 nD1

4 s

2 sC1 K

n s m

1 .2 mn2 / s 2 1 2

1 .2 mn2 / s 2

K

(7.11)

for Re s > 1, where def

K .x/ D

1 2

Z

1 0

exp

x 1 tC t 2 t

1

dt

(7.12)

60

FLOYD L. WILLIAMS

is the Macdonald–Bessel (or K-Bessel) function for 2 C , x > 0. Introducdef P ing the divisor function .n/ D 0 d=2, m > 0, aE D .a1 ; : : : ; ad /, bE D .b1 ; : : : ; bd / 2 R d , ai > 0. The result is, setting Z d D Z d f0g, d=2 .s d2 /md 2s 2 s md=2 s E D p Cp E.s; mI aE; b/ a1 a2 ad .s/ a1 a2 ad .s/ P P P d n2 s d=2 d n2 1 X j j 2 2 2 i jdD1 nj bj e (7.25) K d s 2 m 2 j D1 aj j D1 aj d n E2Z

64

FLOYD L. WILLIAMS

E E; b/, for Re s > d=2. In particular, we shall need the special value E. dC1 2 ; mI a dC1 which we now compute. For s D 2 the first term on the right in (7.25) is Qd

j D1 aj

m 1 2

since

D 1=2 : Also

Kd

2

p

dC1 2

s

.x/ D K

dC1 2

1 .x/ 2

;

D K 1 .x/; 2

x

for x > 0; hence P P d n2 s d=2 d n2 1 j j 2 2 K d s 2 m 2 j D1 aj j D1 aj r P P d n2 1 d n2 1=4 d n2 1 1 P j 2 j j 4 exp 2 m : .2 m/ 2 D 2 j D1 aj j D1 aj j D1 aj by (7.16). This equals

=2x e

1=2

Therefore the second term on the right-hand side of (7.25) is 2

dC1 2 m 1=2

Qd

j D1 aj

That is:

1=2

dC1 2

X

n E 2Z d

P d d n2 1 P 1 j 2 p exp 2 i nj bj exp 2 m : 2 m j D1 j D1 aj

dC1 2 d C1 ; mI aE; bE D E Q 1=2 2 m jdD1 aj

C

m

D

Qd

m

dC1 2

j D1 aj

Qd

1=2

dC1 2

j D1 aj

1=2

dC1 2

dC1 2

X

e

dC1 2

2 i n E bE

exp

n E 2Z d

X

n E 2Z

d

e

2 i n E bE

exp

P d n2 1 j 2 2 m j D1 aj P d n2 1 j 2 : (7.26) 2 m j D1 aj

As a final example we consider the zeta regularization of Casimir energy, after a few general remarks. ˚ def In Lecture 6 it was observed that the sequence En D n C 21 h is the sequence of energy levels of the quantized harmonic oscillator of frequency , where h denotes Planck’s constant. In particular there exists a nonvanishing ground state energy (also called the zero-point energy) given by E0 D 12 h. Zero-point energy is a prevalent notion in physics, from quantum field theory (QFT), where it is also referred to as vacuum energy, to cosmology (concerning


65

issues regarding the cosmological constant, for example), and in between. Based on Planck’s radiation density formula (6.18), A. Einstein and O. Stern concluded (in 1913) that even at zero absolute temperature, atomic systems maintain an energy of the amount E0 D 21 h. It is quite well experimentally established that a vacuum (empty space) contains a large supply of residual energy (zeropoint energy). Vacuum fluctuations is a large scale study. The energy due to vacuum distortion (Casimir energy), for example, was considered by H. Casimir and D. Polder in a 1948 ground-breaking study. Here the vacuum energy was modified by the introduction of a pair of uncharged, parallel, conducting metal plates. A striking prediction emerged: the prediction of the existence of a force of a purely quantum mechanical origin — one arising from zero-point energy changes of harmonic oscillators that make up the normal modes of the electromagnetic field. This force, which has now been measured experimentally by M. Spaarnay, S. Lamoreaux, and others, is called the Casimir force. Casimir energy in various contexts has been computed by many Physicists, including some notable calculations by the co-editor Klaus Kirsten. We refer to his book [22] for much more information on this, and on related matters - a book with 424 references. The author has used the Selberg trace formula for general compact space forms nG=K, mentioned in Lecture 6, of rank-one symmetric spaces G=K to compute the Casimir energy in terms of the Selberg zeta function [40; 39]. This was done by Kirsten and others in some special cases. Consider again a compact smooth Riemannian manifold .M; g/ with discrete spectrum of its Laplacian .g/ given by (7.4). In practice, M is the spatial sector of a space-time manifold R M with metric dt 2 C g. Formally, the Casimir energy in this context is given by the infinite quantity 1 1X 1=2 nj j ; EC D 2

(7.27)

j D1

up to some omitted factors like h. It is quite clear then how to regularize EC . 1=2 Namely, consider j as 1=js for s D 12 and therefore assign to EC the meaning EC D 21 M . 1=2/;

(7.28)

where M .s/ is the spectral zeta function of definition (7.5), meromorphically continued. If dim M is even, for example, the poles of M .s/ are simple, finite in number, and can occur only at one of points s D 1; 2; 3; : : : ; d=2 (see [26]), in which case EC in (7.28) is surely a well-defined, finite quantity. However if dim M is odd, M .s/ will generally have infinitely many simple poles — at the points s D 21 dim M n, for 0 n 2 Z . This would include the point s D 12 if dim M D 5 and n D 3, for example. Assume therefore that dim M is even.

66

FLOYD L. WILLIAMS

When M is one of the above compact space forms, for example, then (based on the results in [41]) EC can be expressed explicitly in terms of the structure of and the spherical harmonic analysis of G=K — and in terms of the Selberg zeta function attached to nG=K, as just mentioned. Details of this are a bit too technical to mention here; we have already listed some references. We point out only that by our assumptions on M , the corresponding Lie group pairs .G; K/ are given by G D SO1 .m; 1/;

K D SO.m/;

G D SU.m; 1/;

K D U.m/;

G D F4.

K D Spin.9/;

G D SP.m; 1/; 20/ ;

m 2;

K D SP.m/ SP.1/;

m 2;

m 2;

where F4. 20/ is a real form of the complex Lie group with exceptional Lie algebra F4 with Dynkin diagram 0—0 H 0—0. More specifically, F4. 20/ is the unique real form for which the difference dim G=K dim K assumes the value 20. In addition to the reference [22], the books [13; 12] are a good source for information on and examples of Casimir energy, and for applications in general of zeta regularization.

Lecture 8. Epstein zeta meets gravity in extra dimensions We compute the Kaluza–Klein modes of the 4-dimensional gravitational potential V4Cd in the presence of d extra dimensions compactified on a d-torus. The result is known of course [3; 21], but we present here an argument based E; bE computed in equation (7.26) of the generon the special value E dC1 2 ; mI a E defined in (7.24). alized Epstein zeta function E.s; mI aE; b/ G. Nordström in 1914 and T. Kaluza (independently) in 1921 were the first to unify Einstein’s 4-dimensional theory of gravity with Maxwell’s theory of electromagnetism. They showed that 5-dimensional general relativity contained both theories, but under an assumption that was somewhat artificial - the socalled “cylinder condition” that in essence restricted physicality of the fifth dimension. O. Klein’s idea was to compactify that dimension and thus to render a plausible physical basis for the cylinder assumption. Consider, for example, the fifth dimension (the “extra dimension”) compactified on a circle . This means that instead of considering the Einstein gravitational field equations Rij .g/

gij R.g/ 2

gij D

8G Tij c4

(8.1)


67

on a 4-dimensional space-time M 4 [8; 11], one considers these equations on the 5-dimensional product M 4 . In (8.1), g D Œgij is a Riemannian metric (the solution of the Einstein equations) with Ricci tensor Rij .g/ and scalar curvature R.g/; is a cosmological constant, and Tij is an energy momentum tensor which describes the matter content of space-time — the left-hand side of (8.1) being pure geometry; G is the Newton constant and c is the speed of light. Given the non-observability of the fifth dimension, however, one takes to be extremely small, say with an extremely small radius R > 0. Geometrically we have a fiber bundle M 4 ! M 4 with structure group .

M4 M4 M4

On all “fields” F.x; / W M 4 R ! C on M 4 R there is imposed, moreover, periodicity in the second variable: F.x; C 2R/ D F.x; /

(8.2)

for .x; / 2 M 4 R . def For n 2 Z and f .x/ on M 4 fixed, the function Fn;f .x; / D f .x/e in=R is an example of a field on M 4 R that satisfies equation (8.2). For a general field F.x; /, subject to reasonable conditions, and the periodicity condition (8.2), one would have a Fourier series expansion F.x; / D

X

n2Z

def

Fn;fn D

X

fn .x/e in=R

(8.3)

n2Z

in which case the functions fn .x/ are called Kaluza–Klein modes of F.x; /. Next we consider d extra dimensions compactified on a d-torus d def

D

1

d;

where the j are circles with extremely small radii Rj > 0, and we consider a field V4Cd .x; y; z; x1 ; : : : ; xd / on .R 3 f0g/ R d . Thus R 3 f0g replaces M 4 ,

68

FLOYD L. WILLIAMS def

replaces , and xE D .x1 ; : : : ; xd / 2 R d replaces in the previous discussion. The field is given by d

def

V4Cd .x; y; z; x1 ; : : : ; xd / D X MG4Cd

1

n ED.n1 ;:::;nd /2Z

d

r2 C

d P

2 nj Rj /2

.xj

j D1

dC1

;

(8.4)

2

which is the gravitational potential due to extra dimensions of an object of mass 1=2 P def M at a distance r 2 C jdD1 xj2 for r 2 D x 2 C y 2 C z 2 ; here G4Cd is the .4Cd/-dimensional Newton constant. Note that, analogously to equation (8.2),

V4Cd .x; y; z; x1 C 2R1 ; : : : ; xd C 2Rd / X def D MG4Cd d P .n1 ;:::;nd /2Z d r 2 C .xj D MG4Cd

X

j D1

.n1 ;:::;nd /2Z d

r2 C

D V4Cd .x; y; z; x1 ; : : : ; xd /;

d P

.xj

1 2.n 1/jRj /2 1 2 nj Rj /2

j D1

dC1 2

(8.5)

dC1 2

where of course we have used that nj 1 varies over Z as nj does. Thus, analogously to equation (8.3), we look for a Fourier series expansion ! X x1 xd fnE .x; y; z/ exp i nE V4Cd .x; y; z; x1 ; : : : ; xd / D ;:::; ; (8.6) R1 Rd n E 2Z d

where the functions fnE .x; y; z/ on R 3 f0g would be called the Kaluza–Klein modes of V4Cd .x; y; z; xE D .x1 ; : : : ; xd //. It is easy, in fact, to establish the expansion (8.6) and to compute the modes fnE .x; y; z/ explicitly. For this, define def def xj aj D .2Rj /2 > 0; bj D ; 2Rj 2 x j nj D aj .bj nj /2 and note that since .xj 2 nj Rj /2 D 2Rj 2Rj we can write, by definition (8.4), X 1 V4Cd .x; y; z; x/ E D MG4Cd dC1 P d n E 2Z d aj .nj bj /2 C r 2 2 D MG4Cd E

j D1

d C1 ; r I aE; bE ; 2

(8.7)


69

by definition (7.24). Thus we are in a pleasant position to apply formula (7.26): For dC1 d Y def 2 2 def d Rj ; ˝d D ˙d D .2/ (8.8) d C1 j D1 2 1=2 Qd D ˙d , and we see that we have j D1 aj V4Cd .x; y; z; x/ E D MG4Cd ˝d X 2r˙d

n E 2Z d

d 2 1 ! P nj 2 xd x1 ;:::; exp r exp i nE 2 R1 Rd j D1 Rj

(8.9)

def

by definition of aj and bj , for r 2 D x 2 Cy 2 Cz 2 . This proves the Fourier series expansion (8.6), where we see that the Kaluza–Klein modes fnE .x; y; z/ are in fact given by d 2 1 P nj 2 MG4Cd ˝d exp r (8.10) fnE .x; y; z/ D 2 2r˙d j D1 Rj

for nE D .n1 ; : : : ; nd / 2 Z d ; .x; y; z/ 2 R 3 f0g. Since V4Cd .x; y; z; x/ E is actually real-valued, we write equation (8.9) as V4Cd .x; y; z; x1 ; : : : ; xd / D d 2 1 ! P nj 2 MG4Cd ˝d X x1 xd exp r cos nE : (8.11) ;:::; 2 R1 Rd 2r˙d j D1 Rj d n E 2Z

Q def Q Since 2Ri is the length of i ; d D diD1 i has volume diD1 2Ri D Q .2/d diD1 Ri . That is, ˙d in definition (8.8) (or in formula (8.11)) is the volume of the compactifying d-torus d . Similarly ˝d in (8.8) or in (8.11), ˚ one knows, is the surface area of the unit sphere x 2 R dC1 j kxk D 1 in R dC1 . In [21], for example, the choice x1 D D xd D 0 is made. Going back to the compactification on a circle, d D 1; R1 D R for example, we can write the sum in (8.11) as 1C

X

e

r jnj=R

n2Z f0g

D 1C 2

1 X

e

nD1

2e

r=R

r=R n

' 1 C 2e r=R (8.12) 1 e r=R for x1 D 0, where we keep in mind that R is extremely small. Thus in (8.12), r=R is extremely large; i.e., e r=R is extremely small. For D 1C

def

Kd D MG4Cd ˝d =2˙d ;

70

FLOYD L. WILLIAMS

we get by (8.11) and (8.12) V5 .x; y; z; x1 / '

K1 .1 C 2e r

r=R

/;

(8.13)

which is a correction to the Newtonian potential V D K1 =r due to an extra dimension. The approximation (8.13) compares with the general deviations from the Newtonian inverse square law that are known to assume the form V D

K .1 C ˛e r

r=

/

for suitable parameters ˛; . Apart from the toroidal compactification that we have considered, other compactifications are important as well [21] — especially Calabi–Yau compactifications. Thus the d-torus d is replaced by a Calabi– Yau manifold — a compact Kähler manifold whose first Chern class is zero.

Lecture 9. Modular forms of nonpositive weight, the entropy of a zero weight form, and an abstract Cardy formula A famous formula of John Cardy [9] computes the asymptotic density of states .L0 / (the number of states at level L0 ) for a general two-dimensional conformal field theory (CFT): For the holomorphic sector .L0 / D e 2

p

cL0 =6

;

(9.1)

where the Hilbert space of the theory carries a representation of the Virasoro algebra V i r with generators fLn gn2Z and central charge c. V i r has Lie algebra structure given by the usual commutation rule c ŒLn ; Lm D .n m/LnCm C n.n2 1/ınCm;0 (9.2) 12 for n; m 2 Z . The CFT entropy S is given by r cL0 : (9.3) S D log .H0 / D 2 6 From the Cardy formula one can derive, for example, the Bekenstein–Hawking formula for BTZ black hole entropy [10]; see also my Speaker’s Lecture presented later. More generally, the entropy of black holes in string theory can be derived — the derivation being statistical in nature, and microscopically based [34]. For a CFT on the two-torus with partition function Z./ D t r ace e 2 i.L0

c 24 /

(9.4)


71

on the upper half-plane C [5], the entropy S can be obtained as follows. Regarding Z./ as a modular form with Fourier expansion X Z./ D cn e 2 i.n c=24/ ; (9.5) n0

one takes S D log cn

(9.6)

for large n. In [5], for example, (also see [4]) the Rademacher–Zuckerman exact formula for cn is applied, where Z./ is assumed to be modular of weight w D 0. This is problematic however since the proof of that exact formula works only for modular forms of negative weight. In this lecture we indicate how to resolve this contradiction (thanks to some nice work of N. Brisebarre and G. Philibert), and we present what we call an abstract Cardy formula (with logarithmic correction) for holomorphic modular forms of zero weight. In particular we formulate, abstractly, the sub-leading corrections to Bekenstein–Hawking entropy that appear in formula (14) of [5]. The discussion in Lecture 4 was confined to holomorphic modular forms of non-negative integral weight. We consider now forms of negative weight w D r for r > 0, where r need not be an integer. The prototypic example def will be the function F0 .z/ D 1=.z/, where .z/ is the Dedekind eta function defined in (3.27), and where it will turn out that w D 21 . We will use, in fact, the basic properties of F0 .z/ to serve as motivation for the general definition of a form of negative weight. We begin with the partition function p.n/ on Z C . For n a positive integer, define p.n/ as the number of ways of writing n as an (orderless) sum of positive integers. For example, 3 is expressible as 3 D 1 C 2 D 1 C 1 C 1, so p.3/ D 3; 4 D 4 D 1 C 3 D 2 C 2 D 1 C 1 C 2 D 1 C 1 C 1 C 1, so p.4/ D 5; 5 D 5 D 2C3 D 1C4 D 1C1C3 D 1C2C2 D 1C1C1C2 D 1C1C1C1C1; def

so p.5/ D 7; similarly p.2/ D 2, p.1/ D 1. We set p.0/ D 1. Clearly p.n/ grows quite quickly with n. A precise asymptotic formula for p.n/ was found by G. Hardy and S. Ramanujan in 1918, and independently by J. Uspensky in 1920: p e 2n=3 p.n/ as n ! 1 (9.7) p 4n 3 (the notation means that the ratio between the two sides of the relation (9.7) tends to 1 as n ! 1). For example, it is known that p.1000/ D 24;061;467;864;032;622;473;692;149;727;991 ' 2:4061 1031 ;

(9.8)

72

FLOYD L. WILLIAMS

whereas for n D 1000 in (9.7) e

q

p 2n 3 =4n 3

' 2:4402 1031 ;

(9.9)

which shows that the asymptotic formula is quite good. L. Euler found the generating function for p.n/. Namely, he showed that 1 X

nD0

1 .1 nD1

p.n/z n D Q1

zn/

(9.10)

for z 2 C with jzj < 1. By this formula and the definition of .z/, we see immediately that 1 F0 .z/ D De .z/ def

i=12

1 X

p.n/e 2 in

(9.11)

nD0

on C . The following profound result is due to R. Dedekind. To prepare the ground, for x 2 R define Œx 21 if x 62 Z , def x ..x// D 0 if x 2 Z , where, as before, Œx denotes the largest integer not exceeding x. def T HEOREM 9.12. Fix D ac db 2 D SL.2; Z /, with c > 0, and define

aCd 1 s.d; c/; 12c 4 where s.d; c/ (called a Dedekind sum) is given by X d def s.d; c/ D : c c S. / D

(9.13)

2Z =c Z

Then, for z 2 C ,

1 1 (9.14) F0 . z/ D e i (S. /C 4 ) i.cz C d/ 2 F0 .z/ for =2 < arg i.cz C d/ < =2, where z is defined in equation (4.3).

The sum in definition (9.13) is over a complete set of coset representatives in Z . The case c D 0 is much less profound; then ˙1 b

D 0 ˙1 (since 1 D det D ad), and

F0 .z ˙ b/ D F0 . z/ D e ib=12 F0 .z/:

(9.15)


In particular we can write F0 .z C 1/ D e i=12 F0 .z/ D e 1 def 1 D 23 e 2 i.1 24 / F0 .z/ D e 2 i˛ F0 .z/ for ˛ D 1 24 24 .

2 i=24C2 i F

73 0 .z/ D

In summary, F0 .z/ D 1=.z/ satisfies the following conditions:

(i) F0 .z/ is holomorphic on C . (This follows from Lecture 3).) (ii) F0 .z C 1/ D e 2 i˛ F0 .z/ for some real ˛ 2 Œ0; 1/ (indeed, with ˛ D 23 24 ). r a b (iii) F0 . z/ D ".a; b; c; d/ i.cz C d/ F0 .z/ with for D c d 2 c > 0, for some r > 0, =2 < arg i.cz C d/ < =2, and for a function ". / D ".a; b; c; d/ on with j". /j D 1 (indeed, for r D 12 and ".a; b; c; d/ D exp i aCd s.d; c/ , by Theorem 9.12). 12c P 2 inz on C for some integer 1 (indeed, (iv) F0 .z/ D e 2 i˛z 1 nD an e for D 1, an D p.n C 1/ for n 1, and an D 0 for n 2, by Euler’s formula (9.11)). def

Note that by conditions (i) and (ii), the function f .z/ D e 2 i˛z F0 .z/ is holomorphic on C , and it satisfies f .z C 1/ DP f .z/. Thus, again by equation (4.1), f .z/ has a Fourier expansion f .z/ D n2Z an e 2 inz on C . That is, conditions (i) and (ii) imply that F0 .z/ has a Fourier expansion P an e 2 inz F0 .z/ D e 2 i˛z n2Z

C,

on and condition (iv) means that we require that a n D 0 for n > , for some positive integer . We abstract these properties of F0 .z/ and, in general, we define a modular form of negative weight r , for r > 0, with multiplier " W ! fz 2 C j jzj D 1g to be a function F.z/ on C that satisfies conditions (i), (ii), (iii), and (iv) for some ˛ and with 0 ˛ < 1, 2 Z , 1. Thus .z/ 1 is a modular form of 1 aCd 23 weight 2 and multiplier ".a; b; c; d/ D exp i 12c s.d; c/ , with ˛ D 24 , D 1, and with Fourier coefficients an D p.nC1/, as we note again. For modular forms of positive integral weight, there are no general formulas available that explicitly compute their Fourier coefficients — apart from Theorem 4.23 for holomorphic Eisenstein series. For forms of negative weight however, there is a remarkable, explicit (but complicated) formula for their Fourier coefficients, due to H. Rademacher and H. Zuckerman [31]; also see [29; 30]. Before stating this formula we consider some of its ingredients. First, we have the modified Bessel function 1 t 2m X def t 2 (9.16) I .t / D 2 m! . C m C 1/ mD0

for t > 0; 2 C ; the series here converges absolutely by the ratio test. Next, for k; h 2 Z with k 1; h 0; .h; k/ D 1, and h < k choose a solution h0 of the

74

FLOYD L. WILLIAMS

congruence hh0 1.mod k/. For example, .h; k/ D 1 means that the equation xh C yk D 1 has a solution .x; y/ 2 Z Z . Then xh D 1 C yk means def that h0 D x is a solution. Since hh0 D 1 C lk for some l 2 Z , we see that .hh0 C 1/=k D l is an integer and 0 0 .hh0 C1/=k h .hh0 C1/=k def h 2 : D 1; so D det k h k h Hence

hh0 C1 ". / D " h0 ; ; k; h (9.17) k is well-defined. Finally, for u; v 2 C we define the generalized Kloosterman sum X 2 i ..u ˛/h0 C.vC˛/h/ def (9.18) ". / 1 e k Ak;u .v/ D Ak .v; u/ D 0h 0, with multiplier ", and with Fourier 1 2 i˛z 2 inz on C given by condition (iv) above, expansion F.z/ De nD an e where 0 ˛ < 1 2 Z . Then for n 0 with not both n; ˛ D 0, an D 1 X X Ak;j .n/ j ˛ r C1 4 2 a j 2 .j ˛/1=2 .n C ˛/1=2 ; (9.21) Ir C1 nC˛ k k j D1

kD1

where Ak;j .n/ is defined in (9.18) and I .t / is the modified Bessel function in (9.16). Note that for 1 j ; j 1 > ˛ ) j ˛ > 0 in equation (9.21). Also n C ˛ > 0 there since n; ˛ 0 with not both n; ˛ D 0. Using the asymptotic result p (9.22) lim 2 t I .t /e t D 1 t!1

for the modified Bessel function I .t / in (9.16), and also the trivial estimate jAk;j .n/j k that follows from (9.18), one can obtain from the explicit formula


(9.21) the following asymptotic behavior of an as n ! 1. Assume that a and define r

def

def

say for n 1, for "0 D " shown that

0 1

¤0

1

a . ˛/ 2 C 4 a .n/ D p "0 exp 4. 2 .n C ˛/ 2r C 34 1

75

1 0

˛/1=2 .n C ˛/1=2 ;

(9.23)

in (9.19). Then in [23], for example, it is

an a1 .n/

as n ! 1

(9.24)

which gives the asymptotic behavior of the Fourier coefficients of a modular form F.z/ of negative weight r with Fourier expansion as in the statement of Theorem 9.20. For forms of zero weight a quite similar result is given in equation (9.30) below. The asymptotic result (9.7) follows from (9.24) applied to F0 .z/, in which case formula (9.21) provides an exact formula (due to Rademacher) for p.n/ [2; 29; 30]. For a; b; k 2 Z with k 1 the classical Kloosterman sum S.a; bI k/ is defined by X 2 i N (9.25) e k .ahCb h/ S.a; bI k/ D h2Z =k Z

.h;k/D1

where hhN 1.mod k/. These sums will appear in the next theorem (Theorem 9.27) that is a companion result of Theorem 9.20. We consider next modular forms F.z/ of weight zero. That is, F.z/ is a holomorphic function on C such that F. z/ D F.z/ for 2 , and with Fourier expansion 1 X an e 2 inz (9.26) F.z/ D nD

C,

on for some positive integer . In case F.z/ is the modular invariant j .z/, for example, this expansion is that given in equation (4.44) with D 1, in which case the an there are computed explicitly by H. Petersson and H. Rademacher [27; 28], independently - by a formula similar in structure to that given in (9.21). For the general case in equation (9.26) the following extension of the Petersson– Rademacher formula is available [6]: T HEOREM 9.27 (N. B RISEBARRE AND G. P HILIBERT ). For a modular form F.z/ of weight zero with Fourier expansion given by equation (9.26), its n-th Fourier coefficient an is given by r 1 p X j X S.n; j I k/ 4 nj a j an D 2 I1 (9.28) n k k j D1

kD1

76

FLOYD L. WILLIAMS

for n 1, where S.n; j I k/ is defined in (9.25) and I1 .t > 0/ in (9.16). M. Knopp’s asymptotic argument in [23] also works for a weight zero form (as he shows), provided the trivial estimate jAk;j .n/j k used above is replaced by the less trivial Weil estimate jS.a; bI k/j C."/.a; b; k/1=2 k 1=2C" ; 8" > 0. The conclusion is that if a ¤ 0, and if 1=4 p def a a1 .n/ D p 3=4 e 4 n ; n 1; 2 n

(9.29)

then an a1 .n/ W lim

n!1

an D 1: a1 .n/

(9.30)

We see that, formally, definition (9.29) is obtained by taking "0 D 1; r D 0; and ˛ D 0 in definition (9.23) - in which case formulas (9.21) and (9.28) are also formally the same. Going back to the Fourier expansion of the modular invariant j .z/ given in equation (4.44), where a D a 1 D 1, we obtain from (9.30) that ([27; 28]) p

e 4 n an p as n ! 1: 2n3=4

(9.31)

A stronger result than (9.31), namely that p

e 4 n an D p 2n3=4

1

:055 3 p C "n ; j"n j n 32 n

(9.32)

(also due to Brisebarre and Philibert [6]) plays a key role in my study of the asymptotics of the Fourier coefficients of extremal partition functions of certain conformal field theories; see Theorem 5-16 of my Speaker’s Lecture (page 345), and the remark that follows it. Motivated by physical considerations, and by equation (9.5) in particular, we consider a modular form of weight zero with Fourier expansion f .z/ D e 2 iz

X

cn e 2 inz

(9.33)

n0

on C , where we assume that is a negative integer. corresponds to c=24 in (9.5), say for a positive central charge c; thus c D 24. /, a case considered def def in my Speaker’s Lecture. D is a positive integer such that for an D cnC , we have (taking cn D 0 for n 1) a n D 0 for n > . Moreover, since


P1

nD0 dn

D

P1

nD dn , 1 P

nD

77

def

we see that for dn D an e 2 inz we have

an e 2 inz D def

D

1 P

an

nD0 1 P

nD0

e

2 i.n /z

cn e 2 i.nC/z D f .z/;

(9.34)

by (9.33). That is, f .z/ has the form (9.26), which means that we can apply formula (9.28), and the asymptotic result (9.30). Assume that c0 ¤ 0, and define jj1=4 def c0 4jj1=2 .nC/1=2 c 1 .n/ D p e 2 .n C /3=4

for n C 1. By definition (9.29), for n a1 .n as a

/ D p

a

2.n

1=4

/3=4

(9.35)

def

D n C 1,

e 4

p

.n / def 1

D c .n/;

def

D c0 ¤ 0. Therefore by (9.30) an 1 n!1 a .n /

1 D lim

cn 1 n!1 c .n/

D lim

W cn c 1 .n/ as n ! 1;

(9.36)

for c 1 .n/ defined in (9.35). Thus (9.36) gives the asymptotic behavior of the Fourier coefficients cn of the modular form f .z/ of weight zero in (9.33). Motivated by equation (9.6), and given the result (9.36) we define entropy function S.n/ associated to f .z/ by def

S.n/ D log c 1 .n/ for n C 1, in case c0 > 0. Also we set p def S0 .n/ D 2 4jj.n C /;

(9.37)

(9.38)

for n C 1. Then for c D 24. / D 24jj, as considered above, (i.e., for 4jj D c=6) S0 .n/ corresponds to the CFT entropy in equation (9.3), where n C corresponds to the L0 there. Moreover, by definition (9.37) we obtain (9.39) S.n/ D S0 .n/ C 14 log jj 34 log.n C / 12 log 2 C log c0 ;

which we can regard as an abstract Cardy formula with logarithmic correction, given by the four terms parenthesized. Note that, apart from, the term log c0 , equation (9.39) bears an exact resemblance to equation (5.22) of my Speaker’s Lecture. We regard S0 .n/ in definition (9.38), of course, as an abstract Bekenstein–Hawking function associated to the modular form f .z/ in (9.33) of zero weight. Equation (9.39) also corresponds to equation (14) of [5].

78

FLOYD L. WILLIAMS

To close things out, we also apply formula (9.28) to f .z/. For n 1, r p X 4 .n /j j S.n ; j I k/ I1 : (9.40) a j cn D an D 2 n k k j D1

P 1 def Use j D1 dj D d C d 1 C : : : C d2 C d1 D j D0 d j and a . j / D cj to write equation (9.40) as p X1 r j S.n ; . j /I k/ 4 .n /. j / cj cn D 2 I1 (9.41) n k k P

j D0 def

where D . For 0 j 1 D 1; j 2 Z ; we have j 0 and j C 1 < 0. Conversely if j 0, j 2 Z , and j C < 0, then as 2 Z we have j C 1, so 0 j 1 D 1: Of course j C < 0 also means that j D j D jj C j. Thus we can write equation (9.41) as r p X 4 .nC/jj Cj jj Cj S.nC; j CI k/ I1 cn D 2 cj nC k k j 0 j C a. Thus f .s/ is well-defined on D. By definition, the integral f .s/ is uniformly convergent on some subset D0 D if to each a number B."/ > a such that for b > ˇ ˇR b " > 0 there corresponds B."/ one has ˇ a F.t; s/ dt f .s/ˇ < " for all s 2 D0 . An equivalent definition


79

is given by the following Cauchy criterion: f .s/ is uniformly convergent on ˇD R 0 if and onlyˇif to each " > 0 there corresponds a number B."/ > a such that ˇ b2 F.t; s/ dt ˇ < " for all b2 > b1 > B."/ and all s 2 D0 . For clearly if f .s/ is b1 uniformly ˇR b convergent on ˇD0 and " > 0 is given, we can choose B."/ > a such that ˇ a F.t; s/ dt f .s/ˇ < "=2 for b > B."/, s 2 D0 . Then for b2 > b1 > B."/ and s 2 D0 , we have Z b1 Z b2 Z b2 F.t; s/ dt f .s/ ; F.t; s/ dt f .s/ F.t; s/ dt D b1

a

a

ˇ ˇR b which implies ˇ b12 F.t; s/ dt ˇ < "=2C"=2 D ". Conversely, assume the alternate condition. Define the sequence ffn .s/gn>a of functions on D0 by Z n def fn .s/ D F.t; s/ dt: a

ˇ ˇR b Given " > 0 we can choose, by hypothesis, B."/ > a such that ˇ b12 F.t; s/ dt ˇ < " for b2 > b1 > B."/ and s 2 D0 . Let N ."/ be an integer ˇ> B."/. Thenˇ for n > ˇm N ."/ and for all s 2 D0 we see that ˇfn .s/ fm .s/ˇ D ˇintegers Rn ˇ F.t; s/ dt ˇ < ". Therefore, by the standard Cauchy criterion, the sequence m ffn .s/gn>a converges uniformly on D0 to a function g.s/ on D0 : For any "1 > 0, there ˇexists an integer 1 / > a such that for ˇ an integer n N ."1 /, one has ˇR." ˇ N n ˇ ˇ ˇ "1 > fn .s/ g.s/ D a F.t; s/ dt f .s/ˇ for all s 2 D0 , since necessarily g.s/ D f .s/. Now let " > 0 be given. Again, by hypothesis, we canˇ choose ˇRb B."/ > a such that for b2 > b1 > B."/ one has that ˇ b12 F.t; s/ dt ˇ < "=2 for all s 2 D0 . Taking the quantity "1 considered a few lines above equal to "=2, we can find ˇR n an integer N ."1 / ˇ> B."1 / such that for an integer n N ."1 /, "=2 D "1 > ˇ a F.t; s/ dt f .s/ˇ for all s 2 D0 . Thus suppose b > N ."1 /. Then, for all s 2 D0 , ˇ ˇ Z N ."1 / ˇ ˇZ b Z b ˇ ˇ ˇ ˇ ˇ F.t; s/ dt f .s/ C F.t; s/ dt f .s/ˇˇ D ˇˇ F.t; s/ dt ˇˇ ˇ a

a

N ."1 /

" " C D "; 2 2

where N ."1 / > a since B."1 / > a, with N ."1 / dependent only on ". This shows that f .s/ is uniformly convergent on D0 . The Cauchy criterion is therefore validated. As an example, we use the Cauchy criterion to prove the following, very useful result: T HEOREM A.1 (W EIERSTRASS M- TEST ). Let M.t / 0 be a function on def Œa; 1/ that is integrable on each Œa; b with b > a. Assume also that I D

80

FLOYD L. WILLIAMS

R1

R1 M.t / dt exists. If jF.t; s/j M.t / on Œa; 1/D0 , thenf .s/ D a F.t; s/ dt converges uniformly on D0 . Again D0 is any subset of D. Rb P ROOF. Let " > 0 be assigned. That ˇ I D R blimb!1 ˇa M.t / dt implies there ˇ exists a number B."/ > a such that ˇI a M.t / dt < "=2 for b > B."/. If b2 > b1 > B."/, then ˇ ˇ Z b2 ˇ ˇ Z b2 Z b1 ˇ " " ˇ ˇ ˇ ˇ ˇ ˇ M.t / dt I C I M.t / dt ˇˇ < C D "I M.t / dt ˇ D ˇ ˇ 2 2 a a b1 ˇ ˇR b ˇ ˇR b ˇ Rb ˇ Rb hence ˇ b12 F.t; s/ dt ˇ b12 ˇF.t; s/ˇ dt b12 M.t / dt D ˇ b12 M.t / dt ˇ (since M.t / 0; b2 > b1 ). But this is less than ", for all s 2 D0 . Theorem A.1 follows, therefore, by the Cauchy criterion. ˜ a

The question of the holomorphicity of f .s/ is settled by the following theorem. T HEOREM A.2. Again let F.t; s/ be defined on Œa; 1/D with D C an open subset. Assume (i) F.t; s/ is continuous on Œa; 1/D (in particular for each s 2 D, t ‘ F.t; s/ is integrable on Œa; b for every b > a); (ii) for every t a fixed, s ‘ F.t; s/ is holomorphic on D; (iii) for everyR s 2 D fixed, t ‘ @F.t; s/=@s is continuous on Œa; 1/; def 1 (iv) f .s/ D a F.t; s/ dt converges for every s 2 D; and (v) f .s/ converges uniformly on compact subsets of D. R1 Then f .s/ is holomorphic on D, and f 0 .s/ D a @F.t; s/=@s dt for every s 2 D. R1 Implied here is the existence of the improper integral a @F.t; s/=@s dt on D. The R 1 idea of the proof is to reduce matters to a situation where the integration Rn over an infinite range is replaced by that over a finite range a a , where holomorphicity is known to follow. This is easily done by considering again the def R n sequence ffn .s/gn>a discussed earlier: fn .s/ D a F.t; s/ dt on D, which is well-defined by (i). If K D is compact, then given (v), the above argument with D0 now taken to be K shows exactly (by way of the Cauchy criterion) that ffn .s/gn>a converges uniformly on K (to f .s/ by (iv)). On the other hand, by (i), (ii), (iii) we have that (i)0 F.t; s/ is continuous on Œa; n D; (ii)0 for every t 2 Œa; n fixed, s ‘ F.t; s/ is holomorphic on D, and (iii)0 for every s 2 D fixed, t ‘ @F.t; s/=@s is continuous on Œa; nI here a < n 2 Z . RGiven (i)0 , (ii)0 n and (iii)0 , it is standard in complex variables texts that fn .s/ D a F.t; s/ dt is R n holomorphic on D and that fn0 .s/ D a @[email protected]; s/ dt . Since we have noted that the sequence ffn .s/gn>a converges uniformly to f .s/ on compact subsets K of D, it follows by the Weierstrass theorem that f .s/ is holomorphic on D, and that fn0 .s/ ‘ f 0 .s/ pointwise on D — with uniform convergence on compact subsets


of D, in fact. That is, f 0 .s/ D limn!1 fn0 .s/ D limn!1 R1 a @F.t; s/=@s dt on D, which proves Theorem A.2.

Rn a

81

@F.t; s/=@s dt D

As an application, we check that the function J.s/ P in definition (1.4) is an def n2 t , for t > 0, entire function. First we claim that the function 0 .t / D 1 nD1 e converges uniformly on Œ1; 1/. This is clear, by the Weierstrass M-test, since 2 for n; t 1, wePhave n2 n, hence n2 t nt n, hence e n t e n , n is a convergent geometric series. Therefore .t / is and moreover 1 0 nD1 e 2 continuous on Œ1; 1/, since theR terms e n t are continuous in t on Œ1; 1/. By def 1 definitions (1.2), (1.4), J.s/ D 1 F.t; s/ dt for F.t; s/ D 0 .t /t s on Œ1; 1/C , where F.t; s/ therefore is also continuous. Again for n; t 1, n2 t nt and 2 also t , so e n t e nt and e t e , so 1 e t 1 e , so 1

1 e

t

1 X

e

1 1 e

D

e

e

def

1

D C:

That is, 0 .t / D

1 X

nD1

e

n2 t

nD1

nt

D

1 X

.e

e

t n

nD1

/ D

1

t

e

t

Ce

t

ˇ ˇ R1 for t 1, so ˇF.t; s/ˇ C e t t Re s on Œ1; 1/ C , where 1 e bt t a dt converges for b > 0; a 2 R . Thus J.s/ converges absolutely for every s 2 C . We see that conditions (i) and (iv) of Theorem A.2 hold. Conditions (ii) and (iii) certainly hold. To check condition (v), let K C be any compact subset. The continuous function s ! Re s on K has an upper bound W Re s on K ) t Re s t on Œ1; 1/ K (since log t 0 for Rt 1). That is, on 1 Œ1; 1/ K the estimate jF.t; s/j C e t t holds where 1 e t t dt < 1, implying that J.s/ converges uniformly on K, by Theorem A.1. Therefore J.s/ is holomorphic on C by Theorem A.2. def

B. A Fourier expansion (or q-expansion). The function q.z/ D e 2 iz is holomorphic and it satisfies the periodicity condition q.z C1/ D q.z/. Suppose f .z/ is an arbitrary holomorphic function defined on an open horizontal strip def

Sb1 ;b2 D fz 2 C j b1 < Im z < b2 g

(B.1)

as indicated in the figure at the top of the next page, where b1 ; b2 2 R , b1 < b2 . Suppose also that f .z/ satisfies the periodicity condition f .z C 1/ D f .z/ on Sb1 ;b2 ; clearly z 2 Sb1 ;b2 implies z C r 2 Sb1 ;b2 for all r 2 R . Then f .z/ has a Fourier expansion (also called a q-expansion) X X f .z/ D an q.z/n D an e 2 inz (B.2) n2Z

n2Z

82

FLOYD L. WILLIAMS

y

y D b2 Sb1 ;b2 y D b1

on Sb1 ;b2 , for suitable coefficients an 2 C ; see Theorem B.7 and equation (B.6) below for an expression of the an . The finiteness of b2 is not essential for the validity of equation (B.2). In fact, one of its most useful applications is in case when Sb1 ;b2 is the upper half plane: b1 D 0, b2 D 1. The Fourier expansion of f .z/ follows from the local invertibility of the function q.z/ and the Laurent expansion of the function .f ı q 1 /.z/. We fill in the details of the proof. Note first that q.z/ is a surjective map of the strip Sb1 ;b2 onto the annulus def def def Ar1 ;r2 D fw 2 C j r1 < jwj < r2 g for r1 D e 2b2 > 0, r2 D e 2b1 > 0 (see figure below). y

Ar1 ;r2 r1

x r2


83

For if z D x C iy 2 Sb1 ;b2 , we have jq.z/j D e 2y and b1 < y < b2 , so e 2b1 > e 2y > e 2b2 , and w D q.z/ 2 Ar1 ;r2 . On the other hand, if w 2 Ar1 ;r2 is given choose t 2 R such that e it D w=jwj (since w ¤ 0), and define def r D log jwj. Then one quickly checks that z D t =2 C i r=. 2/ 2 Sb1 ;b2 such that q.z/ D w, as desired. From f .z C 1/ D f .z/, it follows by induction that f .z C n/ D f .z/ for every positive integer n, and therefore for every negative integer n, f .z/ D f .z C n C . n// D f .z C n/; i.e. f .z C n/ D f .z/ for every n 2 Z . Also since q.z1 / D q.z2 / ( ) e 2 iz1 D e 2 iz2 ( ) e 2 i.z1 z2 / D 1 ( ) z1 D z2 C n for some n 2 Z , the surjectivitiy of q.z/ implies that the equation F.q.z// D f .z/ provides for a well-defined function F.w/ on the annulus Ar1 ;r2 . To check that F.w/ is holomorphic, given that f .z/ is holomorphic, take any w0 2 Ar1 ;r2 and choose z0 2 Sb1 ;b2 such that q.z0 / D w0 , again by the surjectivity of q.z/. Since q 0 .z/ D 2 ie 2 iz implies in particular that q 0 .z0 / ¤ 0, one can conclude that q.z/ is locally invertible at z0 : there exist " > 0 and a neighborhood N of z0 , N Sb1 ;b2 , on which q is injective with i:e:

q.N / D N" .q.z0 // D N" .w0 / D fw 2 C j jw and with q

1

w0 j < "g Ar1 ;r2 ;

holomorphic on N" .w0 /. Thus, on N" .w0 /, F.w/ D F q.q 1 .w// D .f ı q 1 /.w/;

which shows that F is holomorphic on N" .w0 / and thus is holomorphic on Ar1 ;r2 , as w0 2 Ar1 ;r2 is arbitrary. Now F.w/ has a Laurent expansion F.w/ D

1 X

nD0

aQ n w n C

1 Q X bm wm

mD1

on the annulus Ar1 ;r2 where the coefficients aQ n ; bQm are given by Z Z F.w/ dw F.w/ 1 1 Q aQ n D ; bm D dw; nC1 2 i 2 i w w mC1

for any circle in Ar1 ;r2 that separates the circles jwj D r1 , jwj D r2 . We def choose to be the circle centered at w D 0 with radius R D e 2b , given any b with b1 < b < b2 ; r1 < R < r2 . For a continuous function .w/ on the change of variables v.t / D 2 t on Œ0; 1 permits the expression Z 2 Z Z 1 iv iv .w/ dw D .Re /Rie dv D 2 i .Re2 it / Re2 it dt 0

D 2 i

0

Z

0

1

.e 2 i.tCib/ /e 2 i.tCib/ dt;

84

FLOYD L. WILLIAMS

by definition of R. For the choices spectively, one finds that aQ n D bQm D

1

Z

.w/ D F.w/=w nC1 , F.w/=w

F.e 2 i.tCib/ /e

2 in.tCib/

re-

dt;

0

Z

mC1 ,

(B.3)

1

F.e

2 i.tCib/

/e

2 im.tCib/

dt;

0

for n 0; m 1. However t C ib 2 Sb1 ;b2 since b1 < b < b2 so by definition of F.w/ the equations in (B.3) are aQ n D bQm D

Z

Z

1

f .t C ib/e

0 1

f .t C ib/e

0

2 in.tCib/

dt; (B.4)

2 im.tCib/

dt;

for n 0; m 1, and moreover the Laurent expansion of F.w/ has a restatement f .z/ D

1 X

nD0

n

aQ n q.z/ C

1 X

mD1

bQm q.z/m

(B.5)

on Sb1 ;b2 . One can codify the preceding formulas by defining def

an D

Z

1 0

f .t C ib/e

2 in.tCib/

dt

(B.6)

for n 2 Z , again for b1 < b < b2 . Then an D aQ n for n 0 and a n D bQn for n 1. By equation (B.5) we have therefore completed the proof of equation (B.2): T HEOREM B.7 (A F OURIER EXPANSION ). Let f .z/ be holomorphic on the open strip Sb1 ;b2 defined in equation (B.1), and assume that f .z/ satisfies the periodicity condition f .z C 1/ D f .z/ on Sb1 ;b2 . Then f .z/ has a Fourier expansion on Sb1 ;b2 given by equation (B.2), where the an are given by equation (B.6) for n 2 Z , for arbitrary b subject to b1 < b < b2 . Theorem B.7 is valid if Sb1 ;b2 is replaced by the upper half-plane C (with b1 D 0, b2 D 1), for example, as we have indicated. For clearly the preceding arguments hold for b2 D 1. Here, in place of the statement that q W Sb1 ;b2 ! def def Ar1 ;r2 is surjective (again for r1 D e 2b2 , r2 D e 2b1 ; b2 < 1), one simply employs the statement that q W C ! fw 2 C j 0 < jwj < 1g is surjective.


85

C. Poisson summation and Jacobi inversion. The Jacobi inversion formula (1.3) can be proved by a special application of the Poisson summation formula (PSF). The latter formula, in essence, is the statement X X f .n/ D fO.n/; (C.1) n2Z

n2Z

for a suitable class of functions f .x/ and a suitable normalization of the Fourier transform fO.x/ of f .x/. The purpose here is to prove a slightly more general version of the PSF, which applied in a special case, coupled with a Fourier transform computation, indeed does provide for a proof of equation (1.3). For a function h.x/ on R , the definition Z 1 def O h.x/ D h.t /e 2 ixt dt (C.2) 1

will serve as our normalization of its Fourier transform. Here’s what we aim to establish: T HEOREM C.3 (P OISSON SUMMATION ). Let f .z/ be a holomorphic function def on an open horizontal strip Sı D fz 2 C j Pı < Im z < ıg; P ı > 0, say with 1 f jR 2 L1 .R ; dx/. Assume that the series 1 f .z C n/, n/ nD0 nD1 f .z converge uniformly on compact subsets of Sı . Then for any z 2 Sı X X f .z C n/ D e 2 inz fO.n/: (C.4) n2Z

n2Z

In particular for z D 0 we obtain equation (C.1). P ROOF. By the Weierstrass theorem, the uniform convergence of the series P P1 1 f .z C n/ and f n/ on compact subsets of Sı means that the nD0 nD1 .z function def

F.z/ D

P

n2Z

f .z C n/ D

1 P

nD0

f .z C n/ C

1 P

f .z

n/

nD1

P on S is holomorphic. F.z/ satisfies F.z C 1/ D ı n2Z f .z C n C 1/ D P n2Z f .z C n/ D F.z/ on Sı . Therefore, Theorem B.7 of Appendix B is applicable, where the choice b D 0 is made (b1 D ı; b2 D ı): F.z/ has a Fourier expansion P an e 2 inz (C.5) F.z/ D n2Z

def

R1

on Sı , where an D 0 F.t /e 2 int dt for n 2 Z . Since Œ0; 1 Sı is compact, P1 P 2 int and l/e 2 int for n 2 Z fixed the series 1 lD0 f .t C l/e lD1 f .t (whose sum is F.t /e 2 int ) converge uniformly on Œ0; 1 (by hypothesis, given

86

FLOYD L. WILLIAMS

of course that je 2 int j D 1). Therefore an can be obtained by termwise integration; we start by writing Z 1 1 1 P P f .t l/e 2 int dt f .t C l/e 2 int C an D 0

D D

1 P

lD0 0 1 P

lD1

lD0 1

Z

Z

f .t C l/e lC1

f .v/e

lD0 l

2 int

dt C

2 in.v l/

1 P

Z

1

f .t

l/e

2 int

dt

lD1 0

dv C

1 P

lD1

Z

lC1

f .v/e

2 in.vCl/

dv;

l

by the change of variables v.t / D t C l and v.t / D t l on Œl; lC1, Œ l; lC1, R lC1 P f .t /e 2 int dt C respectively (with l 2 Z ), This is further equal to 1 lD0 l R R P1 R lC1 1 0 f .t /e 2 int dt D 0 f .t /e 2 int dt C 1 f .t /e 2 int dt D R 1lD1 l 2 int f .t /e dt D fO.n/, by definition (C.2). That is, by (C.5), for z 2 Sı P1 O def P 2 inz D F.z/ D n2Z f .z C n/, which concludes the proof of n2Z f .n/e Theorem C.3. ˜ Other proofs of the PSF exist. In contrast to the complex-analytic one just presented, a real-analytic proof (due to Bochner) is given in Chapter 14 of [38], for example, based on Fejér’s Theorem, which states that the Fourier series of a continuous, 2-periodic function .x/ on R is Cesàro summable to .x/. 2 def As an example, choose f .z/ D f t .z/ D e z t for t > 0 fixed. In this case f .z/ is an entire function whose restriction to R is Lebesgue integrable; the restriction is in fact a Schwartz function. We claim that the series 1 P

nD0

f .z C n/ and

1 P

f .z

n/

nD1

converge uniformly on compact subsets K of the plane. Since K is compact 2 the continuous functions z ‘ e z t and z ‘ Re z on C are bounded on K W ˇ ˇ ˇ ˇ z 2t ˇ M1 , ˇ Re z ˇ M2 on K for some positive numbers M1 ; M2 . Let n0 ˇe be an integer > 1C 2M2 . Then for n 2 Z with n n0 one has n2 n.1C 2M2 /, 2 2 hence n2 2nM2 n, so that f .z C n/ D e z t e .n C2nz/t for z 2 K. But ˇ .n2 C2nz/t ˇ ˇe ˇ D e .n2 C2n Re z/t e .n2 2nM2 /t D e nt ˇ P 2 ˇ M e nt and ˇe z t ˇ M1 , so jf .z C n/j M1 e nt on K, with 1 P1 nD0 clearly convergent for t > 0. Therefore, by the M -test, nD0 f .z C n/ converges absolutely and uniformly on K. Similarly, for n 2 Z , we have f .z n/ D 2 2 e z t e .n 2nz/t , where for n n0 and z 2 K again n2 2nM2 n, but where we now use the bound Re z M2 : n2 2n Re z n2 2nM2 n


P jf .z n/j M1 e tn on K (for n n0 ), so 1 nD1 f .z def lutely and uniformly on K, which shows that f t .z/ D e the hypotheses of Theorem C.3. The conclusion X X 2 e n t D fOt .n/ n2Z

87

n/ converges abso> 0, satisfies

z 2 t ; t

(C.6)

n2Z

is therefore safe, and the left-hand side here is .t / by definition (1.2). One is therefore placed in the pleasant position of computing the Fourier transform Z 1 2 .C:2/ fOt .x/ D e y t e 2 ixy dy; (C.7) 1

which is a classical computation that we turn to now (for the sake of completeness). p 2 For realpnumbers a; b; c; t withpa < b, t > 0, note that e v e 2 icv= t D 2 c 2 =t e .v ic =t/2 . By the change of variables v.x/ D ep v e 2icv =t p D ep t x on Œa t ; b t , therefore, Z b Z p 2 e c =t b t .v ic p=t/2 x 2 t 2 icx dv: (C.8) e e dx D p e p t a t a Next we show that for b 2 R Z 1 e 1

.xCib/2

dx D

Z

1

e

x2

dx:

(C.9)

1

To do this, assume first that b > 0 and consider the counterclockwise oriented rectangle CR of height b and width 2R W CR D C1 C C2 C C3 C C4 . y

C2

C3

x

C1

C4

88


By Cauchy’s theorem, 0 D IR D R

C1 e

C2 e

z 2 dz

e

z 2 dz

e

z 2 dz

R

R

C3

R

z 2 dz

C4

D

RR

Di D

D

RRb 0

CR e

e

x 2 dx;

e

.RCix/2 dx

RR R

R

R

C2

e

e

z 2 dz.

D ie

Now

R2

Rb 0

.xCib/2 dx;

e

2xRi e x 2 dx;

:

z 2 dz

ˇ ˇR 2R b 2 2 Thus ˇ C2 e z dz ˇ e R 0 e x dx, which tends to 0 as R ! 1. That is, R1 R1 2 2 0 D limR!1 IR D 1 e x dx e .xCib/ dx, which proves equation R 1 1 .xCib/2 R1 2 (C.9) for b > 0. If b < 0, write 1 e dx D 1 e . xCib/ dx D R1 R 2 2 1 .xCi. b// dx D x dx by the previous case, since b > 0. Thus 1e 1e (C.9) holds for all b 2 R (since it clearly holds for b D 0). By (C.8) it then follows that Z 1 Z R 2 2 e x t e 2 icx dx D lim e x t e 2 icx dx 1

R!1

e

R

c 2 =t

D p

t 2

e c =t D p t

lim

R!1

Z

1 1

Z

e

R

p

t

p R t

e

p .xCi. c/ =t/2

dx

2

x2

e c =t p dx D p : t

(C.10)

P ROPOSITION C.11. For c 2 R and t > 0, we have Z 1 Z 1 2 e c =t x 2 t 2 icx x 2 t 2 icx e e dx D e e dx D p : t 1 1 p 2 Hence equation (C.7) is the statement that fOt .x/ D e x =t = t : Having noted that the left-hand side of equation (C.6) is .t /, we see that (C.6) p def p P 2 (by Proposition C.11) now reads .t / D n2Z e n =t = t D . 1t /= t , which proves the Jacobi inversion formula (1.3). D. A divisor lemma and a scholium. The following discussion is taken, nearly word for word, from [38] and thus it has wider applications — for example, applications to the theory of Eisenstein series (see pages 274–276 of that reference). For integers d; n with d ¤ 0 write d j n, as usual, if d P divides n, and def write d - n if d does not divide n. For n 1; 2 C let .n/ D 0 1, is the Riemann (where .s/ D 1 nD1 zeta function) and moreover P1 1 the iterated series nD1 kD1 jd.k; n/k an j converges: 1 1 X X jd.k; n/k an j a. Re /: (D.3) nD1

kD1

By elementary facts regarding double series (found inP advanced it P1calculus texts) a follows that one can conclude that the double series 1 d.k; n/k n nD1 kD1 converges absolutely, and that X 1 1 X 1 X 1 X (D.4) d.k; n/k an : d.k; n/k an D nD1

kD1

kD1

nD1

Similarly for k 1 fixed, equation (D.2) (with fan g1 replaced by fjan jg1 nD1 nD1 ) P1 P P 1 1 Re Re yields mD1 jk akm j D k mD1 jakm j D k lD1 d.k; l/jal j. That is,

90

FLOYD L. WILLIAMS

P1 P jk akm j and moreover the iterated series 1 mD1 kD1 P P1 Re converges, as it equals 1 /. Thus, simkD1 k lD1 d.k; l/jal j a. P Re P 1 ilarly to equation (D.4), one has that the double series 1 kD1 mD1 k akm converges absolutely, and equality of the corresponding iterated series prevails: 1 1 P 1 1 P P P (D.5) k akm : k akm D P1

mD1 jk

a

km j < 1

mD1 kD1

kD1 mD1

Given these observations, we can now state and prove the main lemma regarding the validity of equation (D.1):

D IVISOR L EMMA . Let fan g1 of complex numbers such that the nD1 be a sequence P1 P series nD1 jan j converges. Let .n/ D 0 n. That is, 1 X d.k; n/k D .n/ (D.6) P1

kD1

P for 2 C . The series 1 nD1 Re .n/jan j is, by (D.6), the iterated series P1any P 1 Re ja j , which we have seen converges and is bounded d.k; n/k n kD1 nD1 abovePby . Re / according to (D.3). Clearly j .n/j P Re .n/, so that 1 1 j .n/jja j converges. Again by (D.6), we have also n nD1 .n/an D P1 P1 nD1 (D.4), (D.2), (D.5), suckD1 d.k; n/k an . Now apply equations nD1 P1 P d.k; n/k an D cessively, to express the latter iterated series as 1 nD1 kD1 P1 P1 P1 P1 ka , which proves (D.1). We kD1 mD1 mD1 akm D kD1 k P P1 km have already seen that the double series mD1 1 which equals nD1 m amn P1 P1 (D.5), then one de- kD1 mD1 k akm converges absolutely. By equation P1 P1 rives the equality of the corresponding iterated series mD1 nD1 m amn P1 P1 and nD1 ˜ mD1 m amn . .k/

.k/

Going back to the equality sm D tm of the previous page, we actually have the following fact, recorded for future application:


91

S CHOLIUM . Given a sequence of complex numbers fan g1 and k 2 Z with nD1 P P1 k 1, the series j D1 akj converges if and only if the series j1D1 d.k; j /aj converges, in which case these series coincide. As an example, we use the Divisor Lemma to prove the next lemma, which is important for Lecture 4. L EMMAP D.7. Fix z; k 2 Cwith Im z > 0, ReP k > 2. Then the iterated series P 1 1 1 k 1 e 2 imn exists, the series k 1 e 2 inz conn mD1 nD1 nD1 1 k .n/n verges absolutely, and P 1 1 1 1 P P P k 1 .n/e 2 inz : (D.8) nk 1 e 2 imn D 1 k .n/nk 1 e 2 inz D mD1 nD1

nD1

nD1

n

P ROOF. The last equality comes D .n/. To show the first, P from .n/ P1 def Re k 1 e 2 n Im z converges set an D nk 1 e 2 inz . Then 1 ja j D n nD1 n nD1 by the ratio test, since Im z > 0. Also m1 k amn D m1 k .mn/k 1 e 2 imnz D nk 1 e 2 imnz , where ReP k > 2 ) Re.1 k/ < 1. By the Divisor Lemma (for D 1 k), the series 1 absolutely, the iterated nD1 1 k .n/an converges P P1 P1 P1 1 ka series 1 m converges, and m1 k amn D mn nD1 mD1 nD1 mD1 P1 nD1 1 k .n/an . Substituting the value of an proves the desired equality. ˜ As another example, consider the sequence fan g1 nD1 given by def

an D e ˙2 nxi Ks

1 2

.2 ny/ns

1 2

for x; y 2 R , y > 0, s 2 C fixed, where Z 1 def 1 1 z tC t K .z/ D exp 2 t 2 0

1

dt

(D.9)

P is the K-Bessel function for Re z > 0, 2 C . To see that 1 nD1 jan j < 1, one applies the asymptotic result r p t lim t K .t /e D : (D.10) t!1 2 p ˇ p ˇp =2 ˇ< =2 for t > M ; In particular, choose M > 0 such that ˇ t K .t /e t p that is, jK .t /j < 2 =2t e t for t > M . Then if Ns;y is an integer with Ns;y > Ms 1 =2y we see that for n Ns;y , 2 ny > Ms 1 , so 2

2

ˇ ˇK

s

r ˇ ˇ 1 .2 ny/ < 2 2

p therefore jan j < .nRe s 1 = y/ e by the ratio test since y > 0.

e 2 2 ny 2 ny ,

2 ny

where

1=2

n D p

P1

y

e

2 ny

I

Re s 1 e 2 ny nD1 n

converges

92


Now assume that Re s > 1 so that D 1 2s C 1 satisfies Re < 1. Also def n s 2 for m; n 1. The Divisor Lemma m amn D e ˙2 mnxi Ks 1 .2 mny/ m 2 gives 1 n s 12 1 P ˙2 mnxi P e Ks 1 .2 mny/ 2 m mD1 nD1 1 P 1 D 2sC1 .n/e ˙2 nxi Ks 1 .2 ny/ns 2 ; (D.11) 2

nD1

with absolute convergence of the latter series, and convergence of the iterated series which coincides with the iterated series 1 s 12 1 P ˙2 mnxi P n : e Ks 1 .2 mny/ 2 m nD1 mD1

E. Another summation formula and a proof of formula (2.4). In addition to the useful Poisson summation formula X X f .n/ D fO.n/ (E.1) n2Z

n2Z

of Theorem C.3, there are other very useful, well known summation formulas. The one that we consider here assumes the form P f .n/ D the sum of residues of . cot z/f .z/ at the poles of f .z/; (E.2) n2Z

for a suitable class of functions f .z/. As we applied formula (E.1) to a specific 2 function (namely the function f .z/ D e z t for t > 0 fixed) to prove the Jacobi inversion formula (1.3), we will, similarly, apply formula (E.2) to a specific function (namely the function f .z/ D .z 2 C a2 / 1 for a > 0 fixed) to prove formula (2.4) of Lecture 2. The main observation towards the proof of formula (E.2) is that there is a nice bound for jcot zj on a square CN with side contours RN ; LN and top and bottom contours TN ; BN , as illustrated on the next page, for a fixed integer N > 0. The bound, in fact, is independent of N . Namely, def

jcot zj max.1; B/ < 2

(E.3)

on CN , for B D .1C e /=.1 e /. We begin by checking this known result. For z D x C iy, x; y 2 R , we have iz D y C ix, and simple manipulations give cot z D i

e iz C e e iz e

jcot zj

e y C e y je y e ix e y e

Hence

iz iz

Di

e e

y e ix

C e y e y e ix e y e

ix j

e je

y

ix ix

C e y y e y j

:

(E.4)


y .N C 21 /; N C 21

93

TN

N C 12 ; N C 21

LN

N

N C 21

x

N C1

RN

.N C 12 /; .N C 21 /

BN

N C 12 ; .N C 21 /

ˇ ˇ (since ja bj ˇjaj jbjˇ for a; b 2 C ). Since jaj ˙a for a 2 R , this becomes jcot zj

e y C e y : ˙.e y e y /

(E.5)

Suppose (in general) that y > 21 . Then 2y > , so e 2y < e and 1 e 2y > 1 e , which, by the choice of the minus sign in (E.5), allows us to write y e 2y C 1 e C 1 def e C e y e y D D B; jcot zj < e y e y e y 1 e 1 e 2y for y > 12 . Similarly if y < 21 , then 2y < , so e 2y < e and 1 e 2y > 1 e , and by the choice of the plus sign in (E.5) we get y e C e y e y 1 C e 2y 1C e jcot zj D D B: < e y e y e y 1 e 1 e 2y Thus we see that def

jcot zj < B D

1C e 1 e

(E.6)

for z 2 C with either Im z > 12 or Im z < 12 . In particular on TN , Im z D N C 12 > 12 and on BN , Im z D .N C 21 / < 21 so by (E.6) the estimate jcot zj < B holds on both the contours TN and BN .

(E.7)

94

FLOYD L. WILLIAMS

Regarding the contours RN and LN we have z D N C 21 C iy on RN and z D .N C 12 / C iy on LN . On RN we have e 2 iz D e 2y (since N 2 Z ) and, similarly, e 2 iz D e 2y on LN . Now consider the first equation in (E.4) and multiply the fraction by 1 D iz e =e iz . On both RN and LN this leads to cot z D i

e 2 iz C 1 Di e 2 iz 1

e e

2y 2y

C1 ; 1

and we conclude that j1 e 2 Im z j 1 C e 2 Im z D1 (E.8) 1 C e 2 Im z 1 C e 2 Im z on both contours RN and LN . The inequalities (E.7), (E.8) therefore imply (E.3), as desired, where we note that e t 1C t for t 2 R , so e 1C > 3 ) 2.e 1/ .e C 1/ D e 3 > 0. That is, indeed e C1 e 1 C e def 2> D D B: e 1 e 1 e jcot zj D

We note also that sin z D 0 ( ) z D n 2 Z . That is, since (again) N 2 Z we cannot have sin z D 0 on CN ; in particular CN avoids the poles of cot z D cos z=sin z, and cot z is continuous on CN . Consider now a function f .z/ subject to the following two conditions: C1. f .z/ is meromorphic on C , with only finitely many poles z1 ; z2 ; : : : ; zk , none of which is an integer. C2. There are numbers M; > 0 such that jf .z/j M=jzj2 holds for jzj > . Then:

P T HEOREM E.9. limN !1 N nD N f .n/ exists and equals minus the sum of the residues of the function f .z/ cot z at the poles z1 ; z2 ; : : : ; zk of f .z/. One can replace condition C2, in fact, by the more general condition (E.11) below. P ROOF. Since the poles zj are finite in number we can choose N sufficiently large that CN encloses all of them. The function cot z has simple poles at the integers (again as sin z D 0 ( ) z D n 2 Z ) and the residue at z D n 2 Z is def immediately calculated to be 1. Therefore the residue of .z/ D f .z/ cot z at z D n 2 Z is f .n/. As none of the zj are integers (by C1) the poles of .z/ within CN are given precisely by the set fzj ; n j 1 j k; N n N; n 2 Z g. By the residue theorem, accordingly, we deduce that Z 1 .z/dz D 2 i CN N P f .n/: (E.10) the sum of the residues of .z/ at the zj C nD N


Now if

Z

lim

N !1 CN

.z/dz D 0;

95

(E.11)

we can let N ! 1 in equation (E.10) and conclude the validity of Theorem E.9 (more generally, without condition C2). We check that condition (E.11) is implied by condition C2. Since jzj N C 21 on CN , we have for N C 12 > and z on CN the bound jf .z/j

M M 2 : 2 jzj N C 12

By the main inequality (E.3), we have j cot zj < 2 on CN , so j.z/j D 2 jf .z/ cot zj < 2M= N C 12 on CN . Given that the length of CN is 4 2 N C 21 we therefore have the following estimate (for N C 12 > ): ˇ ˇZ ˇ ˇ 2M 32M ˇ .z/dz ˇˇ 8 N C 12 D ; (E.12) ˇ 2 2N C 1 CN N C 21

where we note that .z/ is continuous on CN , because, as seen, cot z is continuous on CN . The inequality (E.12) clearly establishes the condition (E.11), by which the proof of Theorem E.9 is concluded. ˜ As an example of Theorem E.9 we choose f .z/ D

1 z 2 C a2

D

.z

1 ai/.z C ai/

for a > 0 fixed. Hence f is meromorphic on C with p exactly two simple poles def def z1 D ai, z2 D ai. Suppose, for example, that jzj > 2a: Then ˇ ˇ ˇ ˇ ˇ ˇ z2 ˇ a2 ˇˇ a2 a2 1 ˇ ˇ D ˇ 1 ˇ < 2: ˇ > ; so ˇ1 C 2 ˇ 1 ; so ˇ 2 1 2 2 2 2 jzj z jzj z Ca ˇ ˇ a2 ˇ ˇ1 C 2 ˇ z

2 Therefore is, f .z/ satisfies conditions C1, C2 with M D 2, p jf .z/j < 2=jzj ; that def D 2a. The residue of .z/ D f .z/ cot z at z1 is

lim .z

z!z1

z1 /.z/ D lim

z!ai

cot z cot ai D D z C ai 2ai

coth a; 2a

since cos iw D cosh w, sin iw D i sinh w. Similarly, the residue of .z/ at z2 PN P 1 1 is .=2a/ coth a. As f . z/ D f .z/, N nD1 n2 Ca2 . nD N f .n/ D a2 C 2 Theorem E.9 therefore gives 1

X 1 1 C2 D 2 2 a n C a2 nD1

coth a 2a

coth a D coth a; 2a a

96

FLOYD L. WILLIAMS

which proves the summation formula (2.4). def

F. Generators of SL.2; Z /. Let D SL.2; Z /. To prove that the elements def def T D 10 11 and S D 01 10 2 generate , we start with a lemma. L EMMA F.1. Let D ac db 2 with c 1. Then is a finite product 1 l , where each j 2 has the form j D T nj S mj for some nj ; mj 2 Z . Here for a def group element g and 0 > n 2 Z , we have set g n D .g 1 / n . The proof is by induction on c. If c D 1, then 1 D det D ad bc D ad so b D ad 1, so

D a1 add 1 D 10 a1 01 10 10 d1 10 01 D 1 2

b,

for 1 D 10 a1 01 10 D T a S 1 , 2 D 10 d1 10 01 D T d S 0 . Proceeding by induction, we use the Euclidean algorithm to write d D qc C r for q; r 2 Z with 0 r < c 2, say. If r D 0, then 1 D det D ad bc D .aq b/c, which shows that c is a positive divisor of 1. That is, the contradiction c D 1 implies that r > 0. Now aqCb a aqCb

T q D ac db 10 1q D ac cqCd D c r ;

0 1 aqCb a so T q S D ac aqCb r c , which equals 1 l by induc1 0 D r tion (since 1 r < c), where each j has the form j D T nj S mj for some nj ; mj 2 Z . Consequently,

D 1 l S

1

T q D . 1 l

1 /.T

nl

S ml

1

/T q S 0 ;

which has the desired form for and which therefore completes the induction and the proof of Lemma F.1. T HEOREM F.2. The elements T; S generate : Every 2 is a finite product

1 l where each j 2 has the form j D T nj S mj for some nj ; mj 2 Z . P ROOF. Let D ac db 2 be arbitrary. If c D 0, then 1 D det D ad, so a D d D ˙1, so 1 0 b S 2:

D 10 b1 D T b S 0 or D 10 b1 D 10 b1 0 1 DT

Since the case c 1 is already settled by Lemma F.1, there remains only the case c 1. Then S 2 D 10 01 D ac db D 1 l , by Lemma F.1 (since c 1), where the j have the desired form. Thus D 1 l .T 0 S 2 /, as desired. ˜


97

G. Convergence of the sum of jmCnij ˛ for ˛ > 2. To complete the argument that the Eisenstein series Gk .z/ converge absolutely and uniformly on each of the strips SA;ı in definition (4.5) (for k D 4; 6; 8; 10; 12; : : : ), we must show, according to the inequality (4.10), that the series X 1 def S.˛/ D jm C nij˛ 2 .m;n/2Z

converges for ˛ > 2, where Z 2 D Z Z f.0; 0/g. For n 1; n 2 Z , let n denote the set of integer points on the boundary of the square with vertices .n; n/, . n; n/, . n; n/, .n; n/. As an example, 3 is illustrated here, with 24 D 8 3 points. y

x

.i/

In general n has jn j D 8n points. Also the n partition out all of the nonzero integer pairs: Z Z

is a disjoint union. P L EMMA G.2.

.a;b/2n

1 jaCbij˛

f.0; 0/g D

8 n˛

1

1 S

n

(G.1)

nD1

for ˛ 0; n 1.

P ROOF. For .a; b/ 2 n , either a D ˙n or b D ˙n, according to whether .a; b/ lies on one of the vertical sides of the square (as illustrated above for 3 ), or on one of the horizontal sides, respectively. Thus a2 C b 2 D either n2 C b 2 or

98

FLOYD L. WILLIAMS

a2 Cn2 ) a2 Cb 2 n2 . That is, for .a; b/ 2 n we have jaCbij2 D a2 Cb 2 n2 , so ja C bij˛ n˛ (since ˛ 0). Inverting and summing we get X X 1 1 8n ˛ (by (i)) ˛ ˛ ja C bij n n .a;b/2n

.a;b/2n

D

8

n˛ 1

;

which proves Lemma G.2.

˜

Now use (G.1) and Lemma G.2 to write S.˛/ D

1 X

X

nD1 .a;b/2n

1

X 8 1 ja C bij˛ n˛

for ˛ 0, which proves that S.˛/ < 1 for ˛

1

(G.3)

nD1

1 > 1, as desired.

References [1] T. M. Apostol, Introduction to analytic number theory, Springer, New York, 1976. [2]

, Modular functions and Dirichlet series in number theory, second ed., Graduate Texts in Mathematics, no. 41, Springer, New York, 1990.

[3] N. Arkani-Hamed, S. Dimopoulos, and G. Dvali, Phenomenology, astrophysics, and cosmology of theories with submillimeter dimensions and tev scale quantum gravity, Phys. Rev. D 59 (1999), art. #086004, See also hep-ph/9807344 on the arXiv. [4] D. Birmingham, I. Sachs, and S. Sen, Exact results for the BTZ black hole, Internat. J. Modern Phys. D 10 (2001), no. 6, 833–857. [5] D. Birmingham and S. Sen, Exact black hole entropy bound in conformal field theory, Phys. Rev. D (3) 63 (2001), no. 4, 047501, 3. [6] N. Brisebarre and G. Philibert, Effective lower and upper bounds for the Fourier coefficients of powers of the modular invariant j , J. Ramanujan Math. Soc. 20 (2005), no. 4, 255–282. [7] D. Bump, J. W. Cogdell, E. de Shalit, D. Gaitsgory, E. Kowalski, and S. S. Kudla, An introduction to the Langlands program, Birkhäuser, Boston, MA, 2003. [8] J. J. Callahan, The geometry of spacetime: An introduction to special and general relativity, Springer, New York, 2000. [9] J. L. Cardy, Operator content of two-dimensional conformally invariant theories, Nuclear Phys. B 270 (1986), no. 2, 186–204. [10] S. Carlip, Logarithmic corrections to black hole entropy, from the Cardy formula, Classical quantum gravity 17 (2000), no. 20, 4175–4186. [11] S. Carroll, Spacetime and geometry: An introduction to general relativity, Addison Wesley, San Francisco, 2004.


99

[12] E. Elizalde, Ten physical applications of spectral zeta functions, Lecture Notes in Physics. New Series: Monographs, no. 35, Springer, Berlin, 1995. [13] E. Elizalde, S. D. Odintsov, A. Romeo, A. A. Bytsenko, and S. Zerbini, Zeta regularization techniques with applications, World Scientific, River Edge, NJ, 1994. [14] S. S. Gelbart and S. D. Miller, Riemann’s zeta function and beyond, Bull. Amer. Math. Soc. (N.S.) 41 (2004), no. 1, 59–112. [15] R. C. Gunning, Lectures on modular forms, Annals of Mathematics Studies, no. 48, Princeton University Press, Princeton, N.J., 1962. [16] S. W. Hawking, Zeta function regularization of path integrals in curved spacetime, Comm. Math. Phys. 55 (1977), no. 2, 133–148. ¨ [17] E. Hecke, Uber die Bestimmung Dirichletscher Reihen durch ihre Funktionalgleichung, Math. Ann. 112 (1936), no. 1, 664–699. ¨ [18] , Uber Modulfunktionen und die Dirichletschen Reihen mit Eulerscher Produktentwicklung, I, Math. Ann. 114 (1937), no. 1, 1–28. [19] A. Ivić, The Riemann zeta-function: The theory of the riemann zeta-function with applications, Wiley, New York, 1985. [20] H. Iwaniec, Topics in classical automorphic forms, Graduate Studies in Mathematics, no. 17, American Mathematical Society, Providence, RI, 1997. [21] A. Kehagias and K. Sfetsos, Deviations from the 1=r 2 Newton law due to extra dimensions, Phys. Lett. B 472 (2000), no. 1-2, 39–44. [22] K. Kirsten, Spectral functions in mathematics and physics, Chapman and Hall / CRC, 2002. [23] M. I. Knopp, Automorphic forms of nonnegative dimension and exponential sums, Michigan Math. J. 7 (1960), 257–287. [24] D. Lyon, The physics of the riemann zeta function, online lecture, available at http:// tinyurl.com/yar2znr. [25] J. E. Marsden, Basic complex analysis, W. H. Freeman and Co., San Francisco, Calif., 1973. ˚ Pleijel, Some properties of the eigenfunctions of the [26] S. Minakshisundaram and A. Laplace-operator on Riemannian manifolds, Canadian J. Math. 1 (1949), 242–256. ¨ [27] H. Petersson, Uber die Entwicklungskoeffizienten der automorphen Formen, Acta Math. 58 (1932), no. 1, 169–215. [28] H. Rademacher, The Fourier coefficients of the modular invariant J( ), Amer. J. Math. 60 (1938), no. 2, 501–512. , Fourier expansions of modular forms and problems of partition, Bull. [29] Amer. Math. Soc. 46 (1940), 59–73. [30] , Topics in analytic number theory, Grundlehren math. Wissenschaften, vol. Band 169, Springer, New York, 1973.

100

FLOYD L. WILLIAMS

[31] H. Rademacher and Herbert S. Zuckerman, On the Fourier coefficients of certain modular forms of positive dimension, Ann. of Math. (2) 39 (1938), no. 2, 433–462. ¨ [32] B. Riemann, Uber die Anzahl der Primzahlen unter einer gegebenen Grösse, Mon. Not. Berlin Akad. (1859), 671–680, An English translation by D. Wilkins is available at http://www.maths.tcd.ie/pub/histmath/people/riemann/zeta. [33] P. Sarnak, Some applications of modular forms, Cambridge Tracts in Mathematics, no. 99, Cambridge University Press, Cambridge, 1990. [34] A. Strominger and C. Vafa, Microscopic origin of the Bekenstein–Hawking entropy, Phys. Lett. B 379 (1996), no. 1-4, 99–104. [35] A. Terras, Harmonic analysis on symmetric spaces and applications, I, Springer, New York, 1985. [36] E. C. Titchmarsh, The theory of the Riemann zeta-function, Clarendon Press, Oxford, 1951. [37] H. Weyl, The theory of groups and quantum mechanics, Dover, New York, 1931. [38] F. L. Williams, Lectures on the spectrum of L2 . nG/, Pitman Research Notes in Mathematics Series, no. 242, Longman, Harlow, 1991. , The role of Selberg’s trace formula in the computation of Casimir [39] energy for certain Clifford–Klein space-times, African Americans in mathematics (Piscataway, NJ, 1996), DIMACS Ser. Discrete Math. Theoret. Comput. Sci., no. 34, Amer. Math. Soc., Providence, RI, 1997, pp. 69–82. [40] , Topological Casimir energy for a general class of Clifford–Klein spacetimes, J. Math. Phys. 38 (1997), no. 2, 796–808. [41] , Meromorphic continuation of Minakshisundaram–Pleijel series for semisimple Lie groups, Pacific J. Math. 182 (1998), no. 1, 137–156. , Topics in quantum mechanics, Progress in Mathematical Physics, no. 27, [42] Birkhäuser Boston Inc., Boston, MA, 2003. F LOYD L. W ILLIAMS D EPARTMENT OF M ATHEMATICS AND S TATISTICS L EDERLE G RADUATE R ESEARCH TOWER 710 N ORTH P LEASANT S TREET U NIVERSITY OF M ASSACHUSETTS A MHERST, MA 01003-9305 U NITED S TATES [email protected]


Basic zeta functions and some applications in physics KLAUS KIRSTEN

1. Introduction It is the aim of these lectures to introduce some basic zeta functions and their uses in the areas of the Casimir effect and Bose–Einstein condensation. A brief introduction into these areas is given in the respective sections; for recent monographs on these topics see [8; 22; 33; 34; 57; 67; 68; 71; 72]. We will consider exclusively spectral zeta functions, that is, zeta functions arising from the eigenvalue spectrum of suitable differential operators. Applications like those in number theory [3; 4; 23; 79] will not be considered in this contribution. There is a set of technical tools that are at the very heart of understanding analytical properties of essentially every spectral zeta function. Those tools are introduced in Section 2 using the well-studied examples of the Hurwitz [54], Epstein [38; 39] and Barnes zeta function [5; 6]. In Section 3 it is explained how these different examples can all be thought of as being generated by the same mechanism, namely they all result from eigenvalues of suitable (partial) differential operators. It is this relation with partial differential operators that provides the motivation for analyzing the zeta functions considered in these lectures. Motivations come for example from the questions “Can one hear the shape of a drum?”, “What does the Casimir effect know about a boundary?”, and “What does a Bose gas know about its container?” The first two questions are considered in detail in Section 4. The last question is examined in Section 5, where we will see how zeta functions can be used to analyze the phenomenon of Bose–Einstein condensation. Section 6 will point towards recent developments for the analysis of spectral zeta functions and their applications.

101

102

KLAUS KIRSTEN

2. Some basic zeta functions In this section we will construct analytical continuations of basic zeta functions. From these we will determine the meromorphic structure, residues at singular points and special function values. 2.1. Hurwitz zeta function. We start by considering a generalization of the Riemann zeta function 1 X 1 : (2-1) R .s/ D ns nD1

D EFINITION 2.1. Let s 2 C and 0 < a < 1. Then for Re s > 1 the Hurwitz zeta function is defined by H .s; a/ D

1 X nD0

1 : .n C a/s

Clearly, H .s; 1/ D R .s/. Results for a D 1 C b > 1 follow by observing 1 X

H .s; 1 C b/ D

nD0

1 D H .s; b/ .n C 1 C b/s

1 : bs

In order to determine properties of the Hurwitz zeta function, one strategy is to express it in term of ’known’ zeta functions like the Riemann zeta function. T HEOREM 2.2. For 0 < a < 1 we have 1 X 1 H .s; a/ D s C . 1/k a kD0

.s C k/ k a R .s C k/: .s/k!

P ROOF. Note that for jzj < 1 we have the binomial expansion .1

z/

s

D

1 X kD0

.s C k/ k z : .s/k!

So for Re s > 1 we compute H .s; a/ D

1 1 1 X X 1 1 1 1 1 X C D C . 1/k as ns 1 C na s as ns nD1 nD1 kD0 1

X 1 D sC . 1/k a D

1 C as

kD0 1 X kD0

. 1/k

.s C k/ a k .s/k! n

1 .s C k/ k X 1 a .s/k! nsCk nD1

.s C k/ k a R .s C k/: .s/k!

BASIC ZETA FUNCTIONS AND SOME APPLICATIONS IN PHYSICS

103

From here it is seen that s D 1 is the only pole of H .s; a/ with Res H .1; a/ D 1. In determining certain function values of H .s; a/ the following polynomials will turn out to be useful. D EFINITION 2.3. For x 2 C we define the Bernoulli polynomials Bn .x/ by the equation 1 X ze xz Bn .x/ n D z ; where jzj < 2: (2-2) ez 1 n! nD0

Examples are B0 .x/ D 1 and B1 .x/ D x 12 . The numbers Bn .0/ are called Bernoulli numbers and are denoted by Bn . Thus z ez

1

D

1 X Bn n z ; n!

where jzj < 2:

nD0

L EMMA 2.4. The Bernoulli polynomials satisfy P (1) Bn .x/ D nkD0 kn Bk x n (2) (3)

(2-3)

Bn .x C 1/ .

Bn .x/ D nx n

1/n Bn . Bn .1

(4)

x/ D Bn

k;

1

if n 1;

.x/ C nx n 1 ;

x/ D . 1/n Bn .x/:

E XERCISE 1. Use relations (2-2) and (2-3) to show assertions (1)–(4). We now establish elementary properties of H .s; a/. T HEOREM 2.5. For Re s > 1 we have Z 1 1 H .s; a/ D ts .s/ 0

e

1

1

at

e

dt:

t

(2-4)

Furthermore, for k 2 N0 we have BkC1 .a/ : k C1 P ROOF. We use the definition of the gamma function and have Z 1 Z 1 s 1 u s .s/ D u e du D t s 1 e t dt: H . k; a/ D

0

This shows the first part of the theorem, Z 1 1 X 1 H .s; a/ D t s 1 e t .nCa/ dt D .s/ 0 nD0 Z 1 1 e at D ts 1 dt: .s/ 0 1 e t

(2-5)

0

1 .s/

Z

1

t 0

s 1

1 X nD0

e

t .nCa/

dt

104

KLAUS KIRSTEN

Furthermore we have Z 1 1 t e ta ts 2 dt .s/ 0 1 e t Z 1 1 . t/e t a ts 2 dt C .s/ 0 e t 1

H .s; a/ D D

1 .s/

Z

1

ts

1

2.

t/e

ta

t

1

e

dt:

The integral in the second term is an entire function of s. Given that the gamma function has singularities at s D k, k 2 N0 , only the first term can possibly contribute to the properties H . k; a/ considered. We continue and write 1 .s/

Z

1

t 0

s 2.

e

t/e

ta

t

1

dt D D

1 .s/ 1 .s/

Z

1

t

s 2

0 1 X nD0

1 X Bn .a/ . t/n dt n!

nD0

Bn .a/ . 1/n ; n! s C n 1

which provides the analytical continuation of the integral to the complex plane. From here we observe again Res H .1; a/ D B0 .a/ D 1: Furthermore the second part of the theorem follows: H . k; a/ D lim

"!0

1 BkC1 .a/ . 1/kC1 . k C "/ .k C 1/! "

D lim . 1/k k!" "!0

BkC1 .a/ . 1/kC1 D .k C 1/! "

BkC1 .a/ : k C1

The disadvantage of the representation (2-4) is that it is valid only for Re s > 1. This can be improved by using a complex contour integral representation. The starting point is the following representation for the gamma function [46]. L EMMA 2.6. For z … Z we have .z/ D

1 2i sin.z/

Z C

. t/z

1

e t dt;

where the anticlockwise contour C consists of a circle C3 of radius " < 2 and straight lines C1 , respectively C2 , just above, respectively just below, the x-axis; see Figure 1. P ROOF. Assume Re z > 1. As the integrand remains bounded along C3 , no contributions will result as " ! 0. Along C1 and C2 we parametrize as given in


105

t-plane

6

i u t D e

C3 n

C1C2

-

t D e i u

Figure 1. Contour in Lemma 2.6.

Figure 1 and thus for Re z > 1 Z 0 Z 1 Z . t/z 1 e t dt D e i.z 1/ uz 1 e u du C e i.z lim "!0 C 1 0 Z 1 D uz 1e u e iz e iz du 0 Z 1 D 2i sin.z/ uz 1 e u du;

1/ z 1

u

e

u

du

0

which implies the assertion by analytical continuation.

This representation for the gamma function can be used to show the following result for the Hurwitz zeta function. T HEOREM 2.7. For s 2 C, s … N, we have Z .1 s/ . t/s 1 e H .s; a/ D 2 i 1 e t C

ta

dt;

with the contour C given in Figure 1. P ROOF. We follow the previous calculation to note Z Z 1 . t/s 1 e t a dt D 2i sin.s/ ts t 1 e 0 C and we use [46] sin.s/ .s/ D to conclude the assertion.

e

1

1

ta

e

t

dt;

.1 s/

106

KLAUS KIRSTEN

From here, properties previously given can be easily derived. For s 2 Z the integrand does not have a branch cut and the integral can easily be evaluated using the residue theorem. The only possible singularity enclosed is at t D 0 and to read off the residue we use the expansion s 2.

. t/

t/e

ta

t

1

e

s 2

D . t/

1 X Bn .a/ . t/n : n!

nD0

2.2. Barnes zeta function. The Barnes zeta function is a multidimensional generalization of the Hurwitz zeta function. D EFINITION 2.8. Let s 2 C with Re s > d and c 2 R C , rE 2 R dC . The Barnes zeta function is defined as X 1 B .s; cjEr / D : (2-6) .c C m E rE/s d m2 E N0

E If c D 0 it is understood that the summation ranges over m E ¤ 0. For rE D 1Ed WD .1; 1; : : : ; 1; 1/, the Barnes zeta function can be expanded in terms of the Hurwitz zeta function. E XAMPLE 2.9. Consider d D 2 and rE D .1; 1/. Then B .s; cj1E2 / D

1 1 X X 1 k C1 k Cc C1 c D D .c C m1 C m2 /s .c C k/s .c C k/s 2

X

kD0

m2 E N0

D H .s

1; c/ C .1

kD0

c/H .s; c/:

.d /

E XAMPLE 2.10. Let ek be the number of possibilities to write an integer k as a sum over d non-negative integers. We then can write B .s; cj1Ed / D

X m2 E Nd0

.d/

The coefficient ek

1 X 1 1 .d / D ek : s .c C m1 C C md / .c C k/s kD0

can be determined for example as follows. Consider

1 1 1 D D d 1 x 1 x .1 x/ D

1 X l1 D0

1 X ld D0

x

X 1

x

l1

X 1

l1 D0

l1 CCld

D

ld D0 1 X kD0

.d/

ek x k :

x

ld


107

On the other side, using the binomial expansion 1 X 1 D .1 x/d

1 1 .d C k/ k X .d C k 1/! k X d C k 1 k x D x D x : .d/k! .d 1/!k! d 1

kD0

kD0

kD0

This shows B .s; cj1Ed / D

1 X d Ck kD0

d

1 1

1 ; .c C k/s

which, once the dimension d is specified, allows to write the Barnes zeta function as a sum of Hurwitz zeta functions along the lines in Example 2.9. It is possible to obtain similar formulas for ri rational numbers [27; 28]. For some properties of the Barnes zeta function the use of complex contour integral representations turns out to be the best strategy. T HEOREM 2.11. We have the following representations: Z 1 1 e ct B .s; cjEr / D dt t s 1 Qd rj t / .s/ 0 .1 e j D1 Z .1 s/ e ct D . t/s 1 Qd 2 i C j D1 .1 e

rj t /

dt;

with the contour C given in Figure 1. E XERCISE 2. Use equation (2-5), and again Lemma 2.6, to prove Theorem 2.11. The residues of the Barnes zeta function and its values at non-positive integers are best described using generalized Bernoulli polynomials [70]. .d/

D EFINITION 2.12. We define the generalized Bernoulli polynomials Bn .xjEr / by the equation xt

e Qd

j D1 .1

1 . 1/d X . t/n D Qd n! e rj t / j D1 rj nD0

d

Bn.d/ .xjEr /:

Using Definition 2.12 in Theorem 2.11 one immediately obtains the following properties of the Barnes zeta function. T HEOREM 2.13. (1) Res B .z; cjEr / D

(2)

.z

. 1/dCz .d / Bd z .cjEr /; Qd 1/!.d z/! j D1 rj

B . n; cjEr / D

z D 1; 2; : : : ; d;

. 1/d n! .d/ Bd Cn .cjEr /: Qd .d C n/! j D1 rj

108

KLAUS KIRSTEN

E XERCISE 3. Use the first representation of B .s; cjEr / in Theorem 2.11 together with Definition 2.12 to show Theorem 2.13. Follow the steps of the proof in Theorem 2.5. E XERCISE 4. Use the second representation of B .s; cjEr / in Theorem 2.11 together with Definition 2.12 and the residue theorem to show Theorem 2.13. 2.3. Epstein zeta function. We now consider zeta functions associated with sums of squares of integers [38; 39]. D EFINITION 2.14. Let s 2 C with Re s > d=2 and c 2 R C , rE 2 R dC : The Epstein zeta function is defined as X 1 : E .s; cjEr / D 2 .c C r1 m1 C r2 m22 C C rd m2d /s d m2 E Z

E If c D 0 it is understood that the summation ranges over m E ¤ 0. L EMMA 2.15. For Re s > d=2, we have Z 1 X 1 E .s; cjEr / D ts 1 e .s/ 0 d

t .r1 m21 CCrd m2d Cc/

dt:

m2 E Z

P ROOF. This follows as before from property (2-5) of the gamma function. As we have noted in the proof of Theorem 2.5, it is the small-t behavior of the integrand that determines residues of the zeta function and special function values. The way the integrand is written in Lemma 2.15 this t ! 0 behavior is not easily read off. A suitable representation is obtained by using the Poisson resummation [53]. L EMMA 2.16. Let r 2 C with Re r > 0 and t 2 R C , then r 1 1 X 2 2 X t rl2 e e rt l : D tr lD 1

lD 1

E XERCISE 5. If F.x/ is continuous such that Z 1 jF.x/jdx < 1; 1

then we define its Fourier transform by Z 1 O F .u/ D F.x/e

2 ixu

1

If

Z

1 1

jFO .u/j du < 1;

dx:


109

then we have the Fourier inversion formula Z 1 F.x/ D FO .u/ e 2 ixu du: 1

Show the following Theorem: Let F 2 L1 .R /. Suppose that the series X F.n C v/ n2Z

converges absolutely and uniformly in v, and that X jFO .m/j < 1: m2Z

Then X

F.n C v/ D

n2Z

X

FO .n/e 2 i nv :

n2Z

Hint: Note that G.v/ D

X

F.n C v/

n2Z

is a function of v of period 1. E XERCISE 6. Apply Exercise 5 with a suitable function F.x/ to show the Poisson resummation formula Lemma 2.16. In Lemma 2.16 it is clearly seen that the only term on the right hand side that is not exponentially damped as t ! 0 comes from the l D 0 term. Using the resummation formula for all d sums in Lemma 2.15, after resumming the m E D 0E term contributes Z 1 d=2 1 E 0 e ct dt t s 1 d=2 p E .s; cjEr / D .s/ 0 t r1 rd Z 1 s d2 d=2 d=2 s d=2 1 ct D p t e dt D p : r1 rd .s/ 0 r1 rd .s/c s d=2 All other contributions after resummation are exponentially damped as t ! 0 and can be given in terms of modified Bessel functions [46]. D EFINITION 2.17. Let Re z 2 > 0. We define the modified Bessel function K .z/ by Z 1 z 1 t z 2 1 4t t K .z/ D e dt: 2 2 0 Performing the resummation in Lemma 2.15 according to Lemma 2.16, with Definition 2.17 one obtains the following representation of the Epstein zeta function valid in the whole complex plane [34; 78].

110

KLAUS KIRSTEN

T HEOREM 2.18. We have d 2s s d2 d s 2 s c 4 d=2 c2 C E .s; cjEr / D p p r1 rd .s/ .s/ r1 rd d 1 1 X n2 n2d 2 .s 2 / n2d 2 p n21 1 CC CC K d s 2 c : 2 r1 rd r1 rd E n E 2Z d =f0g

E XERCISE 7. Show Theorem 2.18 along the lines indicated. From Definition 2.17 it is clear that the Bessel function is exponentially damped for large Re z 2 . As a result the representation above is numerically very effective as long as the argument of Kd=2 s is large. The terms involving the Bessel functions are analytic for all values of s, the first term contains poles. As an immediate consequence of the properties of the gamma function one can show the following properties of the Epstein zeta function. T HEOREM 2.19. For d even, E .s; cjEr / has poles at s D d2 ; d2 1; : : : ; 1, whereas for d odd they are located at s D d2 ; d2 1; : : : ; 12 ; 2lC1 2 , l 2 N0 . Furthermore, j

d

Res E .j ; cjEr / D p

E . p; cjEr / D

d

. 1/ 2 Cj 2 c 2 r1 rd

8 ˆ 0 ˆ
12 :

nD1

So here the associated zeta function is a multiple of the zeta function of Riemann, P .s/ D

1 X n nD1

2s

L

D

2s L R .2s/:

E XAMPLE 3.2. The previous example can be easily generalized to higher dimensions. We consider explicitly two dimensions; for the higher dimensional situation see [1]. Let M D f.x; y/jx 2 Œ0; L1 ; y 2 Œ0; L2 g: We consider the boundary value problem with Dirichlet boundary conditions on M , that is @2 @2 P n;m .x; y/ D C c n;m .x; y/ D n;m n;m .x; y/; @x 2 @y 2 n;m .0; y/ D n;m .L1 ; y/ D n;m .x; 0/ D n;m .x; L2 / D 0: Using the process of separation of variables, eigenfunctions are seen to be nx my n;m .x; y/ D A sin sin ; L1 L2 with the eigenvalues n;m D

n L1

2

m C L2

2 C c;

n; m 2 N:

112

KLAUS KIRSTEN

The associated zeta function therefore is s 1 X 1 X n 2 m 2 P .s/ D C Cc ; L1 L2 nD1 mD1

which can be expressed in terms of the Epstein zeta function given in Definition 2.14 as follows: ˇ 2 2 ˇ 1 ; P .s/ D 4 E s; c ˇ L1 L2 ˇ 2 ˇ 2 ˇ ˇ 1 1 C 14 c s : (3-1) 4 E s; c ˇ L 4 E s; c ˇ L 1 2 E XAMPLE 3.3. Similarly one can consider periodic boundary conditions instead of Dirichlet boundary conditions, this means the manifold M is given by M D S 1 S 1 . In this case the eigenfunctions have to satisfy n;m .0; y/ D n;m .L1 ; y/; n;m .x; 0/ D n;m .x; L2 /;

@ @ n;m .0; y/ D n;m .L1 ; y/; @x @x @ @ n;m .x; 0/ D n;m .x; L2 /: @y @y

This shows that n;m .x; y/ D Ae i2 nx=L1 e i2 my=L2 ; which implies for the eigenvalues n;m D

2 n L1

2

C

2 m L2

2

C c;

.n; m/ 2 Z 2 :

The associated zeta function therefore is 2 2 2 2 P .s/ D E s; cjEr ; rE D ; : L1 L2 Clearly, in d dimensions one finds P .s/ D E s; cjEr ;

rE D

2 2 L1

2 2 ;:::; : Ld

E XAMPLE 3.4. As a final example we consider the Schrödinger equation of atoms in a harmonic oscillator potential. In this case M D R 3 , and the eigenvalue equation reads


113

„2 m 2 2 2 C !1 x C !2 y C !3 z n1 ;n2 ;n3 .x; y; z/ 2m 2 D n1 ;n2 ;n3 n1 ;n2 ;n3 .x; y; z/:

This differential equation is augmented by the condition that eigenfunctions must be square integrable, n1 ;n2 ;n3 .x; y; z/ 2 L2 .R 3 /: As is well known, this gives the eigenvalues n1 ;n2 ;n3 D „!1 n1 C 12 C „!2 n2 C 12 C „!3 n3 C 21 ; for .n1 ; n2 ; n3 / 2 N30 . This clearly leads to the Barnes zeta function P .s/ D B .s; cjEr /; where c D

1 2 „.!1 C !2 C !3 /;

rE D „ .!1 ; !2 ; !3 / :

If M D R is chosen the Hurwitz zeta function results. The examples above illustrate how the zeta functions considered in Section 2 are all related in a natural way to eigenvalues of specific boundary value problems. In fact, zeta functions in a much more general context are studied in great detail. For our purposes the relevant setting is the setting of Laplace-type operators on a Riemannian manifold M , possibly with a boundary @M . Laplace-type means the operator P can be written as P D g j k rjV rkV

E;

where g j k is the metric of M , r V is the connection on M acting on a smooth vector bundle V over M , and where E is an endomorphism of V . Imposing suitable boundary conditions, eigenvalues n and eigenfunctions n do exist, Pn .x/ D n n .x/; and assuming n > 0 the zeta function is defined to be P .s/ D

1 X

n s

nD1

for Re s sufficiently large. If there are modes with n D 0 those have to be excluded from the sum. Also, if finitely many eigenvalues are negative the zeta function can be defined by choosing nonstandard definitions of the principal value for the argument of complex numbers, but we will not need to consider those cases.

114

KLAUS KIRSTEN

4. Some motivations to consider zeta functions There are many situations where properties of zeta functions in the above context of Laplace-type operators are needed. In the following we present a few of them, but many more can be found for example in the context of number theory [3; 4; 23; 79] and quantum field theory [8; 14; 15; 16; 26; 30; 31; 33; 41; 42; 57; 74]. 4.1. Can one hear the shape of a drum? Let M be a two-dimensional membrane representing a drum with boundary @M . The drum is fixed along its boundary. Then possible vibrations of the drum and its fundamental tones are described by the eigenvalue problem 2 @2 @ C n .x; y/ D n n .x; y/; n .x; y/j.x;y/2@M D 0: @x 2 @y 2 Here, .x; y/ denotes the variables in the plane, the eigenfunctions n .x; y/ describe the amplitude of the vibrations and n its fundamental tones. In 1966 Kac [56] asked if just by listening with a perfect ear, so by knowing all the fundamental tones n , it is possible to hear the shape of the drum. One problem in answering this question is, of course, that in general it will be impossible to write down the eigenvalues n in a closed form and to read off relations with the shape of the drum directly. Instead one has to organize the spectrum intelligently in form of a spectral function to reveal relationships between the eigenvalues and the shape of the drum. In this context a particularly fruitful spectral function is the heat kernel 1 X K.t/ D e n t ; nD1

which as t tends to zero clearly diverges. Given that some relations between the fundamental tones and properties of the drum are hidden in the t ! 0 behavior let us consider this asymptotic behavior very closely. Before we come back to the setting of the drum, let us use a few examples to get an idea what the structure of the t ! 0 behavior of the heat kernel is expected to be. E XAMPLE 4.1. Let M D S 1 be the circle with circumference L and let P D @2 =@x 2 . Imposing periodic boundary conditions eigenvalues are 2k 2 k D ; k 2 Z; L and the heat kernel reads KS 1 .t/ D

1 X kD 1

e

.2k=L/2 t

:


115

From Lemma 2.16 we find the t ! 0 behavior KS 1 .t/ D p

1 4 t

L C .exponentially damped terms/ :

With the obvious notation this could be written as 1 vol M C .exponentially damped terms/ : KS 1 .t/ D p 4 t E XAMPLE 4.2. The heat kernel for the d-dimensional torus M D S 1 S 1 with P D clearly gives a product of the above and thus KM .t/ D KS 1 .t/ KS 1 .t/ D

1 vol M C e.d.t. .4 t/d=2

E XAMPLE 4.3. To avoid the impression that there is always just one term that is not exponentially damped consider M as above but P D C m2 . Then 1 m2 t m2 t K.t/ D e KM .t/ D e vol M C e.d.t. .4 t/d=2 1 X 1 . 1/` 2` ` d D vol M m t 2 C e.d.t. `! .4/d=2 `D0

In fact, the structure of the heat kernel observed in this last example is the structure observed for the general class of Laplace-type operators. T HEOREM 4.4. Let M be a d-dimensional smooth compact Riemannian manifold without boundary and let P D g j k rjV rkV

E;

where g j k is the metric of M , r V is the connection on M acting on a smooth vector bundle V over M , and where E is an endomorphism of V . Then as t ! 0, 1 X K.t/ ak t k d=2 kD0

with the so-called heat kernel coefficients ak .

P ROOF. See, e.g., [44]. In Example 4.3 one sees that ak D

1 . 1/k 2k m vol M: .4/d=2 k!

116

KLAUS KIRSTEN

In general, the heat kernel coefficients are significantly more complicated and they depend upon the geometry of the manifold M and the endomorphism E [44]. Up to this point we have only considered manifolds without boundary. In order to consider in more detail questions relating to the drum, let us now see what relevant changes in the structure of the small-t heat kernel expansion occur if boundaries are present. E XAMPLE 4.5. Let M D Œ0; L and P D @2 =@x 2 with Dirichlet boundary conditions imposed. Normalized eigenfunctions are then given by r 2 `x '` .x/ D sin L L and the associated eigenvalues are 2 ` ; ` 2 N: ` D L Using Lemma 2.16 this time we obtain 1 K.t/ D p vol M 12 C .exponentially damped terms/: (4-1) 4 t Notice that in contrast to previous results we have integer and half-integer powers in t occurring. E XERCISE 9. There is a more general version of the Poisson resummation formula than the one given in Lemma 2.16, namely r 1 1 X 2 2 X t .`Cc/2 (4-2) e e t ` 2 i`c : D t `D 1

`D 1

Apply Exercise 5 with a suitable function F.x/ to show equation (4-2). E XERCISE 10. Consider the setting described in Example 4.5. The local heat kernel is defined as the solution of the equation @ @2 K.t; x; y/ D 0 @t @x 2 with the initial condition lim K.t; x; y/ D ı.x; y/:

t !0

In terms of the quantities introduced in Example 4.5 it can be written as K.t; x; y/ D

1 X `D1

'` .x/'` .y/e

` t

:


117

Use the resummation (4-2) for K.t; x; y/ and the fact that Z L K.t/ D K.t; x; x/dx 0

to rediscover the result (4-1). E XERCISE 11. Let M D Œ0; L and PD

@2 C m2 @x 2

with Dirichlet boundary conditions imposed. Find the small-t asymptotics of the heat kernel. E XERCISE 12. Let M D Œ0; L S 1 S 1 be a d-dimensional manifold and @2 PD C m2 : @x 2 Impose Dirichlet boundary conditions on Œ0; L and periodic boundary conditions on the circle factors. Find the small-t asymptotics of the heat kernel. As the examples and exercises above suggest, one has the following result. T HEOREM 4.6. Let M be a d-dimensional smooth compact Riemannian manifold with smooth boundary and let P D g j k rjV rkV

E;

where g j k is the metric of M , r V is the connection on M acting on a smooth vector bundle V over M , and where E is an endomorphism of V . We impose Dirichlet boundary conditions. Then as t ! 0, K.t/

1 X

ak t k

d=2

1 kD0; 2 ;1;:::

with the heat kernel coefficients ak . P ROOF. See, e.g., [44].

As for the manifold without boundary case, Theorem 4.4, the heat kernel coefficients depend upon the geometry of the manifold M and the endomorphism E, and in addition on the geometry of the boundary. Note, however, that in contrast to Theorem 4.4 the small-t expansion contains integer and half-integer powers in t.

118

KLAUS KIRSTEN

The same structure of the small-t asymptotics is found for other boundary conditions like Neumann or Robin, see [44], and the coefficients then also depend on the boundary condition chosen. In particular, for Dirichlet boundary conditions one can show the identities (4-3) a0 D .4/ d=2 vol M; a1=2 D .4/ .d 1/=2 41 vol @M; a result going back to McKean and Singer [66]. In the context of the drum, what the formula shows is that by listening with a perfect ear one can indeed hear certain properties like the area of the drum and the circumference of its boundary. But as has been shown by Gordon, Webb and Wolpert [45], one cannot hear all details of the shape. E XERCISE 13. Use Exercise 12 to verify the general formulas (4-3) for the heat kernel coefficients. Instead of using the heat kernel coefficients to make the preceding statements, one could equally well have used zeta function properties for equivalent statements. Consider the setting of Theorem 4.6. The associated zeta function is P .s/ D

1 X

n s ;

nD1

where it follows from Weyl’s law [80; 81] that this series is convergent for Re s > d=2. The zeta function is related with the heat kernel by Z 1 1 P .s/ D t s 1 K.t/dt; (4-4) .s/ 0 where equation (2-5) has been used. This equation allows us to relate residues and function values at certain points with the small-t behavior of the heat kernel. In detail, a.d=2/ z d d 1 1 ; zD ; ;:::; ; 2 2 2 .z/ q P . q/ D . 1/ q!a d Cq ; q 2 N0 :

Res P .z/ D

2

2nC1 ; n 2 N0 ; (4-5) 2 (4-6)

Keeping in mind the vanishing of the heat kernel coefficients ak with half-integer index for @M D ∅, see Theorem 4.4, this means for d even the poles are actually located only at z D d=2; d=2 1; : : : ; 1. In addition, for d odd we get P . q/ D 0 for q 2 N0 . E XERCISE 14. Use Theorem 4.6 and proceed along the lines indicated in the proof of Theorem 2.5 to show equations (4-5) and (4-6).


119

Going back to the setting of the drum properties of the zeta function relate with the geometry of the surface. In particular, from (4-3) and (4-5) one can show the identities Res P .1/ D

vol M ; 4

Res P

1 2

D

vol @M ; 2

and the remarks below equation (4-3) could be repeated. 4.2. What does the Casimir effect know about a boundary? We next consider an application in the context of quantum field theory in finite systems. The importance of this topic lies in the fact that in recent years, progress in many fields has been triggered by the continuing miniaturization of all kinds of technical devices. As the separation between components of various systems tends towards the nanometer range, there is a growing need to understand every possible detail of quantum effects due to the small sizes involved. Very generally speaking, effects resulting from the finite extension of systems and from their precise form are known as the Casimir effect. In modern technical devices this effect is responsible for up to 10% of the forces encountered in microelectromechanical systems [19; 20]. Casimir forces are of direct practical relevance in nanotechnology where, e.g., sticking of mobile components in micromachines might be caused by them [76]. Instead of fighting the occurrence of the effect in technological devices, the tendency is now to try and take technological advantage of the effect. Experimental progress in recent years has been impressive and for some configurations allows for a detailed comparison with theoretical predictions. The best tested situations are those of parallel plates [12] and of a plate and a sphere [20; 21; 62; 63; 69]; recently also a plate and a cylinder has been considered [13; 37]. Experimental data and theoretical predictions are in excellent agreement, see, e.g., [8; 25; 61; 64]. This interplay between theory and experiments, and the intriguing technological applications possible, are the main reasons for the heightened interest in this effect in recent years. In its original form, the effect refers to the situation of two uncharged, parallel, perfectly conducting plates. As predicted by Casimir [17], the plates should attract with a force per unit area, F.a/ 1=a4 , where a is the distance between the plates. Two decades later Boyer [10] found a repulsive pressure of magnitude F.R/ 1=R4 for a perfectly conducting spherical shell of radius R. Up to this day an intuitive understanding of the opposite signs found is lacking. One of the main questions in the context of the Casimir effect therefore is how the occurring forces depend on the geometrical properties of the system considered. Said differently, the question is “What does the Casimir effect know about a boundary?” In the absence of general answers one approach consists in accu-

120

KLAUS KIRSTEN

mulating further knowledge by adding bits of understanding based on specific calculations for specific configurations. Several examples will be provided in this section and we will see the dominant role the zeta functions introduced play. However, before we come to specific settings let us briefly introduce the zeta function regularization of the Casimir energy and force that we will use later. We will consider the Casimir effect in a quantum field theory of a noninteracting scalar field under external conditions. The action in this case is [55] Z 1 ˚.x/ . V .x// ˚.x/ dx; (4-7) SŒ˚ D 2 M describing a scalar field ˚.x/ in the background potential V .x/. We assume the Riemannian manifold M to be of the form M D S 1 Ms , where the circle S 1 of radius ˇ is used to describe finite temperature T D 1=ˇ and Ms , in general, is a d-dimensional Riemannian manifold with boundary. For the action (4-7) the corresponding field equations are .

V .x//˚.x/ D 0:

(4-8)

If Ms has a boundary @Ms , these equations of motion have to be supplemented by boundary conditions on @Ms . Along the circle, for a scalar field, periodic boundary conditions are imposed. Physical properties like the Casimir energy of the system are conveniently described by means of the path-integral functionals Z ZŒV D e S Œ˚ D˚; (4-9) where we have neglected an infinite normalization constant, and the functional integral is to be taken over all fields satisfying the boundary conditions. Formally, equation (4-9) is easily evaluated to be ŒV D ln ZŒV D 21 ln det . C V .x//=2 ; (4-10) where is an arbitrary parameter with dimension of a mass to adjust the dimension of the arguments of the logarithm. E XERCISE 15. In order to motivate equation (4-10) show that for P a positive definite Hermitian .N N /-matrix one has Z e .x;P x/=2 .dx/ D .det P / 1=2 ; Rn

where .dx/ D d n x.2/

n=2

:

For P D C V .x/ and interpreting the scalar product .x; P x/ as an L2 .M /product, one is led to (4-10) by identifying D˚ with .dx/.


121

Equation (4-10) is purely formal, because the eigenvalues n of CV .x/ grow without bound for n ! 1 and thus expression (4-10) needs further explanations. In order to motivate the basic definition let P be a Hermitian .N N /-matrix with positive eigenvalues n . Clearly ln det P D

N X

ln n D

nD1

ˇ N d X s ˇˇ n ˇ D ds sD0 nD1

ˇ ˇ d P .s/ˇˇ ; ds sD0

and the determinant of P can be expressed in terms of the zeta function associated with P . This very same definition, namely ln det P D P0 .0/

(4-11)

with P .s/ D

1 X

n s

(4-12)

nD1

is now applied to differential operators as in (4-10). Here, the series representation is valid for Re s large enough, and in (4-11) the unique analytical continuation of the series to a neighborhood about s D 0 is used. This definition was first used by the mathematicians Ray and Singer [73] to give a definition of the Reidemeister–Franz torsion. In physics, this regularization scheme took its origin in ambiguities of dimensional regularization when applied to quantum field theory in curved spacetime [29; 51]. For applications beyond the ones presented here see, e.g., [14; 15; 26; 30; 31; 41; 42; 74]. The quantity ŒV is called the effective action and the argument V indicates the dependence of the effective action on the external fields. The Casimir energy is obtained from the effective action via ED

@ ŒV D @ˇ

1 @ 0 2 .0/: 2 @ˇ P =

(4-13)

Here, we will only consider the zero temperature Casimir energy ECas D lim E ˇ!1

(4-14)

and we will next derive a suitable representation for ECas . We want to concentrate on the influence of boundary conditions and therefore we set V .x/ D 0. The relevant operator to be considered therefore is PD

@2 @ 2

s ;

122

KLAUS KIRSTEN

where 2 S 1 is the imaginary time and s is the Laplace operator on Ms . In order to analyze the zeta function associated with P we note that eigenfunctions, respectively eigenvalues, are of the form 1 2 i n=ˇ e 'j .y/; ˇ 2 n 2 D C Ej2 ; ˇ

n;j .; y/ D n;j

n 2 Z;

with s 'j .y/ D Ej2 'j .y/; where y 2 Ms . For the non-self-interacting case considered here, Ej are the oneparticle energy eigenvalues of the system. The relevant zeta function therefore has the structure P .s/ D

1 X 1 X 2 n 2

ˇ

nD 1 j D1

C Ej2

s

:

(4-15)

We repeat the analysis outlined previously, namely we use equation (2-5) and we apply Lemma 2.16 to the n-summation. In this process the zeta function Ps .s/ D

1 X

Ej

2s

j D1

and the heat kernel KPs .t/ D

1 X j D1

e

Ej2 t

1 X

ak t k

.d=2/

1 kD0; 2 ;1;:::

of the spatial section are the most natural quantities to represent the answer, 1 Z 1 X 2 1 P .s/ D t s 1 e .2 n=ˇ/ t KPs .t/ dt .s/ nD 1 0 1 Z 1 X s 12 ˇ ˇ 1 ts Dp Ps s 2 C p .s/ .s/ 4 0 nD1

3 2e

n2 ˇ 2 4t KP

s

.t/ dt:


123

For the Casimir energy we need (D D d C 1) P0 =2 .0/ D P0 .0/ C P .0/ ln 2 1 D ˇ FP Ps 2 C 2.1

ln 2/ Res Ps

1 2

1 Z n2 ˇ 2 ˇ X 1 3=2 4t t e KPs .t/ dt Cp 0 nD1 1 1 D ˇ FP Ps p aD=2 .ln 2 / C 2.1 2 4 1 Z 1 X n2 ˇ 2 ˇ 3=2 Cp t e 4t KPs .t/ dt; 0

1 .0/ ln 2 ˇ P

ln 2/

(4-16)

nD1

with the finite part FP of the zeta function and where equations (4-5) and (4-6) together with the fact that KM .t/ D KS 1 .t/ KPs .t/ have been used, in particular Res Ps 21 D

ˇ P .0/ D p aD=2 : (4-17) 4 At T D 0 we obtain for the Casimir energy, see equations (4-13) and (4-14), 1 1 (4-18) ECas D lim E D FP Ps 12 p aD=2 ln Q 2 ; 2 ˇ!1 2 4 with the scale Q D .e=2/. Equation (4-18) implies that as long as aD=2 ¤ 0 the Casimir energy contains a finite ambiguity and renormalization issues need 1 to be discussed. Note from (4-17) that whenever Ps is finite no ambiguity 2 exists because aD=2 D 0. In the specific examples chosen later we will make sure that these ambiguities are absent and therefore a discussion of renormalization will be unnecessary. In a purely formal calculation one essentially is also led to equation (4-18). As mentioned, in the quantum field theory of a free scalar field the eigenvalues of a Laplacian are the square of the energies of the quantum fluctuations. Writing the Casimir energy as (one-half) the sum over the energy of all quantum fluctuations one has 1 1 X 1=2 ECas D k ; (4-19) 2 aD=2 p ; 2

kD0

and a formal identification “shows” that ECas D 21 Ps

1 2

:

(4-20)

124

KLAUS KIRSTEN

Clearly, the expression (4-19) is purely formal as the series diverges. However, 1 when Ps 2 turns out to be finite this formal identification yields the correct result. Otherwise, the ambiguities given in (4-18) remain as discussed above. An alternative discussion leading to definition (4-18) can be found in [7]. As a first example let us consider the configuration of two parallel plates a distance a apart analyzed originally by Casimir [17]. For simplicity we concentrate on a scalar field instead of the electromagnetic field and we impose Dirichlet boundary conditions on the plates. The boundary value problem to be solved therefore is uk .x; y; z/ D k uk .x; y; z/; with uk .0; y; z/ D uk .a; y; z/ D 0. For the time being, we compactify the .y; z/-directions to a torus with perimeter length R and impose periodic boundary conditions in these directions. Later on, the limit R ! 1 is performed to recover the parallel plate configuration. Using separation of variables one obtains normalized eigenfunctions in the form r 2 `x i2`1 y=R i2`2 z=R u`1 `2 ` .x; y; z/ D e e sin a aR2 with eigenvalues 2 2`1 2 2`2 2 ` `1 `2 ` D C C ; .`1 ; `2 / 2 Z 2 ; ` 2 N: R R a This means we have to study the zeta function 1 2 s X X 2`2 2 ` 2`1 2 .s/ D C C : (4-21) R R a 2 .`1 ;`2 /2Z

`D1

As R ! 1 the Riemann sum turns into an integral and we compute using polar coordinates in the .y; z/-plane 1 Z 1 Z 1 2 X 2 s R ` .s/ D k12 C k22 C d k2 d k1 2 a 1 1 `D1 Z 1 1 2 X 2 s R ` 2 D 2 k k C dk 2 a 0 `D1 1 2 sC1 ˇ1 X ˇ 1 ` R2 2 k C D ˇ 2 2.1 s/ a 0 `D1

D

R2 4.1 s/

1 X ` 2. `D1

a

sC1/

D

2 R2 4.1 s/ a

2s

R .2s

2/:


Setting s D

1 2

125

as needed for the Casimir energy we obtain

1 2

D

R2 2 4 3

3 R . 3/ D a

R2 2 : 720a3

(4-22)

The resulting Casimir force per area is @ ECas 2 D : (4-23) @a R2 480a4 Note that this computation takes into account only those quantum fluctuations from between the plates. But in order to find the force acting on the, say, right plate the contribution from the right to this plate also has to be counted. To find this part we place another plate at the position x D L where at the end we take L ! 1. Following the preceding calculation, we simply have to replace a by L a to see that the associated zeta function produces FCas D

1 2

D

R2 2 720.L a/3

and the contribution to the force on the plate at x D a reads FCas D

2 : 480.L a/4

This shows the plate at x D a is always attracted to the closer plate. As L ! 1 it is seen that equation (4-23) also describes the total force on the plate at x D a for the parallel plate configuration. E XERCISE 16. Consider the Casimir energy that results in the previous discussion when the compactification length R is kept finite. Use Lemma 2.18 to give closed answers for the energy and the resulting force. Can the force change sign depending on a and R? More realistically plates will have a finite extension. An interesting setting that we are able to analyze with the tools provided are pistons. These have received an increasing amount of interest because they allow the unambiguous prediction of forces [18; 52; 58; 65; 77]. Instead of having parallel plates let us consider a box with side lengths L1 ; L2 and L3 . Although it is possible to find the Casimir force acting on the plate at x D L1 resulting from the interior of the box, the exterior problem has remained unsolved until today. No analytical procedure is known that allows to obtain the Casimir energy or force for the outside of the box. This problem is avoided by adding on another box with side lengths L L1 ; L2 and L3 such that the wall at x D L1 subdivides the bigger box into two chambers. The wall at x D L1 is assumed to be movable and is called the piston. Each chamber can be dealt

126

KLAUS KIRSTEN

with separately and total energies and forces are obtained by adding up the two contributions. Assuming again Dirichlet boundary conditions and starting with the left chamber, the relevant spectrum reads

`1 `2 `3 D

`1 L1

2

C

`2 L2

2

C

`3 L3

2

;

`1 ; `2 ; `3 2 N;

and the associated zeta function is X ` 2 ` 2 ` 2 s 2 3 1 C C : .s/ D L1 L2 L3

(4-24)

(4-25)

`1 ;`2 ;`3 2N

One way to proceed is to rewrite (4-25) in terms of the Epstein zeta function in Definition 2.14. E XERCISE 17. Use Lemma 2.18 in order to find the Casimir energy for the inside of the box with side lengths L1 ; L2 and L3 and with Dirichlet boundary conditions imposed. Instead of using Lemma 2.18 we proceed as follows. We write first 1 1 X .s/ D 2

1 X `1 2 `2 2 `3 2 C C L1 L2 L3

s

`1 D 1 `2 ;`3 D1

1 2

1 s X `3 2 `2 2 C : L2 L3

(4-26)

`2 ;`3 D1

This shows that it is convenient to introduce 1 s X `3 2 `2 2 C .s/ D C : L2 L3

(4-27)

`2 ;`3 D1

We note that this could be expressed in terms of the Epstein zeta function given in Definition 2.14. However, it will turn out that this is unnecessary. Also, to simplify the notation let us introduce 2`2 `3 D

`2 L2

2

C

`3 L3

2

:

Using equation (2-5) for the first line in (4-26) we continue 1 X 1 .s/ D 2 .s/

Z 1 X

`1 D 1 `2 ;`3 D1

0

1

t

s 1

`1 2 2 C`2 `3 dt exp t L1

1 2 C .s/:


127

We now apply the Poisson resummation in Lemma 2.16 to the `1 -summation and therefore we get 1 X L1 .s/ D p 2 .s/

Z 1 X

`1 D 1 `2 ;`3 D1

1

ts

3=2

0

L21 `21 exp t

t2`2 `3 dt 1 2 C .s/:

(4-28)

The `1 D 0 term gives a C -term, the `1 ¤ 0 terms are rewritten using (2.17). The outcome reads 1 L1 s 21 .s/ D p C s 12 2 C .s/ 2 .s/ sC

1

2L 2 Cp 1 .s/

1 X

`21 2`

2 `3

`1 ;`2 ;`3 D1

1 .s 2

1 2/

K1

2 s

.2L1 `1 `2 `3 /: (4-29)

We need the zeta function about s D 21 in order to find the Casimir energy and Casimir force. Let s D 21 C ". In order to expand equation (4-29) about " D 0 we need to know the pole structure of C .s/. From equation (2.18) it is expected that C .s/ has at most a first order pole at s D 21 and that it is analytic about s D 1. So for now let us simply assume the structure 1 C " D Res C 12 C FP C 21 C O."/; " C . 1 C "/ D C . 1/ C "C0 . 1/ C O."2 /;

C

1 2

where Res C 12 and FP C assumed, we find

1 2

1 2

will be determined later. With this structure

1 L1 1 C" D C . 1/ 12 Res C 2 " 4 1 L1 0 C C . 1/ C C . 1/.ln 4 1/ 2 FP C 4 ˇ ˇ 1 X ˇ `2 `3 ˇ 1 ˇ ˇ K1 2L1 `1 ` ` : 2 3 ˇ ˇ ` `1 ;`2 ;`3 D1

1 2

(4-30)

1

This shows that the Casimir energy for this setting is unambiguously defined 1 only if C . 1/ D 0 and Res C 2 D 0.

128

KLAUS KIRSTEN

E XERCISE 18. Show the following analytical continuation for C .s/: C .s/ D

2s 1 L2 s 12 L3 1 L3 2s R .2s/ C p R .2s 1/ (4-31) 2 2 .s/ sC1=2 X 1 X 1 2L 2L2 `2 `3 `2 L3 s 1=2 K1 s : Cp 2 2 `3 L3 .s/ `2 D1 `3 D1 1 2

Read off that C . 1/ D Res C

D 0.

Using the results from Exercise 18 the Casimir energy, from equation (4-30), can be expressed as L1 0 ECas D . 1/ 8 C

1 4 FP C

1 2

1 2

1 X `1 ;`2 ;`3 D1

ˇ ˇ ˇ `2 `3 ˇ ˇ ˇ ˇ ` ˇ K1 .2L1 `1 `2 `3 /: 1 (4-32)

E XERCISE 19. Use representation (4-31) to give an explicit representation of the Casimir energy (4-32). For the force this shows FCas D

1 0 1 C . 1/ C 8 2

1 X `1 ;`2 ;`3 D1

ˇ ˇ ˇ `2 `3 ˇ @ ˇ ˇ ˇ ` ˇ @L K1 .2L1 `1 `2 `3 /: (4-33) 1 1

E XERCISE 20. Use Definition 2.17 to show that K .x/ is a monotonically decreasing function for x 2 R C . E XERCISE 21. Determine the sign of C0 . 1/. What is the sign of the Casimir force as L1 ! 1? What about L1 ! 0? Remember that the results given describe the contributions from the interior of the box only. The contributions from the right chamber are obtained by replacing L1 with L L1 . This shows for the right chamber ECas D

L

L1 0 C . 1/ 8

1 4 FP C

1 0 1 C . 1/ C 8 2

1 X

ˇ ˇ ˇ `2 `3 ˇ ˇ ˇ ˇ ` ˇ K1 .2.L 1 `1 ;`2 ;`3 D1 1 X ˇˇ ` ` ˇˇ @ ˇ 2 3ˇ ˇ ` ˇ @L K1 .2.L 1 1

1 2 FCas D

1 2

`1 ;`2 ;`3 D1

L1 /`1 `2 `3 /; L1 /`1 `2 `3 /:


129

Adding up, the total force on the piston is ˇ ˇ 1 X ˇ `2 `3 ˇ @ 1 tot ˇ ˇ FCas D ˇ ` ˇ @L K1 .2L1 `1 `2 `3 / 2 1 1 `1 ;`2 ;`3 D1 ˇ ˇ 1 X ˇ `2 `3 ˇ @ 1 ˇ ˇ C ˇ ` ˇ @L K1 .2.L L1 /`1 `2 `3 /: (4-34) 2 1 1 `1 ;`2 ;`3 D1

This shows, using the results of Exercise 20, that the piston is always attracted to the closer wall. Although we have presented the analysis for a piston with rectangular crosssection, our result in fact holds in much greater generality. The fact that we analyzed a rectangular cross-section manifests itself in the spectrum (4-24), namely the part `2 2 `3 2 C L2 L3 is a direct consequence of it. If instead we had considered an arbitrary crosssection C, the relevant spectrum had the form `1 2 `1 i D C 2i ; L1 where, assuming still Dirichlet boundary conditions on the boundary of the cross-section C, 2i is determined from 2 ˇ @2 @ ˇ 2 C .y; z/ D .y; z/; .y; z/ D 0: ˇ i i i i .y;z/2@C @y 2 @z 2 Proceeding in the same way as before, replacing `2 `3 with i and introducing C .s/ as the zeta function for the cross-section, C .s/ D

1 X

i

2s

;

iD1

equation (4-28) remains valid, as well as equations (4-29) and (4-30). So also for an arbitrary cross-section the total force on the piston is described by equation (4-34) with the replacements given and the piston is attracted to the closest wall. E XERCISE 22. In going from equation (4-28) to (4-29) we used the fact that 2` ` > 0. Above we used 2i > 0 which is true because we imposed Dirichlet 2 3 boundary conditions. Modify the calculation if boundary conditions are chosen (like Neumann boundary conditions) that allow for d0 zero modes 2i D 0 [58].

130

KLAUS KIRSTEN

We have presented the piston set-up for three spatial dimensions, but a similar analysis can be performed in the presence of extra dimensions [58]. Once this kind of calculation is fully understood for the electromagnetic field it is hoped that future high-precision measurements of Casimir forces for simple configurations such as parallel plates can serve as a window into properties of the dimensions of the universe that are somewhat hidden from direct observations. As we have seen for the example of the piston, there are cases where an unambiguous prediction of Casimir forces is possible. Of course the set-up we have chosen was relatively simple and for many other configurations even the sign of Casimir forces is unknown. This is a very active field of research; some references are [8; 36; 43; 67; 68; 75]. Further discussion is provided in the Conclusions.

5. Bose–Einstein condensation of Bose gases in traps We now turn to applications in statistical mechanics. We have chosen to apply the techniques in a quantum mechanical system described by the Schrödinger equation „2 C V .x; y; z/ k .x; y; z/ D k k .x; y; z/; (5-1) 2m that is we consider a gas of quantum particles of mass m under the influence of the potential V .x; y; z/. Specifically, later we will consider in detail the harmonic oscillator potential V .x; y; z/ D

m .!1 x 2 C !2 y 2 C !3 z 2 / 2

briefly mentioned in Example 3.4, as well as a gas confined in a finite cavity. Thermodynamic properties of a Bose gas, which is what we shall consider in the following, are described by the (grand canonical) partition sum qD

1 X

ln 1

e

ˇ.k /

;

(5-2)

kD0

where ˇ is the inverse temperature and is the chemical potential. We assume the index k D 0 labels the unique ground state, that is, the state with smallest energy eigenvalue 0 . From this partition sum all thermodynamical properties are obtained. For example the particle number is ˇ 1 X 1 1 @q ˇˇ D ; (5-3) ND ˇ ˇ. ˇ @ T;V e k / 1 kD0


131

where the notation .@q=@jT;V / indicates that the derivative has to be taken with temperature T and volume V kept fixed. The particle number is the most important quantity for the phenomenon of Bose–Einstein condensation. Although this phenomenon was predicted more than 80 years ago [9; 32] it was only relatively recently experimentally verified [2; 11; 24]. Bose–Einstein condensation is one of the most interesting properties of a system of bosons. Namely, under certain conditions it is possible to have a phase transition at a critical value of the temperature in which all of the bosons can condense into the ground state. In order to understand at which temperature the phenomenon occurs a detailed study of N , or alternatively q, is warranted. This is the subject of this section. We first note that from the fact that the particle number in each state has to be non-negative it is clear that < 0 has to be imposed. It is seen in (5-2) that as ˇ ! 0 (high temperature limit) the behavior of q cannot be easily understood. But contour integral techniques together with the zeta function information provided makes the analysis feasible and it will allow for the determination of the critical temperature of the Bose gas. Let us start by noting that from x/ D

ln.1

1 X xn ; n

for jxj < 1;

nD1

the partition sum can be rewritten as qD

1 X 1 X 1 e n

ˇ.k /n

:

(5-4)

nD1 kD0

The ˇ ! 0 behavior is best found using the following representation of the exponential. E XERCISE 23. Given that

1 x

lim j .x C iy/j e 2 jyj jyj 2

jyj!1

and .z/ D

p

2e .z

1 2

/ log z

D

z

p 2;

x; y 2 R ;

.1 C o.1// ;

as jzj ! 1, show that Z Ci1 1 e D a t 2 i i1 valid for > 0, jarg aj < ı, 0 < ı =2. 2 a

.t/ dt;

(5-5)

132

KLAUS KIRSTEN

Before we apply this result to the partition sum (5-4) let us use a simple example to show how this formula allows us to determine asymptotic behavior of certain series in a relatively straightforward fashion. From Lemma 2.16 we know that 1 X

e

ˇ`2

`D1

1 1 X 1 2 e ˇ` 2 2 `D 1 r r X 1 1 1 D C e 2 ˇ 2 ˇ

D

2 2 ˇ `

:

(5-6)

`D1

As ˇ ! 0 it is clear that the series on the left p diverges and Lemma 2.16 shows that the leading behavior is described by a 1= ˇ term, followed by a constant term, followed by exponentially damped corrections. Let us see how we can easily find the polynomial behavior as ˇ ! 0 from (5-5). We first write 1 X

e

ˇ`2

`D1

Z Ci1 1 X 1 .ˇ`2 / D 2 i i1

t

.t/dt:

`D1

Here, > 0 is assumed by Exercise 23. However, in order to be allowed to interchange summation and integration we need to impose > 12 and find 1 X `D1

e

ˇ`2

1 D 2 i

Z

1 D 2 i

Z

Ci1

i1 Ci1

i1

ˇ

t

.t/

1 X

`

2t

dt

`D1

ˇ

t

.t/R .2t /dt:

In order to find the small-ˇ behavior, the strategy now is to shift the contour to the left. In doing so we cross over poles of the integrand generating polynomial contributions in ˇ. For this example, the right most pole is at t D 21 (pole of the zeta function of Riemann) and the next pole is at t D 0 (from the gamma function). Those are all singularities present as R . 2n/ D 0 for n 2 N. Therefore, Z Ci1 1 Q X 1 ˇ`2 1=2 0 1 1 e D ˇ C ˇ 1 .0/ C ˇ t .t/R .2t/dt R 2 2 2 i Q i1 `D1 r Z Ci1 Q 1 1 1 D C ˇ t .t/R .2t /dt; 2 ˇ 2 2 i Q i1 where Q < 0 and where contributions from the horizontal lines between Q ˙ i 1 and ˙i1 are neglected. For the remaining contour integral plus the neglected horizontal lines one can actually show that they will produce the exponentially


133

damped terms as given in (5-6). How exactly this actually happens has been described in detail in [35]. P ˇn˛ ; ˇ > 0; ˛ > 0, behaves as ˇ ! 0 by E XERCISE 24. Argue how 1 nD1 e using the procedure above. Determine the leading three terms in the expansion assuming that the contributions from the contour at infinity can be neglected. E XERCISE 25. Find the leading three terms of the small-ˇ behavior of 1 X

log.1

e

ˇn

/

nD1

assuming that the contributions from the contour at infinity can be neglected. We next apply these ideas to the partition sum (5-4). As a further warmup, for simplicity, let us first set D 0. Not specifying k for now and using .s/ D

1 X

k s

kD0

for Re s > M large enough to make this series convergent, we write Z Ci1 1 X 1 1 1 X 1 ˇk n X X 1 1 q D e D .ˇk n/ t .t/dt n n 2 i i1 nD1 kD0 nD1 kD0 X X Z Ci1 1 1 1 t t 1 t D ˇ .t/ n k dt 2 i i1 nD1 kD0 Z Ci1 1 ˇ t .t/R .t C 1/.t/dt: D 2 i i1 Here > M is needed for the algebraic manipulations to be allowed. It is clearly seen that the integrand has a double pole at t D 0. The right most pole (at M ) therefore comes from .t/, and the location of this pole determines the leading ˇ ! 0 behavior of the partition sum. For the harmonic oscillator potential, in the notation of Example 3.4, the Barnes zeta function occurs and we have Z Ci1 1 qD ˇ t .t/R .t C 1/B .t; cjEr /dt: (5-7) 2 i i1 The location of the poles and its residues are known for the Barnes zeta function, see Definition 2.12 and Theorem 2.13, in particular one has Res B .3; cjEr / D

1 2„3 ˝ 3

;

134

KLAUS KIRSTEN

where, as is common, the geometric mean of the oscillator frequencies ˝ D .!1 !2 !3 /1=3 has been used. The leading order of the partition sum therefore is qD

1 4 C O.ˇ 90 .ˇ„˝/3

2

/:

E XERCISE 26. Use Definition 2.12 and Theorem 2.13 to find the subleading order of the small-ˇ expansion of the partition sum q. E XERCISE 27. Consider the harmonic oscillator potential in d dimensions and find the leading and subleading order of the small-ˇ expansion of the partition sum q. If instead of considering a Bose gas in a trap we consider the gas in a finite threedimensional cavity M with boundary @M we have to augment the Schrödinger equation (5-1) by boundary conditions. We choose Dirichlet boundary conditions and thus the results for the heat kernel coefficients (4-3) are valid. From equation (4-5) we also conclude that the rightmost pole of .s/ is located at s D 3=2 and that a0 vol M Res 23 D I D 3 2 4 2 furthermore the next pole is located at s D 1. For this case, the leading order of the partition sum therefore is 1 R 52 vol M C O.ˇ 1 /: qD 3=2 .4ˇ/ One way to read this result is that the Bose gas does know the volume of its container because it can be found from the partition sum. This is completely analogous to the statement for the drum where we used the heat kernel instead of the partition sum. Subleading orders of the partition sum reveal more information about the cavity, see the following exercise. But as for the drums, the gas does not know all the details of the shape of the cavity because there are different cavities leading to the same eigenvalue spectrum [45]. Those cavities cannot be distinguished by the above analysis. E XERCISE 28. Consider a Bose gas in a d-dimensional cavity M with boundary @M . Use (4-3) and (4-5) to find the leading and subleading order of the smallˇ expansion of the partition sum q. What does the Bose gas know about its container, meaning what information about the container can be read of from the high-temperature behavior of the partition sum?


135

In order to examine the phenomenon of Bose–Einstein condensation we have to consider non-vanishing chemical potential. Close to the phase transition, as we will see, more and more particles have to reside in the ground state and the value of the chemical potential will be close to the smallest eigenvalue, which is the ’critical’ value for the chemical potential, c D 0 . Near the phase transition, for the expansion to be established, it will turn out advantageous to rewrite k such that the small quantity c appears, k

D k

c C c

D k

0 C c

:

Given the special role of the ground state, we separate off its contribution and write 1 X 1 X 1 q D q0 C e n

ˇn.k 0 /

e

ˇn.c /

:

nD1 kD1

Note that the k-sum starts with k D 1, which means that the ground state is not included in this summation. Employing the representation (5-5) only to the first exponential factor and proceeding as before we obtain Z Ci1 1 q D q0 C ˇ t .t/Li1Ct e ˇ.c / 0 .t/dt; (5-8) 2 i i1 with the polylogarithm 1 X x` ; Lin .x/ D `n

(5-9)

`D1

and the spectral zeta function 0 .s/ D

1 X

.k

0 / s :

kD1

In order to determine the small-ˇ behavior of expression (5-8) let us discuss the pole structure of the integrand. Given c > 0, the polylogarithm Li1Ct .e ˇ.c / / does not generate any poles. Concentrating on the harmonic oscillator, we find 1 1 1 1 1 Res 0 .3/ D ; Res 0 .2/ D 2 C C : !1 !2 !1 !3 !2 !3 2.„˝/3 2„ Note that 0 .s/ is the Barnes zeta function as given in Definition 2.8 with c D 0 where we have to exclude m E D 0E from the summation. However, clearly the residues at s D 3 and s D 2 can still be obtained from Theorem 2.13 with c ! 0 taken.

136

KLAUS KIRSTEN

Shifting the contour to the left we now find 1 ˇ.c / q D q0 C Li e 4 .ˇ„˝/3 1 1 1 1 ˇ.c / Li3 e C C C C !1 !2 !1 !3 !2 !3 2.ˇ„/2 In order to find the particle number N we need the relation for the polylogarithm @Lin .x/ 1 D Lin 1 .x/; @x x which follows from (5-9). So 1 ˇ.c / e Li N D N0 C 3 .ˇ„˝/3 1 1 1 1 ˇ.c / C Li2 e C C C !1 !2 !1 !3 !2 !3 2.ˇ„/2 E XERCISE 29. Use (5-5) and (5-9) to show Lin e x D R .n/ xR .n

1/ C

valid for n > 2. What does the subleading term look like for n D 2? As the critical temperature is approached ! c and with Exercise 29 the particle number close to the transition temperature becomes R .2/ 1 1 1 R .3/ C C C C (5-10) N D N0 C .ˇ„˝/3 2.ˇ„/2 !1 !2 !1 !3 !2 !3 The second and third terms give the number of particles in the excited levels (at high temperature close to the phase transition). The critical temperature is defined as the temperature where all excited levels are completely filled such that lowering the temperature the ground state population will start to build up. This means the defining equation for the critical temperature Tc D 1=ˇc in the approximation considered is 1 1 1 1 1 ND R .3/ C R .2/ C C :(5-11) !1 !2 !1 !3 !2 !3 .ˇc „˝/3 2.ˇc „/2 Solving for ˇc one finds Tc D T0 1

R .2/ ıN 3R .3/2=3

1=3

:

Here, T0 is the critical temperature in the bulk limit (N ! 1) 1=3 N T0 D „˝ R .3/


137

and ı D 12 ˝ 2=3

1 1 1 C C : !1 !2 !1 !3 !2 !3

Different approaches can be used to obtain the same answers [47; 48; 49; 50]. If only a few thousand particles are used in the experiment the finite-N correction is actually quite important. For example the first successful experiments on Bose–Einstein were done with rubidium [2] at frequencies p condensates 1 !1 D !2 D 240= 8 s and !3 D 240s 1 . With N D 2000 one finds Tc 31:9nKD 0:93 T0 [59], a significant correction compared to the thermodynamic limit. E XERCISE 30. Consider the Bose gas in a d-dimensional cavity. Find the particle number and the critical temperature along the lines described for the harmonic oscillator. What is the correction to the critical temperature caused by the finite size of the cavity? (For a solution to this problem see [60].)

6. Conclusions In these lectures some basic zeta functions are introduced and used to analyze the Casimir effect and Bose–Einstein condensation for particular situations. The basic zeta functions considered are the Hurwitz, the Barnes and the Epstein zeta function. Although these zeta functions differ from each other they have one property in common: they are based upon a sequence of numbers that is explicitly known and given in closed form. The analysis of these zeta functions and of the indicated applications in physics is heavily based on this explicit knowledge in that well-known summation formulas are used. In most cases, however, an explicit knowledge of the eigenvalues of, say, a Laplacian will not be available and an analysis of the associated zeta functions will be more complicated. In recent years a new class of examples where eigenvalues are defined implicitly as solutions to transcendental equations has become accessible. In some detail let us assume that eigenvalues are determined by equations of the form F` .`;n / D 0

(6-1)

with `; n suitable indices. For example when trying to find eigenvalues and eigenfunctions of the Laplacian whenever possible one resorts to separation of variables and ` and n would be suitable ’quantum numbers’ labeling eigenfunctions. To be specific consider a scalar field in a three dimensional ball of radius R with Dirichlet boundary conditions. The eigenvalues k for this situation,

138

KLAUS KIRSTEN

with k as a multiindex, are thus determined through k .x/ D k k .x/;

k .x/jjxjDR D 0:

In terms of spherical coordinates .r; ˝/, a complete set of eigenfunctions may be given in the form p l;m;n .r; ˝/ D r 1=2 JlC1=2 l;n r Yl;m .˝/; where Yl;m .˝/ are spherical surface harmonics [40], and J are Bessel functions of the first kind [46]. Eigenvalues of the Laplacian are determined as zeroes of Bessel functions. In particular, for a given angular momentum quantum number l, imposing Dirichlet boundary conditions, eigenvalues l;n are determined by p JlC1=2 l;n R D 0: (6-2) Although some properties of the zeroes of Bessel functions are well understood [46], there is no closed form for them available and we encounter the situation described by (6-1). In order to find properties of the zeta function associated with this kind of boundary value problems the idea is to use the argument principle or Cauchy’s residue theorem. For the situation of the ball one writes the zeta function in the form Z 1 X 1 @ .s/ D .2l C 1/ k 2s ln JlC1=2 .kR/ d k; (6-3) 2 i @k lD0

where the contour runs counterclockwise and must enclose all solutions of (6-2). The factor .2l C1/ represents the degeneracy for each angular momentum l and the summation is over all angular momenta. The integrand has singularities exactly at the eigenvalues and one can show that the residues are one such that the definition of the zeta function is recovered. More generally, in other coordinate systems, one would have, somewhat symbolically, Z X 1 @ .s/ D dj k 2s ln Fj .k/ d k; (6-4) 2 i @k j

the task being to construct the analytical continuation of this object. The details of the procedure will depend very much on the properties of the special function Fj that enters, but often all the information needed can be found [57]. Nevertheless, for many separable coordinate systems this program has not been performed but efforts are being made in order to obtain yet unknown precise values for the Casimir energy for various geometries.


139

Acknowledgements This work is supported by the National Science Foundation Grant PHY0757791. Part of the work was done while the author enjoyed the hospitality and partial support of the Department of Physics and Astronomy of the University of Oklahoma. Thanks go in particular to Kimball Milton and his group who made this very pleasant and exciting visit possible.

References [1] J. Ambjorn and S. Wolfram. Properties of the vacuum, 1: Mechanical and thermodynamic. Ann. Phys., 147:1–32, 1983. [2] M. H. Anderson, J. R. Ensher, M. R. Matthews, C. E. Wieman, and E. A. Cornell. Observation of Bose–Einstein condensation in a dilute atomic vapor. Science, 269:198–201, 1995. [3] T. M. Apostol. Introduction to analytic number theory. Springer Verlag, Berlin, 1976. [4] T. M. Apostol. Modular function and Dirichlet series in number theory. Springer Verlag, Berlin, 1990. [5] E. W. Barnes. On the asymptotic expansion of integral functions of multiple linear sequence. Trans. Camb. Philos. Soc., 19:426–439, 1903. [6] E. W. Barnes. On the theory of the multiple gamma function. Trans. Camb. Philos. Soc., 19:374–425, 1903. [7] S. K. Blau, M. Visser, and A. Wipf. Zeta functions and the Casimir energy. Nucl. Phys., B310:163–180, 1988. [8] M. Bordag, U. Mohideen, and V. M. Mostepanenko. New developments in the Casimir effect. Phys. Rept., 353:1–205, 2001. [9] S. N. Bose. Planck’s law and light quantum hypothesis. Z. Phys., 26:178–181, 1924. [10] T. H. Boyer. Quantum electromagnetic zero point energy of a conducting spherical shell and the Casimir model for a charged particle. Phys. Rev., 174:1764–1774, 1968. [11] C. C. Bradley, C. A. Sackett, J. J. Tollett, and R. G. Hulet. Evidence of Bose– Einstein condensation in an atomic gas with attractive interactions. Phys. Rev. Lett., 75:1687–1690, 1995. [12] G. Bressi, G. Carugno, R. Onofrio, and G. Ruoso. Measurement of the Casimir force between parallel metallic surfaces. Phys. Rev. Lett., 88:041804, 2002. [13] M. Brown-Hayes, D. A. R. Dalvit, F. D. Mazzitelli, W. J. Kim, and R. Onofrio. Towards a precision measurement of the Casimir force in a cylinder-plane geometry. Phys. Rev., A72:052102, 2005. [14] I. L. Buchbinder, S. D. Odintsov, and I. L. Shapiro. Effective action in quantum gravity. Hilger, Bristol, 1992.

140

KLAUS KIRSTEN

[15] A. A. Bytsenko, G. Cognola, E. Elizalde, V. Moretti, and S. Zerbini. Analytic aspects of quantum fields. World Scientific, London, 2003. [16] A. A. Bytsenko, G. Cognola, L. Vanzo, and S. Zerbini. Quantum fields and extended objects in space-times with constant curvature spatial section. Phys. Rept., 266:1–126, 1996. [17] H. B. G. Casimir. On the attraction between two perfectly conducting plates. Kon. Ned. Akad. Wetensch. Proc., 51:793–795, 1948. [18] R. M. Cavalcanti. Casimir force on a piston. Phys. Rev., D69:065015, 2004. [19] H. B. Chan, V. A. Aksyuk, R. N. Kleiman, D. J. Bishop, and F. Capasso. Nonlinear micromechanical Casimir oscillator. Phys. Rev. Lett., 87:211801, 2001. [20] H. B. Chan, V. A. Aksyuk, R. N. Kleiman, D. J. Bishop, and F. Capasso. Quantum mechanical actuation of microelectromechanical systems by the Casimir force. Science, 291:1941–1944, 2001. [21] F. Chen, U. Mohidden, G. L. Klimchitskaya, and V. M. Mostepanenko. Experimental test for the conductivity properties from the Casimir force between metal and semiconductor. Phys. Rev., A74:022103, 2006. [22] F. Dalvovo, S. Giorgini, L. Pitaevskii, and S. Stringari. Theory of Bose-Einstein condensation in trapped gases. Rev. Mod. Phys., 71:463–512, 1999. [23] H. Davenport. Multiplicative number theory. Springer Verlag, Berlin, 1967. [24] K. B. Davis, M.-O. Mewes, M. R. Andrews, N. J. van Druten, D. S. Durfee, D. M. Kurn, and W. Ketterle. Bose–Einstein condensation in a gas of sodium atoms. Phys. Rev. Lett., 75:3969–3973, 1995. [25] R. S. Decca, D. López, E. Fischbach, G. L. Klimchitskaya, D. E. Krause, and V. M. Mostepanenko. Precise comparison of theory and new experiment for the Casimir force leads to stronger constraints on thermal quantum effects and long range interaction. Ann. Phys., 318:37–80, 2005. [26] A. Dettki and A. Wipf. Finite size effects from general covariance and Weyl anomaly. Nucl. Phys., B377:252–280, 1992. [27] J. S. Dowker. Effective action in spherical domains. Commun. Math. Phys., 162:633–648, 1994. [28] J. S. Dowker. Functional determinants on spheres and sectors. J. Math. Phys., 35:4989–4999, 1994. [29] J. S. Dowker and R. Critchley. Effective Lagrangian and energy momentum tensor in de Sitter space. Phys. Rev., D13:3224–3232, 1976. [30] G. V. Dunne. Functional determinants in quantum field theory. J. Phys., A41: 304006, 2008. [31] G. V. Dunne, J. Hur, C. Lee, and H. Min. Precise quark mass dependence of instanton determinant. Phys. Rev. Lett., 94:072001, 2005. [32] A. Einstein. Quantentheorie des einatomigen idealen Gases. Sitzungsberichte der Preussichen Akademie der Wissenschaften, 22:261–267, 1924.


141

[33] E. Elizalde. Ten physical applications of spectral zeta functions. Lecture Notes in Physics m35, Springer, Berlin, 1995. [34] E. Elizalde, S. D. Odintsov, A. Romeo, A. A. Bytsenko, and S. Zerbini. Zeta regularization techniques with applications. World Scientific, Singapore, 1994. [35] E. Elizalde and A. Romeo. Rigorous extension of the proof of zeta function regularization. Phys. Rev., D40:436–, 1989. [36] T. Emig, N. Graham, R. L. Jaffe, and M. Kardar. Casimir forces between arbitrary compact objects. Phys. Rev. Lett., 99:170403, 2007. [37] T. Emig, R. L. Jaffe, M. Kardar, and A. Scardicchio. Casimir interaction between a plate and a cylinder. Phys. Rev. Lett., 96:080403, 2006. [38] P. Epstein. Zur Theorie allgemeiner Zetafunctionen. Math. Ann., 56:615–644, 1903. [39] P. Epstein. Zur Theorie allgemeiner Zetafunctionen, II. Math. Ann., 63:205–216, 1907. [40] A. Erdélyi, W. Magnus, F. Oberhettinger, and F. G. Tricomi. Higher transcendental functions. Based on the notes of Harry Bateman, McGraw-Hill, New York, 1955. [41] G. Esposito. Quantum gravity, quantum cosmology and Lorentzian geometries. Lecture Notes in Physics m12, Springer, Berlin, 1994. [42] G. Esposito. Dirac operators and spectral geometry. Cambridge University Press, Cambridge, 1998. [43] S. A. Fulling, L. Kaplan, K. Kirsten, Z. H. Liu, and K. A. Milton. Vacuum stress and closed paths in rectangles, pistons, and pistols. J. Phys., A42:155402, 2009. [44] P. B. Gilkey. Invariance theory, the heat equation and the Atiyah–Singer index theorem. CRC Press, Boca Raton, 1995. [45] C. Gordon, D. Webb, and S. Wolpert. One cannot hear the shape of a drum. Bull. Amer. Math. Soc., 27:134–138, 1992. [46] I. S. Gradshteyn and I. M. Ryzhik. Table of integrals, series and products. Academic Press, New York, 1965. [47] S. Grossmann and M. Holthaus. -transition to the Bose-Einstein condensate. Z. Naturforsch., A50:921–930, 1995. [48] S. Grossmann and M. Holthaus. On Bose-Einstein condensation in harmonic traps. Phys. Lett., A208:188–192, 1995. [49] H. Haugerud, T. Haugset, and F. Ravndal. A more accurate analysis of BoseEinstein condensation in harmonic traps. Phys. Lett., A225:18–22, 1997. [50] T. Haugset, H. Haugerud, and J. O. Anderson. Bose–Einstein condensation in anisotropic harmonic traps. Phys. Rev., A55:2922–2929, 1997. [51] S. W. Hawking. Zeta function regularization of path integrals in curved space-time. Commun. Math. Phys., 55:133–148, 1977.

142

KLAUS KIRSTEN

[52] M. P. Hertzberg, R. L. Jaffe, M. Kardar, and A. Scardicchio. Attractive Casimir forces in a closed geometry. Phys. Rev. Lett., 95:250402, 2005. [53] E. Hille. Analytic function theory, Vol. 2. Ginn, Boston, 1962. P [54] A. Hurwitz. Einige Eigenschaften der Dirichlet’schen Functionen f .s/ D dn n s die bei der Bestimmung der Classenzahlen binärer quadratischer Formen auftreten. Zeitschrift für Math. und Physik, 27:86–101, 1882. [55] C. Itzykson and J.-B. Zuber. Quantum field theory. McGraw-Hill, New York, 1980. [56] M. Kac. Can one hear the shape of a drum? Am. Math. Mon., 73:1–23, 1966. [57] K. Kirsten. Spectral Functions in Mathematics and Physics. Chapman&Hall/CRC, Boca Raton, FL, 2002. [58] K. Kirsten and S. A. Fulling. Kaluza–Klein models as pistons. Phys. Rev., D79: 065019, 2009. [59] K. Kirsten and D. J. Toms. Bose–Einstein condensation of atomic gases in a general harmonic-oscillator confining potential trap. Phys. Rev., A54:4188–4203, 1996. [60] K. Kirsten and D. J. Toms. Bose-Einstein condensation in arbitrarily shaped cavities. Phys. Rev., E59:158–167, 1999. [61] G. L. Klimchitskaya, F. Chen, R. S. Decca, E. Fischbach, D. E. Krause, D. López, U. Mohideen, and V. M. Mostepanenko. Rigorous approach to the comparison between experiment and theory in Casimir force measurements. J. Phys. A: Math. Gen., 39:6485–6493, 2006. [62] S. K. Lamoreaux. Demonstration of the Casimir force in the 0.6 to 6 micrometers range. Phys. Rev. Lett., 78:5–7, 1997. [63] S. K. Lamoreaux. A reply to the comment by Astrid Lambrecht and Serge Renaud. Phys. Rev. Lett., 84:5673–5673, 2000. [64] S. K. Lamoreaux. The Casimir force: background, experiments, and applications. Rep. Prog. Phys., 68:201–236, 2005. [65] V. Marachevsky. Casimir interaction of two plates inside a cylinder. Phys. Rev., D75:085019, 2007. [66] H. P. McKean and I. M. Singer. Curvature and eigenvalues of the Laplacian. J. Diff. Geom., 1:43–69, 1967. [67] K. A. Milton. The Casimir effect: physical manifestations of zero-point energy. World Scientific, River Edge, 2001. [68] K. A. Milton. The Casimir effect: recent controversies and progress. J. Phys., A37:R209–R277, 2004. [69] U. Mohideen and A. Roy. Precision measurement of the Casimir force from 0.1 to 0.9 m. Phys. Rev. Lett., 81:4549–4552, 1998. [70] N. E. Nörlund. Mémoire sur les polynômes de Bernoulli. Acta Math., 43:121–196, 1922.


143

[71] C. J. Pethick and H. Smith. Bose–Einstein condensation in dilute gases. Cambridge University Press, Cambridge, 2002. [72] L. Pitaevskii and S. Stringari. Bose–Einstein condensation. Oxford University Press, Oxford, 2003. [73] D. B. Ray and I. M. Singer. R-torsion and the Laplacian on Riemannian manifolds. Advances in Math., 7:145–210, 1971. [74] I. Sachs and A. Wipf. Finite temperature Schwinger model. Helv. Phys. Acta, 65:652–678, 1992. [75] M. Schaden. Dependence of the direction of the Casimir force on the shape of the boundary. Phys. Rev. Lett., 102:060402, 2009. [76] F. M. Serry, D. Walliser, and G. J. Maclay. The anharmonic Casimir oscillator (ACO): the Casimir effect in a model microelectromechanical system. J. Microelectromech. Syst., 4:193–215, 1995. [77] L. P. Teo. Finite temperature Casimir effect in spacetime with extra compactified dimensions. Phys. Lett., B672:190–195, 2009. [78] A. Terras. Bessel series expansion of the Epstein zeta function and the functional equation. Trans. Amer. Math. Soc., 183:477–486, 1973. [79] E. C. Titchmarsh. The theory of the Riemann zeta function. Oxford Science Publications, Oxford, 1951. [80] H. Weyl. Das asymptotische Verteilungsgesetz der Eigenwerte linearer partieller Differentialgleichungen. Math. Ann., 71:441–479, 1912. [81] H. Weyl. Das asymptotische Verteilungsgesetz der Eigenschwingungen eines beliebig gestalteten elastischen Körpers. Rend. Circ. Mat. Palermo, 39:1–50, 1915. K LAUS K IRSTEN D EPARTMENT OF M ATHEMATICS BAYLOR U NIVERSITY WACO , TX 76798 U NITED S TATES Klaus [email protected]


Zeta functions and chaos AUDREY TERRAS

1. Introduction This paper is an expanded version of lectures given at MSRI in June of 2008. It provides an introduction to various zeta functions emphasizing zeta functions of a finite graph and connections with random matrix theory and quantum chaos. For the number theorist, most zeta functions are multiplicative generating functions for something like primes (or prime ideals). The Riemann zeta is the chief example. There are analogous functions arising in other fields such as Selberg’s zeta function of a Riemann surface, Ihara’s zeta function of a finite connected graph. All of these are introduced in Section 2. We will consider the Riemann hypothesis for the Ihara zeta function and its connection with expander graphs. Chapter 3 starts with the Ruelle zeta function of a dynamical system, which will be shown to be a generalization of the Ihara zeta. A determinant formula is proved for the Ihara zeta function. Then we prove the graph prime number theorem. In Section 4 we define two more zeta functions associated to a finite graph: the edge and path zetas. Both are functions of several complex variables. Both are reciprocals of polynomials in several variables, thanks to determinant formulas. We show how to specialize the path zeta to the edge zeta and then the edge zeta to the original Ihara zeta. The Bass proof of Ihara’s determinant formula for the Ihara zeta function is given. The edge zeta allows one to consider graphs with weights on the edges. This is of interest for work on quantum graphs. See [Smilansky 2007] or [Horton et al. 2006b]. Lastly we consider what the poles of the Ihara zeta have to do with the eigenvalues of a random matrix. That is the sort of question considered in quantum chaos theory. Physicists have long studied spectra of Schrödinger operators and random matrices thanks to the implications for quantum mechanics where eigenvalues are viewed as energy levels of a system. Number theorists such as A. 145

146

AUDREY TERRAS

Odlyzko have found experimentally that (assuming the Riemann hypothesis) the high zeros of the Riemann zeta function on the line Re.s/ D 1=2 have spacings that behave like the eigenvalues of a random Hermitian matrix. Thanks to our two determinant formulas we will see that the Ihara zeta function, for example, has connections with spectra of more that one sort of matrix. References [Terras 2007] and [Terras 2010] may be helpful for more details on some of these matters. The first is some introductory lectures on quantum chaos given at Park City, Utah in 2002. The second is a draft of a book on zeta functions of graphs.

2. Three zeta functions 2.1. The Riemann zeta function Riemann’s zeta function for s 2 C with Re.s/ > 1 is defined to be 1 1 X Y 1 1 .s/ D 1 D : ns ps nD1

p prime

In 1859 Riemann extended the definition of zeta to an analytic function in the whole complex plane except for a simple pole at s D 1. He also showed that there is a functional equation s .s/ D .1 s/: (2-1) .s/ D s=2 2 The Riemann hypothesis, or RH, says that the nonreal zeros of .s/ (equivalently those with 0 < Re s < 1) are on the line Re s D 21 . It is equivalent to giving an explicit error term in the prime number theorem stated below. The Riemann hypothesis has been checked to the 1013 -th zero as of 12 October 2004, by Xavier Gourdon with the help of Patrick Demichel. See Ed Pegg Jr.’s website for an article called the Ten Trillion Zeta Zeros: http://www.maa.org/ editorial/mathgames. Proving (or disproving) the Riemann hypothesis is one of the million-dollar problems on the Clay Mathematics Institute website. There is a duality between the primes and the zeros of zeta, given analytically through the Hadamard product formula as various sorts of explicit formulas. See [Davenport 1980] and [Murty 2001]. Such results lead to the prime number theorem which says x # fp D prime jp x g ; as x ! 1: log x The spacings of high zeros of zeta have been studied by A. Odlyzko; see the page www.dtc.umn.edu/~odlyzko/doc/zeta.htm. He has found that experimentally they look like the spacings of the eigenvalues of random Hermitian matrices (GUE). We will say more about this in the last section. See also [Conrey 2003].

ZETA FUNCTIONS AND CHAOS

147

E XERCISE 1. Use Mathematica to make a plot of the Riemann zeta function. Hint. The function Zeta[s] in Mathematica can be used to compute the Riemann zeta function. There are many other kinds of zeta p function. Onepis the Dedekind zeta of an algebraic number field F such as Q. 2/ D fa C b 2 ja; b 2 Q g, where p primes are replaced by prime ideals p in the ring of integers O (which is Z Œ 2 D F p p fa C b 2 j a; b 2 Zg, if F D Q. 2/). Define the norm of an ideal of OF to be N a D jOF =aj. Then the Dedekind zeta function is defined for Re s > 1 by Y 1 .s; F / D 1 Np s ; p

where the product is over all prime ideals of OF . The Riemann zeta function is .s; Q/. Hecke gave the analytic continuation of the Dedekind zeta to all complex s except for a simple pole at s D 1. And he found the functional equation relating .s; F / and .1 s; F /. The value at 0 involves the interesting number hF =the class number of OF which measures how far OF is from having unique factorization into prime numbers rather than prime ideals (hQ.p2/ D 1/. Also appearing in .0; F / is the regulator which is a determinantpof logarithms of units (i.e., elements u 2 OF such that u 1 2 OF /. For F D Q. 2/, the regulator p is log 1 C 2 . The formula is .0; F / D

hR ; w

(2-2)

p where w is the number of roots of unity in F (w D 2 for F D Q. 2/. One has .0; Q/ D 12 . See [Stark 1992] for an introduction to this subject meant for physicists. 2.2. The Selberg zeta function. This zeta function is associated to a compact (or finite volume) Riemannian manifold. Assuming M has constant curvature 1, it can be realized as a quotient of the Poincaré upper half-plane H D fx C iy j x; y 2 R; y > 0g : The Poincaré arc length element is ds 2 D

dx 2 C dy 2 ; y2

which can be shown invariant under fractional linear transformation z‘

az C b ; cz C d

where a; b; c; d 2 R; ad

bc > 0:

148

AUDREY TERRAS

It is not hard to see that geodesics — curves minimizing the Poincaré arc length — are half-lines and semicircles in H orthogonal to the real axis. Calling these geodesics straight lines creates a model for non-Euclidean geometry since Euclid’s fifth postulate fails. There are infinitely many geodesics through a fixed point not meeting a given geodesic. The fundamental group of M acts as a discrete group of distance-preserving transformations. The favorite group of number theorists is the modular group D SL.2; Z/ of 2 2 matrices of determinant one and integer entries or the quotient D =f˙I g. However the Riemann surface M D SL.2; Z/nH is not compact, although it does have finite volume. Selberg defined primes in the compact Riemannian manifold M D nH to be primitive closed geodesics C in M . Here primitive means you only go around the curve once. Define the Selberg zeta function, for Re.s/ sufficiently large, as Z.s/ D

Y Y

1

e

.sCj /.C /

:

ŒC j 1

The product is over all primitive closed geodesics C in M D nH of Poincaré length .C /. By the Selberg trace formula (which we do not discuss here), there is a duality between the lengths of the primes and the spectrum of the Laplace operator on M . Here 2 @ @2 D y2 C : @x 2 @y 2 Moreover one can show that the Riemann hypothesis (suitably modified to fit the situation) can be proved for Selberg zeta functions of compact Riemann surfaces. E XERCISE 2. Show that Z.s C 1/=Z.s/ has a product formula which is more like that for the Riemann zeta function. The closed geodesics in M D nH correspond to geodesics in H itself. One can show that the endpoints of such geodesics in R (the real line = the boundary of H ) are fixed by hyperbolic elements of ; i.e., the matrices ac db with trace a C d > 2. Primitive closed geodesics correspond to hyperbolic elements that generate their own centralizer in . Some references for this subject are [Selberg 1989] and [Terras 1985].


149

Figure 1. An example of a bad graph for zeta functions.

2.3. The Ihara zeta function. We will see that the Ihara zeta function of a graph has similar properties to the preceding zetas. A good reference for graph theory is [Biggs 1974]. First we must figure out what primes in graphs are. Recalling what they are for manifolds, we expect that we need to look at closed paths that minimize distance. What is distance? It is the number of oriented edges in a path. First suppose that X is a finite connected unoriented graph. Thus it is a collection of vertices and edges. Usually we assume the graph is not a cycle or a cycle with hair (i.e., degree 1 vertices). Thus Figure 1 is a bad graph. We do allow our graphs to have loops and multiple edges however. Let E be the set of unoriented (or undirected) edges of X and V the set of vertices. We orient (or direct) the edges arbitrarily and label them e1 ; e2 ; : : : ; ejEj . An example is shown in Figure 2. Then we label the inverse edges (meaning

e2

e3 e5

e1

e4 Figure 2. We choose an arbitrary orientation of the edges of a graph. Then we label the inverse edges via ej CjEj D ej 1 , for j D 1; : : : ; 5.

150

AUDREY TERRAS

the edge with the opposite orientation) ej CjEj D ej 1 , for j D 1; : : : ; jEj. The oriented edges give an alphabet which we use to form words representing the paths in our graph. Now we can define primes in the graph X . They correspond to closed geodesics in compact manifolds. They are equivalence classes ŒC of tailless primitive closed paths C . We define these last adjectives in the next paragraph. A path or walk C D a1 as , where aj is an oriented or directed edge of X , is said to have a backtrack if aj C1 D aj 1 , for some j D 1; : : : ; s 1. A path C D a1 as is said to have a tail if as D a1 1 . The length of C D a1 as is s D .C /. A closed path means the starting vertex is the same as the terminal vertex. The closed path C D a1 as is called a primitive or prime path if it has no backtrack or tail and C ¤ D f , for f > 1. For the path C D a1 as , the equivalence class ŒC means ŒC D fa1 as ; a2 as a1 ; : : : ; as a1 as

1g :

That is, we call two prime paths equivalent if we get one from the other by changing the starting point. A prime in the graph X is an equivalence class ŒC of prime paths. Examples of primes in a graph. For the graph in Figure 2, we have primes ŒC D Œe2 e3 e5 , ŒD D Œe1 e2 e3 e4 , E D Œe1 e2 e3 e4 e1 e10 e4 . Here e10 D e5 1 and the lengths of these primes are: .C / D 3, .D/ D 4; .E/ D 7. We have infinitely many primes since En D Œ.e1 e2 e3 e4 /n e1 e10 e4 is prime for all n 1. But we don’t have unique factorization into primes. The only nonprimes are powers of primes. D EFINITION 3. The Ihara zeta function is defined for u 2 C, with juj sufficiently small by Y 1 .u; X / D 1 u.P / ; (2-3) ŒP

where the product is over all primes ŒP in X . Recall that .P / denotes the length of P . E XERCISE 4. How small should juj be for convergence of .u; X /? Hint. See formula (3-5) below for log .u; X /: There are two determinant formulas for the Ihara zeta function (see formulas (2-4) and (3-1) below). The first was proved in general by Bass [1992] and Hashimoto [1989], as Ihara considered the special case of regular graphs (those all of whose vertices have the same degree; i.e., the same number of oriented edges coming out of the vertex) and in fact was considering p-adic groups and not graphs. Moreover the degree had to be 1 C p e , where p is a prime number.


151

Figure 3. Part of the 4-regular tree. The tree itself is infinite.

The (vertex) adjacency matrix A of X is a jV jjV j/ matrix whose i; j entry is the number of directed edges from vertex i to vertex j . The matrix Q is defined to be a diagonal matrix whose j -th diagonal entry is 1 less than the degree of the j -th vertex. If there is a loop at a vertex, it contributes 2 to the degree. Then we have the Ihara determinant formula .u; X /

1

D .1

u2 / r

1

det.I

Au C Qu2 /:

(2-4)

Here r is the rank of the fundamental group of the graph. This is r D jEj jV j C 1. In Section 4 we will give a version of Bass’s proof of this formula. In the case of regular graphs, one can prove the formula using the Selberg trace formula for the graph realized as a quotient nT; where T is the universal covering tree of the graph and is the fundamental group of the graph. A graph T is a tree if it is a connected graph without any closed backtrackless paths. For a tree to be regular, it must be infinite. We will discuss covering graphs in the last section of this paper. For a discussion of the Selberg trace formula on nT , see the last chapter of [Terras 1999]. Figure 3 shows part of the 4-regular tree T4 . As the tree is infinite, we cannot put the whole thing on a page. It can be identified with the 3-adic quotient SL.2; Q3 /= SL.2; Z3 /. A finite 4-regular graph X is a quotient of T4 modulo the fundamental group of X . E XAMPLE 5. The tetrahedron graph K4 is the complete graph on 4 vertices and its zeta function is given by .u; K4 /

1

D .1

E XAMPLE 6. Let X D K4 edge e. See Figure 2. Then .u; X /

1

D .1

u2 /.1

u2 /2 .1

u/.1

2u/.1 C u C 2u2 /3 :

e be the graph obtained from K4 by deleting an u/.1 C u2 /.1 C u C 2u2 /.1

u2

2u3 /:

152

AUDREY TERRAS

E XERCISE 7. Compute the Ihara zeta functions of your favorite graphs; e.g., the cube, the icosahedron, the buckyball or soccer ball graph. E XERCISE 8. Obtain a functional equation for the Ihara zeta function of a (q C 1)-regular graph. It will relate .u; X / and .1=qu; X /. Hint. Use the Ihara determinant formula (2-4). There are various possible answers to this question. One answer is: n=2 1 X .u/ D . 1/n X X .u/ D .1 u2 /r 1Cn=2 1 q 2 u2 : qu In the special case of a (q C1)-regular graph the substitution u D q s makes the Ihara zeta more like Riemann zeta. That is we set f .s/ D .q s ; X / when X is (q C 1)-regular. Then the functional equation relates f .s/ and f .1 s/. See Exercise 8. The Riemann hypothesis for Ihara’s zeta function of a (q C 1)-regular graph says that .q

s

; X / has no poles with 0 < Re s < 1 unless Re s D 21 :

(2-5)

It turns out (using the Ihara determinant formula again) that the Riemann hypothesis means that the graph is Ramanujan; i.e., the nontrivial spectrum of the adjacency matrix of the graph is contained in the spectrum of the adjacency p p operator on the universal covering tree which is the interval Œ 2 q; 2 q. This definition was introduced by Lubotzky, Phillips and Sarnak [Lubotzky et al. 1988], who showed that for each fixed degree of the form p e C 1, p Dprime, there is a family of Ramanujan graphs Xn with jV .Xn /j ! 1. Ramanujan graphs are of interest to computer scientists because they provide efficient communication networks. The graph is a good expander. E XERCISE 9. Show that for a (q C 1)-regular graph the Riemann hypothesis is equivalent to saying that the graph is Ramanujan; i.e. if is an eigenvalue of p the adjacency matrix A of the graph such that jj ¤ q C 1, then jj 2 q. Hint. Use the Ihara determinant formula (2-4). What is an expander graph? There are 4 ideas. (1) There is a spectral property of some matrix associated to our finite graph X . Choose one of three matrices: (a) the (vertex) adjacency matrix A, (b) the Laplacian D A or I of degrees of vertices, or

D

1 2

AD

1 2

, where D is the diagonal matrix

(c) the edge adjacency matrix W1 to be defined in the next section.


153

Following [Lubotzky 1995], a graph is Ramanujan if the spectrum of the adjacency matrix for X is inside the spectrum of the analogous operator on the universal covering tree of X . One could ask for the analogous property of the other operators such as the Laplacian or the edge adjacency matrix. (2) X behaves like a random graph in some sense. (3) Information is passed quickly in the gossip network based on X . The graph has a large expansion constant. This is defined by formula (2-6) below. (4) The random walker on the graph gets lost FAST. D EFINITION 10. For sets of vertices S; T of X , define E.S; T / D fe je is edge of X with one vertex in S and the other vertex in T g : D EFINITION 11. If S is a set of vertices of X , we say the boundary is @S D E.S; X S /. D EFINITION 12. A graph X with vertex set V and n D jV j has expansion constant j@Sj : (2-6) h.X / D min jSj S V jS jn=2

The expansion constant is an analog of the Cheeger constant for differentiable manifolds. References for these things include [Chung 2007; Hoory et al. 2006; Terras 1999; 2010]. The first of these references gives relations between the expansion constant and the spectral gap X D min f1 ; 2 n 1 g if 0 D 0 1 n are the eigenvalues of I D 1=2 AD 1=2 . Fan Chung proves that 2 2hX X hX =2. This is an analog of the Cheeger inequality in differential geometry. She also connects these inequalities with webpage search algorithms of the sort used by Google. The possible locations of poles u of .u; X / of a (q C 1)-regular graph can be found in Figure 4. The poles satisfying the Riemann hypothesis are those on p the circle of radius 1= q. Any nontrivial pole; i.e., u ¤ ˙1, ˙1=q, which is not on that circle is a non-RH pole. In the (q C 1)-regular graph case, 1=q is always the closest pole of the Ihara zeta to the origin.

-1

-1/

q

-1/q

0

1/q

1/

q

1

Figure 4. Possible locations of poles of zeta for a (q C 1)-regular graph.

154

AUDREY TERRAS

E XERCISE 13. Show that Figure 4 correctly locates the position of possible poles of the Ihara zeta function of a (q C 1)-regular graph. Hint. Use the Ihara determinant formula (2-4). The Alon conjecture for regular graphs says that the RH is approximately true for “most” regular graphs See [Friedman 2008]for a proof. See [Miller and Novikoff 2008] for experiments leading to the conjecture that the percent of regular graphs exactly satisfying the RH approaches 27% as the number of vertices approaches infinity. The argument involves the Tracy–Widom distribution from random matrix theory. Newland [2005] performed graph analogs of Odlyzko’s experiments on the spacings of imaginary parts of zeros of Riemann zeta. See Figure 5 below and Figure 8 on page 173. An obvious question is: What is the meaning of the RH for irregular graphs? To understand this we need a definition. D EFINITION 14. RX D R is the radius of the largest circle of convergence of the Ihara zeta function.

Figure 5. For a pseudo-random regular graph with degree 53 and 2000 vertices, generated by Mathematica, the top row shows the distributions of the eigenvalues of the adjacency matrix on the left and imaginary parts of the Ihara zeta poles on the right. The bottom row contains their respective level spacings. The red line on the bottom left is the Wigner surmise for 2 the GOE y D ..x=2/ /e x =4 . From [Newland 2005].


155

As a power series in the complex variable u, the Ihara zeta function has nonnegative coefficients. Thus, by a classic theorem of Landau, both the series and the product defining X .u/ will converge absolutely in a circle juj < RX with a singularity (pole of order 1 for connected X ) at u D RX . See [Apostol 1976, p. 237] for Landau’s theorem. Define the spectral radius of a matrix M to be the maximum absolute value of all eigenvalues of M . We will see in the next section that by the Perron– Frobenius theorem in linear algebra (see [Horn and Johnson 1990], for example), 1=RX is the spectral radius of the edge adjacency matrix W1 which will be defined at the beginning of the next section. To apply the theorem, one must show that the edge adjacency matrix of a graph (under our usual assumptions) satisfies the necessary hypotheses. See [Terras and Stark 2007]. It is interesting to see that the quantity RX can be viewed from two points of view; complex analysis and linear algebra. For a (q C1)-regular graph RX D 1=q. If the graph is not regular, one sees by experiment that generally there is no functional equation. Thus when we make the change of variables u D Rs in our zeta, the critical strip 0 Re s 1 is too large. We should only look at half of it and our Riemann hypothesis becomes: The graph theory RH for irregular graphs: .u; X / is pole free in R < juj
1 0 0 0 > ˆ ˆ ˆ > ˆ > ˆ ˆ ˆ > ˆ > : ; : 1 ; 0 0 0 1 0 c2 0 d2

252

GEOFFREY MASON AND MICHAEL TUITE

and the involution is mapped to 2

0 61 ˇD6 40 0

1 0 0 0

0 0 0 1

3 0 07 7: 15 0

(150)

Thus as a subgroup of Sp.4; Z/, G also has a natural action on the Siegel upper half plane H2 as given in (138). This action is compatible with respect to the map (147) which is directly related to the observation that Aa .k; l; a ; "/ of (144) is a modular form of weight k C l for k C l > 2, whereas Aa .1; 1; a ; "/ D "E2 .a / is a quasimodular form. The exceptional modular transformation property of the latter term (29) leads via Theorem 11.7 to the following result: T HEOREM 11.8. F " is equivariant with respect to the action of G; i.e., there is a commutative diagram for 2 G, D"

?

D"

" F-

H2

? " FH2

E XERCISE 11.9. Show that to O."4 / 2 i ˝11 D 2 i 1 C E2 .2 /"2 C E2 .1 /E2 .2 /2 "4 ; 2 i ˝22 D 2 i 2 C E2 .1 /"2 C E2 .1 /2 E2 .1 /2 "4 ; 2 i ˝12 D " C E2 .1 /E2 .2 /"3 : 11.3. The genus-two partition function for the Heisenberg VOA. In this section we define and compute the genus-two partition function for the Heisenberg VOA M0 on the genus-two Riemann surface S.2/ described in the last section. The partition function is defined in terms of the genus-one 1-point functions .1/ ZM0 .v; a / on Sa D C=a for all v 2 M . The rationale behind this definition, which is strongly influenced by ideas in CFT, can be motivated by considering the following trivial sewing of a torus S1 D C=1 to a Riemann sphere CP1 . Let z1 2 S1 and z2 2 CP1 be local coordinates and define the sewing by identifying the annuli ra jza j j"jraN 1 via the sewing relation z1 z2 D " (adopting the same notation as above). The resulting surface is a torus described by the same modular parameter 1 . Let V be a VOA with LiZ metric h ; i and consider an n-point function21 .1/ FV ..v 1 ; x1 /; : : : .v n ; xn /; 1 / for xi 2 A1 , the torus annulus (141). This can 21 Here

and below we include a superscript .1/ to indicate the genus of the Riemann torus.

VERTEX OPERATORS AND MODULAR FORMS

253

be expressed in terms of a 1-point function ([MT1], Lemma 3.1) by .1/

.1/

FV ..v 1;x1 /;:::.v n;xn /;1 / D ZV .Y Œv 1;x1 :::Y Œv n;xn 1;1 / .1/

D ZV .Y Œv 1;x1n :::Y Œv n 1;xn

1n v

n

;1 /; (151)

for xi n D xi xn (see (132)). Denote the square bracket LiZ metric by h ; isq , and choose a basis fug of VŒr with dual basis fug N with respect to h ; isq . Expanding in this basis we find that for any 0 k n 1 X X Y Œv kC1; xkC1 : : : Y Œv n; xn 1 D hu; N Y Œv kC1; xkC1 : : : Y Œv n; xn 1isq u; r 0 u2VŒr

so that .1/

FV ..v 1 ; x1 /; : : : .v n ; xn /; 1 / D

.1/

X X

ZV .Y Œv 1 ; x1 : : : Y Œv k ; xk u; 1 /

r 0 u2VŒr

:hu; N Y Œv kC1 ; xkC1 : : : Y Œv n ; xn 1isq : Using (151) we have .1/

ZV .Y Œv 1 ; x1 : : : Y Œv k ; xk u; 1 / .1/

D Resz1 z1 1 FV ..v 1 ; x1 /; : : : .v k ; xk /; .u; z1 /; 1 /:

(152)

Let us now assume that each v i is quasiprimary of LŒ0 weight wtŒv i and let yi D "=xi 2 CP1 . Then (109), (112), (98) and (103) respectively imply hu; N Y Œv kC1 ; xkC1 : : : Y Œv n ; xn 1isq N sq D h1; Y | Œv n ; xn : : : Y | Œv kC1 ; xkC1 ui D h1; "LŒ0 Y | Œv n ; xn "

LŒ0

: : : "LŒ0 Y | Œv kC1 ; xkC1 " LŒ0 "LŒ0 ui N sq j Y " wtŒv D "r h1; Y Œv n ; yn : : : Y Œv kC1 ; ykC1 ui N sq xj2 kC1j n Y dyj wtŒvj .0/ D "r Resz2 z2 1 ZV ..v n ; yn /; : : : .v kC1 ; ykC1 /; .u; N z2 // : dxj kC1j n

We are also making use here of the isomorphism between the round and square bracket formalisms in the identification of the genus zero correlation function. The result of these calculations is that, for any 0 k n 1, Y i .1/ .1/ FV .v 1 ; : : : ; v n I 1 / FV ..v 1 ; x1 /; : : : ; .v n ; xn /; 1 / dxi wtŒv D 1in

254


vi '$ '$ yi x i

S

CP1

&% &% Figure 3. Equivalent insertion of v i at xi or yi D "=xi .

D

X r 0

"

r

X

u2VŒr

.1/

Resz1 z1 1 FV ..v 1 ; x1 /; : : : .v k ; xk /; .u; z1 /; 1 / .0/

Resz2 z2 1 FV ..v kC1 ; ykC1 /; : : : .v 1 ; y1 /; .u; N z2 // Y Y wtŒv i wtŒv j dxi dyj : 1ik

(153)

kC1j n .1/

Following Exercises 10.10 and 10.11 the (formal) form FV .v 1 ; : : : ; v n I 1 / is invariant with respect to Möbius transformations. (Similarly to Remark 10.12, .1/ we note that FV .v 1 ; : : : v n I 1 / is a conformally invariant global form on S1 for primary v 1; : : : ; v n ). Geometrically, (153) is telling us that we express .1/ FV .v 1; : : : ; v n I 1 / via the sewing procedure in terms of data arising from .1/ .0/ FV .v 1 ; : : : ; v k ; uI 1 / and FV .v kC1 ; : : : ; v 1 ; u/ N (cf. (108)). Furthermore, we may choose to consider the contribution from a quasiprimary vector v i as arising from either an “insertion” at xi 2 S1 or at the identified point yi D "=xi 2 CP1 . A special case of (153) is the partition (0-point) function for which we find the trivial identity X X .1/ .1/ .0/ .1/ ZV .1 / D "r ZV .u; 1 /Resz2 z2 1 FV .u; N z2 / D ZV .1 / C 0; r 0

u2VŒr

(154)

.0/

since FV .u; N z2 / D 0 for uN … VŒ0 . Motivated by this example, we define the genus-two partition function where we effectively replace the Riemann sphere in Figure 3, right, by a second torus S2 D C=2 as described in the Section 11.2. Thus replacing the genus-zero .0/ .1/ 1-point function FV .u; N 0/ of (154) by ZV .u; N 2 / we define the genus-two partition function for a VOA V with a LiZ metric by X X .1/ .2/ .1/ ZV .1 ; 2 ; "/ D "r ZV .u; 1 /ZV .u; N 2 /: (155) r 0

u2VŒr

The inner sum is taken over any basis fug for VŒr with dual basis fug N with respect to the square bracket LiZ metric. Although the definition is associated with the specific genus-two sewing scheme, it is regarded at this stage as a purely formal


255

expression which can be computed to any given order in ". One can also define genus-two correlation functions by inserting appropriate genus-one correlation functions in (155). We do not consider these here. Let us now compute the genus-two partition function for the rank one Heisenberg VOA M0 generated by a of weight 1. We employ the square bracket Fock basis of (124) which we alternatively notate here (cf. (118)) by v D v./ D aŒ 1e1 : : : aŒ pep 1;

(156)

for nonnegative integers ei . We recall that v./ is of square bracket weight P P wtŒv D i i ei and is described by a label set ˚ D f1; : : : ; pg with n D ei elements corresponding to an unrestricted partition D f1e1 : : : p ep g of wtŒv. The Fock vectors (156) form a diagonal basis for the LiZ metric h ; isq with 1 v; ei 1ip . i / ei !

vN D Q

(157)

from (119). Following (155), we find .2/

ZM0 .1 ; 2 ; "/ D

X

"wtŒv .1/ .1/ ZM0 .v; 1 /ZM0 .v; 2 /; e i i . i / ei !

Q v2V

(158)

.2/

where the sum is taken over the basis (156). ZM0 .1 ; 2 ; "/ is given by the following closed formula [MT4]: T HEOREM 11.10. The genus-two partition function for the rank one Heisenberg VOA is 1 .2/ ZM0 .1 ; 2 ; "/ D (159) .det.I A1 A2 // 1=2 ; .1 /.2 / with Aa of (144). P ROOF. The proof relies on an interesting graph-theoretic interpretation of (158). This follows the technique introduced in Theorem 11.3 for graphically .1/ interpreting the genus-one 1-point function ZM0 .v./; 1 / in terms the sum of weights for the -graphs. We sketch the main features of the proof leaving the interested reader to explore the details in [MT4]. Since v./ is indexed by unrestricted partitions D f1e1 ; 2e2 ; : : :g we may write (158) as .2/

ZM0 .1 ; 2 ; "/ D

X

Dfi ei g

Q

1 Q "i ei .1/ .1/ ZM0 .v./; 1 /ZM0 .v./; 2 /: (160) i i ei ! i

256


1s s2 1 1 A 2 s A s 2 1 A 2A 2 As As 1 5 1 Figure 4. A chequered diagram. .1/

Theorem 11.3 implies ZM0 .v./; 1 / D 0 for odd n D .1/

.1/

ZM0 .v./; 1 /ZM0 .v./; 2 / D

P

ei whereas for n even

XX 1 . 1 ; 1 /. 2 ; 2 /; .1 /.2 / 1

2

where 1 ; 2 independently range over the graphs for ˚ . Any pair 1; 2 can be naturally combined to form a chequered diagram D consisting of n verk

l

tices labelled by ˚ of valence 2 with n unoriented edges a consecutively labelled by a D 1; 2 as specified by a D : : : .kl/ : : : . Following Exercise 10.7 there are .n!!/2 chequered diagrams for a given v./. We illustrate an example of such a diagram in Figure 4 for v D aŒ 13 aŒ 22 aŒ 51 with 1 of Figure 1 and a separate choice for 2 with cycle shape .11/.22/.15/ For D f1e1 : : : p ep g the symmetric group ˙.˚ / acts on the chequered diagrams which have ˚ as underlying set of labeled nodes. We define Aut.D/, the automorphism group of D, to be the subgroup of ˙.˚ / which preserves Q node labels. Aut.D/ is isomorphic to ˙e1 ˙ep of order jAut.D/j D i ei !. We may thus express (160) as a sum over the isomorphism classes of chequered diagrams D with .2/

ZM0 .1 ; 2 ; "/ D

X .D/ 1 ; .1 /.2 / jAut.D/j D

and .D/ D

Y "i ei i

i

. 1 ; 1 /. 2 ; 2 /;

(161)

Q where D is determined by 1 ; 2 and noting that i . 1/ei D 1 for n even. From (135) we recall that . a ; 1 / is a product of the weights of the a labelled edges. Then .D/ can be more naturally expressed in terms of a weight function on chequered diagrams defined by .D/ D ˘E .E/;

(162)

where the product is taken over the edges E of D and where for an edge E

VERTEX OPERATORS AND MODULAR FORMS k

257

l

labeled a we define kCl

" 2 .E/ D p C.k; l; a / D Aa .k; l; a ; "/; kl for Aa of (144). Every chequered diagram can be formally represented as a product Y m DD Li i ; i

with D a disjoint union of unoriented chequered cycles (connected diagrams) Li with multiplicity mi (e.g. the chequered diagram of Figure 4 is the product of two disjoint cycles). Then is isomorphic to the direct product of the ˇ Aut.D/ mi mi ˇˇ ˇ groups Aut.Li / of order Aut.Li / D jAut.Li /jmi so that Y jAut.Li /jmi mi !: jAut.D/j D i

But from (162) it is clear that .D/ is multiplicative over disjoint unions of diagrams, and we find X D

YX X .L/ .D/ .L/m D D exp ; jAut.D/j jAut.L/jm m! jAut.L/j L m0

L

where L ranges over isomorphism classes of unoriented chequered cycles. Further analysis shows that [MT4] X L

.L/ D jAut.L/j

so that we find X D

1 2

Tr

X1 .A1 A2 /n D n

1 2

Tr log.1

A1 A2 /;

n1

.D/ D .det.1 jAut.D/j

A1 A2 //

1=2

;

following (145). Thus Theorem 11.10 holds.

The convergence and holomorphy of the determinant is the subject of Theorem 11.6 (b) so that having computed the closed formula (159) we may conclude .2/ that ZM0 .1 ; 2 ; "/ is not just a formal function but can be evaluated on D" to find .2/

T HEOREM 11.11. ZM0 .1 ; 2 ; "/ is holomorphic on the domain D" . .2/

We next consider the automorphic properties of ZM0 .1 ; 2 ; "/ with respect to the modular group G Sp.4; Z/ of (148) which acts on D" . We first recall a little

258


from the classical theory of modular forms (cf. Section 3). For a meromorphic function f . / on H, k 2 Z and D ac db 2 SL.2; Z/, we define the right action f . /jk D f . / .c C d/

k

;

(163)

where, as usual a C b : c C d f . / is called a weak modular form for a subgroup SL.2; Z/ of weight k if f . /jk D f . / for all 2 . We have already discussed the (genus-one) partition function for the rank n Heisenberg VOA V D M0˝n in Section 4.2 (cf. (41)). In particular, for n D 2 we have 1 .1/ .1/ Z 2 . / D ZM0 . /2 D : M0 . /2 Then we find .1/ .1/ Z 2 . /j 1 D . /Z 2 . /; (164)

D

M0

M0

where is a character of SL.2; Z/ of order 12 (cf. Exercise 8.5 and [Se]), and .1/ . / 1 M024

Z

D . /:

(165)

Similarly, we consider the genus-two partition function for the rank two Heisenberg VOA given by .2/ . ; ; "/ M02 1 2

Z

.2/

D ZM0 .1 ; 2 ; "/2 D

1 .1 /2 .2 /2 det.I

A1 A2 /

:

(166)

Analogously to (163), we define for all 2 G f .1 ; 2 ; "/jk D f . .1 ; 2 ; "// det.C ˝ C D/

k

;

(167)

where the action of on the right-hand-side is as in (148) and ˝.1 ; 2 ; "/ is determined by Theorem 11.7. Then (167) defines a right action of G on functions f .1 ; 2 ; "/. We next obtain a natural genus-two extension of (164). Define the a character .2/ of G by .2/ . 1 2 ˇ m / D . 1/m . 1 /. 2 /;

i 2

i;

i D 1; 2:

with i ; ˇ of (149) and (150). .2/ takes values which are twelfth roots of unity. Then, much as for Theorem 11.7, the exceptional transformation law of Aa .1; 1; a ; "/ D E2 .a / implies that T HEOREM 11.12. If 2 G then .2/ . ; ; "/j 1 M02 1 2

Z

.2/ . ; ; "/: M02 1 2

D .2/ . /Z


259

The definition (167) is analogous to that for a Siegel modular form for the symplectic group Sp.4; Z/ defined as follows (e.g. [Fr]). For a meromorphic function F.˝/ on H2 , k 2 Z and 2 Sp.4; Z/, we define the right action F.˝/jk D F. :˝/ det.C ˝ C D/

k

;

(168)

with :˝ of (138). F.˝/ is called a modular form for Sp.4; Z/ of weight k if F.˝/jk D F.˝/ for all 2 . Theorem 11.12 implies that for the rank 24 Heisenberg VOA M024 .2/ . ; ; "/j 12 M024 1 2

Z

.2/ . ; ; "/; M024 1 2

DZ

(169)

for all 2 G. This might lead one to speculate that, analogously to the genus.2/ one case in (165), Z 24 .1 ; 2 ; "/ 1 is a holomorphic Siegel modular form M0

of weight 12. Indeed, there does exist a unique holomorphic Siegel 12 form, .2/ 12 .˝/, such that .2/

12 .˝/ ! .1 /.2 /; .2/ .2/ . ; ; "/ 1 ¤ 12 .˝/. In M024 1 2 action of G on D" to Sp.4; Z/. These

as " ! 0, but explicit calculations show that Z

any case, we cannot naturally extend the observations are strongly expected to be related to the conformal anomaly [BK] in string theory and to the non-existence of a global section for the Hodge line bundle in algebraic geometry [Mu2]. Siegel modular forms do arise in the determination of the genus-two partition function for a lattice VOA VL for even lattice L of rank l (and conjecturally for all rational theories) as follows. We recall the genus-one partition function for VL is (cf. Section 7.3) .1/ .1/ . /L . /; M0l

.1/

ZVL . / D Z

(170)

P .1/ for L . / D ˛ q .˛;˛/=2 . In the genus-two case, we may define the Siegel lattice theta function by [Fr] X .2/ L .˝/ D exp. i ..˛; ˛/˝11 C 2.˛; ˇ/˝12 C .ˇ; ˇ/˝22 //: ˛;ˇ2L .2/

L .˝/ is a Siegel modular form of weight l=2 for a subgroup of Sp.4; Z/. The genus-one result (170) is naturally generalized to find [MT4]: T HEOREM 11.13. For a lattice VOA VL we have .2/

.2/ .2/ . ; ; "/L .˝/: M0l 1 2

ZVL .1 ; 2 ; "/ D Z

260

GEOFFREY MASON AND MICHAEL TUITE .2/

E XERCISE 11.14. Show that ZM0 .1 ; 2 ; "/ to O."4 / is given by 1 1 C 21 E2 .1 /E2 .2 /"2 .1 /.2 / 4 C 83 E2 .1 /2 E2 .2 /2 C 15 2 E4 .1 /E4 .2 / " : E XERCISE 11.15. Verify (159) to O."4 / by showing that det.I

A1 A2 / D 1

E2 .1 /E2 .2 /"2

15 E4 .1 /E4 .2 /"4 C O."6 /:

12. Exceptional VOAs and the Virasoro algebra In this section we review some recent research concerning a rôle played by the Virasoro algebra in certain exceptional VOAs [T2], [T3]. We will mainly concern ourselves here with simple VOAs V of strong CFT-type for which dim V1 > 0. We construct certain quadratic Casimir vectors from the elements of V1 and examine the constraints on V arising from the assumption that the Casimir vectors of low weight are Virasoro descendants of the vacuum. This sort of assumption is similar to that of ‘minimality’ in the holomorphic VOAs V .k/ that we discussed in Section 9.4. In particular we discuss how a special set of simple Lie algebras: A1 ; A2 ; G2 ; D4 ; F4 ; E6 ; E7 ; E8 , known as Deligne’s exceptional series [De], arises in this context. We also show that the genus-one partition function is determined by the same Virasoro condition. These constraints follow from an analysis of appropriate genus zero matrix elements and genus-one 2-point functions. In particular, we will make a relatively elementary use of rational matrix elements, the LiZ metric, the Zhu reduction formula and modular differential equations. As such, this example offers a useful and explicit application of many of the concepts reviewed in these notes. 12.1. Quadratic Casimirs and genus-zero constraints. Consider a simple VOA V of strong CFT-type of central charge c with d D dim V1 > 0. From Theorem 10.15, V possesses an LiZ metric h ; i, i.e., a unique (nondegenerate) normalized symmetric bilinear form. For a; b 2 V1 define Œa; b a0 b. From Exercise 9.7 this defines a Lie algebra on V1 with invariant bilinear form h ; i. We denote this Lie algebra by g. The modes of elements of V1 satisfy the Kac– Moody algebra (cf. Exercise 12.7) Œam ; bn D Œa; bmCn

mha; biımCn;0 :

(171)

which we denote by gO . Let fu˛ j ˛ D 1 : : : dg and fuN ˇ j ˇ D 1 : : : dg denote a g-basis and LiZ dual basis respectively. Define the quadratic Casimir vectors by .n/ D u˛1

N nu

˛

2 Vn ;

(172)


261

where ˛ is summed. Since u˛ 2 V1 is a primary vector it follows that ŒLm ; u˛n D nu˛mCn and hence Lm .n/ D .n

1/.n

m/

for m > 0:

(173)

Let Virc denote the subVOA of V generated by the Virasoro vector !. We then find: L EMMA 12.1. The LiZ metric is nondegenerate on Virc . P ROOF. Let v D L

n1 L n2

:::L

nk 1

2 Virc . Then (105) gives

hv; ai D h1; Lnk : : : Ln2 Ln1 ai D 0; for a 2 V n Virc . Since h ; i is nondegenerate on V it must be nondegenerate on Virc . R EMARK 12.2. This implies from Theorem 10.15 that Virc is simple with c ¤ cp;q of (123). We now consider the constraints on g that follow from assuming that .n/ 2 Virc for small n.22 Firstly let us note [Mat] L EMMA 12.3. If .n/ 2 Virc then .m/ 2 Virc and is uniquely determined for all m n. P P ROOF. If .n/ 2 Virc then .n/ D v2.Virc /n hv; N .n/ iv summing over a basis for .Virc /n . But hv; N .n/ i is uniquely determined by repeated use of (173) and Exercise 12.8. Furthermore, for m n we have .m/ D n 1 1 Ln m .n/ 2 .Virc /m . It follows that .2/ 2 Virc implies .2/ D

2d !; c

(174)

where c ¤ 0 following Remark 12.2. Note that for g simple, this is the standard Sugawara construction for ! of (69). Similarly .4/ 2 Virc implies .4/ D with c ¤ 0;

22 5

3d 4L2 2 1 C .2 C c/ L c .5c C 22/

following Remark 12.2.

41

;

(175)

22 The original motivation, due to Matsuo [Mat], for considering quadratic Casimirs is that both they and Virc are invariant under the automorphism group of V . Matsuo considered VOAs for which the automorphism invariants of Vn consist only of Virasoro descendents for small n. Hence for these VOAs it necessarily follows that .n/ 2 Virc .

262


We next consider the constraints on g if either (174) or (175) hold. We do this by analysing the following genus zero matrix element F.a; bI x; y/ D ha; Y .u˛ ; x/Y .uN ˛ ; y/bi;

(176)

where ˛ is summed and a; b 2 V1 . Using associativity and (172) we find F.a; bI x; y/ D ha; Y .Y .u˛ ; x y/uN ˛ ; y/bi X 1 x y n .n/ ha; o. /bi ; D y .x y/2

(177)

n0

where o..n/ / D n

from (37). Thus Exercise 12.8 implies 1 x y 2 .2/ C : F.a; bI x; y/ D dha; bi C 0 C ha; o. /bi y .x y/2 (178) Alternatively, we also have 1

F.a; bI x; y/ D ha; Y .u˛ ; x/e yL 1 Y .b; y/uN ˛ i D ha; e yL 1 Y .u˛ ; x D he yL1 a; Y .u˛ ; x

y/Y .b; y/uN ˛ i y/Y .b; y/uN ˛ i

D ha; Y .u˛ ; x y/Y .b; y/uN ˛ i 1 X y m ˛ ˛ D 2 ha; um 1 b1 m uN i x y y m0 1 y ˛ ˛ ˛ ˛ C ; D 2 ha; u 1 b1 uN i ha; u0 b0 uN i x y y using skew-symmetry and translation (cf. Exercises 2.15 and 2.16), invariance of the LiZ metric and that a is primary. The leading term is ha; u˛ 1 b1 uN ˛ i D ha; u˛ ihb; uN ˛ i D ha; bi: The next to leading term is ha; u˛0 b0 uN ˛ i D hu˛ ; a0 b0 uN ˛ i D K.a; b/; the Lie algebra Killing form K.a; b/ D T rg .a0 b0 /:

(179)

Thus we have 1 F.a; bI x; y/ D 2 y

ha; bi C K.a; b/

y x

y

C :

(180)


263

From Theorem 10.1 we know that F.a; bI x; y/ is given by a rational function F.a; bI x; y/ D

f .a; bI x; y/ ; x 2 y 2 .x y/2

(181)

where f .a; bI x; y/ is a homogeneous polynomial of degree 4. Furthermore f .a; bI x; y/ is clearly symmetric in x; y so that it may parameterized f .a; bI x; y/ D p.a; b/x 2 y 2 C q.a; b/xy.x

y/2 C r .a; b/.x

y/4 ; (182)

for some bilinears p.a; b/, q.a; b/ and r .a; b/. We find: P ROPOSITION 12.4. p.a; b/; q.a; b/; r .a; b/ are given by p.a; b/ D dha; bi; q.a; b/ D K.a; b/

(183) 2ha; bi;

r .a; b/ D ha; bi:

(184) (185)

P ROOF. Expanding (181) in .x

y/=y we have 1 x y 2 F.a; bI x; y/ D C ; p.a; b/ C q.a; b/ y .x y/2

whereas expanding (181) in y=.x y/ gives 1 y F.a; bI x; y/ D 2 r .a; b/ C . 2r .a; b/ C q.a; b// C : x y y

(186)

(187)

Comparing to (178) and (180) gives the result.

We next show that if .2/ 2 Virc then the Killing form is proportional to the LiZ metric: P ROPOSITION 12.5. If .2/ 2 Virc then K.a; b/ D 2ha; bi

d c

1 ;

(188)

so that 2d f .a; bI x; y/ D ha; bi dx 2 y 2 C xy.x y/2 C .x y/4 : (189) c 2d P ROOF. Equation (174) implies o..2/ / D L . Comparing the next to c 0 leading terms in (178) and (186) we find q.a; b/ D ha; o..2/ /bi D which implies the result.

2d ha; bi; c

264


Since the LiZ metric is nondegenerate, it follows from Cartan’s criterion in Lie theory that g is solvable for d D c and is semisimple for d ¤ c, i.e., g D g1 ˚ g2 ˚ ˚ gr ; for simple components gi of dimension d i . The corresponding Kac–Moody algebra gO i has level l i D 12 h˛ i ; ˛ i i where ˛ i is a long root23 so that the dual Coxeter number is i d h_ 1 : (190) D l i c P Furthermore, (174) implies that ! D 1ir ! i with ! i the Sugawara Virasoro O i . It vector for central charge c i D l i d i =.l i C h_ i / for the simple component g follows that for each component di d D ; i c c

so that

(191)

2d i ! ; c

i.2/ D

(192)

for the quadratic Casimir on gi . We next show that if .4/ 2 Virc then g must be simple. Let Lin denote the P j modes of ! i and Ln D i Lin denote the modes of ! with ŒLim ; Ln D 0 for P i ¤ j . Using .4/ D i i.4/ (for quadratic Casimirs on gi ) it follows from (173) that Li2 .4/ D 3i.2/ : (193) Since Lin satisfies the Virasoro algebra of central charge c i we find Li2 L2 2 1 D 8! i C c i !;

Li2 L

41

D 6! i :

If .4/ 2 Virc then (175) holds and hence Li2 .4/ D

3d .44 C 6c/! i C 4c i ! : c .5c C 22/

Equating to (193) and using (192) implies that !i D

ci !: c

But since the Virasoro vectors ! 1 ; : : : ! r are independent it follows that r D 1; i.e., g is a simple Lie algebra. If (175) holds one also finds that ha; o..4/ /bi D 23 Then .a; b/

i

9d.6 C c/ ha; bi: c .5c C 22/ .li /

ha; bi=li is the unique nondegenerate form on gO i

with normalization .˛i; ˛i /i D 2.


265

Comparing to the corresponding term in (177) this results in a further constraint on the parameters d; c in (189) given by dD

c .5c C 22/ : 10 c

(194)

Notice that the numerator vanishes for c D 0; 22=5, the zeros of the Kac determinant det M4 .c/ (122). For integral d > 0 there are only 42 rational values of c satisfying (194). This list is further restricted by the possible values of d for g simple. The level l is necessarily rational from (190). Restricting l to be integral (for example, if V is assumed to be C2 -cofinite [DM1]) we find that l D 1 and g must be one of Deligne’s exceptional Lie algebras: T HEOREM 12.6. Suppose .4/ 2 Virc . (a) g is a simple Lie algebra. (b) If c is rational and the level l of gO is integral then g D A1 ; A2 ; G2 ; D4 ; F4 ; E6 ; E7 or E8 ; with dual Coxeter number h_ D

d c

1D

12 C 6c ; 10 c

26 for central charge c D 1; 2; 14 5 ; 4; 5 ; 6; 7; 8 respectively and level l D 1.

The simple Lie algebras appearing in Theorem 12.6 are known as Deligne’s exceptional Lie algebras [De]. These algebras are of particular interest because not only is the dimension d of the adjoint representation g described by a rational function of c in (194) but also the dimensions of the irreducible representations that arise in decomposition of up to four tensor products of g. In Deligne’s original calculations, these dimensions were expressed as rational functions of a convenient parameter . In this VOA setting we instead employ the canonical parameter c, where c 10 D : 2Cc E XERCISE 12.7. Verify (171). E XERCISE 12.8. Show that .0/ D d 1 and .1/ D 0. E XERCISE 12.9. Verify (174). E XERCISE 12.10. Verify (175) using (121).

266


12.2. Genus-one constraints from quadratic Casimirs. We next consider the constraints on the genus-one partition function ZV . / that follow if .4/ 2 Virc . We will show that in this case, ZV . / is the unique solution to a second order Modular Linear Differential Equation (MLDE) (cf. Section 8.2). As a consequence, we prove that V D Lg .1; 0/, the level 1 WZW VOA where g is an Deligne exceptional series. To prove this we apply both versions of Zhu’s recursion formulas (Theorems 5.7 and 5.10). In particular, we evaluate the 1point correlation function for a Virasoro descendent of the vacuum from where an MLDE naturally arises. This is similar in spirit to Zhu’s [Z] analysis of correlation functions for the modules of C2 -cofinite VOAs but has the advantage of being considerably less technical. We recall the genus-one partition function ZV . / D TrV .q L0

c=24

/;

the 1-point correlation function for a 2 V ZV .a; / D TrV o.a/q L0

c=24

;

(195)

and the 2-point correlation function which can be expressed in terms of 1-point functions by FV ..a; x/; .b; y/; / D ZV .Y Œa; xY Œb; y1; / D ZV .Y Œa; x

yb; /;

(196) (197)

for square bracket vertex operators Y Œa; z D Y .qzL0 a; qz 1/: We define quadratic Casimir vectors in the square bracket VOA formalism Œn D u˛ Œ1

nuN ˛ 2 VŒn ;

(for ˛ summed) for basis fu˛ g and square bracket LiZ dual basis fuN ˛ g. Consider the genus-one analogue of (176) given by the 2-point function FV ..u˛ ; x/; .uN ˛ ; y/; / D ZV .Y Œu˛ ; xY ŒuN ˛ ; y1; /; (˛ summed). Associativity (196) implies the genus-one analogue of (177) so that X FV ..u˛ ; x/; .uN ˛ ; y/; / D ZV .Œn ; /.x y/n 2 : (198) n0

From Zhu’s first recursion formula (Theorem 5.7) we may alternatively expand F..u˛ ; x/; .uN ˛ ; y/; / in terms of Weierstrass functions as follows:


267

FV ..u˛ ; x/; .uN ˛ ; y/; / D TrV o.u˛ /o.uN ˛ /q L0 c=24 X . 1/mC1 .m/ C P1 .x y; /ZV .u˛ ŒmuN ˛ ; / m! m1 D TrV o.u˛ /o.uN ˛ /q L0 c=24 dP2 .x y; /ZV . /: Recalling Theorem 5.1 we may compare the .x and (198) to obtain

y/2 terms in this expression

ZV .Œ4 ; / D 3dE4 . /ZV . /:

(199)

Since .V; Y . /; 1; !/ is isomorphic to .V; Y Œ ; 1; !/ Q it follows that .n/ 2 Virc iff Œn 2 Virc . Thus assuming .4/ 2 Virc we have 3d ZV .Œ4 ; / D 4ZV .LŒ 22 1; / C .2Cc/ZV .LŒ 41; / ; (200) c.5c C22/ by (175). The Virasoro 1-point functions ZV .LŒ 22 1; /, ZV .LŒ 41; / can be evaluated via Zhu’s second recursion formula (Theorem 5.10). In particular taking u D !Q and v of LŒ0 weight k in (61) we obtain the general Virasoro recursion formula L0 c=24 ZV .LŒ nv; / D ın;2 TrV .o.!/o.v/q Q / X mCn 1 C . 1/m EmCn . /ZV .LŒmv; /: (201) mC1 0mk

But o.!/ Q D L0

c=24 and hence L0 TrV .o.!/o.v/q Q

c=24

/ D ZV .v; /;

where D q d=dq. It follows that ZV .LŒ 2v; / D Dk ZV .v; / C

X

E2Cm . /ZV .LŒmv; /;

(202)

2mk

where Dk D C kE2 . / is the modular derivative (30). (Zhu makes extensive use of the identities (201) and (202) in his analysis of correlation functions for C2 -cofinite VOAs [Z]. This is the origin of MLDEs as discussed in Section 9). We immediately find from (201) that ZV .LŒ 41; / D 0 and ZV .LŒ 22 1; / D D2 ZV .LŒ 21; / C E4 . /ZV .LŒ2LŒ 21; / D D 2 C 21 cE4 . / ZV . /; where D 2 D D2 D0 D .q d=dq/2 C 2E2 . /q d=dq. Substituting into (200) we find ZV . / satisfies the following second order MLDE: D 2 54 c.c C 4/E4 . / ZV . / D 0: (203)

268


(203) has a regular singular point at q D 0 with indicial roots c=24 and .c C 4/=24. Applying (194) it follows that there exists a unique solution with leading q expansion ZV . / D q c=24 .1 C O.q//. Furthermore, since E4 . / is holomorphic then ZV . / is also holomorphic for 0 < jqj < 1. In summary, we find: T HEOREM 12.11. If .4/ 2 Virc then ZV . / is a uniquely determined holomorphic function in H. An immediate consequence of Theorems 12.6 and 12.11 is: T HEOREM 12.12. V D Lg .1; 0/ the level one WZW model generated by g. P ROOF. Clearly Lg .1; 0/ V with !; .2/ ; .4/ 2 Lg .1; 0/. Thus Lg .1; 0/ satisfies the conditions of Theorem 12.11 for the same central charge c. Hence ZLg .1;0/ . / D ZV . / and so Lg .1; 0/ D V . P It is straightforward to substitute Z. / D q c=24 n dim Vn q n into (203) and solve recursively for dim Vn as a rational function in c. In this way we recover dim V1 D d of (194). The next two terms are c.804 C 508c C 175c 2 C 25c 3 / ; 2.22 c/.10 c/ c.33344 C 148872c C 68308c 2 C 10330c 3 C 975c 4 C 125c 5 / dim V3 D : 6.34 c/.22 c/.10 c/

dim V2 D

These dimension formulas can be further refined as follows. Consider the Virasoro decomposition of V2 : V2 D C! ˚ L

1 g ˚ P2 ;

(204)

where P2 is the space of weight two primary vectors. Let p2 D dim P2 . Then dim V2 D 1 C d C p2 with p2 D

5.5c C 22/.c C 2/2 .c 1/ : 2.22 c/.10 c/

(205)

Comparing with Deligne’s analysis of the irreducible decomposition of tensor products of g we find that p2 D dim Y2 ; where Y2 denotes an irreducible representation of g in Deligne’s notation [De]. This is explored further in [T3]. Similarly for V3 we find V3 D CŒL

2 1 ! ˚ L 1 g ˚ L 2 g ˚ L 1 P2 ˚ P3 ;


269

where P3 is the space of weight three primary vectors. Let p3 D dim P3 . Then dim V3 D 1 C 2d C p2 C p3 with p3 D

5c.5c C 22/.c 1/.c C 5/.5c 2 C 268/ D dim X2 C dim Y3 ; 6.34 c/.22 c/.10 c/

where X2 ; Y3 denote two other irreducible representations of g in Deligne’s notation of dimension 5c.5c C22/.c C6/.c 1/ dim X2 D ; 2.10 c/2 5c.5c C22/.c C2/2 .c 8/.5c 2/.c 1/ dim Y3 D : 6.10 c/2 .22 c/.34 c/ 12.3. Higher-weight constructions. We can generalize the arguments given above to consider a VOA V with dim V1 D 0. Here we construct Casimir vectors from the weight two primary space P2 (provided dim P2 > 0) and obtain constraints on V that follow from such Casimirs being Virasoro vacuum descendents. If dim P2 D 0 we consider primaries of weight 3 and so on. In general, let V be a VOA with primary vector space PK of lowest weight K; i.e., Vn D .Virc /n for all n < K, so that X c=24 n K ZV . / D q dim.Virc /n q C O.q / : (206) n0

n0

n 0. The function } is the inverse of the elliptic integral Z 1 1 } 1 .xI g2 ; g3 / D dt (2.6) p x 4t 3 g2 t g3 where g2 ; g3 2 C are known as Weierstrass invariants. Given periods !1 ; !2 , the invariants are g2 D 60

X

.m;n/2Z Z f.0;0/g

g3 D 140

X

.m;n/2Z Z f.0;0/g

1 ; .m!1 C n!2 /4 1 : .m!1 C n!2 /6

(2.7)

Alternately given invariants g2 ; g3 , periods !1 ; !2 can be constructed if the def: discriminant D g23 27g32 is nonzero — that is, when the Weierstrass cubic 4t 3 g2 t g3 does not have repeated roots (see 2173 of [Whittaker and Watson 1927]). Therefore we refer to }.zI !1 ; !2 / as either }.z/ or }.zI g2 ; g3 / and consider only cases where the invariants are such that g23 ¤ 27g32 . By (2.2) and

282

JENNIE D’AMBROISE

(2.6) one can see that the Weierstrass elliptic function satisfies } 0 .z/2 D 4}.z/3

g2 }.z/

g3 :

(2.8)

Note that in Michael Tuite’s lecture in this volume, !m;n there is equal to m!1 C n!2 here with our !1 ; !2 specialized to 2 i and 2 i in his lecture. In the special case that the discriminant > 0, the roots of 4t 3 g2 t g3 are real and distinct, and are conventionally notated by e1 > e2 > e3 for e1 Ce2 Ce3 D 0. In this case 4t 3 g2 t g3 D 4.t e1 /.t e2 /.t e3 /, the Weierstrass invariants are given in terms of the roots by g2 D 4.e2 e3 C e1 e3 C e1 e2 /;

g3 D 4e1 e2 e3 ;

(2.9)

and } can be written in terms of the Jacobi elliptic functions by }.z/ D e3 C 2 ns2 . z; k/; }.z/ D e2 C 2 ds2 . z; k/;

(2.10)

}.z/ D e1 C 2 cs2 . z; k/ def:

for e1 < z 2 R , 2 D e1 e3 and modulus k such that k 2 D ee12 ee33 (similar equations hold if z 2 R and z is in a different range in relation to the real roots e1 ; e2 ; e3 , and alternate relations hold for nonreal roots when < 0; see chapter II of [Greenhill 1959]). Note that the Jacobi elliptic functions solve a differential equation which contains only even powers of f .u/, and } solves an equation with no squared or quartic powers of }. Weierstrass elliptic functions have the advantage of being 3 easily implemented in the case that the cubic 4t g2 tp g3 is not factored in R terms of its roots. Elliptic integrals of the type R.x/= P .x/dx, where P .x/ is a cubic polynomial and R is a rational function of x, can be written in terms of three fundamental Weierstrassian normal elliptic integrals although we will not record the details here (see Appendix of [Byrd and Friedman 1954]). In Section R x p6 we will see a method which allows one to write the elliptic integral x0 1= F.t/dt, for F.t/ a quartic polynomial, in terms of the Weierstrassian normal elliptic integral of the first kind (2.6) by reducing the quartic to a cubic.

3. Jacobi theta functions Jacobi theta functions are functions of two arguments, z 2 C a complex number and 2 H in the upper-half plane. Every elliptic function can be written as the ratio of two theta functions. Doing so elucidates the meromorphic nature of elliptic functions and is useful in the numerical evaluation of elliptic functions. One must be cautious with the notation of theta functions, since many different

APPLICATIONS OF ELLIPTIC AND THETA FUNCTIONS TO FRLW COSMOLOGY

283

conventions are used. We will use the notation of [Whittaker and Watson 1927] to define 1 X def: 1 .z; / D 2q 1=4 . 1/n q n.nC1/ sin..2n C 1/z/; def:

nD0 1 X

def:

nD0 1 X

def:

nD1 1 X

2 .z; / D 2q 1=4 3 .z; / D 1 C 2 4 .z; / D 1 C 2

q n.nC1/ cos..2n C 1/z/;

q

(3.1) n2

cos.2nz/; 2

. 1/n q n cos.2nz/;

nD1 def:

def:

e i

where q D is called the nome. We also define the special values i D i .0; /. In terms of theta functions, the Jacobi elliptic functions are sn.u; k/ D

cn.u; k/ D

dn.u; k/ D

3 1 .u=32 ; / 2 4 .u=32 ; / 4 2 .u=32 ; / 2 4 .u=32 ; / 4 3 .u=32 ; / 3 4 .u=32 ; /

; ;

(3.2)

;

where is chosen such that k 2 D 24 =34 . By [Whittaker and Watson 1927, 2211], if 0 < k 2 < 1 there exists a value of for which the quotient 24 =34 equals k 2 .

4. The FRLW cosmological model The Friedmann–Robertson–Lemaˆıtre–Walker cosmological model assumes that our current expanding universe is on large scales homogeneous and isotropic. On a d-dimensional spacetime this assumption translates into a metric of the form dr 2 2 2 2 2 2 ds D dt C a.t/ Q C r d˝d 2 ; (4.1) 1 k 0r 2 where a.t/ Q is the cosmic scale factor and k 0 2 f 1; 0; 1g is the curvature parameter.

284

JENNIE D’AMBROISE

The Einstein field equations Gij D d Tij C gij then govern the evolution of the universe over time. In these equations, the Einstein tensor def: Gij D Rij 21 Rgij is computed directly from the metric gij by calculating the Ricci tensor Rij and the scalar curvature R. Also d D 8Gd , where Gd is a generalization of Newton’s constant to d-dimensional spacetime and > 0 is the cosmological constant. The form of the energy-momentum tensor Tij depends on what sort of matter content one is assuming, and in this lecture will be that of a perfectfluid — that is, Tij D .p C/gi0 gj 0 Cpgij , where .t/ and p.t/ are the density and pressure of the fluid, respectively. For the metric (4.1), Einstein’s equations are .d 1/.d 2/ k0 H 2 C 2 D d .t/ C ; (i) 2 aQ .d 1/.d 2/ 2 .d 2/.d 3/ k 0 H C D d p.t/ C ; (ii) 2/HP C 2 2 aQ 2 def: P for H .t/ D a.t/= Q a.t/ Q and where dot denotes differentiation with respect to t. In this lecture, only equation (i) will be required to relate a.t/ Q to elliptic and theta functions. We rewrite equation (i) in terms of conformal time by defining the def: conformal scale factor a./ D a.f Q .//, where f ./ is the inverse function of .t/, which satisfies .t/ P D 1=a.t/. Q In terms of a./, (i) becomes .d

4 z a0 ./2 D a./ C zd .f .//a./4

z def: where we use notation D 2=.d 1/.d we take spacetime dimension d > 2.

k 0 a./2 ;

def:

2/, zd D 2d =.d

(4.2) 1/.d

2/ and

5. FRLW and Jacobi elliptic and theta functions In general, if f .u/ is a solution to f 0 .u/2 C af .u/2 C bf .u/4 D c, then g.u/ D ˇf .˛u/ is a solution to g 0 .u/2 C Ag.u/2 C Bg.u/4 D

A2 bc a2 B

(5.1)

p p for ˛ D A=a and ˇ D Ab=.aB/, where we may choose either the positive or negative square root for each of ˛ and ˇ. We will construct solutions to (4.2), given that Jacobi elliptic functions solve (2.4), and also proceed to write these solutions in terms of theta functions by the relations in (3.2).


285

For the special case of density .t/ D D=a.t/ Q 4 with D > 0, (4.2) becomes a0 ./2 C k 0 a./2

4 z a./ D zd D:

(5.2)

z and a; b; c as in the table in Section Therefore in (5.1) we take A D k 0 ; B D , 2. To restrict to real solutions we only consider entries in the table for which the ratios a=A and b=B are positive, and we also restrict D to be such that the right side of (5.2) agrees with the right side of (5.1). 2 k with 0 < k < 1, For positive curvature k 0 D 1, and for D D 2 /2 z .1 C k z d we solve 4 k2 z a0 ./2 C a./2 a./ D (5.3) z C k 2 /2 .1 in conformal time in terms of Jacobi elliptic functions to obtain q 1 = 24 C34 ; k 2 3 q ; asn ./ D p sn p ;k D q 2 2/ z .1Ck

ans ./ D p acd ./ D p adc ./ D p

1 2/ z .1Ck

k 2/ z .1Ck

1 2/ z .1Ck

1Ck

ns

p ;k 1Ck 2

cd

p ;k 1Ck 2

D

dc

p ;k 1Ck 2

D

D

4 4 = 4 C 4 ; z .1C 2 3 2 3 / 4 q 4 4 = 2 C34 ; q 2 3 q ; 4 4 = 4 C 4 ; z .1C 2 3 2 3 / 1 q 2 = 24 C34 ; 2 3 q q ; 4 4 = 4 C 4 ; z .1C 2 3 2 3 / 3 q 3 = 24 C34 ; 2 3 q q ; 4 4 = 4 C 4 ; z .1C 2 3 2 3 / 2

(5.4)

where is chosen such that k 2 D 24 =34 . The first two solutions, asn ./ and ans ./, reduce to hyperbolic trigonometric functions in the case of modulus k D 1, since sn.u; 1/ D tanh.u/ and ns.u; 1/ D coth.u/. That is, two additional solutions in terms of elementary functions are p p 1 1 a2 ./ D p (5.5) a1 ./ D p tanh.= 2/; coth.= 2/: z z 2 2 For these two solutions, one may solve the differential equation .t/ P D 1=a.t/ Q D 1=a..t// for .t/, and therefore obtain the cosmic scale factor a.t/ Q D a..t// which solves the Einstein field equation (i) in Section 4 for .t/ D D=a.t/ Q 4 with zd .1 C k 2 /2 . Doing so, we obtain special value D D k 2 =z q p p 1 z z aQ 1 .t/ D a1 ..t// D p tanh ln e t C e 2 t 1 ; z 2 (5.6) q p p 1 z z t 2 t coth ln e C e C1 ; aQ 2 .t/ D a2 ..t//; D p z 2 for t > 0.

286

JENNIE D’AMBROISE

For negative curvature k 0 D 1, and for DD

1

k2

z zd .k 2

2/2

with 0 < k < 1, equation (5.2) becomes a0 ./2

a./2

4 z a./ D

1 .k 2

k2 z 2/2

:

(5.7)

In terms of Jacobi elliptic functions and theta functions, we obtain the two solutions for the scale factor in conformal time, q q p 2 34 24 3 1 =32 234 24 ; 1 k q ; asc ./ D p sc p 2 ; k D q 2 4 4 2 4 4 z .2 k /

acs ./ D p

1 z .2 k2/

2 k

cs p

z =3 .2 3 2 / 4 2

;k 2 k2

D

2 =32

q

q

23 2 ;

(5.8) 234 24 ;

3 4 q 4 4 = 2 2 4 4 ; z .2 3 3 2 3 2 / 1

;

where is such that k 2 D 24 =34 . Note that by the comments following equation (2.10), it is possible to express the solutions obtained in this section in terms of Weierstrass functions. E. Abdalla and L. Correa-Borbonet [2002] have also considered the Einstein equation (i) with .t/ D D=a.t/ Q 4 , and have found connections with Weierstrass functions in cosmic time (as opposed to the conformal time argument given here). In Section 6 of this lecture we will find more general solutions to the conformal time equation (4.2) in terms of }, for arbitrary curvature k 0 and D-value. We will also consider the density functions .t/ D D=a.t/ Q 3 and .t/ D D1 =a.t/ Q 3 CD2 =a.t/ Q 4 in Section 6.

6. FRLW and Weierstrass elliptic functions In general, if g.0/ D x0 and g.u/ satisfies g 0 .u/2 D F.g.u//

(6.1)

for F.x/ D A4 x 4 C A3 x 3 C A2 x 2 C A1 x C A0 any quartic polynomial with no repeated roots, then the inverse function y.x/ of g.u/ is the elliptic integral Z x dt : (6.2) y.x/ D p F.t/ x0 For the initial condition g 0 .0/ D 0, x0 is a root of the F.x/ by R 1polynomial ıp (6.1). In this case the integral (6.2) can be rewritten as dz P .z/, where D 1=.x x0 / and P .z/ is a cubic polynomial. To do this, first expand F.t/


287

into its Taylor series about x0 and then perform theRchangeıp of variables z D 1 Q. / for D 1=.t x0 /. Furthermore one may obtain the form d 1 1 00 0 3 g2 g3 . This is done by setting 4 F .x0 / C 6 F .x0 / , where Q. / D 4 z D .4 B2 =3/=B3 where B2 D F 00 .x0 /=2 and B3 D F 0 .x0 / are the quadratic and cubic coefficients of P .z/ respectively. Note that since F.x/ has no repeated roots, x0 is not a double root, F 0 .x0 / ¤ 0 and the variable z is well-defined. Therefore we have obtained 6F 0 .x0 / y C x0 D } 1 . /: 24 F 00 .x0 / Writing this in terms of x, setting x D g.u/ and solving for g.u/, one obtains the solution to (6.1): g.u/ D x0 C

F 0 .x0 / 4}.uI g2 ; g3 /

1 00 6 F .x0 /

;

(6.3)

where 1 2 1 4 A1 A3 C 12 A2 1 1 6 A0 A2 A4 C 48 A1 A2 A3

g2 D A0 A4 g3 D

and 1 2 16 A1 A4

2 1 16 A0 A3

3 1 216 A2

(6.4)

are referred to as the invariants of the quartic F.x/. Since F.x/ has no repeated roots, the discriminant D g23 27g32 ¤ 0. Here if x0 D 0, (6.3) becomes g.u/ D

A1 4}.uI g2 ; g3 /

1 3 A2

:

(6.5)

If the initial condition on the first derivative is such that g 0 .0/ ¤ 0 then x0 is not a root of F.x/ and a more general solution to (6.1) is due to Weierstrass. The proof (which we will not include here) was published by Biermann in 1865 (see [Biermann 1865; Reynolds 1989]). The solution is p 1 1 F.x0 /} 0 .u/ C 12 F 0 .x0 / }.u/ 24 F 00 .x0 / C 24 F.x0 /F 000 .x0 / g.u/ D x0 C 2 1 0000 1 2 }.u/ 24 F 00 .x0 / 48 F.x0 /F .x0 / (6.6) where } is formed with the invariants of the quartic seen in (6.4) such that ¤ 0. Here if x0 D 0, (6.6) becomes p 1 A0 } 0 .u/ C 21 A1 }.u/ 12 A2 C 14 A0 A3 g.u/ D : (6.7) 2 1 1 2 }.u/ 12 A2 2 A0 A4 To generate a number of examples, we consider the conformal time Einstein equation (4.2) for density .t/ D D1 =a.t/ Q 3 C D2 =a.t/ Q 4 with D1 ; D2 > 0. In

288

JENNIE D’AMBROISE

this case (4.2) becomes 4 z a0 ./2 D a./

k 0 a./2 C zd D1 a./ C zd D2

(6.8)

z A3 D 0; A2 D k 0 ; A1 D zd D1 and A0 D zd D2 . The most and we take A4 D ; def: general solution seen here is with initial conditions a0 .0/ ¤ 0 so that a.0/ D a0 is not a root of the polynomial z4 F.t/ D t

k 0 t 2 C zd D1 t C zd D2 :

The solution is given by (6.6) as p F.a0 /} 0 ./ C 21 F 0 .a0 / }./ a./ D a0 C 2 1 2 }./ 24 F 00 .a0 /

(6.9)

1 1 00 000 24 F .a0 / C 24 F.a0 /F .a0 / ; 0000 1 F.a /F .a / 0 0 48

(6.10)

with Weierstrass invariants zd D2 C 1 .k 0 /2 and g2 D z 12 zd D2 k 0 1 z z 2 D 2 C g3 D 1 z 6

16

d

1

0 3 1 216 .k /

(6.11)

restricted to be such that D g23 27g32 ¤ 0 so that F.t/ does not have repeated roots. Since D2 ; zd > 0, by (6.9) zero is not a root of F.t/ and therefore for initial condition a0 .0/ ¤ 0 and a0 D 0, the solution to (6.8) is given by (6.7) as p k0 zd D2 } 0 ./ C 21 zd D1 }./ C 12 (6.12) a./ D k0 2 1 z z D 2 }./ C 12 2 d 2 for invariants g2 ; g3 as in (6.11) with ¤ 0. One can compare this with the results in papers by Aurich, Steiner and Then, where curvature is taken to be k 0 D 1 [Aurich and Steiner 2001; Aurich et al. 2004]. For a0 .0/ D 0 and a0 a root of F.t/ in (6.9), the solution to (6.8) is given by (6.3), F 0 .a0 / a./ D a0 C (6.13) 4}./ 16 F 00 .a0 / again with invariants (6.11) such that ¤ 0. For a more concrete example, consider the density function .t/ D D=a.t/ Q 3 for D > 0 so that conformal time equation (4.2) becomes 4 z a0 ./2 D a./

k 0 a./2 C zd Da./:

(6.14)

z A3 D A0 D 0; A2 D k 0 Here zero is a root of the polynomial F.t/ with A4 D ; 0 and A1 D zd D. Therefore with initial conditions a .0/ D a.0/ D 0, the solution


289

to (6.14) is given by (6.5) as a./ D with invariants g2 D

0 2 1 12 .k /

and

3z d D 12}./ C k 0

g3 D

1 z 2 2 1 d D C 216 .k 0 /3 16 z

(6.15)

(6.16)

restricted to be such that ¤ 0. As noted in the comments following equations (2.10), one can write this solution in terms of Jacobi elliptic functions (by using equations (2.10) if the roots of the reduced cubic 4t 3 g2 t g3 are real). To demonstrate this, we choose s 1 2 DD z zd 27 and k 0 D 1, so that g3 D 0 and g2 D becomes

1 12 .

For this positive curvature case, (6.15) p

2

(6.17) ; z 12}./ C 1 3 p p and the reduced cubic isp 4t 3 .1=12/t D 4t.t p 1=4 3/.t C 1=4 3/. Applying (2.10) with e3 D 1=4 3, e2 D 0, e1 D 1=4 3, (6.17) can be equivalently written in terms of Jacobi elliptic functions as p p z z 2= 2= a./ D p p p p p p Dp p 3 3 C 6 ns2 = 2 3; 1= 2 3 C 6 ds2 = 2 3; 1= 2 p z 2= Dp (6.18) p p p 3 C 3 C 6 cs2 = 2 3; 1= 2 ; a./ D p

1 e e 1 since k 2 D 2 3 D and 2 D e1 e3 D p . In terms of theta functions, e1 e3 2 2 3 this is p p p z 2 2 = 2 3 2 ; 2= 3 1 3 a./ D p p p p p 3 3 32 12 = 2 3 32 ; C 6 22 42 = 2 3 32 ; p p p z 4 2 = 2 3 2 ; 2= 3 1 3 Dp p p p p 3 34 12 = 2 3 32 ; C 6 22 42 32 = 2 3 32 ; p p p z 2 2 = 2 3 2 ; 2= 3 1 3 D p p p p p (6.19) 3 C 3 32 12 = 2 3 32 ; C 6 42 22 = 2 3 32 ; where is taken such that

1 2

D 24 =34 .

290

JENNIE D’AMBROISE

As a final example, we return to .t/ D D=a.t/ Q 4 for D > 0 considered in Section 5. That is, we will obtain alternate solutions to equation (5.2). Since z A3 D A1 D D > 0, zero is not a root of the polynomial F.t/ with A4 D ; 0 0; A2 D k and A0 D zd D. Therefore for initial conditions a0 .0/ ¤ 0 and a.0/ D 0, (6.7) gives the solution p zd D} 0 ./ a./ D (6.20) zd D 2.}./ C 1 k 0 /2 1 z 12

2

for invariants zd D C 1 .k 0 /2 g2 D z 12

1 1 z d Dk 0 C 216 .k 0 /3 6 z

g3 D

and

(6.21)

restricted to be such that ¤ 0. (6.20) is more general than the solutions to (5.2) in Section 5, since here the curvature k 0 and the constant D are unspecified. To see this solution expressed in terms of Jacobi elliptic functions, take curvature zd / for 0 < k < 1 so that g3 D 0 and g2 D 1 . Then the k 0 D 1 and D D 1=.36z 9 reduced cubic is 4t 3 19 t D 4 t 16 t C 16 so that e3 D 16 , e2 D 0, e1 D 16 and (6.20) becomes p z } 0 ./ .12= / : (6.22) a./ D 1 2 / 1 144.}./ C 12 By (2.8) and (2.10), we write this solution in terms of Jacobi elliptic functions and obtain r 2 3 sn2 p ; p1 C sn4 p ; p1 3 2 3 2 a./ D p z 2 ns p ; p1 sn p ; p1 6 3

2

1 D p z ds p ; p1 2 3 3

2

3

2

v u u 2 ds2 p ; p1 1 u 3 2 t 2 ds2 p ; p1 C 1 3

cs

p1 p ; 3 2

2

D p r z 1 C cs2 p ; p1 6 1 C 2 cs2 p ; p1 3

2

3

(6.23)

2

where each of the positive and negative square roots solve (5.2) for k 0 D 1, zd / and where 2 D e1 e3 D 1 and k 2 D .e2 e3 /=.e1 e3 / D D 1=.36z 3 p D 12 . Writing (6.23) in terms of theta functions, and defining D = 333 , we


291

find that q 3 1 ./ 224 44 ./ 322 32 12 ./42 ./ C 34 14 ./ a./ D p z 2 4 ./ 2 2 2 ./ 2 2 ./ 6 3 1 2 4 v u 2 2 2 2 u 2 ./ 4 2 ./ 3 1 ./ 3 1 t 2 4 3 D p 4 2 2 2 2 ./ C 2 z 2 3 2 4 3 ./ 3 1 ./ 2 4 3 Dq

4 3 2 ./1 ./ z 2 2 ./ C 2 2 ./ 2 2 ./ C 2 2 2 ./ 6 4 2 3 1 4 2 3 1

(6.24)

by the theta function representations for sn, ds and cs respectively. Here the that forms the theta functions is suppressed and is taken to satisfy 12 D 24 =34 .

7. Summary There are a number of ways to see that elliptic and theta functions solve the d-dimensional Einstein gravitational field equations in a FRLW cosmology with a cosmological constant. Here we considered a scenario with no scalar field and with density functions .t/ D D1 =a.t/ Q 3 C D2 =a.t/ Q 4 , .t/ D D=a.t/ Q 3 and .t/ D D=a.t/ Q 4 scaling in inverse proportion to the scale factor a.t/. Q In PQ 2 D an expression these cases the first Einstein equation (i) takes the form a.t/ containing negative powers of the cosmic scale factor a.t/. Q At this point, one could have introduced the inverse function y.x/ of a.t/ Q to obtain an expression for y.x/ as the integral of a power of x divided by the square root of a polynomial in x. That is, y.x/ would be an elliptic integral that is not normal; other authors have taken this approach [Abdalla and Correa-Borbonet 2002; Kraniotis and Whitehouse 2002]. Here, we switched to conformal time by a change of def: variables a./ D a.f Q .//. This produced an equation of the form a0 ./2 D an expression containing nonnegative powers of the conformal scale factor a./. After reviewing the definitions and properties of elliptic and theta functions in sections 2 and 3, we introduced the FRLW cosmological model in Section 4. In Section 5 for .t/ D D=a.t/ Q 3 , we obtained a differential equation for a./ containing only even powers of a./ and constructed solutions in terms of Jacobi elliptic functions, restricted to particular values of the constant D, parameterized by modulus 0 < k < 1. The equivalent theta function representations for these solutions were recorded, and we noted the special cases for which the elliptic solutions reduce to elementary functions and the corresponding solution in cosmic time was also computed. In Section 6, we considered each of .t/ D D1 =a.t/ Q 3 C D2 =a.t/ Q 4 , .t/ D D=a.t/ Q 3 and .t/ D D=a.t/ Q 4

292

JENNIE D’AMBROISE

with various initial conditions and obtained solutions in terms of Weierstrass functions for general curvature k 0 and constants D1 ; D2 ; D > 0. By considering these solutions restricted to certain D-values (again, parameterized by modulus 0 < k < 1), we wrote a./ equivalently in terms of Jacobi elliptic and theta functions. In current joint work with Floyd Williams [D’Ambroise and Williams 2010], we have seen that elliptic functions also appear in the presence of a scalar field .t/, for both the FRLW and Bianchi I d-dimensional cosmological models with a ¤ 0 and with a similar density function scaling in inverse proportion to a.t/. Q There we note that the equations of each of these cosmological models can be rewritten in terms of a generalized Ermakov–Milne–Pinney differential equation [Lidsey 2004; D’Ambroise and Williams 2007], a type which the square root of the second moment of the wave function of the Bose–Einstein condensate (BEC) also satisfies. On the cosmological side of the FRLW-BEC correspondence, imposing an equation of state .t/ D wp .t/ (w constant) on the density .t/ and pressure p .t/ of the scalar field .t/ allows one to obtain the differential equation for an elliptic function on the side of the BECs.

References [Abdalla and Correa-Borbonet 2002] E. Abdalla and L. Correa-Borbonet, “The elliptic solutions to the Friedmann equation and the Verlinde’s maps”, preprint, 2002. arXiv:hep-th/0212205. [Aurich and Steiner 2001] R. Aurich and F. Steiner, “The cosmic microwave background for a nearly flat compact hyperbolic universe”, Monthly Not. Royal Astron. Soc. 323:4 (2001), 1016–1024. Also see arXiv:astro-ph/0007264. [Aurich et al. 2004] R. Aurich, F. Steiner, and H. Then, “Numerical computation of Maass waveforms and an application to cosmology”, preprint, 2004. arXiv:grqc/0404020. [Basarab-Horwath et al. 2004] P. Basarab-Horwath, W. I. Fushchych, and L. F. Barannyk, “Solutions of the relativistic nonlinear wave equation by solutions of the nonlinear Schrödinger equation”, pp. 81–99 in Scientific works of W. I. Fushchych, vol. 6, 2004. [Biermann 1865] G. G. A. Biermann, Problemata quaedam mechanica functionum ellipticarum ope soluta, Dissertatio Inauguralis, Friedrich-Wilhelm-Universität, 1865. Available at http://edoc.hu-berlin.de/ebind/hdiss/BIERPROB PPN313151385/XML. [Byrd and Friedman 1954] P. F. Byrd and M. D. Friedman, Handbook of elliptic integrals for engineers and physicists, Grundlehren der math. Wissenschaften 67, Springer, Berlin, 1954.


293

[D’Ambroise and Williams 2007] J. D’Ambroise and F. L. Williams, “A non-linear Schrödinger type formulation of FLRW scalar field cosmology”, Int. J. Pure Appl. Math. 34:1 (2007), 117–127. [D’Ambroise and Williams 2010] J. D’Ambroise and F. Williams, “A dynamic correspondence between FRLW cosmology with cosmological constant and Bose– Einstein condensates”, to appear. [Greenhill 1959] A. G. Greenhill, The applications of elliptic functions, Dover, New York, 1959. [Kharbediya 1976] L. I. Kharbediya, “Some exact solutions of the Friedmann equations ˇ 53 (1976), 1145–1152. with the cosmological term”, Astronom. Z. [Kraniotis and Whitehouse 2002] G. V. Kraniotis and S. B. Whitehouse, “General relativity, the cosmological constant and modular forms”, Classical Quantum Gravity 19:20 (2002), 5073–5100. arXiv:gr-qc/0105022. [Lidsey 2004] J. Lidsey, “Cosmic dynamics of Bose–Einstein condensates”, Classical and Quantum Gravity 21 (2004), 777–785. arXiv:gr-qc/0307037. [Reynolds 1989] M. J. Reynolds, “An exact solution in nonlinear oscillations”, J. Phys. A 22:15 (1989), L723–L726. Available at http://stacks.iop.org/0305-4470/22/L723. [Whittaker and Watson 1927] E. T. Whittaker and G. N. Watson, A course of modern analysis, Cambridge Mathematical Library, Cambridge University Press, Cambridge, 1927. J ENNIE D’A MBROISE D EPARTMENT OF M ATHEMATICS AND S TATISTICS L EDERLE G RADUATE R ESEARCH T OWER U NIVERSITY OF M ASSACHUSETTS A MHERST, MA 01003-9305 U NITED S TATES [email protected]


Integrable systems and 2D gravitation: How a soliton illuminates a black hole SHABNAM BEHESHTI

1. Introduction The interlacing of number theory with modern physics has a long and fruitful history. Indeed, the inital seeds were sown by Riemann himself, from paving the way to the Einstein equations with his introduction of the curvature tensor to setting the stage for quantum correction to black hole entropy with his careful study of the zeta function. This note indicates several elegant connections between two-dimensional gravitation, an eigenvalue problem of interest, extended objects (1D bosonic strings), and zeta regularization in 2D quantum gravity; we also note the presence of modular forms when possible and connect our results to the classical three-dimensional theory. It is our aim to find points of tangency with themes from the 2008 MSRI Summer School on Zeta and Modular Physics and motivate the reader for further exploration.

2. JT Gravitation: A simple 2D metric-scalar field theory Consider the vacuum Einstein equations with vanishing cosmological constant R Rij gij D 0; 1 i; j n (1) 2 where Rij and R denote the Ricci tensor and scalar curvature, respectively, and the solution .M n ; g/ is an n-dimensional Riemannian manifold with metric tensor g D gij . A simple calculation shows that for dim M D n D 2, Equation (1) is trivially satisfied. Thus, to make a meaningful interpretation The title of this communication is both in reference and in honour of Geoffrey Mason’s lecture Vertex operators and arithmetic: how a single photon illuminates number theory, at the 2004 ICMS Conference on Moonshine Conjectures and Vertex Algebras.

295

296

SHABNAM BEHESHTI

of these one is compelled to modify the Einstein-Hilbert acp R field equations, n tion M n R.g/ jdet gjd x from which they arise [40]. Introducing a dilaton D .x/ D .x1 ; x2 /, a scalar field in the two variables of the manifold, a potential function V D V .y/, and a nonzero constant m, the modified action becomes Z p R.g/ m2 V ı jdet gj d 2 x: (2) M2

Standard variational principles yield a corresponding set of field equations R.g/ m2 .V 0 ı /D0; m2 ri rj gij .V ı /D0: 1 i; j 2 (3) 2 Here, ri rj denotes the Hessian of the field computed with respect to the metric gij [9]. The first equation in (3) is referred to as the Einstein equation of the system, with the remainder being called equations of motion for the dilaton . In contrast to (1), solutions to the modified two-dimensional model consist of a metric-dilaton pair .g; /. This toy model has proved useful in understanding several key problems of interest, including - relating exact solutions of system (3) to nonlinear equations having known special function solutions [5; 6; 41; 44], - understanding the statistical origin for black hole entropy [20; 30], - studying the endpoint of gravitational collapse [14; 21], - examining the thermodynamics of black hole solutions in two and three dimensions [11; 12; 37], - comparing 2D string gravity with higher dimensional counterparts [28; 45], - providing a stepping stone for finding a consistent theory of quantum gravity (e.g. computing a one-loop effective action in a 2D model [16; 18]). To better understand (3), we will assume the potential takes the form V .y/ D 2y until otherwise indicated. Independently studied by Jackiw [25] in the context of Liouville theory and Teitelboim [36] in the context of Hamiltonian dynamics, this specific case of the action is known as the JT action having JT field equations. Notice the Einstein equation of this system is a constant curvature condition on the manifold, namely R.g/ 2m2 D 0. Having reduced the problem considerably with this choice of potential, we shall state a few solutions without proof. E XAMPLE 1. Let .x1 ; x2 / D .T; r /. If one makes the simplification .T; r / D .r /, then the remaining field equations may be solved to find a static solution to the JT field equations: 2 dsbh D .M

m2 r 2 / d T 2

1 dr 2 ; M m2 r 2

bh .T; r / D mr;

(4)

INTEGRABLE SYSTEMS AND 2D GRAVITATION

297

for M a constant. Excluding the surface D 0, one may show that the Penrose diagram of this metric is identical to a two-dimensional section of the Schwarzschild black hole [1; 11; 27]. We use this to justify using the subscript 2 as a black hole metric. “bh” in (4) and to refer to dsbh On the other hand, setting .x1 ; x2 / D .x; t/ and making the metric ansatz g

ds 2 D cos2

u.x; t/ 2 dx 2

sin2

u.x; t/ 2 dt 2

(5)

for an arbitrary function u D u.x; t/, one finds the Einstein equation in (3) is satisfied if and only if 4u D m2 sin u;

(6)

i.e., u solves the elliptic sine-Gordon equation. This well-studied nonlinear partial differential equation is completely integrable in the sense that it has infinitely many conservation laws, a Lax formulation, a Backlünd Transformation and can be successfully treated with the inverse scattering technique [26; 32; 35; 48]. Consequently, equation (6) falls into a special class of nonlinear equations possessing soliton solutions, or localised wave solutions which maintain their shape and velocity upon collisions. Solving system (3) thus reduces to fixing a soliton solution u of (6) in the metric ansatz above and solving the equations of motion for the dilaton D .x; t/. We thus use the subscript “sol” and refer to 2 dssol as a soliton metric. E XAMPLE 2. Set u to be the simplest nontrivial solution to (6), namely the kink soliton u.x; t/ D 4 arctan e m.x vt /=a , with a; v constants such that a2 D 1 C v 2 . Then a solution of (3) is given by 2 dssol D cos2

u 2 dx 2

sin2

u 2 dt 2

sol .x; t/ D a sech

m .x a

vt/:

(7)

E XAMPLE 3. Choosing a slightly more complicated solution to (6), the oscillating kink-antikink soliton u D u.x; t/ D 4 arctan

v sinh amx ; a cos vmt

with a and v as before, one may verify the pair 2 dssol D cos2

u u 2 dx sin2 dt 2 ; 2 2

sol .x; t/ D

4v 2 am sin vmt sinh amx a2 cos2 vmt Cv 2 sinh2 amx

(8)

solves system (3). Further details as to the derivation of these two examples may be found in [5; 44].

298

SHABNAM BEHESHTI

2 2 Since one expects the two dimensional metrics dsbh and dssol to be locally equivalent, it is reasonable to pose whether it is possible to find an explicit 2 2 correspondence between the solution spaces .dsbh ; bh / and .dssol ; sol /. When 2 M D v in (4), an explicit map is known between the black hole metric and the kink soliton metric described in Example 2 [44]. PDE conditions for a general mapping .x; t/ were later found, establishing a correspondence between 2 .dssol ; .x; t// for an arbitrary solution u.x; t/ of (6) and a generalised black 2 ; .T; r / D mr / of the form hole solution .dsbh 2 dsbh D

jr j2sol ı =m2 d T 2 C m2 = jr j2sol ı dr 2 :

The notation jr j2sol denotes the length of the gradient of with respect to the 2 soliton metric dssol and .T; r / D .x; t/ 1 . To further elevate the status of the dilaton, it is also worth noting that plays a crucial role in determining the geometry of the two-dimensional black hole, as the Killing vectors are known once is given [5; 20]. Remarkably, the mappings ; constructed in [5; 6] turn out to be isometries. Specifically, they are transformations of the solution spaces of the field equations defined by the Laplace Beltrami operators of the soliton and black hole metrics.

3. Application of special functions to JT theory 3.1. Illuminating an eigenvalue problem. We rephrase the final statment of the last section in a particularly useful way. If f D f .T; r /, then a mapping satisfying the PDE system found in [5] satisfies 4sol .f ı / D .4bh f / ı :

(9)

Therefore, 4bh f D f if and only if 4sol F D F with F D f ı. This gives us a mechanism by which to solve eigenvalue problems in soliton coordinates by examining the vastly simpler equation 4bh f D f . Using the separation of variables f .T; r / D e !T h.r /, one obtains a differential equation of hypergeometric type in r .r /h00 .r / C .r /Q .r /h0 .r / C Q .r /h.r / D 0;

(10)

where , and Q and Q are polynomials in r satisfying deg , deg Q 2, deg Q 1. Using the methods in [31; 43], Equation (10) is expressed in canonical form and quantization conditions are derived, from which infinite families of solutions may be written down. In special cases, the final solutions will involve functions such as Jacobi elliptic functions, or Gauss’ hypergeometric functions, among others [5; 6; 43].


299

E XAMPLE 4. Under the mentioned separation of variables, the eigenvalue problem 4bh f D f reduces to B.r /h00 .r /

2m2 rB.r /h0 .r / C .2m2 .B.r /

/h.r / D 0;

where B.r / D M m2 r 2 . Setting D ! 2 , it is possible to reduce the equation 00 0 to one of the form p r .1 r /v .r / C Œ C r .˛ C ˇ C 1/v .r / ˛ˇv.r / D 0, with m0 D 1=.m M /, ˛ D 2 C j!jm0 , ˇ D 1 C j!jm0 , D 1 C j!jm0 and v.r / D 12 h.m2 m0 r C 1/. The equation is now in the standard form of Gauss’ hypergeometric equation, having as solutions generalised hypergeometric functions F.˛; ˇ; I r /: 1 X .˛/n .ˇ/n n F.˛; ˇ; I z/ D z . /n

forjzj < 1; ¤ Z0 ;

(11)

nD0

and where .a/n WD .a C n/= .a/ D a.a C 1/.a C 2/ .a C n 1/ is the Pochhammer symbol [22; 43]. Thus, one solves p the eigenvalue problem as f .T; r / D e !T h.r /, with h.r / D F.˛; ˇ; I mr=.2 M / C 1/, and ˛; ˇ; defined above. It is important to note that properties of this F (e.g., Saalschütz Theorem, Dougall-Ramanujan identity, etc.), and consequently properties of the solution f are intimately dependent on number theoretic and complex analytic results of the Gamma function [15; 34; 39]. Interestingly, hypergeometric differential equations also lend themselves to the study zero-weight modular forms and vertex operator algebras [38; 47]. 3.2. Solitons and black hole entropy. A second way in which one may examine the interplay between the two-dimensional black hole and the soliton gauge, is by computing quantities of physical interest, such as entropy. In explicating 2 2 a correspondence between dsbh and dssol , one finds a relationship between the black hole mass M and several soliton parameters (the constants a; v in (7), for instance). In particular, we find the M is nonnegative in all the cases studied. More, [20; 27] compute the ADM energy p (mM=2G, with G=Newton’s coupling constant), Hawking temperature (m M =2) and associated Bekensteinp 2 Hawking entropy .2 M =G/ for the black hole solutions dsbh given in (4), so each of these quantities may be expressed in terms of the soliton parameters as well. The physical interpretation of these correspondences is still under investigation. Attempts have also been made to recover the asymptotic behaviour of the entropy using N -solitons and partition functions [19]. We shall outline a fairly speculative argument by Gegenberg and Kunstatter here with the hope that further discussion can shed light on the matter. Takhtadjan and Faddeev [35] P compute the total energy for an N -soliton to be E D jND1 m2 =ˇ 2 C pj2 2 , where pj is the canonical momentum of the j th wave packet and ˇ is a constant.

300

SHABNAM BEHESHTI

The rest energy of the state is thus E0 D N m=ˇ. The claim is that “degeneracy of the state comes from the fact that the wave packets of an N -soliton are indistinguishable”; that is, degeneracy is the number of different ways to write N as a sum of non-negative integers. The Hardy–Ramanujan partition function p.N / counts precisely this value and is given asymptotically by p.N /

p 1 p eK N ; 4N 3

p where K D p 2=3; see [23]. Thus, for large N , the entropy grows as S p log p.N / N E0 , up to order one multiplicative constant factors; this coincides with the Bekenstein-Hawking entropy stated above, found in [3; 20]. We remark that in order to make any of the above discussion rigourous, it is first necessary to make an argument which will include solutions of the sine-Gordon equation which do not fit the form of an N -soliton (e.g., breathers, non-soliton solutions). Furthermore, black hole energy has not been proven to be given by the rest energy of the N -soliton solution. We remark that the partition function can be cast as a special case of the Rademacher-Zuckerman formula for the coefficients of a modular form of negative weight - 12 [33]. It is possible that a modular forms perspective will clarify these points.

4. Other two-dimensional considerations: Strings and quantum gravity Two-dimensional theory is not restricted to the study of the JT field equations, of course. We touch upon two possible directions of exploration by altering the potential function V appearing in the action (2) and consequently, the resulting field equations (3). 4.1. 1D bosonic strings. If we now assume V .y/ D y ˛ , the original model not only encompasses the JT Theory ( D 2, ˛ D 1), but several other gravitational theories of interest, including string-inspired gravity and spherically symmetric gravity as well. We shall only discuss the first of these two. Let ˛ D 0 so that V .y/ D 0. Correspondingly, the action appearing in (2) is modified to Z p I Œg; D R.g/ m2 jdet.g/j d 2 x; (12) M2

and first field equation becomes R.g/ D 0. Thus, the metric ansatz in (5) gives rise to the harmonicity condition 4u D 0, rather than the sine-Gordon equation. In this sense, the integrable systems content of the field equations changes qualitatively. However, under the conformal change of coordinates gO D ge ' , one


301

may, in fact, recover the classical Polyakov (bosonic) string action [7; 28; 29] Z p I Œg; O '; ˇ D R.g/ O 4jr'j2gO C ˇ e 2' jdet gj O d 2 x; (13) M2

for ˇ D m2 and D e 2' . The physical and geometric role of the dilaton appears in a new context, as the square of the conformal factor. Although the two actions are related, black hole solutions exist in the string model having vastly different geometry than the static Schwarzschild-type case we have discussed. An example of this follows. E XAMPLE 5. Consider the target space action given in [28] Z p R.g/ 4jr'j2 C jrT j2 C V .T / e 2' jdet gj d 2 x; (14) S.g; '; T / D M2

where g is a two-dimensional metric, ' and T are scalar fields, known as the dilaton field and the tachyon field, respectively, and V is a polynomial potential satisfying V .0/ D 0. Supposing the tachyon field vanishes and '.T; r / D r for some constant , the field equations reduce to an inhomogeneous second order ODE, which yield the metric-dilaton solution ds 2 D .1

ae Qr / d T 2 C

1

1 dr 2 ae Qr

'.T; r / D

Q r; 2

(15)

where Q2 D C is related to the central charge. The asymptotic and topological behaviour of this solution is clearly not of Schwarzschild type. Investigations of this example and string theories in general are detailed in [2; 28; 45]. In connection to the previous section, we comment that the parition function p.N / has been used to count the microstates of a bosonic string; for further details, see [14; 46] and references therein. 4.2. From classical to quantum gravity: Zeta regularization. In a dimensionally reduced model, it is often possible to exactly compute various quantities of interest, both classically and quantum mechanically. We mention the value of two-dimensional models in the context of quantum gravity and zeta functions. Elizalde and Odintsov consider the action Z p 1 R R C jdet gj d 2 x (16) M2 of induced two-dimensional gravity on the background M 2 D R1 S1 ; R D 1 R.g/ is the scalar curvature, is the resolvent operator, and is a constant [16; 18]. Upon consideration of the one-loop gauge-independent effective action, the effective potential V is computed via regularization. Derived in terms of the

302

SHABNAM BEHESHTI

differential operator (see [17]) and a particular constant ˇ, the zeta function is given by Z C1 s s=2 S 1 X 2 2 n D C m2 d k: (17) Cm2 k C ˇ 2 0 nD 1 p Defining the variables x D =4.2 a/ and y D R x, with a Dconstant, the effective potential is computed as p y 1 1 C F.y/ ; (18) V D x 8.2 a/y C .1 ln x/ 8 4 24y for 1 1 X .16/ F.y/ D 4 k!

k

kD0

y

k

1 2

k Y j D1

2

4 .2j 1/

1 X

n

k

3 2 e 2 ny

:

nD1

From this, a minimum for V is found and the authors conclude the compactification is stable [18]. The result is in marked contast to multidimensional quantum gravity on Rd S1 , which is known to be one-loop unstable [10; 24].

5. Relation to the 3D BTZ black hole Two-dimensional models are studied with the ultimate goal of understanding higher dimensional theories. Naturally, we would like to connect the dilaton theory to higher dimensions in an explicit fashion. The Einstein Equations, arising from the classical Einstein-Hilbert action from the first section have also been examinined for three dimensions via Z p .R.g/ 2/ jdet.g/jd 3 x; (19) M3

with x D .x1 ; x2 ; x3 / and a constant. One solution of particular interest is the black hole metric discovered by Bañados, Teitelboim and Zanelli 2 1 2 2 2 dsBT dr 2 C r 2 N .r / d T C d ; (20) Z D N .r / d T C 2 N .r / where x D .T; r; /, N .r / D r 2 M C J 2 =.4r 2 / and N .r / D J =.2r /; see [4]. The constants M and J correspond to the mass and angular momentum of the black hole, respectively. The field equations afforded by the threedimensional case have been carefully studied, with the geometry and physics of the BTZ black hole outlined in [1; 3; 4; 11]. We notice an immediate relationship between the BTZ black hole and the JT black hole from Section 1. Keeping the -coordinate constant and setting J D 0, D m2 , the two-dimensional metric in (4) is recovered as a static slice of (20). This motivates the following


303

discussion: let x D .x1 ; x2 ; / and impose axial symmetry on the 3D metric g as ds 2 D hij .x/ Q dxi dxj C .x/ Q d 2 , for xQ D .x1 ; x2 /, 1 i; j 2. Then (19) reduces to a two-dimensional action from which the JT field equations arise Z p Q (21) I Œg; D .R.h/ 2m2 / .x/ Q jdet.h/j d 2 x; where R.h/ is the scalar curvature of the two dimensional metric h D hij .x1 ; x2 / and .x/ Q D .x1 ; x2 / is the dilaton, as before; compare with (2) for V .y/ D 2y. In this way, the scalar field can be viewed as a radius along the direction of symmetry (the direction) of the the surface defined by the metric h. Clearly then, the soliton content of the three-dimensional case can be considered, as well as pertinent questions on the presence of exact solutions involving special functions and physical quantities of interest. In this context, modular forms of negative weight also appear. The Rademacher-Zuckerman formula asymptotically yields the Cardy entropy formula of conformal field theory and as a special case, the statistical derivation of the Bekenstein-Hawking entropy of the BTZ black hole [8; 13; 14]. Further, quantum correction to entopy can be realised as a deformation of zeta and thus close connections between zeta functions and BTZ black hole thermodynamics have been suggested [42]. It is an interesting question whether the integrability structure in two dimensions sheds any light on the three dimensional case. Such avenues are currently under exploration and will be discussed in a future communication.

6. Conclusion In the context of classical two-dimensional gravitation, we have only touched upon the possible mergers of pure mathematics with black hole physics and cosmology. For further exploration, consult the references.

References [1] Ana Achúcarro and Miguel E. Ortiz. Relating black holes in two and three dimensions. Phys. Rev. D (3), 48(8):3600–3605, 1993. [2] Sergio Albeverio, Jürgen Jost, Sylvie Paycha, and Sergio Scarlatti. A mathematical introduction to string theory, volume 225 of London Mathematical Society Lecture Note Series. Cambridge University Press, Cambridge, 1997. Variational problems, geometric and probabilistic methods. [3] Máximo Bañados, Marc Henneaux, Claudio Teitelboim, and Jorge Zanelli. Geometry of the 2 C 1 black hole. Phys. Rev. D (3), 48(4):1506–1525, 1993. [4] Máximo Bañados, Claudio Teitelboim, and Jorge Zanelli. Black hole in threedimensional spacetime. Phys. Rev. Lett., 69(13):1849–1851, 1992.

304

SHABNAM BEHESHTI

[5] Shabnam Beheshti. Solutions to the dilaton field equations with applications to the soliton-black hole correspondence in generalised JT gravity. Ph.D. Thesis, University of Massachusetts, Amherst, MA (2008). Department of Mathematics. [6] Shabnam Beheshti and Floyd L. Williams. Explicit soliton–black hole correspondence for static configurations. J. Phys. A, 40(14):4017–4024, 2007. [7] A. Belavin and A. Polyakov. Metastable states of two-dimensional isotropic ferromagnets. JETP Letters, 22:245–247, (1975). [8] Danny Birmingham and Siddhartha Sen. Exact black hole entropy bound in conformal field theory. Phys. Rev. D (3), 63(4):047501, 2001. [9] William M. Boothby. An introduction to differentiable manifolds and Riemannian geometry, volume 120 of Pure and Applied Mathematics. Academic Press Inc., Orlando, FL, second edition, 1986. [10] I. L. Buchbinder, P. M. Lavrov, and S. D. Odintsov. Unique effective action in Kaluza-Klein quantum theories and spontaneous compactification. Nuclear Phys. B, 308(1):191–202, 1988. [11] Mariano Cadoni. 2D extremal black holes as solitons. Phys. Rev. D (3), 58(10): 104001, 1998. [12] Mariano Cadoni and Salvatore Mignemi. Nonsingular four-dimensional black holes and the Jackiw-Teitelboim theory. Phys. Rev. D (3), 51(8):4319–4329, 1995. [13] John L. Cardy. Operator content of two-dimensional conformally invariant theories. Nuclear Phys. B, 270(2):186–204, 1986. [14] S. Carlip. Logarithmic corrections to black hole entropy, from the Cardy formula. Classical Quantum Gravity, 17(20):4175–4186, 2000. [15] J. Dougall. On Vandermonde’s theorem and some more general expansions. Proc. Edinburgh Math. Soc., 25:114–132, 1907. [16] E. Elizalde and S. D. Odintsov. Spontaneous compactification in 2D induced quantum gravity. Modern Phys. Lett. A, 7(26):2369–2376, 1992. [17] E. Elizalde and A. Romeo. Regularization of general multidimensional Epstein zeta-functions. Rev. Math. Phys., 1(1):113–128, 1989. [18] Emilio Elizalde. Ten physical applications of spectral zeta functions, volume 35 of Lecture Notes in Physics. New Series m: Monographs. Springer-Verlag, Berlin, 1995. [19] J. Gegenberg and G. Kunstatter. From two-dimensional black holes to sine-Gordon solitons. In Solitons (Kingston, ON, 1997), CRM Ser. Math. Phys., pages 99–106. Springer, New York, 2000. [20] J. Gegenberg, G. Kunstatter, and D. Louis-Martinez. Classical and quantum mechanics of black holes in generic 2-D dilaton gravity. In Heat kernel techniques and quantum gravity (Winnipeg, MB, 1994), volume 4 of Discourses Math. Appl., pages 333–346. Texas A & M Univ., College Station, TX, 1995.


305

[21] J. Gegenberg, G. Kunstatter, and D. Louis-Martinez. Observables for two-dimensional black holes. Phys. Rev. D (3), 51(4):1781–1786, 1995. [22] I. S. Gradshteyn and I. M. Ryzhik. Table of integrals, series, and products. Elsevier/Academic Press, Amsterdam, seventh edition, 2007. Translated from the Russian, Translation edited and with a preface by Alan Jeffrey and Daniel Zwillinger, With one CD-ROM (Windows, Macintosh and UNIX). [23] G. H. Hardy. Ramanujan: twelve lectures on subjects suggested by his life and work. Chelsea Publishing Company, New York, 1959. [24] S. R. Huggins, G. Kunstatter, H. P. Leivo, and D. J. Toms. The Vilkovisky-DeWitt effective action for quantum gravity. Nuclear Phys. B, 301(4):627–660, 1988. [25] R. Jackiw. Liouville field theory: a two-dimensional model for gravity? In Quantum theory of gravity, pages 403–420. Hilger, Bristol, 1984. [26] Peter D. Lax. Integrals of nonlinear equations of evolution and solitary waves. Comm. Pure Appl. Math., 21:467–490, 1968. [27] José P. S. Lemos and Paulo M. Sá. Black holes of a general two-dimensional dilaton gravity theory. Phys. Rev. D (3), 49(6):2897–2908, 1994. [28] Gautam Mandal, Anirvan M. Sengupta, and Spentra R. Wadia. Classical solutions of 2-dimensional string theory. Modern Phys. Lett. A, 6(18):1685–1692, 1991. [29] Robert Marnelius. Canonical quantization of Polyakov’s string in arbitrary dimensions. Nuclear Phys. B, 211(1):14–28, 1983. [30] A. J. M. Medved. Quantum-corrected entropy for .1 C 1/-dimensional gravity revisited. Classical Quantum Gravity, 20(11):2147–2156, 2003. [31] Arnold F. Nikiforov and Vasilii B. Uvarov. Special functions of mathematical physics. Birkhäuser Verlag, Basel, 1988. A unified introduction with applications, Translated from the Russian and with a preface by Ralph P. Boas, With a foreword by A. A. Samarski˘ı. [32] S. Novikov, S. V. Manakov, L. P. Pitaevski˘ı, and V. E. Zakharov. Theory of solitons. Contemporary Soviet Mathematics. Consultants Bureau [Plenum], New York, 1984. The inverse scattering method, Translated from the Russian. [33] Hans Rademacher and Herbert S. Zuckerman. On the Fourier coefficients of certain modular forms of positive dimension. Ann. of Math. (2), 39(2):433–462, 1938. [34] L. Saalschütz. Eine summationsformel. Z. fA˜ 14 r Math. u. Phys., 35:186–188, 1890. [35] L. A. Takhtadzhyan and L. D. Faddeev. Essentially nonlinear one-dimensional model of classical field theory. Theoretical and Mathematical Physics, 22:1046– 1057, February 1975. [36] Claudio Teitelboim. The Hamiltonian structure of two-dimensional space-time and its relation with the conformal anomaly. In Quantum theory of gravity, pages 327– 344. Hilger, Bristol, 1984.

306

SHABNAM BEHESHTI

[37] Jennie Traschen. An introduction to black hole evaporation. In Mathematical methods in physics (Londrina, 1999), pages 180–208. World Sci. Publ., River Edge, NJ, 2000. [38] Hiroyuki Tsutsumi. Modular differential equations of second order with regular singularities at elliptic points for SL2 .Z/. Proc. Amer. Math. Soc., 134(4):931–941, 2006. [39] N. Ja. Vilenkin. Special functions and the theory of group representations. Translated from the Russian by V. N. Singh. Translations of Mathematical Monographs, Vol. 22. American Mathematical Society, Providence, R. I., 1968. [40] Robert M. Wald. General relativity. University of Chicago Press, Chicago, IL, 1984. [41] Floyd Williams. On solitons, non-linear sigma-models and two-dimensional gravity. Fourth International Winter Conference on Mathematical Physics (Rio de Janeiro, 2005). PoS(WC2004)003. http://pos.sissa.it, [42] Floyd Williams. Remarks on the btz instanton with conical singularity. Fifth International Conference on Mathematical Methods in Physics (Rio de Janeiro, 2006). PoS(IC2006)006. http://pos.sissa.it, [43] Floyd Williams. Topics in quantum mechanics, volume 27 of Progress in Mathematical Physics. Birkhäuser Boston Inc., Boston, MA, 2003. [44] Floyd Williams. Further thoughts on first generation solitons and J-T gravity. In Trends in Soliton Research, L. Chen, Ed., pages 1–14. Nova Sci. Publ., New York, 2006. [45] Edward Witten. String theory and black holes. Phys. Rev. D (3), 44(2):314–324, 1991. [46] Donam Youm. Black holes and solitons in string theory. Phys. Rep., 316(1-3):232, 1999. [47] Don Zagier. Modular forms and differential operators. Proc. Indian Acad. Sci. Math. Sci., 104(1):57–75, 1994. K. G. Ramanathan memorial issue. [48] V. E. Zakharov, L. A. Tahtadˇzjan, and L. D. Faddeev. A complete description of the solutions of the “sine-Gordon” equation. Dokl. Akad. Nauk SSSR, 219:1334–1337, 1974. S HABNAM B EHESHTI D EPARTMENT OF M ATHEMATICS 110 F RELINGHUYSEN ROAD RUTGERS U NIVERSITY P ISCATAWAY, NJ 08854-8019 U NITED S TATES [email protected]


Functional determinants in higher dimensions using contour integrals KLAUS KIRSTEN

A BSTRACT. In this contribution we first summarize how contour integration methods can be used to derive closed formulae for functional determinants of ordinary differential operators. We then generalize our considerations to partial differential operators. Examples are used to show that also in higher dimensions closed answers can be obtained as long as the eigenvalues of the differential operators are determined by transcendental equations. Examples considered comprise of the finite temperature Casimir effect on a ball and the functional determinant of the Laplacian on a two-dimensional torus.

1. Introduction Functional determinants of second-order differential operators are of great importance in many different fields. In physics, functional determinants provide the one-loop approximation to quantum field theories in the path integral formulation [21; 48]. In mathematics they describe the analytical torsion of a manifold [47]. Although there are various ways to evaluate functional determinants, the zeta function scheme seems to be the most elegant technique to use [9; 16; 17; 31]. This is the method introduced by Ray and Singer to define analytical torsion [47]. In physics its origin goes back to ambiguities in dimensional regularization when applied to quantum field theory in curved spacetime [11; 29]. For many second-order ordinary differential operators surprisingly simple answers can be given. The determinants for these situations have been related to boundary values of solutions of the operators, see, e.g., [8; 10; 12; 22; 23; 26; 36; 39; 40]. Recently, these results have been rederived with a simple and accessible method which uses contour integration techniques [33; 34; 35]. The main advantage of this approach is that it can be easily applied to general kinds 307

308

KLAUS KIRSTEN

of boundary conditions [35] and also to cases where the operator has zero modes [34; 35]; see also [37; 38; 42]. Equally important, for some higher dimensional situations the task of finding functional determinants remains feasible. Once again closed answers can be found but compared to one dimension technicalities are significantly more involved [13; 14]. It is the aim of this article to choose specific higher dimensional examples where technical problems remain somewhat confined. The intention is to illustrate that also for higher dimensional situations closed answers can be obtained which are easily evaluated numerically. The outline of this paper is as follows. In Section 2 the essential ideas are presented for ordinary differential operators. In Section 3 and 4 examples of functional determinants for partial differential operators are considered. The determinant in Section 3 describes the finite temperature Casimir effect of a massive scalar field in the presence of a spherical shell [24; 25]. The calculation in Section 4 describes determinants for strings on world-sheets that are tori [46; 50] and it gives an alternative derivation of known answers. Section 5 summarizes the main results.

2. Contour integral formulation of zeta functions In this section we review the basic ideas that lead to a suitable contour integral representation of zeta functions associated with ordinary differential operators. This will form the basis of the considerations for partial differential operators to follow later. We consider the simple class of differential operators P WD

d2 C V .x/ dx 2

on the interval I D Œ0; 1, where V .x/ is a smooth potential. For simplicity we consider Dirichlet boundary conditions. From spectral theory [41] it is known that there is a spectral resolution fn ; n g1 nD1 satisfying Pn .x/ D n n .x/;

n .0/ D n .1/ D 0:

The spectral zeta function associated with this problem is then defined by P .s/ D

1 X

n s ;

(2-1)

nD1

where by Weyl’s theorem about the asymptotic behavior of eigenvalues [49] this series is known to converge for Re s > 21 . If the potential is not a very simple one, eigenfunctions and eigenvalues will not be known explicitly. So how can the zeta function in equation (2-1), and in

FUNCTIONAL DETERMINANTS IN HIGHER DIMENSIONS VIA CONTOUR INTEGRALS 309

particular the determinant of P defined via det P D e

0 P .0/

;

be analyzed? From complex analysis it is known that series can often be evaluated with the help of the argument principle or Cauchy’s residue theorem by rewriting them as contour integrals. In the given context this can be achieved as follows. Let 2 C be an arbitrary complex number. From the theory of ordinary differential equations it is known that the initial value problem .P

/u .x/ D 0;

u .0/ D 0;

u0 .0/ D 1;

(2-2)

has a unique solution. The connection with the boundary value problem is made by observing that the eigenvalues n follow as solutions to the equation u .1/ D 0I

(2-3)

note that u .1/ is an analytic function of . With the help of the argument principle, equation (2-3) can be used to write the zeta function, equation (2-1), as Z 1 d P .s/ D d s ln u .1/: (2-4) 2 i d Here, is a counterclockwise contour and encloses all eigenvalues which we assume to be positive; see Figure 1. The pertinent remarks when finitely many eigenvalues are nonpositive are given in [35]. The asymptotic behavior of u .1/ as jj ! 1, namely p sin u .1/ p ; implies that this representation is valid for Re s > 12 . To find the determinant of P we need to construct the analytical continuation of equation (2-4) to a neighborhood about s D 0. This is best done by deforming the contour to enclose 6

-plane

q q q q q q q q q q

Figure 1. Contour used in equation (2-4).

310

KLAUS KIRSTEN

6 -

-plane

q q q q q q q q q q

Figure 2. Contour used in equation (2-4) after deformation.

the branch cut along the negative real axis and then shrinking it to the negative real axis; see Figure 2. The outcome is Z sin s 1 d P .s/ D d s ln u .1/: (2-5) d 0 To see where this representation is well defined notice that for ! 1 the behavior follows from [41] p p sin.i / e u .1/ D p 1 e 2 : p i 2 The integrand, to leading order in , therefore behaves like s 1=2 and convergence at infinity is established for Re s > 21 . As ! 0 the behavior s follows. Therefore, in summary, (2-5) is well defined for 21 < Re s < 1. To shift the range of convergence to the left we add and subtract the leading ! 1 asymptotic behavior of u .1/. The whole point of this procedure will be to obtain one piece that at s D 0 is finite, and another piece for which the analytical continuation can be easily constructed. Given we want to improve the ! 1 behavior without worsening the ! 0 behavior, we split the integration range. In detail we write P .s/ D P;f .s/ C P;as .s/;

(2-6)

where sin s P;f .s/ D

sin s P;as .s/ D

1

Z 0

Z 1

d ln u .1/ d Z sin s 1 d C d s ln u d 1

d

1

s

p .1/2 e

p

;

(2-7)

p

d

s

d e ln p : d 2

(2-8)


By construction, P;f .s/ is analytic about s D 0 and its derivative at s D 0 is trivially obtained, 0 P;f .0/ D ln u

1 .1/

ln u0 .1/

ln u

1 .1/2e

1

D

ln

2u0 .1/ : e

(2-9)

Although the representation (2-8) is only defined for Re s > 21 , the analytic continuation to a meromorphic function on the complex plane is found using Z 1 1 d ˛ D for Re ˛ > 1: ˛ 1 1 This shows that

sin s 1 P;as .s/ D 2 s 1=2

1 ; s

and furthermore 0 P;as .0/ D 1:

Adding up, the final answer reads P0 .0/ D

ln.2u0 .1//:

(2-10)

For the numerical evaluation of the determinant, not even one eigenvalue is needed. The only relevant information is the boundary value of the unique solution to the initial value problem d2 C V .x/ u0 .x/ D 0; u0 .0/ D 0; u00 .0/ D 1: 2 dx General boundary conditions can be dealt with as easily. The best formulation results by rewriting the second-order differential equation as a first-order system in the usual way. Namely, we define v .x/ D du .x/=dx such that the differential equation (2-2) turns into d u .x/ u .x/ 0 1 D : (2-11) V .x/ 0 v .x/ dx v .x/ Linear boundary conditions are given in the form u .0/ u .1/ 0 M CN D ; v .0/ v .1/ 0

(2-12)

where M and N are 2 2 matrices whose entries characterize the nature of the boundary conditions. For example, the previously described Dirichlet boundary conditions are obtained by choosing 1 0 0 0 MD ; ND : 1 0 0 0

312

KLAUS KIRSTEN

In order to find an implicit equation for the eigenvalues like equation (2-3) we .1/ .2/ use the fundamental matrix of (2-11). Let u .x/ and u .x/ be linearly independent solutions of (2-11). Suitably normalized, these define the fundamental matrix ! .1/ .2/ u .x/ u .x/ H .x/ D ; H .0/ D Id22 : .1/ .2/ v .x/ v .x/ The solution of (2-11) with initial value .u .0/; v .0// is then obtained as u .x/ u .0/ D H .x/ : v .x/ v .0/ The boundary conditions (2-12) can therefore be rewritten as u .0/ 0 .M C NH .1// D : v .0/ 0

(2-13)

This shows that the condition for eigenvalues to exist is det.M C NH .1// D 0; which replaces (2-3) in case of general boundary conditions. The zeta function associated with the boundary condition (2-12) therefore takes the form Z 1 d d s ln det.M C NH .1// P .s/ D 2 i d and the analysis proceeds from here depending on M and N . If P represents a system of operators one can proceed along the same lines. Note that we have replaced the task of evaluating the determinant of a differential operator by one of computing the determinant of a finite matrix. The procedure just outlined is by no means confined to be applied to ordinary differential operators only. In fact, the zeta function associated with many boundary value problems allowing for a separation of variables can be analyzed using this contour integral technique. In more detail, starting off with some coordinate system (see [43], for example), eigenvalues are often determined by Fj .j ;n / D 0; where j is a suitable quantum number depending on the coordinate system considered and Fj is a given special function depending on the coordinate system; e.g. for ellipsoidal coordinate systems the relevant special function is the Mathieu function. Continuing along the lines described above, denoting by dj an appropriate degeneracy that might be present, we write somewhat symbolically Z X 1 d P .s/ D dj d s ln Fj ./; (2-14) 2 i d j


the task being to construct the analytical continuation of this object to s D 0. The details of the procedure will depend very much on the properties of the special function Fj that enters. For example, on balls Bessel functions are relevant [4; 6; 7], the spherical suspension [3], or sphere-disc configurations [27; 32], involve Legendre functions, ellipsoidal boundaries involve Mathieu functions etc. For many examples relevant properties of Fj ./ are not available in the literature and need to be derived using techniques of asymptotic analysis [41; 44; 45]. For quite common coordinate systems like the polar coordinates this is not necessary. When the asymptotics is known, the relevant integrals resulting in (2-14) need to be evaluated and closed expressions representing the determinant of partial differential operators are found. Although the remaining sums in general cannot be explicitly performed, the results obtained are very suitable for numerical evaluation.

3. Finite temperature Casimir energy on the ball Let us now apply the above remarks about higher dimensions using the general formalism described in [14]. As a concrete example we consider the finite temperature theory of a massive scalar field on the three dimensional ball. Using the zeta function scheme we have to consider the eigenvalue problem d2 2 P .; x/ E WD C m .; x/ E D 2 .; x/; E (3-1) d 2 where is the imaginary time and xE 2 B 3 WD fxE 2 R 3 j jxj E 1g. We have written 2 for the eigenvalues to avoid the occurrence of square roots in arguments of Bessel functions later on. For finite temperature theory we impose periodic boundary conditions in the imaginary time, .; x/ E D . C ˇ; x/; E where ˇ is the inverse temperature, and for simplicity we choose Dirichlet boundary conditions on the boundary of the ball, ˇ D 0: .; x/ E ˇjxjD1 E The zeta function associated with this boundary value problem is then X P .s/ D 2s ;

(3-2)

and the energy of the system is defined by E WD

1 @ 0 2 .0/; 2 @ˇ P =

(3-3)

314

KLAUS KIRSTEN

where is an arbitrary parameter with dimension of a mass introduced in order to get the correct dimension for the energy. For a full discussion of its relevance in the renormalization process in this model at zero temperature see [5]. That discussion remains completely unchanged at finite temperature and we will put D 1 henceforth. Given the radial symmetry of the problem we separate variables in polar coordinates according to 1 .; r; ; '/ D p e i.2 n=ˇ/ J`C1=2 .!`j r /Y`m .; '/; r with the spherical surface harmonics Y`m .; '/ [20] solving @2 sin2 @' 2 1

1 @ @ sin Y`m .; '/ D `.` C 1/Y`m .; '/; sin @ @

and with the Bessel function J .z/, which is the regular solution of the differential equation [28] 2 d 2 J .z/ 1 dJ .z/ C 1 C J .z/ D 0: z dz dz 2 z2 Imposing the boundary condition on the unit sphere, J`C1=2 .!`j / D 0; determines the eigenvalues. Namely, 2 n 2 2 2 C !`j C m2 ; n`j D ˇ

(3-4)

n 2 Z ; ` 2 N0 ; j 2 N:

(3-5)

This leads to the analysis of the zeta function P .s/ D

1 1 X 1 X X

2 .2` C 1/ pn2 C !`j C m2

s

;

(3-6)

nD 1 `D0 j D1

where we have used the standard abbreviation pn D 2 n=ˇ. The factor 2` C 1 represents the multiplicity of eigenvalues for angular momentum `. The zeroes !`j of the Bessel functions J`C1=2 .!`j / are not known in closed form and thus we represent the j -summation using contour integrals. Starting with equation (3-4) and following the argumentation of the previous section, this gives the identity Z 1 1 X X s d d 2 pn C2 Cm2 ln J`C1=2 ./; (3-7) P .s/ D .2`C1/ d

2 i nD 1 `D0


valid for Re s > 2. The contour runs counterclockwise and must enclose all the solutions of (3-4) on the positive real axis. The next step is to shift the contour and place it along the imaginary axis. As ! 0 we observe that to leading order J ./ =.2 . C 1// such that the integrand diverges in this limit. Therefore, we include an additional factor ` 1=2 in the logarithm in order to avoid contributions coming from the origin. Because there is no additional pole enclosed, this does not change the result. Furthermore we should note that the integrand has branch cuts starting at D ˙i .pn2 C m2 /. Leaving out the n; ` summations for the moment and considering the -integration alone, we then obtain, with D ` C 12 , d 2 d .pn C2 Cm2 / s ln J ./ d

2 i Z 1 d sin s ln k d k .k 2 pn2 m2 / s D p 2 dk pn Cm2

P;n` .s/ WD

Z

I .k/ ; (3-8)

where J .i k/ D e i J . i k/ and I .k/ D e i=2 J .i k/ has been used [28]. The next step is to add and subtract the asymptotic behavior of the integrand in (3-8). The relevant uniform asymptotics, after substituting k D z in the integral, is the Debye expansion of the Bessel functions [1]. We have 1 X e uk .t/ I .z/ p 1C ; k 2 .1 C z 2 /1=4 1

(3-9)

kD1

p p p with t D 1= 1 C z 2 and D 1 C z 2 C ln z=.1 C 1 C z 2 / . The first few coefficients are listed in [1], higher coefficients are immediately obtained by using the recursion [1] 1 ukC1 .t/ D t 2 .1 2

t 2 /u0k .t/ C

1 8

Z 0

t

d .1

5 2 /uk . /;

(3-10)

starting with u0 .t/ D 1. As is clear, all the uk .t/ are polynomials in t. The same holds for the coefficients Dn .t/ defined by X 1 1 X uk .t/ Dn .t/ ln 1 C : k n

kD1

(3-11)

nD1

The polynomials uk .t/ as well as Dn .t/ are easily found with the help of a simple computer program. As we will see below, we need the first three terms

316

KLAUS KIRSTEN

in the expansion (3-11). Explicitly, D1 .t/ D 81 t

5 3 24 t ;

D2 .t/ D

1 2 16 t

D3 .t/ D

25 3 384 t

3 4 5 6 8 t C 16 t ;

(3-12)

531 5 221 7 640 t C 128 t

1105 9 1152 t :

Adding and subtracting these terms in (3-8) allows us to rewrite the zeta function as P .s/ D P;f .s/ C P;as .s/; where Z 1 1 1 s sin s X X P;f .s/ D dz z 2 2 pn2 m2 .2`C1/ p nD 1 pn2 Cm2 = `D0 z e D1 .t/ D2 .t/ D3 .t/ d ln z I .z/ ln p ; (3-13) dz 2 3 2.1Cz 2 /1=4 Z 1 1 1 s sin s X X P;as .s/ D .2`C1/ p dz z 2 2 pn2 m2 2 nD 1 pn Cm2 = `D0 D1 .t/ D2 .t/ D3 .t/ z e d C C ln p C : (3-14) dz 2 3 2.1Cz 2 /1=4 The number of terms subtracted in (3-13) is chosen so that P;f .s/ is analytic about s D 0. The contributions from the asymptotics collected in (3-14) are simple enough for an analytical continuation to be found. Although it would be possible to proceed just with the contribution from inside the ball, in order to make the calculation as transparent and unambiguous as possible (as far as the interpretation of results goes) let us add the contribution from outside the ball. The exterior of the ball, once the free Minkowski space contribution is subtracted, yields the starting point (3-8) with the replacement k I ! k K [5]. In this case the relevant uniform asymptotics is [1] r 1 X e k uk .t/ K .z/ 1C . 1/ ; (3-15) 2 .1 C z 2 /1=4 k kD1

where the notation is as in (3-9). This produces the analogous splitting of the zeta function for the exterior space. Due to the characteristic sign changes in the asymptotics of I and K , adding up the interior and exterior contributions several cancellations take place. As a result, the zeta function for the total space has the form tot .s/ D tot;f .s/ C tot;as .s/


with Z 1 1 1 s sin s X X .2`C1/ p dz z 2 2 pn2 m2 tot;f .s/ D nD 1 pn2 Cm2 = `D0 d ln I .z/K .z/ C ln.2/ C 12 ln.1 C z 2 / 2 2 D2 .t/ ; (3-16) dz Z 1 1 1 s sin s X X tot;as .s/ D .2`C1/ p dz z 2 2 pn2 m2 nD 1 pn2 Cm2 = `D0 d ln.2/ 21 ln.1 C z 2 / C 2 2 D2 .t/ : (3-17) dz By construction, tot;f .s/ is analytic about s D 0 and one immediately finds 0 tot;f .0/ D

1 X 1 X

.2` C 1/ ln I .z/K .z/ C ln.2/

nD 1 `D0

C 21 ln.1Cz 2 /

ˇˇ 2 2 D2 .t/ ˇˇ

(3-18) zD

p

;

pn2 Cm2 =

p with t D 1= 1 C z 2 as defined earlier. Although one could use (3-18) for numerical evaluation, further simplifications are possible. Following [14] we rewrite this expression according to pn2 C m2 pn2 m2 2 1Cz D 1C D 1C 2 1C 2 : (3-19) 2 C pn2 The advantage of the right-hand side is that it can be expanded further for 2 ! 1 or pn2 ! 1 or both. This will allow us to subtract exactly the behavior that makes the double series convergent; the oversubtraction immanent in (3-18) can then be avoided. It is expected that expanding the rightmost factor further for 0 2 C pn2 1 leads to considerable cancellations when combined with tot;as .0/ [14]. We split the asymptotic terms in (3-18) into those strictly needed to make the sums convergent and those that ultimately will not contribute. For example, we expand according to ˇ ˇ ln.1 C z 2 /ˇp 2 pn Cm2 =

pn2 C m2 pn2 m2 D ln 1 C D ln 1 C 2 C ln 1 C 2 2 C pn2 pn2 m2 m2 m2 D ln 1 C 2 C 2 C ln 1 C 2 : C pn2 C pn2 2 C pn2

318

KLAUS KIRSTEN

The first two terms have to be subtracted in (3-18) in order to make the summations convergent. The terms in brackets are of the order O.1=. 2 C pn2 /2 / and even after performing the summations in (3-18) a finite result follows. Thus the first two terms represent a minimal set of terms to be subtracted in (3-18) in order to make thepsums finite. This minimal set of necessary terms will be asym;.1/ .i pn2 Cm2 /. The last two terms can be summedpseparately called ln f` asym;.2/ .i pn2 Cm2 /. yielding a finite answer; they are summarized under ln f` One can proceed along the same lines for all other terms. With the definition asym p ln f` .i pn2 Cm2 / ˇ ˇ D ln.2/ 12 ln.1 C z 2 / C 2 2 D2 .t/ˇ p 2 2 zD pn Cm = p asym;.2/ asym;.1/ p 2 .i pn2 Cm2 / (3-20) D ln f` .i pn Cm2 / C ln f` the splitting is asym;.1/

ln f`

asym;.2/

ln f`

p 1 pn2 1 m2 .i pn2 Cm2 / D ln.2/ ln 1C 2 2 2 2 Cpn2 2 1 5 pn2 3 pn2 1 3 pn2 2 C 2 1C 2 1C 2 1C 2 C ; 8 16 16 1 1 m2 m2 ln 1 C 2 2 C 2 2 2 Cpn2 Cpn 2 1 p2 1 m2 C 2 1C 2n 1C 2 2 Cpn 16 3 pn2 2 m2 1C 2 1C 2 2 8 Cpn 5 p2 3 m2 C 1C 2n 1C 2 2 16 Cpn

(3-21)

p .i pn2 Cm2 / D

1

1

2

1 3

1 :

(3-22)

We have used the given notation for the asymptotics to make a comparison with 0 [14] as easy as possible. With these asymptotic quantities we rewrite tot;f .0/ as 1 X 1 X

0 tot;f .0/ D

nD

C

p p .2` C 1/ ln.I . pn2 Cm2 /K . pn2 Cm2 // 1 `D0 asym;.1/ p 2 ln f` .i pn Cm2 /

1 X 1 X

asym;.2/

.2` C 1/ ln f`

nD 1 `D0

p .i pn2 Cm2 /:

(3-23)

FUNCTIONAL DETERMINANTS IN HIGHER DIMENSIONS VIA CONTOUR INTEGRALS 319 0 Let us next analyze tot;as .0/. To further analyze tot;as .s/, equation (3-17), we use the integrals

Z 1 p 2

du .u2

pn2

m2 /

s

Z 1 p 2

du .u2

pn2

m2 /

s

pn Cm2

pn Cm2

D

d u2 ln 1 C 2 D .m2 C 2 C pn2 / s ; du sin s d u2 1C 2 du

sin.s/

s C N2 N 2

N=2

.s/

2s

1C

pn2 Cm2 2

s

N 2

;

which are the relevant ones after substituting z D u. This shows that tot;as .s/ D

1 P

1 P

p 2 Cm2 s 1C n 2 nD 1 `D0 1 1 P P pn2 Cm2 s 1 2s 1 1 s 1 C 4 2 nD 1 `D0 1 1 P P p 2 Cm2 s 2 2s 1 1 C n 2 C 23 s.sC1/ nD 1 `D0 1 1 P P pn2 Cm2 2s 1 5 1 C s.sC1/.sC2/ 8 2 nD 1 `D0 1

2s

s 3

: (3-24)

To each of these terms we apply the rewriting (3-19). Intermediate expressions are relatively lengthy and we explain details only for the first term. We proceed as for the splitting in (3-21) and (3-22) and write 1 X 1 X nD 1 `D0 1 X 1 X

1

2s

1C

pn2 Cm2 2

s

s pn2 s m2 1 C 2 2 Cpn2 nD 1 `D0 1 X 1 s X m2 m2 D 1 C 1 C s C1 2 Cpn2 2 Cpn2 . 2 C pn2 /s nD 1 `D0 1 X 1 s X m2 m2 D 1C 2 2 1Cs 2 2 Cpn Cpn . 2 C pn2 /s nD 1 D

1

2s

1C

m2 s 2 2 Cpn

`D0

1 1 1 1 1 X X 2 sm2 X X 2 C : 2 2 s 2 2 nD 1 2 nD 1 . C pn / . C pn2 /sC1 `D0

`D0

(3-25)

320

KLAUS KIRSTEN

The first line is seen to be analytic about s D 0. We have subtracted the minimal number of terms to make the sums convergent. The remaining terms represent zeta functions of Epstein type, E .k/ .s; a/ D

1 X 1 X

2

nD 1 `D0

k ; . 2 C a2 n2 /s

(3-26)

the analytical continuation of which is well understood. Performing a Poisson resummation on the n-summation [2; 15; 30] yields p s 12 2 .k/ E .s; a/ D H 2s k 2; 21 a .s/ 1 1 s X X 8 n kC.3=2/ s s 1=2 C n K 2 : (3-27) 1=2 s a .s/asC1=2 `D0

nD1

The first line has poles at s D 21 j , j 2 N0 , and for s D 12 .k C 3/, the second line is analytic for s 2 C . In terms of these Epstein functions, in equation (3-25) we have shown that 1 X 1 X nD

sm2 2 2 s; C E .0/ sC1; ˇ ˇ 2 1 `D0 1 X 1 s X m2 m2 1C 2 2 1 C s 2 2 : (3-28) Cpn Cpn . 2 C pn2 /s nD 1 1

2s

1C

pn2 Cm2 2

s

D

1 .0/ 2E

`D0

Noting from equation (3-27) that E .0/ .s; a/ and E .0/ .sC1; a/ are analytic about s D 0, we get 1 1 d X X 1 ds nD 1 `D0

2s

p 2 Cm2 s 2 2 1C n 2 D 12 E .0/0 0; C 21 m2 E .0/0 1; ˇ ˇ 1 X 1 X m2 m2 ln 1 C 2 2 C 2 2 : (3-29) Cpn Cpn nD 1 `D0

asym;.2/ p 2 The last term on the right cancels the first line from ln f` .i pn Cm2 / in equation (3-22), the remaining terms are easily found from (3-27). One can proceed in exactly this way for the other terms in tot;as .s/; there are asym;.2/ p 2 always terms that cancel with terms from ln f` .i pn Cm2 / in (3-22) and terms expressible using the Epstein type zeta functions given in (3-26). Adding up all contributions, the second line in (3-23) completely cancels and we obtain


the following closed form for the finite-T zeta function: 1 X 1 X p p 0 2 ln I . pn2 Cm2 /K . pn2 Cm2 / C ln. 2 C pn2 / tot .0/ D nD 1 `D0 1 1 5 4 m2 6 2 C 1 C 2 2 Cpn 8 2 Cpn2 . 2 Cpn2 /2 . 2 Cpn2 /3 2 2 1 .0/0 1 2 .0/ E m E 0; C 1; 2 2 ˇ ˇ 2 2 5 .4/ C 34 E .2/ 2; 3; : (3-30) 8E ˇ ˇ From (3-27) it is clear that the Epstein type zeta functions contain zero temperature contributions to the Casimir energy (first line in (3-27)) and exponentially damped contributions for small temperature described by the Bessel functions (second line in (3-27)). As it turns out, the zero temperature contributions from the Epstein type zeta functions in (3-30) all vanish. The remaining zero temperature contributions in (3-30) are found replacing the Riemann sum over n by an integral, Z 1 1 X ˇ dpf .p/: f .n/ 2 1 nD 1 As ˇ ! 0 this shows that 1 0 .0/ ˇ tot D

1 2

Z

1

dp 1

C

1 X

p p 2 ln I . p 2 Cm2 /K . p 2 Cm2 / C ln. 2 Cp 2 /

`D0 m2

2 C p2

1 1 1 8 2 C p2

6 2 5 4 C . 2 Cp 2 /2 . 2 Cp 2 /3

;

(3-31)

from which the Casimir energy (3-3) is trivially obtained. The result is much simpler than previous results given [24; 5] and a numerical evaluation could easily be performed.

4. Functional determinant on a two dimensional torus As our next example let us consider a two dimensional torus S 1 S 1 . For convenience we choose the perimeter of the circles to be 1. The relevant eigenvalue problem to be considered then is @2 @2 P .x; y/ WD .x; y/ D 2 .x; y/; @x 2 @y 2

322

KLAUS KIRSTEN

and we choose periodic boundary conditions .x; y/ D .x C 1; y/;

@ .x; y/ @ .x C 1; y/ D ; @x @x

.x; y/ D .x; y C 1/;

@ .x; y/ @ .x; y C 1/ D : @y @y

The eigenfunctions and eigenvalues clearly are m;n .x; y/ D e

2 imx

e

2 i ny

2 D .2/2 .m2 C n2 /;

;

n; m 2 Z :

The related zeta function then reads P .s/ D .2/

X

2s

.m2 C n2 / s I

(4-1)

2

.m;n/2Z nf.0;0/g

note that the zero mode m D n D 0 has to be omitted in the summation to make P .s/ well defined. The zeta function in equation (4-1) is an Epstein zeta function and P0 .0/ can be evaluated using the Kronecker limit formula [18; 19]. Here, we apply the contour approach previously outlined which simplifies the calculation. Instead of using the fact that the eigenvalues can be given in closed form, we proceed differently. We say that 2 D .2/2 .n2 C k 2 /;

n 2 Z;

where k is determined as a solution to the equation e ik

e

ik

D 0:

(4-2)

Of course, solutions are given by k 2 Z and the correct eigenvalues follow. Using equation (4-2) determining the eigenvalues in the way we have used equations (2-3) and (3-4), the zeta function can be represented as the contour integral ik 1 Z X dk e e ik 2s 2 2 s d P .s/ D 4 .2/ .n C k / ln dk 2 i k

2 i nD1

C 4.2/

2s

R .2s/:

(4-3)

The last term represents the part where one of the two indices m or n is zero in equation (4-1). The first line represents the remaining contributions. The factor of 4 is a result of summing over positive n only and because the contour is supposed to enclose positive integers only. The reason that we have used e ik e ik 2 i k


instead of equation (4-2) is that e ik e ik D 1; 2 i k k!0 lim

which will allow us to shift the contour in a way as to include the origin; see the discussion below equation (3-7). Let us evaluate the contour integral Z dk d e ik e ik n .s/ D .2/ 2s .n2 C k 2 / s ln : dk 2 i k

2 i p Substituting k D z and deforming the contour to the negative real axis along the lines described previously, an intermediate result is p p Z 1 e z e z 2s sin s 2 s d : (4-4) n .s/ D .2/ dz.z n / ln p dz 2 z n2 From the behavior of the integrand as z ! 1 and z ! n2 this representation is seen to be valid for 21 < Re s < 1. In order to construct the analytical continuation to a neighborhood of s D 0 we note that e

p z

e p

p

z

2 z

p

e z D p 1 2 z

2

e

p z

:

We therefore write p

d e z dz.z n2 / s ln p dz 2 z n2 Z 1 sin s C .2/ 2s dz.z n2 / n2

sin s n .s/ D .2/ 2s

Z

1

The first line is evaluated using Z 1 .z n2 / dz p z n2 Z 1 .z n2 / dz z n2 With the identity [28] n .s/ D

sin s

2sC1 n 1 4 .2/

n1 2s D p .1

s

D .1 1 2

1 2s

p

s

1 2

s/

d ln 1 dz

e

2

e

2

p

z

:

z

:

Cs ;

n 2s : sin s 1 , this produces .s/

s/ D Cs

2s 1 n 2s 2 .2/

.s/

C .2/

s

2s sin s

Z

1

dz.z n2

n2 /

s

d ln 1 dz

p

324

KLAUS KIRSTEN

This is the form that allows the sum over n to be (partly) performed and it shows that P .s/ D 4.2/

2s sin s

2.2/

1 Z X

dz.z

n2 /

R .2s/ C .2/

2sC1

n2

nD1 2s

1

s

d ln 1 dz p

1 2

2

e

Cs

p

z

.s/

R .2s

1/:

(4-5)

This form allows for the evaluation of P0 .0/. From known elementary properties of the -function and the zeta function of Riemann [28] we obtain ˇ d ˇˇ 0 .2/ 2s R .2s/ D 2 ln.2/R .0/ C 2R .0/ D 0; ˇ ds sD0 ˇ 1 d ˇˇ 2 Cs 2sC1 R .2s 1/ D : .2/ p ds ˇsD0 3 .s/ The first line in (4-5) is also easily evaluated because ˇ d ˇˇ 4.2/ ds ˇsD0

2s sin s

D4

1 Z X

nD1

n2 /

dz.z

n2

nD1

1 Z X

1

1

d ln 1 dz

dz

n2

e

s

d ln 1 dz

2

p

z

2

e

D 4

p

1 X

z

ln 1

e

2 n

:

nD1

This can be reexpressed using the Dedekind eta function . / WD e i=12

1 Y

1

e 2 i n

nD1

for 2 C , Re > 0. The relation relevant for us follows by setting D i : 1

ln j.i /j4 D

X C4 ln 1 3

e

2 n

:

nD1

Adding up all contributions for P0 .0/, the final answer reads P0 .0/ D

3

4

1 X

ln 1

e

2 n

D

ln j.i /j4 ;

nD1

in agreement with known answers; see [46; 50], for example.

(4-6)


5. Conclusions We have shown that contour integrals are very useful and effective tools for the evaluation of determinants of differential operators. Although the results look very simple only in one dimension — see equation (2-10) — , for particular configurations also in higher dimensions closed answers can be found suitable for numerical evaluation, as in equations (3-31) and (4-6). Here we have provided answers only for the torus and a spherically symmetric situation. But the same ideas should apply when separability of the partial differential equations in other coordinate systems is possible. Results in this direction will be presented elsewhere.

Acknowledgement Kirsten acknowledges support by the NSF through grant PHY-0757791. Part of the work was done while the author enjoyed the hospitality and partial support of the Department of Physics and Astronomy of the University of Oklahoma. Thanks go in particular to Kimball Milton and his group who made this very pleasant and exciting visit possible.

References [1] M. Abramowitz and I. A. Stegun. Handbook of mathematical functions. Dover, New York, 1970. [2] J. Ambjorn and S. Wolfram. Properties of the vacuum. 1. Mechanical and thermodynamic. Ann. Phys., 147:1–32, 1983. [3] A. O. Barvinsky, A. Yu. Kamenshchik, and I. P. Karmazin. One loop quantum cosmology: Zeta function technique for the Hartle–Hawking wave function of the universe. Ann. Phys., 219:201–242, 1992. [4] M. Bordag, E. Elizalde, and K. Kirsten. Heat kernel coefficients of the Laplace operator on the D-dimensional ball. J. Math. Phys., 37:895–916, 1996. [5] M. Bordag, E. Elizalde, K. Kirsten, and S. Leseduarte. Casimir energies for massive fields in a spherical geometry. Phys. Rev., D56:4896–4904, 1997. [6] M. Bordag, B. Geyer, K. Kirsten, and E. Elizalde. Zeta function determinant of the Laplace operator on the D-dimensional ball. Commun. Math. Phys., 179:215–234, 1996. [7] M. Bordag, K. Kirsten, and J. S. Dowker. Heat kernels and functional determinants on the generalized cone. Commun. Math. Phys., 182:371–394, 1996. [8] D. Burghelea, L. Friedlander, and T. Kappeler. On the determinant of elliptic differential and finite difference operators in vector bundles over S 1 . Commun. Math. Phys., 138:1–18, 1991.

326

KLAUS KIRSTEN

[9] A. A. Bytsenko, G. Cognola, L. Vanzo, and S. Zerbini. Quantum fields and extended objects in space-times with constant curvature spatial section. Phys. Rept., 266:1– 126, 1996. [10] S. Coleman. Aspects of symmetry: Selected lectures of Sidney Coleman. Cambridge University Press, Cambridge, 1985. [11] J. S. Dowker and R. Critchley. Effective Lagrangian and energy momentum tensor in de Sitter space. Phys. Rev., D13:3224–3232, 1976. [12] T. Dreyfuss and H. Dym. Product formulas for the eigenvalues of a class of boundary value problems. Duke Math. J., 45:15–37, 1978. [13] G. V. Dunne and K. Kirsten. Functional determinants for radial operators. J. Phys., A39:11915–11928, 2006. [14] G. V. Dunne and K. Kirsten. Simplified vacuum energy expressions for radial backgrounds and Domain Walls. J. Phys. A, 42:075402, 2009. [15] E. Elizalde. On the zeta-function regularization of a two-dimensional series of Epstein-Hurwitz type. J. Math. Phys., 31:170–174, 1990. [16] E. Elizalde. Ten physical applications of spectral zeta functions. Lecture Notes in Physics m35, Springer, Berlin, 1995. [17] E. Elizalde, S. D. Odintsov, A. Romeo, A. A. Bytsenko, and S. Zerbini. Zeta regularization techniques with applications. World Scientific, Singapore, 1994. [18] P. Epstein. Zur Theorie allgemeiner Zetafunctionen. Math. Ann., 56:615–644, 1903. [19] P. Epstein. Zur Theorie allgemeiner Zetafunctionen II. Math. Ann., 63:205–216, 1907. [20] A. Erdélyi, W. Magnus, F. Oberhettinger, and F. G. Tricomi. Higher transcendental functions. Based on the notes of Harry Bateman, McGraw-Hill Book Company, New York, 1955. [21] R. P. Feynman and A. R. Hibbs. Quantum mechanics and path integrals. McGrawHill, New York, 1965. [22] R. Forman. Functional determinants and geometry. Invent. Math., 88:447–493, 1987; erratum in 108:453-454, 1992. [23] R. Forman. Determinants, finite-difference operators and boundary value problems. Commun. Math. Phys., 147:485–526, 1992. [24] M. De Francia. Free energy for massless confined fields. Phys. Rev., D50:2908– 2919, 1994. [25] M. De Francia, H. Falomir, and M. Loewe. Massless fermions in a bag at finite density and temperature. Phys. Rev., D55:2477–2485, 1997. [26] I. M. Gelfand and A. M. Yaglom. Integration in functional spaces and its applications in quantum physics. J. Math. Phys., 1:48–69, 1960.


[27] P. B. Gilkey, K. Kirsten, and D. V. Vassilevich. Heat trace asymptotics with transmittal boundary conditions and quantum brane world scenario. Nucl. Phys., B601:125–148, 2001. [28] I. S. Gradshteyn and I. M. Ryzhik. Table of integrals, series and products. Academic Press, New York, 1965. [29] S. W. Hawking. Zeta function regularization of path integrals in curved space-time. Commun. Math. Phys., 55:133–148, 1977. [30] K. Kirsten. Topological gauge field mass generation by toroidal space-time. J. Phys. A, 26:2421–2435, 1993. [31] K. Kirsten. Spectral functions in mathematics and physics. Chapman&Hall/CRC, Boca Raton, FL, 2001. [32] K. Kirsten. Heat kernel asymptotics: more special case calculations. Nucl. Phys. B .Proc. Suppl./, 104:119–126, 2002. [33] K. Kirsten and P. Loya. Computation of determinants using contour integrals. Am. J. Phys., 76:60–64, 2008. [34] K. Kirsten and A. J. McKane. Functional determinants by contour integration methods. Ann. Phys., 308:502–527, 2003. [35] K. Kirsten and A. J. McKane. Functional determinants for general Sturm–Liouville problems. J. Phys. A, 37:4649–4670, 2004. [36] H. Kleinert. Path integrals in quantum mechanics, statistics, polymer physics, and financial markets. World Scientific, Singapore, 2006. [37] H. Kleinert and A. Chervyakov. Simple explicit formulas for Gaussian path integrals with time-dependent frequencies. Phys. Lett. A, 245:345–357, 1998. [38] H. Kleinert and A. Chervyakov. Functional determinants from Wronskian Green functions. J. Math. Phys., 40:6044–6051, 1999. [39] M. Lesch. Determinants of regular singular Sturm–Liouville operators. Math. Nachr., 194:139–170, 1998. [40] M. Lesch and J. Tolksdorf. On the determinant of one-dimensional elliptic boundary value problems. Commun. Math. Phys., 193:643–660, 1998. [41] B. M. Levitan and I. S. Sargsjan. Introduction to spectral theory: Selfadjoint ordinary differential operators. Translations of Mathematical Monographs 39. AMS, Providence, R. I., 1975. [42] A. J. McKane and M. B. Tarlie. Regularisation of functional determinants using boundary perturbations. J. Phys. A, 28:6931–6942, 1995. [43] P. Moon and D. E. Spencer. Field theory handbook. Springer, Berlin, 1961. [44] F. W. Olver. The asymptotic expansion of Bessel functions of large order. Philos. Trans. Roy. Soc. London, A247:328–368, 1954. [45] F. W. J. Olver. Asymptotics and special functions. Academic Press, New York, 1974.

328

KLAUS KIRSTEN

[46] J. Polchinski. String theory, 1: An introduction to the bosonic string. Cambridge University Press, Cambridge, 1998. [47] D. B. Ray and I. M. Singer. R-torsion and the Laplacian on Riemannian manifolds. Advances in Math., 7:145–210, 1971. [48] L. S. Schulman. Techniques and applications of path integration. Wiley, New York, 1981. [49] H. Weyl. Das asymptotische Verteilungsgesetz der Eigenwerte linearer partieller Differentialgleichungen. Math. Ann., 71:441–479, 1912. [50] F. Williams. Topics in quantum mechanics. Birkhäuser, New York, 2003. K LAUS K IRSTEN D EPARTMENT OF M ATHEMATICS BAYLOR U NIVERSITY WACO , TX 76798 U NITED S TATES Klaus [email protected]


The role of the Patterson–Selberg zeta function of a hyperbolic cylinder in three-dimensional gravity with a negative cosmological constant FLOYD L. WILLIAMS

To the memory of Kenneth Hoffman

1. Introduction A few years ago, the author took note of certain sums that appeared in the physics literature in connection with thermodynamic of the BTZ black hole — a three-dimensional solution discovered by M. Bañados, C. Teitelboim, and J. Zanelli [1] of the Einstein gravitational field equations Rij

1 2 Rgij

gij D 0

(1.1)

with negative cosmological constant . Here Rij D Rij .g/; R D R.g/ are the Ricci tensor and Ricci scalar curvature, respectively, of the solution metric g D Œgij . We describe the BTZ metric in equation (2.1) below. These sums were used to express, for example, the nondivergent part of the effective BTZ action, or corrections to classical Bekenstein–Hawking entropy [4; 13; 15] — sums that physicists evidently did not realize were related to the Patterson–Selberg zeta function Z .s/ of a hyperbolic cylinder. The paper [21], for example, was written to point out this relation and thus to establish a thermodynamics-zeta function connection. Another such connection appears in my Lecture 6 of this volume. In [23; 25; 26], for example, we see that the Mann–Solodukhin quantum correction to black hole entropy [15] is expressed, in fact, in terms of a suitable “deformation” of Z .s/. It is also possible to keep track of a corresponding deformation of the black hole topology. We review the deformation of zeta, and of the BTZ topology, in Section 4 below where we use it to set up a one-loop 329

330

FLOYD L. WILLIAMS

determinant formula (or an effective action formula) in the presence of conical singularities. In Section 3 we express the one-loop quantum field partition function, the one-loop gravity partition function , and the full gravity partition function all in terms of the zeta function Z .s/. Using the holomorphic sector of the oneloop gravity partition function and the classical elliptic modular function j . /, one can build up (with the help of Hecke operators) modular invariant partition functions of proposed holomorphic conformal field theories with central charge 24k, where k is a positive integer — theories first defined by G. Höhn [12] and proposed by E. Witten [28] as the holographic dual of pure 2 C 1 gravity. That is, these partition functions exist even if the theories do not (although for k D 1 existence has been established by I. Frenkel, J. Lepowsky, and A. Meurman in [8]), and in Section 5 we take a close look (in Theorem 5.16, for example) at their Fourier coefficients — the asymptotics of which provide for quantum corrections to holomorphic sector black hole entropy. The lecture, after this introduction, consists of four sections and an appendix:

The BTZ black hole Patterson–Selberg zeta function and a one-loop determinant formula Determinant formula in the presence of conical singularities Extremal partition functions of conformal field theories with central charge 24k Appendix to Section 5: Computation of Zk . / for k D 2; 3 References

The author dedicates this lecture to the memory of Professor Kenneth Hoffman. His kind support and friendliness to me, as a young MIT postdoc, remains most highly appreciated these many years later.

2. The BTZ black hole The BTZ metric that solves the vacuum Einstein equations (1.1) in three dimensions with < 0 is given (in Euclidean form) by 2 dsBTZ D

N1 .r /2 Cr 2 N2 .r /2 d 2 CN1 .r /

2

dr 2 C2r 2 N2 .r / d d Cr 2 d 2 ; (2.1)

in coordinates .r; ; / on a region of anti-deSitter space where for mass and angular momentum parameters M > 0; J 0, respectively, def:

N1 .r /2 D M

r 2

J 2 =4r 2 ;

N2 .r / D J =2r 2 :

(2.2)

In equation (2.1), one has periodicity of the Schwarzschild variable ; i.e. there is the identification C 2 n for n 2 Z, the ring of integers. We return

PATTERSON–SELBERG ZETA FUNCTION AND THREE-DIMENSIONAL GRAVITY

331

to this important point shortly. Not all scalar curvatures are created equal. So in equation (1.1), our sign convention is such that R D 6 > 0; in particular 2 2 dsBTZ is a constant curvature solution. The metric dsBTZ is also a black hole solution with outer and inner event radii rC ; r given by J 2 1=2 J i M2 2 ; (2.3) rC D 1C 1C 2 2 ; r D 2 2rC M def: p where i 2 D 1 and D 1= > 0. Here rC > 0, but r 2 i R is pure imaginary (since we are working with the Euclidean form of BTZ). Note that J 2 1=2 M2 J 2 r D 1 1C 2 2 ; jr j D D ir : (2.4) 2 2rC M Of course r D 0 is equivalent to J D 0, which is the case of the nonspinning black hole, in which case there is a single event horizon. Given the periodicity C 2 n, n 2 Z, of the Schwarzschild variable , as mentioned earlier, one can describe the topology of the space-time (where 2 dsBTZ lives, with regarded as a time variable) as a quotient space B D

n H3

(2.5)

def:

where H3 D f.x; y; z/ 2 R3 j z > 0g is hyperbolic 3-space, and where aCib e 0 def: n ; D .a;b/ D f j n 2 Zg for D 0 e .aCib/ def:

(2.6)

def:

with a D rC = > 0, and b D jr j= D J =2r C 0; see equations (2.3). Thus SL.2; C/ is the cyclic subgroup with generator 2 SL.2; C/. The action of on H3 is given by n .x; y; z/ D .x 0 ; y 0 ; z 0 / for x 0 De 2an .x cos 2bn

y sin 2bn/;

0

2an

.x sin 2bn C y cos 2bn/;

0

2an

z:

y De z De

A fundamental domain F for this action is given by p ˚ F D .x; y; z/ 2 H3 j 1 < x 2 C y 2 C z 2 < e 2a ;

(2.7)

(2.8)

a proof of which is given in Appendix A3 of [27], for example. In particular SL.2; C/ is a Kleinian subgroup; that is, F has infinite hyperbolic volume: Z def: vol F D dx dy dz=z 3 D 1: (2.9) F

The description (2.5) is derived by way of a suitable change of variables 2 .r; ; / ! .x; y; z/; z > 0, whereby (remarkably) the BTZ metric dsBTZ in

332

FLOYD L. WILLIAMS

equation (2.1) is transformed, in fact, to a multiple ds 2 of the standard hyperbolic metric .dx 2 C dy 2 C dz 2 /=z 2 on H3 : ds 2 D 2 .dx 2 C dy 2 C dz 2 /=z 2

(2.10)

where 2 D 1=. /, by definition (2.3); see [7; 18], for example.

3. Patterson–Selberg zeta function and a one-loop determinant formula Going back to the fact that is Kleinian, we can assign to the black hole 3 B D nH a natural zeta function (an Euler product) def:

Z .s/ D

1 Y

.e 2bi /k1 .e

1

2bi k2

/ e

.k1 Ck2 Cs/2a

;

(3.1)

0k1 ;k2 2Z

which is the Patterson–Selberg zeta function attached to the hyperbolic cylinder nH3 ; see [16; 21]. Z .s/ is an entire function whose zeros are the numbers 2bi 2 ni C 2a 2a for k1 ; k2 ; n 2 Z, k1 ; k2 0, that come from the zeros of its factors. In particular, Z .s/ ¤ 0 for Re s > 0. In fact, for Re s > 0, Z .s/ D e log Z .s/ , where def:

Nk1 ;k2 ;n D .k1 C k2 / C .k1

def:

log Z .s/ D D

1 X nD1 1 X nD1

e

k2 /

.s 1/2an

4n sinh2 .an/ C sin2 .bn/ e .s 1/2an : 2n cosh.2an/ cos.2bn/

(3.2)

In [10], the authors study the one-loop partition function of a free quantum field propagating in a locally anti-de Sitter background. The results they obtain cover not only the BTZ case, but higher genus generalizations of it, as well as the case of nonscalar fields (say gauge and graviton excitations). In the special BTZ case with a scalar field, for example, the one-loop determinant formula (equation (4.9) of [10]) log det D

1 X e nD1

p 2 n 1Cm2 Im

2nj sin n j2

(3.3)

def:

1 is derived, where now D 2 . C iˇ/ denotes the modular parameter corresponding to the anti-de Sitter temperature ˇ 1 and angular potential , where

K.t; r / D

e

.m2 C1/t r 2 =4t

.4 t/3=2

r sinh r

(3.4)


333

is the scalar heat kernel on H3 , and where in (3.3) the divergent contribution proportional to vol F (see equation (2.9)) is disregarded. We indicate how formula (3.3) also follows quite quickly from our result in [3], and we point out that the right-hand side of (3.3) in fact coincides with the special value p 2 log Z .1C 1 C m2 / (where we identify ˇ=2 with a and =2 with b) — this observation being an example of our initial remarks regarding sums appearing in the physics literature that are expressible in terms of the zeta function Z .s/ — a point unnoticed by physicists. We start with the reminder that for z D x C iy 2 C, sin z D sin x cosh y C i cos x sinh y: In particular j sin zj2 D sin2 x cosh2 y C cos2 x sinh2 y D sin2 x 2 2 2 cosh2 y C .1 sin2 x/ sinh2 y Dsin2x.cosh2 y sinh y/ C sinh y D sin x C def: ˇn C sinh2 2 ) the right-hand side of sinh2 y ) j sin nj2 D sin2 n 2 def: ˇ 2 )

formula (3.3) is (since Im D 2

1 X

p

e

1Cm2 ˇn

∴

h i D 2 log Z .1 C 2 ˇn 2 n 4n sinh C sin nD1 2 2

p

1 C m2 /;

(3.5)

by definition (3.2). On the other hand, we have considered in [3; 22] a truncated heat kernel def:

K t .e p1; p e2 / D

X

K t .p1 ; n p2 /

(3.6)

n2Z f0g

for B D nH3 , t > 0, where p ej 2 B denotes the n

p2 is given by definition (2.7), and where def: e

K t .p1 ; p2 / D

t d.p1 ;p2 /2 =4t

.4 t/3=2

-orbit of pj 2 H3 , j D 1; 2,

d.p1 ; p2 / sinh.d.p1 ; p2 //

(3.7)

(compare equation (3.4)) for d.p1 ; p2 / the hyperbolic distance between p1 and p2 , given by cosh.d.p1 ; p2 // D 1 C

.x1

x2 /2 C .y1 y2 /2 C .z1 2z1 z2

z2 / 2

(3.8)

for pj D .xj ; yj ; zj /. The expression K t .pz1 ; pz2 / gives rise to the theta function (or heat trace) ZZZ def: .t/ D trace K t D K t .p; z p/ z dv.p/ (3.9) F

334

FLOYD L. WILLIAMS

where dv D dx dy dz=z 3 is the hyperbolic volume element; see (2.8) and (2.9). We regard the integral Z 1 dt 2 def: I.m/ D e m t trace K t (3.10) t 0 as the meaning of the expression log det in the left-hand side of (3.3), with the understanding that by restricting the summation to n ¤ 0 in (3.6), we disregard the divergent term Z Z Z Z 1 Z 1 vol F e t dt 2 m2 t e dv D e .1Cm /t e 3=2 1 dt: 3=2 3=2 t .4/ 0 0 F .4 t/ (3.11) Namely, n D 0 implies K t .p; n p/ D K t .p; p/ D e t .4 t/

3=2

;

by (3.7) and (3.8), and this expression is independent of p 2 H3 . Thus if we were to the term for n D 0 in (3.6), there would be a manifest contribution RRRinclude t .4 t/ 3=2 dv to (3.9), which in turn would lead to the contribution e R 1F m2 t RRR e t 0 e F .4 t / 3=2 dv dt =t to (3.10). This explains the divergent term mentioned in (3.11), where one notes not only the “infrared” divergence vol F (by (2.9)), but also the “ultraviolet” divergence reflected by the negative 3=2 2 def: R 1 in the integral J.m/ D 0 e .1Cm /t t 3=2 1 dt . Given the formula Z 1 .v/ 2 (3.12) e .1Cm /t t v 1 dt D .1 C m2 /v 0 for positive v, the authors in [10] (also compare [5]) remove the ultraviolet divergence by assigning to J.m/ the value p . 3=2/ 4 D .1 C m2 /3=2 : 3 .1 C m2 / 3=2 Thus, in summary, the divergent term being disregarded is (by (3.11)) equal to p .1 C m2 /3=2 vol F 4 2 3=2 .1 C m / D vol F; 6 .4/3=2 3 and we regard the one-loop determinant formula (3.3) as the statement that Z 1 p dt 2 .3:10/ I.m/ D e m t trace K t D 2 log Z .1 C 1 C m2 /; (3.13) t 0 since we have noted that the right-hand side of (3.3) is the right-hand side of equation (3.5).


335

Now formula (3.13) is easy to prove since the theta function .t/ D trace K t was computed in [3]; also compare [4; 15; 17]. Namely .t/ D p

a

1 X

4 t

nD1

e

.t Cn2 a2 =t /

sinh2 .na/ C sin2 .nb/

Also by formula 32. on page 1145 of [11] r Z 1 3=2 A=4t Bt e t e e dt D 2 A 0

:

(3.14)

.AB/1=2

(3.15)

for A > 0; B 0. Commutation of the integration in (3.13) with the summation in (3.14) is okay: Z 1 1 a X 1 2 2 2 I.m/ D p t 3=2 e 4n a =.4t / e .1Cm / dt 2 2 4 nD1 sinh .na/ C sin .nb/ 0 r 1 X 1 2 2 2 1=2 .3:15/ a D p e Œ4n a .1Cm / 2 2 2 2 2 4n a sinh .na/ C sin .nb/ 4 nD1

D2

1 X

e

p 1Cm2 2an

4n sinh2 .na/ C sin2 .nb/ p D 2 log Z .1 C 1 C m2 /; nD1

(3.16)

as desired (again by (3.2)). Given formula (3.13), we can go a step further and obtain in terms of Z .s/ 1-loop the one-loop partition function denoted by Zscalar .; / in [10]. By definition, it equals .det / 1=2 , which we take to mean .e I.m/ / 1=2 . Thus, by (3.13), 1 : (3.17) p Z .1 C 1 C m2 / p def: def: Let q D e 2 i D e 2bi 2a , qN D e 2bi 2a , h D .1 C 1 C m2 /=2, and note that for 0 k1 ; k2 2 Z one has 1-loop

Zscalar .; / D

q k1 Ch .q/ N k2 Ch D e .2bi

2a/.k1 Ch/ . 2bi 2a/.k2 Ch/

e

D .e 2bi /k1 .e

2bi k2

/ e

.k1 Ck2 C2h/2a

:

Therefore by definition (3.1) we can also write Z .1 C

1 p

1 C m2 /

D

1

Y 0k1 ;k2 2Z

1

q k1 Ch .q/ N k2 Ch

(3.18)

336

FLOYD L. WILLIAMS

where, as noted in [10], the right-hand side has the form trace q L0 qN L0 for Virasoro operators L0 ; L0 that generate scale transformations (in the language of boundary conformal field theory). 1-loop The one-loop gravity partition function Zgravity . / is also computed in [10]. The result is 1-loop Zgravity . /

D

1 Y mD2

1 Y 1 D j1 q m j2 j1 mD2

1 e 2mbi e 2am j2

;

(3.19)

from which one obtains the full gravity partition function 2k

Zgravity . / D jqj

1-loop

Zgravity . /:

(3.20)

for the Chern–Simon coupling constant k D =16G (see (2.3)), G being the 1-loop Newton constant; see [10; 14; 28]. We claim that Zgravity . / can also be expressed in terms of the zeta function Z .s/. We have a factorization 1-loop

Zgravity . / D Zhol . /Z hol . /

(3.21)

for def:

Zhol . / D

1 Y

1 1 qm

mD2

(3.22)

its holomorphic sector. P Since a > 0, we have jq m j D e 1 mn =n. That is, hence log.1 q m / D nD1 q log Zhol . / D

1 X

log.1

mD2

D

1 X nD1

D

1 1 X 1 X n m .q / q /D n

1 q 2n 1 n .1 q n / 1

nD1

< 1 for m > 0;

m

mD2

nD1

1 X e 4bni e

2am

qN n D qN n

4an

1 X

4an .1

e 4bni e

nj1

nD1

e 2bni e nj1 q n j2

4an e 2an

e 2bni e q n j2

:

2an /

(3.23)

On the other hand, sin2 .bn/ C sinh2 .an/ D 12 cosh.2an/ cos.2bn/ — this identity was used in (3.2) and will be used later in (4.8). Thus j1 q n j2 1 D 4jqjn

qn

D

1 2an 4 .e

D

1 2

qN n C jqj2n 1 D 4jqjn

2an

e 4e

2bni e 2an C e 4an 2an

2an

/ cos.2bn/ D sin2 .bn/ C sinh2 .an/:

2 cos.2bn/ C e

cosh.2an/

e 2bni e

(3.24)


1 D That is, j1 q n j2 e write

2an 4

1 , which by (3.23) lets us sinh2 .an/Csin2 .bn/

1 X e 4bni e

log Zhol . / D

nD1

2an

e 2bni e

4an

4n sinh2 .an/C sin2 .bn/

1 X e

D

337

. 2b a iC1/2an

e

. b a iC2/2an

4n sinh2 .an/ C sin2 .bn/ 2b i C log Z 3 log Z 2 a

nD1

D

b i a

(3.25)

by definition (3.2). By definition (3.1), Z .s/ D Z .Ns /. By equation (3.25) we have therefore established: . b 2b b ai T HEOREM 3.26. For D C , Zhol . / D Z 3 i Z 2 i . In a a particular, by (3.21), b b Z 3 i Z 3C i a a 1-loop Zgravity . / D (3.27) ; 2b 2b Z 2 i Z 2C i a a and thus by equation (3.20), Zgravity . / also has the explicit expression e 4ak (the right-hand side of equation (3.27)) in terms of the Patterson–Selberg zeta function Z .s/.

4. Determinant formula in the presence of conical singularities We will now extend the one-loop determinant formula (3.13) to the BTZ black holes B .˛/ with conical singularities. Here we fix 0 < ˛ 1 and for i ˛ 0 def: e

˛ D 0 e i ˛ we define in (2.6):

.˛/

to be the subgroup of SL.2; C/ generated by and ˛ , for as .˛/ def:

D f n ˛m j n; m 2 Zg:

.˛/

(4.1)

acts on H3 by . n ˛m / .x; y; z/ D .x 0 ; y 0 ; z 0 /, where x 0 D e 2an x cos 2.bn C ˛m/ y 0 D e 2an

y sin 2.bn C ˛m/ ; x sin 2.bn C ˛m/ C y cos 2.bn C ˛m/ ;

z 0 D e 2an z:

(4.2)

338

FLOYD L. WILLIAMS

This action, like that defined in equation (2.7), is the restriction of the standard action of SL.2; C/ on H3 . We take B

.˛/

def:

D

.˛/

nH3 :

If ˛ D 1, then .˛/ D and B .˛/ D B . However, in general, B .˛/ is not a smooth manifold since the action of .˛/ is not free. For example each point .0; 0; z/; z > 0, on the positive z-axis is a fixed point of ˛m , by definition (4.2). To understand the topology of B .˛/ a little better, consider the action of Z on R2 given by x def: x cos.2 m˛/ y sin.2 m˛/ D m x sin.2 m˛/ C y cos.2 m˛/ y cos.2 m˛/ sin.2 m˛/ x D (4.3) sin.2 m˛/ cos.2 m˛/ y for m 2 Z and x; y 2 R. Thus the action is a rotation, the angle of rotation being 2 m˛. Let .ZnR2 /.˛/ denote the corresponding quotient space, and let S 1 D fz 2 C j jzj D 1g denote the unit circle. In [24] we construct a well-defined surjective homeomorphism 2 .˛/ S 1 . In fact, given .x; y; z/ 2 H3 define ˛ W B .˛/ ! .ZnR / def:

r Dr .x; y; z/ D

B

log z (since z > 0), a def: x rb y rb C sin ; uDu.x; y; z/ D cos z z def: x rb y rb vDv.x; y; z/ D sin C cos : z z

(4.4)

A

If .x; y; z/ 2 B .˛/ denotes the .˛/ -orbit of .x; y; z/ 2 H3 , and .u; v/ 2 .ZnR2 /.˛/ denotes the Z-orbit of .u; v/ 2 R2 , then

B

˛ ..x; y; z//

def:

A

D ..u; v/;

e i r D e i a log z /:

(4.5)

Similarly, the inverse function ˛ 1 W .ZnR2 /.˛/ S 1 ! B .˛/ is explicated in [24]. If ˛ D 1= l with 2 l 2 Z, for example, then one computes that a fundamental domain for the Z action in (4.3) is given by a cone in R2 with vertex at .0; 0/, and with opening angle 2= l D 2 ˛. Given that the black holes B .˛/ have the topology .ZnR2 /.˛/ S 1 , as just indicated, we see that they have conical singularities. In particular B has the topology R2 S 1 , as is well-known. The family fB .˛/ g0 0) K t

.˛/

def:

.pz1 ; pz2 / D

l 1 X mD0

K t

B

pz1 ; ˛m p2 ;

(4.9)

340

FLOYD L. WILLIAMS

which equals l 1 X 1 .4 t/3=2

X

e

t d.p1 ;. n ˛m /p2 /2 =4t

mD0 n2Z f0g

d.p1 ; . n ˛m / p2 / sinh d.p1 ; . n ˛m / p2 /

for B .˛/ , from which we can define (compare definition (3.9)) the theta function (for t > 0) ZZZ .˛/ def: .˛/ .˛/ .t/ D trace K t D K t .p; z p/ z dv.p/; (4.10) F .˛/

where F .˛/ H3 is defined in terms of spherical coordinates x D sin cos , y D sin sin , z D cos , with 0, 0 < 2, 0 < =2: 1 0: T HEOREM 4.12. For ˛ D 1= l, with 1 l 2 Z, one has

.˛/

˛a .t/ D p 2 4 t

e

X n2Z f0g m2Z 0ml 1

t a2 n2 =t

sinh2 .an/ C sin2 .bn C ˛m/

:

This theorem generalizes the trace formula (3.14). Similarly the following theorem generalizes the one-loop determinant formula (3.13): T HEOREM 4.13. For ˛ D 1= l, with 1 l 2 Z, one has Z 1 p 2 .˛/ dt D 2 log Z .˛/ .1 C 1 C m2 /: e m t trace K t t 0

(4.14)

P ROOF. We follow the argument above in the proof of (3.13), given Theorem 4.12. Z 1 2 .˛/ dt e m t trace K t D t 0 Z 1 X ˛a 1 2 2 2 t 3=2 e 4n a =4t e .1Cm /t dt p 2 2 2 4 n¤0 sinh .an/ C sin .bnC ˛m/ 0 0ml 1

˛a D p 2 4

.3:15/

1

X

sinh2 .an/ C sin2 .bnC ˛m/

n¤0 0ml 1

r 2

p 4n2 a2 .1Cm2 / e 2 2 4n a


˛ D 2

X

˛ D 2

X

e

341

p 2jnja 1Cm2

jnj 2 sinh2 .an/ C sin2 .bn C ˛m/

n¤0 0ml 1

p

2

e 2jnja 1Cm : jnj cosh.2an/ cos 2.bn C ˛m/

(4.15)

n¤0 0ml 1

To proceed further we employ the identity1 l 1 X mD0

cosh u

1 cos.v

2 m= l/

def:

D

l sinh.lu/ : sinh u cosh.lu/ cos.lv/

(4.16)

def:

We apply it with u D 2an and v D 2bn to rewrite the last sum in (4.15) as ˛ 2

p

X n2Z f0g

2

e 2jnja 1Cm l sinh.l2an/ jnj sinh.2an/ cosh.l2an/ cos.l2bn/ 1 X

p 1Cm2

sinh.2an=˛/ e 2an D2 2n sinh.2an/ cosh.2an=˛/ nD1 p D 2 log Z .˛/ .1 C 1 C m2 /;

cos.2bn=˛/

(4.17)

by definition (4.8), which concludes the proof of Theorem 4.13.

In the effective action formula (4.14) we have assumed that ˛ 1 2 Z. This assumption can be removed and thus a more general formula can be presented if we appeal to an old contour integral formula that goes back to A. Sommerfeld in 1897, in his amazing diffraction studies. The reader can consult the references [13; 15], for example, on this point — references which of course do not employ the zeta function Z .˛/ .s/.

5. Extremal partition functions of conformal field theories with central charge 24k In this section we consider the modular invariant partition function Zk . / of a holomorphic conformal field theory (CFT) with central charge c D 24k, k D 1; 2; 3; 4; : : : . Such a theory was introduced by G. Höhn [12], and is called an extremal CFT (ECFT) — which according to a bold proposal of E. Witten [28] is the dual to 3-dimensional pure gravity with a negative cosmological constant; 1 Formula

(4.16) corrects a misprint in [27]. Namely, the expression sin.lu/ in formula (4.10) of [27] .˛/ should read sinh.lu/. Also in equations (4.5) and (4.6) of [27] the often occurring expression m should read ˛m .

342

FLOYD L. WILLIAMS

also compare [14]. Apart from the case k D 1, however, there is uncertainty regarding the existence of ECFT’s. I. Frenkel, J. Lepowsky, and A. Meurman (FLM) [8] have indeed constructed a holomorphic CFT with central charge c D 24 (i.e., with k D 1) and with Z1 . / D j . / 744, where j . / is the classical elliptic modular invariant (see (5.6) below). An important point regarding the FLM construction is monster symmetry: the states of the theory transform as a representation of the finite, simple, Fischer– Griess group M , of order jM j D 246 320 59 76 112 132 17 19 23 29 31 41 47 59 71 1054 , called the monster (or the friendly giant). However for k D 2, D. Gaiotto [9] has slain the “two-headed monster”: there exists no self-dual ECFT for c D 48 with monster symmetry. We begin by indicating how Zk . / (defined for Im > 0) can be explicitly constructed from the FLM Z1 . / and the one-loop partition function Zhol . / of definition (3.22), with help of Hecke operators. def: Fix k D 1; 2; 3; 4; : : : , and for 2 C with Im > 0 set q D q. / D e 2 i , so jqj < 1. Define def:

Z0 . / D q

k

def:

Zhol . / D q

k

1 Q

1 1

nD2

qn

I

(5.1)

compare definition (3.22). The full gravity partition function of definition (3.20) therefore admits the factorization Zgravity . / D Z0 . /Z 0 . /;

(5.2)

by equation (3.21). Let p denote the partition function on Z C ; that is, p.n/ is the number of ways of writing a positive integer n as a sum of positive integers, without regard to order. Euler’s formula (equation (9.10) of my introductory lectures, page 72) says that 1 P 1 Q1 D p.n/z n (5.3) n/ .1 z nD0 nD1 def:

for jzj < 1, where p.0/ D 1. Therefore we can write Z0 . / D q D

k

1 P nD0

.1

q/

p.n/q n

1 Q

1 qn

nD1 1

1 P

k

Dq

k

.1

q/

1 P

p.n/q n

nD0

p.n/q nC1

k

:

nD0

Collecting coefficients here we see that Z0 . / D

1 P rD k

ar .k/q r

(5.4)


343

for def:

ar .k/ D p.r C k/

p.r C k

1/;

r k;

(5.5)

def:

where we set p. 1/ D 0. As mentioned in my introductory lectures (equations (4.44) and (4.45) on page 42), the modular j -invariant has a q-expansion with all Fourier coefficients cn 2 Z. That is, defining def:

j . / D

1728.60G4 . //3 ; .60G4 . //3 27.140G6 . //2

we have

(5.6)

1

1 X j . / D C cn q n q

with

nD0

c0 D 744; c1 D 196;884; (5.7)

c2 D 21;493;760; c3 D 864;299;970; c4 D 20;245;856;256; c5 D 333;202;640;600; c6 D 4;252;023;300;096:

The denominator in (5.6) is the Dedekind–Klein discriminant form . /, and Gl .z/ is the holomorphic Eisenstein series of weight l given in definition (4.4) of the introductory lectures (page 31). Here we depart from convention in using J. / not in the classical sense but to denote the function j . / c0 : def:

J. / D j . /

744 D

1 C 196;884 q C 21;493;760 q 2 C q 864;299;970 q 3 C 20;245;856;256 q 4 C :

(5.8)

Now recall the n-th Hecke operator T .n/ of weight w acting on a function f . /, Im > 0, where n; w 2 Z, n 1, w 0. As seen in (3.22) of the introductory lectures (page 28), it is given by def: w 1

.T .n/f /. / D n

X dX1

d

w

f

d>0 aD0 d jn

In particular

def:

n.T .n/f /. / D

X dX1 d>0 aD0 d jn

f

n C da : d2

n C da d2

(5.9)

(5.10)

344

FLOYD L. WILLIAMS

for w D 0, which is the only case we will need, since J. / in (5.8) has weight zero. Of course T .1/f D f for any weight w. We can now define the main object, where we have fixed an integer k > 0: def:

Zk . / D a0 .k/C

k X

a

r .k/r .T .r /J /. /

r D1

D p.k/

p.k 1/C

k X

p.k r /

p.k r 1// r .T .r /J /. / (5.11)

r D1

by definition (5.5), where r .T .r /J /. / is given by equation (5.10) applied to f D J . Since p.0/ D p.1/ D 1, p. 1/ D 0, and T .1/f D f we see that Z1 . / J. /, which (as remarked on earlier) is the partition function of the FLM holomorphic CFT of central charge c D 24, with monster symmetry. Frenkel, Lepowsky, and Meurman also conjecture that this ECFT is unique — a result that remains unproved at the present time. To be a bit more precise, these authors construct a graded, infinite-dimensional M -module V \ D V0 ˚ V1 ˚ V2 ˚ V3 ˚ V4 ˚ (the moonshine module), where V0 is the trivial representation of M , V1 D f0g, V2 D 1 ˚ 196;833 , V3 D 1 ˚ 196;883 ˚ 21;296;876 , and so on; d is the irreducible representation of M of degree d, for d 1. A remarkable observation, first made by John McKay in 1978, is that the early Fourier coefficients cn in (5.7) are integral linear combinations of the degrees d; thus c1 D 196;884 D 1 C 196; 883, c2 D 21;493;760 D 1 C 196;883 C 21;296;876, and c3 D 864;299;970 D 2 1 C 2 .196;883/ C 21;296;876 C 842;609;326: V \ has the structure, in fact, of a vertex operator algebra (VOA), a subject thoroughly discussed by G. Mason and M. Tuite in their lectures in this book. The submodule V2 is actually an algebra (which is commutative but not associative), the Griess algebra, which has the monster M as its full symmetry group (i.e., as its automorphism group). P By equations (5.7), (5.8) we have the Fourier expansion J. / D n 1 cn q n , where c 1 D 1; c0 D 0. Accordingly, .T .n/J /. / has Fourier expansion X .n/ m .T .n/J /. / D cm q ; m n

where .n/ cm D

X cmn=d 2

d >0 djn; d jm

d

; m 1I

.n/

c0 D c0

X 1 ∴ D 0 D c .n/ m ; 1 m < nI d d>0 djn 1 c .n/ n D : (5.12) n


345

That is, .T .n/J /. / D

1 X

n

q n

C

.n/ m cm q

(5.13)

mD1

which we use in definition (5.11): Zk . / D a0 .k/ C

k X

a r .k/ q

r

Cr

r D1

1 X

cn.r / q n

:

nD1

Collecting coefficients here, we see that Zk . / D a

k .k/q

k

CCa

2 .k/q

2

Ca

1 .k/q

1

C a0 .k/ C

1 X

bk;n q n ;

(5.14)

nD1

where a

k .k/

D 1;

def:

bk;n D

k X

ra

.r / r .k/cn ;

n 1;

(5.15)

r D1 .r /

with P a r .k/ D p.k r / p.k r 1/ for 1 r k given by (5.5), and cn D d cr n=d 2 =d given by (5.12). Before commenting on the important physical significance of the coefficients bk;n in (5.15), we state the following result: T HEOREM 5.16. For k; n 2 Z; k; n 1, let p p 1 def: bk;n D ke 4 k n = 2.k n/3=4 : Then bk;n equals 3 1 bk;n 1 p C "k n C T .k; n/ 32 k n k 1 1 X r 1=4 a r .k/ C 1=4 1 p p p 4 n. k r/ k e r D1

3 p C "r n C T .r; n/ ; 32 r n

(5.17)

(5.18)

where j"m j :055=mpfor integer m 1, and 0 p T .r; n/ is bounded above by 3 3 2 r n 3=2 2 r n / for 1 r k, where .s/ 3=2 both r 2 =.2e / and n 2 =.2e is the Riemann zeta function. In setting up the proof of Theorem 5.16, the author relied heavily on the following result of N. Brisebarre and G. Philibert [2] (as mentioned in equation (9.32) on page 76 of my introductory lectures): For m 1 p e 4 m 3 1 cm D p (5.19) p C "m ; 32 m 2m3=4

346

FLOYD L. WILLIAMS

where again j"m j :055=m. Equation (5.19) immediately implies the weaker asymptotic result (see equation (9.31) on page 76) p p (5.20) cm e 4 m = 2m3=4 as m ! 1; due to H. Petersson in 1932 and H. Rademacher in 1938, who was unaware of Petersson’s proof. Similarly, Theorem 5.16 immediately implies the weaker asymptotic result p p 1 def: bk;n bk;n D ke 4 kn = 2.k n/3=4 as n ! 1 (5.21) for every fixed k, as observed by E. Witten in Section 3 of [28]. Actually Witten assumes that k is large with n=k fixed, but we see that this assumption is unnecessary for the statement (5.21). Now from log bk;n , say for n sufficiently large, one obtains both the classical, p holomorphic sector Bekenstein–Hawking black hole entropy Shol D 4 k n (the leading asymptotic term) and corrections (subleading asymptotic terms) to that entropy: p 1 log bk;n D 4 k n C 14 log k 34 log n 12 log 2 ; (5.22) by (5.21). We offer further explanation regarding p equation (5.22). In particular we explain why the leading term Shol D 4 k n there was referred to as the holomorphic sector entropy. In formulas (2.3), (2.4), the outer and inner black hole radii for the BTZ metric in Euclidean form (2.1) are given by r 2 M J 2 2 r˙ D 1˙ 1C : M 2 For convenience, we also consider the Lorentzian form of the metric 2 dsL D N1 .r /2 C r 2 N2 .r /2 dt 2 C N1 .r / 2 dr 2 C 2r 2 N2 .r / d dt C r 2 d 2 ;

(5.23)

where now 16G 2 2 1=2 r2 C J ; (5.24) 2 r2 with the gravitational constant G also included for generality. We omit the definition of N2 .r /, which will not be needed. The corresponding radii, which we again denote by r˙ , are (by definition) solutions of the quartic equation N1 .r / D 0 W r J 2 2 2 r˙ D 4GM 1 ˙ 1 : (5.25) M N1 .r / D

8GM C


347

2 can have a black hole structure) that 1 .J =M /2 0, Here we assume (so dsL which is to say jJ j M ; then r˙ 0. The key point here is that the classical Bekenstein–Hawking entropy, given by

SBH D

rC ; 2G

(5.26)

is also given by the formula of J. Cardy [6] (see equation (9.3) on page 70) r r cL0 cL0 SBH D 2 C 2 ; (5.27) 6 6 where L0 and L0 are eigenvalues of the holomorphic and antiholomorphic Virasoro generators, respectively, and where the central charge c equals 3=2G. To check equation (5.27) we use the equalities L0 D .M C J /=2;

L0 D .M

J /=2:

(5.28)

2 By definition (5.25) we can write rC C r 2 D 8GM 2 and rC r D 4GJ ; therefore .rC ˙ r /2 D 8GM 2 ˙ 8GJ D 16G .M ˙ J /=2, which yields

L0 D .rC C r /2 =16G;

L0 D .rC

r /2 =16G:

(5.29)

By (5.26), the right-hand side of equation (5.27) is then r c rC C r C rC r 2 D r C =2G D SBH p 6 4 G for c D 3=2G, which verifies equation (5.27). Recall the Chern–Simon coupling constant k D =16G following equation (3.20). Since c D 3=2G, we see that c D 24k. Thus our ongoing assumption c 2 24 ZC amounts now to the “quantization”pof k; that is, k D =16G is a positive integer. If we call the first term 2 cL0 =6 in equation (5.27) the holomorphic sector (for obvious reasons), then for c D 24k we p entropy Shol p have Shol D 2 24kL0 =6 D 4 kL0 , which is the leading asymptotic term in equation (5.22), where n there is identified with the Virasoro eigenvalue L0 . This is a justification for referring to that leading term as holomorphic sector entropy.

Appendix to Section 5: Computation of Zk ./ for k D 2; 3 The explicit formulas (5.14) and (5.15) are sufficient for the direct computation of the initial terms of Zk . /, say for small values of k. One could employ

348

FLOYD L. WILLIAMS

a computer program to deal with larger values of k. For example, take k D 2. Then, by (5.5) and (5.15), a

1 .2/

D p.2

a

2 .2/

D 1;

a0 .2/ D p.2/

1/

p.2

2/ D 1

1 D 0;

p.1/ D 1:

Also (5.15) gives, for n 1, b2;n D

2 X

r .2/

ra

r D1

X 1 X 1 2c2n ∴ cr n=d 2 D 2 c2n=d 2 D 2c2n C cn=2 d d d >0 dj2; d jn

d>0 d jr; d jn

if 2 - n, if 2 j n,

leading to b2;1 D 2c2 D 42;987;520; b2;2 D 2c4 C c1 D 40;491;909;396; b2;3 D 2c6 D 8;540;046;600;192; by (5.7). Therefore by (5.14) Z2 . / D q

2

C 1 C 42;987;520q C 40;491;909;396q 2 C 8;504;046;600;192q 3 C :

Of course, def:

def:

Z1 . / D J. / D j . / q

1

744 D

C 196;884q C 21;493;760q 2 C 864;299;970q 3 C 20;245;856;256q 4 C ;

by equation (5.8). Similarly for k D 3 we have a0 .3/ D 1, a so that b3;n D

3 X

ra

r D1

r .3/

1 .3/ D 1,

a

2 .3/ D 0,

a

3 .3/ D 1,

X 1 X 1 cr n=d 2 D cn C 3 c 2; d d 3n=d

d>0 djr; d jn

d>0 dj3; d jn

which leads to b3;1 D c1 C 3c3 D 2;593;096;794; b3;2 D c2 C 3c6 D 12;756;091;394;048; and hence Z3 . / D q

3

Cq

1

C 1 C 2;593;096;794q C 12;756;091;394;048q 2 C :


349

References [1] Máximo Bañados, Claudio Teitelboim, and Jorge Zanelli, Black hole in three-dimensional spacetime, Phys. Rev. Lett. 69 (1992), no. 13, 1849–1851. [2] Nicolas Brisebarre and Georges Philibert, Effective lower and upper bounds for the Fourier coefficients of powers of the modular invariant j , J. Ramanujan Math. Soc. 20 (2005), no. 4, 255–282. [3] A. A. Bytsenko, M. E. X. Guimarães, and F. L. Williams, Remarks on the spectrum and truncated heat kernel of the BTZ black hole, Lett. Math. Phys. 79 (2007), no. 2, 203–211. [4] Andrei A. Bytsenko, Luciano Vanzo, and Sergio Zerbini, Quantum correction to the entropy of the .2 C 1/-dimensional black hole, Phys. Rev. D (3) 57 (1998), no. 8, 4917–4924. [5] Roberto Camporesi, Harmonic analysis and propagators on homogeneous spaces, Phys. Rep. 196 (1990), no. 1-2, 1–134. [6] John L. Cardy, Operator content of two-dimensional conformally invariant theories, Nuclear Phys. B 270 (1986), no. 2, 186–204. [7] Steven Carlip and Claudio Teitelboim, Aspects of black hole quantum mechanics and thermodynamics in 2 C 1 dimensions, Phys. Rev. D (3) 51 (1995), no. 2, 622–631. [8] Igor B. Frenkel, James Lepowsky, and Arne Meurman, A moonshine module for the Monster, Vertex operators in mathematics and physics (Berkeley, Calif., 1983), Math. Sci. Res. Inst. Publ., vol. 3, Springer, New York, 1985, pp. 231–273. [9] D. Gaiotto, Monster symmetry and extremal CFTs, preprint, 2008, arXiv:0801.0988. [10] Simone Giombi, Alexander Maloney, and Xi Yin, One-loop partition functions of 3D gravity, J. High Energy Phys. (2008), no. 8, 007, 25. [11] I. Gradshteyn and I. Ryzhik, Table of integrals, series, and products, corrected and enlarged ed., Academic Press, 1980. [12] Gerald Höhn, Selbstduale Vertexoperatorsuperalgebren und das Babymonster, Bonner Mathematische Schriften, no. 286, Universität Bonn Mathematisches Institut, Bonn, 1996, Dissertation, Rheinische Friedrich-Wilhelms-Universität Bonn, 1995. [13] Ikuo Ichinose and Yuji Satoh, Entropies of scalar fields on three-dimensional black holes, Nuclear Phys. B 447 (1995), no. 2-3, 340–370. [14] A. Maloney and E. Witten, Quantum gravity partition functions in three dimensions, preprint, 2007, arXiv:0712.0155. [15] Robert B. Mann and Sergey N. Solodukhin, Quantum scalar field on a threedimensional (BTZ) black hole instanton: heat kernel, effective action, and thermodynamics, Phys. Rev. D (3) 55 (1997), no. 6, 3622–3632. [16] S. J. Patterson, The Selberg zeta-function of a Kleinian group, Number theory, trace formulas and discrete groups (Oslo, 1987), Academic Press, Boston, MA, 1989, pp. 409–441.

350

FLOYD L. WILLIAMS

[17] P. Perry, Heat trace and zeta function for the hyperbolic cylinder in three dimensions, four page fax based in part on notes of F. Williams, 2001. [18] P. Perry and F. Williams, Selberg zeta function and trace formula for the BTZ black hole, Int. J. Pure Appl. Math. 9 (2003), no. 1, 1–21. ¨ [19] Hans Petersson, Uber die Entwicklungskoeffizienten der automorphen Formen, Acta Math. 58 (1932), no. 1, 169–215. [20] Hans Rademacher, The Fourier Coefficients of the Modular Invariant J(), Amer. J. Math. 60 (1938), no. 2, 501–512. [21] F. Williams, A zeta function for the BTZ black hole, Internat. J. Modern Phys. A 18 (2003), no. 12, 2205–2209. , BTZ black hole and Jacobi inversion for fundamental domains of infinite [22] volume, Algebraic structures and their representations, Contemp. Math., vol. 376, Amer. Math. Soc., Providence, RI, 2005, pp. 385–391. , Conical defect zeta function for the BTZ black hole, One hundred years [23] of relativity: Proceedings of the Einstein Symposium (Iais, Romania, 2005), 2005, Scientific Annals of Alexandru Ioan Cuza Univ., pp. 54–58. [24] , Topology of the BTZ black hole with a conical singularity, unpublished lecture notes, 2005. [25] , Note on quantum correction to BTZ instanton entropy, Proceedings of Science (IC 2006) (2006), no. 006, Available at http://pos.sissa.it/archive/conferences/ 031/006/IC2006˙006.pdf. [26] , A deformation of the Patterson–Selberg zeta function, Actas del XVI ´ Coloquio Latinoamericano de Algebra (Colonia del Sacramento, Uruguay, August 2005) (W. Santos, G. González-Sprinberg, A. Rittatore, and A. Solotar, eds.), Rev. Mat. Iberoamericana, Madrid, 2007, pp. 109–114. [27] , A resolvent trace formula for the BTZ black hole with conical singularity, Council for African American Researchers in the Mathematical Sciences. Vol. V, Contemp. Math., vol. 467, Amer. Math. Soc., Providence, RI, 2008, pp. 49–62. [28] E. Witten, Three-dimensional gravity revisited, preprint, 2007, arXiv:0706.3359.

Added in proof The author has discovered that a (slightly incorrect) version of formula (3.27), page 337, has been obtained, independently, by A. Bytsenko and M. Guimarães, (see formula (4.13) in their Truncated heat kernel and one-loop determinants for the BTZ geometry, Eur. Phys. J. C 58 (2008), pp. 511–516). The following reference provides for further connections of the Patterson–Selberg zeta function to BTZ physics: D. Diaz, Holographic formula for the determinant of the scattering operator in thermal AdS, preprint, arXiv:0812.2158v3 (2009).


F LOYD L. W ILLIAMS D EPARTMENT OF M ATHEMATICS AND S TATISTICS L EDERLE G RADUATE R ESEARCH T OWER 710 N ORTH P LEASANT S TREET U NIVERSITY OF M ASSACHUSETTS A MHERST, MA 01003-9305 U NITED S TATES [email protected]

351

A Window into Zeta and Modular Physics (Mathematical Sciences Research Institute Publications 57)

A Window Into Zeta and Modular Physics (Mathematical Sciences Research Institute Publications 57)

Assessing Mathematical Proficiency (Mathematical Sciences Research Institute Publications)

Galois Groups and Fundamental Groups (Mathematical Sciences Research Institute Publications)

Tight and Taut Submanifolds (Mathematical Sciences Research Institute Publications)

Algebraic Topology and Its Applications (Mathematical Sciences Research Institute Publications)

Exterior Differential Systems (Mathematical Sciences Research Institute Publications)

Global Analysis on Foliated Spaces (Mathematical Sciences Research Institute Publications)

Topology of Stratified Spaces (Mathematical Sciences Research Institute Publications 58)

New Directions in Hopf Algebras (Mathematical Sciences Research Institute Publications)

Topology of Stratified Spaces (Mathematical Sciences Research Institute Publications 58)

Flavors of Geometry (Mathematical Sciences Research Institute Publications)

Convex Geometric Analysis (Mathematical Sciences Research Institute Publications)

Probability, geometry, and integrable systems: for Henry McKean's 75th birthday (Mathematical Sciences Research Institute Publications 55)

Probability, Geometry and Integrable Systems (Mathematical Sciences Research Institute Publications 55)

Heegner Points and Rankin L-Series (Mathematical Sciences Research Institute Publications)

Current Topics in Complex Algebraic Geometry (Mathematical Sciences Research Institute Publications)

Generic Polynomials : Constructive Aspects of the Inverse Galois Problem (Mathematical Sciences Research Institute Publications)

The Eightfold Way: The Beauty of Klein's Quartic Curve (Mathematical Sciences Research Institute Publications)

Games of No Chance 3 (Mathematical Sciences Research Institute Publications - Volume 56)

Galois Groups over Q: Proceedings of a Workshop Held March 23-27, 1987 (Mathematical Sciences Research Institute Publications)

Determinants and Their Applications in Mathematical Physics (Applied Mathematical Sciences)

Problems of Mathematical Physics (Selected Russian Publications in the Mathematical Sciences)

Problems of Mathematical Physics (Selected Russian Publications in the Mathematical Sciences)

Zeta Functions, Topology and Quantum Physics

Physics Letters B vol 57

Perspectives in Mathematical Sciences (Interdisciplinary Mathematical Sciences)

Analysis and mathematical physics

Analysis and Mathematical Physics

Trade, Doha, and Development: A Window into the Issues

57)

A Window into Zeta and Modular Physics (Mathematical Sciences Research Institute Publications 57)