Lecture Notes in Physics Editorial Board R. Beig, Wien, Austria W. Beiglböck, Heidelberg, Germany W. Domcke, Garching, Germany B.G. Englert, Singapore U. Frisch, Nice, France P. Hänggi, Augsburg, Germany G. Hasinger, Garching, Germany K. Hepp, Zürich, Switzerland W. Hillebrandt, Garching, Germany D. Imboden, Zürich, Switzerland R. L. Jaffe, Cambridge, MA, USA R. Lipowsky, Golm, Germany H. v. Löhneysen, Karlsruhe, Germany I. Ojima, Kyoto, Japan D. Sornette, Nice, France, and Zürich, Switzerland S. Theisen, Golm, Germany W. Weise, Garching, Germany J. Wess, München, Germany J. Zittartz, Köln, Germany
The Lecture Notes in Physics The series Lecture Notes in Physics (LNP), founded in 1969, reports new developments in physics research and teaching – quickly and informally, but with a high quality and the explicit aim to summarize and communicate current knowledge in an accessible way. Books published in this series are conceived as bridging material between advanced graduate textbooks and the forefront of research to serve the following purposes: • to be a compact and modern uptodate source of reference on a welldefined topic; • to serve as an accessible introduction to the field to postgraduate students and nonspecialist researchers from related areas; • to be a source of advanced teaching material for specialized seminars, courses and schools. Both monographs and multiauthor volumes will be considered for publication. Edited volumes should, however, consist of a very limited number of contributions only. Proceedings will not be considered for LNP. Volumes published in LNP are disseminated both in print and in electronic formats, the electronic archive is available at springerlink.com. The series content is indexed, abstracted and referenced by many abstracting and information services, bibliographic networks, subscription agencies, library networks, and consortia. Proposals should be sent to a member of the Editorial Board, or directly to the managing editor at Springer: Dr. Christian Caron Springer Heidelberg Physics Editorial Department I Tiergartenstrasse 17 69121 Heidelberg/Germany [emailprotected]
Joachim Asch Alain Joye (Eds.)
Mathematical Physics of Quantum Mechanics Selected and Refereed Lectures from QMath9
ABC
Editors Joachim Asch Université du Sud Toulon Var Centre de physique théorique Département de Mathématiques BP 20132 F83957 La Garde Cedex France Email: [emailprotected]
Alain Joye Institut Fourier Université Grenoble 1 BP 74 38402 SaintMartind’Hères Cedex France Email: [emailprotected]
J. Asch and A. Joye, Mathematical Physics of Quantum Mechanics, Lect. Notes Phys. 690 (Springer, Berlin Heidelberg 2006), DOI 10.1007/b11573432
Library of Congress Control Number: 2005938945 ISSN 00758450 ISBN10 3540310266 Springer Berlin Heidelberg New York ISBN13 9783540310266 Springer Berlin Heidelberg New York This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable for prosecution under the German Copyright Law. Springer is a part of Springer Science+Business Media springer.com c SpringerVerlag Berlin Heidelberg 2006 Printed in The Netherlands The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Typesetting: by the authors and TechBooks using a Springer LATEX macro package Printed on acidfree paper
SPIN: 11573432
54/TechBooks
543210
Preface
The topics presented in this book were discussed at the conference “QMath9” held in Giens, France, September 12th16th 2004. QMath is a series of meetings whose aim is to present the state of the art in the Mathematical Physics of Quantum Systems, both from the point of view of physical models and of the mathematical techniques developed for their study. The series was initiated in the early seventies as an attempt to enhance collaboration between mathematical physicists from eastern and western European countries. In the nineties it took a worldwide dimension. At the same time, due to engineering achievements, for example in the mesoscopic realm, there was a renewed interest in basic questions of quantum dynamics. The program of QMath9, which was attended by 170 scientists from 23 countries, consisted of 123 talks grouped by the topics: Nanophysics, Quantum dynamics, Quantum ﬁeld theory, Quantum kinetics, Random Schr¨ odinger operators, Semiclassical analysis, Spectral theory. QMath9 was also the frame for the 2004 meeting of the European Research Group on “Mathematics and Quantum Physics” directed by Monique Combescure. For a detailed account of the program, see http://www.cpt.univ.mrs.fr/qmath9. Expanded versions of several selected introductory talks presented at the conference are included in this volume. Their aim is to provide the reader with an easier access to the sometimes technical state of the art in a topic. Other contributions are devoted to a pedagogical exposition of quite recent results at the frontiers of research, parts of which were presented in “QMath9”. In addition, the reader will ﬁnd in this book new results triggered by discussions which took place at the meeting. Hence, while based on the conference “QMath9”, this book is intended to be a starting point for the reader who wishes to learn about the current research in quantum mathematical physics, with a general perspective. Effort has been made by the authors, editors and referees in order to provide contributions of the highest scientiﬁc standards to meet this goal. We are grateful to Yosi Avron, Volker Bach, Stephan De Bi`evre, Laszlo Erd¨ os, Pavel Exner, Svetlana Jitomirskaya, Fr´ed´eric Klopp who mediated the scientiﬁc sessions of “QMath9”. We should like to thank all persons and institutions who helped to organize the conference locally: Sylvie Aguillon, JeanMarie Barbaroux,
VI
Preface
Nils Berglund, JeanMichel Combes, Elisabeth Elophe, JeanMichel Ghez, Corinne Roux, Corinne Vera, Universit´e du Sud Toulon–Var and Centre de Physique Th´eorique Marseille. We gratefully acknowledge ﬁnancial support from: European Science Foundation (SPECT), International Association of Mathematical Physics, Minist`ere de l’Education Nationale et de la Recherche, Centre National de la Recherche Scientiﬁque, R´egion ProvenceAlpesCˆote d’Azur, Conseil G´en´eral du Var, Centre de Physique Th´eorique, Universit´e du Sud Toulon–Var, Institut Fourier, Universit´e Joseph Fourier. Toulon Grenoble January 2006
Joachim Asch Alain Joye
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
Part I Quantum Dynamics and Spectral Theory Solving the Ten Martini Problem A. Avila and S. Jitomirskaya . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Rough Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Analytic Extension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 The Liouvillian Side . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Gaps for Rational Approximants . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Continuity of the Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 The Diophantine Side . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Reducibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Localization and Reducibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 A Localization Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5 5 6 8 9 9 10 10 11 12 12 14
Swimming Lessons for Microbots Y. Avron . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 LandauZener Formulae from Adiabatic Transition Histories V. Betz and S. Teufel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Exponentially Small Transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 The Hamiltonian in the SuperAdiabatic Representation . . . . . . . . . . 4 The Scattering Regime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
19 19 22 25 27 31
Scattering Theory of Dynamic Electrical Transport M. B¨ uttiker and M. Moskalets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 1 From an Internal Response to a Quantum Pump Eﬀect . . . . . . . . . . 33 2 Quantum Coherent Pumping: A Simple Picture . . . . . . . . . . . . . . . . . 36
VIII
Contents
3
Beyond the Frozen Scatterer Approximation: Instantaneous Currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 The LandauerB¨ uttiker Formula and Resonant Quantum Transport H.D. Cornean, A. Jensen and V. Moldoveanu . . . . . . . . . . . . . . . . . . . . . . 1 The LandauerB¨ uttiker Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Resonant Transport in a Quantum Dot . . . . . . . . . . . . . . . . . . . . . . . . . 3 A Numerical Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
45 45 47 48 53
Point Interaction Polygons: An Isoperimetric Problem P. Exner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 The Local Result in Geometric Terms . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Proof of Theorem 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 About the Global Maximizer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Some Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
55 55 56 58 61 62 64
Limit Cycles in Quantum Mechanics S.D. Glazek . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Deﬁnition of the Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Renormalization Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Limit Cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Marginal and Irrelevant Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Tuning to a Cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Generic Properties of Limit Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
65 65 67 69 71 73 74 75 76 76
Cantor Spectrum for QuasiPeriodic Schr¨ odinger Operators J. Puig . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 The Almost Mathieu Operator & the Ten Martini Problem . . . . . . . 1.1 The ids and the Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Sketch of the Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Reducibility of QuasiPeriodic Cocycles . . . . . . . . . . . . . . . . . . . . 1.4 End of Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Extension to Real Analytic Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Cantor Spectrum for Speciﬁc Models . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
79 79 80 83 84 86 87 88 90
Contents
IX
Part II Quantum Field Theory and Statistical Mechanics Adiabatic Theorems and Reversible Isothermal Processes W.K. AbouSalem and J. Fr¨ ohlich . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 A General “Adiabatic Theorem” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 The “Isothermal Theorem” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 (Reversible) Isothermal Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
95 95 97 99 101 104
Quantum Massless Field in 1+1 Dimensions J. Derezi´ nski and K.A. Meissner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Poincar´e Covariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Changing the Compensating Functions . . . . . . . . . . . . . . . . . . . . . . . . . 5 Hilbert Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Fields in Position Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 The SL(2, R) × SL(2, R) Covariance . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Normal Ordering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Classical Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Algebraic Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Vertex Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Fermions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 Supersymmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
107 107 108 111 112 113 115 116 117 118 120 122 123 125 126
Stability of MultiPhase Equilibria M. Merkli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Stability of a SinglePhase Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 The Free Bose Gas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Spontaneous Symmetry Breaking and MultiPhase Equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Return to Equilibrium in Absence of a Condensate . . . . . . . . . . 1.4 Return to Equilibrium in Presence of a Condensate . . . . . . . . . 1.5 Spectral Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Stability of MultiPhase Equilibria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Quantum Tweezers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 NonInteracting System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Interacting System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Stability of the Quantum Tweezers, Main Results . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
129 129 129 133 135 135 136 137 138 141 146 147 148
X
Contents
Ordering of Energy Levels in Heisenberg Models and Applications B. Nachtergaele and S. Starr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Proof of the Main Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 The TemperleyLieb Basis. Proof of Proposition 1 . . . . . . . . . . . . . . . 3.1 The Basis for Spin 1/2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 The Basis for Higher Spin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 The Spin 1/2 SUq (2)symmetric XXZ Chain . . . . . . . . . . . . . . . 4.2 Higher Order Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Diagonalization at Low Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 The Ground States of Fixed Magnetization for the XXZ Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Aldous’ Conjecture for the Symmetric Simple Exclusion Process . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
149 149 152 158 158 160 163 163 165 165 165 166 167 169
Interacting Fermions in 2 Dimensions V. Rivasseau . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Fermi Liquids and Salmhofer’s Criterion . . . . . . . . . . . . . . . . . . . . . . . . 3 The Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 A Brief Review of Rigorous Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Multiscale Analysis, Angular Sectors . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 One and Two Particle Irreducible Expansions . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
171 171 171 173 174 175 176 178
On the Essential Spectrum of the Translation Invariant Nelson Model J. SchachMøller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 The Model and the Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 A Complex Function of Two Variables . . . . . . . . . . . . . . . . . . . . . . . . . 3 The Essential Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A Riemannian Covers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
179 179 182 189 194 195
Part III Quantum Kinetics and BoseEinstein Condensation BoseEinstein Condensation as a Quantum Phase Transition in an Optical Lattice M. Aizenman, E.H. Lieb, R. Seiringer, J.P. Solovej and J. Yngvason . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
Contents
2 Reﬂection Positivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Proof of BEC for Small λ and T . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Absence of BEC and Mott Insulator Phase . . . . . . . . . . . . . . . . . . . . . 5 The NonInteracting Gas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Long Time Behaviour to the Schr¨ odinger–Poisson–X α Systems ´ S´ O. Bokanowski, J.L. L´ opez, O. anchez and J. Soler . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 On the Derivation of the Slater Approach . . . . . . . . . . . . . . . . . . . . . . 3 Some Results Concerning Well Posedness and Asymptotic Behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Existence and Uniqueness of Physically Admissible Solutions . 3.2 Minimum of Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Optimal Kinetic Energy Bounds . . . . . . . . . . . . . . . . . . . . . . . . . . 4 LongTime Behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 On the General X α Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Towards the Quantum Brownian Motion L. Erd˝ os, M. Salmhofer and H.T. Yau . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Statement of Main Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Sketch of the Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Renormalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 The Expansion and the Stopping Rules . . . . . . . . . . . . . . . . . . . . 3.3 The L2 Norm of the NonRepetitive Wavefunction . . . . . . . . . . 3.4 Sketch of the Proof of the Main Technical Theorem . . . . . . . . . 3.5 Point Singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Computation of the Main Term and Its Convergence to a Brownian Motion . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . BoseEinstein Condensation and Superradiance J.V. Pul´e, A.F. Verbeure and V.A. Zagrebnov . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Solution of the Model 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 The Eﬀective Hamiltonian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 The Pressure for Model 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Model 2 and MatterWave Grating . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
XI
203 205 210 213 214 214
217 217 220 223 223 225 227 228 230 231 233 233 237 242 242 243 245 250 253 255 257 259 259 263 263 270 272 276 277
XII
Contents
Derivation of the GrossPitaevskii Hierarchy B. Schlein . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 The Main Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Sketch of the Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
279 279 286 290 292
Towards a Microscopic Derivation of the Phonon Boltzmann Equation H. Spohn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Microscopic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Kinetic Limit and Boltzmann Equation . . . . . . . . . . . . . . . . . . . . . . . . 4 Feynman Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
295 295 296 298 300 304
Part IV Disordered Systems and Random Operators On the Quantization of Hall Currents in Presence of Disorder J.M. Combes, F. Germinet and P.D. Hislop . . . . . . . . . . . . . . . . . . . . . . . 1 The Edge Conductance and General Invariance Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Regularizing the Edge Conductance in Presence of Impurities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Generalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 A Time Averaged Regularization for a Dynamically Localized System . . . . . . . . . . . . . . . . . . . . . . . 2.3 Regularization Under a Stronger form of Dynamical Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Localization for the Landau Operator with a HalfPlane Random Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 A Large Magnetic Field Regime . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 A Large Disorder Regime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Equality of the Bulk and Edge Hall Conductances in 2D A. Elgart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction and Main Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Proof of σB = σE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Some Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Convergence and Trace Class Properties . . . . . . . . . . . . . . . . . . . 2.3 Edge – Bulk Interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 σE = σB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
307 307 310 310 312 314 317 317 319 321 325 325 329 329 330 330 331 332
Contents
XIII
Generic Subsets in Spaces of Measures and Singular Continuous Spectrum D. Lenz and P. Stollmann . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Generic Subsets in Spaces of Measures . . . . . . . . . . . . . . . . . . . . . . . . . 3 Singular Continuity of Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Selfadjoint Operators and the Wonderland Theorem . . . . . . . . . . . . . 5 Operators Associated to Delone Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
333 333 334 334 336 338 341
Low Density Expansion for Lyapunov Exponents H. SchulzBaldes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Model and Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Result on the Lyapunov Exponent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Result on the Density of States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
343 343 344 346 347 349 350
Poisson Statistics for the Largest Eigenvalues in Random Matrix Ensembles A. Soshnikov . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Wigner Random Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Band Random Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Sample Covariance Random Matrices . . . . . . . . . . . . . . . . . . . . . . 1.4 Universality in Random Matrices . . . . . . . . . . . . . . . . . . . . . . . . . 2 Wigner and Band Random Matrices with Heavy Tails of Marginal Distributions . . . . . . . . . . . . . . . . . . . . . 3 Real Sample Covariance Matrices with Cauchy Entries . . . . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
351 351 351 354 355 356 356 360 362 363
Part V Semiclassical Analysis and Quantum Chaos Recent Results on Quantum Map Eigenstates S. De Bi`evre . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Perturbed CAT Maps: Classical Dynamics . . . . . . . . . . . . . . . . . . . . . . 3 Quantum Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 What is Known? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Perturbed Cat Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
367 367 368 370 371 377 380
XIV
Contents
Level Repulsion and Spectral Type for OneDimensional Adiabatic QuasiPeriodic Schr¨ odinger Operators A. Fedotov and F. Klopp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 A Heuristic Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Mathematical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 The Periodic Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 A “Geometric” Assumption on the Energy Region Under Study . . . . . . . . . . . . . . . . . . . . . . . 2.3 The Deﬁnitions of the Phase Integrals and the Tunelling Coeﬃcients . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Ergodic Family . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 A Coarse Description of the Location of the Spectrum in J . . 2.6 A Precise Description of the Spectrum . . . . . . . . . . . . . . . . . . . . . 2.7 The Model Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.8 When τ is of Order 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.9 Numerical Computations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Low Lying Eigenvalues of Witten Laplacians and Metastability (After HelﬀerKleinNier and HelﬀerNier) B. Helﬀer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Main Goals and Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Saddle Points and Labelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Rough SemiClassical Analysis of Witten Laplacians and Applications to Morse Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Previous Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Witten Laplacians on pForms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Morse Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Main Result in the Case of Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 About the Proof in the Case of Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Witten Complex, Reduced Witten Complex . . . . . . . . . . . . . . . . 5.3 Singular Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 The Main Result in the Case with Boundary . . . . . . . . . . . . . . . . . . . . 7 About the Proof in the Case with Boundary . . . . . . . . . . . . . . . . . . . . 7.1 Deﬁne the Witten Complex and the Associate Laplacian . . . . . 7.2 Rough Localization of the Spectrum of this Laplacian on 1Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3 Construction of WKB Solutions Attached to the Critical Points of Index 1 . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
383 383 387 388 388 389 391 392 393 399 400 400 401
403 403 404 406 406 407 407 407 408 408 409 409 410 411 411 412 412 414
Contents
XV
The Mathematical Formalism of a Particle in a Magnetic Field M. M˘ antoiu and R. Purice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 The Classical Particle in a Magnetic Field . . . . . . . . . . . . . . . . . . . . . . 2.1 Two Hamiltonian Formalisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Magnetic Translations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 The Quantum Picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 The Magnetic Moyal Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 The Magnetic Moyal Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 The Twisted Crossed Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Abstract Aﬃliation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 The Limit → 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 The Schr¨odinger Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Representations of the Twisted Crossed Product . . . . . . . . . . . . 5.2 Pseudodiﬀerential Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 A New Justiﬁcation: Functional Calculus . . . . . . . . . . . . . . . . . . 5.4 Concrete Aﬃliation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Applications to Spectral Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1 The Essential Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 A NonPropagation Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
417 417 418 418 421 422 422 423 424 426 426 427 427 429 429 430 431 431 432 433
Fractal Weyl Law for Open Chaotic Maps S. Nonnenmacher . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Generalities on Resonances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Trapped Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Fractal Weyl Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4 Open Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 The Open Baker’s Map and Its Quantization . . . . . . . . . . . . . . . . . . . 2.1 Classical Closed Baker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Opening the Classical Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Quantum Baker’s Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Resonances of the Open Baker’s Map . . . . . . . . . . . . . . . . . . . . . . 3 A Solvable Toy Model for the Quantum Baker . . . . . . . . . . . . . . . . . . 3.1 Description of the Toy Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . N as a WalshQuantized Baker . . . . . . . . . . . 3.2 Interpretation of C 3.3 Resonances of CN =3k . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
435 435 435 436 437 438 439 439 440 441 442 446 446 446 447 449
Spectral Shift Function for Magnetic Schr¨ odinger Operators G. Raikov . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Auxiliary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Notations and Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
451 451 453 453
XVI
Contents
2.2 A. Pushnitski’s Representation of the SSF . . . . . . . . . . . . . . . . . Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Singularities of the SSF at the Landau Levels . . . . . . . . . . . . . . 3.2 Strong Magnetic Field Asymptotics of the SSF . . . . . . . . . . . . . 3.3 High Energy Asymptotics of the SSF . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
453 455 455 460 462 463
Counting String/M Vacua S. Zelditch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Type IIb Flux Compactiﬁcations of String/M Theory . . . . . . . . . . . . 3 Critical Points and Hessians of Holomorphic Sections . . . . . . . . . . . . 4 The Critical Point Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Statement of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Comparison to the Physics Literature . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Sketch of Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Other Formulae for the Critical Point Density . . . . . . . . . . . . . . . . . . 9 Black Hole Attractors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
467 467 468 470 471 473 474 475 476 480 481
3
List of Contributors
W.K. AbouSalem Institute for Theoretical Physics, ETHH¨ onggerberg CH8093 Z¨ urich, Switzerland [emailprotected]
de Math´ematiques et Laboratoire Painlev´e, 59655 Villeneuve d’Ascq Cedex, France [emailprotected]
M. Aizenman Departments of Mathematics and Physics Jadwin Hall Princeton University P. O. Box 708, Princeton New Jersey, 08544 USA
O. Bokanowski Laboratoire Jacques Louis Lions (Paris VI) UFR Math´ematique B. P. 7012 & Universit´e Paris VII Paris, Paris Cedex 05 France [emailprotected]
A. Avila CNRS UMR 7599 Laboratoire de Probabilit´es et Mod`eles al´eatoires Universit´e Pierre et Marie Curie–Boite courrier 188, 75252–Paris Cedex 05 France [emailprotected] V. Betz Institute for Biomathematics and Biometry GSF Forschungszentrum Postfach 1129 D85758 Oberschleißheim Germany [emailprotected] S. De Bi` evre Universit´e des Sciences et Technologies de Lille, UFR
M. B¨ uttiker D´epartement de Physique Th´eorique Universit´e de Gen`eve CH1211 Gen`eve 4, Switzerland [emailprotected] J.M. Combes D´epartement de Math´ematiques Universit´e de ToulonVar BP 132, 83957 La Garde C´edex France [emailprotected] H.D. Cornean Department of Mathematical Sciences Aalborg University Fredrik Bajers Vej 7G 9220 Aalborg, Denmark [emailprotected]
XVIII List of Contributors
J. Derezi´ nski Department of Mathematics and Methods in Physics Warsaw University Ho˙za 74 00682 Warsaw Poland [emailprotected]
J. Fr¨ ohlich Institute for Theoretical Physics ETHH¨onggerberg CH8093 Z¨ urich Switzerland [emailprotected]
A. Elgart Department of Mathematics Stanford University Stanford, CA 943052125 USA [emailprotected]
F. Germinet D´epartement de Math´ematiques Universit´e de CergyPontoise Site de SaintMartin 2 avenue Adolphe Chauvin 95302 CergyPontoise C´edex France [emailprotected]
L. Erd˝ os Institute of Mathematics University of Munich Theresienstr. 39 D80333 Munich Germany [emailprotected]. de P. Exner Department of Theoretical Physics Nuclear Physics Institute Academy of Sciences ˇ z near Prague 25068 Reˇ Czechia and Doppler Institute Czech Technical University Bˇrehov´ a 7, 11519 Prague, Czechia [emailprotected] A. Fedotov Department of Mathematical Physics St Petersburg State University 1 Ulianovskaja 198904 St PetersburgPetrodvorets, Russia mailto:[emailprotected] [emailprotected]
S.D. Glazek Institute of Theoretical Physics Warsaw University 00681 Hoza 69 Poland [emailprotected] B. Helﬀer D´epartement de Math´ematiques Bˆat. 425, Universit´e Paris Sud UMR CNRS 8628, F91405 Orsay Cedex France [emailprotected] P.D. Hislop Department of Mathematics University of Kentucky Lexington KY 405060027 USA [emailprotected] A. Jensen Department of Mathematical Sciences Aalborg University Fredrik Bajers Vej 7G 9220 Aalborg, Denmark [emailprotected]
List of Contributors
S. Jitomirskaya University of California Irvine Department of Mathematics 243 Multipurpose Science & Technology Building Irvine, CA 926973875 USA [emailprotected] F. Klopp LAGA, Institut Galil´ee U.M.R 7539 C.N.R.S Universit´e ParisNord Avenue J.B. Cl´ement F93430 Villetaneuse, France [emailprotected] D. Lenz Fakult¨ at f¨ ur Mathematik Technische Universit¨at 09107 Chemnitz, Germany d.lenz@@mathematik.tuchemnitz.de E.H. Lieb Departments of Mathematics and Physics Jadwin Hall Princeton University P. O. Box 708, Princeton New Jersey, 08544 USA J.L. L´ opez Departamento de Matem´atica Aplicada Facultad de Ciencias Universidad de Granada Campus Fuentenueva 18071 Granada Spain [emailprotected] M. M˘ antoiu “Simion Stoilow” Institute of Mathematics, Romanian Academy 21, Calea Grivitei Street 010702Bucharest, Sector 1 Romania [emailprotected]
XIX
K.A. Meissner Institute of Theoretical Physics Warsaw University Ho˙za 69 00681 Warsaw Poland [emailprotected] M. Merkli Department of Mathematics and Statistics, McGill University 805 Sherbrooke W., Montreal QC Canada, H3A 2K6 and Centre des Recherches Math´ematiques, Universit´e de Montr´eal, Succursale centreville Montr´eal QC Canada, H3C 3J7 [emailprotected] V. Moldoveanu National Institute of Materials Physics, P.O. Box MG7 Magurele, Romania [emailprotected] J.S. Møller Aarhus University Department of Mathematical Sciences 8000 Aarhus C, Denmark [emailprotected] M. Moskalets Department of Metal and Semiconductor Physics National Technical University “Kharkiv Polytechnic Institute” 61002 Kharkiv, Ukraine [emailprotected] B. Nachtergaele Department of Mathematics University of California Davis One Shields Avenue Davis, CA 956168366, USA [emailprotected]
XX
List of Contributors
S. Nonnenmacher Service de Physique Th´eorique CEA/DSM/PhT, Unit´e de recherche associ´ee au CNRS, CEA/Saclay 91191 GifsurYvette France [emailprotected] J. Puig Departament de Matem`atica Aplicada I Universitat Polit`ecnica de Catalunya. Av. Diagonal 647, 08028 Barcelona, Spain
M. Salmhofer Theoretical Physics University of Leipzig Augustusplatz 10 D04109 Leipzig Germany and Max–Planck Institute for Mathematics, Inselstr. 22 D04103 Leipzig Germany [emailprotected]
J.V. Pul´ e Department of Mathematical Physics University College Dublin, Belﬁeld, Dublin 4 Ireland [emailprotected]
´ S´ O. anchez Departamento de Matem´atica Aplicada, Facultad de Ciencias Universidad de Granada Campus Fuentenueva 18071 Granada Spain [emailprotected]
R. Purice “Simion Stoilow” Institute of Mathematics, Romanian Academy 21, Calea Grivitei Street 010702Bucharest, Sector 1 Romania [emailprotected]
B. Schlein Department of Mathematics Stanford University Stanford, CA 94305, USA [emailprotected]
G. Raikov Facultad de Ciencias Universidad de Chile Las Palmeras 3425 Santiago, Chile [emailprotected] V. Rivasseau Laboratoire de Physique Th´eorique CNRS, UMR 8627 Universit´e de ParisSud 91405 Orsay France [emailprotected]
H. SchulzBaldes Mathematisches Institut FriedrichAlexanderUniversit¨ at ErlangenN¨ urnberg Bismarckstr. 1 D91054, Erlangen [emailprotected] R. Seiringer Departments of Mathematics and Physics, Jadwin Hall Princeton University P. O. Box 708, Princeton New Jersey, 08544 USA
List of Contributors
J. Soler Departamento de Matem´atica Aplicada, Facultad de Ciencias Universidad de Granada Campus Fuentenueva 18071 Granada Spain [emailprotected] J.P. Solovej Department of Mathematics University of Copenhagen Universitetsparken 5 DK2100 Copenhagen, Denmark and Institut f¨ ur Theoretische Physik Universit¨at Wien Boltzmanngasse 5 A1090 Vienna, Austria [emailprotected] A. Soshnikov University of California at Davis Department of Mathematics Davis, CA 95616, USA [emailprotected] H. Spohn Zentrum Mathematik and Physik Department TU M¨ unchen D85747 Garching Boltzmannstr 3, Germany [emailprotected]
XXI
S. Teufel Mathematisches Institut Universit¨at T¨ ubingen Auf der Morgenstelle 10 72076 T¨ ubingen Germany [emailprotected] A.F. Verbeure Instituut voor Theoretische Fysika, Katholieke Universiteit Leuven Celestijnenlaan 200D, 3001 Leuven, Belgium [emailprotected]. ac.be H.T. Yau Department of Mathematics Stanford University, CA94305, USA [emailprotected] J. Yngvason Institut f¨ ur Theoretische Physik Universit¨at Wien Boltzmanngasse 5 A1090 Vienna, Austria and Erwin Schr¨ odinger Institute for Mathematical Physics Boltzmanngasse 9 A1090 Vienna, Austria
S. Starr Department of Mathematics University of California Los Angeles, Box 951555 Los Angeles, CA 900951555 USA [emailprotected]
V.A. Zagrebnov Universit´e de la M´editerran´ee and Centre de Physique Th´eorique, LuminyCase 907, 13288 Marseille, Cedex 09 France [emailprotected]
P. Stollmann Fakult¨ at f¨ ur Mathematik Technische Universit¨at 09107 Chemnitz, Germany [emailprotected]
S. Zelditch Department of Mathematics Johns Hopkins University Baltimore MD 21218 USA [emailprotected]
Introduction
QMath9 gave a particular importance to summarize the state of the art of the ﬁeld in a perspective to transmit knowledge to younger scientists. The main contributors to the ﬁeld were gathered in order to communicate results, open questions and motivate new research by the confrontation of diﬀerent view points. The edition of this book follows this spirit; the main eﬀort of the authors and editors is to help ﬁnding an access to the variety of themes and the sometimes very sophisticated literature in Mathematical Physics. The contributions of this book are organized in ﬁve topical groups: Quantum Dynamics and Spectral Theory, Quantum Field Theory and Statistical Mechanics, Quantum Kinetics and BoseEinstein Condensation, Disordered Systems and Random Operators, Semiclassical Analysis and Quantum Chaos. This splitting is admittedly somewhat arbitrary, since there are overlaps between the topics and the frontiers between the chosen groups may be quite fuzzy. Moreover, there are close connections between the tools and techniques used in the analysis of quite diﬀerent physical phenomena. An introduction to each theme is given. This plan is intended as a readers guide rather than as an attempt to put contributions into well deﬁned categories.
Solving the Ten Martini Problem Artur Avila1 and Svetlana Jitomirskaya2 1
2
CNRS UMR 7599, Laboratoire de Probabilit´es et Mod`eles al´eatoires Universit´e Pierre et Marie Curie–Boite courrier 188 75252–Paris Cedex 05, France [emailprotected] University of California, Irvine, California The research of S.J. was partially supported by NSF under DMS0300974 [emailprotected]
Abstract. We discuss the recent proof of Cantor spectrum for the almost Mathieu operator for all conjectured values of the parameters.
1 Introduction The almost Mathieu operator (a.k.a. the Harper operator or the Hofstadter model) is a Schr¨ odinger operator on 2 (Z), (Hλ,α,θ u)n = un+1 + un−1 + 2λ cos 2π(θ + nα)un , where λ, α, θ ∈ R are parameters (the coupling, the frequency and the phase). This model ﬁrst appeared in the work of Peierls [21]. It arises in physics literature as related, in two diﬀerent ways, to a twodimensional electron subject to a perpendicular magnetic ﬁeld [15, 23]. It plays a central role in the Thouless et al theory of the integer quantum Hall eﬀect [27]. The value of λ of most interest from the physics point of view is λ = 1. It is called the critical value as it separates two diﬀerent behaviors as far as the nature of the spectrum is concerned. If α = pq is rational, it is well known that the spectrum consists of the union of q intervals possibly touching at endpoints. In the case of irrational α the spectrum (which then does not depend on θ) has been conjectured for a long time to be a Cantor set for all λ = 0 [7]. To prove this conjecture has been dubbed the Ten Martini problem by Barry Simon, after an oﬀer of Kac in 1981, see Problem 4 in [25]. In 1984 Bellissard and Simon [8] proved the conjecture for generic pairs of (λ, α). In 1987 Sinai [26] proved Cantor spectrum for a.e. α in the perturbative regime: for λ = λ(α) suﬃciently large or small. In 1989 HelﬀerSj¨ostrand proved Cantor spectrum for the critical value λ = 1 and an explicitly deﬁned generic set of α [16]. Most developments in the 90s were related to the following observation. For α = pq the spectrum of Hλ,α,θ can have at most q − 1 gaps. It turns out that all these gaps are open, except for the middle one for even q [11, 20]. Choi, Eliott, and Yui obtained in fact an exponential A. Avila and S. Jitomirskaya: Solving the Ten Martini Problem, Lect. Notes Phys. 690, 5–16 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
6
A. Avila and S. Jitomirskaya
lower bound on the size of the individual gaps from which they deduced Cantor spectrum for Liouville (exponentially well approximated by the rationals) α [11]. In 1994 Last, using certain estimates of Avron, van Mouche and Simon [6], proved zero measure Cantor spectrum for a.e. α (for an explicit set that intersects with but does not contain the set in [16]) and λ = 1 [18]. Just extending this result to the case of all (rather than a.e.) α was considered a big challenge (see Problem 5 in [25]). A major breakthrough came recently with an inﬂux of ideas coming from dynamical systems. Puig, using Aubry duality [1] and localization for θ = 0 and λ > 1 [13], proved Cantor spectrum for Diophantine α and any noncritical λ [22]. At about the same time, Avila and Krikorian proved zero measure Cantor spectrum for λ = 1 and α satisfying a certain Diophantine condition, therefore extending the result of Last to all irrational α [3]. The solution of the Ten Martini problem as originally stated was ﬁnally given in [2]: Main Theorem [2]. The spectrum of the almost Mathieu operator is a Cantor set for all irrational α and for all λ = 0. Here we present the broad lines of the argument of [2]. For a much more detailed account of the history as well as of the physics background and related developments see a recent review [19]. While the ten martini problem was solved, a stronger version of it, dubbed by B. Simon the Dry Ten Martini problem is still open. The problem is to prove that all the gaps prescribed by the gap labelling theorem are open. This fact would be quite meaningful for the QHE related applications [4]. Dry ten martini was only established for Liouville α [2, 11] and for Diophantine α in the perturbative regime [22], using a theorem of Eliasson [12]. 1.1 Rough Strategy The history of the Ten Martini problem we described shows the existence of a number of diﬀerent approaches, applicable on diﬀerent parameter ranges. Denote by Σλ,α the union over θ ∈ R of the spectrum of Hλ,α,θ (recall that the spectrum is actually θindependent if α ∈ R \ Q). Due to the obvious symmetry Σλ,α = −Σ−λ,α , we may assume that λ > 0. Aubry duality gives a much more interesting symmetry, which implies that Σλ,α = λΣλ−1 ,α . The critical coupling λ = 1 separates two very distinct regimes. The transition at λ = 1 can be clearly seen by consideration of the Lyapunov exponent L(E) = Lλ,α (E), for which we have the following statement. Theorem 1. [9] Let λ > 0, α ∈ R \ Q. For every E ∈ Σλ,α , Lλ,α (E) = max{ln λ, 0}. With respect to the frequency α, one can broadly distinguish two approaches, applicable depending on whether α is well approximated by rationals or not (the Liouville and the Diophantine cases):
Solving the Ten Martini Problem
7
1. In the Liouvillian region, one can try to proceed by rational approximation, exploiting the fact that a signiﬁcant part of the behavior at rational frequencies is accessible by calculation (this is a very special property of the cosine potential). 2. In the Diophantine region, one can attemp to solve two small divisor problems that have been linked with Cantor spectrum. (a) Localization (for large coupling), whose relevance to Cantor spectrum was shown in [22]. (b) Floquet reducibility (for small coupling), which is connected to Cantor spectrum in [12, 22]. Although Aubry duality relates both problems for λ = 1, it is important to notice that the small divisor analysis is much more developed in the localization problem, where powerful nonperturbative methods are currently available. To decide whether α should be considered Liouville or Diophantine for the Ten Martini problem, we introduce a parameter β = β(α) ∈ [0, ∞]: β = lim sup n→∞
ln qn+1 , qn
(1)
where pqnn are the rational approximations of α (obtained by the continued fraction algorithm). As β grows, the Diophantine approach becomes less and less eﬃcient, until it ceases to work, while the opposite happens for the Liouville approach. As discussed before, those lines of attack lead to the solution of the Ten Martini problem in a very large region of the parameters, which is both generic and of full Lebesgue measure. However there is no reason to expect that one could cover the whole parameter range by this Liouville/Diophantine dichotomy. Actually our analysis seems to indicate the existence of a critical range, β ≤  ln λ ≤ 2β, where one is close enough to the rationals to make the small divisor problems intractable (so that, in particular, localization does not hold in the full range of phases for which it holds for larger λ), but not close enough so that one can borrow their gaps. In order to go around the (seemingly) very real issues present in the critical range, we will use a somewhat convoluted argument which proceeds by contradiction. The contradiction argument allows us to exploit the following new idea: roughly, absence of Cantor spectrum is shown to imply much better, irrealistically good estimates. Still, those “ﬁctitious” estimates are barely enough to cover the critical range of parameters, and we are forced to push the more direct approaches close to their technical limits. We will need to apply this trick both in the Liouvillian side and in the Diophantine side. In the Liouvillian side, it implies improved continuity estimates for the dependence of the spectrum on the frequency. In the Diophantine side, it immediately solves the “noncommutative” part of Floquet
8
A. Avila and S. Jitomirskaya
reducibility: what remains to do is to solve the cohom*ological equation. Unfortunately, this can not be done directly. Instead, what we pick up from the (“soft”) analysis of the cohom*ological equation is used to complement the (“hard”) analysis of localization. In the following sections we will succesively describe the analytic extension trick, the Liouville estimates, the two aspects of the Diophantine side (reducibility and localization), and we will conclude with some aspects of the proof of localization.
2 Analytic Extension In Kotani theory, the complex analytic properties of Weyl’s mfunctions are used to describe the absolutely continuous component of the spectrum of an ergodic Schr¨ odinger operator. However, it can also be interpreted as a theory about certain dynamical systems, cocycles. We restrict to the case of the almost Mathieu operator. A formal solution of Hλ,α,θ u = Eu, u ∈ CZ , satisﬁes the equation E − 2λ cos 2π(θ + nα) −1 un un+1 · = . (2) 1 0 un−1 un
E − 2λ cos 2πx −1 , the importance of the products 1 0 Sλ,E (θ + (n − 1)α) · · · Sλ,E (θ) becomes clear. Since Sλ,E are matrices in ab · z = az+b SL(2, C), which has a natural action on C, cz+d , this leads cd to the consideration of the dynamical system Deﬁning Sλ,E (x) =
(α, Sλ,E ) : R/Z × C → R/Z × C (x, w) → (x + α, Sλ,E (x) · w) ,
(3)
which is the projective presentation of the almost Mathieu cocycle. An invariant section for the cocycle (α, Sλ,E ) is a function m : R/Z → C such that Sλ,E (x) · m(x) = m(x + α). The existence of a (suﬃciently regular) invariant section is of course a nice feature, as it in a sense means that the cocycle does not see the whole complexity of the group SL(2, C): the cocycle is conjugate to a cocycle in a simpler group (of triangular matrices). The existence of two distinct invariant sections means that the simpler group is isomorphic to an even simpler, abelian group (of diagonal matrices). It turns out that the cocycle is well behaved when E belongs to the resolvent set C \ Σλ,α : it is hyperbolic, which in particular means the existence of two continuous invariant sections. Moreover, the dependence of the invariant sections on E is analytic. Kotani showed that the existence of an open interval J in the spectrum where the Lyapunov exponent is zero allows one to use
Solving the Ten Martini Problem
9
the Schwarz reﬂection principle with respect to E, and to conclude that the invariant sections can be analytically continued through J. Thus for E ∈ J, there are still two continuous invariant sections. A crucial new idea is that those invariant sections are actually analytic also in the other variable. Theorem 2. Let 0 < λ ≤ 1, α ∈ R \ Q. Let J ∈ Σλ,α be an open interval. For E ∈ J, there exists an analytic map BE : R/Z → SL(2, R) such that BE (x + α) · Sλ,E (x) · BE (x)−1 ∈ SO(2, R) ,
(4)
that is, (α, Sλ,E ) is analytically conjugate to a cocycle of rotations. Moreover, (x, E) → BE (x) is analytic for (x, E) ∈ R/Z × J. The proof uses the analyticity of the almost Mathieu cocycle (α, Sλ,E ) coupled with an analytic extension (Hartogs) argument.
3 The Liouvillian Side The rational approximation argument centers around two estimates, on the size of gaps for rational frequencies, and on the modulus of continuity (in the Hausdorﬀ topology) of the spectrum as a function of the frequency. 3.1 Gaps for Rational Approximants The best eﬀective estimate for the size of gaps had been given in [11], which established that all gaps of Σλ, pq (except the central collapsed gap for q even) have size at least C(λ)−q , where C(λ) is some explicit constant (for instance, C(1) = 8). Such eﬀective constants are not good enough for our argument (for instance, it is important to have C(λ) close to 1 when λ is close to 1). On the other hand, we only need asymptotic estimates, addressing rationals p q approximating some given irrational frequency for which we want to prove Cantor spectrum. Theorem 3. Let α ∈ R \ Q, λ > 0. For every > 0, if pq is close enough to α then all open gaps of Σλ, pq have size at least e−( ln λ+)q/2 . It was pointed out to us by Bernard Helﬀer (during the Qmath9 conference) that this asymptotic estimate does not hold under the sole assumption of q → ∞, as is demonstrated by the analysis of Helﬀer and Sjostrand, so it is important to only consider approximations of a given irrational frequency. The proof starts as in [11], which gives a global inequality relating all bands in the spectrum. We then use the integrated density of states to get a better (asymptotic) estimate on the position of bands in the spectrum. Using the Thouless formula, we get an asymptotic estimate for the size of gaps near a given frequency α and near a given energy E ∈ Σλ,α in terms of the Lyapunov exponent Lλ,α (E). Theorem 1 then leads to the precise estimate above.
10
A. Avila and S. Jitomirskaya
3.2 Continuity of the Spectrum The best general result on continuity of the spectrum was obtained in [6], 1/2H¨ older continuity. Coupled with the gap estimate for rational approximants, we get the following contribution to the Dry Ten Martini problem. Theorem 4. If e−β < λ < eβ then all gaps of Σλ, pq are open. Unfortunately this cannot be complemented by any Diophantine method that in one way or another requires localization, as it would miss the parameters such that  ln λ = β > 0. Indeed, there are certain reasons to believe that, for any θ operator Hλ,α,θ has no exponentially decaying eigenfunctions for λ ≤ eβ . Better estimates on continuity of the spectrum were obtained by [14] in the Diophantine range, but these estimates get worse in the critical range and can not be used. What we do instead is a “ﬁctitious” improvement based on Theorem 2. Theorem 5. Let α ∈ R \ Q, λ > 0. If J ⊂ int Σλ,α is a closed interval then there exists C > 0 such that for every E ∈ J, and for every α ∈ R, there exists E ∈ Σλ,α with E − E  < Cα − α . This estimate, Lipschitz continuity, is obtained in the range 0 < λ ≤ 1 using Theorem 2 and a direct dynamical estimate on perturbations of cocycles of rotations. This result can be applied in an argument by contradiction: Theorem 6. If e−2β < λ < e2β then Σλ,α is a Cantor set.
4 The Diophantine Side The Diophantine side is ruled by small divisor considerations. Two traditional small divisor problems are associated to quasiperiodic Schr¨ odinger operators: localization for large coupling and Floquet reducibility for small coupling. Those two problems are largely related by Aubry duality. While originally both problems were attacked by perturbative methods (very large coupling for localization and very small coupling for reducibility, depending on speciﬁc Diophantine conditions), powerful nonperturbative estimates are now available for the localization problem. For this reason, all the eﬀective “hard analysis” we will do will be concentrated in the localization problem. However, those estimates by themselves are insuﬃcient. We will need an additional soft analysis argument (again analytic extension), carried out for the reducibility problem under the assumption of nonCantor spectrum, to improve (irrealistically) the localization results.
Solving the Ten Martini Problem
11
4.1 Reducibility We say that (α, Sλ,E ) is reducible if it is analytically conjugate to a constant cocycle, that is, there exists an analytic map B : R/Z → SL(2, R) such that B(x + α) · Sλ,E (x) · B(x)−1 is a constant A∗ . An important idea is that (α, Sλ,E ) is much more likely to be reducible if one assumes that E ∈ int Σλ,α , 0 < λ ≤ 1. Indeed most of reducibility is taken care by Theorem 2, which simpliﬁes the problem to proving reducibility for an analytic cocycle of rotations. This is a much easier task, which reduces to consideration of the classical cohom*ological equation φ(x) = ψ(x + α) − ψ(x) ,
(5)
which can be analysed via Fourier series: one has an explicit formula for 1 ˆ ˆ φ(k). The small divisors arise when the Fourier coeﬃcients ψ(k) = e2πikα −1 qαR/Z is small, where · R/Z denotes the distance to the nearest integer. This easily takes care of the case β = 0, but for β > 0 the information given by Theorem 2 is not quantitative enough to conclude. The analysis of the cohom*ological equation gives still the following interesting qualitative information. Theorem 7. Let α ∈ R \ Q and let 0 < λ ≤ 1. Assume that β < ∞. Let Λλ,α be the set of E ∈ Σλ,α such that (α, Sλ,E ) is reducible. If Λλ,α ∩ int Σλ,α has positive Lebesgue measure then Λλ,α has nonempty interior. The proof of this theorem uses again ideas from analytic extension. Let N = Nλ,α : R → [0, 1] be the integrated density of states. One of the key ideas of [22] is that if (α, Sλ,E ) is reducible for some E ∈ Σλ,α such that N (E) ∈ αZ + Z then E is the endpoint of an open gap. The argument is particular to the cosine potential, and involves Aubry duality. It, in fact, extends to the case of any analytic function such that the dual model (which in general will be longrange) has simple spectrum. Since an open subset of Σλ,α must intersect {E ∈ Σλ,α , N (E) ∈ αZ + Z}, we immediately obtain Cantor spectrum in the entire range of β = 0 just from the reducibility considerations alone. Note that β = 0 is strictly stronger than the Diophantine condition, and we did not use any localization result. As noted above, this β = 0 result extends to quasiperiodic potentials deﬁned by analytic functions under the condition that the Lyapunov exponent is zero on the spectrum1 and that the dual model has simple spectrum (it is actually enough to require that spectral multiplicities are nowhere dense). For 0 < β < ∞ it follows similarly that the hypothesis of the previous theorem must fail: Corollary 1. Let α ∈ R \ Q and let 0 < λ ≤ 1. If β < ∞ then Λλ,α ∩ int Σλ,α has zero Lebesgue measure. 1
This condition holds for all analytic functions for suﬃciently small λ (in a nonperturbative way) so that the result of [10] applies, thus by [9] L(E) is zero on the spectrum for all irrational α.
12
A. Avila and S. Jitomirskaya
4.2 Localization and Reducibility Aubry duality gives the following relation between reducibility and local/ αZ + Z then the following are ization. If E ∈ Σλ,α is such that N (E) ∈ equivalent: 1. (α, Sλ,E ) is reducible, 2. There exists θ ∈ R, such that 2θ ∈ ±N (E) + 2αZ + 2Z and λ−1 E is a localized eigenvalue (an eigenvalue for which the corresponding eigenfunction exponentially decays) of Hλ−1 ,α,θ . Remark 1. When N (E) ∈ αZ + Z, (1) still implies (2), but it is not clear that (2) implies (1) unless β = 0 (which covers the case treated in [22]). This is not however the main reason for us to avoid treating directly the case N (E) ∈ αZ + Z. Remark 2. The approach of [22] is to obtain a dense subset of {E ∈ Σλ,α , N (E) ∈ αZ + Z} for which (α, Sλ,E ) is reducible, for 0 < λ < 1 and α satisfying the Diophantine condition ln qn+1 = O(ln qn ), as a consequence of localization for Hλ−1 ,α,0 and Aubry duality. Such a localization result (for θ = 0) is however not expected to hold in the critical range of α, see more discussion in the next section. Thus proving localization of Hλ−1 ,α,θ for a large set of θ allows one to conclude reducibility of (α, Sλ,E ) for a large set of E. Coupled with Corollary 1, we get the following criterium for Cantor spectrum. Theorem 8. Let α ∈ R \ Q, 0 < λ ≤ 1. Assume that β < ∞. If Hλ−1 ,α,θ displays localization for almost every θ ∈ R then Σλ,α (and hence Σλ−1 ,α ) is a Cantor set.
5 A Localization Result In order to prove the Main Theorem, it remains to obtain a localization result that covers the pairs α ∈ R \ Q and λ > 1 which could not be treated by the Liouville method, namely the parameter region ln λ ≥ 2β. In proving localization of Hλ,α,θ , two kinds of small divisors intervene, 1. The usual ones for the cohom*ological equation, arising from q ∈ Z \ {0} for which qαR/Z is small, 2. Small denominators coming from q ∈ Z such that 2θ + qαR/Z is small. Notice that for any given α, a simple BorelCantelli argument allows one to obtain that for almost every θ the small denominators of the second kind satisfy polynomial lower bounds: 2θ + qαR/Z >
κ . (1 + q)2
(6)
Solving the Ten Martini Problem
13
When θ = 0, or more generally 2θ ∈ αZ + Z, which is the case linked to Cantor spectrum in [22], the small divisors of the second type are exactly the same as the ﬁrst type.2 When β > 0, where the small denominators of the ﬁrst type can be exponentially small, θ = 0 is thus much worse behaved than almost every θ, leading to a smaller range where one should be able to prove localization. More precisely, one expects that localization holds for almost every θ if and only if ln λ > β, and for θ = 0 if and only if ln λ > 2β. Even with all the other tricks, this would leave out the parameters such that ln λ = 2β > 0. In any case, the following localization result is good enough for our purposes. Theorem 9. Let α ∈ R \ Q. Assume that ln λ > localization for almost every θ ∈ R.
16 9 β.
Then Hλ,α,θ displays
This is the most technical result of [2]. We use the general setup of [13], however our key technical procedure is quite diﬀerent. It is well known that to prove localization of Hλ,α,θ it suﬃces to prove that all polynomially bounded solutions of Hλ,α,θ Ψ = EΨ decay exponentially. We will use the notation G[x1 ,x2 ] (x, y) for matrix elements of the Green’s function (H − E)−1 of the operator Hλ,α,θ restricted to the interval [x1 , x2 ] with zero boundary conditions at x1 − 1 and x2 + 1. It can be checked easily that values of any formal solution Ψ of the equation HΨ = EΨ at a point x ∈ I = [x1 , x2 ] ⊂ Z can be reconstructed from the boundary values via Ψ (x) = −GI (x, x1 )Ψ (x1 − 1) − GI (x, x2 )Ψ (x2 + 1) .
(7)
The strategy is to ﬁnd, for every large integer x, a large interval I = [x1 , x2 ] ⊂ Z containing x such that both G(x, x1 ) and G(x, x2 ) are exponentially small (in the length of I). Then, by using the “patching argument” of multiscale analysis, we can prove that Ψ (x) is exponentially small in x. (The key property of Ψ , that it is a generalized eigenfunction, is used to control the boundary terms in the blockresolvent expansion.) Fix m > 0. A point y ∈ Z will be called (m, k)regular if there exists an interval [x1 , x2 ], x2 = x1 + k − 1, containing y, such that G[x1 ,x2 ] (y, xi ) < e−my−xi  , and dist(y, xi ) ≥
1 k; i = 1, 2 . 40
We now have to prove that every x suﬃciently large is (m, k)regular for appropriate m and k. The precise procedure to follow will depend strongly on the position of x with respect to the sequence of denominators qn (we 8/9 1 assume that x > 0 for convenience). Let bn = max{qn , 20 qn−1 }. Let n be such that bn < x ≤ bn+1 . We distinguish between the two cases: 2
Actually there is an additional very small denominator, 0 of the second type, which leads to special considerations, but is not in itself a show stopper.
14
A. Avila and S. Jitomirskaya
1. Resonant: meaning x − qn  ≤ bn for some ≥ 1 and 2. Nonresonant: meaning x − qn  > bn for all ≥ 0. Theorem 9 is a consequence then of the following estimates: Lemma 1. Assume that θ satisﬁes (6). Suppose x is nonresonant. Let s ∈
N ∪ {0} be the largest number such that sqn−1 ≤ dist(x, {qn }≥0 ). Then for any > 0 for suﬃciently large n,
qn 1. If s ≥ 1 and ln λ > β, x is (ln λ − qlnn−1 − , 2sqn−1 − 1)regular. qn−1 2. If s = 0 then x is either (ln λ − , 2[ 2 ] − 1) or (ln λ − , 2[ q2n ] − 1) or (ln λ − , 2qn−1 − 1)regular.
Lemma 2. Let in addition ln λ > 16 9 β. Then for suﬃciently large n, every ln λ resonant x is ( 50 , 2qn − 1)regular. Each of those estimates is proved following a similar scheme, though the proof of Lemma 2 needs additional bootstrapping from the proof of Lemma 1. All small denominators considerations are entirely captured through the following concept: We will say that the set {θ1 , . . . , θk+1 } is uniform if max
max
z∈[−1,1] j=1,...,k+1
k+1 =1 =j
z − cos 2πθ ) < ek .  cos 2πθj − cos 2πθ )
(8)
The uniformity of some speciﬁc sequences can then be used to show that some y ∈ Z is regular following the scheme of [13]. In this approach, the goal is to ﬁnd two nonintersecting intervals, I1 around 0 and I2 around y, of combined length I1  + I2  = k + 1, such that we can establish the uniformity of {θi } where θi = θ + (x + k−1 2 )α, i = 1, . . . , k + 1, for x ranging through I1 ∪ I2 . The actual proof of uniformity depends on the careful estimates of trigonometric products along arithmetic progressions θ + jα. Since ln E − cos 2πθdθ = − ln 2 for any E ≤ 1 such estimates are equivalent to the analysis of large deviations in the appropriate ergodic theorem. A simple trigonometric expansion of (8) shows that uniformity involves equidistribution of the θi along with cumulative repulsion of ±θi (mod 1)’s, and thus involves both kinds of small divisors previously mentioned.
References 1. S. Aubry, The new concept of transition by breaking of analyticity. Solid State Sci. 8, 264 (1978). 2. A. Avila and S. Jitomirskaya, The ten martini problem. Preprint (www.arXiv. org).
Solving the Ten Martini Problem
15
3. A. Avila and R. Krikorian, Reducibility or nonuniform hyperbolicity for quasiperiodic Schr¨ odinger cocycles. Preprint (www.arXiv.org). To appear in Annals of Math. 4. J. E. Avron, D. Osadchy, and R. Seiler, A topological look at the quantum Hall eﬀect, Physics today, 38–42, August 2003. 5. J. Avron and B. Simon, Almost periodic Schr¨ odinger operators. II. The integrated density of states. Duke Math. J. 50, 369–391 (1983). 6. J. Avron, P. van Mouche, and B. Simon, On the measure of the spectrum for the almost Mathieu operator. Commun. Math. Phys. 132, 103–118 (1990). 7. M. Ya. Azbel, Energy spectrum of a conduction electron in a magnetic ﬁeld. Sov. Phys. JETP 19, 634–645 (1964). 8. J. Bellissard, B. Simon, Cantor spectrum for the almost Mathieu equation. J. Funct. Anal. 48, 408–419 (1982). 9. J. Bourgain, S. Jitomirskaya, Continuity of the Lyapunov exponent for quasiperiodic operators with analytic potential, Dedicated to David Ruelle and Yasha Sinai on the occasion of their 65th birthdays. J. Statist. Phys. 108, 1203–1218 (2002). 10. J. Bourgain, S. Jitomirskaya, Absolutely continuous spectrum for 1D quasiperiodic operators, Invent. math. 148, 453–463 (2002). 11. M.D. Choi, G.A. Eliott, N. Yui, Gauss polynomials and the rotation algebra. Invent. Math. 99, 225–246 (1990). 12. L. H. Eliasson, Floquet solutions for the onedimensional quasiperiodic Schr¨ odinger equation. Commun. Math. Phys. 146, 447–482 (1992). 13. S. Jitomirskaya, MetalInsulator transition for the almost Mathieu operator. Annals of Math. 150, 1159–1175 (1999). 14. S. Ya. Jitomirskaya, I. V. Krasovsky, Continuity of the measure of the spectrum for discrete quasiperiodic operators. Math. Res. Lett. 9, no. 4, 413–421 (2002). 15. P.G. Harper, Single band motion of conduction electrons in a uniform magnetic ﬁeld, Proc. Phys. Soc. London A 68, 874–892 (1955). 16. B. Helﬀer and J. Sj¨ ostrand, Semiclassical analysis for Harper’s equation. III. Cantor structure of the spectrum. Mm. Soc. Math. France (N.S.) 39, 1–124 (1989). 17. S. Kotani, Lyapunov indices determine absolutely continuous spectra of stationary random onedimensional Schr¨ odinger operators. Stochastic analysis (Katata/Kyoto, 1982), 225–247, NorthHolland Math. Library, 32, NorthHolland, Amsterdam, 1984. 18. Y. Last, Zero measure of the spectrum for the almost Mathieu operator, CMP 164, 421–432 (1994). 19. Y. Last, Spectral theory of SturmLiouville operators on inﬁnite intervals: a review of recent developments. Preprint 2004. 20. P.M.H. van Mouche, The coexistence problem for the discrete Mathieu operator, Comm. Math. Phys., 122, 23–34 (1989). 21. R. Peierls, Zur Theorie des Diamagnetismus von Leitungselektronen. Z. Phys., 80, 763–791 (1933). 22. J. Puig, Cantor spectrum for the almost Mathieu operator, Comm. Math. Phys, 244, 297–309 (2004). 23. A. Rauh, Degeneracy of Landau levels in chrystals, Phys. Status Solidi B 65, K131135 (1974). 24. B. Simon, Kotani theory for onedimensional stochastic Jacobi matrices, Comm. Math. Phys. 89, 227–234 (1983).
16
A. Avila and S. Jitomirskaya
25. B. Simon, Schr¨ odinger operators in the twentyﬁrst century, Mathematical Physics 2000, Imperial College, London, 283–288. 26. Ya. Sinai, Anderson localization for onedimensional diﬀerence Schr¨ odinger operator with quasiperiodic potential. J. Stat. Phys. 46, 861–909 (1987). 27. D.J. Thouless, M. Kohmoto, M.P. Nightingale and M. den Nijs, Quantised Hall conductance in a two dimensional periodic potential, Phys. Rev. Lett. 49, 405–408 (1982).
Swimming Lessons for Microbots Y. Avron Dept of Physics, Technion, Haifa, Israel
The powerpoint ﬁle of this lecture is available at http://physics.technion.ac.il/∼avron/ﬁles/ppt/robotschaos5.htm
Y. Avron: Swimming Lessons for Microbots, Lect. Notes Phys. 690, 17–17 (2006) www.springerlink.com
LandauZener Formulae from Adiabatic Transition Histories Volker Betz1 and Stefan Teufel2 1
2
Institute for Biomathematics and Biometry, GSF Forschungszentrum, Postfach 1129, D85758 Oberschleißheim, Germany [emailprotected] Mathematisches Institut, Universit¨ at T¨ ubingen, Auf der Morgenstelle 10, 72076 T¨ ubingen, Germany [emailprotected]
Abstract. We use recent results on precise coupling terms in the optimal superadiabatic basis in order to determine exponentially small transition probabilities in the adiabatic limit of timedependent twolevel systems. As examples, we discuss the LandauZener and the RosenZener models. Key words: Superadiabatic basis, exponential asymptotics, Darboux principle.
1 Introduction Transitions between separated energy levels of slowly timedependent quantum systems are responsible for many important phenomena in physics, chemistry and even biology. In the mathematical model the slow variation of the Hamiltonian is expressed by the smallness of the adiabatic parameter ε in the Schr¨ odinger equation (1) i∂s − H(εs) φ(t) = 0 , where H(t) is a family of selfadjoint operator on a suitable Hilbert space. In order to see in (1) nontrivial eﬀects from the timevariation of the Hamiltonian, one has to follow the solutions up to times s of order ε−1 . Alternatively one can transform (1) to the macroscopic time scale t = εs, resulting in the equation (2) iε∂t − H(t) φ(t) = 0 , and study solutions of (2) for times t of order one. Often one is interested in the situation where the Hamiltonian is timeindependent for large negative and positive times. Then one can consider the scattering limit and the aim is to compute the scattering amplitudes. In the simplest and at the same time paradigmatic example the Hamiltonian is just a 2 × 2 matrix Z(t) X(t) H(t) = , X(t) −Z(t) which can be chosen real symmetric and traceless without essential loss of generality [1]. With this choice for H(t), the Schr¨ odinger equation (2) is V. Betz and S. Teufel: LandauZener Formulae from Adiabatic Transition Histories, Lect. Notes Phys. 690, 19–32 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
20
V. Betz and S. Teufel
just an ordinary diﬀerential equation for the C2 valued function φ(t). But even this simple system displays a very interesting behavior, of which we will give an informal description here in the introduction. The mathematical mechanism which generates this behavior will be explained in the main body of this paper. We will assume that H(t) has two distinct eigenvalues {E+ (t), E− (t)} for any t and approaches constant matrices as t → ±∞. Then also the eigenvalues {E+ (t), E− (t)} and the orthonormal basis {v+ (t), v− (t)} of R2 consisting of the real eigenvectors of H(t) have limits as t → ±∞. By deﬁnition, the transition probability from the “upper” to the “lower” eigenstate is given by P = lim φ− (t)2 := lim  v− (t), φ(t)C2 2 , t→∞
(3)
t→∞
where φ(t) is a solution of (2) with lim φ− (t)2 = 1 − lim φ+ (t)2 = 0 .
t→−∞
(4)
t→−∞
Despite the presence of a natural small parameter, the adiabatic parameter ε 1, it is far from obvious how to compute P even to leading order in ε. This is because the transition amplitudes connecting diﬀerent energy levels are exponentially small with respect to ε, i.e. of order O(e−c/ε ) for some c > 0, and thus have no expansion in powers of ε. The result of a numerical computation of φ− (t) for a typical Hamiltonian H(t) is displayed in Fig. 1a. After rising to a value which is of order ε, φ−  falls oﬀ again and ﬁnally, in the regime where H(t) is approximately constant, settles for a value of order e−c/ε . It is no surprise that supt∈R φ− (t) is of order ε: this is just a consequence of the proof of the adiabatic theorem [8], and in fact we perform the relevant calculation in Sect. 2. There we see that the size of φ− (t) is determined by Φ
(a)
t
(b)
Φn t 0.0035
0.08
0.003 0.0025
0.06
0.002 0.04
0.0015 0.001
0.02
0.0005 10
5
5
10
15
20
t
4
2
2
4
t
Fig. 1. This ﬁgure shows the lower components of a numerical solution of (5) for ε = 1/6. In (a), the lower component in the adiabatic basis rises to a value order ε before approaching its exponentially small asymptotic value. In (b), the lower component in the optimal superadiabatic basis rises monotonically to its ﬁnal value. Note the diﬀerent axes scalings, as the asymptotic values in both pictures agree
LandauZener Formulae from Adiabatic Transition Histories
21
the size of the oﬀdiagonal elements of the adiabatic Hamiltonian Had (t). The latter is obtained by expressing (2) in the adiabatic basis {v+ (t), v− (t)}. More precisely, let U0 (t) be the orthogonal matrix that takes the adiabatic basis into the canonical basis. Then multiplication of (2) with U0 (t) from the left leads to iε∂t − Had (t) φad (t) := U0 (t) iε∂t − H(t) U0∗ (t)U0 (t)φ(t) = 0 , (5) where Had (t) = diag(E+ (t), E− (t)) − iεU0 (t)U˙ 0∗ (t). Clearly, the oﬀdiagonal elements of the matrix Had are of order ε, and φ− (t) is just the second component of φad (t). However, the O(ε) smallness of the coupling in the adiabatic Hamiltonian does not explain the exponentially small scattering regime in Fig. 1a. In the adiabatic basis, there is no easy way to see why this eﬀect should take place, although with some goodwill it may be guessed by a heuristic calculation to be presented in the next section. A natural strategy to understand the exponentially small scattering amplitudes goes back to M. Berry [1]: the solution of (2) with initial condition (4) remains in the positive adiabatic subspace spanned by v+ (t) only up to errors of order ε. Hence one should ﬁnd a better subspace, the optimal superadiabatic subspace, in which the solution remains up to exponentially small errors for all times. Since we are ultimately interested in the transition probabilities, at the same time this subspace has to coincide with the adiabatic subspace as t → ±∞. One way to determine the superadiabatic subspaces is to optimally truncate the asymptotic expansion of the true solution in powers of ε, as Berry [1] did. Alternatively one can look for a timedependent basis of C2 such that the analogues transformation to (5) yields a Hamiltonian with exponentially small oﬀdiagonal terms. To do so, one ﬁrst constructs the nth superadiabatic basis recursively from the adiabatic basis for any n ∈ N. Let us write Uεn (t) for the transformation taking the nth superadiabatic basis into the canonical one. Then as in (5) the Schr¨ odinger equation takes the form iε∂t − Hεn (t) φn (t) = 0 , (6) where Hεn (t)
=
ρnε (t)
εn+1 cnε (t)
εn+1 c¯nε (t) −ρnε (t)
n
and φ (t) =
Uεn (t)φ(t)
=
φn+ (t) φn− (t)
.
(7) Above, ρnε = 12 + O(ε2 ). While the oﬀdiagonal elements of Hεn indeed are of order εn+1 , the nth superadiabatic coupling function cnε grows like n! so that the function n → εn+1 cnε will diverge for each ε as n → ∞. However, for each ε > 0 there is an nε ∈ N such that εn+1 cnε takes its minimal value for n = nε . This deﬁnes the optimal superadiabatic basis. In this basis the oﬀdiagonal elements of Hεn (t) are exponentially small for all t. As a consequence, also the lower component φn− (t) of the solution with limt→−∞ φn− (t) = limt→−∞ φ− (t) = 0 is exponentially small, as illustrated in
22
V. Betz and S. Teufel
Fig. 1b, and one can compute the scattering amplitude by ﬁrst order perturbation theory. Berry and Lim [1, 2] showed on a nonrigorous level that φn− (t) is not only exponentially small in ε but has the universal form of an error function, a feature also illustrated in Fig. 1(b). A rigorous derivation of the optimal superadiabatic Hamiltonian and of the universal transition histories has been given recently in [3] and [4]. The aim of this note is to explain certain aspects of the results from [4] and to show how to obtain scattering amplitudes from them. In Sect. 2 we basically give a more detailed and also more technical introduction to the problem of exponentially small nonadiabatic transitions. Section 3 contains a concise summary of the results obtained in [4]. In order to apply these results to the scattering situation, we need some control on the time decay of the error estimates appearing in our main theorem. In Sect. 4 we use standard Cauchy estimates to obtain such bounds and give a general recipe for obtaining rigorous proofs of scattering amplitudes. We close with two examples, the LandauZener model and the RosenZener model. While the LandauZener model displays, in a sense to be made precise, a generic transition point, the RosenZener model is of a nongeneric type, which is not covered by existing rigorous results.
2 Exponentially Small Transitions From now on we study the Schr¨odinger equation (2) with the Hamiltonian Z(t) X(t) cos θph (t) sin θph (t) . (8) = ρ(t) Hph (t) = X(t) −Z(t) sin θph (t) −cos θph (t) Thus Hph (t) is a traceless
realsymmetric 2 × 2matrix, and the eigenvalues of Hph (t) are ±ρ(t) = ± X(t)2 + Z(t)2 . We assume that the gap between them does not close, i.e. that 2ρ(t) ≥ g > 0 for all t ∈ R. As to be detailed below, we assume that X and Z are realvalued on the real axis and analytic on a suitable domain containing the real axis. Moreover, in order to be able to consider the scattering limit it is assumed that Hph (t) approaches limits H± suﬃciently fast as t → ±∞. Before proceeding we simplify (8) by switching to the natural time scale τ (t) = 2
t
ds ρ(s) .
(9)
Since ρ(t) is assumed to be strictly positive, the map t → τ is a bijection of R. In the natural time scale the Schr¨ odinger equation (2) becomes iε∂τ − Hn (τ ) φ(τ ) = 0 (10)
LandauZener Formulae from Adiabatic Transition Histories
with Hamiltonian 1 Hn (τ ) = 2
cos θn (τ ) sin θn (τ ) sin θn (τ ) −cos θn (τ )
23
,
(11)
where θn (τ ) = θph (t(τ )). As a consequence we now deal with a Hamiltonian with constant eigenvalues equal to ± 12 , which is completely deﬁned through the single realanalytic function θn . The transformation (5) to the adiabatic basis, i.e. the orthogonal matrix that diagonalizes Hn (τ ), is cos(θn (τ )/2) sin(θn (τ )/2) . (12) U0 (τ ) = sin(θn (τ )/2) − cos(θn (τ )/2) odinger equation in Multiplying (10) from the left with U0 (τ ) yields the Schr¨ the adiabatic representation iε∂τ − Hεa (τ ) φa (τ ) = 0 , (13) where
iε θ (τ ) 2 n Hεa (τ ) = 1 iε − θn (τ ) − 2 2 1 2
and
φa (τ ) = U0 (τ )φ(τ ) =
φ+ (τ ) φ− (τ )
.
(14) θn is called the adiabatic coupling function. The exponentially small scattering amplitude in Fig. 1a) can be guessed iτ by a heuristic calculation. We solve (13) for φ− (τ ) using φ+ (τ ) = e− 2ε +O(ε), which holds according to the adiabatic theorem [8], and variation of constants, i.e. iσ i iτ τ iε − 2ε 2ε dσ e φ− (τ ) = e − θn (σ) φ+ (σ) ε 2 −∞ τ iσ 1 iτ = e 2ε dσ θn (σ) e− ε + O(ε) . (15) 2 −∞ Integration by parts yields φ− (τ ) =
iε iε iτ θn (τ ) − e 2ε 2 2
τ
−∞
dσ θn (σ) e− ε + O(ε) . iσ
(16)
The ﬁrst term in this expression is of order ε and not smaller. This strongly suggests that the O(ε) error estimate in the adiabatic theorem is optimal, which we have seen to be indeed the case. However, no conclusion can be inferred from (16) for the scattering regime τ → ∞ since θn vanishes there. The key to the heuristic treatment of the scattering amplitude is to calculate the integral in (15) not by integration by parts but by contour integration
24
V. Betz and S. Teufel
in the complex plane. For the sake of a simple argument let us assume here that θn is a meromorphic function. Let τc be the location of the pole in the lower complex half plane closest to the real line and γ its residue, then from (15) and contour integration around the poles in the lower half plane we read oﬀ 2Imτc (17) lim φ− (τ )2 = π 2 γ 2 e− ε + O(ε2 ) . τ →∞
Strictly speaking (17) tells us nothing new: while the explicit term is exponentially small in ε, the error term is of order ε2 and thus the statement is not better than what we know from the adiabatic theorem already. Nevertheless it turns out that the exponential factor appearing here actually yields the correct asymptotic behavior of the transition probability. Our heuristic argument also correctly attributes the dominant part of the transition to the pole of θn closest to the real axis. The prefactor in (17), however, is wrong, the correct answer being πγ 2Imτc e− ε (1 + O(εα )) , lim φ− (τ )2 = 4 sin2 (18) τ →∞ 2 for some α > 0. Expression (18) is a generalization of the LandauZener formula and was ﬁrst rigorously derived in [7]. The problem to solve when trying to rigorously treat exponentially small transitions and to arrive at the correct result (18) is to control the solution of (2) up to errors that are not only exponentially small in ε, but smaller than the leading order transition probability. As a consequence a naive perturbation calculation in the adiabatic basis will not do the job. The classical approach [7] to cope with this is to solve (2) not on the real axis but along a certain path in the complex plane, where the lower component of the solution is always exponentially small. The comparison with the solution on the real line is made only in the scattering limit at τ = ±∞. The trick is to choose the path in such a way that it passes through the relevant singularity of θn in the complex plane. In a neighborhood of the singularity one can solve (13) explicitly and thereby determine the leading order contribution to the transition probability. Moreover, away from the transition point the path must be chosen such that the lower component φ− (τ ) remains smaller than the exponentially small leading order contribution from the transition point for all τ along this path. There are two drawbacks of this approach: the technical one is that there are examples (see the RosenZener model below), where such paths do not exist. On the conceptual side, this approach yields only the scattering amplitudes, but gives no information whatsoever about the solution for ﬁnite times. Our approach is motivated by the ﬁndings of Berry [1] and of Berry and Lim [2]. Instead of solving (13) along a path in the complex plane we solve the problem along the real axis but in a superadiabatic basis instead of the adiabatic one, i.e. we solve (6) with the Hamiltonian (7) and the optimal n(ε). While the oﬀdiagonal elements of the Hamiltonian in the adiabatic basis are
LandauZener Formulae from Adiabatic Transition Histories
25
only of order ε, cf. (14), the oﬀdiagonal elements of the Hamiltonian in the optimal superadiabatic basis are exponentially small, i.e. of order e−c/ε . In order to control the exponentially small transitions, we will give precise exponential bounds on the coupling εnε +1 cnε ε (τ ) away from the transition regions and explicitly determine the asymptotic form of cε (τ ) within each transition region. Since the superadiabatic bases agree asymptotically for t → ±∞ with the adiabatic basis, the scattering amplitudes agree in all these bases. In the optimal superadiabatic basis the correct transition probabilities (18) now follow from a ﬁrst order perturbation calculation analogous to the one leading to (17) in the adiabatic basis. However, in addition to the scattering amplitudes we obtain approximate solutions for all times, i.e. “histories of adiabatic quantum transitions” [1]. As is illustrated in Fig. 1b, these are monotonous and asymptotically take the form of an error function.
3 The Hamiltonian in the SuperAdiabatic Representation In [4] we formulate our results for the system (10) and (11). However, we have to keep in mind that (10) and (11) arise from the physical problem (2) and (8) through the transformation to the natural time scale (9). Therefore, to be physically relevant, the assumptions must be satisﬁed by all θn arising from generic Hamiltonians of the form (8). As observed in [2], see also [4], for such θn the adiabatic coupling function θn is real analytic and at its complex singularities z0 closest to the real axis it has the form −iγ + (z − z0 )−αj hj (z − z0 ) , z − z0 j=1 N
θn (z − z0 ) =
(19)
where Imz0  > 0, γ ∈ R, αj < 1 and hj is analytic in a neighborhood of 0 for j = 1, . . . , N . The following norms on the real line capture exactly the behavior (19) of the complex singularities of θn . They are at the heart of the analysis in [4]. Deﬁnition 1. Let τc > 0, α > 0 and I ⊂ R be an interval. For f ∈ C ∞ (I) we deﬁne τc α+k ≤∞ (20) f (I,α,τc ) := sup sup ∂ k f (t) Γ (α + k) t∈I k≥0 and
Fα,τc (I) = f ∈ C ∞ (I) : f (I,α,τc ) < ∞ .
The connection of these norms with (19) relies on the Darboux Theorem for power series and is described in [4]. Let us just note here that θn as given in (19) is an element of F1,τc ({τr }) for τc = Im(z0 ) and τr = Re(z0 ), while
26
V. Betz and S. Teufel
the second term of (19) is in Fβ,τc ({τr }) with β = maxj αj . In order to control the transitions histories, the real line will be segmented into intervals I, which are either considered to be a small neighborhood of a transition point or to contain no transition point. Assumption 1 below thus applies to intervals without transition point and Assumption 2 generically holds near a transition point. In the rest of this section we drop the subscript n for the natural time scale in order not to overburden our notation. Assumption 1: For a compact interval I and δ ≥ 0 let θ (τ ) ∈ F1,τc +δ (I). Assumption 2: For γ, τr , τc ∈ R let 1 1 − θ0 (t) = i γ τ − τr + iτc τ − τr − iτc be the sum of two complex conjugate ﬁrst order poles located at τr ± iτc with residues ∓iγ. On a compact interval I ⊂ [τr − τc , τr + τc ] with τr ∈ I we assume that θ (τ ) = θ0 (τ ) + θr (τ )
with
θr (τ ) ∈ Fα,τc (I)
(21)
for some γ, τc , τr ∈ R, 0 < α < 1. It turns out that under Assumption 2 the optimal superadiabatic basis is given as the nth ε superadiabatic basis where 0 ≤ σε < 2 is such that nε =
τc − 1 + σε ε
is an even integer.
(22)
The two main points of the following theorem are: outside the transition regions, the oﬀdiagonal elements of the Hamiltonian in the optimal superadiabatic basis are bounded by (24), while within each transition region they are asymptotically equal to g(ε, τ ) as given in (ii). Theorem 1. (i) Let H satisfy Assumption 1. Then there exists ε0 > 0 such that for all ε ∈ (0, ε0 ] and all τ ∈ I the elements of the superadiabatic Hamiltonian (7) and the unitary Uεnε (τ ) with nε as in (22) satisfy n ρε ε (τ ) − 1 ≤ ε2 φ1 θ (23) (I,1,τc +δ) 2 n +1 n √ τ τ +δ ε ε cε ε (τ ) ≤ ε e− εc (1+ln cτc ) φ1 θ (24) (I,1,τc +δ) and
Uεnε (τ ) − U0 (τ ) ≤ εφ1 θ (I,1,τc +δ) .
(25)
Here φ1 : R+ → R+ is a locally bounded function with φ1 (x) = O(x) as x → 0 which is independent of I and δ.
LandauZener Formulae from Adiabatic Transition Histories
27
(ii) Let H satisfy Assumption 2 and deﬁne g(ε, τ ) = 2i
2ε πτc
sin
πγ 2
e− ε e− τc
(τ −τr )2 2ετc
cos
τ −τr ε
−
(τ −τr )3 3ετc 2
+
σε τ τc
.
There exists ε0 > 0 and a constant C < ∞ such that for all ε ∈ (0, ε0 ] and all τ ∈ I n +1 n ε ε cε ε (τ ) − g(ε, τ ) ≤ Cε 32 −α e− τεc . (26) Furthermore, the assertions of part (i) hold with δ = 0. Remark 1. In [4] we show in addition that the error bounds in Theorem 1 are locally uniform in the parameters α, γ and τc . This generality is not needed here and thus omitted from the statement. In order to pass to the scattering limit it is now necessary to show that the errors in part (i) of Theorem 1, i.e. in the regions away from the transition points, are integrable.
4 The Scattering Regime We will treat the scattering regime by using ﬁrst order perturbation theory on the equation in the optimal superadiabatic basis. As in (15), variation of constants yields τ σ i i i τ φn−ε (τ ) = e ε −∞ dσ ρ(σ) dσ e− ε −∞ dν ρ(ν) c(nε , σ) φn+ε (σ) , (27) ε −∞ where we put c(nε , τ ) = εnε +1 cnε ε (τ ). We now replace ρnε ε (τ ) and c(nε , τ ) in (27) by the explicit asymptotic values given in Theorem 1, and use the iτ adiabatic approximation φn+ε (τ ) = e− 2ε + O(ε). To this end we assume that θn has k poles of the form (19) at distance τc from the real axis and none closer to the real axis. Let gj (ε, τ ) be the associated coupling functions of Theorem 1 for j = 1, . . . , k and f1 (τ ) = ρnε ε (τ ) − Then i
eε
k iτ 1 , f2 (τ ) = c(nε , τ )− gj (ε, τ ), f3 (τ ) = φn+ε (τ )−e− 2ε . 2 j=1 τ −∞
dσ ρ(σ)
τ iτ = e 2ε 1 + O ε dσ f1 (σ) −∞ =:F1 (τ )
and
28
V. Betz and S. Teufel
φn−ε (τ ) =
τ iσ i iτ e 2ε 1 + O(εF1 (τ )) dσ e− 2ε 1 + O(εF1 (σ)) ε −∞ k gj (ε, σ) − f2 (σ) φn+ε (σ) × j=1
=
k
τ iσ i iτ e 2ε dσ e− ε gj (ε, σ) ε −∞ j=1 τ τ + O (F1 ∞ + ε−1 f3 ∞ ) dσc(nε , σ) + ε−1 dσf2 (σ) . −∞
−∞
Assuming integrability of the error terms in (23) and (24), the following lemma can be established by straightforward computations. Proposition 1. Let θn (τ ) be as above and let τ → θn ({τ },1,τc +δ) be integrable outside of some bounded interval and for some δ > 0. Then τ k τc iσ 1 i iτ dσ e− ε gj (ε, σ) + O(ε 2 −α e− ε ) . φn−ε (τ ) = e 2ε ε −∞ j=1 τc
Note that the leading term in Proposition 1 is of order e− ε . Thus for α ≥ 12 the estimate is too weak. However, a more careful analysis of the error near 1 the transition points allows one to replace ε 2 −α by ε1−α in Proposition 1, see [5], and thus to obtain a nontrivial estimate for all α < 1. Since the functions gj (ε, τ ) are explicitly given in Theorem 1, the leading order expression for φn−ε (τ ) can be computed explicitly as well. A simple computation, c.f. [3], yields for k = 1 that τ τc iσ 1 i iτ dσ e− ε g(ε, σ) + O(ε 2 −α e− ε ) φn−ε (τ ) = e 2ε ε −∞ πγ τc iτ τ τc 1 = sin e− ε e 2ε erf √ + 1 + O(ε 2 −α e− ε ) . 2 2ετc For more than one transition point the same computation reveals interference eﬀects, c.f. [5]. In the limit τ → ∞ we recover the LandauZener formula for the transition probability: πγ 2τc 2τc 1 e− ε + O(ε 2 −α e− ε ) . (28) φn−ε (∞)2 = 4 sin2 2 Proposition 1 yields the transition histories as well as the transition probabilities in the scattering limit for a large class of Hamiltonians under the assumption that θn ({τ },1,τc +δ) is integrable at inﬁnity for some δ > 0. At ﬁrst sight it might seems hard to establish integrability of this norm, since it involves derivatives of θn of all orders. However, the following proposition shows that θn ({τ },1,τc +δ) can be bounded by the supremum of the function θn in a ball around τ with radius slightly larger than τc + δ.
LandauZener Formulae from Adiabatic Transition Histories
29
Proposition 2. Let α > 0 and r > 0. Assume for some δ > 0 that f is analytic on Br+δ = {z ∈ C : z ≤ r + δ} . Then f ({0},α,r) ≤
rα sup f (z) . e ln ((r + δ)/r) z∈Br+δ
Proof. Put M = supz∈Br+δ f (z). By the Cauchy formula, f (z) ∂tk f (0) = k! dz k+1 ≤ 2π k! M (r + δ)−k . z z=r+δ Therefore ∂tk f (0)
Γ (1 + k) rα+k ≤ M rα Γ (α + k) Γ (α + k)
r r+δ
k .
(29)
The kdependent part of the right hand side above is obviously maximal for α = 0, and then is equal to φ(k) := k(r/(r + δ))k . φ(k) is maximal at k = 1/ ln((r + δ)/r) with value 1/(e ln((r + δ)/r), and the claim follows by taking the supremum over k in (29). Hence, integrability of θn ({τ },1,τc +δ) follows if we can establish suﬃcient decay of supz−τ  0. Thus ρ2 (t) = a2 + t2 , and the transformation to the natural time scale reads t
τ (t) = 2 a2 + s2 ds . (31) 0
From (30) one reads oﬀ that complex zeros of ρ give rise to complex singularities of θn . In the LandauZener model, ρ has two zeros at tc = ±ia. Thus (31) 2 yields τc = a2π , and expansion of θn (τ ) around τc shows γ = 13 and α = 13 , cf. [2,4]. We now apply Proposition 1 in order to pass to the scattering limit. According to Proposition 2 we need to control the decay of θn  in a ﬁnite strip around the real axis for large τ . From (31) one reads oﬀ that
30
V. Betz and S. Teufel
τ (t) ≤ 2 · 2t a2 + t2 ≤ 4(a2 + t2 ) , and thus t2  ≥ τ /4 − a2 . From (30) and the estimates above we infer for τ  suﬃciently large that θn (τ ) =
a a a ≤ ≤ . 2a2 + t(τ )2 3/2 2(t(τ )2 − a2 )3/2 2(τ /4 − 2a2 )3/2
Consequently, Proposition 2 yields for every r, δ > 0 and τ ∈ R suﬃciently large that θn ({τ },1,r) ≤
a r . e ln((r + δ)/r) ((τ  − r − δ)/4 − 2a2 )3/2
Thus τ → θn ({τ },1,r) is integrable at inﬁnity for any r > 0, and in particular for r = τc + 2δ. According to (28), we have therefore shown the classical LandauZener formula φn−ε (∞)2 = e−
a2 π ε
+ O(ε 6 e− 1
a2 π ε
).
For the LandauZener model, the transition probabilities can also be proved by the method of [7]. There, the antiStokes lines, i.e. the level lines Im(τ (t)) = Im(τ (tc )), play an essential role. In particular, the method requires that an antiStokes line emanating from the critical point tc of τ stays in a strip of ﬁnite width around the real axis as Re(t) → ±∞. As shown in Fig. 2, this is the case in the present example. The previous example also shows a useful general strategy: One can use (30) in order to ﬁnd upper bounds on τ (t), which in turn yield lower bounds on the inverse function t(τ ). These can then be used in (30) to estimate the decay at inﬁnity of θn in a strip around the real axis. It is clear that this strategy also works in cases where the Hamiltonian is not given in closed form. Of course, things are much easier when we know θn explicitly. This is the case in the following example. Example 2 (RosenZener model). In this model X(t) = 2(t21+1) and Z(t) = t 2(t2 +1) . Therefore τ (t) = arsinh(t), τc = Im(arsinh(i)) = π, and (30) yields θn (τ ) = 1/ cosh(τ ) in the natural time scale. It is immediate that θn (τ ) ≤ c exp(−τ ) for large τ  in each ﬁxed strip around the real axis, and that γ = 1. Since θn is meromorphic, 0 < α can be chosen arbitrarily small. In summary, Propositions 1 and 2 yield φn−ε (∞)2 = 4e− ε + O(ε 2 −α e− ε ) . π
1
π
Although the RosenZener example is very easy in our picture, it is not clear how to prove it using the methods of [7]. The reason is that there are no antiStokes lines emanating from the singularity of θn and staying in a bounded strip around the real axis as Re(t) → ±∞. In fact, the only relevant antiStokes line remains on the imaginary axis, cf. Fig. 2.
LandauZener Formulae from Adiabatic Transition Histories
31
1.5
(a)
1
0.5
2
1.5
1
0.5
0.5
1
1.5
2
0.5
1
1.5
2
1.5
(b)
1
0.5
2
1.5
1
0.5
Fig. 2. This ﬁgure shows the Stokes and antiStokes lines for τ (t) in the LandauZener model (Fig. 2a) and the RosenZener model (Fig. 2b). Level lines of Reτ (t) are grey, while the lines of Imτ (t) are black. The fat lines correspond to the Stokes and antiStokes lines emanating from the critical point tc of τ (t) in the upper complex halfplane. In both examples, tc = i. While in the LandauZener model, two antiStokes lines remain in a ﬁnite strip around the real axis, the antiStokes line of the RosenZener model remains on the imaginary axis
References 1. M.V. Berry. Histories of adiabatic quantum transitions, Proc. R. Soc. Lond. A 429, 61–72 (1990). 2. M.V. Berry and R. Lim. Universal transition prefactors derived by superadiabatic renormalization, J. Phys. A 26, 4737–4747 (1993). 3. V. Betz and S. Teufel. Precise coupling terms in adiabatic quantum evolution, to appear in Annales Henri Poincar´e (2005). 4. V. Betz and S. Teufel. Precise coupling terms in adiabatic quantum evolution: the generic case, to appear in Commun. Math. Phys. (2005). (2004). 5. V. Betz and S. Teufel. Exponentially small transitions in adiabatic quantum evolution, in preparation. 6. G. Hagedorn and A. Joye. Time development of exponentially small nonadiabatic transitions, Commun. Math. Phys. 250, 393–413 (2004). 7. A. Joye. Nontrivial prefactors in adiabatic transition probabilities induced by high order complex degeneracies, J. Phys. A 26, 6517–6540 (1993).
32
V. Betz and S. Teufel
8. T. Kato. On the adiabatic theorem of quantum mechanics, Phys. Soc. Jap. 5, 435–439 (1950). 9. S. Teufel. Adiabatic perturbation theory in quantum dynamics, Springer Lecture Notes in Mathematics 1821, 2003. 10. C. Zener. Nonadiabatic crossing of energy levels, Proc. Roy. Soc. London 137 696–702, (1932).
Scattering Theory of Dynamic Electrical Transport M. B¨ uttiker1 and M. Moskalets2 1
2
D´epartement de Physique Th´eorique, Universit´e de Gen`eve, CH1211 Gen`eve 4, Switzerland [emailprotected] Department of Metal and Semiconductor Physics, National Technical University “Kharkiv Polytechnic Institute”, 61002 Kharkiv, Ukraine [emailprotected]
Abstract. We have developed a scattering matrix approach to coherent transport through an adiabatically driven conductor based on photonassisted processes. To describe the energy exchange with the pumping ﬁelds we expand the Floquet scattering matrix up to linear order in driving frequency. Key words: Emissivity; Instantaneous currents; Internal response; Photon–assisted transport; Quantum pump eﬀect; Scattering matrix
1 From an Internal Response to a Quantum Pump Eﬀect The possibility to vary several parameters at the same frequency but with diﬀerent phases [1] of a coherent (mesoscopic) system opens up new prospects for the investigation of dynamical quantum transport. The adiabatic variation of parameters is of particular interest since at small frequencies the conductor stays close to an equilibrium state: the opening of inelastic conduction channels is avoided and quantum mechanical phase coherence is preserved to the fullest extend possible. The relevant physics has a simple and transparent explanation within the scattering matrix approach. The variation of parameters leads to a dynamic scattering geometry. Quite generally we consider changes in the scattering geometry as an internal response [2] in contrast to the external response generated by voltages applied to the contacts of the conductor, see Fig. 1. In general a linear response consists both of a response to an external potential oscillations, and response to internal potentials. The internal response can be expressed with the help of the emissivity ν(α, r). The emissivity ν(α, r) is the portion of the density of states at r of carriers that will exit the conductor through contact α. The emissivity relates the amplitude Iα (ω) of the current in lead α to the amplitude U (r, ω) of a small and slowly oscillating internal potential. At zero temperature we have [2] 2 Iα (ω) = ie ω d3 rν(α, r)U (r, ω) . (1) M. B¨ uttiker and M. Moskalets: Scattering Theory of Dynamic Electrical Transport, Lect. Notes Phys. 690, 33–44 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
34
M. B¨ uttiker and M. Moskalets
external response
V(ω) e iωt
internal response
I(ω) ~ U(ω)
I(ω) ~ V(ω)
Ue iωt
Fig. 1. External and internal response: An ac current with amplitude I(ω), say, in the left lead can arise as a response either to an oscillating potential V (t) = to an V (ω)e−iωt + V (−ω)eiωt at one of the external reservoirs, or as a response oscillating potential proﬁle δU (r, t) = δu(r) U (ω)e−iωt + U (−ω)eiωt inside the mesoscopic sample
Here e is an electron charge and i2 = −1. The integral in (1) runs over the region in which the potential deviates from its equilibrium value (typically the volume occupied by the scatterer). The emissivity is expressed in terms of the scattering matrix S (of the stationary scatterer) and its functional to the internal derivative with respect potential variation δU (r, t) = δu(r) U (ω)e−iωt + U (−ω)eiωt : ν(α, r) = −
Nr ∗ δSαβ δSαβ 1 ∗ − Sαβ . Sαβ 4πi δeu(r) δeu(r)
(2)
β=1
The scattering matrix is evaluated at the Fermi energy E = µ. The summation runs over all the leads (for simplicity here assumed to be single channel) connecting the sample to external Nr = 1, 2, 3... reservoirs. Applying the inverse Fourier transformation to (1) gives the current Iα (t) ﬂowing in response to a timedependent internal potential U (r, t). Equation (1) can easily be generalized to ﬁnd the response to an arbitrary ﬁeld or parametric variation of the scattering geometry [3]. To arrive at a general expression we remember that the scattering matrix depends on the internal potential U . That in turn makes S timedependent, S(t) ≡ S[U (t)]. Thus alternatively we can express (1) in the form Iα (t) =
Nr ie ∗ dSαβ . Sαβ 2π dt
(3)
β=1
Originally in [2] the potential U (r, t) is the selfconsistent Coulomb potential. However the form of (3) tells us that the current Iα (t) can arise in response to a slow variation of any quantity (parameter) which aﬀects the scattering properties of a mesoscopic sample. For instance the current generated by a slowly varying vector potential [4] permits to derive the Landauer dcconductance from (3). But (3) is not limited to the linear response regime.
Scattering Theory of Dynamic Electrical Transport
35
The current Iα (t) given by (3) is a nonlinear functional of the scattering matrix. Therefore, the mesoscopic system can exhibit an internal rectiﬁcation eﬀect, i.e., an oscillating internal potential (or any appropriate oscillating parameter) can result in a dc current Idc . Since the elements of the scattering matrix are quantum mechanical amplitudes the dccurrent is the result of a quantum rectiﬁcation process. This is a quantum pump eﬀect [3–9]. An approach to quantum pumping based on (1) and (3) was put forth by Brouwer [3]. Since the time derivative enters (3), quantum rectiﬁcation will work only under special conditions [3]. Let one or several parameters aﬀecting the scattering properties of a mesoscopic sample change adiabatically and periodically in time. Then the scattering matrix changes periodically as well. Consider the point representing the scattering matrix in the space of all scattering matrices. During the completion of one time period this point will move along a closed line L. Then the dc current Idc,α , which is the current Iα (t) averaged over the time period T = 2π/ω, can be represented as a contour integral in the above mentioned abstract space [9]: ieω dSS† αα . Idc,α = (4) 2 4π L The dc current Idc,α , is nonzero if and only if the line L encloses a nonvanishing area F. The easy way to see this is to consider a two parameter space with parameters being S and S† . Since S and S† depend on the same set of parameters, S = S({Xj}), S† = S† ({Xj}), j = 1, 2, . . . , Np , (which includes, for instance, the shape of a sample, the internal potential, the magnetic ﬁeld, the temperature, the pressure, the Fermi energy, etc.) then to get the cycle with F = 0 it is necessary to have at least two parameters X1 (t) = X1 cos(ωt + ϕ1 ) and X2 (t) = X2 cos(ωt + ϕ2 ) varying with the same frequency ω and with a phase lag ∆ϕ ≡ ϕ1 − ϕ2 = 0. In particular, if the oscillating amplitudes are small, i.e., if the scattering matrix changes only a little across F, then the pumped current is proportional to the square of the cycle area [3]: Idc,α =
∗ Nr ∂Sαβ ∂Sαβ eω sin(∆ϕ)X1 X2 . 2π ∂X1 ∂X2 X1 =0,X2 =0
(5)
β=1
The very simple and compact expression (4) allows to ﬁnd the pumped current for a wide range of situations. Illustrative examples can be found in [10]. In the following we will now discuss the pumping process from the point of view of photonassisted transport through a mesoscopic system. We will show how the interlay between photon–assisted transport and quantum mechanical interference results in a quantum pump eﬀect [11, 12].
36
M. B¨ uttiker and M. Moskalets
2 Quantum Coherent Pumping: A Simple Picture By nature, the quantum pump eﬀect is a rectiﬁcation eﬀect. A single parameter variation only leads to an accurrent. In a two parameter variation the modulation of the scatterer due to the second parameter rectiﬁes the accurrents generated by the ﬁrst parameter. Rectiﬁcation is only achieved if the driven system can scatter electrons in an asymmetric way. Here the asymmetry means that the probability Tαβ for an electron to pass through the sample, say, from the lead β to the lead α and the probability Tβα to transit the scatterer in the reverse direction diﬀer, Tαβ = Tβα . Then the ﬂux of electrons entering the scatterer through lead α and the ﬂux of electrons scattered and leaving the system through the same lead α diﬀer from each other, resulting in a net electron ﬂow in lead α. The resulting current can be viewed as a result of an asymmetrical redistribution of incoming ﬂows between the outgoing leads. To clarify the physical mechanism which can lead to asymmetric scattering we now emphasize the essential diﬀerence between the driven scatterer and a stationary one. The key diﬀerence is the possibility of photon–assisted transport. In the stationary case if an electron with energy E enters the phase coherent system then it leaves the system with the same energy E. In contrast, in the driven case while traversing the system an electron can absorb (or emit) energy quanta nω and thus it can leave the system with an energy En = E + nω. It is important that the electron changes its energy interacting with a system which is modulated deterministically. As a consequence the inelastic processes is coherent. If there are several possibilities for transmission through the system absorbing or emitting the same energy (say, nω) the corresponding quantum–mechanical amplitudes will interfere. Such an interference of photon–assisted amplitudes can lead to directional asymmetry of electron propagation through a driven mesoscopic sample. To illustrate this process we consider a simple but generic example. It is a system consisting of two regions with oscillating potentials V1 (t) = 2V cos(ωt + ϕ1 ) and V2 (t) = 2V cos(ωt + ϕ2 ) separated by the distance L. For the sake of simplicity we assume that both potentials oscillate with the same small amplitude 2V . Consider an electron with energy E incident on the system. In leading order in the oscillating amplitudes only absorption/emission of a single energy quantum ω needs to be taken into account. So, there are only three scenarios to traverse the system. In the ﬁrst case, an electron does not change its energy, the outgoing energy is E (out) = E. In the second case, it absorbs one energy quantum, E (out) = E + ω. In the third case, it emits an energy ω, E (out) = E − ω. Since all these processes correspond to diﬀerent ﬁnal states (which diﬀer in energy E (out) from each other) then the full probability T to pass through the system is a sum of three contributions: T = T (0) (E; E) + T (+) (E + ω; E) + T (−) (E − ω; E) .
(6)
Scattering Theory of Dynamic Electrical Transport
37
V1(t) V2(t) E
L
E ± hω
V1(t) E
= V2(t)
E ± hω
+
E
E ± hω
Fig. 2. For the process in which a carrier with energy E traverses two oscillating potentials V1 (t) and V2 (t), a distance L apart, and absorbs/emits an energy quantum ω there are two interfering alternatives. The modulation quantum can be either absorbed/emitted at the ﬁrst barrier or at the second one
Here the ﬁrst argument is an outgoing electron energy while the second argument is an incoming energy. The probability T (0) , like the tranmission probability of stationary scatterer, is insensitive to the propagation direction. In contrast T (+) and T (−) are directionally sensitive. Therefore we concentrate on the last two. Let us consider T (+) . There are two possibilities to pass through the system and to absorb an energy ω, see, Fig. 2. The ﬁrst possibility is to absorb an energy due to the oscillation of V1 (t). The second possibility is to absorb an energy interacting with V2 (t). In these two processes an electron has the same initial state and the same ﬁnal state. Therefore, we can not distinguish between these two possibilities and according to quantum mechanics to calculate the corresponding probability we ﬁrst have to add up the corresponding amplitudes and only then take the square. Let us denote the amplitude corresponding to the propagation through the system with absorbing ω at Vj as A(j,+) , j = 1, 2. Then the corresponding full probability is: 2 T (+) = A(1,+) + A(2,+) . (7) (2,+) ) is a product of two terms, the amEach amplitude (either A(1,+) or A√ (free) ikL (E) = e (here k = 2mE/ is an electron wave number) of plitude A (+) free propagation in between the potentials and the amplitude Aj describing the absorption of an energy quantum ω at the potential Vj . The amplitude (+) Aj is proportional to the corresponding Fourier coeﬃcient of Vj (t). The pro
portionality constant is denoted by α. Therefore we have Aj = αVj e−iϕj . The probability for an electron going from the left to the right is denoted (+) (+) . The probability corresponding to the reverse direction – by T← . by T→ Our aim is to show that (+) (+) = T← . (8) T→ (+)
38
M. B¨ uttiker and M. Moskalets
(+) First we consider T→ . Scattering from the left to the right an electron ﬁrst meets the potential V1 (t) and only then the potential V2 (t). Therefore if an electron absorbs the energy ω at the ﬁrst potential it traverses the remaining part of the system with enhanced energy E+1 = E + ω. The (+) corresponding amplitude is A(1,+) = A1 A(free) (E+1 ). While if an electron → absorbed ω at the second potential then it goes through the system with the initial energy E. The quantum mechanical amplitude corresponding to such (+) = A(free) (E)A2 . If the energy quantum is much a process reads: A(2,+) → smaller then the electron energy, ω E, then we can expand the phase factor corresponding to free propagation with enhanced energy E+1 to ﬁrst order in the driving frequency: k+1 L ≈ k + ωv L, here v = k/m is an electron velocity. Thus we have:
A(1,+) ≈ αV e−iϕ1 ei(k+ v )L , → ω
A(2,+) →
=e
ikL
αV e
−iϕ2
(9)
.
Substituting these amplitudes into (7) we obtain the probability to pass through the system from the left to the right with the absorption of an energy quantum ω: ! ωL (+) = 2α2 V 2 1 + cos ϕ1 − ϕ2 − T→ . (10) v (+) . Going from the right to the left an elecNow we consider the probability T← tron ﬁrst meets the potential V2 (t) and then the potential V1 (t). Therefore, the corresponding amplitudes are:
= eikL αV e−iϕ1 , A(1,+) ← (11) ≈ αV e−iϕ2 ei(k+ v )L A(2,+) ← ω
Using (7) and (11) we ﬁnd: (+) T←
2
= 2α V
2
! ωL 1 + cos ϕ1 − ϕ2 + . v
(12)
Comparing (10) and (12) we see that the transmission probability depends on the direction of electron propagation as we announced in (8). Let us characterize the asymmetry in transmission probability by the (+) (+) − T← : diﬀerence ∆T (+) = T→ ωL ∆T (+) = 4α2 V 2 sin(∆ϕ) sin . (13) v In our simple case the emission leads to the same asymmetry in the photon– assisted transmission probability: ∆T (−) = ∆T (+) . Therefore if there are the
Scattering Theory of Dynamic Electrical Transport
39
same electron ﬂows I0 coming from the left and from the right, then a net current I = I0 ∆T (−) + ∆T (+) is generated. We assume a positive current to be directed to the right. We see that the induced current I depends separately on ∆ϕ = ϕ1 − ϕ2 and on the factor ωL/v. This is an additional dynamical phase due to absorption of an energy quantum ω. Such a separation of phase factors can be interpreted in the following way. The presence of the phase lag ∆ϕ (by modulo 2π) between the oscillating potentials V1 (t) and V2 (t) breaks the time reversal invariance of a problem and hence potentially permits the existence of a steady particle ﬂow. The second term emphasizes a spatial asymmetry of a model consisting of two inequivalent oscillating regions separated by a distance L, and tells us that the interference of photon–assisted amplitudes is the mechanism inducing electron ﬂow. Neither a phase lag nor a photonassisted process can separately lead to a dc current. The simple model presented here reproduces several generic properties of a periodically driven quantum system. First, a driven spatially asymmetric system can pump a current between external electron reservoirs to which this system is coupled. Second, the pumped current is periodic in the phase lag between the driving parameters. Third, at small driving frequency, ω → 0, the current generated is linear in ω. In contrast the oscillatory dependence on ω of (13) is a special property of the simple resonant tunneling structure [13].
3 Beyond the Frozen Scatterer Approximation: Instantaneous Currents In experiments the external electrical circuit to which the pump is connected is important [14, 15]. Consequently, it is of interest to understand the workings of a quantum pump connected to contacts which are not at equilibrium but support dcvoltages and acpotentials. We now present a number of results of a rigorous calculation of dynamically generated currents within the scattering matrix approach for spinless noninteracting particles [16]. These results permit the investigation of pumping also in experimentally more realistic nonideal situations. We generalize the approach of [2] to the case of strong periodic driving when many photon processes are of importance. To take them into account we use the Floquet scattering matrix whose elements SF,αβ (En , E) are the
√ quantum mechanical amplitudes (times kn /k, where kn = 2mEn /, En = E+nω) for scattering of an electron with energy E from lead β to lead α with absorption (n > 0) or emission (n < 0) of n energy quanta ω. Like in [2] we deal with low frequencies (adiabatic driving) and calculate the current linear in ω. In this limit we can expand the Floquet scattering matrix in powers of ω. To zeroth order the Floquet submatrices SF (En , E) are merely the matrices of Fourier coeﬃcients Sn of the stationary scattering matrix with time
40
M. B¨ uttiker and M. Moskalets
dependent (pump)parameters, S(t) ≡ S({Xj (t)}). The matrix S(t) is the frozen scattering matrix, in the sense that it describes the time moment t and hence stationary scattering. If all the reservoirs are kept at the same conditions (potential, temperature..) then the knowledge of only the frozen scattering matrix is suﬃcient to calculate the current ﬂowing through the scatterer, see (3). Under more general conditions knowledge of the frozen scattering matrix is not suﬃcient. We stress that the frozen scattering matrix does not describe the scattering of electrons by a timedependent scatterer: only the Floquet scattering matrix does. Since we are interesting in a current linear in ω then, in general, it is necessary to know the Floquet scattering matrix with the same accuracy. Thus we have to go beyond the frozen scattering matrix approximation and to take into account the corrections of order ω. As we illustrated in Sect. 2 such corrections are due to interference between photon–assisted amplitudes. We use the following ansatz: SF (En , E) = Sn (E) +
nω ∂Sn (E) + ωAn (E) + O(ω 2 ) . 2 ∂E
(14)
The matrix A(E, t) [with Fourier coeﬃcients An (E)] introduced here is a key ingredient. As we will see this matrix reﬂects the asymmetry in scattering from one lead to the other and back. The unitarity condition for the Floquet scattering matrix leads to the following equation for the matrix A(E, t): 1 ω S† (E, t)A(E, t) + A† (E, t)S(E, t) = P{S† ; S} , 2
(15)
where P is the Poisson bracket with respect to energy and time † ∂S ∂S ∂S† ∂S − P{S† ; S} = i . ∂t ∂E ∂E ∂t The matrix A can not be expressed in terms of the frozen scattering matrix S(t) and it has to be calculated (like S itself) in each particular case. Nevertheless there are several advantages in using (14). First, the matrix A has a much smaller number of elements than the Floquet scattering matrix. The matrix A depends on only one energy, E, and, therefore, it has Nr × Nr elements like the stationary scattering matrix S. In contrast, the Floquet scattering matrix SF depends on two energies, E and En , and, therefore, it has (2nmax + 1) × Nr × Nr relevant elements. Here nmax is the maximum number of energy quanta ω absorbed/emitted by an electron interacting with the scatterer which we have to take into account to correctly describe the scattering process. For small amplitude driving we have nmax ≈ 1, whereas if the parameters vary with a large amplitude then nmax 1. Second, the Floquet scattering matrix has no deﬁnite symmetry with respect to a magnetic ﬁeld H reversal. In contrast both the frozen scattering
Scattering Theory of Dynamic Electrical Transport
41
matrix S and A have. The analysis of the microreversibility of the equations of motion gives the following symmetry: S(−H) = ST (H) , (16) A(−H) = −AT (H) , where the upper index “T” denotes the transposition. In the absence of a magnetic ﬁeld, H = 0, the matrix A is antisymmetric in lead indices, Aαβ = −Aβα . Next, using the adiabatic expansion, (14), we calculate the full timedependent current Iα (t) ﬂowing in lead α as follows [16]: Iα (t) =
∞
dE
" β
2 [f0,β − f0,α ] Sαβ (E, t)
e h
#
∂ f0,β − e ∂t
dNαβ (E,t) dE
(17)
!
$ + f0,β
dIαβ (E,t) dE
.
Here f0,α is the Fermi distribution function for electrons in reservoir α. We assume a current in a lead to be positive if it is directed from the scatterer to the corresponding reservoir. Equation (17) generalizes (3) to the case with external reservoirs being nonidentical (e.g., having diﬀerent chemical potentials, temperatures, etc.). The three parts in the curly brackets on the RHS of (17) can be interpreted as follows: The ﬁrst part deﬁnes the currents injected from the external reservoirs. It depends on the timedependent conductance 2 Gαβ (t) = eh Sαβ (t)2 of the frozen scatterer and hence it describes a classical rectiﬁcation contribution to the dc current Idc,α . The second part deﬁnes the current generated by the oscillating charge Q(t) of the scatterer:
∞
Q(t) = e
α
β
dEf0,β
dNαβ (E, t) . dE
(18)
Here dNαβ /dE is the global partial density of states for a frozen scatterer: ∗ ∂Sαβ i dNαβ ∗ ∂Sαβ = Sαβ − Sαβ . (19) dE 4π ∂E ∂E Apparently this part gives no contribution to the dc current. The third part describes the currents generated by the oscillating scatterer. The ability to generate these ac currents diﬀerentiates a nonstationary dynamical scatterer from a merely frozen scatterer. The instantaneous spectral currents dIαβ /dE pushed by the oscillating scatterer from lead β to lead α read: e 1 dIαβ ∗ ∗ = Aαβ ] + P{Sαβ ; Sαβ } . (20) 2ωRe[Sαβ dE h 2
42
M. B¨ uttiker and M. Moskalets
The two terms in this equation have diﬀerent symmetry properties with respect to the interchange of lead indices. That is most evident in the absence of a magnetic ﬁeld, H = 0. In this case the ﬁrst term on the RHS of (20) is antisymmetric in lead indices while the second term has the symmetry of the stationary scattering matrix S and is thus symmetric. Therefore, the matrix A is responsible for the directional asymmetry of dynamically generated currents: dIβα dIαβ = . (21) dE dE We remark, that if one calculates the dc current generated in the particular case when all the reservoirs are at same potentials and temperatures (f0,α = f0,β , ∀ α, β) then (17) (after averaging over a pump period) generates Brouwer’s result, (4). In this case the matrix A plays no role. Diﬀerent contributions of this matrix can be combined as in (15). In contrast, the matrix A is important in less symmetrical situations, when the electron ﬂows arriving at the scatterer from diﬀerent leads are diﬀerent. The spectral currents dIαβ /dE are subject to the following conservation law: Nr dIαβ (E, t) =0. (22) dE α=1 Such a property supports the point of view that these currents arise inside the scatterer. They are generated by the nonstationary scatterer without any external current source. The appearance of currents subject to the conservation law (22) can be easily illustrated within the quasi–particle picture [12]. Let all the reservoirs have the same chemical potential µα = µ, α = 1, . . . , Nr . We introduce quasi–particles: the quasi–electrons corresponding to ﬁlled states with energy E > µ and holes corresponding to empty states with energy E < µ. Then at zero temperature there are no incoming quasi–particles. In other words, from each lead the vacuum of quasi–particles is falling upon the scatterer. Interacting with the oscillating scatterer, the system of (real) electrons can gain, say, n energy quanta ω. In the quasi–particle picture this process corresponds to the creation of a quasi–electron–hole pair with energy nω. The pair dissolves and the quasi–particles are scattered separately to the same or diﬀerent leads, see, Fig. 3. If the scattering matrix depends on energy then the quasi–electron and hole are scattered, on average, into diﬀerent leads since they have diﬀerent energies. Suppose the electron leaves the scattering region through lead α and the hole leaves through lead β. Since electrons and holes have opposite charge the current pulses created in the leads α and β have diﬀerent sign. As a result a current pulse arises between the α and β reservoirs. In this picture it is evident that there is no incoming current and the sum of outgoing currents does satisfy the conservation law (22).
Scattering Theory of Dynamic Electrical Transport
43
nhω
Fig. 3. Interacting with an oscillating potential the electron system gains modulation quanta of energy. Absorption of an energy nω leads to the creation of a nonequilibrium quasi–electron–hole pair. The quasi–electron (black circle) and hole (open circle) can leave the scattering region through diﬀerent leads. This leads to current pulses of diﬀerent signs in the corresponding leads
Note, from (15) and (20) it is obvious that a conductor with strictly energy independent scattering matrix does not produce current and thus it does not show a quantum pump eﬀect. The current Iα (t), calculated with (17), satisﬁes the continuity equation: α
Iα (t) +
∂Q(t) =0, ∂t
(23)
and thus conserves charge. " this we use the unitarity of the "To demonstrate frozen scattering matrix, α Sαβ 2 = β Sαβ 2 = 1, and the deﬁnition of the charge Q(t) of the scatterer, see (18). It follows from (22) that the dynamically generated currents dIαβ /dE do not contribute to (23). Therefore they have nothing to do with charging/discharging of a scatterer. The existence of these currents is an intrinsic property of dynamical scatterer. In conclusion, we clariﬁed the role played by photon–assisted processes in adiabatic electron transport through a periodically driven mesoscopic system. The interference between the corresponding photon–assisted amplitudes makes the transmission probability dependent on the electron transmission direction in striking contrast with stationary scattering. To consider properly this eﬀect we introduced an adiabatic (in powers of a driving frequency) expansion of the Floquet scattering matrix and demonstrated that already linear in ω terms exhibit the required asymmetry. The ability to generate currents is only one interesting aspect of quantum pumps. Recent works point to the possibility of dynamical controlled generation of entangled electronhole states [17, 18]. This brings into focus the dynamic quantum state of pumps. This and other properties will likely assure a continuing lively interest in quantum pumping.
44
M. B¨ uttiker and M. Moskalets
References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18.
M. Switkes, C.M. Marcus, K. Campman et al: Science 283, 1905 (1999) M. B¨ uttiker, H. Thomas, A. Prˆetre: Z. Phys. B 94, 133 (1994) P.W. Brouwer: Phys. Rev. B 58, R10135 (1998) D. Cohen: Phys. Rev. B 68, 201303 (2003) D.J. Thouless: Phys. Rev. B 27, 6083 (1983) B. Spivak, F. Zhou, M.T. Beal Monod: Phys. Rev. B 51, 13226 (1995) I.L. Aleiner, A.V. Andreev: Phys. Rev. Lett. 81, 1286 (1998) F. Zhou, B. Spivak, B. Altshuler: Phys. Rev. Lett. 82, 608 (1999) J.E. Avron, A. Elgart, G.M. Graf et al: Phys. Rev. B 62, R10618 (2000) J.E. Avron, A. Elgart, G.M. Graf et al: J. Stat. Phys. 116, 425 (2004) B. Wang, J. Wang, H. Guo: Phys. Rev. B 65, 073306 (2002) M. Moskalets, M. B¨ uttiker: Phys. Rev. B 66, 035306 (2002) M. Moskalets, M. B¨ uttiker: Phys. Rev. B 66, 205320 (2002) M.L. Polianski, P.W. Brouwer: Phys. Rev. B 64, 075304 (2001) M.G. Vavilov, L. DiCarlo, C.M. Marcus: Phys. Rev. B 71, 241309 (2005) M. Moskalets, M. B¨ uttiker: Phys. Rev. B 69, 205316 (2004) P. Samuelsson, M. B¨ uttiker: Phys. Rev. B 71, 245317 (2005) C.W.J. Beenakker, M. Titov, B. Trauzettel: Phys. Rev. Lett. 94, 186804 (2005)
The LandauerB¨ uttiker Formula and Resonant Quantum Transport Horia D. Cornean1 , Arne Jensen2 , and Valeriu Moldoveanu3 1
2
3
Department of Mathematical Sciences, Aalborg University, Fredrik Bajers Vej 7G, 9220 Aalborg, Denmark [emailprotected] Department of Mathematical Sciences, Aalborg University, Fredrik Bajers Vej 7G, 9220 Aalborg, Denmark [emailprotected] National Institute of Materials Physics, P.O.Box MG7, Magurele, Romania [emailprotected]
We give a short presentation of two recent results. The ﬁrst one is a rigorous proof of the LandauerB¨ uttiker formula, and the second one concerns resonant quantum transport. The detailed results are in [2]. In the last section we present the results of some numerical computations on a model system. Concerning the literature, then see the starting point of our work, [6]. In [4] a related, but diﬀerent, problem is studied. See also [5] and the recent work [1].
1 The LandauerB¨ uttiker Formula We start by introducing the notation and the assumptions. The model used here describes a ﬁnite sample coupled to a ﬁnite number of leads. The leads may be ﬁnite or semiinﬁnite. We use a discrete model, i.e. the tightbinding approximation. The sample is modeled by a ﬁnite set Γ ⊂ Z2 . Each lead is modeled by N = {0, 1, . . . , N } ⊆ N. The case N = N (N = +∞) is the semiinﬁnite lead. We assume that we have M ≥ 2 leads. The oneparticle Hilbert space is then H = 2 (Γ ) ⊕ 2 (N ) ⊕ · · · ⊕ 2 (N ) .
(1)
M copies
The Hamiltonian is denoted by H. It is the sum of the following components. For the sample we can take any selfadjoint operator H S on 2 (Γ ). In each lead we take the discrete Laplacian with Dirichlet boundary conditions. The leads are numbered by α ∈ {1, 2, . . . , M }. Thus HL =
M α=1
HαL ,
HαL =
tL (nα nα + 1 + nα nα − 1)
(2)
nα ∈N
H.D. Cornean et al.: The LandauerB¨ uttiker Formula and Resonant Quantum Transport, Lect. Notes Phys. 690, 45–53 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
46
H.D. Cornean et al.
Functions in 2 (N ) are by convention extended to be zero at −1 and N + 1. The parameter tL is the hopping integral. The coupling between the leads and the sample is described by the tunneling Hamiltonian H T = H LS + H SL ,
where H LS = τ
M
0α S α  ,
(3)
α=1
and H SL is the adjoint of H LS . Here 0α denotes the ﬁrst site on lead α, and S α is the contact site on the sample. The parameter τ is the coupling constant. It is arbitrary in this section, but will be taken small in the next section. The total oneparticle Hamiltonian is then H = HS + HL + HT
on H .
(4)
First we consider electronic transport through the system. Initially the leads are ﬁnite, all of length N , with N arbitrary. We work exclusively in the grand canonical ensemble. Thus our system is in contact with a reservoir of energy and particles. We study the linear response of a system of noninteracting Fermions at temperature T and with chemical potential µ. The system is subjected adiabatically to a perturbation, deﬁned as follows. Let χη be a smooth switching function, i.e. 0 ≤ χη (t) ≤ 2, χη (t) = eηt for t ≤ 0, while χη (t) = 1 for t > 1. The timedependent perturbation is then given by M V (N, t) = χη (t) Vα Iα (N ) . α=1
"N
Here Iα (N ) = nα =0 nα nα  is the identity on the αcopy of 2 (N ). This perturbation models the adiabatic application of a constant voltage Vα on lead α, which will generate a charge transfer between the leads via the sample. We are interested in deriving the current response of the system due to the perturbation. In the grand canonical ensemble we need to look at the second quantized operators. We omit the details and state the result. The current at time t = 0 in lead α is given by Iα (0) =
M
gαβ (T, µ, η, N )Vβ + O(V 2 ) .
(5)
β=1
The gαβ (T, µ, η, N ) are the conductance coeﬃcients [3]. It is clear from the above formula that we work in the linear response regime. Below we are going to take the limit N → ∞, followed by the limit η → 0. The limits have to be taken in this order, since the error term is in fact O(V 2 /η 2 ). The next step is to look at the transmittance, which is obtained from scattering theory, applied to the pair of operators (K, H0 ), where H0 = H L (N = +∞ case) and K = H0 + H S + H T . Properly formulated this is done in the two space scattering framework, see [7]. Since the perturbation
The LandauerB¨ uttiker Formula and Resonant Quantum Transport
47
H S + H T is of ﬁnite rank, and since we have explicitly a diagonalization of the operator H0 , the stationary scattering theory gives an explicit formula for the scattering matrix, which is an M × M matrix, depending on the spectral parameter λ = 2tL cos(k) of H0 . The T operator is then given by an M × M matrix tαβ (λ), and the transmittance is given by Tαβ (λ) = tαβ (λ)2 .
(6)
It follows from the explicit formulas that Tαβ (λ) is real analytic on (−2tL , 2tL ), and zero outside this interval. With these preparations we can state the main result. Theorem 1. Consider α = β, T > 0, µ ∈ (−2tL , 2tL ), and η > 0. Assume that the point spectrum of K (corresponding to the N = +∞ case) is disjoint from {−2tL , 2tL }. Then taking ﬁrst the limit N → ∞, and then η → 0, we have % & gα,β (T, µ) = lim lim gα,β (T, µ, η, N ) η→0 N →∞ 2tL ∂fF−D (λ) 1 Tαβ (λ)dλ . (7) =− 2π −2tL ∂λ Here fF−D (λ) = 1/(e(λ−µ)/T + 1) is the FermiDirac function. If we ﬁnally take the limit T → 0, we obtain the Landauer formula gα,β (0+ , µ) =
1 Tαβ (µ) . 2π
(8)
The proof of this main result is quite long and technical. One has to study the two sides of the equality above. The scattering part (the transmittance) is quite straightforward, using the Feshbach formula. The conductance part is a fairly long chain of arguments, as is the proof of the equality statement in the theorem. We refer to [2] for the details.
2 Resonant Transport in a Quantum Dot In the previous section we have allowed the coupling constant τ (see (3)) to be arbitrarily large. The only assumption was that {−2tL , 2tL } was not in the point spectrum of K. We now look at the small coupling case, τ → 0. In this case we will assume that the sample Hamiltonian H S does not have eigenvalues {−2tL , 2tL }. It then follows from a perturbation argument, using the Feshbach formula, that the same is true for K, provided τ is suﬃciently small. Since H S is an operator on the ﬁnite dimensional space 2 (Γ ), is has a purely discrete spectrum. We enumerate the eigenvalues in the interval (−2tL , 2tL ):
48
H.D. Cornean et al.
σ(H S ) ∩ (−2tL , 2tL ) = {E1 , . . . , EJ } . Let β = γ be two diﬀerent leads. The conductance between these two is now denoted by Tβ,γ (λ, τ ), making the dependence on the coupling constant explicit, see (6). Theorem 2. Assume that the eigenvalues {E1 , . . . , EJ } are nondegenerate, and denote by φ1 , . . . φJ the corresponding normalized eigenfunctions. We then have the following results: (i) For every λ ∈ (−2tL , 2tL ) \ {E1 , . . . , EJ } we have lim Tβ,γ (λ, τ ) = 0 .
τ →0
(9)
(ii) Let λ = Ej . If either S β , φj = 0 or S γ , φj = 0, then lim Tβ,γ (Ej , τ ) = 0 .
τ →0
(10)
(iii)Let λ = Ej . If both S β , φj = 0 and S γ , φj = 0, then there exist positive constants C(Ej ), such that S β , φ · S γ , φ 2 j j lim Tβ,γ (Ej , τ ) = C(Ej ) "M . τ →0  S α , φj 2
(11)
α=1
This result can be interpreted as follows. Case (i): If the energy of the incident electron is not close to the eigenvalues of H S , it will not contribute to the current. Case (ii): If the incident energy is close to some eigenvalue of H S , but the eigenfunction is not localized along both contact points S β and S γ , again there is no current. Case (iii): In order to have a peak in the current it is necessary for H S to have extended edge states, which couple to several leads.
3 A Numerical Example We end this contribution with some numerical results on the transport through a noninteracting quantum dot described by a discrete lattice containing 20 × 20 sites and coupled to two leads connected to two opposite corners. The magnetic ﬂux is ﬁxed and measured in arbitrary units, while the leaddot coupling was set to τ = 0.2. The sample Hamiltonian H S is given by the Dirichlet restriction to the above mentioned ﬁnite domain of Bm (E0 + Vg )m, n m, n + t1 (e−i 2 m, n m, n + 1 + h.c.) H S (Vg ) = (m,n)∈Z2
+ t2 (e−i
Bn 2
m, n m + 1, n + h.c.) .
(12)
The LandauerB¨ uttiker Formula and Resonant Quantum Transport
49
Here h.c. means hermitian conjugate, E0 is the reference energy, B is a magnetic ﬁeld, from which the magnetic phases appear (the symmetric gauge was used), while t1 and t2 are hopping integrals between nearest neighbor sites. The constant denoted Vg adds to the onsite energies E0 , simulating the socalled “plunger gate voltage” in terms of which the conductance is measured in the physical literature. The variation of Vg has the role to “move” the dot levels across the ﬁxed Fermi level of the system (recall that the latter is entirely controlled by the semiinﬁnite leads). Otherwise stated, the eigenvalues of H S (Vg ) equal the ones of H S (Vg = 0) (we denote them by uttiker formula (8), and {Ei }), up to a global shift Vg . Using the LandauerB¨ the formulas (3.8) and (4.6) in [2], it turns out that the computation of the conductivity between the two leads (or equivalently, of T12 ) reduces to the inversion of an eﬀective Hamiltonian. Moreover, when Vg is ﬁxed such that there exists an eigenvalue Ei of H S (Vg = 0) obeying Ei + Vg = EF , the transmittance behavior is described by (11). Thus one expects to see a series of peaks as Vg is varied. Here the Fermi level was ﬁxed to EF = 0.0 and the hopping constants in the lattice t1 = 1.01 and t2 = 0.99. We have taken the magnetic ﬂux (which equals B for the unit lattice) to 0.15. Then the resonances appear, whenever Vg = −Ei (since the spectrum of our discrete operator H S (0) is a subset of [−4, 4], the suitable interval for varying Vg is the same). Before discussing the resonant transport let us analyze the spectrum of our dot at Vg = 0, in order to emphasize the role of the magnetic ﬁeld. We recall that we used Dirichlet boundary conditions (DBC) and the magnetic ﬁeld appears in the Peierls phases of H S (see (12)). In Fig. 1 we plot the ﬁrst 200 eigenvalues (this suﬃces since the spectrum is symmetrically located with respect to 0, i.e both Ei and −Ei belong to σ(HS (0))). One notices two things. 0 0.5
Energy
1 1.5 2 2.5 3 3.5
50
100 Eigenvalue label
Fig. 1. The dot spectrum
150
200
50
H.D. Cornean et al.
First, there are two narrow energy intervals ([−3.17, −3.16] and [−1.75, 1.65]) covered by many eigenvalues (∼33 and 45 respectively). Secondly, the much larger ranges [−3.16, −1.72] and [−1.65, −0.8] contain only 25 and 30 eigenvalues. This particular structure of the spectrum is due to both the magnetic ﬁeld and the DBC. The dense regions are reminescences of the Landau levels of the inﬁnite system while the largely spaced eigenvalues appear between the Landau levels due to the DBC. As we shall see below their corresponding eigenfunctions are mostly located on the edge of the sample. As the energy approaches zero, the distinction between edge and bulk states is not anymore clear and one can have quite complex topologies for eigenfunctions. We point out that the “clarity”, the length, and the number of edge states regions intercalated in the Landau gaps, increase as the sample gets bigger. Now let us again comment on (11). Here Ei must be replaced by Ei + Vg , where Ei are eigenvalues of H S (0). Remember that we took µ = 0. The number of leads is M = 2. By inspecting formula (4.6) in [2], one can show that the constant C(Ei + Vg ) will always equal 4 (we have kµ = π/2 and tL = 1). Therefore, each time we fulﬁll the condition Vg = −Ei , we obtain a peak in the transmittance, which for small τ should be close to S 1 , φ · S 2 , φ 2 j j 4 "2 (13) ≤1. α=1  S α , φj 2 We have equality with 1, if and only if  S 1 , φj  =  S 2 , φj , and this does not depend on the magnitude of these quantities. Therefore, even for weakly coupled, but completely symmetric eigenfunctions, we can expect to have a strong signal. In fact, in this case the relevant parameter is !  S 1 , φj   S 2 , φj  , min . (14)  S 2 , φj   S 1 , φj  Now let us investigate how the transmittance behaves, when Vg is varied. Figure 2a shows the peaks corresponding to the ﬁrst six (negative) eigenvalues of H S (Vg = 0). Their amplitude is very small because the associated eigenvectors are (exponentially) small at the contact sites, and not completely symmetric (since t1 = t2 ). In fact, a few eigenvectors with more symmetry do generate some small peaks. The spatial localisation of the second and the sixth eigenvector is shown in Figs. 2b,c. The peak aspect changes drastically at lower gate potentials as the Fermi level encounters levels whose eigenstates have a strong component on the contact subspace (see Figs. 3b and 3c for the spatial localisation of the 38th and the 49th eigenstate). The transmittance is close to unity in this regime, since the parameter in (14) is also nearly one. This is explained by the fact that t1 and t2 have very close values, and the relative perturbation induced by the lack of symmetry is much smaller than for the bulk states. One notices that the width of the peaks increases as Vg is decreased as well as their
The LandauerB¨ uttiker Formula and Resonant Quantum Transport 0.14 0.12
Transmittance
0.1 0.08 0.06 0.04 0.02 0 3.1676
3.1678
3.168
3.1682
3.1684
3.1686
3.1688
Gate potential
2nd state
0.06 0.05 0.04 0.03 0.02 0.01 0
0.016 0.014 0.012 0.01 0.008 0.006 0.004 0.002 0
2
2
4
4
6
6
8
8
10
10
12
12
14
14
16
16
18
18
2
2
4
4
6
6
8
8
10
12
14
16
18
20
6th state
10
12
14
Fig. 2. Top to bottom: parts a, b, c
16
18
20
3.169
51
H.D. Cornean et al. 1 0.9 0.8 0.7 Transmittance
52
0.6 0.5 0.4 0.3 0.2 0.1 0 3
3.02
3.04
3.06
3.08
3.1
3.12
3.14
Gate potential
38th state
0.012 0.01 0.008 0.006 0.004 0.002 2
4
6
8
10
12
14
16
18
2
4
6
8
10
12
14
16
18
20
49th state
0.012 0.01 0.008 0.006 0.004 0.002 0
2
4
6
8
10
12
14
16
18
2
4
6
8
10
12
14
16
Fig. 3. Top to bottom: parts a, b, c
18
20
3.16
The LandauerB¨ uttiker Formula and Resonant Quantum Transport
53
separation. In Figs. 3b,c we have plotted the 38th eigenfunction, which gives the ﬁrst peak on the right of Fig. 3a, and the 49th eigenfunction associated to the peak around Vg = 3.06.
References 1. W. Aschbacher, V. Jaksic, Y. Pautrat, C.A. Pillet: Introduction to nonequilibrium quantum statistical mechanics Preprint, 2005. 2. H.D. Cornean, A. Jensen, and V. Moldoveanu: A rigorous proof of the LandauerB¨ uttiker formula. J. Math. Phys. 46, 042106 (2005). 3. S. Datta: Electronic transport in mesoscopic systems Cambridge University Press, 1995. 4. V. Jaksic, C.A. Pillet, Comm. Math. Phys. 226, no. 1, 131–162 (2002). 5. V. Jaksic, C.A. Pillet, J. Statist. Phys. 108 no. 56, 787–829 (2002). 6. V. Moldoveanu, A. Aldea, A. Manolescu, M. Nit¸a ˘, Phys. Rev. B 63, 045301045309 (2001). 7. D.R. Yafaev: Mathematical Scattering Theory, Amer. Math. Soc. 1992.
Point Interaction Polygons: An Isoperimetric Problem Pavel Exner Department of Theoretical Physics, Nuclear Physics Institute, Academy of ˇ z near Prague, Czechia, and Doppler Institute, Czech Sciences, 25068 Reˇ Technical University, Bˇrehov´ a 7, 11519 Prague, Czechia [emailprotected]
We will discuss a new type of an isoperimetric problem concerning a Hamiltonian with N point interactions in Rd , d = 2, 3, all with the same coupling constant, placed at vertices of an equilateral polygon PN . We show that the ground state energy is locally maximized by a regular polygon and conjecture that the maximum is global; on the way we encounter an interesting geometric inequality. We will also mention some extensions of this problem.
1 Introduction In my talk at the QMath9 conference I discussed three recent mathematical results inspired by investigations of quantum waveguides. Two of them are either already published or will be soon [2, 11] while the third one [6] is still not at the time when I am writing this text; this is the reason why I have chosen to devote this proceedings contribution to it. In addition, the problem is amusing and has various extensions worth to explore. Relations between geometrical conﬁgurations and extremal values of spectral quantities are a trademark topic of mathematical physics. Let me recall the FaberKrahn inequality [12, 14] which states that among all regions of a ﬁxed volume the principal eigenvalue of the Dirichlet Laplacian is minimized by a ball. What is intriguing in this and similar results, however, is that they are sensitive to the topology. If you consider nonsimply connected regions of annular type, for instance, then a rotationally symmetric shape on the contrary typically maximizes the groundstate energy [8, 13]. The question is what happens if the particle conﬁnement is much weaker than a hard wall modelled by Dirichlet boundary condition, so that speaking about topology loses meaning. Since I do not adhere to the Bourbaki philosophy and believe that a good example is vital for any theory, I am going to discuss a simple model situation in which the conﬁnement is extremely weak being realized by a loopshaped “polymertype” point interaction array deﬁned as in [1]. It has been demonstrated recently that spectral properties of such arrays depend substantially on their geometry, in particular, that curvature leads to an eﬀective attractive interaction [4, 10]. In this context it is natural to ask whether the mentioned result about Dirichlet annuli [8] has P. Exner: Point Interaction Polygons: An Isoperimetric Problem, Lect. Notes Phys. 690, 55–64 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
56
P. Exner
an analogue in the situation when the point interactions are arranged along a closed curve of a ﬁxed length. We will state below this isoperimetric problem properly and show that the shape close as possible to the circular one, namely a regular polygon, is a local maximizer for the principal eigenvalue. I am unable for the time being to answer the question about the global uniqueness of this maximizer. What I can do, however, and what I will demonstrate is its reformulation in terms of a purely geometric problem which you would expect to be solved in Euclid’s Στ oιχειoν, or at least not much later. Surprisingly, it seems to be open, and it has an appeal which problems expressible in terms of grammarschool mathematics usually have. At the same time, the problem has various extensions which are no less interesting.
2 The Local Result in Geometric Terms Let PN ⊂ Rd , d = 2, 3, be a polygon which we will for the present purpose identify with an ordered set of its vertices, PN = {y1 , . . . , yN }; if the vertex indices exceed this range they are understood mod N . We suppose that PN is equilateral, yi+1 − yi  = for a ﬁxed > 0 and any i. By P˜N we denote a regular polygon of the edge length , which means planar (this is trivial, of π −1 course, if d = 2) with vertices lying on a circle of radius 2 sin N . We are interested in the operator −∆α,PN in L2 (Rd ) with N point interactions in the sense of [1], all of the same coupling α, placed at the vertices of PN . We suppose that it has a nonempty discrete spectrum,
1 ≡ 1 (α, PN ) := inf σ (−∆α,PN ) < 0 , which is satisﬁed for any α ∈ R if d = 2, while in the case d = 3 it is true below a certain critical value of α – cf. [1, Sect. II.1]. The main result which I am going to report here can be then stated as follows. Theorem 1. Under the stated conditions, 1 (α, PN ) is for ﬁxed α and locally sharply maximized by a regular polygon, PN = P˜N . It goes without saying that speaking about uniqueness of the maximizer, we have always in mind the family of regular polynomials related mutually by Euclidean transformations of the underlying space Rd . To prove Theorem 1 we are going to show ﬁrst that the task can be reformulated in purely geometric terms. Using k = iκ with κ > 0, we ﬁnd the eigenvalues −κ2 solving the following spectral condition, det Γk = 0
with
k (Γk )ij := (α − ξ k )δij − (1 − δij )gij ,
k where gij := Gk (yi − yj ), or equivalently ' 1 2π K0 (κyi − yj ) k gij = e−κyi −yj  4πyi −yj 
...
d=2
...
d=3
(1)
Point Interaction Polygons: An Isoperimetric Problem
57
and the regularized Green’s function at the interaction site is ' κ 1 ln 2 + γE ... d = 2 − 2π k ξ = κ − 4π ... d = 3 The matrix Γiκ has N eigenvalues counting multiplicity which are increasing in (0, ∞) as functions of κ – see [15] and recall that they are realanalytic and nonconstant in view of their known asymptotic behavior [1]. The sought quantity, 1 (α, PN ), corresponds to the point κ where the lowest of these eigenvalues vanishes. Consequently, we have to check that min σ(Γi˜κ1 ) < min σ(Γ˜i˜κ1 )
(2)
κ21 = 1 (α, P˜N ). holds locally for PN = P˜N , where −˜ Next we notice that the lowest eigenvalue of Γ˜i˜κ1 corresponds to the eigenvector φ˜1 = N −1/2 (1, . . . , 1). Indeed, by [1] there is a bijective correspondence between an eigenfunction c = (c1 , . . . , cN ) of Γiκ at the point, where the corresponding eigenvalue equals "N zero, and the corresponding eigenfunction of −∆α,PN given by c ↔ j=1 cj Giκ (· − yj ), up to a normalization. Again by [1], the principal eigenvalue of −∆α,PN is simple, so it has to be associated with a onedimensional representation of the corresponding discrete symmetry group of P˜N ; it follows that c1 = · · · = cN . Hence min σ(Γ˜i˜κ1 ) = (φ˜1 , Γ˜i˜κ1 φ˜1 ) = α − ξ i˜κ1 −
2 i˜κ1 g˜ . N i<j ij
(3)
On the other hand, for the l.h.s. of (2) we have a variational estimate 2 i˜κ1 g , min σ(Γi˜κ1 ) ≤ (φ˜1 , Γi˜κ1 φ˜1 ) = α − ξ i˜κ1 − N i<j ij and therefore it is certainly suﬃcient to check that the inequality Giκ (yi − yj ) > Giκ (˜ yi − y˜j ) i<j
(4)
i<j
holds for all κ > 0 and PN = P˜N in the vicinity of the regular polygon P˜N . For brevity we introduce the symbol ij for the diagonal length yi − yj  and ˜ij := ˜ yi − y˜j . We deﬁne the function F : (R+ )N (N −3)/2 → R by
[N/2]
F ({ij }) :=
#
$ Giκ (ij ) − Giκ (˜ij ) ;
m=2 i−j=m
notice that m = 1 does not contribute because i,i+1 = ˜i,i+1 = by assumption. Our aim is to show that F ({ij }) > 0 except if {ij } = {˜ij }. We use
58
P. Exner
the fact that the function Giκ (·) is convex for any ﬁxed κ > 0 and d = 2, 3 as it can be seen from cf. (1); this yields the inequality [N/2] 1 F ({ij }) ≥ νm Giκ ij − Giκ (˜1,1+m ) , ν m m=2 i−j=m
where νn is the number of the appropriate diagonals, ' & % N . . . m = 1, . . . , 12 (N − 1) νm := 1 . . . m = 12 N for N even 2N At the same time, Giκ (·) is monotonously decreasing in (0, ∞), so the sought claim would follow if we demonstrate the inequality 1 ˜1,m+1 ≥ νn
ij
i−j=m
and show that it is sharp for at least one value of m if PN = P˜N . In this way we have reduced the task to veriﬁcation of a geometric inequality. Since it may be of independent interest we will state it more generally, without dimensional restrictions. Let PN be an equilateral polygon in Rd , d ≥ 2. Given a ﬁxed integer m = 2, . . . , [ 12 N ] we denote by Dm the sum of lengths of all mdiagonals, i.e. the diagonals jumping over m vertices. (Pm ) The quantity Dm is, in the set of equilateral polygons PN ⊂ Rd with ˜ m referring to the a ﬁxed edge length > 0, uniquely maximized by D ˜ (family of) regular polygon(s) PN .
3 Proof of Theorem 1 We have thus to demonstrate the following claim for d = 2, 3. Theorem 2. The property (Pm ) holds locally for any m = 2, . . . , [ 12 N ]. Proof. Let us look for local maxima of the function fm : fm (y1 , . . . , yN ) =
N 1 yi − yi+m  N i=1
under the constraints gi (y1 , . . . , yn ) = 0, where gi (y1 , . . . , yn ) := − yi − yi+1  ,
i = 1, . . . , N .
Point Interaction Polygons: An Isoperimetric Problem
59
There are (N − 2)(d − 1) − 1 independent variables here because 2d − 1 parameters are related to Euclidean transformations and can be ﬁxed. We put N Km (y1 , . . . , yN ) := fm (y1 , . . . , yN ) + λr gr (y1 , . . . , yn ) (5) r=1
and compute the derivatives ∇j Km (y1 , . . . , yN ) which are equal to ! yj − yj+m yj − yj−m yj − yj+1 yj − yj−1 1 + − λj−1 . − λj N yj − yj+m  yj − yj−m  We want to show that these expressions vanish for a regular polygon. Let us introduce a parametrization for any planar equilateral polygon. Without loss of generality we may suppose that it lies in the plane of the ﬁrst two axes. The other coordinates are then zero and we neglect them writing j−1 n j−1 n yj = cos βi − ϕ , sin βi − ϕ , (6) n=0
n=0
i=1
i=1
where ϕ ∈ R is a free parameter and βi is the “bending angle” at the ith vertex (modulo 2π); the family of these angles satisﬁes naturally the condition N
βi = 2πw
(7)
i=1
for some w ∈ Z \ {0}. Choosing ϕ˜ =
Then we have
2πi N ,
we get in particular
m−1 π π = ± cos (2n + 1), sin (2n + 1) . N N n=0 n=0
y˜±m
and β˜i =
π N
m−1
˜ yj − y˜j±m  =
m−1
2
cos
n=0
π (2n + 1) N
+
m−1 n=0
2
sin
π (2n + 1) N
=: Υm ,
and consequently, ∇j Km (˜ y1 , . . . , y˜N ) = 0 holds for j = 1, . . . , N if we choose all the Lagrange multipliers λr in (5) equal to the expression λ=
σm N Υm
"m−1 with
σm :=
n=0
π sin N (2n + 1) sin2 πm N = . π π sin N sin2 N
(8)
The second partial derivatives, ∇k,r ∇j,s Km (y1 , . . . , yN ), are computed to be
60
P. Exner
1 N
'
δkj − δk,j+m (yj − yj+m )r (yj − yj+m )s (δkj − δk,j+m ) δrs − yj − yj+m  yj − yj+m 3
δkj − δk,j−m (yj − yj−m )r (yj − yj−m )s (δkj − δk,j−m ) δrs − yj − yj−m  yj − yj−m 3 , λ + δk,j+m + δk,j−m − 2δkj δrs . +
This allows us to evaluate the Hessian at the stationary point corresponding to P˜N . After a long but straightforward calculation we arrive at the expression ∇k,r ∇j,s Km (˜ y1 , . . . , y˜N )ξk,r ξj,s (9) k,j,r,s
=
! N 1 (ξj − ξj+m , y˜j − y˜j+m )2 2 − σ ξ − ξ  ξj − ξj+m 2 − . m j j+1 N Υm j=1 ˜ yj − y˜j+m 2
We observe that the form depends on vector diﬀerences only, in other words, it is invariant with respect to translations and transforms naturally when the system is rotated. Furthermore, the sum of the ﬁrst two terms in the bracket at the r.h.s. of (9) is nonnegative by Schwarz inequality. Since the second term in nonpositive, it will be suﬃcient to establish negative deﬁniteness of the “reduced” quadratic form . ξj − ξj+m 2 − σm ξj − ξj+1 2 (10) ξ → Sm [ξ] := j
on RN d . Moreover, it is enough to consider here the case d = 1 only because Sm is a sum of its “component” forms. We observe that the matrices corresponding to the two parts of (10) can be simultaneously diagonalized; sin 2πr (µr j)}N the corresponding eigenfunctions are { cos j=1 , where µr = N and r = 0, 1, . . . , m − 1. Taking the corresponding eigenvalues we see that it is necessary to establish the inequalities πmr πr − σm sin2 4 sin2 Um−1 cos . N N One can check this inequality directly, because (12) is equivalent to
Point Interaction Polygons: An Isoperimetric Problem
sin
π πr πmr πm sin > sin sin , N N N N
2≤r<m≤
N 2
61
.
We have sin x sin(η 2 /x) ≥ sin η for a ﬁxed η ∈ (0, 12 π) and 2η 2 /π ≤ x ≤ 12 π, and moreover, this inequality is sharp if x = η, hence the desired assertion 2 π sin Nπx ≥ 0 valid for x ∈ (0, 12 π). follows from the inequality sin2 x − sin N This concludes the proof of Theorem 2, and by that also of Theorem 1.
4 About the Global Maximizer I have said that the question whether the maximizer represented by regular polygons is global remain open. Using the “geometrization” argument described above, we see that it is necessary to prove the following claim. Conjecture 1. The property (Pm ) holds globally for any m = 2, . . . , [ 12 N ]. Let us look at the problem in more detail in the particular case of planar polygons, d = 2. We employ a parametrization analogous to (6): for a ﬁxed i we identify yi with the origin and set for simplicity ϕ = 0, i.e. m−1 n m−1 n cos βj+i , sin βj+i ; yi+m = 1 + n=1
n=1
j=1
j=1
in addition to the “winding number” condition (7) we require naturally also that yi = yi+N , or in other words 1+
N −1 n=1
cos
n j=1
βj+i =
N −1 n=1
sin
n
βj+i = 0
(13)
j=1
for any i = 1, . . . , N . The mean length of all mdiagonals is easily found,
Mm
2 2 1/2 N m−1 n m−1 n = cos βj+i + sin βj+i , 1+ N i=1 n=1 n=1 j=1 j=1
or alternatively
Mm
1/2 N m−1 n n m + 2 = cos βj+i . N i=1 n=1 r=1 j=r
This result allows us to prove the claim in the simplest nontrivial case. Proposition 1. The property (P2 ) holds globally if d = 2.
(14)
62
P. Exner
Proof. By (14) the mean length of the 2diagonals equals √ N N 2 2 βi (1 + cos βi )1/2 = cos ; M2 = N i=1 N i=1 2 notice that cos β2i > 0 because βi ∈ (−π, π). Using now convexity of the function u → − cos u2 in (−π, π) together with the condition (7) we ﬁnd N N βi π βi ≥ −N cos = −N cos , − cos 2 2 N i=1 i=1 π ˜ 2 . Moreover, since the said function is and therefore M2 ≤ 2 cos N = M strictly convex, the inequality is sharp unless all the βi ’s are the same.
For m ≥ 3 the situation is more complicated and one has to take into account also the condition (13); for the moment the problem remains open.
5 Some Extensions The main task which this lecture raises is, of course, to prove Conjecture 1 and by that the global uniqueness of the maximizer. At the same time, the present problem oﬀers various other extensions. One may ask, for instance, what will be the maximizer when we replace the equilaterality by a prescribed ordered N tuple of polygon lengths {j } and/or coupling constants {αj }. In both cases the task becomes more diﬃcult because we loose the ground state symmetry which yielded the relation (3) and consequently the geometric reformulation based on the inequality (4). Using a perturbative approach, one may expect that for small symmetry violation the maximizer will not be far from P˜N , while in the general case we have no guiding principle. One can also attempt to extend the result to point interaction family of point interactions in R3 placed on a closed surface. In this case, however, there is no unique counterpart to the equilaterality and one has to decide ﬁrst what the “basic cell” of such a polyhedron surface should be. Another extensions of our isoperimetric problem concern “continuous” versions of the present situation, i.e. Schr¨ odinger operators with singular interactions supported by closed curves or surfaces – cf. [5, 9] and references therein – or with a “transverse” potential well extended along a closed curve. Let us restrict here to describing the situation in the simplest case when we have an operators in L2 (R2 ) given formally by the expression Hα,Γ = −∆ − αδ(x − Γ ) ,
(15)
where α > 0 and Γ is a C 2 smooth loop in the plane of a ﬁxed length L > 0 without cuspshaped selfintersections; for a proper deﬁnition of the operator (15) see [9]. In analogy with Theorem 1 we have the following claim.
Point Interaction Polygons: An Isoperimetric Problem
63
Theorem 3. Within the speciﬁed class of curves, 1 (α, Γ ) is for any ﬁxed α > 0 and L > 0 locally sharply maximized by a circle. One can also conjecture that the circle is a sharp global maximizer, even under weaker regularity assumptions. The proof is based on the generalized BirmanSchwinger principle [3] which makes it again possible to rewrite the problem in purely geometric terms – see [7] for details – namely as the task to check the inequality
L
Γ (s+u) − Γ (s)p ds ≤ 0
πu L1+p sinp πp L
(16)
for p = 1 and all u ∈ (0, 12 L] and to show that is sharp unless Γ is a circle. Moreover, it follows from the convexity of the function x → x2 that it is suﬃcient to prove the inequality (16) with p = 2 instead. While we believe that these inequalities hold globally, we have at the moment a proof only for a particular case when u is small enough; the situation is similar to that in Proposition 1. To prove the local result, one has to express Γ through its (signed) curvature γ := Γ˙2 Γ¨1 − Γ˙1 Γ¨2 ; up to a Euclidean transformations we have s s Γ (s) = cos β(s ) ds , sin β(s ) ds , (17) 0
s
where β(s) := 0 γ(s ) ds is the bending angle relative to the tangent at the chosen initial point, s = 0; to ensure that the curve is closed, the conditions L L cos β(s ) ds = 0 sin β(s ) ds = 0 must be satisﬁed. Then the lefthand 0 side of the inequality (16) with p = 2 becomes 2 2 2 1 L
s+u
cos β(s ) ds
s+u
+
s
sin β(s ) ds
ds := c2Γ (u) ;
(18)
s
the relation (17) is analogous to (6), and similarly for the chord length expressions. A sequence of integral transformations described in detail in [7] brings the quantity in question to the form 1 u
L
dx (u − x)
c2Γ (u) = 2 0
z+ 2 x
dz cos 0
γ(s) ds
.
(19)
z− 12 x
Gentle deformations of a circle can be characterized by the curvature γ(s) = 2π and small in the sense that g∞ L−1 and L +g(s), where g is continuous L satisﬁes the condition 0 g(s) ds = 0. Expanding in powers of g and using the last named condition, we ﬁnd c2Γ (u) =
πu L3 − Ig (u) + O(g 3 ) , sin2 π2 L
64
P. Exner
where the error term is a shorthand for O(Lg3∞ ). One has to show that Ig (u) > 0 unless g = 0 identically which is trivial for u ≤ 14 L while for u ∈ ( 14 L, 12 L] one can employ expansion of g into a Fourier series – cf. [7]. Naturally, a further extension of the present problem concerns a maximizer for the generalized Schr¨ odinger operator in R3 with an attractive δ interaction supported by a closed surface of a ﬁxed area A, and its generalization to closed hypersurfaces of codimension one in Rd , d > 3. In the case of d = 3 we have again a heuristic argument relying on [5, 8] which suggests that the problem is solved by the sphere provided the discrete spectrum is not empty, of course, which is a nontrivial assumption in this case. The BirmanSchwinger reduction of the problem can be performed again leading to to new modiﬁcations of the geometric inequalities discussed above.
Acknowledgments The research has been partially supported by ASCR within the project IRP AV0Z10480505, its presentation by the QMath9 organizing committee.
References 1. S. Albeverio, F. Gesztesy, R. HøeghKrohn, H. Holden: Solvable Models in Quantum Mechanics, 2nd edn (AMS Chelsea, Providence, R.I., 2005). 2. D. Borisov, P. Exner: J. Phys. A: Math. Gen. 37, 3411 (2004). ˇ 3. J.F. Brasche, P. Exner, Yu.A. Kuperin, P. Seba: J. Math. Anal. Appl. 184, 112 (1994). 4. P. Exner: Lett. Math. Phys. 57, 87 (2001). 5. P. Exner: Spectral properties of Schr¨ odinger operators with a strongly attractive δ interaction supported by a surface. In Contemporary Mathematics, vol 339, ed by P. Kuchment (AMS, Providence, R.I., 2003) pp 25–36. 6. P. Exner: An isoperimetric problem for point interactions, J. Phys. A: Math. Gen. 38, 4795 (2005). 7. P. Exner: An isoperimetric problem for leaky loops and related meanchord inequalities, J. Math. Phys. 46, 062105 (2005). 8. P. Exner, E.M. Harrell, M. Loss: Optimal eigenvalues for some Laplacians and Schr¨ odinger operators depending on curvature. In Operator Theory : Advances and Applications, vol 108, ed by J. Dittrich et al. (Birkh¨ auser, Basel 1998) pp 47–53. 9. P. Exner, T. Ichinose: J. Phys. A: Math. Gen. 34, 1439 (2001). 10. P. Exner, K. Nˇemcov´ a: J. Phys. A: Math. Gen. 36, 10173 (2003). 11. P. Exner, O. Post: Convergence of spectra of graphlike thin manifolds, J. Geom. Phys. 54, 77 (2005). 12. G. Faber: Sitzungber. der math.phys. Klasse der Bayerische Akad. der Wiss. zu M¨ unchen, 169 (1923). 13. E.M. Harrell, P. Kr¨ oger, K. Kurata: SIAM J. Math. Anal. 33, 240 (2001). 14. E. Krahn: Ann. Math. 94, 97 (1925). 15. M.G. Krein, G.K. Langer: Sov. J. Funct. Anal. Appl. 5, 59 (1971).
Limit Cycles in Quantum Mechanics Stanislaw D. Glazek Institute of Theoretical Physics, Warsaw University, Poland [emailprotected]
1 Introduction This lecture concerns limit cycles in renormalization group (RG) behavior of quantum Hamiltonians. Cyclic behavior is perhaps more common in quantum mechanics than the ﬁxedpoint behavior which is wellknown from critical phenomena in classical statistical mechanics. We discuss a simple Hamiltonian model that exhibits limit cycle behavior. The value of the model is its simplicity. The model can serve as a basis for further mathematical studies of the cycle for the purpose of solving complex physical theories. For example, the limit cycle may be related to existence of a set of bound states with binding energies forming a geometric sequence converging to zero. The model oﬀers a possibility of studying in detail how Hamiltonians with bound states behave as operators under the RG transformation in the vicinity of the cycle. This lecture reports on results concerning behavior of marginal and irrelevant operators in the model. The possibility of limit cycle behavior in RG calculations was originally pointed out by Wilson [1] in the context of theory of strong interactions in particle physics. The classiﬁcation of operators near a ﬁxed point as relevant, marginal, and irrelevant, was introduced by Wegner [2] in the context of critical phenomena and corrections to scaling in condensed matter physics. Recent interest in the RG limit cycle stems from the structure of threenucleon boundstate spectrum in the case of twobody nucleonnucleon interactions that have a very short range in comparison to the scattering length. This structure was ﬁrst noticed by Thomas [3] and subsequently discussed by Eﬁmov [4]. It was only recently associated with ultraviolet regularization dependence of parameters in the nuclear potentials by Bedaque, Hammer, and van Kolck [5]. Interactions of atoms, especially in Helium trimers, may exhibit a similar cycle structure even more transparently than eﬀective nuclear forces because more bound states may exist for atoms than for nuclei. In addition, atomic interactions are easier to tune close to the cycle structure by changing experimental conditions than nuclear forces. The cyclic fewbody atomic interactions also contribute to manybody dynamics. Braaten, Hammer, and Kusunoki discussed some eﬀects in a BoseEinstein condensate [6]. LeClair, Roman, and Sierra discussed a theoretical possibility of variation in the superconductivity mechanism [7]. On the other hand, the S.D. Glazek: Limit Cycles in Quantum Mechanics, Lect. Notes Phys. 690, 65–78 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
66
S.D. Glazek
range of phenomena that may exhibit cycle structure extends also to subnuclear domain. Namely, Braaten and Hammer observed that masses of the up and down quarks appear to be close to special values at which an infrared cyclic behavior may develop in quantum chromodynamics [8]. From a mathematical point of view, it is also interesting that the cyclic behavior occurs in the elementary case of twobody quantum mechanics with the wellknown potential r−2 , as recently described by Braaten and Phillips [9]. Readers interested in the literature that discusses quantum mechanics with shortrange interactions without direct reference to the RG limit cycle can consult a review article by Nielsen, Fedorov, Jensen and Garrido [10]. Braaten and Hammer recently wrote a review article on universal cyclic properties of physical systems with twobody short range potentials and large scattering length in the context of renormalization of the potentials [11]. See also [12]. It should be mentioned that examples of renormalization group maps, which have critical attractors that are not simple ﬁxed points, have been reported in the literature concerning classical dynamical systems. The author has learned about existence of a number of works during the process of preparation of this report [13–20]. The model Hamiltonian reviewed here has been discovered in our RG studies of Hamiltonians that were focused on hadronic parton dynamics in the inﬁnite momentum frame [21]. We have recognized that the model has a RG cycle in a later article [22, 23]. We have subsequently found an analytic solution for the RG behavior of Hamiltonians in the vicinity of the cycle in the model and this enabled us to discuss the RG universality of quantum Hamiltonians with a limit cycle using the model [24]. The purpose of my talk is to describe the cycle in our model, discuss corrections due to irrelevant operators, and show tuning to criticality, which allows the cycle spectrum to clearly appear even in approximations in which the basis of the space of quantum states is limited to about 70. This is attractive because one can inspect what happens in the model using a computer. A machine can provide numerical facts that help in developing intuition which otherwise is not available. I also address issues that are not fully understood and invite more research on mathematics of the RG limit cycle in quantum mechanics. Since atomic interactions can be strongly inﬂuenced by varying external ﬁelds (e.g., see the recent work of Roberts, Claussen, Cornish, and Wieman on magnetic ﬁeld dependence of ultracold inelastic atomic collisions [25]), better understanding of quantum Hamiltonians near a RG limit cycle is a prerequisite to systematic experimental searches for cycles in real systems. The distinct feature to look for is the geometric sequence of bound states converging at threshold.
Limit Cycles in Quantum Mechanics
67
2 Deﬁnition of the Model Our model can be deﬁned in more than one way. Let us consider a nonrelativistic oneparticle Hamiltonian in 2 dimensions in the presence of a pointlike potential H=
−∆r − gδ 2 (r ) , 2µ
(1)
and write its eigenfunctions in momentum space as d2 r −i p r φ(p ) = e ψ(r ) . (2π)2 The momentumspace eigenvalue equation reads p2 g φ(p ) − d2 q φ(q ) = E φ(p ) . 2µ (2π)2
(2)
(3)
φ can be composed of angular momentum eigenfunctions eimϕ with coeﬃcients φm that depend on p = p , φ(p ) =
∞
ei m ϕ φm (p ) ,
(4)
m=−∞
and it turns out that wave functions with m = 0 are the same as in the case with no interaction. The wave function with m = 0 satisﬁes the equation ∞ p2 g φ 0 (p ) − dq q φ 0 (q ) = E φ 0 (p ) . (5) 2µ 2π 0 Further discussion concerns only the component with m = 0. Using notation p2 → z ,
φ 0 (p ) → φ(z) ,
µ→
1 , 2
g →g , 4π
one obtains the eigenvalue equation of the form ∞ z φ(z) − g dz φ(z ) = E φ(z) .
(6)
(7)
If there is a bound state with E = −EB and EB > 0, the solution takes the wellknown shape for factorizable interactions: φ(z) = c/(z − E) with some constant c, and the eigenvalue condition for EB is ∞ dz , (8) 1=g z + EB 0 which would be ﬁne if not the fact that the integral diverges. This divergence and a cure for it are well known [26–34]. The resulting eigenvalue equations
68
S.D. Glazek
are studied in mathematical physics [35, 36]. The essence of the cure is to regulate the integral somehow and make the coupling constant g depend on the regularization in such a way that the result for EB does not depend on the regularization when it is being removed. Then, other observables become also independent of the regularization when it is being removed and instead they depend on EB . I will focus on a RG calculation for Hamiltonians in Wilson’s sense of the word [37, 38] and this point of view will be applied in discussion of the limit cycle. The point is that every interval 2n−1 ≤ z ≤ 2n contributes the same 2n amount 2n−1 dz z −1 = ln 2 to the complete integral and there are inﬁnitely many such intervals to add. The coupling constant g must be inversely proportional to the number of such intervals included in the integration. Since each of the intervals contributes the same amount, we can consider every interval as a single degree of freedom. The essence of regularization is to limit the number of degrees of freedom. The problem with the divergence is that all these degrees of freedom are coupled with equal strength and the correlation length among an inﬁnite number of degrees of freedom on the energy scale is inﬁnite. In order to deal with the large number of degrees of freedom it is enough to consider a discrete model in which the continuous variable z is replaced by a set of values zn = bn with some b > 1 (b replaces the number 2 in the integration interval discussed above). Instead of φ(z) we have φ(zn ) = φn , ∞ Λ and dz = zn dn ln b. The regularization replaces the integral 0 by ε . In N the discrete version, we have ε = bM with M large negative "N and Λ = b with N large positive. The integral is replaced by a sum n=M since dn = 1. The dependence of g on Λ means that in the discrete model we expect g to depend on N (dependence on ε or M does not require a separate RG study in the model). We denote g ln b with appropriate factors including π by gN . Finally, we obtain a Hamiltonian matrix of the form N 0 ... 0 b 1 1 ... 1 1 1 ... 1 0 bN −1 ... 0 − b k+l , 2 g (9) Hkl = N . . . ... 1 . ... 0 1 1 ... 1 kl 0 0 ... bM kl that acts on the state vectors with components χl = bl/2 φl . Our model with a limit cycle is deﬁned by the replacement 1 1 ... 1 1 1 ... 1 0 1 ... 1 1 ... 1 1 1 ... 1 −1 0 ... gN . . ... 1 → gN . . ... 1 + ihN . . ... 1 1 ... 1 1 1 ... 1 −1 −1 ...
1 1 . 1 0
(10)
This means that we complement a real symmetric interaction matrix that requires renormalization due to the large number of equal entries with an imaginary skewsymmetric matrix of similarly equal entries. Such imaginary part
Limit Cycles in Quantum Mechanics
69
is allowed in quantum mechanics. It turns out that the ultravioletdiverging Hamiltonian structure with nonzero imaginary part leads to a limit cycle, or even more chaotic RG behavior [22].
3 Renormalization Group Our model Hamiltonian matrix is Hkl (N, gN , hN ) = b
k+l 2
( δkl − gN − ihN skl ) ,
(11)
where skl = 1 for k > l and slk = −skl . In order to introduce the RG procedure let us ﬁrst set hN = 0. Then, the eigenvalue equation for the matrix "N Hkl can be rewritten, introducing ψk = bk/2 χk , k = Eb−k , and σN = l=M ψl , as the following set of equations: (1 − N ) ψN = gN σN , (1 − N −1 ) ψN −1 = gN σN , : (1 − M ) ψM = gN σN .
(12)
The ﬁrst equation allows us to express ψN by all other components with m smaller than N , ψN =
gN σN −1 , 1 − N − gN
(13)
and this value of ψN can be inserted into the remaining equations (1 − N −1 ) ψN −1 = gN σN −1 + gN ψN gN = gN 1 + σN −1 1 − N − gN gN (1 − N −2 ) ψN −2 = gN 1 + σN −1 1 − N − gN : gN (1 − M ) ψM = gN 1 + σN −1 . 1 − N − gN
(14)
This set can be written as (1 − N −1 ) ψN −1 = gN −1 σN −1 , (1 − N −2 ) ψN −2 = gN −1 σN −1 , : (1 − M ) ψM = gN −1 σN −1 ,
(15)
70
S.D. Glazek
when one deﬁnes
gN −1 = gN 1 +
gN 1 − N − gN
.
(16)
Thus, the Hamiltonian matrix of (11) is turned into a new matrix, smaller by one row and one column of highest energy bN , k, l ≤ N − 1, Hkl (N − 1, gN −1 ) = b
k+l 2
( δkl − gN −1 ) .
(17)
The new ultraviolet cutoﬀ is ΛN −1 = bN −1 = Λ/b, and there is a new coupling constant gN −1 . This process can be repeated and one obtains a recursion gn−1 = gn +
gn2 . 1 − gn − n
(18)
Since no other change occurs in the Hamiltonian matrix, (18) fully describes the RG transformation for Hamiltonians in our model. Suppose that we consider n such that the eigenvalue E can be neglected in comparison to bn . Then the RG transformation simpliﬁes to gn−1 =
gn . 1 − gn
(19)
If we denote gn−1 = RG(gn ), then RG(f ) = f /(1 − f ). This recursion has a solution fn = 1/(n + c) with arbitrary constant c, which means that gn = 1/(c + ln Λn / ln b) and the larger is the cutoﬀ Λn , the smaller is gn . This phenomenon is called asymptotic freedom [39, 40]. It is associated with the ﬁxed point, f ∗ , of the RG transformation, RG(f ∗ ) = f ∗ , which is f ∗ = 0. When one reverses the RG transformation to calculate gN for increasing N in terms of gn0 for some ﬁnite n0 , gN approaches 0 for asymptotically large N . The coupling constant gN vanishes as an inverse of a logarithm of the ultraviolet cutoﬀ Λ. Let us now speculate a bit about what may happen when the eigenvalue E is not neglected in the recursion. The RG transformation becomes related to the spectrum of the Hamiltonian. In our discrete model, after repeating the transformation N − M times, which can be a very large number, one obtains one last equation 1 − gM (E) − E/bM = 0 ,
(20)
which must be satisﬁed if E is to be an eigenvalue. But instead of a recursion for a sequence of coupling constants, there appears now a recursion for a sequence of functions gn (E) = fn (εn ) in our model. The recursion is fn−1 (bx) = RG[fn (x)] with RG[f (x)] = (1 − x)f (x)/[1 − x − f (x)]. If the repeated application of the RG transformation carried every starting function fN (x) over to a ﬁxedpoint function f ∗ (x) after many iterations, very many eigenvalues would be given by one equation
Limit Cycles in Quantum Mechanics
1 − f ∗ ( M ) − M = 0 ,
71
(21)
provided that M is suﬃciently large not to produce results sensitive to the infrared cutoﬀ M . The ﬁxedpoint function in our model with hN = 0 satisﬁes the equation f ∗ (bx) = (1 − x)f ∗ (x)/[1 − x − f ∗ (x)], which has a solution −1 1 1 1 ∗ + + +y , (22) f (x) = . . . + 1 − b2 x 1 − bx 1 − x with arbitrary number y (y is determined by one eigenvalue). Most of the spectrum is determined not by the initial function fN (x) but by the ﬁxedpoint function f ∗ (x) with some y. The largest eigenvalues may be sensitive to the initial function in the Hamiltonian with ultraviolet cutoﬀ N , which can be just a constant. The lowest eigenvalues are sensitive to the infrared cutoﬀ M because the series that extends to M = −∞ in the ﬁxedpoint solution is cut oﬀ in a model with a ﬁnite M . But a large number of eigenvalues is determined by the behavior of f ∗ (x) as function of x in the region between bM −N and 1. This behavior depends on the properties of the RG transformation rather than on the details of the initial Hamiltonian. Thus, all Hamiltonians that lead to essentially the same RG recursion produce a universal answer for the spectrum. One can also ask if the initial Hamiltonian may be tuned so that a small number of initial iterations of the RG transformation brings the Hamiltonian right to the vicinity of the ﬁxedpoint structure so that further iterations have universal properties. In such case, the spectrum may more clearly exhibit the universal structure in a larger subset of eigenvalues than in the case where the initial Hamiltonian is far away from the criticality that allows the RG transformation to quickly reach the vicinity of the ﬁxedpoint. Since the ﬁxedpoint is not changed by the RG transformation, it corresponds to a theory with an inﬁnite cutoﬀ. A theory with ﬁnite cutoﬀs must be tuned to yield the ﬁxedpoint structure. We can now proceed to our model with the limit cycle where such universality scenario exists in an explicit form and produces a geometric sequence of bound states [24]. The sequence is related to the fact that there is not just one ﬁxed point, but a whole cycle of them.
4 Limit Cycle We now consider the Hamiltonian matrix in (11) with hN = 0. By eliminating component after component as in the previous section, we arrive at the RG transformation that transforms a Hamiltonian matrix Hkl (n, gn , hn ) = b
k+l 2
( δkl − gn − ihn skl ) ,
into the one with one row and one column less and [24]
(23)
72
S.D. Glazek
gn2 + h2n , 1 − gn − n = hn ≡ hN = h .
gn−1 = gn +
(24)
hn−1
(25)
The eigenvalue condition at the end of iteration takes the form of (20). The exact RG recursion can be rewritten as αn−1 = αn + βn ,
(26)
when one introduces angles gn , h h βn = arctan . 1 − n
αn = arctan
(27) (28)
When the eigenvalue E is set to 0 (there may exist such eigenvalue), we have βn = arctan h ≡ β, and the successive coupling constants are given by the formula gn−p = h tan (αn + p β) ,
(29)
which implies that gn−p = gn when β = π/˜ p with integer p˜ ≥ 3 and p = p˜. The coupling constant goes through a cycle and returns to the same value because the function tan is periodic with period π. If p˜ were irrational, one would obtain a chaotic RG behavior [22]. Note that for E = 0 we have gn−p = gn regardless of the value of gn . Therefore, one can seek a simpliﬁed RG relationship between gn−p and gn for E near 0. For this purpose, we introduce functions fn ( n ) = gn ( n )/h and calculate the RG transformation for p successive steps of elimination of rows and columns one after another. This way we obtain fn−p (rx) = Rp [fn (x)] ,
(30)
where x = n , n−p = r n , and r = bp . There are p − 1 diﬀerent intermediate functions in this recursion but we focus on the transformation over the entire cycle because it has a ﬁxed point and we can take advantage of the universality that develops around ﬁxed points. The transformation Rp takes the form fn−p (xn−p ) =
fn (xn ) + zp (xn ) , 1 − fn (xn ) zp (xn )
(31)
where the function zp (0) = 0. For example, when p = 3, (7/4)x(1 − x) 1 + x3 − (7/4)x(1 + x) 7 21 2 343 3 = x+ x + x + O(x4 ) . 4 16 64
z3 (x)/h =
(32) (33)
Limit Cycles in Quantum Mechanics
73
The transformation Rp for p = 3 will be discussed in a later section of this lecture. A complete analysis for arbitrary integer p ≥ 3 and to all orders of expansion in powers of x is given in the literature [24].
5 Marginal and Irrelevant Operators Let us consider ﬁrst two terms in the function f (x) = f0 + f1 x + O(x2 ) ,
(34)
and ﬁnd the ﬁxedpoint values of the coeﬃcients f0 and f1 . The ﬁxed point condition is f ∗ (x) = Rp [f ∗ ] = f ∗ (x/r) + (1 + f0∗ 2 ) zp(1) x/r ,
(35)
(1) (1) and the solution is given by f0∗ = g ∗ /h and f1∗ = 1 + f0∗ 2 zp /(r − 1). zp denotes the ﬁrst term in the Taylor expansion of zp (x) in powers of x. Now we can write the function f (x) as a ﬁxedpoint function plus an inﬁnitesimal correction, f (x) = f ∗ (g ∗ , x) + df (g ∗ , x) ,
(36)
and ﬁnd the RG evolution of the corrections df (g ∗ , x) by inspecting the transformation Rp [f ∗ + df ] = f ∗ (x/r) + df (x/r) + [1 + f0∗ 2 + 2f0∗ df0 ] zp(1) x/r ,
(37)
in the linearized approximation, Lp (df ) = w df (x), where Lp denotes the derivative of Rp at the ﬁxed point and w is the Wegner eigenvalue [2]. This way we ﬁnd that in df (x) = c0 + c1 x the numbers c0 and c1 must simultaneously satisfy equations c0 = w c0 1 1 c1 + 2zp(1) f0∗ c0 = w c1 . r r
(38) (39)
The two solutions for the eigenvalues are w0 = 1 and w1 = r−1 = 1/bp < 1. The marginal operator corresponds to the eigenvalue 1 (it does not change under the RG transformation). In this case c0 is arbitrary and c1 is determined by c0 . All coeﬃcients in the expansion of the marginal operator in powers of x are determined by c0 through the eigenvalue condition [24]. The critical exponent for the marginal Wegner operator, deﬁned by the scaling relation w0 = rλ0 , is λ0 = 0. The ﬁrst irrelevant operator has c0 equal to zero and c1 arbitrary. All coeﬃcients in the ﬁrst irrelevant operator are determined by c1 in it through
74
S.D. Glazek
the eigenvalue condition. The critical exponent for the ﬁrst irrelevant Wegner operator, deﬁned by the relation w1 = rλ1 , is λ1 = −1. The linearized RG transformation allows us to ﬁnd all Wegner’s eigenoperators using condition df (x, g ∗ ) + zp (x)[df (x, g ∗ )f (rx, g ∗ ) + f (x, g ∗ )wdf (rx, g ∗ )] = wdf (rx, g ∗ ) , (40) and we ﬁnd that the operators df whose Taylor expansion starts with terms ∼xl are irrelevant with eigenvalues wl = r−l . This analysis allows us to understand how one should choose the initial Hamiltonian in order to bring it close to criticality.
6 Tuning to a Cycle We want to make sure that the initial H transforms under the ﬁrst few RG transformations to such a matrix that the marginal operator takes some convenient value in a cycle and the ﬁrst few irrelevant operators vanish. The latter condition can be realized by eliminating the ﬁrst few terms in the Taylor expansion in powers of x. We consider tuning of our model Hamiltonian to a limit cycle using example with p = 3, b = 2, N = 17, and M = −51. In a physical system in which external conditions deﬁne the range of available options for tuning the system to a cycle, one has to proceed by analogy and identify the key factors that can expose the cycle within a limited window of energy scales available experimentally. The idea we use here in the mathematical model is to modify only the four upperleft (highenergy) entries in the matrices Hkl in (9) and (10) according to the rule N tb 0 0 . 0 t + iv g + ih . 0 tbN −1 0 . t − iv 0 g + ih . − b k+l 2 Hkl → N −2 0 g − ih g − ih g . . (41) 0 b . . . . . kl . . . . kl √ We are lucky doing this, because for v = 3 and t = 20/3 the nonlinear exact RG transformation produces in two iterations gN −2 (x) = −7 + 13 x −
65 2 x + o(x3 ) . 3
(42)
This is precisely the ﬁrst three terms in the marginal operator solution from the previous section. In our example, here we have x = E/215 . The ﬁrst two irrelevant operators with eigenvalues w1 = 1/8 and w2 = 1/64 are absent. Corrections to the limit cycle spectrum die out at the rate given by the third
Limit Cycles in Quantum Mechanics
75
irrelevant operator with eigenvalue w3 = 1/512. This is illustrated for bound state eigenvalues in Table 1 (they are found with help of a simple numerical routine). Note that in order to get rid of the infrared cutoﬀ M we had to extend the matrix to M = −60 to see precisely how the eigenvalue 1/512 appears in the cyclic spectrum. Table 1. The boundstate spectrum of a quantum Hamiltonian with a cycle tuned to criticality in marginal and ﬁrst two irrelevant operators. The ratio of successive corrections to the cycle period 23 = 8 for boundstate energies approaches the inverse of Wegner’s eigenvalue for third irrelevant operator with critical exponent −3, 1/8−3 = 512, and the entry ∼400 turns to 511.6 when the infrared cutoﬀ M = −51 is changed to M = −60. Without tuning, the ﬁrst irrelevant operator with critical exponent −1 and eigenvalue 8−1 would slow down the approach to the cycle 64 times E −111890.939577163000 −13606.149375382800 −1700.369094987040 −212.546021531960 −26.568252662770 −3.321031582839
Ratio 8.223556605927 8.001879953885 8.000004341325 8.000000008649 8.000000000017
w3−1 119 433 502 511 ∼ 400
7 Generic Properties of Limit Cycles In this section we indicate which features of the model cycle are generic and will occur in other cases and which are speciﬁc and cannot be guaranteed to occur elsewhere. See [24] for details. We cannot guarantee that bound states must appear. But we can say that if bound states do appear, they must form a geometric series converging on zero. Hamiltonians in the cycle vary but critical exponents do not vary on the cycle. This is a generic feature that should be helpful in tuning. A marginal operator is guaranteed in continuum, because the operator that changes the Hamiltonian in the direction of the cycle will be a derivative with respect to a continuous parameter that labels the cycle, as the coupling constant g ∗ did in the discrete model. But since tuning depends on nonlinear mappings from HΛ to coeﬃcients in Wegner’s operators, there is no guarantee that one will be able to eliminate any given set of Wegner’s operators. The generic properties of cyclic behavior can be studied in the model without extensive eﬀort required in realistic cases. These properties can be searched for in real physical systems using exact analytic insight into the cycle structure that the model provides.
76
S.D. Glazek
8 Conclusion Perhaps the most important question regarding the model that has no answer so far is the question about the scattering states in the continuum limit, when b tends to 1. Another question concerns the ratios of positive eigenvalues. For example, for p = 3 and b = 2,√the successive positive eigenvalues appear with ratio En /En−1 very close to 8, but not exactly. It is not clear to the author what this ratio pshould be as function of b and p beyond the suggestion that it should be b p−1 . Precise conditions required for existence of bound states in a cyclic Hamiltonian spectrum are not known. The existence of limit cycles in quantum mechanics has a philosophical consequence: in search for an ultimate theory we always have to keep in mind that the “ultimate” may mean not a ﬁxedpoint structure with parameters reaching their ultimate values but that we may rather wonder endlessly through a sequence of patterns. This comment is particularly interesting in the context of asymptotic freedom as an example of ultimate understanding of microscopic phenomena. The cyclic model shows that asymptotic freedom may be merely an intermediate part of a larger cycle. Besides, since we know that real symmetric Hamiltonians are of measure 0 in the space of complex Hermitean Hamiltonians, one may immediately suggest that the cyclic or even chaotic RG behavior of quantum Hamiltonians is more common than the ﬁxedpoint behavior known in classical theories. Hopefully, the concrete example of such surprising mathematical transparency, combined with a considerable scope of possible interpretations and applications by analogy in various branches of physics, will inspire further interest and the unknowns of the model will be resolved.
Acknowledgments I am grateful to Ken Wilson for illuminating discussions during the course of our work on the model described in this lecture, and on other subjects. I would like to thank the organizers of Qmath9 for inviting me to speak.
References 1. K.G. Wilson. Renormalization group and strong interactions. Phys. Rev., D3:1818, 1971. 2. F.J. Wegner. Corrections to scaling laws. Phys. Rev., B5:4529, 1972. 3. L.H. Thomas. The interaction between a neutron and a proton and the structure of H3. Phys. Rev., 47:903, 1935. 4. V. Eﬁmov. Energy levels arising from resonant twobody forces in a threebody system. Phys. Lett., 33B:563, 1970. 5. P.F. Bedaque, H.W. Hammer, and U. van Kolck. Renormalization of the threebody system with shortrange interactions. Phys. Rev. Lett., 82:463, 1999.
Limit Cycles in Quantum Mechanics
77
6. E. Braaten, H.W. Hammer, and M. Kusunoki. Eﬁmov states in a BoseEinstein condensate near a feshbach resonance. Phys. Rev. Lett., 90:170402, 2003. 7. A. LeClair, J.M. Roman, and G. Sierra. Russian doll renormalization group and superconductivity. Phys. Rev., B69:20505, 2004. 8. E. Braaten and H.W. Hammer. An infrared renormalization group limit cycle in qcd. Phys. Rev. Lett., 91:102002, 2003. 9. E. Braaten and D. Phillips. Renormalizationgroup limit cycle for the r−2 potential. Phys. Rev., A70:052111, 2004. 10. E. Nielsen, D.V. Fedorov, A.S. Jensen, and E. Garrido. The threebody problem with shortrange interactions. Phys. Rept., 347:373, 2001. 11. E. Braaten and H.W. Hammer. Universality in fewbody systems with large scattering length, 2004, INTPUB0427, condmath/0410417. 12. R.F. Mohr Jr. Quantum mechanical three body problem with short range interactions. Ph.D. Thesis, Ohio State U., Advisor: R. J. Perry, UMI3109134mc (microﬁche), 2003. See also R.F. Mohr, R.J. Furnstahl, R.J. Perry, K.G. Wilson, H.W. Hammer, Precise numerical results for limit cycles in the quantum threebody problem, Annals Phys. 321:225, 2006. 13. O.E. Landford III. Renormalization group methods for circle mappings, in statistical mechanics and ﬁeld theory: Mathematical aspects, edited by T. C. Dorlas, N. M. Hugenholtz, M. Winnik, Springer Vg. Lecture Notes in Physics, 257:176, 1986. 14. O.E. Landford III. Renormalization group methods for circle mappings, in G. Gallavotti, P.F. Zweifel (eds.). Nonlinear Evolution and Chaotic Phenomena, Plenum Press, p. 25, 1998. 15. M. Yampolsky. Hyperbolicity of renormalization of critical circle maps. Publ. Math. Inst. Hautes Etudes Sci., 96:1, 2002. 16. M. Yampolsky. Global renormalization horseshoe for critical circle maps. Commun. Math. Physics, 240:75, 2003. 17. R.S. MacKay. Renormalization in areapreserving maps. World Scientiﬁc, 1993. 18. J.M. Green and J. Mao. Higherorder ﬁxed points of the renormalisation operator for invariant circles. Nonlinearity, 3:69, 1990. 19. C. Chandre and H.R. Jauslin. Renormalizationgroup analysis for the transition to chaos in Hamiltonian systems. Phys. Rep., 365:1, 2002. 20. C. Chandre and H.R. Jauslin. Critical attractor and universality in a renormalizationgroup scheme for three frequency Hamiltonian systems. Phys. Rev. Lett., 81:5125, 1998. 21. S.D. Glazek and K.G. Wilson. Renormalization of overlapping transverse divergences in a model light front Hamiltonian. Phys. Rev., D47:4657, 1993. 22. S.D. Glazek and K.G. Wilson. Limit cycles in quantum theories. Phys. Rev. Lett., 89:230401, 2002. 23. S.D. Glazek and K.G. Wilson. Erratum: Limit cycles in quantum theories [Phys. Rev. Lett. 89, 230401 (2002)]. Phys. Rev. Lett., 92:139901, 2004. 24. S.D. Glazek and K.G. Wilson. Universality, marginal operators, and limit cycles. Phys. Rev., B69:094304, 2004. 25. J.L. Roberts, N.R. Claussen, S.L. Cornish, and C.E. Wieman. Magnetic ﬁeld dependence of ultracold inelastic collisions near a Feshbach resonance. Phys. Rev. Lett., 85:728, 2000. 26. E.E. Salpeter. Wave functions in momentum space. Phys. Rev., 84:1226, 1951. 27. T.D. Lee. Some special examples in renormalizable ﬁeld theory. Phys. Rev., 95:1329, 1954.
78
S.D. Glazek
28. Y. Yamaguchi. Twonucleon problem when the potential is nonlocal but separable. Phys. Rev., 95:1628, 1954. 29. W. Heisenberg. Lee model and quantisation of non linear ﬁeld equations. Nucl. Phys., 4:532, 1957. 30. Ya. B. Zel’dovich. Journal of theoretical and experimental physics. Soviet Physics JETP, 11:594, 1960. 31. F.A. Berezin and L.D. Faddeev. A remark on Schroedinger’s equation with a singular potential. Dokl. Akad. Nauk USSR, 137:1011, 1961. English translation in: Sov. Math., Dokl. 2, 372 (1961). 32. R. Jackiw. Delta function potentials in twodimensional and threedimensional quantum mechanics. In A. Ali and P. Hoodbhoy, editors, M. A. B. B´ eg Memorial Volume, Singapore, 1991. World Scientiﬁc. 33. R.L. Jaﬀe and L.R. Williamson. The Casimir energy in a separable potential. Ann. Phys., 282:432, 2000. 34. J. Derezi´ nski and R. Fr¨ uboes. Renormalization of the Friedrichs Hamiltonian. Rep. Math. Phys., 50:433, 2002. 35. D.R. Yafaev. General Theory, volume 105 of Amer. Math. Soc. Trans. Amer. Math. Soc., Providence RI, 1992. 36. S. Albeverio and P. Kurasov. Singular Perturbations of Diﬀerential Operators, volume 271 of Lond. Math. Soc. Lect. Note Ser. Cambridge University Press, Cambridge, 2000. 37. K.G. Wilson. Model Hamiltonians for local quantum ﬁeld theory. Phys. Rev., 140:B445, 1965. 38. K.G. Wilson. Model of couplingconstant renormalization. Phys. Rev., D2:1438, 1970. 39. D.J. Gross and F. Wilczek. Ultraviolet behavior of nonabelian gauge theories. Phys. Rev. Lett., 30:1343, 1973. 40. H.D. Politzer. Reliable perturbative results for strong interactions? Phys. Rev. Lett., 30:1346, 1973.
Cantor Spectrum for QuasiPeriodic Schr¨ odinger Operators Joaquim Puig Departament de Matem` atica Aplicada I. Universitat Polit`ecnica de Catalunya Av. Diagonal 647, 08028 Barcelona, Spain.
Abstract. We present some results concerning the Cantor structure of the spectrum of quasiperiodic Schr¨ odinger operators. These are obtained studying the dynamics of the corresponding eigenvalue equations, specially the notion of reducibility and Floquet theory. We will deal with the Almost Mathieu case, and the solution of the “Ten Martini Problem” for Diophantine frequencies, as well as other models.1
In recent years there has been substantial progress in the understanding of the structure of the spectrum of Schr¨ odinger operators with quasiperiodic potential. Here we will concentrate on onedimensional, real analytic quasiperiodic potentials with one or more Diophantine frequencies. We will begin with the best studied of such operators: the Almost Mathieu operator and the solution of the “Ten Martini Problem” which asks for the Cantor structure of its spectrum. Secondly we will see how this Cantor structure is generic in the set of quasiperiodic real analytic potentials. We will end introducing a diﬀerent approach to this problem which is helpful to study the phenomenon of “gap opening”.
1 The Almost Mathieu Operator & the Ten Martini Problem The Almost Mathieu operator is probably the best studied model among quasiperiodic Schr¨ odinger operators. It is the following secondorder diﬀerence operator: AM x)n = xn+1 + xn−1 + b cos(2πωn + φ)xn , (Hb,ω,φ
n∈Z,
(1)
where b is a real parameter (a coupling parameter, since for b = 0 the operator is trivial), ω is the frequency, which we assume to be an irrational number (in most of what follows, also Diophantine) and φ ∈ T = R/(2πZ) will be called the phase. 1
This work was done while the author was at the Departament de Matem` atica Aplicada i An` alisi of the Universitat de Barcelona. It has been supported by grants DGICYT BFM200309504C0201 (Spain) and CIRIT 2001 SGR70 (Catalonia). J. Puig: Cantor Spectrum for QuasiPeriodic Schr¨ odinger Operators, Lect. Notes Phys. 690, 79–91 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
80
J. Puig
Considered as an operator on l2 (Z), the Almost Mathieu operator is bounded and selfadjoint. The reason for its name comes from the fact that its eigenvalue equation, namely xn+1 + xn−1 + b cos(2πωn + φ)xn = axn ,
n∈Z,
(sometimes called Harper’s equation), is a discretization of the classical Mathieu equation, x + (a + b cos(t)) x = 0 . which is a secondorder periodic diﬀerential equation (see Ince [9]). The analogies between the Harper equation and Mathieu equation are quite striking and their comparison illustrates the diﬀerences between periodic and quasiperiodic Schr¨ odinger operators. 1.1 The ids and the Spectrum The Integrated Density of States (ids) is a very convenient object for the description of the spectrum of quasiperiodic Schr¨ odinger operators which can be extended to more general operators. Here we introduce it in the case of the Almost Mathieu operator for the sake of concreteness. AM,L as the Fix some b ∈ R and φ ∈ T. For any L ∈ N we deﬁne Hb,ω,φ restriction of the Almost Mathieu operator to the interval {1, . . . , L − 1} with zero boundary conditions at 0 and L. Let 1 AM,L AM,L # eigenvalues of Hb,ω,φ (a) = ≤a . kb,ω,φ (L − 2) Then Avron & Simon [2] prove that lim k AM,L (a) L→∞ b,ω,φ
AM = kb,ω (a) ,
which is called integrated density of states, ids for short. This limit is independent of the value of φ and it is a continuously increasing function of a. The ids can be used to describe the spectrum of quasiperiodic Schr¨ odinger AM is precisely the operators in a very nice way. Indeed, the spectrum of Hb,ω,φ set of points of increase of the map AM (a) a → kb,ω
so that the intervals of constancy belong to the resolvent set of the operator (and are called the spectral gaps). In particular, this characterization shows that the spectrum of the Almost Mathieu operator (and in general any quasiperiodic Schr¨ odinger operator with irrational frequencies) does not depend on φ. Therefore, we write AM AM . = Spec Hb,ω,φ σb,ω
Cantor Spectrum for QuasiPeriodic Schr¨ odinger Operators
81
Remark 1. The ids for onedimensional quasiperiodic Schr¨ odinger operators has many other characterizations. It can be linked to the rotation number of the corresponding eigenvalue equation. This object was introduced by Johnson & Moser [12] in the continuous case (see Delyon & Souillard [7] for the adaption to the discrete case and Johnson [11] for a review on these diﬀerent characterizations). The ids can also be used to “label” the spectral gaps of the Almost Mathieu operator. This is the contents of the Gap Labelling Theorem, by Johnson & Moser [12]: if I is a spectral gap (an interval of constancy of the ids) then there is an integer n ∈ Z such that (modulus Z)
AM (a) = nω, kb,ω
for all a ∈ I. Figure 1 displays the gap labelling for the Almost Mathieu operator at “critical coupling” b = 2.
3 2 1 4 4 1 2 3 0 5
4
3
2
1
1
2
3
4
5
AM Fig. 1. Schematic view of Gap Labelling for H2,ω,φ and b = 2. The ids is plot as a function of a. Integers in the vertical direction correspond to values n such that the ids equals nω modulus Z
The Gap Labelling Theorem motivates the following deﬁnitions. For any n ∈ Z let . AM (a) − nω ∈ Z} . I(n) = a ∈ R; kb,ω If I(n) = [an− , an+ ] for some an− < an+ then we will say that (an− , an+ ) is a noncollapsed or open spectral gap. If an− = an+ then we will call {an− } a collapsed or closed spectral gap. Note that noncollapsed spectral gaps are subsets of
82
J. Puig
the resolvent whereas collapsed spectral gaps belong to the spectrum. In both cases, the endpoints of gaps belong to the spectrum. In the quasiperiodic case, when ω is an irrational frequency, the possible values of the ids at gaps deﬁne the set of labels M(ω) = {m + nω,
n, m ∈ Z} ∩ [0, 1] ,
which is dense in [0, 1]. Since the ids is a continuously increasing function of a, the spectrum of the Almost Mathieu operator is a Cantor set if all spectral gaps are open. For a general quasiperiodic Schr¨ odinger operator, gaps can be collapsed and, in fact, the spectrum may contain intervals. Figure 2 displays a numerical computation of some of the gaps of the Almost Mathieu operator. None of them appears to be collapsed. 3 2.5 2 1.5 1 0.5 0 0
0.5
1
1.5
2
2.5
3
3.5
4
Fig. 2. Numerical computation of the ten biggest spectral gaps for the Almost Mathieu operator. Coupling parameter b is in the vertical direction, whereas the spectral one a is in the horizontal one. Shaded regions correspond to gaps
Due to all this, and to some physical arguments, Simon [18], after an oﬀer by Kac, posed the following problems on the Cantor structure for the spectrum of the Almost Mathieu operator. The ﬁrst one is the “Ten Martini Problem”: for ω irrational and b = 0 prove that the spectrum of the Almost Mathieu operator is a Cantor set. The second one, which implies the ﬁrst, is the “Strong (or Dry) Ten Martini Problem” and, under the same hypothesis, asks if all gaps, as predicted by the Gap Labelling Theorem, are open. Concerning the Ten Martini Problem, we can prove the following [14]. Corollary 1. Assume that ω ∈ R is Diophantine, that is, there exist positive constants c and r > 1 such that
Cantor Spectrum for QuasiPeriodic Schr¨ odinger Operators
sin 2πnω >
83
c nr
AM for all n = 0. Then, the spectrum of the Almost Mathieu operator, σb,ω , is a Cantor set if b = 0, ±2.
Remark 2. Very recently Avila & Jitomirskaya have proved Cantor structure for all irrational frequencies. The set of Diophantine frequencies is a total measure subset of the real numbers. Concerning the dry version of the Ten Martini Problem, using a reducibility theorem by Eliasson [8], one can also say something in the perturbative regime [14] Corollary 2. Let ω ∈ R be Diophantine. Then, there is a constant C = C(ω) > 0 such that if 0 < b < C or 4/C < b < ∞ all the spectral gaps of AM are open. σb,ω In the remaining of this section we will sketch the reason why Corollary 1 is an (almost) direct consequence of the following nonperturbative localization result due to Jitomirskaya [10]. AM has Theorem 1. Let ω be Diophantine. Then, if b > 2 the operator Hb,ω,0 only pure point spectrum with exponentially decaying eigenfunctions.
1.2 Sketch of the Proof AM Let b > 2 and ω Diophantine. Jitomirskaya proves that, in this case, Hb,ω,0 has purepoint spectrum with exponentially decaying eigenfunctions. In particular (and this is everything that we will need from her result), there exists AM AM of point eigenvalues of Hb,ω,0 whose eigenvectors are a dense subset in σb,ω exponentially localized. Let a be one of these eigenvalues and ψ = (ψn )n∈Z its exponentially localized eigenvector. We are going to see that a is the endpoint of a noncollapsed spectral gap. From this the Cantor structure of the spectrum follows immediately. AM and ψ ∈ l2 (Z) satisfy the Harper equation By hypothesis a ∈ σb,ω
ψn+1 + ψn−1 + b cos(2πωn)ψn = aψn ,
n∈Z,
with some constants A, β > 0 such that ψn  ≤ A exp(−βn),
n∈Z.
The very special form of the Almost Mathieu operator makes that the Fourier transform of ψ, ˜ ψ(θ) = ψn einθ , θ∈T, n∈Z
84
J. Puig
which is real analytic in Imθ < β, deﬁnes the following quasiperiodic Bloch wave n∈Z, xn = ψ˜ (2πωn + θ) , and that this, satisﬁes the equation (xn+1 + xn−1 ) +
4 2a cos(2πωn + θ)xn = xn , b b
n∈Z.
(2)
for any θ ∈ T. Note that this is again a Harper equation whose parameters have changed 2a 4 , (a, b) → . b b This invariance of the Almost Mathieu operator under Fourier transform is known as Aubry duality [1]. Although the argument above requires the existence of a point eigenvalue, Avron & Simon [2] proved the following form of Aubry duality in terms of the ids 2a AM AM (a) = k4/b,ω . kb,ω b In particular, if we prove that 2a/b is the endpoint of a noncollapsed gap of AM we are done. σ4/b,ω 1.3 Reducibility of QuasiPeriodic Cocycles Our main tool will be to use the dynamics of the eigenvalue equation (2) to prove that a is an endpoint of a noncollapsed gap. To do so, it is convenient to write down (2) as a ﬁrst order system k 2a 4 xn xn+1 b − b cos θn −1 = , θn+1 = θn + 2πω , (3) xn xn−1 1 0 with θn ∈ T. Such ﬁrstorder systems are usually called quasiperiodic skewproducts. The evolution of the vector vn = (xn+1 , xn )T and the angle θn can be seen as the iteration of a quasiperiodic cocycle on SL(2, R) × T AM (v, θ) ∈ R2 × T → AAM , 2a/b,4/b,ω (θ), ω) (v, θ) = A2a/b,4/b,ω (θ)v, θ + 2πω setting AAM 2a/b,4/b,ω (θ)
2a =
b
−
cos θ −1 1 0
2a
4 b
.
That is, 2a vn+1 =
b
−
4 b
cos(2πωn + φ) −1 1 0
···
b
−
cos(φ) −1 1 0
4 b
· v0
Cantor Spectrum for QuasiPeriodic Schr¨ odinger Operators
85
and θn = 2πωn + θ0 . When the frequency ω is rational the skewproduct is periodic and, thanks to Floquet theory, it can be reduced to a skewproduct with constant matrix by means of a periodic transformation. Quasiperiodic reducibility tries to extend this theory to the quasiperiodic case. Let us now introduce some basic notions. Two cocycles (A, ω) and (B, ω) of SL(2, R)×T (not necessarily associated to the Harper equation) are conjugated if there exists a continuous Z : T → SL(2, R) such that A(θ)Z(θ) = Z(θ + 2πω)B(θ),
θ∈T.
In this case the corresponding quasiperiodic skewproducts un+1 = A(θ)un ,
θn+1 = θn + 2πω
vn+1 = B(θ)vn ,
θn+1 = θn + 2πω
and are conjugated through the change u = Zv. Particularly important to our purposes is the case of cocycles which are conjugated to a constant cocycles. A cocycle (A, ω) is reducible to constant coeﬃcients if it is conjugated to a cocycle (B, ω) with B not depending on θ. Remark 3. B is called the Floquet matrix. Neither B nor Z are unique. The fundamental solution of a reducible system Xn (φ) has the following Floquet representation: Xn (φ) = Z(2πnω + φ)B n Z(φ)−1 X0 (φ) .
(4)
In particular, and this is an important observation, if B = I then all solutions of the corresponding skewproduct are quasiperiodic with frequency ω. If the cocycle comes from a Harper’s equation, then all the solutions of this equation are quasiperiodic Bloch waves. Now let us go back to our dual Harper’s equation. In terms of ψ˜ we have that the relation 2a 4 ˜ ˜ ψ(2πω + θ) ψ(4πω + θ) − b cos θ −1 b = ˜ ˜ 1 0 ψ(θ) ψ(2πω + θ) holds for all θ ∈ T. The following Lemma shows that, in this situation, the Almost Mathieu cocycle is reducible to constant coeﬃcients. Lemma 1. Let A : T → SL(2, R) be a real analytic map and ω be Diophantine. Assume that there is a nonzero real analytic map v : T → R2 , such that the relation
86
J. Puig
v(θ + 2πω) = A(θ)v(θ) holds for all θ ∈ T. Then, the quasiperiodic cocycle (A, ω) is reducible to constant coeﬃcients by means of a quasiperiodic transformation which is real analytic. Moreover the Floquet matrix can be chosen to be of the form 1 c B= (5) 0 1 for some c ∈ R. 1.4 End of Proof Classical Floquet theory for periodic Hill’s equation relates endpoints of gaps to the corresponding Floquet matrices (which always exist because the system is periodic). It turns out that, if an Almost Mathieu cocycle is reducible to constant coeﬃcients such characterization also holds. In fact, one can prove that if an Almost Mathieu cocycle (or any other quasiperiodic Schr¨ odinger cocycle), for some a, b, ω ﬁxed, is reducible to constant coeﬃcients with Floquet matrix B then a is at the endpoint of a spectral gap of the operator if, and only if, trace B = ±2. Moreover the gap is collapsed if and, only if, B = ±I. Therefore, if 1 c B= 0 1 then the gap is collapsed if, and only if, c = 0. Summing up, we have that 2a/b AM is a noncollapsed spectral gap of σ4/b,ω if, and only if, the Floquet matrix of the corresponding cocycle, is the identity. Now we can use an adaption of Ince’s argument [9] for the classical Mathieu equation to our case. If B was the identity then, as we learned from Floquet representation (4), there would be two linearly independent real analytic quasiperiodic Bloch waves of Harper’s equation xn+1 + xn−1 +
4 2a cos(2πωn + φ)xn = xn , b b
n∈Z.
Passing to the dual, this would tell us that xn+1 + xn−1 + b cos(2πωn)xn = axn ,
n∈Z.
has two linearly independent solutions in l2 (Z). This is a contradiction with the limitpoint character of the Almost Mathieu operator (or the preservation of the Wronskian for the diﬀerence equation). Therefore B = I (c = 0) so that 2a/b is the endpoint of a noncollapsed AM . Since such endpoints are dense in the spectrum, this must be gap of σ4/b,ω a Cantor set for all b = 0, ±2 and Diophantine frequencies.
Cantor Spectrum for QuasiPeriodic Schr¨ odinger Operators
87
2 Extension to Real Analytic Potentials In the proof of the Ten Martini Problem that we have presented above, there are some features which are speciﬁc of the Almost Mathieu operator. Some other, however, can be extended to more general potentials. Let us try to reproduce the proof for a real analytic potential V : T → R instead of b cos θ. The corresponding Schr¨ odinger operators are of the form (HV,ω,φ x)n = xn+1 + xn−1 + V (2πωn + φ)xn . The dual model of this operator is the following longrange operator, (LV,ω,φ x)n = Vk xn+k + 2 cos(2πωn + φ)xn k∈Z
so that analytic quasiperiodic Bloch waves of HV,ω,φ correspond to exponentially localized eigenvectors of LV,ω,φ . Bourgain & Jitomirskaya [3] proved that, for some ε > 0, LV,ω,φ has purepoint spectrum with exponentially localized eigenfunctions for almost all φ ∈ T if V ρ := sup V (θ) < ε Imθ 0. The proper Poincar´e group in 1+1 dimension can be naturally embedded in the direct product of two copies of A+ (1, R), one for the right movers and one for the left movers. The inﬁnitesimal generators of the right A+ (1, R) group will be denoted HR (the right Hamiltonian) and DR (the right generator of dilations) and they satisfy the commutation relations [DR , HR ] = iHR . The representation of these operators in terms of the creation and annihilation operators is given by dk † a (k)aσR (k) , HR = 2π σR i dk † DR = aσR (k)∂k aσR (k) − ∂k a†σR (k) aσR (k) . 2 2π (16) Their action on ﬁelds is given by [HR , φ(gR , gL )] = −iφ(∂t gR , 0) , [DR , φ(gR , gL )] = iφ(∂t tgR , 0) , and in the exponentiated form by eisHR φ(gR , gL )e−isHR = φ(gR (· − s), gL eisDR φ(gR , gL )e−isDR = φ(e−s gR (e−s ·), gL ) . For (a, b) ∈ A+ (1, R) we set ra,b g(t) := a−1 g(a−1 (t − b)) and RR (a, b) = eilnaDR eibHR . RR is a unitary representation of A+ (1, R), which acts naturally on the ﬁelds: RR (a, b)φ(gR , gL )RR (a, b)† = φ(ra,b gR , gL ) .
(17)
Note, however, that ra,b does not preserve the indeﬁnite scalar product (9) unless we impose the constraint gˆ(0)=0: g1∗ (0)ˆ g2 (0) . (ra,b g1 ra,b g2 ) = (g1 g2 ) − lnaˆ Similarly we introduce the left Hamiltonian HL and the left generator of dilations DL satisfying analogous commutation relations and the representation of the left A+ (1, R).
112
J. Derezi´ nski and K.A. Meissner
The Poincar´e group generators are the Hamiltonian H = HR + HL , the momentum P = HR − HL and the boost operator Λ = DR − DL (the only Lorentz generator in 1+1 dimensions). The elements of the Poincar´e group are of the form (a, bR ), (a−1 , bL ) ∈ A+ (1, R) × A+ (1, R) . The scalar product (gR1 gR2 ) + (gL1 gL2 ) is invariant wrt the proper Poincar´e group.
4 Changing the Compensating Functions It is important to discuss the dependence of the whole construction on the choice of compensating functions σR and σL . ˜L be another pair of real functions satisfying (11). Set Let σ ˜R and σ ˜R (x)−σR (x), ξL (x) := σ ˜L (x)−σL (x). Note that ξˆR (0) = ξˆL (0) = 0. ξR (x) := σ Deﬁne dk ˆ∗ iχξR (k)aR (k) + iχξˆR (k)a†R (k) U (ξR , ξL ) = exp 2πk 1 ∗ ∗ (k)ˆ σR (k) − ξˆR (k)ˆ σR (k))+ R → L . (18) + χ2 (ξˆR 2 Using the formula 1 eA Be−A = B + [A, B] + [A, [A, B]] + · · · (19) 2 we have U aR (k)U −1 = aR (k) − iχξˆR (k) , ∗ U a†R (k)U −1 = a†R (k) + iχξˆR (k) , −1 ˆ U aL (k)U = aL (k) − iχξL (k) , ∗ U a†L (k)U −1 = a†L (k) + iχξˆL (k) , −1 U χU = χ
and U pU
−1
(20)
dk ˆ∗ − ξR (k)aR (k) − ξˆR (k)a†R (k) 2πk ∗ ∗ −iχξˆR (k)ˆ σR (k) + iχξˆR (k)ˆ σR (k) +R → L) .
= p+
(21)
Using these relations we get for example U φσ (gR , gL )U −1 = φσ˜ (gR , gL ) , U a†σR U −1 = a†σ˜ R , U HσR U −1 = Hσ˜ R ,
(22)
˜ . Thus the where we made explicit the dependence of φ and HR on σ and σ two constructions – with σ and with σ ˜ – are unitarily equivalent.
Quantum Massless Field in 1+1 Dimensions
113
5 Hilbert Space The Hilbert space of the system is the product of three spaces: H = HR ⊗ HL ⊗ H0 . HR is the bosonic Fock space spanned by the creation operators a†R (k) acting on the vacuum vector ΩR . Analogously HL is the bosonic Fock space spanned by the creation operators a†L (k) acting on the vacuum vector ΩL . With the third sector H0 we have essentially two options. If we take the usual choice H0 = L2 (R, dχ) then we can deﬁne the vacuum state (vacuum expectation value) but there does not exist a vacuum vector. On the other hand, we can take H0 = l2 (R), i.e. the space with the scalar product f ∗ (χ)g(χ) , (23) (f g) = χ∈R
which is a nonseparable space. It may sound as a nonstandard choice, it has however the advantage of possessing a vacuum vector. The orthonormal basis in the latter space consists of the Kronecker delta functions δχ for each χ ∈ R. In the nonseparable case, the operator p, and therefore also φ(gR , gL ), cannot be deﬁned. But there exist operators eisp , for s ∈ R, and also eiφ(gR ,gL ) . The commutation relations for these exponential operators follow from the commutation relations for p and φ(gR , gL ) described above. In such a space the vacuum vector is given by Ω = ΩR ⊗ ΩL ⊗ δ0 .
(24)
This vector is invariant under the action of the Poincar´e group and the action of the gauge group U . We now prove that it is the unique vector with the lowest energy. Note ﬁrst that H is diagonal in χ ∈ R. Now for an arbitrary Φ ∈ HR ⊗ HL and χ1 ∈ R,
Φ ⊗ δχ1 HΦ ⊗ δχ1 5 6 dk ∗ Φ a†R (k) + iχ1 σ + R → L (25) ˆR (k) aR (k) − iχ1 σ ˆR (k) Φ = 2π For any χ1 , the expression (25) is nonnegative. If χ1 = 0, it has a unique ground state ΩR ⊗ ΩL . If χ1 = 0, then (25) has no ground state. In fact, it is well known that a ground state of a quadratic Hamiltonian is a coherent state, that is given by a vector of the form dk βR , βL = C exp βR (k)a†R (k) + βL (k)a†L (k) ΩR ⊗ ΩL (26) 2πk and C is the normalizing constant dk 1 2 2 βR (k) + βL (k) C = exp − . 2 2πk
(27)
114
J. Derezi´ nski and K.A. Meissner
If we set Φ = βR , βL in (25), then we obtain 2
Φ ⊗ δχ1 HΦ ⊗ δχ1 = βR (k) + iχ1 σR (k) + R → L . that takes the minimum for βR (k) = −iχ1 σ ˆR (k),
βL (k) = −iχ1 σ ˆL (k) .
But for χ1 = 0, βR , βL is not well deﬁned as a vector in the Hilbert space. To see this we can note that the normalizing constant C equals zero, because then dk ˆ σR (k)2 + ˆ σL (k)2 = ∞ . χ21 2πk (The fact that operators of the form (25) have no ground state is well known in the literature, see eg. [7]). In the nonseparable case H0 = l2 (R), the expectation value
Ω · Ω =: ω(·) is a Poincar´einvariant state (positive linear functional) on the algebra of observables. If we take the separable case H0 = L2 (R, dχ), the state ω can also be given a meaning, even though the vector Ω does not exist (since then δ0 is not well deﬁned). Note that in the nonseparable case the state ω can act on an arbitrary bounded operator on H. In the separable case we have to restrict ω to a smaller algebra of operators, say, the algebra (or the C ∗ algebra) spanned by the operators of the form eiφ(gR ,gL ) . The expectation values of the exponentials of the 1+1dimensional massless ﬁeld make sense and can be computed, both in the separable and nonseparable case: 1 dk ˆ gR (k)2 + ˆ gL (k)2 ω (exp (iφ(gR , gL ))) = exp − . (28) 2 2πk Note that the integral in the exponent of (28) is the usual integral of a positive function, and not its regularization as in (9). Therefore, if gˆ(0) = 0, then this integral equals +∞ and (28) equals zero. The “twopoint functions” of massless ﬁelds in 1+1 dimension, even smeared out ones, are not well deﬁned. Formally, they are introduced as ω (φ(gR,1 , gL,1 )φ(gR,2 , gL,2 )) .
(29)
If we use the nonseparable H0 = l2 (R), then ﬁeld operators φ(gR , gL ) is not well deﬁned if gˆ(0) = 0, and thus (29) is not deﬁned. If we use the separable Hilbert space H0 = L2 (R, dχ), then φ(gR,1 , gL,1 )φ(gR,2 , gL,2 ) are unbounded operators and there is no reason why the state ω could act on them. Thus (29) a priori does not make sense. In the usual free quantum ﬁeld theory,
Quantum Massless Field in 1+1 Dimensions
115
if mass is positive or dimension more than 2, the expectation values the exponentials depend on the smeared ﬁelds analytically, and by taking their second derivative at (gR , gL ) = (0, 0), one can introduce the 2point function. This is not the case for massless ﬁeld in 1+1 dimension. The above discussion shows that the problem of the nonpositive definiteness of the twopoint function, so extensively discussed in the literature [4, 9, 11–13] does not exist in our formalism. It should be noted that massless ﬁelds in 1+1 dimension do not satisfy the Wightman axioms [14]. In the separable case there is no vacuum vector in the Hilbert space; in the nonseparable case there is a vacuum vector, but there are no ﬁelds φ(f ), only the “Weyl operators” eiφ(f ) .
6 Fields in Position Representation So far in our discussion we found it convenient to use the momentum representation. The position representation is, however, better suited for many purposes. Let W(t) denote the Fourier transform of the appropriately regularized distribution θ(k) 2πk , that is dk ikt 1 lim e + ln( /µ) W(t) = (30) 2π →0 k> k 1 iπ = −γE − lnµt + sgn(t) = W ∗ (−t) 2π 2 where γE is the Euler’s constant. We can rewrite (9) as ∞ (g1 g2 ) = dtdsg1∗ (t)W(t − s)g2 (s)
(31)
−∞
To describe massless ﬁeld in the position representation we introduce the operators ψR (t) deﬁned as dk † a (k)e−ikt + aR (k)eikt , ψR (t) = 2πk R and similarly for R → L. Note that it is allowed to smear ψR (t) and ψL (t) only with test functions satisfying g(t)dt = 0 . Note that W(t−s) and 2i sgn(t−s) are the correlator and the commutator functions for ψR (t):
116
J. Derezi´ nski and K.A. Meissner
ΩR ψR (t)ψR (s)ΩR = W(t − s) , [ψR (t), ψR (s)] =
i sgn(t − s) , 2
and similarly for R → L. We introduce also ψσR (t) =
dk † dk −ikt a (k)e aσR (k)eikt + 2πk σR 2πk χ = ψR (t) + dsσR (s)sgn(t − s) , 2
as well as R → L. It is perhaps useful to note that formally we can write ψσR (t) = YR ψR (t)YR† , where
YR := exp iχ
dtσR (t)ψR (t)
Note that YR is not a well deﬁned operator, since σ ˆR (0) = 0. Expressed in position representation the ﬁelds are given by φ(gR , gL ) = dt(gR (t) − gˆ(0)σR (t))ψσR (t) + R → L + gˆ(0)p .
(32)
Since (gR (t) − gˆ(0)σR (t))dt = 0 the whole expression is well deﬁned. The commutator of two ﬁelds equals [φ(f1 ), φ(f2 )] = i dt1 dt2 dx1 dx2 f1 (t1 , x1 )f2 (t2 , x2 ) × (sgn(t1 −t2 +x1 −x2 )+sgn(t1 −t2 −x1 +x2 )) Note that the commutator of ﬁelds is causal – it vanishes if the supports of f1 and f2 are spatially separated.
7 The SL(2, R) × SL(2, R) Covariance Massless ﬁelds in 1+1 dimension satisfying the constraint (2) actually possess much bigger symmetry than just the A+ (1, R)×A+ (1, R) symmetry, they are covariant wrt the action of SL(2, R) × SL(2, R) (for right and left movers). We will restrict ourselves to the action of SL(2, R) for, say, right movers. First we consider it on the level of test functions. We assume that test functions satisfy gˆ(0) = 0 and g(t) = O(1/t2 ),
t → ∞ .
(33)
Quantum Massless Field in 1+1 Dimensions
Let C=
ab ∈ SL(2, R) cd
(i.e. ad − bc = 1). We deﬁne the action of C on g by dt − b −2 (rC g)(t) = (−ct + a) g . −ct + a
117
(34)
(35)
Note that (35) preserves (33) and the scalar product (rC g1 rC g2 ) = (g1 g2 ) , and is a representation, that is rC1 rC2 = rC1 C2 . We second quantize rC by introducing the unitary operator RR (C) on HR ﬁxed uniquely by the conditions RR (C)ΩR = ΩR , at + b RR (C)ψR (t)RR (C)† = ψR . ct + d Note that RR (C)
† dtgR (t)ψR (t) RR (C) = dt(rC gR )(t)ψR (t).
(36)
(37)
C → RR (C) is a representation in HR . Thus the operators RR (C) act naturally on ﬁelds satisfying (2) (and hence also gˆR (0) = 0). The ﬁelds without the constraint (2) are not covariant with respect to SL(2, R) × SL(2, R), since this symmetry fails even at the classical level. What remains is the A+ (1, R) × A+ (1, R) symmetry described in (17). Note that A+ (1, R) can be viewed as a subgroup of SL(2, R): 1/2 −1/2 a ba A+ (1, R) (a, b) → C ∈ SL(2, R) . (38) 0 a−1/2 Clearly, on the restricted Hilbert space, under the identiﬁcation (38), RR (a, b) coincides with RR (C).
8 Normal Ordering In the theory without the compensating sector the normal ordering can be introduced in a standard way. In particular we have : eiφ(gR ,gL ) : = e 2 (gR gR )+ 2 (gL gL ) eiφ(gR ,gL ) 1
1
(39)
If the compensating sector is present then the theory does not act in the Fock space any longer and we do not have an invariant particle number
118
J. Derezi´ nski and K.A. Meissner
operator. It is however possible (and useful) to introduce the notion of the normal ordering. For Weyl operators it is by deﬁnition given by (39). For an arbitrary operator, we ﬁrst decompose it in terms of Weyl operators, and then we apply (39). Note that our deﬁnition has an invariant meaning wrt the change of the compensating function: in the notation of (22) we have U : eıφσ (gR ,gL ) : U −1 =: eıφσ˜ (gR ,gL ) : . Normal ordering is Poincar´e invariant but suﬀers anomalies under remaining A+ (1, R) × A+ (1, R) transformations (because the prefactor on the rhs of (39) is invariant only under the Poincar´e group). If the constraint (2) is satisﬁed, then normal ordering is SL(2, R) × SL(2, R) covariant.
9 Classical Fields In order to better understand massless quantum ﬁelds in 1+1 dimension it is useful to study the underlying classical system, that is the wave equation in 1+1 dimension (4). From the general representation of any classical solution φ(t, x) = φR (t − x) + φL (t + x)
(40)
we get (in notation where f (±∞) stands for limt→±∞ f (t)) φ(t, ∞) + φ(t, −∞) = φR (−∞) + φL (∞) + φR (∞) + φL (−∞) = φ(∞, x) + φ(−∞, x)
(41)
It will be convenient to denote by the space of Schwartz functions on R by S and by ∂0−1 S the space of functions whose derivatives belong to S and satisfy the condition f (∞) = −f (−∞). We are interested only in those solutions that restricted to lines of constant time and lines of constant position belong to ∂0−1 S (we will denote them as F11 ). Neglecting a possible global constant shift we therefore assume that they satisfy φ(t, ∞) + φ(t, −∞) = φ(∞, x) + φ(−∞, x) = 0
(42)
F11 is characterized by two numbers lim φ(t, x) = − lim φ(t, x) =: c0 ,
t→∞
t→−∞
lim φ(t, x) = − lim φ(t, x) =: c1 .
x→∞
x→−∞
It is natural to distinguish the following subclasses of solutions to (4): – F00 – solutions that restricted to lines of constant time and to lines of constant position belong to S i.e. c0 = c1 = 0.
Quantum Massless Field in 1+1 Dimensions
119
– F10 – solutions that restricted to lines of constant time belong to S and restricted to lines of constant position belong to ∂0−1 S i.e. c1 = 0. – F01 – solutions that restricted to lines of constant position belong to S and restricted to lines of constant time belong to ∂0−1 S i.e. c0 = 0. There are several useful ways to parametrize elements of F11 . 1. Initial conditions at t = 0: f0 (x) = φ(0, x) , f1 (x) := ∂t φ(0, x) . Here, f0 ∈ ∂0−1 S, f1 ∈ S. Note that 1 c0 = f1 (x)dx, 2
(43)
c1 = f0 (∞) .
2. Derivatives of right/left movers: 1 1 gR (t) := − f0 (−t) + f1 (−t) , 2 2 1 1 gL (t) := f0 (t) + f1 (t) . 2 2 Note that gR , gL ∈ S and they satisfy gL (t)dt = c0 + c1 . gR (t)dt = c0 − c1 ,
(44)
3. Right/left movers: 1 φR (t) = gR (t − u)sgn(u)du 2 1 gL (t − u)sgn(u)du . φL (t) = 2
(45)
Note that φR , φL ∈ ∂0−1 S and they satisfy 1 (c0 − c1 ) , 2 1 φL (∞) = −φL (−∞) = (c0 + c1 ) . 2
φR (∞) = −φR (−∞) =
We can go back from (gR , gL ) to (f0 , f1 ) by 1 1 f0 (x) = gR (s − x)sgn(−s)ds + gR (s + x)sgn(−s)ds , 2 2 f1 (x) = gR (−x) + gL (x) . We can go back from (φR , φL ) to (gR , gL ) by
(46)
120
J. Derezi´ nski and K.A. Meissner
gR = φR ,
gL = φL .
The unique solution of (4) with the initial conditions (43) equals φ(t, x) = φR (t − x) + φL (t + x) . It will be sometimes denoted by φ(gR , gL ). In the literature, one can ﬁnd all three parametrizations of solutions of the wave equation. In particular, note that 3. is especially useful in the case of F00 , since then φR , φL ∈ S. Note that in our paper we use 2. as the standard parametrization of solutions of the wave equation. We are interested primarily in the space F10 . Note that F10 are the solutions to the wave equation with f0 , f1 ∈ S. Equivalently, for F10 , the functions gR , gL satisfy gR (t)dt = gL (t)dt . (47) We equip the space F11 with the Poisson bracket, which we write for all three parametrizations: {φ(gR1 , gL1 ), φ(gR2 , gL2 )} = f01 (x)f12 (x)dx − f02 (x)f11 (x)dx (48) = gR1 (t)sgn(s − t)gR1 (s)dtds + gL1 (t)sgn(s − t)gL1 (s)dtds = Im(gR1 gR2 ) + Im(gL1 gL2 ) 1 1 = ∂t φR1 (t)φR2 (t)dt + ∂t φL1 (t)φL2 (t)dt . 2 2
(49) (50)
Above, (f0i , f1i ) and (φRi , φLi ) correspond to (gRi , gLi ). The formula in (48) is the usual Poisson bracket for the space of solutions of relativistic 2nd order equations (both wave and KleinGordon equations). (49) we have already seen in (15). The Poisson bracket in F11 is invariant wrt to the conformal group ﬁxing the inﬁnities preserving separately the orientation of right and left movers, that is Diﬀ + (R) × Diﬀ + (R). In the case of F00 we can extend this action to the full orientation preserving conformal group, that is Diﬀ + (S 1 )×Diﬀ + (S 1 ), where we identify R together with the point at inﬁnity with the unit circle.
10 Algebraic Approach Among mathematical physicists, it is popular to use the formalism of C ∗ algebras to describe quantum systems. A description of massless ﬁelds in 1+1 dimension within this formalism is sketched in this section.
Quantum Massless Field in 1+1 Dimensions
121
To quantize the space F11 , we consider formal expressions eiφ(gR ,gL )
(51)
equipped with the relations eiφ(gR1 ,gL1 ) eiφ(gR2 ,gL2 ) = eiIm(gR1 gR2 )+iIm(gL1 gL2 ) eiφ(gR1 +gR2 ,gL1 +gL2 ) ; (eiφ(gR ,gL ) )† = eiφ(−gR ,−gL ) . Linear combinations of (51) form a ∗algebra, which we will denote Weyl (F11 ). (If we want, we can take its completion in the natural norm and obtain a C ∗ algebra). Note that the group Diﬀ + (R)×Diﬀ + (R) acts on Weyl (F11 ) by ∗automorphisms. In other words, we have two actions Diﬀ + (R) F → αR (F ) ∈ Aut(Weyl(F11 )) , Diﬀ + (R) F → αL (F ) ∈ Aut(Weyl(F11 )) , commuting with one another given by αR (F ) eiφ(gR ,gL ) = eiφ(rF gR ,gL ) , αL (F ) eiφ(gR ,gL ) = eiφ(gR ,rF gL ) .
(52)
Above, Aut(Weyl(F11 )) denotes the group of ∗automorphisms of the algebra Weyl(F)) and rF g(t) := F 1(t) g(F −1 (t)). Similarly Diﬀ + (S 1 ) × Diﬀ + (S 1 ) acts on Weyl(F00 ) by ∗automorphisms. The state ω given by (28) is invariant wrt A+ (1, R) × A+ (1, R) on Weyl(F11 ) and wrt SL(2, R) × SL(2, R) on Weyl(F00 ). In our paper we restricted ourselves to Weyl(F10 ). The constructions presented in this paper give representations of Weyl(F10 ) in a Hilbert space H and two commuting with one another strongly continuous unitary representations A+ (1, R) (a, b) → RR (a, b) ∈ U (H) , A+ (1, R) (a, b) → RL (a, b) ∈ U (H) . implementing the automorphisms (52): αR (a, b)(A) = RR (a, b)ARR (a, b)† , αL (a, b)(A) = RL (a, b)ARL (a, b)† .
(53) (54)
In the case of the algebra Weyl(F00 ) the same is true for SL(2, R). In Sect. 5 we described two representations that satisfy the above mentioned conditions. The ﬁrst, call it πI , represents Weyl(F10 ) in a separable Hilbert space. Its drawback is the absence of a vacuum vector – a Poincar´e invariant vector. The second, call it πII , represents Weyl(F10 ) in a nonseparable Hilbert space. It has an invariant vector Ω.
122
J. Derezi´ nski and K.A. Meissner
We can perform the GNS construction with ω. As a result we obtain the representation πII together with the cyclic invariant vector Ω. The description of this construction for massless ﬁelds in 1+1 dimension can be found in [1], Sect. III D, and [2] Sect. 4. Note, however, that we have not seen the representation πI in the literature, even though one can argue that it is in some ways superior to πII . Let us make a remark concerning the role played by the functions (σR , σL ). We note that F00 is a subspace of F10 of codimension 1. Fixing (σR , σL ) satisfying (11) allows us to identify F10 with F00 ⊕ R. Thus any (gR , gL ) g (0)σR , gL −ˆ g (0)σL ) satisfying (47) is decomposed into the direct sum of (gR −ˆ and gˆ(0)(σR , σL ). Of course, similar constructions can be performed for the algebra Weyl(F11 ) or Weyl(F01 ). In the literature, algebras of observables based on F01 appear in the context of “DoplicherHaagRoberts charged sectors” in [5, 6, 15].
11 Vertex Operators Finally, let us make some comments about the socalled vertex operators, often used in string theory [8]. Let δy denote the delta function at y ∈ R. Let tR1 , . . . , tRn ∈ R correspond to insertions for right movers and tL1 , . . . , tLm ∈ R correspond to insertions for left movers. Suppose that the complex numbers βR1 , . . . , βRn , and βL1 , . . . , βLm denote the corresponding insertion amplitudes and satisfy βRi = βLj . Then the corresponding vertex operator is formally deﬁned as V (tR1 , βR1 ; . . . ; tRn , βRn ; tL1 , βL1 ; . . . ; tLm , βLm ) = : exp (iφ(gR , gL )) : , (55) where gR = βR1 δtR1 + · · · + βRn δtRn ,
(56)
gL = βL1 δtL1 + · · · + βLm δtLm .
(57)
Strictly speaking, the rhs of (55) does not make sense as an operator in the Hilbert space. In fact, in order that eiφ(gR ,gL ) be a well deﬁned operator, we need that dk dk 2 2 ˆ gR (k) − gˆ(0)ˆ ˆ gL (k) − gˆ(0)ˆ σR (k) + σL (k) < ∞. (58) 2πk 2πk This is not satisﬁed if gR or gL are as in (56) and (57). Nevertheless, proceeding formally, we can deduce various identities. For instance, we have the Poincar´e covariance:
Quantum Massless Field in 1+1 Dimensions
123
RR (a, bR )RL (a−1 , bL ) × V (tR1 , βR1 ; . . . ; tRn , βRn ; tL1 , βL1 ; . . . ) † † × RL (a−1 , bL )RR (a, bR )
= V (atR1 + bL , βR1 ; . . . ; a−1 tRn + bR , βRn ; a−1 tL1 + bL , βL1 ; . . . ) . " If in addition βRi = 0, then a similar identity is true for SL(2, R) × SL(2, R). Clearly, we have ω (V (tR1 , βR1 ; . . . ; tRn , βRn ; tL1 , βL1 ; . . . ; tLm , βLm )) '
"
1,
=
"
0,
βRi = 0; βRi = 0.
.
(59)
The following identities are often used in string theory for the calculation of onshell amplitudes. Suppose that tR1 , . . . , tRn are distinct, and the same is true for tL1 , . . . , tLn . Then, using (30), we obtain V (tR1 , βR1 ; tL1 , βL1 ) · · · V (tRn , βRn ; tLn , βLn )
=e
"
W(tRi −tRj ))βRi βRj +W(tLi −tLj )βLi βLj
i<j
×V (tR1 , βR1 ; . . . ; tRn , βRn ; tL1 , βL1 ; . . . ; tLn , βLn ) =
tRi − tRj −βRi βRj /2π tLi − tLj −βLi βLj /2π i<j
iµeγE
iµeγE
×V (tR1 , βR1 ; . . . ; tRn , βRn ; tL1 , βL1 ; . . . ; tLn , βLn ) .
12 Fermions Massless fermions in 1+1 dimension do not pose as bosons. such problems λR (t, x) The ﬁelds are spinors, they will be written as . They satisfy the λL (t, x) Dirac equation ∂t − ∂ x λR (t, x) 0 =0. 0 ∂t + ∂x λL (t, x) We will also use the ﬁelds smeared with real functions f , where the condition (2) is not needed any more: λR (t, x) λR (f ) = f (t, x)dtdx . λL (f ) λL (t, x)
124
J. Derezi´ nski and K.A. Meissner
Because of the Dirac equation, they can be written as λR (f ) = λR (gR ), λL (f ) = λL (gL ) , where gR and gL where introduced when we discussed bosons. For k > 0, we introduce fermionic operators (for right and left sectors) bR (k) and bL (k) satisfying the anticommutation relations {bR (k), b†R (k )} = 2πδ(k − k ) , {bL (k), b†L (k )} = 2πδ(k − k ), with all other anticommutators vanishing. Now dk ∗ λR (gR ) = gR (k)bR (k) + gR (k)b†R (k) 2π dk ∗ gL (k)bL (k) + gL (k)b†L (k) λL (gL ) = 2π The anticommutation relations for the smeared ﬁelds read dk ∗ ∗ gˆ (k)ˆ (t)gR2 (t)dt = gR2 (k) {λ†R (gR1 ), λR (gR2 )} = gR1 2π R1
(60)
(61)
(62)
and similarly for the left sector. Note the diﬀerence of the fermionic scalar product (62) and the bosonic one (··). In terms of spacetime smearing functions these anticommutation relations read † {λR (f1 ), λR (fR )} = 2 dtdxδ(t + x)f1 (t, x)f2 (t, x) , {λ†L (f1 ), λR (fL )} = 2 dtdxδ(t − x)f1 (t, x)f2 (t, x) . Fermionic ﬁelds are covariant with respect to the group A+ (1, R) × A+ (1, R). We will restrict ourselves to discussing the covariance for say, right movers. The right Hamiltonian and the right dilation generator are dk † f kb (k)bR (k) HR = 2π R dk † i f DR bR (k)k∂k bR (k) − k∂k b†R (k) bR (k) . = 2 2π f f We have the usual commutation relations for HR and DR and their action on the ﬁelds is anomalyfree: % f & HR , λR (gR ) = −iλ(∂t gR ) , % f & DR , λR (gR ) = iλ (t∂t + 1/2)gR .
Quantum Massless Field in 1+1 Dimensions
125
We have also the covariance with respect to the conformal group SL(2, R)× SL(2, R). We need to assume that test functions satisfy t → ∞ .
g(t) = O(1/t), We deﬁne the action of
(63)
ab C= ∈ SL(2, R) cd
on g by f (rC g)(t)
−1
= (−ct + a)
g
(64)
dt − b −ct + a
.
(65)
Note that (65) has a diﬀerent power than (35). It is a unitary representation for the scalar product (··)f . f on the fermionic Fock space by introducing the We second quantize rC f unitary operator RR (C) ﬁxed uniquely by the conditions f RR (C)ΩR = ΩR , f f (C)λR (t)RR (C)† = (ct + d)−1 λR RR
at + b ct + d
.
(66)
f (C) is a unitary representation and it acts naturally on Note that C → RR ﬁelds: f f f RR (C)λR (gR )RR (C)† = λR (rC gR ),
(67)
13 Supersymmetry In this section we consider both bosons and fermions. Thus our Hilbert space is the tensor product of the bosonic and fermionic part. We assume that the bosonic and fermionic operators commute with one another. Clearly, our theory is A+ (1, R) × A+ (1, R) covariant. In fact, the right Hamiltonian and f and the generator of dilations for the combined theory are equal to HR + HR f DR + DR . In the case of the theory with the constraint (2), we have also the SL(2, R) × SL(2, R). covariance. On top of that, the combined theory is supersymmetric. The supersymmetry generators QR , QL are deﬁned as dk † aσR (k)bR (k) + aσR (k)b†R (k) , QR = 2π dk † QL = (68) aσL (k)bL (k) + aσL (k)b†L (k) . 2π
126
J. Derezi´ nski and K.A. Meissner
They satisfy the basic supersymmetry algebra relations without the central charge f {QR , QR } = 2(HR + HR ), f {QL , QL } = 2(HL + HL ) ,
{QR , QL } = 0.
(69)
The action of the supersymmetric charge transforms bosons into fermions and vice versa: [QR , φ(gR , gL )] = λR (gR ) , [QL , φ(gR , gL )] = λL (gL ) , [QR , λR (gR )] = φ(∂t gR , 0) , [QL , λL (gL )] = φ(0, ∂t gL ) . QR The pair of operators behaves like a spinor under the Poincar´e QL group. Even more is true: we have the covariance under the group A+ (1, R) × A+ (1, R), which for the right movers can be expressed in terms of the following commutation relations: [HR , QR ] = 0 ,
i [DR , QR ] = − QR . 2
Acknowledgement J.D. was partly supported by the European Postdoctoral Training Program HPRNCT20020277, the Polish KBN grant SPUB127 and 2 P03A 027 25. K.A.M. was partially supported by the Polish KBN grant 2P03B 001 25 and the European Programme HPRN–CT–2000–00152. J.D. would like to thank S. DeBi`evre, C. G´erard and C. J¨ akel for useful discussions.
References 1. Acerbi, F., Morchio, G., Strocchi, F.: Infrared singular ﬁelds and nonregular representations of canonical commutation relation algebras, Journ. Math. Phys. 34 (1993) 899–914 2. Acerbi, F., Morchio, G., Strocchi, F.: Theta vacua, charge conﬁnement and charged sectors from nonregular representations of CCR algebras, Lett. Math. Phys. 27 (1993) 1–11. 3. Brattelli, O., Robinson D. W.: Operator Algebras and Quantum Statistical Mechanics, Volume 2, SpringerVerlag, Berlin, second edition 1996.
Quantum Massless Field in 1+1 Dimensions
127
4. S. De Bi`evre and J. Renaud: A conformally covariant quantum ﬁeld in 1+1 dimension, J. Phys. A34 (2001) 10901–10919. 5. Buchholz, D.: Quarks, gluons, colour: facts or ﬁction?, Nucl. Phys. B469 (1996) 333–356. 6. D. Buchholz and R. Verch, Scaling algebras and renormalization group in algebraic quantum ﬁeld theory. II. Instructive examples, Rev. Math. Phys. 10 (1998) 775–800. 7. J. Derezi´ nski: Van Hove Hamiltonians – exactly solvable models of the infrared and ultraviolet problem, Ann. H. Poincar´e 4 (2003) 713–738. 8. M.B. Green, J.H. Schwarz and E. Witten: Superstring theory, Cambridge Univ. Press, Cambridge 1987. 9. G.W. Greenberg, J.K. Kang and C.H. Woo: Infrared regularization of the massless scalar free ﬁeld in twodimensional spacetime via Lorentz expansion, Phys. Lett. 71B (1977) 363–366. 10. C. Itzykson and J.B. Zuber: Quantum Field Theory, McGrawHill, 1980, Chap. 11. 11. G. Morchio, D. Pierotti and F. Strocchi: Infrared and vacuum structure in twodimensional local quantum ﬁeld theory models. The massless scalar ﬁeld, Journ. Math. Phys. 31 (1990) 1467–1477. 12. N. Nakanishi: Free massless scalar ﬁeld in twodimensional spacetime, Prog. Theor. Phys. 57 (1977) 269–278. 13. N. Nakanishi: Free massless scalar ﬁeld in twodimensional spacetime: revisited, Z. Physik C. Particles and Fields 4 (1980) 17–25. 14. R.F. Streater and A.S. Wightman: PCT, spin and statistics and all that, W.A.Benjamin, New YorkAmsterdam 1964. 15. R.F. Streater and I.F. Wilde: Fermion states of a boson ﬁeld, Nucl. Phys. B24 (1970) 561–575.
Stability of MultiPhase Equilibria Marco Merkli1,2 1
2
Department of Mathematics and Statistics, McGill University, 805 Sherbrooke W., Montreal QC, Canada, H3A 2K6 [emailprotected] Centre des Recherches Math´ematiques, Universit´e de Montr´eal, Succursale centreville, Montr´eal, QC, Canada, H3C 3J7
1 Stability of a SinglePhase Equilibrium We expect equilibrium states of very large systems to have a property of dynamical stability, called the property of return to equilibrium. This means that if the system is initially in a state that diﬀers only a bit (say locally in space) from an equilibrium state, then it approaches that equilibrium state in the large time limit. One may view this irreversible process as a consequence of the dispersiveness of the dynamics, and the inﬁnite size of the system: the local spatial disturbance of the equilibrium state, deﬁning the initial state, propagates out of any bounded region if one waits long enough. Strictly speaking we expect return to equilibrium only for spatially inﬁnitely extended systems. If one desires to observe localized events of a large, but ﬁnite systems (a laboratory) then the above description is a good approximation on an intermediate time scale (time should be large so the system can settle towards the equilibrium state, but not too large as to avoid recurrences in the ﬁnite system). A quantum system for which the property of return to equilibrium is easy to examine is the free Bose gas. If the gas has a BoseEinstein condensate then the system has many coexisting phases (many equilibrium states at the same temperature). We will explain what the property of return to equilibrium translates into for systems with multiple (multiphase) equilibria. If a free Bose gas, modeling an “environment” or a “heat bath”, is coupled to a ﬁnite system, e.g. to a spin, then it is not so easy any more to show the property of return to equilibrium. This question has been examined, for singlephase equilibria, in a variety of recent publications, [3, 6, 10, 14, 15, 27]. The case when the gas is in a condensate state is investigated in [18]. We refer to [11] for a discussion of the above mentioned validity of the inﬁnite volume approximation of ﬁnite systems for intermediate times. 1.1 The Free Bose Gas Let Λ = [−L/2, L/2]3 ⊂ R3 be a box in physical space. Pure states of the bose gas localized inside Λ are represented by vectors in the bosonic Fock M. Merkli: Stability of MultiPhase Equilibria, Lect. Notes Phys. 690, 129–148 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
130
M. Merkli
space FΛ =
∞ 7
P+ L2 (Λ, d3 x)⊗n ,
(1)
n=0
where P+ projects onto functions which are symmetric under permutations of arguments. In order to describe an inﬁnitely extended Bose gas we want to increase the volume Λ ↑ R3 . One may replace Λ by R3 in expression (1) thereby getting a Hilbert space F whose vectors represent states (where particles are not constrained to any bounded volume). If the system is in the state ψ = {ψn }∞ n=0 (a normalized vector in F) then the probability of ﬁnding n particles is given by pn = ψn 2L2 (R3n ,d3n x) . Since these probabilities must add up to one we have pn → 0, as n → ∞. This indicates that if we would like to describe an inﬁnitely extended Bose gas with a ﬁxed (average) density, say one particle per unit volume, then Fock space cannot be the right state space. In order to describe the inﬁnite system we should take the inﬁnite volume limit of the expectation functional (“averages”) deﬁned by a vector, or a density matrix on FΛ . To this end it is useful to introduce the Weyl algebra WΛ = W(L2 (Λ, d3 x)). This is the C ∗ algebra generated by the unitary Weyl operators, (2) W (f ) = eiϕ(f ) , f ∈ L2 (Λ, d3 x), where ϕ(f ) = √12 (a∗ (f ) + a(f )), and a∗ (f ), a(f ) are the usual creation and annihilation operators on FΛ . The Weyl algebra provides us with a rich class of “observables”. Namely, W(L2 (Λ, d3 x)) is dense (in the weak operator topology) in B(FΛ ), the set of all bounded operators on FΛ . (This is a consequence of the fact that the Weyl algebra acts irreducibly on Fock space.) The Weyl operators satisfy the Canonical Commutation Relations (CCR) i (3) W (f )W (g) = e− 2 Imf,g W (f + g) . A state of the system is given by a positive linear functional ωΛ : WΛ → C, normalized as ωΛ (1l) = 1. In view of (3) it is not very surprising that any state on the Weyl algebra is entirely determined by its value on the generators of the algebra, i.e., by the (nonlinear) expectation functional L2 (Λ, d3 x) f → EΛ (f ) := ωΛ (W (f )) .
(4)
The converse is true too: if EΛ : L (Λ, d x) → C is a functional satisfying certain compatibility conditions, then it deﬁnes uniquely a state ωΛ on WΛ , via ωΛ (W (f )) = EΛ (f ). The dynamics of Weyl operators is given by 2
3
t → αtΛ (W (f )) = W (e−itHΛ f ) ,
(5)
where the one particle Hamiltonian HΛ is −∆ (Laplace operator acting on L2 (Λ, d3 x) with some classical boundary condition, or a function of this operator).
Stability of MultiPhase Equilibria
131
The procedure of ﬁnding (deﬁning!) the inﬁnitely extended system is now reduced to ﬁnding the limiting expectation functional E = limΛ↑R3 EΛ . We are interested in constructing equilibrium states (w.r.t. the dynamics (5)). Those are special states, characterized by an inverse temperature β and a mean density ρ, which satisfy the socalled KMS condition, [4]. Formally, a state ω on a C ∗ algebra A is a (β, αt )KMS state (where αt is a ∗automorphism group of A, and 0 < β < ∞) if the KMS condition is satisﬁed, ω(Aαt (B)) = ω(αt−iβ (B)A) ,
(6)
for suitable observables A, B ∈ A. It is convenient and standard to formulate the theory in Fourier space, where the periodic Laplacian is diagonalized. In the inﬁnite volume limit, the test functions (wave functions of a single particle) f in (2)–(5) are elements of L2 (R3 , d3 k). In the following we place ourselves in this setting (all results can be expressed in direct space at the expense of a more cumbersome notation). The Araki–Woods Construction, [2] Let R3 k → ρ(k) > 0 be a given function (the “continuous momentumdensity distribution”), and ρ0 ≥ 0 a ﬁxed number (the “condensate density”). Araki and Woods obtain a state of the Bose gas by the following procedure. Put L3 ρ0 particles in the ground state of the one particle Hamiltonian HΛ , and a discrete distribution of particles in excited states. Then take the limit L → ∞ while keeping ρ0 ﬁxed and letting the discrete distribution of excited states tend to ρ(k). In this way one obtains the generating functional
1 AW 3 2(2π)3 ρ0 f (0) , (7) Eρ,ρ0 (f ) = exp − f, 1 + 2(2π) ρ f J0 4
π dθ −i(α cos θ+β sin θ) where J0 ( α2 + β 2 ) = −π 2π e , α, β ∈ R (Bessel function). AW As mentioned above, Eρ,ρ0 deﬁnes a state of the inﬁnitely extended Bose gas. The physical interpretation is that this state describes a free Bose gas where a sea of particles, all being in the same state (corresponding to the ground state of the ﬁnitevolume Hamiltonian), form a condensate with density ρ0 , which is immersed in a gas of particles where ρ(k) particles per unit volume have momentum in the inﬁnitesimal volume d3 k around k ∈ R3 . If the Hamiltonian in the ﬁnite box is taken with periodic boundary conditions the condensate is hom*ogeneous in space (the ground state wave function is a constant in position space). The resulting state is an equilibrium state (satisﬁes the KMS condition (6) for A, B Weyl operators with test functions f ∈ L2 (R2 , d3 k) and w.r.t. the dynamics t → W (e−itω f )) if the momentum density distribution is given by 1 ρ(k) = (2π)−3 βω(k) , (8) e −1
132
M. Merkli
corresponding to Planck’s law of black body radiation, and the condensate density ρ0 ≥ 0 is arbitrary. We consider dispersion relations ω(k) = k2 or ω(k) = k (nonrelativistic Bosons or massless relativistic ones). The GrandCanonical Construction, [16] The density matrix (acting on Fock space FΛ ) for the local grandcanonical equilibrium system is Λ = σβ,z Λ
e−β(HΛ −µΛ NΛ ) , Tr e−β(HΛ −µΛ NΛ )
(9)
where µΛ ∈ R is the chemical potential, zΛ = eβµΛ is the fugacity, and NΛ is the number operator. For a ﬁxed inverse temperature 0 < β < ∞ deﬁne the critical density by d3 k ρcrit (β) = (2π)−3 , (10) βω(k) e −1 and denote by ρ ≥ 0 the total (mean) density of the gas, whose value we are at liberty to choose, NΛ Λ ρ = Tr σβ,z . (11) Λ L3 For each ﬁxed L, this determines the value of zΛ (as a function of L, ρ and β). One then performs the thermodynamic limit of the expectation functional, Λ GC Λ Eβ,ρ (f ) = lim Eβ,z (f ) := lim Tr σβ,z W (f ) , (12) Λ L→∞
L→∞
where the limit L → ∞ is taken under the constraint (11). The limiting generating functional is # $ ' 1 2 z∞ 1 − 4 f e exp −
f, f , ρ ≤ ρcrit (β) GC 2 eβω −z∞ Eβ,ρ (f ) = (13) con ρ ≥ ρcrit (β) Eβ,ρ (f ), where, with ρ0 := ρ − ρcrit (β) ≥ 0, . 1 con (f ) = exp − f, (1 + 2(2π)3 ρ)f exp −4π 3 ρ0 f (0)2 , Eβ,ρ 4
(14)
and ρ = ρ(k) is given in (8). For subcritical density, ρ ≤ ρcrit (β), the number z∞ ∈ [0, 1] is determined by the equation z∞ ρ = (2π)−3 d3 k . (15) eβω(k) − z∞ In the supercritical case, ρ ≥ ρcrit (β), we have z∞ = 1 which corresponds to a vanishing chemical potential, µ∞ = 0.
Stability of MultiPhase Equilibria
133
The Canonical Construction, [5] The density matrix of the canonical local Gibbs state is µΛ β,ρ =
e−βHΛ PρL3 , Tr e−βHΛ PρL3
(16)
and PρL3 is the projection onto the subspace of Fock space with ρL3 particles (if ρL3 is not an integer take a convex combination of canonical states with integer values ρ1 L3 and ρ2 L3 extrapolating ρL3 ). The limiting generating functional is given by # $ ' 1 2 z∞ 1 − 4 f e exp −
f, f , ρ ≤ ρcrit (β) C 2 eβω −z∞ Eβ,ρ (17) (f ) = AW ρ ≥ ρcrit (β) . Eρ,ρ0 (f ), It coincides with the grandcanonical generating functional in the subcritical case, and with the ArakiWoods generating functional with ρ given by (8) and ρ0 = ρ − ρcrit in the supercritical case. The grandcanonical and the canonical generating functionals are linked, [5], in the supercritical case, ρ0 > 0, by the Laplace transform ∞ con C Eβ,ρ (f ) = K(r; ρ)Eβ,r (f )dr , (18) 0
where the Kac density K(r; ρ) is K(r; ρ) =
e−(r−ρcrit )/ρ0 /ρ0 , r > ρcrit 0, r ≤ ρcrit .
(19)
This means that the grandcanonical equilibrium state with supercritical mean density ρ is a superposition of canonical equilibrium states with supercritical densities r, weighted with the Kac density K(r, ρ). From now on, we focus on the inﬁnite volume equilibrium state ωβ,ρ determined by the thermodynamic limit of the canonical expectation functional, (17). (All that follows can be carried out for the grandcanonical expectation functional, (13), see [18], but is notationally less cumbersome in the canonical case.) 1.2 Spontaneous Symmetry Breaking and MultiPhase Equilibrium We denote by EρAW (f ) the ArakiWoods expectation functional (7) at critical density, where ρ(k) is given by (8), and ρ0 = 0. The corresponding equilibrium state is denoted by ωβ (it is a state on W ≡ W(D), where
134
M. Merkli
D ⊂ L2 (R3 , d3 k)
(20)
consists of functions s.t. the right side of (7) is deﬁned). With the expansion of the Bessel function J0 given after (7) we can write the canonical expectation functional, for ρ ≥ ρcrit (β), as a superposition dθ −iΦ(f,θ) AW C e (f ) = Eρ (f ) , (21) Eβ,ρ S 1 2π where the real phase Φ is given by
Φ(f, θ) = (2π)3/2 2ρ0 (Ref (0)) cos θ + (Imf (0)) sin θ . Correspondingly, we deﬁne the states
ωβθ
(22)
on W ≡ W(D) by
ωβθ (W (f )) = e−iΦ(f,θ) ωβ (W (f )) .
(23)
The point of this exercise is to notice that for each θ, (23) deﬁnes a βKMS state w.r.t. the dynamics αt (W (f )) = W (e−itω f ). Thus the supercritical equilibrium state ωβ,ρ corresponding to (21) is a uniform superposition of the equilibrium states ωβθ , θ ∈ S 1 . Of course, we can now take any probability measure dµ on S 1 and deﬁne a (β, αt )KMS state on W by µ ω (·) = dµ(θ)ωβθ (·) . (24) S1
One easily shows that the states (23) are factor states, so they are extremal. We also point out that they are not invariant under the gauge group σs (W (f )) = W (eis f ), s ∈ R, which is a symmetry group of the dynamics αt (meaning that αt ◦ σs = σs ◦ αt , s, t ∈ R). The existence of equilibrium states which have “less symmetry” than the dynamics is called spontaneous symmetry breaking. We close this paragraph with some observations on space mixing properties. Given a vector a ∈ R3 we deﬁne τa (W (f )) := W (fa ), where fa (x) := f (x − a) is the (directspace) translate of f by a. τa deﬁnes a (three parameter) group of automorphisms on W. A state ω on W is called strongly mixing w.r.t. space translations if lima→∞ ω(W (f )τa (W (g))) = ω(W (f ))ω(W (g)), for any f, g ∈ D. This means that if two observables (W (f ) and W (g)) are spatially separated far from each other then the expectation of the product of the observables is close to the product of the expectation values (independence of random variables). Intuitively, this means that the state ω has a certain property of locality in space: what happens far out in space does not inﬂuence events taking place, say, around the origin. For the equilibrium state ωβ,ρ determined by (17), it is easy to show that lim ωβ,ρ W (f )τa (W (g)) a→∞ ' 1, # $ ρ ≤ ρcrit (β) = ωβ,ρ (W (f ))ωβ,ρ (W (g)) exp −8π 3 ρ0 Re(f (0)g(0)) , ρ ≥ ρcrit (β) .
Stability of MultiPhase Equilibria
135
Consequently, this equilibrium state is strongly mixing w.r.t. space translations if and only if ρ0 = 0, i.e., if and only if there is no condensation. In presence of a condensate, the system exhibits long range correlations (what happens far out does inﬂuence what happens say at the origin). On the other hand, it is easily veriﬁed that each state ωβθ is strongly mixing. 1.3 Return to Equilibrium in Absence of a Condensate Consider the equilibrium state ωβ,ρ deﬁned by the expectation functional (17), in the regime ρ ≤ ρcrit (β). We say that ωβ,ρ has the property of return to equilibrium iﬀ lim ωβ,ρ (B ∗ αt (A)B) = ωβ,ρ (B ∗ B)ωβ,ρ (A) ,
t→∞
(25)
for all A, B ∈ W. (This means that all states which are normal w.r.t. ωβ,ρ converge to ωβ,ρ in the long time limit.) Let us show that lim ωβ,ρ W (g)αt (W (f ))W (h) = ωβ,ρ W (g)W (h) ωβ,ρ (W (f )) , (26) t→∞
for all f, g, h ∈ D. (26) implies (25). Using the CCR, (3), we obtain W (g)W (eiωt f )W (h) = e− 2 Im[g,e i
iωt
f +g+eiωt f ,h]
W (eiωt f + g + h) .
(27)
Using the RiemannLebesgue lemma, the ﬁrst factor on the r.h.s. of (27) is i seen to have the limit e− 2 Img,h , as t → ∞. From (17) we obtain ωβ,ρ (W (eiωt f + g + h)) =e
− 14 eiωt f +g+h2
1
89 8 2 iωt 82 z∞ 18 8 e f +g+h 8 exp − 8 8 , (28) 2 eβω(k) − z∞
and another application of the RiemannLebesgue lemma shows that the r.h.s. of (28) tends to ωβ,ρ (W (f ))ωβ,ρ (W (g + h)), as t → ∞. That’s it as for return to equilibrium for ρ ≤ ρcrit (β)! 1.4 Return to Equilibrium in Presence of a Condensate Of course one can do the calculation of the previous section in the case ρ > ρcrit by using explicitly (17) again. One will then notice that due to the presence of the Bessel function J0 in (7), expressions do not split so nicely into products any longer. A better way of doing things is to realize that the extremal states (23) do have the property of return to equilibrium, as is obvious from the calculation in the previous paragraph and the fact that Φ(eiωt f, θ) = Φ(f, θ). This leads immediately to the following expression for the asymptotic state of an initial condition which is a local perturbation of (normal w.r.t.) a general mixture, (24), of the extremal equilibria:
136
M. Merkli
lim ω µ (B ∗ αt (A)B) =
t→∞
S1
dµ(θ)ωβθ (B ∗ B)ωβθ (A) .
(29)
Hence the time asymptotic state of the system depends on the state it was initially in. We may interpret this “memory eﬀect” as a consequence of the fact that the system has longrange correlations. Although the perturbation propagates away from any bounded set (the dynamics is dispersive), correlations survive in the limit of large times. If B is such that ωβθ (B ∗ B) = 1 for all θ in the support of dµ (e.g. if B is unitary) then the asymptotic state is just ω µ again. In general however, the limit state is a diﬀerent mixture of the extremal equilibria, with a distribution given by the measure dµB (θ) = ωβθ (B ∗ B)dµ(θ) ,
(30)
another probability measure, which is absolutely continuous w.r.t. dµ(θ). In particular, the time asymptotic state (29) is normal w.r.t. ω µ . 1.5 Spectral Approach Take a heat reservoir and bring it into contact with a small system, say an N level system (e.g. a spin, or an idealized atom in a cavity). The interaction gives rise to emission and absorption processes, where particles in the reservoir (Bosons) are swallowed by the small system, which thereby increases its energy, or where the small system releases a particle by lowering its energy. It is a well know fact that the coupled system has an equilibrium state (Araki’s structural stability of equilibria, see also [7]). We want to show return to equilibrium for the coupled system. Let us outline here a strategy for doing so, introduced in [14, 15], and further developed in [3, 6, 9, 10, 17, 18]. Assume ω is a (β, αt )equilibrium state on a C ∗ algebra A, where αt is a dynamics on A (a group of ∗automorphisms). The GNS construction gives a Hilbert space representation (H, π, Ω) of the pair (A, ω), ω(A) = Ω, π(A)Ω ,
(31)
where H is the representation Hilbert space, Ω ∈ H, and π maps A into bounded operators on H. There is a unique selfadjoint operator L on H satisfying π(αt (A)) = eitL π(A)e−itL , LΩ = 0 , (32) for all A ∈ A, t ∈ R. L is called the (standard) Liouvillian. We have ω(B ∗ αt (A)B) = ω(α−iβ (B)B ∗ αt (A)) = Ω, π (α−iβ (B)B ∗ ) eitL π(A)Ω .
(33)
We use here that ω satisﬁes the KMS condition (6), and (31), (32). It becomes now apparent how we can link the longtime behaviour of (33) to the spectral
Stability of MultiPhase Equilibria
137
properties of L. If the kernel of L has dimension one (its dimension is at least one due to (32)), and if L has purely absolutely continuous spectrum on R\{0}, then the r.h.s. of (33) converges to
Ω, π (α−iβ (B)B ∗ ) Ω Ω, π(A)Ω = ω(α−iβ (B)B ∗ )ω(A) ,
(34)
as t → ∞. The ﬁrst term on the r.h.s. is just ω(B ∗ B) (use again the KMS condition (6) with t = 0). The combination of (34) and (33) shows that lim ω(B ∗ αt (A)B) = ω(B ∗ B)ω(A) ,
t→∞
(35)
provided KerL = CΩ and the spectrum of L on R\{0} is purely absolutely continuous. This shows return to equilibrium (c.f. (25))! A more modest version is return to equilibrium in the ergodic average sense, where the limit in (35) is understood in the ergodic mean, 1 T ω(B ∗ αt (A)B)dt = ω(B ∗ B)ω(A) . (36) lim T →∞ T 0 Relation (36) follows from (33) and the von Neumann ergodic theorem, provided KerL = CΩ.
2 Stability of MultiPhase Equilibria The discussion of Sect. 1.4 motivates the following more abstract Deﬁnitions. 1. Let ω be a state on a C ∗ algebra A, invariant w.r.t. a ∗automorphism group αt of A. We say that ω is asymptotically stable (w.r.t. αt ) if (37) lim ω(B ∗ αt (A)B) = ω(B ∗ B)ω(A) , t→∞
for any A, B ∈ A. 2. Let ωξ , ξ ∈ X (a measurable space), be a measurable family of states on a C ∗ algebra A (in the sense that ξ → ωξ (A) is measurable for all A ∈ A) and let αt be a ∗automorphism group of A. Given any probability measure µ on X we deﬁne the state dµ(ξ) ωξ . (38) ωµ = X
We say that the family ωξ is asymptotically stable (w.r.t. αt ) if, for any µ, A, B, we have dµ(ξ) ωξ (B ∗ B) ωξ (A) . (39) lim ω µ (B ∗ αt (A)B) = t→∞
X
3. If ω in 1. is a (β, αt )KMS state then we say ω has the property of Return to Equilibrium. Similarly, if the ωξ in 2. are (β, αt )KMS states (then so is ω µ ) we say the family ωξ has the property of Return to Equilibrium.
138
M. Merkli
In the above deﬁnitions the dynamics of the system is given by a ∗automorphism group αt of a C ∗ algebra A. While this description applies to free Fermionic or Bosonic heat reservoirs it does not in case a Bosonic reservoir is coupled to a small system. The problem is that one does not know how to deﬁne the dynamics for the coupled system as a ∗automorphism group of the C ∗ algebra of observables (unless the algebra is modiﬁed, see [9]). One circumvents this issue by deﬁning the interacting dynamics as a ∗automorphism group of the von Neumann algebra associated with a reference state (e.g. the uncoupled equilibrium state). We shall therefore adapt the above deﬁnitions to a setting where the dynamics is not deﬁned on the level of the C ∗ algebra of observables, but is rather expressed as a (“Schr¨ odinger”) dynamics of states. Deﬁnitions. 1’. Let ω be a state on a C ∗ algebra A and denote by (Hω , πω , Ωω ) its GNS representation, ω(A) = Ωω , πω (A)Ωω . Suppose σt is a ∗automorphism group of the von Neumann algebra πω (A) . We say that ω is asymptotically stable (w.r.t. σt ) if lim Ωω , πω (B ∗ )σt (πω (A))πω (B)Ωω = ω(B ∗ B)ω(A) ,
t→∞
(40)
for all A, B ∈ A. 2’. Let ωξ , ξ ∈ X (a measurable space), be a measurable family of states on a C ∗ algebra A and denote their GNS representations by (Hξ , πξ , Ωξ ). Suppose that, for each ξ, σtξ is a ∗automorphism group of the von Neumann algebra πξ (A) , s.t. ξ → σtξ (A)BΩξ is measurable, for all A, B ∈ πξ (A) , t ∈ R. ( AΩ = Ω, AΩ.) We say that the family ωξ is asymptotically stable (w.r.t. σtξ ) if, for any A, B ∈ A, we have lim dµ(ξ) σtξ (πξ (A))πξ (B)Ωξ = dµ(ξ) ωξ (B ∗ B) ωξ (A) , (41) t→∞
X
X
where µ is an arbitrary probability measure on X. 3’. If ω in 1’. is a (β, σt )KMS state of πω (A) then we say ω has the property of Return to Equilibrium. Similarly, if the ωξ in 2’. are (β, σtξ )KMS states of πξ (A) we say the family ωξ has the property of Return to Equilibrium.
3 Quantum Tweezers We investigate a Bose gas in a state with BoseEinstein condensate coupled to a small system. Description of Model and Stability Result The small system with which the Bose gas with condensate interacts can trap ﬁnitely many Bosons – we call it therefore a quantum dot. One can imagine
Stability of MultiPhase Equilibria
139
the use of such a trap to remove single (uncharged) particles from a reservoir, hence the name quantum tweezers (see also [8] and the references therein). The pure states of the small system are given by normalized vectors in Cd . We interpret [1, 0, . . . , 0] as the ground state (or “vacuum state”), [0, 1, 0, . . . , 0] as the ﬁrst excited state, e.t.c. The Hamiltonian is given by the diagonal matrix (42) H1 = diag(0, 1, 2, . . . , d − 1) . Our method applies to any selfadjoint diagonal matrix with nondegenerate spectrum. We introduce the raising and lowering operators, G+ and G− , 0 0 ··· 0 . 1 0 . . . .. , G− = (G+ )∗ , G+ = . (43) .. . . . . . . 0 0 ··· 1 0 (G+ has ones on its subdiagonal) which satisfy H1 G± = G± (H1 ± 1). The action of G+ (G− ) increases (decreases) the excitation level of the quantum dot by one. The dynamics of an observable A ∈ B(Cd ) (bounded operators on Cd ) is given by α1t (A) = eitH1 Ae−itH1 ,
t∈R.
(44)
The observable algebra of the combined system is the C ∗ algebra A = B(Cd ) ⊗ W(D) ,
(45)
where D ⊂ L2 (R3 , d3 k) (Fourier space) consists of f ∈ L2 (R3 , (1 + ρ)d3 k) which are continuous at zero. The noninteracting dynamics is α0t = α1t ⊗ α2t ,
(46)
α2t (W (f )) = W (eitω f ) ,
(47)
where ω(k) = k2 or k, is the free ﬁeld dynamics. Denote by ω1,β the Gibbs state of the quantum dot, and let ωβθ be given as in (23). Then θ := ω1,β ⊗ ωβθ ωβ,0
(48)
is a (β, α0t )KMS state. We can form diﬀerent equilibrium states by mixing such states according to any probability measure µ on S 1 . Let us now introduce an interaction operator, formally given by the expression (49) λ G+ ⊗ a(g) + G− ⊗ a∗ (g) , where λ ∈ R is a coupling constant, the G± are the raising and lowering operators, (43), and a# (g) are creation (# = ∗) and annihilation operators
140
M. Merkli
of the heat bath, smeared out with a function g ∈ D, called a form factor. The operator G+ ⊗ a(g) destroys a Boson and traps it in the quantum dot (whose excitation level is thereby increased by one) and similarly, the eﬀect of dot. The total number of G− ⊗ a∗ (g) is to release a Boson from the quantum particles, measured by the “observable” H1 + R3 a∗ (k)a(k)d3 k, is preserved by the interaction (meaning that (49) commutes with this operator). Since the quantum dot can absorb only ﬁnitely many Bosons, the interacting equilibrium state is a (local) perturbation of the noninteracting one. A physically diﬀerent situation occurs when the condensate is coupled to another reservoir. Then the timeasymptotic states are nonequilibrium stationary states. As we show in Sect. 3.2, the system has equilibrium states w.r.t. the interacting dynamics, and there is again a special family among them (extremal θ , labelled by θ ∈ S 1 , compare with (48). factorial ones), ωβ,λ Let µ be a probability measure on S 1 and set θ ωµ = dµ(θ) ωβ,λ . (50) S1
Our weak coupling result on Return of Equilibrium (Theorem 1 below) says that for all µ, A, B 1 λ→0 T →∞ T
lim lim
T
dt ω µ (B ∗ σλt (A)B) =
S1
θ θ dµ(θ) ωβ,0 (B ∗ B)ωβ,0 (A) ,
(51)
where σλt is the interacting dynamics. (The expression σλt (A) has to be understood cum grano salis, in the sense of Deﬁnition 2’, as σλt can only be deﬁned on the von Neumann algebra of observables, c.f. Sect. 3.2). We prove (51) under a condition of regularity and “eﬀectiveness” of the interaction. Let us close this section by discussing the physical meaning of the latter condition. Consider ﬁrst the Bose gas at critical density ρcrit (β) for some ﬁxed temperature 1/β (so that there is no condensate, ρ0 = 0). Heuristically, the probability of trapping a Boson in a state f in the quantum dot is given by 2 e−itHλ (ϕ ⊗ Ω) , (52) (G+ ⊗ a(f ))(ϕ ⊗ Ω), where ϕ an eigenstate of the quantum dot Hamiltonian and the Bose gas is in (for the calculation, we put the system in a box and the equilibrium state Ω Ω is a vector in Fock space with Bosons distributed according to a discrete distribution approaching the Planck distribution as the box size increases). The interacting Hamiltonian is Hλ = H0 + λ(G+ ⊗ a(g) + G− ⊗ a∗ (g)). The second order contribution in λ to (52), for large values of t, is P2 = C
λ2 f (1)g(1)2 , (eβω(1) − 1)2
(53)
Stability of MultiPhase Equilibria
141
where we assume that f (r), g(r) are radially symmetric, and where C is a constant independent of β, f, g. P2 gives the probability of the second order process where a Boson gets trapped in the quantum dot; the excitation energy is 1 (the quantum dot Hamiltonian (42) has equidistant eigenvalues) and the probability density of ﬁnding a Boson with energy ω(1) = 1 per unit volume is ∝ (eβ − 1)−1 , according to (8). In order not to suppress this trapping process at second order in the coupling constant we assume that g(1) = 0 (“eﬀective coupling”). Next let us investigate the inﬂuence of the condensate. For this we ﬁx a density ρ0 of the Bose gas and consider very low temperatures (β → ∞), so that most particles are in the condensate. For an explicit calculation we take in (52) is taken to consist only of particles in the a pure condensate, i.e., Ω ground state (with the prescribed density ρ0 ). We calculate the second order in λ of (52) to be Q2 (t) = C(1 − cos t)λ2 ρ20 f (0)g(0)2 .
(54)
We see from (54) that if g(0) = 0 then there is no coupling to the modes of the condensate: a physically trivial situation where the condensate evolves freely and the small system coupled to the “excited modes” undergoes return to equilibrium. We outline in this note results established in [18], which include the case g(0) = 0, a situation which could not be handled by approaches developped so far. 3.1 NonInteracting System The states of the small system are determined by density matrices ρ on the ﬁnite dimensional Hilbert space Cd . A density matrix is a positive traceclass operator, normalized as Tr ρ = 1, and the corresponding state ωρ (A) = Tr (ρA),
A ∈ B(Cd )
(55)
is a normalized positive linear functional on the C ∗ algebra B(Cd ) of all bounded operators on Cd , which we call the algebra of observables. The (Heisenberg) dynamics of the small system is given by (44). Denote the normalized eigenvector of H1 corresponding to Ej = j by ϕj . Given any inverse temperature 0 < β < ∞ the Gibbs state ω1,β is the unique βKMS state on B(Cd ) associated to the dynamics (44). The corresponding density matrix is e−βH1 . (56) ρβ = Tr e−βH1 Let ρ be a density matrix of rank d (equivalently, ρ > 0) and let {ϕj }d−1 j=0 be an orthonormal basis of eigenvectors of ρ, corresponding to eigenvalues " d 0 < pj < 1, j pj = 1. The GNS representation of the pair (B(C ), ωρ ) is given by (H1 , π1 , Ω1 ), where the Hilbert space H1 and the cyclic (and separating) vector Ω1 are
142
M. Merkli
H1 = Cd ⊗ Cd , √ pj ϕj ⊗ ϕj ∈ Cd ⊗ Cd , Ω1 =
(57) (58)
j
and the representation map π1 : B(Cd ) → B(H1 ) is π1 (A) = A ⊗ 1l .
(59)
We introduce the von Neumann algebra M1 = B(Cd ) ⊗ 1lCd ⊂ B(H1 ) .
(60)
The modular conjugation operator J1 associated to the pair (M1 , Ω1 ) is given by (61) J1 ψ ⊗ ψr = C1 ψr ⊗ C1 ψ , " " where C1 is the antilinear involution C1 j zj ϕj = j zj ϕj (complex conjugate). According to (58) and (56) the vector Ω1,β representing the Gibbs state ω1,β is given by 1 Ω1,β = √ e−βEj /2 ϕj ⊗ ϕj ∈ H1 . −βH 1 Tr e j
(62)
Denote by ωρ,ρ0 the state on W(D) whose generating functional is (7), where ρ(k) is given in (8), and ρ0 ≥ 0. The GNS representation of the pair (W(D), ωρ,ρ0 ) has been given in [2] as the triple (H2 , π2 , Ω2 ), where the representation Hilbert space is H2 = F ⊗ F ⊗ L2 (S 1 , dσ) ,
(63)
F = F(L2 (R3 , d3 k)) is the Bosonic Fock space over L2 (R3 , d3 k) and L2 (S 1 , dσ) is the space of L2 functions on the circle, with uniform normalized measure dσ (= (2π)−1 dθ, when viewed as the space of periodic functions of θ ∈ [−π, π]). The cyclic vector is (64) Ω 2 = ΩF ⊗ ΩF ⊗ 1 where ΩF is the vacuum in F and 1 is the constant function in L2 (S 1 , dσ). The representation map π2 : W(D) → B(H2 ) is given by
√ π2 (W (f )) = WF ( 1 + ρf ) ⊗ WF ( ρf ) ⊗ e−iΦ(f,θ) , (65) where WF (f ) = eiϕF (f ) is a Weyl operator in Fock representation and the ﬁeld operator ϕF (f ) is 1 ϕF (f ) = √ (a∗F (f ) + aF (f )) 2
(66)
and a∗F (f ), aF (f ) are the smeared out creation, annihilation operators satisfying the commutation relations
Stability of MultiPhase Equilibria
[aF (f ), a∗F (g)] = f, g, [aF (f ), aF (g)] = [a∗F (f ), a∗F (g)] = 0.
143
(67)
Our convention is that f → aF (f ) is an antilinear map. The phase Φ ∈ R is given in (22). In the absence of a condensate (ρ0 = 0 ⇒ Φ = 0) the third factor in (63)–(65) disappears and the representation reduces to the “ArakiWoods representation” in the form it has appeared in a variety of recent papers. We denote this representation by π0 . More precisely, the GNS representation of (W(D), ωρ,ρ0 =0 ) is given by (F ⊗ F, π0 , Ω0 ), where
√ (68) π0 (W (f )) = WF ( 1 + ρf ) ⊗ WF ( ρf ) , Ω0 = ΩF ⊗ ΩF .
(69)
Let us introduce the von Neumann algebras M0 = π0 (W(D)) ⊂ B(F ⊗ F) M2 = π2 (W(D)) ⊂ B(H2 )
(70) (71)
which are the weak closures (double commutants) of the Weyl algebra represented as operators on the respective Hilbert spaces. M2 splits into a product M2 = M0 ⊗ M ⊂ B(F ⊗ F) ⊗ B(L2 (S 1 , dσ)) ,
(72)
where M is the abelian von Neumann algebra of all multiplication operators on L2 (S 1 , dσ). It satisﬁes M = M. Relation (72) follows from this: clearly we have M0 ⊗ M ⊂ M2 , so taking the commutant gives M0 ⊗ M ⊃ M2 .
(73)
The reverse inclusion is obtained from 1lF ⊗F ⊗ M ⊂ M2 and M0 ⊗ 1lL2 (S 1 ) ⊂ M2 (see [2]). It is well known that M0 , the von Neumann algebra corresponding to the situation without condensate, is a factor. That means that its center is trivial, Z(M0 ) = M0 ∩M0 ∼ = C. However, we have Z(M2 ) = (M0 ⊗M)∩(M0 ⊗M), Z(M2 ) = 1lF ⊗F ⊗ M ,
(74)
so the von Neumann algebra M2 is not a factor. One can decompose M2 into a direct integral of factors, or equivalently, one can decompose ωρ,ρ0 into an integral over factor states. The Hilbert space (63) is the direct integral ⊕ dθ F ⊗F , (75) H2 = 2π [−π,π] and the formula (see (64), (65), (68), (69)) π dθ −iΦ(f,θ) ωρ,ρ0 (W (f )) = Ω2 , π2 (W (f ))Ω2 = e
Ω0 , π0 (W (f ))Ω0 −π 2π (76)
144
M. Merkli
shows that π2 is decomposed as
⊕
π2 = [−π,π]
dθ πθ , 2π
(77)
where πθ : W(D) → B(F ⊗ F) is the representation deﬁned by πθ (W (f )) = e−iΦ(f,θ) π0 (W (f )) . For each ﬁxed θ,
πθ (W(D)) = M0
is a factor. Accordingly we have the factor decomposition M2 = Consider the equilibrium state of the uncoupled system
(78)
⊕
dθ [−π,π] 2π
con ωβ,0 = ω1,β ⊗ ω2,β ,
(79) M0 . (80)
where ω1,β is the Gibbs state of the quantum dot (see (62)), and where ω2,β = ωρ,ρ0 is the equilibrium state of the heat bath at inverse temperature β and above critcal density, introduced after (62). The index 0 in (80) indicates the absence of an interaction between the two systems. The GNS representation con ) is just (H, π, Ω), where of (A, ωβ,0 H = H1 ⊗ H2 π = π1 ⊗ π 2 con Ωβ,0
= Ω1,β ⊗ Ω2 .
(81) (82)
The free dynamics is α0t , (46). Let Mcon := π(A) = M1 ⊗ M2 = β
⊕
[−π,π]
dθ M1 ⊗ M0 ⊂ B(H) 2π
(83)
be the von Neumann algebra obtained by taking the weak closure of all observables of the combined system, when represented on H. We have ⊕ dθ −iΦ(f,θ) e π2 (α2t (W (f ))) = π0 (W (eiωt f )) . (84) [−π,π] 2π It is well known and easy to verify that for A ∈ A, (π1 ⊗ π0 )(α0t (A)) = eitL0 (π1 ⊗ π0 )(A)e−itL0 ,
(85)
where the selfadjoint L0 on H1 ⊗ F ⊗ F is given by L0 = L1 + L2 , L1 = H1 ⊗ 1lCd − 1lCd ⊗ H1 ,
(86) (87)
L2 = dΓ (ω) ⊗ 1lF − 1lF ⊗ dΓ (ω) .
(88)
Stability of MultiPhase Equilibria
145
Here dΓ (ω) is the second quantization of the operator of multiplication by ω on L2 (R3 , d3 k). We will omit trivial factors 1l or indices Cd , F whenever we have the reasonable hope that no confusion can arise. It follows from (83)–(88) that the uncoupled dynamics α0t is unitarily implemented in H by π(α0t (A)) = eitL0 π(A)e−itL0 ,
(89)
where the standard, noninteracting Liouvillian L0 is the selfadjoint operator on H with constant (θindependent) ﬁber L0 , ⊕ dθ L0 . (90) L0 = [−π,π] 2π The r.h.s. of (89) extends to a ∗automorphism group σ0t of Mcon β , which we write ⊕ dθ t t σ0,θ , (91) σ0 = 2π [−π,π] t where σ0,θ is the ∗automorphism group of M1 ⊗ M0 generated by L0 . As is well known, (92) Ωβ,0 = Ω1,β ⊗ Ω0 t is a (β, σ0,θ )KMS state of M1 ⊗ M0 . The modular conjugation operator J associated to (M0 , Ω1,β ⊗ Ω0 ) is
J = J1 ⊗ J0 ,
(93)
where J1 is given by (61) and where the action of J0 on F ⊗ F is deter√ mined by antilinearly extending the relation J0 π0 (W (f ))Ω0 = WF ( ρf ) ⊗ √ WF ( 1 + ρ f )Ω0 . J0 deﬁnes an antilinear representation of the Weyl algebra according to W (f ) → J0 π0 (W (f ))J0 , which commutes with the representation π0 given in (68). We view this as a consequence of the TomitaTakesaki theory which asserts that M0 = J0 M0 J0 . It follows from (82), (83), (91) that ⊕ dθ con Ωβ,0 = (94) Ωβ,0 2π [−π,π] is a (β, σ0t )KMS state on Mcon β , and that the modular conjugation operator con J associated to (Mcon β , Ωβ,0 ) is given by J =
⊕
[−π,π]
dθ J1 ⊗ J0 . 2π
(95)
The standard Liouvillian L0 , (90), satisﬁes the relation J L0 = −L0 J and con =0. L0 Ωβ,0
(96)
146
M. Merkli
3.2 Interacting System The ﬁeld operator ϕ(f ) = 1i ∂t t=0 π(W (tf )) in the representation π, (81), is ⊕ dθ ϕθ (f ), where easily calculated to be ϕ(f ) = [−π,π] 2π
√ ϕθ (f ) = ϕF ( 1 + ρf ) ⊗ 1l + 1l ⊗ ϕF ( ρf ) − Φ(f, θ) ,
(97)
where Φ(f, θ) is given in (22), and where ϕF (f ) is given in (66). The creation operator a∗θ (f ) := √12 (ϕθ (f ) − iϕθ (if )) has the expression
√ √ a∗θ (f ) = a∗F ( 1 + ρf ) ⊗ 1lF + 1lF ⊗ aF ( ρf ) − (2π)−3/2 ρ0 f (0)e−iθ . (98) Using these expressions it is not diﬃcult to evaluate π G+ ⊗ a(g) + adjoint (apply (81) to (49)), and to see that the standard interacting Liouvillian is given by ⊕ dθ Lλ,θ , (99) Lλ = 2π [−π,π] where the selfadjoint operator Lλ,θ is Lλ,θ = L0 + λIθ .
(100)
Here L0 is given in (86) and we deﬁne Iθ = I + Kθ ,
(101)
∗ √ I = G+ ⊗ 1lCd ⊗ aF ( 1 + ρ g) ⊗ 1lF + 1lF ⊗ aF ( ρ g) + adj. (102)
√ −1lCd ⊗ C1 G+ C1 ⊗ a∗F ( ρg) ⊗ 1lF + 1lF ⊗ aF ( 1 + ρ g) + adj.
Kθ = Kθ1 ⊗ 1lCd ⊗ 1lF ⊗F − 1lCd ⊗ C1 Kθ1 C1 ⊗ 1lF ⊗F √ Kθ1 = −(2π)−3/2 ρ0 G+ g(0)eiθ + G− g(0)e−iθ
(103)
with C1 , Φ deﬁned in (61), (22) and where the creation and annihilation operators a∗F , aF are deﬁned by (67). The operator Lλ,θ generates a Heisenberg t on the von Neumann algebra M1 ⊗ M0 . It is convenient to dynamics σλ,θ write (compare with (91)) ⊕ dθ t σλ,θ . (104) σλt = [−π,π] 2π To the interacting dynamics (104) corresponds a βKMS state on Mcon β , the equilibrium state of the interacting system. It is given by the vector ⊕ dθ θ con con −1 Ωβ,λ , (105) Ωβ,λ = (Zβ,λ ) 2π [−π,π] con con where Zβ,λ is a normalization factor ensuring that Ωβ,λ = 1, and where
Stability of MultiPhase Equilibria θ θ Ωβ,λ = (Zβ,λ )−1 e−β(L0 +λIθ, )/2 Ωβ,0 ∈ H1 ⊗ F ⊗ F .
147
(106)
θ Zβ,λ is again a normalization factor, and Iθ, is obtained by dropping the terms coming with a minus sign in the r.h.s. of both (102) and (103). The fact −β(L0 +λIθ, )/2 , that Ωβ,0 , (92), √ is in the domain of the unbounded operator e provided g/ ωL2 (R3 ) < ∞, can be seen by expanding the exponential in a Dyson series and verifying that the series applied to Ωβ,0 converges, see e.g. [3]. It then follows from the generalization of Araki’s perturbation theory θ t is a (β, σλ,θ )KMS state on M1 ⊗ M0 , of KMS states, given in [7], that Ωβ,λ denoted by θ θ θ (·) = Ωβ,λ , · Ωβ,λ , (107) ωβ,λ
and that θ Lλ,θ Ωβ,λ =0.
(108)
con con We conclude that Ωβ,λ is a (β, σλt )KMS state on Mcon β , and that Lλ Ωβ,λ = 0.
3.3 Stability of the Quantum Tweezers, Main Results We make two assumptions on the form factor g determining the interaction. (A1) Regularity. The form factor g is a function in C 4 (R3 ) and satisﬁes
√ (1 + 1/ ω)(k · ∇k )j 1 + ρ gL2 (R3 ,d3 k) < ∞ , (109) √ 2 for j = 0, . . . , 4, and (1 + ω) 1 + ρ gL22(R3 ,d3 k) < ∞. (A2) Eﬀective coupling. We have S 2 dσg(1, σ) = 0, where g is represented in spherical coordinates. Remarks. 1) The operator k · ∇k emerges because we apply the positive commutator method with conjugate operator 12 (k · ∇k + ∇k · k) (dilation generator). It is important to notice that in order to couple the particles of the BoseEinstein condensate to the quantum dot, we must treat the case g(0) = 0. If the dispersion relation is given by ω(k) ∼ ks , as k ∼ 0, then (109) is satisﬁed for s < 3/2. (This does not include nonrelativistic Bosons, for which s = 2.) 2) Condition (A2) is often called the Fermi Golden Rule Condition. It guarantees that the processes of absorption and emission of ﬁeld quanta by the small system, which are the origin of the stability of the equilibrium, are eﬀective (see the discussion at the beginning of Sect. 3). We integrate over a sphere of radius one since the gap between neighbouring eigenvalues of H1 is equal to one (Bohr frequency). Let B ∈ A, µ a probability measure on S 1 , and deﬁne a state on A µ θ θ θ θ by ωB (A) = S 1 dµ(θ)ωβ,λ (B ∗ AB), where ωβ,λ (A) = Ωβ,λ , Aθ Ωβ,λ , with (107), (59), (78)). We introduce the suggestive noAθ := (π1 ⊗ πθ )(A) (see µ t (σλt (A)) := S 1 dµ(θ) σλ,θ (Aθ )Bθ Ωβ,λ , where Aψ = ψ, Aψ, and tation ωB θ t where σλ,θ is given in (104).
148
M. Merkli
Theorem 1 (Stability of equilibrium with condensate, [18]). Assume conditions (A1) and (A2). Let A, B ∈ A, and let µ be a probability measure on S 1 . Then 1 T µ t θ θ lim lim ωB (σλ (A))dt = dµ(θ)ωβ,0 (B ∗ B)ωβ,0 (A) , (110) λ→0 T →∞ T 0 S1 θ is given in (48). where ωβ,0
The proof of Theorem 1 follows the ideas of the spectral method outlined in Subsection 1.5. It is based on the following result, which in turn is proved using positive commutator techniques. Theorem 2 (Structure of kernel, [18]). Assume Conditions (A1) and θ the projection onto the subspace spanned by the interacting (A2) and let Pβ,λ θ KMS state Ωβ,λ , (106). Let θ ∈ [−π, π] be ﬁxed. Any normalized element ψλ ∈ ⊥ θ Ker(Lλ,θ ) ∩ RanPβ,λ converges weakly to zero, as λ → 0. The convergence is uniform in θ ∈ [−π, π] and in β ≥ β0 , for any β0 > 0 ﬁxed. I would like to express my thanks to Joachim Asch and to Alain Joye for having organized Qmath9, and for having invited me to make a contribution. My thanks also go to the referee for his careful reading of the manuscript.
References 1. W. Amrein, A. Boutet de Monvel, V. Georgescu: C0 Groups, Commutator Methods and Spectral Theory of N body Hamiltonians. BaselBostonBerlin: Birkh¨ auser, 1996. 2. H. Araki, E. Woods: J. Math. Phys. 4, 637 (1963). 3. V. Bach, J. Fr¨ ohlich, I.M. Sigal, I.M.: J. Math. Phys. 41, no. 6, 3985 (2000). 4. Bratteli, O., Robinson, D.W., Operator Algebras and Quantum Statistical Mechanics I, II. Texts and Monographs in Physics, SpringerVerlag, 1987. 5. J.T. Cannon: Commun. Math. Phys., 29, 89 (1973). 6. J. Derezi´ nski, J., V. Jak˘si´c: Ann. Henri Poincar´e 4, no. 4, 739 (2003). 7. J. Derezi´ nski, V. Jak˘si´c, C.A. Pillet: Rev. Math. Phys. 15, no. 5, 447 (2003). 8. R.B. Diener, B. Wu, M.G. Raizen, Q. Niu: Phys. Rev. Lett. 89, 070401 (2002). 9. J. Fr¨ ohlich, M. Merkli: Math. Phys. Anal. Geom. 7, no. 3, 239 (2004). 10. J. Fr¨ ohlich, M. Merkli: Commun. Math. Phys. 251, no. 2, 235 (2004). 11. J. Fr¨ ohlich, M. Merkli, D. Ueltschi: Ann. Henri Poincar´e 4, no. 5, 897 (2003). 12. J. Fr¨ ohlich, M. Merkli, I.M. Sigal: J. Statist. Phys. 116, no. 1–4, 311 (2004). 13. V. Georgescu, C. G´erard: Commun. Math. Phys. 208, 275 (1999). 14. V. Jak˘si´c, C.A. Pillet: Commun. Math. Phys. 176, 619 (1996). 15. V. Jak˘si´c, C.A. Pillet: Commun. Math. Phys. 178, 627 (1996). 16. J.T. Lewis, J.V. Pul´e: Commun. Math. Phys. 36, 1 (1974). 17. M. Merkli: Commun. Math. Phys. 223, 327 (2001). 18. M. Merkli: Commun. Math. Phys. 257, 621 (2005)
Ordering of Energy Levels in Heisenberg Models and Applications Bruno Nachtergaele1 and Shannon Starr2 1
2
Department of Mathematics, University of California, Davis, One Shields Avenue, Davis, CA 956168366, USA [emailprotected] Department of Mathematics, University of California, Los Angeles, Box 951555, Los Angeles, CA 900951555, USA [emailprotected]
Abstract. In a recent paper [17] we conjectured that for ferromagnetic Heisenberg models the smallest eigenvalues in the invariant subspaces of ﬁxed total spin are monotone decreasing as a function of the total spin and called this property ferromagnetic ordering of energy levels (FOEL). We have proved this conjecture for the Heisenberg model with arbitrary spins and coupling constants on a chain [17, 20]. In this paper we give a pedagogical introduction to this result and also discuss some extensions and implications. The latter include the property that the relaxation time of symmetric simple exclusion processes on a graph for which FOEL can be proved, equals the relaxation time of a random walk on the same graph with jump rates given by the coupling constants, i.e., the relaxation time is independent of the number of particles. Therefore, our results also provide a proof of Aldous’ Conjecture in one dimension.
1 Introduction The ferromagnetic Heisenberg model is the primordial quantum spin model. It has been studied almost continuously since it was introduced by Heisenberg in 1926. In the course of its long history, this model has inspired an amazing variety of new developments in both mathematics and physics. The Heisenberg Hamiltonian is one of the basic, nontrivial quantum manybody operators, and understanding its spectrum has been a guiding problem of mathematical physics for generations. A lot of attention has been given to the BetheAnsatz solvable onedimensional spin1/2 model, which has an inﬁnitedimensional algebra of symmetries [9]. The results we will discuss here are not related to exact solutions but there is an essential connection with the SU (2) symmetry of the model, much in the spirit of the famous result by Lieb and Mattis ( [15], see also [14, footnote 6]). The LiebMattis Theorem proves “ordering of energy levels” for a large class of antiferromagnetic Heisenberg models on bipartite ∗
c 2006 by the authors. This paper may be reproduced, in its enCopyright tirety, for noncommercial purposes. B. Nachtergaele and S. Starr: Ordering of Energy Levels in Heisenberg Models and Applications, Lect. Notes Phys. 690, 149–170 (2006) www.springerlink.com
150
B. Nachtergaele and S. Starr
lattices. Namely, if the two sublattices are A and B, and all interactions within A and B are ferromagnetic while interactions in between A and B are antiferromagnetic, then the unique ground state multiplet has total spin equal to SA − SB , where SA and SB are the maximum total spins on the two sublattices. Moreover, the minimum energy in the invariant subspace of total spin S, for S ≥ SA − SB , is monotone increasing as a function of S. The most important example where this theorem provides useful information is the usual antiferromagnet on a bipartite lattice with equalsize sublattices. Then the ground state is a unique spin singlet, and the minimum energy levels for each possible total spin S, are monotone increasing in S. Our aim is a similar result for ferromagnets. To be able to state the ferromagnetic ordering of energy levels (FOEL) property precisely, we ﬁrst give some deﬁnitions. Let Λ be a ﬁnite connected graph with a set of vertices or sites, x, that we will also denote by Λ and a set E of unoriented edges, or bonds, (xy). We will often write x ∼ y ∈ Λ to signify that the edge (xy) is present in Λ. In many physical examples one has Λ ⊂ Zd . Each site x ∈ Λ has a quantum spin of magnitude sx ∈ {1/2, 1, 3/2, . . .}, associated with it. The state space at x is 2sx + 1dimensional and we denote acting on the xth tensor by Sxi , i = 1, 2, 3, the standard spins : x matrices 2sx +1 C . The isotropic (also called factor in the Hilbert space H = x∈Λ XXX) ferromagnetic Heisenberg Hamiltonian on Λ is given by Jxy S x · S y , (1) HΛ = − x∼y∈Λ
where the real numbers Jxy are the coupling constants, which we will always assume to be strictly positive (that they are positive is what it means to have the ferromagnetic Heisenberg model). This model is widely used to describe ferromagnetism at the microscopic level whenever itinerant electron eﬀects can be ignored. Examples are magnetic domain walls and their properties and a variety of dynamical phenomena. The spin matrices generate an irreducible representation of SU (2) at each vertex. This representation is conventionally denoted by D(sx ) . An important feature of the Hamiltonian (1) is that it commutes with SU (2) via the representation ; D(sx ) (2) x∈Λ
or, equivalently, with the total spin matrices deﬁned by i SΛ = Sxi , i = 1, 2, 3 . x∈Λ
and hence also with the Casimir operator given by C = SΛ · SΛ .
Ordering of Energy Levels in Heisenberg Models and Applications
151
"
The eigenvalues of C are S(S + 1), S = Smin , Smin + 1, . . . , Smax ≡ x∈Λ sx , which are the spin labels of the irreducible representations that occur in the direct sum decomposition of the tensor product representation (2) into irreducible components. The value of Smin is usually 0 or 1/2, but may be larger if one of the sx is greater than Smax /2. The decomposition into irreducible components can be obtained by repeated application of the ClebschGordan series: D(s1 ) ⊗ D(s2 ) ∼ = D(s1 −s2 ) ⊕ D(s1 −s2 +1) · · · ⊕ D(s1 +s2 ) .
(3)
The label S is called the total spin, and the eigenvectors of the eigenvalue S(S+1) of C, are said to have total spin S. Let H(S) denote the corresponding eigenspace. Since C commutes with HΛ , the spaces H(S) are invariant subspaces for HΛ . For any hermitian matrix H leaving the spaces H(S) invariant we deﬁne E(H, S) = min spec HH(S) . By Ferromagnetic Ordering of Energy Levels (FOEL) we mean the property E(H, S) ≤ E(H, S ), if S < S . for all S and S in the range [Smin , Smax ], and we will speak of strict FOEL if the inequality is strict. In particular, if HΛ has the FOEL property it follows that its ground state energy is E(HΛ , Smax ), which is indeed wellknown to be the case for the Heisenberg ferromagnets. Moreover, since the multiplet of maximal spin is unique, FOEL also implies that the gap above the ground state is E(HΛ , Smax − 1) − E(HΛ , Smax ), which is wellknown for translation invariant Heisenberg ferromagnets on Euclidean lattices. Conjecture 1. All ferromagnetic Heisenberg models have the FOEL property. The FOEL property and the LiebMattis theorem applied to a spin1 chain of 5 sites is illustrated in Fig. 1. Our main result is a proof of this conjecture for the special case of arbitrary ferromagnetic Heisenberg models on chains, i.e., onedimensional model [17, 18, 20]. Theorem 1. Strict FOEL holds for ferromagnetic XXX spin chains, i.e., for all L−1 1 H=− Jx,x+1 S x · S x+1 − 1 , (4) sx sx+1 x=1 for any choice of sx ∈ {1/2, 1, 3/2, . . .} and Jx,x+1 > 0.
152
B. Nachtergaele and S. Starr Spectrum of SU(2) − symmetric Heisenberg model : j = 1, L = 5 10
1 0 2
Energy
3
5
4
0 1
2 3 4 5
5 −5
−4
−3
−2
−1 0 1 2 (3) S −eigenvalue / M−sector
3
4
5
Fig. 1. The spectrum of a ferromagnetic Heisenberg chain consisting of 5 spin1 spins, and with constant couplings. On the horizontal axis we have plottted the eigenvalue of the third component of the total spin. The spectrum is oﬀset so that the ground state energy vanishes. The arrows on the right, with label S, indicate the multiplets of eigenvalues E(H, S), i.e., the smallest eigenvalue in the subspace of total spin S. The monotone ordering of the spin labels is the FOEL property. On the left, we have indicated the largest eigenvalues for each value of the total spin. The monotone ordering of their labels in the range 1, . . . , 5, is the content of the LiebMattis theorem applied to this system
2 Proof of the Main Result Our proof of Theorem 1 proceeds by a ﬁnite induction argument for a sequence of models with Hamiltonians Hk = Hk∗ , 1 ≤ k ≤ N , on Hilbert spaces Hk , with the following properties: (i) There is a unitary representation of SU (2), Uk , on Hk , that commutes with Hk . (ii) There are isometries Vk : Hk+1 → Hk ⊗ C2 , interwining the representations Uk+1 and Uk ⊗ D(1/2) , i.e., Vk Uk+1 (g) = (Uk (g) ⊗ D(1/2) (g))Vk , for all g ∈ SU (2), and such that Hk+1 ≥ Vk∗ (Hk ⊗ 1l)Vk
Ordering of Energy Levels in Heisenberg Models and Applications
153
(iii) H1 has the FOEL property. (iv) For every S = 0, 1/2, 1, 3/2, . . ., for which H(S) = {0}, we have E(Hk+1 , S + 1/2) < E(Hk , S) . To prove FOEL and not necessarily strict FOEL, one only needs to have ≤ here. We will ﬁrst present the induction argument using the assumptions (i)– (iv), and then construct the sequence Hk satisfying these four assumptions. This argument is a generalization of results in [12] and [17]. The sequence of Hamiltonians will, roughly speaking, be a sequence of systems of increasing size, starting with the trivial system of a single spin. Property (i) simply means that all models will have isotropic interactions. Property (ii) will closely guide the construction of our sequence. Property (iii) will be trivial in practice, since H1 will be a multiple of the identity on H1 in our applications. Property (iv) has a nice physical interpretation at least in some of the examples we will consider (see Sect. 5.2). It is our (in) ability to prove (iv) that limits the range of models for which we can prove FOEL. Theorem 2. Let (Hk )1≤k≤N , be a sequence of Hamiltonians satisfying properties (i)–(iv). Then, for all k, Hk has the FOEL property. Proof. Since H1 has the FOEL property by assumption, it is suﬃcient to prove the induction step. Consider the following diagram: E(Hk , S)
>1
E(Hk , S + 1)
>
E(Hk , S + 2) >
>
>
≥
≥
3 2 2 E Hk+1 , S + 12 >4 E Hk+1 , S + 32 > E Hk+1 , S + 52
The inequality labeled 1 is FOEL for Hk , and inequality 2 is property (iv) assumed in the theorem. We will prove inequality 3 (using inequality 1) and, combined with inequality 2 this implies inequality 4, which is the induction step. As before, we use superscripts to Hilbert spaces to denote their subspaces of ﬁxed total spin. To prove inequality 3, we start from the variational principle: E(Hk+1 , S + 1/2) = ≥
,φ=1
inf
(S+1/2)
φ∈Hk+1
≥
φ, Hk+1 φ
inf
(S+1/2)
φ∈Hk+1
φ, Vk∗ (Hk ⊗ 1l2 )Vk φ
,Vk φ=1
inf
ψ, (Hk ⊗ 1l2 )ψ
ψ∈(Hk ⊗C2 )(S+1/2) ,ψ=1
The ﬁrst inequality uses the fact that Vk is an isometry and property (ii). For the second inequality we enlarged the subspace over which the inﬁmum is taken.
154
B. Nachtergaele and S. Starr
Now, we use the ClebschGordan series (3) to see that (Hk ⊗C2 )(S+1/2) ⊂ (S+1) ⊕ Hk ) ⊗ C2 . Therefore
(S) (Hk
E(Hk+1 , S + 1/2) ≥ min{E(Hk , S), E(Hk , S + 1)} = E(Hk , S + 1) . (S)
(S+1)
Clearly, Hk ⊗ 1l2 restricted to (Hk ⊕ Hk ) ⊗ C2 has the same spectrum as (S) (S+1) Hk restricted to Hk ⊕Hk . The last equality then follows from inequality 1, i.e., the induction hypothesis. This concludes the proof of Theorem 2. For the proof of Theorem 1 we will apply Theorem 2 to the sequence (Hk )1≤k≤N , with N = 2Smax , H1 = 0, and HN = H, constructed as follows: for each k = 1, . . . , N −1, the model with Hamiltonian Hk+1 is obtained from Hk in one of two ways: either a new spin 1/2 is added to right of the chain, or the magnitude of the rightmost spin is increased by 1/2. In both cases, Smax "L goes up by 1/2 at each step, hence N = 2 x=1 sx . Each Hk is of the form (4), and we have written the interactions in such a way that the coupling constants Jx,x+1 can be taken to be independent of k, although this is not crucial since all arguments work for any choice of positive coupling constants at each step. The parameters that change with k are thus L and the set of spin magnitudes (sx )L x=1 . To be explicit, the two possible ways of deriving Hk+1 from Hk are summarized in Table 1. Table 1. Summary of the kdependence of the sequence of models used in the proof by induction of Theorem 2 parameter L {sx } H
Case I
Case II
Lk+1 = Lk + 1 Lk+1 = Lk sLk +1 (k) = 0, sLk +1 (k + 1) = 1/2 sLk (k + 1) = sLk (k) + 1/2 Hk+1 = V (Hk ⊗ C2 ) Hk+1 = Hk ⊗ C2
The Hamiltonians are of the form L k −1 1 S x · S x+1 − 1 , Jx,x+1 Hk = − sx (k)sx+1 (k) x=1
(5)
where Sxi , i = 1, 2, 3, are the 2sx (k)+1 dimensional spin matrices. To simplify the notation, the dependence on k will often be omitted further on. We now have a uniquely deﬁned sequence of Hamiltonians (Hk )1≤l≤N , with H1 = 0 and HN = H. Next, we proceed to proving the properties (i)– (iv). Property (i) is obvious by construction. Property (iii) is trivial since H1 = 0. To verify property (ii), we need to distinguish the two cases for the relation between Hk and Hk+1 , as given in Table 1. For Case I, Uk+1 = Uk ⊗ D(1/2) and we can take the identity map for V . Property (ii) follows from the positivity of the additional interaction term in Hk+1 :
Ordering of Energy Levels in Heisenberg Models and Applications
Hk+1 = Hk + JLk ,Lk +1
1 SL sLk · (1/2) k
155
· S Lk +1 − 1 .
For Case II, we have Hk = Hl ⊗ C2sLk +1 and Hk+1 = Hl ⊗ C2sLk+1 +1 , for some l < k, possibly l = 0, H0 = C. Since sLk+1 = sLk + 1/2, there is a (up to a phase) unique SU (2) intertwining isometry W : C2sLk+1 → C2sLk +1 ⊗ C2 , namely the W that identiﬁes the spin sLk+1 subrepresentation in D(sLk ) ⊗ D(1/2) . From the intertwining property, the irreducibility of the spin representations, and the SU (2) commutation relations one deduces that there is a constant c such that W ∗ (SLi k (k) ⊗ 1l)W = cSLi k (k + 1),
i = 1, 2, 3 .
The constant c is most easily determined by calculating the left and right hand sides on a highest weight vector (a simultaneous eigenvector of C and S 3 with eigenvalues S(S + 1) and S, respectively). One ﬁnds c=
sLk (k) . sLk (k + 1)
Now, take V = 1lHl ⊗ W . It is then straightforward to check that 1 1 S Lk−1 · S Lk ⊗ 1l2 V = SL V∗ · S Lk , sLk−1 sLk (k) sLk−1 sLk (k + 1) k−1 where the spin matrices on the left hand side are of the magnitude determined by sLk−1 and sLk (k) , while on the right hand side they are the magnitudes of the spins are sLk−1 and sLk (k + 1). To prove Property (iv), we start by observing that spec(Hk H(S) ) = spec(Hk V (S) ) k
k
(S)
where Vk is the subspace of Hk of all highest weight vectors of weight S. This is an invariant subspace for Hk and for every eigenvalue of Hk H(S) there k
(S)
(S)
is at least one eigenvector in Vk . Let d(k, S) denote the dimension of Vk . Property (iv) will be obtained as a consequence of the following proposition and a version of the PerronFrobenius Theorem. (S)
Proposition 1. We have d(k +1, S +1/2) ≥ d(k, S) and there are bases Bk (S) for Vk such that the matrices A(k,S) of Hk V (S) with respect to these bases k have the following properties: (k,S)
Aij (k+1,S+1/2)
Aij
≤ 0, for 1 ≤ i = j ≤ d(k, S), 1 ≤ k ≤ N
(k,S)
≤ Aij
, for 1 ≤ i, j ≤ d(k, S), 1 ≤ k ≤ N − 1 .
156
B. Nachtergaele and S. Starr
For reasons of pedagogy and length, we will give the complete proof of this proposition only for the spin 1/2 chain. The proposition provides the assumptions needed to apply a slightly extended PerronFrobenius theorem (see, e.g., [23]), which we state below. The standard PerronFrobenius Theorem makes several statements about square matrices with all entries nonnegative, which we will call a nonnegative matrix for short. Recall that a nonnegative matrix A is called irreducible if there exists an integer n ≥ 1 such that the matrix elements of An are all strictly positive. The standard results are the following: (i) every nonnegative matrix has a nonnegative eigenvalue equal to its spectral radius (hence it has maximal absolute value among all eigenvalues), and there is a corresponding nonnegative eigenvector (i.e., with all components nonnegative); (ii) if A is an irreducible nonnegative matrix there is a unique eigenvalue with absolute value equal to the spectral radius of A, which is strictly positive and has algebraic (and hence geometric) multiplicity 1. Its corresponding eigenvector can be chosen to have all strictly positive components. If A is a square matrix with all oﬀdiagonal matrix elements nonpositive, we will call A irreducible if there exists a constant c such that c1l − A is irreducible according to the previous deﬁnition. From the standard PerronFrobenius Theorem it immediately follows that the eigenvalue with smallest real part of an irreducible matrix in the last sense is real, has algebraic (and hence geometric) multiplicity 1, and that the corresponding eigenvector can be chosen to have all components strictly positive. In the following, we will repeatedly use the information provided by the standard PerronFrobenius Theorem as described above without further reference. Let specrad(A) denote the spectral radius of a square matrix A. Lemma 1. Let A = (aij ) and B = (bij ) be nonnegative n × n matrices, and assume that aij ≤ bij , for all 1 ≤ i, j ≤ n. Then specrad(A) ≤ specrad(B) .
(6)
If B is irreducible and there is at least one pair ij such that aij < bij , then specrad(A) < specrad(B) .
(7)
Since the spectral radii are also the eigenvalues of maximal absolute value, the same relations holds for these eigenvalues. Proof. Let r = specrad(A). Then A has a nonnegative eigenvector, say v, with eigenvalue r. If aij ≤ bij , for all 1 ≤ i, j ≤ n, it is clear that there is a nonnegative vector w such that Bv = rv + w .
(8)
This relation implies that B k ≥ rk , for all positive integers k and, hence, specrad(B) ≥ r. This proves (6).
Ordering of Energy Levels in Heisenberg Models and Applications
157
To prove (7) for irreducible B such that aij ≤ bij for at least one pair of indices, let k be a positive integer such that B k is strictly positive. This implies that B k v has all strictly positive components. From this it is easy to see that the nonnegative w such that B k v = rk v + w cannot be the zero vector. Therefore there is z ∈ R with all strictly positive components such that B k+1 v = rk+1 v + z . Since z is strictly positive, there exists ε > 0 such that, v ≤ z componentwise, and therefore we can ﬁnd δ > 0 such that B k+1 v = (r + δ)k+1 v + z , with z nonnegative. We conclude that specrad(B) ≥ r + δ > specrad(A). Note that the argument that proves this lemma could also be used to give a lower bound for the diﬀerence of the spectral radii. Since we do not need it, we will not pursue this here. The next theorem is an extension of Lemma 1. Theorem 3. Let A = (aij ) and B = (bij ) be two square matrices of size n and m, respectively, with n ≤ m, both with all oﬀdiagonal matrix elements nonpositive, and such that bij ≤ aij , for 1 ≤ i, j ≤ n. Then inf spec(B) ≤ inf spec(A) .
(9)
If B is irreducible and either (i) there exists at least one pair ij, 1 ≤ i, j ≤ n, such that bij < aij ; or (ii) bij < 0, for at least one pair ij with at least one of the indices i or j > n, then inf spec(B) < inf spec(A) .
(10)
Proof. Let c ≥ 0 be a constant such that the matrices A = (aij ) = c1ln − A and B = (bij ) = c1lm − B are nonnegative. Deﬁne A to be the m × m matrix obtained by extending A with zeros: A 0 (aij ) = A = . 0 0 It is easy to see that aij ≤ bij , for 1 ≤ i, j ≤ m. Therefore, we can apply Lemma 1 with A playing the role of A, and B playing the role of B. Clearly, specrad(A ) = c − inf spec(A) and specrad(B ) = c − inf spec(B). Therefore, this proves (9). Similarly, (10) follows from the additional assumptions and (7). Proof of Theorem 1: The remaining point was to prove property (iv) needed in the assumptions of Theorem 2. We use Proposition 1, which we will prove in the next section, and apply Theorem 3 with A = A(k,S) and B = A(k+1,S+1/2) . This completes the proof.
158
B. Nachtergaele and S. Starr
3 The TemperleyLieb Basis. Proof of Proposition 1 In the proof of Theorem 1 in the previous section we used the matrix representation of the Hamiltonians restricted to the highest weight spaces given by Proposition 1. We now give the complete proof of that proposition for the spin 1/2 chain and sketch the proof in the general case. The main issue is to ﬁnd a basis of the highest weight spaces with the desired properties. Fortunately for us, such a basis has already been constructed and we only need to show that it indeed had the properties claimed in Proposition 1. For the spin 1/2 chain we will use the TemperleyLieb basis [22], and for the general case its generalization to arbitrary spin representations introduced by Frenkel and Khovanov [6]. 3.1 The Basis for Spin 1/2 We start with the spin 1/2 chain, i.e., sx = 1/2, for all x. In this case Smax = (S) k/2 and Vk is the subspace of (C2 )⊗k consisting of all vectors ψ such that 3 S ψ = Sψ and S + ψ = 0. Let n be the “spindeviation” deﬁned as S = k/2 − n. Then, n is a nonnegative integer. The case n = 0 is trivial since (k/2) dim Vk = 1, namely just the mutiples of the vector + ⊗ + ⊗ · · · ⊗ +, where ± is the basis of C2 that diagonalizes S 3 . For n ≥ 1, the basis vectors are a tensor product of n singlet vectors ξ = + ⊗ − − − ⊗ +, accounting for two sites each, and k −2n factors equal to +. Such vectors are sometimes called Hulth´en brackets. It is clear that any such factor is a highest weight vector of weight k/2 − n, just calculate the action of S 3 and S + on such a vector. They are not linearly independent however, except in the trivial case k = 2. The contribution of Temperley and Lieb was to show how to select a complete and linearly independent subset, i.e., a basis. How to select the TemperleyLieb basis, is most easily explained by representing the vectors by conﬁgurations of n arcs on the k vertices 1, . . . , n. The arcs are drawn above the line of vertices as shown in Fig. 2. Each arc represents a spin singlet ξ, and each unpaired vertex represents a factor +. The vectors (conﬁgurations of arcs) selected for the basis are those that satisfy two properties: (i) the arcs are noncrossing, (ii) no arc spans an unpaired vertex. The resulting set is a (nonorthogonal) basis. E.g., the basis for k = 5 and n = 2 is shown in Fig. 2. We will use, α, β, . . ., to denote arc conﬁgurations that obey these rules, and by the corresponding basis vectors will be denoted by α , β , . . .. We will use the notation [xy] ∈ α to denote that the arc connecting x and y is present in α. Proof of Proposition 1 for the spin 1/2 chain. The action of the Hamiltonian on the basis vectors has an appealing graphical representation. We can write the Hamiltonian as Hk = −2
k−1 x=1
Jx,x+1 Ux,x+1
Ordering of Energy Levels in Heisenberg Models and Applications
159
Fig. 2. The possible conﬁgurations of 2 arcs on 5 vertices
where Ux,x+1 = −ξ ⊗ξ ∗ which, up to a factor −2, is the orthogonal projection onto the singlet vector acting on the xth and x + 1st factor in the tensor product. The Ux,x+1 form a representation of the TemperleyLieb algebra with parameter q = 1 (see, e.g, [10]). It is a straightforward calculation to verify the action of Ux,x+1 on a basis vector α: (i) if both x and x + 1 are unpaired vertices in α, Ux,x+1 α = 0; (ii) if [x, x + 1] ∈ α, we have Ux,x+1 α = −2 α; (iii) if [uv] ∈ α, with exactly one of the vertices u and v equal to x or x + 1, we have Ux,x+1 α = β, where β is obtained form α by removing [uv] and adding [x, x + 1]; (iv) if [ux] and [x + 1, v] are both present in α, we have Ux,x+1 α = β, where β is obtained form α by removing [ux] and x + 1, v], and adding [uv] and [x, x + 1]. The action of Ux,x+1 on the vector α can be graphically represented by placing the diagram shown in Fig. 3 under the diagram for α, and read oﬀ the result using the graphical representation of the rules (i)–(iv) shown in Fig. 4. The action of the Hamiltonian is then obtained by summing over x as shown in Figs. 5 and 6. x
x+1
Fig. 3. The graphical representation of Ux,x+1
The important observation is the action of the Hamiltonian on a basis vector α yields a linear combination of basis vectors with nonpositive coefﬁcients except possibly for the coeﬃcient of α itself, which has the opposite sign resulting from the “bubble” in the graphical representation. This means that all oﬀdiagonal matrix elements are nonpositive as claimed for the matrices Ak,S in the proposition. The second will follow from the observation that Ak,S is a submatrix of (S) (S+1/2) k+1,S+1/2 . Note that the spin deviation for Vk and Vk+1 is the same, A (S+1/2)
say n. Let us order the basis elements of Vk+1 so that all α where the last vertex, k + 1, is unpaired, are listed ﬁrst, and consider the αβ matrix element of Hk+1 for such α and β. Then, it is easy to see that there are no
160
B. Nachtergaele and S. Starr
=
= 2
(i)
(ii)
= (iii)
= (iv) Fig. 4. The graphical rules (i)–(iv) for the action of Ux,x+1 on a TemperleyLieb basis vector
+
+
+
Fig. 5. Action of the Hamiltonian of the spin1/2 XXX or XXZ chain on a generalized Hulth´en bracket, for L = 4, k = 1
contributions from the k, k + 1 term in Hamiltonian, since its action results in nonzero coeﬃcients only for conﬁgurations where k + 1 belongs to an arc. This means that these matrix elements are identical to those computed for Hk for basis vectors labeled α and β obtained from α and β by dropping the last vertex, k + 1 which is unpaired. This completes the proof of Proposition 1 in the case of the pure spin 1/2 chain. Q.E.D. 3.2 The Basis for Higher Spin We are looking for a basis of the space of highest weight vectors of weight S of the spin chain with Hilbert space Hk . Equivalently, we may look for a basis of the SU (2) intertwiners D(S) → Hk . There is a graphical algebra of such intertwiners with a very convenient basis, the dual canonical basis,
Ordering of Energy Levels in Heisenberg Models and Applications
161
+
+
+
+ Fig. 6. Action of the Hamiltonian of the spin1/2 XXX or XXZ chain on a generalized Hulth´en bracket, for k = 6, n = 2
introduced by Frenkel and Khovanov [6]. This is the basis we will use, but (S) we will present it as a basis for the subspaces Vk of highest weight vectors. The state space at site x can be thought of as the symmetric part of 2sx spins 21 . We can label the 2sx + 1 states by the Ising conﬁgurations ↑↑ · · · ↑ , ↓↑ · · · ↑ , ↓↓↑ · · · ↑ , . . . , ↓↓↓ · · · ↓ . where each conﬁguration stands for the equivalence class up to reordering of all conﬁgurations with the same number of down spins. E.g., ↓↓↑↑↑ is the vector normally labelled as j, m = 5/2, 1/2, and not the tensor ↓ ⊗ ↓ ⊗ ↑ ⊗ ↑ ⊗ ↑. The states for a chain of L spins of magnitudes s1 , . . . , sL are then tensor products of these conﬁgurations. We shall call such vectors ordered Ising conﬁgurations. These tensor product vectors, in general, are not eigenvectors of the Casimir operator S, i.e., they are not of deﬁnite total spin. Suitable linear combinations that do have deﬁnite total spin are obtained by extending the Hulth´en bracket idea to arbitrary spin as follows. Start from any ordered Ising conﬁguration such that 2M = # ↑ −# ↓. Then, look for the leftmost ↓ that has a ↑ to its left, and draw an arc connecting this ↓ to the rightmost ↑, left of it. At this point, one may ignore the paired spins, and repeat the procedure until there is no remaining unpaired ↓ with an unpaired ↑ to its left. This procedure guarantees that no arcs will cross and no arc will span an unpaired spin. The result, when ignoring all paired spins, is an ordered Ising conﬁguration of a single spin. See Fig. 7 for an example of this procedure. The result is a basis for the spin chain consisting entirely of simultaneous eigenvectors of the total spin and its third component, with eigenvalues S and M , respectively. The value of M is 1/2 times the diﬀerence between the number of up spins and the number of down spins in the ordered
162
B. Nachtergaele and S. Starr
Fig. 7. Construction of a basis vector from an ordered Ising conﬁguration
Ising conﬁguration. The totalspin S is equal to S minus the number of pairs. Clearly, the highest weight vectors are then those that have no unpaired ↓, i.e., the ordered Ising conﬁguration consists exclusively of up spins. The vectors can be expanded in the tensor product basis by the following procedure: each arc is replaced by the spin singlet ↑ ⊗ ↓ − ↓ ⊗ ↑, and the unpaired spins are replaced by their tensor products. Finally, one symmetrizes in each block. Next, we brieﬂy sketch how the properties claimed in Proposition 1 can be veriﬁed. To do this we have to calculate the action of the Hamiltonian on the highest weight vector constructed in the previous paragraph. This is most easily accomplished by deriving a graphical representation for the action of each term in the Hamiltonian as we did in the case of the pure spin 1/2 chain. The Heisenberg interaction for arbitrary spins of magnitude sx and sx+1 can be realized as an interaction between spin 12 ’s making up the spin sx and sx+1 , conjugated with the projections onto the symmetric vectors. The result is the following:
−hx,x+1
1 = 2
2sx 1 ... S x · S x+1 − 1 = sx sx+1 2s
2sx+1
... 2sx+1 .
x
Here, the rectangles with label 2s represent the symmetrizing projections on the space of 2s spin 12 variables. The fundamental algebraic property that allows us to calculate the matrix elements of Hk graphically is the JonesWenzl relation (c.f., [10] and references therein): 2s ...
1
2s + 1 ... 2s
1
2s ...
2s 1 . . . 2s 2s − 1 1 + 2s + 1 . . .
2s
1
2s ... =
1
2s
1
For any element of the basis introduced above one can now compute the action of the Hamiltonian and write it as a linear combination of the same basis vectors. From the grahical rules it is easy to observe that all oﬀdiagonal matrix elements are nonpositive.
Ordering of Energy Levels in Heisenberg Models and Applications
163
(S)
As before, it is straighforward to identify the basis for Vk with a subset (S+1/2) of the basis for Vk+1 . The label of the rightmost box in any basis vector for the system k is raised by one but the number of arcs remains unchanged. The crucial property that allows us to compare the two Hamiltonians is the following. When Hk+1 acts on a basis vector obtained from a corresponding Hk vector as we have just described, the only possible new terms that are generated are oﬀdiagonal terms, which do not contain a bubble and, hence, are negative. The details of the calculation of these matrix elements and further applications will appear elsewhere [18].
4 Extensions A highly desirable extension of our main result, of course, would be the proof of Conjecture 1 for all ferromagnetic isotropic Heisenberg models on an arbitrary graph! While we have been able to prove some partial results for the spin1/2 model on an arbitrary tree and a few other graphs, we do not have an argument that works for arbitrary graphs [18]. But there are a few other directions in which one might extend the ordering of energy levels property. The aim of this section is to discus two such generalizations. In the ﬁrst, the group SU (2) is replaced by the quantum group SUq (2), 0 < q < 1. This only works on the chain and, as far as we are aware, leads to information about physically interesting models in the case of the spin 1/2 chain, namely the XXZ chain. The second generalization we consider is isotropic ferromagnetic models with higher order nearest neighbor interaction terms, such as (S x · S x+1 )2 . For this to be relevant, the spins have to be of magnitude ≥ 1. 4.1 The Spin 1/2 SUq (2)symmetric XXZ Chain It is wellknown that the translation invariant spin1/2 XXZ chain with a particular choice of boundary ﬁelds is SUq (2) invariant [21]. This SUq (2) symmetry can be exploited in much the same way as the SU (2) symmetry of the isotropic model [12, 19]. Here we will show how it leads to a natural SUq (2) analogue of the FOEL property. The Hamiltonian of the SUq (2)invariant ferromagnetic spin1/2 chain of length L ≥ 2 is given by HL = −
L−1
1 2 3 [∆−1 (Sx1 Sx+1 + Sx2 Sx+1 ) + (Sx3 Sx+1 − 1/4)]
x=1
−A(∆)(SL3 − S13 ). where ∆ > 1, and A(∆) =
1
1 − 1/∆2 2
(11)
164
B. Nachtergaele and S. Starr
This model commutes with one of the two natural representation of SUq (2) on (C2 )L , with q ∈ (0, 1), such that ∆ = (q + q −1 )/2. Concretely, this means that HL commutes with the three generators of this representation deﬁned as follows: S3 =
L
1l1 ⊗ · · · ⊗ Sx3 ⊗ 1lx+1 ⊗ · · · 1lL
x=1
S+ =
L
t1 ⊗ · · · ⊗ tx−1 ⊗ Sx+ ⊗ 1lx+1 ⊗ · · · 1lL
x=1
S− =
L
−1 1l1 ⊗ · · · ⊗ Sx− ⊗ t−1 x+1 ⊗ · · · tL
x=1
where t = q −2S = 3
q −1 0 0 q
.
The SUq (2) commutation relations are q 2S − q −2S [S , S ] = q − q −1 3
±
±
[S , S ] = ±S , 3
3
−
+
.
Note that one recovers the SU (2) deﬁnitions and commutation relations in the limit q → 1. HL also commutes with the Casimir opeator for SUq (2), given by (qT )−1 + qT , T = t ⊗ t ⊗ ··· ⊗ t . C = S+S− + (q −1 − q)2 The eigenvalues of C are q −(2S+1) + q 2S+1 , (q −1 − q)2
S = 0, 1/2, 1, 3/2, . . .
and play the same role as S for the XXX model, e.g., they label the irreducible representations of SUq (2). The eigenspaces of C are invariant subspaces of HL and, as before, we denote the smallest eigenvalues of HL restricted to these invariant subspaces by E(HL , S). Note that the subspaces depend on q, but their dimensions are constant for 0 < q ≤ 1. Theorem 4. E(HL , S + 1) < E(HL , S),
for all S ≤ L/2 − 1 .
The proof of this theorem is identical to the one for the isotropic spin1/2 chain up to substitution of the singlet vector ξ by the SUq (2) singlet ξq = q + ⊗ − − − ⊗ +, and changing the scalar value of the “bubble” to −(q + q −1 ). The details are given in [17].
Ordering of Energy Levels in Heisenberg Models and Applications
165
4.2 Higher Order Interactions For spins of magnitude greater than 1/2 the Heisenberg interaction is not the only SU (2) invariant nearest neighbor interactions. It is easy to show that the most general SU (2) invariant interaction of two spins of magnitudes s1 and s2 , i.e., any hermitian matrix commuting with the representation D(s1 ) ⊗ D(s2 ) , is an arbitrary polynomial of degree ≤ 2 min{s1 , s2 } in the Heisenberg interaction with real coeﬃcients: 2 min{s1 ,s2 }
h12 =
J (m) (S 1 · S 2 )m .
(12)
m=0
The deﬁnition of the FOEL property only uses SU (2)invariance and therefore applies directly to any Hamiltonian for a quantum spin system on a graph with at each edge an interaction of the form (12). We believe it is possible to determine the exact range of coupling constants J (m) such that FOEL holds for spin s chains with translation invariant interactions. So far, we have carried this out only for the spin1 chain. Theorem 5. FOEL holds for the spin1 chains with Hamiltonian HL =
L−1
(1 − S x · S x+1 ) + β(1 − S x · S x+1 )2 )
x=1
with 0 ≤ β ≤ 1/3. Level crossings occur at β = 1/3 and FOEL does not hold, in general, for β > 1/3. The overall method of proof is the same as for the standard Heisenberg model. Theorem 2 applies directly since only the SU (2) symmetry is used in its proof. The only diﬀerence is in the proof of Proposition 1. The same basis for the highest weight spaces is used but verifying the signs of the matrix elements is more involved.
5 Applications In this section we discuss a number of results that are either consequences of the FOEL property, or other applications of the properties of the Heisenberg Hamiltonian that allowed us to prove FOEL. 5.1 Diagonalization at Low Energy The most direct applications of the FOEL property are its implications for the lowlying spectrum of the Hamiltonian. FOEL with strict inequality implies that the ground states are the multiplet of maximal spin which, of course,
166
B. Nachtergaele and S. Starr
is not a new result. Since the maximal spin multiplet is unique, the ﬁrst excited state must belong to less than maximal spin and therefore, by FOEL, to Smax − 1. In the case of models for which this second eigenvalue can be computed, such as translation invariant models on a lattice, this is again consistent with a wellknown fact, namely that the lowest excitation are simple spinwaves. But in the case of arbitrary coupling constants and spin magnitudes in one dimension it proves that the ﬁrst excited state is represented in the subspace of “one overturned spin” (with respect to the fully polarized 3 − 1, which is a new result. ground state), i.e., S 3 = Smax More generally, the FOEL property can help with determining the spectrum of the Heisenberg model at low energies, whether by numerical or other means, in the following way. Suppose H is a Hamiltonian with the FOEL property. Diagonalize H in the subspaces H(Smax −n) , for n = 0, 1, . . . N , and select those eigenvalues that are less or equal than E(H, Smax − N ). It is easy to see that the FOEL property implies that this way you have obtained all eigenvalues of the full H below ≤ E(H, Smax − N ). This is interesting because you only had to diagonalize the Hamiltonian in an invariant subspace that is explicitly known (by the representation theory of SU (2)) and of relatively low dimension: dim(H(Smax −n) ) is O(Ln ), while the full Hilbert space has dimension (2J + 1)L for L spin J variables. 5.2 The Ground States of Fixed Magnetization for the XXZ Chain The spin 1/2 XXZ ferromagnetic chain with suitable boundary conditions, or deﬁned on the appropriate inﬁnitechain Hilbert space has lowenergy states that can be interpreted as welldeﬁned magnetic domains in a background of opposite magnetization [11, 19]. Using the techniques we have used for proving FOEL, we can rigorously determine the dispersion relation of a ﬁnite droplet of arbitrary size. The spin 1/2 XXZ chain can, in principle, be diagonalized using the Bethe Ansatz [13]. There are two complications that may prevent one from obtaining the desired information about its spectrum. The ﬁrst is that a complete proof of completeness of the Bethe Ansatz eigenstates has been obtained and published only for the XXX chain (q = ∆ = 1), although the corresponding result for the XXZ chain has been announced quite some time ago [7]. The second problem is that the eigenvalues are the solutions of complicated sets of equations, such that proving statements as the one we discuss here, may be very hard. For brevity, let us consider the XXZ Hamiltonian for the iniﬁnite chain deﬁned on the Hilbert space generated by the orthonormal set of vectors representing n down spins in an inﬁnite “sea” of up spins, and let us denote this space by Hn . Deﬁne E(n) = inf spec(HHn ) .
Ordering of Energy Levels in Heisenberg Models and Applications
167
As before, the relation between q ∈ (0, 1) and the anisotropy parameter ∆ in the XXZ Hamiltonian (11) is given by ∆ = (q + q −1 )/2. Theorem 6. For n ≥ 1, we have E(n) =
(1 − q 2 )(1 − q n ) . (1 + q 2 )(1 + q n )
Moreover, E(n) belongs to the continuous spectrum and is the bottom of a band of width 1 − q2 . 4q n (1 + q n )(1 − q n ) The states corresponding to this band can be interpreted as a droplet of size n with a deﬁnite momentum. The formula for the width indicates that the “mass” of a droplet diverges as n → ∞. The proof of this result will appear in a separate paper [16]. If one looks back at the ﬁnitevolume eigenvalues E(Hk , k/2−n) that converge to the bottom of the band in the inﬁnitevolume limit (k → ∞, n ﬁxed), the property (iv) amounts to property that the ground state energy of a droplet of ﬁxed size n is strictly monotone decreasing in the volume. Moreover the ﬁnitevolume eigenvalues can be related to E(n), in the above limit, by using the generalization of the PerronFrobenius result stated in Theorem 3. 5.3 Aldous’ Conjecture for the Symmetric Simple Exclusion Process The Symmetric Simple Exclusion Process (SSEP) is a Markov process deﬁned on particle conﬁgurations on a ﬁnite graph Λ. For our purposes it is convenient to deﬁne the process as a semigroup on HΛ ∼ = l2 (ΩΛ ), where ΩΛ is the space of conﬁgurations η : Λ → {0, 1}. One thinks of η(x) = 1 to indicate the presence of a particle at the vertex x. Let L be deﬁned in HΛ by the formula (Lf )(η) = Jxy (f (η) − f (η xy )) (13) x∼y η
where η xy denotes the conﬁguration obtained form η by interchanging the values of η(x) and η(y). The parameters Jxy are positive numbers representing the jump rates at the edges x ∼ y ∈ Λ. Clearly, the number of particles is a conserved quantity of the process. Concretely, this means that HΛ decomposes into a direct sum of invariant (n) (n) subspaces HΛ , n = 0, . . . , Λ, where HΛ consists of all " functions supported on conﬁgurations η that have extactly n particles, i.e., x η(x) = n. 0 is a simple eigenvalue of each of the restrictions LH (n) , and L is nonnegative Λ deﬁnite. For each n, the corresponding invariant measure is the uniform distribution on nparticle conﬁgurations.
168
B. Nachtergaele and S. Starr
Let λ(n) denote the smallest positive eigenvalue of LH (n) . Since the dyΛ
namics of the SSEP is given by the semigroup {e−tL }t≥0 , λ(n), determines the speed of relaxation to the invariant measure. The following conjecture is known as Aldous’ Conjecture but on his website [1] Aldous states that it arose in a conversation with Diaconis. So, maybe it should be called the AldousDiaconis Conjecture. Conjecture 2. λ(n) = λ(1),
for all 1 ≤ n ≤ Λ − 1 .
Apart from being a striking property, namely that the relaxation rate should be independent of the number of particles, it could also be very useful. The SSEP for one particle is just a random walk on the graph Λ, and many powerful techniques are available to study the relaxation rate of random walks. In physical terms one would say that the conjectured property reduces the manybody problem of ﬁnding the relaxation rate for n particles to a single oneparticle problem. Proposition 2. If the ferromagnetic spin1/2 Heisenberg model with coupling constants Jxy on a graph Λ satisﬁes FOEL, then Conjecture 2 holds for the SSEP on Λ with jump rates Jxy /2. Proof. The proof is based on the unitary equivalence of L and the ferromagnetic spin 1/2 Heisenberg Hamiltonian H given by 1 1l − S x · S y . Jxy H= 4 x∼y∈Λ
The unitary transformation U : L2 (ΩΛ ) → HΛ = (C2 )⊗Λ , that relates L and H is explicitly given by f (η) η , where Sx3 η = (ηx − 1/2) η . L2 (ΩΛ ) f → U f = ψ = η
To see this note that 1/4 − S x · S y = (1 − txy )/2 , where txy interchanges the states at x and y in any tensor product vector. Then 1 f (η)Jxy (1 − txy ) η Hψ = 2 x∼y η 1 = Jxy (f (η) − f (η xy )) η 2 x∼y η 1 = (Lf )(η) η . 2 η
Ordering of Energy Levels in Heisenberg Models and Applications
169
Therefore, HU f = U Lf , for all f ∈ L2 (ΩΛ ). Under this unitary transformation, the particle number becomes the third component of the total spin: 3 = −Λ/2 + n . Stot
The unique invariant measure of SSEP for n particles is the uniform measure " on {η ∈ ΩΛ  x ηx = n}. The corresponding state for the spin model belongs to the unique multiplet of maximal total spin, i.e., is a ground state. λ(n) is the next eigenvalue of H is the same value of total S 3 . Since the ﬁrst excited state of H, by FOEL, is a multiplet of total spin Smax − 1, this eigenvalue has an eigenvalue with any value of S 3 in the range −Smax + 1, . . . , Smax − 1. We have Smax = Λ/2. Therefore, this corresponds to the range 1 ≤ n ≤ Λ − 1. Hence, λ(n) is independent of n in this range. In combination with Theorem 1, this proposition has the following corollary. Corollary 1. Conjecture 2 holds for chains. Our partial result for trees (not discussed here) also implies Conjecture 2 for arbitrary ﬁnite trees as well as some graphs derived from trees. These cases of the AldousDiaconis conjecture were previously know [2, 8], as well as some other examples where one can compute λ(n) exactly [3–5]. Needless to say, a full proof of FOEL, the AldousDiaconis Conjecture, or even a proof for additionial special cases, would be of great interest. An interesting direction for generalization considered by Aldous is to also establish the analogous formula for the spectral gap for a cardshuﬄing model with full SU (n) symmetry, which restricts to the SSEP when one considers cards of only two colors.
Acknowledgement This work was supported in part by the National Science Foundation under Grant # DMS0303316. B.N. also thanks the Erwin Schr¨ odinger Institute, Vienna, where part of this work was done, for ﬁnancial support and the warm and eﬃcient hospitality it oﬀers.
References 1. D. Aldous, http://statwww.berkeley.edu/users/aldous/problems.ps. 2. R. Bacher, Valeur propre minimale du laplacien de Coxeter pour le groupe sym´etrique, J. Algebra 167 (1994), 460–472. 3. P. Diaconis and L. SaloﬀCoste, Comparison techniques for random walk on ﬁnite groups, Ann. Prob. 21 (1993), 2131–2156.
170
B. Nachtergaele and S. Starr
4. P. Diaconis and M. Shahshahani, Generating a random permutation with random transpositions, Z. Wahrsch. Verw. Geb. 57 (1981), 159–179. 5. L. Flatto, A.M. Odlyzko, and D.B. Wales, Random shuﬄes and group representations, Ann. Prob. 13 (1985), 154–178. 6. I.B. Frenkel and M.G. Khovanov, Canonical bases in tensor products and graphical calculus for uq (sl2 ), Duke Math. J. 87 (1997), 409–480. 7. E. Gutkin, Plancherel formula and critical spectral behaviour of the inﬁnite XXZ chain, Quantum symmetries (Clausthal, 1991), World Scientiﬁc, River Edge, NJ, 1993, pp. 84–98. 8. S. Handjani and D. Jungreis, Rate of convergence for shuﬄing cards by transpositions, J. Theor. Prob. 9 (1996), 983–993. 9. M. Jimbo and T. Miwa, Algebraic analysis of solvable lattice models, Regional Conference Series in Mathematics, American Mathematical Society, Providence, RI, 1995. 10. L.H. Kauﬀman and S.L. Lins, TemperleyLieb recoupling theory and invariants of 3manifolds, Princeton University Press, 1994. 11. T. Kennedy, Expansions for droplet states in the ferromagnetic XXZ Heisenberg chain, Markov Processes and Rel. Fields 11 (2005) 223–236 12. T. Koma and B. Nachtergaele, The spectral gap of the ferromagnetic XXZ chain, Lett. Math. Phys. 40 (1997), 1–16. 13. V.E. Korepin, N.M. Bogoliubov, and A.G. Izergin, Quantum inverse scattering method and correlation functions, Cambridge Monographs on Mathematical Physics, Cambridge University Press, Cambridge, England, 1993. 14. E.H. Lieb, Two theorems on the Hubbard model, Phys. Rev. Lett. 62 (1989), 1201–1204. 15. E.H. Lieb and D. Mattis, Ordering energy levels of interacting spin systems, J. Math. Phys. 3 (1962), 749–751. 16. B. Nachtergaele, W. Spitzer, and S. Starr, Droplet Excitations for the Spin1/2 XXZ Chain with Kink Boundary Conditions, arXiv::mathph/0508049. , Ferromagnetic ordering of energy levels, J. Stat. Phys. 116 (2004), 17. 719–738. 18. B. Nachtergaele and S. Starr, in preparation. , Droplet states in the XXZ Heisenberg model, Commun. Math. Phys. 19. 218 (2001), 569–607, mathph/0009002. 20. B. Nachtergaele and S. Starr, Ferromagnetic LiebMattis theorem, Phys. Rev. Lett. 94 (2005), 057206, arXiv:mathph/0408020. 21. V. Pasquier and H. Saleur, Common structures between ﬁnite systems and conformal ﬁeld theories through quantum groups, Nucl. Phys. B330 (1990), 523– 556. 22. H.N.V. Temperley and E.H. Lieb, Relations between the “percolation” and “colouring” problem and other graphtheoretical problems associated with regular planar lattices: some exact results for the “percolation” problem, Proc. Roy. Soc. A322 (1971), 252–280. 23. H. Wielandt, Unzerlegbare, nicht negative Matrizen, Math. Z. 52 (1950), 642– 648.
Interacting Fermions in 2 Dimensions Vincent Rivasseau Laboratoire de Physique Th´eorique, CNRS, UMR 8627, Universit´e de ParisSud, 91405 Orsay [emailprotected]
Abstract. We provide an introduction to the constructive results on interacting Fermions in two dimensions at thermal equilibrium above the critical temperature of pair condensation.
1 Introduction Do interacting Fermions in 2 dimensions (above any lowtemperature phase) resemble more three dimensional Fermions, i.e. the Fermi liquid, or one dimensional Fermions, i.e. the Luttinger liquid? The short answer to this controversial question is that it depends on the shape of the Fermi surface. Interacting Fermions with a round Fermi surface behave more like three dimensional Fermi liquids, whether interacting Fermions with the square Fermi surface of the Hubbard model at halfﬁlling behave more like a onedimensional Luttinger liquid. This statement has been now proved in full mathematical rigor, beyond perturbation theory, using the mathematically precise criterion of Salmhofer. In this lecture we give an introduction to this result, obtained in the series of papers [1]– [5]. I also take this occasion to thank my colleagues and friends M. Disertori, S. Afchain and J. Magnen for their very pleasant collaborations on this subject.
2 Fermi Liquids and Salmhofer’s Criterion The textbooks deﬁnition of condensed matter physicists [6] for Fermi liquid simply does not work for interacting models which are parity invariant i.e. in absence of magnetic ﬁeld. Indeed the textbook deﬁnition states that at zero temperature an interacting Fermi liquid should exhibit a discontinuity in its density of states at a certain Fermi surface, just like the free Fermi liquid. A curve in the style of Fig. 1 is usually shown as typical of such a liquid. Many condensed matter theorists know that this deﬁnition is at best a “ﬁgment of imagination” to use Professor Anderson’s words. Indeed KohnLuttinger singularities are known to be generic. This means that for any ﬁxed bare generic interaction, going towards zero temperature at ﬁxed interaction V. Rivasseau: Interacting Fermions in 2 Dimensions, Lect. Notes Phys. 690, 171–178 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
172
V. Rivasseau
Σ
pF
p
Fig. 1. The wrong picture, still found in many textbooks, for the alleged discontinuity of the selfenergy Σ at the Fermi surface pF
cannot probe the famous discontinuity supposed to deﬁne the Fermi liquid phase: the system instead undergoes a transition into a diﬀerent condensed phase (such as the BCS phase). To conclude from this fact that interacting Fermi liquids simply do not exist would be wrong because condensed matter physicists tell us that they know “deep in their bones” that Fermi liquids do “exist” physically. . . So a better solution is to work out a correct deﬁnition, that will allow interacting Fermi liquids to exist mathematically as well. Such a mathematically correct formulation is ultimately needed for the theory of Fermi liquids just like for thermodynamic limits, Gibbs states or phase transitions. It was found by M. Salmhofer a few years ago [7]– [8]. He proposed to characterize the Fermi liquid behavior by moving a system simultaneously towards λ = T = 0, so as to remain above any low temperature phase Kohn Luttinger singularities. This can be accomplished for instance by sending λ and T both to 0 along a curve λ log T p = c, where c is constant and p is an integer, typically 1 or 2 (see Fig. 2). Of course no singularity will be met, but Salmhofer proposed to call the system a Fermi liquid if the second derivative of the selfenergy, as function of the momentum, remains bounded along such a curve [7]. This criterion works at thermal equilibrium and it has been chosen so that it is not obeyed by the one dimensional Luttinger liquid, for which indeed this second derivative along such a curve blows up. To discuss the physical relevance of Salmhofer’s criterion would be too long, and I do not feel competent at all, in particular for the experimental aspects (is it possible for example to turn knobs to move at will both temperature and interaction in as Fermion system? may be yes, by varying temperature and pressure?...) I just remark that this rigorous criterion is certainly up to now the best for parity invariant interacting systems, at least because it is the only one.
Interacting Fermions in 2 Dimensions
173
T Τc = e
a
−1/λ
Τc = e
λ
−1/λ
b
Fig. 2. (a): Lowering the temperature at ﬁxed coupling; (b): Lowering the temperature according to Salmhofer
3 The Models We consider a gas of Fermions in thermal equilibrium at temperature T , with coupling constant λ. The free propagator for these models is Cˆa,b (k) = δa,b
1
→ , − ik0 − e( k )
(1)
→ − → − → − with e( k ) = ( k ) − µ, ( k ) being the kinetic energy and µ the chemical potential. The relevant values for k0 are discrete and called the Matsubara frequencies : π k0 = (2n + 1), n ∈ Z , (2) β where β = (kT )−1 . When T → 0+ (which means β → +∞), k0 becomes a continuous variable, the corresponding discrete sum becomes an integral, and the corresponding propagator C0 (x) becomes singular on the Fermi surface deﬁned by k0 = 0 → − and e(k) = 0. This Fermi surface depends on the kinetic energy ( k ) of the → − → − model. For rotation invariant models, ( k ) = k 2 /2m and the Fermi surface is √ a circle in two dimensions and a sphere in three dimensions, with radius 2mµ. These two rotation invariant models, or jellium models are respectively nicknamed J2 and J3 . For the halfﬁlled Hubbard model, nicknamed → − H2 , x lives on the lattice ZZ2 , and ( k√) = cos k1 + cos k2 so that at µ = 0 the Fermi surface is a square of side size 2π, joining the points (±π, 0), (0, ±π) in the ﬁrst Brillouin zone. It is also possible to interpolate continuously between H2 and J2 by varying the ﬁlling factor of the Hubbard model. Lattice models with nextnearest neighbor hopping are also interesting. In contrast with the propagator, the interaction is almost unique. Indeed we are interested in longrange physics, so we should start from a quasilocal bare interaction. But there is a unique exactly local such interaction, namely 2 d3 x ψ a (x)ψa (x) , (3) SV = λ V
a∈{↑,↓}
174
V. Rivasseau
where V := [−β, β[×V and V is an auxiliary volume cutoﬀ in two dimensional space, that will be sent to inﬁnity in the thermodynamic limit. Indeed any local polynomial of higher degree is zero since Fermionic ﬁelds anticommute.
4 A Brief Review of Rigorous Results What did the programs of rigorous mathematical study of interacting Fermi systems accomplish until now? Recall that in dimension 1 there is neither superconductivity nor extended Fermi surface, and Fermion systems have been proved to exhibit Luttinger liquid behavior [9]. The initial goal of the studies in two or three dimensions was to understand the low temperature phase of these systems, and in particular to build a rigorous constructive BCS theory of superconductivity. The mechanism for the formation of Cooper pairs and the main technical tool to use (namely the corresponding 1/N expansion, where N is the number of sectors which proliferate near the Fermi surface at low temperatures) have been identiﬁed [10]. But the goal of building a completely rigorous BCS theory ab initio remains elusive because of the technicalities involved with the constructive control of continuous symmetry breaking. So the initial goal was replaced with a more modest one, still important in view of the controversies over the nature of two dimensional “Fermi liquids” [11], namely the rigorous control of what occurs before pair formation. The last decade has seen excellent progress in this direction. As is well known, suﬃciently high magnetic ﬁeld or temperature are the two diﬀerent ways to break the Cooper pairs and prevent superconductivity. Accordingly two approaches were devised for the construction of “Fermi liquids”. One is based on the use of nonparity invariant Fermi surfaces to prevent pair formation. These surfaces occur physically when generic magnetic ﬁelds are applied to two dimensional Fermi systems. The other is based on Salmhofer’s criterion [7], in which temperature is the cutoﬀ which prevents pair formation. In a large series of papers [12], the construction of two dimensional Fermi liquids for a wide class of nonparity invariant Fermi surfaces has been completed in great detail by Feldman, Kn¨ orrer and Trubowitz. These papers establish Fermi liquid behavior in the traditional sense of physics textbooks, namely as a jump of the density of states at the Fermi surface at zero temperature, but they do not apply to the simplest Fermi surfaces, such as circles or squares, which are parity invariant. An other program in recent years was to explore which models satisfy Salmhofer’s criterion. Of particular interest to us are the three most “canonical” models in more than one dimension namely: • the jellium model in two dimensions, with circular Fermi surface, nicknamed J2 ,
Interacting Fermions in 2 Dimensions
175
• the halfﬁlled Hubbard model in two dimensions, with square Fermi surface, nicknamed H2 , • and the jellium model in three dimensions, with spherical Fermi surface, nicknamed J3 . The study of each model has been divided into two main steps of roughly equal diﬃculty, the control of convergent contributions and the renormalization of the two point functions. In this sense, ﬁve of the six steps of our program are now completed. J2 is a Fermi liquid in the sense of Salmhofer [1]– [2], H2 is not, and is a Luttinger liquid with logarithmic corrections, according to [3, 4] and [5]. Results similar to [1]  [2] have been also obtained for more general convex curves not necessarily rotation invariant such as those of the Hubbard model at low ﬁlling, where the Fermi surface becomes more and more circular, including an improved treatment of the four point functions leading to better constants [13]. Therefore as the ﬁlling factor of the Hubbard model is moved from halfﬁlling to low ﬁlling, we conclude that there must be a crossover from Luttinger liquid behavior to Fermi liquid behavior. This solves the controversy [11] over the Luttinger or Fermi nature of twodimensional manyFermion systems above their critical temperature. Up to now only the convergent contributions of J3 , which is almost certainly a Fermi liquid, have been controlled [14]. The renormalization of the two point functions for J3 , the last sixth of our program, remains still to be done. This last part is diﬃcult since the cutoﬀs used in [14] do not conserve momentum. This means that the two point functions that have to be renormalized in this formalism are not automatically one particle irreducible, as is the case both in [2] and in this paper. This complicates their analysis.
5 Multiscale Analysis, Angular Sectors For any twodimensional model built until now in the constructive sense, the strategy is the same. It is based on some kind of multiscale expansion, which keeps a large fraction of the theory in unexpanded determinants. The global bound on these determinant (using determinant inequalities such as Gram inequality) is much better than if the determinant was expanded into Feynman graphs which would then be bounded one by one, and the bounds summed. The bound obtained in this way would simply diverge at large order (i.e. not prove any analyticity at all in the coupling constant) simply because there are too many Feynman graphs at large order. But the divergence of a bound does not mean the divergence of the true quantity if the bound is bad. Constructive analysis, which “keeps loops unexpanded” is the correct way to obtain better bounds, which do prove that the true series in fact does not diverge, i.e. has a ﬁnite convergence radius in the coupling constant. This
176
V. Rivasseau
radius howerver shrink when the temperature goes to 0, and a good constructive analysis should establish the correct shrinking rate, which is logarithmic. This is were multiscale rather than single scale constructive analysis becomes necessary. The basic idea of the multiscale analysis is to slice the propagator according to the size of its denominator so that the slice number i corresponds to → − k0 − e( k ) M −i , where M is some ﬁxed constant. This multiscale analysis is supplemented within each scale by an angular “sector analysis”. The number of sectors should be kept as small as possible, so each sector should be as large as possible in the directions tangent to the Fermi surface in three dimensions, or to the Fermi curve in two dimensions. What limits however the size of these sectors is the curvature of the surface, so that stationary phase method could still relate the spatial decay of a propagator within a sector to its dual size in momentum space. In the case of a circle, the number of sectors at distance ε of the singularity grows therefore at least like ε−1/2 , hence like a power of ε. However for the halfﬁlled Hubbard model, since the curvature is “concentrated at the corners” the number of sectors grows only like  log ε. In one dimension there are really only two sectors since the Fermi singularity is made of two points. A logarithm is closer to a constant than to a power; this observation is the main reason for which the halfﬁlled Hubbard model is closer to the onedimensional Luttinger liquid than to the three dimensional Fermi liquid. Momentum conservation rules for sectors which meet at a given vertex in general are needed to ﬁx the correct power counting of the subgraphs of the model. In the Hubbard case at half ﬁlling, these rules are needed only to ﬁx the correct logarithmic power counting, since the growth of sectors near the singularity is only logarithmic. In both cases the net eﬀect in two dimensions of these conservation rules is to roughly identify two pairs of “conserved” sectors at any vertex, so that in each slice the model resembles an N component vector model, where N is the number of sectors in the slice. The multiscale renormalization group analysis of the model then consists essentially in selecting, for any graph, a tree which is a subtree in each of the “quasilocal” connected components of the graph according to the momentum slicing. These connected components are those for which all internal lines are farther from the Fermi surface than all external lines. The selection of this tree can be performed in a constructive manner, keeping the remaining loop ﬁelds in a determinant. The combinatoric diﬃculty related to the fact that a graph contains many trees has been tackled once and for all thanks to forest formulas such as those of [15].
6 One and Two Particle Irreducible Expansions Salmhofer’s criterion is stated for the selfenergy, i.e. the sum of all oneparticle irreducible graphs for the two point function. Its study requires the
Interacting Fermions in 2 Dimensions
177
correct renormalization of these contributions. Since angular sectors in a graph may vary from one propagator to the next in a graph, and since diﬀerent sectors have diﬀerent decays in diﬀerent directions, we are in a delicate situation. In order to prove that renormalization indeed does the good that it is supposed to do, one cannot simply rely on the connectedness of these selfenergy graphs, but one must use their particle irreducibility explicitly. So the proof requires a constructive particle irreducible analysis of the selfenergy. The following theorem summarizes the results of [1]DR2 Theorem 1. The radius of convergence of the jellium twodimensional model perturbative series for any thermodynamic function is at least c/ log T , where T is the temperature and c some numerical constant. As T and λ jointly tend to 0 in this domain, the selfenergy and its ﬁrst two momentum derivatives remain uniformly bounded so that the model is a Fermi liquid in the sense of Salmhofer. In the case of the jellium model J2, this analysis can be performed at the level of oneparticle irreducible graphs [2]. The halfﬁlled Hubbard model, however, is more diﬃcult. It came as a surprise to us that although there is no real divergence of the selfenergy (the associated counterterm is zero thanks to the particle hole symmetry of the model at halfﬁlling) one really needs a twoparticle and onevertex irreducible constructive analysis to establish the necessary constructive bounds on the selfenergy and its derivatives [4]. For parity reasons, the selfenergy graphs of the model are in fact not only oneparticle irreducible but also two particle and one vertex irreducible, so that this analysis is possible. This analysis according to the line form of Menger’s theorem ( [16]) leads to the explicit construction of three linedisjoint paths for every selfenergy contribution, in a way compatible with constructive bounds. On top of that analysis, another one which is scaledependent is performed: after reduction of some maximal subsets provided by the scale analysis, two vertexdisjoint paths are selected in every selfenergy contribution. This requires a second use of Menger’s theorem, now in the vertex form. This construction allows to improve the power counting for two point subgraphs, exploiting the particlehole symmetry of the theory at halfﬁlling, and leads to the desired analyticity result. Finally an upper bound for the self energy second derivative is combined with a lower bound for the explicit leading self energy Feynman graph [5]. This completes the proof that the Hubbard model violates Salmhofer’s criterion, hence is not a Fermi liquid, in contrast with the jellium two dimensional model. More precisely the following theorem summarizes the results of [3–5]. Theorem 2. The radius of convergence of the Hubbard model perturbative series at halfﬁlling is at least c/ log2 T , where T is the temperature and c some numerical constant. As T and λ jointly tend to 0 in this domain, the
178
V. Rivasseau
selfenergy of the model does not display the properties of a Fermi liquid in the sense of Salmhofer, since the second derivative is not uniformly bounded.
References 1. M. Disertori and V. Rivasseau, Interacting Fermi liquid in two dimensions at ﬁnite temperature, Part I: Convergent Attributions, Commun. Math. Phys. 215, 251 (2000). 2. M. Disertori and V. Rivasseau, Interacting Fermi liquid in two dimensions at ﬁnite temperature, Part II: Renormalization, Commun. Math. Phys. 215, 291 (2000). 3. V. Rivasseau, The two dimensional Hubbard Model at halfﬁlling: I. Convergent Contributions, Journ. Stat. Phys. 106, 693–722, (2002). 4. S. Afchain, J. Magnen and V. Rivasseau, Renormalization of the 2point function of the Hubbard Model at halfﬁlling, condmat/0409231 (2004), to appear in Annales Henri Poincar´e. 5. S. Afchain, J. Magnen and V. Rivasseau, The Hubbard Model at halfﬁlling, part III: the lower bound on the selfenergy, condmat/0412401 (2004), to appear in Annales Henri Poincar´e. 6. See e.g. E. Fradkin, “Field Theories of Condensed Matter Systems”, Frontiers in Physics, 1991 7. M. Salmhofer, Continuous renormalization for Fermions and Fermi liquid theory, Commun. Math. Phys. 194, 249 (1998). 8. M. Salmhofer, Renormalization, an introduction, Springer Verlag, 1999. 9. G. Benfatto and G. Gallavotti, Renormalization Group, Physics Notes, Chap. 11 and references therein, Vol. 1 Princeton University Press, 1995 10. J. Feldman, J. Magnen, V. Rivasseau and E. Trubowitz, An Intrinsic 1/N Expansion for Many Fermion System, Europhys. Letters 24, 437 (1993). 11. P.W. Anderson, Luttinger liquid behavior of the normal metallic state of the 2D Hubbard model, Phys Rev Lett. 64 1839–1841 (1990). 12. Joel Feldman, H. Kn¨ orrer and E. Trubowitz, A two dimensional Fermi Liquid, series of papers in Commun. Math. Phys. 247, 1–319, 2004 and Reviews in Math. Physics, 15, 9, 949–1169, (2003). 13. G. Benfatto, A. Giuliani and V. Mastropietro, Low temperature Analysis of TwoDimensional Fermi Systems with Symmetric Fermi surface, Ann. Henri Poincar´e, 4 137, (2003). 14. M. Disertori, J. Magnen and V. Rivasseau, Interacting Fermi liquid in three dimensions at ﬁnite temperature, part I: Convergent Contributions, Ann. Henri Poincar´e, 2 733–806 (2001). 15. A. Abdesselam and V. Rivasseau, Trees, Forests and Jungles: A Botanical Garden for Cluster Expansions, in “Constructive Physics”, LNP 446, Springer Verlag, 1995. 16. J.A. Bondy and U.S.R. Murty, Graph theory with applications, NorthHolland Editions, 1979.
On the Essential Spectrum of the Translation Invariant Nelson Model Jacob SchachMøller Aarhus University, Department of Mathematical Sciences, 8000 Aarhus C, Denmark [emailprotected] Abstract. Let Rν ξ → Σess (ξ) denote the bottom of the essential spectrum for the ﬁber Hamiltonians of the translation invariant massive Nelson model, which describes a νdimensional electron linearly coupled to a scalar massive radiation ﬁeld. We prove that, away from a locally ﬁnite set, Σess is an analytic function of total momentum.
1 The Model and the Result Let hph := L2 (Rνk ) and F = Γ (hph ) denote the bosonic Fock space constructed from hph . We write p = −i∇x for the momentum operator in K := L2 (Rνx ). The translation invariant Nelson Hamiltonian describing a νdimensional electron (or positron) linearly coupled to a massive scalar radiation ﬁeld has the form H := Ω(p) ⊗ 1lF + 1lK ⊗ dΓ (ω) + V , on K ⊗ F , where
V := Rν
e−ik·x v(k) 1lK ⊗ a∗ (k) + eik·x v(k) 1lK ⊗ a(k) dk .
We assume that the form factor v satisﬁes v ∈ L2 (Rνk ) , v real valued , v = 0 a.e. and ∀O ∈ O(ν) : v(Ok) = v(k) a.e. ,
(1)
which implies a UVcutoﬀ. Here O(ν) denotes the orthogonal group. The physically interesting relations Ω and ω are Ω(η) =
choices for the dispersion √ η 2 /2M , Ω(η) = η 2 + M 2 and ω(k) = k 2 + m2 , where M, m > 0 are the electron and boson masses. We will however work with general forms of both Ω and ω. As for ω, this is partly motivated by the similarity with the Polaron model, cf. [5, 11], where ω is not explicitly known. We make no attempt here to say anything about the Polaron model. The operator H commutes < with the total momentum p ⊗ 1lF + 1lK ⊗ dΓ (k) and hence ﬁbers as H ∼ Rν H(ξ)dξ, where the ﬁber Hamiltonians H(ξ), ξ ∈ Rν , are operators on F given by J.S. Møller: On the Essential Spectrum of the Translation Invariant Nelson Model, Lect. Notes Phys. 690, 179–195 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
180
J. SchachMøller
H(ξ) = H0 (ξ) + Φ(v) where H0 (ξ) = dΓ (ω) + Ω(ξ − dΓ (k)) and the interaction is
Φ(v) =
. v(k) a∗ (k) + v(k) a(k) dk .
(2)
(3)
Rν
We formulate precise assumptions on Ω and ω, which are satisﬁed by the examples mentioned above. We use the standard notation t := (1 + t2 )1/2 . Condition 1 (The particle dispersion relation) Let Ω ∈ C ∞ (Rν ). There exists sΩ ∈ {0, 1, 2} such that (i) (ii) (iii) (iv)
There exists C > 0 such that Ω(η) ≥ C −1 ηsΩ − C. For any multiindex α there exists Cα such that ∂ α Ω(η) ≤ Cα ηsΩ −α . Ω is rotation invariant, i.e., Ω(Oη) = Ω(η), for all O ∈ O(ν). The function η → Ω(η) is real analytic.
Condition 2 (The photon dispersion relation) Let ω ∈ C ∞ (Rν ) satisfy (i) (ii) (iii) (iv) (v) (vi)
There exists m > 0, the photon mass, such that inf k∈Rν ω(k) = ω(0) = m. ω(k) → ∞, in the limit k → ∞. There exist sω ≥ 0, Cω > 0, and for any multiindex α with α ≥ 1, a Cα such that ω(k) ≥ Cω−1 ksω − Cω and ∂kα ω(k) ≤ Cα ksω −α . ω is rotation invariant, i.e., ω(Oη) = ω(η), for all O ∈ O(ν). ω is real analytic. ω is strictly subadditive, i.e. ω(k1 ) + ω(k2 ) > ω(k1 + k2 ) for all k1 , k2 ∈ Rν .
Remarks: (1) The assumption that the photons are massive is essential. (2)√ One could weaken the assumption v ∈ L2 (Rν ) by taking instead v/ ω ∈ L2 (Rν ). This is a weaker ultraviolet condition, which still allows for the construction of the Hamiltonian. See [2]. (3) The subadditivity assumption is discussed at the end of this section. (4) Condition 2 vi) follows from subadditivity ω(k1 ) + ω(k2 ) ≥ ω(k1 + k2 ) together with Condition 2 (i), (iv), and (v). (5) The assumptions (1), Condition 1 (ii), Condition 2 (i), (ii), and iii) can be relaxed, cf. [10]. We introduce the bottom of the spectrum and essential spectrum as functions of total momentum Σ0 (ξ) := inf σ H(ξ) and Σess (ξ) := inf σess H(ξ) . The energy of a system of n noninteracting bosons, with momenta k ∈ Rnν , k = (k1 , . . . , kn ), and one interacting electron with momentum ξ −k (n) , where k (n) = k1 + · · · + kn , is
On the Essential Spectrum of the Translation Invariant Nelson Model n (n) Σ0 (ξ; k) := Σ0 ξ − k (n) + ω(kj )
181
(4)
j=1
and the smallest of such energies (n)
Σ0 (ξ) :=
(n)
inf Σ0 (ξ; k) ,
k∈Rnν
(5)
which is a threshold energy for the model. Due to the assumption of strict subadditivity of ω, Condition 2 (vi), we have (n)
(n )
Σ0 (ξ) < Σ0
(ξ) , for n < n .
(6)
The function Σess can be expressed in terms of Σ0 (1)
Σess (ξ) = Σ0 (ξ) .
(7)
This is the content of the HVZ theorem, see [9, Theorem 1.2 and Corollary 1.4] and [11, Sect. 4]. Write I0 := {ξ ∈ Rν Σ0 (ξ) < Σess (ξ)} (ξ’s with an isolated (1) groundstate) and for ξ ∈ Rν : I0 (ξ) := {k ∈ Rν : ξ − k ∈ I0 }. We recall that H(ξ) is selfadjoint on D = D(H0 (ξ)), which is independent (n) of ξ. The functions ξ → Σ0 (ξ), Σess (ξ), Σ0 (ξ) are Lipschitz continuous, rotation invariant, and go to inﬁnity at inﬁnity. For a treatment of the second quantization formalism used in the formulation of the model see [1] or the brief overviews given in [4] and [9]. See also the recent monograph [12] by Spohn, for up to date material on models of nonrelativistic QED. The authors talk at the QMATH9 meeting, was devoted to an overview of results for the spectral functions introduced above. Drawing mostly on work of Fr¨ ohlich [6, 7], Spohn [11], and the author [9]. One of these results, [9, Theorem 1.9], states that R t → Σess (tu) is a real analytic function away from a closed countable set, under the additional assumption that ω is also convex. Here u is an arbitrary unit vector. This prompted the following question from Heinz Siedentop: “Is this optimal?”. Here is the answer: Theorem 1. Fix a unit vector u ∈ Rν . Suppose (1) and Conditions 1 and 2. Then there exists a locally ﬁnite set S ⊂ R such that R\S t → Σess (tu) is analytic. For any connected component I = (a, b) of R\S we have either: Σess (1) is constant on I, or there exists an analytic function I t → θ(t) ∈ I0 (tu) such that for t ∈ I (1) Σess tu = Σ0 tu; θ(t)u and ∇Σess tu = ∇ω θ(t)u . In the latter case, there furthermore exist integers 1 ≤ p, q < ∞ such that the functions (a, a + δ) t → θ(a + (t − a)p )) and (b − δ, b) t → θ(b − (b − t)q ) extend analytically through a respectively b. (Here δ is chosen such that a + δ p , b − δ q ∈ I.)
182
J. SchachMøller
Remarks: (1) The part of (1) requiring v to be real is an input to a PerronFrobenius argument, see [7, Sect. 2.4] and [9, Sect. 3.3], which ensures that the groundstate of each H(ξ) is nondegenerate. This, together with analytic perturbation theory, implies that ξ → Σ0 (ξ) is analytic in I0 , see [7, Theorem 3.6]. This is in fact the information we need to make the proof of Theorem 1 work. Hence the conclusion of the theorem remains true also for the uncoupled system (v ≡ 0) although (1) is not satisﬁed in this case. (2) If subadditivity of ω is not assumed we are faced with two problems: (I) We would need to understand the breakup of degenerate critical points (n) of k → Σ0 (ξ; k) for any n, not just n = 1. This is a much more diﬃcult (n) problem (but probably doable). (II) The crossing of thresholds Σ0 (ξ) may be associated with the disappearance of the groundstate Σ0 (ξ) into the essential spectrum. The strategy of the proof below would require that Σ0 (ξ) (suitably reparameterized as in Theorem (1) continues analytically into the essential spectrum. This is beyond current technology. In Sect. 2 below, we study analytic functions of two complex variables (1) which are of the form of f (x − y) + g(y), cf. the deﬁnition (4) of Σ0 (·; ·). In Sect. 3 we apply the results of Sect. 2 to prove Theorem 1. In Appendix A we recall basic properties of Riemannian covering spaces.
2 A Complex Function of Two Variables We write (Rp , πp ) for the p’th Riemannian cover over C\{0}. See Appendix A. For z ∈ C and r > 0 we introduce the notation D(z, r) := {z ∈ C : z − z  < r} , D (z, r) := D(z, r)\{z} , Dp (r) := πp−1 (D (0, r)) ⊂ Rp . We furthermore use the abbreviations D(r) ≡ D(0, r) and D (r) = D (0, r). Fix x0 , y0 ∈ C and r0 > 0. Let f, g be analytic in D(x0 − y0 , 2r0 ) and D(y0 , r0 ) respectively. We deﬁne: H(x, y) := f (x − y) + g(y) for (x, y) ∈ D := D(x0 , r0 ) × D(y0 , r0 ) . The function H is an analytic function of two variables in the polydisc D. We suppose that y → H(x0 , y) has a critical point at y = y0 , that is: (∂y H)(x0 , y0 ) = 0 .
(8)
The aim of this section is to catalogue the breakup of the critical point, counting multiplicity, when x0 is replaced by an x near x0 . We wish to determine the sets (9) Θ(x) := {y ∈ D(y0 , ry ) : (∂y H)(x, y) = 0} , for x ∈ D (x0 , rx ) and rx , ry small enough.
On the Essential Spectrum of the Translation Invariant Nelson Model
183
We write f and g as convergent power series in the discs D(x0 − y0 , 2r0 ) and D(y0 , r0 ) respectively f (z) =
∞
∞
fk (z − (x0 − y0 ))k and g(z) =
k=0
gk (y − y0 )k .
(10)
k=0
We can without loss of generality assume f0 = g0 = 0, and (8) implies f1 = g1 , and hence we can in addition assume f1 = g1 = 0. (The critical points are independent of f0 , f1 , g0 and g1 ). To summarize f0 = f1 = g0 = g1 = 0 .
(11)
For a function h, analytic in an open set U ⊂ C, we recall that h has a zero of order k at z0 if (z − z0 )−k h is analytic near z0 , and (z − z0 )−k−1 h is singular at z0 . Equivalently h has a zero of order k at z0 if ∂∂zh (z0 ) = 0, for k
0 ≤ < k, and ∂∂zhk (z0 ) = 0. We will use the following notation for roots of unity. For k ≥ 1, we write the k solutions of αk = 1 as αk := ei2π/k , for ∈ {1, . . . , k} .
(12)
By the p’th root of a complex nonzero constant C ∼ (C, arg(C)), where 0 ≤ arg(C) < 2π, we understand C 1/p := C1/p ei arg C/p .
(13)
In the following we write κH for the order of the zero y0 for the analytic function (∂y H)(x0 , ·), κg for the order of the zero y0 of y → g(y), and κf for the order of the zero x0 − y0 of z → f (z) (recall (11)). We furthermore abbreviate ∂ κf f (x0 ∂z κf
κg
∂ g − y0 ) κg (y0 ) = κf fκf , G := ∂z = κg gκg , F := (14) (κf − 1)! (κg − 1)! (∂yκH +1 H)(x0 , y0 ) = (κH + 1)((−1)κH +1 fκH +1 + gκH +1 ) = 0 . M := κH !
Proposition 1. Let f, g, x0 , y0 and r0 be as above, with κg , κf , κH < ∞. Then there exist 0 < rx , ry ≤ r0 , such that for x ∈ D (x0 , rx ) the set of solutions Θ(x) ⊂ D (y0 , ry ) consists of precisely κH distinct points, all of which are zeroes of order 1 for (∂y H)(x, ·). We have the following description of Θ(x): I The case κg ≤ κH : There are analytic functions θ : D(rx ) → D(ry ), ∈ {1, . . . , κg − 2}, and θ : Dκ H −κg +2 (rx ) → D (ry ), such that Θ(x) = κ −2
g {θ (x−x0 )})∪θ(πκ−1 ({x−x0 })). We have the asymptotics y0 +(∪=1 H −κg +2
184
J. SchachMøller κ −1
θ (x) = θ(x) =
α g
x + κ −1 α g − 1 CI x1/(κH −κg +2)
O x2 , ∈ {1, . . . , κg − 2} , + O x−2/(κH −κg +2)
where CI = [−(κg − 1)G/M ]1/(κH −κg +2) . II The case κg = κH + 1: Here κf ≥ κg and we write κf − 1 = pκH + q and d = (q, κH ) (the greatest common divisor), where p ≥ 1, 0 ≤ q < κH . There exists d analytic functions θ : Dκ H /d (rx ) → D (ry ) such that Θ(x) = y0 + ∪d=1 θ (πκ−1 ({x − x0 })). We have two possible asymptotics: H /d If κf = κg (and hence q = 0, d = κH , and D1 (rx ) ≡ D (rx )) then θ (x) =
CII ακH x + O x2 , ∈ {1, . . . , κH } , CII ακH − 1
where CII = ((−1)κH F/G)1/κH = 1. If κf > κg then 1 q θ (x) = CII αd πκH /d (x)p x κH /d d + O xκf /κH , ∈ {1, . . . , d} , = (F/G)1/κH . where CII III The case κg > κH + 1: Write κg − 1 = pκH + q and d = (κH , q), where p ≥ 1 and 0 ≤ q < κH . There are d analytic functions θ : Dκ H /d (rx ) → D (ry ) such that Θ(x) = y0 + ∪d=1 θ (πκ−1 ({x − x0 })). We have the H /d,x0
asymptotics, with CIII = (−G/M )1/κH and ∈ {1, . . . , d}, 1 q θ (x) = πκH /d (x) + CIII αd πκH /d (x)p x κH /d d + O xκg /κH . If q = 0 and hence d = κH in II and III, then the maps θ , a priori deﬁned on D1 (rx ) ≡ D (0, rx ), extend to analytic maps from D(0, rx ) by the prescription θ (0) := y0 . (Note the convention (0, p) = p for p = 0.) Remarks: (1) If κf = ∞, then H(x, y) = g(y), and θ(x) ≡ y0 is the solution to (8). If κg = ∞ then g = 0 and θ(x) = x is the solution to (8). If κH = ∞ then H(x0 , y) ≡ H(x0 , y0 ) and hence g(y) = H(x0 , y0 ) − f (x − y). (2) If κH = 1 We get an analytic solution x → θ(x) of (8) from the implicit function theorem, cf. [8, Theorem I.B.4]. Proposition 1 II and III then states the possible asymptotics. (3) A particular consequence is that degenerate critical points are isolated. (4) In the proof we handle the error term by a ﬁxed point argument. This implies the following important observation. If x0 , y0 ∈ R and f and g are real analytic. Then a branch R\{x0 } x → θ(x) ∈ Θ(x) is real valued if and only if θ(x) is real to leading order (the order needed to uniquely determine θ.) Proof. The plan of the proof is as follows. First we identify enough terms in ˜ + z(x) of the critical points so that we an asymptotic expansion θ(x) = θ(x)
On the Essential Spectrum of the Translation Invariant Nelson Model
185
can separate them. Secondly we use a ﬁxed point argument to show that the ˜ remainder, x → z(x), vanishes at a faster rate than the leading order term θ. Note that it is a general result that for x close to x0 , ∂y H(x, ·) has precisely κH zeroes counting their orders. See [8, Lemma 1.B.3]. Our task is to account for those κH zeroes. We remark that we could have simply postulated the ˜ but at the cost of transparency. form of the leading order terms θ, We can assume without loss of generality that x0 = y0 = 0. We wish to solve, for a ﬁxed x in a neighbourhood of 0, (∂y H)(x, y) = 0 .
(15)
We begin by collecting some facts. Compute ∀ : (∂y H)(0, 0) = (−1)
∂f ∂g (0) + (0) . ∂y ∂y
(16)
We thus ﬁnd from (8) and (10), recall (11), that ∀ ≤ κH : (−1)+1 f = g ,
(17)
and from the deﬁnition of κg that ∀ < min{κH + 1, κg } : f = g = 0 .
(18)
Below we will use the following notation for remainders in expansions. Let function in a disc D(z0 , rh ) with expansion h(z) = "∞ h be an analytic k h (z − z ) . We write for z ∈ D(z0 , rh ) k 0 k=0 Rh (z) :=
∞
k hk (z − z0 )k−−1 .
(19)
k=+1
Note that Rh are bounded analytic functions in D(z0 , rh /2). We separate into the three cases I, II, and III. Case I (κg ≤ κH ): Expand the lefthand side of (15), using (17), (18), and the notation (19): (∂y H)(x, y) =
κ H +1
& % − f (x − y)−1 + g y −1
=κg
− RfκH +1 (x − y) (x − y)κH +1 + RgκH +1 (y) y κH +1 =
κH
% & g y −1 − (y − x)−1
=κg
+ M y κH + (κH + 1) fκH +1 ((x − y)κH − y κH ) − RfκH +1 (x − y) (x − y)κH +1 + RgκH +1 (y) y κH +1 .
(20)
186
J. SchachMøller
First we look for solutions to (15) with asymptotics θ(x) ∼ x. That is, the leading order term should solve y κg −1 − (y − x)κg −1 = 0, i.e. if we put y = βx then β should solve (β/(β − 1))κg −1 = 1. This gives the following κg − 2 solutions for β κ −1
β =
α g
κ −1
α g
−1
, for ∈ {1, κg − 2} .
(21)
κ −1
Note that ακgg −1 = 1 does not give rise to a solution. We thus ﬁnd in this case θ˜ (x) = β x. Recall the notation αk from (12). Secondly we look for solutions to (15) with asymptotics θ(x) ∼ xρ for some 0 < ρ < 1. Expanding the terms (y−x)−1 in binomial series we identify the highest order terms and are led to require (κg − 1)Gxy κg −2 + M y κH = 0. This gives the equation y κH −κg +2 = −[(κg − 1)G/M ]x. (We note that the κg −2 zero solutions are the ones we identiﬁed in the ﬁrst step above.) We use the map RpI x → x1/pI introduced in (32), pI = κH − κg + 2, to express the solution θ(x) = CI x1/pI , where CI = [−(κg − 1)G/M ]1/pI .
(22)
Case II (κg = κH + 1): We expand again (∂y H)(x, y) = − F (x − y)κf −1 + G y κg −1 κ +1
− Rf f
(x − y) (x − y)κf + Rgκg +1 (y) y κg .
(23)
First we consider the case κf = κg = κH + 1. This is similar to the ﬁrst step above. We look for solutions with the asymptotics θ(x) ∼ x, and put ˜ θ(x) = βx. We get the equation (−1)κH +1 F (β − 1)κH + Gβ κH = 0. This leads us to consider the equation (β/(β − 1))κH = (−1)κH F/G, which has κH solutions β =
CII ακH , for ∈ {1, . . . , κH } . CII ακH − 1
(24)
Here CII := ((−1)κH F/G)1/κH . We note that since (−1)κH +1 F +G = M = 0, cf. (14), we must have 0 < arg(CII ) < 2π/κH . This observation ensures that we avoid any singularity in the case F  = G. Secondly we assume κf > κg and look for solutions to (15) with asymptotics θ(x) ∼ xρ , for some ρ > 1. This leads to the equation − F xκf −1 + G y κg −1 = 0 . (Here M = G.) Let κf − 1 = pκH + q and d = (q, κH ) as in the statement of the proposition. We express the solutions as d analytic maps from RκH /d to C\{0}
On the Essential Spectrum of the Translation Invariant Nelson Model
187
αd πκH /d (x)p (x1/(κH /d) )q/d , CII = (F/G)1/κH . θ˜ (x) = CII
(25)
Case III (κg > κH + 1): Here we must have κf = κH + 1. We write down the expansion (∂y H)(x, y) = M (y − x)κH + G y κg −1 − RfκH +2 (x − y) (x − y)κH +1 + Rgκg +1 (y) y κg . Here the asymptotics is the same to leading order, namely y ∼ x. Write ˜ ˆ and look for θˆ with the asymptotics θˆ ∼ xρ , ρ > 1. This gives θ(x) = x + θ, the equation for θˆ M θˆκH + G xκg −1 = 0 , As above let κg − 1 = pκH + q and d = (q, κH ) . We express the solutions as d analytic maps from RκH /d to C\{0}, with CIII = (−G/M )1/κH , θ˜ (x) = πκH /d (x) + CIII αd πκH /d (x)p (x1/(κH /d) )q/d .
(26)
We have now determined the leading order term in all cases. We proceed to show by a ﬁxed point argument that indeed there is a zero of order 1 near each of the terms identiﬁed above. We introduce function spaces . Zp (ρ, C) := z ∈ C(Dp (rx ) ; C)  z(x) ≤ C xρ , equipped with supnorm. If p = 1 we identify D1 (rx ) ≡ D (rx ). We now describe the procedure which we follow below, so as to cut short the individual arguments. First we write the actual branch of critical points as a sum θ(x) = ˜ θ(x) + z(x), where θ˜ is the leading order term as derived above and z is an element of a suitable Zp . We plug this into an expansion of (∂y H)(x, y) and identify leading order terms. These are of two types. One is linear in z and the others are independent of z. The term linear in z is used to construct ˜ + z(x))/h(x), if hz is maps T on Zp , by (T z)(x) = z(x) − (∂y H)(x, θ(x) the term linear in z. The remaining leading order terms now become leading order terms for T z and their decay determine the decay of the remainder and hence ρ. The constant C is chosen such that T maps Zp into itself. Finally, since terms in T z depending on z are of higher order, we can choose rx small enough such that T becomes a contraction. Its unique ﬁxed point z0 is the desired correction to the leading order contribution found above. Note that ˜ + z0 (x)) = 0. a ﬁxed point satisﬁes (∂y H)(x, θ(x) Case I: Write θ (x) = β x + z(x), cf. (21), and look for z vanishing faster than x at 0. The leading order term linear in z in (20) is & % (κg − 1) G (β x)κg −2 − ((β − 1)x)κg −2 z = (κg − 1) G γ xκg −2 z κ −2
A computation yields γ := β g
−(β −1)κg −2 = 0. Deﬁne maps on Z1 (2, C)
(T z)(x) := z(x) −
(∂y H)(x, β x + z(x)) . (κg − 1)Gγ xκg −2
188
J. SchachMøller
The contributions to T z which scale as x2 (the slowest appearing rate) are % κ & (κg + 1)gκg +1 β g − (β − 1)κg 2 − x , for κg < κH , (κg − 1)Gγ M βκH + (κH + 1)fκH +1 [(1 − β )κH − βκH ] 2 − x , for κg = κH . (κg − 1)Gγ We now choose C large enough such that the norm of the coeﬃcients above are less than C. Choosing rx suﬃciently small turns T into contractions on Z1 (2, C). Now write θ(x) = CI x1/pI + z(x) where z ∈ ZpI (κg /pI , C). The term in (20) which is linear in z and of leading order is & % (κg − 2) (κg − 1) G πpI (x) (CI x1/pI )κg −3 + κH M (CI x1/pI )κH −1 z = pI M CIκH −1 (x1/pI )κH −1 z , where we used (33). Note that the coeﬃcient is nonzero. Deﬁne a map (T z)(x) := z(x) −
(∂y H)(πpI (x), CI x1/pI + z(x)) . pI M (CI x1/pI )κH −1
We wish to show that T : ZpI (κg /pI , C) → ZpI (κg /pI , C) if C is large enough. Let C(κg ) = fκg +1 if κg < κH and C(κg ) = gκH +1 if κg = κH . The term in T z which vanish to lowest order is 2 (κg + 1) C(κg ) πpI (x) (CI x1/pI )κg −1 pI = O x , pI M (CI x1/pI )κH −1 Choosing C and rx as above ﬁnishes case I. Case II: Consider the case κf = κg = κH + 1. Let θ˜ (x) = β x, where β is as in (24). We deﬁne maps on Z1 (2, C) (T z)(x) := z −
(∂y H)(x, β x + z(x)) , (κg − 1) A xκg −2
which is well deﬁned since A := (−1)κH +1 F (β − 1)κH −1 + G βκH −1 κH −1 CII − (ακH )κH −1 = 0 . = − G (β − 1)κH β−1 CII Here we used (24), the deﬁnition of CII , and that 0 < arg(CII ) < 2π/κH . The leading order contributions to T z are κ +1 κ [(κg − 1) A ]−1 Rf f (1 − β )κg + Rgκg +1 β g x2 = O(x2 ) . As above this estimate suﬃce.
On the Essential Spectrum of the Translation Invariant Nelson Model
189
Next we turn to the case κf > κg . Let θ˜ be as in (25) and deﬁne maps on ZκH /d (κf /κH , C) by (T z)(x) := z −
(∂y H)(πκH /d (x), θ˜ (x) + z(x)) . (κg − 1)Gθ˜ (x)κg −2
Let ρII := min{2κf − κg − 1, κf + κg − 2}. The leading order terms in T z are κ +1
(κf − 1)F πκH /d (x)κf −2 θ˜ (x) − Rf f − (κg − 1)Gθ˜ (x)κg −2
πκH /d (x)κf
= O xρII .
Since 2κf − κg − 1 ≥ κf and κf + κg − 2 ≥ κf , we have ρII ≥ κf and conclude, as above, case II. Case III: Let θ˜ be as in (26) and write θ(x) = θ˜ (x) + z(x), where we take z from ZκH /d (κg /κH , C). We deﬁne maps (T z)(x) := z −
(∂y H)(πκH /d (x), θ˜ (x) + z(x)) κH M [θ˜ (x) − πκ /d (x)] H
Let ρIII := min{2κg − κH − 2, κH + κg − 1}. The leading order terms in T z are κ +1 (κg − 1)GπκH /d (x)κg −2 θ˜ (x) + Rg g πκH /d (x)κg = O xρIII . κH M [θ˜ (x) − πκH /d (x)] Since 2κg − κH − 2 ≥ κg and κg + κH − 1 ≥ κg , we conclude, as above, case III. Finally we address analyticity of the a priori continuous solutions found above. Maps from D (rx ) are analytic by the analytic implicit function theorem [8, Theorem I.B.4], and maps from Riemanian covers are locally analytic by the same argument, and hence analytic. That the maps above deﬁned on D (rx ) extend to analytic functions on the whole disc D(rx ) follows from [3, Theorem V.1.2].
3 The Essential Spectrum In this section we use Proposition 1 to prove Theorem 1. Let u ∈ Rν be a unit vector. We introduce σ(t) := Σ0 (tu) and σ (1) (t; s) := σ(t − s) + ω(s) , where we abuse notation and identify ω(s) ≡ ω(su). We furthermore write σ (1) (t) := inf σ (1) (t; s) s∈R
(1) I0 (t)
(1) I0 (tu)}.
:= {s ∈ Rsu ∈ and We begin with three lemmata and a proposition. The ﬁrst lemma is a special case of [9, Sect. 3.2].
190
J. SchachMøller
Lemma 1. Assume v ∈ L2 (Rν ) and Conditions 1 (i), (ii) and 2 (i), (ii), (vi). (1) (2) (1) Let ξ ∈ Rν and k ∈ Rν . If Σ0 (ξ; k) < Σ0 (ξ), then k ∈ I0 (ξ). Using this lemma, cf. (6), we ﬁnd that (1)
Σess (ξ) = min{E  (ξ, E) ∈ T0 } ,
(27)
(1)
where T0 is the set of thresholds coming from onephoton excitations of the ground state. It is deﬁned by (1) (1) (1) . T0 := (ξ, E) ∈ Rν+1  ∃k ∈ Rν : E = Σ0 (ξ; k) and (ξ; k) ∈ Crit0 , . (1) (1) (1) Crit0 := (ξ, k) ∈ R2ν  k ∈ I0 (ξ) and ∇k Σ0 (ξ; k) = 0 . There are obvious extensions to higher photon number, which must be included in (27), if ω is not subadditive. The next lemma is the key to the applicability of Proposition 1. (1)
Lemma 2. Let (ξ, k) ∈ Crit0 , such that ξ = 0 and ∇ω(k) = 0. Then there exists θ ∈ R such that k = θξ. (1)
Proof. Let ξ = 0 and k ∈ Rν be a critical point ∇k Σ0 (ξ; k) = 0. Then ∇Σ0 (ξ − k) = ∇ω(k)
(28)
Write ∇ω(k) = c1 k and ∇Σ0 (ξ − k) = c2 (ξ − k), using rotation invariance. Here c1 = 0. From (28) we ﬁnd cξ = (1 + c)k, where c = c2 /c1 . Since ξ = 0, we conclude the result. We write in the following, for r ≥ 0, B(r) = {k ∈ Rν k = r}. For a unit vector u and radii r1 , r2 ≥ 0 we write for t ∈ R Ct ≡ Ct (r1 , r2 ; u) := {k ∈ B(r1 )  t u − k ∈ B(r2 )} . We leave the proof of the following lemma to the reader. (Draw a picture.) It deals with the stability of critical points which are not covered by Lemma 2. Lemma 3. Let r1 , r2 ≥ 0, u ∈ Rν be a unit vector, and assume ν ≥ 2. Suppose t0 ∈ R and Ct0 = ∅. There exists a neighbourhood U of t0 , such that: i) If Ct0 ⊂ {−r1 u, +r1 u} and t ∈ U, then Ct = ∅. ii) If Ct0 ⊂ {−r1 u, +r1 u} , then Ct0 = {k0 }. Let σ = u · k0 (t0 − u · k0 ) ∈ {−r1 r2 , 0, +r1 r2 } (σ = 0 iﬀ either r1 or r2 equals 0) and t ∈ U\{t0 }. If σ(t − t0 ) > 0 then Ct = ∅, and if σ(t − t0 ) ≤ 0, then Ct = ∅. (2)
Proposition 2. Let t0 ∈ R be such that σ (1) (t0 ) < Σ0 (t0 u). There exist 0 < δ ≤ 1 and an analytic function Uδ \{t0 } t → θ(t), where Uδ = (t0 − δ, t0 + δ), such that σ (1) (t) = σ (1) (t; θ(t)), for t ∈ Uδ \{t0 }. Furthermore, there exist integers 1 ≤ p , pr < ∞, such that the functions (t0 − δ, t0 ) t → θ(t0 − (t0 − t)p ) and (t0 , t0 + δ) t → θ(t0 + (t − t0 )pr ) extend analytically through t0 .
On the Essential Spectrum of the Translation Invariant Nelson Model
191 (2)
Remark: The reason for only studying σ (1) where it is smaller than Σ0 , is (1) the need for having global minima of s → σ (1) (t; s) in I0 (t), cf. Lemma 1. This may not be true in general. If ν = 1, 2 or ω is convex, this consideration is unnecessary. See [9, Theorem 1.5 i)] and Lemma 2. (2) ˜ For t ∈ U ˜, let Proof. Pick δ˜ > 0 such that σ (1) (t) < Σ0 (tu), for t − t0  ≤ δ. δ (1) (1) Gt = {s ∈ Rσ (t; s) = σ (t)} be the set of global minima for s → σ (1) (t; s). We recall from [9, Proof of Theorem 1.9] (an application of Lemma 1) that the sets Gt are ﬁnite, and all s ∈ Gt are zeros of ﬁnite order for the analytic (1) function I0 (t) s → ∂s σ (1) (t; s). Secondly we remark that for any t˜ > 0 and σ ¯ ∈ R, the set {(t, s) ∈ R2 t ≤ (1) ˜ ¯ } is compact. From this remark and the ﬁniteness of the t and σ (t; s) ≤ σ Gt ’s we conclude from Proposition 1 and a compactness argument that the set
S := {t ∈ Uδ˜∃s ∈ Gt , n ∈ N s.t. ∂s2 σ (1) (t; s) = 0 and ∂ n σ(t − s) = 0} is locally ﬁnite. In particular, for t ∈ Uδ˜\S the global minima s ∈ Gt are all either simple zeroes of s → ∂s σ (1) (t; s), or zeroes of inﬁnite order for s → ∂σ(t − s). For s0 ∈ Gt which are simple zeroes of s → Suppose ﬁrst that t0 ∈ S. 0 (1) ∂s σ (t; s) we obtain from the analytic implicit function theorem analytic solutions θ to ∂s σ (1) (t; θ(t)) = 0, deﬁned in a neighbourhood of t0 . For s0 ∈ Gt0 which are zeroes of inﬁnite order of s → ∂σ(t0 − s) (and not already included in the ﬁrst case), we take θ(t) ≡ s0 which solves ∂s σ (1) (t; θ(t)) = 0 near t0 . See Remark 1), with κf = ∞, after Proposition 1. We have thus for some 0 < δ < δ˜ constructed Gt0  analytic functions θ deﬁned in Uδ , such that Gt0 = {θ (t0 )} and Gt ⊂ {θ (t)} (by continuity) for t ∈ Uδ . Hence σ (1) (t) = min1≤≤Gt0  σ (1) (t; θ (t)), for t ∈ Uδ . If Gt0  = 1 take δ = δ and θ = θ1 . If Gt0  > 1 choose 0 < δ < δ , 1 , and 2 such that the following choice works: θ(t) = θ1 (t), for t0 − δ < t < t0 , and θ(t) = θ2 (t), for t0 < t < t0 + δ. This proves the result if t0 ∈ S. Here we get from Proposition 1 a 0 < δ < δ˜ and It remains to treat t0 ∈ S. right two families of analytic functions {θleft } and {θ } deﬁned in (t0 − δ , t0 ) and (t0 , t0 + δ ) respectively, which parameterize the critical points for t near t0 , which comes from Gt0 . Furthermore Gt ⊂ {θleft (t)}, t0 − δ < t < t0 and right Gt ⊂ {θ (t)}, t0 < t < t0 + δ . Note that the number of branches to the left and to the right need not be the same, but both are ﬁnite. We are ﬁnished if we can prove that, for = , the function σ (1) (t; θleft (t)) − σ (1) (t; θleft (t)) is either identically zero, or it does not van ish on a sequence of t’s converging to t0 from the left. Similarly for the right of t0 . (If there is only one branch, then it continues analytically through t0 and we are done).
192
J. SchachMøller
In the following we work to the left of t0 and drop the superscript “left”. The region to the right of t0 can be treated similarly. There exists 1 ≤ p, p < ∞ and analytic functions θ : Rp → C and θ : Rp → C such that θ and θ are branches of θ and θ respectively. See Proposition 1. That is, there exist 0 ≤ q < p and 0 ≤ q < p such that
θ (t) = θ(Rpρ (t0 − t)) and θ (t) = θ (Rpρ (t0 − t)) , where ρ = π + 2πq and ρ = π + 2πq . Recall notation from (30) and (31). Here we used the canonical embedding R\{0} r → (t, 0) ∈ Rp , for any p. Since z → θ(Rpρ (Pp (z))) and z → θ (Rpρ (Pp (z))) are analytic and bounded functions from D (r) (for some r > 0), they extend to analytic functions on D(r). We can hence deﬁne an analytic function
h(z) = σ (1) (z; θ(Rpρ (Pp ((t0 − z)p )))) − σ (1) (z; θ (Rpρ (Pp ((t0 − z)p ))))
in D(r1/pp ). The function h is either identically zero or has only isolated zeroes (one is at t0 ). This now implies that σ (1) (t; θ (t)) − σ (1) (t; θ (t)) = h(t0 − (t0 − t)1/pp ) is either identically zero or has ﬁnitely many zeroes near t0 . We can now choose 0 < δ < δ and θ as above. This concludes the proof. Proof of Theorem 1: Proposition 2 covers the case ν = 1. In the following we assume ν ≥ 2. It suﬃces to prove the theorem locally near any t0 ∈ R. For the global minima at t ∈ R we write Mt :=

(1)
(1)
k ∈ Rν  Σ0 (tu; k) = Σ0 (tu)
.
and we introduce two subsets . . Mt := k ∈ Mt  k u and M0t := k ∈ Mt  ∇ω(k) = 0 .
We begin with the following note. Let t0 ∈ R. If Mt0 = ∅ (M0t0 = ∅) then there exists a neighbourhood U t0 such that for t ∈ U we have Mt = ∅ 0 (Mt = ∅). Let t0 ∈ R. First consider the case M0t0 = ∅. For t ∈ U, chosen as above, we have Σess (tu) = σ (1) (t), which by Proposition 2 concludes the proof. We can now assume that M0t0 = ∅. By analyticity and rotation invariance, the set of k’s such that ∇ω(k) = 0 is a set of concentric balls, with a locally ﬁnite set of radii. If k ∈ M0t0 then Ok ∈ M0t0 for any O ∈ O(ν; u), where O(ν; u) := {O ∈ O(ν)Ou = u}. Let . =0t = k ∈ I (1) (t)  (tu, k) ∈ Crit(1) and ∇ω(k) = 0 , M 0 0 (1)
σ ˜ (1) (t) := min Σ0 (tu; k) , =0 k∈M t
On the Essential Spectrum of the Translation Invariant Nelson Model
193
=0t = ∅. Then by Lemma 2 with the convention that σ ˜ (1) (t) = +∞ if M (1)
˜ (1) (t)} . Σ0 (tu) = min{σ (1) (t), σ We work only to the left of t0 , i.e. we take t ≤ t0 . The case t ≥ t0 can be treated similarly. We proceed to ﬁnd a δ > 0 such that σ (1) is either constant equal to (1) (1) Σ0 (t0 u), for t0 − δ < t ≤ t0 or it satisﬁes σ ˜ (1) (t) ≥ σ (1) (t) = Σ0 (tu), for t0 − δ < t ≤ t0 . This concludes the result, since both σ (1) and a constant function, suitably reparameterized, continues analytically though t0 . See Proposition 2. First we consider the case where: (A) Σ0 is not constant on the connected component of I0 containing t0 u−k0 , for any k0 ∈ M0t0 . (B) For any k0 ∈ M0t0 we have (with r1 = k0  and r2 = t0 u − k0 ): Ct0 (r1 , r2 ; u) ⊂ {−r1 u, r1 u} and u · k0 (t0 − u · k0 ) ≤ 0. Assuming (A) and (B) we have by Lemma 3 ii) a δ > 0, such that Ct (r1 , r2 ; u) = ∅ , for t0 − δ < t < t0 . (1)
(29)
We proceed to argue that (A) and (B) implies Σ0 (tu) = σ (1) (t), for t < t0 . It suﬃces to ﬁnd a δ > 0 such that M0t = ∅, t0 −δ < t < t0 . Suppose to the contrary that there exists a sequence tn → t0 , with tn < t0 and M0tn = ∅. Let kn ∈ M0tn . We can assume, by possibly passing to a subsequence, that kn → k∞ . Here we used that ω(k) → ∞ as k → ∞. Clearly k∞ ∈ M0t0 , and hence k∞ ∈ Ct0 (r1 , r2 ; u) for some r1 , r2 . Since the possible r1 ’s and r2 ’s are isolated, we must have a n ¯ such that ∀n > n ¯ : kn ∈ Ctn (r1 , r2 ; u). This contradicts (29). For the remaining case we assume one of the following: (C) There exists k0 ∈ M0t0 such that Σ0 is constant on the connected component of I0 containing t0 u − k0 . (The converse of ( A) above.) (D) There exists k0 ∈ M0t0 such that either Ct0 (r1 , r2 ; u) ⊂ {−r1 u, r1 u} or Ct0 (r1 , r2 ; u) ⊂ {−r1 u, r1 u} and u · k0 (t0 − u · k0 ) > 0. Again r1 = k0  and r2 = t0 u − k0 . (The converse of B) above.) There exists δ > 0 such that: In the case (C), for t0 − δ < t < t0 , =0t with Σ (1) (tu; k) = Σ (1) (t0 u). In case (D) we have there exists k ∈ M 0 0 =0t with by Lemma 3 (i) and (ii), for t0 − δ < t < t0 , likewise k ∈ M (1) (1) Σ0 (tu; k) = Σ0 (t0 u). Hence, if either (C) or (D) are satisﬁed we have (1) σ ˜ (1) (t) ≤ Σ0 (t0 u), for t0 − δ < t < t0 . (1) In order to show the converse inequality σ ˜ (1) (t) ≥ Σ0 (t0 u), we assume =0 such that to the contrary that there exists a sequence tn → t0 and kn ∈ M tn (1) (1) Σ0 (tn u; kn ) < Σ0 (t0 u). As above we can assume kn → k∞ ∈ M0t0 . If Σ0 (1) is constant on the connected component of t0 u − k∞ , then Σ0 (tn u; kn ) = (1) Σ0 (t0 u) is a constant sequence for n large enough, which is a contradiction. If Σ0 is not constant on the connected component of t0 u − k∞ , then
194
J. SchachMøller
kn  = k∞  and tn u − kn  = t0 u − k∞ , for n large enough, and again we (1) (1) conclude Σ0 (tn u; kn ) = Σ0 (t0 u) is a constant sequence, which is a contradiction.
A Riemannian Covers Let Rp = (0, ∞) × R/2πpZ, equipped with the product topology. We write z = (z, arg(z)) for elements of Rp and introduce the pcover (Rp , πp ) of C\{0} by πp : Rp → C\{0} where πp (z) := z ei arg(z) . Note that πp is locally a homeomorphism and thus provides a chart which turns Rp into an analytic surface. If U ⊂ Rp is such that πp is 1–1 on U, write πU−1 for the inverse homeomorphism from πp (U) to U. We will use the concept of analytic functions to, from, and between cover spaces. This is just special cases of what it means to be an analytic map between two analytic surfaces, cf. [3, Sects. IX.6 and IX.7] Let V ⊂ C, Vp ⊂ Rp , and Vq ⊂ Rq be open sets, and f1 : V → Rp , f2 : Vp → Rq , and f3 : Vq → C continuous maps. We say f1 is analytic if for any z0 ∈ V there exists an open set U ⊂ V with z0 ∈ U, such that the map U z → πp (f1 (z)) is analytic in the usual sense. We say f2 is analytic if for any z0 ∈ Vp there exists an open set Up ⊂ Vp with z0 ∈ Up and πp : Up → C 1–1, such that πp (Up ) z → πq (f2 (πU−1 (z))) p is analytic in the usual sense. We say f3 is analytic if for any z0 ∈ Vq there exists an open set Uq ⊂ Vq with z0 ∈ Uq and πq : Uq → C 11, such that the map πq (Uq ) z → (z)) is analytic in the usual sense. f3 (πU−1 q With this deﬁnition it is easy to check that f2 ◦ f1 , f3 ◦ f2 and f3 ◦ f2 ◦ f1 are analytic maps. We give three examples which we use in Sect. 2. The ﬁrst example is Pp : C\{0} → Rp , deﬁned by Pp (z) := (zp , p arg(z)) .
(30)
Second example: Let ρ ∈ R. We deﬁne a map Rpρ : Rp → Rp by Rpρ (z) := (z, arg(z) + ρ mod 2πp) .
(31)
The third example is the map Rp z → z1/p ∈ C\{0} deﬁned by z1/p := z1/p ei arg(z)/p .
(32)
The three examples above are all analytic and in addition bijections. We have Pp (z1/p ) = (Pp (z))1/p = z and (z1/p )p = πp (z) .
(33)
On the Essential Spectrum of the Translation Invariant Nelson Model
195
References 1. F.A. Berezin, The method of second quantization, 1 ed., Academic Press, New York, San Francisco, London, 1966. 2. L. Bruneau and J. Derezi´ nski, PauliFierz Hamiltonians deﬁned as quadratic forms, Rep. Math. Phys., 54 (2004), 169–199. 3. J.B. Conway, Functions of one complex variable, 2 ed., Graduate texts in mathematics, vol. 11, SpringerVerlag, New York, 1978. 4. J. Derezi´ nski and C. G´erard, Asymptotic completeness in quantum ﬁeld theory. massive PauliFierz Hamiltonians, Rev. Math. Phys. 11 (1999), 383–450. 5. R.P. Feynman, Statistical mechanics. A set of lectures, Frontiers in physics, W.A. Benjamin, Inc., Reading, Massechusets, 1972. 6. J. Fr¨ ohlich, On the infrared problem in a model of scalar electrons and massless scalar bosons, Ann. Inst. Henri Poincar´e 19 (1973), 1–103. , Existence of dressed oneelectron states in a class of persistent models, 7. Fortschr. Phys. 22 (1974), 159–198. 8. R.C. Gunning and H. Rossi, Analytic functions of several complex variables, Series in modern analysis, PrenticeHall, Inc., Englewood Cliﬀs, N. J., 1965. 9. J.S. Møller, The translation invariant massive Nelson model: I. The bottom of the spectrum, Ann. Henri Poincar´e 6 (2005) 1091–1135. 10. “The Fr¨ ohlich polaron revisited”. Submitted. 11. H. Spohn, The polaron at large total momentum, J. Phys. A 21 (1988), 1199– 1211. , Dynamics of charged particles and their radiation ﬁeld, Cambridge 12. University Press, 2004.
BoseEinstein Condensation as a Quantum Phase Transition in an Optical Lattice∗ M. Aizenman1 , E.H. Lieb1 , R. Seiringer1 , J.P. Solovej2 and J. Yngvason3,4 1
2
3
4
Departments of Mathematics and Physics, Jadwin Hall, Princeton University, P. O. Box 708, Princeton, New Jersey 08544 Department of Mathematics, University of Copenhagen, Universitetsparken 5, DK2100 Copenhagen, Denmark Institut f¨ ur Theoretische Physik, Universit¨ at Wien, Boltzmanngasse 5, A1090 Vienna, Austria Erwin Schr¨ odinger Institute for Mathematical Physics, Boltzmanngasse 9, A1090 Vienna, Austria
Abstract. One of the most remarkable recent developments in the study of ultracold Bose gases is the observation of a reversible transition from a Bose Einstein condensate to a state composed of localized atoms as the strength of a periodic, optical trapping potential is varied. In [1] a model of this phenomenon has been analyzed rigorously. The gas is a hard core lattice gas and the optical lattice is modeled by a periodic potential of strength λ. For small λ and temperature BoseEinstein condensation (BEC) is proved to occur, while at large λ BEC disappears, even in the ground state, which is a Mottinsulator state with a characteristic gap. The interparticle interaction is essential for this eﬀect. This contribution gives a pedagogical survey of these results.
1 Introduction One of the most remarkable recent developments in the study of ultracold Bose gases is the observation of a reversible transition from a BoseEinstein condensate to a state composed of localized atoms as the strength of a periodic, optical trapping potential is varied [2, 3]. This is an example of a quantum phase transition [4] where quantum ﬂuctuations and correlations ∗ Contribution to the proceedings of QMath9, Giens, France, Sept. 12–16, c 2004 by the authors. This paper may be 2004. Talk given by Jakob Yngvason. reproduced, in its entirety, for noncommercial purposes.
Work supported in part by US NSF grants PHY 9971149 (MA), PHY 0139984A01 (EHL), PHY 0353181 (RS) and DMS0111298 (JPS); by an A.P. Sloan Fellowship (RS); by EU grant HPRNCT200200277 (JPS and JY); by FWF grant P17176N02 (JY); by MaPhySto – A Network in Mathematical Physics and Stochastics funded by The Danish National Research Foundation (JPS), and by grants from the Danish research council (JPS). M. Aizenman et al.: BoseEinstein Condensation as a Quantum Phase Transition in an Optical Lattice, Lect. Notes Phys. 690, 199–215 (2006) www.springerlink.com
200
M. Aizenman et al.
rather than energyentropy competition is the driving force and its theoretical understanding is quite challenging. The model usually considered for describing this phenomenon is the BoseHubbard model and the transition is interpreted as a transition between a superﬂuid and a Mott insulator that was studied in [5] with an application to He4 in porous media in mind. The possibility of applying this scheme to gases of alkali atoms in optical traps was ﬁrst realized in [6]. The article [7] reviews these developments and many recent papers, e.g., [8–16] are devoted to this topic. These papers contain also further references to earlier work along these lines. The investigations of the phase transition in the BoseHubbard model are mostly based on variational or numerical methods and the signal of the phase transition is usually taken to be that an ansatz with a sharp particle number at each lattice site leads to a lower energy than a delocalized Bogoliubov state. On the other hand, there exists no rigorous proof, so far, that the true ground state of the model has oﬀdiagonal long range order at one end of the parameter regime that disappears at the other end. In this contribution, which is based on the paper [1], we study a slightly diﬀerent model where just this phenomenon can be rigorously proved and which, at the same time, captures the salient features of the experimental situation. Physically, we are dealing with a trapped Bose gas with short range interaction. The model we discuss, however, is not a continuum model but rather a lattice gas, i.e., the particles are conﬁned to move on a ddimensional, hypercubic lattice and the kinetic energy is given by the discrete Laplacian. Moreover, when discusssing BEC, it is convenient not to ﬁx the particle number but to work in a grandcanonical ensemble. The chemical potential is ﬁxed in such a way that the average particle number equals half the number of lattice sites, i.e., we consider half ﬁlling. (This restriction is dictated by our method of proof.) The optical lattice is modeled by a periodic, onebody potential. In experiments the gas is enclosed in an additional trap potential that is slowly varying on the scale of the optical lattice but we neglect here the inhom*ogeneity due to such a potential and consider instead the thermodynamic limit. In terms of bosonic creation and annihilation operators, a†x and ax , our Hamiltonian is expressed as H=−
1 † (ax ay + ax a†y ) + λ (−1)x a†x ax + U a†x ax (a†x ax − 1) . (1) 2 x x xy
The sites x are in a cube Λ ⊂ Zd with opposite sides identiﬁed (i.e., a ddimensional torus) and xy stands for pairs of nearest neighbors. Units are chosen such that 2 /m = 1. " The ﬁrst term in (1) is the discrete Laplacian xy (a†x − a†y )(ax − ay ) " minus 2d x a†x ax , i.e., we have subtracted a chemical potential that equals d. The optical lattice gives rise to a potential λ(−1)x which alternates in sign between the A and B sublattices of even and odd sites. The interatomic
BoseEinstein Condensation as a Quantum Phase Transition
201
onsite repulsion is U , but we consider here only the case of a hardcore interaction, i.e., U = ∞. If λ = 0 but U < ∞ we have the BoseHubbard model. Then all sites are equivalent and the lattice represents the attractive sites of the optical lattice. In our case the adjustable parameter is λ instead of U and for large λ the atoms will try to localize on the B sublattice. The Hamiltonian (1) conserves the particle number N and it is shown in [1], Appendix A, that, for U = ∞, the lowest energy is obtained uniquely for N = 12 Λ, i.e., half the number of lattice sites. Because of the periodic potential the unit cell in this model consists of two lattice sites, so that we have on average one particle per unit cell. This corresponds, physically, to ﬁlling factor 1 in the BoseHubbard model. For given temperature T , we consider grandcanonical thermal equilibrium states, described by the Gibbs density matrices Z −1 exp(−βH) with Z the normalization factor (partition function) and β = 1/T the inverse temperature. Units are chosen so that Boltzmann’s constant equals 1. The thermal expectation value of some observable O will be denoted by O = Z −1 Tr O exp(−βH). In the proof of BEC we focus on dimensions d ≥ 3, but, using the technique employed in [17], an extension to the ground state in two dimensions is possible. Our main results about this model can be summarized as follows: 1. If T and λ are both small, there is BoseEinstein condensation. In this parameter regime the onebody density matrix γ(x, y) = a†x ay has exactly one large eigenvalue (in the thermodynamic limit), and the corresponding condensate wave function is φ(x) = constant. 2. If either T or λ is big enough, then the onebody density matrix decays exponentially with the distance x − y, and hence there is no BEC. In particular, this applies to the ground state T = 0 for λ big enough, where the system is in a Mott insulator phase. 3. The Mott insulator phase is characterized by a gap, i.e., a jump in the chemical potential. We are able to prove this, at halfﬁlling, in the region described in item 2 above. More precisely, there is a cusp in the dependence of the ground state energy on the number of particles; adding or removing one particle costs a nonzero amount of energy. We also show that there is no such gap whenever there is BEC. 4. The interparticle interaction is essential for items 2 and 3. Noninteracting bosons always display BEC for low, but positive T (depending on λ, of course). 5. For all T ≥ 0 and all λ > 0 the diagonal part of the onebody density matrix a†x ax (the oneparticle density) is not constant. Its value on the A sublattice is constant, but strictly less than its constant value on the B sublattice and this discrepancy survives in the thermodynamic limit. In contrast, in the regime mentioned in item 1, the oﬀdiagonal longrange order is constant, i.e., a†x ay ≈ φ(x)φ(y)∗ for large x − y with φ(x) =constant.
202
M. Aizenman et al.
λ Mott insulator (at low T) exp. decay of correlations no BEC
BEC T Fig. 1. Schematic phase diagram at halfﬁlling
Because of the hardcore interaction between the particles, there is at most one particle at each : site and our Hamiltonian (with U = ∞) thus acts on the Hilbert space H = x∈Λ C2 . The creation and annihilation operators can be represented as 2 × 2 matrices with 01 00 10 a†x ↔ , ax ↔ , a†x ax ↔ , 00 10 00 for each x ∈ Λ. More precisely, these matrices act on the tensor factor associated with the site x while a†x and as the identity on the other : ax act 2 C . Thus the operators at diﬀerfactors in the Hilbert space H = x∈Λ ent sites commute as appropriate for Bosons, but on each site they satisfy anticommutation relations, reﬂecting the hard core condition. The Hamiltonian can alternatively be written in terms of the spin 1/2 operators 1 01 1 0 −i 1 1 0 1 2 3 S = , S = , S = . 2 10 2 i 0 2 0 −1 The correspondence with the creation and annihilation operators is a†x = Sx1 + iSx2 ≡ Sx+ ,
ax = Sx1 − iSx2 ≡ Sx− ,
and hence a†x ax = Sx3 + 12 . (This is known as the MatsubaraMatsuda correspondence [18].) Adding a convenient constant to make the periodic potential positive, the Hamiltonian (1) for U = ∞ is thus equivalent to % & 1 + − 1 H=− (Sx Sy + Sx− Sy+ ) + λ + (−1)x Sx3 2 2 x xy % & 1 1 2 2 x 3 1 (2) =− (Sx Sy + Sx Sy ) + λ 2 + (−1) Sx . xy
x
BoseEinstein Condensation as a Quantum Phase Transition
203
Without loss of generality we may assume λ ≥ 0. This Hamiltonian is well known as a model for interacting spins, referred to as the XY model [19]. The last term has the interpretation of a staggered magnetic ﬁeld. We note that BEC for the lattice gas is equivalent to oﬀdiagonal long range order for the 1 and 2components of the spins. The Hamiltonian (2) is clearly invariant under simultaneous rotations of all the spins around the 3axis. In particle language this is the U (1) gauge symmetry associated with particle number conservation of the Hamiltonian (1). Oﬀdiagonal long range order (or, equivalently, BEC) implies that this symmetry is spontaneously broken in the state under consideration. It is notoriously diﬃcult to prove such symmetry breaking for systems with a continuous symmetry. One of the few available techniques is that of reﬂection positivity (and the closely related property of Gaussian domination) and fortunately it can be applied to our system. For this, however, the hard core and halfﬁlling conditions are essential because they imply a particlehole symmetry that is crucial for the proofs to work. Naturally, BEC is expected to occur at other ﬁllings, but no one has so far found a way to prove condensation (or, equivalently, longrange order in an antiferromagnet with continuous symmetry) without using reﬂection positivity and infrared bounds, and these require the addtional symmetry. Reﬂection positivity was ﬁrst formulated by K. Osterwalder and R. Schrader [20] in the context of relativistic quantum ﬁeld theory. Later, J. Fr¨ ohlich, B. Simon and T. Spencer used the concept to prove the existence of a phase transition for a classical spin model with a continuous symmetry [21], and E. Lieb and J. Fr¨ ohlich [22] as well as F. Dyson, E. Lieb and B. Simon [19] applied it for the analysis of quantum spin systems. The proof of oﬀdiagonal long range order for the Hamiltonian (2) (for small λ) given here is based on appropriate modiﬁcations of the arguments in [19].
2 Reﬂection Positivity In the present context reﬂection positivity means the following. We divide the torus Λ into two congruent parts, ΛL and ΛR , by cutting it with a hyperplane orthogonal to one of the d directions. (For this we assume that the side length of Λ is even.) This induces a factorization of the Hilbert space, H = HL ⊗HR , with ; C2 . HL,R = x∈ΛL,R
There is a natural identiﬁcation between a site x ∈ ΛL and its mirror image ϑx ∈ ΛR . If F is an operator on H = HL we deﬁne its reﬂection θF as an operator on HR in the following way. If F = Fx operates nontrivially only on one site, x ∈ ΛL , we deﬁne θF = V Fϑx V † where V denotes the unitary particlehole transformation or, in the spin language, rotation by π around
204
M. Aizenman et al.
the 1axis. This deﬁnition extends in an obvious way to products of operators on single sites and then, by linearity, to arbitrary operators on HL . Reﬂection positivity of a state · means that
F θF ≥ 0
(3)
for any F operating on HL . Here F is the complex conjugate of the operator F in the matrix representation deﬁned above, i.e., deﬁned by the basis where the operators Sx3 are diagonal. We now show that reﬂection positivity holds for any thermal equilibrium state of our Hamiltonian. We can write the Hamiltonian (2) as H = HL + HR −
1 2
(Sx+ Sy− + Sx− Sy+ ) ,
(4)
xy∈M
where HL and HR act nontrivially only on HL and HR , respectively. Here, M denotes the set of bonds going from the left sublattice to the right sublattice. (Because of the periodic boundary condition these include the bonds that connect the right boundary with the left boundary.) Note that HR = θHL , and (Sx+ Sy− + Sx− Sy+ ) = (Sx+ θSx+ + Sx− θSx− ) . xy∈M
xy∈M
For these properties it is essential that we included the unitary particlehole transformation V in the deﬁnition of the reﬂection θ. For reﬂection positivity it is also important that all operators appearing in H (4) have a real matrix representation. Moreover, the minus sign in (4) is essential. Using the Trotter product formula, we have Tr F θF e−βH = lim Tr F θF Zn n→∞
with
Zn = e− n βHL θe− n βHL 1
1
xy∈M
n % & β S + θS + + Sx− θSx− ) . 1+ 2n x x
(5)
Observe that Zn is a sum of terms of the form > (6) i Ai θAi , 1 β + β − with Ai given by either e− n βHL or 2n Sx or 2n Sx . All the Ai are real matrices, and therefore $ # > > > > 2 TrH F θF i Ai θAi = TrH F i Ai θ F j Aj = TrHL F i Ai ≥ 0 . (7) Hence Tr F θF Zn is a sum of nonnegative terms and therefore nonnegative. This proves our assertion.
BoseEinstein Condensation as a Quantum Phase Transition
205
3 Proof of BEC for Small λ and T The main tool in our proof of BEC are infrared"bounds. More precisely, for p ∈ Λ∗ (the dual lattice of Λ), let Sp# = Λ−1/2 x Sx# exp(ip · x) denote the Fourier transform of the spin operators. We claim that T 1 )≤ , (Sp1 , S−p 2Ep
(8)
"d with Ep = i=1 (1 − cos(pi )). Here, pi denotes the components of p, and ( , ) denotes the Duhamel two point function at temperature T , deﬁned by 1 Tr Ae−sβH Be−(1−s)βH ds/Tr e−βH (9) (A, B) = 0
for any pair of operators A and B. Because of invariance under rotations around the S 3 axis, (8) is equally true with S 1 replaced by S 2 , of course. The crucial lemma (Gaussian domination) is the following. Deﬁne, for a complex valued function h on the bonds xy in Λ, Z(h) = Tr exp [−βK(h)] ,
(10)
with K(h) the modiﬁed Hamiltonian K(h) =
% 2 2 & 1 + 1 Sx − Sy− − hxy + Sx− − Sy+ − hxy +λ +(−1)x Sx3 . 2 4 x xy
(11) Note that for h ≡ 0, K(h) agrees with the Hamiltonian H, because (S ± )2 = 0. We claim that, for any real valued h, Z(h) ≤ Z(0) .
(12)
The infrared bound then follows from d2 Z(εh)/dε2 ε=0 ≤ 0, taking hxy = exp(ip · x) − exp(ip · y). This is not a real function, though, but the negativity of the (real!) quadratic form d2 Z(εh)/dε2 ε=0 for real h implies negativity also for complexvalued h. The proof of (12) is very similar to the proof of the reﬂection positivity property (3) given above. It follows along the same lines as in [19], but we repeat it here for convenience of the reader. The intuition behind (12) is the following. First, in maximizing Z(h) one ˆx − h ˆ y for some function h ˆ x on Λ. (This can restrict to gradients, i.e., hxy = h follows from stationarity of Z(h) at a maximizer hmax .) Reﬂection positivity implies that AθB deﬁnes a scalar product on operators on HL , and hence there is a corresponding Schwarz inequality. Moreover, since reﬂection positivity holds for reﬂections across any hyperplane, one arrives at the socalled chessboard inequality, which is simply a version of Schwarz’s inequality for
206
M. Aizenman et al.
multiple reﬂections across diﬀerent hyperplanes. Such a chessboard estimate ˆx implies that in order to maximize Z(h) it is best to choose the function h to be constant. In the case of classical spin systems [21], this intuition can be turned into a complete proof of (12). Because of noncommutativity of K(h) with K(0) = H, this is not possible in the quantum case. However, one can proceed by using the Trotter formula as follows. Let hmax be a function that maximizes Z(h) for real valued h. If there is more than one maximizer, we choose hmax to be one that vanishes on the largest number of bonds. We then have to show that actually hmax ≡ 0. If hmax ≡ 0, we draw a hyperplane such that hxy = 0 for at least one pair xy crossing the plane. We can again write K(h) = KL (h)+KR (h)+
1 4
(Sx+ − Sy− − hxy )2 + (Sx− − Sy+ − hxy )2 .
xy∈M
(13) Using the Trotter formula, we have Z(h) = limn→∞ αn , with n + − 2 − + 2 αn = Tr e−βKL /n e−βKR /n e−β(Sx −Sy −hxy ) /4n e−β(Sx −Sy −hxy ) /4n . xy∈M
(14) For any matrix, we can write −D 2
e
−1/2
= (4π)
dk eikD e−k
2
/4
.
(15)
R
If we apply this to the last two factors in (14), and note that Sy− = θSx+ if
xy ∈ M . Denoting by x1 , . . . , xl the points on the left side of the bonds in M , we have that # $ + + 1/2 1/2 αn = (4π)−nl d2nl k Tr e−βKL /n e−βKR /n eik1 (Sx1 −θSx1 )β /2n . . . R2nl −k2 /4 −ik1 hx1 ϑx1 β 1/2 /2n1/2 ···
×e
e
.
(16)
" 2 Here we denote k 2 = ki for short. Since matrices on the right of M commute with matrices on the left, and since all matrices in question are real, we see that the trace in the integrand above can be written as $ # $ # + 1/2 1/2 + 1/2 1/2 Tr e−βKL /n eik1 Sx1 β /2n . . . Tr e−βKR /n eik1 θSx1 β /2n . . . . (17) Using the Schwarz inequality for the k integration, and “undoing” the above step, we see that
BoseEinstein Condensation as a Quantum Phase Transition
207
αn 2 ≤
2 (4π)−nl R2nl d2nl k e−k /4 # $ + + 1/2 1/2 × Tr e−βKL /n e−βθKL /n eik1 (Sx1 −θSx1 )β /2n . . . 2 × (4π)−nl R2nl d2nl k e−k /4 # $ + + 1/2 1/2 × Tr e−βθKR /n e−βKR /n eik1 (Sx1 −θSx1 )β /2n . . . .
(18)
In terms of the partition function Z(h), this means that Z(hmax )2 ≤ Z(h(1) )Z(h(2) ) ,
(19)
where h(1) and h(2) are obtained from hmax by reﬂection across M in the following way: hxy if x, y ∈ ΛL hϑxϑy if x, y ∈ ΛR (20) = h(1) xy 0 if xy ∈ M and h(2) is given by the same expression, interchanging L and R. Therefore also h(1) and h(2) must be maximizers of Z(h). However, one of them will contain strictly more zeros than hmax , since hmax does not vanish identically for bonds crossing M . This contradicts our assumption that hmax contains the maximal number of zeros among all maximizers of Z(h). Hence hmax ≡ 0 identically. This completes the proof of (12). The next step is to transfer the upper bound on the Duhamel two point function (8) into an upper bound on the thermal expectation value. This involves convexity arguments and estimations of double commutators like in Sect. 3 in [19]. For this purpose, we have to evaluate the double commutators 2 1 2 ]]+[Sp2 , [H, S−p ]] = − Sx3 Sy3 cos p·(x−y) . H− 12 λΛ+2 [Sp1 , [H, S−p Λ xy
(21) Let Cp denote the expectation value of this last expression, 1 2 Cp = [Sp1 , [H, S−p ]] + [Sp2 , [H, S−p ]] ≥ 0 .
The positivity of Cp can be seen from an eigenfunctionexpansion of the trace. From [19, Corollary 3.2 and Theorem 3.2] and (8) we infer that B Cp 1 1 2 + Sp2 S−p ≤ coth β 2 Cp Ep /4 . (22)
Sp1 S−p 2 Ep Using coth x ≤ 1 + 1/x and Schwarz’s inequality, we obtain for the sum over all p = 0,
208
M. Aizenman et al.
1 1 1 1 2
Sp1 S−p + Sp2 S−p ≤ + β Ep 2
p=0
p=0
1 1/2 Cp Ep
p=0
1/2 . (23)
p=0
" We have p∈Λ∗ Cp = −2 H+λΛ, which can be bounded from above using the following lower bound on the Hamiltonian: H≥−
&1/2 1 Λ % d(d + 1) + 4λ2 + λΛ . 4 2
(24)
This inequality follows from the fact that the lowest eigenvalue of 2d 2d 1 1 1 2 − Sx1 Syi − Sx2 Syi + λSx3 2 2 i=1 i=1
(25)
is given by − 14 [d(d + 1) + 4λ2 ]1/2 . This can be shown exactly in the same way as [19, Theorem C.1]. Since the Hamiltonian H can be written as a sum of terms like (25), with yi the nearest neighbors of x, we get from this fact the lower bound (24). With the aid of the sum rule
1 2
Sp1 S−p + Sp2 S−p =
p∈Λ∗
Λ 2
(which follows from (S 1 )2 = (S 2 )2 = 1/4), we obtain from (23) and (24) the following lower bound in the thermodynamic limit: 1 1 1 2 2
S0 S0 + S0 S0 Λ→∞ Λ 1/2 1 & 1 1 1 % 2 1/2 ≥ − d(d + 1) + 4λ c − cd , d 2 2 2 β lim
with cd given by
1 cd = (2π)d
dp [−π,π]d
1 . Ep
(26)
(27)
This is our ﬁnal result. Note that cd is ﬁnite for d ≥ 3. Hence the right side of (26) is positive, for large enough β, as long as λ2
3. The connection with BEC is as follows. Since H is real, also γ(x, y) is real and we have
BoseEinstein Condensation as a Quantum Phase Transition
209
γ(x, y) = Sx+ Sy− = Sx1 Sy1 + Sx2 Sy2 . Hence, if ϕ0 = Λ−1/2 denotes the constant function,
ϕ0 γϕ0 = S01 S01 + S02 S02 , and thus the bound (26) implies that the largest eigenvalue of γ(x, y) is bounded from below by the right side of (26) times Λ. In addition one can show that the infrared bounds imply that there is at most one large eigenvalue (of the order Λ), and that the corresponding eigenvector (the “condensate wave function”) is strictly constant in the thermodynamic limit [1]. The constancy of the condensate wave function is surprising and is not expected to hold for densities diﬀerent from 12 , where particlehole symmetry is absent. In contrast to the condensate wave function the particle density shows the staggering of the periodic potential [1, Theorem 3]. It also contrasts with the situation for zero interparticle interaction, as discussed at the end of this paper. In the BEC phase there is no gap for adding particles beyond half ﬁlling (in the thermodynamic limit): The ground state energy, Ek , for 12 Λ + k particles satisﬁes (const.) (28) 0 ≤ Ek − E0 ≤ Λ (with a constant that depends on k but not on Λ.) The proof of (28) is by a variational calculation, with a trial state of the form (S0+ )k 0, where 0 denotes the absolute ground state, i.e., the ground state for half ﬁlling. (This is the unique ground state of the Hamiltonian, as can be shown using reﬂection positivity. See Appendix A in [1].) Also, in the thermodynamic limit, the energy per site for a given density, e(#), satisﬁes 2 e(#) − e( 12 ) ≤ const. # − 12 .
(29)
Thus there is no cusp at # = 1/2. To show this, one takes a trial state of the form " 2 ψy = eiε x Sx (Sy1 + 12 )0 . (30) The motivation is the following: we take the ground state and ﬁrst project onto a given direction of S 1 on some site y. If there is longrange order, this should imply that essentially all the spins point in this direction now. Then we rotate slightly around the S 2 axis. The particle number should then go up by εΛ, but the energy only by ε2 Λ. We refer to [1, Sect. IV] for the details. The absence of a gap in the case of BEC is not surprising, since a gap is characteristic for a Mott insulator state. We show the occurrence of a gap, for large enough λ, in the next section.
210
M. Aizenman et al.
4 Absence of BEC and Mott Insulator Phase The main results of this section are the following: If either – λ ≥ 0 and T > d/(2ln2), or – T ≥ 0 and λ ≥ 0 such that λ + e(λ) > d, with e(λ) = ground state energy per site, then there is exponential decay of correlations: γ(x, y) ≤ (const.) exp(−κx − y)
(31)
with κ > 0. Moreover, for T = 0, the ground state energy in a sector of ﬁxed particle number N = 12 Λ + k, denoted by Ek , satisﬁes Ek + E−k − 2E0 ≥ (λ + e(λ) − d)k .
(32)
I.e, for large enough λ the chemical potential has a jump at half ﬁlling. The derivation of these two properties is based on a path integral representation of the equilibrium state at temperature T , and of the ground state which is obtained in the limit T → ∞. The analysis starts from the observation that the density operator e−βH has nonnegative matrix elements in the basis in which {Sx3 } are diagonal, i.e. of states with speciﬁed particle occupation numbers. It is convenient to focus on the dynamics of the ‘quasiparticles’ which are deﬁned so that the presence of one at a site x signiﬁes a deviation there from the occupation state which minimizes the potentialenergy. Since the Hamiltonian is H = H0 + λW , with H0 the hopping term in (2) and W the staggered ﬁeld, we deﬁne the quasiparticle number operators nx as: † for x even ax ax , 1 nx = + (−1)x Sx3 = . (33) 2 1 − a†x ax , for x odd Thus nx = 1 means presence of a particle if x is on the A sublattice (potential maximum) and absence if x is on the B sublattice (potential minimum). The collection of the joint eigenstates of the occupation numbers, {{nx }}, provides a convenient basis for the Hilbert space. The functional integral representation of {nx } e−β(H0 +λW ) {nx } involves an integral over conﬁgurations of quasiparticle loops in a space × time for which the (imaginary) “time” corresponds to a variable with period β. The fact that the integral is over a positive measure facilitates the applicability of statisticalmechanics intuition and tools. One ﬁnds that the quasiparticles are suppressed by the potential energy, but favored by the entropy, which enters this picture due to the presence of the hopping term in H. At large λ, the potential suppression causes localization: long ‘quasiparticle’ loops are rare, and the amplitude for long paths decays exponentially in the distance, both for paths which may occur spontaneously and for paths whose presence is forced through the insertion of sources, i.e., particle creation and annihilation operators. Localization
BoseEinstein Condensation as a Quantum Phase Transition
211
is also caused by high temperature, since the requirement of periodicity implies that at any site which participates in a loop there should be be at least two jumps during the short ‘time’ interval [0, β) and the amplitude for even a single jump is small, of order β. The path integral described above is obtained through the Dyson expansion B(tm ) · · · B(t1 )dt1 · · · dtm (34) et(A+B) = etA m≥0
0≤t1 ≤t2 ≤···≤tm ≤t
for any matrices A and B and t > 0, with B(t) = e−tA BetA . (The m = 0 term in the sum is interpreted here as 1.) In evaluating the matrix elements of e−βH = e−β(H0 +λW ) , in the basis {{nx }}, we note that W is diagonal and {nx }H0 {nx } is nonzero only if the conﬁgurations {nx } and {nx } diﬀer at exactly one nearest neighbor pair of sites where the change corresponds to either a creation of a pair of quasiparticles or the annihilation of such a pair. I.e., the matrix elements are zero unless nx = nx for all x except for a nearest neighbor pair xy, where nx = ny , nx = ny , and nx + nx = 1. In this case, the matrix element equals −1/2. Introducing intermediate states, the partition function can thus be written as follows: ∞ −βH = Tr e m=0
0≤t1 ≤t2 ≤···≤tm ≤β
× exp −λ
m
(i)
{nx }, 1≤i≤m
dt1 · · · dtm (ti − ti−1 ) x i=1 (m) (m) (m−1) ×(−1)m {n(1) } x }H0 {nx } {nx }H0 {nx (m−1) (m−2) (2) (1) × {nx }H0 {nx } · · · {nx }H0 {nx } nx(i)
(35)
with the interpretation t0 = tm − β. Note that the factor in the last two lines of (35) equals (1/2)m if adjacent elements in the sequence of conﬁgurations (i) {nx } diﬀer by exactly one quasiparticle pair, otherwise it is zero. Expansions of this type are explained more fully in [23]. A compact way of writing (35) is: Tr e−βH =
v(dω)e−λω .
(36)
Here the “path” ω stands for a set of disjoint oriented loops in the “spacetime” Λ × [0, β], with periodic boundary conditions in “time””. Each ω is parametrized by a number of jumps, m, jumping times 0 ≤ t1 ≤ t2 ≤ · · · ≤ (i) tm ≤ β, and a sequence of conﬁgurations {nx }, which is determined by (1) the initial conﬁguration {nx } plus a sequence of “rungs” connecting nearest neighbor sites, depicting the creation or annihilation of a pair of neighboring
212
M. Aizenman et al.
β
A
B
A
B
A
B
A
B
A
Fig. 2. Loop gas describing paths of quasiparticles for particle number N = Λ/2− 1. A line on an A site means presence of a particle, while on a B site it means absence. The horizontal rungs correspond to hopping of a particle
quasiparticles (see Fig. 2). As in Feynman’s picture of QED, it is convenient to regard such an event as a jump of the quasiparticle, at which its timeorientation is also reversed. The length of ω, denoted by ω, is the sum of the vertical lengths of the loops. The measure v(dω) is determined (i) by (35); namely, for a given sequence of conﬁgurations {nx }, 1 ≤ i ≤ m, the integration takes places over the times of the jumps, with a measure (1/2)m dt1 · · · dtm . One may note that the measure v(dω) corresponds to a Poisson process of random conﬁgurations of oriented “rungs”, linking neighboring sites at random times, and signifying either the creation or the annihilation of a pair of quasiparticles. The matrix element {nx }e−βH {nx } gets no contribution from rung conﬁgurations that are inconsistent, either internally or with the boundary conditions corresponding to the speciﬁed state vectors. A consistent conﬁguration yields a family of nonoverlapping loops which describe the motion of the quasiparticles in in the “spacetime” Λ × [0, β). Each such conﬁguration contributes with weight e−λω to the above matrix element (another positive factor was absorbed in the measure v(dω)). One may note that long paths are suppressed in the integral (38) at a rate which increases with λ. Likewise, for x = y, we can write v(dω)e−λω , (37) Tr a†x ay e−βH = A(x,y)
where A(x,y) denotes the set of all loops that, besides disjoint closed loops, contain one curve which avoids all the loops and connects x and y at time zero. The oneparticle density matrix can thus be written −λω (x,y) v(dω)e . (38) γ(x, y) = A −λω v(dω)e
BoseEinstein Condensation as a Quantum Phase Transition
213
For an upper bound, we can drop the condition in the numerator that the loops and the curve from x to y do not intersect. The resulting measure space is simply a Cartesian product of the measure space appearing in the denominator and the space of all curves, ζ, connecting x and y, both at time 0. Denoting the latter by B(x, y), we thus get the upper bound v(dζ)e−λζ . (39) γ(x, y) ≤ B(x,y)
The integral over paths is convergent if either λ or T is small enough, and away from the convergence threshold the resulting amplitude decays exponentially. A natural random walk estimate, see [1, Lemma 4], leads to the claimed exponential bound provided (40) d 1 − e−βλ < λ . This includes, in particular, the cases T > d for any λ, and λ > d for any T . Exponential decay actually holds for the larger range of parameters where (41) d 1 − e−β(λ−f ) < λ − f , where f = f (β, λ) = −(βΛ)−1 lnTr e−βH is the free energy per site. Note that f < 0. This condition can be obtained by a more elaborate estimate than the one used in obtaining (39) from (38), as shown in [1, Lemma 3]. The argument there uses reﬂection positivity of the measure v(dω). Using simple bounds on f one can then obtain from (41) the conditions stated in the beginning of this section. Pk e−βH The proof of the energy gap is based on an estimate for the ratio Tr Tr P0 e−βH where Pk projects onto states in Fock space with particle number N = 12 Λ+ k, expressing numerator and denominator in terms of path integrals. The integral for the numerator is over conﬁgurations ω with a nontrivial winding number k. Each such conﬁguration includes a collection of “noncontractible” loops with total length at least βk. An estimate of the relative weight of such loops yields the bound k Tr Pk e−βH k 1−(const.)β e ≤ (const. )(Λ/k) (42) Tr P0 e−βH which gives for β → ∞ Ek − E0 ≥ (const. )k
(43)
independently of Λ. We refer to [1] for details.
5 The NonInteracting Gas The interparticle interaction is essential for the existence of a Mott insulator phase for large λ. In case of absence of the hardcore interaction, there is
214
M. Aizenman et al.
BEC for any density and any λ at low enough temperature (for d ≥ 3). To see this, we have to calculate the spectrum of the oneparticle Hamiltonian − 12 ∆ + V (x), where ∆ denotes the discrete Laplacian and V (x) = λ(−1)x . The spectrum can be easily obtained by noting that V anticommutes with the oﬀdiagonal part of the Laplacian, i.e., {V, ∆ + 2d} = 0. Hence
2 2 1 − ∆ − d + V (x) = − 12 ∆ − d + λ2 , 2
(44)
so the spectrum is given by d±
(
" i
2
cos pi ) + λ2 ,
(45)
where p ∈ Λ∗ . In particular, E(p) − E(0) ∼ 12 d(d2 + λ2 )−1/2 p2 for small p, and hence there is BEC for low enough temperature. Note that the condensate wave function is of course not constant in this case, but rather given by the eigenfunction corresponding to the lowest eigenvalue of − 21 ∆ + λ(−1)x .
6 Conclusion In this paper a lattice model is studied, which is similar to the usual BoseHubbard model and which describes the transition between BoseEinstein condensation and a Mott insulator state as the strength λ of an optical lattice potential is increased. While the model is not soluble in the usual sense, it is possible to prove rigorously all the essential features that are observed experimentally. These include the existence of BEC for small λ and its suppression for large λ, which is a localization phenomenon depending heavily on the fact that the Bose particles interact with each other. The Mott insulator regime is characterized by a gap in the chemical potential, which does not exist in the BEC phase and for which the interaction is also essential. It is possible to derive bounds on the critical λ as a function of temperature.
References 1. M. Aizenman, E.H. Lieb, R. Seiringer, J.P. Solovej, and J. Yngvason, BoseEinstein quantum phase transition in an optical lattice model, Phys. Rev. A 70, 0236121–12 (2004). 2. M. Greiner, O. Mandel, T. Esslinger, T.E. H¨ ansch, I. Bloch, Quantum phase transition from a superﬂuid to a Mott insulator in a gas of ultracold atoms, Nature 415, 39 (2002). 3. M. Greiner, O. Mandel, T.E. H¨ ansch, I. Bloch, Collapse and revival of the matter wave ﬁeld of a BoseEinstein condensate, Nature 419, 51 (2002). 4. S. Sachdev, Quantum Phase Transitions, Cambridge University Press, 1999.
BoseEinstein Condensation as a Quantum Phase Transition
215
5. M.P.A. Fisher, P.B. Weichman, G. Grinstein, D.S. Fisher, Boson localization and the superﬂuidinsulator transition, Phys. Rev. B 40, 546–570 (1989). 6. D. Jaksch, C. Bruder, J.I. Cirac, C.W. Gardiner, P. Zoller, Cold bosonic atoms in optical lattices, Phys. Rev. Lett. 81, 3108–3111 (1998). 7. W. Zwerger, MottHubbard transition of cold atoms in optical lattices, Journal of Optics B 5, 9–16 (2003). 8. J.J. GarciaRipoll, J.I. Cirac, P. Zoller, C. Kollath, U. Schollwoeck, J. von Delft, Variational ansatz for the superﬂuid Mottinsulator transition in optical lattices, Optics Express, 12, 42–54 (2004). 9. K. Ziegler, Phase Transition of a Bose Gas in an Optical Lattice, Laser Physics 13, 587–593 (2003). 10. Z. Nazario, D.I. Santiago, Quantum States of Matter of Simple Bosonic Systems: BEC’s, Superﬂuids and Quantum Solids, Arxiv: condmat/0308005 (2003). 11. G.M. Genkin, Manipulating the superﬂuid – Mott insulator transition of a BoseEinstein condensate in an amplitudemodulated optical lattice, Arxiv: condmat/ 0311589 (2003). 12. K. Ziegler, Twocomponent Bose gas in an optical lattice at singleparticle ﬁlling, Phys. Rev. A 68, 053602 (2003). 13. D.B.M. Dickerscheid, D. van Oosten, P.J.H. Denteneer, H.T.C. Stoof, Ultracold atoms in optical lattices, Phys. Rev. A 68, 043623 (2003). 14. A.M. Rey, K. Burnett, R. Roth, M. Edwards, C.J. Williams, C.W. Clark, Bogoliubov approach to superﬂuidity of atoms in an optical lattice, J. Phys. B 36, 825–841 (2003). 15. O. Morsch, E. Arimondo, Ultracold atoms and BoseEinstein condensates in optical lattices, Lecture Notes in Physics Vol. 602, Springer (2002). 16. E. Altman, A. Auerbach. Oscillating Superﬂuidity of Bosons in Optical Lattices, Phys. Rev. Lett. 89, 250404 (2002). 17. T. Kennedy, E.H. Lieb, S. Shastry, The XY Model has LongRange Order for all Spins and all Dimensions Greater than One, Phys. Rev. Lett. 61, 2582–2584 (1988). 18. T. Matsubara, H. Matsuda, A lattice model of liquid helium, Progr. Theor. Phys. 16, 569–582 (1956). 19. F.J. Dyson, E.H. Lieb, B. Simon, Phase Transitions in Quantum Spin Systems with Isotropic and Nonisotropic Interactions, J. Stat. Phys. 18, 335–383 (1978). 20. K. Osterwalder, R. Schrader, Axioms for Euclidean Green’s Functions, Commun. Math. Phys. 31, 83–112 (1973); Commun. Math. Phys. 42, 281–305 (1975). 21. J. Fr¨ ohlich, B. Simon, T. Spencer, Phase Transitions and Continuous Symmetry Breaking, Phys. Rev. Lett. 36, 804 (1976); Infrared bounds, phase transitions and continuous symmetry breaking, Commun. Math. Phys. 50, 79 (1976). 22. J. Fr¨ ohlich, E.H. Lieb, Phase Transitions in Anisotropic Lattice Spin Systems, Commun. Math. Phys. 60, 233–267 (1978). 23. M. Aizenman, B. Nachtergaele, Geometric Aspects of Quantum Spin States, Commun. Math. Phys. 164, 17–63 (1994).
Long Time Behaviour to the Schr¨ odinger–Poisson–X α Systems ´ Olivier Bokanowski1 , Jos´e L. L´opez2 , Oscar S´ anchez2 , and Juan Soler2 1
2
Lab. Jacques Louis Lions (Paris VI), UFR Math´ematique B. P. 7012 & Universit´e Paris VII [emailprotected] Departamento de Matem´ atica Aplicada, Facultad de Ciencias, Universidad de Granada, Campus Fuentenueva, 18071 Granada [emailprotected] [emailprotected] [emailprotected]
1 Introduction This paper1 is intended to constitute a review of some mathematical theories incorporating quantum corrections to the Schr¨ odingerPoisson (SP) system. More precisely we shall focus our attention in the electrostatic Poisson potential with corrections of power type. The SP system is a simple model used for the study of quantum transport in semiconductor devices that can be written as a system of inﬁnitely many coupled equations (mixedstate) as γ 2 ∂ψj =− ∆x ψj + ∗ nψ ψj , j ∈ N , (1) i ∂t 2m 4πx ψj = ψj (x, t) being the jth component of the vector wave function and where t > 0 is the time variable, x ∈ R3 is the position variable, " ψj (x, 0) = φj (x) is the jth component of the initial condition and nψ := j∈N λj ψj 2 holds denoting the for the charge density. Here, {λj } is a nonnegative sequence " occupation probabilities such that the normalization property j∈N λj = 1 is fulﬁlled. In (1), and m stand for the Planck constant and the particle γ ∗ nψ , where ∗ stands mass respectively. The Poisson potential is Vψ := 4πx for the xconvolution, γ = 1 for repulsive interactions and γ = −1 for attractive interactions. In the sequel we are only concerned with the repulsive system modeling electrostatic forces. The solutions to the SP system have been proved to be strongly dispersive in the sense 1
This work was partially supported by the European Union, TMR–contract IHP HPRN–CT–2002–00282. J.L.L., O.S. and J.S. were also partially sponsored by DGES (Spain), Project MCYT BFM2002–00831. O. Bokanowski et al.: Long Time Behaviour to the Schr¨ odinger–Poisson–X α Systems, Lect. Notes Phys. 690, 217–232 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
218
O. Bokanowski et al.
ψ(·, t)Lp (R3 ) ≤
C(φ) , tβ(p)
p>2,
β(p) > 0 ,
(2)
ﬁrst in the singlestate case λ = (1, 0, 0, . . . ) in [16] and later in the mixedstate case in [7]. These bounds were recently improved in [33]. Here, the key point is the positive character of the total energy 1 1 λj ∇x ψj 2 dx + ∇x Vψ 2 dx . 2 2 3 3 R R j∈N
As consequence, other behaviours like steadystates, breathers, . . . are not allowed by the SP system in contrast to experimental evidence (see [29]). The motivation for the X α approach 1 ∂ψj 2 i =− ∆x ψj + ∗ nψ ψj − CS nα j∈N, (3) ψ ψj , ∂t 2m 4πx as a correction to the Hartree–Fock system stems from experimentation, which evidences that eﬀects due to the Coulomb charging energy could strongly modify the electron tunneling. In this direction, the Schr¨ odinger– Poisson–Slater (SPS) system incorporates quantum corrections of power type 1
to the SP system with α = 13 . The nonlinear Slater term −CS nψ3 comes out as a local correction to the Fock term Vex (see Sect. 2) and should be understood as a quantum eﬀect, following [10, 34] (see also [1, 4, 13] for a mathematical approach), contrary to the Poisson term which has a classical counterpart. Physically, the Slater constant C" S is positive for the case of electrons. Notice that if a diﬀerent normalization j∈N λj = a > 0 is assumed, then we are led to an analogous problem with modiﬁed constants. In particular, the value of CS is genuinely relevant for the subsequent dynamical stability analysis. One important feature of the SPS system is that the total energy operator E[ψ] = EK [ψ] + EP [ψ] , with
(4)
1 λj ∇x ψj 2 dx kinetic energy , 2 3 R j∈N 4 1 3 EP [ψ] = ∇x Vψ 2 dx − CS nψ3 dx potential energy , 2 R3 4 R3
EK [ψ] =
can reach negative values. This implies some relevant consequences: (a) The minimum of the total energy operator is negative. (b) There are steadystate (standing wave) solutions, i.e. solutions with constant density. (c) There are solutions (even with positive energy) which preserve the Lp norm and do not decay with time. Thus, there are solutions that do not exhibit strong dispersive character in the sense of (2).
Long Time Behaviour to the Schr¨ odinger–Poisson–X α Systems
219
(d) Otherwise, in certain regime it can be proved that the solutions are dispersive in a statistical sense (∆x)2 = O(t2 ) i. e., the variance associated with the solution grows with time. This may aim to the study of other type of weaker dispersion properties implying loss of charge at t → ∞. To carry on this analysis the preserved quantities and the Galilean invariance of the system are exploited. These features show important qualitative diﬀerences concerning the dynamics of solutions to the SPS system when compared to the behaviour of solutions to Schr¨ odinger–Poisson or Hartree–Fock systems (see [7,15,16,18]). Accordingly, the SPS system seems to be best realistic and ﬁt properly to semiconductor and heterostructure modeling. Another justiﬁcation for this approach stems from the ambit of quantum chemistry, especially due to the high number of calculations necessary to evaluate the Fock term (usually of order N 4 , N being the number of particles). Several X α approaches to Fock’s correction (for example the Slater approach) have been proved relevant in diﬀerent contexts. The quantum cor1
rection nψ3 ψ is also known as the Dirac exchange term. Another interesting approach comes up from the limit of heavy atoms, i.e. the high–charge–of– nuclei limit. This leads to so–called Thomas–Fermi correction (α = 23 ) of the kinetic energy (see [19, 20]), which can be alternatively seen as a correction to Fock’s interaction showing up as a repulsive potential (see [20]).These local approximations to nonlocal interaction terms yield excellent results when studying stationary states, for example in the frame of quantum chemistry (see [11] and [8,19,23] for details on derivations and analysis of these systems). Then, the number of calculations is reduced from N 4 to N 3 , even leaving a margin to improvements. On the contrary, there is no rigorous foundation of X α models in the timedependent case. Following the classical ideas of thermodynamical limits in statistical mechanics (see [12]), some progress is being currently done on this subject via continuum and mean–ﬁeld limits of the N quantum–particle system (see for instance [2]). In the singlestate case λ = (1, 0, 0, . . . ) the system is reduced to only one equation (for the wave function ψ(x, t) with normalized constants = m = 1). In this case the model belongs to a wider class of nonlinear Schr¨ odinger equations with power nonlinearities, already analyzed in [9]. We also note that the case of N coupled equations, with λj = N1 if 1 ≤ j ≤ N and λj = 0 otherwise, corresponds to the X α correction in Kohn–Sham equations [17]. This model belongs to the density–functional theory approach in Molecular Quantum Chemistry (see [26]) and constitutes a local approximation of the timedependent Hartree–Fock system (see Sect. 2). Concerning the mixedstate case, an existence and uniqueness theory in L2 and H 1 has been recently dealt with in [5] for the SPS system, being the basic ingredient a generalization of Strichartz’ inequalities to mixed states (already exploited in [7]) and Pazy’s ﬁxed–point techniques [27].
220
O. Bokanowski et al.
Our ﬁrst aim in these notes is to set up the origins of the mixed SPS system and of other related models, as well as to show that they are well– posed in R3 . The analysis of some of the qualitative diﬀerences between X α models and the Schr¨ odinger–Poisson and Hartree–Fock systems shall also be dealt with. We are mainly interested in the exponent α = 13 studied in semiconductor theory, which is derived from the Fock term by means of a low density limit (see [5]). Some of our results hold true under precise hypotheses relating the values of the Slater constant, the mass and the energy of the system. It is known that the Slater constant is a characteristic of the component metals in the semiconductor device (see [14]) when interpreting the exchange–correlation potential of Kohn–Sham type. Our analysis covers the whole range of variation for these constants and the relations among them appear in a natural way, not being a restriction from a physical point of view. In this spirit, we shall brieﬂy report on the connections among the physical constants and the minimum of energy in Sect. 5. Attractive interactions are also of interest in applications, for instance to quantum gravity (see [28]). We ﬁnally remark that most of our results are valid for other X α approaches else than Slater’s.
2 On the Derivation of the Slater Approach 1
We now describe how −CS nψ3 enters as a local approximation to the Hartree– Fock exchange potential, which is itself an approximation of the N electron problem. In this case, the system evolution is represented by an N electron state wave function Ψ = Ψ (x1 , . . . , xN , t) (xi ∈ Ω = R3 , 1 ≤ i ≤ N ) that CN belongs to L2a (Ω N ) = i=1 L2 (Ω), that is the space of L2 functions which are antisymmetric with respect to the exchange of any two variables xi and xj . The antisymmetry assumption stems from the Pauli exclusion principle, which establishes that two electrons cannot share the same position, so that the probability density Ψ 2 vanishes as xi = xj . However, for a rigorous treatment the spin variables of the electrons must be considered. We can then assume that the evolution of the system is described by the Schr¨ odinger equation N 2 ∂Ψ = HΨ = − ∆x + i Vext (xj ) + Vee Ψ , ∂t 2m j=1 where Vext is an external potential (such as interactions with nuclei) and Vee :=
1≤i<j≤N
1 xi − xj 
is the electron–electron Coulomb potential, which mixes the variables xi . Then, the total Coulomb energy is given by U (Ψ ) := Ψ, Vee Ψ L2 (Ω N ) .
Long Time Behaviour to the Schr¨ odinger–Poisson–X α Systems
221
In the Hartree–Fock system, determinantal wave functions 1 Ψ (x1 , . . . , xN , t) = √ det(ψi (xj , t)) N! are used to represent the electrons and guarantee that the system obeys the Pauli principle. This expression contains the one–electron wave functions ψi . Now, we assume ψi , ψj = δij , 1 ≤ i, j ≤ N , and skip t for simplicity. Then, classical quantum calculations (see [26]) give U (Ψ ) = Ecoul (nψ ) + Eex (ψ) , "N where ψ = (ψ1 , ψ2 , . . . , ψN ) and nψ = j=1 ψj 2 is the density, nψ (x)nψ (y) 1 dx dy Ecoul (nψ ) = 2 Ω×Ω x − y is the direct Coulomb energy and where the exchange energy is deﬁned by 1 Eex (ψ) = − 2
Ω×Ω
Dψ (x, y)2 dx dy , x − y
Dψ (x, y) :=
N
ψj (x)ψj (y) .
j=1
The Hartree–Fock equations can then be obtained as extrema points of the energy functional 2
Ψ, HΨ = 2m j=1 N
∇x ψj  dx + 2
Ω
nψ Vext dx + U (Ψ ) Ω
under the constraint ψi , ψj L2 (Ω) = δij , which leads to (see [23, 25]) 2 1 ∆x ψj + Vext ψj + ∗ nψ ψj + Vex • ψj = j ψj , 1 ≤ j ≤ N , − 2m x where the term Vex is a nonlocal operator deﬁned by 1 Dψ (x, y)ψj (y, t) dy (Vex • ψj )(x, t) := − x − y Ω
(5)
and the j are constants. The timedependent Hartree–Fock equations are 2 1 ∂ψj =− ∆x ψj + Vext ψj + ∗ nψ ψj + (Vex • ψj ) , 1 ≤ j ≤ N . i ∂t 2m x 1 ∗nψ as well as the exchange operator The Poisson (multiplicative) operator x Vex can be obtained as functional derivatives of the Coulomb and exchange energies Ecoul (nψ ) and Eex (ψ), respectively. Existence and uniqueness of solutions to this problem can be found in [25]. We also refer to [18] for a general existence result for a system coupling the quantum Hartree–Fock equations for electrons with a classical dynamics for the nuclei. For the semiclassical and longtime asymptotics of the Hartree–Fock system, see [15].
222
O. Bokanowski et al.
Remark 1. In>a simpler model for which antisymmetry might not be assumed, N such as Ψ = j=1 ψ(xj ) (with ψ2 dx = 1), we would ﬁnd N − 1 ψ(x)2 ψ(y)2 N (N − 1) dx dy = Ecoul (nψ ) U (Ψ ) = 2 x − y N Ω×Ω with nψ = N ψ2 , thus no exchange term is expected to appear. There is no exact local expression for Vex . However, Vex can be well ap1 proximated by the multiplicative operator −CS nψ3 (for some constant CS , that can be adjusted to ﬁt experimental evidence) as proposed by Slater [34]. This is also called the “X α method”. We refer to [26] for a review of such sort of approximations and numerical experiments. Justiﬁcations of the Slater approximation can be found in [1] (with Ω = R3 ) and [3, 4, 6, 13] for a periodic model (when Ω is a box), in the limit N → ∞ of a large number of particles. We give now an idea of Slater’s proof. Firstly we search for a local ap(V •ψj )(x) is not a local proximation (Vex • ψj )(x) ∼ Vav (x)ψj (x) . However, exψj (x) operator. Following [34], we can approximate Vex by averaging the Hartree– Fock exchange potential by the weighted densities ψj 2 , i.e. Vav (x) =
N (Vex • ψj )(x) ψj (x)2
ψj (x)
j=1
nψ (x)
=− Ω
Dψ (x, y)2 dy . x − y nψ (x)
(6)
Note that the formal approximation Vex ∼ Vav is an exact integral identity, since when the exchange energy is evaluated one ﬁnds Eex =
1 2
1 2 i=1 N
Vav (x)nψ (x) dx = Ω
(Vex • ψi )(x)ψi (x) dx . Ω
The next step is to use a planewave approximation (see [13]). We assume now that Ω = [− L2 , L2 ]3 is a box with periodic boundary conditions. Denote N n0 := Ω the averaged electron density in the box of volume Ω = L3 . Now we take as ﬁrstorder approximations to the singleparticle wave functions the planewave states 1 ψj (x) =
eikj ·x , Ω 3 with kj ∈ 2π L Z and kj ∈ BR , BR being the (euclidean) ball centered at 0 with smallest radius R in order to minimize the kinetic energy
Ekin =
N j=1
In particular
4 3 3 πR
∼
Also ψi , ψj L2 (Ω) = δij
3
Ω
∇x ψj 2 dx =
N
kj 2 .
j=1 1
N = (2π)3 n0 , so that R ∼ Cn03 with C > 0. "N N and nψ = j=1 ψj 2 = Ω = n0 . As consequence, 2π L
Long Time Behaviour to the Schr¨ odinger–Poisson–X α Systems
1 ikj ·(x−y) 1 Dψ (x, y) =
e ∼ (2π)3 Ω j=1 N
223
eih·(x−y) dh .
(7)
BR
This (nontrivial) continuous approximation is related to numbertheory results (see [13]). Actually, the righthand side of (7) is reduced to Dψ (x, y) = CR3 f (R(x − y)) where f (t) =
sin(t) − t cos(t) . t3
Finally, by using a change of variables in (6) and the periodic boundary conditions, we get Dψ (x, x + h)2 f (t)2 1 Vav ∼ − dh ∼ − CR dt . n0 Ω h t RΩ If we assume R → ∞, we ﬁnd that Vav can be expressed as the product of 1 1 R = Cn03 times a convergent integral: Vav ∼ −Cn03 . This approach depends 1
on the dimension. In dimension d we would have Vav ∼ −Cd n0d , although for 1 . d = 3 we should also possibly consider other interaction than x To conclude, we note that it is also important to understand if the above approximation still holds for varying densities nψ (x) = n0 . We refer to [4, 6] 1
for a proof of Vav ∼ − CS nψ3 in a ﬁrstorder approximation, for densities nψ (x) close to n0 and in the limit N → ∞. Remark 2. The Slater and Dirac approximations of Eex are related through 4 "N Dirac Eex (ψ) = − CD Ω nψ3 dx, for some CD > 0 (see [10]) and nψ = j=1 ψj 2 . 1 Dirac Then, Dψi Eex (ψ) = − 43 CD nψ3 ψi .
3 Some Results Concerning Well Posedness and Asymptotic Behaviour In the sequel we shall assume = m = 1 and denote U (t) the propagator of the free Schr¨ odinger Hamiltonian. 3.1 Existence and Uniqueness of Physically Admissible Solutions Deﬁnition 1. Let 1 ≤ p, q ≤ ∞ and T > 0. q,p q q p p (a) We deﬁne Lq,p T (λ) := L ([0, T ]; L (λ)) , Lloc (λ) := Lloc ((0, ∞); L (λ)) q,p q p and L (λ) := L ((0, ∞); L (λ)), where 12 Lp (λ) := φ = {φj }j∈N , φLp (λ) = λj φj 2Lp (R3 ) < ∞ , j∈N
224
O. Bokanowski et al.
(b) If
1 p
+
1 p
=
1 q
+
1 q
= 1, we deﬁne
q,p XTq,p := L∞,2 T (λ) ∩ LT (λ) ,
YTq,p := LT1,2 (λ) + LqT ,p (λ) ,
equipped with the norms ψXTq,p := ψL∞,2 (λ) + ψLq,p T (λ) T
and ψYTq,p :=
inf
ψ1 L1,2 (λ) + ψ2 Lq,p (λ) T
ψ1 +ψ2 =ψ
.
T
In the case T = ∞ we omit the index T and denote X q,p := L∞,2 (λ) ∩ Lq,p loc (λ) ,
Y q,p := L1,2 (λ) + Lqloc,p (λ) .
q,p Also, in the singlestate case we omit the index λ and denote Lq,p T , Lloc and q,p q,p Lq,p instead of LT (λ), Lloc (λ) and Lq,p (λ).
We now introduce the concept of solutions we shall deal with.
Deﬁnition 2 (Mild solution). Given T > 0, we say that ψ ∈ XTq,p is a mild solution of the SPS system if it solves the integral equation t 1 ψ(x, t) = U (t)φ(x) − i U (t − s) Vψ ψ − CS nψ3 ψ (x, s) ds , (8) 0
where V (ψ) =
1 4πx
∗ nψ and nψ =
" j∈N
λj ψj 2 .
It can be seen that (8) makes sense when ψ ∈ XTq,p for some well chosen (q, p) (see [5]). Given φ ∈ L2 (λ), we ﬁrst aim to search for a ﬁxed point of
t
Γ [ψ](x, t) := U (t)φ(x) − i 0
1 U (t − s) Vψ ψ − CS nψ3 ψ (x, s) ds .
For that, we use the deﬁnition of admissible pair given in [9]. Deﬁnition 3. For d = 3, we say that a pair (q, p) is admissible (and denote it by (q, p) ∈ S) if 2 ≤ p < 6 and 2q = 3( 12 − p1 ) . Existence and uniqueness of solutions to the initial value problem for the SPS system (see [5]) is established in the following Theorem 1. Let (q, p) ∈ S with p > 3 and φ ∈ L2 (λ). We have (a) There exists a unique global mild solution ψ ∈ X q,p ∩ C 0 ([0, ∞); L2 (λ)) of the SPS system. Furthermore,
Long Time Behaviour to the Schr¨ odinger–Poisson–X α Systems
225
(i) ∀t > 0, ∀j, k ∈ N, ψj (·, t), ψk (·, t)L2 (R3 ) = φj , φk L2 (R3 ) . In particular, ψ(·, t)L2 (λ) = φL2 (λ) (charge preservation). (ii) If ψ 1 , ψ 2 solve the SPS system with initial data φ1 , φ2 ∈ L2 (λ), then ψ 2 − ψ 1 XTq,p ≤ CT φ2 − φ1 L2 (λ)
∀T > 0 ,
where CT > 0 is a constant that only depends on T and (φj L2 (λ) )j=1,2 . (b) If φ ∈ H 1 (λ), then ∇x ψ ∈ X q,p and the total energy E[ψ](t) is time preserved (c) If CS = 0 the above results still hold true for p > 2 and (q, p) ∈ S. Notice that under the assumption φ ∈ H 1 (λ) there are also other preserved quantities: the linear momentum p(t) := ψ, −i∇x ψL2 (λ) , the angular momentum j(t) := ψ, −i(x ∧ ∇x )ψL2 (λ) and the boost operator
b(t) := ψ, (x − it∇x )ψL2 (λ) (see [9] for the singlestate case). These invariances result via Noether’s theorem in conservation laws of the system and play a relevant role in the analysis of the behaviour of solutions. 3.2 Minimum of Energy For the sake of simplicity we shall deal with the singlestate case from now on. It is not surprising that the Slater term gives rise to some qualitative diﬀerences in the behaviour of the solutions to the SPS system when compared to solutions to the SP system. While the Schr¨ odinger–Poisson energy operator is positive in the repulsive case, it may become negative when the Slater nonlinearity is also considered. Indeed, an application of the inequality 4
8
∇x Vψ 2L2 ≤ CψL3 2 ψ 3 8 L3
8
∀ ψ ∈ L2 (R3 ) ∩ L 3 (R3 ) ,
(9)
C being the best possible constant, yields negativeness for all times of the potential energy associated with the solution ψ(x, t) under the following assumption on the initial data φ of the SPS initial value problem: CS 2C ≤ . 4 3 φL3 2 (R3 ) Remark 3. The inequality (9) as well as an upper bound for the sharp constant C were found by Lieb and Oxford in [22]. In our context, this bound takes the value C = 1.092 2π = 0.1737. Furthermore, since the potential energy might be initially negative in the repulsive case, we can ﬁnd initial data for which the total energy is also 3 negative by scaling arguments (ψσ (x) = σ 2 ψ(σx)).
226
O. Bokanowski et al.
The fact that the total energy can reach negative values allows to establish signiﬁcant diﬀerences in the asymptotic behaviour of solutions to the SP and the SPS systems. For the repulsive SP system it is well known the strong dispersion property (see [7, 16]): given φ ∈ L2 , ψ(·, t)Lp tends to zero asymptotically in time for p ∈ (2, 6]. At variance, ψ(·, t) 83 cannot go L to zero as t → ∞ if ψ is a solution to the repulsive SPS system such that E[φ] < 0. This is because E[ψ] is time preserved and the Slater term is the only nonpositive contribution to the total energy. Motivated by these estimates, one can proceed to analyze the following minimization problem associated with the total energy of the SPS system . (10) IM = inf E[ψ]; ψ ∈ H 1 (R3 ), ψL2 = M . Existence of a minimizer for (10) implies two interesting consequences. (a) The existence of stationary proﬁles, which are periodicintime solutions to the SPS system preserving the density. (b) The derivation of optimal bounds for the kinetic energy linked to solutions ψ for which E[ψ] is welldeﬁned. Observe that things are sensibly diﬀerent if we deal with the repulsive SP system, since in this case the minimization problem has no solution because the energy inﬁmum is always set to 0, which is not a minimum except for the case M = 0. In particular, stationary solutions do not exist. The technical diﬃculties arising in the (nonconvex) minimization problem (10) stem from the invariance of E[ψ] by the noncompact group of translations as well as by the change of sign in the terms contributing to the potential energy, which prevents rearrangement techniques to be used. The possible loss of compactness due to that invariance has to be detected by the techniques used in the proofs. In this way, two methods are commonly used in the literature to solve this kind of problems: the concentration–compactness method [24] and the method of nonzero weak convergence after translations [21]. In fact, it can be proved that every minimizing sequence is in essence relatively compact provided that a certain subadditivity property is strict. This condition implies that a minimizing sequence is concentrated in a bounded domain. By considering this sort of arguments (under certain technical assumptions) the results stated in the sequel were all proved in [32]. Theorem 2. Let CS , M > 0 be such that M
0. Then, the system expands unboundedly for large times and the position dispersion x2 (t) grows like O(t2 ). We can straightforwardly deduce lower bounds for the Lp norm of the solutions, which are either positive constants or coincide with the usual decay rates of the free Schr¨ odinger equation, depending on a relation linking the total energy, the mass and the linear momentum.
Long Time Behaviour to the Schr¨ odinger–Poisson–X α Systems
229
Corollary 1. Let ψ(x, t) be a solution of the SPS system with initial data φ ∈ Σ such that 1  p[φ]2 . (18) E[φ] < 2 φL2 Then, there exist positive constants C, C and C depending on φL2 , E[φ],  p(0)[φ]2 and p such that #8 $ ψ(t)Lp (R3 ) ≥ C, EP [ψ] ≤ − C , ∀ t ≥ 0 , ∀ p ∈ , 6 . (19) 3 In the case that E[φ] ≥
1  p[φ]2 , 2 φL2
(20)
the following lower bound ψ(·, t)Lp (R3 ) ≥
C t
3p−6 2p
,
∀t > ξ > 0 ,
∀ p ∈ [2, 6] ,
(21)
holds. Furthermore, if E[φ] < 0 and φL2 = 1 we have (∆x)2 = O(t2 ) . The proof combines the dispersion properties with the Galilean invariance of the system (see [5]). In fact, this property guarantees that if ψ(x, t) is a solution to the SPS system with initial data φ, then the solution corresponding to the initial data ψN (x, 0) = eiN x φ(x), with N ∈ R3 , is 2 ψN (x, t) = eiN x−itN ψ(x − 2tN, t). The inequalities (18) and (20) are motivated by this property. We also remark that these solutions verify ψN (·, t)Lp = φLp ,
p≥2
and EP (ψN ) = EP (φ). The next result provides a positive rateofdecay estimate for the potential energy. Recall that the potential energy may be negative as shown before. In the case that the initial condition satisﬁes (20) we ﬁnd additional information on the potential energy in the following Proposition 3. Let ψ be the (unique) solution to the SPS system with initial data φ ∈ Σ. Then Cξ ∀t ≥ ξ > 0 , (22) EP [ψ](t) ≤ t where Cξ is a positive constant depending on ξ. Consider now the function
∇x V (x, t)2 dx fψ (t) = (x + it∇)ψ(·, t)2L2 + t2 R3 8 3 2 ψ(x, t) 3 dx . − CS t 2 3 R
(23)
230
O. Bokanowski et al.
From (17) and (22) we get fψ (t) ≤ C + ξ
t
Cξ s ds ≤ Cξ t ∀ t ≥ ξ > 0 . s
The evolution of fψ (more precisely, the evolution of its sign) implies qualitative diﬀerences in the behaviour of ψ. The following result provides either a decay estimate for EP [ψ] (in the attractive case) either a weak decay property for some Lp,q norms of ψ. Corollary 2. Let fψ be as in (23). If there exists t0 ∈ R+ such that fψ (t0 ) < 0, then fψ (t) < 0 for all t ≥ t0 and fψ (t0 ) 1 2EP [ψ] ≤ < 0 ∀ t ≥ t0 . t0 t Otherwise, we have ξ
∞
4p
ψ(s)L3(p−2) p (R3 ) ds ≤ C
∀ p ∈ (2, 6] ,
where C is a positive constant depending on p, φL2 , xφL2 and ξ > 0.
5 On the General X α Case As before, E[ψ] also reaches negative values when the Poisson potential is coupled with power nonlinearities Cα ψα ψ with α ∈ (0, 43 ]. Even combinations of some of these terms (with plus or minus sign) make this property to hold (see [19]), leading eventually to a convex energy functional to be minimized via standard techniques for convex $operators. Let us remark that in the interval 0, 43 there are two critical (with respect to the minimization of the energy functional) exponents. The Slater exponent α = 13 is critical because for all 0 < α < 13it is easy to prove that the energy may reach negative values. However, in 13 , 23 it can be proved that the energy also assumes negative values if a positive constant K(α) depending upon α exists such that the interaction constant Cα and the total charge φL2 fulﬁll the following relation Cα 4(1−2α) φL2
> K(α) > 0 .
The other critical case for the energy minimization is α = 23 because the total energy operator is not bounded from below when it reaches negative values. This opens the problem of studying the balance of the kinetic energy against the potential energy [31].
Long Time Behaviour to the Schr¨ odinger–Poisson–X α Systems
231
References 1. V. Bach: Accuracy of mean ﬁeld approximations for atoms and molecules. Commun. Math. Phys. 155, 295 (1993) . 2. C. Bardos: The weak coupling limit of systems of N quantum particles. In: Euroconference on Asymptotic Methods and Applications in Kinetic and QuantumKinetic Theory’, ed. by L.L. Bonilla, J. Soler, J.L. V´ azquez (Granada 2001). 3. O. Bokanowski, B. Gr´ebert, N. Mauser: Approximations de l’´energie cin´etique en fonction de la densit´e pour un mod`ele de Coulomb p´eriodique. C.R. Acad. Sci., Math. Phys. 329 85 (1999). 4. O. Bokanowski, B. Gr´ebert, N. Mauser: Local density approximations for the energy of a periodic Coulomb model. Math. Mod. Meth. Appl. Sci. 13 1185 (2003). 5. O. Bokanowski, J.L. L´ opez, J. Soler: On an exchange interaction model for quantum transport: the Schr¨ odinger–Poisson–Slater system. Math. Mod. Meth. Appl. Sci. 13 1 (2003). 6. O. Bokanowski, N. Mauser: Local approximation for the Hartree–Fock exchange potential: a deformation approach. Math. Mod. Meth. Appl. Sci. 9(6) 941 (1999). odinger–Poisson system: existence, unique7. F. Castella: L2 solutions to the Schr¨ ness, time behaviour and smoothing eﬀects. Math. Mod. Meth. Appl. Sci. 8 1051 (1997). 8. I. Catto, P.L. Lions: Binding of atoms and stability of molecules in Hartree and Thomas–Fermi type theories. Part 1: A necessary and suﬃcient condition for the stability of general molecular systems. Commun. Partial Diﬀ. Equ. 17 1051 (1992). 9. T. Cazenave: An Introduction to Nonlinear Schr¨ odinger Equations, 2nd edn (Textos de M´etodos Matem´ aticos 26, Universidade Federal do Rio de Janeiro 1993). 10. P.A.M. Dirac: Note on exchange phenomena in the Thomas–Fermi atom. Proc. Cambridge Philos. Soc. 26 376 (1931). 11. R.M. Dreizler, E.K.U. Gross: Density Functional Theory (Springer, Berlin Heidelberg New York 1990). 12. R.S. Ellis: Entropy, Large Deviations and Statistical Mechanics (Springer, Berlin Heidelberg New York 1985). 13. G. Friesecke: Pair correlation and exchange phenomena in the free electron gas. Comm. Math. Phys. 184 143 (1997). 14. M.K. Harbola, V. Sahni: Quantummechanical interpretation of the exchangecorrelation potential of Kohn–Sham density–functional theory. Phys. Rev. Lett. 62 448 (1989). 15. I. Gasser, R. Illner, P.A. Markowich et al.: Semiclassical, t → ∞ asymptotics and dispersive eﬀects for Hartree–Fock systems. M2 AN 32(6) 699 (1998). 16. R. Illner, P. Zweifel, H. Lange: Global existence, uniqueness and asymptotic behaviour of solutions of the Wigner–Poisson and Schr¨ odinger–Poisson systems. M2 AS 17 349 (1994). 17. W. Kohn, L.J. Sham: Selfconsistent equations including exchange and correlation eﬀects. Phys. Rev. 140 A 1133 (1965).
232
O. Bokanowski et al.
18. C. LeBris, E. Canc`es: On the timedependent Hartree–Fock equations coupled with a classical nuclear dynamics. Math. Mod. Meth. Appl. Sci. 9(7) 963 (1999). 19. E.H. Lieb: Thomas–Fermi and related theories of atoms and molecules. Rev. Mod. Phys. 53 603 (1981). 20. E.H. Lieb: Thomas–Fermi theory. In: Kluwer Encyclopedia of Mathematics, suppl. Vol 2 (2000) pp. 311–313. 21. E.H. Lieb, M. Loss: Analysis (Graduate Studies in Mathematics, Vol. 14. Amer. Math. Soc., Providence, Rhode Island 2001). 22. E.H. Lieb, S. Oxford: Improved lower bound on the indirect Coulomb energy. Int. J. Quant. Chem. 19 427 (1981). 23. E.H. Lieb, B. Simon: The Hartree–Fock theory for Coulomb systems. Comm. Math. Phys. 53 185 (1977). 24. P.L. Lions: The concentrationcompactness principle in the calculus of variations. The locally compact case. Ann. Inst. H. Poincar´e 1 109 & 223 (1984). 25. P.L. Lions: Solutions of Hartree–Fock equations for Coulomb systems. Comm. Math. Phys. 109, 33 (1987). 26. R.G. Parr, W. Yang: Density Functional Theory of Atoms and Molecules (Oxford University Press 1989). 27. A. Pazy: Semigroups of Linear Operators and Applications to Partial Diﬀerential Equations (Springer, Berlin Heidelberg New York 1983). 28. R. Penrose: On gravity’s role in quantum state reduction. Gen. Rel. Grav. 28, 581 (1996). 29. A. Puente, L. Serra: Oscillation modes of twodimensional nanostructures within the timedependent localspindensity approximation. Phys. Rev. Lett. 83, 3266 (1999). 30. E. Ru´ız Arriola, J. Soler: A variational approach to the Schr¨ odinger–Poisson system: asymptotic behaviour, breathers and stability. J. Stat. Phys. 103, 1069 (2001). 31. O. S´ anchez, in preparation. 32. O. S´ anchez, J. Soler: Longtime dynamics of the Schr¨ odinger–Poisson–Slater system. J. Stat. Phys. 114, 179 (2004). 33. O. S´ anchez, J. Soler: Asymptotic decay estimates for the repulsive Schr¨ odinger– Poisson system. Math. Mod. Appl. Sci. 27, 371 (2004). 34. J.C. Slater: A simpliﬁcation of the Hartree–Fock method. Phys. Rev. 81(3), 385 (1951).
Towards the Quantum Brownian Motion L´aszl´o Erd˝ os1 , Manfred Salmhofer2 , and HorngTzer Yau3 1
2
3
Institute of Mathematics, University of Munich, Theresienstr. 39, D80333 Munich [emailprotected] Theoretical Physics, University of Leipzig, Augustusplatz 10, D04109 Leipzig, and Max–Planck Institute for Mathematics, Inselstr. 22, D04103 Leipzig [emailprotected] Department of Mathematics, Stanford University, CA94305, USA [emailprotected]
Abstract. We consider random Schr¨ odinger equations on Rd or Zd for d ≥ 3 with uncorrelated, identically distributed random potential. Denote by λ the coupling constant and ψt the solution with initial data ψ0 . Suppose that the space and time variables scale as x ∼ λ−2−κ/2 , t ∼ λ−2−κ with 0 < κ ≤ κ0 , where κ0 is a suﬃciently small universal constant. We prove that the expectation value of the Wigner distribution of ψt , EWψt (x, v), converges weakly to a solution of a heat equation in the space variable x for arbitrary L2 initial data in the weak coupling limit λ → 0. The diﬀusion coeﬃcient is uniquely determined by the kinetic energy associated to the momentum v.
1 Introduction Brown observed almost two centuries ago that the motion of a pollen suspended in water was erratic. This led to the kinetic explanation by Einstein in 1905 that Brownian motion was created by the constant “kicks” on the relatively heavy pollen by the light water molecules. Einstein’s theory, based upon Newtonian dynamics of the particles, in fact postulated the emergence of the Brownian motion from a classical nondissipative reversible dynamics. Einstein’s theory became universally accepted after the experimental veriﬁcation by Perrin in 1908, but it was far from being mathematically rigorous. The key diﬃculty is similar to the justiﬁcation of Boltzmann’s molecular chaos assumption (Stoßzahlansatz) standing behind Boltzmann’s derivation of the Boltzmann equation. The point is that the dissipative character emerges only in a scaling limit, as the number of degrees of freedom goes to inﬁnity. The ﬁrst mathematical deﬁnition of the Brownian motion was given in 1923 by Wiener, who constructed the Brownian motion as a scaling limit of random walks. This construction was built upon a stochastic microscopic dynamics which by itself are dissipative. The derivation of the Brownian motion from a Hamiltonian dynamics was not seriously investigated until the end of the seventies, when several L. Erd˝ os et al.: Towards the Quantum Brownian Motion, Lect. Notes Phys. 690, 233–257 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
234
L. Erd˝ os et al.
results came out almost simultaneously. Kesten and Papanicolaou [16] proved that the velocity distribution of a particle moving in a random scatterer environment (socalled Lorenz gas with random scatterers) converges to the Brownian motion in a weak coupling limit for d ≥ 3. The same result was obtained in d = 2 dimensions by D¨ urr, Goldstein and Lebowitz [8]. In this model the bath of light particles is replaced with random static impurities. In a very recent work [18], Komorowski and Ryzhik have controlled the same evolution on a longer time scale and proved the convergence to Brownian motion of the position process as well. BunimovichSinai [5] proved the convergence of the periodic Lorenz gas with a hard core interaction to a Brownian motion. In this model the only source of randomness is the distribution of the initial condition. Finally, D¨ urrGoldsteinLebowitz [7] proved that the velocity process of a heavy particle in a light ideal gas converges to the OrnsteinUhlenbeck process that is a version of the Brownian motion. This model is the closest to the one in Einstein’s kinetic argument. An analogous development happened around the same time towards the rigorous derivation of the Boltzmann equation. It was proved by Gallavotti [14], Spohn [26] and Boldrighini, Bunimovich and Sinai [3] that the dynamics of the Lorenz gas with random scatterers converges to the linear Boltzmann equation at low density on the kinetic time scale. Lanford [19] has proved that a truly manybody classical system, a low density gas with hardcore interaction, converges to the nonlinear Boltzmann equation for short macroscopic times. Brownian motion was discovered and theorized in the context of classical dynamics. Since it postulates a microscopic Newtonian model for atoms and molecules, it is natural to replace the Newtonian dynamics with the Schr¨ odinger dynamics and investigate if Brownian motion correctly describes the motion of a quantum particle in a random environment as well. One may of course take ﬁrst the semiclassical limit, reduce the problem to the classical dynamics and then consider the scaling limit. This argument, however, does not apply to particles (or Lorenz scatterers) of size comparable with the Planck scale. It is physically more realistic and technically considerably more challenging to investigate the scaling limit of the quantum dynamics directly without any semiclassical limit. We shall prove that Brownian motion also describes the motion of a quantum particle in this situation. It is remarkable that the Schr¨ odinger evolution, which is time reversible and describes wave phenomena, converges to a Brownian motion. The random Schr¨ odinger equation, or the quantum Lorentz model, is given by the evolution equation: i∂t ψt (x) = Hψt (x),
1 H = Hω = − ∆x + λVω (x) 2
where λ > 0 is the coupling constant and Vω is the random potential.
(1)
Towards the Quantum Brownian Motion
235
The ﬁrst time scale with a nontrivial limiting dynamics is the weak coupling limit, λ → 0, where space and time are subject to kinetic scaling and the coupling constant scales as √ (2) t → tε−1 , x → xε−1 , λ = ε . Under this limit, the appropriately rescaled phase space density (Wigner distribution, see (10) later) of the solution to the Schr¨ odinger evolution (1) converges weakly to a linear Boltzmann equation. This was ﬁrst established by Spohn (1977) [25] if the random potential is a Gaussian random ﬁeld and the macroscopic time is small. This method was extended to study higher order correlations by Ho, Landau and Wilkins [13]. A diﬀerent method was developed in [10] where the short time restriction was removed. This method was also extended to the phonon case in [9] and to the lattice case in [6]. For longer time scales, one expects a diﬀusive dynamics since the long time limit of a Boltzmann equation is a heat equation. We shall therefore take a time scale longer than in the weak coupling limit (2), i.e. we set t ∼ λ−2−κ , κ > 0. Our aim is to prove that the limiting dynamics of the Schr¨ odinger evolution in a random potential under this scaling is governed by a heat equation. This problem requires to control the Schr¨ odinger dynamics up to a time scale λ−2−κ . This is a much harder task than ﬁrst deriving the Boltzmann equation from Schr¨ odinger dynamics on the kinetic scale and then showing that Boltzmann equation converges to a diﬀusive equation under a diﬀerent limiting procedure. Quantum correlations that are small on the kinetic scale and are neglected in the ﬁrst limit, may contribute on the longer time scale. We consider two models in parallel. In the discrete setup we put the Schr¨ odinger equation (1) on Zd , i.e. we work with the Anderson model [2]. Thus the kinetic energy operator on 2 (Zd ) is given by (∆f )(x) := 2d f (x) − f (x + e) (3) e=1
and the random potential is given by Vω (x) = Vγ (x) ,
Vγ (x) := vγ δ(x − γ)
(4)
γ∈Zd
where vγ are real i.i.d. random variables and δ is the lattice delta function, δ(0) = 1 and δ(y) = 0, y = 0. In the continuum model we consider the usual Laplacian, − 12 ∆x , as the kinetic energy operator on L2 (Rd ). The random potential is given by B(x − y)dµω (y) , (5) Vω (x) = Rd
where µω is a Poisson process {yγ : γ" = 1, 2, . . .} on Rd with unit density and i.i.d. random masses, vγ , i.e. µω = γ vγ δ(· − yγ ), and B : Rd → R is a smooth, radially symmetric function with rapid decay, with 0 in the support H of B.
236
L. Erd˝ os et al.
Since we investigate large distance phenomena, there should be no physical diﬀerence between the continuum and discrete models. On the technical level, the discrete model is more complicated due to the nonconvexity of the energy surfaces of the discrete Laplacian in momentum space. However, the continuum model also has an additional technical diﬃculty: the large momentum regime needs a separate treatment. Our proof builds upon the method initiated in [10]. In that paper the continuum model with a Gaussian random ﬁeld was considered. Here we also consider the discrete model and nonGaussian randomness, in order to demonstrate that these restrictions are not essential. On the Boltzmann scale this extension has also been achieved by Chen [6]. The other reason for working on the lattice as well is to make a connection with the extended state conjecture in the Anderson model. We recall that the Anderson model was invented to describe the electric conduction properties of disordered metals. It was postulated by Anderson that for localized initial data the wave functions for large time are localized for large coupling constant λ and are extended for small coupling constant (away from the band edges and in dimension d ≥ 3). The localization conjecture was ﬁrst established rigorously by Goldsheid, Molchanov and Pastur [15] in one dimension, by Fr¨ ohlichSpencer [12], and later by AizenmanMolchanov [1] in several dimensions, and many other works have since contributed to this ﬁeld. The extended state conjecture, however, has remained a diﬃcult open problem and only very limited progress has been made. Most approaches on extended states focused on the spectral property of the random Hamiltonian. It was proved by Klein [17] that all eigenfuctions are extended on the Bethe lattice. In Euclidean space, Schlag, Shubin and Wolﬀ [24] proved that the eigenfunctions cannot be localized in a region smaller than λ−2+δ for some δ > 0 in d = 2. Chen [6], extending the method of [10] to the lattice case, proved that the eigenfunctions cannot be localized in a region smaller than λ−2 in any dimension d ≥ 2 with logarithmic corrections. Lukkarinen and Spohn [21] have employed a similar technique for studying energy transport in a harmonic crystal with weakly perturbed random masses. A special class of random Schr¨ odinger equation was proposed to understand the dynamics in the extended region. Instead of random potential with i.i.d. random variables, one considers a random potential Vω (x) with a power law decay, i.e., h(x) ∼ x−η Vω (x) = h(x)ωx , where ωx are mean zero i.i.d. random variables and η > 0 is a ﬁxed parameter. If η ≥ 1 a standard scattering argument yields that for λ small enough Hω has absolutely continuous spectrum. Using cancellation properties of the random potential, Rodnianski and Schlag [22] have improved the same result to η > 3/4 in d ≥ 2 and recently, J. Bourgain [4] has extended it to η > 1/2.
Towards the Quantum Brownian Motion
237
For η > 1/2 the particle becomes essentially ballistic at large distances and there are only ﬁnitely many eﬀective collisions. In summary, in all known results [4,6,22,24] for the Anderson model (or its modiﬁcation) in Euclidean space the number of eﬀective collisions are ﬁnite. In the scaling of the current work (13), the number of eﬀective scatterings goes to inﬁnity in the scaling limit, as it should be the case if we aim to obtain a Brownian motion. As in [6], our dynamical result also implies that the eigenfunctions cannot be localized in a region smaller than λ−2−δ for some δ > 0 and dimension d ≥ 3 (one can choose δ = κ/2 with κ from Theorem 1). Though this result is the strongest in the direction of eigenfunction delocalization, we do not focus on it here. Our main result is that the time reversible Schr¨ odinger evolution with random impurities on a time scale λ−2−κ is described by a dissipative dynamics. In fact, this work is the ﬁrst rigorous result where a heat equation is established from a time dependent quantum dynamics without ﬁrst passing through a semiclassical limit. In this contribution we explain the result and the key ideas in an informal manner. The complete proof is given in [11].
2 Statement of Main Result We consider the discrete and the continuum models in paralell, therefore we work either on the ddimensional lattice, Zd , or on the continuous space, Rd . We always assume d ≥ 3. Let 1 Hω := − ∆ + λVω 2
(6)
denote a random Schr¨ odinger operator acting on H = l2 (Zd ), or H = L2 (Rd ). The kinetic energy operator and the random potential are deﬁned in (3)–(5). We assume that Evγ = Evγ3 = 0, Evγ2 = 1 and Evγ2d < ∞. In the discrete case, the Fourier transform is given by e−2πip·x f (x) , fH(p) ≡ (Ff )(p) := x∈Zd
where p = (p(1) , . . . , p(d) ) ∈ Td := [− 12 , 12 ]d . Sometimes an integral notation will be used for the normalized summation over any lattice (δZ)d : (· · · )dx := δ d (· · · ) . x∈(δZ)d
The inverse Fourier transform is given by
238
L. Erd˝ os et al.
(F −1 gH)(x) = . . .
gH(p)e2πip·x dp . (T/δ)d
In the continuous case the Fourier transform and its inverse are given by e−2πip·x f (x)dx , (F −1 gH)(x) = gH(p)e2πip·x dp . (Ff )(p) := Rd
Rd
We will discuss thetwo cases in parallel, in particular we will use the uniﬁed integral notations (· · · )dx and (· · · )dp. The letters x, y, z will always be used for position space coordinates (hence elements of (δZ)d or Rd ). The letters p, q, r, u, v, w denote for ddimensional momentum variables (elements of (T/δ)d or Rd ). The Fourier transform of the kinetic energy operator is given by # 1 $ F − ∆ f (p) = e(p)fH(p) . 2 The dispersion law, e(p), is given by e(p) :=
d (1 − cos(2πp(i) )),
and
i=1
e(p) :=
1 2 p 2
in the discrete and in the continuous case, respectively. For h : Td → C and an energy value e ∈ [0, 2d] we introduce the notation dν(q) (7) h(q) [h](e) := h(v)δ(e − e(v))dv := ∇e(q) Σe where dν(q) = dνe (q) is the restriction of the ddimensional Lebesgue measure to the level surface Σe := {q : e(q) = e} ⊂ Td . By the coarea formula it holds that 2d
[h](e)de =
h(v)dv .
(8)
We deﬁne the projection onto the energy space of the free Laplacian by [h](e) , where Φ(e) := [1](e) = δ(e − e(u))du . (9)
h(v) e := Φ(e) In the continuous case we deﬁne analogous formulas for any function h : Rd → C and energy value e ≥ 0. Deﬁne the Wigner transform of a function ψ ∈ L2 (Zd ) or ψ ∈ L2 (Rd ) via its Fourier transform by w H w dw . Wψ (x, v) := e2πiw·x ψH v − ψ v+ 2 2
Towards the Quantum Brownian Motion
239
In the lattice case the integration domain is the double torus (2T)d and x runs over the reﬁned lattice, x ∈ (Z/2)d . For ε > 0 deﬁne the rescaled Wigner distribution as X ,V . (10) Wψε (X, V ) := ε−d Wψ ε (with X ∈ (εZ/2)d in the lattice case). The weak coupling limit is deﬁned by the following scaling: T := εt,
X := εx,
ε = λ2 .
(11)
In the limit ε → 0 the Wigner distribution Wψε −1 (X , V) converges weakly ε T to the Boltzmann equation ( [6, 10]) # $ 1 ∇e(V ) · ∇X FT (X , V ) = dU σ(U, V ) FT (X , U ) − FT (X , V ) ∂T + 2π (12) 1 where 2π ∇e(V ) is the velocity. The collision kernel is given by σ(U, V ) := 2πδ(e(U ) − e(V )) H − V )2 δ(e(U ) − e(V )) σ(U, V ) := 2πB(U
discrete case continuous case .
Note that the Boltzmann equation can be viewed as the generator of a Markovian semigroup on phase space. In particular, the validity of the Boltzmann equation shows that all correlation eﬀects become negligible in this scaling limit. Now we consider the long time scaling, i.e. with some κ > 0, x = λ−κ/2−2 X = ε−1 X,
t = λ−κ−2 T = ε−1 λ−κ/2 T,
ε = λκ/2+2 .
(13)
Theorem 1. [Quantum Diﬀusion on Lattice] Let d = 3 and ψ0 ∈ 2 (Zd ) λ solve the be an initial wave function with ψH0 ∈ C 1 (Td ). Let ψ(t) = ψt,ω v) be a function on Rd ×Td whose Fourier Schr¨ odinger equation (1). Let O(x, transform in the ﬁrst variable, denoted by O(ξ, v), is a C 1 function on Rd ×Td and dξ dvO(ξ, v)ξ ≤ C . (14) Rd
Fix e ∈ [0, 2d]. Let f be the solution to the heat equation ∂T f (T, X, e) = ∇X · D(e)∇X f (T, X, e) with the initial condition
# $ f (0, X, e) := δ(X) ψH0 (v)2 (e)
and the diﬀusion matrix D
(15)
240
L. Erd˝ os et al.
I Dij (e) :=
sin(2πv (i) ) · sin(2πv (j) ) 2π Φ(e)
J e
i, j = 1, 2, 3 .
(16)
Then for κ < 1/2000 and ε and λ related by (13), the Wigner distribution satisﬁes ε ˜ lim dX dv O(X, v)EWψ(λ (17) −κ−2 T ) (X, v) ε→0
(εZ)d
Rd
=
˜ dv O(X, v)f (T, X, e(v)) .
dX Rd
Rd
By the symmetry of the measure ·e under each sign ﬂip vj → −vj we see that D(e) is a constant times the identity matrix: I 2 J sin (2πv (1) ) e , De := Dij (e) = De δij , 2π Φ(e) in particular we see that the diﬀusion is nondegenerate. The diﬀusion matrix can also be obtained from the long time limit of the Boltzmann equation (12). For any ﬁxed energy e, let Le f (v) := du σ(u, v)[f (u) − f (v)], e(v) = e , (18) be the generator of the momentum jump process on Σe with the uniform stationary measure ·e . The diﬀusion matrix in general is given by the velocity autocorrelation function ∞ I J dt sin(2πv (i) (t)) · sin(2πv (j) (0)) e , (19) Dij (e) = 0
where v(t) is the process generated by Le . Since the collision kernel σ(U, V ) is uniform, the correlation between v(t) and v(0) vanishes after the ﬁrst jump and we obtain (16), using du σ(u, v) = 2πΦ(e) , e(v) = e . The result in the continuum case is analogous. The diﬀusion matrix is again a constant times the identity matrix, Dij (e) = De δij , and De is again given by the velocity autocorrelation function ∞ 1 De := dt v(t) · v(0)e (20) 3(2π)2 0 using the spatial isotropy. In this case De cannot be computed as a simple integral since the outgoing velocity u in the transition kernel σ(u, v) of the momentum process depends on the direction of v.
Towards the Quantum Brownian Motion
241
Theorem 2. [Quantum Diﬀusion on Rd ] Let d = 3 and ψ0 ∈ L2 (Rd ) be an initial wave function with ψH0 (v)2 vN ∈ L2 for a suﬃciently large N . λ v) be a funcLet ψ(t) = ψt,ω solve the Schr¨ odinger equation (1). Let O(x, tion whose Fourier transform in x, denoted by O(ξ, v), is a C 1 function on Rd × Rd and dξdvO(ξ, v)ξ ≤ C . (21) Let e > 0 and let f be the solution to the heat equation ∂T f (T, X, e) = De ∆X f (T, X, e)
(22)
with diﬀusion constant De given in (20) and with the initial condition $ # f (0, X, e) := δ(X) ψH0 (v)2 (e) . Then for κ < 1/500 and ε and λ related by (13), the Wigner distribution satisﬁes ε lim dXdv O(X, v)EWψ(λ (23) −κ−2 T ) (X, v) ε→0
Rd ×Rd
= Rd ×Rd
dXdv O(X, v)f (T, X, e(v)) .
The main tool of our proof is to use the Duhamel expansion to decompose the wave function into elementary wavefuctions characterized by their collision histories with the random obstacles. Assume for the moment that the randomness is Gaussian and high order expectations can be computed by Wick pairing. The higher order cumulants arising from a nonGaussian randomness turn out to be negligible by a separate argument. Therefore, when computing the expectation of a product involving ψ and ψ¯ (e.g. E Wψ ), we pair the obstacles in the collision histories of ψ and ψ¯ and we thus generate Feynman graphs. If we take only the Laplacian as the free part in the expansion, even the amplitudes of individual graphs diverge in the limit we consider. However, this can be remedied by a simple resummation of all twolegged insertions caused by the lowest order selfenergy contribution. The resummation is performed by choosing an appropriate reference Hamiltonian H0 for the expansion. After this rearrangement, all graphs have a ﬁnite amplitude in our scaling limit, and the socalled ladder graphs give the leading contribution. Each nonladder graph has a vanishing amplitude as λ → 0 due to oscillatory integrals, in contrast to the ladder graphs where no oscillation is present. However, the number of nonladder graphs grows as k!, where k ∼ λ2 t ∼ λ−κ is the typical number of collisions. To beat this factorial growth, we need to give a very sharp bound on the individual graphs.
242
L. Erd˝ os et al.
We give a classiﬁcation of arbitrary large graphs, based on counting the number of vertices carrying oscillatory eﬀects. The number of these vertices is called the degree of the graph. For the ladder graphs, the degree is zero. For general graphs, the degree is roughly the number of vertices after removing all ladder and antiladder subgraphs. We thus obtain an extra λc power (for some c > 0) per non(anti) ladder vertex. This strong improvement is suﬃcient to beat the growth of the combinatorics in the time scale we consider. To our knowledge, nothing like this has been done in a graphical expansion before. For a comparison, the unperturbed Green functions in the perturbation expansion for the manyfermion systems for small temperature and for the random Schr¨ odinger equation for large time are given by 1 , ip0 + p2 − µ
1 . p2 − α + iη
In the manyfermion case, p0 ∈ MF = { πβ (2n + 1) : n ∈ Z} where β ∼ T −1 is the inverse temperature. In the random Schr¨ odinger case, η ∼ t−1 . Their L2 properties are diﬀerent: −2 −2 1 2 ∼  log β, dpp2 − α + iη ∼ η −1 dp ip0 + p − µ β p0 ∈MF
Notice the divergence is more severe for the random Schr¨odinger equation case. Finally we note that the threshold κ < 1/2000 in our theorem can be signiﬁcantly improved with more detailed arguments. However, one cannot go beyond κ = 2 with only improvements on estimates of the individual graphs. The Duhamel formula must be expanded at least up to k = λ2 t = λ−κ , which is the typical number of collisions up to time t. Even if one proves for most graphs the best possible estimate, λ2k , it cannot beat the k! combinatorics when k λ−2 , i.e., λ2k k! 1 for k λ−2 . A diﬀerent resummation procedure is needed beyond this threshold to exploit cancellations among these graphs.
3 Sketch of the Proof We present the main ideas of the proof for the lattice case and comment on the modiﬁcations for the continuous case. 3.1 Renormalization Before expanding the solution of the Schr¨ odinger equation (1) via the Duhamel formula, we perform a renormalization of the “oneparticle propagator” by splitting the Hamiltonian as H = H0 + V˜ , with H0 already containing
Towards the Quantum Brownian Motion
243
the part of the selfenergy produced by immediate recollisions with the same obstacle. This eﬀectively resums all such immediate recollisions. Let θ(p) := Θ(e(p)), where Θ(α) := limε→0+ Θε (α) and 1 Θε (α) := dq . (24) α − e(q) + iε We have Im Θ(α) = −πΦ(α)
(25)
with Φ deﬁned in (9). We rewrite the Hamiltonian as H = H0 + V with V := λV − λ2 θ(p) .
H0 := ω(p) := e(p) + λ2 θ(p),
(26)
Our renormalization includes only the lowest order selfenergy. This suﬃces on the time scales we consider. 3.2 The Expansion and the Stopping Rules Iterating the Duhamel formula −itH
e
−itH0
=e
t
−i
ds e−i(t−s)H V e−isH0
(27)
gives for any ﬁxed integer N ≥ 1 ψt := e−itH ψ0 =
N −1
ψn (t) + ΨN (t) ,
(28)
n=0
with ψn (t) := (−i)n
t
[dsj ]n+1 e−isn+1 H0 V e−isn H0 V . . . V e−is1 H0 ψ0 1
(29)
being the fully expanded terms and t ΨN (t) := (−i) ds e−i(t−s)H V ψN −1 (s)
(30)
is the nonfully expanded or error term. We used the shorthand notation t t t n n n [dsj ]1 := ... dsj δ t − sj . 0
j=1
j=1
2 Since " each potential V in (29), (30) is a summation itself, V = −λ θ(p) + λ γ Vγ , both of these terms in (29) and (30) are actually big summations
244
L. Erd˝ os et al.
over socalled elementary wavefunctions, which are characterized by their collision history, i.e. by a sequence of obstacles labelled by γ ∈ Zd and a label ϑ corresponding to an insertion of −λ2 θ(p). Because this expansion is generated by iteration of (27), the sequences deﬁning collision histories can be obtained recursively. This allows us to reﬁne the Duhamel expansion by using stopping rules that depend on the type of collision history. We call a sequence nonrepetitive if the only repetitions in potential labels γ occur in gates (immediate recollisions). The iteration of (27) is stopped when adding a new entry to the sequence makes it violate this condition. This can happen because of a recollision, a nested recollision, or a triple collision. The precise deﬁnition of these recollision types is given in [11]. The iteration is also stopped when the last entry in the sequence causes the total number of gates and ϑ’s to reach 2. If the sequence stays nonrepetitive and the total number of gates and ϑ’s stays below 2, the iteration is stopped when the number of nongate potential labels reaches K = λ−δ (λ2 t) .
(31)
Note that K is much bigger than the expected typical number of collisions, λ2 t. We denote the sum of the truncated elementary nonrepetitive wave functions with at most one λ2 power from the nonskeleton indices or ϑ’s and with (≤1),nr K skeleton indices by ψ∗s,K . The superscript (≤ 1) refers to the number of gates and ϑ’s, each of which gives a factor λ2 . By this splitting, we arrive at the following modiﬁed Duhamel formula, in which all nonerror terms are nonrepetitive. Proposition 1. [Duhamel formula] For any K ≥ 1 we have ψt = e−itH ψ0 =
K−1
(≤1),nr
ψt,k
k=0
+ 0
t
' ds e−i(t−s)H ψ∗s,K
(≤1),nr
+
K
(2),last
ψ∗s,k
(≤1),rec
+ψ∗s,k
(≤1),nest
+ψ∗s,k
(32),
(≤1),tri
+ψ∗s,k
k=0
The terms under the integral correspond to the various stopping criteria indicated above. For the precise deﬁnition of the corresponding wave functions, see [11]. The main contribution comes from the nonrepetitive sequences with k < K, i.e. from the ﬁrst term in (32). The estimate of the terms in the second line (32) ﬁrst uses the unitarity of the full evolution 8 8 t 8 8 ds e−i(t−s)H ψs# 8 ≤ t · sup ψs# . 8 0
s≤t
(33)
Towards the Quantum Brownian Motion
245
For # = rec, nest, tri we will use the fact the Feynman graphs arising in the expectation Eψs# 2 contain an additional oscillatory factor, which renders them smaller than the corresponding nonrepetitive term. It turns out that the oscillation eﬀect from one single recollision, nest or triple collision is already suﬃcient to overcome the additional factor t arising from the crude bound (33). This fact relies on estimates on singular integrals concentrating on the energy level sets Σe . It is a wellknown fact from harmonic analysis, that such singular integrals can more eﬀectively be estimated for convex level sets. This is why the nonconvexity of the energy shells is a major technical complication for the discrete model in comparison with the continuous case, where the level sets are spheres. Nonskeleton labels also give rise to a smallness eﬀect due to a cancellation between gates and ϑ’s, however, one such cancellation would not be suﬃcient to beat the t–factor. This is why at least two such cancellations are necessary (2),last (≤1),nr in ψ∗s,k . Finally, the term ψ∗s,K is small because it has unusually many collisions, thanks to the additional factor λ−δ in the deﬁnition of K. (≤1),nr In this exposition we focus only on the nonrepetitive terms, ψt,k , because estimating them involves the main new ideas. The error terms are estimated by laborious technical modiﬁcations of these ideas. 3.3 The L2 Norm of the NonRepetitive Wavefunction We ﬁrst estimate the L2 norm of the fully expanded wave function with no (0),nr gates or ϑ, ψt,k . This is the core of our analysis. Feynman Graphs The wavefunction t (0),nr = [dsj ]k+1 e−isk+1 H0 Vγk e−isk H0 Vγk−1 . . . e−is2 H0 Vγ1 e−is1 H0 ψ0 ψt,k 1 γ
where the summation is over all sequences for which the potential labels γi are all diﬀerent. Therefore every term in (0),nr Eψt,k 2 = E ψt,γ ψt,γ γ,γ
has 2k potential terms, and their expectation, E Vγ1 Vγ2 . . . Vγk Vγ1 Vγ2 . . . Vγk , is zero, using EVγ = 0, unless the potentials are paired. Since there is no repetition within γ and γ , all these pairings occur between γ and γ , therefore every pairing corresponds to a permutation on {1, 2, . . . , k}. The set of such
246
L. Erd˝ os et al.
permutations is denoted by Pk and they can be considered as a map between the indices of the γ and γ labels. We recall the following identity from Lemma 3.1 of [10]
t
[dsj ]k+1 1 0
k+1
e−isj ω(pj ) =
j=1
ieηt 2π
dα e−iαt
R
k+1 j=1
1 α − ω(pj ) + iη
for any η > 0. We will choose η := t−1 . Therefore, we have λ2k e2tη (0),nr 2 Eψt,k = dpd˜ p δ(pk+1 − p˜k+1 ) (2π)2 γ1 ,...,γk σ∈Pk
γi =γj
(34) ×E
k
˜ , η)ψH0 (p1 )ψH0 (˜ VHγj (pj+1 − pj )VHγj (˜ pσ(j)+1 − p˜σ(j) )M (k, p, p p1 )
j=1
with p = (p1 , p2 , . . . , pk+1 ), and d˜ p, and
dp :=
i(α−β)t
˜ ) := Mη (k, p, p
dαdβ e R
(Td )k+1
k+1 j=1
˜ dp1 dp2 . . . dpk+1 , similarly for p
1 1 α − ω(pj ) − iη β − ω(˜ pj ) + iη
. (35)
We compute the expectation: E
k
VHγj (pj+1 − pj )VHγj (˜ pσ(j)+1 − p˜σ(j) )
j=1
=
k γ1 ,...,γk γi =γj
eiγj (pj+1 −pj −(p˜σ(j)+1 −p˜σ(j) )) .
(36)
j=1
H In the continuous model this formula also contains a product of Bterms, where B was the single site potential function in (5). These factors are included into the deﬁnition of Mη . Most importantly, they provide the necessary decay in the momentum variables in the case of noncompact momentum space. A similar idea was used in [10]. Due to the restriction γi = γj , (36) is not a simple product of delta functions in the momenta. We have to use a connected graph expansion that is well known in the polymer expansions of ﬁeld theory (see, e.g. [23]). We do not give the details here, we only note that the result is a weighted sum over partitions of the index set {1, . . . , k}. Each term in the sum is a product of delta functions labelled by the lumps of the partition and each delta funtion imposes the Kirchoﬀ Law for the incoming and outgoing momenta of the lump and its σimage. The trivial partition, where each lump has a single
Towards the Quantum Brownian Motion
247
element, carries the main contribution. Estimating the terms with nontrivial partitions can be reduced to estimates for the trivial partition [11]. We there(0),nr 2 fore discuss " only the contribution from the trivial partition to Eψt,k , given by σ∈Pk Vη (k, σ), where Vη (k, σ) :=
λ2k e2tη (2π)2 ×
˜ )δ(˜ pk+1 − pk+1 )ψH0 (p1 )ψH0 (˜ p1 ) dpd˜ pMη (k, p, p
k δ pi+1 − pi − (˜ pσ(i)+1 − p˜σ(i) )
(37)
i=1
This complicated formula can be encoded by a Feynman graph and Vη (k, σ) is called the value or amplitude of the graph. The Feynman graph for the trivial partition corresponds to the usual Feynman graphs for the Gaussian case discussed in [10] and we brieﬂy describe their construction. A Feynman graph consists of two directed horizontal lines (upper and lower) with k collision vertices on each that represent the collision histories of ψ¯ and ψ, respectively. These two lines are joined at the two ends. This corresponds to evaluating the L2 norm on one end and inserting the initial wavefunction ψ0 on the other end. Each horizontal segment carries a momentum, p1 , p2 , . . . pk+1 and p˜1 , p˜2 , . . . p˜k+1 and a corresponding (renormalized) propagator, (α − ω(pj ) − pj ) + iη)−1 . Here α and β are the dual variables to the time iη)−1 and (β − ω(˜ on each line and they will be integrated out, see (35). Finally, the collision vertices are paired. Each pairing line joins an upper and a lower vertex and thus can be encoded with a permutation σ ∈ Pk . It is useful to think of the momenta as ﬂowing through the lines of the graph. The delta function associated to each pairing line in the value of the graph (37) then expresses the Kirchhoﬀ Law for the ﬂow of momenta adjacent to the two vertices. A typical graph with trivial partition is shown on Fig. 1. For the special case of the identity permutation σ = id we obtain the socalled ladder graph (Fig. 2). The following proposition shows that the ladder gives the main contribution.
p1
p1
p2
ψ
p2
ψ
Fig. 1. Typical Feynman graph with no lumps
p k+1
p k+1
248
L. Erd˝ os et al.
Fig. 2. Ladder graph
The Main Contribution is the Ladder Proposition 2 (L2 estimate). Let η −1 := t, t = O(λ2+κ ) and k ≤ K := λ−δ (λ2 t). For suﬃciently small λ, κ and δ there exists a positive number c1 (κ, δ) such that (0),nr (38) Eψt,k 2 = Vη (k, id) + Oδ λc1 (κ,δ) . The threshold values for κ, δ and the explicit form of c1 (κ, δ) are found in [11]. Sketch of the proof. As mentioned above, we discuss only how to estimate the contributions from the trivial partition, but for an arbitrary permutation σ. Given a permutation σ ∈ Pk , we deﬁne a (k+1)×(k+1) matrix M = M (σ) as follows if σ ˜ (j − 1) < i ≤ σ ˜ (j) 1 if σ ˜ (j) < i ≤ σ ˜ (j − 1) (39) Mij (σ) := −1 0 otherwise where, by deﬁnition, σ ˜ is the extension of σ to a permutation of {0, 1, . . . , k + 1} by σ ˜ (0) := 0 and σ ˜ (k + 1) := k + 1. It is easy to check that k+1 k+1 λ2k e2tη ˜) Vη (k, σ) := δ p˜i − Mij pj , (40) dpd˜ p Mη (k, p, p (2π)2 i=1 j=1 in other words, the matrix M encodes the dependence of the p˜momenta on the pmomenta. This rule is transparent in the graphical representation of the Feynman graph: the momentum pj appears in those p˜i ’s which fall into its “domain of dependence”, i.e. the section between the image of the two endpoints of pj , and the sign depends on the ordering of these images (see Fig. 3). The matrix M (σ) has several properties that follow easily from this structure: Lemma 1. For any permutation σ ∈ Pk the matrix M (σ) is (i) invertible; (ii) totally unimodular, i.e. any subdeterminant is 0 or ±1.
Towards the Quantum Brownian Motion p
p
j
j
These momenta equal + p j+ ...
249
These momenta equal p j+ ...
Fig. 3. Domain of momenta dependencies
The following deﬁnition is crucial. It establishes the necessary concepts to measure the complexity of a permutation. Deﬁnition 1 (Peak, valley and slope). Given a permutation σ ∈ Pk let σ ˜ be its extension. A point (j, σ(j)), j ∈ Ik := {1, 2, . . . , k}, on the graph of σ is called peak if σ ˜ (j − 1) > σ(j) < σ ˜ (j + 1), it is called valley if σ ˜ (j − 1) < σ(j) > σ ˜ (j + 1), otherwise it is called slope. Additionally, the point (k + 1, k + 1) is also called valley. The set I = {1, 2, . . . , k + 1} is partitioned into three disjoint subsets, I = Ip ∪ Iv ∪ Is , such that i ∈ Ip , Iv or Is depending on whether (˜ σ −1 (i), i) is a peak, valley or slope, respectively. σ −1 (i)−˜ σ −1 (i−1) = 1. Finally, an index i ∈ Iv ∪Is is called ladder index if ˜ The set of ladder indices is denoted by I ⊂ I and their cardinality is denoted by = (σ) := I . The number of nonladder indices, d(σ) := k + 1 − (σ) is called the degree of the permutation σ. Remarks: (i) The terminology of peak, valley, slope, ladder comes from the graph of the permutation σ ˜ viewed as a function on {0, 1, . . . , k + 1} in a coordinate system where the vertical axis is oriented downward. (ii) For σ = id we have Ip = ∅, Is = {1, 2, . . . , k}, Iv = {k + 1} and I = {1, 2, . . . , k + 1}. In particular, d(id) = 0 and d(σ) > 0 for any other permutation σ = id. The following theorem shows that the degree of the permutation d(σ) measures the size of Vη (k, σ). This is the key theorem in our method and we will sketch its proof separately in Sect. 3.4. Theorem 3. Let η −1 := t, t = O(λ2+κ ) with a suﬃciently small κ. Let σ ∈ Pk and assume that k ≤ K = λ−δ (λ2 t). For suﬃciently small κ and δ there exists c2 (κ, δ) > 0 such that d(σ) , Vη (k, σ) ≤ Cλc2 (κ,δ)
λ1.
This theorem is complemented by the following lemma:
(41)
250
L. Erd˝ os et al.
Lemma 2. Let k = O(λ−κ−δ ), d > 0 integer and let γ > κ + δ. Then λγd(σ) ≤ O λd(γ−κ−δ)
(42)
σ∈Pk d(σ)≥d
for all suﬃciently small λ. The proof follows from the combinatorial estimate on the number of permutations with a given degree: #{σ ∈ Pk : d(σ) = d} ≤ (Ck)d . From Theorem 3 and Lemma 2 we immediately obtain an estimate on the (0),nr contribution of the trivial lumps to Eψt,k 2 if κ and δ are suﬃciently small: (43) Vη (k, σ) ≤ Oδ λc3 (κ,δ) σ∈Pk σ=id
with some appropriate c3 (κ, δ) > 0. 3.4 Sketch of the Proof of the Main Technical Theorem In this section we explain the proof of Theorem 3. We set Eη (M ) : = λ
2k
4d
dαdβ −4d
×
k+1 j=1
dp
k+1 i=1
1 α − ω(pi ) − iη
1 . "k+1 β − ω( =1 Mj p ) + iη
(44)
H factors to ensure the For the continuous model, the deﬁnition includes the B integrability for the large momentum regime. It is easy to check that Vη (k, σ) is estimated by Eη (M (σ)) modulo constant factors and negligible additive terms coming from the regime where α or β is big. The denominators in this multiple integral are almost singular in certain regimes of the high dimensional space of all momenta. The main contribution comes from the overlap of these singularities. The overlap structure is encoded in the matrix M , hence in the permutation σ, in a very complicated entangled way. Each variable pj may appear in many denominators in (44), so successive integration seems very diﬃcult. We could not ﬁnd the exact order (as a power of λ) of this multiple integral but we conjecture that true order is essentially λ2d(σ) . Our goal in Theorem 3 is to give a weaker bound of order λcd(σ) , i.e. that is still a λpower linear in the degree, but the coeﬃcient considerably smaller than 2. Notice that the αdenominators in (44) correspond to the columns of M and the βdenominators corresponds to the rows. For this presentation we
Towards the Quantum Brownian Motion
251
will use j to label row indices and i to label column indices. We recall the sets Iv , Ip , I from Deﬁnition 1 and we will view these sets as subsets of the row indices of M . First we notice that if j ∈ (I \ Iv ), i.e. j is a nonvalley ladder row, then there exists a column index i = c(j) such that the momentum pi appears only in the jth βdenominator. In other words, the ith column of M has a single nonzero element (that is actually ±1) and it is in the jth row. Therefore the dpi integral can be performed independently of the rest of the integrand by using the following elementary but quite involved bound for small κ: λ2 ≤ 1 + O(λ1/4 ) . (45) dpi sup α − ω(pi ) − iη β − ω(±pi + w) + iη w,α,β Td Note that the constant of the main term is exactly 1. This fact is important, since in graphs with low degree this estimate has to be raised to a power I \ Iv  that may be comparable with k. Clearly for k ≤ K ∼ λ−κ−δ and κ + δ < 1/4 we have k (46) 1 + O(λ1/4 ) ≤ const , but had 1 been replaced with a bigger constant in (45), we would obtain an exponentially big factor (const)k that would not be aﬀordable. The precise constant 1 in the estimate (45) is related to the appropriate choice of the renormalization θ(p) in ω(p). After the nonvalley ladder rows are integrated out, and the corresponding rows and columns are removed from the matrix M , we obtain a smaller matrix M (1) describing the remaining denominators. In M (1) we keep the original labelling of the rows from M . Now we estimate some of the βdenominators in (44) by L∞ norm, i.e. by −1 η . This is a major overestimate, but these denominators are chosen in such a way that the entangled structure imposed by M becomes much simpler and many other denominators can be integrated out by L1 bounds that are only logarithmic in λ. We start with estimating all βdenominators in rows j ∈ Ip by the trivial L∞ norm. The corresponding rows are removed from M (1) , in this way we obtain a matrix M (2) . Let I ∗ := I \ Ip ∪ (I \ Iv ) be the remaining row indices after removing the peaks and the nonvalley ladders. Then we inspect the remaining rows j ∈ I ∗ of M (2) in increasing order. The key observation is that for each j ∈ I ∗ there exists a column index, i = c(j), such that the variable pi appears only in the jth βdenominator, provided that all βdenominators with j < j have already been integrated out. In view of the structure of M (1) , it means that for any j ∈ I ∗ there exists
252
L. Erd˝ os et al.
a column i = c(j) such that the only nonzero element among {Mij : j ≥ j} (2)
(2)
is Mij . This fact follows from the structure of M (σ) and from the fact that all rows with j ∈ Ip have been removed. This property allows us to remove each remaining βdenominator, one by one, by estimating integrals of the type 1 Cη −τ 1 ≤ , (47) dpi α − ω(pi ) − iη β − ω(±pi + w) + iη w Td where w is a linear combination of momenta other than pi . The absolute value w is interpreted as the distance of w from the nearest critical point of the dispersion relation e(p). The variable pi at this stage of the procedure appears only in these two denominators. The exponent τ can be chosen zero (with logarithmic corrections) for the continuous model and this fact has already been used in [10]. For the discrete model we can prove (47) with τ = 3/4 + 2κ and we know that the exponent cannot be better than 1/2. The reason for the weaker estimate is the lack of convexity of the level set Σe . Replacing ω(p) with e(p) for a moment, the inequality (47) with τ = 0 essentially states that the level set {α = e(p)} and its shifted version {β = e(p + w)} intersect each other transversally, unless w is close to zero. Indeed, the transversal intersection guarantees that the volume of the p values, where both denominators are of order η, is of order η 2 . Then a standard argument with dyadic decomposition gives the result with a logarithmic factor. For translates of spheres the transversal intersection property holds, unless w ∼ 0. However, in certain points of the level sets Σe of the discrete dispersion relation the curvature vanishes, in fact Σe even contains straight lines for 2 ≤ e ≤ 4. The transversal intersection fails in certain regions and results in a weaker bound. Neglecting the point singularity w−1 in (47) for a moment (see Sect. 3.5 later), we easily see that with this algorithm one can bound Eη (M (σ)) by λ2(k−q) η −p η −τ (k−p−q) , modulo logarithmic factors, where p = Ip  is the number of peak indices and q := I \Iv  is the number of nonvalley ladder indices. From the deﬁnitions it follows that the sets Iv , Ip and I \ Iv are disjoint and Iv  = p + 1. Thus we have 2p + 1 + q ≤ k + 1. Therefore λ2(k−q) η −p(1−τ )−τ (k−q) ≤ (λ4 tτ +1 )(k−q)/2 ≤ (λ4 tτ +1 ) 2 [d(σ)−1] 1
(48)
since q ≤ . If τ < 1, then with a suﬃciently small κ we see that λ4 tτ +1 is a positive power of λ. Thus we obtain a bound where the exponent of λ is linear in d(σ). With a more careful estimate one can remove the additional −1 in the exponent. In particular, for the continuous case with τ = 0 this argument works up to κ < 2. We end this section with a remark. Apparently the bound κ < 2 (or, equivalently, t λ−4 ) shows up in two diﬀerent contexts in this argument. To avoid misunderstandings, we explain brieﬂy that neither of these two
Towards the Quantum Brownian Motion
253
appearences is the genuine signature of the expected threshold κ = 2 for our expansion method to work. The true reason is the one mentioned in the introduction: even the best possible bound, λ2d(σ) , on the graph with permutation σ, cannot beat the k! combinatorics of the graphs beyond κ = 2. In the argument above, on one hand, κ < 2 is related to the error term in the ladder calculation (45). This error term can be improved to λ2  log λ and it is apparently due to the fact that the renomalization term θ(p) was solved only up to lowest order. An improvement may be possible by including more than the lowest order of the selfenergy. The second apperance of κ < 2, or t λ−4 , at least for the continuous model, is in (48) and it is due to the fact that certain βdenominators are overestimated by L∞ . This is again a weakness of our method; we did overestimates in order to simplify the integrand. 3.5 Point Singularities The argument in the previous section has neglected the point singularity arising from (47). While a point singularity is integrable in d ≥ 3 dimensions, it may happen that exactly the same linear combinations of the independent variables keep on accumulating by the repeated use of the bound (47). In that case at some point a high negative power of w needs to be integrated. While it is possible to improve the estimate (47) by changing the denominator on the right hand side to w + η, this would still yield further negative ηpowers. It is easy to see that this phenomenon does occur. Primarily this would have occurred if we had not treated the ladders separately: if pi ’s are ladder variables, then the corresponding w momenta in (47) are indeed the same. Although we have removed the ladders beforehand, the same phenomenon occurs in case of a graph which contains ladder only as a minor but not as a subgraph. Our separate ladder integration procedure (45) can be viewed as a very simple renormalization of the ladder subgraphs. The correct procedure should renormalize all ladder minors as well. To cope with this diﬃculty, we have to follow more precisely the point singularities. To this end, we deﬁne the following generalization of Eη (M ). For any index set I ⊂ I = {1, 2, . . . , k + 1}, any I  × (k + 1) matrix M , any ν integer and any ν × (k + 1) matrix E we deﬁne Eη (I , M, E) := λ2k e2tη ×
4d
dαdβ
dp
−4d
1 1 "k+1 α − ω(pi ) − iη β − ω j=1 Mij pj + iη
ν
Eµj pj  (49) We follow the same procedure as described in Sect. 3.4, but we also keep track of the evolution of the point singularity matrix E. At the beginning I = I, ν = 0 and E is not present. After the ﬁrst nonladder type integration, a i∈I
µ=1

1 "k+1 j=1
254
L. Erd˝ os et al.
point singularity will appear from (47). Some of the point singularities may get integrated out later as one of their variables become integration variable. Therefore we will need the following generalization of (47): Lemma 3. There exists a constant C such that for any index set A 1 1 dp sup α − ω(p) − iη β − ω(r + p) + iη r a + p α,β≤4d
≤ Cη −τ  log η3
a∈A
a ∈A a =a
a∈A
1 1 . ra − ra  r
+ *
For the continuous model τ = 0 while for the discrete model τ =
(50) 7 8
+ 2κ.
Using this lemma, we can keep record of the evolution of the point singularity matrix E at an intermediate step of our integration algorithm. These matrices change by simple operations reminiscent to the Gaussian elimination. Three complications occur along this procedure, we brieﬂy describe how we resolve them: (1) The inequality (50) does not allow higher order point singularities. Although it is possible to generalize it to include such singularities as well, we followed a technically simpler path. In addition to the indices j ∈ Ip , we select further βdenominators that we estimate by the trivial L∞ bound. These additional indices are chosen in such a way, that (i) the number of remaining rows be at least 13 d(σ); (ii) the point singularity matrix be of full rank at every step of the algorithm. This second criterion guarantees that no higher order point singularities occur. Since every E can be derived from M by a procedure that is close to Gaussian elimination and M is invertible (Lemma 1), the fullrank property is relatively easy to guarantee. (2) The fullrank property actually needs to be guaranteed in a quantitative way, at least the entries of E needs to be controlled. These entries appear in the point singularity denominators of (50) and their inverses would appear in the estimate. The key observation is that each entry of every matrix E along the procedure is always 0, 1 or −1. It is actually easier to prove a stronger statement, namely that every E is a totally unimodular matrix. The proof follows from the fact that every E can be derived from M by elementary Gaussian elimination steps plus zeroing out certain rows and columns. Such steps preserve total unimodularity and M is totally unimodular by Lemma 1. (3) After all βdenominators are eliminated, we are left with an integral of the form 4d ν 1 1 " dα dpi Eη (J, ∅, E) := α − ω(pi ) − iη µ=1  i∈J Eµi pi  −4d i∈J
i∈J
(51)
Towards the Quantum Brownian Motion
255
for some index set J and some point singularity matrix obtained along the integration procedure. Without the point singularities, this integral could be estimated by the Jth power of  log η. Since E is totally unimodular, a similar estimate can be obtained for (51) as well.
4 Computation of the Main Term and Its Convergence to a Brownian Motion Our goal is to compute the Wigner distribution EWψεt (X, v) with t = λ−2−κ T and ε = λ2+κ/2 . From Proposition 2, and similar bounds on the repetitive terms in (32), we can restrict our attention to the ladder graph. The following lemma is a more precise version of the ladder integration Lemma (45) and it is crucial to this computation. We present it for the more complicated discrete case. The proof is a tedious calculation in [11]. Lemma 4. Suppose f (p) is a C 1 function on Td . Recall 0 < κ < 1/16 and deﬁne γ := (α + β)/2. Let η satisfy λ2+4κ ≤ η ≤ λ2+κ . Then for r ≤ λ2+κ/4 we have, λ2 f (p) dp (52) (α − ω(p − r) − iη)(β − ω(p + r) + iη) f (p) δ(e(p) − γ) dp+O(λ1/2−8κ  log λ). =−2πiλ2 (α − β) + 2(∇e)(p) · r − 2i[λ2 ImΘ(γ) + η] Since the Boltzmann collision kernel is uniform on the energy shell, the calculation of EWψεt (X, v) is more straightforward for the discrete case. We present the sketch of this calculation, the continuous model requires a little more eﬀort at this stage. Let ε = λ2+κ/2 be the space scale. After rescaling the Wigner function K (εξ, v) tested against a smooth, decaying function at time t, we compute W O(ξ, v). In particular ξ is of order 1. After the application of Lemma 4 (with v = vk+1 ) and change of variables a := (α + β)/2 and b := λ−2 (α − β), we obtain dαdβ K := dvdξ O(ξ, v)EW K (εξ, v) = dv eit(α−β)+2ηt
O, EW 2 2 R (2π) λ k≤K
K0 (εξ, v1 ) × O(ξ, vk+1 )W
k+1
k≤K
R
α − ω(vi +
K0 (εξ, v1 )O(ξ, vk+1 ) , ×W
εξ 2 )
− iη
λ2
β − ω(vi − dadb itλ2 b −2πi δ(e(vj ) − a)dvj e (2π)2 b + λ−2 ε∇e(vj ) · ξ − 2iI(a) j j=1
≈
εξ 2 )
+ iη
256
L. Erd˝ os et al.
where we deﬁned I(γ) := ImΘ(γ) for brevity. We used η = λ2+κ to estimate the error terms. The main term (left hand side above) however, is independent of η, so we can choose η = λ2+4κ for the rest of the calculation and we note that Lemma 4 holds for this smaller η as well. This is the reason why the e2ηt factor is negligible. We expand the fraction up to second order in ε, we get 1 −i −i λ−2 ε∇e(vj ) · ξ ≈ 1 − b + λ−2 ε∇e(vj ) · ξ − 2iI(a) b − 2iI(a) b − 2iI(a) 2 λ−4 ε2 [∇e(vj ) · ξ]2 + (b − 2iI(a))2 By symmetry of the measure 2πδ(e(v)−a)dv under the sign ﬂip, v → −v and using (∇e)(v) = −∇e(−v), we see that the ﬁrst order term vanishes after the integration. We also deﬁne the matrix 1 ∇e(v) ∇e(v) D(a) := ⊗ dµa (v) 2 I(a) 2π 2π After integrating out all momenta and changing the b variable we obtain 2I(a)da db 2iλ2 tbI(a) K K
O, EW ≈
O(ξ, ·)a W0 (εξ, ·)a e dξ 2π 2π R R k≤K
1 2k+1 −i k+1 1 (2π)2 ε2 λ−4 ξ, D(a)ξ · × × 1+ b−i 2I(a) (b − i)2 We sum up the geometric series and perform a residue calculation to evaluate the db integral. We obtain that the main contribution comes from k ∼ 2λ2 tI(e), so the truncation k ≤ K can be neglected and the result of the db integration is (2π)2 ε2 t ξ, D(e)ξ db · · · ≈ exp − λ2 R 2π To obtain a nontrivial limit, ε2 t/λ2 ∼ 1 is necessary. Noting that t = λ−2−κ T with T = O(1), we see that indeed the space must be rescaled by ε = λ2+κ/2 . Finally 2I(a)da K K0 (εξ, ·)a exp −(2π)2 T ξ, D(a)ξ .
O(ξ, ·)a W
O, EW ≈ dξ 2π R Since exp[−(2π)2 T ξ, D(a)ξ] is the fundamental solution to the heat equation (15), from the deﬁnition of ·, and after inverse Fourier transform we obtain (17). This completes the sketch of the calculation of the main term.
Towards the Quantum Brownian Motion
257
Acknowledgements L. Erd˝ os was partially supported by NSF grant DMS0200235 and EUIHP Network “Analysis and Quantum” HPRNCT20020027. H.T. Yau was partially supported by NSF grant DMS0307295 and MacArthur Fellowship.
References 1. M. Aizenman and S. Molchanov: Commun. Math. Phys. 157, 245 (1993). 2. P. Anderson: Phys. Rev. 109, 1492 (1958). 3. C. Boldrighrini, L. Bunimovich and Y. Sinai: J. Statis. Phys. 32 no. 3, 477 (1983). 4. J. Bourgain: Random lattice Schr¨ odinger operators with decaying potential: some higher dimensional phenomena (Lecture Notes in Mathematics Vol. 1807 (Springer, Berlin Heidelberg New York 2003) pp 70–99. 5. L. Bunimovich, Y. Sinai: Commun. Math. Phys. 78, 479 (1980/81). 6. T. Chen: Localization Lengths and Boltzmann Limit for the Anderson Model at Small Disorders in Dimension 3. Preprint (2003) http://xxx.lanl.gov/ abs/mathph/0305051. 7. D. D¨ urr, S. Goldstein, J. Lebowitz: Commun. Math. Phys. 78, 507 (1980/81). 8. D. D¨ urr, S. Goldstein, J. Lebowitz: Commun. Math. Phys. 113, 209 (1987). 9. L. Erd˝ os: J. Statis. Phys. 107, 1043 (2002). 10. L. Erd˝ os and H.T. Yau: Commun. Pure Appl. Math. LIII, 667 (2000). 11. L. Erd˝ os, M. Salmhofer and H.T. Yau: Quantum diﬀusion of the random Schr¨ odinger evolution in the scaling limit. Preprints (2005) http://xxx.lanl.gov/abs/mathph/0502025, 0512014, 0512015 12. J. Fr¨ ohlich and T. Spencer: Commun. Math. Phys. 88, 151 (1983). 13. T.G. Ho, L.J. Landau and A.J. Wilkins: Rev. Math. Phys. 5, 209 (1993). 14. G. Gallavotti: Rigorous theory of the Boltzmann equation in the Lorentz model. Nota interna n. 358, Physics Department, Universita di Roma, pp. 1–9 (1972) mp [emailprotected], # 93–304. 15. I.Ya. Goldsheid, S.A. Molchanov and L.A. Pastur: Funct. Anal. Appl. 11, 1 (1997). 16. H. Kesten, G. Papanicolaou: Comm. Math. Phys. 78, 19 (1980/81). 17. A. Klein: Math. Res. Lett. 1, 399 (1994). 18. T. Komorowski, L. Ryzhik: Diﬀusion in a weakly random Hamiltonian ﬂow. Preprint (2004). 19. O. Lanford: Ast´erisque 40, 117 (1976). 20. P.A. Lee and T.V. Ramakrishnan: Rev. Mod. Phys. 57, 287 (1985). 21. J. Lukkarinen and H. Spohn. Kinetic limit for wave propagation in a random medium. Preprint. http://xxx.lanl.gov/mathph/0505075 22. I. Rodnianski and W. Schlag: Int. Math. Res. Not. 5, 243 (2003). 23. E. Seiler, Gauge Theories as a Problem of Constructive Quantum Field Theory and Statistical Mechanics. Lecture Notes in Physics 159 (Springer, Berlin Heidelberg New York 1982). 24. W. Schlag, C. Shubin, T. Wolﬀ: J. Anal. Math. 88, 173 (2002). 25. H. Spohn: J. Statist. Phys. 17, 385 (1977). 26. H. Spohn: Commun. Math. Phys. 60, 277 (1978). 27. D. Vollhardt and P. W¨ olﬂe: Phys. Rev. B 22, 4666 (1980).
BoseEinstein Condensation and Superradiance∗ J.V. Pul´e1 , A.F. Verbeure2 and V.A. Zagrebnov3 1
2
3
Department of Mathematical Physics, University College Dublin, Belﬁeld, Dublin 4, Ireland [emailprotected] Instituut voor Theoretische Fysika, Katholieke Universiteit Leuven, Celestijnenlaan 200D, 3001 Leuven, Belgium [emailprotected] Universit´e de la M´editerran´ee and Centre de Physique Th´eorique, LuminyCase 907, 13288 Marseille, Cedex 09, France [emailprotected]
Abstract. We consider two models which exhibit equilibrium BEC superradiance. They are related to two diﬀerent types of superradiant scattering observed in recent experiments. The ﬁrst one corresponds to the ampliﬁcation of matterwaves due to Raman superradiant scattering from a cigarshaped BE condensate, when the recoiled and the condensed atoms are in diﬀerent internal states. The main mechanism is stimulated Raman scattering in twolevel atoms, which occurs in a superradiant way. Our second model is related to the superradiant Rayleigh scattering from a cigarshaped BE condensate. This again leads to a matterwaves ampliﬁcation but now with the recoiled atoms in the same state as the atoms in the condensate. Here the recoiling atoms are able to interfere with the condensate at rest to form a matterwave grating (interference fringes) which has been recently observed in experiments.
1 Introduction We report here our recent results on the equilibrium BoseEinstein Condensation (BEC) superradiance motivated by discovery of the Dicke superradiance and BEC matter waves ampliﬁcation [1]– [5]. In these experiments the condensate is illuminated with a laser beam, the so called dressing beam. The BEC atoms then scatter photons from this beam and receive the corresponding recoil momentum producing coherent fourwave mixing of light and atoms [5]. The aim of our project is the construction of soluble statistical mechanical models for these phenomena. In the ﬁrst paper [6], motivated by the principle of fourwave mixing of light and atoms [5], we considered two models with a linear interaction between Bose atoms and photons, one with a global gauge symmetry and another one in which this symmetry is broken. We proved that there is ∗
Talk given by Valentin Zagrebnov.
J.V. Pul´ e et al.: BoseEinstein Condensation and Superradiance, Lect. Notes Phys. 690, 259–278 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
260
J.V. Pul´e et al.
equilibrium superradiance and also that there is an enhancement of condensation compared with that occurring in the case of the free Bose gas. In the second paper [7] we formalized the ideas described in [4, 5] by constructing a thermodynamically stable model whose main ingredient is the twolevel internal states of the Bose condensate atoms. We showed that our model is equivalent to a bosonized Dicke maser model. Besides determining its equilibrium states, we computed and analyzed the thermodynamic functions, again ﬁnding the existence of a cooperative eﬀect between BEC and superradiance. In our last paper [8] we study the eﬀect of momentum recoil which was omitted in [6] and [7]. There we consider two models motivated by two diﬀerent types of superradiant scattering observed in recent experiments carried out by the MIT group, see e.g. [1]– [3]. Our ﬁrst model (Model 1 ) corresponds to the Raman superradiant scattering from a cigarshaped BE condensate considered in [1]. This leads to the ampliﬁcation of matter waves (recoiled atoms) in the situation when ampliﬁed and condensate atoms are in diﬀerent internal states. The main mechanism is stimulated Raman scattering in twolevel atoms, which occurs in a way similar to Dicke superradiance [7]. Our second model (Model 2 ) is related to the superradiant Rayleigh scattering from a cigarshaped BE condensate [2,3]. This again leads to a matterwave ampliﬁcation but now with recoiled atoms in the same state as the condensate at rest. This is because the condensate is now illuminated by an oﬀresonant pump laser beam, so that for a longpulse the atoms remain in their lower level states. In this case the (nonDicke) superradiance is due to selfstimulated Bragg scattering [3]. From a theoretical point of view both models are interesting as they describe hom*ogeneous systems in which there is spontaneous breaking of translation invariance. In the case of the Rayleigh superradiance this means that the phase transition corresponding to BEC is at the same time also a transition into a matterwave grating i.e. a “frozen” spatial density wave structure, see Sect. 3. The fact that recoiling atoms are able to interfere with the condensate at rest to form a matterwave grating (interference fringes) has been recently observed experimentally, see [3]– [5] and discussion in [9] and [10]. In the case of the Raman superradiance there is an important diﬀerence: the internal atomic states for condensed and recoiled bosons are orthogonal. Therefore these bosons are diﬀerent and consequently cannot interfere to produce a matterwave grating as in the ﬁrst case. Thus the observed spatial modulation is not in the atomic density of interfering recoiled and condensed bosons, but in the oﬀdiagonal coherence and photon condensate producing a onedimensional (corrugated ) optical lattice, see Sect. 3 and discussion in Sect. 4. To make the deﬁnition of our models more exact consider a system of identical bosons of mass m enclosed in a cube Λ ⊂ Rν of volume V = Λ
BoseEinstein Condensation and Superradiance
261
centered at the origin. We impose periodic boundary conditions so that the momentum dual set is Λ∗ = {2πp/V 1/ν p ∈ Zν }. In Model 1 the bosons have an internal structure which we describe by considering them as twolevel atoms, the two levels being denoted by σ = ±. For momentum k and level σ, a∗k,σ and ak,σ are the usual boson creation and annihilation operators with [ak,σ , a∗k ,σ ] = δk,k δσ,σ . Let (k) = k2 /2m be the single particle kinetic energy and Nk,σ = a∗k,σ ak,σ the operator for the number of particles with momentum k and level σ. Then the total kinetic energy is T1,Λ =
(k)(Nk,+ + Nk,− ) (1) k∈Λ∗
and the total number operator is N1,Λ = the Hamiltonian H1,Λ for Model 1 by
"
k∈Λ∗ (Nk,+
+ Nk,− ). We deﬁne
H1,Λ = T1,Λ + U1,Λ
(2)
g λ 2 U1,Λ = Ω b∗q bq + √ (a∗q+ a0− bq + aq+ a∗0− b∗q ) + N , 2V 1,Λ 2 V
(3)
where
g > 0 and λ > 0. Here bq , b∗q are the creation and annihilation operators of the photons, which we take as a onemode boson ﬁeld with [bq , b∗q ] = 1 and a frequency Ω. g is the coupling constant of the interaction of the bosons with the photon external ﬁeld which, without loss of generality, we can take to be positive as we can always incorporate the sign of g into b. Finally the λterm is added in (2) to obtain a thermodynamical stable system and to ensure the right thermodynamic behaviour. In Model 2 we consider the situation when the excited atoms have already irradiated photons, i.e. we deal only with deexcited atoms σ = −. In other words, we neglect the atom excitation and consider only elastic atomphoton scattering. This is close to the experimental situation [3]– [5], in which the atoms in the BE condensate are irradiated by oﬀresonance laser beam. Assuming that detuning between the optical ﬁelds and the atomic twolevel resonance is much larger that the natural line of the atomic transition (superradiant Rayleigh regime [2, 3]) we get that the atoms always remain in their lower internal energy state. We can then ignore the internal structure of the atoms and let a∗k and ak be the usual boson creation and annihilation operators for momentum k with [ak , a∗k ] = δk,k , Nk = a∗k ak the operator for the number of particles with momentum k,
(k)Nk (4) T2,Λ = k∈Λ∗
" the total kinetic energy, and N2,Λ = k∈Λ∗ Nk the total number operator. (2) We then deﬁne the Hamiltonian HΛ for Model 2 by
262
where
J.V. Pul´e et al.
H2,Λ = T2,Λ + U2,Λ
(5)
g λ 2 U2,Λ = Ω b∗q bq + √ (a∗q a0 bq + aq a∗0 b∗q ) + N . 2V 2,Λ 2 V
(6)
Our main results concerning the Models 1 and 2 can be summarized as follows: – We give a complete rigorous solution of the variational principle for the equilibrium state of Model 1, and we compute the pressure as a function of the temperature and the chemical potential. We show (see Sect. 2) that Model 1 manifests the Raman superradiance and we prove for this model the occurrence of spontaneous breaking of translation invariance of the equilibrium state. – The analysis of the Model 2 for the Rayleigh superradiance is very similar and therefore we do not repeat it but simply state the results in Sect. 3. – In Sects. 3 and 4 we relate the spontaneous breaking of translation invariance of the equilibrium state with spatial modulation of matterwaves (matterwave grating) in both models. We ﬁnish this introduction with the following comments concerning interpretation of the Models 1 and 2 : – In our models we do not use for eﬀective photonboson interaction the fourwave mixing principle, see [5, 6, 11]. The latter seems to be important for the geometry, when a linearly polarized pump laser beam is incident in a direction perpendicular to the long axis of a cigarshaped BE condensate, inducing the “45◦ recoil pattern” picture [1]– [3]. Instead as in [7], we consider a minimal photonatom interaction only with superradiated photons, cf [12]. This corresponds to superradiance in a “onedimensional” geometry, when a pump laser beam is collimated and aligned along the long axis of a cigarshaped BE condensate, see [10, 13]. – In this geometry the superradiant photons and recoiled matterwaves propagate in the same direction as the incident pump laser beam. If one considers it as a classical “source” (see [5]), then we get a minimal photonatom interaction [7] generalized to take into account the eﬀects of recoil. Notice that the further approximation of the BEC operators by cnumbers leads to a bilinear photonatom interaction studied in [5, 6]. – Here we study equilibrium BEC superradiance while the experimental situation (as is the case with Dicke superradiance [14]) is more accurately described by nonequilibrium statistical mechanics. However we believe that for the purpose of understanding the quantum coherence interaction between light and the BE condensate our analysis is as instructive and is in the same spirit as the rigorous study of the Dicke model in thermodynamic equilibrium, see e.g. [15]– [17].
BoseEinstein Condensation and Superradiance
263
– In spite of the simplicity of our exactly soluble Models 1 and 2 they are able to demonstrate the main features of the BEC superradiance with recoil: the photonboson condensate enhancement with formation of the light corrugated optical lattice and the matterwave grating. The corresponding phase diagrams are very similar to those in [7]. However, though the type of behaviour is similar, this is now partially due to the momentum recoil and not entirely to the internal atomic level structure.
2 Solution of the Model 1 2.1 The Eﬀective Hamiltonian We start with the stability of Hamiltonian (2). Consider the term U1,Λ in (3). This gives g g g2 √ aq+ a∗0− √ a∗q+ a0− − N0 U1,Λ = Ω b∗q + bq + 4Ω 2 V 2Ω V 2Ω V λ 2 λ 2 g2 N1,Λ ≥ N1,Λ − N0− (Nq+ + 1) . (7) −(Nq+ + 1) + 2V 2V 4Ω 2 V On the basis of the trivial inequality 4ab ≤ (a + b)2 , the last term in the lower bound in (7) is dominated by the ﬁrst term if λ > g 2 /8Ω, that is if the stabilizing coupling λ is large with respect to the coupling constant g or if the external frequency is large enough. We therefore assume the stability condition: λ > g 2 /8Ω. Since we want to study the equilibrium properties of the model (2) in the grandcanonical ensemble, we shall work with the Hamiltonian H1,Λ (µ) = H1,Λ − µN1,Λ
(8)
where µ is the chemical potential. Since T1,Λ and the interaction U1,Λ conserve the quasimomentum, Hamiltonian (2) describes a hom*ogeneous (translation invariant) system. To see this explicitly, notice that the external laser ﬁeld possesses a natural quasilocal structure as the Fourier transform of the ﬁeld operator b(x): 1 dx eiq·x b(x) . (9) bq = √ V Λ If for z ∈ Rν , we let τx be the translation automorphism (τz b)(x) = b(x + z), then τz bq = eiq·z bq and similarly τz ak,σ = eik·z ak,σ . Therefore, the Hamiltonian (2) is translation invariant. Because the interaction (3) is not bilinear or quadratic in the creation and annihilation operators the system cannot be diagonalized by a standard symplectic or Bogoliubov transformation. One way to solve the Model 1 is by applying the socalled eﬀective Hamiltonian method, which is based on
264
J.V. Pul´e et al.
the fact that an equilibrium state is not determined by the Hamiltonian but by its Liouvillian [18]. The best route to prove the exactness of the eﬀective Hamiltonian method in our case (see also [19]) is to use the characterization of the equilibrium state by means of the correlation inequalities [18, 20]: A state ω is an equilibrium state for H1,Λ (µ) at inverse temperature β, if and only if for all local observables A, it satisﬁes lim βω ([A∗ , [H1,Λ (µ), A]]) ≥ ω(A∗ A)ln
V →∞
ω(A∗ A) . ω(AA∗ )
(10)
Clearly only the Liouvillian [H1,Λ (µ), ·] of the Hamiltonian enters into these inequalities and therefore we can replace H1,Λ (µ) by a simpler Hamiltonian, the eﬀective Hamiltonian, which gives in the limiting state ω the same Liouvillian as H1,Λ (µ) and then look for the equilibrium states corresponding to it. Now in our case for an extremal or mixing state ω we deﬁne the eﬀective eﬀ (µ, η, ρ) such that for all local observables A and B Hamiltonian H1,Λ eﬀ (µ, η, ρ), B] . lim ω (A, [H1,Λ (µ), B]) = lim ω A, [H1,Λ
V →∞
V →∞
(11)
The signiﬁcance of the parameters η and ρ will become clear below. One can then replace (10) by ω(A∗ A) eﬀ . (µ, η, ρ), A]] ≥ ω(A∗ A)ln lim βω [A∗ , [H1,Λ V →∞ ω(AA∗ )
(12)
eﬀ We can choose H1,Λ (µ, η, ρ) so that it can be diagonalized and thus (12) can be solved explicitly. For a given chemical potential µ, the inequalities (12) can have more than one solution. We determine the physical solution by minimizing the free energy density with respect to the set of states or equivalently by maximizing the grand canonical pressure on this set. Let
g eﬀ H1,Λ (µ, η, ρ) = (λρ − µ + (q))a∗q+ aq+ + (λρ − µ)a∗0− a0− + (ηa∗q+ bq 2 √ V g ¯ ∗ + T + (λρ − µ)N (,13) ζa0− + ζa + η¯aq+ b∗q ) + Ω b∗q bq + 0− 1,Λ 1,Λ 2 where
T1,Λ =
(k)Nk,+ +
k∈Λ∗ , k=q = N1,Λ
(k)Nk,− ,
(14)
Nk,− ,
(15)
k∈Λ∗ , k=0
Nk,+ +
k∈Λ∗ , k=q
k∈Λ∗ , k=0
η and ζ are complex numbers and ρ is a positive real number. Then one can easily check that (11) is satisﬁed if η=
ω(a0− ) √ , V
ζ=
ω(a∗q+ bq ) V
and
ρ=
ω(N1,Λ ) , V
(16)
BoseEinstein Condensation and Superradiance
265
where the state ω coincides with the equilibrium state ·H1,Λ eff (µ,η,ρ) deﬁned eﬀ by the eﬀective Hamiltonian H1,Λ (µ, η, ρ). By virtue of (11) and (16) we then obtain the selfconsistency equations
1 ∗ 1 1
aq+ bq H1,Λ
N1,Λ H1,Λ η = √ a0− H1,Λ eff (µ,η,ρ) , ζ = eff (µ,η,ρ) , ρ = eff (µ,η,ρ) . V V V (17) Note that since ζ is a function of η and ρ through (17), we do not need to label the eﬀective Hamiltonian by ζ. The important simpliﬁcation here is eﬀ (µ, η, ρ) can be diagonalized: that H1,Λ eﬀ (µ, η, ρ) = E+,Λ (µ, η, ρ)α1∗ α1 + E−,Λ (µ, η, ρ)α2∗ α2 H1,Λ + (λρ − µ)α3∗ α3 + T1,Λ + (λρ − µ)N1,Λ +
(18)
g V ζ , 4(µ − λρ) 2
2
where 1
(Ω + µ − λρ − (q))2 + g 2 η2 , 2 1
(Ω + µ − λρ − (q))2 + g 2 η2 , 2 (19) √ g Vζ , α2 = aq+ sin θ−bq cos θ, α3 = a0− + 2(λρ − µ) (20)
1 (Ω − µ + λρ + (q)) + 2 1 E−,Λ (µ, η, ρ) = (Ω − µ + λρ + (q)) − 2 E+,Λ (µ, η, ρ) =
α1 = aq+ cos θ+bq sin θ, and
tan 2θ = −
gη . Ω + µ − λρ − (q)
(21)
Note that the correlation inequalities (12) (see [20]) imply that lim ω (A∗ , [H1,Λ (µ), A]) ≥ 0
(22)
V →∞
for all observables A. Applying (22) with A = a∗0+ , one gets λρ − µ ≥ 0. Similarly, one obtains the condition λρ + (q) − µ ≥ 0 by applying (22) to A = a∗q+ . We also have that E+,Λ (µ, η, ρ) ≥ E−,Λ (µ, η, ρ) and E−,Λ (µ, η, ρ) = 0 when η2 = 4Ω(λρ+ (q)−µ)/g 2 and then E+,Λ (µ, η, ρ) = Ω −µ+λρ+ (q). Thus we have the constraint: η2 ≤ 4Ω(λρ + (q) − µ)/g 2 . We shall need this information to make sense of the thermodynamic functions below. The (17) can be made explicit using (18): η=
g ζ, 2(µ − λρ)
1 gη ζ= 2 V (E+ − E− )
1 eβE+ − 1
(23) −
1 eβE− − 1
! (24)
266
and
J.V. Pul´e et al.
! 1 1 1 1 1 + + V e−β(µ−λρ) − 1 2V eβE+ − 1 eβE− − 1 ! 1 1 (µ − λρ − (q) + Ω) − − 2V (E+ − E− ) eβE+ − 1 eβE− − 1 1 1 1 1 + + . β((k)−µ+λρ) β((k)−µ+λρ) V e − 1 V k∈Λ∗ , k=0 e −1 k∈Λ∗ , k=q
ρ = η2 +
Combining (23) and (24) we obtain the consistency equation: g 2 η η= 4(µ − λρ)V (E+ − E− )
1 eβE+ − 1
−
1 eβE− − 1
! .
(25)
So, the equilibrium states are determined by the limiting form of the consistency equations (25) and (25). Clearly if E− does not tend to zero as V → ∞ then the righthand side of (25) tends to zero and η = 0. For η = 0 we must have E− → 0, that is η2 → 4Ω(λρ + (q) − µ)/g 2 . In fact for a ﬁnite limit, the largevolume asymptotic should to be 4Ω 1 η2 ≈ 2 λρ + (q) − µ − , (26) g βV τ where τ > 0. This implies that E+ → Ω − µ + λρ + (q), and (25) becomes in the limit: η 1−
Ω βV τ (Ω − µ + λρ + (q))
E− ≈
g2 τ 4(λρ − µ)Ω
(27)
=0.
(28)
The last equation has solutions: η = 0,
or
τ=
For µ < 0, let ε0 (µ) = ρ0 (µ) = and
1 (2π)3
d3 k R3
1 (2π)3
1 p0 (µ) = − (2π)3
4(λρ − µ)Ω . g2
R3
d3 k R3
(k) − µ , −1
eβ((k)−µ)
1 eβ((k)−µ) − 1
d3 kln(1 − e−β((k)−µ) ) ,
(29)
(30) (31)
(32)
BoseEinstein Condensation and Superradiance
267
that is the grandcanonical energy density, the particle density and the pressure for the free Bosegas. Let s0 (µ) = β(ε0 (µ) + p0 (µ)) ,
(33)
and note that s0 (µ) is an increasing function of µ. We shall denote the free Bosegas critical density by ρc := ρ0 (0). Recall that ρc is inﬁnite for ν < 3 and ﬁnite for ν ≥ 3. Now we analyze in detail the solutions of (28) and compute their thermodynamic functions. We consider three cases: Case 1. Suppose that in the thermodynamic limit one has: η = 0 and λρ − µ > 0. By virtue of (24) in this case we have ζ = 0 , i.e. there is no condensation: 1 ∗ 1 ∗
a0− a0− H1,Λ
aq+ aq+ H1,Λ eff (µ,η,ρ) = lim eff (µ,η,ρ) V →∞ V V →∞ V 1 ∗
bq bq H1,Λ = lim eff (µ,η,ρ) = 0 , V →∞ V lim
and the photon and boson subsystems are decoupled. Now (25) takes the form ρ = 2ρ0 (µ − λρ) ,
(34)
the energy density is given by lim
V →∞
1 1 2
H1,Λ (µ)H1,Λ λρ eff (µ,η,ρ) = 2ε0 (µ − λρ) − V 2
and the entropy density is equal to s(µ) = 2s0 (µ − λρ) .
(35)
Since the grandcanonical pressure is given by p(µ) =
1 1 s(µ) − lim
H1,Λ (µ)H1,Λ eff (µ,η,ρ) , V →∞ V β
(36)
then
1 (37) p(µ) = 2p0 (µ − λρ) + λρ2 . 2 Case 2. Now suppose that in this case the thermodynamic limit gives: η = 0, λρ − µ = 0. Then the solution of (25) has the asymptotic form: 1 1 µ + o ρV = + , (38) λ V λτ V with some τ ≥ 0. So, (25) yields the identity ρ = τ + 2ρc ,
(39)
268
J.V. Pul´e et al.
which implies that ρ ≥ 2ρc . Note that this case is possible only if ν ≥ 3. By explicit calculations one gets: lim
V →∞
1 ∗
a a0− H1,Λ eff (µ,η,ρ) = τ = ρ − 2ρc , V 0−
(40)
i.e., there is a possibility of the meanﬁeld condensation of zeromode nonexcited bosons σ = −. Notice that the gauge invariance implies √ lim a0− H1,Λ V =0 eff (µ,0,ρ) / V →∞
Furthermore, we get also: lim
V →∞
1 ∗ 1 ∗
a aq+ H1,Λ
bq bq H1,Λ eff (µ,0,ρ) = lim eff (µ,0,ρ) = 0 , V →∞ V V q+
(41)
i.e., there is no condensation in the q = 0 modes and the laser boson ﬁeld. Hence again the contribution from the interaction term vanishes, i.e. the photon and boson subsystems are decoupled. In this case the energy density is given by: 1 µ2
H1,Λ (µ)H1,Λ (42) lim eff (µ,η,ρ) = 2ε0 (0) − V →∞ V 2λ and the entropy density has the form: s(µ) = 2s0 (0) = 2β(ε0 (0) + p0 (0)) .
(43)
Thus for the pressure one gets: p(µ) = 2p0 (0) +
µ2 . 2λ
(44)
Case 3. Suppose that η = 0. Then by diagonalization of (13) we obtain a simultaneous condensation of the excited/nonexcited bosons and the laser photons in the qmode: lim
V →∞
1 ∗ 4Ω(λρ + (q) − µ) 2
a a0− H1,Λ , eff (µ,η,ρ) = η = V 0− g2
(45)
1 ∗ 4(λρ − µ)Ω
a aq+ H1,Λ , eff (µ,η,ρ) = τ = V q+ g2
(46)
1 ∗ 4(λρ + (q) − µ)(λρ − µ)
bq bq H1,Λ . eff (µ,η,ρ) = V g2
(47)
lim
V →∞
lim
V →∞
Equation (25) becomes: ρ=
8Ω (λρ − µ + (q)/2) + 2ρ0 (µ − λρ) . g2
Using the diagonalization of (13) one computes also
(48)
BoseEinstein Condensation and Superradiance
lim
V →∞
1 ∗
a bq eff V q+ H1,Λ (µ,η,ρ)
4(λρ − µ) Ω(λρ + (q) − µ) = . g2
Then using (46) and (47), one obtains 2 1 lim a∗q+ bq H1,Λ eff (µ,η,ρ) V →∞ V 1 ∗ 1 ∗
aq+ aq+ H1,Λ
bq bq H1,Λ = lim eff (µ,η,ρ) lim eff (µ,η,ρ) . V →∞ V V →∞ V
269
(49)
(50)
In this case the energy density is given by: lim
1
V →∞ V
H1,Λ (µ)H1,Λ eff (µ,η,ρ)=
4Ω(λρ + (q) − µ)(λρ − µ) 1 +2ε0 (µ−λρ)− λρ2 . 2 g 2
The entropy density is again given by s(µ) = 2s0 (µ − λρ)
(51)
1 4Ω(λρ + (q) − µ)(λρ − µ) p(µ) = 2p0 (µ − λρ) + λρ2 − . 2 g2
(52)
and the pressure becomes
Before proceeding to a detailed study of the pressure, we would like to comment an important conclusion which is implied by the fact that instead of a standard inequality we obtain the equality in (50). Proposition 1. The equality (50) implies that the limiting Gibbs state manifests a spontaneous space translation invariance breaking in direction of the vector q. The proof follows from a direct application of arguments proving Theorem III.3 in [21]. These arguments are developed to prove the spontaneous breaking of the gauge invariance, but by inspection one ﬁnds that they cover also the case of translation invariance. Instead of repeating them for our case, we prefer to describe our situation in an explicit manner, cf [21]: For simplicity we take q = (2π/γ)e1 , where e1 = (1, 0, . . . , 0) ∈ Rν and γ > 0. Let ω(·) = limV →∞ ·H1,Λ eff (µ,η,ρ) denote our equilibrium state determined by the parameter η, the chemical potential µ and the density ρ, where in fact η and ρ are functions of µ through the selfconsistency equations. Since the initial Hamiltonian (2) is translation invariant, so is the state ω. Note that (45), (46), (47) and (49) are expectation values in this state of translation invariant operators a∗0− a0− , a∗q+ aq+ , b∗q bq and a∗q+ bq . However the single operators aq+ , and bq are not translation invariant and their averages in the state ω are modulated with period γ in the e1 direction. Let ωγ denote an equilibrium state periodic in the e1 direction with period γ. Then (50) expresses that for ωγ the analogue of the mixing property for ω takes the following form:
270
J.V. Pul´e et al.
1 1 ω(a∗q+ bq ) = lim ωγ (a∗q+ bq ) V →∞ V V →∞ V 1 1 = lim √ ωγ (a∗q+ ) lim √ ωγ (bq ) , V →∞ V →∞ V V lim
(53)
implying that both factors in the righthand side are nonzero. Similarly, we get, for example, that lim
1 1 ω(a0− bq ) = lim ωγ (a0− bq ) V →∞ V V 1 1 = lim √ ωγ (a0− ) lim √ ωγ (bq ) . V →∞ V →∞ V V
V →∞
(54)
To get decoupling in (53) and (54) note that the quasilocal structure (9) implies that 1 1 √ bq = dx eiq·x b(x) (55) V Λ V is an operator spaceaverage, which in the limit is a cnumber in the periodic state ωγ . Therefore the emergence of macroscopic occupation of the laser qmode (47) is accompanied by the creation of a onedimensional optical lattice in the e1 direction with period γ. We can then reconstruct the translation invariant state ω and thus recover (50), by averaging ωγ : 1 γ dξ ωγ ◦ τe1 ξ , (56) ω= γ 0 over elementary interval of the length γ. 2.2 The Pressure for Model 1 Having discussed the three Cases 1–3 we give the values of the chemical potential µ when they occur. This analysis involves a detailed study of the pressure. Let κ = 8Ωλ/g 2 − 1 and α = (q)(κ + 1)/2. From the condition for thermodynamic stability we know that κ > 0. Let x0 be the unique value of x ∈ [0, ∞) such that 2λρ0 (−x) = κ and let µ0 = 2λρ0 (−x0 ) + κx0 . Note that µ0 < 2λρc . – The case when µ0 + α ≥ 2λρc is easy. In this situation Case 1 applies for µ ≤ 2λρc and there exists µ1 > µ0 + α such that Case 2 applies for 2λρc < µ < µ1 and Case 3 for µ ≥ µ1 . – When µ0 + α < 2λρc the situation is more subtle. We show that there exists µ1 > µ0 + α such that Case 3 applies for µ ≥ µ1 . However we are not able to decide on which side of 2λρc , the point µ1 lies. If µ1 > 2λρc the situation is as in the previous subcase, while if µ0 + α < µ1 < 2λρc the intermediate phase where Case 2 obtains is eliminated. This the situation is similar to [7], where one has α = 0.
BoseEinstein Condensation and Superradiance
271
– Note that for ν < 3, Case 1 applies when µ < µ1 and Case 3 when µ ≥ µ1 . Let x = λρ − µ and recall that κ = 8Ωλ/g 2 − 1 and α = (q)(κ + 1)/2. In terms of x and η the above classiﬁcation takes the form: Case 4. η = 0 and x > 0. Then (34) becomes 2λρ0 (−x) − x = µ .
(57)
For µ ≤ 2λρc this has a unique solution in x, denoted by x1 (µ), while for µ > 2λρc it has no solutions. Let p1 (x, µ) := 2p0 (−x) +
(x + µ)2 . 2λ
(58)
Then p(µ) = p1 (x1 (µ), µ) .
(59)
Case 5. η = 0 and x = 0. For µ > 2λρc p(µ) = p2 (µ) := 2p0 (0) +
µ2 . 2λ
(60)
Case 6. η = 0. Then (48) becomes 2λρ0 (−x) + κx + α = µ .
(61)
Recall that x0 is the unique value of x ∈ [0, ∞) such that 2λρ0 (−x) = κ, µ0 = 2λρ0 (−x0 ) + κx0 and that µ0 < 2λρc . Then for µ < µ0 + α, (61) has no solutions. For µ0 + α ≤ µ ≤ 2λρc + α ˜3 (µ) < x3 (µ) if this equation has two solutions: x ˜3 (µ) and x3 (µ), where x ˜3 (µ0 + α) = x3 (µ0 + α). Finally for µ > 2λρc + α it has a µ = µ0 + α, and x unique solution x3 (µ). Let p3 (x, µ) := 2p0 (−x) +
{(x + µ)2 − (κ + 1)x2 − 2αx} . 2λ
(62)
Then dp3 (˜ x ˜3 (µ) + µ x3 (µ) + µ dp3 (x3 (µ), µ) x3 (µ), µ) = < = dµ λ λ dµ
(63)
for µ = µ0 + α. Since p3 (˜ x3 (µ0 + α), µ0 + α) = p3 (x3 (µ0 + α), µ0 + α), p3 (˜ x3 (µ), µ) < p3 (x3 (µ), µ)
(64)
for µ0 + α < µ ≤ 2λρc + α. Therefore p(µ) = p3 (x3 (µ), µ) for all µ ≥ µ0 + α. Note that x ˜3 (2λρc + α) = 0 so that
(65)
272
J.V. Pul´e et al.
p3 (˜ x3 (2λρc + α), 2λρc + α) = p1 (x1 (2λρc ), 2λρc ) = p2 (2λρc ) .
(66)
Therefore x3 (µ0 + α), µ0 + α) < 2p0 (0) + 2λρ2c . p3 (x3 (µ0 + α), µ0 + α) = p3 (˜
(67)
Also for large µ, p3 (x3 (µ), µ) ≈ (µ2 /2λ)((κ + 1)/κ) while p2 (µ) ≈ (µ2 /2λ), so that p3 (x3 (µ), µ) > p2 (µ) eventually. We remark ﬁnally that the slope of p3 (x3 (µ), µ) is greater than that of p2 (µ), dp3 (x3 (µ), µ) x3 (µ) + µ µ dp2 (µ) = > = , dµ λ λ dµ
(68)
so that the corresponding curves intersect at most once. The case α = 0, i.e. (q = 0) = 0, has been examined in [7]. For the case α > 0 we have two subcases: The subcase µ0 + α ≥ 2λρc is easy. In this situation Case 1 applies for µ ≤ 2λρc . From (67) we see that p3 (x3 (µ0 + α), µ0 + α) < 2p0 (0) + 2λρ2c < p2 (µ0 + α)
(69)
and therefore from the behaviour for large µ we can deduce that there exists µ1 > µ0 + α such that Case 2 applies for 2λρc < µ < µ1 and Case 3 for µ ≥ µ1 . The subcase µ0 + α < 2λρc is more complicated. We know that x3 (2λρc ), 2λρc ) < p3 (˜ x3 (2λρc + α), 2λρc + α) = p1 (x1 (2λρc ), 2λρc ) . (70) p3 (˜ x3 (µ), µ) is greater than the slope of Therefore since the slope of p3 (˜ p1 (x1 (µ), µ) for µ0 + α < µ < 2λρc : x ˜3 (µ) + µ x1 (µ) + µ dp1 (x1 (µ), µ) dp3 (˜ x3 (µ), µ) = > = , dµ λ λ dµ
(71)
we can conclude that p3 (x3 (µ0 +α), µ0 +α) = p3 (˜ x3 (µ0 +α), µ0 +α) < p1 (x1 (µ0 +α), µ0 +α) . (72) We also know by the arguments above that there exists µ1 > µ0 +α such that Case 3 applies for for µ ≥ µ1 . However we do know on which side of 2λρc , the point µ1 lies. If µ1 > 2λρc the situation is as in the previous subcase while if µ0 + α < µ1 < 2λρc the intermediate phase where Case 2 obtains is eliminated.
3 Model 2 and MatterWave Grating The analysis for this model is very similar to that of Model 1. Therefore, we only brieﬂy summarize the main results. For Model 2 the eﬀective Hamiltonian is
BoseEinstein Condensation and Superradiance
273
g eﬀ H2,Λ (µ, η, ρ) = (λρ − µ + (q))a∗q aq + (λρ − µ)a∗0 a0 + (ηa∗q bq + η¯aq b∗q ) 2 √ g V ¯ ∗ ) + T + (λρ − µ)N +Ω b∗q bq + (ζa0 + ζa (73) 0 2,Λ 2,Λ 2 where
T2,Λ =
(k)Nk ,
(74)
(k)Nk .
(75)
k∈Λ∗ , k=0 k=q
= N2,Λ
k∈Λ∗ , k=0 k=q
The parameters η, ζ and ρ satisfy the selfconsistency equations: 1 1 ∗ 1
aq bq H2,Λ
N2,Λ H2,Λ η = √ a0 H2,Λ eff (µ,η,ρ) , ζ = eff (µ,η,ρ) , ρ = eff (µ,η,ρ) . V V V (76) Solving these equations we again have three cases: Case 7. ζ = η = 0 and λρ − µ > 0. In this case there is no condensation: lim
V →∞
1 ∗ 1 ∗
a a0 eff
a aq eff = lim V 0 H2,Λ (µ,η,ρ) V →∞ V q H2,Λ (µ,η,ρ) 1 ∗
bq bq H2,Λ = lim eff (µ,η,ρ) = 0 , V →∞ V
the density equation is ρ = ρ0 (µ − λρ)
(77)
and the pressure is 1 p(µ) = p0 (µ − λρ) + λρ2 . 2 Case 8. η = 0, λρ − µ = 0. Here ρ ≥ ρc and lim
V →∞
1 ∗
a a0 eff = ρ − ρc . V 0 H2,Λ (µ,η,ρ)
There is condensation in the k = 0 mode but there is no condensation in the k = q mode and the photon laser ﬁeld: 1 ∗ 1 ∗
aq aq H2,Λ
bq bq H2,Λ eff (µ,η,ρ) = lim eff (µ,η,ρ) = 0 . V →∞ V V →∞ V lim
(78)
The pressure density is given by p(µ) = p0 (0) +
µ2 . 2λ
(79)
Case 9. η = 0. There is simultaneous condensation of the zeromode and the qmode bosons as well as the laser qmode photons: lim
V →∞
1 ∗ 4Ω(λρ + (q) − µ)
a a0 eff = , V 0 H2,Λ (µ,η,ρ) g2
(80)
274
J.V. Pul´e et al.
lim
V →∞
1 ∗ 4(λρ − µ)Ω
a aq eff = , V q H2,Λ (µ,η,ρ) g2
1 ∗ 4(λρ + (q) − µ)(λρ − µ)
b bq eff = , V q H2,Λ (µ,η,ρ) g2
4(λρ − µ) Ω(λρ + (q) − µ) 1 ∗
aq bq H2,Λ lim eff (µ,η,ρ) = V →∞ V g2 lim
V →∞
and
2 1 lim a∗q bq H2,Λ eff (µ,η,ρ) V →∞ V
(81) (82) (83)
(84)
1 ∗ 1 ∗
aq aq H2,Λ
bq bq H2,Λ eff (µ,η,ρ) lim eff (µ,η,ρ) . V →∞ V V →∞ V
= lim
The density is given by ρ=
8Ω (λρ − µ + (q)/2) + ρ0 (µ − λρ) g2
and pressure is 1 4Ω(λρ + (q) − µ)(λρ − µ) p(µ) = p0 (µ − λρ) + λρ2 − . 2 g2 Note that relations between the values of µ and the three cases above are exactly the same as for Model 1 apart from the fact that 2ρ0 is now replaced by ρ0 and 2ρc by ρc . For that one has to compare the kinetic energy operators (1) and (4). The recently observed phenomenon of periodic spacial variation in the bosondensity is responsible for the light and matterwave ampliﬁcation in superradiant condensation, see [2–4, 13]. This so called matterwave grating is produced by the interference of two diﬀerent macroscopically occupied momentum states: the ﬁrst corresponds to a macroscopic number of recoiled bosons and the second to residual BE condensate at rest. To study the possibility of such interference in Model 1 we recall that for system (2), with two species of boson atoms the local particle density operator has the form ρ(x) :=
1 V
ρk, σ eikx ,
(85)
k∈Λ∗ , σ=±
where the Fourier transforms of the local particle densities for two species are ρk, σ := a∗p+k, σ ap, σ . (86) p∈Λ∗
If the limiting equilibrium states generated by the Hamiltonian (2) is translation invariant, the momentum conservation law yields:
BoseEinstein Condensation and Superradiance
lim ω(a∗p+k, σ ap, σ ) = δk, 0 lim ω(a∗p, σ ap, σ ) .
V →∞
V →∞
275
(87)
So, in this state the equilibrium expectation of the local density lim ω(ρ(x)) =
V →∞
lim
V →∞
1 V
eikx
k∈Λ∗ ,σ
(88)
ω(a∗p+k,σ ap,σ ) = lim ω(ρ(0)) V →∞
p∈Λ∗
is a constant. Since condensation may break the translation invariance in one direction (see Proposition 1), we get a corresponding nonhom*ogeneity (grating) of the equilibrium total particle density in the extremal ωγ state. This means that in the integral sum (88) over k, the ±qmode terms survive in the thermodynamic limit. By (53) and (54) we know that condensation occurs only in the zero mode for the σ = − bosons and in the qmode for the σ = + bosons. and therefore we have the following relations: 1 1 1 ωγ (a∗q,+ a0,+ ) = lim √ ωγ (a∗q,+ ) lim √ ωγ (a0,+ ) , V →∞ V →∞ V V V 1 1 1 ∗ ∗ ωγ (aq,− a0,− ) = lim √ ωγ (aq,− ) lim √ ωγ (a0,− ) . lim V →∞ V V →∞ V →∞ V V lim
V →∞
(89) (90)
To get decoupling √ in (89) and√(90) one has to note that (as in (53), (54)) the operators a0,± / V , or a∗q,± / V , are spaceaverages, which in the limit are cnumbers in the periodic state ωγ . Since there is no condensation of σ = ± bosons except in those two modes, the righthand sides of both (89) and (90) are equal to zero. Noting that lim
V →∞
1 ωγ (a∗p+k,σ ap,σ ) = 0 V
(91)
for any other mode, one gets the space hom*ogeneity of the equilibrium particle density in the extremal ωγ state: limV →∞ ωγ (ρ(x)) = const. So, for the Model 1 we have no particle density space variation even in the presence of the light corrugated lattice of condensed photons, cf Proposition 1. Let us now look at the corresponding situation in Model 2. The important diﬀerence is that in this model the same boson atoms may condense in two states (see Case 3): 1 lim √ ωγ (a∗q ) = 0 V
V →∞
and
1 lim √ ωγ (a0 ) = 0 . V
V →∞
(92)
Then (90) implies lim
V →∞
1 1 1 ωγ (a∗q a0 ) = lim √ ωγ (a∗q ) lim √ ωγ (a0 ) ≡ ξ = 0 . V →∞ V →∞ V V V
(93)
Therefore, the bosons of these two condensates may interfere. By virtue of (88) and (93) this gives the matterwave grating formed by two macroscopically occupied momentum states:
276
J.V. Pul´e et al.
1 V →∞ V
lim ωγ (ρ(x)) = lim
V →∞
k∈Λ∗ ,σ
eikx
ωγ (a∗p+k,σ ap,σ )
(94)
p∈Λ∗
1 ikx e ωγ (ρ◦k ) , V →∞ V
= (ξeiqx + ξe−iqx ) + lim
k=±q
where
ρ◦k :=
a∗p+k ap .
(95)
p=0,q
Notice that by (93) and by (94) there is no matterwave grating in the Case 2, when one of the condensates (for example the qcondensate) is empty, see (41).
4 Conclusion It is clear that the absence of the matterwave grating in Model 1 and its presence in Model 2 provides a physical distinction between Raman and Rayleigh superradiance. Note ﬁrst that matterwave ampliﬁcation diﬀers from light ampliﬁcation in one important aspect: a matterwave ampliﬁer has to possess a reservoir of atoms. In Models 1 and 2 this is the BE condensate. In both models the superradiant scattering transfers atoms from the condensate at rest to a recoil mode. The gain mechanism for the Raman ampliﬁer is superradiant Raman scattering in a twolevel atoms, transferring bosons from the condensate into the recoil state [1]. The Rayleigh ampliﬁer is in a sense even more eﬀective. Since now the atoms in a recoil state interfere with the BE condensate at rest, the system exhibits a space matterwave grating and the quantummechanical amplitude of transfer into the recoil state is proportional to the product N0 (Nq + 1). Each time the momentum imparted by photon scattering is absorbed by the matterwave grating by the coherent transfer of an atom from the condensate into the recoil mode. Thus, the variance of the grating grows, since the quantum amplitude for scattered atom to be transferred into a recoiled state is increasing [2–4, 13]. At the same time the dressing laser beam prepares from the BE condensate a gain medium able to amplify the light. The matterwave grating diﬀracts the dressing beam into the path of the probe light resulting in the ampliﬁcation of the latter [5]. In the case of equilibrium BEC superradiance the ampliﬁcation of the light and the matter waves manifests itself in Models 1 and 2 as a mutual enhancement of the BEC and the photons condensations, see Cases 3 in Sects. 2 and 3. Note that the corresponding formulæ for condensation densities for Model 1 (45)–(47) and for Model 2 (80), (81), (82) are identical. The same is true for the bosonphoton correlations (entanglements) between recoiled bosons and photons, see (49), (83), as well as between photons and the BE condensate at rest:
BoseEinstein Condensation and Superradiance
1 ∗ 1 ∗
a0− bq H1,Λ
a bq eff eff (µ,η,ρ) = V →∞ V V 0 H2,Λ (µ,η,ρ) √ 4Ω(λρ + (q) − µ) λρ − µ = , g2 lim
277
(96)
and for the oﬀdiagonal coherence between recoiled atoms and the condensate at rest: lim
1 1
a0− a∗q+ H1,Λ
a0 a∗q H2,Λ eff (µ,η,ρ) = eff (µ,η,ρ) V V
4Ω (λρ + (q) − µ)(λρ − µ) . = g2
V →∞
(97)
As we have shown above, the diﬀerence between Models 1 and 2 becomes visible only on the level of the wavegrating or spatial modulation of the local particle density (94).
Acknowledgement We wish to thank Alain Joye and Joachim Asch for invitation to give this lecture on QMath9.
References 1. D. Schneble, G.K. Campbell, E.W. Streed et al. Phys. Rev. A 69, 041601 (2004). 2. S. Inouye, A.P. Chikkatur, D.M. StamperKurn et al. Science 285, 571 (1999). 3. D. Schneble, Y. Toril, M. Boyd et al. Science 300, 475 (2003). 4. W. Ketterle, S. Inouye: Phys. Rev. Lett. 89, 4203 (2001). 5. W. Ketterle, S. Inouye: C.R. Acad. Sci. Paris s´erie IV 2, 339 (2001). 6. J.V. Pul´e, A. Verbeure, V.A. Zagrebnov: J. Phys. A (Math. Gen.) 37, L321 (2004). 7. J.V. Pul´e, A.F. Verbeure, V.A. Zagrebnov: J. Stat. Phys. 118 (2005). 8. J.V. Pul´e, A.F. Verbeure, V.A. Zagrebnov: J. Phys. A (Math. Gen.) (2005). 9. N. Piovella, R. Bonifacio, B.W.J. McNeil et al. Optics Commun. 187, 165 (2001). 10. R. Bonifacio, F.S. Cataliotti, M. Cola et al. Optics Commun. 233, 155 (2004). 11. M.G. Moore, P. Meystre: Phys. Rev. Lett. 86, 4199 (2001). 12. M.G. Moore, P. Meystre: Phys. Rev. Lett. 83, 5202 (1999). 13. M. Kozuma, Y. Suzuki, Y. Torii et al. Science 286, 2309 (1999). 14. R.H. Dicke: Phys. Rev. 93, 99 (1954). 15. K. Hepp, E.H. Lieb: Ann. Phys. 76, 360 (1973). 16. J.G. Brankov, V.A. Zagrebnov, N.S. Tonchev: Theor. Math. Phys. 22, 13 (1975). 17. M. Fannes, P.N.M. Sisson, A. Verbeure et al. Ann. Phys. 98, 38 (1976).
278
J.V. Pul´e et al.
18. O. Bratteli, D.W. Robinson: Operator Algebras and Quantum Statistical Mechanics II, 2nd edn (Springer, Berlin Heidelberg New York 1997). 19. M. Fannes, H. Spohn, A. Verbeure: J. Math. Phys. 21, 355 (1980). 20. M. Fannes, A. Verbeure: Commun. Math. Phys. 57, 165 (1977). 21. M. Fannes, J.V. Pul´e, A. Verbeure: Helv.Phys.Acta 55, 391 (1982).
Derivation of the GrossPitaevskii Hierarchy Benjamin Schlein∗ Department of Mathematics, Stanford University, Stanford, CA 94305, USA [emailprotected]
Abstract. We report on some recent results regarding the dynamical behavior of a trapped BoseEinstein condensate, in the limit of a large number of particles. These results were obtained in [4], a joint work with L. Erd˝ os and H.T. Yau.
1 Introduction In the last years, progress in the experimental techniques has made the study of dilute Bose gas near the ground state a hot topic in physics. For the ﬁrst time, the existence of BoseEinstein condensation for trapped gases at very low temperatures has been veriﬁed experimentally. The experiments were conducted observing the dynamics of Bose systems, trapped by strong magnetic ﬁeld and cooled down at very low temperatures, when the conﬁning traps are switched oﬀ. It seems therefore important to have a good theoretical description of the dynamics of the condensate. Already in 1961 Gross [7,8] and Pitaevskii [14] proposed to model the many body eﬀects in a trapped dilute Bose gas by a nonlinear onsite self interaction of a complex order parameter (the condensate wave function ut ). They derived the GrossPitaevskii equation (1) i∂t ut = −∆ut + 8πa0 ut 2 ut for the evolution of ut . Here a0 is the scattering length of the pair interaction. A mathematically rigorous justiﬁcation of this equation is still missing. The aim of this article is to report on recent partial results towards the derivation of (1) starting from the microscopic quantum dynamics in the limit of a large number of particles. Here we only expose the main ideas: for more details, and for all the proofs, we refer to [4]. Also in the mathematical analysis of dilute bosonic systems some important progress has recently been made. In [13], Lieb and Yngvason give a rigorous proof of a formula for the leading order contribution to the ground state energy of a dilute Bose gas (the correct upper bound for the energy was already obtained by Dyson in [2], for the case of hard spheres). This important result inspired a lot of subsequent works establishing diﬀerent properties of the ground state of the Bose system. In [12], the authors give a proof of ∗
Supported by NSF Postdoctoral Fellowship
B. Schlein: Derivation of the GrossPitaevskii Hierarchy, Lect. Notes Phys. 690, 279–293 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
280
B. Schlein
the asymptotic exactness of the GrossPitaevskii energy functional for the computation of the ground state energy of a trapped Bose gas. In [10], the complete condensation of the ground state of a trapped Bose gas is proven. For a review of recent results on dilute Bose systems we refer to [11]. All these works investigate the properties of the ground state of the system. Here, on the other hand, we are interested in the dynamical behavior. Next, we want to describe our main result in some details. To this end, we need to introduce some notation. From now on we consider a system of N bosons trapped in a box Λ ⊂ R3 with volume one and we impose periodic boundary conditions. In order to describe the interaction among the bosons, we choose a positive, smooth, spherical symmetric potential V (x) with compact support and with scattering length a0 . Let us brieﬂy recall the deﬁnition of the scattering length a0 of the potential V (x). To deﬁne a0 we consider the radial symmetric solution f (x) of the zero energy oneparticle Schr¨odinger equation 1 (2) −∆ + V (x) f (x) = 0 , 2 with the condition f (x) → 1 for x → ∞. Since the potential has compact support, we can deﬁne the scattering length a0 associated to V (x) by the equation f (x) = 1 − a0 /x for x outside the support of V (x) (this deﬁnition can be generalized by a0 = limr→∞ r(1 − f (r)), if V has unbounded support but still decays suﬃciently fast at inﬁnity). Another equivalent characterization of the scattering length is given by the formula dx V (x)f (x) = 8πa0 . (3) Physically, a0 is a measure of the eﬀective range of the potential V (x). The Hamiltonian of the N boson system is then given by H=−
N
∆j +
j=1
Va (xi − xj )
(4)
i<j
with Va (x) = (a0 /a)2 V ((a0 /a)x). By scaling, Va has scattering length a. In the following we keep a0 ﬁxed (of order one) and we vary a with N , so that when N tends to inﬁnity a approaches zero. In order for the GrossPitaevskii theory to be relevant we have to take a of order N −1 (see [12] for a discussion of other possible scalings). In the following we choose therefore a = a0 /N , and thus Va (x) = N 2 V (N x). Note that, with this choice of a, the Hamiltonian (4) can be viewed as a special case of the meanﬁeld Hamiltonian Hmf = −
N j=1
∆j +
1 3 β V (β(xi − xj )) . N i<j
(5)
Derivation of the GrossPitaevskii Hierarchy
281
The GrossPitaevskii scaling is recovered when β = N . We study the dynamics generated by (5) for other choices of β (β = N α , with α < 3/5) in [3]. Since we have N particles in a box of volume one, the density is given by ρ = N . Hence, the total number of particles interacting at a given time with a ﬁxed particle in the system is typically of order ρa3 N −2 1 . This means that our system is actually a very dilute gas, scaled so that the total volume remains ﬁxed to one. The dynamics of the N boson system is determined by the Schr¨odinger equation (6) i∂t ψN,t = HψN,t for the wave function ψN,t ∈ L2 (R3N , dx). Instead of describing the quantum mechanical system through its wavefunction we can describe it by the corresponding density matrix γN,t = ψN,t ψN,t  which is the orthogonal projection onto ψN,t . We choose the normalization so that Tr γN,t = 1. The Schr¨ odinger equation (6) takes the form i∂t γN,t = [H, γN,t ] .
(7)
For large N this equation becomes very diﬃcult to solve, even numerically. Therefore, it is desirable to have an easier description of the dynamics of the system in the limit N → ∞, assuming we are only interested in its macroscopic behavior, resulting from averaging over the N particles. In order to investigate the macroscopic dynamics, we introduce the marginal distributions associated to the density matrix γN,t . The kparticle marginal distribution (k) γN,t is deﬁned by taking the partial trace over the last N − k variables. That (k)
is, the kernel of γN,t is given by (k) γN,t (xk ; xk ) = dxN −k γN,t (xk , xN,k ; xk , xN −k ) where γN,t (x; x ) denotes the kernel of the density matrix γN,t . Here and in the following we use the notation x = (x1 , . . . , xN ), xk = (x1 , . . . , xk ), xN −k = (xk+1 , . . . , xN ), and analogously for the primed variables. By deﬁnition, the kparticle marginal distributions satisfy the normalization condition (k)
Tr γN,t = 1
for all k = 1, . . . , N .
In contrast to the density matrix γN,t , one can expect that, for ﬁxed k, the (k) (k) marginal distribution γN,t has a well deﬁned limit γ∞,t for N → ∞ (with respect to some suitable weak topology), whose dynamics can be investigated. In particular, the GrossPitaevskii equation (1) is expected to describe the (1) time evolution of the limit γ∞,t of the oneparticle marginal distribution, (1)
provided γ∞,t = ut ut  is a pure state. Equation (1) can be generalized, for (1) γ∞,t
describing a mixed state, to
282
B. Schlein
i∂t γ∞,t (x; x ) = (−∆ + ∆ )γ∞,t (x; x ) (1) (1) (1) + 8πa0 γ∞,t (x; x) − γ∞,t (x ; x ) γ∞,t (x; x ) , (1)
(1)
(8)
which we again denote as the GrossPitaevskii equation. To understand the origin of (8), we start from the dynamics of the mar(k) ginals γN,t , for ﬁnite N . From the Schr¨ odinger equation (7), we can easily derive a hierarchy of N equations, commonly called the BBGKY hierarchy, (k) describing the evolution of the distributions γN,t , for k = 1, . . . , N : i∂t γN,t (xk ; xk ) = (k)
k
(−∆xj + ∆xj )γN,t (xk ; xk ) (k)
j=1
+
k
(Va (xj − x ) − Va (xj − x ))γN,t (xk ; xk ) (k)
j=
+ (N − k)
k
(9)
dxk+1 (Va (xj − xk+1 ) − Va (xj − xk+1 ))
j=1
×
(k+1) γN,t (xk , xk+1 ; xk , xk+1 )
.
(k)
Here we use the convention that γN,t = 0, for k > N . Hence, the oneparticle (1)
marginal density γN,t satisﬁes i∂t γN,t (x1 ; x1 ) = (−∆x1 + ∆x1 )γN,t (x1 ; x1 ) (10) (2) + (N − 1) dx2 (Va (x1 − x2 ) − Va (x1 − x2 )) γN,t (x1 , x2 ; x1 , x2 ) . (1)
(1)
(1)
In order to get a closed equation for γN,t we need to assume some relation (1)
(2)
between γN,t and γN,t . The most natural assumption consists in taking the two particle marginal to be the product of two identical copies of the one particle marginal. Although this kind of factorization cannot be true for ﬁnite (k) N , it may hold in the limit N → ∞. We suppose therefore that γ∞,t , for (k)
k = 1, 2, is a limit point of γN,t , with respect to some weak topology, with the factorization property γ∞,t (x1 , x2 ; x1 , x2 ) = γ∞,t (x1 ; x1 )γ∞,t (x2 ; x2 ) . (2)
(1)
(1)
Under this assumption we could naively guess that, in the limit N → ∞, (10) takes the form i∂t γ∞,t (x1 ; x1 ) = (−∆x1 + ∆x1 )γ∞,t (x1 ; x1 ) (1)
(1)
+ (Qt (x1 ) − Qt (x1 )) γ∞,t (x1 ; x1 ) (1)
(11)
Derivation of the GrossPitaevskii Hierarchy
with
283
(1) Qt (x1 ) = lim N dx2 Va (x1 − x2 )γ∞,t (x2 ; x2 ) N →∞ (1) = lim dx2 N 3 V (N (x1 − x2 ))γ∞,t (x2 ; x2 ) N →∞
(1)
= b0 γ∞,t (x1 ; x1 ) where we deﬁned b0 = dx V (x). Using the last equation, (11) can be rewritten as i∂t γ∞,t (x1 ; x1 ) = (−∆x1 + ∆x1 )γ∞,t (x1 ; x1 ) (1) (1) (1) + b0 γ∞,t (x1 ; x1 ) − γ∞,t (x1 ; x1 ) γ∞,t (x1 ; x1 ) (1)
(1)
(12)
which is exactly the GrossPitaevskii equation (8), but with the wrong coupling constant in front of the nonlinear term (b0 instead of 8πa0 ). The fact that we get the wrong coupling constant suggests that something was not completely correct with the naif argument leading from (10) to (12). Reconsidering the argument, the origin of the error is quite clear: when passing to (2) (2) the limit N → ∞ we ﬁrst replaced γN,t with γ∞,t and only after this replacement we took the limit N → ∞ in the potential. This procedure gives the (2) wrong result because the marginal distribution γN,t has a short scale structure living on the scale 1/N , which is the same length scale characterizing (2) the potential Va (x). The short scale structure of γN,t (which describes the correlations among the particles) disappears when the weak limit is taken, (2) so that γ∞,t lives on a length scale of order one. Therefore, in (12) we get the wrong coupling constant because we erroneously disregarded the eﬀect (2) of the correlations present in γN,t . It is hence clear that in order to derive the GrossPitaevskii equation (8) with the correct coupling constant 8πa0 , (2) we need to take into account the short scale structure of γN,t . To this end we begin by studying the ground state of the system. A good approximation for the ground state wave function of the N boson system is given by N W (x) = f (N (xi − xj )) i<j
where f (x) is deﬁned by (2) (then f (N x) solves the same equation (2) with V replaced by Va ). Since we assumed the potential to be compactly supported (let R denote the radius of its support), we have f (x) = 1−a0 /x, for x > R, and thus f (N x) = 1 − a0 /N x = 1 − a/x, for x > Ra. A similar ansatz for the ground state wave function was already used by Dyson in [2] to prove his upper bound on the ground state energy. In order to describe states of the condensate, it seems appropriate to consider wave functions of the form
284
B. Schlein
ψN (x) = W (x)φN (x) where φN (x) varies over distances of order one, and is approximately factor>N ized, that is φN (x) j=1 φ(xj ). Assuming for the moment that this form is preserved under the timeevolution we have γN,t (x1 , x2 ; x1 , x2 ) f (N (x1 − x2 ))f (N (x1 − x2 ))γN,t (x1 ; x1 )γN,t (x2 ; x2 ) . (2)
(1)
(1)
(2)
Thus, for ﬁnite N , γN,t is not exactly factorized and has a short scale structure given by the function f (N x). When we consider the limit N → ∞ of the second term on the right hand side of (10) we obtain (2) lim N dx2 Va (x1 − x2 )γN,t (x1 , x2 ; x1 , x2 ) N →∞ (1) (1) = lim N 3 dx2 V (N (x1 − x2 ))f (N (x1 − x2 ))γ∞,t (x1 ; x1 )γ∞,t (x2 ; x2 ) N →∞
= 8πa0 γ∞,t (x1 ; x1 )γ∞,t (x1 ; x1 ) (1)
(1)
(13) (1)
where we used (3) and the fact that γN,t lives on a scale of order one (and (1)
thus we can replace it by γ∞,t without worrying about the correlations). This (1)
leads to the GrossPitaevskii equation for γ∞,t , (1) (1) i∂t γ∞,t (x1 ; x1 ) = −∆x1 + ∆x1 γ∞,t (x1 ; x1 ) (1) (1) (1) + 8πa0 γ∞,t (x1 ; x1 ) − γ∞,t (x1 ; x1 ) γ∞,t (x1 ; x1 ) which has the correct coupling constant in front of the nonlinear term. Note that the factorization γ∞,t (x1 , x2 ; x1 , x2 ) = γ∞,t (x1 ; x1 )γ∞,t (x2 ; x2 ) (2)
(1)
(1)
(2)
still holds true, because the short scale structure of γN,t vanishes when the weak limit N → ∞ is taken. The short scale structure only shows up in the GrossPitaevskii equation due to the singularity of the potential. In order to make this heuristic argument for the derivation of the GrossPitaevskii equation rigorous, we are faced with two major steps. (i) In the ﬁrst step we have to prove that the kparticle marginal density in the limit N → ∞ really has the short scale structure we discussed above. That is we have to prove that, for large N , k+1 (k+1) f (N (xi − xj ))f (N (xi − xj )) γN,t (xk+1 ; xk+1 ) i<j
× γ∞,t (xk+1 ; xk+1 ) (14) (k+1)
Derivation of the GrossPitaevskii Hierarchy (k+1)
285
(k+1)
where γ∞,t is the limit of γN,t with respect to some suitable weak topology (in the heuristic argument above we considered the case k = 1, here k is an arbitrary ﬁxed integer k ≥ 1). Equation (14) would then imply that, as N → ∞, the last term on the r.h.s. of the BBGKY hierarchy (9) converges to lim N
N →∞
dxk+1 Va (xj − xk+1 )γN,t (xk+1 ; xk+1 ) (k+1)
= 8πa0 γ∞,t (xk , xj ; xk , xj ) . (k+1)
Therefore, if we could also prove that the second term on the r.h.s. of (9) vanishes in the limit N → ∞ (as expected, because formally of the (k) order N −1 ), then it would follow that the family γ∞,t satisﬁes the GrossPitaevskii hierarchy i∂t γ∞,t (xk ; xk ) = (k)
k
(k) −∆j + ∆j γ∞,t (xk ; xk )
j=1
+ 8πa0
k
dxk+1 δ(xk+1 − xj ) − δ(xk+1 − xj )
j=1
×
(k+1) γ∞,t (xk , xk+1 ; xk , xk+1 )
(15)
for all k ≥ 1. We already know that this inﬁnite hierarchy of equation (k) has a solution. In fact the factorized family of densities γ∞,t (xk ; xk ) = >k (1) (1) j=1 γ∞,t (xj ; xj ) is a solution of (15) if and only if γ∞,t solves the GrossPitaevskii equation (8). (k) (ii) Secondly, we need to prove that the densities γ∞,t factorize, that is, that, for all k ≥ 1, k (k) (1) γ∞,t (xj ; xj ) . (16) γ∞,t (xk ; xk ) = j=1 (1)
Then, from (15) and (16), it would follow that γ∞,t is a solution of the GrossPitaevskii equation (8). Note that, since we already know that (15) has a factorized solution, in order to prove (16) it is enough to prove the uniqueness of the solution of the inﬁnite hierarchy (15). Unfortunately, due to the singularity of the δfunction, we are still unable to prove that (15) has a unique solution and thus we cannot prove part (ii) (the best result in this direction is the proof of the uniqueness for the hierarchy with a Coulomb singularity, see [5]). On the other hand we can complete (k) part i) of our program, that is, we can prove that any limit point {γ∞,t }k≥1 (k)
of the family {γN,t }N k=1 (with respect to an appropriate weak topology), satisﬁes the inﬁnite hierarchy (15), provided we replace the original Hamiltonian
286
B. Schlein
where we artiﬁcially modify the interH with a slightly modiﬁed version H, action when a large number of particles come into a region with diameter much smaller than the typical interparticle distance. Since H agrees with apart in the very rare event (rare with respect to the expected typical H, distribution of the particles) that many particles come very close together, we don’t expect this modiﬁcation to change the macroscopic dynamics of the system: but unfortunately we cannot control this eﬀect rigorously. Note that the GrossPitaevskii equation (1) is a nonlinear Hartree equation i∂t ut = −∆ut + (V ∗ ut 2 )ut (17) in the special case V (x) = 8πa0 δ(x). In the literature there are several works devoted to the derivation of (17) from the N body Schr¨ odinger equation. The ﬁrst results were obtained by Hepp in [9], for a smooth potential V (x), and by Spohn in [15], for bounded V (x). Later, Ginibre and Velo extended these results to singular potentials in [6]: their result is limited to coherent initial states, for which the number of particles cannot be ﬁxed. In [5], Erd˝ os and Yau derived (17) for the Coulomb potential V (x) = ±1/x. More recently, Adami, Bardos, Golse and Teta obtained partial results for the potential V (x) = δ(x), which leads to the GrossPitaevskii equation, in the case of onedimensional systems; see [1].
2 The Main Result In this section we explain how we need to modify the Hamiltonian and then we state our main theorem. In order to derive (15) it is very important to ﬁnd a good approximation for the wave function of the ground state of the N boson system. We need an approximation which reproduces the correct short scale structure and, at the same time, does not become too singular (so that error terms can be controlled). Our ﬁrst guess fa (xi − xj ) = f (N (xi − xj )) (18) W (x) = i<j
i<j
is unfortunately not good enough. First of all we need to cutoﬀ the correlations at large distances (we want fa (x) = 1 for x a). To this end we ﬁx a length scale 1 a, and we consider the Neumann problem on the ball {x : x ≤ 1 } (we will choose 1 = N −2/3+κ for a small κ > 0). We are interested in the solution of the ground state problem (−∆ + 1/2Va (x))(1 − w(x)) = e1 (1 − w(x)) on {x : x ≤ 1 }, with the normalization condition w(x) = 0 for x = 1 . Here e1 is the lowest possible eigenvalue. It is easy to check that, up to contributions of lower order, e1 3a/31 . We can extend w(x) to be identically zero, for x ≥ 1 . Then
Derivation of the GrossPitaevskii Hierarchy
(−∆ + 1/2Va (x))(1 − w(x)) = q(x)(1 − w(x)), 3a q(x) 3 χ(x ≤ 1 ) . 1
287
with (19)
For a x 1 , the function 1 − w(x) still looks very much like 1 − a/x, but now it equals one, for x ≥ 1 . Replacing fa (xi − xj ) by 1 − w(xi − xj ) in (18) is>still not suﬃcient for our purposes. The problem is that the wave function i<j (1 − w(xi − xj )) becomes very singular when a large number of particles come very close together. We introduce another cutoﬀ to avoid this problem. We ﬁx a new length scale 1 a, such that N −1/3 (that is is still much smaller than the typical interparticle distance: we will choose = N −2/5−κ for a small κ > 0). Then, for ﬁxed indices i and j, and for an arbitrary ﬁxed number K ≥ 1, we cutoﬀ the correlation between particles i and j (that is we replace 1 − w(xi − xj ) by one) whenever at least K other particles come inside a ball of radius around i and j. In order to keep our exposition as clear as possible we choose K = 1, that is we cutoﬀ correlations if at least three particles come very close together. But there is nothing special about K = 1: what we really need to avoid are correlations among a macroscopic number of particles, all very close together. To implement our cutoﬀ we introduce, for ﬁxed indices i, j, a function Fij (x) with the property that xi − xm  xj − xm 
Fij (x) ∼ = 1 if
for all m = i, j
Fij (x) ∼ = 0 otherwise. > Instead of using the wave function i<j (1 − w(xi − xj )) we will approximate the ground state of the N boson system by (1 − w(xi − xj )Fij (x)) , (20) W (x) = i<j
(the exact deﬁnition of W (x) is a little bit more complicated; see [4], Sect. 2.3). The introduction of the cutoﬀs Fij in the wave function W (x) forces us to modify the Hamiltonian H. To understand how H has to be modiﬁed, we compute its action on W (x). We have, using (19), q(xi − xj ) W (x)−1 (HW )(x) = i,j
+
((1/2)Va (xi − xj ) − q(xi − xj )) (1 − Fij (x))
i,j
+ lower order contributions. The “lower order contributions” are terms containing derivatives of Fij : they need some control, but they are not very dangerous for our analysis. On the other hand, the second term on the r.h.s. of the last equation, whose presence
288
B. Schlein
is due to the introduction of the cutoﬀs Fij , still contains the potential Va and unfortunately we cannot control it with our techniques. Therefore, we by artiﬁcially remove it, deﬁning a new Hamiltonian H, =H− H ((1/2)Va (xi − xj ) − q(xi − xj )) (1 − Fij (x)) . i,j
equals the physical Hamiltonian H unless Note that the new Hamiltonian H three or more particles come at distances less than N −1/3 . This is a rare event, and thus we don’t expect the modiﬁcation of the Hamiltonian H to change in a macroscopic relevant way the dynamics of the system. Before stating our main theorem, we still have to specify the topology we (k) use in taking the limit N → ∞ of the marginal distributions γN,t . It is easy to check that, for every k ≥ 1, γN,t (xk ; xk ) ∈ L2 (Λk × Λk ). This motivates L the following deﬁnition. For Γ = {γ (k) }k≥1 ∈ k≥1 L2 (Λk × Λk ), and for a ﬁxed ν > 1, we deﬁne the two norms ν −k γ (k) 2 and Γ + := sup ν k γ (k) 2 Γ − := (21) (k)
k≥1
k≥0
where .2 denotes the L2 norm on Λk × Λk . We have to introduce the pa(k) rameter ν > 1 to make sure that, for ΓN,t = {γN,t }N k=1 , the norm ΓN,t − is ﬁnite (choosing ν large enough, we ﬁnd ΓN,t − ≤ 1, uniformly in N and t). We also deﬁne the Banach spaces 7 L2 (Λk × Λk ) : Γ − < ∞} H− := {Γ = {γ (k) }k≥0 ∈ k≥1
and H+ := {Γ = {γ (k) }k≥0 ∈
7
L2 (Λk × Λk ) : lim ν k γ (k) 2 = 0} . k→∞
k≥1
Then we have (H− , .− ) = (H+ , .+ )∗ . This induces a weak* topology on H− , with respect to which the unit ball B− of H− is compact (BanachAlaouglu Theorem). Since the space H+ is separable, the weak* topology on the unit ball B− is metrizable: we can ﬁnd a metric ρ on H− , such that a sequence Γn ∈ B− converges with respect to the weak* topology if and only if it converges with respect to the metric ρ. For a ﬁxed time T , we will also consider the space C([0, T ], B− ) of functions of t ∈ [0, T ], with values in the unit ball B− ⊂ H− , which are continuous with respect to the metric ρ (or equivalently with respect to the weak* topology of H− ). We equip C([0, T ], B− ) with the metric ρ(Γ1 (t), Γ2 (t)) = sup ρ(Γ1 (t), Γ2 (t)) . t∈[0,T ]
Derivation of the GrossPitaevskii Hierarchy
289
(k)
In the following we will consider the families ΓN,t = {γN,t }N k=1 as elements of C([0, T ], B− ), and we will study their convergence and their limit points with respect to the metric ρ. We are now ready to state our main theorem. Theorem 1. Assume a = a0 /N , 1 = N −2/3+κ , = N −2/5−κ , for some suﬃciently small κ > 0. Assume 2 ψN,0 ) ≤ CN 2 , (ψN,0 , H where (., .) denotes the inner product on L2 (R3N , dx). Let ψN,t , for t ∈ [0, T ], be the solution of the Schr¨ odinger equation N,t i∂t ψN,t = Hψ
(22)
with initial data ψN,0 . Then, if α = (V 1 + V ∞ ) is small enough (of order one) and ν > 1 is large enough (recall that ν enters the deﬁnition of the norms (21)), we have: (k)
(i) ΓN,t = {γN,t }N k=1 has at least one (nontrivial) limit point Γ∞,t = (k)
{γ∞,t }k≥1 ∈ C([0, T ], B− ) with respect to the metric ρ. (k)
(ii)For any limit point Γ∞,t = {γ∞,t }k≥1 and for all k ≥ 1, there exists a constant C such that (k)
Tr (1 − ∆i )(1 − ∆j )γ∞,t ≤ C
(23)
for all i = j, t ∈ [0, T ]. (iii)Any limit point Γ∞,t satisﬁes the inﬁnite GrossPitaevskii hierarchy (15) when tested against a regular function J (k) (xk ; xk ): (k)
(k)
J (k) ,γ∞,t = J (k) , γ∞,0 − i
k j=1
− 8iπa0
k t j=1
ds
t
(k) ds J (k) , (−∆j + ∆j )γ∞,s
dxk dxk J (k) (xk ; xk )
(24) dxk+1
(k+1) × δ(xj − xk+1 ) − δ(xj − xk+1 ) γ∞,s (xk , xk+1 ; xk , xk+1 ) . Here we use the notation J (k) , γ (k) =
dxk dxk J (k) (xk ; xk ) γ (k) (xk ; xk ).
Remarks. (i) The main assumption of the theorem is the requirement that the expec 2 at t = 0 is of order N 2 . One can prove that this condition is tation of H satisﬁed for ψN,0 (x) = W (x)φN (x) and φN suﬃciently smooth (see [4], Lemma D1). Physically, this assumption guarantees that the initial wave function ψN,0 (x) has the shortscale structure characteristic of W (x) and, hence, that it describes, locally, a condensate.
290
B. Schlein
(ii) It is a priori not clear that the action of the deltafunctions in the GrossPitaevskii hierarchy (24) is well deﬁned. This fact follows by the bound (k) (23), which makes sure that γ∞,t is suﬃciently smooth. (iii) We also need to assume that α = (V ∞ + V 1 ) is small enough (but still of order one). This technical assumption is needed in the proof of the energy estimate, Proposition 1.
3 Sketch of the Proof In this section we explain some of the main ideas used in the proof of Theodinger equation (22) (with the orem 1. Let ψN,t be the solution of the Schr¨ We can decompose ψN,t as modiﬁed Hamiltonian H). ψN,t (x) = W (x)φN,t (x) , where W (x) is the approximation for the ground state wave function deﬁned in (20). This decomposition is always possible because W (x) is strictly positive. The main tool in the proof of Theorem 1 is an estimate for the L2 norm of the second derivatives of φN,t . This bound follows from the following energy estimate. Proposition 1. Assume a = a0 /N , 1 = N −2/3+κ and = N −2/5−κ for κ > 0 small enough, and suppose α = (V 1 + V ∞ ) is suﬃciently small. Then there exists a constant C > 0 such that N ˜ φ)(x)2 ≥ (C − o(1)) dx W 2 (x)∇i ∇j φ(x)2 dx (HW − o(1) N
i,j=1 N
2
2
dx W (x)∇i φ(x) + N
2
2
2
dx W (x)φ(x)
,
i=1
where o(1) → 0 as N → ∞. Remark. The proof of this proposition is the main technical diﬃculty in our analysis. It is in order to prove this proposition that we need to introduce the cutoﬀs Fij in the approximate ground state wave function W (x), and that we need to modify the Hamiltonian. 2 ψN,0 ) ≤ CN 2 , the conUsing the assumption that, at t = 0, (ψN,0 , H servation of the energy, and the symmetry with respect to permutations, we immediately get the following corollary. Corollary 1. Suppose the assumptions of Proposition 1 are satisﬁed. Suppose moreover that the initial data ψN,0 is symmetric with respect to permutations 2 ψN,0 ) ≤ CN 2 . Then there exists a constant C such that and (ψN,0 , H
Derivation of the GrossPitaevskii Hierarchy
291
W 2 (x)∇i ∇j φN,t (x)2 ≤ C
(25)
for all i = j, t and all N large enough. Remark. The bound (25) is not an estimate for the derivatives of the whole wave function ψN,t . The inequality dx ∇i ∇j ψ(x)2 < C (26) 2 ψ) ≤ CN 2 . In fact, in order for (ψ, H 2 ψ) to be is wrong, if ψ satisﬁes (ψ, H 2 of order N , the wave function ψ(x) needs to have the short scale structure characterizing W (x). This makes (26) impossible to hold true uniformly in N . Only after the singular part W (x) has been factorized out, we can prove bounds like (25) for the derivatives of the remainder. One of the consequences of our energy estimate, and one of the possible interpretation of our result, is that the separation between the singular part of the wave function (living on the scale 1/N ) and its regular part is preserved by the time evolution. Next we show how the important bound (25) can be used to prove Theorem 1. According to the decomposition ψN,t (x) = W (x)φN,t (x), we deﬁne, (k) for k = 1, . . . , N , the densities UN,t (xk ; xk ), for k = 1, . . . , N , to be, roughly, the kparticle marginal density corresponding to the wave function φN,t (the exact deﬁnition is a little bit more involved, see [4], Sect. 4). The estimate (25) for the second derivatives of φN,t translates into a bound for the densities (k) UN,t : (k)
Tr (1 − ∆i )(1 − ∆j )UN,t ≤ C
(27)
for all i, j ≤ N with i = j, for all t and for all N large enough. Moreover, we can show that, for ν > 1 large enough (recall that the parameter ν enters the deﬁnition of the norms (21)), the families UN,t = (k) {UN,t }N k=1 deﬁne an equicontinuous sequence in the space C([0, T ], B− ) (this follows from a careful analysis of the BBGKY hierarchy associated to the Schr¨ odinger equation (22); see [4], Sects 9.1 and 9.2 for more details). Applying standard results (ArzelaAscoli Theorem), it follows that the sequence (k) UN,t has at least one limit point, denoted U∞,t = {U∞,t }k≥1 , in the space C([0, T ], B− ). The bound (27) can then be passed to the limit N → ∞, and we obtain (k) Tr (1 − ∆i )(1 − ∆j )U∞,t ≤ C for all i = j and t ∈ [0, T ]. (k) Next we go back to the family ΓN,t = {γN,t }N k=1 . Clearly, the densities (k)
(k)
γN,t do not satisfy the estimate (27). In fact, γN,t still contains the short scale structure of W (x) (which, on the contrary, has been factorized out (k) from UN,t ), and thus cannot have the smoothness required by (27).
292
B. Schlein
It is nevertheless clear that the short scale structure of ΓN,t disappears when we consider the limit N → ∞ (in the weak sense speciﬁed by Theorem 1). In fact, one can prove the convergence of an appropriate subsequence of ΓN,t to the limit point U∞,t of UN,t . In other words one can show that limit points of ΓN,t , denoted by Γ∞,t , coincide with the limit points of UN,t . Therefore, even though ΓN,t , for ﬁnite N , does not satisﬁes the bound (27), (k) its limit points Γ∞,t = {γ∞,t }k≥1 do. For every k ≥ 1 we have (k)
Tr (1 − ∆i )(1 − ∆j )γ∞,t ≤ C
(28)
for all i = j and t ∈ [0, T ]. This proves part i) and ii) of Theorem 1 (the (1) nontriviality of the limit follows by showing that Tr γ∞,t = 1). The bound (28) can then be used to prove part iii) of Theorem 1, that is to prove that the family Γ∞,t satisﬁes the inﬁnite GrossPitaevskii hierarchy (24). In fact, (k) having control over the derivatives of γ∞,t allows us to prove the convergence of the potential to a deltafunction (that is, it allows us to make (13) rigorous). To this end we use the following lemma (see [4], Sect. 8). Lemma 1. Suppose δβ (x) = β −3 h(x/β), for some regular function h, with h(x) = 1. Then, for any 1 ≤ j ≤ k, and for any regular function J(xk ; xk ), we have dxk dxk dxk+1 J(xk ; xk )(δβ (xj − xk+1 ) − δ(xj − xk+1 )) × γ (k+1) (xk , xk+1 ; xk , xk+1 )
≤ CJ β Tr (1 − ∆j )(1 − ∆k+1 )γ (k+1) . Part (iii) of Theorem 1 can then be proven combining this lemma with the estimates (27) and (28) (see [4], Sect. 9.4, for more details).
References 1. R. Adami, C. Bardos, F. Golse and A. Teta: Towards a rigorous derivation of the cubic nonlinear Schr¨ odinger equation in dimension one. Asymptot. Anal. (2) 40 (2004), 93–108. 2. F.J. Dyson: Groundstate energy of a hardsphere gas. Phys. Rev. (1) 106 (1957), 20–26. 3. A. Elgart, L. Erd˝ os, B. Schlein, and H.T. Yau: GrossPitaevskii equation as the mean ﬁeld limit of weakly coupled bosons. Preprint mathph/0410038. To appear in Arch. Rat. Mech. Anal. 4. L. Erd˝ os, B. Schlein, and H.T. Yau: Derivation of the Gross–Pitaevskii hierarchy for the dynamics of a BoseEinstein condensate. Preprint mathph/0410005.
Derivation of the GrossPitaevskii Hierarchy
293
5. L. Erd˝ os and H.T. Yau: Derivation of the nonlinear Schr¨ odinger equation from a many body Coulomb system. Adv. Theor. Math. Phys. (6) 5 (2001), 1169– 1205. 6. J. Ginibre and G. Velo: The classical ﬁeld limit of scattering theory for nonrelativistic manyboson systems. I and II. Commun. Math. Phys. 66 (1979), 37–76 and 68 (1979), 45–68. 7. E.P. Gross: Structure of a quantized vortex in boson systems. Nuovo Cimento 20 (1961), 454–466. 8. E.P. Gross: Hydrodynamics of a superﬂuid condensate. J. Math. Phys. 4 (1963), 195–207. 9. K. Hepp: The classical limit for quantum mechanical correlation functions. Commun. Math. Phys. 35 (1974), 265–277. 10. E.H. Lieb and R. Seiringer: Proof of BoseEinstein Condensation for Dilute Trapped Gases. Phys. Rev. Lett. 88 (2002), 17040914. 11. E.H. Lieb, R. Seiringer, J.P. Solovej, and J. Yngvason: The QuantumMechanical ManyBody Problem: Bose Gas. Preprint mathph/0405004. 12. E.H. Lieb, R. Seiringer, J. Yngvason: Bosons in a Trap: A Rigorous Derivation of the GrossPitaevskii Energy Functional. Phys. Rev A 61 (2000), 043602. 13. E.H. Lieb and J. Yngvason: Ground State Energy of the low density Bose Gas. Phys. Rev. Lett. 80 (1998), 2504–2507. 14. L.P. Pitaevskii: Vortex lines in an imperfect Bose gas. Sov. Phys. JETP 13 (1961), 451–454. 15. H. Spohn: Kinetic Equations from Hamiltonian Dynamics. Rev. Mod. Phys. 52 no. 3 (1980), 569–615.
Towards a Microscopic Derivation of the Phonon Boltzmann Equation Herbert Spohn Zentrum Mathematik and Physik Department, TU M¨ unchen, 85747 Garching, Boltzmannstr. 3, Germany [emailprotected]
1 Introduction The thermal conductivity of insulating (dielectric) crystals is computed almost exclusively on the basis of the phonon Boltzmann equation. We refer to [1] for a discussion more complete than possible in this contribution. On the microscopic level the starting point is the BornOppenheimer approximation (see [2] for a modern version), which provides an eﬀective Hamiltonian for the slow motion of the nuclei. Since their deviation from the equilibrium position is small, one is led to a wave equation with a weak nonlinearity. As already emphasized by R. Peierls in his seminal work [3], physically it is of importance to retain the structure resulting from the atomic lattice, which forces the discrete wave equation. On the other hand, continuum wave equations with weak nonlinearity appear in the description of the waves in the upper ocean and in many other ﬁelds. This topic is referred to as weak turbulence. Again the theoretical treatment of such equations is based mostly on the phonon Boltzmann eqation, see e.g. [4]. In these applications one considers scales which are much larger than the atomistic scale, hence quantum eﬀects are negligible. For dielectric crystals, on the other side, quantum eﬀects are of importance at low temperatures. We refer to [1] and discuss here only the classical discrete wave equation with a small nonlinearity. If one considers crystals with a single nucleus per unit cell, then the displacement ﬁeld is a 3vector ﬁeld over the crystal lattice Γ . The nonlinearity results from the weakly nonquadratic interaction potentials between the nuclei. As we will see, the microscopic mechanism responsible for the validity of the Boltzmann equation can be understood already in case the displacement ﬁeld is declared to be scalar, the nonlinearity to be due to an onsite potential, and the lattice Γ = Z3 . This is the model I will discuss in my notes. As the title indicates there is no complete proof available for the validity of the phonon Boltzmann equation. The plan is to explain the kinetic scaling and to restate our conjecture in terms of the asymptotics of certain Feynman diagrams. H. Spohn: Towards a Microscopic Derivation of the Phonon Boltzmann Equation, Lect. Notes Phys. 690, 295–304 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
296
H. Spohn
2 Microscopic Model We consider the simple cubic crystal Z3 . The displacement ﬁeld is denoted by (1) qx ∈ R , x ∈ Z3 , with the canonically conjugate momenta px ∈ R ,
x ∈ Z3 .
(2)
We use units in which the mass of the nuclei is m = 1. The particles interact harmonically and are subject to an onsite potential, which is divided into a quadratic part and a nonquadratic correction. Thus the Hamiltonian of the system reads 1 1 2 px +ω02 qx2 + α(x−y)qx qy + V (qx ) = H0 + V (qx ) . H= 2 2 3 3 3 3 x∈Z
x,y∈Z
x∈Z
x∈Z
(3) The coupling constants have the properties α(x) = α(−x) ,
(4)
α(x) ≤ α0 e−γx
(5)
for suitable α0 , γ > 0, and
α(x) = 0 ,
(6)
x∈Z3
because of the invariance of the interaction between the nuclei under the translation qx qx + a. For the anharmonic onsite potential we set V (u) =
√ 1 3 ε λu + ε(λ2 /18ω02 )u4 , u ∈ R . 3
(7)
ε is the dimensionless scale parameter, eventually ε → 0. The quartic piece is added so to make sure that H ≥ 0. In the limit ε → 0 its contribution will vanish and for simplicity of notation we will omit it from the outset. Then the equations of motion are d qx (t) = px (t) , dt √ d px (t) = − α(y − x)qy (t) − ω02 qx (t) − ελqx (t)2 , dt 3
x ∈ Z3 . (8)
y∈Z
We will consider only ﬁnite energy solutions. In particular, it is assumed that px  → 0, qx  → 0 suﬃciently fast as x → ∞. In fact, later on there will
Towards a Microscopic Derivation of the Phonon Boltzmann Equation
297
be the need to impose random initial data, which again are assumed to be supported on ﬁnite energy conﬁgurations. In the kinetic limit the average energy will diverge as ε−3 . It is convenient to work in Fourier space. For f : Z3 → R we deﬁne fH(k) = e−i2πk·x fx , (9) x∈Z3
k ∈ T3 = [− 12 , 12 ]3 , with inverse fx =
dkei2πk·x fH(k) ,
(10)
T3
dk the 3dimensional Lebesgue measure. The dispersion relation for the harmonic part H0 is 1/2 H(k) ≥ ω0 > 0 , (11) ω(k) = ω02 + α since α H(k) > 0 for k = 0 because of the mechanical stability of the harmonic lattice with vanishing onsite potential. In Fourier space the equations of motion read ∂ qH(k, t) = pH(k, t) , ∂t ∂ pH(k, t) = −ω(k)2 qH(k, t) ∂t √ − ελ dk1 dk2 δ(k − k1 − k2 )H q (k1 , t)H q (k2 , t)
(12)
T6
with k ∈ T3 . Here δ is the δfunction on the unit torus, to say, δ(k ) carries a point mass whenever k ∈ Z3 . It will be convenient to concatenate qx and px into a single complexvalued ﬁeld. We set 1 1 √ (13) ωH q (k) + i √ pH(k) a(k) = √ ω 2 with the inverse 1 1 1 √ qH(k) = √ √ a(k) + a(−k)∗ , pH(k) = √ i ω − a(k) + a(−k)∗ . (14) ω 2 2 To have a concise notation, we introduce a(k, +) = a(k)∗ , a(k, −) = a(k) .
(15)
Then the aﬁeld evolves as
√ ∂ a(k, σ, t) = iσω(k)a(k, σ, t) + iσ ελ dk1 dk2 ∂t 6 σ ,σ =±1 T 1
−1/2
(8ω(k)ω(k1 )ω(k2 ))
2
δ(−σk + σ1 k1 + σ2 k2 )a(k1 , σ1 , t)a(k2 , σ2 , t) . (16)
298
H. Spohn
3 Kinetic Limit and Boltzmann Equation The kinetic limit deals with a special class of initial probability measures. Their displacement ﬁeld has a support of linear size ε−1 and average energy of order ε−3 . More speciﬁcally, these probability measures have the property of being locally Gaussian and almost stationary under the dynamics. Because of the assumed slow variation in space the covariance of such probability measures changes only slowly, i.e. on the scale ε−1 , in time. Let us assume then that the initial data for (16) are random and speciﬁed by a Gaussian probability measure on phase space. It is assumed to have mean (17)
a(k, σ)G ε =0, and for the covariance we set
a(k, σ)a(k , σ)G ε =0,
W ε (y, k) = ε3 (T/ε)3
dηei2πy·η a(k − εη/2, +)a(k + εη/2, −)G ε ,
(18) (19)
y ∈ (εZ)3 , which deﬁnes the Wigner function rescaled to the lattice (εZ)3 . Local stationarity is ensured by the condition lim W ε (,rε , k) = W 0 (r, k) ,
ε→0
(20)
where ,rε denotes integer part modulo ε. Note that W ε is normalized as ε dkW (y, k) = dk a(k, +)a(k, −)G (21) ε . y∈(εZ)3
T3
T3
The condition that the limit in (20) exists thus implies that the average phonon number increases as ε−3 , equivalently the average total energy increases as 4G 3 −3 d3 kω(k)a(k, +)a(k, −) = H0 G ). (22) = O(ε T3
ε
Let ·t be the timeevolved measure at time t. Its rescaled Wigner function is dηei2πy·η a(k − εη/2, +)a(k + εη/2, −)t/ε . (23) W ε (y, k, t) = ε3 (T/ε)3
Kinetic theory claims that lim W ε (,rε , k, t) = W (r, k, t) ,
ε→0
where W (r, k, t) is the solution of the phonon Boltzmann equation
(24)
Towards a Microscopic Derivation of the Phonon Boltzmann Equation
299
∂ 1 W (r, k, t) + ∇ω(k) · ∇r W (r, k, t) ∂t 2π π = λ2 dk1 dk2 (ω(k)ω(k1 )ω(k2 ))−1 δ(ω(k) + σ1 ω(k1 ) + σ2 ω(k2 )) 2 σ ,σ =±1 T6 1 2 δ(k + σ1 k1 + σ2 k2 ) W (r, k1 , t)W (r, k2 , t) +σ1 W (r, k, t)W (r, k2 , t) + σ2 W (r, k, t)W (r, k1 , t) (25) to be solved with the initial condition W (r, k, 0) = W 0 (r, k). The free streaming part is an immediate consequence of the evolution of W as generated √ by H0 . The strength of the cubic nonlinearity was assumed to be of order ε, which results in an eﬀect of order 1 on the kinetic time scale. The speciﬁc form of the collision operator will be explained in the following section. It can be brought into a more familiar form by performing the sum over σ1 , σ2 . Then the collision operator has two terms. The ﬁrst one describes the merging of two phonons with wave number k and k1 into a phonon with wave number k2 = k + k1 , while the second term describes the splitting of a phonon with wave number k into two phonons with wave numbers k1 and k2 , k = k1 + k2 . In such a collision process energy is conserved and wave number is conserved modulo an integer vector. In (25) the summand with σ1 = 1 = σ2 vanishes trivially. However it could be the case that the condition for energy conservation, ω(k) + ω(k ) − ω(k + k ) = 0 ,
(26)
has also no solution. If so, the collision operator vanishes. In fact, for nearest neighbor coupling only, α(0) = 6, α(e) = −1 for e = 1, α(x) = 0 otherwise, it can be shown that (26) has no solution whenever ω0 > 0. To have a nonzero collision term we have to require dk dk δ(ω(k) + ω(k ) − ω(k + k )) > 0 , (27) which is an implicit condition on the couplings α(x). A general condition to ensure (27) is not known. A simple example where (27) can be checked by hand is 3 (1 − cos(2πk α )) , k = (k 1 , k 2 , k 3 ) . (28) ω(k) = ω0 + α=1
It corresponds to suitable nearest and next nearest neighbor couplings. There is a second more technical condition which requires that sup dk δ(ω(k) + ω(k ) − ω(k + k )) = c0 < ∞ . (29) k
It holds for the dispersion relation (28). This uniform bound allows for a simple proof that the Boltzmann equation has a unique solution for short times provided W 0 (r, k) is bounded.
300
H. Spohn
4 Feynman Diagrams Denoting by ·t the average with respect to the measure at time t (in microscopic units), the starting point of the timedependent perturbation series is the identity M M n N NG n n a(kj , σj ) = exp it σj ω(kj ) a(kj , σj ) j=1
t
√ +i ε
t
n
j=1
n
σj ω(kj )
j=1
=1 σ ,σ =±1
ds exp i(t − s)
M
n
j=1 j=
Here
j=1
σ
T6
dk dk φ(k , k , k )δ(−σ k + σ k + σ k )
N .
a(kj , σj ) a(k , σ )a(k , σ )
(30)
s
φ(k, k , k ) = λ(8ω(k)ω(k )ω(k ))−1/2
(31)
One starts with n = 2 and σ1 = 1, σ2 = −1. Then on the right hand side of (30) there is the product of three a’s. One resubstitutes (30) with n = 3, etc. Thereby one generates an inﬁnite series, in which only the average over the initial Gaussian measure ·G appears. To keep the presentation transparent, let me assume that ·G is a translation invariant Gaussian measure with
a(k, ±)G = 0 , a(k, σ)a(k , σ)G = 0 ,
a(k, +)a(k , −)G = δ(k − k )W (k) .
(32)
Then the measure at time t is again translation invariant. Kinetic scaling now merely amounts to considering the long times t/ε. The Wigner function at that time is then represented by the inﬁnite series ∞ ε
a(q, −)a(p, +)t/ε = δ(q − p) W (q) + Wn (q, t) . (33) n=1
The inﬁnite sum is only formal. Taking naively the absolute value at iteration 2n one ﬁnds that Wnε (q, t) ≤ εn (t/ε)2n ((2n)!)−1 (2n)!c2n ((2n + 2)!/2n+1 (n + 1)!) . (34) √ Here εn = ( ε)2n , (t/ε)2n /(2n)! comes from the time integration, (2n)! from the sum over in (30), c2n from the kintegrations and the initial W (k), and
Towards a Microscopic Derivation of the Phonon Boltzmann Equation
301
the factor (2n + 2)!/2n+1 (n + 1)! from the Gaussian pairings in the initial measure. Thus even at ﬁxed ε there are too many terms in the sum. Since no better estimate is available at present, we concentrate on the structure of a single summand Wnε (q, t). δ(q −p)Wnε (q, t) is a sum of integrals. The summation comes from – the sum over σ , σ in (30) – the sum over in (30) – the sum over all pairings resulting from the average with respect to the initial Gaussian measure ·G . Since each single integral has a rather complicated structure, it is convenient to visualize them as Feynman diagrams. A Feynman diagram is a graph with labels. Let us ﬁrst explain the graph. The graph consists of two binary trees. It is convenient to draw them on a “backbone” consisting of 2n + 2 equidistant horizontal level lines which are labelled from 0 (bottom) to 2n + 1 (top). The two roots of the tree are two vertical bonds from line 2n + 1 to level line 2n. At level m there is exactly one branch point with two branches in either tree. Thus there are exactly 2n branch points. At level 0 there are then 2n + 2 branches. They are connected according to the pairing rule, see ﬁgure below. In the Feynman graph each bond is oriented with arrows pointing either up (σ = +1) or down (σ = −1). The left root is down while the right root is up. If there is no branching the orientation is inherited from the upper level. At a pairing the orientation must be maintained. Thus at level 0 a branch with an up arrow can be paired only with a branch with a down arrow, see (32). Every internal line in the graph must terminate at either end by a branch point. Every such internal line admits precisely two orientations.
q
5
? 2 1 6 3 ?4
4 3 2 1 0
p
7
8 6 ?
6
6 6 5 6 ? ?
t t4 t3 t2 t1 0
302
H. Spohn
Next we insert the labels. The level lines 0 to 2n + 1 are labelled by times 0 < t1 . . . < t2n < t. The left root carries the label q while the right root carries the label p. Each internal line is labelled with a wave number k. To each Feynman diagram one associates an integral through the following steps. (i) The time integration is over the simplex 0 ≤ t1 . . . ≤ t2n ≤ t as dt1 . . . dt2n . (ii) The wave number integration is over all internal lines as dk1 . . . dkκ , where κ = 3n − 1 is the number of internal lines. (iii) One sums over all orientations of the internal lines. The integrand is a product of three factors. (iv) There is a product over all branch points. At each branchpoint there is a root, say wave vector k1 and orientation σ1 , and there are two branches, say wave vectors k2 , k3 and orientations σ2 , σ3 . Then each branch point carries the weight (35) δ(−σ1 k1 + σ2 k2 + σ3 k3 )σ1 φ(k1 , k2 , k3 ) . If one regards the wave vector k as a current with orientation σ, then (35) expresses Kirchhoﬀ’s rule for conservation of the current. (v) By construction each bond carries a time diﬀerence tm+1 − tm , a wave vector k, and an orientation σ. Then to this bond one associates the phase factor (36) exp[i(tm+1 − tm )σω(k)/ε] . The second factor is the product of such phase factors over all bonds. (vi) The third factor of the integrand is given by n+1
W (kj ) ,
(37)
j=1
where k1 , . . . , kn+1 are the wave numbers of the bonds between level 0 and level 1. (vii) Finally there is the prefactor (−1)n ε−n . To illustrate these rules we give an example for n = 2, see ﬁgure above. The associated integral is given by, more transparently keeping the δfunctions from the pairings, −2 ε dt1 . . . dt4 dk1 . . . dk8 0≤t1 ≤...≤t4 ≤t
T24
δ(q + k1 − k2 )δ(k2 + k3 + k4 )δ(−p − k5 − k6 )δ(−k1 + k7 − k8 ) φ(q, k1 , k2 )φ(k2 , k3 , k4 )φ(p, k5 , k6 )φ(k1 , k7 , k8 ) δ(k7 − k6 )W (k7 )δ(k8 − k3 )W (k8 )δ(k4 − k5 )W (k4 ) % exp {i(t − t4 )(−ωq + ωp ) + i(t4 − t3 )(ω1 − ω2 + ωp ) +i(t3 − t2 )(ω1 + ω3 + ω4 + ωp ) + i(t2 − t1 )(ω1 + ω3 + ω4 − ω5 − ω6 ) & +it1 (ω7 − ω8 + ω3 + ω4 − ω5 − ω6 )}/ε (38) with ωq = ω(q), ωp = ω(p), ωj = ω(kj ).
Towards a Microscopic Derivation of the Phonon Boltzmann Equation
303
δ(q − p)Wnε (q, t) is the sum over all Feynman diagrams with 2n + 2 levels and thus is a sum of oscillatory integrals. In the limit ε → 0 only a few leading terms survive while all remainders vanish. E.g., the Feynman diagram above is subleading. In fact, the conjecture of kinetic theory can be stated rather concisely: Kinetic Conjecture: In a leading Feynman diagram the Kirchhoﬀ rule never forces an identiﬁcation of the form δ(kj ) with some wave vector kj . In addition, the sum of the 2(n − m + 1) phases from the bonds between level lines 2m and 2m+1 vanishes for every choice of internal wave numbers. This cancellation must hold for m = 0, . . . , n. Since we assumed the initial data to be spatially hom*ogeneous, the phonon Boltzmann equation (25) simpliﬁes to ∂ W (k, t) ∂t = 4πλ2 σ1 ,σ2 =±1
T6
dk1 dk2 φ(k, k1 , k2 )2 δ(ω(k) + σ1 ω(k1 ) + σ2 ω(k2 ))
δ(k + σ1 k1 + σ2 k2 ) W (k1 , t)W (k2 , t) + 2σ2 W (k, t)W (k1 , t) ,
(39)
where we used the symmetry with respect to (k1 , σ1 ) and (k2 , σ2 ). To (39) we associate the Boltzmann hierarchy ∂ fn = Cn,n+1 fn+1 , n = 1, 2, . . . , ∂t
(40)
acting on the symmetric functions fn (k1 , . . . , kn ) with Cn,n+1 fn+1 (k1 , . . . , kn ) = 4πλ
2
n
=1 σ ,σ =±1 ω(k ))δ(k +
T6
dk dk φ(k , k , k )2
δ(ω(k ) + σ ω(k ) + σ σ k + σ k ) [fn+1 (k1 , . . . , k , k+1 , . . . , k ) + 2σ fn+1 (k1 , . . . , kn , k )] .
(41)
Under the condition (29) and provided W ∞ < ∞, the hierarchy (40) has a unique solution for short times. In case fn (k1 , . . . , kn , 0) =
n
W (kj ) ,
(42)
j=1
the factorization is maintained in time and each factor agrees with the solution of the Boltzmann equation (39). From (40) one easily constructs the perturbative solution to (39) with the result
304
H. Spohn
W (k, t) = W (k) +
∞ 1 n t (C1,2 . . . Cn,n+1 W ⊗n+1 )(k) n! n=1
= W (k) +
∞
Wn (k, t) .
(43)
n=1
The series in (43) converges for t suﬃciently small. For n = 1, 2 the oscillating integrals can be handled by direct inspection with the expected results limε→0 W1ε (k, t) = W1 (k, t), limε→0 W2ε (k, t) = W2 (k, t). If the leading terms are as claimed in the Kinetic Conjecture, then they agree with the series (43). The complete argument is a somewhat tricky counting of diagrams, which would lead us too far astray. Thus the most immediate project is to establish that all subleading diagrams vanish in the limit ε → 0. This would be a step further when compared to the investigations [5], [6]. Of course a complete proof must deal with the uniform convergence of the series in (33).
Acknowledgements I thank Jani Lukkarinen for instructive discussions and Gianluca Panati for a careful reading.
References 1. H. Spohn, The Phonon Boltzmann Equation, Properties and Link to Weakly Anharmonic Lattice Dynamics, preprint. 2. S. Teufel, Adiabatic Perturbation Theory in Quantum Dynamics, Lecture Notes in Mathematics 1821, SpringerVerlag, Berlin, Heidelberg, New York 2003. 3. R.E. Peierls, Zur kinetischen Theorie der W¨ armeleitung in Kristallen, Annalen Physik 3, 1055–1101 (1929). 4. V.E. Zakharov, V.S. L’vov, and G. Falkovich, Kolmogorov Spectra of Turbulence: I Wave Turbulence. Springer, Berlin 1992. 5. D. Benedetto, F. Castella, R. Esposito, and M. Pulvirenti, Some considerations on the derivation of the nonlinear quantum Boltzmann equation, J. Stat. Phys. 116, 381–410 (2004). 6. L. Erd¨ os, M. Salmhofer, and H.T. Yau, On the quantum Boltzmann equation, J. Stat. Phys. 116, 367–380 (2004).
On the Quantization of Hall Currents in Presence of Disorder JeanMichel Combes1 , Fran¸cois Germinet2 , and Peter D. Hislop3 1
2
3
D´epartement de Math´ematiques, Universit´e de ToulonVar, BP 132, 83957 La Garde C´edex, France [emailprotected] D´epartement de Math´ematiques, Universit´e de CergyPontoise, Site de SaintMartin, 2 avenue Adolphe Chauvin, 95302 CergyPontoise C´edex, France [emailprotected] Department of Mathematics, University of Kentucky, Lexington, KY 405060027, USA [emailprotected]
Abstract. We review recent results of two of the authors concerning the quantization of Hall currents, in particular a general quantization formula for the diﬀerence of edge Hall conductances in semiinﬁnite samples with and without a conﬁning wall. We then study the case where the Fermi energy is located in a region of localized states and discuss new regularizations. We also sketch the proof of localization for 2Dmodels with constant magnetic ﬁeld with random potential located in a halfplane in two diﬀerent situations: (1) with a zero potential in the other half plane and for energies away from the Landau levels and (2) with a conﬁning potential in the other half plane and on an interval of energies that covers an arbitrary number of Landau levels.
1 The Edge Conductance and General Invariance Principles Quickly after the discovery of the integer quantum Hall eﬀect (IQHE) by von Klitzing et. al. [33], then Halperin [28] put the accent on the crucial role of quantum currents ﬂowing at the edges of the (ﬁnite) sample. Such edge currents, carried by edge states, should be quantized, and the quantization should agree with the one of the transverse (Hall) conductance. While edge currents have been widely studied in the physics literature since the early eighties, e.g. [1, 11, 20, 35, 40] (see also [34, 37] and references therein), it is only recently that a mathematical understanding of the existence of such edge currents has been obtained [9,10,13,17–19,21]. The study of the quantization of the edge Hall conductance at a mathematical level is even more recent [3, 15, 16, 31, 32, 38]. We consider the simplest model for quantum devices exhibiting the IQHE. This consists of an electron conﬁned to the 2dimensional plane considered as the union of two complementary semiinﬁnite regions supporting potentials V1 and V2 , respectively, and under the inﬂuence of a constant magnetic ﬁeld J.M. Combes et al.: On the Quantization of Hall Currents in Presence of Disorder, Lect. Notes Phys. 690, 307–323 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
308
J.M. Combes et al.
B orthogonal to the sample. In the absence of potentials V1 and V2 , the free electron is described by the free Landau Hamiltonian HL = p2x + (py − Bx)2 . The spectrum of HL consists of the wellknown Landau levels BN = (2N − 1)B, N ≥ 1, with the convention B0 = −∞. To introduce the halfplane potentials, we let 1− and 1+ be the characteristic functions of, respectively, {x ≤ 0} and {x > 0}. Then, if V1 , V2 are two potentials bounded from below and in the Kato class [12], the Hamiltonian of the system is given, in suitable units and Landau gauge, by H(V1 , V2 ) := HL + V1 1− + V2 1+ ,
(1)
as a selfadjoint operator acting on L2 (R2 , dxdy), where HL = H(0, 0) in this notation. For technical reasons it is convenient to assume that V1 , respectively V2 , does not grow faster than polynomially as x → −∞, respectively, as x → +∞. We shall say that V1 is a left conﬁning potential with respect to the interval I = [a, b] ⊂ R if, in addition to the previous conditions, the following holds: There exists R > 0, s.t. ∀x ≤ −R, ∀y ∈ R, V1 (x, y) > b .
(2)
The “hard wall” case, i.e. V1 = +∞ and H = HL + V2 acting on L2 (R+ × R, dxdy) with Dirichlet boundary conditions, can be considered as well. As typical examples for H(V1 , V2 ) one may think of the right potential V2 as an impurity potential and the left potential V1 as either a wall, conﬁning the electron to the right halfplane and generating an edge current near x = 0, or as the zero potential. In this latter case, the issue is to determine whether or not V2 is strong enough to create edge currents by itself. We will discuss this in Sect. 3.1. Another example is the strip geometry, where both V1 and V2 are conﬁning potentials outside of a strip x ∈ [−R, R], where the electron is localized. We deﬁne a “switch” function as a smooth real valued increasing function equal to 1 (resp. 0) at the right (resp. left) of some bounded interval. Following [3, 15, 31, 38], we deﬁne the (Hall) edge conductance as follows. Deﬁnition 1. Let X ∈ C ∞ (R2 ) be a xtranslation invariant switch function with suppX ⊂ R × [− 14 , 14 ], and let −g ∈ C ∞ (R) be switch a function with supp g ⊂ I = [a, b] a compact interval. The edge conductance of H(V1 , V2 ) in the interval I is deﬁned as σe (g, V1 , V2 ) = −tr(g (H(V1 , V2 ))i[HL , X ])
(3)
whenever the trace is ﬁnite (we shall also use the notation σe (g, H) = σe (g, V1 , V2 ) if H = H(V1 , V2 )). Note that in the situations of interest σe (g, V1 , V2 ) will turn out to be independent of the particular shape of the switch function X and also of the switch function g, provided supp g does not contain any Landau level.
On the Quantization of Hall Currents in Presence of Disorder
309
We turn to the description of the results of [3]. Let us assume that I lies in between two successive Landau levels, say the N th and the (N + 1)th . While clearly σe (g, 0, 0) = 0, for any g as above since g (HL ) = 0, a straightforward computation shows that σe (g, V1 , 0) = N , provided V1 (x, y) = V1 (x) is such that limx→−∞ V1 (x) > b (see, for example, [3, Proposition 1]). The ﬁrst result tells us that the edge conductance is stable under a perturbation by a potential W located in a strip [L1 , L2 ] × R of ﬁnite width. Theorem 1. ( [3, Theorem 1]) Let H = H(V1 , V2 ) be as in (1), and let W be a bounded potential supported in a strip [L1 , L2 ] × R, with −∞ < L1 < L2 < +∞. Then the operator (g (H + W ) − g (H))i[HL , X ] is trace class, and tr((g (H + W ) − g (H))i[HL , X ]) = 0 .
(4)
As a consequence: (i) σe (g, HL + W ) = 0. (ii) Assume V1 is a yinvariant potential, i.e. V1 (x, y) = V1 (x), that is left conﬁning with respect to I ⊃ supp g . If I ⊂]BN , BN +1 [, for some N ≥ 0, then (5) σe (g, HL + V1 + W ) = N . We note that Theorem 1 extends perturbations W that decay polynomially fast in the xdirection. In particular, it allows for more general conﬁning potentials than yinvariant ones. But, it is easy to see that Theorem 1 does not hold for all perturbations in the x direction. For example, if supp g ⊂]BN , BN +1 [, then σe (g, 0, 0) = 0, so for if W (x, y) = ν0 1[0,] (x), with ν0 > BN +1 and 0 < < ∞, then σe (g, 0, W ) = 0. On the other hand, a simple calculation shows (e.g. [3]) that for W∞ = ν0 1[0,∞[ (x), we have σe (g, 0, W∞ ) = −N . However, one has the following invariance principle, which is a consequence of a more general sum rule given in [3, Theorem 2]. Theorem 2. ( [3, Corollary 3]) Let g be s.t. supp g ⊂]BN , BN +1 [, for some N ≥ 0. Let V1 be a yinvariant left conﬁning potential with respect to suppg . Then the operator {g (H(V1 , V2 )) − g (H(0, V2 ))}i[HL , X ] is trace class and − tr({g (H(V1 , V2 )) − g (H(0, V2 ))}i[HL , X ]) = N .
(6)
In particular, if either σe (g, V1 , V2 ) or σe (g, 0, V2 ) is ﬁnite, then both are ﬁnite, and (7) σe (g, V1 , V2 ) − σe (g, 0, V2 ) = N . An immediate, but important, consequence of Theorem 2, if V2 ∞ < B, then σe (g, V1 , V2 ) = N , whenever I ⊂]BN + V2 ∞ , BN +1 − V2 ∞ [, recovering a recent result of Kellendonk and SchulzBaldes [32]. In general, neither term in (6) is separately trace class but a meaning can be given to each term through an appropriate regularization. Various regularizations of the edge conductance were discussed in [3], and two others are
310
J.M. Combes et al.
presented in Sect. 2. It is proved in [3] that the regularized edge conductance σereg (g, V0 , V ) satisﬁes a sum rule similar to (7): σereg (g, V0 , V ) = N + σereg (g, 0, V ) .
(8)
With reference to this, note that σereg (g, 0, V ) = 0 would imply the existence of current carrying states solely due to the impurity potential. Since (8) would yield σe (g, V1 , V2 ) = N , we see that such “edge currents without edges” are responsible for the deviation of the Hall conductance from its ideal value N . Typically this is expected to happen in a regime of strong disorder (with respect to the magnetic strength B). As an example of this phenomenon, a model studied by S. Nakamura and J. Bellissard [36] is revisited in [3] and it is shown that in this case σe (g, V0 , V2 ) = 0 and thus σe (g, 0, V2 ) = −N . In Sect. 3.2, we present another example for which localization in the strong disorder regime implies that σereg (0, V ) is quantized so that there are edge currents without edges. As a counterpart, in the weak disorder regime (i.e. weak impurities in the region x1 ≥ 0 and no electric potential in the left halfplane), one expects that no current will ﬂow near the region x1 = 0. After a regularizing procedure, we argue in Sect. 3.1 that this is exactly what happens with the model studied in [5,25,39] if I lies in a region of the spectrum where localization has been shown. Notation: Throughout this note 1X = 1(x,y) will denote the characteristic function of a unit cube centered X = (x, y) ∈ Z2 . If A is a subset of R2 , then 1A will denote the characteristic function of this set. We recall that 1− and 1+ stand, respectively, for 1x≤0 and 1x>0 .
2 Regularizing the Edge Conductance in Presence of Impurities 2.1 Generalities Let V2 = V be a potential located in the region x ≥ 0. If the operator H(0, V ) has a spectral gap and if the interval I falls into this gap, then the edge conductance is quantized as mentioned above. However such a situation is not physically relevant, since the quantization of the Hall conductance can only be related to the quantum Hall eﬀect in presence of impurities that create the wellknown “plateaus” [2, 34]. If I falls into a region of localized states of H(0, V ), then the conductances may not be welldeﬁned, and a regularization is needed. In this section, we brieﬂy recall the regularization procedure described in [3], and we then propose new candidates. Assume supp g ⊂ I ⊂]BN , BN +1 [. Let (JR )R>0 be a family of operators s.t. C1. JR = 1 and limR→∞ JR ψ = ψ, for all ψ ∈ EH(0,V ) (I)L2 (R2 ).
On the Quantization of Hall Currents in Presence of Disorder
311
C2. JR regularizes H(0, V ) in the sense that g (H(0, V ))i[HL , X ]JR is trace class for all R > 0, and limR→∞ tr(g (H(0, V ))i[HL , X ]JR ) exists and is ﬁnite. Then if V1 = V0 is a yinvariant left conﬁning potential with respect to I, it follows from Theorem 2 that lim −tr ({g (H(V0 , V )) − g (H(0, V ))}i[HL , X ]JR ) = N .
R→∞
In other terms, if C1 and C2 hold, then JR also regularizes H(V0 , V ). Deﬁning the regularized edge conductance by σereg (g, V1 , V2 ) := − lim tr(g (H(V1 , V2 ))i[HL , X ]JR ) , R→∞
(9)
whenever the limit exists, we get the analog of Theorem 2: σereg (g, V0 , V ) = N + σereg (g, 0, V ) .
(10)
In particular, if we can show that σereg (g, 0, V ) = 0, for instance, under some localization property, then the edge quantization for H(V0 , V ) follows: σereg (g, V0 , V ) = − lim tr(g (H(V0 , V ))i[HL , X ]JR ) = N . R→∞
Let us now consider Hω = H(0, Vω,+ ) = HL + Vω,+ ,
Vω,+ =
ωi u(x − i) ,
(11)
(12)
i∈Z+∗ ×Z
a random Schr¨ odinger operator modeling impurities located on the positive halfplane (the (ωi )i are i.i.d. (independent, identically distributed) random variables, and u is a smooth bump function). If Hω has pure point spectrum in I for Pa.e. ω, then denoting by (ϕω,n )n≥1 a basis of orthonormal eigenfunctions of Hω with energies Eω,n ∈ supp g ⊂ I, one has, whenever the regularization holds, σereg (g, 0, Vω,+ ) = − lim g (Eω,n ) ϕω,n , i[Hω , X ]JR ϕω,n . (13) R→∞
n
If JR = 1x≤R , the limit (13) actually exists [3, Propostion 2], but it is very likely that it will not be zero, even under strong localization properties of the eigenfunctions such as (SULE) (see [14]) or (SUDEC) (see Deﬁnition 2 below and [26]). We refer to [16] for a concrete example. This can be understood as follows: Because the cutoﬀ JR (even a smooth version of it) cuts classical orbits living near x = R, it will create spurious contributions to the total current, and the latter will no longer be zero. The quantum counter part of this picture is that although the expectation of i[Hω , X ] in an eigenstate of Hω is zero by the Virial Theorem, this is no longer true if this commutator is multiplied by JR . Of course, the sum in (13) is zero if JR commutes with Hω , as in [3, Theorem 3]. In the next two sections we investigate new regularizations that commute with Hω only asymptotically (as R → ∞).
312
J.M. Combes et al.
2.2 A Time Averaged Regularization for a Dynamically Localized System We assume that the operator H = H(0, V ) exhibits dynamical localization in an open interval I ∈]BN , BN +1 [. This means that for any p ≥ 0, there exists a nonnegative constant Cp < ∞ such that for any Borel function f on I, with f  ≤ 1, and for any X1 , X2 ∈ R2 , sup 1X1 f (H)e−itH 1X2 2 ≤ Cp min(1, X1 − X2 )−p ) .
(14)
t∈R
We used the HilbertSchmidt norm. For random Schr¨ odinger operators Hω , this assumption is one of the standard conclusions of multiscale analysis [23, 24]. We show in Sect. 3.1 that as long as I ∈]BN , BN +1 [ such an analysis applies to the Hamiltonian Hω = H(0, Vω ) as in (12). Note that it follows from (14) that if X ∈ R2 and A is a subset of R2 (A may contain X) then, for any p > 0, there exists a (new) constant 0 ≤ Cp < ∞, such that sup 1A f (H)e−itH 1X 2 ≤ Cp min(1, dist({X}, A)−p ) .
(15)
t∈R
For R < ∞, η > 0 and γ > 0, we set, with H = H(0, V ) and X = (x, y), ∞ JR = η EH (I)eitH 1x≤R e−itH EH (I)e−ηt dt, with R = η −γ . (16) 0
Theorem 3. Let JR as in (16), with γ ∈]0, 1[. Assume that H(0, V ) exhibits dynamical localization (i.e. (14)) in I ⊂]BN , BN +1 [ for some N ≥ 0. Then JR regularizes H(0, V ), and thus also H(V0 , V ), in the sense that C1 and C2 hold. Moreover the edge conductances take the quantized values: σereg (g, 0, V ) = 0 and σereg (g, V0 , V ) = N . Remark 1. In [16], a similar regularization is considered, where γ = 1 and H is the bulk Hamiltonian H(V, V ). We also note that if R and η are independent variables, then one recovers the regularization [3, (7.13)], see [3, Remark 13]. Proof. Since JR ≤ 1, to get C1 it is enough to check limR→∞ JR EH (I)ψEH (I)ψ for compactly supported states, and it is thus enough to note that by (15), ∞ sup 1x>R e−itH EH (I)10 e−ηt dt ≤ Cp R−p . (17) (1 − JR )EH (I)10 ≤ η 0
t
We turn to C2. As in [3], we write i[H, X ] = i[H, X ]1y≤ 12 , with 1y≤ 12 = " x2 ∈Z 1(x2 ,0) . Note that, by hypothesis on I, g (H) = g (H) − g (HL ), so that terms that are far in the left half plane will give small contributions. To see this, we develop
On the Quantization of Hall Currents in Presence of Disorder
8 8 8(g (H) − g (HL ))i[HL , X ]EH (I)e−itH 1x≤−R 8 1
313
(18)
using the HelﬀerSj¨ ostrand formula [29, 30] and the resolvent identity with RL (z) = (HL − z)−1 and R(z) = (H − z)−1 . It is thus enough to control terms of the form, with Imz = 0, x1 , y1 , x2 ∈ Z, X1 = (x1 , y1 ), 8 8 8 8 (19) 8R(z)V 1x≥0 RL (z)i[HL , X ]1y≤ 12 EH (I)e−itH 1x≤−R 8 ≤ 1
x1 ≥0,y1 ,x2
× R(z)V 1X1 2 1X1 RL (z)i[HL , X ]1(x2 ,0) 1(x2 ,0) EH (I)e−itH 1x≤−R 2 But, as a wellknown fact, R(z)1X1 is HilbertSchmidt in dimension 2, e.g. [25, Lemma A.4], and its HilbertSchmidt norm is bounded by C[dist(z, σ(H))]−1 ≤ Cz −1 , uniformly in X1 . Then, for some κ ≥ 1, (19) ≤
C (z)κ +
≤
1X1 RL (z)i[HL , X ]1(x2 ,0)
(20)
x1 ≥0,y1 ,x2 ≤− R 2
C (z)κ
1(x2 ,0) EH (I)e−itH 1x≤−R 2
(21)
x2 ≥− R 2
C R−p , (z)κ
(22)
where the latter follows from (15) and from the fact that 1X1 RL (z) i[HL , X ]1(x2 ,0) decays faster than any polynomial in X1 − X2 , as can be seen by a CombesThomas estimate together with standard computations (e.g. [3, Lemma 3]). As a consequence the trace norm (18) is ﬁnite and goes to zero as R → ∞, uniformly in η. The next step is to control contributions coming from terms living far from the support of X , i.e. terms s.t. y ≥ Rν with ν ∈]0, 1]. Set S(R, ν) = {(x, y) ∈ R2 , x ≤ R, y ≥ Rν }. Then, using (15), 8 8 8g (H)i[HL , X ]EH (I)e−itH 1S(R,ν) 8 (23) 1 8 8 8 8 −itH 8g (H)i[HL , X ]1(x ,0) 8 81(x ,0) EH (I)e ≤ 1S(R,ν) 82 (24) 2 2 2 x2 ∈Z
≤C
min(1, dist({(x2 , 0)}, S(R, ν))−p ) ≤ C(2R)(Rν )−p .
(25)
x2 ∈Z
Contributions from (23) are thus negligible as R → ∞, uniformly in η. Letting KR,ν denote the compact set KR,ν = {(x, y) ∈ R2 , x ≤ R, y ≤ Rν } , we are left so far with the evaluation of ∞ η e−ηt dt g (H)i[HL , X ]EH (I)eitH 1KR,ν e−itH EH (I) , 0
(26)
314
J.M. Combes et al.
which is clearly now a trace class operator (the integral is absolutely convergent in trace norm). In other terms g (H)i[HL , X ]JR is thus trace class. It remains to show that its trace goes to zero as R goes to inﬁnity. But on the account of (18) and (23), it remains to show that the trace of (26) goes to zero. By cyclicity, ∞ . e−ηt tr g (H)X e−itH i[HL , EH (I)1KR,ν EH (I)]eitH dt tr(26) = −η ∞0 . d =η e−ηt tr g (H)X e−itH EH (I)1KR,ν EH (I)eitH dt dt 0  . = ηtr g (H)X EH (I)1KR,ν EH (I) ∞ . e−ηt tr g (H)X e−itH EH (I)1KR,ν EH (I)eitH EH (I) dt. −η 2 0
Thus tr(26) ≤ CηKR,ν  = CηR1+ν . Since R = η −γ , the trace goes to zero 1 . if γ < 1+ν 2.3 Regularization Under a Stronger form of Dynamical Localization In this section, we consider JR = EH(0,V ) (I)1x≤R EH(0,V ) (I) .
(27)
Note that the regularization (16) studied in Sect. (2.2) is the time average of (27). The eﬀect of the time averaging is to provide a control on the cross terms arising in (27) if one expands EH(0,V ) (I) over a basis of eigenfunctions. In [3, (7.13)], cross terms were suppressed from the very deﬁnition of JR . By showing that JR , given in (27), regularizes H(0, Vω ) under, basically, the same assumption as in [3, Theorem 3], we strengthen [3]’s result. Let T be the multiplication operator by T (X) = Xν , ν > d2 = 1, with 1
X = (1 + X2 ) 2 , for X ∈ R2 . It is well known for Schr¨ odinger operators that tr(T −1 EH(0,V ) (I)T −1 ) < ∞, if I is compact (e.g. [25]). Deﬁnition 2 (SUDEC). Assume H has pure point spectrum in I with eigenvalues En and corresponding normalized eigenfunctions ϕn , listed with multiplicities. We say that H has Summable Uniform Decay of Eigenfunction Correlations (SUDEC) in I, if there exist ζ ∈]0, 1[ and a ﬁnite constant c0 > 0 such that for any En ∈ I and X1 , X2 ∈ Z2 , 1X1 ϕn 1X2 ϕn ≤ c0 αn T 1X1 2 T 1X2 2 e−X1 −X2  , ζ
(28)
where αn = T −1 ϕn 2 . Note that,
n
αn = tr(T −1 EH(0,V ) (I)T −1 ) < ∞ .
(29)
On the Quantization of Hall Currents in Presence of Disorder
315
Remark 2. Property (28) (or a modiﬁed version of it) was called (WULE) in [3] and was introduced in [22]. The more accurate acronym (SUDEC) comes from [26] and Property (SUDEC) is used in [27] as a very natural signature of localization in order to get the quantization of the bulk conductance. Theorem 4. Assume that H(0, V ) has (SUDEC) in I ⊂]BN , BN +1 [ for some N ≥ 0. Then JR , given in (27), regularizes H(0, V ), and thus also H(V0 , V ), in the sense that C1 and C2 hold. Moreover the edge conductances take the quantized values: σereg (g, 0, V ) = 0 and σereg (g, V0 , V ) = N . Proof. That the operator g (H(0, V ))i[H(0, V ), X ]JR is trace class follows from the comparison g (H(0, V )) = g (H(0, V )) − g (HL ). In order to control the region x ≤ 0, and the immediate estimate, let Pn be the eigenprojector on the eigenfunction ϕn , and write 1X Pn 1Y 2 = 1X ϕn 1Y ϕn (30) 1X EH(0,V ) (I)1Y 2 ≤ n
≤ c0
αn
n
T 1X 2 T 1Y 2 e−X−Y  , ζ
(31)
n
where we used the assumption (28) (and recall (29)). We proceed and set Λ2,R = 1x≤R . We are looking at (reg)
σE
(g, R) = tr(g (H(0, V ))i[H(0, V ), X ]JR ) .
(32)
The operator being trace class, we expand the trace in the basis of eigenfunctions of H(0, V ) in the interval I and get (reg) σE (g, R) = g (En )(En − Em ) ϕn , X ϕm ϕm , Λ2,R ϕn (33) n=m ≤ C(g, I)  ϕn , X ϕm  ϕm , Λ2,R ϕn  . (34) n=m
It remains to show that the double sum in (34) is convergent (i.e. the trace is absolutely convergent). If it is so, then we can interchange the limit in R and the double sum to get zero due to the orthogonality of the eigenfunctions. In full generality, dynamical localization is not enough to show that (34) is absolutely convergent. This is the case if H = H(0, 0) = HL and I contains a Landau level. The sum (34) diverges, even though HL exhibits dynamical localization. But if one has (SULE) or (SUDEC), then the sum converges absolutely. We have, writing Λ2 instead of Λ2,R :
316
J.M. Combes et al.
(34)  ϕn , X ϕm   ϕm , Λ2 ϕn  ≤
(35) (36)
n=m
=
1
1
1
1
 ϕn , X ϕm  2  ϕn , (1 − X )ϕm  2  ϕm , Λ2 ϕn  2  ϕm , (1 − Λ2 )ϕn  2
n=m
≤
√ √ √ √ X ϕn X ϕm 1 − X ϕn 1 − X ϕm
(37)
n=m
12
× Λ2 ϕn Λ2 ϕm 1 − Λ2 ϕn 1 − Λ2 ϕm 12 √
√ X ϕn 1 − X ϕn Λ2 ϕn 1 − Λ2 ϕn ≤
(38) (39)
n
×
12 √
√ X ϕm 1 − X ϕm Λ2 ϕm 1 − Λ2 ϕm
(40)
m
It remains to show that (28) implies 12 √
√ X ϕn 1 − X ϕn Λ2 ϕn 1 − Λ2 ϕn b, if I = [0, b] is the interval where we would like to prove localization. As in Sect. 3.1, the modiﬁcations of the Wegner estimate, of the starting estimates of the the multiscale analysis (MSA), and of the MSA itself, have to be checked separately. While the comments made in Sect. 3.1 concerning the MSA are still valid, the new geometry requires new speciﬁc arguments for the Wegner estimate and the starting estimate. The Wegner estimate: Its proof can no longer be borrowed from [5] as in Sect. 3.1, and one has to explicitly take into account the eﬀect of the conﬁning potential V0 1− . We shall modify the argument given in [8] as follows. The only case we have to discuss is the one of a box ΛL (X), with x < L2 so that it overlaps both types of potentials. We set V˜L = ωi u(X − i) . i∈ΛL (X)∩Z2
Let HL denote the restriction of H(V0 , Vω ) to the box ΛL (X) with selfadjoint boundary conditions (e.g. [27]). By Chebychev’s inequality, the proof of the Wegner estimate is reduced to an upper bound on the expectation of the trace of the spectral projector EHL (I) for the interval I. Following [7], we write trEHL (I) = tr1ΛL (X) EHL (I) 1 λ ˜ ≤ trV0 1ΛL (X) 1− EHL (I) + trVL EHL (I) V0 C0 1 λ ˜ ≤ tr1− HL EHL (I) + trVL EHL (I) V0 C0 b λ ˜ ≤ trEHL (I) + trVL EHL (I) , V0 C0 so that, with V0 > b by assumption, −1 λ b trV˜L EHL (I) . trEHL (I) ≤ 1− C0 V0
(51) (52) (53) (54)
(55)
At this point, the proof follows the usual strategy, as in [6–8]. The starting estimate: The initial estimate follows from the analysis, at large disorder, given in [25, Section 3]. Since in the left halfplane, the potential is already very high (V0 > b), it is enough to estimate the probability
On the Quantization of Hall Currents in Presence of Disorder
321
that all the random variables ωi in the right part of the box is higher than say b/2. Doing this creates a gap in the spectrum of the ﬁnite volume operator HL . This spectral gap, occurring with good probability, can be used to obtain the exponential decay of the (ﬁnite volume) resolvent thanks to a CombesThomas argument.
References 1. Avron, J.E., Seiler, R.: Quantization of the Hall conductance for general, multiparticle Schr¨ odinger operators, Phys. Rev. Lett. 54 259–262 (1985) 2. Bellissard, J., van Elst, A., SchulzBaldes, H.: The non commutative geometry of the quantum Hall eﬀect. J. Math. Phys. 35, 5373–5451 (1994) 3. Combes, J.M., Germinet, F.: Edge and Impurity Eﬀects on Quantization of Hall Currents, Commun. Math. Phys. 256, 159–180 (2005) 4. Combes, J.M., Germinet, F., Hislop, P.D: in preparation. 5. Combes, J.M., Hislop, P.D.: Landau Hamiltonians with random potentials: localization and the density of states Commun. Math. Phys. 177, 603–629 (1996) 6. Combes, J.M., Hislop, P.D., Klopp, F. Nakamura, S.: The Wegner estimate and the integrated density of states for some random operators. Spectral and inverse spectral theory (Goa, 2000), Proc. Indian Acad. Sci. Math. Sci. 112, 31–53 (2002) 7. Combes, J.M., Hislop, P.D., Klopp, F.: H¨ older continuity of the integrated density of states for some random operators at all energies. IMRN 4, 179–209 (2003) 8. Combes, J.M., Hislop, P.D., Nakamura, S.: The Lp theory of the spectral shift function, the Wegner estimate, and the integrated density of states for some random operators Commun. Math. Phys. 218, 113–130 (2001) 9. Combes, J.M., Hislop, P.D., Soccorsi, E.: Edge states for quantum Hall Hamiltonians. Mathematical results in quantum mechanics (Taxco, 2001), 69–81, Contemp. Math., 307, Amer. Math. Soc., Providence, RI, 2002 10. Combes, J.M., Hislop, P.D., Soccorsi, E.: in preparation 11. Cresti, A., Fardrioni, R., Grosso, G., Parravicini, G.P.: Current distribution and conductance quantization in the integer quantum Hall regime, J. Phys. Condens. Matter 15, L377–L383 (2003) 12. Cycon, H.L., Froese, R.G., Kirsch, W., Simon, B.: Schr¨ odinger operators. Heidelberg: SpringerVerlag, 1987 13. De Bi`evre, S., Pul´e, J.: Propagating Edge States for a Magnetic Hamiltonian. Math. Phys. Elec. J. vol. 5, paper 3 14. Del Rio, R., Jitomirskaya, S., Last, Y., Simon, B.: Operators with singular continuous spectrum. IV. Hausdorﬀ dimensions, rank one perturbations, and localization. J. Anal. Math. 69, 153–200 (1996) 15. Elbau, P., Graf., G.M.: Equality of Bulk and Edge Hall Conductance Revisited. Commun. Math. Phys. 229, 415–432 (2002) 16. Elgart, A., Graf, G.M., Schenker, J.: Equality of the bulk and edge Hall conductances in a mobility gap. Comm. Math. Phys. 259, 185–221 (2005) 17. Exner, P., Joye, A., Kovarik, H.: Edge currents in the absence of edges, Physics Letters A, 264, 124–130 (1999)
322
J.M. Combes et al.
18. Exner, P., Joye, A., Kovarik, H.: Magnetic transport in a straight parabolic channel, J. of Physics A: Math. Gen., 34, 9733–9752 (2001) 19. Ferrari, C., Macris, N.: Intermixture of extended edge and localized bulk levels in macroscopic Hall systems. J. Phys. A: Math. Gen. 35, 6339–6358 (2002) 20. Fleischmann, R., Geisel, T,. Ketzmerick, R., Phys. Rev. Lett. 68, 1367 (1992). 21. Fr¨ ohlich, J., Graf, G.M., Walcher, J.: On the extended nature of edge states of quantum Hall Hamiltonians. Ann. H. Poincar´e 1, 405–444 (2000) 22. Germinet, F.: Dynamical localization II with an Application to the Almost Mathieu Operator. J. Stat. Phys. 95, 273–286 (1999) 23. Germinet, F., De Bi`evre, S.: Dynamical localization for discrete and continuous random Schr¨ odinger operators. Commun. Math. Phys. 194, 323–341 (1998) 24. Germinet, F., Klein, A.: Bootstrap Multiscale Analysis and Localization in Random Media. Commun. Math. Phys. 222, 415–448 (2001) 25. Germinet, F, Klein, A.: Explicit ﬁnite volume criteria for localization in continuous random media and applications. Geom. Funct. Anal. 13, 1201–1238 (2003) 26. Germinet, F, Klein, A.: New characterizations of the region of dynamical localization for random Schr¨ odinger operators, to appear in J. Stat. Phys. 27. Germinet, F, Klein, A., Schenker, J.: Dynamical delocalization in random Landau Hamiltonians, to appear in Annals of Math. 28. Halperin, B.I.: Quantized Hall conductance, current carrying edge states and the existence of extended states in a twodimensional disordered potential. Phys. Rev. B 25, 2185–2190 (1982) ´ 29. Helﬀer, B., Sj¨ ostrand, J.: Equation de Schr¨ odinger avec champ magn´etique et ´equation de Harper, in Schr¨ odinger operators, H. Holden and A. Jensen eds., LNP 345, 118–197 (1989) 30. Hunziker W., Sigal, I.M.: Timedependent scattering theory for N body quantum systems. Rev. Math. Phys. 12, 1033–1084 (2000) 31. Kellendonk, J., Richter, T., SchulzBaldes, H.: Edge Current channels and Chern numbers in the integer quantum Hall eﬀect. Rev. Math. Phys. 14, 87–119 (2002) 32. Kellendonk, T., SchulzBaldes, H.: Quantization of Edge Currents for continuous magnetic operators. J. Funct. Anal. 209, 388–413 (2004) 33. von Klitzing, K, Dorda, G, Pepper, N.: New method for highaccuracy determination of the ﬁne structure constant based on quantized Hall resistance, Phys. Rev. Lett 45, 494 (1980). 34. von Klitzing, 25 years of QHE, in S´eminaire Henri Poincar´e (2004) 35. Mac Donald, A.H., Streda, P.: Quantized Hall eﬀect and edge currents. Phys. Rev. B 29, 1616–1619 (1984) 36. Nakamura, S., Bellissard, J.: Low Energy Bands do not Contribute to Quantum Hall Eﬀect. Commun. Math. Phys. 131, 283–305 (1990) 37. Prange, Girvin, The Quantum Hall Eﬀect, Graduate texts in contemporary Physics, SpringerVerlag, N.Y. 1987 38. SchulzBaldes, H., Kellendonk, J., Richter, T.: Simultaneous quantization of edge and bulk Hall conductivity. J. Phys. A 33, L27–L32 (2000)
On the Quantization of Hall Currents in Presence of Disorder
323
39. Wang, W.M.: Microlocalization, percolation, and Anderson localization for the magnetic Schr¨ odinger operator with a random potential. J. Funct. Anal. 146, 1–26 (1997) 40. Zozoulenko, I.V., Maao, F.A., Hauge, E.H., Phys. Rev. B53, 7975 (1996).
Equality of the Bulk and Edge Hall Conductances in 2D A. Elgart Department of Mathematics, Stanford University, Stanford, CA 943052125 [emailprotected]
1 Introduction and Main Result Von Klitzing [15] observed that a two dimensional electron gas at very low temperatures and strong magnetic ﬁeld displays a quantization of the Hall conductance, that is the conductance measured in the direction transversal to the applied current. Speciﬁcally, the conductance plotted as a function of the magnetic ﬁeld shows extremely ﬂat plateaux at integer multiples of e2 /h (e is a charge of electron and h is Planck’s constant). Two pictures were introduced for a description of the Quantum Hall Eﬀect: “Edge currents picture” and “Bulk currents picture”. The edge current picture suggests that the Hall current ﬂows in the narrow regions along the sample boundaries (we will denote the corresponding conductance by σE ), so that the Hall voltage drops entirely in these regions. On the other hand, the description in terms of bulk currents suggests that the Hall voltage drops gradually across the sample (and let σB denote the Hall conductance associated with this regime). It was proposed by Halperin [13] that in reality one should expect an intermix of these two pictures, and that σE = σB . In [14], σE,B were linked for a Harper’s model with rational ﬂux. In more general setup, the equality of the edge and the bulk conductances was recently rigorously established [10,16,18], provided that there is a spectral gap ∆ at Fermi energy of the singleparticle (bulk) Hamiltonian HB . The aim of this article is to report on a recent joint result with G. M. Graf and J. H. Schenker [11], and to expose some of the ideas involved. There we proved this equality in the more general setting in which HB exhibits Anderson localization in ∆ – more precisely, strong dynamical localization (see (2) below). The result applies to Schr¨ odinger operators which are random, but does not depend on that property. We therefore formulate the result for deterministic operators. A full account of this work with detailed proofs of all the results stated below is provided in [11]. The bulk system is described by the lattice Z2 x = (x1 , x2 ) with ∗ on 2 (Z2 ). We assume its matrix elements HB (x, x ), Hamiltonian HB = HB 2 x, x ∈ Z , to be of short range, namely HB (x, x ) (eµx−x  − 1) =: C1 < ∞ (1) sup x∈Z2
x ∈Z2
A. Elgart: Equality of the Bulk and Edge Hall Conductances in 2D, Lect. Notes Phys. 690, 325–332 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
326
A. Elgart
for some µ > 0, where x = x1  + x2 . Our hypothesis on the bounded open interval ∆ ⊂ R is that for some ν ≥ 0 sup g(HB )(x, x ) (1 + x)−ν eµx−x  =: C2 < ∞ (2) g∈B1 (∆)
x,x ∈Z2
where B1 (∆) denotes the set of Borel measurable functions g which are constant in {λλ < ∆} and in {λλ > ∆} with g(x) ≤ 1 for every x. In particular C2 is a bound when g is of the form gt (λ) = e−itλ E∆ (λ) and the supremum is over t ∈ R, which is a statement of dynamical localization. By the RAGE theorem this implies that the spectrum of HB is pure point in ∆. We denote the corresponding eigenprojections by E{λ} (HB ) for λ ∈ E∆ , the set of eigenvalues λ ∈ ∆. We assume that no eigenvalue in E∆ is inﬁnitely degenerate, (3) dim E{λ} (HB ) < ∞ , λ ∈ E∆ . The validity of these assumptions is discussed below. The zero temperature bulk Hall conductance at Fermi energy λ is deﬁned by the KuboStˇreda formula [5] σB (λ) = −i tr Pλ [ [Pλ , Λ1 ] , [Pλ , Λ2 ] ] ,
(4)
where Pλ = E(−∞,λ) (HB ) and Λi (x) is the characteristic function of . x = (x1 , x2 ) ∈ Z2  xi < 0 . Under the above assumptions σB (λ) is welldeﬁned for λ ∈ ∆, and shows a plateau (see [11] for further details). We remark that (3) is essential for a plateau: for the Landau Hamiltonian (though deﬁned on the continuum rather than on the lattice) (1, 2) hold if properly interpreted, but (3) fails in an interval containing a Landau level, where indeed σB (λ) jumps. The edge system is described as a halfplane Z × Za , where Za = {n ∈ Z  n ≥ −a}, with the height −a of the edge eventually tending to −∞. The Hamiltonian Ha = Ha∗ on 2 (Z × Za ) is obtained by restriction of HB under some largely arbitrary boundary condition. More precisely, we assume that (5) Ea = Ja Ha − HB Ja : 2 (Z × Za ) → 2 (Z2 ) satisﬁes sup x∈Z2
Ea (x, x ) eµ(x2 +a+x1 −x1 ) ≤ C3 < ∞ ,
(6)
x ∈Z×Za
where Ja : 2 (Z × Za ) → 2 (Z2 ) denotes extension by 0. For instance with Dirichlet boundary conditions, Ha = Ja∗ HB Ja , we have Ea = (Ja Ja∗ − 1)HB Ja ,
Equality of the Bulk and Edge Hall Conductances in 2D
i.e.,
' −HB (x, x ) , Ea (x, x ) = 0,
327
x2 < −a , x2 ≥ −a ,
whence (6) follows from (1). We remark that (1) is inherited by Ha with a constant C1 that is uniform in a, but not so for (2) as a rule. The deﬁnition of the edge Hall conductance requires some preparation. The current operator across the line x1 = 0 is −i [Ha , Λ1 ]. Matters are simpler if we temporarily assume that ∆ is a gap for HB , i.e., if σ(HB ) ∩ ∆ = ∅, in which case one may set [18] σE := −i tr ρ (Ha ) [Ha , Λ1 ] , where ρ ∈ C ∞ (R) satisﬁes
' 1, λ∆.
(7)
(8)
The operator in (7) is trace class essentially because i [H, Λ1 ] is relevant only on the states near x1 = 0, and ρ (Ha ) only near the edge x2 = −a, so that the intersection of the two strips is compact (see Sect. 2 for more details). In the situation (2) considered in this paper the operator appearing in (7) is not trace class, since the bulk operator may have spectrum in ∆, which can cause the above stated property to fail for ρ (Ha ). In search of a proper deﬁnition of σE , we consider only the current ﬂowing across the line x1 = 0 within a ﬁnite window −a ≤ x2 < 0 next to the edge. This amounts to modifying the current operator to be −
i i (Λ2 [Ha , Λ1 ] + [Ha , Λ1 ] Λ2 ) = − { [Ha , Λ1 ] , Λ2 } , 2 2
(9)
with which one may be tempted to use i lim − tr ρ (Ha ) { [Ha , Λ1 ] , Λ2 } a→∞ 2
(10)
as a deﬁnition for σE . Though we show that this limit exists, it is not the physically correct choice. In particular, the edge conductance should be nonzero on average for the Harper Hamiltonian with an i.i.d. random potential [11]. The basic fact that the net current of a bound eigenstate ψλ of HB is zero, (11) − i (ψλ , [HB , Λ1 ] ψλ ) = 0 , can be preserved by the regularization provided the spatial cutoﬀ Λ2 is time averaged. In fact, let 1 T iHa t −iHa t e Xe dt (12) AT,a (X) = T 0
328
A. Elgart
be the time average over [0, T ] of a (bounded) operator X with respect to the Heisenberg evolution generated by Ha , with a ﬁnite or a = B. If a limit Λ∞ 2 = limT →∞ AT,B (Λ2 ) were to exist, it would commute with HB so that i − (ψλ , { [HB , Λ1 ] , Λ∞ 2 } ψλ ) = 0 . 2 This motivates our deﬁnition, i 1 σE := lim lim − tr ρ (Ha ) { [Ha , Λ1 ] , AT,a (Λ2 )} . T →∞ a→∞ 2
(13)
We have the following result [11]. Theorem 1. Under the assumptions (1, 2, 3, 6, 8) the limit in (13) exists, and 1 σE = σB . In particular (13) depend neither on the choice of ρ nor on that of Ea . Remark 1. (i) The hypotheses (1, 2) hold almost surely for ergodic Schr¨odinger operators whose Green’s function G(x, x ; z) = (HB − z)−1 (x, x ) satisﬁes a moment condition [3] of the form s (14) sup lim sup E G(x, x ; E + iη) ≤ Ce−µx−x  E∈∆
η→0
for some s < 1. It implies the dynamical localization bound E
sup
g(HB )(x, x )
≤ Ce−µx−x  ,
(15)
g∈B1 (∆)
although (2) has also been obtained by diﬀerent means, e.g., [12]. The implication (14) ⇒ (15) was proved in [1] (see also [2, 4, 9]). The bound (15) may be better known for supp g ⊂ ∆, but is true as stated since it also holds [2, 6] for the projections g(HB ) = Pλ , Pλ⊥ = 1 − Pλ , (λ ∈ ∆). (ii) Condition (3), in fact simple spectrum, follows from the arguments in [19], at least for operators with nearest neighbor hopping, HB (x, y) = 0 if x − y > 1. (iii) When σ(HB ) ∩ ∆ = ∅, the operator appearing in (7) is known to be 1 coincides with σE deﬁned in (7). trace class. In this case, the conductance σE This statement follows from Theorem 1 and the known equality σE = σB [10, 18], but can also be proven directly. (iv) A recent paper of Combes and Germinet [7] contains results which are topically related to but substantially diﬀerent from the ones presented here. The detailed comparison can be found in [11]. In the second part of the paper we will outline the simple proof of equality σB = σE , with σE deﬁned in (7), when ∆ is a spectral gap for HB . Even though this result was established before by a number of authors, the new proof highlights some of the ideas used in our generalization of this result, namely in [11].
Equality of the Bulk and Edge Hall Conductances in 2D
329
2 Proof of σB = σE 2.1 Some Preliminaries We are going to use the HelﬀerSj¨ostrand representation for function ρ(Ha ): 1 ρ(Ha ) = dm(z)∂z¯ρ(z)Ra (z) (16a) 2π 1 dm(z)∂z¯ρ(z)Ra2 (z) , ρ (Ha ) = − (16b) 2π with Ra (z) = (Ha − z)−1 . The integral is over z = x + iy ∈ C with measure dm(z) = dxdy, ∂z¯ = ∂x + i∂y , and ρ(z) is a quasianalytic extension of ρ(x) which for given n can be chosen so that
dm(z) ∂z¯ρ(z) y−p−1 ≤ C
n+2 8
8 8 (k) 8 8ρ 8
k=0
k−p−1
(17)
k for p = 1, ..., n, provided the appearing norms f k = dx(1 + x2 ) 2 f (x) are ﬁnite. This is the case for ρ with ρ ∈ C0∞ (R). For p = 1 this shows that (16b) is norm convergent. The integral (16a), which would correspond to the case p = 0, is nevertheless a strongly convergent improper integral, see e.g., (A.12) of [10]. In particular, we obtain 1 dm(z)∂z¯ρ(z)Ra (z) [Ha , Λ1 ] Ra (z) [ρ(Ha ), Λ1 ] = − (18a) 2π 1 (18b) ρ (Ha ) [Ha , Λ1 ] = − dm(z)∂z¯ρ(z)Ra2 (z) [Ha , Λ1 ] . 2π A further preliminary is the CombesThomas bound [8] 8 8 8 δ(x) 8 Ra (z)e−δ(x) 8 ≤ 8e
C , Imz
(19)
where δ can be chosen as δ −1 = C 1 + Imz−1
(20)
for some (large) C > 0 and (x) is any Lipschitz function on Z2 with (x) − (y) ≤ x − y .
(21)
tr AB = tr BA if AB , BA ∈ I1 .
(22)
We will also use
330
A. Elgart
2.2 Convergence and Trace Class Properties Let RB (z) := (HB − z)−1 . Note that Ja Ra (z)Ja∗ − RB (z) = − (Ja Ra (z)Ea∗ + 1 − Ja Ja∗ ) RB (z) −−−→ 0 (23) s
a→∞
s
s
a→∞
a→∞
because Ea∗ −−−→ 0 by (6) and because 1 − Ja Ja∗ −−−→ 0 is the projection onto states supported in {x2 < −a}. This implies [17, Theorem VIII.20] slim Ja f (Ha )Ja∗ = f (HB ) a→∞
(24)
for any bounded continuous function f . Moreover, using CombesThomas bound (19) and Ja∗ Ja = 1, together with Holmgren’s bound one can infer that1 Lemma 1. For any a, a [Ha , Λ1 ] Λ2 ∈ I1 , ρ (Ha ) [Ha , Λ1 ] ∈ I1 , [ρ(HB ), Λ1 ] [ρ(HB ), Λ1 ] ∈ I1 ; (25) tr ρ (Ha ) [Ha , Λ1 ] = tr ρ (Ha ) [Ha , Λ1 ] = lim tr ρ (Ha ) [Ha , Λ1 ] Λ2 . a→∞
(26) 2.3 Edge – Bulk Interpolation We can now use (18a), (18b) and cyclicity of the trace to rewrite the expression appearing in the last equation as 1 tr ρ (Ha ) [Ha , Λ1 ] Λ2 = − dm(z)∂z¯ρ(z) tr Ra (z) [Ha , Λ1 ] Λ2 Ra (z) 2π 1 = tr [ρ(Ha ), Λ1 ] Λ2 − dm(z)∂z¯ρ(z) tr R(z) 2π [Ha , Λ1 ] Ra (z) [Ha , Λ2 ] Ra (z) . (27) The ﬁrst contribution, computed in the position basis, is seen to vanish. As for the second contribution, one can use (23) and CombesThomas estimate to obtain lim tr Ra (z) [Ha , Λ1 ] Ra (z) [Ha , Λ2 ] Ra (z)
a→∞
= tr RB (z) [HB , Λ1 ] RB (z) [HB , Λ2 ] RB (z) , (28) for Imz = 0. We deduce that σE = −i tr ρ (Ha ) [Ha , Λ1 ] i dm(z)∂z¯ρ(z) tr RB (z) [HB , Λ1 ] RB (z) [HB , Λ2 ] RB (z) . (29) = 2π 1
See [11] for details.
Equality of the Bulk and Edge Hall Conductances in 2D
331
2.4 σE = σB To ﬁnish this end, i 2π i = 2π i + 2π
the proof we need to reduce the rhs of (29) into the form of (4). To we write dm(z)∂z¯ρ(z) tr RB (z) [HB , Λ1 ] RB (z) [HB , Λ2 ] RB (z) dm(z)∂z¯ρ(z) tr Pλ RB (z) [HB , Λ1 ] RB (z) [HB , Λ2 ] RB (z)Pλ dm(z)∂z¯ρ(z) tr P¯λ RB (z) [HB , Λ1 ] RB (z) [HB , Λ2 ] RB (z)P¯λ , (30)
with P¯λ := 1 − Pλ . The ﬁrst term is equal to i dm(z)∂z¯ρ(z) tr Pλ RB (z) [HB , Λ1 ] RB (z) [HB , Λ2 ] RB (z)Pλ 2π i = dm(z)∂z¯ρ(z) tr (Pλ Λ1 RB (z) [HB , Λ2 ] RB (z)Pλ 2π + Pλ RB (z)Λ1 [HB , Λ2 ] RB (z)Pλ ) i dm(z)∂z¯ρ(z) tr Pλ Λ1 [RB (z), Λ2 ] Pλ =− 2π = −i tr Pλ Λ1 [ρ(HB ), Λ2 ] Pλ % & = i tr Pλ Λ1 P¯λ , Λ2 Pλ = i tr Pλ Λ1 P¯λ Λ2 Pλ , (31) where the term on the third line vanishes by integration by parts since Pλ RB (z) = RB (z)Pλ is analytic on the support of 1 − ρ(z). Similar consideration shows that i dm(z)∂z¯ρ(z) tr P¯λ RB (z) [HB , Λ1 ] RB (z) [HB , Λ2 ] RB (z)P¯λ 2π = −i tr P¯λ Λ1 [Pλ , Λ2 ] P¯λ = −i tr P¯λ Λ1 Pλ Λ2 P¯λ . (32) Since Pλ Λ1,2 P¯λ = Pλ [ρ(HB ), Λ1,2 ] P¯λ , one can deduce from (25) that both contribution that appear in (31) and (32) are trace class, and one can use cyclicity of the trace and Pλ2 = Pλ to conclude that i 2π
dm(z)∂z¯ρ(z) tr RB (z) [HB , Λ1 ] RB (z) [HB , Λ2 ] RB (z) = i tr Pλ Λ1 P¯λ Λ2 Pλ − i tr P¯λ Λ1 Pλ Λ2 P¯λ = i tr Pλ Λ1 P¯λ Λ2 Pλ − i tr Pλ Λ2 P¯λ Λ2 Pλ = σB (λ) . (33)
332
A. Elgart
References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19.
M. Aizenman: Rev. Math. Phys. 6, 1163 (1994). M. Aizenman and G. M. Graf: J. Phys. A 31, 6783 (1998). M. Aizenman and S. Molchanov: Commun. Math. Phys. 157, 245 (1993). M. Aizenman, J. H. Schenker, R. M. Friedrich, and D. Hundertmark: Commun. Math. Phys. 224, 219 (2001). J.E. Avron, R. Seiler, and B. Simon: Comm. Math. Phys. 159, 399 (1994). J. Bellissard, A. van Elst, and H. SchulzBaldes: J. Math. Phys. 35, 5373 (1994). J.M. Combes and F. Germinet: Comm. Math. Phys., to appear. J.M. Combes and L. Thomas: Comm. Math. Phys. 34, 251 (1973). R. del Rio, S. Jitomirskaya, Y. Last, and B. Simon: J. Anal. Math. 69, 153 (1996). P. Elbau and G.M. Graf: Comm. Math. Phys. 229, 415 (2002). A. Elgart, G.M. Graf and J.H. Schenker: Comm. Math. Phys., to appear. F. Germinet and S. De Bi`evre: Comm. Math. Phys. 194, 323 (1998). B.I. Halperin: Phys. Rev. B 25, 2185 (1982). Y. Hatsugai: Phys. Rev. L. 71, 3697 (1993). K.v. Klitzing, G. Dorda, and M. Pepper: Phys. Rev. Lett. 45, 494 (1980). N. Macris: preprint, 2003. M. Reed and B. Simon: Methods of modern mathematical physics. I, 2nd edn (Academic Press Inc., New York 1980). H. SchulzBaldes, J. Kellendonk, and T. Richter: J. Phys. A 33, L27 (2000). B. Simon: Rev. Math. Phys. 6, 1183 (1994).
Generic Subsets in Spaces of Measures and Singular Continuous Spectrum Daniel Lenz and Peter Stollmann Fakult¨ at f¨ ur Mathematik, Technische Universit¨ at, 09107 Chemnitz, Germany [emailprotected] d.lenz@@mathematik.tuchemnitz.de Abstract. We discuss recent results of ours showing that geometric disorder leads to some purely singularly continuous spectrum generically. This is based on a slight extension of Simons Wonderland theorem. Our approach to this theorem relies on the study of generic subsets of certain spaces of measures. In this article, we elaborate on this purely measure theoretic basis of our approach.
Key words: Random Schr¨ odinger operators, Delone sets, spaces of measures, 81Q10,35J10,82B44, 28A33,28C15
1 Introduction In this article we review our work [8] and elaborate on certain measure theoretic parts of it. The starting point of [8] is the study of operators Hω := −∆ + v(· − x) x∈ω
on L2 (Rd ), for suitable functions v and certain uniformly discrete subsets ω of Rd called Delone sets. Such operators arise in the study of disordered solids. More precisely, they can be thought to model geometric disorder. It is shown that these operators have a purely singularly continuous spectral component generically. Here, generic refers to a topology on the set of all Delone sets (see below for details). The abstract operator theoretic tool behind our reasoning is a slight strengthening of a result due to B. Simon called the “Wonderland theorem” [11]. Our method of proving this is diﬀerent from Simons. It consists of two steps. We ﬁrst prove that certain subsets of spaces of measures are generic. This generalizes the corresponding results of Simon [11] and Zamﬁrescu [13] for R to rather general measure spaces. In the second step, the Wonderland theorem follows by considering spectral measures as continuous maps from the space of selfadjoint operators to measures on the real line. Given our version of the Wonderland theorem, our result on generic singularly continuous spectrum follows from geometric considerations. There, we approximate arbitrary ω by essentially periodic ones in various ways. D. Lenz and P. Stollmann: Generic Subsets in Spaces of Measures and Singular Continuous Spectrum, Lect. Notes Phys. 690, 333–341 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
334
D. Lenz and P. Stollmann
In the present article, we particularly focus on the measure theoretic side of things. This is discussed in the next two sections. In particular, Sect. 3 elaborates on the last part of Sect. 2 of [8] and presents a generalization of Corollary 2.8 given there. The subsequent discussion of the Wonderland theorem in Sect. 4 and the application to geometric disorder in Sect. 5 then follows [8] quite closely.
2 Generic Subsets in Spaces of Measures We will be concerned with subsets of the set of positive, regular Borel measures M+ (S) on some locally compact, σcompact, separable metric space S. The closed ball around x ∈ S with radius r is denoted by Br (x). The space M+ (S) is endowed with the weak topology from Cc (S), also called the vague topology. We refer the reader to [3] for standard results concerning the space of measures. We will use in particular that the vague topology is metrizable such that M+ (S) becomes a complete metric space. Thus, the Baire category theorem becomes applicable. For the application to spectral theory, S is just an open subset U of the real line. We call a measure µ ∈ M+ (S) continuous if its atomic or pure point part vanishes, i.e. if µ({x}) = 0 for every x ∈ S. (We prefer the former terminology in the abstract framework and the latter for measures on the real line.) The set of all continuous measures on S is denoted by Mc (S). A measure µ is called a point measure if there exists a countable set Y in X with µ(X \ Y ) = 0. The set of all point measures on S is denoted by Mp (S). This set is dense in M+ (S). Two measures are said to be mutually singular, µ ⊥ ν, if there exists a set C ⊂ S such that µ(C) = 0 = ν(S \ C). The set of all measures on S which are singular with respect to a measure λ is denoted by Mλ,sing (S). Our main generic result on spaces of measures reads as follows. Theorem 1. Let S be a locally compact, σcompact, separable complete metric space. Then, the following holds: (1) The set Mc (S) is a Gδ set in M+ (S). (2) For any λ ∈ M+ (S), the set Mλ,sing (S) is a Gδ set in M+ (S). (3) For any closed F ⊂ S the set {µ ∈ M+ (S) F ⊂ supp(µ)} is a Gδ set in M+ (S).
3 Singular Continuity of Measures As a ﬁrst application of Theorem 1, we obtain a result on genericity of singular continuous spectrum. This generalizes the corresponding results of Simon [11] and Zamﬁrescu [13, 14] for measures on the real line. The ﬁrst result of this
Generically Singular Continuous Spectrum
335
section elaborates on Corollary 2.8 of [8] and generalizes the result mentioned there. The second result seems to be new. Theorem 2. Let S be be locally compact, σcompact, separable complete metric space. Let λ be a continuous measure on S with supp λ = S. Then, {µ ∈ M+ (S) µ continuous and µ ⊥ λ} is a dense Gδ set in M+ (S). Proof. The set in question is the intersection of Mc (S) and Mλ,sing (S). By Theorem 1, Mc (S) and Mλ,sing (S) are Gδ sets. Thus, by the Baire category theorem, it remains to show that Mc (S) and Mλ,sing (S) are dense. Denseness of these two sets follows from continuity of λ: By continuity of λ, Mλ,sing (S) contains all point measures and is therefore dense. Also, by continuity of λ, Mc (S) is dense (see e.g. (ii) of the next theorem). The previous theorem assumes the existence of a suitable measure λ. Such a λ does not always exist. Instead the following is valid. Theorem 3. Let S be a locally compact σcompact separable complete metric space. Then, the following assertions are equivalent: (i) There exists a continuous measure λ on S with supp λ = S. (ii) The set Mc (S) is dense in M+ (S). (iii) S has no isolated points. Proof. (i)=⇒ (ii): Let x ∈ S be arbitrary. Then, for any n ∈ N, the measure µn with µn := λ(B1/n (x))−1 λB1/n (x) is continuous. Moreover, the sequence (µn ) converges towards the unit point mass at x. Thus, every point measure can be approximated by continuous measures. As the point measures are dense, so are the continuous measures. (ii)=⇒ (iii): This is clear. (iii)=⇒ (i): We start with the following intermediate result: Claim. For each closed ball B in S with positive radius, there exists a continuous probability measure µB whose support is contained in B. Proof of claim. Denote the metric on S by d. Let x ∈ S and r > 0 with B = Br (x) be given. As S is locally compact, we can assume without loss of generality that B is compact. Deﬁne δ0 := r/4 and consider Bδ0 (x). By (iii), Bδ0 (x) contains two dif(1) (1) (1) (1) ferent points x1 and x2 . Set δ1 := 1/4 d(x11 , x2 ) and consider Bδ1 (x2 ) (1) and Bδ1 (x2 ). These balls are disjoint and by (ii) each contains two diﬀerent (2) (2) (1) (2) (2) (1) points, i.e. there exist x1 , x2 in Bδ1 (x1 ) and x3 , x4 in Bδ1 (x2 ).
336
D. Lenz and P. Stollmann (2)
(2)
(2)
Set δ2 := 1/4 min{d(xi , xj ) : i, j = 1, . . . , 4}. Then, the balls Bδ2 (xj ), j = 1, . . . , 4, are disjoint. Proceeding inductively, we can construct for each n ∈ N a δn > 0 and a set (n)
(n)
Xn := {x1 , . . . , x2n } with 2n elements such that δn , Xn ⊂ B, 2 and for each y ∈ B, and n ≥ k, the ball Bδk (y) contains at most 2n−k points of Xn . Now, consider, for each n ∈ N, the measure δn+1 ≤
n
µn :=
2 1 δ (n) . 2n j=1 xj
Then each µn is a probability measure supported in B and µn (Bδk (y)) ≤
1 × 2n−k = 2−k 2n
(1)
for every k ≤ n and y ∈ S. By the Theorem of Banach/Alaoglu, the sequence (µn ) has a converging subsequence. Thus, without loss of generality, we can assume that the sequence itself converges to a measure µ. As each µn is a probability measure supported in B and B is compact, µ is a probability measure supported in B as well. Moreover, by (1), µ is continuous. This ﬁnishes the proof of the claim. Let D be a countable dense subset of S. For each x ∈ D and n ∈ N, we can ﬁnd a continuous probability measure µn,x := µB 1 (x) supported in n
B n1 (x) according to the claim. Now, choose numbers cn,x > 0, x ∈ D and n ∈ N, with cn,x < ∞. n∈N,x∈D
Then, λ :=
cn,x µn,x
n∈N,x∈D
is a continuous measure whose support contains all x ∈ D. As D is dense, the support of λ equals S and (i) follows.
4 Selfadjoint Operators and the Wonderland Theorem In this section, we discuss a consequence of Theorem 2 on generic appearance of a purely singular continuous component in the spectrum for certain Hamiltonians.
Generically Singular Continuous Spectrum
337
This provides a slight strengthening of Simon’s “Wonderland Theorem”from [11]. The main point of our discussion, however, is not this strengthening but rather the new proof we provide. In order to formulate our result, let us introduce the following notation: For a ﬁxed separable Hilbert space H consider the space S = S(H) of selfadjoint operators in H. For ξ ∈ H and A ∈ S let the spectral measure ρA ξ be deﬁned by ρA ξ (ϕ) := ξ, ϕ(A)ξ for each continuous ϕ on R with compact support. We endow S with the strong resolvent topology τsrs , the weakest topology for which all the mappings S → C, A → (A + i)−1 ξ
(ξ ∈ H)
are continuous. Therefore, a sequence (An ) converges to A w.r.t. τsrs if and only if (An + i)−1 ξ → (A + i)−1 ξ for all ξ ∈ H. Thus, for each ξ ∈ H, the mapping ρξ : S −→ M(R), A → ρA ξ , is continuous. The spectral subspaces of A ∈ S are deﬁned by Hac (A) = {ξ ∈ H ρA ξ is absolutely continuous} Hsc (A) = {ξ ∈ H ρA ξ is singular continuous}, Hc (A) = {ξ ∈ H ρA ξ is continuous} Hpp (A) = Hc (A)⊥ , Hs (A) = Hac (A)⊥ . These subspaces are closed and invariant under A. Hpp (A) is the closed linear hull of the eigenvectors of A. Recall that the spectra σ∗ (A) are just the spectra of A restricted to H∗ (A). Theorem 4. Let (X, ρ) be a complete metric space and H : (X, ρ) → (S, τsrs ) a continuous mapping. Assume that, for an open set U ⊂ R, (1) the set {x ∈ X σpp (H(x)) ∩ U = ∅} is dense in X, (2) the set {x ∈ X σac (H(x)) ∩ U = ∅} is dense in X, (3) the set {x ∈ X U ⊂ σ(H(x))} is dense in X. Then, the set {x ∈ X U ⊂ σ(H(x)), σac (H(x)) ∩ U = ∅, σpp (H(x)) ∩ U = ∅} is a dense Gδ set in X.
338
D. Lenz and P. Stollmann
A proof can be sketched as follows (see [8] for details): By assumption H is continuous. Furthermore, the restriction rU : M(R) −→ M(U ), µ → µU , can easily be seen to be continuous as well. Finally, as discussed above, for each ξ ∈ H, the map ρξ is continuous. Thus, the composition H(x)
µξ := rU ◦ ρξ ◦ H : X −→ M(U ), x → ρξ
U ,
is a continuous map. Thus, the inverse image of a Gδ set in M(U ) under µξ is a Gδ set in X. Thus, by Theorem 1, the sets {x ∈ S : µξ (x) is continuous}, {x ∈ S : µξ (x) is singular w.r.t. Lebesgue measure} and {x ∈ S : supp µξ (x) contains U } are all Gδ sets. Moreover, by assumption they are dense. Thus, their intersection is a dense Gδ set by the Baire category theorem. One more intersection over a countable dense subset of ξ ∈ H yields the desired result. This ﬁnishes the proof of the theorem. We say that the spectrum of A ∈ S is pure point (purely absolutely continuous, purely singularly continuous) on U ⊂ R if the restrictions ρA ξ U have the corresponding properties. Of course, if A has pure point (purely absolutely continuous) spectrum on U it does not have any absolutely continuous (pure point) spectrum on U . As in [11], the theorem has then the following immediate but remarkable corollary. Corollary 1. Let (X, ρ) be a complete metric space and H : (X, ρ) → (S, τsrs ) a continuous mapping. Assume that, for an open set U ⊂ R, (1) the set {x ∈ X H(x) has pure point spectrum in U } is dense in X, (2) the set {x ∈ X H(x) has purely absolutely continuous spectrum in U } is dense in X, (3) the set {x ∈ X U ⊂ σ(H(x))} is dense in X. Then, the set {x ∈ X U ⊂ σ(H(x)), σac (H(x)) ∩ U = ∅, σpp (H(x)) ∩ U = ∅} is a dense Gδ set in X.
5 Operators Associated to Delone Sets In this section we discuss an application of Theorem 4 to geometric disorder. We start by recalling the necessary notation. A key notion is the notion of Delone set, named after B.N. Delone (Delaunay), [6]. The Euclidean norm on Rd is denoted by · . We replace B by U to denote open balls. Deﬁnition 1. A set ω ⊂ Rd is called an (r, R)set if – ∀x, y ∈ ω, x = y : x − y > r, – ∀p ∈ Rd ∃x ∈ ω : x − p ≤ R.
Generically Singular Continuous Spectrum
339
By Dr,R (Rd ) = Dr,R we denote the set of all (r, R)sets. We say that ω ⊂ Rd is a P Delone set, if it is an (r, R)set for some 0 < r ≤ R so that D(Rd ) = D = 0 0 an L > l such that the ωn ∩ UL (0) converge to ω ∩ UL (0) with respect to the Hausdorﬀ distance as n → ∞. Given the lemma, it is not hard to see that the map H : Dr,R (Rd ) −→ S(L2 ((Rd )), ω → H(ω), is continuous. Finally, we recall that a Delone set γ on Rd is called crystallographic if the set of its periods P er(γ) := {t ∈ Rd : t + γ = γ} is a lattice of full rank in Rd . Now our result on generic singularly continuous spectrum can be stated as follows. Theorem 5. Let r, R > 0 with 2r ≤ R and v be given such that there exist crystallographic γ, γ˜ ∈ Dr,R with σ(H(γ)) = σ(H(˜ γ )). Then U := (σ(H(γ))◦ \ σ(H(˜ γ ))) ∪ (σ(H(˜ γ ))◦ \ σ(H(γ))) is nonempty and there exists a dense Gδ set Ωsc ⊂ Dr,R such that for every ω ∈ Ωsc the spectrum of H(ω) contains U and is purely singular continuous in U .
340
D. Lenz and P. Stollmann
A proof can be sketched as follows (see [8] for details): γ )) and U2 := σ(H(˜ γ ))◦ \ σ(H(γ)). Since We let U1 := σ(H(γ))◦ \ σ(H(˜ γ, γ˜ are crystallographic, the corresponding operators are periodic and their spectra are consequently purely absolutely continuous and consist of a union of closed intervals with only ﬁnitely many gaps in every compact subset of the reals. Hence, under the assumption of the theorem U1 or U2 is nonempty. Thus, U is nonempty. We now consider the case that U1 is nonempty. We will verify conditions (1)(3) from Theorem 4. Ad (1): Fix ω ∈ Dr,R . For n ∈ N consider νn := ω ∩ Q(n). We can then periodically extend νn , i.e. we ﬁnd crystallographic ωn in Dr,R with ωn ∩ Q(n) = νn . For given L > 0 we get that ωn ∩ UL (0) = ω ∩ UL (0) if n is large enough. Therefore, by Lemma 1, we ﬁnd that ωn → ω with respect to the natural topology. On the other hand, σpp (H(ωn )) = ∅ since the potential of H(ωn ) is periodic. Consequently, {ω ∈ Dr,R σpp (H(ω)) ∩ U1 = ∅} is dense in Dr,R . Ad (2): We have to show denseness of ω for which σac (H(ω)) ∩ U1 = ∅. Thus, ﬁx ω ∈ Dr,R . Then, we can construct ωn which agree with ω around zero and with γ away from zero. More precisely, for n ∈ N large enough, ﬁnd ωn ∈ Dr,R such that ωn ∩ Un (0) = ω ∩ Un (0) and ωn ∩ U2n (0)c = γ ∩ U2n (0)c . In virtue of the last property, H(ωn ) and H( γ ) only diﬀer by a compactly γ )) ⊂ U1c . Again, supported, bounded potential, so that σac (H(ωn )) = σac (H( ωn → ω yields condition (2) of Theorem 4. Ad (3): This can be checked with a similar argument as (2), this time with γ˜ instead of γ. More precisely, ﬁx ω ∈ Dr,R . For n ∈ N large enough, we ﬁnd ωn ∈ Dr,R such that ωn ∩ Un (0) = ω ∩ Un (0) and ωn ∩ U2n (0)c = γ ∩ U2n (0)c . In virtue of the last property, H(ωn ) and H(γ) only diﬀer by a compactly supported, bounded potential, so that σac (H(ωn )) = σac (H(γ)) ⊃ U1 . By ωn → ω, we obtain (3) of Theorem 4. As a consequence of these considerations, Theorem 4 gives that {ω ∈ Dr,R  σpp (H(ω)) ∩ U1 = ∅, σac (H(ω)) ∩ U1 = ∅, U1 ⊂ σ(H(ω))} is a dense Gδ set if U1 is not empty. An analogous argument shows the same statement with U2 instead of U1 . This proves the assertion if only one of the Ui , i = 1, 2, is not empty. Otherwise, the assertion follows after intersecting the two dense Gδ ’s. This ﬁnishes the proof of the theorem.
Generically Singular Continuous Spectrum
341
Acknowledgment The research presented above has been partly supported by the DFG. Useful comments of J. Voigt are gratefully acknowledged. We would also like to take this opportunity to thank the organizers of QMath9 for the stimulating and relaxed atmosphere.
References 1. M. Baake, R.V. Moody (eds), Directions in mathematical quasicrystals, CRM Monogr. Ser., 13, Amer. Math. Soc., Provicence, RI, 2000. ¨ 2. S. Banach, Uber die Bairesche Kategorie gewisser Funktionenmengen. Studia Math. 3 (1931), 174–179. 3. H. Bauer, Maß und Integrationstheorie, De Gruyter, Berlin, 1990. 4. J. Bellissard, D.J.L. Hermann, and M. Zarrouati, Hulls of Aperiodic Solids and Gap Labeling Theorem, In: [1], pp. 207–258. 5. D. Buschmann, G. Stolz, Twoparameter spectral averaging and localization for nonmonotonic random Schr¨ odinger operators, Trans. Amer. Math. Soc. 353 (2001), 635–653. 6. B. Delaunay [B.N. Delone], Sur la sph`ere vide, Izvestia Akad Nauk SSSR Otdel. Mat. Sov. Nauk. 7, 1934, pp. 793800. 7. D. Lenz and P. Stollmann, Delone dynamical systems and associated random operators, Proceedings, Constanta (Romania), 2001, eds. J.M. Combes et al., Theta Foundation, eprint: arXiv:mathph/0202042. 8. D. Lenz and P. Stollmann, Generic sets in spaces of measures and generic singular continuous spectrum for Delone Hamiltonians, eprint: arXiv:mathph/0410021, to appear in Duke Math. J.. 9. S. Mazurkiewicz, Sur les fonctions non d´erivables, Studia Math. 3 (1931), 92–94. 10. R.V. Moody (ed.), The mathematics of longrange aperiodic order. NATO Advanced Science Institutes Series C: Mathematical and Physical Sciences, 489. Kluwer Academic Publishers Group, Dordrecht, 1997. 11. B. Simon, Operators with singular continuous spectrum: I. General operators. Annals of Math. 141, 131–145 (1995). 12. J. Weidmann, Lineare Operatoren in Hilbertr¨ aumen. Teil I: Grundlagen. B.G. Teubner, Stuttgart, 2000. 13. T. Zamﬁrescu, Most monotone functions are singular, Amer. Math. Monthly 88 (1981), 456–458. 14. T. Zamﬁrescu, Typical monotone continuous functions, Arch. Math. 42 (1984), 151–156.
Low Density Expansion for Lyapunov Exponents Hermann SchulzBaldes Mathematisches Institut, Universit¨ at ErlangenN¨ urnberg [emailprotected]
1 Introduction A perturbative formula for the Lyapunov exponent of a onedimensional random medium for weakly coupled disorder was ﬁrst given by Thouless [12] and then proven rigorously by Pastur and Figotin [9]. Anomalies in the perturbation theory at the band center were discovered by Kappus and Wegner [7] and further discussed by various other authors [2,3,11]. The Lyapunov exponent is then identiﬁed with the inverse localization length of the system. This short note concerns the behavior of the Lyapunov exponent for a low density of impurities, each of which may, however, be large. The presented method is as [6,10,11] a further application of diagonalizing the transfer matrices without perturbation (here the low density of impurities) and then rigorously controlling the error terms by means of oscillatory sums of rotating modiﬁed Pr¨ ufer phases. Some of the oscillatory sums remain large if the rotation phases (here the quasimomenta) are rational. This leads to supplementary contributions of the KappusWegner type. The calucalations are carried through explicitely for the onedimensional Anderson model, but the method transposes also to more complicated models with a periodic background as well as lowdensity disorder with correlations similar to the random polymer model [6]. Extension to a quasionedimensional situation as in [11] should be possible, but is even more cumbersome on a calculatory level. It is also straightforward to calculate and control higher order terms in the density. As one motivation for this study (apart from a mathematical one), let us indicate that a low density of strong impurities seems to describe materials like carbon nanotubes more adequately than a small coupling limit of the Anderson model. Indeed, these materials have perfect cristaline structure over distances of microns which leads to a ballisistic transport over such a distance [5]. The existing few defects are, on the other hand, quite large. Coherent transport within a oneparticle framework should then be studied by a model similar to the one considered here. However, it is possible that the impurties rather play the role of quantum dots so that Coulomb blockade is the determining eﬀect for the transport properties [8] rather than the coherent transport studied here. H. SchulzBaldes: Low Density Expansion for Lyapunov Exponents, Lect. Notes Phys. 690, 343–350 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
344
H. SchulzBaldes
2 Model and Preliminaries The standard onedimensional Anderson Hamiltonian is given by (Hω ψ)n = − ψn+1 − ψn−1 + vn ψn ,
ψ ∈ 2 (Z) .
Here ω = (vn )n∈Z is a sequence of independent and identically distributed real random variables. The model is determined by their probability distribution p depending on a given density ρ ∈ [0, 1]: ˜, p = (1 − ρ) δ0 + ρ p
(1)
˜ is a ﬁxed compactly supported probability measure on R. This meawhere p sure may simply be a Dirac peak if there is only one type of impurity, but ˜ = supp(˜ diﬀerent from δ0 . Set Σ = supp(p) and Σ p). The expectation w.r.t. ˜ v is ˜ while Ev and E ˜ ’s by E, the p’s will be denoted by E, that w.r.t. the p ˜ respectively over one random variable v ∈ Σ the expectation w.r.t. p and p only. In order to deﬁne the Lyapunov exponent, one rewrites the Schr¨ odinger equation Hω ψ = Eψ using transfer matrices vn − E −1 ψn ψn+1 E E = Tn . Tn = . ψn ψn−1 1 0 We also write TvE for TnE if vn = v. Then the Lyapunov exponent at energy E ∈ R associated to products of random matrices chosen independently according to p from the family (TvE )v∈Σ of SL(2, R)matrices is given by N 8 8 1 8 E8 . (2) E log 8 Tn 8 γ(ρ, E) = lim N →∞ N n=1 The aim is to study the asymptotics of γ(ρ, E) in small ρ for E < 2. In order to state our results, let us introduce, for E = −2 cos(k) with k ∈ (0, π), the basis change M ∈ SL(2, R) and the rotation matrix Rk by the quasimomentum k: 1 sin(k) 0 cos(k) − sin(k) M =
, Rk = . sin(k) cos(k) sin(k) − cos(k) 1 It is then a matter of computation to verify M TvE M −1
= Rk (1 +
PvE )
,
PvE
v = − sin(k)
00 10
.
ˆ= Next we introduce another auxiliary family of random matrices. Set Σ ˜ and, for (ψ, v) ∈ Σ: ˆ [− π2 , π2 ) × Σ
Low Density Expansion for Lyapunov Exponents
345
E Tˆψ,v = Rψ M TvE M −1 .
ˆ ˆ∞ = The following probability measures on Σ will play a role in the sequel: p " q dψ 1 ˆ q = q p=1 δ π ( p − q+1 ) ⊗˜ p for q ∈ N. The Lyapunov exponents p and p π ⊗˜ 2 q 2q associated to these families of random matrices are denoted by γˆ∞ (E) and γˆq (E) respectively. It is elementary to check that the subgroups generated ˆ q are noncompact ˆ ∞ and p by matrices corresponding to the supports of p and strongly irreducible, which implies [1] that the corresponding Lyapunov exponents are strictly positive. E induce actions SE,v and SˆE,ψ,v on R via The matrices TvE and Tˆψ,v eSE,v (θ) =
M TvE M −1 eθ , M TvE M −1 eθ
eSˆE,ψ,v (θ) =
E Tˆψ,v eθ . E Tˆ eθ
(3)
ψ,v
where the freedom of phase is ﬁxed by SE,0 (θ) = θ+k and SˆE,ψ,0 (θ) = θ+k+ψ E and νˆqE for these as well as the continuity in v. Invariant measures ν E , νˆ∞ ˆ q are deﬁned by ˆ ∞ and p actions and the probability measures p, p π π dν E (θ) f (θ) = dν E (θ) Ev f (SE,v (θ)modπ) , f ∈ C(R/πZ) , 0
E and similar formulas for νˆ∞ and νˆqE . Due to a theorem of Furstenberg [1], the invariant measures exist and are unique whenever the associated Lyapunov exponent is positive. Let us note that one can easily verify that the invariant E E is simply given by the Lebesgue measure dθ ˆ∞ measure νˆ∞ π . Furthermore ν E E and νˆq do not depend on ρ (but ν does). Next let us write out a more explicit formula for the new Lyapunov exponent γˆ∞ (E). First of all, according to Furstenberg’s formula [1, 6], π cos(θ) E E ˆ ψ,v γˆ∞ (E) = E dˆ ν∞ (θ) log(Tˆψ,v eθ ) , where eθ = . sin(θ) 0 E is the Lebesgue measure, rotations are orthogonal and the integrand As νˆ∞ is πperiodic, one gets 2π dθ 1 ˜ log eθ (M TvE M −1 )∗ (M TvE M −1 )eθ . (4) γˆ∞ (E) = Ev 2 2π 0
Now (M TvE M −1 )∗ (M TvE M −1 ) = 1 + PvE 2 is a positive matrix with eigen2 2 a values λv ≥ 1 and 1/λv where λv = 1 + 2 + a + a4 with a = sinv2 (k) . As it can be diagonalized by an orthogonal transformation leaving the Lebesgue measure invariant, we deduce that 2π dθ 1 ˜ 1 log λv cos2 (θ) + sin2 (θ) γˆ∞ (E) = Ev 2 2π λv 0 1 + λ2 1 v = . d˜ p(v) log 2 2λv
346
H. SchulzBaldes
˜ = δ0 (in which This formula shows immediately that γˆ∞ (E) > 0 unless p case λv = λ0 = 1).
3 Result on the Lyapunov Exponent Theorem. Let E = −2 cos(k) and k ∈ (0, π) with either satisfying the weak diophantine condition 1 − e2ımk ≥ c e−ξ m , ∀ m∈Z, for some c > 0 and ξ > 0. Then ρ γˆ∞ (E) + O(ρ2 ) γ(ρ, E) = ρ γˆ (E) + O(ρ2 ) q
k π
rational or k (5)
k satisﬁes (5) , k=π
p q
,
˜, where p and q are relatively prime. Furthermore, for ξ depending only on p  γˆq (E) − γˆ∞ (E)  ≤ c e−ξq . The result can be interpreted as follows: if the density of the impurities is small, then the incoming (Pr¨ ufer) phase at the impurity is uniformly distrubuted for a suﬃciently irrational rotation angle (i.e. quasimomentum) because the sole invariant measure of an irrational rotation is the Lebesgue measure. For a rational rotation, the mixing is to lowest order in ρ perfect over the orbits of the rational rotation, which leads to the deﬁnition of the E ˆ q . As indicated above, the proof )(p,σ)∈Σˆq and its distribution p family (Tˆp,σ that this is the correct image is another simple application of modiﬁed Pr¨ ufer phases and an oscillatory sum argument. Let us note that γˆq (E) = γˆ∞ (E); more detailed formulas for the diﬀerence are given below. As a result, one can expect a numerical curve of the energy dependence of the Lyapunov exponent at a given ﬁxed low density to have spikes at energies corresponding to rational quasimomenta with small denominators. Moreover, the invariant measures ν E and νˆqE are not close to the Lebesgue measure, but have higher harmonics as is typical at KappusWegner anomalies. Furthermore, let us add that at the band center E = 0 the identity γ(ρ, 0) = ρ γˆ0 (0) holds with no higher order correction terms and where γˆ0 (0) is the center of band Lyapunov exponent of the usual Anderson ˜. model with distribution p Finally, let us compare the above result with that obtained for a weakcoupling limit of the Anderson model [6, 9]: the Lyapunov exponent grows quadratically in the coupling constant of the disordered potential, while it grows linearly in the density. The reason is easily understood if one thinks of the change of the coupling constant also rather as a change of the probability
Low Density Expansion for Lyapunov Exponents
347
distribution on the space of matrices. At zero coupling, the measure is supported on one critical point (or more generally, on a commuting subset), and the weight in its neighborhood grows as the coupling constant grows. In (1) the weight may grow far from the critical point, and this leads to the faster increase of the Lyapunov exponent.
4 Proof For ﬁxed energy E, conﬁguration (vn )n∈N and (ψn )n∈N , as well as an initial condition θ0 , let us deﬁne iteratively the seqences θˆn = SˆE,ψn ,vn (θˆn−1 ) = SE,vn (θˆn−1 ) + ψn . (6)
θn = SE,vn (θn−1 ) ,
When considered modulo π, these are also called the modiﬁed Pr¨ ufer phases. They can be eﬃciently used in order to calculate the Lyapunov exponent as well as the density of states. For the Lyapunov exponent, let us ﬁrst note that one can make a basis change in (2) at the price of boundary terms vanishing at the limit, and furthermore, that according to [1, A.III.3.4] it is possible to introduce an arbitrary initial vector, so that N 8 8 1 8 8 E −1 (7) eθ 8 . E log 8 M Tn M γ(ρ, E) = lim N →∞ N n=1 Now using the modiﬁed Pr¨ ufer phases with initial condition θ0 = θ, this can be developed into a telescopic sum: N 8 8 1 8 8 E log 8 M TnE M −1 eθn−1 8 N →∞ N n=1
γ(ρ, E) = lim
N 8 8 1 8 ˜ v log 8 E E 8 M TvE M −1 eθn−1 8 , N →∞ N n=1
= ρ lim
where in the second step we have evaluated the partial expectation over the last random variable vn by using the fact that for vn = 0 the contribution ˜ v log(M T E M −1 eθ ) vanishes. Next let us note that the function eıθ → E v has an analytic extension to C\{0}, contains only even frequencies so that its Fourier series 8 8 8 ˜ v log 8 am e2ımθ , E 8 M TvE M −1 eθ 8 = m∈Z
has coeﬃcients satisfying for any ξ > 0 a Cauchy estimate of the form am  ≤ cξ e−ξm . Comparing with (4), we deduce
(8)
348
H. SchulzBaldes
a0 = γˆ∞ (E) . Introducing now the oscillatory sums N 1 2ımθn Im (N ) = E e , N n=1
N 1 2ımθˆn ˆ ˆ Im (N ) = E e , N n=1
the Lyapunov exponent now reads γ(ρ, E) = ρ am lim Im (N ) . N →∞
m∈Z
(9)
Hence we need to calculate Im (N ) perturbatively in ρ. Clearly I0 (N ) = 1. Furthermore, integrating over the initial condition w.r.t. the invariant measure gives for all N ∈ N E dν (θ) Im (N ) = dν E (θ) e2ımθ . Hence calculating Im (N ) perturbatively also gives the harmonics of ν E perturbatively (similar statements hold for Iˆm (N ), of course). Going back in history once, one gets Im (N ) =
N 1 ˜ v (e2ımSE,v (θn−1 ) ) (1 − ρ) e2ımk e2ımθn−1 + ρ E E N n=1
= (1 − ρ) e2ımk Im (N ) + O(ρ, N −1 ) . For k satisfying (5), one deduces Im (N ) ≤
1 O(ρ, N −1 ) ≤ c eξ m O(ρ, N −1 ) . 2ımk 1 − (1 − ρ) e 
Replacing this and (8) with ξ > ξ into (9) concludes the proof in this case because only the term m = 0 gives a contribution to order ρ. If now k = π pq , the same argument implies Inq+r (N ) = O(ρ, N −1 ) , Setting ˜ v (e2ımSE,v (θ) ) = E
∀ n ∈ Z , r = 1, . . . , q − 1 .
(m) 2ı(m+l)θ
bl
e
,
l∈Z
and
ˆ ψ,v (e2ımSˆE,ψ,v (θ) ) = E
l∈Z
one deduces for the remaining cases
ˆb(m) e2ı(m+l)θ , l
(10)
Low Density Expansion for Lyapunov Exponents
Inq (N ) = (1 − ρ) Inq (N ) + O(N −1 ) + ρ
349
(nq)
bl
Inq+l (N ) + O(N −1 ) .
l∈Z
Due to (10), this gives the following equations (nq) Inq (N ) = brq I(n+r)q (N ) + O((ρN )−1 , ρ) . r∈Z
They determine the invariant measure ν E to lowest order in ρ. This shows, in particular, that ν E is already to lowest order not given by the Lebesgue measure. We will not solve these equations, but rather show that the oscillatory sums Iˆnq (N ) satisfy the same equations, and hence, up to higher order corrections, the measure νˆqE can be used instead of ν E in order to calculate the Lyapunov exponent. Indeed, it follows directly from (6) and the deﬁnition ˆ q that of p (m) ˆb(m) = δm . modq=0 bl l In particular, Iˆm (N ) = 0 if m mod q = 0. Thus we deduce (nq) ˆ I(n+r)q (N ) + O(N −1 ) . Iˆnq (N ) = brq r∈Z
Comparing the equations for Inq (N ) and Iˆnq (N ) (which have a unique solution becaue the invariant measures are unique by Furstenberg’s theorem), it follows that Iˆnq (N ) = Inq (N ) + O(ρ, (ρN )−1 ) . Replacing this into (9), one deduces γ(ρ, E) = ρ am dˆ νqE (θ) e2ımθ + O(ρ2 ) m∈Z
=ρ
8 8 8 ˜ v log 8 dˆ νqE (θ) E 8 M TvE M −1 eθ 8 + O(ρ2 ) .
Now due to the orthogonality of rotations one may replace M TvE M −1 by E Tˆψ,v , and then the r.h.s. contains exactly the Furstenberg formula for γˆq (E) as claimed. The estimate comparing γˆq (E) and γˆ∞ (E) follows directly from the Cauchy estimate (8).
5 Result on the Density of States Another ergodic quantity of interest is the integrated density of states, deﬁned by 1 E # neg. eigenvalues of Hω − E on 2 ({1, . . . , N }) . N (ρ, E) = lim N →∞ N
350
H. SchulzBaldes
Recall, in particular, that N (0, E) = k if E = −2 cos(k). Deﬁning the mean phase shift of the impurities by ˜ v (SE,v (θ) − θ) , ϕ(θ) ˜ = E the low density expansion of the density of states reads as follows: dθ ϕ(θ) ˜ + O(ρ2 ) k satisﬁes (5) , (1 − ρ) k + ρ 2π N (ρ, E) = E ˜ + O(ρ2 ) k = π pq (1 − ρ) k + ρ dˆ νq (θ) ϕ(θ) with p and q relatively prime. The proof of this is completely analogous to the above when the rotation number calculation (e.g. [6] for a proof) is applied: N (ρ, E) = lim
N →∞
N 1 E SE,v (θn−1 ) − θn−1 N n=1
= (1 − ρ) k + ρ lim
N →∞
N 1 E ϕ(θ ˜ n−1 ) . N n=1
References 1. P. Bougerol, J. Lacroix, Products of Random Matrices with Applications to Schr¨ odinger Operators, (Birkh¨ auser, Boston, 1985). 2. A. Bovier, A. Klein, Weak disorder expansion of the invariant measure for the onedimensional Anderson model, J. Stat. Phys. 51, 501–517 (1988). 3. M. Campanino, A. Klein, Anomalies in the onedimensional Anderson model at weak disorder, Commun. Math. Phys. 130, 441–456 (1990). 4. B. Derrida, E.J. Gardner, Lyapunov exponent of the one dimensional Anderson model: weak disorder expansion, J. Physique 45, 1283 (1984). 5. S. Frank, P. Poncharal, Z.L. Wang, W. DeHeer, Carbon Nanotube Quantum Resistors, Science 280, 1744–1746 (1998). 6. S. Jitomirskaya, H. SchulzBaldes, G. Stolz, Delocalization in random polymer chains, Commun. Math. Phys. 233, 27–48 (2003). 7. M. Kappus, F. Wegner, Anomaly in the band centre of the onedimensional Anderson model, Z. Phys. B 45, 15–21 (1981). 8. P.L. McEuen, M. Bockrath, D.H. Cobden, Y.G. Yoon, S. Louie, Disorder, pseudospins, and backscattering in carbon nanotubes, Phys. Rev. Lett. 83, 5098 (1999). 9. L. Pastur, A. Figotin, Spectra of Random and AlmostPeriodic Operators, (Springer, Berlin, 1992). 10. R. Schrader, H. SchulzBaldes, A. Sedrakyan, Perturbative test of single parameter scaling for 1D random media, Ann. H. Poincar´e, 5, 1159–1180 (2004). 11. H. SchulzBaldes, Perturbation theroy for Lyapunov exponents of an Anderson model on a strip, GAFA 14, 1089–1117 (2004). 12. D.J. Thouless, in IllCondensed Matter, Les Houches Summer School, 1978, edited by R. Balian, R. Maynard, G. Toulouse (NorthHolland, New York, 1979).
Poisson Statistics for the Largest Eigenvalues in Random Matrix Ensembles Alexander Soshnikov University of California at Davis, Department of Mathematics, Davis, CA 95616, USA [emailprotected]
1 Introduction The two archetypal ensembles of random matrices are Wigner real symmetric (Hermitian) random matrices and Wishart sample covariance real (complex) random matrices. In this paper we study the statistical properties of the largest eigenvalues of such matrices in the case when the second moments of matrix entries are inﬁnite. In the ﬁrst two subsections we consider Wigner ensemble of random matrices and its generalization – band random matrices. 1.1 Wigner Random Matrices A real symmetric Wigner random matrix is deﬁned as a square symmetric n × n matrix with i.i.d. entries up from the diagonal A = (ajk ), ajk = akj , 1 ≤ j ≤ k ≤ n, {ajk }j 0. We then concluded, that with probability going to 1, there is no row i such 1 " 2 +2δ 1 +δ aij  ≥ bn . To establish (31), it is enough to prove that that 2 j:aij ≤bn
for any ﬁxed row i Pr
3 4 +2δ
aij  ≥ bn
< exp(−n ) ,
(32)
3 +δ 1/2+δ j:bn ≤aij ≤bn4
for suﬃciently small positive . The proof is very similar to the argument presented in Lemma 5, and is left to the reader.
360
A. Soshnikov
3 Real Sample Covariance Matrices with Cauchy Entries Let A be a rectangular m×n matrix with independent identically distributed entries with the marginal probability distribution of matrix entries satisfying (21). Based on the results in the last section, one can expect that the largest eigenvalues have Poisson statistics as well. At this point, we have been able to prove it only in a weak form, and only when matrix entries have Cauchy distribution. We recall, that the probability density of the Cauchy distribution is given 1 by the formula f (x) = π(1+x 2 ) . Cauchy distribution is very important in probability theory (see e.g. [15]). In particular, Cauchy distribition is a (1,1,0) stable distribution, i.e. the scale parameter is 1, the index of the distribution α = 1 and the symmetry parameter is zero (see [18], [23]). The following theorem was proven by Fyodorov and Soshnikov in [39] Theorem 2. Let A be a random rectangular m × n matrix (m ≥ n) with i.i.d. Cauchy entries and let z be a complex number with a positive real part. Then, as n → ∞ we have ∞ −1/2 z 2√ t = exp − z = E (1 + zxi )−1/2 , lim E det 1 + 2 2 A A n→∞ m n π i=1 √ (33) √ where we consider the branch of z on D = {z : /z > 0} such that 1 = 1, E denotes the mathematical expectation with respect to the random matrix ensemble deﬁned above, E denotes the mathematical expectation with respect to the inhom*ogeneous Poisson random point process on the positive halfaxis with the intensity πx13/2 , and the convergence is uniform inside D (i.e. it is unform on compact subsets of D). For a real positive z = t2 , t ∈ R1 , one can estimate the rate of convergence, namely −1/2 t2 2 = exp − t 1 + o(n−1/2+ ) lim E det 1 + 2 2 At A , n→∞ m n π (34) where is an arbitrary small positive number and the convergence is uniform on the compact subsets of [0, +∞). The result of Theorem 2 allows a generalization to the case of a sparse random matrix with Cauchy entries. Let, as before, {ajk }, 1 ≤ j ≤ m, 1 ≤ k ≤ n, be i.i.d. Cauchy random variables, and B = (bjk ) be a m × n nonrandom rectangular 0 − 1 matrix such that the number of nonzero entries in each column is ﬁxed and equals to dn . Let dn grow polynomially, i.e. bn ≥ nα , for some 0 < α ≤ 1. Also assume that lnm grows much slower than than any power of n. Deﬁne a m × n rectangular matrix A with the entries Ajk = bjk ajk , 1 ≤ j ≤ m, 1 ≤ k ≤ n. Let λ1 ≥ λ2 . . . ≥ λn denote the eigenvalues of At × A. The appropriate rescaling for the largest eigenvalues in this case is λ˜i = λi m2 d2 , i = 1, . . . , n. n
Poisson Statistics in Random Matrices
361
Theorem 3. Let A be a sparse random rectangular m × n matrix (m ≥ n) deﬁned as above, and let z be a complex number with a positive real part. Then, as n → ∞ we have
−1/2 n z t = lim E (1 + z λ˜i )−1/2 (35) lim E det 1 + 2 2 A A n→∞ n→∞ m dn i=1 ∞ 2√ = exp − z = E (1 + zxi )−1/2 , (36) π i=1 √ where, as √ in Theorem 1.1, we consider the branch of z on D = {z : /z > 0} such that 1 = 1. E denotes the mathematical expectation with respect to the inhom*ogeneous Poisson random point process on the positive halfaxis with the intensity πx13/2 , and the convergence is uniform inside D (i.e. it is unform on the compact subsets of D). For a real positive z = t2 , t ∈ R1 , one can get an estimate on the rate of convergence, namely
−1/2 t2 2 t −1/2+ E det 1 + 2 2 A A = exp − t 1 + o(dn ) , m dn π
(37)
where is an arbitrary small positive number and the convergence is uniform on the compact subsets of [0, +∞). The proof relies on the following property of the Gaussian integrals: −1/2 = det(B)
N 1 x exp −xBxt dN , π
(38)
where B is an N dimensional matrix with a positive deﬁnite Hermitian part (i.e. all eigenvalues of B + B ∗ are positive), x = (x1 , . . . , xN ) ∈ RN , and " t xBx = ij bij xi xj . Id tiA Let B = B(t) = . Then, one can write tiAt Id
−1/2 det(1 + t2 At A) =
det
1 tiA tiAt 1
−1/2
−1/2 = det(B) ,
(39)
and apply (38) to the r.h.s. of (39). Assuming that the entries of A are independent, one can signiﬁcantly simplify the expression, using the fact that entries of A appear linearly in B(t) (see Proposition 1 of [39]). In the Cauchy case, one can simplify the calculations even further, and prove that −1/2 √ exists and equals exp − π2 z . limn→∞ E det(1 + m2zd2 At A) n On the other side, for Poisson >∞ random point processes the mathematical expectations of the type E i=1 (1 + f (xi )) can be calculated explicitely
362
A. Soshnikov
E
∞
(1 + f (xi )) = 1 +
i=1
∞ k=1
E
k
f (xij )
(40)
1≤i1 2, such maps nevertheless have surprisingly rich properties. Indeed, in that case A has two real eigenvectors v± and two real eigenvalues exp ±γ0 , with the socalled Lyapounov exponent γ0 > 0. This implies the dynamical system is exponentially unstable. Roughly speaking, if x, x ∈ T2 are a small distance
apart, then generically At x and At x will be a distance exp γ0 t apart: the dynamics displays sensitive dependence on initial conditions, a well known source of chaotic behaviour. To put it graphically, if you know that initially the system point is in some set of linear size , all you can say at a time
Recent Results on Quantum Map Eigenstates
369
of order γ10 ln  later is that the system is in a set of size 1, which is not saying much more than that it is somewhere in the torus, and that is not saying very much at all. In other words, after a time of order γ10 ln  you lost all information on the whereabouts of the system. This is a very short time indeed, as you can see by imagining = 0 10−k : it then grows only linearly in k. This instability has the consequence that the dynamics is exponentially mixing. This means that ∀f1 , f2 ∈ C ∞ (T2 ),  (f1 ◦ At )(x)f2 (x)dx − f1 (x)dx f2 (x)dx ≤ CA,f1 ∇f2 1 e−γ0 t . T2
T2
T2
Replacing f1 and f2 by characteristic functions of subsets B1 and B2 of the torus, this can be interpreted as follows: at large times, the probability that a trajectory starting in some set B2 ends up in B1 converges to the size of B1 , independently of where B1 and B2 are located on the torus. This is again a way of saying that all information on the whereabouts of the system is rapidly lost over time. Exponential mixing plays an important role in the following analysis. It implies furthermore that the dynamics is ergodic, meaning that for almost all initial conditions x0 ∈ T2 , T 1 f (At x0 ) → f (x)dx . T t=1 T2 Note that this can not possibly be true for all initial conditions since there exists a dense set of periodic points for the dynamics: every point on the torus that has two rational coordinates is periodic. All these periodic orbits are unstable, of course. The simplest one is the ﬁxed point x = 0. These discrete dynamical systems are arguably the simplest fully chaotic Hamiltonian systems one can ﬁnd: in particular, the ergodic properties I mentioned are readily proved by some simple Fourier analysis. This is in sharp contrast to what happens for other systems. Proving the Sinai or Bunimovich billiards are ergodic (let alone mixing) is very hard work and the geodesic ﬂow on a negative curvature Riemannian manifold is also not a particularly simple object to study. This is presumably why Hannay and Berry [18] came up with the idea of constructing a quantum mechanical equivalent to these simple systems, that I will describe in the next section. Before turning to that task, let me point out that all the above properties of the dynamics are preserved if it is perturbed. Let g be a ﬁxed smooth function on T2 and consider Φ = Φg ◦ A . For small , this is still hyperbolic and exponentially mixing, with exponent γ , lim→0 γ = γ0 [3].
370
S. De Bi`evre
3 Quantum Maps What are the discrete quantum dynamical systems that correspond to the discrete classical dynamical systems described in the previous section? I will give a much simpliﬁed description and refer for details to [5] [11]. To avoid inessential technicalities, I will consider only the case where A is of the form (m ∈ N∗ ) 2m 1 A= . 4m2 − 1 2m Introducing for each a = (a1 , a2 ) ∈ R2 the phase space translation operator U (a)ψ(y) = e− 2 a1 a2 e a2 y ψ(y − a1 ) = e− (a1 P −a2 Q) ψ(y) , i
i
i
the quantum Hilbert space of the system is deﬁned by H = {ψ ∈ S (R)  U (1, 0)ψ = ψ = U (0, 1)ψ} . This space is N dimensional when 2πN = 1 and zerodimensional otherwise. Then any ψ ∈ H can be written as follows: c δ y − ψ(y) = ; c+N = c . N ∈Z
If A were not of the above form, the deﬁnition of H involves some extra phases that clutter up the page, which is the only reason why I avoid treating the general case here. That H is ﬁnite dimensional is not surprising: since the classical phase space has ﬁnite area, and since the uncertainty principle tells us every quantum state “takes up” an area 2π, it is normal the Hilbert space is ﬁnite dimensional. This is reminiscent also of what happens when treating spin, where the classical phase space is a sphere, and the quantum Hilbert spaces are also ﬁnite dimensional. The relation 2πN = 1 will always be assumed to hold from now on. So when both and N show up in the same formula, they are always related in that manner. Consequently, the semiclassical limit corresponds to taking N to inﬁnity. In this spirit, I will occasionally ﬁnd it useful to write HN rather than H . The standard Weyl quantization is readily adapted to the present situation. For f ∈ C ∞ (T2 ), x = (q, p) ∈ T2 , write fn e−i2π(n1 p−n2 q) f (x) = n∈Z2
and deﬁne OpW f = fˆ =
n∈Z2
fn e−i2π(n1 P −n2 Q) =
n∈Z2
fn U
n N
: H → H .
Recent Results on Quantum Map Eigenstates
371
The quantum dynamics is now deﬁned as follows. For A ∈ SL(2, Z),  TrA > 2, construct 1/2 2 2 i i e 2 (2my −yy +2my ) ψ(y )dy . M (A)ψ(y) = 2π R Then, for all t ∈ Z, M (A)H = H
and
M (A)−t OpW f M (A)t − OpW (f ◦ At ) = 0 .
Now, for > 0 deﬁne the unitary operator U = e− Op
W
i
g
M (A) : H → H .
This is the quantum map we wish to study. It is naturally related to the discrete Hamiltonian dynamics on T2 obtained by iterating Φ = Φg ◦ A. It acts on the N dimensional spaces H and we are interested in the behaviour of its eigenfunctions and eigenvalues in the N → ∞ limit: (N )
U ψj
(N )
(N )
= eiθj ψj
,
j = 1...N .
4 What is Known? Let me start by recalling a precise statement of the Schnirelman theorem in the present context: Theorem 1. [5] Let ≥ 0 and small. Then, for “almost all” sequences (N )
(N )
ψjN ∈ H , so that U ψjN = e (N )
ψjN
(N )
iθj
(N )
ψjN , (N ) N →+∞  OpW f  ψjN → f (x)dx, N
T2
∀f ∈ C ∞ (T2 ) .
(1)
Note that here, for each N , a ﬁxed basis of eigenfunctions has been chosen, and jN is then an index in the set {1, . . . N }. The “almost all” statement refers to the fact that one has to choose the jN from a subset GN ⊂ {1, . . . , N } such that #GN /N → 1. As I already mentioned, this result is ‘robust’: it holds quite generally whenever the classical system is ergodic. It is proven in each diﬀerent situation by adapting arguments that are used to prove it for other systems, such as LaplaceBeltrami operators on negatively curved manifolds. This is why it holds also for = 0 and on higher dimensional tori. With (sometimes considerable) additional work, the proof can be even be adapted for maps that are not continuous, such as the Baker map [12, 14]. A version of the theorem for certain systems with a mixed phase space also exists [25]. As pointed out in the introduction, this theorem invites some immediate and obvious questions:
372
S. De Bi`evre
Question 1. Do there exist exceptional sequences of eigenfunctions meaning ones that do not converge semiclassically to Lebesgue measure (this means (1) does not hold). Question 2. Can you characterize all possible semiclassical measures? Here any measure ν on the torus for which there exists some sequence of eigenfunctions ψN satisfying N →+∞
ψN OpW f ψN → f (x)dν, ∀f ∈ C ∞ (T2 ) (2) T2
is called a semiclassical measure. Note that in this deﬁnition the sequence is not necessarily picked from a preassigned basis as is the case in the Schnirelman theorem. There are of course still other questions that readily come to mind, related to the speed of convergence in the Schnirelman theorem, as well as to the ﬂuctuations of the matrix elements that I shall not discuss here. Answers to the above two questions are not available in general systems. Only in some restricted classes of models have partial answers been obtained. For the (Hecke) eigenfunctions of the LaplaceBeltrami operator of a (class) of constant negative curvature surfaces the answer to this ﬁrst question has been proven to be “NO!” [24]. On the other hand, for quantized hyperbolic toral automorphisms, the answer has been proven to be “YES!”: Theorem 2. [15] Let = 0. Let 0 ≤ α ≤ 12 , then there exists Nk → ∞ and eigenfunctions ψNk ∈ HNk so that N →+∞
ψNk OpW f ψNk k → αf (0) + (1 − α) f (x)dx, ∀f ∈ C ∞ (T2 ) . (3) T2
This theorem holds for any choice of A with TrA > 2, not just for the particular ones discussed here. The sequence Nk depends on A. The construction (sketched below) yields such a sequence of eigenfunctions for any sequence of (N ) eigenangles θjN k of the quantum maps. k Note that this theorem is only stated for the quantized cat maps themselves, not for their perturbations. I strongly believe the result does not survive perturbation, a claim I will try to corroborate below. Let me ﬁrst sketch some of the ingredients that go into the proof of the result. I will indicate how the eigenfunctions with the above property are constructed in the case α = 1/2. First of all, it is well known that the quantum maps M (A) have a period: for each N , there exists an integer TN so (N ) that M (A)TN = eiϕN 1lHN . This implies that the eigenvalues exp iθj have the property: ! 2π ϕN (N ) θj ∈ +  ∈ Z . TN TN (N )
Consequently, the N eigenangles θj can take on at most TN distinct equally spaced values. Let me note in passing that it can be shown that the eigenvalues equidistribute around the circle in the classical limit [6]: this is the
Recent Results on Quantum Map Eigenstates
373
analog of Weyl’s law in the present context. Now it is easy to see that Pk =
1 TN
TN /2
e
− Ti (2πjk+ϕN j) N
M (A)j
j=−TN /2+1
is a projector. I am supposing TN is even; if it is odd, you sum from −(TN − 1)/2 to (TN −1)/2. If this projector is nonzero, it projects onto the eigenspace i (2πk+ϕN ) . In particular, if you can ﬁnd corresponding to the eigenvalue e TN some nonzero vector φN for which Pk φN = 0, it certainly is an eigenvector. This suggests it is easy to construct at least some eigenvectors. Now, we are hunting for eigenvectors that are well localized around the origin in phase space. The idea is then to take for φN a coherent state centered on the origin. Recall that in the ordinary quantum mechanics of a system with one degree of freedom, those are of the form ηz (y) =
Imz π
1/4 exp i
z 2 y , 2
where z is a complex number in the upper half plane. By periodicizing this Gaussian in both the y variable and the dual (Fourier) variable, one obtains a family 0, z of states in HN that have two crucial properties for our purposes. First, they are localized at the origin in the sense that, as N → ∞,  0, zOpW f 0, z − f (0) ≤ Cf
1 + z2 . ImzN
(4)
Next, they behave nicely under the dynamics: M (A)0, z = eiσN (A) 0, A · z .
(5)
Here σN (A) is a phase that we will not play any role in what follows and A·z =
2mz + (4m2 − 1) . z + 2m
Let us now compute the matrix element of an observable OpW f in an evolved coherent state: 0, At · zOpW f 0, At · z. Since Im(At · z) behaves like exp −2γ0 t as t → ∞ (the real part being uniformly bounded in time), the upper bound in (4) then guarantees that, for times t satisfying t < 12 τN , where 1 τN = lnN , γ0 this still converges to f (0):
0, zM (A)−t OpW f M (A)t 0, z
N →+∞
→
f (0) .
(6)
374
S. De Bi`evre
But what happens for longer times? It was proven in [8–10] that, for times t between 12 τN and 32 τN , one has instead
0, zM (A)−t OpW f M (A)t 0, z → f (x)dx . (7) T2
This is a consequence of the familiar “spreading of the wavepacket”, together with the exponential instability of the dynamics. One can indeed √ think of a coherent state as being concentrated in a region of linear size in both the position and momentum variables. Now in the semiclassical regime, evolving the coherent state with the exponentially unstable √ dynamics A stretches its support along the unstable direction v+ to a size exp γ0 t: this is no longer microscopic if t is of order 2γ10 lnN or more so that the estimate (6) breaks down. Since the unstable manifold through 0, which is just the line Rv+ through the origin, wraps itself ergodically around the torus, estimate (7) follows. To understand the reason for its breakdown beyond 32 τN we need to return to the quantum periods TN . The quantum periods TN are very irregular functions of N , that grow roughly linearly in N [20], but that can, for some N , be very much smaller than that. It is shown in [8] that there exists a sequence Nk for which this period behaves like 2τN . But this means that, for those values of N
0, zM (A)−(2τN −s) OpW f M (A)2τN −s 0, z = 0, zM (A)s OpW f M (A)−s 0, z and consequently, for times s so that s < 12 τN , and N → ∞,
0, zM (A)−(2τN −s) OpW f M (A)2τN −s 0, z → f (0) . This explains the breakdown of (7) for times beyond 32 τN . This breakdown can also be understood intuitively [8]. The exponential instability of the classical dynamics explained in Sect. 2 implies that in the na¨ıvest semiclassical picture of the quantum dynamics the support of the Wigner function of the evolved coherent states is stretched out over √ exp γ0 t along the unstable direction. Now, this supa length Lt = port is wound around the torus, and so successive windings are a distance Dt = −1/2 exp −γ0 t apart. But on the torus, for any pair of conjugate variables Q, P , the uncertainty principle says ∆Q∆P ≥ , and, since we are on the torus ∆Q, ∆P ≤ 1, so that both ∆Q and ∆P are greater than ! This means, that on the torus, the Wigner function can not resolve details ﬁner than on a scale . Hence one may expect that the above simple semiclassical picture of the evolution will break down when Dt is of order , and this is precisely when t is of order 32 τN . At this time, interference eﬀects start to take over. In the case of CAT maps, this leads eventually to the very “unclassical” recurrence at the quantum period, in which the original wave packet is reconstituted completely for t = TN , and which may be as short as 2τN . Let me now write
Recent Results on Quantum Map Eigenstates
ψN =
375
TN Pk 0, z .
These are our√candidate eigenfunctions that will be proven to satisfy (3). The extra factor TN is a normalization factor the origin of which will become clear in a second. It is enough to look at (4) and (7) to understand (i) why this ought to work when N = Nk and (ii) what the remaining problems are. First of all, let’s be optimistic and suppose that in computing the matrix element ψN , OpW f ψN we have to deal only with the diagonal matrix elements. Let us write TN k
ψNk Op f ψNk = W
1
2
TNk t=−
TN k 2
0, At · zOpW f 0, At · z + OﬀDTerms . +1
(8) Now taking N → ∞ and using (6) and (7), the sum of the diagonal terms above converges to the desired 12 f (0) + 12 T2 f (x)dx. This then proves Theorem 2 provided one can control the oﬀ diagonal terms, which requires more work. One further idea that goes into this, and that is easily understood intuitively, is the following. For positive times t+ larger than 12 τN , the evolved coherent states stretches along the unstable manifold of x = 0 in the sense that its Wigner or Husimi function has a support that is concentrated along √ Rv+ over a length Lt+ = exp γo t+ . Similarly, in the past, for negative times −t− so that t− ≥ 12 τN , the evolved coherent state stretches along the stable manifold Rv− over a length Lt− . Now these two intervals intersect, for each such time in only a ﬁnite (but growing!) set of hom*oclinic points. Since a ﬁnite number of points on a line should carry no weight (if all goes well), this suggests that the evolved states M (A)t+ 0, z and M (A)−t− 0, z are almost orthogonal. This is indeed what is proven in [15], for times t− + t+ < 2τN . It should by now be clear why I claim the construction of strongly scarred eigenstates is particular to quantum CAT maps. Without the existence of quantum periods (which is already exceptional), and without the rather special arithmetic properties guaranteeing the existence of very short quantum periods, the construction would generate only quasimodes, not eigenvectors. The construction of strongly scarred eigenstates even breaks down on higher dimensional tori, since then the maps with short quantum periods do not exist [7]. All of this indicates that these strongly scarred eigenstates on isolated unstable periodic orbits are a particular feature of the unperturbed twodimensional cat maps. Having answered Question 1 positively, we can now turn to Question 2: what measures on T2 can be obtained as limits of the type ψN OpW f ψN , for some sequence of eigenfunctions ψN . It is easy to see that such measures have to be invariant under the dynamics generated by A. Elaborating a little
376
S. De Bi`evre
on the above ideas, one can show that, if µ is an Ainvariant probability measure on T2 that is singular with respect to Lebesgue measure any probability measure on T2 of the form αµ + (1 − α)dx
(9)
is a limit measure provided α ≤ 1/2. This follows from a diagonalization trick and the well known fact that all invariant probability measures are weak limits of pure point measures [16]. This is optimal in the following sense: Theorem 3. [9, 16] Let = 0. Let µ be an Ainvariant pure point measure. Let α > 12 . Then there does not exist Nk → ∞ and eigenfunctions ψNk ∈ HNk so that N →+∞
ψNk , OpW f ψNk k → αµ(f ) + (1 − α) f (x)dx, ∀f ∈ C ∞ (T2 ) . T2
This result was obtained in [9] with a suboptimal bound α > 2/3, which was then optimized in [16]. Still, this does not completely answer Question 2. In particular, it is not known what happens to this result if µ is purely singular continuous or, more speciﬁcally, if a purely singular measure can be a semiclassical limit of the above type. This theorem does continue to hold on higher dimensional tori, but it is not clear if it is then optimal, for the reasons explained above. The sequence Nk in the above construction is very sparse. It behaves roughly like exp(γ0 k/2). It is legitimate to ask what happens if you stay away from this sequence: are exceptional sequences of eigenfunctions then still possible? First, one could consider sequences of N along which TN behaves like στN , with σ > 2. In that cases it is tempting to guess that any measure of the form (9) is now a semiclassical measure provided α ≤ σ −1 . The idea is that, in the eigenfunction construction sketched above, the fraction of terms that concentrates at the origin is σ −1 . Note however that our estimates do not imply such a result immediately, since we did not control the dynamics for times longer than 2τN . Still following the same line of thought, whenever the sequence of N is chosen so that TN /lnN → ∞, one expects that the system should be uniquely quantum ergodic along that sequence. Here is a result that corroborates this conjecture: Theorem 4. [22] If A ∈ SL(2, Z) is hyperbolic and a11 a12 ≡ 0 ≡ a21 a22 ˜ )∈N along mod 2, then there exists a density one sequence of integers (N which ˜ →+∞ N f (x)dx, ∀f ∈ C ∞ (T2 )
ψN˜ OpW f ψN˜ → T2
for any sequence of eigenfunctions ψN˜ . This is a generalization of the result of [13], where a result of this type is ˜ showing up proven along a subsequence of primes. Clearly, the sequence N
Recent Results on Quantum Map Eigenstates
377
here is disjoint from the sequence Nk in Theorem 2. In fact TN˜ is larger than
˜ . This result is of course a considerable strengthening of the Schnirelman N theorem for the sequence of N concerned, because it says equidistribution holds for all sequences of eigenfunctions, not necessarily picked from a preassigned basis, and without any “density one” restriction, as opposed to what happens in the statement of the Schnirelman theorem. To end this discussion, I would like to point out one further result that is fun to compare to Theorem 2: Theorem 5. [23] If A ∈ SL(2, Z) is hyperbolic and A ≡ I2 mod 4, then there exists for each N a basis {ψ1 , ψ2 , . . . ψN } of eigenfunctions of M (A) so that N →+∞ W f (x)dx, ∀f ∈ C ∞ (T2 ) (10)
ψjN Op f ψjN → T2
holds for any sequence ψjN . This again is an obvious strengthening of the Schnirelman theorem for the particular class of A considered. Note that the result holds for all N , without exceptions, so that a superﬁcial reading of this result makes it look as if it is in contradiction with Theorem 2. The point is that the basis for which the result holds is explicitly described in the paper. Now, the eigenvalues of M (A) may be degenerate so that it is possible that exceptional sequences of eigenfunctions not belonging to the above basis have a diﬀerent semiclassical limit. This is precisely what happens with Theorem 2. Actually, since the quantum Hilbert spaces are N dimensional, the fact that the period is of order lnNk for the special values of N implies that at least some of the eigenvalues must be highly degenerate in this case, with degeneracies of order Nk /lnNk . Although our construction of the exceptional eigenfunctions does not exploit this degeneracy explicitly, it clearly does so implicitly. Note ﬁnally ˜ of Theorem 4 is such that the quantum periods T ˜ are that the sequence N N
˜ longer than N . In fact, the authors of that paper show that, whenever the periods are at least this long along a sequence of integers, quantum unique ergodicity holds along that sequence. This implies that degeneracies as large
˜ do not suﬃce to produce strong scarring. as N
5 Perturbed Cat Maps CAT maps are rather special: their linearity makes them easy to treat classically and endows their quantum counterparts with a number of arithmetic properties that can be exploited to prove a variety of results going beyond the Schnirelman theorem, as in [13, 22, 23] and to a lesser degree in [8, 9, 15, 16]. Since such results are not available in more complicated systems, the CAT maps provide a theoretical laboratory in which one can investigate with relative ease a certain number of questions in quantum chaos. But a real understanding of the issues of quantum chaos requires proofs that rely exclusively
378
S. De Bi`evre
on semiclassical analysis and on the ergodic properties of the underlying classical dynamics of the systems considered. Indeed, only such proofs have a ﬁghting chance to give insight into more complicated models, such as billiards or geodesic ﬂows on negatively curved manifolds, and ultimately more realistic models with mixed phase space. Before attacking these systems, a good starting point is to look at the perturbed CAT maps described in Sect. 2. They are nonlinear and preserve none of the simple arithmetic properties of the CAT maps themselves, once quantized. Their classical mechanics is suﬃciently complicated that detailed information on their classical mixing rate has become available only rather recently [3]. On the quantum side, and contrary to what happens for quantized CAT maps, their eigenvalue statistics has been shown numerically to be generic [21]. Beyond the Schnirelman theorem, the only theorem in the previous section the proof of which has a chance to generalize to the case of perturbed maps is Theorem 3. The simplest version one could hope to prove would be: If ψN ∈ HN is a sequence of eigenfunctions of U , then there exist a function f so that lim ψN , OpW f ψN = f (0).
N →∞
In other words, this statement says that for a perturbed cat map, the delta measure at the origin is not a semiclassical measure of the system. A stragegy of proof of such a result is suggested by the main ingredient of the proof of Theorem 3 in [10]. The following result is proven there: Theorem 6. Let = 0. Let a0 ∈ T2 . If ϕN ∈ H is some sequence (not necessarily eigenfunctions!) with the property that, for all f ∈ C ∞ (T2 )
ϕN OpW f ϕN
N →+∞
→
f (a0 )
then there exists a sequence of times tN → ∞ so that N →+∞
ϕN U0−tN OpW f U0tN ϕN → f (x)dx . T2
If you can prove this result for = 0, then that would imply immediately that a sequence of eigenfunctions can not concentrate on x = 0, since for eigenfunctions
ψN OpW f ψN = ψN M (A)−t OpW f M (A)t ψN . The time scales tN involved in the theorem are logarithmic in , more precisely, they are of the form (1 + δ) γ10 lnN . So if you want to prove something like this for = 0, you need to be able to control UtN at such time scales. Recall that for the Schnirelman theorem, such long time control is not needed since there, you take → 0 ﬁrst, and only after that exploit ergodicity by taking t → ∞.
Recent Results on Quantum Map Eigenstates
379
In [4], we made some progress on this question. To state the result, we need a deﬁnition: Deﬁnition 1. A sequence ϕN ∈ H is said to localize at a0 ∈ T2 if N →+∞
ϕN OpW f ϕN −→ f (a0 ), This implies there exists r → 0 so that dx  ϕN , xz2 2π x−a≥r
∀f ∈ C ∞ (T2 ) .
N →+∞
−→ 0 .
Here x, z is the coherent state localized at x ∈ T2 . Theorem 7. Let U be as before. There exists δ0 > 0 with the following property. Let ψN be a sequence of eigenfunctions of U that localizes at some 1 point a ∈ T2 . Then r ≥ 2 −δ0 . One can paraphrase this result as follows: Eigenfunctions do it slowly (if they do it at all). In other words, a sequence of eigenfunctions can concentrate on a point only if it does so slowly, meaning that the region in which it concentrates is “large”, 1 i.e. of linear size of order 2 −δ0 . The strategy of the proof is the following: we ﬁrst prove an Egorov theorem for the system with control on the error term for time of order 3γ2 ln. We then use this result to get information on the propagation of coherent states, and ﬁnally use this to obtain the result. This result is of course far from optimal: this is largely due to the fact that we control the dynamics for insuﬃciently long times. Roughly speaking, and modulo some other technical problems, to show the desired result along those lines, one expects to need to control the dynamics up to times (1 + δ) γ1 ln. This is consistent with the observation that such time scales are used in the proof of Theorem 6 for the unperturbed case. An improvement on the above result has been announced in [17]: Nonnenmacher has announced he can take δ0 ( ) in such a way that it converges to 1/2 as goes to zero. To do this, he controls the relevant dynamics for times up to (1 − δ( )) γ1 ln. A diﬀerent approach to characterizing semiclassical measures has very recently been proposed in [1, 2]. In [1], the author investigated the semiclassical measures on manifolds of negative curvature and shows, modulo an as yet unproven hypothesis on the eigenfunctions, that their support can not have small topological entropy. That result implies in particular that they can not concentrate on a (ﬁnite union of) periodic orbit(s). Recently, Anantharaman has been able to get rid of the extra unproven assumption [2]. If it is conﬁrmed that this last argument does indeed work, then I expect it to go through also for the perturbed cat maps, with the same result: no semiclassical measures concentrated on a ﬁnite set of periodic orbits. In addition, since Anantharaman obtains these results with a control on the quantum
380
S. De Bi`evre
evolution up to times that are only of the form δ γ1 ln, this is very good news for quantum chaos in general: it would mean very strong results on the eigenfunction behaviour can be obtained with rather minimal control on the long time quantum evolution, contrary to the general intuition conﬁrmed by the work described in this note. Although I have tried to explain that Theorem 2 is particular to quantized CAT maps and that, in my opinion, strong scars on isolated hyperbolic periodic orbits are not likely to show up in other systems, for those wanting to prove unique quantum ergodicity in exponentially unstable systems, there is still a lesson to be learned from Theorem 2: any general strategy based on semiclassical analysis and ergodic properties of the underlying dynamical system alone will work for CAT maps as well. But Theorem 2 shows CAT maps are not uniquely ergodic.
Acknowledgments It is a pleasure to thank S. Nonnenmacher and Z. Rudnick for helpful comments and remarks.
References 1. N. Anantharaman, The eigenfunctions of the Laplacian do not concentrate on sets of small topological entropy, preprint june 2004. 2. N. Anantharaman, private communication, march 2005. 3. M. Blank, G. Keller, C. Liverani, RuellePerronFrobenius spectrum for Anosov maps, Nonlinearity 15, no. 6, 1905–1973 (2002). 4. J. M. Bouclet, S. De Bi`evre, Long time propagation and control on scarring for perturbed quantized hyperbolic toral automorphisms, mp− arc 04305, Annales H. Poincar´e, to appear. 5. A. Bouzouina, S. De Bi`evre, Equipartition of the eigenfunctions of quantized ergodic maps on the torus, Comm. Math. Phys. 178, 83–105 (1996). 6. A. Bouzouina, S. De Bi`evre, Equidistribution des valeurs propres et ergodicit´e semiclassique de symplectomorphismes du tore quantiﬁ´ es, C. R. Acad. Sci. Paris, S´erie I t.326, 1021–1024 (1998). 7. P. Corvaja, Z. Rudnick, U. Zannier, A lower bound for periods of matrices, Comm. Math. Phys. 252, 13, 535–541 (2004). 8. F. Bonecchi, S. De Bi`evre, Exponential mixing and ln time scales in quantized hyperbolic maps on the torus, Comm. Math. Phys. 211, 659–686 (2000). 9. F. Bonecchi, S. De Bi`evre, Controlling strong scarring for quantized ergodic toral automorphisms, Duke Math. J, Vol. 117, No. 3, 571–587 (2003). 10. F. Bonecchi, S. De Bi`evre, Controlling strong scarring for quantized ergodic toral automorphisms, Section 5, mp arc 02–81 (2002). 11. S. De Bi`evre, Quantum chaos: a brief ﬁrst visit, mp− arc 01207, Contemporary Mathematics 289, 161–218 (2001).
Recent Results on Quantum Map Eigenstates
381
12. S. De Bi`evre, M. Degli Esposti, Egorov Theorems and equipartition of eigenfunctions for quantized sawtooth and Baker maps on the torus, mp− arc 96210, Ann. Inst. H. Poincar´e 69, 1, 1–30 (1998). 13. M. Degli Esposti, S. Graﬃ, S. Isola, Stochastic properties of the quantum Arnold cat in the classical limit, Commun. Math. Phys. 167, 471–509 (1995). 14. M. Degli Esposti, S. Nonnenmacher, B. Winn, Quantum variance and ergodicity for the baker’s map, preprint (2004), Commun. Math. Phys., to appear. 15. F. Faure, S. Nonnenmacher, S. De Bi`evre, Scarred eigenstates for quantum cats of minimal periods, Commun. Math. Phys. 239, 449–492 (2003). 16. F. Faure, S. Nonnenmacher, On the maximal scarring for quantum cat map eigenstates, Commun. Math. Phys. 245, 201–214 (2004). 17. F. Faure, S. Nonnenmacher, contribution at the Workshop on Random Matrix theory and Arithmetic Aspects of Quantum Chaos, Newton Institute, Cambridge, june 2004. 18. J.H. Hannay, M.V. Berry, Quantization of linear mapsFresnel diﬀraction by a periodic grating, Physica D 1, 267–290 (1980). 19. E.J. Heller, Boundstate eigenfunctions of classically chaotic hamiltonian systems: scars of periodic orbits, Phys. Rev. Lett. 53, 16, 1515–1518 (1984). 20. J.P. Keating, Asymptotic properties of the periodic orbits of the cat maps, Nonlinearity 4, 277–307 (1990). 21. J.P. Keating, F. Mezzadri, Pseudosymmetries of Anosov maps and Spectral statistics, Nonlinearity 13, 747–775 (2000). 22. P. Kurlberg, Z. Rudnick, Hecke theory and equidistribution for the quantization of linear maps of the torus, Duke Math. J. 103, 1, 47–77 (2000). 23. P. Kurlberg, Z. Rudnick, On quantum ergodicity for linear maps of the torus, Comm. Math. Phys. 222, 1, 201–227 (2001). 24. E. Lindenstrauss, Invariant measures and arithmetic quantum unique ergodicity, Annals of Math., to appear. 25. J. Marklof, S. O’Keefe, Weyl’s law and quantum ergodicity for maps with divided phase space, mp− arc 04122, Nonlinearity 18, 277–304 (2005). 26. S. Zelditch, Quantum ergodicity and mixing of eigenfunctions, mathph/ 0503026, preprint (2005), Elsevier Encyclopedia of Mathematical Physics, to appear.
Level Repulsion and Spectral Type for OneDimensional Adiabatic QuasiPeriodic Schr¨ odinger Operators Alexander Fedotov1 and Fr´ed´eric Klopp2 1
2
Department of Mathematical Physics, St Petersburg State University, 1, Ulianovskaja, 198904 St PetersburgPetrodvorets, Russia [emailprotected] LAGA, Institut Galil´ee, U.M.R 7539 C.N.R.S, Universit´e ParisNord, Avenue J.B. Cl´ement, F93430 Villetaneuse, France [emailprotected]
1 A Heuristic Description This report is devoted to the study of the spectral properties of the family of onedimensional quasiperiodic Schr¨ odinger operators acting on L2 (R) deﬁned by Hz,ε ψ = −
d2 ψ(x) + (V (x − z) + α cos(εx))ψ(x) , dx2
(1)
where (H1) V : R → R is a non constant, locally square integrable, 1periodic function; (H2) ε is a small positive number chosen such that 2π/ε be irrational; (H3) z is a real parameter indexing the operators; (H4) α is a strictly positive parameter. Quasiperiodic operators of the type (1) (and their discrete analogs) are standard models used to described the motion of an electron in a deterministic disordered system (see [3, 18, 20]). It is well known that the spectral type of these operators depends on the energies at which one looks at them ( [7]); Anderson transitions do occur in some regions ( [10]) whereas other regions display “essentially” purely absolutely continuous ( [9]) or purely singular spectrum ( [8]). In the present note, we review an energy region where the spectrum of the operator displays energy level repulsion and show how this repulsion, more precisely its strength relates to the spectral nature of the operator. The material is taken from [5,6]. To describe the energy region where we work, consider the spectrum of the periodic Schr¨ odinger operator (on L2 (R)) H0 = −
d2 + V (x) dx2
(2)
A. Fedotov and F. Klopp: Level Repulsion and Spectral Type for OneDimensional Adiabatic QuasiPeriodic Schr¨ odinger Operators, Lect. Notes Phys. 690, 383–402 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
384
A. Fedotov and F. Klopp E−α
E
E +α
Fig. 1. Bands in interaction
We assume that two of its spectral bands are interacting through the perturbation α cos i.e., that the relative position of the spectral window F(E) := [E − α, E + α] and the spectrum of the unperturbed operator H0 is that shown in Fig. 1. In such an energy region, the spectrum is localized (l) (l ) near two sequences of quantized energy values, say (E0 )l and (Eπ )l (see Theorem 1); each of these sequences is “generated” by one of the ends of the neighboring spectral bands of H0 . In [6], we study neighborhoods of such quantized energy values that are not resonant i.e. to neighborhoods of points (l) (l ) Eµ that are not “too” close to the points (Eν )l for {µ, ν} = {0, π}. The distance between the two sequences inﬂuences the nature and location of the spectrum: a weak level repulsion arises due to “weakly resonant tunneling”. Similarly to what happens in the standard “double well” case (see [13, 19]), the resonant tunneling begins to play an important role when the two energies, each generated by one of the quantization conditions, are suﬃciently close to each other. In the resonant case, we ﬁnd a strong relationship between the level repulsion and the nature of the spectrum. Recall that the latter is determined by the nature of the genralized eignefunctions i.e. the solutions to (Hz,ε − E)ψ = 0 (see [12]): localized or delocalized. Thus, it is very natural that the two characteristics are related: the slower the decay of the generalized eigenfunctions, the larger the overlap between generalized eigenfunction corresponding to close energy levels, hence, the larger the tunneling between these levels and thus the repulsion between them. Let us now brieﬂy describe the various situations we encounter. Let J be an interval of energies such that, for all E ∈ J, the spectral window F(E) covers the edges of two neighboring spectral bands of H0 and the gap located between them (see Fig. 1 and assumption (TIBM)). Under this assumption, consider the real and complex isoenergy curves associated to (1). Denoted respectively by ΓR and Γ , they are deﬁned by ΓR := {(κ, ζ) ∈ R2 , E(κ) + α · cos(ζ) = E} ,
(3)
Γ := {(κ, ζ) ∈ C , E(κ) + α · cos(ζ) = E} .
(4)
2
where E(κ) be the dispersion relation associated to H0 . These curves are roughly depicted in Fig. 2. They are periodic both in the ζ and κ variables. Consider one of the periodicity cells of ΓR . It contains two tori. They are denoted by γ0 and γπ and shown in full lines. Integrating 1/2 times the ˇ0 and fundamental 1form on Γ along γ0 and γπ , one deﬁnes the phases Φ
Adiabatic QuasiPeriodic Schr¨ odinger Operators
385
Fig. 2. The adiabatic phase space
ˇπ (see Theorem 1). Each of these phases deﬁnes a sequence of “quantized Φ energies” in J, say, for (l, m) ∈ N2 , 1ˇ (l) Φ0 (E0 ) = ε 1ˇ Φπ (Eπ(m) ) = ε
π + lπ , 2 π + mπ . 2
(5) (6)
Each of the dashed lines in Fig. 2 represents a loop on Γ that connects certain connected components of ΓR ; one can distinguish between the “horizontal” loops and the “vertical” loops. There are two special horizontal loops denoted by γh,0 and γh,π ; the loop γh,0 (resp. γh,π ) connects γ0 to γπ − (2π, 0) (resp. γ0 to γπ ). In the same way, there are two special vertical loops denoted by γv,0 and γv,π ; the loop γv,0 (resp. γv,π ) connects γ0 to γ0 + (0, 2π) (resp. γπ to γπ + (0, 2π)). To each of these complex loops, one associates an action obtained by integrating −i/2 times the fundamental 1form on Γ along the loop. For a ∈ {0, π} and b ∈ {v, h}, we denote by Sa,b the action associated to γa,b . For E ∈ R, all these actions are real. One orients the loops so that they all be positive. Finally, we deﬁne tunneling coeﬃcients as ta,b = e−Sa,b /ε ,
a ∈ {0, π}, b ∈ {v, h} . (l)
In Theorem 1, we prove that, for a ∈ {0, π}, near each Ea , there is an expo(l) nentially small interval Ia such that the spectrum of Hz,ε in J is contained in the union of these intervals. The precise description of the spectrum in the (l) interval Ia depends on whether it intersects another such interval or not. (l) (l) Note that an interval of type I0 can only intersect intervals of type Iπ and vice versa. (l) (l ) In the present report, we consider two intervals I0 and Iπ that do intersect and describe the spectrum in the union of these two intervals. It is
386
A. Fedotov and F. Klopp
useful to keep in mind that this union is exponentially small (when ε goes to 0). There are two main parameters controlling the spectral type: B B tv,0 tv,π max(tv,0 , tv,π )2 and ρ=2 τ =2 th,0 th,π th,0 th,π As we are working inside an exponentially small interval, the point inside this interval at which we compute the tunneling coeﬃcients does not really matter: over this interval, the variation of any of the actions is exponentially small. We essentially distinguish three regimes: – τ 1, – ρ 1 and τ 1, – ρ 1. The symbol 1 (resp. 1) mean that the quantity is exponentially small (resp. large) in 1ε as ε goes to 0, the exponential rate being arbitrary. In each of the three cases, we consider two energies, say E0 and Eπ , satisfying respectively (5) and (6), and describe the evolution of the spectrum as E0 and Eπ become closer to each other. As the quantization conditions (5) and (6) show, this can be achieved by reducing ε somewhat. As noted above, when moving E0 and Eπ closer together, we stay in the same regime for τ and ρ. In the case τ 1, when E0 and Eπ are still “far” away from each other, one sees two intervals containing spectrum, one located near each energy; they contain the same amount of spectrum i.e. the measure with respect to the density of states of each interval is ε/2π; and the Lyapunov exponent is positive on both intervals (see Fig. 5(a)). When E0 and Eπ approach each other, the picture does not change except for the fact that the intervals intersect, so only a single interval is seen (see Fig. 5(b)); its density of states measure is ε/π and the Lyapunov exponent is positive on this interval. There is no gap separating the intervals of spectrum generated by the two quantization conditions. This can be interpreted as a consequence of the positivity of the Lyapunov exponent : the states are well localized so the overlapping is weak and there no level repulsion. Nevertheless, the resonance has one eﬀect: when the two intervals intersect, it gives rise to a sharp drop of the Lyapunov exponent (which stays positive) in the middle of the interval containing spectrum. Over a distance exponentially small in ε, the Lyapunov exponent drops by an amount of order one. In the other extreme, in the case ρ 1, the starting geometry of the spectrum is essentially the same as in the previous case, namely, two well separated intervals containing each an ε/2π “part” of spectrum. The main diﬀerence is that each of these intervals now essentially only contains absolutely continuous spectrum. As the energies E0 and Eπ approach each
Adiabatic QuasiPeriodic Schr¨ odinger Operators
387
other, so do the intervals until they roughly reach an interspacing of size √ th ; during this process, the size of the intervals which, at the start, was √ roughly of order tv,0 + th and tv,π + th grew to reach the order of th (this number is much larger than any of the other two as ρ 1). When E0 and Eπ move closer to each √ other, the intervals containing spectrum stay “frozen” at a distance of size th from each other, and their sizes do not vary noticeably either (see Fig. 6). They start moving and changing size again √when E0 and Eπ again become separated by an interspacing of size at least th . So, in this case, we see a very strong repulsion preventing the intervals of spectra from intersecting each other. The interspacing is quite similar to that observed in the case of the standard double well problem (see [13, 19]). In the last case, when ρ 1 and τ 1, we see an intermediate behavior. For the sake of simplicity, let us assume that tv,π tv,0 . Starting from the situation when E0 and Eπ are far apart, we see two intervals, one around each point E0 and Eπ and this as long as E0 − Eπ  tv,π (see Fig. 7(a)). As in the ﬁrst case, the Lyapunov exponent varies by an amount of order one over each of the exponentially small intervals. The diﬀerence is that it need not stay positive: at the edges of the two intervals that are facing each other, it becomes small and even can vanish. These edges seem more prone to interaction. When now one moves E0 towards Eπ , the lacuna separating the two intervals stays open and starts moving with E0 ; it becomes roughly centered at E0 , stays of ﬁxed size (of order th /tv,π ) and moves along with E0 as E0 crosses Eπ and up to a distance roughly tv,π on the other side of Eπ (see Fig. 7(b)). Then, when E0 moves still further away from Eπ , it becomes again the center of some interval containing spectrum that starts moving away from the band centered at Eπ . We see that, in this case as in the case of strong repulsion, there always are two intervals separated by a gap; both intervals contain a ε/2π “part” of spectrum. But, now, the two intervals can become exponentially larger than the gap (in the case of strong interaction, the length of the gap was at least of the same order of magnitude as the lengths of the bands). Moreover, on both intervals, the Lyapunov exponent is positive near the outer edges i.e. the edges that are not facing each other; it can become small or even vanish on the inner edges. So, there may be some Anderson transitions within the intervals. We see here the eﬀects of a weaker form of resonant tunneling and a weaker repulsion.
2 Mathematical Results We now state our assumptions and results in a precise way.
388
A. Fedotov and F. Klopp
2.1 The Periodic Operator This section is devoted to the description of elements of the spectral theory of onedimensional periodic Schr¨ odinger operator H0 that we need to present our results. For proofs and more details, we refer to [4, 11]. The Spectrum of H0 The spectrum of the operator H0 deﬁned in (2) is a union of countably many intervals of the real axis, say [E2n+1 , E2n+2 ] for n ∈ N , such that E1 < E2 ≤ E3 < E4 . . . E2n ≤ E2n+1 < E2n+2 ≤ . . . , En → +∞,
n → +∞.
This spectrum is purely absolutely continuous. The points (Ej )j are the eigenvalues of the selfadjoint operator obtained by considering the diﬀerential polynomial (2) acting in L2 ([0, 2]) with periodic boundary conditions (see [4]). For n ∈ N, the intervals [E2n+1 , E2n+2 ] are the spectral bands, and the intervals (E2n , E2n+1 ), the spectral gaps. When E2n < E2n+1 , one says that the nth gap is open; when [E2n−1 , E2n ] is separated from the rest of the spectrum by open gaps, the nth band is said to be isolated. Generically all the gap are open. From now on, to simplify the exposition, we suppose that (O) all the gaps of the spectrum of H0 are open. The Bloch QuasiMomentum Let x → ψ(x, E) be a non trivial solution to the periodic Schr¨ odinger equation H0 ψ = Eψ such that ψ (x + 1, E) = µ ψ (x, E), ∀x ∈ R, for some µ ∈ C∗ independent of x. Such a solution is called a Bloch solution to the equation, and µ is the Floquet multiplier associated to ψ. One may write µ = exp(ik) where, k is the Bloch quasimomentum of the Bloch solution ψ. It appears that the mapping E → k(E) is analytic and multivalued; its branch points are the points {En ; n ∈ N}. They are all of “square root” type. The dispersion relation k → E(k) is the inverse of the Bloch quasimomentum. 2.2 A “Geometric” Assumption on the Energy Region Under Study Let us now describe the energy region where our study is valid. Recall that the spectral window F(E) is the range of the mapping ζ ∈ R → E − α cos(ζ). In the sequel, J always denotes a compact interval such that, for some n ∈ N∗ and for all E ∈ J, one has
Adiabatic QuasiPeriodic Schr¨ odinger Operators
389
˙ (TIBM) [E2n , E2n+1 ] ⊂ F(E) and F(E) ⊂]E2n−1 , E2n+2 [. ˙ where F(E) is the interior of F(E) (see Fig. 1). Remark 1. As all the spectral gaps of H0 are assumed to be open, as their length tends to 0 at inﬁnity, and, as the length of the spectral bands goes to inﬁnity at inﬁnity, it is clear that, for any non vanishing α, assumption (TIBM) is satisﬁed in any gap at a suﬃciently high energy; it suﬃces that this gap be of length smaller than 2α. 2.3 The Deﬁnitions of the Phase Integrals and the Tunelling Coeﬃcients The Complex Momentum and its Branch Points The phase integrals and the tunelling coeﬃcients are expressed in terms of integrals of the complex momentum. Fix E in J. The complex momentum ζ → κ(ζ) is deﬁned by κ(ζ) = k(E − α cos(ζ)) .
(7)
As k, κ is analytic and multivalued. The set Γ deﬁned in (4) is the graph of the function κ. As the branch points of k are the points (Ei )i∈N , the branch points of κ satisfy (8) E − α cos(ζ) = Ej , j ∈ N∗ . As E is real, the set of these points is symmetric with respect to the real axis, to the imaginary axis and it is 2πperiodic in ζ. All the branch points of κ lie on arccos(R) which consists of the real axis and all the translates of the imaginary axis by a multiple of π. As the branch points of the Bloch quasimomentum, the branch points of κ are of “square root” type. Due to the symmetries, it suﬃces to describe the branch points in the halfstrip {ζ; Imζ ≥ 0, 0 ≤ Reζ ≤ π}. These branch points are described in detail in section 7.1.1 of [6]. In Fig. 3, we show some of them. The points ζj being deﬁned by (8), one has 0 < ζ2n < ζ2n+1 < π, 0 < Imζ2n+2 < Imζ2n+3 < · · · , 0 < Imζ2n−1 < · · · < Imζ1 . The Contours To deﬁne the phases and the tunneling coeﬃcients, we introduce various integration contours in the complex ζplane. These loops are shown in Fig. 3 and 4. The loops γ˜0 , γ˜π , γ˜h,0 , γ˜h,π , γ˜v,0 and γ˜v,π are simple loops, respectively, going once around the intervals [−ζ2n , ζ2n ], [ζ2n+1 , 2π −ζ2n+1 ], [−ζ2n+1 , −ζ2n ], [ζ2n , ζ2n+1 ], [ζ2n−1 , ζ2n−1 ] and [ζ2n+2 , ζ2n+2 ]. On each of the above loops, one can ﬁx a continuous branch of the complex momentum.
390
A. Fedotov and F. Klopp ζ2n+2
ζ2n−1 ζ2n+1
γ˜0 0
π
ζ2n
γ˜π
Fig. 3. The loops for the phase integrals
Fig. 4. The loops for the action integrals
Consider Γ , the complex isoenergy curve deﬁned by (4). Deﬁne the projection Π : (ζ, κ) ∈ Γ → ζ ∈ C. The fact that, on each of the loops γ˜0 , γ˜π , γ˜h,0 , γ˜h,π , γ˜v,0 and γ˜v,π , one can ﬁx a continuous branch of the complex momentum implies that each of these loops is the projection on the complex plane of some loop in Γ (see [6]). The curves γ0 , γπ , γh,0 , γh,π , γv,0 and γv,π represented in Fig. 2 are mapped onto the curves γ˜0 , γ˜π , γ˜h,0 , γ˜h,π , γ˜v,0 and γ˜v,π respectively by the projector Π : (κ, ζ) ∈ Γ → ζ ∈ Γ . The Phase Integrals, the Action Integrals and the Tunneling Coeﬃcients Let ν ∈ {0, π}. To the loop γν , we associate the phase integral Φν deﬁned as 1 Φν (E) = κ dζ , (9) 2 γ˜ν where κ is a branch of the complex momentum that is continuous on γ˜ν . The function E → Φν (E) is real analytic and does not vanish on J. The loop γ˜ν is oriented so that Φν (E) be positive. One shows that, for all E ∈ J, Φ0 (E) < 0
and
Φπ (E) > 0 .
(10)
To the loop γv,ν , we associate the vertical action integral Sv,ν deﬁned as i Sv,ν (E) = − κdζ , (11) 2 γ˜v,ν
Adiabatic QuasiPeriodic Schr¨ odinger Operators
391
where κ is a branch of the complex momentum that is continuous on γ˜v,ν . The function E → Sv,ν (E) is real analytic and does not vanish on J. The loop γ˜v,ν is oriented so that Sv,ν (E) be positive. The vertical tunneling coeﬃcient is deﬁned to be 1 tv,ν (E) = exp − Sv,ν (E) . (12) ε The index ν being chosen as above, we deﬁne horizontal action integral Sh,ν by i κ(ζ) dζ , (13) Sh,ν (E) = − 2 γ˜h,ν where κ is a branch of the complex momentum that is continuous on γ˜h,ν . The function E → Sh,ν (E) is real analytic and does not vanish on J. The loop γ˜h,ν is oriented so that Sh,ν (E) be positive. The horizontal tunneling coeﬃcient is deﬁned as 1 (14) th,ν (E) = exp − Sh,ν (E) . ε As the cosine is even, one has Sh,0 (E) = Sh,π (E)
and
th,0 (E) = th,π (E) .
(15)
Finally, one deﬁnes Sh (E) = Sh,0 (E) + Sh,π (E)
and
th (E) = th,0 (E) · th,π (E) .
(16)
In (9), (11), and (13), only the sign of the integral depends on the choice of the branch of κ; this sign was ﬁxed by orienting the integration contour. 2.4 Ergodic Family Before discussing the spectral properties of Hz,ε , we recall some general results from the spectral theory of ergodic operators. As 2π/ε is supposed to be irrational, the operators deﬁned by (1) form an ergodic family (see [17]). The ergodicity has the following consequences: 1. the spectrum of Hz,ε is almost surely independent of z ( [18]); in fact, in the case of the quasiperiodic operators, it is independent of z ( [1]); 2. the absolutely continuous spectrum and the singular spectrum are almost surely independent of z ( [18]); in fact, in the case of the quasiperiodic operators, they are independent of z ( [15]); 3. for almost all z, the discret spectrum is empty ( [18]);
392
A. Fedotov and F. Klopp
4. the Lyapunov exponent exists for almost all z and is independent of z ( [18]); it is deﬁned in the following way: let x → ψ(x) be the solution to the Cauchy problem Hz,ε ψ = Eψ,
ψx=0 = 0,
ψx=0 =1,
the following limit (when it exists) deﬁnes the Lyapunov exponent:
log ψ(x, E, z)2 + ψ (x, E, z)2 Θ(E) = Θ(E, ε) := lim . x→+∞ x 5. the absolutely continuous spectrum is the essential closure of the set of E where Θ(E) = 0 (the IshiiPasturKotani Theorem, see [18]); 6. the density of states exists for almost all z and is independent of z ( [18]); it is deﬁned in the following way: for L > 0, let Hz,ε;L be the operator Hz,ε restricted to the interval [−L, L] with the Dirichlet boundary conditions; for E ∈ R; the following limit (when it exists) deﬁnes the density of states: #{ eigenvalues of Hz,ε;L less then or equal toE} ; L→+∞ 2L
Nε (E) := lim
7. the density of states is non decreasing; the spectrum of Hz,ε is the set of points of increase of the density of states. 2.5 A Coarse Description of the Location of the Spectrum in J Henceforth, we assume that the assumptions (H) and (O) are satisﬁed and that J is a compact interval satisfying (TIBM). We assume that max max(Sh (E), Sv,0 (E), Sv,π (E)) (T)
E∈J
< 2π · min min(Imζ2n−2 (E), Imζ2n+3 (E)) . E∈J
Note that (T) is veriﬁed if the spectrum of H0 has two successive bands that are suﬃciently close to each other and suﬃciently far away from the remainder of the spectrum (this can be checked numerically on simple examples, see Sect. 2.9). Deﬁne 1 (17) δ0 := inf min(Sh (E), Sv,0 (E), Sv,π (E)) > 0 . 2 E∈J To locate the spectrum roughly, we prove Theorem 1 ( [6]). Fix E∗ ∈ J. For ε suﬃciently small, there exists V∗ ⊂ C, ˇ0 (E, ε) and a neighborhood of E∗ , and two real analytic functions E → Φ ˇ E → Φπ (E, ε), deﬁned on V∗ satisfying the uniform asymptotics
Adiabatic QuasiPeriodic Schr¨ odinger Operators
ˇ0 (E, ε) = Φ0 (E) + o(ε), Φ
Φˇπ (E, ε) = Φπ (E) + o(ε)
393
when ε → 0 ,
(18)
(l)
such that, if one deﬁnes two ﬁnite sequences of points in J ∩V∗ , say (E0 )l := (l) (l ) (l ) (E0 (ε))l and (Eπ )l := (Eπ (ε))l , by 1ˇ π (l) Φ0 (E0 , ε) = + πl ε 2
1ˇ π Φπ (Eπ(l ) , ε) = + πl , ε 2
and
(l, l ) ∈ N2 , (19)
then, for all real z, the spectrum of Hz,ε in J ∩ V∗ is contained in the union of the intervals I0 := E0 + [−e−δ0 /ε , e−δ0 /ε ] (l)
that is
(l)
Iπ(l ) := Eπ(l ) + [−e−δ0 /ε , e−δ0 /ε ] (20) O (l) O O (l ) I0 Iπ σ(Hz,ε ) ∩ J ∩ V∗ ⊂ . and
l
l
In the sequel, to alleviate the notations, we omit the reference to ε in the ˇπ . ˇ0 and Φ functions Φ By (10) and (18), there exists C > 0 such that, for ε suﬃciently small, the points deﬁned in (19) satisfy 1 (l) (l−1) ε ≤ E0 − E0 ≤ Cε , C 1 ε ≤ Eπ(l) − Eπ(l−1) ≤ Cε . C
(21) (22) (l)
Moreover, for ν ∈ {0, π}, in the interval J ∩ V∗ , the number of points Eν is of order 1/ε. (l) (l) In the sequel, we refer to the points E0 (resp. Eπ ), and, by extension, (l) (l) to the intervals I0 (resp. Iπ ) attached to them, as of type 0 (resp. type π). By (21) and (22), the intervals of type 0 (resp. π) are two by two disjoints; any interval of type 0 (resp. π) intersects at most a single interval of type π (resp. 0). 2.6 A Precise Description of the Spectrum As pointed out in the introduction, the present paper deals with the resonant (l) (l ) case that is we consider two energies, say E0 et Eπ , that satisfy
Eπ(l ) − E0  ≤ 2e− (l)
(l)
(l )
δ0 ε
.
(23)
This means that the intervals I0 and Iπ intersect each other. Moreover, by (21) and (22), these intervals stay at a distance at least C −1 ε of all the other intervals of the sequences deﬁned in Theorem 1. We now describe the (l) (l ) spectrum of Hz,ε in the union I0 ∪ Iπ .
394
A. Fedotov and F. Klopp
To simplify the exposition, we set (l)
E0 := E0 ,
Eπ := Eπ(l ) ,
(l)
I0 := I0 ,
and
Iπ := Iπ(l ) .
(24)
In the resonant case, the primary parameter controlling the location and the nature of the spectrum is B ¯ tv,π (E) ¯ tv,0 (E) ¯ = E π + E0 . (25) where E τ =2 ¯ 2 th (E) As tunneling coeﬃcients are exponentially small, one typically has either τ 1 or τ 1. We will give a detailed analysis of these cases. More precisely, we ﬁx δ > 0 arbitrary and assume that either ∀E ∈ V∗ ∩ J,
Sh (E) − Sv,0 (E) − Sv,π (E) ≥ δ ,
(26)
Sh (E) − Sv,0 (E) − Sv,π (E) ≤ −δ .
(27)
or ∀E ∈ V∗ ∩ J,
The case τ 0 1 is more complicated and satisﬁed by less energies. We discuss it brieﬂy later. To describe our results, it is convenient to introduce the following “local variables” ¯ E − Eν Φˇ (E) where ν ∈ {0, π} . (28) · ξν (E) = ν ¯ ε tv,ν (E) When τ is Large Let us now assume τ 1. The location of the spectrum in I0 ∪Iπ is described by Theorem 2 ( [5]). Assume we are in the case of Theorem 1. Assume (26) is satisﬁed. Then, there exist ε0 > 0 and a nonnegative function ε → f (ε) tending to zero as ε → 0 such that, for ε ∈ (0, ε0 ], the spectrum of Hz,ε in I0 ∪ Iπ is located in two intervals Iˇ0 and Iˇπ deﬁned by Iˇ0 = {E ∈ I0 : ξ0 (E) ≤ 1 + f (ε)} and Iˇπ = {E ∈ Iπ : ξπ (E) ≤ 1 + f (ε)}. ε Let dNε (E) be the density of states measure of Hz,ε ; then dNε (E) = . π Iˇ0 ∪Iˇπ Moreover, if Iˇ0 ∩ Iˇπ = ∅, then ε (29) dNε (E) = dNε (E) = 2π ˇ ˇ I0 Iπ The Lyapunov exponent on Iˇ0 ∪ Iˇπ satisﬁes
ε Θ(E, ε) = log τ 1 + ξ0 (E) + ξπ (E) + o(1) , π where o(1) → 0 when ε → 0 uniformly in E, E0 and Eπ .
(30)
Adiabatic QuasiPeriodic Schr¨ odinger Operators
395
By (29), if Iˇ0 and Iˇπ are disjoint, they both contain spectrum of Hz,ε ; if not, one only knows that their union contains spectrum. Let us analyze the results of Theorem 2. The location of the spectrum. By (28), the intervals Iˇ0 and Iˇπ deﬁned in Theorem 2 are respectively “centered” at the points E0 and Eπ . Their lengths are given by Iˇ0  ∼
ε→0
2ε · tv,0 (E0 ) and Φˇ0 (E0 )
Iˇπ  ∼
ε→0
2ε · tv,π (Eπ ) . Φˇπ (Eπ )
Depending on Eπ − E0 , the picture of the spectrum in I0 ∪ Iπ is given by Fig. 5, case (a) and (b). The repulsion eﬀect observed in the non resonant case (see [6]) does not exist here: it is negligible with respect to the length of the intervals Iˇ0 and Iˇπ . (l)
(l )
∼ 2tv,π (Eπ )
(l )
∼ 2tv,0 (E0 )
∼ 2tv,π (Eπ )
(l)
(l)
(l)
(a) Eπ − E0
(l)
E0 Eπ
Eπ
E0
(b) Eπ − E0
max(tvπ tv0 )
max(tv,π , tv,0 )
(l )
∼ 2tv,π (Eπ ) (l)
(l)
E0 Eπ
Fig. 5. The location of the spectrum for τ large
The nature of the spectrum. In the intervals Iˇ0 and Iˇπ , according to (26) and (30), the Lyapunov exponent is positive. Hence, by the IshiiPasturKotani Theorem ( [17]), in both Iˇ0 and Iˇπ , the spectrum of Hz,ε is singular. The Lyapunov exponent Θ(E, ε) on the spectrum. The general formula (30) can be simpliﬁed in the following way:
ε Θ(E, ε) = log τ 1 + ξ0 (E) + o(1) when E ∈ Iˇπ , π and Θ(E, ε) =
ε log τ 1 + ξπ (E) + o(1) π
when E ∈ Iˇ0 .
If Eπ − E0  max(tv,π , tv,0 ), then the Lyapunov exponent stays essentially constant on each of the intervals Iˇ0 and Iˇπ . On the other hand, if Eπ −E0 
396
A. Fedotov and F. Klopp
max(tv,π , tv,0 ), then, on these exponentially small intervals, the Lyapunov exponent may vary by a constant. To see this, let us take a simple example. Assume that tv,0 tv,π , more precisely, that there exists δ > 0 such that ∀E ∈ V∗ ∩ J,
Sv,0 (E) > Sv,π (E) + δ .
If E0 and Eπ coincide, then Iˇ0 ⊂ Iˇπ , and, near the center of Iˇπ , the Lyapunov exponent assumes the value Θ(E, ε) =
1 ε ¯ − Sv,π (E) ¯ − Sv,0 (E) ¯ + o(1) . log τ + o(1) = Sh (E) π 2π
Near the edges of Iˇπ , its value is given by ε ε ¯ v,0 (E)) ¯ + o(1) log τ + log(tv,π (E)/t π 2π 1 ¯ − 2Sv,π (E) ¯ + o(1) . Sh (E) = 2π
Θ(E, ε) =
1 ¯ − Sv,π (E)) ¯ So the variation of the Lyapunov exponent is given by 2π (Sv,0 (E) on an exponentially small interval. One sees a sharp drop of the Lyapunov exponent on the interval containing spectrum when going from the edges of Iˇπ towards Eπ .
When τ is Small We now assume that τ 1, i.e., that (27) is satisﬁed. Then, the spectral behavior depends on the value of the quantity Λn (V ) deﬁned in [5]. We only recall that Λn (V ) depends solely on V and on the index of the gap separating the two interacting bands; moreover, it generically satisﬁes Λn (V ) > 1 .
(31)
Below, we only consider this generic case. There are diﬀerent possible “scenarii” for the spectral behavior when τ 1. Before describing them in detail, we start with a general description of the spectrum. We prove Theorem 3 ( [5]). Assume we are in the case of Theorem 1. Assume that (27) and (31) are satisﬁed. Then, there exists ε0 > 0 and a non negative function ε → f (ε) tending to zero when ε → 0 such that, for ε ∈]0, ε0 [, the (l) (l ) spectrum of Hz,ε in I0 ∪ Iπ is contained in Σ(ε), the set of energies E satisfying 2 τ ξ0 (E)ξπ (E) + 2Λn (V ) ≤ 2 + τ 2 ξ0 (E) + τ 2 ξπ (E) (1 + f (ε)) . (32) The set Σ(ε) is the union of two disjoint compact intervals, say I0 and Iπ ; ¯ both intervals are strictly contained inside the (2e−δ0 /ε )neighborhood of E.
Adiabatic QuasiPeriodic Schr¨ odinger Operators
397
One checks the Theorem 4 ( [5]). In the case of Theorem 1, if dNε (E) denotes the density of states measure of Hz,ε , then ε . dNε (E) = dNε (E) = 2π I0 Iπ Hence, each of the intervals I0 and Iπ contains some spectrum of Hz,ε . This implies that, when τ 1 and Λn (V ) > 1, there is a “level repulsion” or a “splitting” of resonant intervals. As for the nature of the spectrum, one shows the following results. The behavior of the Lyapunov exponent is given by Theorem 5 ( [5]). In the case of Theorem 1, on the set Σ(ε), the Lyapunov exponent of Hz,ε satisﬁes ε log τ 2 (ξ0 (E) + ξπ (E)) + 1 + o(1) , (33) Θ(E, ε) = 2π where o(1) → 0 when ε → 0 uniformly in E, E0 and Eπ . For c > 0, one deﬁnes the set
Ic+ = E ∈ Σ(ε) : ε log τ ξ0 (E) + ξπ (E) > c .
(34)
Theorem 5 and the IshiiPasturKotani Theorem imply Corollany 1. [5] In the case of Theorem 1, for ε suﬃciently small, the set Σ(ε) ∩ Ic+ only contains singular spectrum. Deﬁne
(35) Ic− = E ∈ Σ(ε) : ε log τ ξ0 (E) + ξπ (E) < −c . Theorem 3 implies that, for suﬃciently small ε, the set Ic− is contained in the set ˜ac (ε) = {E ∈ R : τ 2 ξ0 (E)ξπ (E) + 2Λn (V ) ≤ 2(1 + g(ε))} , (36) Σ where ε → g(ε) is independent of c and satisﬁes the estimate g = o(1) as ˜ac (ε) consists of I˜0 and I˜π , two disjoint intervals, and the ε → 0. The set Σ
¯ distance between these intervals is greater or equal to C ε th (E). Let Σac denote the absolutely continuous spectrum of Hz,ε . One shows Theorem 6 ( [5]). Pick ν ∈ {0, π}. In the case of Theorem 1, there exists η > 0 and D ⊂ (0, 1), a set of Diophantine numbers such that –
mes (D ∩ (0, ε)) = 1 + o e−η/ε when ε → 0 . ε – for ε ∈ D suﬃciently small, if Ic− ∩ I˜ν = ∅, then mes (I˜ν ∩ Σac ) = mes (Iν ) (1 + o(1)) , where o(1) → 0 when ε → 0 uniformly in E0 and Eπ .
398
A. Fedotov and F. Klopp
Possible Scenarii When τ is Small Assume that τ 1 and Λn (V ) > 1. Essentially, there are two possible cases for the location and the nature of the spectrum of Hz,ε . Deﬁne B max(tv,π , tv,0 ) max(tv,π , tv,0 ) √ ρ := . (37) ¯ =τ min(tv,π , tv,0 ) ¯ th E=E E=E
Note that ρ ≥ τ . We only discuss the cases when ρ is exponentially small or exponentially large.
¯ Σ(ε) is the 1. We now discuss the case ρ 1. If E√π − E0  ε th (E), union of two intervals √ of length roughly ε th ; they are separated by¯ a gap The of length roughly ε th (see Fig. 6); this gap is centered the point E. length of the intervals containing spectrum, as well as the length and center of the lacuna essentially do not change as the distance Eπ −E0 increases upto
¯ after that, the intervals containing spectrum begin a size of order ε th (E); to move away from each other. √ ε th
√ ε th
√ ε th
E0 Eπ Fig. 6. When ρ 1
As for the nature of the spectrum, when ρ is exponentially small and when
¯ the intervals containing spectrum are contained in the Eπ −E0  ε th (E), set Ic− ; so, most of the spectrum in these intervals is absolutely continuous (if ε satisﬁes the Diophantine condition of Theorem 6). 2. Consider the case ρ 1. For sake of deﬁniteness, assume that tv,0 tv,π . Then, there exists an interval, say Iπ , that is asymptotically centered at Eπ ¯ and that contains spectrum. The length of this interval is of order εtv,π (E). One can distiguish two cases: 1. if E0 belongs to Iπ and if the distance from E0 to the edges of Iπ is of the same order of magnitude as the length of Iπ , then Σ(ε) consists ¯ v,π (E) ¯ and of the interval Iπ without a “gap” of length roughly εth (E)/t containing E0 (see Fig. 7(b)). Moreover, the distance from E0 to any edge ¯ v,π (E). ¯ of the gap is also of order εth (E)/t 2. if E0 is outside Iπ and at a distance from Iπ at least of the same order of magnitude as the length of Iπ , then Σ(ε) consists in the union of Iπ and an interval I0 (see Fig. 7(a)). The interval I0 is contained in ¯ neighborhood of E0 of size roughly ε2 th (E)/E 0 − Eπ . The length of I0 2 ¯ ¯ is of size ε th (E)/E0 − Eπ  + εtv,0 (E).
Adiabatic QuasiPeriodic Schr¨ odinger Operators
399
0 ε(th /E0 − Eπ  + tv,0 ) 0 εtv,π Eπ
E0 ε(th /E0 − Eπ  + tv,0 )
ε εtv,π
E0
(a) When Eπ − E0
th tv,π
∼ εtv,π E0
Eπ
¯ εtv,π (E)
(b) When Eπ − E0
Eπ
¯ εtv,π (E)
Fig. 7. The locus of the spectrum when τ 1 and ρ 1
When ρ is exponentially large, the Lyapunov exponent may vary very quickly ¯ on the intervals containing spectrum. Consider the case Eπ −E0  εtv,π (E). For E close to the gap surrounding E0 , τ 2 ξ0 (E) is of order 1 whereas τ 2 ξπ (E) is exponentially small. Hence, Theorem 5 implies that Θ(E, ε) = o(1). On the other hand, at the external edges of the intervals containing spectrum, τ 2 ξ0 (E) is of size roughly ρ2 ; this factor being exponentially large, at such energies, the Lyapunov exponent is positive and given by Θ(E, ε) =
ε log ρ + o(1) . π
This phenomenon is similar to that observed for τ 1 except that, now, the Lyapunov exponent sharply drops to a value that is small and that may even vanish. On most of Σ(ε), the Lyapunov exponent stays positive and, the spectrum is singular (by Corollary 1); near the lacuna surrounding E0 , neither Corollary 1, nor Theorem 6 apply. These zones are similar to the zones where asymptotic Anderson transitions were found in [10]. 2.7 The Model Equation To obtain the results described in Sect. 2.5 and 2.6, the study of the spectrum of Hz,ε is reduced to the study of the ﬁnite diﬀerence equation: Ψk+1 = M (kh + z, E)Ψk ,
Ψk ∈ C2 ,
k∈Z,
(38)
where h = 2π ε mod 1, and (z, E) → M (z, E) is a function taking values in SL(2, C) (the monodromy matrix, see [10]). The behavior of the solutions to (38) mimics that of those to Hz,ε ψ = Eψ (see [10]). Equation (38) in which the matrix is replaced with its principal term is a model equation of our system. All the eﬀects we have described can be seen when analyzing this model equation.
400
A. Fedotov and F. Klopp
The asymptotic of M is decribed very precisely in [5,6]; here, we only write ¯ ≥ min tv,ν (E). ¯ down its leading term. Assume additionally that τ 2 (E) ν∈{0,π}
Then, for E ∈ I0 ∪ Iπ , M has a quite simple asymptotic 2 τ g0 (z + h/2) τ g0 (z + h/2)gπ (z) + θn−1 , M (z, E) ∼ M0 (z, E) := θn θn τ gπ (z) (39) where gν = ξν + sin(2πz),
ν ∈ {0, π} ,
and θn is the solution of 2Λn = θn + θn−1 in [1, +∞[. Introducing the monodromy matrix for the Harper equation (see [2]) τ gν (z) −1 Nν (z) = , 1 0 one notes that M0 (z, E) = N0 (z + h/2, E)Nπ (z, E) . So as suggested by the phase space picture, Fig. 2, at the energies under consideration, the system behaves like two interacting Harper equations. 2.8 When τ is of Order 1 When τ is of order 1, the principal term of the monodromy matrix asymptotics given by (39). If τ and ξ0  and ξπ  are of order of 1, the principal term does not contain any asymptotic parameter. This regime is similar to that of the asymptotic Anderson transitions found in [10]. If at least one of the “local variables” ξν (E) becomes large, then, the spectrum can again be analyzed with the same precision as in the cases τ 1 and τ 1. 2.9 Numerical Computations We now turn to some numerical results showing that the multiple phenomena described in Sect. 2.6 do occur. All these phenomena only depend on the values of the actions Sh , Sv,0 , Sv,π . We pick V to be a twogap potential; for such potentials, the Bloch quasimomentum k (see Sect. 2.1) is explicitly given by a hyperelliptic integral ( [14, 16]). The actions then become easily computable numerically. As the spectrum of H0 = −∆ + V only has two gaps, we write σ(H0 ) = [E1 , E2 ] ∪ [E3 , E4 ] ∪ [E5 , +∞). In the computations, we take the values E1 = 0, E2 = 3.8571, E3 = 6.8571, E4 = 12.1004, et E5 = 100.7092 .
Adiabatic QuasiPeriodic Schr¨ odinger Operators
401
E = E4 − α
ρ
1
ρ
1, τ
τ
1
E = E3 − α 1
E = E2 + α
E = E1 + α Fig. 8. Comparing τ and ρ to 1
On the Fig. 8, we represented the part of the (α, E)plane where the condition (TIBM) is satisﬁed for n = 1, say Λ. Its boundary consists of the straight lines E = E1 +α, E = E2 +α, E = E3 −α and E = E4 −α. The computations show that (T) is satisﬁed in the whole of ∆. As n = 1, one has E2n−2 = −∞. It suﬃces to check (T) for ζ2n+3 = ζ5 . (T) can then be understood as a consequence of the fact that E5 − E4 is large. On Fig. 8, we show the zones where τ and ρ are large and small. So, for carefully chosen α, all the phenomena described in Sect. 2.6, that is in Figs. 5, 6 and 7, do occur.
References 1. J. Avron and B. Simon. Almost periodic Schr¨ odinger operators, II. the integrated density of states. Duke Mathematical Journal, 50:369–391, 1983. 2. V. Buslaev and A. Fedotov. Spectral properties of the monodromy matrix for ´ Harper equation. In Journ´ees “Equations aux D´ eriv´ees Partielles” (SaintJean´ deMonts, 1996), pages Exp. No. IV, 11. Ecole Polytech., Palaiseau, 1996. 3. H.L. Cycon, R.G. Froese, W. Kirsch, and B. Simon. Schr¨ odinger Operators. Springer Verlag, Berlin, 1987. 4. M. Eastham. The spectral theory of periodic diﬀerential operators. Scottish Academic Press, Edinburgh, 1973. 5. A. Fedotov and F. Klopp. Strong resonant tunneling, level repulsion and spectral type for onedimensional adiabatic quasiperiodic Schr¨ odinger operator. In progress. 6. A. Fedotov and F. Klopp. Weakly resonant tunneling interactions for adiabatic quasiperiodic Schr¨ odinger operators. Preprint mathph/0408006, 2004. http://arxiv.org/abs/mathph/0408006.
402
A. Fedotov and F. Klopp
7. A. Fedotov and F. Klopp. The spectral theory of adiabatic quasiperiodic operators on the real line. Markov Process. Related Fields, 9(4):579–614, 2003. 8. A. Fedotov and F. Klopp. On the singular spectrum of one dimensional quasiperiodic Schr¨ odinger operators in the adiabatic limit. Ann. H. Poincar´e, 5:929– 978, 2004. 9. A. Fedotov and F. Klopp. On the absolutely continuous spectrum of one dimensional quasiperiodic Schr¨ odinger operators in the adiabatic limit. To appear in Trans. A.M.S, 2005. 10. Alexander Fedotov and Fr´ed´eric Klopp. Anderson transitions for a family of almost periodic Schr¨ odinger equations in the adiabatic case. Comm. Math. Phys., 227(1):1–92, 2002. 11. N.E. Firsova. On the global quasimomentum in solid state physics. In Mathematical methods in physics (Londrina, 1999), pp. 98–141. World Sci. Publishing, River Edge, NJ, 2000. 12. D. Gilbert and D. Pearson. On subordinacy and analysis of the spectrum of onedimensional Schr¨ odinger operators. Journal of Mathematical Analysis and its Applications, 128:30–56, 1987. 13. B. Helﬀer and J. Sj¨ ostrand. Multiple wells in the semiclassical limit I. Communications in Partial Diﬀerential Equations, 9:337–408, 1984. 14. A.R. It.s and V.B. Matveev. Hill operators with a ﬁnite number of lacunae. Funkcional. Anal. i Priloˇzen., 9(1):69–70, 1975. 15. Yoram Last and Barry Simon. Eigenfunctions, transfer matrices, and absolutely continuous spectrum of onedimensional Schr¨ odinger operators. Invent. Math., 135(2):329–367, 1999. 16. H. McKean and P. van Moerbeke. The spectrum of Hill’s equation. Inventiones Mathematicae, 30:217–274, 1975. 17. L. Pastur and A. Figotin. Spectra of Random and AlmostPeriodic Operators. Springer Verlag, Berlin, 1992. 18. Leonid Pastur and Alexander Figotin. Spectra of random and almostperiodic operators, volume 297 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. SpringerVerlag, Berlin, 1992. 19. Barry Simon. Instantons, double wells and large deviations. Bull. Amer. Math. Soc. (N.S.), 8(2):323–326, 1983. 20. M. Wilkinson. Critical properties of electron eigenstates in incommensurate systems. Proc. Roy. Soc. London Ser. A, 391(1801):305–350, 1984.
Low Lying Eigenvalues of Witten Laplacians and Metastability (After HelﬀerKleinNier and HelﬀerNier) Bernard Helﬀer∗ D´epartement de Math´ematiques Bˆ at. 425, Universit´e Paris Sud, UMR CNRS 8628, F91405 Orsay Cedex (FRANCE) [emailprotected] Abstract. The aim of this lecture is to present the recent results obtained in collaboration with M. Klein and F. Nier on the low lying eigenvalues of the Laplacian attached to the Dirichlet form : ∇v(x)2 e−2f (x)/h dx , C0∞ (Ω) v → h2 Ω ∞
where f is a C Morse function on Ω and h > 0. We give in particular an optimal asymptotics as h → 0 of the lowest strictly positive eigenvalue, which will hold under generic assumptions. We discuss also some aspects of the proof.
1 Main Goals and Assumptions We are interested in the low lying, actually exponentially small, eigenvalues of the Dirichlet realization of the semiclassical Witten Laplacian on 0forms (0)
2
∆f,h = −h2 ∆ + ∇f (x) − h∆f (x) .
(1)
This question appears quite naturally when analyzing the asymptotic behav(0) ior as t → +∞ of exp −t∆f,h . We would like to describe the recent results concerning the lowest strictly positive eigenvalue of this Laplacian. We will discuss brieﬂy three cases : – the case of Rn , which was analyzed by BovierEckhoﬀGayrardKlein [1], BovierGayrardKlein [2], and by HelﬀerKleinNier [6]; – the case of a compact riemannian manifold of dimension n ( [6]); – The case of a bounded set Ω in Rn with regular boundary treated by HelﬀerNier [8]) (in this case, we consider the Dirichlet realization of this operator). In all these contributions, the goal is to get the optimal accuracy asymptotic (0) formulas for the m0 ﬁrst eigenvalues of the Dirichlet realization of ∆f,h . ∗
Partially supported by the programme SPECT (ESF) and by the PROGRAMME HPRNCT200200277 B. Helﬀer: Low Lying Eigenvalues of Witten Laplacians and Metastability (After HelﬀerKleinNier and HelﬀerNier), Lect. Notes Phys. 690, 403–415 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
404
B. Helﬀer
This question was already addressed a long time ago, via a probabilistic approach, sometimes in relation with the problem of the simulated annealing. Here we would like to mention FreidlinWentcel [4], HolleyKusuokaStrook [11], Miclo [14], Kolokoltsov [13], BovierEckhoﬀGayrardKlein [1] and BovierGayrardKlein [2], but the proof, as mentioned for example in [13], of the optimal accuracy (except may be for the case of dimension 1) was left open. (0) The Witten Laplacian ∆f,h is associated to the Dirichlet form 2 ∞ (h∇ + ∇f )u(x) dx . C0 (Ω) u → Ω
Note that the probabilists look equivalently at: 2 ∞ 2 C0 (Ω) v → h ∇v(x) e−2f (x)/h dx . Ω
Let us now describe the main assumptions. In the whole paper, we assume that: Assumption 1 The function f is a C ∞  function on Ω and a Morse function on Ω. In the case when Ω = Rn , the operator deﬁned on C0∞ is essentially selfadjoint in L2 and if Assumption 2 lim inf ∇f (x)2 > 0 ,
x→+∞
and Dxα f  ≤ Cα (∇f 2 + 1) , for α = 2, is satisﬁed, Pearson’s theorem permits to prove that the bottom of the essential spectrum is, for h small enough, strictly positive. In the case with boundary, we will assume: Assumption 3 The function f has no critical points at the boundary and the function f/∂Ω is a Morse function on ∂Ω. The last assumption, which appears to be generic, is more diﬀﬁcult to describe and will be presented in the next section.
2 Saddle Points and Labelling The presentation of the results involves a right deﬁnition for the saddle points (or critical points of index 1). If for a point in Ω, we take the usual deﬁnition
Witten Laplacians and Metastability
405
(the index at a critical point U being deﬁned as the number of negative eigenvalues of the Hessian of f at U ), we shall say that a point U at the boundary is a critical point of index 1 if U is a local minimum of f/∂Ω and if the external normal derivative of f is strictly positive. Our statements also involve a labelling of the local minima, which is inspired by previous works by probabilists [1, 4, 14] and [2]. The existence of such a labelling is an assumption which is generically satisﬁed. This discussion can be done in the three cases. Let us focus on the case of a compact connected oriented Riemannian manifold Ω = Ω ∪∂Ω with boundary and that the function f satisﬁes Assumptions 1 and 3 and refer to the original papers [6, 8] (and references therein) for the other cases or for details. The set of critical points with index 1 is denoted by U (1) and its cardinal by m1 . For the deﬁnition of the saddle set (or set of saddle points), we need some notations. For any A, B ⊂ Ω, H(A, B) denotes the least height to be reached to go continuously from A to B. When A and B are closed nonempty subsets of Ω, one can show that H(A, B) is a minimum. We next need two deﬁnitions. Deﬁnition 1. Let A and B be two closed subsets of Ω. We say that Z ⊂ Ω is a saddle set for (A, B) , if it is not empty and satisﬁes the following conditions: Z ⊂ U (1) ∩ f −1 ({H(A, B)}) , . C ∈ Conn f −1 ((−∞, H(A, B)]) \ Z , C ∩ A = ∅, C ∩ B = ∅ = ∅ . Deﬁnition 2. Let A, B be closed nonempty disjoint subsets of Ω . The point z ∈ U (1) is said to be a unique (one point)saddle set2 for the pair (A, B) if # $ (∩C∈C(A,B) C) ∩ U (1) ∩ f −1 (H(A, B)) = {z} , where C(A, B) denotes the set of closed connected sets C ⊂ f −1 ((−∞, H (A, B)]) , such that C ∩ A = ∅ and C ∩ B = ∅. We now give the main assumption which ensures that each exponentially (0) small eigenvalue of ∆f,h is simple, with a diﬀerent asymptotic behavior. Assumption 4 There existsa labelling of the set of the local minima of f (0) (0) in Ω, U (0) := U1 , . . . , Um0 , such that, by setting (0) (0) C0 = ∂Ω and Ck = Uk , . . . , U1 ∪ C0 , for k ≥ 1 , we have: 2
Or more shortly, a unique saddle point,
406
B. Helﬀer (0)
(i) For all k ∈ {1, . . . , m0 } , Uk
is the unique minimizer of
H(U, Ck \ {U }) − f (U ), U ∈ Ck \ C0 . (0) (ii) For all k ∈ {1, . . . , m0 } , the pair {Uk }, Ck−1 admits a unique (one point)saddle set {zk∗ } . It is possible to check that this hypothesis is generically satisﬁed. More precisely, it is satisﬁed if all the critical values of f are distinct and all the quantities f (U (1) ) − f (U (0) ), with U (1) ∈ U (1) and U (0) ∈ U (0) are distinct. By its deﬁnition, the point zk∗ is a critical point with index 1, zk∗ ∈ U (1) . Deﬁnition 3. (The map j) If the critical points of index 1 are denoted by (1) Uj , j = 1, . . . , m1 , we deﬁne the application k → j(k) on {1, . . . , m0 } by: Uj(k) = zk∗ . (1)
(2)
With these deﬁnitions, one can prove : Proposition 1. Under Assumption 4,the following properties are satisﬁed : (1) (0) (a) The sequence f (Uj(k) ) − f (Uk ) is strictly decreasing. k∈{1,...,m0 }
(b) The application j : {1, . . . , m0 } → {1, . . . , m1 } is injective.
3 Rough SemiClassical Analysis of Witten Laplacians and Applications to Morse Theory 3.1 Previous Results It is known (see [10,17,18] and more recently [3,8]) that the Witten Laplacian 6 (0) on functions ∆f,h admits in the interval [0, h 5 ] and for h > 0 small enough exactly m0 eigenvalues, where m0 is the number of local minima of f in Ω . (0) This is easy to guess by considering, near each of the local minima Uj , of f the function (0)
wkb,(0)
uj
(x) := χj (x) exp −
f (x) − f (Uj ) , h (0)
(3)
where χj is a suitable cutoﬀ function localizing near Uj as suitable quasimode. This shows, via the minimax principle, that these eigenvalues are actually exponentially small as h → 0 . Note that we consider the Dirichlet problem. So the ﬁrst part of Assumption 3 implies that the eigenfunctions corresponding to low lying eigenvalues are localized far from the boundary.
Witten Laplacians and Metastability
407
3.2 Witten Laplacians on pForms (0)
Although we are mainly interested in the operator ∆h,f , we will also meet in the proof other Laplacians. The spectral analysis can be extended (see Simon [17], Witten [18], HelﬀerSj¨ ostrand [9], ChangLiu [3]) to Laplacians on pforms, p ≥ 1. We recall that Witten is ﬁrst introducing a dixtorted De Rham complex d : f f df,h = exp − (hd) exp = hd + df ∧ , h h
(4)
(p)
The restriction of df,h to pforms is denoted by df,h . Witten then associates to this complex the Laplacian : ∆f,h = (df,h + d∗f,h )2 ,
(5)
where d∗f,h is its L2 adjoint. By restriction to the pforms, one gets the Witten (p)
Laplacian ∆f,h . 3.3 Morse Inequalities In the compact case, the analysis of the low lying eigenvalues of the Witten Laplacians was the main point of the semiclassical proof suggested by Witten of the Morse inequalities [10, 17, 18]. Each of the Witten Laplacians is essentially selfadjoint and an analysis based on the harmonic approximation (consisting in replacing f by its quadratic approximation at a critical point of f ) shows that the dimension 6 of the eigenspace corresponding to [0, h 5 ] is, for h small enough, equal to mp the number of critical points of index p. (p) Note that the dimension of the kernel of ∆h,f being equal to the Betti number bp , this gives immediately the so called “ Weak Morse Inequalities” : bp ≤ mp , for all p ∈ {0, . . . , n} .
(6)
4 Main Result in the Case of Rn In the case of Rn and under Assumptions 1, 2 and 4, the main result is the following : (0)
Theorem 5. The ﬁrst eigenvalues λk (h), k ∈ {2, . . . , m0 }, of ∆f,h have the following expansions : B (0) det(Hessf (Uk )) (1) h H λk (h) = λ1 (U ) π
× exp − h2 with r1 (h) = o(1).
j(k)
(1)
det(Hessf (Uj(k) ))
(1) (0) f (Uj(k) ) − f (Uk ) × (1 + r1 (h)) ,
408
B. Helﬀer (0)
Here the Uk denote the local minima of f ordered in some speciﬁc way (see (1) Sect. 2), the Uj(k) are “saddle points” (critical points of index 1), attached to (0) H1 (U (1) ) is the negative eigenvalue of Hessf (U (1) ). U via the map j, and λ k
j(k)
j(k)
Actually, the estimate 1
r1 (h) = O(h 2  log h) , is obtained in [2] (under weaker assumptions on f ) and the complete asymptotics, r1j hj , r1 (h) ∼ j≥1
is proved in [6]. In the above statement, we have left out the case k = 1, which leads to a speciﬁc assumption (see Assumption 2) in the case of Rn for f at ∞. This (0) implies that ∆f,h is essentially selfadjoint and that the bottom of the essential spectrum is bounded below by some 0 > 0 (independently of h ∈]0, h0 ], h0 small enough). If the function exp − fh is in L2 , then λ1 (h) = 0 . In this case, denoting by Π0 the orthogonal projector on exp − fh , the main motivation for having a good control of λ2 (h) is the estimate, for t > 0, (0)
exp −t∆h,f − Π0 L(L2 (Ω)) ≤ exp −tλ2 (h) . In other words, the estimate of λ2 (h) permits to measure the rate of the return to equilibrium. Note ﬁnally that other examples like f (x) = −(x2 − 1)2 (with n = 1) are interesting and an asymptotic of λ1 (h) can be given for this example.
5 About the Proof in the Case of Rn 5.1 Preliminaries The case of Rn requires some care (see [5, 12] or [7]) for controlling the problem at inﬁnity. The approach given in [6] intensively uses, together with the techniques of [10], the two following facts : – The Witten Laplacian is associated to a cohom*ology complex. – The function exp − f (x) h is, as a distribution, in the kernel of the Witten Laplacian on 0−forms. This permits to construct very easily and eﬃciently – and this is speciﬁc (0) of the case of ∆f,h – quasimodes. We note that we have between diﬀerential operators acting on C0∞ the relations
Witten Laplacians and Metastability (0)
(0)
(1) (0)
df,h ∆f,h = ∆f,h df,h , and
(0) (0),∗
(0),∗
409
(7)
(1)
∆f,h df,h = df,h ∆f,h ,
(8) (0)
This shows in particular that if u is an eigenvector of ∆h,f for an eigenvalue (1)
λ = 0, then dh,f u is an eigenvector of ∆f,h for λ. 5.2 Witten Complex, Reduced Witten Complex It is more convenient to consider the singular values of the restricted diﬀerential df,h : F (0) → F (1) . The space F () is the m dimensional spectral () subspace of ∆f,h , ∈ {0, 1}, ()
F () = Ran 1I(h) (∆f,h ) , 6
with I(h) = [0, h 5 ] and the property (1)
(0)
1I(h) (∆f,h )df,h = df,h 1I(h) (∆f,h ) . We will analyze : ()
()
βf,h := (df,h )/F () . We will mainly concentrate on the case = 0. 5.3 Singular Values In order to exploit all the information which can be extracted from well (0) chosen quasimodes, working with singular values sj of βf,h happens to be more eﬃcient than considering their squares, that is the eigenvalues λj = s2j of (0)
∆f,h . The main point3 is probably that one can choose suitable approximate well localized “almost” orthogonal basis of F (0) and F (1) separately and that the errors appear “multiplicatively” when computing the singular values (0) of βf,h . By this we mean : sj = sapp j (1 + ε1 (h)) , instead of additively 2 λj := s2j = (sapp j ) + ε2 (h) ,
as for example in [10]. Here sjapp will be explicitly obtained from the WKB analysis. In the ﬁrst case, it is actually enough to prove that ε1 (h) = O(h∞ ). 3
See also [15] for a pedagogical discussion,
410
B. Helﬀer
In the second case, the analysis of [10] gives a control of ε2 (h) in O(exp − Sh ), (1) (0) with S > 2 inf j,k (f (Uj ) − f (Uk )), which is enough for estimating the highest low lying eigenvalue (see [7]) but could be unsatisfactory for the lowest strictly positive eigenvalue, as soon as the number of local minima is > 2. Although it is not completely hopeless to have a better control of ε2 (h) by improving the analysis of [10] and introducing a reﬁned notion of non resonant wells, the approach developed in [8] appears to be simpler.
6 The Main Result in the Case with Boundary In the case with boundary, the function exp − fh , which is the only distrib(0) ution in the kernel of ∆h,f , does not satisfy the Dirichlet condition, so the smallest eigenvalue can not be 0. The estimate of λ1 (h) permits consequently (0) to measure the decay of  exp −t∆h,f  ≤ exp −tλ1 (h) as t → +∞. For this case, a starting reference is the book by FreidlinWentzel [4], which says (in particular) that, if f has a unique critical point, corresponding (0) to a non degenerate local minimum Umin , then the lowest eigenvalue λ1 (h) (0) of the Dirichlet realization ∆f,h in Ω satisﬁes : (0)
lim −h log λ1 (h) = 2 inf (f (x) − f (Umin )) .
h→0
x∈∂Ω
Other results are given in this book for the case of many local minima but they are again limited to the determination of logarithmic equivalents. We have explained in Sect. 2 that, under Assumption 4, one can label the m0 local minima and associate via the map j from the set of the local minima into the set of the m1 (generalized) saddle points of the Morse functions in Ω of index 1 . The main theorem of HelﬀerNier [8] is : Theorem 6. Under Assumptions 1, 3 and 4, there exists h0 such that, for 3 (0) h ∈ (0, h0 ] , the spectrum in [0, h 2 ) of the Dirichlet realization of ∆f,h in Ω, consists of m0 eigenvalues λ1 (h) < . . . < λm0 (h) of multiplicity 1, which are exponentially small and admit the following asymptotic expansions : B (0) det(Hessf (Uk )) (1) h H 1 + hc1 (h) λk (h) = λ1 (U ) π
× exp − h2
j(k)
(1) (0) f (Uj(k) ) − f (Uk ) ,
and (1)
λk (h) =
2h1/2 ∇f (Uj(k) ) π 1/2
k
(1)
det(Hessf (Uj(k) ))
B
(1)
if Uj(k) ∈ Ω ,
(0) det(Hessf (Uk )) (1) det(Hessf (Uj(k) ))
(1) (0) × exp − h2 f (Uj(k) ) − f (Uk ) ,
1 + hc1k (h)
∂Ω
(1)
if Uj(k) ∈ ∂Ω .
Witten Laplacians and Metastability
411
Here c1k (h) admits a complete expansion : c1k (h) ∼
∞
hm ck,m .
m=0
This theorem extends to the case with boundary the previous results of [2] and [6] (see also the books [4] and [13] and references therein).
7 About the Proof in the Case with Boundary As in [10], the proof is deeply connected with the analysis of the small eigenvalues of a suitable realization (which is not the Dirichlet realization) of the Laplacian on the 1forms. In order to understand the strategy, three main points have to be explained. 7.1 Deﬁne the Witten Complex and the Associate Laplacian The case of a compact manifold without boundary was treated in the foundational paper of Witten [18]. A ﬁner (and rigorous) analysis is then given in [10] and further developments appear in [6]. The case with boundary creates speciﬁc new problems. Our starting problem being the analysis of the Dirichlet realization of the Witten Laplacian on functions, we were let to ﬁnd the right realization of the Witten Laplacian on 1forms in the case with boundary in order to extend the commutation relation (7) in a suitable “strong” sense (at the level of the selfadjoint realizations). The answer was actually present in the literature [3] in connection with the analysis of the relative cohom*ology and the proof of the Morse inequalities. Let us explain how we can guess the right condition by looking at the eigenvectors. (0) If u is eigenvector of the Dirichlet realization of ∆f,h , then by commu(0)
tation relation, df,h u (which can be identically 0) should be an eigenvector (1)
(0)
in the domain of the realization of ∆f,h . But df,h u does not necessarily satisfy the Dirichlet condition in all its components, but only in its tangential components. This is the natural condition that we keep in the deﬁnition of the varia(1) (0) ∗ tional domain to take for the quadratic form ω → df,h ω2 + df,h ω2 . (1) DT
The selfadjoint realization ∆f,h obtained as the Friedrichs extension associated to the quadratic form gives the right answer. (0),∗ (0) Observing also that df,h (df,h u) = λu (with λ = 0), we get the second natural (Neumann type)boundary condition saying that a one form ω in the (1) DT domain of the operator ∆f,h should satisfy
412
B. Helﬀer (0),∗
df,h ω/∂Ω = 0 .
(9)
So we have shown that the natural boundary conditions for the Witten Laplacians are (9) together with ω/∂Ω = 0 .
(10)
The associated cohom*ology is called relative (see for example [16]). 7.2 Rough Localization of the Spectrum of this Laplacian on 1Forms (p)
The analysis of ∆f,h was performed in [3], in the spirit of Witten’s idea, extending the so called Harmonic approximation. But these authors, because they were interested in the Morse theory, used the possibility to add symplifying assumptions on f and the metric near the boundary. We emphasize that [8] treats the general case. May be one could understand what is going on at the boundary by analyzing the models corresponding to f (x , xn ) = 12 x 2 + xn , with = ±1 in Rn+ = {xn > 0}. The analysis in this case is easily reduced to the analysis of the one dimension case on R+ (together with the standard analysis of Rn−1 ). The Dirichlet Laplacian to analyze is simply : d2 −h2 2 + 1 , dx + on R , which is strictly positive, but the Laplacian on 1forms is u(x) dx → (−h2 u (x) + u(x)) dx , on R+ , but with the boundary condition : hu (0) − u(0) = 0 . Depending on the sign of , the bottom of the spectrum is 0 if < 0 or 1 if
> 0. This explains our deﬁnition of critical point of index 1 at the boundary. 7.3 Construction of WKB Solutions Attached to the Critical Points of Index 1 The construction of the approximate basis of F (0) and F (1) is obtained through WKB constructions. The constraints are quite diﬀerent in the two cases. For F (0) , we need rather accurate quasimodes but can take advantage of their simple structure given in (3). The diﬃculty is concentrated in the choice of χj . For F (1) , it is enough to construct quasimodes lecalized in a small neighborhood of a critical point of index 1. This was done in [10] for the case without boundary, as an extension of previous constructions of [9].
Witten Laplacians and Metastability
413
The new point is the construction of WKB solutions near critical points of the restriction of the Morse function at the boundary, which is done in [8] for 1forms. Let us explain the main lines of the construction. The construction is done locally around a local minimum U0 of f ∂Ω with ∂n f (U0 ) > 0 . The function Φ is a local solution of the eikonal equation 2
2
∇Φ = ∇f  , which also satisﬁes Φ=f
on
∂Ω
and ∂n Φ = −∂n f
on
∂Ω
and we normalize f so that f (U0 ) = f (0) = 0 . near the point x = 0 of We ﬁrst consider a local solution uwkb 0 e h ∆f,h uwkb = O(h∞ ) , 0 Φ
with uwkb in the form 0
(0)
uwkb = a(x, h)e− h , 0 aj (x)hj , a(x, h) ∼ Φ
j≥0
and the condition at the boundary f
a(x, h)e− h = e− h Φ
on
∂Ω ,
which leads to the condition a(x, h)∂Ω = 1 . In order to verify locally the boundary condition for our future uwkb 1 , we f −h and still obtain substract e f
e h ∆f (uwkb − e− h ) = O(h∞ ) . 0 Φ
(0)
(11)
We now deﬁne the WKB solution uwkb by considering : 1 f
uwkb := df,h uwkb = df,h (uwkb − e− h ) . 1 0 0 The 1form uwkb = df,h uwkb satisﬁes locally the Dirichlet tangential condition 1 0 on the boundary (10) and, using (11), (modulo O(h∞ )) the Neumann type gives a good approximation for a ground condition (9) is satisﬁed. So uwkb 1 (1) state of a suitable realization of ∆f,h in a neighborhood of this boundary critical point.
414
B. Helﬀer
Acknowledgements The author would like to thank F. Nier for fruitful collaboration on this subject and useful comments on the manuscript. He also thanks the organizers of the meeting J. Asch and A. Joye.
References 1. A. Bovier, M. Eckhoﬀ, V. Gayrard, and M. Klein : Metastability in reversible diﬀusion processes I. Sharp asymptotics for capacities and exit times. J. of the Europ. Math. Soc. 6 (4), pp. 399–424 (2004). 2. A. Bovier, V. Gayrard, and M. Klein. Metastability in reversible diﬀusion processes II Precise asymptotics for small eigenvalues. J. Eur. Math. Soc. 7, pp. 69–99 (2005). 3. Kung Ching Chang and Jiaquan Liu. A cohom*ology complex for manifolds with boundary. Topological methods in non linear analysis. Vol. 5, pp. 325– 340 (1995). 4. M.I. Freidlin and A.D. Wentzell. Random perturbations of dynamical systems.Transl. from the Russian by Joseph Szuecs. 2nd ed. Grundlehren der Mathematischen Wissenschaften. 260. New York (1998). 5. B. Helﬀer. Semiclassical analysis, Witten Laplacians and statistical mechanics. World Scientiﬁc (2002). 6. B. Helﬀer, M. Klein, and F. Nier. Quantitative analysis of metastability in reversible diﬀusion processes via a Witten complex approach. Proceedings of the Symposium on Scattering and Spectral Theory. Proceedings of the Symposium on Scattering and Spectral Theory. Matematica Contemporˆ anea (Brazilian Mathematical Society), Vol. 26, pp. 41–86 (2004). 7. B. Helﬀer and F. Nier. Hypoellipticity and spectral theory for FokkerPlanck operators and Witten Laplacians. Lecture Notes in Mathematics 1862, Springer Verlag (2005). 8. B. Helﬀer and F. Nier. Quantitative analysis of metastability in reversible diﬀusion processes via a Witten complex approach : the case with boundary. To appear in M´emoires de la SMF (2006) 9. B. Helﬀer and J. Sj¨ ostrand. Multiple wells in the semiclassical limit I, Comm. Partial Diﬀerential Equations 9 (4), pp. 337–408 (1984). 10. B. Helﬀer and J. Sj¨ ostrand. Puits multiples en limite semiclassique IV Etude du complexe de Witten . Comm. Partial Diﬀerential Equations 10 (3), pp. 245– 340 (1985). 11. R. Holley, S. Kusuoka, and D. Stroock. Asymptotics of the spectral gap with applications to the theory of simulated annealing. J. Funct. Anal. 83 (2), pp. 333–347 (1989). 12. J. Johnsen. On the spectral properties of Witten Laplacians, their range projections and BrascampLieb’s inequality. Integral Equations Operator Theory 36(3), pp. 288–324 (2000). 13. V.N. Kolokoltsov. Semiclassical analysis for diﬀusions and stochastic processes. Lecture Notes in Mathematics 1724. Springer Verlag, Berlin 2000. 14. L. Miclo. Comportement de spectres d’op´erateurs a ` basse temp´erature. Bull. Sci. Math. 119, pp. 529–533 (1995).
Witten Laplacians and Metastability
415
15. F. Nier. Quantitative analysis of metastability in reversible diﬀusion processes via a Witten complex approach. Proceedings of the PDE conference in ForgesLesEaux (2004). 16. G. Schwarz. Hodge decomposition. A method for solving boundary value problems. Lect. Notes in Mathematics 1607, Springer (1995). 17. B. Simon. Semiclassical analysis of low lying eigenvalues, I.. Nondegenerate minima: Asymptotic expansions. Ann. Inst. H. Poincar´e (Section Phys. Th´eor.) 38, pp. 296–307 (1983). 18. E. Witten. Supersymmetry and Morse inequalities. J. Diﬀ. Geom. 17, pp. 661– 692 (1982).
The Mathematical Formalism of a Particle in a Magnetic Field Marius M˘ antoiu1 and Radu Purice2 1
2
“Simion Stoilow” Institute of Mathematics, Romanian Academy [emailprotected] “Simion Stoilow” Institute of Mathematics, Romanian Academy [emailprotected]
1 Introduction In this review article we develop a basic part of the mathematical theory involved in the description of a particle (classical and quantal) placed in the Euclidean space RN under the inﬂuence of a magnetic ﬁeld B, emphasising the structure of the family of observables. The classical picture is known, see for example [21]; we present it here for the convenience of the reader, in a form wellﬁtted for the passage to the quantum counterpart. In doing this we shall emphasize a manifestly gauge invariant Hamiltonian description [27] that is less used, although it presents many technical advantages and a good starting point for quantization. The main contribution concernes the quantum picture. Up to our knowledge, until recently the single right attitude towards deﬁning quantum observables when a nonconstant magnetic ﬁeld is present can be found in a remarkable old paper of Luttinger [14] (we thank Gh. Nenciu for pointing it out to us); still this was undeveloped and with a limited degree of generality. In recent years, the solution to this problem appeared in two related forms: (1) a gauge covariant pseudodiﬀerential calculus in [8, 9, 17, 19] and (2) a C ∗ algebraic formalism in [16] and [19]. We cite here also the results in [22], where a gauge independent perturbation theory is elaborated for the resolvent of a magnetic Schr¨odinger Hamiltonian, starting from an observation in [3]. For the classical picture, we deﬁne a perturbed symplectic form on phase space [27] and study the motions deﬁned by classical Hamiltonians with respect to the associated perturbed Poisson algebra. The usual magnetic momenta appear then as momentum map for the associated ’symplectic translations’. The quantum picture is treated in detail; two points of view are addopted: The ﬁrst is to preserve (essentially) the same set of functions as observables, but with a diﬀerent algebraic structure. The main input is a new, (B, )dependent multiplication law associated to the perturbed symplectic form deﬁned for the classical theory. This new product converges in a suitable sense to poinwise (classical) multiplication when → 0. And it collapses for B = 0 to the symbol multiplication of Weyl and Moyal, familiar from pseudodiﬀerential theory. It depends on no choice of a vector potential, so it M. M˘ antoiu and R. Purice: The Mathematical Formalism of a Particle in a Magnetic Field, Lect. Notes Phys. 690, 417–434 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
418
M. M˘ antoiu and R. Purice
is explicitely gauge invariant. Asside the pseudodiﬀerential form, we present also a form comming from the theory of twisted crossed product C ∗ algebras and justiﬁed by interpreting our physical system as a dynamical system given by an action twisted by the magnetic ﬁeld. The second point of view, more conventional, is in terms of selfadjoint operators in some Hilbert space. One achieves this by representing the previously mentioned intrinsic structures, and this is done by choosing vector potentials A generating the magnetic ﬁeld B. For diﬀerent but equivalent choices one gets unitarily equivalent representations, a form of what is communly called “gauge covariance”. The represented form is bestsuited to the interpretation in terms of magnetic canonical commutation relations. A functional calculus is associated to this highly noncommutative family of operators. Actually, the twisted dynamical system mentioned above (a sort of twisted imprimitivity system) is equivalent to these commutation rules. The limit → 0 of the quantum system was studied in [18], in the framework of Rieﬀel’s strict deformation quantization. To show that the formalism is useful in applications, we dedicate a section to spectral theory for anisotropic magnetic operators, following [20]. This relies heavily on an aﬃliation result, saying that the resolvent family of a magnetic Schr¨odinger Hamiltonian belongs to a suitable C ∗ algebra of magnetic pseudodiﬀerential operators. Recently, the usual pseudodiﬀerential theory has been generalized to a groupoid setting, cf. [13, 23] and references therein; this is in agreement with modern trends in deformation quantization, cf. [10] for example. The right concept to include magnetic ﬁelds should be that of twisted groupoid, as appearing in [30], accompanied by the aﬀerent C ∗ algebras. Let us also mention here the possibility to use our general framework in dealing with nonabelian gauge theories.
2 The Classical Particle in a Magnetic Field In this section we shall give a classical bakground for our quantum formalism. We use the setting and ideas in [21] but develop the gauge invariant Poisson algebra feature. We begin by very brieﬂy recalling the usual Hamiltonian formalism for classical motion in a magnetic ﬁeld and then change the point of view by perturbing the canonical symplectic structure. 2.1 Two Hamiltonian Formalisms The basic fact provided by physical measurements is that the magnetic ﬁeld in R3 may be described by a function B : R3 → R3 with divB = 0, such that the motion R t → q(t) ∈ R3 of a classical particle (mass m and electric charge e) is given by the equation of motion deﬁned by the Lorentz force:
The Mathematical Formalism of a Particle in a Magnetic Field
m¨ q (t) = eq(t) ˙ × B(q(t))
419
(1)
where × is the antisymmetric vector product in R3 and the point denotes derivation with respect to time. An important fact about this equation of motion is that it can be derived from a Hamilton function, the price to pay being the necessity of a vector potential, i.e. a vector ﬁeld A : R3 → R3 such that B = rotA, that is unfortunately not uniquely determined. Let us very brieﬂy recall the essential facts concerning the Hamiltonian formalism. Given a smooth manifold X we associate to it its “phase space” deﬁned as the cotangent bundle T∗ X on which we have a canonical symplectic form, that we shall denote by σ. If we set Π : T[T∗ X] → T∗ X and π ˜ : T∗ X → ∗ ˜, X the canonical projections and π ˜∗ : T[T X] → TX the tangent map of π then σ := dβ where β(ξ) := [Π(ξ](˜ π∗ (ξ)), for ξ a smooth section in T[T∗ X]. A Hamiltonian system is determined by a Hamilton function h : T∗ X → R (supposed to be smooth) such that the vector ﬁeld associated to the law of motion of the system (R t → x(t) ∈ T∗ X) is given by the following ﬁrst order diﬀerential equation ξσ − dh = 0, where ξσ is the oneform deﬁned by (ξσ)(η) := σ(ξ, η), for any η smooth section in T[T∗ X]. Let us take X = R3 such that all the above bundles are trivial and we have canonical isomorphisms T∗ X ∼ = X ×X ∗ (that we shall also denote by Ξ) ∗ ∗ and T[T X] ∼ = (X × X ) × (X × X ∗ ), deﬁned by the usual transitive action of translations on X; we can view any two sections ξ and η as functions ξ(q, p) = (x(q, p), k(q, p)), η(q, p) = (y(q, p), l(q, p)) and we can easily verify that σ(ξ, η) = k · y − l · x, with ξ · y the canonical pairing X ∗ × X → R. Moreover, the equations of motion deﬁned by a Hamilton function h become: q˙j = ∂h/∂pj , p˙j = −∂h/∂qj .
(2)
Then (1) may be written in the above form if one chooses a vector potential A such that B = rotA and deﬁnes the Hamilton function hA (q, p) := (2m)−1
3
(pj − eAj (q))2 .
j=1
Although very useful, this Hamiltonian description has the drawback of involving the choice of a vector potential. Two diﬀerent choices A and A have to satisfy rot(A − A ) = 0. Since R3 is simply connected, there exists a function ϕ : R3 → R with A = A + ∇ϕ and any such choice is admissible. We call these changes of descriptions “gauge transformations”; the “gauge group” is evidently C ∞ (X) and the action of the gauge group is given by hA → hA . An interesting fact is that we can actually obtain an explicitely gauge invariant description by using a perturbed symplectic form on T∗ X [27]. For that it is important to notice that the magnetic ﬁeld may in fact be described as a 2form (a ﬁeld of antisymmetric bilinear functions on R3 ), due to the obvious isomorphism between R3 and the space of antisymmetric matrices
420
M. M˘ antoiu and R. Purice
on R3 (just take Bjk := jkl Bl with jkl the completely antisymetric tensor of rank 3 on R3 ). Thus from now on we shall consider the magnetic ﬁeld B given by a smooth section of the vector bundle Λ2 X → X (the ﬁbre at x being T∗x X ∧T∗x X ∼ = [Tx X ∧Tx X]∗ ). Due to the canonical global trivialisation discussed above (deﬁned by translations) we can view B as a smooth map B : X → X∗ ∧ X∗ ∼ = (X ∧ X)∗ . Then a vector potential is described by a ∗ 1form A : X → X such that B = dA where d is the exterior diﬀerential. This also allow us to consider the case X = RN for any natural number N . Any kform on X may be considered as a kform on T∗ X. Explicitely, using the projection π ˜ : T∗ X → X, we may canonically deﬁne the pull∗ back π ˜ B of B and the “perturbed symplectic form” on T∗ X deﬁned by the ˜ ∗ B. magnetic ﬁeld B as σB := σ + π Now let us brieﬂy recall the construction of the Poisson algebra associated to a symplectic form. We start from the trivial fact that any nondegenerate bilinear form Σ on the vector space Ξ deﬁnes a canonical isomorphism iΣ : Ξ → Ξ ∗ by the equality [iΣ (x)](y) := Σ(x, y). Then we deﬁne the follow−1 ing composition law on C ∞ (X): {f, g}B := σB (i−1 σB (df ), iσB (dg)), called the Poisson braket. The case B = 0 gives evidently the canonical Poisson braket {., .} on the cotangent bundle. A computation gives immediately {f, g}B =
N
N ∂pj f ∂qj g − ∂qj f ∂pj g + e Bjk (·) ∂pj f ∂pk g .
j=1
(3)
j,k=1
For the usual Hamilton function of the free classical particle h(p) := N " (2m)−1 p2j , we can write down the Poisson form of the equation of motion: j=1
q˙j = {h, qj }B =
1 m pj
p˙j = −{h, pj }B =
e m
, N "
Bkj (q)pk ,
(4)
k=1
that combine to the equation of motion (1) deﬁned by the Lorentz force. We remark ﬁnally that in the present formulation the Hamilton function of " the free particle h(q, p) = (2m)−1 p2j is no longer privileged; any Hamilton function is now a candidate for a Hamiltonian system in a magnetic ﬁeld just by considering it on the phase space endowed with the magnetic symplectic form. The relativistic kinetic energy h(p) := (p2 + m2 )1/2 is a physically interesting example. Remark. The real linear space C ∞ (Ξ; R) endowed with the usual product of functions and the magnetic Poisson braket {., .}B form a Poisson algebra (see [10, 18]), i.e. (C ∞ (Ξ; R), ·) is a real abelian algebra and {., .}B : C ∞ (Ξ; R) × C ∞ (Ξ; R) → C ∞ (Ξ; R) is an antisymetric bilinear composition law that satisﬁes the Jacobi identity and is a derivation with respect to the usual product.
The Mathematical Formalism of a Particle in a Magnetic Field
421
2.2 Magnetic Translations For the perturbed symplectic form on T∗ X, the usual translations are no longer symplectic. We intend to deﬁne “magnetic symplectic translations” and compute the associated momentum map. Using the canonical global trivialisation, we are thus looking for an action X x → αx ∈ Diﬀ(X × X ∗ ) having the form αx (q, p) = (q + x, p + τx (q, p)). A group action clearly imposes the 1cocycle condition: τx+y (q, p) = τx (q, p) + τy (q + x, p + τx (q, p)). The symplectic condition reads: (α−x )∗ σB = σB . A simple computation gives us for any (q, p) ∈ Ξ: ∗
[α−x ] =
1
∗
[τ−x ]X
0 ∗ 1 + [τ−x ]X ∗
∧2 : Λ2(q,p) (Ξ) → Λ2(q+x,p+τx (q,p)) (Ξ) , (5)
where we identiﬁed all the cotangent ﬁbres T∗(q,p) Ξ ∼ = T∗q X ⊕ T∗p (T∗q X) ∼ = T∗q X ⊕ T∗p X ∗ ∗
[τ−x ]X : T∗ X ∗ → T∗ X,
∗
[τ−x ]X ∗ : T∗ X ∗ → T∗ X ∗ .
(6) (7)
Finally we obtain: . ∗ [α−x ] σB − σB (q+x,p+τ
x (q,p))
N
=
(8)
{[T −x (q, p)]jk dqj ∧ dqk + [S −x (q, p)]jk dqj ∧ dpk } ,
j,k=1
with (T x (q, p))jk = = (∂/∂qj )(τx (q, p))k − (∂/∂qk )(τx (q, p))j + eB(q)jk − eB(q + x)jk , (S x (q, p))jk = (∂/∂pj )(τx (q, p))k .
(9) (10)
Asking for αx to be symplectic implies that S = 0, hence τx does not depend on p. If we ﬁx a point q0 ∈ X we can deﬁne the function a(x) := τx (q0 ) ∈ X ∗ and the condition imposed on τx (q) for having a group action leads to τx (q + q0 ) = a(x + q) − a(q). Chosing q0 = 0 and a vector potential A for B, the ﬁrst equation in (9) implies (τx (q)) := eA(q + x) − eA(q). " Let us compute the associate diﬀerential action. We set [(DA(q))∗· x]j := k [∂k Aj (q)]xk and for x ∈ X we deﬁne the vector ﬁeld in T(X × X ): tx (q, p) := (∂/∂t) t=0 α−tx (q, p) =
(11)
= (−x, (∂/∂t) t=0 τ−tx (q)) = (−x, e(DA(q)) · x) . Let us ﬁnd the associated momentum map. A computation using the deﬁnition above (see also [18]) gives: [iσB ](x, l) = (l + exB, −x), where (xB)(y) := B(x, y). Then we obtain
422
M. M˘ antoiu and R. Purice
[iσB ](tB x )(q,p) = (e(DA(q)) · x − exB, x)(q,p) = (−d(eA(q) · x), x)(q,p) , with A(q) · x =
N " j=1
A Aj (q)xj . It follows then that [iσB ](tB x ) = dγx , where
γxA (q, p) := x · p − eA(q) · x and thus for any direction ν ∈ X (ν = 1) we have deﬁned the inﬁnitesimal observable magnetic momentum along ν to be γνA (q, p) := ν · (p − eA(q)). The momentum map ( [21]) is thus given by µA : T∗ X → X ∗ ,
[µA (q, p)](x) := γxA (q, p) ,
(12)
i.e. µA (q, p) = p − eA(q).
3 The Quantum Picture A guide in guessing a quantum multiplication for observables is the WeylMoyal product of symbols, valid for B = 0 and underlying the Weyl form of pseudodiﬀerential theory. A replacement of σ by σB , as suggested by Sect. 2.1, triggers a formalism which will be exposed in the following sections. Here we examine a way to extend the multiplication, put it into a form suited for dynamical systems and C ∗ norms and study how unbounded observables may be expressed by means of bounded ones. 3.1 The Magnetic Moyal Product The wellknown formula of symbol composition in the usual Weyl quantization can be expressed in terms of the canonical symplectic form. Assume for simplicity that f, g ∈ S(Ξ); then Weyl and Moyal proposed the multiplication
(f ◦ g)(ξ) = (2/)
2N
dζ exp {−(2i/)σ(η, ζ)} f (ξ − η)g(ξ − ζ),
dη Ξ
Ξ
where ξ = (q, p), η = (y, k), ζ = (z, l). By a simple calculation, one gets
(f ◦ g)(ξ) = (2/)2N
dη dζ exp −(i/) Ξ Ξ
T (ξ,η,ζ)
σ
f (ξ − η)g(ξ − ζ),
in terms of the ﬂux of σ through the triangle in phase space T (ξ, η, ζ) :=< (q − y − z, p − k − l), (q + y − z, p + k − l), (q + z − y, p + l − k) > . ∞ (X), A magnetic ﬁeld B is turned on, with components supposed of class Cpol i.e. indeﬁnitely derivable and each derivative polynomially bounded. Taking into account the formalism of Sect. 2.1, it is natural to replace σ by σB :
The Mathematical Formalism of a Particle in a Magnetic Field
2N (f ◦B g)(ξ) = (2/) dη dζ exp −(i/) Ξ Ξ
σB
T (ξ,η,ζ)
423
f (ξ −η)g(ξ −ζ) . (13)
This leads readily to the formula. (f ◦B g)(ξ) = 2N −(2i/)σ(η,ζ) = (2/) dη dζ e exp −(i/) Ξ Ξ
(14)
T (q,y,z)
B
f (ξ−η)g(ξ−ζ) ,
where the triangle T (q, y, z) :=< q − y − z, q + y − z, q + z − y > is the projection of T (ξ, η, ζ) on the conﬁguration space. We call the composition law ◦B : S(Ξ)×S(Ξ) → S(Ξ) the magnetic Moyal product. It is welldeﬁned, associative, noncommutative and satisﬁes f ◦B g = g ◦B f . It oﬀers a way to compose observables in a quantum theory of a particle placed in the magnetic ﬁeld. It is expressed only in terms of B; no vector potential is needed. 3.2 The Magnetic Moyal Algebra The ∗algebra S(Ξ) is much too small for most of the applications. Extensions by absolutely convergent integrals still give rather poor results. One method to get much larger algebras (classes of H¨ormander symbols) is by oscillatory integrals. This requires somewhat restricted conditions on the magnetic ﬁeld, but leads to a powerful ﬁltred symbolic calculus that we intend to develop in a forthcoming paper. Here we indicate an approach by duality. So let us keep the mild assumption that the components of the magnetic ∞ (X)functions. The duality approach is based on the observation ﬁeld are Cpol [17, Lem. 14] : For any f, g in the Schwartz space S(Ξ), we have dξ (f ◦B g)(ξ) = dξ (g ◦B f )(ξ) = dξ f (ξ) g(ξ) = f , g ≡ (f, g) . Ξ
Ξ
Ξ
As a consequence, if f, g and h belong to S(Ξ), the equalities (f ◦B g, h) = (f, g ◦B h) = (g, h ◦B f ) hold. This suggests Deﬁnition 1. For any distribution F ∈ S (Ξ) and any function f ∈ S(Ξ) we deﬁne (F ◦B f, h) := (F, f ◦B h),
(f ◦B F, h) := (F, h ◦B f )
for all h ∈ S(Ξ) .
The expressions F ◦B f and f ◦B F are a priori tempered distributions. The Moyal algebra is precisely the set of elements of S (Ξ) that preserves regularity by composition.
424
M. M˘ antoiu and R. Purice
Deﬁnition 2. The Moyal algebra M(Ξ) ≡ MB (Ξ) is deﬁned by . M(Ξ) := F ∈ S (Ξ)  F ◦B f ∈ S(Ξ) and f ◦B F ∈ S(Ξ) for all f ∈ S(Ξ) . For two distributions F and G in M(Ξ), the Moyal product is extended by (F ◦B G, h) := (F, G ◦B h) for all h ∈ S(Ξ). The set M(Ξ) with this composition law and the complex conjugation F → F is a unital ∗ algebra. Actually, this extension by duality also gives compositions M(Ξ) ◦B S (Ξ) ⊂ S (Ξ) and S (Ξ) ◦B M(Ξ) ⊂ S (Ξ). An important result [17, Prop. 23] concerning the Moyal algebra is that it contains ∞ (Ξ), the space of inﬁnitely derivable complex functions on Ξ having Cpol,u polynomial growth at inﬁnity uniformly for all the derivatives. This duality strategy is often substantiated in calculations by regularization techniques. Further properties of ◦B and M(Ξ) can be found in [17]. 3.3 The Twisted Crossed Product One thing missing in the pseudodiﬀerential setting is a “good norm” on suitable subclasses of M(Ξ). We can introduce some useful norms after a partial Fourier transformation 1 ⊗ F : S(Ξ) ≡ S(X × X " ) → S(X × X). Setting (1 ⊗ F)(f ◦B g) =: [(1 ⊗ F)f ] 1B [(1 ⊗ F)f ], one gets for ϕ = (1 ⊗ F)f, ψ = (1 ⊗ F)g in S(X × X) the multiplication law ϕ 1B ψ (q; x) := (15) dy ϕ q − (x − y); y) ψ q + y; x − y e−(i/)ΦB (q,x,y) 2 2 X where ΦB (q, x, y) is the ﬂux of B through the triangle deﬁned by the points q− 2 x, q− 2 x+y and q+ 2 x. The partial Fourier transformation also converts the complex conjugation f → f into the involution ϕ → ϕ , with ϕ (q; x) := ϕ(q; −x). Thus one gets a new ∗algebra S(X × X), 1B , , isomorphic with the previous one. This also can be extended in various ways; in particular, there are Moyal type algebras M(X × X) ≡ MB (X × X) in this setting too. But it is important to note that (15) is just a particular instance of a general mathematical object, the twisted crossed product. We give here the main ideas and refer to [25] and [26] for the full theory and to [16] and especially [19] for a comprehensive treatment of its relevance to quantum magnetic ﬁelds. Let A be a unital C ∗ algebra composed of bounded, uniformly continuous functions on X; this algebra is supposed to contain the “admissible” potentials. The idea behind this algebra is that for many problems it is more adequate to consider the whole algebra generated by a potential function and its translations. We shall always assume that A contains the constant functions as well as the ideal C0 (X) := {a : X → C  f is continuous and a(x) → 0 for x → ∞} (in fact this hypothesis is not necessary everywhere) and is stable by translations, i.e. θx (a) := a(· + x) ∈ A for all a ∈ A and x ∈ X.
The Mathematical Formalism of a Particle in a Magnetic Field
425
Such a C ∗ algebra will be called admissible. Thus, for any = 0, one can deﬁne the continuous action of X by automorphisms of A: % & θx (a) (y) := a(y + x) . θ : X → Aut(A), θ is a group morphism and the maps X x → θx (a) ∈ A are all continuous. We suppose B to have components Bjk in A and we deﬁne the map: (q, x, y) → ωB (q; x, y) := e−(i/)ΓB () ,
where ΓB (< q, q + x, q + x + y >) denotes the ﬂux of the magnetic ﬁeld B through the triangle deﬁned by the vertices q, q+x, % q+x+y & in X. It can be : X × X → C(X; T), ωB (x, y) (q) := ωB (q; x, y) interpreted as a map ωB with values in the set of continuous functions on X taking values in the 1torus T := {z ∈ C  z = 1}. It is easy to see by Stokes Theorem and the satisﬁes the 2cocycle condition equation dB = 0 that ωB % & (y, z) ωB (x, y + z), ∀x, y, z ∈ X . ωB (x, y)ωB (x + y, z) = θx ωB (x, 0) = 1 = ωB (0, x), ∀x ∈ X. It is also normalized, i.e. ωB The quadruplet (A, θ , ωB , X) is a magnetic example of an abelian twisted C ∗ dynamical system (A, θ, ω, X). In the general case X is an abelian second countable locally compact group, A is an abelian C ∗ algebra, θ is a continuous morphism from X to the group of automorphisms of A and ω is a continuous 2cocycle with values in the group of all unitary elements of A. Given any abelian twisted C ∗ dynamical system, a natural C ∗ algebra can be deﬁned. We recall its construction. Let L1 (X; A) be the set of Bochner integrable functions on X with values in A, with the L1 norm ϕ1 := X dx ϕ(x)A . For any ϕ, ψ ∈ L1 (X; A) and x ∈ X, we deﬁne the product dy θ y−x [ϕ(y)] θ y2 [ψ(x − y)] θ− x2 [ω(y, x − y)] (ϕ 1 ψ)(x) := X
2
and the involution φ (x) := θ− x2 [ω(x, −x)−1 ]φ(−x)∗ . In this way, one gets a Banach ∗algebra. Deﬁnition 3. The enveloping C ∗ algebra of L1 (X, A) is called the twisted 1 crossed product and is denoted by Aω θ X. It is the completion of L (X; A) ∗ under the C norm ϕ := sup{ π(ϕ) B()  π : L1 (X; A) → B(H) representation} . It is easy to see that, with θ = θ , ω = ωB , one gets exactly the structure ω
exposed above restricted to S(X × X) ⊂ L1 (X; A). The C ∗ algebra AθB X h (x, −x) = 1. will be denoted simply by CB (A). In the magnetic case ωB ∗ After a partial Fourier transformation we get the C algebra BB (A) := (1 ⊗ F −1 )CB (A), which is another extension of the ∗subalgebra S(Ξ) endowed with complex conjugation and the multiplication (14).
426
M. M˘ antoiu and R. Purice
3.4 Abstract Aﬃliation When working with a selfadjoint operator H in a Hilbert space H, it might be useful to know that the functional calculus of H (its resolvent for example) belongs to some special C ∗ algebra of B(H). Our representationfree approach forces us to use an abstract version, borrowed from [1]. Deﬁnition 4. An observable aﬃliated to a C ∗ algebra C is a morphism Φ : C0 (R) → C. ∞ ∗ Recall that a function
h ∈ C (X ) is called an elliptic symbol of type α s ∈ R if (with p := 1 + p2 ) (∂ h)(p) ≤ cα ps−α for all p ∈ X ∗ , α ∈ NN and there exist R > 0 and c > 0 such that c ps ≤ h(p) for all p ∈ ∞ (Ξ), thus in X ∗ and p ≥ R. Such a function is naturally contained in Cpol,u M(Ξ). For any z ∈ R, we also set rz : R → C by rz (t) := (t − z)−1 . BC ∞ (X) is the space of all functions in C ∞ (X) with bounded derivatives of any order.
Theorem 1. Assume that B is a magnetic ﬁeld whose components belong to A ∩ BC ∞ (X). Then each real elliptic symbol h of type s > 0 deﬁnes an observable ΦB,h aﬃliated to BB (A), such that for any z ∈ R one has (h − z) ◦B ΦB,h (rz ) = 1 = ΦB,h (rz ) ◦B (h − z) .
(16)
In fact one has ΦB,h (rz ) ∈ (1 ⊗ F) L1 (X; A) ⊂ S (Ξ), so the compositions can be interpreted as M(Ξ) × S (Ξ) → S (Ξ) and S (Ξ) × M(Ξ) → S (Ξ). The proof can be found in [20] and consists in starting with the usual inverse for function multiplication and control the corrections using the L1 norm in the algebra CB (A). This result is basic for our approach to spectral analysis for Hamiltonians with magnetic ﬁelds in Sections 6.1 and 6.2. A represented version will be found in Sect. 5.3.
4 The Limit → 0 The quantum and classical descriptions we have given for a particle in a magnetic ﬁeld, can be gathered into a common “continuous” structure indexed by the Plank’ constant ∈ [0, 0 ], by the procedure of strict deformation quantization. Our strategy follows [10] and the details may be found in our paper [18]. The main idea is to deﬁne for each value of ∈ [0, 0 ] an algebra of bounded observables and using a common dense subalgebra, to prove that the family is in fact a continuous ﬁeld of C ∗ algebras (see [28, 29]). So far we have deﬁned for > 0 a C ∗ algebra BB (A) describing the observables of the quantum particle in a magnetic ﬁeld B. Let us deﬁne now for = 0 the C ∗ algebra B0B (A) := C(X ∗ ; A) with the usual commutative product of functions (f ◦0B g := f g) and the involution deﬁned by complex
The Mathematical Formalism of a Particle in a Magnetic Field
427
conjugation. Setting A∞ := {a ∈ A ∩ C ∞ (X)  ∂ α a ∈ A, ∀α ∈ NN } one veriﬁes that the linear space A := S(X ∗ ; A) is closed for any Moyal product ◦B , also for = 0. For ∈ [0, 0 ] we denote by . the C ∗ norm in BB (A). Moreover let us remark that the real algebra A0 := {f ∈ A  f = f } is a Poisson subalgebra of C ∞ (Ξ; R) endowed with the magnetic Poisson braket associated to the magnetic ﬁeld B. It is easy to verify that one has the • Completness condition: A = C ⊗ A0 is dense in each C ∗ algebra BB (A). The following convergences are proved by direct computation [18]: • von Neumann condition: For f and g in A0 one has lim
→0
1 f ◦B g + g ◦B f − f g = 0 . 2
• Dirac condition: For f and g in A0 one has lim
→0
1 f ◦B g − g ◦B f − {f, g}B = 0 . i
An argument using a theorem in [24], concerning continuous ﬁelds of twisted crossedproducts, allows to prove the following continuity result [18]: • Rieﬀel condition: For f ∈ A0 the map [0, 0 ] → f ∈ R is continuous. Following [10,28,29] we say that we have a strict deformation quantization of the Poisson algebra A0 .
5 The Schr¨ odinger Representation A complete overview of the formalism is achieved only after representations in Hilbert spaces are also outlined. This will put forward magnetic potentials, but in a gauge covariant way. We obtain integrated forms of covariant representations as well as the magnetic version of pseudodiﬀerential operators. Unbounded pseudodiﬀerential operators have their resolvents in wellcontrolled C ∗ algebras composed of bounded ones, as a consequence of Sect. 3.4; this is basic to the spectral results of Sect. 6. 5.1 Representations of the Twisted Crossed Product Fortunately, nondegenerate representations of twisted crossed product C ∗ algebras admit a complete classiﬁcation. We recall that the representation ρ : C → B(H) of the C ∗ algebra C in the Hilbert space H is called nondegenerate if ρ(C)H generates H. Since Aω θ X was obtained from the twisted C ∗ dynamical system (A, θ, ω, X), one may expect that the representations of A ω θ X can be deduced from a certain kind of Hilbert representations of the system (A, θ, ω, X).
428
M. M˘ antoiu and R. Purice
Deﬁnition 5. Given a twisted dynamical system (A, θ, ω, X), we call covariant representation a Hilbert space H together with two maps r : A → B(H) and U : X → U(H) satisfying: • r is a nondegenerate representation, • U is strongly continuous and U (x)U (y) = r[ω(x, y)]U (x+y) ∀x, y ∈ X, • U (x)r(a)U (x)∗ = r[θx (a)], ∀x ∈ X, a ∈ A. It can be shown that there is a onetoone correspondence between covariant representations of (A, θ, ω, X) and nondegenerate representations of A ω θ X. The following evident statement will be needed. Lemma 1. For (H, r, U ) covariant representation of (A, θ, ω, X), the map r U deﬁned on L1 (X; A) by the formula & % (r U )ϕ := dx r θx/2 ϕ(x) U (x) X
extends to a representation of A ω θ X, called the integrated form of (r, U ). For our magnetic C ∗ dynamical systems one constructs covariant representations by choosing vector potentials. We shall call them and their integrated forms Schr¨ odinger representations, inspired by the case B = 0.For A such that B = dA and for points x, y ∈ X, we deﬁne ΓA ([x, y]) := [x,y] A the circulation of A through the segment [x, y] := {sx + (1 − s)y  s ∈ [0, 1]}. By Stokes Theorem we have ΓB (< q, q + x, q + x + y >) = = ΓA ([q, q + x])ΓA ([q + x, q + x + y])ΓA ([q + x + y, q]) , leading to % &−1 ωB (q; x, y) = λA (q; x)λA (q + x; y) λA (q; x + y) ,
(17)
where we set λA (q; x) := exp {−(i/)ΓA ([q, q + x])} . We deﬁne H := L2 (X), r : A → B[L2 (X)], r(a) :=the operator of multiplication by a ∈ A and % & UA (x)u (q) := λA (q; x)u(q + x), ∀q, x ∈ X, ∀u ∈ L2 (X) . It follows easily that (H, r, UA ) is a covariant representation of (A, θ , ωB , X). ) is Rep ≡ r U : C (A) → The integrated form associated to (H, r, U A A A B % & B L2 (X) , given explicitely on L1 (X; A) by $ # x+y y−x , dy e(i/)ΓA ([x,y]) ϕ u(y) . (18) RepA (ϕ)u (x) = −N 2 X
The Mathematical Formalism of a Particle in a Magnetic Field
429
5.2 Pseudodiﬀerential Operators Let us compose RepA with the partial Fourier transformation in order to get a representation OpA := RepA ◦ (1 ⊗ F) : BB (A) → B(H). A calculation on suitable subsets of BB (A) (on S(Ξ) for example) gives the explicit action # $ OpA (f )u (x) = (19) =
−N
(i/)(x−y)·k −(i/)ΓA ([x,y])
dy dk e X
e
X
f
x+y , k u(y) . 2
We call OpA (f ) the magnetic pseudodiﬀerential operator associated to the symbol f. A posteriori, one may say that la raison d’ˆetre of the composition (13) is to ensure the equality: OpA (f )OpA (g) = OpA (f ◦B g). One also has OpA (f )∗ = OpA (f ). Some properties of OpA can be found in [17] and [19]. Now it is easy to see what gauge covariance is at the level of the two representations RepA and OpA . If two 1forms A and A are equivalent (A = A + dρ) then one will get unitarily equivalent representations: OpA (f ) = e(i/)ρ OpA (f )e(−i/)ρ and RepA (ϕ) = e(i/)ρ RepA (ϕ)e(−i/)ρ . We refer to [17] for a comparaison with a quantization procedure f → Op,A (f ), combining (in an inappropriate order) the usual, nonmagnetic calculus with the minimal coupling rule (x, p) → (x, p − A(x)). It is not gaugecovariant, so that it is not suitable as a real quantization procedure. Finally let us quote a result linking M(Ξ) with OpA [17, Prop. 21] : ∞ (X), OpA is an isomorphism of ∗ algebras For any vector potential A in Cpol between M(Ξ) and L[S(X)] ∩ L[S (X)], where L[S(X)] and L[S (X)] are, respectively, the spaces of linear continuous operators on S(X) and S (X). 5.3 A New Justiﬁcation: Functional Calculus We give here a new justiﬁcation of our formalism. It is obvious that if one gives some convincing reason for working with (19), then the remaining part can be deduced as a necessary consequence, by reversing the arguments. Let us accept that our quantum particle placed in a magnetic ﬁeld is de scribed by the family of elementary operators Q1 , . . . , QN ; (ΠA )1 , . . . , (ΠA )N , where Qj is the operator of multiplication by xj and (ΠA )j := Pj − Aj = −i∂j − Aj is the j’th component of the magnetic momentum deﬁned by a vector potential A with dA = B (these may be considered as quantum observables associated to the position and the momentum map for the translation ) group). Then OpA should be a functional calculus f → OpA (f ) ≡ f (Q, ΠA for this family of noncommuting selfadjoint operators. The scheme is: (i) , (ii) condense them consider the commutation relations satisﬁed by Q, ΠA in a global, exponential form, (iii) deﬁne OpA (f ) by decomposing f as a
430
M. M˘ antoiu and R. Purice
continuous linear combination of exponentials. We mention that exactly this argument leads to the usual Weyl calculus (B = 0). So let us take into account the following commutation relations, easy to , Qk ] = δj,k , i[ΠA,j , ΠA,k ] = Bkj (Q), ∀j, k = check: i[Qj , Qk ] = 0, i[ΠA,j 1, . . . , N . A convenient global form maybe given in terms of the magnetic Weyl system. Recall the unitary group eiQ·p p∈X of the position as well , given explicitely in the as the magnetic translations UA (q) := eiq·ΠA q∈X
Hilbert space H := L2 (X) by
UA (x) = e−(i/)ΓA ([Q,Q+x]) eix·P , (20) which is just another way to write (17). The family UA (x) x∈X satisﬁes UA (x)UA (x ) = ωB (Q; x, x )UA (x + x ),
x, x ∈ X ,
>) where we set ωB (q; x, x ) := e−(i/)ΓB ( 0, if A, Aν are continuous vector potentials respectively for B, Bν ≡ BFν , one has % & O )] . σess h(ΠA ) = σ[h(ΠA ν
(21)
ν The operators h(ΠA ) are the asymptotic operators mentioned earlier. ν All the spectra in (21) are only depending on the respective magnetic ﬁelds. Examples may be found in [20], see also [15]. Some related results may be found in the recent paper [11].
6.2 A NonPropagation Result We ﬁnally describe, following [20], how the localization results proved in [2] in the case of Schr¨ odinger operators without magnetic ﬁeld can be extended to the situation where a magnetic ﬁeld is present. Once again, the algebraic formalism and the aﬃliation criterion introduced above play an essential role in the proofs. For any quasiorbit F , let NF be the family of sets of the form W = W ∩ X, where W is any element of a base of neighbourhoods of F in SA . We write χW for the characteristic function of W . Theorem 3. Let B be a magnetic ﬁeld whose components belong to A ∩ BC ∞ (X) and let h be a real elliptic symbol of type s > 0. Assume that potentials for B F ⊂ FA is a quasiorbit. Let A, AF be continuous vector % & ) = ∅, then and BF , respectively. If η ∈ C0 (R) with supp8(η) ∩ σ h(ΠA % F &8 ) 8 ≤ . In for any > 0 there exists W ∈ NF such that 8χW (Q) η h(ΠA particular, the following inequality holds uniformly in t ∈ R and u ∈ H: 8 % & 8 8χW (Q) e−ith(ΠA ) η h(ΠA ) u8 ≤ u . The last statement of this theorem gives a precise meaning to the notion of nonpropagation. We refer to [2] for physical explanations and interpretations of this result as well as for some examples.
Acknowledgements This work was deeply inﬂuenced by Vladimir Georgescu to whom we are grateful. Part of the results described in this paper were obtained while
The Mathematical Formalism of a Particle in a Magnetic Field
433
the authors were visiting the University of Geneva; many thanks are due to Werner Amrein for his kind hospitality. We are greateful to Serge Richard for enjoyable collaboration. M.M. thanks Joseph Avron for a stimulating discussion and encouragements in developing this work. We also acknowledge partial support from the CERES Program of the Romanian Ministery of Education and Research.
References 1. W.O. Amrein, A. Boutet de Monvel, V. Georgescu: C0 Groups, Commutator Methods and Spectral Theory of NBody Hamiltonians, (Birkh¨ auser Verlag, 1996). 2. W.O. Amrein, M. M˘ antoiu, R. Purice: Ann. Henri Poincar´e 3, 1215 (2002). 3. H. D. Cornean, G. Nenciu: Ann. Henri Poincar´e 1, 203 (2000). 4. V. Georgescu, A. Iftimovici: preprint mparc0199. 5. V. Georgescu, A. Iftimovici: Commun. Math. Phys. 228, 519 (2002). 6. V. Georgescu, A. Iftimovici: C ∗ Algebras of Quantum Hamiltonians. In Operator Algebras and Mathematical Physics, Constant¸a, 2001, ed by J.M. Combes, J. c*ntz, G. A. Elliott et al, (Theta Foundation, 2003). 7. B. Helﬀer, A. Mohamed: Ann. Inst. Fourier 38, 95 (1988). 8. M. V. Karasev, T. A. Osborn: J. Math. Phys. 43, 756 (2002). 9. M. V. Karasev, T. A. Osborn: J. Phys. A 37, 2345 (2004). 10. N. P. Landsman: Mathematical Topics Between Classical and Quantum Mechanics, (SpringerVerlag, NewYork, 1998), 529 pp. 11. Y. Last, B. Simon: preprint mparc 05112. 12. R. Lauter, V. Nistor: Analysis of Geometric Operators on Open Manifolds: a Groupoid Approach. In Quantization of Singular Symplectic Quotients, ed by N. P. Landsman, M. Pﬂaum and M. Schlichenmaier, (Birkh¨ auser, Basel, 2001). 13. R. Lauter, B. Monthubert and V. Nistor: Documenta Math. 5, 625 (2000). 14. J. M. Luttinger: Phys. Rev. 84, 814 (1951). 15. M. M˘ antoiu: J. reine angew. Math. 550, 211 (2002). 16. M. M˘ antoiu, R. Purice: The Algebra of Observables in a Magnetic Field. In: Mathematical Results in Quantum Mechanics (Taxco, 2001), ed by R. Weder, P. Exner and B. Gr´ebert (Amer. Math. Soc., Providence, RI, 2002) pp 239–245. 17. M. M˘ antoiu, R. Purice: J. Math.Phys. 45, 1394 (2004). 18. M. M˘ antoiu, R. Purice: Strict deformation quantization for a particle in a magnetic ﬁeld, to appear in J. Math. Phys. 2005. 19. M. M˘ antoiu, R. Purice, S. Richard: preprint mparc 0476. 20. M. M˘ antoiu, R. Purice, S. Richard: preprint mparc 0584. 21. J. E. Marsden, T. S. Rat¸iu: Introduction to Mechanics and Symmetry, 2nd edn, (SpringerVerlag, Berlin, New York, 1999), 582 pp. 22. G. Nenciu: preprint mparc 0096. 23. V. Nistor, A. Weinstein, P. Xu: Paciﬁc J. Math., 189, 117 (1999). 24. M. Nilsen: Indiana Univ. Math. J. 45, 436 (1996). 25. J.A. Packer, I. Raeburn: Math. Proc. Camb. Phil. Soc. 106, 293 (1989). 26. J.A. Packer, I. Raeburn: Math. Ann. 287, 595 (1990). 27. G. D. Raikov, M. Dimassi: Cubo Mat. Educ. 3, 317 (2001).
434
M. M˘ antoiu and R. Purice
28. M. Rieﬀel: Math. Ann. 283, 631 (1989). 29. M. Rieﬀel: Memoirs of the AMS, 106, 93 pp (1993). 30. J. Renault: A Groupoid Approach to C ∗ Algebras, (Springer, Berlin, 1980) 160 pp.
Fractal Weyl Law for Open Chaotic Maps St´ephane Nonnenmacher Service de Physique Th´eorique, CEA/DSM/PhT, Unit´e de recherche associ´ee au CNRS, CEA/Saclay , 91191 GifsurYvette, France [emailprotected]
1 Introduction We summarize our work in collaboration with Maciej Zworski [16], on the semiclassical density of resonances for a quantum open system, in the case when the associated classical dynamics is uniformly hyperbolic, and the set of trapped trajectories is a fractal repeller. The system we consider is not a Hamiltonian ﬂow, but rather a “symplectic map with a hole” on a compact phase space (the 2torus). Such a map can be considered as a model for the Poincar´e section associated with a scattering Hamiltonian on R2 , at some positive energy; the “hole” represents the points which never return to the Poincar´e section, that is, which are scattered to inﬁnity. We then quantize this open map, obtaining a sequence of subunitary operators, the eigenvalues of which are interpreted as resonances. We are especially interested in the asymptotic density of “longliving resonances”, representing metastable states which decay in a time bounded away from zero (as opposed to “short resonances”, associated with states decaying instantaneously). Our results (both numerical and analytical) support the conjectured fractal Weyl law, according to which the number of longliving resonances scales as −d , where d is the (partial) fractal dimension of the trapped set. 1.1 Generalities on Resonances A Hamiltonian dynamical system (say, H(q, p) = p2 + V (q) on R2n ) is said to be “closed” at the energy E when the energy surface ΣE is a compact subset of the phase space. The associated quantum operator H = −2 ∆ + V (q) then admits discrete spectrum near the energy E (for small enough ). Furthermore, if E is nondegenerate (meaning that the ﬂow of H has no ﬁxed point on ΣE ), then the semiclassical density of eigenvalues is given by the celebrated Weyl’s law [11]: 1 # { Spec(H) ∩ [E − δ, E + δ] } = dq dp + O(1−n ) . (2π)n H(q,p)−E 0, ΣE is unbounded, and H admits absolutely continuous spectrum on [0, ∞). However, one can meromorphically continue the resolvent (z − H )−1 across the real axis from the upper halfplane into the lower halfplane. In general, this continuation will have discrete poles {zj = Ej − Iγj } with “widths” γj > 0, which are the resonances of H . Physically, each resonance is associated with a metastable state: a (not squareintegrable) solution of the Schr¨ odinger equation at the energy zj , which decays like e−tγj / when t → +∞. In spectroscopy experiments, one measures the energy dependence of some scattering crosssection σ(E). Each γ resonance zj imposes a Lorentzian component (E−Ejj)2 +γ 2 on σ(E); a resj
onance zj will be detectable on the signal σ(E) only if its Lorentzian is wellseparated from the ones associated with nearby resonances of comparable widths, therefore iﬀ Ej − Ej  γj . This condition of “wellseparability” is NOT the one we will be interested in here. We will rather consider the order of magnitude of each resonance lifetime /γj , independently of the nearby ones, in the semiclassical r´egime: a resonant state will be “visible”, or “longliving”, if γj = O(). Our objective will be to count the number of resonances zj in boxes of the type {Ej − E ≤ C , γj ≤ C}, or equivalently {zj − E ≤ C}. 1.2 Trapped Sets Since resonant states are “invariant up to rescaling”, it is natural to relate them, in the semiclassical spirit, to invariant structures of the classical dynamics. For a scattering system, the set of points (of energy E) which don’t escape to inﬁnity (either in past or future) is called the trapped set at energy E, and denoted by K(E). The textbook example of a radiallysymmetric potential shows that this set may be empty (if V (r) decreases monotonically from r = 0 to r → ∞), or have the same dimension as ΣE (if V (r) has a maximum V (r0 ) > 0 before decreasing as r → ∞). For n = 2 degrees of freedom, the geometry of the trapped set can be more complex. Let us consider the wellknown example of 2dimensional scattering by a set of nonoverlapping disks [5,9] (a similar model was studied in [2,26]). When the scatterer is a single disk, the trapped set is obviously empty. The scattering by two disks admits a single trapped periodic orbit, bouncing back and forth between the disks. Since the evolution between
Fractal Weyl Law for Open Chaotic Maps
437
two bounces is “trivial”, it is convenient to represent the scattering system through the bounce map on the reduced phase space (position along the boundaries × velocity angle). This map is actually deﬁned only on a fraction of this phase space, namely on those points which will bounce again at least once. For the 2disk system, this map has a unique periodic point (of period 2), which is of hyperbolic nature due to the curvature of the disks. The trapped set K of the map (“reduced” trapped set) reduces to this pair of points; it lies at the intersection of the forward trapped set Γ− (points trapped as t → +∞) and the backward trapped set Γ+ (points trapped as t → −∞). The addition of a third disk generates a complex bouncing dynamics, for which the trapped set is a fractal repeller [9]. We will explain in the next section how such a structure arises in the case of the open baker’s map. As in the 2disk case, the bounce map is uniformly hyperbolic; each forward trapped point x ∈ Γ− admits a stable manifold W− (x) (and viceversa for x ∈ Γ+ ). One can show that Γ− is fractal along the unstable direction W+ : Γ− ∩W+ has a Hausdorﬀ dimension 0 < d < 1 which depends on the positions and sizes of the disks. Due to timereversal symmetry, Γ+ ∩ W− has the same Hausdorﬀ dimension. Finally, the reduced trapped set K = Γ+ ∩ Γ− is a fractal of dimension 2d, which contains inﬁnitely many periodic orbits. The unreduced trapped set K(E) ⊂ ΣE has one more dimension corresponding to the direction of the ﬂow, so it is of dimension D = 2d + 1. 1.3 Fractal Weyl Law We now relate the geometry of the trapped set K(E), to the density of resonances of the quantized Hamiltonian Hh in boxes { z − E ≤ C }. The following conjecture (which dates back at least to the work of Sj¨ ostrand [21]) relates this density with the “thickness” of the trapped set. Conjecture 1. Assume that the trapped set K(E) at energy E has dimension 2dE + 1. Then, the density of resonances near E grows as follows in the semiclassical limit: . # Res(H ) ∩ { z : z − E < r } h→0 −→ cE (r) , (2) ∀r > 0, −dE for a certain “shape function” 0 ≤ cE (r) < ∞. We were voluntarily rather vague on the concept of “dimension” (a fractal set can be characterized by many diﬀerent dimensions). In the case of a closed system, K(E) has dimension 2n − 1, so we recover the Weyl law (1). If K(E) consists in one unstable periodic orbit, the resonances form a (slightly deformed) rectangular lattice of sides ∝ , so each box contains at most ﬁnitely many resonances [22]. For intermediate situations (0 < dE < n − 1), one has only been able to prove one half of the above estimate, namely the upper bound for this
438
S. Nonnenmacher
resonance counting [10, 21, 24, 27]. The dimension appearing in these upper bounds is the Minkowski dimension deﬁned by measuring neighborhoods of K(E). In the case we will study, this dimension is equal to the Hausdorﬀ one. Some lower bounds for the resonance density have been obtained as well [23], but are far below the conjectured estimate. Several numerical studies have attempted to conﬁrm the above estimate for a variety of scattering Hamiltonians [10,12–14], but with rather inconclusive results. Indeed, it is numerically demanding to compute resonances. One method is to “complex rotate” the original Hamiltonian into a nonHermitian operator, the eigenvalues of which are the resonances. Another method uses the (approximate) relationship between, on one side, the resonance spectrum of H , one the other side, the set of zeros of some semiclassical zeta function, which is computed from the knowledge of classical periodic orbits [5, 14]. In the case of the geodesic ﬂow on a convex cocompact quotient of the Poincar´e disk (which has a fractal trapped set), the resonances of the Laplace operator are exactly given by the zeros of Selberg’s zeta function. Even in that case, it has been diﬃcult to check the asymptotic Weyl law (2), due to the necessity to reach suﬃciently high values of the energy [10]. 1.4 Open Maps Confronted with these diﬃculties to deal with open Hamiltonian systems, we decided to study semiclassical resonance distributions for toy models which have already proven eﬃcient to modelize closed systems. In the above example of obstacle scattering, the bounce map emerged as a way to simplify the description of the classical dynamics. It acts on a reduced phase space, and gets rid of the “trivial” evolution between bounces. The exact quantum problem also reduces to analyzing an operator acting on wavefunctions on the disk boundaries, but this operator is inﬁnitedimensional, and extracting its resonances is not a simple task [2, 9]. Canonical maps on the 2torus were often used to mimic closed Hamiltonian systems; they can be quantized into unitary matrices, the eigenphases of which are to be compared with the eigenvalues e−iEj / of the propagator e−iH / (see e.g. [7] and references therein for a mathematical introduction on quantum maps). We therefore decided to construct a “toy bounce map” on T2 , with dynamics similar to the original bounce map, and which can be easily quantized into an N × N subunitary matrix (where N = (2π)−1 ). This matrix is then easily diagonalized, and its subunitary eigenvalues {λj } should be compared with the set {e−izj / }, where the zj are the resonances of H near some positive energy E. We cannot prove any direct correspondence between, on one side the eigenvalues of our quantized map, on the other side resonances of a bona ﬁde scattering Hamiltonian. However, we expect a semiclassical property like the fractal Weyl law to be robust, in the sense that it should be shared by all types of “quantum models”. To support this claim, we notice
Fractal Weyl Law for Open Chaotic Maps
439
that the usual, “closed” Weyl law is already (trivially) satisﬁed by quantized maps: the number of eigenphases θj on the unit circle (corresponding to an energy range ∆E = 2π) is exactly N = (2π)−1 , which agrees with the Weyl law (1) for n = 2 degrees of freedom. Testing Conjecture 1 in the framework of quantum maps should therefore give a reliable hint on its validity for more realistic Hamiltonian systems. Schomerus and Tworzydlo recently studied the quantum spectrum of an open chaotic map on the torus, namely the open kicked rotator [20]; they obtain a good agreement with a fractal Weyl law for the resonances (despite the fact that the geometry of the trapped set is not completely understood for that map). The authors also provide a heuristic argument to explain this Weyl law. We believe that this argument, upon some technical improvement, could yield a rigorous proof of the upper bound for the fractal Weyl law in case of maps. We preferred to investigate that problem using one of the best understood chaotic maps on T2 , namely the baker’s map.
2 The Open Baker’s Map and Its Quantization 2.1 Classical Closed Baker The (closed) baker’s map is one of the simplest examples of uniformly hyperbolic, strongly chaotic systems (it is a perfect model of Smale’s horseshoe). The “3baker’s map” B on T2 ≡ [0, 1) × [0, 1) is deﬁned as follows: p if 0 ≤ q < 1/3 , (3q, 3 ) p+1 2 (3) T (q, p) → B(q, p) = (3q − 1, 3 ) if 1/3 ≤ q < 2/3 , (3q − 2, p+2 ) if 2/3 ≤ q < 1. 3 This map preserves the symplectic form dq ∧ dp on T2 , and is invertible. Compared with a generic Anosov map, it has the particularity to be linear by parts, and its linearized dynamics (welldeﬁned away from its lines of discontinuities) is independent of the point x ∈ T2 . As a consequence, the stretching exponent is constant on T2 , as well as the unstable/stable directions (horizontal/vertical). This map admits a very simple Markov partition, made of the three vertical rectangles Rj = {q ∈ [j/3, (j + 1)/3), p ∈ [0, 1)}, j = 0, 1, 2 (see Fig. 1). Any biinﬁnite sequence of symbols . . . −2 −1 · 0 1 2 . . . (where each
i ∈ {0, 1, 2}) will be associated with the unique point x s.t. B t (x) ∈ Rt for all t ∈ Z. This is the point of coordinates (q, p), where q and p admit the ternary decompositions i−1 def q = 0 · 0 1 . . . = , p = 0 · −1 −2 . . . . 3i i≥1
440
S. Nonnenmacher
The baker’s map B simply acts as a shift on this symbolic sequence: B(x = . . . −2 −1 · 0 1 2 . . .) = . . . −2 −1 0 · 1 2 . . . .
(4)
2.2 Opening the Classical Map We explained above that the bounce maps associated with the 2 or 3disk systems were deﬁned only on parts of the reduced phase space, namely on those points which bounce at least one more time. The remaining points, which escape to inﬁnity right after the bounce, have no image through the map. Hence, to open our baker’s map B, we just decide to restrict it on a subset S ⊂ T2 , or equivalently we send points in T2 \ S to inﬁnity. We obtain an Anosov map “with a hole”, a class of dynamical systems recently studied in the literature [4]. The study is simpler when the hole corresponds to a Markov rectangle [3], so this is the choice we will make (we expect the fractal Weyl law to hold for an arbitrary hole as well). Let us choose for the hole the second Markov rectangle R1 , so that S = R0 ∪ R2 . Our open map C = BS reads (see Fig. 1): C(q, p) =
(3q, p3 ) if q ∈ R0 , (3q − 2, p+2 ) if q ∈ R2 . 3
(5)
This map is canonical on S, and its inverse C −1 is deﬁned on the set C(S).
1 R0
R1
R2 C
p
q
1
Fig. 1. Open baker’s map C. The points in the middle rectangle are sent to inﬁnity.
Our choice for S coincides with the points x = (q, p) satisfying 0 (x) ∈ {0, 2} (equivalently, points s.t. 0 (x) = 1 are sent to inﬁnity through C). This allows us to characterize the trapped sets very easily: –
the forward trapped set Γ− (see ﬁg. 2) is made of the points x which will never fall in the strip R1 for times t ≥ 0: these are the points s.t.
Fractal Weyl Law for Open Chaotic Maps
441
i ∈ {0, 2} for all i ≥ 0, with no constraint on the i for i < 0. This set is of the form Γ− = Can × [0, 1), where Can is the standard 1/3Cantor set on the unit interval. As a result, the intersection Γ− ∩ W+ ≡ Can has 2 the Hausdorﬀ (or Minkowski) dimension d = log log 3 . – the backward trapped set Γ+ is made of the points satisfying i ∈ {0, 2} for all i < 0, and is given by [0, 1) × Can. – the full trapped set K = Can × Can.
1
p
q
1
Fig. 2. Iterative construction of the forward trapped set Γ− for the open baker’s map C: we remove from T2 the points leaving to inﬁnity at times t = 1, 2, 3, 4 (from light grey to dark grey) etc. At the end, there remains the fractal set Γ− = Can × [0, 1).
2.3 Quantum Baker’s Map We now describe in some detail the quantization of the above maps. We recall [6, 7] that a nontrivial quantum Hilbert space can be associated with the phase space T2 only for discrete values of Planck’s constant, namely = (2πN )−1 , N ∈ N0 . In that case (the only one we will consider), this space HN is of dimension N . It admits the “position” basis {Qj , j = 0, . . . , N − 1} made of the “Dirac combs” 1 j − ν) . Qj (q) = √ δ(q − N N ν∈Z
This basis is connected to the “momentum” basis {Pk , k = 0, . . . , N − 1} through the discrete Fourier transform:
Pk Qj = (FN )kj =
e−2iπN kj √ , N
j, k ∈ {0, . . . , N − 1} ,
(6)
442
S. Nonnenmacher
where the Fourier matrix FN is unitary. Balazs and Voros [1] proposed to quantize the closed baker’s map B as follows, when N is a multiple of 3 (a condition we will always assume): in the position basis, it takes the block form FN/3 −1 . FN/3 (7) BN = FN FN/3 This matrix is obviously unitary, and exactly satisﬁes the Van Vleck formula (the semiclassical expression for a quantum propagator, in terms of the classical generating function). In the semiclassical limit N → ∞, it was shown [8] that these matrices classically propagate Gaussian coherent states supported far enough from the lines of discontinuities. As usual, discontinuities of the classical dynamics induce diﬀraction eﬀects at the quantum level, which have been partially analyzed for the baker’s map [25] (in particular, diﬀractive t )). orbits have to be taken into account in the Gutzwiller formula for tr(BN We believe that these diﬀractive eﬀects should only induce lowerorder corrections to the Weyl law (9). We are now ready to quantize our open baker’s map C of (5): since the classical map sends points in R1 to inﬁnity and acts through B on S = R0 ∪ R2 , the quantum propagator should kill states microsupported on R1 , and act as BN on states microsupported on S. Therefore, in the position basis we get the subunitary matrix FN/3 −1 . 0 (8) CN = FN FN/3 A very similar open quantum baker was constructed in [18], as a quantization of Smale’s horseshoe. In Fig. 3 (left) we represent the moduli of the matrix elements (CN )nm . The largest elements are situated along the “tilted diagonals” n = 3m, n = 3(m − 2N/3), which correspond to the projection on the qaxis of the graph of C. Away from these “diagonals”, the amplitudes of the elements decrease relatively slowly (namely, like 1/n − 3m). This slow decrease is due to the diﬀraction eﬀects associated with the discontinuities of the map. 2.4 Resonances of the Open Baker’s Map We numerically diagonalized the matrices CN , for larger and larger Planck’s constants N . First of all, we notice that the subspace Span{Qj , j = N/3, . . . , 2N/3 − 1}, made of position states in the “hole”, is in the kernel of CN . Therefore, it is suﬃcient to diagonalize the matrix obtained by removing the corresponding lines and columns. Upon a slight modiﬁcation of the quantization procedure [17], one obtains for CN a matrix covariant w.r.to parity, allowing for a separation of the even and odd eigenstates, and therefore reducing the dimension of each part by 2. This is the quantization we used for
Fractal Weyl Law for Open Chaotic Maps
443
N (10). Each grey Fig. 3. Graphical representation of the matrices CN (8) and C √ square represents the modulus of a matrix element (white=0, black=1/ 3)
our numerics: we only plot the evenparity resonances (the distribution of the oddparity ones is very similar). In Fig. 4 we show the evenparity spectra of the matrix CN for N = 35 and N = 38 . Although we could not detect exact null states for the reduced matrix, many among the N/3 eigenvalues had very small moduli: for large values of N , the spectrum of CN accumulates near the origin. This accumulation is an obvious consequence of the fractal Weyl law we want to test: Conjecture 2. For any radius 1 > r > 0 and N ∈ N0 , 3N , let us denote 1
1 be81
be2187
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0.2
0.2
0.4
0.4
0.6
0.6
0.8
0.8
1
1 1
0.8
0.6
0.4
0.2
0.2
0.4
0.6
0.8
1
1
0.8
0.6
0.4
0.2
0.2
0.4
0.6
Fig. 4. Evenparity spectrum of the matrices CN for N/3 = 81, 2187
0.8
1
444
S. Nonnenmacher
. def n(N, r) = # λ ∈ Spec(CN ) ∩ { λ ≥ r } . In the semiclassical limit, this counting function behaves as n(N, r)
N →∞
−→ c(r) ,
log 2
N log 3
(9)
with a “shape function” 0 ≤ c(r) < ∞. To test this conjecture, we proceed in two ways: • In a ﬁrst step, we select some discrete values for r, and plot n(N, r) for an arbitrary sequence of N , in a loglog plot (see Fig. 5). We observe that the Even eigenvalues of open 3baker Counting eigenvalues in logarithmic scale 7 N=200*3^k, r=0.5 N random, r=0.5 N=27x3^k, r=0.5 N=27x3^k, r=0.15 N=27x3^k, r=0.03 theoretical slope
6
log(n(N,r))
5
4
3
2
1
3
4
5
7
6
8
9
10
log(N) Fig. 5. Checking the N dependence of n(N, r) for various values of r, along geometric and arbitrary sequences for N . The thick curve has the slope log 2/ log 3
2 slope of the data nicely converges towards the theoretical one log log 3 (thick line), all the more so along geometric subsequences N = 3k No , and for relatively large values of the radius (r = 0.5). For the smaller value r = 0.03, the annulus {z ≥ r} still contains “too many resonances” and the asymptotic r´egime is not yet reached. log 2 • In a second step, conﬁdent that n(N, r) scales like N log 3 , we try to extract the shape function c(r). For an arbitrary sequence of values of N , we
Fractal Weyl Law for Open Chaotic Maps
445
Spectrum of the open 3−baker
Spectrum of the open 3−baker
Even spectrum, random sequence
Even spectrum, random sequence, rescaled counting
600
5
500
Rescaled eigenvalue number
Eigenvalue number = n(N,r)
4
400
300
200
3
2
1 100
0.2
0.4
0.6
Eigenvalue modulus = r
0.8
1
0.2
0.4
0.6
0.8
Eigenvalue modulus = r
Fig. 6. On the left, we plot the number n(N, r) of even eigenvalues of CN of modulus ≥ r. On the right plot, we rescale those functions by the factors N
2 − log log 3
plot the function n(N, r) (Fig. 6, left), and then rescale the vertical coordinate log 2 by a factor N − log 3 (right). The rescaled curves do roughly superpose on one another, supporting the conjecture. However, there remains relatively large ﬂuctuations, even for large values of N . The curves corresponding to a geometric sequence N = 3k No , k = 0, 1, . . . tend to be nicely superposed to one another, but slightly diﬀer from one sequence to another. Similar plots were given in [20] in the case of the kicked rotator; the shape function c(r) is conjectured there to correspond to some ensemble of random subunitary matrices. Our data are too unprecise to perform such a check. The fact that the spectra of the matrices CN “behave nicely” along geometric sequences, while they ﬂuctuate more strongly between successive values of N , is not totally unexpected (similar phenomena had been noticed for the quantizations BN of the closed baker [1]). In view of Fig. 6, our conjecture (9) may be too strict if we apply it to a general sequence of N.. At least, it seems to be satisﬁed along geometric sequences 3k No , k ∈ N , with shape functions cNo (r) slightly depending on the sequence.
1
446
S. Nonnenmacher
3 A Solvable Toy Model for the Quantum Baker 3.1 Description of the Toy Model In an attempt to get some analytical grip on the resonances, we tried to simplify the quantum matrix CN , keeping only its “backbone” along the tilted diagonals and removing the oﬀdiagonal components. We obtained the “toyofthetoy model” given by the following matrices (the moduli of the components are shown on right plot of Fig. 3): 1000001 0 0 1 0 0 0 0 0 ω2 0 0 1 0 0 0 0 0 ω 0 0 0 1 0 0 0 0 0 1 0 1 N =9 = √ 0 1 0 0 0 0 0 ω 2 0 , ω = e2πi/3 . C (10) 3 0 1 0 0 0 0 0 ω 0 0 0 1 0 0 0 0 0 1 0 0 1 0 0 0 0 0 ω2 0010000 0 ω N for N an arbitrary From this example, it is pretty clear how one constructs C multiple of 3. A similar quantization of the closed 2baker was introduced in [19]. Before describing the spectra of these matrices, we describe their propagation properties. Removing the “oﬀdiagonal” elements, we have eliminated the eﬀects of diﬀraction due to the discontinuities of C. However, this elimination is so abrupt that it modiﬁes the semiclassical transport. Indeed, a coherent state situated at a point x away from the discontinuities will not N into a single coherent state (as does CN ), but rather be transformed by C into a linear combination of 3 coherent states, shifted vertically by 1/3 from N do not quantize the open baker C one another. Therefore, the matrices C of (5), but rather the following multivalued (“raysplitting”) map: p) = C(q,
p+2 (3q, p3 ) ∪ (3q, p+1 if q ∈ R0 , 3 ) ∪ (3q, 3 ) p p+1 (3q − 2, 3 ) ∪ (3q − 2, 3 ) ∪ (3q − 2, p+2 ) if q ∈ R2 . 3
(11)
This modiﬁcation of the classical dynamics is rather annoying. Still, the dy shares some common features with that of C: the forward trapped namics C is the same as for C, that is the set Γ− described in Fig. 2. On the set for C other hand, the backward trapped set is now the full torus T2 . N as a WalshQuantized Baker 3.2 Interpretation of C N A possible way to avoid this modiﬁed classical dynamics is to interpret C as a “Walshquantized map” (this interpretation makes sense when N = 3k ,
Fractal Weyl Law for Open Chaotic Maps
447
k ∈ N). To introduce this Walsh formalism, let us ﬁrst write the Hilbert space as a tensor product HN = (C3 )⊗k , where we take the ternary decomposition of discrete positions Nj = 0 · 0 1 · · · k−1 into account. If we call {e0 , e1 , e2 } the canonical basis of C3 , each position state Qj ∈ HN can be represented as the tensor product state Qj = e0 ⊗ e1 ⊗ · · · ⊗ ek−1 . In the language of quantum computing, each tensor factor C3 is the Hilbert space of a “qutrit” associated with a certain scale [19]. The Walsh Fourier transform is a modiﬁcation of the discrete Fourier transform (6), which ﬁrst appeared in signal theory, and has been recently used as a toy model for harmonic analysis [15]. Its major advantage is the possibility to construct states compactly supported in both position and “Walsh momentum”. In our ﬁnitedimensional framework, this Walsh transform is the matrix 2iπ
(Qj ) (Qj ) , j, j = 0, . . . , N −1 , (WN )jj = 3−k/2 exp − 3 + =k−1
and acts as follows on tensor product states: WN (v0 ⊗v1 ⊗· · · vk−1 ) = F3 vk−1 ⊗· · · F3 v1 ⊗F3 v0 ,
v ∈ C3 , = 0, . . . , k−1 .
N can be expressed as Now, in the case N = 3k , our toy model C WN/3 N = W −1 . 0 C N WN/3 N One can show that “Walsh coherent states” are propagated through C according to the map C. Hence, as opposed to what happens in “standard” N Walshquantizes the open baker C. quantum mechanics, C N =3k 3.3 Resonances of C 3k to analytically We now use the very peculiar properties of the matrices C compute their spectra. From the expressions in last section, one can see that N acts very simply on tensor product states: the toy model C N v0 ⊗ v1 ⊗ · · · vk−1 = v1 ⊗ · · · vk−1 ⊗ F −1 π02 v0 , C 3
(12)
where π02 projects C3 orthogonally onto Span { e0 , e2 }. Like its classical N realizes a symbolic shift between the diﬀerent scales. It also counterpart, C sends the ﬁrst symbol 0 to the “end of the queue”, after a projection and a Fourier transform. The projection π02 kills the states Qj localized in the
448
S. Nonnenmacher 1 k=10 0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1 1
0.8
0.6
0.4
0.2
0.2
0.4
0.6
0.8
1
N for N = 310 (circles), N = 315 (crosses). The Fig. 7. Spectra of the matrices C 1 large circles have radii λ = 1, λ+ , λ+ λ−  2 , λ− 
rectangle R1 . The vector F3−1 e0 in the last qutrit induces a localization in the momentum direction, near the momentum p = 0 · 0 . N )k acts By iterating this expression k times, we see that the operator (C −1 3 independently on each tensor factor C , through the matrix F3 π02 . The latter has three eigenvalues: N )k kills any state Qj for which at – it kills the state e1 , implying that (C least one of the symbols (Qj ) is equal to 1. These 3k − 2k position states are localized “outside” of the trapped set Γ− , which explains why they are killed by the dynamics. – its two remaining eigenvalues λ± have moduli λ+  ≈ 0.8443, λ−  ≈ N , 0.6838. They build up the (2k dimensional) nontrivial spectrum of C which has the form of a “lattice” (see Fig. 7): N is the set Proposition 1. For N = 3k , the nonzero spectrum of C O j 1−p/k p/k { λ+ } ∪ { λ− } e2iπ k λ+ λ− : 1 ≤ p ≤ k − 1 , 0 ≤ j ≤ k − 1 . Most of these eigenvalues are highly degenerate (they span a subspace of dimension 2k ). When k → ∞, the highest degeneracies occur when p/k ≈ 2, which results in the following asymptotic distribution:
Fractal Weyl Law for Open Chaotic Maps
∀f ∈ C(R2 ),
lim
k→∞
1 2k
k )\0 λ∈Spec(C 3
2π
f (λ− λ+ 1/2 , θ)
mult(λ) f (λ) = 0
449
dθ . 2π
N along the geometric sequence The last formula shows that the spectrum of C {N = 3k , k ∈ N} satisﬁes the fractal Weyl law (9), with a shape function in form of an abrupt step: c(r) = Θ(λ+ λ− 1/2 − r). Although the above spectrum seems very nongeneric (lattice structure, singular shape function), it is the ﬁrst example (to our knowledge) of a quantum open system proven to satisfy the fractal Weyl law.
Acknowledgments We beneﬁted from insightful discussions with Marcos Saraceno, Andr´e Voros, Uzy Smilansky, Christof Thiele and Terry Tao. Part of the work was done while I was visiting M. Zworski in UC Berkeley, supported by the grant DMS0200732 of the National Science Foundation.
References 1. N.L. Balazs and A. Voros The quantized baker’s transformation, Ann. Phys. 190 (1989), 1–31. 2. R. Bl¨ umel and U. Smilansky, A simple model for chaotic scattering, Physica D 36 (1989), 111–136. 3. N. Chernov and R. Markarian, Ergodic properties of Anosov maps with rectangular holes, Boletim Sociedade Brasileira Matematica 28 (1997), 271–314. 4. N. Chernov, R. Markarian and S. Troubetzkoy, Conditionally invariant measures for Anosov maps with small holes, Ergod. Th. Dyn. Sys. 18 (1998), 1049– 1073. 5. P. Cvitanovi´c and B. Eckhardt, Periodicorbit quantization of chaotic systems, Phys. Rev. Lett. 63 (1989) 823–826 6. S. De Bi`evre, Recent results on quantum map eigenstates, these Proceedings. 7. M. Degli Esposti and S. Graﬃ, editors The mathematical aspects of quantum maps, volume 618 of Lecture Notes in Physics, Springer, 2003. 8. M. Degli Esposti, S. Nonnenmacher and B. Winn, Quantum variance and ergodicity for the baker’s map, to be published in Commun. Math. Phys. (2005), arXiv:mathph/0412058. 9. P. Gaspard and S.A. Rice, Scattering from a classically chaotic repellor, J. Chem. Phys. 90 (1989), 2225–2241; ibid, Semiclassical quantization of the scattering from a classically chaotic repellor, J. Chem. Phys. 90 (1989), 2242–54; ibid, Exact quantization of the scattering from a classically chaotic repellor, J. Chem. Phys. 90 (1989), 2255–2262; Errata, J. Chem. Phys. 91 (1989), 3279– 3280. 10. L. Guillop´e, K. Lin, and M. Zworski, The Selberg zeta function for convex cocompact Schottky groups, Comm. Math. Phys, 245(2004), 149–176.
450
S. Nonnenmacher
11. V. Ivrii, Microlocal Analysis and Precise Spectral Asymptotics, Springer Verlag, 1998. 12. K. Lin, Numerical study of quantum resonances in chaotic scattering, J. Comp. Phys. 176(2002), 295–329. 13. K. Lin and M. Zworski, Quantum resonances in chaotic scattering, Chem. Phys. Lett. 355(2002), 201–205. 14. W. Lu, S. Sridhar, and M. Zworski, Fractal Weyl laws for chaotic open systems, Phys. Rev. Lett. 91(2003), 154101. 15. C. Muscalu, C. Thiele, and T. Tao, A Carlesontype theorem for a Cantor group model of the Scattering Transform, Nonlinearity 16 (2003), 219–246. 16. S. Nonnenmacher and M. Zworski, Distribution of resonances for open quantum maps, preprint, 2005. 17. M. Saraceno, Classical structures in the quantized baker transformation, Ann. Phys. (NY) 199 (1990), 37–60. 18. M. Saraceno and R.O. Vallejos, The quantized Dtransformation, Chaos 6 (1996), 193–199. 19. R. Schack and C.M. Caves, Shifts on a ﬁnite qubit string: a class of quantum baker’s maps, Appl. Algebra Engrg. Comm. Comput. 10 (2000) 305–310 20. H. Schomerus and J. Tworzydlo, Quantumtoclassical crossover of quasibound states in open quantum systems, Phys. Rev. Lett. 93 (2004), 154102. 21. J. Sj¨ ostrand, Geometric bounds on the density of resonances for semiclassical problems, Duke Math. J., 60 (1990), 1–57. 22. J. Sj¨ ostrand, Semiclassical resonances generated by a nondegenerate critical point, Lecture Notes in Math. 1256, 402–429, Springer (Berlin), 1987. 23. J. Sj¨ ostrand and M. Zworski, Lower bounds on the number of scattering poles, Commun. PDE 18 (1993), 847–857. 24. J. Sj¨ ostrand and M. Zworski, Geometric bounds on the density of semiclassical resonances in small domains, in preparation (2005). 25. F. Toscano, R.O. Vallejos and M. Saraceno, Boundary conditions to the semiclassical traces of the baker’s map, Nonlinearity 10 (1997), 965–978. 26. G. Troll and U. Smilansky, A simple model for chaotic scattering, Physica D 35 (1989), 34–64. 27. M. Zworski, Dimension of the limit set and the density of resonances for convex cocompact Riemann surfaces, Inv. Math. 136 (1999), 353–409.
Spectral Shift Function for Magnetic Schr¨ odinger Operators Georgi Raikov Facultad de Ciencias, Universidad de Chile, Las Palmeras 3425, Santiago, Chile [emailprotected]
1 Introduction In this survey article based on the papers [7, 10], and [8], we consider the 3D Schr¨odinger operator with constant magnetic ﬁeld of intensity b > 0, perturbed by an electric potential V which decays fast enough at inﬁnity, and discuss various asymptotic properties of the corresponding spectral shift function. More precisely, let H0 = H0 (b) := (i∇+A)2 −bbe the unperturbed operator, essentially selfadjoint on C0∞ (R3 ). Here A = − bx22 , bx21 , 0 is the magnetic potential which generates the constant magnetic ﬁeld B = curl A = (0, 0, b), b > 0. It is wellknown that σ(H0 ) = σac (H0 ) = [0, ∞) (see [1]), where σ(H0 ) stands for the spectrum of H0 , and σac (H0 ) for its absolutely continuous spectrum. Moreover, the socalled Landau levels 2bq, q ∈ Z+ := {0, 1, . . .}, play the role of thresholds in σ(H0 ). For x = (x1 , x2 , x3 ) ∈ R3 we denote by X⊥ = (x1 , x2 ) the variables on the plane perpendicular to the magnetic ﬁeld. Throughout the paper we assume that the electric potential V satisﬁes V (x) ≤ C0 X⊥ −m⊥ x3 −m3 ,
x = (X⊥ , x3 ) ∈ R3 , (1) with C0 > 0, m⊥ > 2, m3 > 1, and x := (1 + x2 )1/2 , x ∈ Rd , d ≥ 1. Some of our results hold under a more restrictive assumption than (1), namely V ≡ 0,
V ∈ C(R3 ),
V ≡ 0,
V ∈ C(R3 ),
V (x) ≤ C0 x−m0 ,
m0 > 3,
x ∈ R3 .
(2)
Note that (2) implies (1) with any m3 ∈ (0, m0 ) and m⊥ = m0 − m3 . In particular, we can choose m3 ∈ (1, m0 − 2) so that m⊥ > 2. On the domain of H0 deﬁne the operator H = H(b) := H0 + V . Obviously, inf σ(H) ≤ inf σ(H0 ) = 0. Moreover, if (1) holds, then for E < inf σ(H) we have (H − E)−1 − (H0 − E)−1 ∈ S1 where S1 denotes the trace class. Hence, there exists a unique function ξ = ξ(·; H, H0 ) ∈ L1 (R; (1 + E 2 )−1 dE) vanishing on (−∞, inf σ(H)), such that the Lifsh*tsKrein trace formula ξ(E; H, H0 )f (E)dE Tr (f (H) − f (H0 )) = R
C0∞ (R)
(see the original works [20, 22], the survey article holds for each f ∈ [5], or Chapter 8 of the monograph [44]). The function ξ(·; H, H0 ) is called G. Raikov: Spectral Shift Function for Magnetic Schr¨ odinger Operators, Lect. Notes Phys. 690, 451–465 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
452
G. Raikov
the spectral shift function (SSF) for the operator pair (H, H0 ). If E < 0 = inf σ(H0 ), then the spectrum of H below E could be at most discrete, and for almost every E < 0 we have ξ(E; H, H0 ) = −N (E; H)
(3)
where N (E; H) denotes the number of eigenvalues of H lying in the interval (−∞, E), and counted with their multiplicities. On the other hand, for almost every E ∈ [0, ∞), the SSF ξ(E; H, H0 ) is related to the scattering determinant det S(E; H, H0 ) for the pair (H, H0 ) by the BirmanKrein formula det S(E; H, H0 ) = e−2πiξ(E;H,H0 ) (see [4] or [44, Sect. 8.4]). A survey of various asymptotic results concerning the SSF for numerous quantum Hamiltonians is contained in [39]. A priori, the SSF ξ(E; H, H0 ) is deﬁned for almost every E ∈ R. In this article we will identify this SSF with a representative of its equivalence class which is welldeﬁned on R \ 2bZ+ , bounded on every compact subset of R \ 2bZ+ , and continuous on R \ (2bZ+ ∪ σpp (H)) where σpp (H) denotes the set of the eigenvalues of H. In the case of V of deﬁnite sign this representative is described explicitly in Subsect. 2.2 below (see, in particular, (5)). In the case of nonsigndeﬁnite V its description can be found in [7, Sect. 3]. In the present article we investigate the behaviour of the SSF in several asymptotic regimes: – First, we analyse the singularities of the SSF at the Landau levels. In other words, we ﬁx b > 0 and q ∈ Z+ , and investigate the behaviour of ξ(2bq + λ; H, H0 ) as λ → 0. – Further, we study the strongmagneticﬁeld asymptotics of the SSF, i.e. the behaviour of the SSF as b → ∞. Here we distinguish between the asymptotics far from the Landau levels, and the asymptotics near a given Landau level. – Finally, we obtain a Weyl type formula describing the high energy asymptotics of the SSF. The paper is organised as follows. Section 2 contains some auxiliary results: in Subsect. 2.1 we introduce some basic concepts and notations used throughout the paper, while In Subsect. 2.2 we describe the representation of the SSF in the case of perturbations of ﬁxed sign, due to A. Pushnitski (see [29]). In Sect. 3 we formulate our main results, and discuss brieﬂy on them. More precisely, Subsect. 3.1 contains the results on the singularities of the SSF at the Landau levels, Subsect. 3.2 is devoted to the strong magnetic ﬁeld asymptotics of the SSF, and Subsect. 3.3 – to its high energy behaviour. We refer the reader to the original articles [7, 10], and [8], for the proofs of the results presented in this survey.
Spectral Shift Function for Magnetic Schr¨ odinger Operators
453
2 Auxiliary Results 2.1 Notations and Preliminaries In this subsection we give deﬁnitions and introduce notations used throughout the paper. We denote by S∞ the class of linear compact operators acting in a given Hilbert space. Let T = T ∗ ∈ S∞ . Denote by PI (T ) the spectral projection of T associated with the interval I ⊂ R. For s > 0 set n± (s; T ) := rank P(s,∞) (±T ) . For an arbitrary (not necessarily selfadjoint) operator T ∈ S∞ put n∗ (s; T ) := n+ (s2 ; T ∗ T ),
s>0.
Further, we denote by Sp , p ∈ (0, ∞), the Schattenvon Neumann class of ∞ 1 compact operators for which the functional T p : = p 0 sp−1 n∗ (s; T ) ds p is ﬁnite. It is wellknown that if p ∈ [1, ∞), then Sp are Banach spaces with norm ·p . Finally, if T is a bounded linear operator acting in a given Hilbert 1 (T − T ∗ ). space, we set Re T := 12 (T + T ∗ ), Im T := 2i 2.2 A. Pushnitski’s Representation of the SSF In this subsection we introduce a representation of the SSF ξ(·; H, H0 ) in the case of signdeﬁnite potentials V dt Let I ∈ R be a Lebesgue measurable set. Set µ(I) := π1 I 1+t 2. Lemma 1. [29, Lemma 2.1] Let T1 = T1∗ ∈ S∞ and T2 = T2∗ ∈ S1 . Then 1 n± (s1 + s2 ; T1 + t T2 ) dµ(t) ≤ n± (s1 ; T1 ) + T2 1 , s1 , s2 > 0 . (4) πs2 R Suppose that V satisﬁes (1), and ±V ≥ 0; in this case we will write H = H± = H0 ± V . For z ∈ C, Im z > 0, set T (z) := V 1/2 (H0 − z)−1 V 1/2 ∈ S∞ . Proposition 1. [7, Lemma 4.2] Assume that (1) holds, and E ∈ R \ 2bZ+ . Then the operator limit T (E + i0) := n − limδ↓0 T (E + iδ) exists, and we have T (E + i0) ≤ C1 (dist (E, 2bZ+ ))
−1/2
with C1 independent of E and b. Moreover, 0 ≤ Im T (E + i0) ∈ S1 , and if E < 0 then Im T (E + i0) = 0, while for E ∈ (0, ∞) \ 2bZ+ we have [ 2b ] b (E − 2bl)−1/2 V (x)dx Im T (E + i0)1 = Tr Im T (E + i0) = 4π R3 E
l=0
where [x] denotes the integer part of x ∈ R.
454
G. Raikov
By Lemma 1 and Proposition 1, the quantity ˜ ξ(E; H± , H0 ) = ± n∓ (1; Re T (E + i0) + t Im T (E + i0)) dµ(t)
(5)
R
is welldeﬁned for every E ∈ R \ 2bZ+ , and bounded on every compact subset ˜ H± , H0 ) is continuous on of R \ 2bZ+ . Moreover, by [7, Proposition 2.5], ξ(·; R \ {2bZ+ ∪ σpp (H± )}. On the other hand, the general result of [29, Theo˜ rem 1.2] implies ξ(E; H± , H0 ) = ξ(E; H± , H0 ) for almost every E ∈ R. As explained in the introduction, in the case of signdeﬁnite perturbations we ˜ H± , H0 ), while in the case of will identify the SSF ξ(E; H± , H0 ) with ξ(E; nonsigndeﬁnite perturbations, we will identify it with the generalisation of ˜ ξ(E; H± , H0 ) described in [7, Sect. 3] on the basis of the general results of [14] and [30], using the concept of the index of orthogonal projections (see [2]). Here it should be underlined that in contrast to the case b = 0, we cannot rule out the possibility that the operator H has inﬁnite discrete spectrum, or eigenvalues embedded in the continuous spectrum by imposing shortrange conditions on V . First, it is wellknown that if V satisﬁes V (x) ≤ −Cχ(x),
x ∈ R3 ,
(6)
where C > 0, and χ is the characteristic function of a nonempty open subset of R3 , then the discrete spectrum of H is inﬁnite (see [1, Theorem 5.1], [38, Theorem 2.4]). Further, assume that V is axisymmetric, i.e. V = V (X⊥ , x3 ). It is wellknown (see e.g. [1]) that in this case the operators H0 and H are " " (m) unitarily equivalent to the orthogonal sums m∈Z ⊕H (m) and m∈Z ⊕H0 respectively, where the operators (m)
H0
:= −
∂2 1 ∂ ∂ # − + # ∂# ∂# ∂x23
b# m + 2 #
2 − b,
(m)
H (m) := H0
+ V (#, x3 ) ,
(7) are selfadjoint in L2 (R+ × R; #d#dx3 ). Assume moreover that V satisﬁes (6). Then the operator H (m) with m ≥ 0 has at least one eigenvalue in the interval (2bm − V L∞ (R3 ) , 2bm), and hence the operator H has inﬁnitely many eigenvalues embedded in its continuous spectrum (see [1, Theorem 5.1]). Suppose now that V is axisymmetric and satisﬁes the estimate V (X⊥ , x3 ) ≤ −Cχ⊥ (X⊥ ) x3 −m3 ,
(X⊥ , x3 ) ∈ R3 ,
(8)
where C > 0, χ⊥ is the characteristic function of a nonempty open subset of R2 , and m3 ∈ (0, 2) which is compatible with (1) if m3 ∈ (1, 2). Then, using the argument of the proof of [1, Theorem 5.1] and the variational principle, we can easily check that for each m ≥ 0 the operator H (m) has inﬁnitely many discrete eigenvalues which accumulate to the inﬁmum 2bm of its essential spectrum. Hence, if V is axisymmetric and satisﬁes (8), then below each Landau level 2bq, q ∈ Z+ , there exists an inﬁnite sequence of
Spectral Shift Function for Magnetic Schr¨ odinger Operators
455
ﬁnitemultiplicity eigenvalues of H, which converges to 2bq. Note however that the claims in [10, p. 385] and [8, p. 3457] that [1, Theorem 5.1] implies the same phenomenon for axisymmetric nonpositive potentials compactly supported in R3 , are not justiﬁed. The challenging and interesting problem about the accumulation at a given Landau level of embedded eigenvalues and/or resonances of H will be considered in a future work. Finally, we note that generically the only possible accumulation points of the eigenvalues of H are the Landau levels (see [1, Theorem 4.7], [13, Theorem 3.5.3 (iii)]). Further information on the location of the eigenvalues of H can be found in [7, Proposition 2.6].
3 Main Results 3.1 Singularities of the SSF at the Landau Levels Introduce the Landau Hamiltonian 2 2 ∂ bx2 ∂ bx1 h(b) := i − + i + −b, ∂x1 2 ∂x2 2 essentially selfadjoint on C0∞ (R2 ). It is wellknown that σ(h(b)) = ∪∞ q=0 {2bq} and each eigenvalue 2bq, q ∈ Z+ , has inﬁnite multiplicity (see e.g. [1]). Denote by pq = pq (b) the orthogonal projection onto the eigenspace Ker (h(b) − 2bq), q ∈ Z+ . The estimates of the SSF for energies near the Landau level 2bq, q ∈ Z+ , will be given in the terms of traces of functions of Toeplitz operators pq U pq where U : R2 → R decays in a certain sense at inﬁnity. Lemma 2. [31, Lemma 5.1], [10, Lemma 3.1] Let U ∈ Lr (R2 ), r ≥ 1, and q ∈ Z+ . Then pq U pq ∈ Sr . Assume that (1) holds. For X⊥ ∈ R2 set W (X⊥ ) := R V (X⊥ , x3 )dx3 . Since V satisﬁes (1), we have W ∈ L1 (R2 ), and Lemma 2 with U = W implies pq W pq ∈ S1 , q ∈ Z+ . Evidently, pq W pq ≥ 0, and it follows from V ≡ 0 and V ∈ C(R2 ), that rank pq W pq = ∞ for all q ∈ Z+ (see below Lemma 5). If, moreover, V satisﬁes (2), then 0 ≤ W (X⊥ ) ≤ C0 X⊥ −m0 +1 , X⊥ ∈ R2 , with C0 = C0 R x−m0 dx. In the following two theorems we assume that V has a deﬁnite sign. As in Subsect. 2.2, if ±V ≥ 0, we will write H = H± = H0 ± V . Theorem 1. [10, Theorem 3.1] Assume that (2) is valid, and ±V ≥ 0. Let q ∈ Z+ , b > 0. Then the asymptotic estimates ξ(2bq − λ; H+ , H0 ) = O(1) , √ −n+ ((1 − ε)2 λ; pq W pq ) + O(1) ≤ ξ(2bq − λ; H− , H0 ) ≤ √ − n+ ((1 + ε)2 λ; pq W pq ) + O(1), hold as λ ↓ 0 for each ε ∈ (0, 1).
(9)
456
G. Raikov
Suppose that V satisﬁes (1). For λ ≥ 0 deﬁne the matrixvalued function w11 w12 Wλ = Wλ (X⊥ ) := , X⊥ ∈ R2 , w21 w22
where w11 :=
R
w12 = w21 :=
R
√ √ V (X⊥ , x3 ) cos ( λx3 ) sin ( λx3 )dx3 ,
w22 :=
√ V (X⊥ , x3 ) cos2 ( λx3 )dx3 ,
R
√ V (X⊥ , x3 ) sin2 ( λx3 )dx3 .
It is easy to check that for λ ≥ 0 and q ∈ Z+ the operator pq Wλ pq : L2 (R2 )2 → L2 (R2 )2 satisﬁes 0 ≤ pq Wλ pq ∈ S1 , and rank pq Wλ pq = ∞. Theorem 2. [10, Theorem 3.2] Assume that (2) is valid, and ±V ≥ 0. Let q ∈ Z+ , b > 0. Then the asymptotic estimates √ 1 ± Tr arctan (((1 ± ε)2 λ)−1 pq Wλ pq ) + O(1) ≤ ξ(2bq + λ; H± , H0 ) ≤ π √ 1 (10) ± Tr arctan (((1 ∓ ε)2 λ)−1 pq Wλ pq ) + O(1) π hold as λ ↓ 0 for each ε ∈ (0, 1). Relations (9) and (10) allow us to reduce the analysis of the behaviour as λ → 0 of ξ(2bq + λ; H± , H0 ), to the study of the asymptotic distribution of the eigenvalues of Toeplitztype operators pq U pq . The following three lemmas concern the spectral asymptotics of such operators. Lemma 3. [31, Theorem 2.6] Let the function 0 ≤ U ∈ C 1 (R2 ) satisfy the estimates U (X⊥ ) = u0 (X⊥ /X⊥ )X⊥ −α (1 + o(1)), ∇U (X⊥ ) ≤ C1 X⊥ −α−1 ,
X⊥  → ∞ ,
X⊥ ∈ R2 ,
where α > 0, and u0 is a continuous function on S1 which is nonnegative and does not vanish identically. Then for each q ∈ Z+ we have n+ (s; pq U pq ) =
. b X⊥ ∈ R2 U (X⊥ ) > s (1 + o(1)) 2π
= ψα (s) (1 + o(1)),
s↓0,
where . denotes the Lebesgue measure, and b ψα (s) := s−2/α u0 (t)2/α dt, 4π S1
s>0.
(11)
Spectral Shift Function for Magnetic Schr¨ odinger Operators
457
Lemma 4. [38, Theorem 2.1, Proposition 4.1] Let 0 ≤ U ∈ L∞ (R2 ). Assume that ln U (X⊥ ) = −µX⊥ 2β (1 + o(1)), X⊥  → ∞ , for some β ∈ (0, ∞), µ ∈ (0, ∞). Then for each q ∈ Z+ we have n+ (s; pq U pq ) = ϕβ (s)(1 + o(1)),
s↓0,
where
b 1/β if 0 < β < 1 , 2µ1/β lns 1 lns if β = 1, ϕβ (s) := ln(1+2µ/b) β (lnlns)−1 lns if 1 < β < ∞, β−1
s ∈ (0, e−1 ) .
(12)
Lemma 5. [38, Theorem 2.2, Proposition 4.1] Let 0 ≤ U ∈ L∞ (R2 ). Assume that the support of U is compact, and there exists a constant C > 0 such that U ≥ C on an open nonempty subset of R2 . Then for each q ∈ Z+ we have n+ (s; pq U pq ) = ϕ∞ (s) (1 + o(1)), where
ϕ∞ (s) := (lnlns)−1 lns,
s↓0,
s ∈ (0, e−1 ) .
(13)
Employing Lemmas 3, 4, 5, we easily ﬁnd that asymptotic estimates (9) and (10) entail the following Corollary 1. [10, Corollaries 3.1 – 3.2] Let (2) hold with m0 > 3. i) Assume that the hypotheses of Lemma 3 hold with U = W and α > 2. Then √ b ξ(2bq − λ; H− , H0 ) = − X⊥ ∈ R2 W (X⊥ ) > 2 λ (1 + o(1)) 2π √ = −ψα (2 λ) (1 + o(1)), λ ↓ 0 , (14) √ b ξ(2bq + λ; H± , H0 ) = ± 2 arctan ((2 λ)−1 W (X⊥ ))dX⊥ (1 + o(1)) 2π R2 √ 1 ψα (2 λ) (1 + o(1)), λ ↓ 0 . =± 2 cos (π/α) the function ψα being deﬁned in (11). ii) Assume that the hypotheses of Lemma 4 hold with U = W . Then √ ξ(2bq − λ; H− , H0 ) = −ϕβ (2 λ) (1 + o(1)), λ ↓ 0, β ∈ (0, ∞) , the functions ϕβ being deﬁned in (12). If, in addition, V satisﬁes (1) for some m⊥ > 2 and m3 > 2, we have ξ(2bq + λ; H± , H0 ) = ±
√ 1 ϕβ (2 λ) (1 + o(1)), 2
λ ↓ 0,
β ∈ (0, ∞) .
458
G. Raikov
iii) Assume that the hypotheses of Lemma 5 hold with U = W . Then √ ξ(2bq − λ; H− , H0 ) = −ϕ∞ (2 λ) (1 + o(1)), λ ↓ 0 , the function ϕ∞ being deﬁned in (13). If, in addition, V satisﬁes (1) for some m⊥ > 2 and m3 > 2, we have ξ(2bq + λ; H± , H0 ) = ±
√ 1 ϕ∞ (2 λ) (1 + o(1)), 2
λ↓0,
the function ϕ∞ being deﬁned in (13). In particular, we ﬁnd that lim λ↓0
ξ(2bq − λ; H− , H0 ) 1 = π ξ(2bq + λ; H− , H0 ) 2 cos α
(15)
if W has a powerlike decay at inﬁnity (more precisely, if the assumptions of Corollary 1 i) hold), or lim λ↓0
1 ξ(2bq − λ; H− , H0 ) = ξ(2bq + λ; H− , H0 ) 2
(16)
if W decays exponentially or has a compact support (more precisely, if the assumptions of Corollary 1 ii)  iii) are fulﬁlled). Relations (15) and (16) could be interpreted as analogues of the classical Levinson formulae (see e.g. [39]). Remarks: √ i) Since the ranks of pq W pq√and pq Wλ pq are inﬁnite, the quantities n+ (s2 λ; pq W pq ) and Tr arctan ((s2 λ)−1 pq Wλ pq ) tend to inﬁnity as λ ↓ 0 for every s > 0. Therefore, Theorems 1 and 2 imply that the SSF ξ(·; H± , H0 ) has a singularity at each Landau level. The existence of singularities of the SSF at strictly positive energies is in sharp contrast with the nonmagnetic case b = 0 where the SSF ξ(E; −∆ + V, −∆) is continuous for E > 0 (see e.g. [39]). The main reason for this phenomenon is the fact that the Landau levels play the role of thresholds in σ(H0 ) while the free Laplacian −∆ has no strictly positive thresholds in its spectrum. It is conjectured that the singularity of the SSF ξ(·; H± (b), H0 (b)), b > 0, at a given Landau level 2bq, q ∈ Z+ , could be related to a possible accumulation of resonances and/or eigenvalues of H at 2bq. Here it should be recalled that in the case b = 0 the high energy asymptotics (see [27]) and the semiclassical asymptotics (see [28]) of the derivative of the SSF for appropriate compactly supported perturbations of the Laplacian, are related by the BreitWigner formula to the asymptotic distribution near the real axis of the resonances deﬁned as poles of the meromorphic continuation of the resolvent of the perturbed operator. ii) In the case q = 0, when by (3) we have ξ(−λ; H− , H0 ) = −N (−λ; H− ) for λ > 0, asymptotic relations of the type of (14) have been known since long ago (see [42], [41], [17,31,43]). An important characteristic feature of the methods
Spectral Shift Function for Magnetic Schr¨ odinger Operators
459
used in [31], and later in [38], is the systematic use, explicit or implicit, of the connection between the spectral theory of the Schr¨ odinger operator with constant magnetic ﬁeld, and the theory of Toeplitz operators acting in holomorphic spaces of FockSegalBargmann type and the related pseudodiﬀerential operators with generalised antiWick symbols (see [3, 12, 15, 40]). Various important aspects of the interaction between these two theories have been discussed in [37] and [7, Sect. 9]. The Toeplitzoperator approach turned to be especially fruitful in [38] where electric potentials decaying rapidly at inﬁnity (i.e. decaying exponentially, or having compact support) were considered (see Lemmas 4  5). It is shown in [11] that the precise spectral asymptotics for the Landau Hamiltonian perturbed by a compactly supported electric potential U of ﬁxed sign recovers the logarithmic capacity of the support of U. iii) Let us mention several other existing extensions of Lemmas 3 – 5. Lemmas 3 and 5 have been generalised to the multidimensional case where pq is the orthogonal projection onto a given eigenspace of the Schr¨ odinger operator with constant magnetic ﬁeld of full rank, acting in L2 (R2d ), d > 1 (see [31] and [25] respectively). Moreover, Lemma 5 has been generalised in [25] to a relativistic setting where pq is an eigenprojection of the Dirac operator. Finally, in [36] Lemmas 3 – 5 have been extended to the case of the 2D Pauli operator with variable magnetic ﬁeld from a certain class including the almost periodic ﬁelds with nonzero mean value (in this case the role of the Landau levels is played by the origin), and electric potentials U satisfying the assumptions of Lemmas 3 – 5. In the case of compactly supported U of deﬁnite sign, [11] contains a more precise version of the corresponding result of [36], involving again the logarithmic capacity of the support of U . iv) To the author’s best knowledge, the singularities at the Landau levels of the SSF for the 3D Schr¨ odinger operator in constant magnetic ﬁeld have been investigated for the ﬁrst time in [10]. However, it is appropriate to mention here the article [19] where axisymmetric potentials V have been considered. For a ﬁxed magnetic quantum number m ∈ Z the authors of [19] studied the (m) (m) (m) behaviour of the SSF ξ(E; H , H0 ) for the operator pair H (m) , H0 deﬁned in (7) at energies E near the Landau level 2m if m > 0, and near the origin if m ≤ 0, and of the classical Levinson formulae, deduced analogues (m) (m) . Later, the methods in [19] were developed concerning the pair H , H0 in [23] and [24]. However, it is not possible to recover the results of Theorem 1, Theorem 2 and Corollary 1 from the results of [19], [23], and [24] even in the case of axisymmetric V . v) Finally, [16] contains general bounds on the SSF for appropriate pairs of magnetic Schr¨odinger operators. These bounds are applied in order to deduce Wegner estimates of the integrated density of states for some random alloytype models.
460
G. Raikov
3.2 Strong Magnetic Field Asymptotics of the SSF Our ﬁrst theorem in this subsection treats the asymptotics as b → ∞ of ξ(·; H(b), H0 (b)) far from the Landau levels. Theorem 3. [7, Theorem 2.1] Let (1) hold. Assume that E ∈ (0, ∞) \ 2Z+ , and λ ∈ R. Then ξ(Eb + λ; H(b), H0 (b)) =
[E/2] b1/2 −1/2 (E − 2l) V (x)dx + O(1), 4π 2 R3
b→∞.
l=0
(17) The following two theorems concern the asymptotics of the SSF near a given Landau level. In order to formulate our next theorem, we introduce the following selfadjoint operators χ0 := −d2 /dx23 ,
χ = χ(X⊥ ) := χ0 + V (X⊥ , .),
X⊥ ∈ R2 ,
which are deﬁned on the Sobolev space H2 (R), and depend on the parameter X⊥ ∈ R2 . If (1) holds, then (χ(X⊥ ) − λ0 )−1 − (χ0 − λ0 )−1 ∈ S1 for each X⊥ ∈ R2 and λ0 < inf σ(χ(X⊥ )). Hence, the SSF ξ(.; χ(X⊥ ), χ0 ) is welldeﬁned. Set Λ: = minX⊥ ∈R2 inf σ(χ(X⊥ )). Evidently, Λ ∈ [−C0 , 0]. Moreover, Λ = limb→∞ inf σ(H(b)) (see [1, Theorem 5.8]). Proposition 2. (cf. [7, Proposition 2.2]) Assume that (1) holds. i) For each λ ∈ R \ {0} we haveξ(λ; χ(.), χ0 ) ∈ L1 (R2 ). ii) The function (0, ∞) λ → R2 ξ(λ; χ(X⊥ ), χ0 )dX⊥ is continuous, while the nonincreasing function (−∞, 0) λ → ξ(λ; χ(X⊥ ), χ0 )dX⊥ = − N (λ; χ(X⊥ ))dX⊥ R2
R2
(see (3)), is continuous at the point λ < 0 if and only if {X⊥ ∈ R2 λ ∈ σ(χ(X⊥ ))} = 0 . (18) iii) Assume ±V ≥ 0. If λ > Λ, λ = 0, then ± R2 ξ(λ; χ(X⊥ ), χ0 )dX⊥ > 0. Remark: The third part of Proposition 2 is not included in [7, Proposition 2.2], but it follows easily from A. Pushnitski’s representation of the SSF (see [29]), and the hypotheses V ≡ 0 and V ∈ C(R3 ). Theorem 4. [7, Theorem 2.3] Assume that (1) holds. Let q ∈ Z+ , λ ∈ R \ {0}. If λ < 0, suppose also that (18) holds. Then we have 1 −1 lim b ξ(2bq + λ; H(b), H0 (b)) = ξ(λ; χ(X⊥ ), χ0 ) dX⊥ . (19) b→∞ 2π R2
Spectral Shift Function for Magnetic Schr¨ odinger Operators
461
By Proposition 2 iii), if ±V ≥ 0, then the r.h.s. of (19) is diﬀerent from zero if λ > Λ, λ = 0. Unfortunately, we cannot prove the same for general nonsigndeﬁnitepotentials V . On the other hand, it is obvious that for arbitrary V we have R2 ξ(λ; χ(X⊥ ), χ0 )dX⊥ = 0 if λ < Λ. The last theorem of this subsection contains a more precise version of (19) for the case λ < Λ. Theorem 5. [7, Theorem 2.4] Let (1) hold. i) Let λ < Λ. Then for suﬃciently large b > 0 we have ξ(λ; H(b), H0 (b)) = 0. ii) Let q ∈ Z+ , q ≥ 1, λ < Λ. Assume that the partial derivatives of x3 m3 V with respect to the variables X⊥ ∈ R2 exist, and are uniformly bounded on R3 . Then we have lim b−1/2 ξ(2bq + λ; H(b), H0 (b)) =
b→∞
q−1 1 −1/2 (2(q − l)) V (x)dx . (20) 4π 2 R3 l=0
Remarks: i) Relations (17), (19), and (20) can be uniﬁed into a single asymptotic formula. In order to see this, notice that a general result on the highenergy asymptotics of the SSF for 1D Schr¨ odinger operators (see e.g. [39]) implies, in particular, that 1 1/2 V (X⊥ , x3 ) dx3 , X⊥ ∈ R2 . lim E ξ(E; χ(X⊥ ), χ0 ) = E→∞ 2π R Then relation (17) with 0 < E ∈ 2Z+ , or relations (19) and (20) with E = 2q, q ∈ Z+ , entail ξ(Eb + λ; H(b), H0 (b)) [E/2] b ξ(b(E − 2l) + λ; χ(X⊥ ), χ0 )dX⊥ (1 + o(1)), b → ∞ . (21) = 2π R2 l=0
ii) By (3) for λ < 0 we have ξ(λ; H(b), H0 ) = −N (λ; H(b)). The asymptotics as b → ∞ of the counting function N (λ; H0 (b)) with λ < 0 ﬁxed, has been investigated in [32] under considerably less restrictive assumptions on V than in Theorems 3–5. The asymptotic properties as λ ↑ 0, and as λ ↓ Λ if 1 N (λ; χ(X⊥ ))dX⊥ which appears Λ < 0, of the asymptotic coeﬃcient − 2π R2 at the r.h.s. of (19) in the case of a negative perturbation, have been studied in [33]. The asymptotic distribution of the discrete spectrum for the 3D magnetic Pauli and Dirac operators in strong magnetic ﬁelds has been considered in [34] and [35] respectively. The main purpose in [32, 34], and [35] was to obtain the main asymptotic term (without any remainder estimates) of the corresponding counting function under assumptions close to the minimal ones which guarantee that the Hamiltonians are selfadjoint, and the asymptotic coeﬃcient is welldeﬁned. Other results which again describe the asymptotic distribution of the discrete spectrum of the Schr¨ odinger and Dirac operator in strong magnetic ﬁelds, but contain also sharp remainder estimates, have
462
G. Raikov
been obtained [9, 17], and [18] under assumptions on V which, naturally, are more restrictive than those in [32, 34], and [35]. iii) Generalizations of asymptotic relation (17) in several directions can be found in [26]. In particular, [26, Theorem 4] implies that if V ∈ S(R3 ), then the SSF ξ(Eb+λ; H(b), H0 (b)), E ∈ (0, ∞)\2Z+ , λ ∈ R, admits an asymptotic expansion of the form ξ(Eb + λ; H(b), H0 (b)) ∼
∞
cj b
1−2j 2
b→∞.
,
j=0
iv) Together with the pointwise asymptotics as b → ∞ of the SSF for the pair (H0 (b), H(b)) (see (17), (19), or (20)), it also is possible to consider its weak asymptotics, i.e. the asymptotics of the convolution of the SSF with an arbitrary ϕ ∈ C0∞ (R). Results of this type are contained in [6]. 3.3 High Energy Asymptotics of the SSF Theorem 6. [8, Theorem 2.1] Assume that V satisﬁes (1). Then we have 1 −1/2 lim E ξ(E; H, H0 ) = V (x)dx, r ∈ (0, b) , (22) E→∞,E∈Or 4π 2 R3 where Or := {E ∈ (0, ∞)dist(E, 2bZ+ ) ≥ r}. Remarks: i) It is essential to avoid the Landau levels in (22), i.e. to suppose that E ∈ Or , r ∈ (0, b), as E → ∞, since by Theorems 1  2, the SSF has singularities at the Landau levels, at least in the case ±V ≥ 0. ii) For E ∈ R set Θ(E − p + A(x)2 ) − Θ(E − p + A(x)2 − V (x)) dxdp = ξcl (E) := T ∗ R3
4π 3
R3
3/2
3/2
E+ − (E − V (x))+
dx
0 if s ≤ 0 , is the Heaviside function. Note that ξcl (E) 1 if s > 0, is independent of the magnetic ﬁeld b ≥ 0. Evidently, under the assump−1/2 ξcl (E) = 2π R3 V (x)dx. Hence, if tions of Theorem 6 we have limE→∞ E V (x)dx = 0, then (22) is equivalent to R3
where Θ(s) :=
ξ(E; H, H0 ) = (2π)−3 ξcl (E)(1 + o(1)),
E → ∞,
E ∈ Or ,
r ∈ (0, b) .
iii) As far as the author is informed, the highenergy asymptotics of the SSF for 3D Schr¨ odinger operators in constant magnetic ﬁelds was investigated for the ﬁrst time in [8]. Nonetheless, in [19] the asymptotic behaviour as E → ∞, (m) (m) E ∈ Or , of the SSF ξ(E; H (m) , H0 ) for the operator pair (H (m) , H0 ) (see (7)) with ﬁxed m ∈ Z has been been investigated. It does not seem possible to deduce (22) from the results of [19] even in the case of axisymmetric V .
Spectral Shift Function for Magnetic Schr¨ odinger Operators
463
Acknowledgements I would like to thank my coauthors Vincent Bruneau, Claudio Fern´ andez, and Alexander Pushnitski for having let me present our joint results in this survey article. The major part of this work has been done during my visit to the Institute of Mathematics of the Czech Academy of Sciences in December 2004. I am sincerely grateful to Miroslav Engliˇs for his warm hospitality. The partial support by the Chilean Science Foundation Fondecyt under Grant 1050716 is acknowledged.
References 1. J. Avron, I. Herbst, B. Simon, Schr¨ odinger operators with magnetic ﬁelds. I. General interactions, Duke Math. J. 45 (1978), 847–883. 2. J. Avron, R. Seiler, B. Simon, The index of a pair of projections, J. Funct. Anal. 120 (1994), 220–237. 3. F.A. Berezin, M.A. Shubin, The Schr¨ odinger Equation. Kluwer Academic Publishers, Dordrecht, 1991. ˇ Birman, M.G. Kre˘in, On the theory of wave operators and scattering 4. M.S. operators, Dokl. Akad. Nauk SSSR 144 (1962), 475–478 (Russian); English translation in Soviet Math. Doklady 3 (1962). ˇ Birman, D.R. Yafaev, The spectral shift function. The papers of M. G. 5. M.S. Kre˘in and their further development, Algebra i Analiz 4 (1992), 1–44 (Russian); English translation in St. Petersburg Math. J. 4 (1993), 833–870. 6. V. Bruneau, M. Dimassi, Weak asymptotics of the spectral shift function in strong constant magnetic ﬁeld, Preprint 2004 (to appear in Math. Nachr.) 7. V. Bruneau, A. Pushnitski, G.D. Raikov, Spectral shift function in strong magnetic ﬁelds, Algebra i Analiz 16 (2004), 207  238; see also St. Petersburg Math. Journal 16 (2005), 181–209. 8. V. Bruneau, G. Raikov, High energy asymptotics of the magnetic spectral shift function, J. Math. Phys. 45 (2004), 3453–3461. 9. M. Dimassi, D´eveloppements asymptotiques de l’op´ erateur de Schr¨ odinger avec champ magn´ etique fort, Commun. Part. Diﬀ. Eqns 26 (2001), 595–627. ´ndez, G.D. Raikov, On the singularities of the magnetic spectral 10. C. Ferna shift function at the Landau levels, Ann. Henri Poincar´e 5 (2004), 381–403. 11. N. Filonov, A. Pushnitski, Spectral asymptotics for Pauli operators and orthogonal polynomials in complex domains, Preprint 2005; available at http://arxiv.org/pdf/math.SP/0504044. (to appear in Comm. Math. Phys.) 12. V.Fock, Bemerkung zur Quantelung des harmonischen Oszillators im Magnetfeld, Z. Physik 47 (1928), 446–448. ´rard, I. Laba, Multiparticle Quantum Scattering in Constant Mag13. C. Ge netic Fields, Mathematical Surveys and Monographs, 90, AMS, Providence, RI, 2002. 14. F. Gesztesy, K. Makarov, The Ξ operator and its relation to Krein’s spectral shift function, J. Anal. Math. 81 (2000), 139–183. 15. B.C. Hall, Holomorphic methods in analysis and mathematical physics, In: First Summer School in Analysis and Mathematical Physics, Cuernavaca Morelos, 1998, 1–59, Contemp.Math. 260, AMS, Providence, RI, 2000.
464
G. Raikov
´, 16. D. Hundertmark, R. Killip, S. Nakamura, P. Stollmann, I. Veselic Bounds on the spectral shift function and the density of states, Comm. Math. Phys. 262 (2006), 489–503. 17. V. Ivrii, Microlocal Analysis and Precise Spectral Asymptotics, Springer monographs in Math. Springer, Berlin, 1998. 18. V. Ivrii, Sharp spectral asymptotics for magnetic Schr¨ odinger operator with irregular potential., Russian Journal of Mathematical Physics, 11 (2004), 415– 428. 19. V. Kostrykin, A. Kvitsinsky, S. Merkuriev, Potential scattering in constant magnetic ﬁeld: spectral asymptotics and Levinson formula, J. Phys. A 28 (1995), 3493–3509. 20. M.G. Krein, On the trace formula in perturbation theory, Mat. Sb. 33 (1953), 597–626 (Russian). 21. L. Landau, Diamagnetismus der Metalle, Z. Physik 64 (1930), 629–637. 22. I.M. Lifsh*ts, On a problem in perturbation theory, Uspekhi Mat. Nauk 7 (1952), 171–180 (Russian). 23. M. Melgaard, New approach to quantum scattering near the lowest Landau threshold for a Schr¨ odinger operator with a constant magnetic ﬁeld, FewBody Systems 32 (2002), 1–22. 24. M. Melgaard, Quantum scattering near the lowest Landau threshold for a Schr¨ odinger operator with a constant magnetic ﬁeld, Cent. Eur. J. Math. 1, (2003) 477–509. 25. M. Melgaard, G. Rozenblum, Eigenvalue asymptotics for weakly perturbed Dirac and Schr¨ odinger operators with constant magnetic ﬁelds of full rank, Commun. Part. Diﬀ. Eqns 28 (2003), 697–736. 26. L. Michel, Scattering amplitude and scattering phase for the Schr¨ odinger equation with strong magnetic ﬁeld, J. Math. Phys. 46 (2005) 043514, 18 pp. 27. V. Petkov, M. Zworski, BreitWigner approximation and the distribution of resonances Comm. Math. Phys. 204 (1999) 329–351; Erratum: Comm. Math. Phys. 214 (2000) 733–735. 28. V. Petkov, M. Zworski, Semiclassical estimates on the scattering determinant, Ann. Henri Poincar´e 2 (2001) 675–711. 29. A. Pushnitski˘i, A representation for the spectral shift function in the case of perturbations of ﬁxed sign, Algebra i Analiz 9 (1997), 197–213 (Russian); English translation in St. Petersburg Math. J. 9 (1998), 1181–1194. 30. A. Pushnitski, The spectral shift function and the invariance principle, J. Funct. Anal. 183 (2001), 269–320. 31. G. D. Raikov, Eigenvalue asymptotics for the Schr¨ odinger operator with hom*ogeneous magnetic potential and decreasing electric potential. I. Behaviour near the essential spectrum tips, Commun. P.D.E. 15 (1990), 407–434; Errata: Commun. P.D.E. 18 (1993), 1977–1979. 32. G.D. Raikov, Eigenvalue asymptotics for the Schr¨ odinger operator in strong constant magnetic ﬁelds, Comm. Partial Diﬀerential Equations 23 (1998), no. 9–10, 1583–1619. 33. G.D. Raikov, Asymptotic properties of the magnetic integrated density of states, Electron. J. Diﬀ. Eq. 1999, No. 13, 27 pp. (1999). 34. G.D. Raikov, Eigenvalue asymptotics for the Pauli operator in strong nonconstant magnetic ﬁelds, Ann.Inst.Fourier, 49 (1999), 1603–1636. 35. G.D. Raikov, Eigenvalue asymptotics for the Dirac operator in strong constant magnetic ﬁelds, Math.Phys.Electr.J., 5 (1999), No.2, 22 pp.
Spectral Shift Function for Magnetic Schr¨ odinger Operators
465
36. G.D. Raikov, Spectral asymptotics for the perturbed 2D Pauli operator with oscillating magnetic ﬁelds. I. Nonzero mean value of the magnetic ﬁeld, Markov Process. Related Fields 9 (2003) 775–794. 37. G.D. Raikov, M. Dimassi, Spectral asymptotics for quantum Hamiltonians in strong magnetic ﬁelds, Cubo Mat. Educ. 3 (2001), 317  391. 38. G.D. Raikov, S. Warzel, Quasiclassical versus nonclassical spectral asymptotics for magnetic Schr¨ odinger operators with decreasing electric potentials, Rev. Math. Phys. 14 (2002), 1051–1072. 39. D. Robert, Semiclassical asymptotics for the spectral shift function, In: Diﬀerential Operators and Spectral theory, AMS Translations Ser. 2 189, 187–203, AMS, Providence, RI, 1999. 40. M.A. Shubin, Pseudodiﬀerential Operators and Spectral Theory, Berlin etc.: SpringerVerlag. (1987). 41. A.V. Sobolev, Asymptotic behavior of energy levels of a quantum particle in a hom*ogeneous magnetic ﬁeld, perturbed by an attenuating electric ﬁeld. I, Probl. Mat. Anal. 9 (1984), 67–84 (Russian); English translation in: J. Sov. Math. 35 (1986), 2201–2212. 42. S.N. Solnyshkin, Asymptotic behaviour of the energy of bound states of the Schr¨ odinger operator in the presence of electric and hom*ogeneous magnetic ﬁelds, Probl. Mat. Fiziki, Leningrad University 10 (1982), 266–278 (Russian); Engl. transl. in Selecta Math. Soviet. 5 (1986), 297–306. 43. H. Tamura, Asymptotic distribution of eigenvalues for Schr¨ odinger operators with hom*ogeneous magnetic ﬁelds, Osaka J. Math. 25 (1988), 633–647. 44. D.R. Yafaev, Mathematical scattering theory. General theory. Translations of Mathematical Monographs, 105 AMS, Providence, RI, 1992.
Counting String/M Vacua Steve Zelditch Department of Mathematics, Johns Hopkins University, Baltimore, MD 21218, USA∗ [emailprotected]
Abstract. We report on some recent work with M. R. Douglas and B. Shiﬀman on vacuum statistics for ﬂux compactiﬁcations in string/M theory.
1 Introduction According to string/M theory, the vacuum state of our universe is a 10 dimensional spacetime of the form M 3,1 × X, where M 3,1 is Minkowski space and X is a small 3complex dimensional CalabiYau manifold X known as the “small” or “extra” dimensions [6, 28]. The vacuum selection problem is that there are many candidate vacua for the CalabiYau 3fold X. Here, we report on recent joint work with B. Shiﬀman and M. R. Douglas devoted to counting the number of supersymmetric vacua of type IIb ﬂux compactiﬁcations [12–14]. We also describe closely related the physics articles of AshokDouglas and DenefDouglas [2, 8, 11] on the same problem. At the time of writing of this article, vacuum statistics is being intensively investigated by many string theorists (see for instance [1, 7, 10, 19, 30] in addition to the articles cited above). One often hears that the number of possible vacua is of order 10500 (see e.g. [4]). This large ﬁgure is sometimes decried (at this time) as a blow to predictivity of string/M theory or extolled as giving string theory a rich enough “landscape” to contain vacua that match the physical parameters (e.g. the cosmological constant) of our universe. However, it is very diﬃcult to obtain suﬃciently accurate results on vacuum counting to justify the claims of 10500 total vacua, or even the existence of one vacuum which is consistent with known physical parameters. The purpose of our work is to develop methods and results relevant to accurate vacuum counting. From a mathematical viewpoint, supersymmetric vacua are critical points ∇G W (Z) = 0
(1)
of certain holomorphic sections WG called ﬂux superpotentials of a line bundle L → C over the moduli space C of complex structures on X × T 2 where T 2 = R2 /Z2 . Flux superpotentials depend on a choice of ﬂux G ∈ ∗
Research partially supported by NSF grant DMS0302518
S. Zelditch: Counting string/M vacua, Lect. Notes Phys. 690, 467–483 (2006) c SpringerVerlag Berlin Heidelberg 2006 www.springerlink.com
468
S. Zelditch
√ H 3 (X, Z⊕ −1Z). There is a constraint on G called the “tadpole constraint’, so that G is a lattice point lying in a certain hyperbolic shell 0 ≤ Q[G] ≤ L in H 3 (X, C) (16). Our goal is to count all critical points of all ﬂux superpotentials WG in a given compact set of C as G ranges over such lattice points. Thus, counting vacua in K ⊂ C is a combination of an equidistribution problem for projections of lattice points and an equidistribution problem for critical points of random holomorphic sections. The work we report on gives a rigorous foundation for the program initiated by M. R. Douglas [11] to count vacua by making an approximation to the Gaussian ensembles the other two authors were using to study statistics of zeros of random holomorphic sections (cf. [3, 27]). The results we describe here are the ﬁrst rigorous results on counting vacua in a reasonably general class of models (type IIb ﬂux compaticiﬁcations). They are admittedly still in a rudimentary stage, in particular because they are asymptotic rather than eﬀective. We will discuss the diﬃculties in making them eﬀective below. This report is a written version of our talk at the QMath9 conference in Giens in October, 2004. A more detailed expository article with background on statistical algebraic geometry as well as string theory is given in [32], which was based on the author’s AMS address in Atlanta, January 2005.
2 Type IIb Flux Compactiﬁcations of String/M Theory The string/M theories we consider are type IIb string theories compactiﬁed on a complex 3dimensional CalabiYau manifold X with ﬂux [2, 18–20, 23]. We recall that a CalabiYau 3fold is a compact complex manifold X of dimension 3 with trivial canonical bundle KX , i.e. c1 (X) = 0 [21, 22]. Such X possesses a unique Ricci ﬂat K¨ ahler metric in each K¨ahler class. In what follows, we ﬁx the K¨ ahler class, and then the CY metrics correspond to the complex structures on X. We denote the moduli space of complex structures on X by MC . In addition to the complex structure moduli on X there is an extra parameter τ called the dilaton axion, which ranges over complex structure moduli on T 2 = R2 /Z2 . Hence, the full conﬁguration space C of the model is the product C = MC × E, (Z = (z, τ ); z ∈ MC , τ ∈ E)
(2)
where E = H/SL(2, Z) is the moduli space of complex 1tori (elliptic curves). One can think of C as a moduli space of complex structures on the CY 4fold X × T 2. By “ﬂux” is meant a complex integral 3form √ G = F + iH ∈ H 3 (X, Z ⊕ −1Z) . (3) The ﬂux superpotential WG (Z) corresponding to G is deﬁned as follows: On a CalabiYau 3fold, the space Hz3,0 (X) of holomorphic (3, 0)forms for each
Counting String/M Vacua
469
complex structure z on X has dimension 1, and we denote a holomorphically varying family by Ωz ∈ Hz3,0 (X). Given G as in (3) and τ ∈ H, physicists deﬁne the superpotential corresponding to G, τ by: WG (z, τ ) = (F − τ H) ∧ Ωz . (4) X
This is not welldeﬁned as a function on C, since Ωz is not unique and τ corresponds to the holomorphically varying form ωτ = dx + τ dy ∈ Hτ1,0 (T 2 ) which is not unique either. To be more precise, we deﬁne WG to be a holomorphic section of a line bundle L → C, namely the dual line bundle to the 4,0 = Hz3,0 (X) ⊗ H 1,0 (T 2 ) → C. We form the 4form on Hodge line bundle Hz,τ 2 X ×T ˜ = F ∧ dy + H ∧ dx G and deﬁne a linear functional on Hz3,0 (X) ⊗ Hτ1,0 (T 2 ) by ˜ ∧ Ωz ∧ ω τ . G
WG (z, τ ), Ωz ∧ ωτ =
(5)
X×T 2
When ωτ = dx+τ dy we obtain the original formula. As Z = (z, τ ) ∈ C varies, (5) deﬁnes a holomorphic section of the line bundle L dual to Hz3,0 ⊗Hτ1,0 → C. The Hodge bundle carries a natural Hermitian metric hW P (Ωz ∧ ωτ , Ωz ∧ ωτ ) = Ωz ∧ ωτ ∧ Ωz ∧ ωτ X×T 2
known as the WeilPetersson metric, and an associated metric (Chern) conahler potential of the WeilPetersson metric on MC nection by ∇W P . The K¨ is deﬁned by K = ln −iln
Ω∧Ω
.
(6)
X
There is a similar deﬁnition on E and we take the direct sum to obtain a K¨ ahler metric on C. We endow L with the dual WeilPetersson metric and connection. The hermitian line bundle (H 4,0 , hW P ) → MC is a positive line bundle, and it follows that L is a negative line bundle. The vacua we wish to count are the classical vacua of the eﬀective supergravity Lagrangian of the string/M model, which is derived by “integrating out” the massive modes (cf. [28]). The only term relevant of the Lagrangian to our counting problem is the scalar potential [31] VG (Z) = ∇WG (Z)2 − 3W (Z)2 ,
(7)
where the connection and hermitian metric are the WeilPetersson ones. We only consider the supersymmetric vacua here, which are the special critical points Z of VG satisfying (1).
470
S. Zelditch
3 Critical Points and Hessians of Holomorphic Sections We see that type IIb ﬂux compactiﬁcations involve holomorphic sections of hermitian holomorphic line bundles over complex manifolds. Thus, counting ﬂux vacua is a problem in complex geometry. In this section, we provide a short review from [12, 13]. Let L → M denote a holomorphic line bundle over a complex manifold, and endow L with a hermitian metric h. In a local frame eL over an open set U ⊂ M , one deﬁnes the K¨ ahler potential K of h by eL (Z)2h = e−K(Z) .
(8)
We write a section s ∈ H 0 (M, L) locally as s = f eL with f ∈ O(U ). We further choose local coordinates z. In this frame and local coordinates, the covariant derivative of a section s takes the local form m m ∂f ∂ −K ∂K e −f eK f dZj ⊗ eL . (9) dZj ⊗ eL = ∇s = ∂Zj ∂Zj ∂Zj j=1 j=1 The critical point equation ∇s(Z) = 0 thus reads, ∂f ∂K −f =0. ∂Zj ∂Zj It is important to observe that although s is holomorphic, ∇s is not, and the critical point equation is only C ∞ and not holomorphic. This is ∂K , which is only smooth. Connection critical points of s due to the factor ∂Z j are the same as ordinary critical points of log s(Z)h . Thus, the critical point equation is a system of real equations and the number of critical points varies with the holomorphic section. It is not a topological invariant, as would be the number of zeros of m sections in dimension m, even on a compact complex manifold. This is one reason why counting critical points, hence vacua, is so complicated. We now consider the Hessian of a section at a critical point. The Hessian of a holomorphic section s of a general Hermitian holomorphic line bundle (L, h) → M at a critical point Z is the tensor D∇W (Z) ∈ T ∗ ⊗ T ∗ ⊗ L where D is a connection on T ∗ ⊗ L. At a critical point Z, D∇s(Z) is independent of the choice of connection on T ∗ . The Hessian D∇W (Z) at a critical point determines the complex symmetric matrix H c (which we call the “complex Hessian’). In an adapted local frame (i.e. holomorphic derivatives vanish ahler normal coordinates, it takes the form at Z0 ) and in K¨ H −f (Z0 )Θ H H c H := = , (10) H H H −f (z0 )Θ
Counting String/M Vacua
whose components are given by ∂ ∂ ∂K ∂K − − Hjq = f (Z0 ) , ∂Zj ∂Zj ∂Zq ∂Zq ∂ 2 K Hjq =−f = −f (Z0 )Θjq . ∂Z ∂ Z¯ j
Here, Θh (z0 ) =
" j,q
471
(11) (12)
q Z0
Θjq dZj ∧ dZ¯q is the curvature.
4 The Critical Point Problem We can now deﬁne the critical point equation (1) precisely. We deﬁne a supersymmetric vacuum of the ﬂux superpotential WG corresponding to the ﬂux G of (3) to be a critical point ∇W P WG (Z) = 0 of WG relative to the WeilPetersson connection on L. We obtain a local formula by writing WG (Z) = fG (Z)eZ where eZ is local frame for L → C. We choose the local frame eZ to be dual to Ωz ⊗ ωτ , and then fG (z, τ ) is given by the formula (4). The E component of ∇W P is ∂ 1 ∂τ − τ −¯ τ . The critical point equation is the system: ∂Ω ∂K X (F − τ H) ∧ { ∂zjz + ∂zj Ωz } = 0 , (13) (F − τ ¯ H) ∧ Ω = 0 , z X where K is from (6). Using the special geometry of C ( [5, 29]), one ﬁnds that the critical point equation is equivalent to the following restriction on the Hodge decomposition of H 3 (X, C) at z: ∇W P WG (z, τ ) = 0 ⇐⇒ F − τ H ∈ Hz2,1 ⊕ Hz0,3 .
(14)
Here, we recall that each complex structure z ∈ MC gives rise to a Hodge decomposition H 3 (X, C) = Hz3,0 (X) ⊕ Hz2,1 (X) ⊕ Hz1,2 (X) ⊕ Hz0,3 (X)
(15)
into forms of type (p, q). In the case of a CY 3fold, h3,0 = h0,3 = 1, h1,2 = h2,1 and b3 = 2 + 2h2,1 . Next, we specify the tadpole constraint. We deﬁne the real symmetric bilinear form on H 3 (X, C) by 3 ψ ∧ ϕ¯ . (16) Q(ψ, ϕ) = i X
The HodgeRiemann bilinear relations for a 3fold say that the form Q is deﬁnite in each Hzp,q (X) for p + q = 3 with sign alternating + − +− as one moves left to right in (15). The tadpole constraint is that
472
S. Zelditch
¯≤L. G∧G
Q[G] = i3
(17)
X
Here, L is determined by X in a complicated way (it equals χ(Z)/24 where Z is CY 4fold which is an elliptic ﬁbration over X/g, where χ(Z) is the Euler characteristic and where g is an involution of X). Although Q is an indeﬁnite symmetric bilinear form, we see that Q >> 0 on Hz2,1 (X) ⊕ Hz0,3 for any complex structure z. We now explain the sense in which we are dealing with a lattice point problem. The deﬁnition of WG makes sense for any G ∈ H 3 (X, C), so we obtain a real (but not complex) linear embedding H 3 (X, C) ⊂ H 0 (C, L). Let us denote the image by F and call it the space of complexvalued ﬂux superpotentials with dilatonaxion. The set of WG with G ∈ H 3 (X, Z ⊕ √ −1Z) is then a lattice FZ ⊂ F, which we will call the lattice of quantized (or integral) ﬂux superpotentials. Each integral ﬂux superpotential WG thus gives rise to a discrete set of of critical points Crit(WG ) ⊂ C, any of which could be the vacuum state √ the universe. Moreover, the ﬂux G can be any element of H 3 (X, Z ⊕ −1Z) satisfying the tadpole constraint (17). Thus, the set of possible vacua is the union O Crit(WG ) . (18) VacuaL = √ G∈H 3 (X,Z⊕ −1Z), 0≤Q[G]≤L
Our purpose is to count the number of vacua #VacuaL ∩ K in any given compact subset K ⊂ C. More generally, we wish to consider the sums
CN , ψ , (19) Nψ (L) = √ N ∈H 3 (X,Z⊕ −1Z):Q[N ]≤L
where
CN , ψ =
ψ(N, z, τ ) ,
(20)
(z,τ ):∇N (z,τ )=0
and where ψ is a reasonable function on the incidence relation I = {(W ; z, τ ) ∈ F × C : ∇W (z, τ ) = 0} .
(21)
We often write Z = (z, τ ) ∈ C. Points (W, Z) such that Z is a degenerate critical point of W cause problems. They belong to the discriminant variety ⊂ I of singular points of the projection π : I → F. We note that π −1 (W ) = D {(W, Z) : Z ∈ Crit(W )}. This number is constant on each component of F \D but jumps as we cross over D. where D = π(D) To count critical points in a compact subset K ⊂ C of moduli space, we would put ψ = χK (z, τ ). We often want to exclude degenerate critical points and then use test functions ψ(W, Z) which are hom*ogeneous of degree 0 in Another important example is the cosmological constant W and vanish on D ψ(W, z, τ ) = VW (z, τ ), i.e. the value of the potential at the vacuum, which is hom*ogeneous of degree 2 in W .
Counting String/M Vacua
473
5 Statement of Results We ﬁrst state an initial estimate which is regarded as “trivial’ in lattice counting problems. In pure lattice point problems it is sharp, but we doubt that it is sharp in the vacuum counting problem because of the “tilting” of the projection I → C. We denote by χQ the characteristic function of the hyperbolic shell 0 < QZ [W ] < 1 ⊂ F and by χQZ the characteristic function of the elliptic shell 0 < QZ [W ] < 1 ⊂ FZ . Proposition 1. Suppose that ψ(W, Z) = χK where K ⊂ I is an open set with smooth boundary. Then: b3 c Nψ (L) = L ψ(W, Z) det H W (Z)χQZ dW dVW P + RK (L) , C
FZ
where
then RK (L) = O L−1/2 . 1. If K is disjoint from the D, 2. If K is a general compact set (possibly intersecting the discriminant locus), then RK (L) = O L−1/2
Here, b3 = dim H3 (X, R), Qz,τ = QFz,τ and χQz,τ (W ) is the characteristic function of {Qz,τ ≤ 1} ⊂ Fz,τ , H c W (Z) is the complex Hessian of W at the critical point Z in the sense of (10). We note that the integral converges since {QZ ≤ 1} is an ellipsoid of ﬁnite volume. This is an asymptotic formula which is a good estimate on the number of vacua when L is large (recall that L is a topological invariant determined by X). The reason for assumption (1) is that number of critical points and the summand CW , ψ jump across D, so in Nψ (L) we are summing a discontinuous function. This discontinuity could cause a relatively large error term in the asymptotic counting. However, superpotentials of physical interest have nondegenerate supersymmetric critical points. Their Hessians at the critical points are ‘fermionic mass matrices’, which in physics have only nonzero eigenvalues (masses), so it is reasonable assume that suppψ is disjoint from D. Now we state the main result. Theorem 1 Suppose ψ(W, z, τ ) ∈ Cb∞ (F × C) is hom*ogeneous of degree 0 in W , with ψ(W, z, τ ) = 0 for W ∈ D. Then 1 Nψ (L) = Lb3
C
Fz,τ
ψ(W, z, τ ) det H c W (z, τ )χQz,τ (W )dW dVW P (z, τ )
+O L Here, Cb∞ denotes bounded smooth functions.
2b3 3 +1
− 2b
2 .
474
S. Zelditch
There is a simple generalization to hom*ogeneous functions of any degree such as the cosmological constant. The formula is only the starting point of a number of further versions which will be presented in Sect. 8 in which we “pushforward” the dW integral under the Hessian map, and then perform an ItzyksonZuberHarishChandra transformation on the integral. The latter version gets rid of the absolute value and seems to most useful for numerical studies. Further, one can use the special geometry of moduli space to simplify the resulting integral. Before discussing them, we pause to compare our results to the expectations in the string theory literature.
6 Comparison to the Physics Literature The reader following the developments in string theory may have encountered discussions of the “string theory landscape” (see e.g. [4, 30]). The multitude of superpotentials and vacua is a problem for the predictivity of string theory. It is possible that a unique vacuum will distinguish itself in the future, but until then all critical points are candidates for the small dimensions of the universe, and several groups of physicists are counting or enumerating them in various models (see e.g. [7, 8, 10]). The graph of the scalar potential energy may be visualized as a landscape [30] whose local minima are the possible vacua. It is common to hear that there are roughly 10500 possible vacua. This heuristic ﬁgure appears to originate in the following reasoning: assuming b3 ∼ 250, the potential energy VG (Z) is a function roughly 500 variables (including ﬂuxes G). The critical point equation for a function of m variables is a system of m equations. Naively, the number of solutions should grow like dm where d is the number of solutions of the jth equation with the other variables held ﬁxed. This would follow from B´ezout’s formula if the function was a polynomial and if we were counting complex zeros. Thus, if the “degree” of VG were a modest ﬁgure of 10 we would obtain the heuristic ﬁgure. Such an exponential growth rate of critical points in the number of variables also arises in estimates of the number of metastable states (local minima of the Hamiltonian) in the theory of spin glasses. In fact, an integral similar to that in Theorem 1 arises in the formula for the expected number of local minima of a random spin glass Hamiltonian. Both heuristic and rigorous calculations lead to an exponential growth rate of the number of local minima as the number of variables tends to inﬁnity (see e.g. [17] for a mathematical discussion and references to the literature). The mathematical similarity of the problems at least raises the question whether the number of string/M vacua should grow exponentially in the number 2b3 of variables (G, Z), i.e. in the “topological complexity” of the CalabiYau manifold X. Our results do not settle this problem, and indeed it seems to be a diﬃcult question. Here are some of the diﬃculties: First, in regard to the B´ezout estimate, the naive argument ignores the fact that the critical point equation is
Counting String/M Vacua
475
a real C ∞ equation, not a holomorphic one and so the B´ezout estimate could be quite inaccurate. Moreover, a ﬂux superpotential is not a polynomial and it is not clear what “degree” it has, as measured by its number critical points. In simple examples (see e.g. [2, 8, 10], the superpotentials do not have many critical points and it is rather the large number of ﬂuxes satisfying the tadpole constraint which produces the leading term Lb3 . This is why the ﬂux G has to be regarded as one of the variables if one wants to rescue the naive counting argument. In addition, the tadpole constraint has a complicated dimensional dependence. It induces a constraint on the inner integral in Theorem 1 to an ellipse in b3 dimensions, and the volume of such a domain shrinks at the rate 1/(b3 )!. Further, the volume of the CalabiYau moduli space is not known, and could be very small. Thus, there are a variety of competing inﬂuences on the growth rate of the number of vacua in b3 which all have a factorial dependence on the dimension. To gain a better perspective on these issues, it is important to estimate the integral giving the leading coeﬃcient and the remainder in Theorem 1. The inner integral is essentially an integral of a hom*ogeneous function of degree b3 over an ellipsoid in b3 dimensions, and is therefore very sensitive to the size of b3 . The full integral over moduli space carries the additional problem of estimating its volume. Further, one needs to estimate how large L is for a given X. Without such eﬀective bounds on L, it is not even possible to say whether any vacua exist which are consistent with known physical quantities such as the cosmological constant.
7 Sketch of Proofs The proof of Theorem 1 is in part an application of a lattice point result to the lattice of ﬂux superpotentials. In addition, it uses the formalism on the density of critical points of Gaussian random holomorphic sections in [12]. The lattice point problem is to study the distribution of radial projections of lattice points in the shell 0 ≤ Q[G] ≤ L on the surface Q[G] = 1. Radial projections arise because the critical point equation ∇WG = 0 is hom*ogeneous in G. Thus, we consider the model problem: Let Q ⊂ Rn (n ≥ 2) be a smooth, starshaped set with 0 ∈ Q◦ and whose boundary has a nondegenerate second fundamental form. Let XQ denote the norm of X ∈ Rn deﬁned by n Q= √{X ∈ R : XQ < 1} . In the following, we denote the large parameter by L to maintain consistency with Theorem 1. Theorem 2 [14] If f is hom*ogeneous of degree 0 and f ∂Q ∈ C0∞ (∂Q), then n n n 2 f (k) = L f dX + O(L 2 − n+1 ), L → ∞ . Sf (L) := √ k∈Zn ∩ LQ\{0}
Q
476
S. Zelditch
Although we have only stated it for smooth f , the method can be generalized to f ∂Q = χK where K is a smooth domain in ∂Q [33]. However, the remainder then depends on K and reﬂects the extent to which projections of lattice points concentrate on ∂K ⊂ ∂Q. The asymptotics are reminiscent of the the result of van der Corput, Hlawka, Herz and Randol on the number of lattice points in dilates of a convex set, but as of this time of writing we have not located any prior studies of the radial projection problem. Number theorists have however studied the distribution of lattice points lying exactly on spheres (Linnik, Pommerenke). We also refer the interested reader to [15] for a recent article counting lattice points in certain rational cones using methods of automorphic forms, in particular Lfunctions. We thank B. Randol for some discussions of this problem; he has informed the author that the result can also be extended to more general kinds of surfaces with degenerate second fundamental forms. Applying Theorem 2 to the string/M problem gives that 1 2 2b3 − 2b +1 b3
CW , ψ dW + O L 3 . (22) Nψ (L) = L {Q[W ]≤1}
We then write (22) as an integral over the incidence relation (21) and change the order of integration to obtain the leading coeﬃcient
CW , ψ dW {Q[W ]≤1} = ψ(W, z, τ ) det H c W (z, τ )χQz,τ dW dVW P (z, τ ) (23) C
Fz,τ
in Theorem 1. Heuristically, the integral on the left side is given by ψ(W, Z) det H c W (Z)δ(∇W (z))χQ (Z)dW dVW P (Z) . F
(24)
C
The factor  det H c W (Z) arises in the pullback of δ under ∇W (Z) for ﬁxed W , since it weights each term of (20) by  det H 1c W (Z) . We obtain the stated form of the integral in (23) by integrating ﬁrst in W and using the formula for the pullback of a δ function under a linear submersion. That formula also 1 where A(Z) = ∇Zj ∇Zk ΠZ (Z , Z )Z =Z =Z , contains another factor det A(Z) where ΠZ is the Szeg¨o kernel of FZ , i.e. the orthogonal projection onto that subspace. Using special geometry, the matrix turns out to be just I and hence the determinant is one.
8 Other Formulae for the Critical Point Density In view of the diﬃculty of estimating the leading term in Theorem 1, it is useful to have alternative expressions. We now state two of them.
Counting String/M Vacua
477
The ﬁrst method is to change variables to the Hessian H c W (Z) under the Hessian map HZ : SZ → Sym(m, C) ⊕ C,
HZ (W ) = H c W (Z) ,
(25)
where m = dim C = h2,1 +1. It turns out that Hessian map is an isomorphism to a real b3 dimensional space HZ ⊕ C, where ! 0 ej 0 iej HZ = spanR . (26) , etj F j (z) ietj −iF j (z) j=1,...,h2,1 2,1
Here, ej is thejth standard basis element of Ch
and F j (z) ∈ Sym(h2,1 , C)
j is the matrix Fik (z) whose entries deﬁne the socalled “Yukawa couplings” (see [5, 29] for the deﬁnition). We deﬁne the positive deﬁnite operator CZ : HZ → HZ by: (27) (CZ−1 HZ W, HZ W ) = QZ (W, W ) . j The entries in CZ are quadratic expressions in the Fik (see [14]).
Theorem 3 We have: −(C −1 H,H)+x2 ) 1 ∗ 2 Z e Kcrit (Z) = b3 ! det dH dx , C HZ ⊕C det H H − x I Z
=
1 det CZ
HZ ⊕C
det H ∗ H − x2 I χC (H, x)dHdx, Z
where χCZ is the characteristic function of the ellipsoid {(CZ H, H) + x2 ) ≤ 1} ⊂ HZ . Finally, we give formula of ItzyksonZuber type as in [13, Lemma 3.1], which is useful in that it has a ﬁxed domain of integration. Theorem 4 Let ΛZ = CZ ⊕ I on HZ ⊕ C and let PZ denote the orthogonal projection from Sym(m, C) onto HZ . Then: Kcrit (Z) = cm lim+ lim+ ε →0
×
Rm ε→0
>
Rm
U(m)
iξ,λ −εξ2 −ε λ2
∆(ξ) ∆(λ)  j λj  e e 9 $ # H +I det iΛZ PZ ρ(g)∗ D(ξ)ρ(g)
where: – m = h2,1 + 1, cm =
(−i)m(m−1)/2 >m 2m π 2m j=1 j!
;
– ∆(λ) = Πi<j (λi − λj ), – dg is unit mass Haar measure on U(m),
dg dξ dλ ,
478
S. Zelditch
H – D(ξ) is the Hermitian operator on Sym(m, C) ⊕ C given by " ξj + ξk m H D(ξ) (Hjk ), x = Hjk , − ξ , q=1 q x 2 – ρ is the representation of U(m) on Sym(m, C) ⊕ C given by ρ(g)(H, x) = (gHg t , x) . – HZ is a real (but not complex) subspace of Sym(m, C). The proof is similar to the one in [13], but we sketch the proof here to provide a published reference. Some care must be taken since the Gaussian integrals are over real but not complex spaces of complex symmetric matrices. Proof. We ﬁrst rewrite the integral in Theorem 3 as a Gaussian integral over HZ ⊕ C (viewed as a real vector space): Kcrit (Z) =  det H ∗ H − x2 I χ{Λ−1 H,H≤1} dHdx Z HZ ⊕C √ m det ΛZ π I(Z) , = b3 ! where I(Z) =
1 det(HH ∗ − x2 I) π det ΛZ HZ ×C × exp − Λ−1 Z (H, x), (H, x) dHdx . √ m
(28)
Here, H is a complex m × m symmetric matrix, so H ∗ = H. The inner product in the exponent is the real part of the HilbertSchmidt inner product,
A, B = ReTrAB ∗ . As in [13], we rewrite the integral as I(Z) = lim lim Iε,ε (Z) , ε →0 ε→0
where Iε,ε (Z) is the absolutely convergent integral, Iε,ε (Z) =
1 (2π)m2 π m
√
det Λ ∗
∗
det P  e−εT rΞ Ξ−ε T rP P eiΞ,P −HH Hm Hm HZ ×C × exp − Λ−1 Z (H, x), (H, x) dH dx dP dΞ .
∗
+x2 IHS
(29)
Here, Hm denotes the space of all Hermitian matrices of rank m, and
, HS is the HilbertSchmidt inner product T rAB ∗ . Formula 29 is valid,
Counting String/M Vacua
479
since as ε → 0, the dΞ integral converges to the delta function δHH ∗ −x2 I (P ). Then, as ε → 0, the dP integral evaluates the integrand at P = HH ∗ − x2 I and we retrieve the original integral I(Z). By the same manipulations as in [13], we obtain: Iε,ε (Z) =
(−i)m(m−1)/2 √ >m (2π)m ( j=1 j!)π m det ΛZ 2 2 ∆(λ)∆(ξ) det(D(λ)) eiλ,ξ e−ε(ξ +λ ) U(m)
HZ ×C
× eiD(ξ),x
2
Rm
Rm
∗ ∗
I−gHH g HS −Λ−1 (H,x),(H,x)
dξ dλ dH dx dg .
(30)
Further we observe that the dHdx integral is a Gaussian integral. Simplifying the phase as in [13] using
D(ξ), gHH ∗ g ∗ − x2 IHS = T r(D(ξ)gHg t g¯H ∗ g ∗ ) − T rD(ξ) x2 5 6 H = D(ξ)ρ(g)(H, x), ρ(g)(H, x) HS
H where D(ξ) and ρ(g) are as in the statement of the theorem, the HZ × C integral becomes 5 6 % H Iξ,g (Z) : = exp − i D(ξ)ρ(g)(H, x), ρ(g)(H, x) HS HZ ×C & −1 − ΛZ (H, x), (H, x) dH dx . (31) The only new points in the calculation are that this Gaussian integral is over the Hessian space HZ rather than over the full space of complex symmetric matrices of this rank, and that it is a real subspace a complex vector space. Hence the Gaussian integral is a real one albeit with a complex quadratic form. We denote by PZ the orthogonal projection PZ : Sym(m, C) → HZ and then we have: πm
√
1 1 1 Iξ,g (Z) = √ det ΛZ det ΛZ H + Λ−1 det[iPZ ρ(g)∗ D(ξ)ρ(g) Z ] = 9
1 # $ . H + Im det iΛZ PZ ρ(g)∗ D(ξ)ρ(g)
(32)
Substituting (32) into (30), we obtain the desired formula. We now recall that Λ = C ⊕ 1. It follows that m ∗H ∗ ξq , ΛZ PZ ρ(g) D(ξ)ρ(g) + Im = C PZ ρ(g) D(ξ)ρ(g) + Ih21 ) ⊕ (1 − Z
q=1
480
where
S. Zelditch
D(ξ)(Hjk ) =
ξj + ξk Hjk . 2
Hence, its determinant equals h21 1 − ξq det(CZ PZ ρ(g)∗ D(ξ)ρ(g) + Ih21 ) . q=1
9 Black Hole Attractors We close this survey with a discussion of a simpler problem analogous to counting ﬂux vacua that arises in the quantum gravity of black holes [16,28], namely counting solutions of the blackhole attractor equation. For a mathematical introduction to this equation, we refer the reader to [26]. The attractor equation is the same as the critical point equation for ﬂux superpotentials except that C = M and G ∈ H 3 (X, R). Physically, Nψ (S) counts the socalled dualityinequivalent, regular, spherically symmetric BPS black holes with entropy S ≤ S∗ . The charge of a black hole is an element Q = N α Σα ∈ H 3 (X, Z). The central charge Z = Q, Ω plays the role of the superpotential. There are two main diﬀerences to the vacuum counting problems for ﬂux superpotentials. First, the reality of the ﬂux G in the blackhole attractor equation ∇WG (z) = 0 forces G ∈ Hz3,0 ⊕ Hz0,3 rather than G ∈ Hz2,1 ⊕ Hz0,3 as in the ﬂux vacua equation. The space H 3,0 ⊕H 0,3 is only 2dimensional and that drastically simpliﬁes the problem. Second, by a wellknown computation due to Strominger, the Hessian D∇G(z) of Z2 at a critical point is always a scalar multiple xΘ of the curvature form of the line bundle, which is the WeilPetersson (1, 1) form. We now state the analogue of Theorem 3 in the black hole attractor case (see also [8]). The new feature is that the image of Hessian map from the space Sz of WG with a critical point at z is the onedimensional space of Hessians of the form 0 −xΘ , (33) −xΘ 0 and hence the pushforward under the Hessian map truly simpliﬁes the integral in Theorem 3. The formula for the blackhole density becomes crit Kγ,∇ (z) = x2b3 χQz (x)dx . C
We note that the diﬃcult absolute value in Theorem 3 simpliﬁes to a perfect square in the black hole density formula and can therefore be evaluated as a
Counting String/M Vacua
481
Gaussian integral. Additionally, the onedimensionality of the space of Hessians has removed the complexity of the b3 dimensional integral in the ﬂux vacuum setting. We can further simplify the integral by removing Qz , which is a scalar multiple of the Euclidean x2 . The scalar multiple involves the orthogonal projection ΠSz (z, w) onto
the space of Sz for the inner product Qz . If we change variables x → ΠSz (z, z), we get crit Kγ,∇ (z) = ΠSz (z, z) x2b3 e−x,x dx. C
In K¨ ahler normal coordinates, use of special geometry shows that Πz (z, z) = 1. A simple calculation shows: Proposition 2. The density of extremal black holes is given by: crit (z) = Kγ,∇
1 dVW P =⇒ b3
Nψ (L) ∼ Lb3 V olW P (M) .
The analogy between the black hole density and ﬂux vacuum critical point density should be taken with some caution since the simplifying features are likely to have oversimpliﬁed the problem. We therefore mention another modiﬁed ﬂux vacuum problem in which the oﬀdiagonal entries xΘ of the matrix vanish, so that the Hessian matrix is purely holomorphic and Hessian det H ∗ H − x2 Θ = det H ∗ H again becomes a perfect square which can be evaluated by the Wick method. Namely, if one uses a ﬂat meromorphic connection ∇ rather than the WeilPetersson connection, the curvature vanishes away from the polar divisor. The WeilPetersson connection arises naturally in string/M theory [6], but one may view a meromorphic connection as an approximation in which the “Planck mass” is inﬁnitely large. In any case, it would be interesting to evaluate the density of critical points relative to meromorphic connections since they are more calculable and should have the same complexity as those for WeilPetersson connections.
References 1. N. ArkaniHamed, S. Dimopoulos and Shamit Kachru, Predictive Landscapes and New Physics at a TeV (hepth/0501082). 2. S. Ashok and M. Douglas, Counting Flux Vacua, J. High Energy Phys. 0401 (2004) 060 hepth/0307049. 3. P. Bleher, B. Shiﬀman, and S. Zelditch, Universality and scaling of correlations between zeros on complex manifolds. Invent. Math. 142 (2000), no. 2, 351–395. 4. R. Bousso and J. Polchinski, The String Theory Landscape, Scientiﬁc American, September 2004 issue. 5. P. Candelas, X. C. de la Ossa, Moduli space of CalabiYau manifolds. Nuclear Phys. B 355 (1991), no. 2, 455–481.
482
S. Zelditch
6. P. Candelas, G. Horowitz, A. Strominger, and E. Witten, Vacuum conﬁgurations for superstrings. Nuclear Phys. B 258 (1985), no. 1, 46–74. 7. J. P. Conlon and F. Quevedo, On the explict construction and statistics of CalabiYau vacua, Jour. High Energy Physics 0410:039 (2004) (ePrint Archive: hepth/0409215). 8. F. Denef and M. R. Douglas, Distributions of ﬂux vacua, J. High Energy Phys. JHEP05(2004)072 (arxiv preprint hepth/0404116.) 9. F. Denef and M. R. Douglas, Enumerating ﬂux vacua with enhanced symmetries. J. High Energy Phys. 2005, no. 2, 037, ePrint Archive: hepth/0411183. 10. O. DeWolfe, A. Giryavets, S. Kachru and W. Taylor, Enumerating ﬂux vacua with enhanced symmetries. J. High Energy Phys. 2005, no. 2, 037, (hepth/0411061). 11. M. R. Douglas, The statistics of string/M theory vacua, J. High Energy Phys. 2003, no. 5, 046, 61 pp. (hepth/0303194). 12. M. R. Douglas, B. Shiﬀman and S. Zelditch, Critical points and supersymmetric vacua I, Communications in Mathematical Physics 252, Numbers 13 (2004), pp. 325  358 (arxiv preprint arxiv.org/math.CV/0402326). 13. M. R. Douglas, B. Shiﬀman and S. Zelditch, Critical Points and supersymmetric vacua II: asymptotics and extremal metrics, math.CV/0406089, to appear in Jour. Diﬀ. Geom. 14. M. R. Douglas, B. Shiﬀman and S. Zelditch, Critical Points and supersymmetric vacua III: string/M models, to appear in Comm. Math. Phys. 15. W. Duke and O. Imamoglu, Lattice points in cones and Dirichlet series, IMRN 53 (2004) 16. S. Ferrara, G. W. Gibbons, and R. Kallosh, Black holes and critical points in moduli space. Nuclear Phys. B 500 (1997), no. 13, 75–93. 17. Y. V. Fyodorov, Complexity of random energy landscapes, glass transition, and absolute value of the spectral determinant of random matrices. Phys. Rev. Lett. 92 (2004), no. 24, 240601. 18. S.B. Giddings, S. Kachru, and J. Polchinski, Hierarchies from ﬂuxes in string compactiﬁcations. Phys. Rev. D (3) 66 (2002), no. 10, 106006. 19. A. Giryavets, S. Kachru and P. K. Tripathy, On the taxonomy of ﬂux vacua, JHEP 0408:002,2004 ePrint Archive: hepth/0404243. 20. A. Giryavets, S. Kachru, P. K. Tripathy, and S. P. Trivedi, Flux Compactiﬁcations on CalabiYau threefolds, JHEP 0404:003,2004 ePrint Archive: hepth/0312104. 21. M. Gross, CalabiYau manifolds and mirror symmetry. CalabiYau manifolds and related geometries, in [22], 69–159. 22. M. Gross, D. Huybrechts, and D. Joyce, CalabiYau manifolds and related geometries. Lectures from the Summer School held in Nordfjordeid, June 2001. Universitext. SpringerVerlag, Berlin, 2003. 23. S. Gukov, C. Vafa, and E. Witten, CFT’s from CalabiYau fourfolds. Nuclear Phys. B 584 (2000), no. 1–2, 69–108. 24. S. Kachru, R. Kallosh, A. Linde, and S. P. Trivedi, de Sitter Vacua in String Theory, Phys.Rev. D68 (2003) 046005 (hepth/0301240). 25. S. Kachru, M. Schulz, and S. Trivedi, Moduli Stabilization from Fluxes in a Simple IIB Orientifold, JHEP 0310 (2003) 007 (hepth/0201028). 26. S.D. Miller and G. Moore, LandauSiegel zeros and black hole entropy, ePrint Archive: hepth/9903267.
Counting String/M Vacua
483
27. B. Shiﬀman and S. Zelditch, Distribution of zeros of random and quantum chaotic sections of positive line bundles. Comm. Math. Phys. 200 (1999), no. 3, 661–683. 28. A. Strominger, KaluzaKlein compactiﬁcations, supersymmetry and CalabiYau manifolds, p. 10911115 of Quantum ﬁelds and strings: a course for mathematicians. Vol. 1, 2. Material from the Special Year on Quantum Field Theory held at the Institute for Advanced Study, Princeton, NJ, 1996–1997. Edited by Pierre Deligne, Pavel Etingof, Daniel S. Freed, Lisa C. Jeﬀrey, David Kazhdan, John W. Morgan, David R. Morrison and Edward Witten. American Mathematical Society, Providence, RI; Institute for Advanced Study (IAS), Princeton, NJ, 1999. 29. A. Strominger, Special geometry. Comm. Math. Phys. 133 (1990), no. 1, 163– 180. 30. L. Susskind, The anthropic landscape of string theory, ePrint Archive: hepth/0302219. 31. J. Wess and J. Bagger, Supersymmetry and Supergravity, Second Edition, Princeton Series in Physics, Princeton U. Press (1992). 32. S. Zelditch, Random Complex Geometry and Vacua, or: How to count universes in string/M theory (preprint, 2005). 33. S. Zelditch, Angular distribution of lattice points (in preparation).
Lecture Notes in Physics For information about earlier volumes please contact your bookseller or Springer LNP Online archive: springerlink.com Vol.641: A. Lalazissis, P. Ring, D. Vretenar (Eds.), Extended Density Functionals in Nuclear Structure Physics Vol.642: W. Hergert, A. Ernst, M. Däne (Eds.), Computational Materials Science Vol.643: F. Strocchi, Symmetry Breaking Vol.644: B. Grammaticos, Y. KosmannSchwarzbach, T. Tamizhmani (Eds.) Discrete Integrable Systems Vol.645: U. Schollwöck, J. Richter, D. J. J. Farnell, R. F. Bishop (Eds.), Quantum Magnetism Vol.646: N. Bretón, J. L. CervantesCota, M. Salgado (Eds.), The Early Universe and Observational Cosmology Vol.647: D. Blaschke, M. A. Ivanov, T. Mannel (Eds.), Heavy Quark Physics Vol.648: S. G. Karshenboim, E. Peik (Eds.), Astrophysics, Clocks and Fundamental Constants Vol.649: M. Paris, J. Rehacek (Eds.), Quantum State Estimation Vol.650: E. BenNaim, H. Frauenfelder, Z. Toroczkai (Eds.), Complex Networks Vol.651: J. S. AlKhalili, E. Roeckl (Eds.), The Euroschool Lectures of Physics with Exotic Beams, Vol.I Vol.652: J. Arias, M. Lozano (Eds.), Exotic Nuclear Physics Vol.653: E. Papantonoupoulos (Ed.), The Physics of the Early Universe Vol.654: G. Cassinelli, A. Levrero, E. de Vito, P. J. Lahti (Eds.), Theory and Appplication to the Galileo Group Vol.655: M. Shillor, M. Sofonea, J. J. Telega, Models and Analysis of Quasistatic Contact Vol.656: K. Scherer, H. Fichtner, B. Heber, U. Mall (Eds.), Space Weather Vol.657: J. Gemmer, M. Michel, G. Mahler (Eds.), Quantum Thermodynamics Vol.658: K. Busch, A. Powell, C. Röthig, G. Schön, J. Weissmüller (Eds.), Functional Nanostructures Vol.659: E. Bick, F. D. Steffen (Eds.), Topology and Geometry in Physics Vol.660: A. N. Gorban, I. V. Karlin, Invariant Manifolds for Physical and Chemical Kinetics Vol.661: N. Akhmediev, A. Ankiewicz (Eds.) Dissipative Solitons Vol.662: U. CarowWatamura, Y. Maeda, S. Watamura (Eds.), Quantum Field Theory and Noncommutative Geometry Vol.663: A. Kalloniatis, D. Leinweber, A. Williams (Eds.), Lattice Hadron Physics Vol.664: R. Wielebinski, R. Beck (Eds.), Cosmic Magnetic Fields
Vol.665: V. Martinez (Ed.), Data Analysis in Cosmology Vol.666: D. Britz, Digital Simulation in Electrochemistry Vol.667: W. D. Heiss (Ed.), Quantum Dots: a Doorway to Nanoscale Physics Vol.668: H. Ocampo, S. Paycha, A. Vargas (Eds.), Geometric and Topological Methods for Quantum Field Theory Vol.669: G. AmelinoCamelia, J. KowalskiGlikman (Eds.), Planck Scale Effects in Astrophysics and Cosmology Vol.670: A. Dinklage, G. Marx, T. Klinger, L. Schweikhard (Eds.), Plasma Physics Vol.671: J.R. Chazottes, B. Fernandez (Eds.), Dynamics of Coupled Map Lattices and of Related Spatially Extended Systems Vol.672: R. Kh. Zeytounian, Topics in Hyposonic Flow Theory Vol.673: C. Bona, C. PalenzulaLuque, Elements of Numerical Relativity Vol.674: A. G. Hunt, Percolation Theory for Flow in Porous Media Vol.675: M. Kröger, Models for Polymeric and Anisotropic Liquids Vol.676: I. Galanakis, P. H. Dederichs (Eds.), Halfmetallic Alloys Vol.678: M. Donath, W. Nolting (Eds.), LocalMoment Ferromagnets Vol.679: A. Das, B. K. Chakrabarti (Eds.), Quantum Annealing and Related Optimization Methods Vol.680: G. Cuniberti, G. fa*gas, K. Richter (Eds.), Introducing Molecular Electronics Vol.681: A. Llor, Statistical Hydrodynamic Models for Developed Mixing Instability Flows Vol.682: J. Souchay (Ed.), Dynamics of Extended Celestial Bodies and Rings Vol.683: R. Dvorak, F. Freistetter, J. Kurths (Eds.), Chaos and Stability in Planetary Systems Vol.685: C. Klein, O. Richter, Ernst Equation and Riemann Surfaces Vol.686: A. D. Yaghjian, Relativistic Dynamics of a Charged Sphere Vol.687: J. W. LaBelle, R. A. Treumann (Eds.), Geospace Electromagnetic Waves and Radiation Vol.688: M. C. Miguel, J. M. Rubi (Eds.), Jamming, Yielding, and Irreversible Deformation in Condensed Matter Vol.689: W. Pötz, J. Fabian, U. Hohenester (Eds.), Quantum Coherence Vol.690: J. Asch, A. Joye (Eds.), Mathematical Physics of Quantum Mechanics