Microscopic Chaos, Fractals and Transport in Nonequilibrium Statistical Mechanics (Advanced Series in Nonlinear Dynamics 24)

Microscopic Chaos, Fractals and Transport in Nonequilibrium Statistical Mechanics ADVANCED SERIES IN NONLINEAR DYNA...

Author: Rainer Klages

42 downloads 725 Views 12MB Size Report

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

Report copyright / DMCA form

DOWNLOAD PDF

Microscopic Chaos,

Fractals and Transport in Nonequilibrium

Statistical Mechanics

ADVANCED SERIES IN NONLINEAR DYNAMICS* Editor-in-Chief: R. S. MacKay (Univ. Warwick) Published Vol. 6

Renormalization in Area-Preserving Maps R. S. MacKay

Vol. 7

Structure & Dynamics of Nonlinear Waves in Fluids eds. A. Mielke & K. Kirchgässner

Vol. 8

New Trends for Hamiltonian Systems & Celestial Mechanics eds. J. Llibre & E. Lacomba

Vol. 9

Transport, Chaos and Plasma Physics 2 S. Benkadda, F. Doveil & Y. Elskens

Vol. 10 Renormalization and Geometry in One-Dimensional and Complex Dynamics Y.-P. Jiang Vol. 11 Rayleigh–Bénard Convection A. V. Getling Vol. 12 Localization and Solitary Waves in Solid Mechanics A. R. Champneys, G. W. Hunt & J. M. T. Thompson Vol. 13 Time Reversibility, Computer Simulation, and Chaos W. G. Hoover Vol. 14 Topics in Nonlinear Time Series Analysis – With Implications for EEG Analysis A. Galka Vol. 15 Methods in Equivariant Bifurcations and Dynamical Systems P. Chossat & R. Lauterbach Vol. 16 Positive Transfer Operators and Decay of Correlations V. Baladi Vol. 17 Smooth Dynamical Systems M. C. Irwin Vol. 18 Symplectic Twist Maps C. Gole Vol. 19 Integrability and Nonintegrability of Dynamical Systems A. Goriely Vol. 20 The Mathematical Theory of Permanent Progressive Water-Waves H. Okamoto & M. Shoji Vol. 21 Spatio-Temporal Chaos & Vacuum Fluctuations of Quantized Fields C. Beck Vol. 22 Energy Localisation and Transfer eds. T. Dauxois, A. Litvak-Hinenzon, R. MacKay & A. Spanoudaki Vol. 23 Geometrical Theory of Dynamical Systems and Fluid Flows T. Kambe

*For the complete list of titles in this series, please write to the Publisher.

Lakshmi - Microscopic Chaos.pmd

2

7/12/2007, 12:09 PM

Microscopic Chaos,

Fractals and Transport in Nonequilibrium

Statistical Mechanics

Rainer Klages Queen Mary, University of London, UK

N E W JERSEY

-

vp World Scientific LONDON

*

SINGAPORE

*

BElJlNG

SHANGHAI

*

HONG KONG

*

TAIPEI

*

CHENNAI

Published by World Scientific Publishing Co. Pte. Ltd. 5 Toh Tuck Link, Singapore 596224 USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library.

MICROSCOPIC CHAOS, FRACTALS AND TRANSPORT IN NONEQUILIBRIUM STATISTICAL MECHANICS Copyright © 2007 by World Scientific Publishing Co. Pte. Ltd. All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.

ISBN-13 978-981-256-507-5 ISBN-10 981-256-507-8

Printed in Singapore.

Lakshmi - Microscopic Chaos.pmd

1

7/12/2007, 12:09 PM

May 8, 2007

4:1

World Scientific Book - 9in x 6in

(1) (2) (3) (4)

Logic is the mathematics of truth and falseness. Mathematics is the theory of structures. Theory is the art of the true and the false. Art is the perception of figures through the creation of figures.

The background image shows the woodcut In the Hollow of a Wave off the Coast at Kanagawa by Katsushika Hokusai (10x15 inches, from the series Thirty-Six Views of Mount Fuji; the original is at the Hakone Museum in Japan). The citation is from C.F. von Weizsaecker, Zeit und Wissen (DTV, Munich, 1995; translated from German to English)

v

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

vi

F¨ ur meine Eltern

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Preface

The field of research presented in this book is born from a blending of ideas and techniques from statistical physics, dynamical systems theory, stochastic processes and computational physics. This combination suitably amends traditional statistical mechanics, which is a remarkably expedient tool for coping effectively with the presence of many degrees of freedom by applying probabilistic concepts. For a long time statistical mechanics focussed on a rather restricted class of systems in relatively simple situations such as fluids and crystalline solids at or near equilibrium. Today it is realized that systems involving only a few variables also exhibit complex behavior in the form of bifurcations, chaotic dynamics and fractal geometry characterized by sensitive dependence on initial conditions and parameters, and by random-looking evolution in time and space reminiscent of manybody systems. Such systems can be viewed as a “laboratory” in which questions that for a long time remained unsolved in statistical mechanics, such as the origin of irreversible entropy production and linear response, can be raised on a new basis and brought to their solution. Since chaos appears to be ubiquitous at the level of the microscopic dynamics of single particles it should determine to a large extent macroscopiclevel behavior, from the shape of probability distributions and values of transport coefficients in simple models to diffusion in nanopores, transport on vibratory conveyors and the active Brownian motion of biological entities. That way, the very origin of irreversibility and transport can be intimately related to the intrinsic complexity due to nonlinear interactions between the individual constituents forming a macroscopic system. This poses the challenge to modern statistical mechanics of explaining the macroscopic properties of matter starting from microscopic deterministic chaos in the equations of motion of many-particle systems. vii

klages˙book

May 8, 2007

4:1

viii

World Scientific Book - 9in x 6in

klages˙book

Preface

This book highlights two very recent approaches towards a microscopic theory of macroscopic transport: One focuses on Hamiltonian dynamical systems under nonequilibrium boundary conditions, the other proposes a non-Hamiltonian approach to nonequilibrium situations created by, for example, electric fields and temperature or velocity gradients. A surprising result related to the former approach is that, in lowdimensional periodic structures, transport coefficients can be fractal functions of control parameters. These fractal transport coefficients are the first central theme of our book. We exemplify this phenomenon by deterministic diffusion in a chaotic map, which generalizes the well-known problem of a random walk on the line by inclusion of memory effects. We then outline an arsenal of analytical and numerical methods for computing deterministic transport coefficients such as electrical conductivities and chemical reaction rates. These methods are applied to a hierarchy of nonlinear dynamical systems which successively become more complex, starting from abstract one-dimensional maps up to particle billiards that are directly accessible to experiments. In all cases the resulting transport coefficients turn out to be either fractal or to be at least profoundly irregular. We furthermore provide physical explanations for these fractalities. The second central theme is a critical assessment of a non-Hamiltonian approach to nonequilibrium transport. Here we deal with situations where nonequilibrium constraints pump energy into a system, hence there must be some thermal reservoir that prevents a system from heating up. Very practical modelings of thermal reservoirs, which are widely used in molecular dynamics computer simulations, are Gaussian and Nos´e-Hoover thermostats. Surprisingly, these approaches yield simple relations between fundamental quantities of nonequilibrium statistical mechanics and dynamical systems theory. This is due to the fact that these thermostats are deterministic and time-reversible, thus carrying over respective properties of Newton’s equations of motion to nonequilibrium situations. This is in contrast to stochastic thermostats, which are recovered from them in case of memory loss at subsystem–reservoir couplings. Our goal is to critically assesses the universality of these results. As a vehicle of demonstration we employ the driven periodic Lorentz gas, a toy model for the classical dynamics of an electron in a metal under application of an electric field. Applying different types of thermal reservoirs to this system, we compare the resulting nonequilibrium steady states with each other. Along the same lines we discuss an interacting many-particle system under shear and heat. We then describe an unexpected relationship between deterministic thermostats and

May 8, 2007

4:1

Preface

World Scientific Book - 9in x 6in

klages˙book

ix

active Brownian particles modeling the motility of biological entities. Related to these two main themes the book features a third part in which we outline very recent developments in this field. There is thus a chapter with mini-reviews on fluctuation relations, Lyapunov modes, Fourier’s law and pseudochaotic transport connecting this work directly with highly active research areas that we feel are particularly interesting. We argue that the theory presented in this book is particularly useful for the description of a novel class of systems, which is right in-between high- and low-dimensional, interacting and non-interacting many-particle systems. Such intermediate statistical dynamics may display both ordinary thermodynamic behavior as well as nontrivial nonlinear properties. Crosslinks to experiments are discussed throughout the whole work. Most of these experiments are on systems defined on meso to nanoscales. We thus presume that the theory outlined here should have further applications in the currently emerging field of the nanosciences. Understanding the basic physical principles of this theory may indeed help to “design” systems exhibiting specifically desired non-trivial properties. This point will be worth future exploration which, however, goes beyond the scope of this work. A detailed outline of the contents of this book is presented in the introductory Chapter 1. Section 1.3 contains a guideline of how to read this book, some remarks on the style in which it is written and advice on required background knowledge. With this recipe, we hope that our book will provide a useful introduction for newcomers to this field as well as a reference and source of inspiration for established researchers. This book is a profoundly amended and updated version of the author’s habilitation thesis [Kla04a] summarizing about ten years of research. Such work would not have been possible without inspiring and fruitful collaboration with a large number of colleagues. First of all, I would like to express my sincere appreciation and gratitude to Profs. S.Hess, J.R.Dorfman, T.T´el, P.Gaspard, and G.Nicolis, who allowed me to learn from them during my formation as a scientist. I owe them very much for their guidance. I furthermore wish to thank Prof. P.Fulde for supporting profoundly my research activities at the Max Planck Institute for the Physics of Complex Systems (MPIPKS) in Dresden. Sincere thanks go to Drs. J.Groeneveld, T.Harayama, P.Dieterich, A.Chechkin as well as to Profs. C.Dellago and W.G.Hoover with whom I had, or have, the great pleasure of collaborating on topics presented in this book. I also thank my former Ph.D. students Drs. K.Rateitschak and N.Korabel as well as my former coworkers Drs. L.M´ aty´ as, I.F.Barna and

May 8, 2007

4:1

x

World Scientific Book - 9in x 6in

klages˙book

Preface

Dipl.-Math. T.Klauss for joint research with me on chaotic transport. It was furthermore a great pleasure to organize two international conferences on the field of research partly covered by this book, pursued together with Profs. H.van Beijeren, P.Gaspard, J.R.Dorfman, I.M.Sokolov and G.Radons. Thanks go to all of them for their enthusiasm, help and advice. Profs. P.Cvitanovic, W.Ebeling, C.Grebogi, D.J.Evans, P.H¨ anggi, G.Keller, H.Kantz, I.Melbourne, A.Pikovsky, H.A.Posch, T.Prellberg, E.Sch¨ oll as well as Drs. N.Baba, C.Dettmann, K.Gelfert, W.Just, D.Panja, A.Riegert, F.Schm¨ user and A.Sch¨ uring helped with many valuable hints and discussions during my research. Special thanks go to Dr. P.Howard for his time-consuming proof-reading of the whole work. Last but not least I would like to thank the editor of the Advanced Series in Nonlinear Dynamics, Prof. R.S.MacKay, for immediately accepting this book, and Ms. L.Narayan, the editor at World Scientific, for her patience and support regarding the completion of this work. Most of this book was written while I was a Distinguished Postdoctoral Fellow at the MPIPKS Dresden. Financial support of my previous research by many grants from German research organizations, notably the DAAD and the DFG, is gratefully acknowledged. I finally wish to express my gratitude to my colleagues at Queen Mary, University of London for a warm welcome as a lecturer, to the British EPSRC for a research grant and generally to the United Kingdom for offering a long-term scientific perspective to me. London, Winter 2006/2007

Rainer Klages

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

Contents

Preface 1.

∗

vii

Introduction and outline

1.1 1.2 1.3

1

Hamiltonian dynamical systems approach to rium statistical mechanics . . . . . . . . . . . Thermostated dynamical systems approach to rium statistical mechanics . . . . . . . . . . . The red thread through this book . . . . . . .

nonequilib. . . . . . . nonequilib. . . . . . . . . . . . . .

Part 1: Fractal transport coefficients 2.

∗

Deterministic diffusion

2.1 2.2 2.3

3.3 3.4 3.5

7 11

15 17

A simple model for deterministic diffusion . . . . . . . . . A parameter-dependent fractal diffusion coefficient . . . . Summary . . . . . . . . . . . . . . . . . . . . . . . . . . .

3. Deterministic drift-diffusion 3.1 3.2

2

17 22 28 29

Drift-diffusion model: mathematical definition . . . . . . + Calculating deterministic drift and diffusion coefficients 3.2.1 Twisted eigenstate method . . . . . . . . . . . . . 3.2.2 Transition matrix methods . . . . . . . . . . . . . 3.2.3 Numerical comparison of the different methods . . The phase diagram . . . . . . . . . . . . . . . . . . . . . . Simple maps as deterministic ratchets . . . . . . . . . . . ∗ Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

29 32 33 37 39 40 49 54

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

xii

Contents

4. Deterministic reaction-diffusion 4.1

4.2

4.3

4.4

55

A reactive-diffusive multibaker map . . . . . . . . . . 4.1.1 Deterministic models of reaction-diffusion . . . 4.1.2 The Frobenius-Perron operator . . . . . . . . . Diffusive dynamics . . . . . . . . . . . . . . . . . . . . 4.2.1 + Diffusive modes of the dyadic multibaker . . 4.2.2 The parameter-dependent diffusion coefficient Reactive dynamics . . . . . . . . . . . . . . . . . . . . 4.3.1 + Reactive modes of the dyadic multibaker . . 4.3.2 The parameter-dependent reaction rate . . . . ∗ Summary . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

5. Deterministic diffusion and random perturbations 5.1 5.2 5.3

83

Disordered dynamical systems . . . . . . . . . . . . . . . . Noisy dynamical systems . . . . . . . . . . . . . . . . . . ∗ Summary . . . . . . . . . . . . . . . . . . . . . . . . . . .

6. From normal to anomalous diffusion 6.1 6.2 6.3

Deterministic diffusion and bifurcations . . . . . . . . . . 99 Anomalous diffusion in intermittent maps . . . . . . . . . 107 ∗ Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 121

Correlated random walks in maps . . . . . . . . . . . . . . 121 Correlated random walks in billiards . . . . . . . . . . . . 128 ∗ Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

8. Designing billiards with irregular transport coefficients 8.1 8.2 8.3 8.4

Diffusion in the flower-shaped billiard . + Random and correlated random walks Diffusion in porous solids . . . . . . . . ∗ Summary . . . . . . . . . . . . . . . . .

. . . .

. . . .

. . . .

. . . .

137 . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

9. Deterministic diffusion of granular particles 9.1 9.2 9.3 9.4

83 89 98 99

7. From diffusive maps to Hamiltonian particle billiards 7.1 7.2 7.3

55 56 60 62 62 64 70 70 75 81

Resonances and diffusion in the bouncing ball + Diffusion by correlated random walks . . . . Vibratory conveyors . . . . . . . . . . . . . . ∗ Summary . . . . . . . . . . . . . . . . . . . .

137 141 148 150 153

billiard . . . . . . . . . . . . . . .

. . . .

. . . .

153 157 160 161

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

xiii

Contents

Part 2: Thermostated dynamical systems

163

10. Motivation: coupling a system to a thermal reservoir 10.1 10.2 10.3 10.4 10.5 11.

∗

165

∗

Why thermostats? . . . . . . . . . . . . . . . . . . . . . . Modeling thermal reservoirs: the Langevin equation . . . Equilibrium velocity distributions for thermostated systems Applying thermostats: the periodic Lorentz gas . . . . . . ∗ Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . ∗

The Gaussian thermostat

11.1 11.2

11.3

185

Construction of the Gaussian thermostat . . . . . . . . . Chaos and transport in Gaussian thermostated systems 11.2.1 Phase space contraction and entropy production 11.2.2 Lyapunov exponents and transport coefficients . 11.2.3 Nonequilibrium fractal attractors . . . . . . . . 11.2.4 Electrical conductivity . . . . . . . . . . . . . . Summary . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . .

12. The Nos´e-Hoover thermostat 12.1 12.2

12.3

12.4

12.5

The dissipative Liouville equation . . . . . . . . . Construction of the Nos´e-Hoover thermostat . . . 12.2.1 Heuristic derivation . . . . . . . . . . . . 12.2.2 Physics of this thermostat . . . . . . . . Properties of the Nos´e-Hoover thermostat . . . . 12.3.1 Chaos and transport . . . . . . . . . . . 12.3.2 + Generalized Hamiltonian formalism . . 12.3.3 Fractals and transport . . . . . . . . . . + Subtleties of Nos´e-Hoover dynamics . . . . . . . 12.4.1 Necessary conditions and generalizations 12.4.2 Thermal reservoirs in nonequilibrium . . ∗ Summary . . . . . . . . . . . . . . . . . . . . . .

185 189 189 190 193 198 202 205

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

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

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

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

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

13. Universalities in Gaussian and Nos´e-Hoover dynamics? 13.1 13.2 13.3 13.4 13.5

165 167 173 179 183

Non-Hamiltonian nonequilibrium steady states . . . . . Phase space contraction and entropy production . . . . Transport coefficients and dynamical systems quantities Fractal attractors for nonequilibrium steady states . . . Nonlinear response in the driven periodic Lorentz gas .

205 208 208 210 213 213 215 218 222 222 226 227 231

. . . . .

231 235 240 247 251

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

xiv

Contents

13.6

∗

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 253

14. Gaussian and Nos´e-Hoover thermostats revisited 14.1 14.2 14.3 14.4

Non-ideal Gaussian thermostat . . Non-ideal Nos´e-Hoover thermostat + Further alternative thermostats . ∗ Summary . . . . . . . . . . . . . .

. . . .

. . . .

. . . .

. . . .

257 . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

257 261 264 266

15. Stochastic and deterministic boundary thermostats 15.1 15.2 15.3 15.4

15.5

15.6

269

Stochastic boundary thermostats . . . . . . . . . . . . . . Deterministic boundary thermostats . . . . . . . . . . . . + Boundary thermostats from first principles . . . . . . . . Deterministic boundary thermostats for the driven periodic Lorentz gas . . . . . . . . . . . . . . . . . . . . . . . 15.4.1 Phase space contraction and entropy production . 15.4.2 Attractors, bifurcations and conductivity . . . . . 15.4.3 Lyapunov exponents . . . . . . . . . . . . . . . . . Hard disk fluid under shear and heat flow . . . . . . . . . 15.5.1 Homogeneously and inhomogeneously driven shear and heat flows . . . . . . . . . . . . . . . . . . . . 15.5.2 Shear and heat flows thermostated by deterministic scattering . . . . . . . . . . . . . . . . . . . . . ∗ Summary . . . . . . . . . . . . . . . . . . . . . . . . . . .

270 271 273 279 280 283 286 287 288 291 300

16. Active Brownian particles and Nos´e-Hoover dynamics 16.1 16.2 16.3 16.4

Brownian motion of migrating cells? . . . . . . . . . . . . + Moving biological entities as active Brownian particles . + Bimodal velocity distributions and Nos´e-Hoover dynamics ∗ Summary . . . . . . . . . . . . . . . . . . . . . . . . . . .

Part 3: Outlook and conclusions 17. Further topics in chaotic transport theory 17.1

303

Fluctuation relations . . . . . . . . . . . . . . . . . . 17.1.1 Entropy fluctuation in nonequilibrium steady 17.1.2 The Gallavotti-Cohen fluctuation theorem . 17.1.3 The Evans-Searles fluctuation theorem . . . 17.1.4 Jarzynski work relation and Crooks relation

304 306 308 314

317 319 . . . 320 states 320 . . . 321 . . . 327 . . . 328

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

xv

Contents

17.2 17.3

17.4

17.5 18.

∗

Lyapunov modes . . . . . . . . . . . . . . . . . . . . . Fourier’s law . . . . . . . . . . . . . . . . . . . . . . . 17.3.1 The basic problem . . . . . . . . . . . . . . . . 17.3.2 Heat conduction in anharmonic chaotic chains 17.3.3 Heat conduction in chaotic particle billiards . Pseudochaotic diffusion . . . . . . . . . . . . . . . . . 17.4.1 Microscopic chaos and diffusion? . . . . . . . . 17.4.2 Polygonal billiard channels . . . . . . . . . . . ∗ Summary . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . .

. . . . . . . . .

Conclusions

Microscopic chaos and nonequilibrium statistical mechanics: the big picture . . . . . . . . . . . . . . . . . . . . . . 18.2 Assessment of the main results . . . . . . . . . . . . . . . 18.2.1 Existence of fractal transport coefficients . . . . . 18.2.2 Universalities in thermostated dynamical systems? 18.3 Important open questions . . . . . . . . . . . . . . . . . . 18.3.1 Fractal transport coefficients . . . . . . . . . . . . 18.3.2 Thermostated dynamical systems . . . . . . . . . Note added in proof . . . . . . . . . . . . . . . . . . . . . . . . .

331 337 338 340 344 347 348 352 364 367

18.1

367 371 371 374 376 377 379 380

Bibliography

381

Index

435

This page intentionally left blank

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Chapter 1 ∗

Introduction and outline

Statistical mechanics endeavours to understand the origin of macroscopic properties of matter starting from the microscopic equations of motion of single atoms or molecules. This program traces back to the founders of statistical mechanics, Boltzmann, Maxwell and Gibbs [Bol64; Max65a; Max65b; Gib60]. For many-particle systems in thermal equilibrium it was pursued with remarkable success [Tol38; Ehr59; Rei65; Hua87; Tod92]. However, in nonequilibrium situations, that is for systems under constraints such as external fields or by imposing temperature or velocity gradients, statistical mechanical theories appear to be rather incomplete: In contrast to the equilibrium case there is no generally accepted definition of a nonequilibrium entropy, and there is no general agreement on nonequilibrium ensembles that might replace the equilibrium ones [Pen79; Eva90b; Ger99; Rue99b; Gal99]. Fresh input concerning these fundamental problems came from the side of dynamical systems theory, in particular by works of mathematicians like Sinai, Ruelle, Bowen and others over the past decades [Sin91; Sin00; Rue78; Bow75]. Indeed, SRB measures1 appear to be good candidates for taking over the role of the Gibbs ensemble in nonequilibrium [Gas98a; Dor99; Gal99; Rue99b; You02; Gal03b]. Additionally, the advent of powerful computers made it possible to numerically solve the nonlinear equations of motion of many-particle systems [All87; Eva90b; Hoo91; Hoo99]. This enabled to investigate the interplay between microscopic chaos in the collisions of single particles and transport properties on macroscopic scales in much more detail than it was possible to the times of the founders of statistical mechanics. This book focuses on two basic approaches that evolved over the past 1 The

acronym stands for the initials of Sinai, Ruelle and Bowen. 1

klages˙book

May 8, 2007

4:1

2

World Scientific Book - 9in x 6in

klages˙book

∗ Introduction

and outline

two decades trying to develop a concise picture of nonequilibrium statistical mechanics by employing methods of dynamical systems theory. In the following two introductory sections we summarize important features of these two directions of research. We then sketch how this book is embedded into the existing literature, outline in more detail its contents and say some words about the style in which it is written. The hurried reader may wish to immediately pick up the red thread through this book provided in Section 1.3.

1.1

Hamiltonian dynamical systems approach to nonequilibrium statistical mechanics

In recent work, Gaspard, Nicolis and Dorfman studied nonequilibrium situations by imposing specific boundary conditions onto spatially extended chaotic Hamiltonian (-like) dynamical systems. A typical example is diffusion processes due to concentration gradients at the boundaries. By this approach the macroscopic transport properties of deterministic dynamical systems could be linked to the underlying microscopic chaos in the equations of motion of the single particles in two ways: The escape rate formalism considers dynamical systems with absorbing boundaries [Gas90; Gas92c; Gas93; Gas95b; Dor95; Gas95c; Gas98a; Dor99]. Here the escape rate determined by a statistical physical transport equation such as, for example, the diffusion equation, is matched to the one calculated from the Liouville equation of the dynamical system. This procedure yields simple formulas linking transport coefficients to dynamical systems quantities, which are here the positive Lyapunov exponents and the Kolmogorov-Sinai entropy of the dynamical system or the fractal dimension of the repeller associated with the open system. A second, conceptually related approach applies to closed systems with periodic boundary conditions. It has been worked out for diffusion [Gil01; Gas01] and for reaction-diffusion [Cla02] in low-dimensional models. In this case the decay rate to thermal equilibrium obtained from the reactiondiffusion equation is related to the fractal dimension of the corresponding hydrodynamic mode in the Liouville equation of the dynamical system. The transport coefficients can thus be expressed as functions of the system’s largest Lyapunov exponent combined with the Hausdorff dimension of this mode. Both approaches can consistently be derived by using Ruelle’s thermodynamic formalism [Gas95c; Gas01]. Along similar lines the origin of nonequilibrium entropy production in

May 8, 2007

4:1

World Scientific Book - 9in x 6in

1.1. Hamiltonian dynamical systems approach to nonequilibrium statistical mechanics3

chaotic dynamical systems has been investigated. Here the analysis of transport in two-dimensional multibaker maps, introduced by Gaspard [Gas92a], played a crucial role. Multibakers are deterministic versions of random walks on the line, where stochasticity is replaced by microscopic chaos. On the other hand, these maps share generic properties with Hamiltonian particle billiards of which Lorentz gases [Lor05] are typical examples. In these models a moving point particle collides elastically with circular scatterers distributed randomly or periodically in space. The periodic version with applied external field is also known as the Galton board [Gal77]. Both multibakers and Lorentz gases became paradigmatic models in the field of chaos and transport [Gas98a; Dor99; Hoo99; Sza00; Kar00; Gar02; Vol02]. For nonequilibrium transport in such systems, T´el, Vollmer, Breymann and M´ aty´ as [Bre96; T´el96; Vol97; Bre98; Vol98; T´el00; Vol00; M´ at01; Vol02; Vol04; M´ at04b] as well as Gaspard, Tasaki, Dorfman and Gilbert [Gas97b; Gil99a; Tas99; Tas00; Gil00a; Gil00b; Dor02] proposed new concepts for defining coarse-grained Gibbs entropies leading to a nonequilibrium entropy production that is in agreement with irreversible thermodynamics. The former group attributed the source of irreversible entropy production to the chaotic mixing of dynamical systems and to the associated loss of information due to coarse graining. The latter authors argued that the singularity of the SRB measures exhibited by these nonequilibrium systems already necessitates a respective coarse graining for mathematical reasons. In both approaches, the source of irreversible entropy production is thus identified with the fractal character of these SRB measures. Cohen and Rondoni, on the other hand, criticized both theories starting from the fact that the simple models to which these analyses were applied consist of moving point particles that do not interact with each other but only with fixed scatterers [Coh98; Ron00a; Coh02; Ron02b]. In their view these systems are non-thermodynamic models, which do not allow to identify local thermodynamic equilibrium or any proper source of thermodynamic entropy production. These arguments have been critically analyzed in Refs. [Vol02; Dor02; T´el02; Gas02a; Gas03]. Another important topic associated with the Hamiltonian approach to nonequilibrium transport concerns the parameter dependence of transport coefficients. It originated from studying deterministic transport in simple one-dimensional chaotic maps periodically continued on the line [Sch89; Kla96b; Gas98a; Dor99; Cvi07]. Specific types of such models can straight-

klages˙book

May 8, 2007

4:1

4

World Scientific Book - 9in x 6in

klages˙book

∗ Introduction

and outline

forwardly be derived from the multibaker maps mentioned above [Gas92a; Gas98a; Dor99]. There exists quite an arsenal of methods to compute deterministic diffusion coefficients for these maps, such as transition matrix methods [Gas92a; Cla93; Kla95; Kla96b; Kla97; Gas98c; Gas98a; Kla99a; Koz99; Kla02a; Yos06; Bar05; Bar06], systematic evaluations starting from Taylor-Green-Kubo formulas [Kla96b; Dor99; Gas98c; Kla02d; Kor02; Kor04b], rigorous mathematical treatments related to kneading sequences [Gro02], cycle expansion methods [Art91; Art93; Art94; Tse94; Che95a; Gas98a; Cvi07; Cri06] and techniques employing the thermodynamic formalism [Sto94; Sto95a; Sto95c; Sto95b; Rad97]. Using such methods to calculate the parameter-dependent deterministic diffusion coefficient of a simple one-dimensional map, the result was found to be a fractal function [Kla03; Koz04; Kel07] of a control parameter [Kla95; Kla96b; Kla99a]. This forms the main theme of Chapter 2. The origin of this fractality can be traced back to the existence of long-range dynamical correlations in the microscopic dynamics of the moving particle [Kla96b; Kla02d]. These correlations are topologically unstable and change in a complicated way under parameter variation, a phenomenon that is known as “pruning” in periodic orbit theory [Cvi07]. Starting from this observation, both the electrical conductivity [Gro02] and the chemical reaction rate [Gas98c] of related maps were found to be fractal functions of control parameters, see Chapters 3 and 4. The driftdiffusion coefficient furthermore exhibits phase locking, and the nonlinear response turns out to be partly negative. The latter fact can be understood by relating the biased model to deterministic ratchets, see Section 3.4, for which the existence of current reversals under parameter variation is a characteristic property [Jun96; H¨ an96; Rei02]. Similar fractal transport properties were detected for phase diffusion in a time-discrete model of a driven nonlinear pendulum [Kor02; Kor04b], see Chapter 6. In this case bifurcation scenarios generate a complicated interplay between normal and anomalous diffusive dynamics. Analogous equations have been studied in the context of experiments on Josephson junctions [Jac81; Cir82; Mir85; Mar89; Wei00; Tan02d], on superionic conductors [Ful75; Bey76; Mar86] and on systems exhibiting charge-density waves [Bro84]. A related phenomenon has recently been discussed for a toy model of nanoporous material, which motivated the introduction of different notions of complexity [Jep06]. Purely anomalous dynamics is the subject of the second half of Chap-

May 8, 2007

4:1

World Scientific Book - 9in x 6in

1.1. Hamiltonian dynamical systems approach to nonequilibrium statistical mechanics5

ter 6. There it is shown that an intermittent subdiffusive map exhibits fractal parameter dependencies of a suitably generalized diffusion coefficient [Kor05]. This establishes an interesting crosslink to the very active field of anomalous transport for which a rich arsenal of stochastic methods is available while approaches by dynamical systems theory are currently being developed [Gas88; Wan89b; Wan89a; Wan93; Sto95b; Art93; Art97a; Cvi07; Det97a; Det98; Art03; Art04; Tas02; Tas04]. In view of experimental situations one may ask to which extent fractal transport coefficients are robust with respect to imposing random perturbations in space and time onto the underlying models. This connects the present line of research to the fields of disordered [Rad96; Rad99; Rad04; Fic05], respectively noisy [Gei82; Rei94; Rei96a; Rei96b; Fra91; Wac99; Cvi00] dynamical systems. A first answer to this question is provided by Chapter 5 in that the oscillatory structure of the diffusion coefficient of a simple model gradually “smoothes out” by increasing the perturbation strength, eventually recovering precise agreement with results from stochastic theory [Kla02b; Kla02c]. In order to move fractal transport coefficients to more realistic physical systems, studies of deterministic transport in Hamiltonian particle billiards were put forward. In the periodic Lorentz gas long-range dynamical correlations have again a profound impact on the parameter dependence of the diffusion coefficient [Kla00a; Kla02d]. This effect was found to be even more pronounced in a billiard with scatterers of flower-shaped geometry [Har02] and in a periodically corrugated floor, where particles move under the action of a gravitational force [Har01]. This forms the contents of Chapters 7 and 8. We remark that billiards of Lorentz gas-type can actually be manufactured and studied experimentally in form of semiconductor antidot lattices, where noninteracting electrons move quasi-classically under application of external electric and magnetic fields [Wei91; Lor91; Wei95; Wei97]. Indeed, the measured magnetoresistance of such systems is wellknown to be a highly irregular function of the field strength. For specific settings this quantity has theoretically been predicted to be fractal [Wie01]. Similar periodic structures are present in meso-, micro and nanoporous crystalline solids, which have wide-ranging industrial applications, for example, as molecular sieves, adsorbants and catalysts [K¨ ar92; Sch03a]. Techniques and results that are connected to the ones summarized in Chapter 7 are indeed available in literature on diffusion in zeolites, see Chapter 8 for details.

klages˙book

May 8, 2007

4:1

6

World Scientific Book - 9in x 6in

klages˙book

∗ Introduction

and outline

Another link between fractal transport coefficients and physical reality is provided by the bouncing ball billiard introduced in Chapter 9, which models deterministic diffusion of granular particles on a vibrating periodically corrugated floor. Simulations of this system revealed again highly irregular diffusion coefficients, which are here particularly due to phase locking and related resonances [M´ at04a; Kla04b]. This model was constructed in order to be relevant to recent experiments on granular material diffusing on vibrating surfaces [Far99; Pre02]. It also describes the very practical problem of transport of granular particles on vibratory conveyor belts [Per92; Han01; Gro03; Gro04]. To finish this brief outline of a Hamiltonian dynamical systems approach to nonequilibrium statistical mechanics, as far as it relates to this book, we refer to an interesting experiment by Gaspard et al. [Gas98b; Bri01]. Its purpose was to verify the existence of microscopic deterministic chaos in the Brownian motion of an interacting many-particle system. Long trajectories of a tracer particle suspended in a fluid were recorded, and the data analysis yielded a non-zero sum of positive Lyapunov exponents. Hence the existence of an exponential dynamical instability underlying many-particle diffusion, that is, chaos in the sense of Lyapunov, was concluded. This finding motivated the study of diffusion in models without Lyapunov dynamical instabilities. These “nonchaotic” models yielded results that were indistinguishable from the experimental data and were thus presented as counterexamples [Det99b; Gra99; Det00b]. As we will show in Chapter 17, such models belong to an interesting class of systems exhibiting what one may call pseudochaotic transport [Zas03], where dynamical randomness is generated by mechanisms that are weaker than exponential dynamical instabilities. Other prominent examples are polygonal billiard channels [Alo02; Jep06; San06], for which both normal diffusion and normal heat conduction have been verified numerically [Li02; Li03a]. The latter studies point towards another broad field of research, which tries to learn about the necessary and sufficient conditions for the existence of Fourier’s law [Bon00b; Lep03; Pro05]. Starting from harmonic and anharmonic chains up to thermal conduction in particle billiards like Lorentz and rotating disk channels, we will review some aspects of this research in the same chapter.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

1.2. Thermostated dynamical systems approach to nonequilibrium statistical mechanics7

1.2

Thermostated dynamical systems approach to nonequilibrium statistical mechanics

A nontrivial limitation of the Hamiltonian approach to chaotic transport is that it excludes nonequilibrium constraints generating a continuous flux of energy into the system as, for example, the application of external fields.2 Such situations necessitate the modeling of an infinite dimensional thermal reservoir , which is able to continuously absorb energy in order to prevent a subsystem from heating up [Pen79; Rue99b; Gal99; Dor99; Ron02a]. The need to model these situations emerged particularly from nonequilibrium molecular dynamics computer simulations, which focus on simulating heat or shear flow of many-particle systems or currents under application of external fields [Eva90b; Hoo91; Hes96a; Mor98; Hoo99; Det00a; Mun00; Tuc00]. A well-known example for a modeling of thermal reservoirs is provided by the Langevin equation [Lan08] yielding the interaction with a heat bath by a combination of Stokes friction and stochastic forces [Wax54; Rei65; Pat88; Kub92; Zwa01]. One way to derive generic types of Langevin equations starts from a Hamiltonian formulation for a heat bath consisting of infinitely many harmonic oscillators. This heat bath suitably interacts with a subsystem that consists of a single particle [Zwa73; For87; Kub92; Stu99; Zwa01]. In the course of the derivation the detailed bath dynamics is eliminated resulting in an equation of motion for the subsystem that is non-Hamiltonian. The Langevin equation thus nicely illustrates Ruelle’s statement “if we want to study non-equilibrium processes we have thus to consider an infinite system or non-Hamiltonian forces” [Rue99a], see Chapter 10.3 As we will argue in the second part of this book, there is nothing 2 Note that the use of Helfand moments enables an indirect treatment of such situations similar to the use of equilibrium time correlation functions related to Green-Kubo formulas [Dor95; Gas98a]. 3 For a related statement see, e.g., Smale [Sma80]: “We would conclude that theoretical physics and statistical mechanics should not be tied to Hamiltonian equations so absolutely as in the past. On physical grounds, it is certainly reasonable to expect physical systems to have (perhaps very small) non-Hamiltonian perturbations due to friction and driving effects from outside energy absorbtion. Today also mathematical grounds suggest that it is reasonable to develop a more non-Hamiltonian approach to some aspects of physics.” Smale further suggests to “revive the ergodic hypothesis via introduction of a dissipative/forcing term” into Hamiltonian equations of motion, since in his view dissipative dynamical systems have a better chance to be ergodic than Hamiltonian ones, which usually exhibit profoundly non-ergodic dynamics due to a mixed phase space.

klages˙book

May 8, 2007

4:1

8

World Scientific Book - 9in x 6in

klages˙book

∗ Introduction

and outline

mysterious in modeling thermal reservoirs with non-Hamiltonian equations of motion, in line with Refs. [Pen79; Sma80; Che95b; Rue96; Gal99; Lie99; Ron02a]. In case of thermostated systems the non-Hamiltonianity straightforwardly results from projecting out spurious reservoir degrees of freedom. Early nonequilibrium molecular dynamics computer simulations employed stochastic models of heat baths [And80; Cic80; Sch78; Ten82; All87; Nos91], which were partly considered to be inefficient. Infinite-dimensional Hamiltonian thermal reservoirs, on the other hand, can very well be modeled and analyzed analytically [Eck99b; Eck99a; Zwa01], but on a computer the number of degrees of freedom must, for obvious reasons, remain finite. These constraints provided a very practical motivation for constructing nonequilibrium steady states on the basis of finite-dimensional, deterministic, non-Hamiltonian equations of motion. About twenty-five years ago Hoover et al. [Hoo82] and Evans [Eva83a] came up with a strikingly simple non-Hamiltonian modeling of a thermal reservoir, which they coined the Gaussian thermostat [Eva83b]. This scheme introduces a (Gaussian) constraint in order to keep the temperature for a given subsystem strictly constant in nonequilibrium at any time step. A few years later Nos´e invented a closely related non-Hamiltonian thermal reservoir, which was able to thermostat the velocity distribution of a given subsystem onto the canonical one in equilibrium and to keep the energy of a subsystem constant on average in nonequilibrium [Nos84a; Nos84b]. His formulation was simplified by Hoover [Hoo85] leading to the famous Nos´e-Hoover thermostat [Eva90b; Hoo91; Hes96a; Mor98; Hoo99; Det00a; Mun00; Ron02a]. Suitable adaptations of these schemes to nonequilibrium situations such as, e.g., shear flows yielded results that were well in agreement with predictions of irreversible thermodynamics and linear response theory [Eva90b; Sar98]. Hence, these thermostats became widely accepted tools for performing nonequilibrium molecular dynamics computer simulations. Eventually, they were successfully applied even to more complex fluids such as, for example, polymer melts, liquid crystals and ferrofluids [Hes96a; Hes96b; Hes97], to proteins in water and to chemical processes in the condensed matter phase [Tuc00]. These two schemes are introduced and analyzed in detail in Chapters 11 and 12. Soon it was realized that a non-Hamiltonian modeling of thermal reservoirs not only enabled the efficient construction of nonequilibrium steady states on the computer but also that it made them amenable to an analysis by dynamical systems theory [Eva90b; Hoo91; Mar92a;

May 8, 2007

4:1

World Scientific Book - 9in x 6in

1.2. Thermostated dynamical systems approach to nonequilibrium statistical mechanics9

Mar97; T´el98; Mor98; Hoo99]. First of all, in contrast to the stochastic Langevin equation Gaussian and Nos´e-Hoover thermostats preserve the deterministic nature of the underlying Newtonian equations of motion. Furthermore, although the resulting dynamical systems are dissipative, surprisingly the thermostated equations of motion are still timereversible hence yielding a class of systems characterized by the, at first view, contradictory properties of being time-reversible, dissipative and, under certain circumstances, even being ergodic [Che95b; Che97; Hoo96b]. Computer simulations furthermore revealed that subsystems thermostated that way contract onto fractal attractors [Hol87; Mor87a; Mor87b; Hoo87; Pos88; Mor89a; Pos89] with an average rate of phase space contraction that is identical to the thermodynamic entropy production [Hol87; Pos88; Che93a]. This led researchers to conclude that in thermostated dynamical systems the phase space contraction onto fractal attractors is at the origin of the second law of thermodynamics [Hol87; Rue96; Rue97c; Rue97b; Hoo99; Gal98; Gal99]. Interestingly, the average rate of phase space contraction plays the same role in linking statistical physical transport properties to dynamical systems quantities as the escape or decay rates in the Hamiltonian approach to nonequilibrium [T´el96; Rue96; Bre98; Gil01]. The key observation is that, on the one hand, the average phase space contraction rate is identical to the sum of Lyapunov exponents of a dynamical system, whereas, on the other hand, for Gaussian and Nos´e-Hoover thermostats it equals the thermodynamic entropy production. For thermostated dynamical systems this again furnishes a relation between transport coefficients and dynamical systems quantities [Mor87a; Pos88; Eva90a; Van92]. A suitable reformulation of these equations makes them formally analogous to the ones obtained from the Hamiltonian approach to transport. These results were considered as an indication of the existence of a specific backbone of nonequilibrium transport in terms of dynamical systems theory [T´el96; Bre98; Gas98a; Gil01]. This discussion is summarized in Chapter 13. Another interesting feature of Gaussian and Nos´e-Hoover thermostated dynamical systems is the existence of generalized Hamiltonian and Lagrangian formalisms from which the thermostated equations of motion can be deduced, which involve non-canonical transformations of the phase space variables [Nos84a; Nos84b; Det96b; Det97b; Mor98; Cho98]. Similarly to Hamiltonian dynamics, deterministically thermostated systems often share a certain symmetry in the spectrum of their Lyapunov exponents known

klages˙book

May 8, 2007

4:1

10

World Scientific Book - 9in x 6in

klages˙book

∗ Introduction

and outline

as the conjugate pairing rule, which was widely studied in the recent literature [Dre88; Pos88; Eva90a; Mor98; Sea98; Rue99b]. That is, all Lyapunov exponents of a given dynamical system can be grouped into pairs such that each pair sums up to the same value, which in nonequilibrium is non-zero. In most cases Lyapunov exponents of thermostated systems can only be calculated numerically. For Lorentz gases and related systems, however, an analytical kinetic theory approach is available [vB95; Lat97; vB97; Del97b; vB98; M¨ ul04]. Furthermore, in recent computer simulations of interacting many-particle systems Posch et al. [Mil98a; Pos00a; Mil02; Hoo02c; For04] observed the existence of Lyapunov modes in thermal equilibrium indicating that the microscopic contributions to the Lyapunov instability of a many-particle fluid form specific modes of instability, quite in analogy to the well-known hydrodynamic modes governing macroscopic transport [Eck00; McN01; Tan02c; Mar04; dW04a; Tan05a; Tan05b]. We will further elaborate on these findings in Chapter 17. Such interesting properties inspired mathematicians to look at these systems from a rigorous point of view. A cornerstone is the proof by Chernov et al. of the existence of Ohm’s law for the periodic Lorentz gas driven by an external electric field and connected to a Gaussian thermostat [Che95b; Che97]. Another important development was the chaotic hypothesis by Gallavotti and Cohen [Gal95a; Gal95b; Gal98; Gal99], which was motivated by results from computer simulations on thermostated dynamical systems [Eva93]. This fundamental assumption generalizes Boltzmann’s ergodic hypothesis in summarizing some general expectations on the chaotic nature of interacting many-particle systems which, if fulfilled, considerably facilitate calculations of nonequilibrium properties; see the beginning of Chapter 18 for details and for what may represent a general picture of a chaotic dynamical systems approach to nonequilibrium statistical mechanics. A rapidly developing field of research is that of nonequilibrium fluctuation relations, which establish simple symmetry relations between positive and negative fluctuations of nonequilibrium entropy production. A first version of such laws again came up within the framework of nonequilibrium molecular dynamics computer simulations for thermostated systems [Eva93; Eva94; Eva02b]. Later on different formulations of fluctuation theorems were proven on the basis of the chaotic hypothesis [Gal95a; Gal95b], for stochastic systems [Kur98; Leb99] and for Gibbs measures [Mae99]. Another type of fluctuation relations emerged at first quite independently from these theorems: The Jarzynski work relation connects equilibrium with nonequilibrium statistical mechanics by suggesting to ex-

May 8, 2007

4:1

World Scientific Book - 9in x 6in

1.3. The red thread through this book

klages˙book

11

tract equilibrium free energy differences from irreversible work performed in a nonequilibrium situation [Jar97b]. A link between this relation and fluctation theorems was later on realized in form of the Crooks relation [Cro99; Cro00]. Meanwhile fluctuation relations have been verified for many different systems in many different ways analytically and in computer simulations [Kur05; Eva02b; Ron02a; Jar02] as well as in experiments [Cil98; Wan02; Rit03; Bus05]. It appears that they belong to the rather few general results characterizing nonequilibrium steady states very far from equilibrium thus generalizing Green-Kubo formulas and Onsager reciprocity relations, which can be derived from them close to equilibrium. Chapter 17 provides a short introduction to this interesting new research topic.

1.3

The red thread through this book

The emphasis of this book is on two themes, which are intimately connected with the two directions of research outlined above: Part 1 elaborates on the calculation and explanation of fractal transport coefficients in low-dimensional deterministic dynamical systems. Part 2 deals with the construction and analysis of nonequilibrium steady states in dissipative dynamical systems associated with thermal reservoirs. On the basis of these discussions, Part 3 provides an outlook towards further important directions of research within the field of chaos and transport, summarizes all results and concludes with a number of open questions. For a quick reading one is recommended all parts marked with a star (*), supplemented by studying the introductory remarks preceding all the single chapters. Sections labeled by a plus (+) contain advanced material that may be left for thorough studies. We highly recommend inclusion of the full Chapter 2 in such a quick reading. This chapter summarizes the major finding of the author’s Ph.D. thesis work, which is the existence of a fractal diffusion coefficient in a very simple deterministic map. This map is a paradigmatic model motivating many of the investigations reported in Part 1. It furthermore exemplifies a deterministic approach towards nonequilibrium transport. The central aim of the work compiled in Chapters 3 to 9 is to bring fractal diffusion coefficients to physical reality. This requires a sharpening of the theoretical methods, which goes hand in hand with studying a series of increasingly more complex models. In the course of these efforts, analogous fractal properties were discovered for additional

May 8, 2007

4:1

12

World Scientific Book - 9in x 6in

klages˙book

∗ Introduction

and outline

deterministic transport coefficients and in more complicated settings. Our main conclusion is that fractal transport coefficients are typical for a specific class of physical dynamical systems, and that they should be seen in experiments. To systematically argue for this statement is the main task of Part 1. Concerning Part 2, we suggest that one goes at least through Section 10.1, which motivates the physics of thermal reservoirs in an intuitive way. The patient reader may wish to consult as well Section 10.2 for a more detailed motivation, which starts from the well-known Langevin equation by pointing towards some fundamental problems in modeling thermal reservoirs. If yet more time is left, we recommend taking a look at Chapter 11, which describes a paradigmatic modeling of a deterministic and timereversible thermal reservoir. Applying this method to a simple model in a nonequilibrium situation, we summarize the resulting chaos and transport properties of the combined system in this chapter. Some researchers have argued that these properties should be universal for thermostated dynamical systems in nonequilibrium steady states altogether, see Chapters 11 to 13. Hence, the main theme of Part 2 is to critically assess the universality of these results as obtained from a non-Hamiltonian approach to nonequilibrium steady states. In Chapters 14 to 15 we show that there exist thermal reservoirs yielding counterexamples to most of these claims of universality. Chapter 16 then establishes a previously unexpected relationship between deterministic thermal reservoirs and simple models for the motility of biological entities such as, e.g., migrating cells. The first chapter of the final Part 3 contains a very brief non-technical outline of four different research topics, which emerged particularly over the past few years. We emphasize that the corresponding single sections are not designed to provide in-depth reviews of these very active research areas. However, they might be of interest for researchers who are willing to take a bird’s-eye view on what we consider to be very stimulating new developments in this field. These introductions might also be suitable for graduate students who want to enter these topics, e.g., in order to prepare for seminar talks or as a starting point for some research project. All four sections can be read independently. As far as the style of this book is concerned we remark that all three parts are of a somewhat different nature. In a way, there exists a gradient of technicality, which is reflected in the degree of difficulty the non-expert may expect by reading through this book: Part 1 is the most technical one. Here the reader may enjoy learning about deterministic transport along the

May 8, 2007

4:1

World Scientific Book - 9in x 6in

1.3. The red thread through this book

klages˙book

13

lines of hermeneutics: That is, the main models, methods and findings are presented from successively different points of views thus providing themes and variations that the reader will hopefully find interesting. Part 2 is written in a more pedagocial way. Here we try to keep things as simple as possible and do not present more technical details than absolutely necessary. Consequently there are only few formulas in this part but many figures and a lot of text. Part 3 is even less technical than Part 2. We remark that the first two parts contain a number of previously unpublished results, see particularly Chapter 3 and Sections 14.1 as well as 16.3. All three parts presuppose some basic knowledge of (nonequilibrium) statistical mechanics [Rei65; Hua87] and of dynamical systems theory [Sch89; Eck85; Ott93; Bec93; All97; T´el06]. We touch upon some rigorous mathematics particularly in Section 3.2.1, otherwise our work represents a generic theoretical physicist’s approach towards chaos and transport in nonequilibrium statistical mechanics, which does not require a detailed knowledge of mathematical dynamical system’s theory. In this respect, we remark that for Part 2 we do not explicitly develop concepts such as SRB measures and Anosov systems, despite the fact that all issues discussed in this part are intimately related to them. We emphasize that these objects do play a crucial role for building up a more mathematical theory of nonequilibrium steady states [Gas98a; Dor99; Gal99; Rue99b]. However, in Part 2 we work on a level that is more applied consisting of straightforward physical examples and demonstrations combined with simple calculations and results from computer simulations. It is our hope that this approach still suffices to make the reader familiar with what we believe are some central problems in the field of chaos and transport in thermostated dynamical systems. Books and reviews that are closely related to the topics covered by this work are particularly Refs. [Kla96b; Gas98a; Nic98; Dor99; Cvi07] for Part 1 and Refs. [Eva90b; Hes96a; Mor98; Hoo99; Det00a; Ron02a] for Part 2. There also exists a number of conference proceedings and related collections of articles, which the reader may wish to consult [Mar92a; Mar97; T´el98; Kar00; Sza00; Gar02; Kla04c]. We tried to be quite exhaustive as far as literature is concerned that we feel is especially relevant to the problems highlighted in this book, which resulted in the compilation of more than 800 references. However, it is impossible to be complete, and we apologize in advance for any important work that may have escaped our attention. We hope that this long bibliography will serve as a useful source of information for scientists doing research in this field. We also remark that in this book

May 8, 2007

4:1

14

World Scientific Book - 9in x 6in

klages˙book

∗ Introduction

and outline

we usually do not intend to give historical accounts of developments in chaos theory and nonequilibrium statistical mechanics; for this purpose see, e.g., Refs. [Bru76; Gle88; Cvi07; Uff01]. Only on rare occasions do we go a little bit more into historical depth. Finally, there is a website available for this book, which will be kept up to date concerning comments, amendments or possible corrections, please see www.maths.qmul.ac.uk/∼klages/cftbook for further information.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

PART 1

Part 1: Fractal transport coefficients

15

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

This page intentionally left blank

16

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Chapter 2 ∗

Deterministic diffusion

In this chapter we analyze a paradigmatic model for deterministic diffusion, which is a chaotic map acting on the real line. In Section 2.1 we explain the basic idea of deterministic diffusion starting from the famous drunken sailor’s problem and introduce our model. Section 2.2 outlines a method of how to exactly calculate the parameter-dependent diffusion coefficient for this dynamical system. This physical quantity turns out to be a fractal function, a result that is expected to be typical for a wide class of dynamical systems. To the end of this section we supply some intuition about the physical origin of this fractality. The research presented in this chapter has been performed together with J.R.Dorfman [Kla95; Kla96b; Kla99a]. For a mathematical underpinning of these results see Refs. [Gro02; Kel07]; at the beginning we partially follow Ref. [Kla96b]. 2.1

A simple model for deterministic diffusion

In a first approach one may think of diffusion as a simple random walk. Fig. 2.1 shows quite a realistic example of such a process: Let us imagine that some evening a sailor wants to go home, but he is so completely drunk that he has absolutely no control over his single steps. For sake of simplicity let us imagine that he moves in one dimension. He starts at a lamppost at position x = 0 and then makes steps of a certain step length s to the left and to the right. However, since he is completely drunk he looses all memory between any single steps, that is, all steps are uncorrelated. It is like tossing a coin in order to decide whether to go to the left or to the right at the next step. One may then ask for the probability to find the sailor after n steps a distance x away from his starting point. 17

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

∗ Deterministic

18

diffusion

x

n Fig. 2.1 The “problem of the random walk” [Pea05] in terms of a drunken sailor at a lamppost. The space-time diagram shows an example of a trajectory for such a drunken sailor, where n ∈ N holds for discrete time and x ∈ R for the position of the sailor on a discrete lattice of spacing s.

This “problem of considerable interest” was first formulated (in three dimensions) by Karl Pearson in a letter to Nature [Pea05]. He asked for a solution, which was provided by Lord Rayleigh [Ray05] referring to older work by himself. Pearson then concluded: “The lesson of Lord Rayleigh’s solution is that in open country the most probable place to find a drunken man, who is at all capable of keeping on his feet, is somewhere near his starting point” [Pea05]. This refers to the Gaussian probability distributions for the sailor’s positions, which are obtained in a suitable scaling limit from a Gedankenexperiment with an “ensemble of sailors” starting from the same lamppost. Fig. 2.2 sketches the spreading of such a diffusing distribution of sailors in time [Rei65]. Ref. [Wei94] provides a short outline about the history of the random walk problem. In contrast to this traditional picture of diffusion as an uncorrelated random walk, the theory of dynamical systems makes it possible to treat diffusion as a deterministic dynamical process. We may now replace the poor sailor by a point particle. Instead of coin tossing, the orbit of such a particle starting at initial condition x0 may then be generated by a chaotic dynamical system xn+1 = M (xn ) . (2.1) M (x) is here a one-dimensional map that determines how a particle gets

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

19

2.1. A simple model for deterministic diffusion

ρ (x) n n1 n2 n3 x Fig. 2.2 Probability distribution functions to find a sailor after time n at position x on the line calculated for an ensemble of sailors starting at the same lamppost, cf. Fig. 2.1. Shown are three probability densities after different numbers of iteration n 1 < n2 < n3 .

mapped at discrete time steps n ∈ N from position xn to position xn+1 . A simple example for such a map will be introduced in detail below. There we will also explain in which sense we wish the map to be chaotic. Defining M (x) together with Eq. (2.1) yields the full microscopic equations of motion of the system. Note that this way the complete memory of a particle is taken into account, that is, all steps are fully correlated. The decisive new fact that distinguishes this dynamical process from the one of a simple uncorrelated random walk is hence that xn+1 is uniquely determined by xn , rather than having a random distribution of xn+1 for a given xn . If the resulting dynamics of an ensemble of particles, each governed by the same deterministic dynamical system M (x), has the property that a diffusion coefficient D exists, this process is called deterministic diffusion [Fuj82; Gei82; Sch82]. The diffusion coefficient may be defined by the Einstein formula [Ein05; Wax54] D = lim

n→∞

< (xn − x0 )2 > 2n

,

(2.2)

where < . . . > represents the average over an initial ensemble of values of x0 . We will typically choose the initial ensemble as one in which the points are distributed uniformly over a unit interval on the real line. In order to obtain a representative value for the diffusion coefficient one must argue that D is largely independent of the choice of the initial ensemble.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

∗ Deterministic

20

diffusion

Ma(x) 3

2

a 1

x 0

1

2

3

Fig. 2.3 A simple model for deterministic diffusion. The dashed line depicts the orbit of a diffusing particle in the so-called cobweb plot [All97]. The slope a, here a = 3, serves as a control parameter in the periodically continued piecewise linear map.

Fig. 2.3 shows the model that shall predominately be studied throughout Part 1. As far as we know, this map was first introduced by Grossmann and Fujisaka [Gro82; Fuj82]. It depicts a “chain of boxes” of chainlength L ∈ N, which continues periodically in both directions to infinity, and the orbit of a moving point particle. Let Ma : R → R , xn 7→ Ma (xn ) = xn+1 , n ∈ N0

(2.3)

be a map modelling the chain of boxes introduced above, i.e., a periodic continuation of discrete one-dimensional piecewise linear expanding [Bec93] maps with uniform slope. The index a denotes a control parameter, which is the slope of the map. If |a| > 1 the map is expanding, and its Lyapunov exponent λ := ln a is greater than zero. Thus Ma (x) is dynamically unstable, that is, nearby orbits diverge exponentially from each other, and

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

21

2.1. A simple model for deterministic diffusion

may in this sense be called chaotic [Ott93].1 Note that this way we are introducing an element of dynamical randomness into the system. The term “chain” in the characterization of Ma (x) can be made more precise as a lift of degree one, Ma (x + 1) = Ma (x) + 1 ,

(2.4)

for which the acronym old has been introduced [Kat95]. This means that Ma (x) is to a certain extent translational invariant. Being old, the full map Ma (x) is generated by the map of one box, e.g., on the unit interval 0 < x ≤ 1, which will be referred to as the box map. In this chapter it shall be assumed that the graph of this box map is point symmetric with respect to the center of the box at (x, y) = (0.5, 0.5). This implies that the graph of the full map Ma (x) is anti-symmetric with respect to x = 0, Ma (x) = −Ma (−x)

,

(2.5)

so that there is no “drift” in the chain of boxes. The case with drift will be discussed in Chapter 3. In Fig. 2.3, which contains a section of a simple map of this type, the box map has been chosen as ax , 0 < x ≤ 21 , a≥2 , (2.6) Ma (x) = ax + 1 − a , 21 < x ≤ 1 cf. Refs. [Gro82; Gas92c; Ott93]. The chaotic dynamics of this type of map is generated by a “stretch-split-merge”-mechanism for a density of points on the real line [Pei92]. For sake of brevity, Eq. (2.6), together with Eqs. (2.3) and (2.4), and trivially fulfilling Eq. (2.5), may be referred to as map L. Related maps have been discussed in Refs. [Gro82; Fuj82; Art91; Gas92c; Kla96b; Gro83; Tho83; Kla97]. It has been proposed [Fuj82] to look at the dynamics in this chain of boxes in analogy to Brownian motion [Wax54; vK92]: If a particle stays in a box for a few iterations, its internal box motion is supposed to be getting randomized and may resemble the microscopic fluctuations of a Brownian particle, whereas its external jumps between the boxes could be interpreted as sudden “kicks” the particle suffers by some strong collisions. This suggests that “jumps between boxes” contribute most to the actual value of the diffusion coefficient. Indeed, in Section 3.4 it will be shown that maps of this type are related to time-discrete versions of overdamped Langevin equations, which are standard models of Brownian motion. 1 This

is of course a very loose meaning of deterministic chaos. For mathematically rigorous definitions, which usually involve topological transitivity, see, e.g., Refs. [Dev89; Rob95; Kol97]; see also Section 17.4.2 for further details.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

∗ Deterministic

22

diffusion

This dynamics is usually described in statistical physics by introducing some stochasticity into the equations that model a diffusion process. The main advantage of the simple model discussed here is that diffusion can be treated by taking the full dynamics of the system into account, i.e., the complete orbit of the moving particle is considered, without any additional approximations. This is another way to understand the notion of deterministic diffusion in contrast to diffusion as obtained from stochastic approaches. That is, in the purely deterministic case the orbit of the particle is immediately fixed by determining its initial condition. One should note that the strength of diffusion, and therefore the magnitude of the diffusion coefficient, are related to the probability of the particle to escape out of a box, i.e., to perform a jump into another box. This escape probability, however, as well as the average distance a particle travels by performing such a jump, changes by varying the system parameter. The problem that will be solved in the following is to develop a general method for computing parameter-dependent diffusion coefficients D(a) for piecewise linear maps on the line. Here map L will serve as a simple example. However, the methods to be presented should work as well for any other piecewise linear map of the type introduced above, evidently with analogous results [Kla96b; Kla97].

2.2

A parameter-dependent fractal diffusion coefficient

In order to describe the dynamical behaviour of an arbitrary initial density for a set of particles on some interval of the real line, we need the Liouville (continuity) equation of the dynamical system, which is usually called Frobenius-Perron equation in the case of time-discrete maps. This equation is given by [Las94; Ott93; Gas98a; Dor99; Bec93] ρn+1 (x) =

Z

dy ρn (y) δ(x − Ma (y))

,

(2.7)

where ρn (x) is the probability density for points on the line and Ma (y) is the map under consideration. We suppose that the motion takes place on an interval 0 < x < L and we impose periodic boundary conditions, i.e. ρn (0) = ρn (L), or absorbing boundary conditions, ρn (x) = 0 for x = 0, L, for all n [vK92]. Next we use the argument of Gaspard and coworkers [Gas90; Dor95; Gas95c; Gas92a; Gas92b; Gas93; Gas92c] to relate the eigenmodes

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

23

2.2. A parameter-dependent fractal diffusion coefficient

of the Frobenius-Perron equation to the solution of the diffusion equation ∂ 2 %(x, t) ∂%(x, t) =D , (2.8) ∂t ∂x2 where %(x, t) is the macroscopic density of particles at a point x at time t, and D is the diffusion coefficient defined by this equation. If for large L and large n the first few eigenmodes of the Frobenius-Perron equation are identical to those of the diffusion equation, the diffusion coefficient can be obtained by matching eigenmodes in an appropriate scaling limit. More explicitly, for periodic boundary conditions %(0, t) = %(L, t), one can calculate that for large times %(x, t) has the form [Kla96b; Kla99a] %(x, t) = const. + A exp −4π 2 Dt/L2 ± i2πx/L . (2.9)

Consequently, if one can find a solution of Eq. (2.7), for large L and n, in the form of ρ(x, n) = const. + A0 exp (−γp (a)n ± i2πx/L)

,

(2.10)

one can relate the decay rate γp (a) to D by 2

D(a) = lim (L/2π) γp (a) . L→∞

(2.11)

For absorbing boundary conditions the diffusion coefficient is linked to the escape rate from the system by an equation similar to Eq. (2.11) [Kla96b; Kla99a]. The escape rate formalism of chaotic dynamics shows that the escape rate from a system with absorbing boundaries is equal to the Lyapunov exponent minus the Kolmogorov-Sinai entropy for particles trapped within the system whose trajectories lie on a fractal repeller [Kan85; Dor99; Gas98a]. For the case of maps with slopes of uniform magnitude considered here, the Komolgorov-Sinai entropy for the fractal repeller hKS (a) is identical with the topological entropy of points on the repeller and can be computed from γp (a) as hKS (a) = log a − γp /4 + O(L−3 ) [Gas92a; Gas92b; Gas93]. The use of maps with uniform slope is not an essential ingredient in the method for calculating D(a) that we are now going to describe, which can be applied to more general linear maps. The main idea is that the Frobenius-Perron equation can be written as a matrix equation whenever the parameters of the map are such that one can construct a Markov partition of the interval (0, L). A Markov partition has the property that parts of the partion get mapped onto parts of the partition, or onto unions of parts of the partition [Rue89; dM93; Bec93]. For such values of a, Eq. (2.7) can be written as 1 ρn+1 = M ρn , (2.12) a

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

∗ Deterministic

24

diffusion

Ma (x) 3

2

a 1

x 0

1

2

3

Fig. 2.4 Illustration of the dynamical system denoted as map L, cf. to Eqs. (2.3) to (2.6), for the particular slope a = 3. The Markov partition given by the dashed grid leads to the construction of the transition matrix in Eq. (2.13).

where ρn is a column vector of the probability densities in each of the Markov partition regions at time n and M is a topological transition matrix, whose elements Mij are unity if points in region j can be mapped into region i and are zero otherwise. As a simple example we consider the form of the matrix M when a = 3 for map L, and periodic boundary conditions are used on an interval of length L. In this case the regions of the partition are all of length 1/2, as illustrated in Fig. 2.4. Then M is a 2L x 2L matrix of the form   1 1 0 0 ··· 0 0 1 0 1 1 0 1 0 0 ··· 0 0    1 0 1 1 0 0 ··· 0 0      0 0 1 1 0 1 0 0 · · ·  .  (2.13) M =  0 0 1 0 1 1 0 0 ···   ..  ..  .. . .  . 0 1 0 0 ··· 0 0 1 1

In the limit n → ∞, for any L and for any “Markov partition” value of a, the Frobenius-Perron equation can be solved in terms of the eigenmodes of M for any initial value ρ0 (x) that is uniform in each of the Markov partition

May 8, 2007

4:1

World Scientific Book - 9in x 6in

2.2. A parameter-dependent fractal diffusion coefficient

klages˙book

25

regions. For periodic boundary conditions, M is always a (block)circulant [Dav79], the largest eigenvalue of M is precisely a according to the PerronFrobenius theorem [Gan71], and the corresponding eigenmode is a constant representing the equilibrium state. The rate of decay to equilibrium is obtained as γp (a) = log (a/χ1 ), where χ1 is the next largest eigenvalue of M [Gas92a; Gas92b; Gas93]. Analytical expressions for D(a) can be derived for all integer values of a ≥ 2. For even integers, results of Grossmann and Fujisaka [Gro82; Fuj82] are recovered, D(a) = (a−1)(a−2)/24, and for odd integers we find D(a) = (a2 − 1)/24. To obtain D(a) for a general Markov partition value of a, one can use numerical methods [Kla96b; Kla99a]. We remark that by methods of dynamical systems theory and ergodic theory it can be proven that the Markov partition parameter values are dense on the parameter line for all a > 2, that for all these parameter values the diffusion coefficient exists and that diffusion in map L obeys a central limit theorem [Gro02; Kel07]. Fig. 2.5 (a) shows the results for the diffusion coefficient of the dynamical system map L, Eqs.(2.3) to (2.6), for values of a in the range 2 ≤ a ≤ 8. In Fig. 2.5 (b)-(d) we present magnifications of three small regions in this interval.2 One can see clearly that D(a) has a complicated fractal structure with regions exhibiting self similar-like details. Quantifying the irregularity of these graphs in terms of fractal dimensions turned out to be a highly delicate matter. Very recently it has been proven rigorously that the box counting dimension [Man82; Ott93; Pei92] of D(a) is equal to one on any parameter subinterval [Kel07]. This follows from the existence of a modified version of Lipschitz continuity for D(a) featuring multiplicative logarithmic corrections, which furthermore implies that the graph is H¨ older continuous and hence continuous. The result verifies a conjecture of Ref. [Koz04], where the application of a different concept of fractal dimension already revealed the existence of logarithmic terms yielding integer values for the respective dimensionality. These findings correct and amend Ref. [Kla03], where fractal dimensions for D(a) were numerically computed both from the box counting method and by evaluating the autocorrelation function of this graph [Fal90; Tri95]. Due to undetected logarithmic corrections, in this work both dimensions were found to be close but not equal to one for large parameter intervals, see also Ref. [Yos06] for analogous results. However, both quantities were also reported to be themselves fractal functions of 2 For

a chainlength of L = 100, the numerical precision for each D(a) is always better than 0.1% with respect to the limit in Eq.(2.11). Therefore, errorbars do not appear in the diagrams.

4:1

World Scientific Book - 9in x 6in

klages˙book

∗ Deterministic

26

diffusion

2 (a) 1.05

1.5

D(a)

D(a)

1

1

(c)

0.95 0.9 0.85

0.5

0.8 5

5.2

5.4

5.6

5.8

6

a

0 2

3

4

5

6

7

8

a 0.835

0.365

(b)

0.36

(d)

0.83

0.355

0.825

0.35

D(a)

D(a)

May 8, 2007

0.345

0.82 0.815

0.34

0.81

0.335 0.33 3

3.02

3.04

3.06 a

3.08

3.1

0.805 5.6

5.61

5.62

5.63

5.64

5.65

5.66

a

Fig. 2.5 Diffusion coefficient D(a) computed for the dynamical system map L, see Eqs. (2.3) to (2.6), and some enlargements. Graph (a) consists of 7908 single data points. In graph (b)-(d), the dots are connected with lines. The number of data points is 476 for (b), 1674 for (c), and 530 for (d).

the control parameter if computed locally on a uniform grid of small but finite subintervals. Correspondingly, D(a) was characterized as a fractal fractal function in Ref. [Kla03]. Very recent results [Kel07] indicate that the latter finding reflects a strong parameter dependence of the exponent of the logarithmic terms, similar to the one indicated in Ref. [Koz04]. The diffusion coefficient displayed in Fig. 2.5 thus exhibits a very strange, novel type of fractality. Fig. 2.6 shows an enlargement of the region for 2 ≤ a ≤ 3. The dashed line is the prediciton of D(a) for a simple random-walk model [Sch82; Fuj82; Kla97; Kla02a; Kla02d]. Note that the model correctly accounts for the behaviour of D(a) near a = 2. The fine structure in this graph can be understood by considering the transport of particles from one unit interval to another. These regions are coupled to each other by turnstiles, where points in one unit interval get mapped outside that particular interval into

4:1

World Scientific Book - 9in x 6in

klages˙book

27

2.2. A parameter-dependent fractal diffusion coefficient

0.35 series α series β series γ

0.3 0.25

D(a)

May 8, 2007

4 3

0.2

2

2

0.15 2

1

0.1

1

2

2 2

0.05

1

2

0 2

2.2

2.4

2.6

2.8

3

a Fig. 2.6 Enlargement of the region of slope 2, lea ≤ 3 with the solution for a simple random walk model (dashed line) and labels for the values which are significant for turnstile dynamics (see text). For some points, the turnstile coupling is shown by pairs of boxes. The graph consists of 979 single data points.

another unit interval. As in the case of two-dimensional twist maps, such as the sawtooth map, these turnstiles are crucial for large-scale transport [Wig92; Mac84; Che89; Che90; Mei92]. The region 2 ≤ a ≤ 3 can be analyzed by studying the interaction of turnstiles. One recognizes three distinct series of values of a, each of which provides a cascade of seemingly self-similar regions of decreasing size as the limits a → 2 or a → 3 are approached. To understand these series, consider the trajectory of a point that starts just to the left at x = 1/2. The first iterate of x = 1/2 is in the second interval, (1, 2). The series α values of a are defined by the condition that the second iterate of x = 1/2 is at the leftmost point of the upward turnstile in the second interval (1, 2) (a = 2.732), or that the third iterate is at the corresponding point in the third interval (a = 2.920), etc. The numbers on the graph refer to the number of intervals the image of x = 1/2 has travelled before it gets to the appropriate point on the turnstiles. Series β points are defined in a similar way, but they are allowed to have two or more internal reflections within an interval before reaching the left edge of a turnstile. Series γ points

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

∗ Deterministic

28

diffusion

are defined by the condition that some image of x = 1/2 has reached the rightmost edge of an upward turnstile (i.e. some point x = n + 1/2, where n is an integer), and consequently an increase in a will lead to a decrease in D(a). These cascades provide a basis for a physical understanding of the structure of D(a) in this region: Particles leave a particular unit interval through a turnstile and undergo a number of iterations before they are within another turnstile. Whether they continue to move in the same or the reverse direction at the next and later turnstiles is a sensitive function of the slope of the map. Thus the fractal structure of the D(a) curve is due to the effects of long-range dynamical correlations among turnstiles and these correlations lead to changes of D(a) on an infinitely fine scale. A similar argument can be employed to explain, at least qualitatively, the fractal structure of D(a) for higher values of the slope [Kla96b].3 2.3

Summary

(1) We have motivated a deterministic chaos approach towards diffusive transport by generalizing the concept of diffusion as a simple random walk on the line. As an example, we have introduced a piecewise linear one-dimensional chaotic map modeling a diffusion process by including long-range deterministic dynamical correlations. (2) For such types of dynamical systems we have presented a method that enables the computation of the full parameter-dependent diffusion coefficient numerically exactly. This method is based on solving the eigenvalue problem of the Liouville (Frobenius-Perron) equation of the dynamical system by constructing the Liouville operator in terms of topological transition matrices. The diffusion coefficient can then be related to specific eigenvalues of this matrix. The result turns out to be a fractal function of the control parameter of the map. The fractal (box counting) dimension of this curve is furthermore found to be a fractal function of the control parameter. For a certain parameter region, a heuristic explanation of the self similar-like structure is given in terms of specific microscopic scattering processes that change in a very complicated way under parameter variation.

3 These qualitative explanations are not sufficient to get the exact values for all the local extrema of D(a). Compare, e.g., a = 3, 5, 7, . . .

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Chapter 3

Deterministic drift-diffusion

Here we put the analysis started in the previous chapter on more mathematical grounds. In Section 3.1 we rigorously define a slightly generalized model, which consists of the simple map studied before amended by a second parameter mimicking the action of a bias. We then introduce the quantities we want to compute, namely the cumulants which yield in particular the current and the diffusion coefficient as functions of the two control parameters. In Section 3.2 we outline two further methods which both enable one to compute exactly the values of these quantities for our model. Section 3.3 is devoted to an analysis of the phase diagram of this system. Its fundamental structure is explained by simple heuristic arguments. We then numerically evaluate the current and the diffusion coefficient on the basis of these methods and discuss the results with respect to the phase diagram. In Section 3.4 we establish a link between these simple maps and ratchets. The latter systems are usually modeled by overdamped Langevin equations and have been widely studied both in theory and in experiments. This chapter profoundly reflects a cooperation with J.Groeneveld [Gro02], to whom the author is very much indebted for many invaluable explanations.1

3.1

Drift-diffusion model: mathematical definition

The model studied in the following is defined by a map M of the line R onto itself having the property that the translation by M , i.e. the function M (x) − x, is periodic in x with period one so that it satisfies, for all x ∈ R, 1 But of course J.Groeneveld is not responsible for any mistakes contained in this chapter.

29

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

30

klages˙book

Deterministic drift-diffusion

the relation M (x + 1) = 1 + M (x)

,

(3.1)

which has been coined old, cf. Eq. (2.4). To such a map M there is associated a map m of a half open/half closed interval or, equivalently, of the circle, onto itself, which can be made precise as follows: We first define within R two fundamental sets, i.e. the two half open/half closed unit intervals I0− = [− 21 , 12 ) and I0+ = (− 12 , 12 ]. To say that such an interval I0 is a fundamental set of R with respect to its normal subgroup Z only means that any real number x ∈ R has the unique decomposition x =y+n

(3.2) I0 .

into an integer n ∈ Z and an element y ∈ Accordingly, we may decompose, for any fixed = ±, a function M satisfying the relation Eq.(3.1) uniquely as a sum of a map m of I0 onto itself, and an integer valued function z with period one, as follows: (1) For x ∈ I0 we define i(x) and z(x) by the relation M (x) = i(x) + z(x)

(3.3)

and the conditions that i(x) ∈ I0 and z(x) ∈ Z. (2) This definition is extended to all of R by imposing the relations i(x + 1) = 1 + i(x) and z(x + 1) = z(x). The result is that Eq.(3.3) holds true for any x ∈ R. It is now obvious that m not only follows linearly from M in a unique manner, but also that this is an idempotent operation, i.e. a projection. We now specialize our M to being piecewise linear, and that of the simplest kind in that the number of its laps per unit interval equals one. We then arrive at the following specification of M (x) for |x| < 12 : M (x) = ax + b ,

(3.4)

where a and b are two real parameters. This definiton is completed by assigning to M ( 12 ) a respective value. This map is a straightforward generalization of the map Eq.(2.6) from the previous chapter for which b = 0. Related maps with broken symmetry have been studied in Refs. [Bar97; Bar98; Art06]. Our aim here is to determine the long-time behavior of the position xn at time n of a point particle starting at time n = 0 at x0 which moves in discrete time steps n = 0, 1, . . . according to the equation of motion

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

31

3.1. Drift-diffusion model: mathematical definition

xn+1 = M (xn ), see Eq.(2.1). The most general case would be to assume for the initial position x0 any probability distribution or measure µ0 restricted to either one of the fundamental intervals I0 . As a consequence of Eq.(2.1), xn will have a distribution µn whose time dependence is governed by the Frobenius-Perron operator LF corresponding to the map M , symbolically written as µn+1 = LM µn

.

(3.5)

The distribution µn of xn can be gathered from the characteristic function Qn (u) of µn defined by Qn (u) =< euxn >0

(3.6)

R∞

with < . . . >0 = −∞ dµ0 . . ., where u is a complex number the real part <(u) of which is sufficiently small so that the integral converges. The above would be equivalent to knowing the logarithm of Qn (u), which is the cumulant generating function [vK92], ∞ X

ui Ki (n) = log Qn (u) ,

(3.7)

i=1

or, equivalently, to knowing the set of cumulants Ki (n) (i = 1, 2, . . .) of µn . The first two of these are the most important ones to describe particle transport after long times and are expressed in terms of the moments of xn according to K1 (n) =< xn >0

(3.8)

and K2 (n) =< x2n >0 − < xn >20

.

(3.9)

In terms of these, the current J and the diffusion coefficient D can be expressed, if these limits exist, by 1 K1 (n) n

(3.10)

1 K2 (n) . 2n

(3.11)

J ≡ c1 = lim

n→∞

and D ≡ c2 = lim

n→∞

c1 was already considered by Poincar´e [Ito81] and is also known as the rotation number or winding number. The quantity c2 for this model at b = 0 was studied in the previous section. These quantities ck for any k will

May 8, 2007

4:1

World Scientific Book - 9in x 6in

32

klages˙book

Deterministic drift-diffusion

be referred to as the cumulant rates or scaled cumulants of the dynamical system. For k ≥ 3 they yield the higher-order Burnett coefficients [Dor77]. The limits in Eqs.(3.10) and (3.11) depend in general on the initial measure µ0 . For example, if µ0 is chosen to be concentrated on a single point x0 , then, due to the property of sensitive dependence on initial conditions for |a| > 1, stationary measures do not exist and, even worse, the corresponding measures do not represent the physical properties one is interested in. Hence, it seems necessary to restrict the measures we will consider to a smaller class. A minimal requirement for such a class should be the invariance under the Frobenius-Perron operator of the dynamical map M of the problem at hand. In particular, one would like it to contain the natural Lebesgue measure on the space considered, and one would like it to be small enough such that the Frobenius-Perron operator would only have a single invariant vector. In addition, one would like to be able to do analysis, and so one would need a certain topology or even a metric on the function space. In the present problem there are several choices which meet all these requirements. One choice is the following: We consider the space of probability distributions with piecewise constant densities ρ(x) : R → R, which are related to the measure by dµ = ρ(x)dx

.

(3.12)

Firstly, this choice is physically reasonable. And secondly, as one can prove [Gro], the values of the above cumulant rates c1 , c2 , . . . do not depend on the particular choice of dµ0 from this class. Therefore, the values allowed this way are arguably the physical relevant ones.

3.2

+

Calculating deterministic drift and diffusion coefficients

In this section we review two basic methods to compute various dynamical properties of the model, such as the cumulant rates ck and some time dependent quantities. These are (1) the twisted eigenstate method2 [Gro02; Gro], which is rigorous and explicit enough so as to allow for further mathematical analysis, and (2) numerical transition matrix methods, 2 Here

we outline this method as understood by the present author on the basis of kind explanations by its creator J.Groeneveld. For more detailed expositions see the references cited above.

May 8, 2007

4:1

3.2.

World Scientific Book - 9in x 6in

+ Calculating

deterministic drift and diffusion coefficients

klages˙book

33

which represent refinements of the approach described in Refs. [Kla96b; Kla97]. 3.2.1

Twisted eigenstate method

The aim of this method is to obtain the large time behavior of the probability distribution of xn , which is tantamount to computing the characteristic function Qn (u) of xn with respect to the initial measure µ0 . For this purpose, one takes advantage of the following two basic features: 1) The discrete translational symmetry in the problem as given by Eq.(3.1) can be expressed by saying that the map M (x) and the discrete translation operator T (x) = 1 + x commute, MT = T M

.

(3.13)

This allows one to split the function space on which the Frobenius-Perron operator LM of Eq. (3.5) acts into eigenspaces of a corresponding translation operator LT . The elements of these latter spaces can be reinterpreted as functions on a unit interval with properly modified boundary conditions, which are also called twisted boundary conditions in the literature [Gro81]. In the present case one such space, which is characterized by a complex number u called the twist parameter, consists of functions satisfying ψ(x + 1) = e−u ψ(x)

(3.14)

where ψ(x) is a C-valued function on the real line. In addition, these functions are required to be of bounded variation. The corresponding FrobeniusPerron operator will then be denoted with LM (u). It should be remarked that, in the general case of eu 6= 1, probability is no longer conserved globally, although it still is locally. This is expressed by the fact that the stationary eigenstate of LM (u) obeying LM (u)|Φ >= λ(u)|Φ > ,

(3.15)

which approaches the equilibrium one if u → 0, has an eigenvalue λ0 (u) which is no longer equal to one. It is just this quantity which contains all information we need to solve our problem, that of calculating the cumulant rates ck . One has that their generating function ∞ X c(u) = uk c k (3.16) k=1

is given by

c(u) = log λ0 (u) ,

(3.17)

May 8, 2007

4:1

World Scientific Book - 9in x 6in

34

klages˙book

Deterministic drift-diffusion

where in turn λ0 (u) = lim

n→∞

1 log Qn (u) . n

(3.18)

2) As a calculational device, the problem is reinterpreted in terms of one-dimensional electrostatics: The probability density can then be considered as an electric field strength, and a jump at a discontinuity of such a density is therefore regarded as a charge at that point. This is an entirely equivalent description of the problem, which has the virtue of facilitating the calculations of the spectrum of LM (u) of which we need λ0 (u) for our present purpose. Only if we want to calculate time dependent quantities, such as velocity autocorrelation functions, as can also be done within the present formalism, do we need the remaining parts of the spectrum of the operator. In this picture, the effect of LM (u) on the point charges is to move them according to the map M as if they were particles, and to decrease their strengths by multiplying them with a factor of 1/a. There is then a simple procedure to obtain an equation which determines λ0 (u) uniquely for sufficiently small u and hence all coefficients ck in Eq.(3.16). To obtain that crucial formula, the action of LM on the set of piecewise constant functions defined above must be translated respectively. We note that the latter are characterized by a set of point charges such that the sum of the positive ones within either of the intervals I0± is finite, as well as that of the negative ones (the sum of the absolute magnitudes of these two total charges is just the ‘bounded variation’-norm alluded to earlier). However, before letting the map M act on these charges in the way mentioned above, the Frobenius-Perron operator makes a small modification of such a state: Step A: modify At every point where M is discontinuous a double layer is introduced consisting of two equal charges of opposite signs such that in-between these charges the electric field (i.e., the probability density) vanishes. This can always be done and corresponds to shifting charges by an infinitesimal amount only; which implies that our present calculation of an equation for λ0 (u) is still rigorous and precise. An important consequence of these modifications is that afterwards the total charge in each unit interval vanishes. Step B: move and reduce In this next step all charges are moved and reduced in size as explained above. The action of LM in the language of electrostatics then consists of these two parts A and B, or if one wishes: LM (u) = LM,B (u)LM,A (u). The

May 8, 2007

4:1

3.2.

World Scientific Book - 9in x 6in

+ Calculating

klages˙book

35

deterministic drift and diffusion coefficients

above techniques allow to derive an equation for λ0 (u) which will turn out to determine, together with the condition λ0 (u) → 1 as u → 0 ,

(3.19)

the function λ0 (u) uniquely. The argument can be given in three steps: (1) One assumes that λ ≡ λ(u) 6= 0 is an eigenvalue of LM (u) according to Eq.(3.15) and writes down the equations for the coordinates of all the charges which may occur in such an eigenstate, and for their strengths. For this purpose, we recall that = ±, I0+ = (− 12 , 21 ], and I0− = [− 12 , 12 ), and start with y0 = . (3.20) 2 Next we define for all r = 0, 1, . . . the numbers nr and yr by the relation nr + yr+1 = ayr + b

nr

yr

(3.21)

r+1 I0α ,

and the conditions ∈ Z and ∈ where α is a constant defined by α = sign(a). (2) The effect of LM,B on an eigenstate is then just to ‘normalize’ such a state in the sense that it is a projection operator: L2M,B = LM,B . Hence, if |Φ > is an eigenstate with eigenvalue λ, so then is LM,B |Φ >. (3) Consequently, the total charge of an eigenstate, as far as it is located within either of the two fundamental intervals I0 , must vanish. The strengths of these charges can be calculated recursively and involves all powers of λ. Requiring their sum to be zero then leads to a consistency equation of the form C(λ, u) = 0 . The left hand side of this ‘consistency function’ is defined by ∞ X X C(λ, u) = (λa)−r euNr . =±

(3.22)

(3.23)

r=0

This equation for λ as a function of u is the central formula of our approach. It turns out that the consistency function is analytic in λ and u near the set λ = 1 , <(u) = 0. Furthermore, a simple argument shows that this equation has only a single root λ0 which satisfies the further condition Eq.(3.19). This root even turns out to be analytic in u. At this point all that remains to be done is to expand Eq.(3.22) into powers of log λ and u. Here one encounters certain ‘N -quantities’ defined for r = 0 by N0 = − 2 and for r = 0, 1, . . . by Nr+1 = Nr + nr

.

(3.24)

May 8, 2007

4:1

World Scientific Book - 9in x 6in

36

klages˙book

Deterministic drift-diffusion

We note that Nr ∈ Z + 21 . In addition, we will encounter certain ‘N moments’ defined by Nk,l =

∞ 1 X X −r k l a (Nr ) r k!l! =± r=0

.

(3.25)

Expanding into powers of c ≡ log λ and u by taking into account also Eq.(3.16) and Eq.(3.17), the above implicit equation for λ0 is reduced to an explicit equation for each of the scaled cumulants ck (k = 1, 2, . . .). These expressions turn out to be amenable to iterative calculations on a computer. Here we display only the resulting formulas for c1 and c2 in the case of general (a, b) and for c2 and c4 in the bias-free case b = 0. They are: (1) For general (a, b), c1 and c2 are given by c1 = N2,0 /N1,1 c2 = (N3,0 − N2,1 c1 +

(3.26) N1,2 c21 )/N1,1

.

(3.27)

(2) For b = 0, the odd order ck ’s vanish and the expressions for c2 and c4 simplify to c2 = N3,0 /N1,1 c4 = (N5,0 − N3,1 c2 +

(3.28) N1,2 c22 )/N1,1

.

(3.29)

There are also alternative expressions in terms of the y-quantities instead of the above ones which are in terms of the N -quantities. Of these we display only the following intriguingly simple one for c1 . It can be derived from Eq. (3.26) by using the relations (3.21), (3.24) and (3.25), J = c1 = b + (a − 1)y (2) /y (1)

,

(3.30)

where the y (k) ’s are defined, for k = 1, 2, by y (k) =

∞ 1 X X −r k a (yr ) k! =± r=0

.

(3.31)

Finally, we remark that it follows as a direct consequence of the PerronFrobenius theorem that the quantities N1,1 and y (1) , which are related by (a − 1)N1,1 = ay (1)

,

(3.32)

are both positive for all real a and b with |a| > 1. This is comforting since it shows consistency with the fact that the above expressions require division by either of these quantities. Hence these divisions can cause no problem.

May 8, 2007

4:1

3.2.

World Scientific Book - 9in x 6in

+ Calculating

deterministic drift and diffusion coefficients

klages˙book

37

The above expressions can be reformulated in terms of some simple, fast recursive schemes suitable for numerical computations [Gro02; Koz04]. Additionally, we remark that by employing the twisted eigenstate method proofs of several important properties of our model can be obtained, such as [Gro]: • The Markov partition points are dense in the chaotic part (|a| > 1) of the parameter plane. • For typical values of a, the ratio J(a, b)/b has no limit as b → 0. • D(a, 0) is continuous in a. Both the first property for b = 0 and the last property were originally conjectured in Refs. [Kla95; Kla96b; Kla99a] on the basis of detailed numerical evidence; for proofs of the first result for b = 0 and of the last result see also Ref. [Kel07]. Interestingly, in the case of b = 0 very recently another exact solution for all parameter values of the diffusion coefficient has been derived by generalizing cycle expansion techniques [Cri06]. Similarly to the transition matrix methods reported in Chapter 2 (see also below), previously periodic orbit theory seemed restricted to Markov partition values of the slope [Cvi07]. The precise relation between the set of equations calculated in Ref. [Cri06] and the solution above is currently unclear, but it may be conjectured that they can be transformed into each other. 3.2.2

Transition matrix methods

A more physically intuitive way to solve the Frobenius-Perron equation Eq. (3.5) for our model starts from its formulation as a matrix equation, see Eq. (2.12). Here the probability density is given by a column vector at discrete time steps defined with respect to a suitable partitioning in phase space, and the Frobenius-Perron operator is given by a topological transition matrix. Instead of solving the eigenvalue problem of the transition matrix, however, as was outlined in the previous chapter, solutions of this equation can also be obtained by straightforward numerical iteration of this equation [Kla96b; Kla97; Gas98c; Kla02a]. In both cases the fundamental problem is how to construct the transition matrix of Eq. (2.12). We elaborate on this by first discussing in more detail the method of construction used in Chapter 2, which employs Markov partitions. We then point out the limited applicability of this method in the

May 8, 2007

4:1

38

World Scientific Book - 9in x 6in

Deterministic drift-diffusion

case of our present two-parameter model. This leads to two improvements of the original method: one that incorporates a more efficient approach to obtain Markov partition points and a second one which is not restricted to Markov partitions anymore. 1. Markov partition method: In Chapter 2 we have computed the diffusion coefficient D as a function of the slope a for our map Ma,b specialized to the case b = 0. Here the fact has been used that for parameters where a Markov partition for the given map exists, the topological transition matrix of Eq. (2.12) can be constructed in a straightforward way. Thus, the applicability of this method, which we call the Markov partition method, depends, firstly, on whether a Markov partition exists or not; secondly, on whether these partitions are dense in the parameter space; and thirdly, on whether the quantity to be calculated is continuous in the parameters. Thus, an important sub-problem of this approach is that of finding parameter values corresponding to Markov partitions close enough to the values one is interested in. In the case of the one-parameter model at b = 0, one can solve this sub-problem by defining a partition via the iteration of the critical point of the map. Markov parameter values of the slope a can then be computed as the roots of algebraic equations corresponding to certain classes of such orbits [Kla96b; Kla99a]. However, in the case of our two-parameter model considered here, two such discontinuous algebraic equations must be solved simultaneously. This makes the method technically more difficult and numerically cumbersome. For this reason, a more efficient method was looked for to compute Markov partition parameter values. 1.a Renormalized iteration method: This method has the property that it converges to a Markov partition parameter pair nearby an arbitrary initial parameter pair (a, b). The method starts by constructing transition matrices according to the prescription of the Markov partition method, but stopping this process after a finite number of steps irrespective of whether a Markov partition has been found or not. This implies making a small error in the construction of the matrices such that only an approximate matrix for a given parameter pair is obtained. A consequence is that this approximate matrix in the Frobenius-Perron equation Eq. (2.12) is no longer probability conserving. The loss of probability is then compensated by renormalizing the Frobenius-Perron equation after each time step, which amounts to calculating a new slope which is closer to a Markov partition value. Under numerical iteration the probability density of this renormalized equation converges to the one of a Markov partition parameter pair nearby the orig-

klages˙book

May 8, 2007

4:1

3.2.

World Scientific Book - 9in x 6in

+ Calculating

deterministic drift and diffusion coefficients

klages˙book

39

inal one. The parameter values of this Markov partition can then easily be determined.3 A disadvantage of this renormalized iteration method is that, still, it only solves the Frobenius-Perron equation for Markov partition parameter pairs. Therefore, it was desirable to develop a matrix method which would not be restricted to Markov partitions anymore. 1.b Approximate iteration method: Here approximate transition matrices are constructed for arbitrary parameter pairs of our model in a way that the norm of the probability density in the Frobenius-Perron equation is preserved. The construction is such that for a given size of the partition, i.e. for a given number of its parts, and by continuously changing the two parameters, the matrices yield linear interpolations between single Markov partition parameter pairs approximating the exact solution of the Frobenius-Perron equation. Thus, at any Markov partition parameter pair the method gives the exact results for the respective Markov partition. Since the Markov partitions are dense in the chaotic part of the parameter plane [Gro], by allowing the size of the partition to increase this method converges everywhere to the exact results. In low order, that is, for small partition size, the method can be applied analytically. In higher order, that is, for larger partitions, it was implemented numerically. Related transition matrix methods have recently been applied to the periodically continued map Eq. (3.4) in the case of b = 0 within the setting of the thermodynamic formalism [Yos06]. Here the fractality of the diffusion coefficient has been coined a “complex parameter dependence”. A similar problem has been studied within the context of electrical engineering [Bar05; Bar06], see Section 4.2.2 for further details. 3.2.3

Numerical comparison of the different methods

Both the improved transition matrix methods and the twisted eigenstate method have been implemented numerically for our two-parameter model. The twisted eigenstate method yields analytically exact expressions for the first four cumulant rates ck , but in principle it is not restricted to these. The two improved transition matrix methods provide a direct picture of the spreading of the probability density in time and space starting from some initial density. In this sense, the matrix methods give access to more information than the twisted eigenstate method, which currently focuses 3 Note that apart from renormalizing the slope in the Frobenius-Perron equation the bias parameter needs to be optimized by a suitable variation.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

40

Deterministic drift-diffusion

on the asymptotic behavior of the cumulant rates only. That is, via the time dependent probability density all relevant time-dependent quantities can be obtained such as higher cumulant rates and velocity autocorrelation functions [Kla96b]. However, the matrix methods are not yet optimized for efficiently computing the values of the cumulant rates in the limit of infinite time. Such an optimization could be obtained by combining them with Green-Kubo formulas [Kla96b; Gas98c]. Here the twisted eigenstate method is numerically far more efficient and precise. Apart from that, the twisted eigenstate method is mathematically rigorous and allows one to obtain proofs of important general properties of the present model. The numerical differences may be illustrated by the following results: We chose some typical values in the parameter plane. We then applied our different methods to compute the first four cumulant rates at these parameters as described above. To iterate transition matrices up to a size of about 2500 x 2500 elements over 40 time steps, the matrix methods required computing times of about 0.4 to 4 CPU seconds per parameter pair on a SUN Ultra-5 and on DEC-Alphas. For the renormalized iteration method 1.a, this yielded a precision of about 12 digits behind the dot in the cumulant rates, for the approximate iteration method 1.b the precision was about 5 digits. The twisted eigenstate method required for the same parameter values and for a precision of much more than 12 digits behind the dot CPU times in the order of 10−3 to 10−4 seconds. Through application of these different methods we have verified that the numerical error is less than visible in the following Figs. 3.2, 3.3, 3.6 and 3.8. In the case of the other figures it has been checked via convergence of the data that the numerical error is less than visible.

3.3

The phase diagram

In this section we explore the parameter plane (a, b) of our model. We characterize the different regions in it according to certain topological properties of the map M and by studying the qualitative behavior of certain dynamical order parameters of the system such as the Lyapunov exponent and the cumulant rates J = c1 and D = c2 . We start by discussing what may be termed the phase diagram of our two-parameter model, by which we understand a kind of a map indicating the different dynamical phases distinguished by the above criteria, see

klages˙book

4:1

World Scientific Book - 9in x 6in

klages˙book

41

3.3. The phase diagram

unidirectional jumps

0.5 nonchaotic

chaotic

0.4 0.3

0

1

2

3

4

5

6

7

b

May 8, 2007

number of fixed points of map on the circle

0.2 0.1 nonchaotic

0 0

chaotic

1

2

3

4

5

6

7

8

localization regimes

a Fig. 3.1 The most significant part of the positive a-region of the parameter plane of the map Ma,b . Parts for other values of b follow by the reflection and translational symmetries of the system. At the zig-zag lines, as parameters are varied, fixed points of the projected map ma,b of the circle onto itself are generated or disappear. These straight lines thus separate regions characterized by a particular value of the number of fixed points of the circle map ma,b . Also shown is one of the Arnol’d tongues characterized by 2 > a > 1 and J = 0 for b → 0.

Fig. 3.1. Since our model M ≡ Ma,b obeys the translation symmetry as expressed by Ma,b + 1 = Ma,b+1 , cf. Eq.(3.1), and the reflection symmetry as expressed by Ma,b (x) = −Ma,−b (x), it follows straightforwardly that the cumulant rates ck ≡ ck (a, b) satisfy ck (a, b + 1) = ck (a, b) + δk,1 and ck (a, −b) = (−1)k ck (a, b), for k = 1, 2, . . .. Keeping these simple relations in mind it follows that one may, without loss of generality, restrict the discussion of the parameter plane to a fundamental strip, e.g., to the region defined by 0 ≤ b ≤ 21 . In Fig. 3.1 the positive a part of the fundamental strip of the parameter plane is represented. First, there is the Lyapunov exponent which, for the present model, is equal to ln |a|. This fact implies that the phase diagram can be subdivided into three connected regions according to the sign of the Lyapunov expo-

May 8, 2007

4:1

World Scientific Book - 9in x 6in

42

klages˙book

Deterministic drift-diffusion

nent: One non-chaotic region given by |a| < 1 and two chaotic regions defined by a < −1 and a > 1 respectively. In the case of the positive-a part shown in the figure, this separation of the parameter plane is indicated by the dotted vertical line at a = 1. As will be discussed below, the picture of the various phases in the region close to this transition line turns out to be quite complicated. A second way of characterizing the dynamical system is in terms of topological invariants: As a first and simplest example we take the property whether or not the movements of the particles are restricted to be possible only in the direction of the bias b. In the case b > 0, which we shall without loss of generality assume, this is equivalent to the assumption that our map Ma,b (x) satisfies for all x the condition M (x) > x. Because of the assumed periodicity with period one of M (x) − x, cf. Eq.(3.1), this is equivalent to the condition that for |x| < 21 , then ax + b > x, which yields 2b > |a − 1| .

(3.33)

The latter inequality defines a region of the parameter plane confined by two straight lines intersecting at a = 1, b = 0, as shown in Fig. 3.1. In the positive a-part of the fundamental parameter strip depicted in that figure, regions are then further subdivided by the values attained by a more general topological invariant, viz. the number of fixed points of the associated circle map ma,b . A simple reasoning shows that the zig-zag line in that figure can be obtained from the single straight line in the figure given by the equation 2b = 1 − a, which is just one of the two straight lines determined by Eq.(3.33), by means of the symmetry operations b → b + 1 (translation symmetry) and b → −b (reflection symmetry). It follows that whenever one crosses this zig-zag line, the number of fixed points of the projected map m changes by one. On the other hand, from Fig. 3.2 we observe that just at these lines the surfaces representing the current J and the diffusion coefficient D exhibit certain ridges. These lines are related to the process in which certain Markov partitions are generated [Kla]: In Fig. 3.1, each line of the zig-zag curve corresponds to the situation that one of the two critical points of m is mapped, in a single iteration, onto an endpoint x = ± 21 . Wherever two such curves intersect, as e.g., at a = 3, 5, 7, . . . and b = 0, 0.5, one finds oneself at a Markov partition point. Thus, we observe that at places where such a “half-way Markov condition” as meant above is satisfied, this appears to be reflected in the structure of the cumulant rates J and D. In addition, upon closer scrutiny of Fig. 3.2 further such topological subdivisions of the

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

43

3.3. The phase diagram

(a) J(a,b) 0.4 0.2 0

0.5 0.4 0.3 0.2 b

2 4 a

0.1

6 80

(b)

D(a,b)

2

0.5 0.4 0.3 0.2 b

1 0 2 4 a

0.1

6 80

Fig. 3.2 Three-dimensional plots of (a) the current J(a, b), and (b) the diffusion coefficient D(a, b), as functions of the control parameters a and b. Computations were performed on the basis of the explicit formulas for these quantities derived by means of the twisted eigenstate method, see Section 3.2. One notices that both surfaces are “scarred” along just the same straight lines as drawn in Fig. 3.1.

primary regimes of Fig. 3.1 can be detected. Such higher-order topological effects may eventually form the key to understand the full structure in the parameter dependence of the cumulant rates. We note that in the case of b = 0 a closely related argument has been used to understand the complicated structure of the diffusion coefficient D(a, 0) for our model in

May 8, 2007

4:1

World Scientific Book - 9in x 6in

44

Deterministic drift-diffusion

the region 2 < a < 3, cf. Section 2.2. We furthermore remark that J(a, b) = b for a ∈ N and J(a, b) = b for 2b ∈ N, which yields the exact boundaries in Fig. 3.2 (a). Note that in a small spot near the point (a = 2, b = 0.5) the function J(a, b) goes down far too steeply for it to be visible, in the chosen perspective, that at the boundary it indeed ends up at the value J(a, 0.5) = 0.5. At the boundary of Fig. 3.2 (b) at a = 1, then D = 0 for all b, and at a = 8 it exactly equals 7/4 for all b, as can easily be verified analytically. A third way in which we have classified points of the parameter plane, which is more physically relevant but also more difficult than the previous analysis, is a subdivision of the plane according to the qualitative behavior of quantities such as J and/or D. An example of this is the subdivision of the chaotic region for a < 2 into Arnold tongues, which are well-known from the sine circle map [Din87; Hao89]. In these subregions there is no diffusion, and the current J is fixed to a rational value. In Fig. 3.1 we have depicted part of the boundary curve of just one of them, which is easily calculated from an exact algebraic equation. However, this is only a first example of a region consisting of points for which the corresponding dynamical system has the property that there exist certain intervals on the line from which a particle cannot escape once it is inside them. In these situations the diffusion coefficient evidently vanishes, D(a, b) = 0. Further such Arnold tongue regions can be discerned in Fig. 3.3 (a) to (c), which display D(a, b) as a function of b for three fixed values of a. The transition line at a = 1 is approached in two ways: (1) Almost everywhere in the parameter plane where the diffusion coefficient does not vanish,4 this coefficient appears to be a fractal5 function. (2) The regions with D = 0 seem to be getting more and more numerous and more densely packed the closer a gets to one [Gro02]. Note also the critical behavior around these regions as magnified, for a typical example, in Fig. 3.3 (d). The dynamical explanation of these Arnold tongues is displayed in Figs. 3.4 and 3.5, where a uniform probability density on the unit interval was chosen as the initial condition. Fig. 3.4 shows the time evolution of this density in a chain of L boxes after a certain number of iterations. The parameter pair was situated within an Arnold tongue. One observes that this 4 By

of 5

“almost everywhere” we mean except at integer values of the slope and multiples for the bias, in which case the map yields a Bernoulli process. For a definition of the term “fractal function” see, e.g., Refs. [Ber80; Man82; Kla03].

1 2

klages˙book

4:1

World Scientific Book - 9in x 6in

klages˙book

45

3.3. The phase diagram

(a)

0.002

0.00015

D(a,b)

D(a,b)

0.0002

(c)

0.0015

0.001

0.0001

0.00005

0.0005

0

0 0

0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

0

0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

b

b

(d)

0.0008

0.00035

0.0007

0.0003

0.0006

0.00025

0.0005

D(a,b)

(b)

D(a,b)

May 8, 2007

0.0004 0.0003

0.0002 0.00015

0.0002

0.0001

0.0001

0.00005

0 0

0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 b

slope a=1.125 slope a=1.075 slope a=1.0375

0 0.32

0.325

0.33

0.335

0.34

0.345

b

Fig. 3.3 Diffusion coefficient D(a, b) as a function of the shift b for three fixed values of the slope a: (a) a = 1.125, (b) a = 1.075, (c) a = 1.0375. With a approaching the line a = 1 where a transition to non-chaotic behavior occurs, one observes more and more b-intervals with vanishing diffusion coefficient. In these regions mode locking occurs as is shown in Figs. 3.4 and 3.5. These are the well-known Arnold tongues whose number increases to infinity as a → 1. Their boundaries can be calculated exactly. The black boxes in (a), (b) and (c) indicate the region which is magnified in part (d).

density does not spread out anymore yielding D = 0, whereas it continues periodically moving to the right with a non-zero fluctuating velocity. In Fig. 3.5 the second cumulants for two parameter values in different Arnold tongues are displayed as functions of time. They clearly depict the mode locking behavior of the second cumulant K2 (n) implying D = 0. Finally, regions in the parameter plane can be classified with respect to the sign of the current: Fig. 3.6 depicts the current J as a function of the slope a for three different values of the bias b approaching zero. Firstly, one clearly sees the highly irregular structure of the current which becomes wilder and wilder the closer b comes to zero. Again, these irregularities, which persist on finer and finer scales, are suggestive of the current being a fractal function of the two parameters a and b almost everywhere (see

4:1

World Scientific Book - 9in x 6in

46

klages˙book

Deterministic drift-diffusion

(a)

4

(c)

3

3

2.5

2.5

ρ (x) n

ρ (x) n

4 3.5

3.5

2 1.5

2 1.5

1

1

0.5

0.5

0 12.5

13

13.5

14

14.5

0 12.5

15

13

13.5

x

(b)

4

(d)

14.5

15

14

14.5

15

4 3.5 3

2.5

2.5

ρ (x) n

3

2 1.5

2 1.5

1

1

0.5

0.5

0 12.5

14

x

3.5

ρ (x) n

May 8, 2007

13

13.5

14

x

14.5

15

0 12.5

13

13.5

x

Fig. 3.4 Probability density ρn (x) at a = 1.125 and b = 0.333 as obtained from the approximate iteration method described in Section 3.2 after (a) n = 40, (b) n = 41, (c) n = 42, (d) n = 43 iterations starting from a uniform density on the unit interval. The order of the partition is 35 parts per unit interval determining the accuracy of the results. Since the parameter pair does not correspond to a Markov partition, the exact solution should have infinitely many “steps”, part of which can be seen in the figure. One furthermore observes that the probability distribution has developed a periodic pattern repeating itself after three time steps, apart from a translation over a distance of one. Hence, the current J is mode locked to the value 1/3 and the diffusion coefficient vanishes.

footnote 4) in the parameter plane. The region of a < 2 with J = 0 is the localization region indicated in Fig.3.1. Secondly, for b → 0, and on a line with constant a there appear more and more regions where the current becomes negative. Indeed, for small enough b the regions with negative currents form an intriguing fractal structure in the parameter plane as is shown in Fig.3.7 [Gro02]. More recently, the fractality and the negativity of the current were corroborated in Ref. [Gil04] by studying a related model. Negative currents in similar one-dimensional maps have meanwhile also been reported in Refs. [Art06; SG06]. In Fig. 3.8 such regions are explored in more detail for three fixed values

4:1

World Scientific Book - 9in x 6in

klages˙book

47

3.3. The phase diagram

0.115 0.11 0.105 K2(n)

May 8, 2007

a=1.125, b=0.333 a=1.05, b=0.2

0.1 0.095 0.09 0.085 20

40

60

80 n

100

120

140

Fig. 3.5 Time-dependence of the second cumulant K2 (n) defined by Eq.(3.9) for two specific pairs of parameter values situated in different Arnold tongues. In both cases the initial density was uniform on the unit interval. Both curves appear to remain bounded for all times implying a vanishing diffusion coefficient, and to become periodic (with periods three and five, respectively) as n → ∞, a sure sign of the phenomenon of mode locking.

of a and by varying b in the region 10−8 ≤ b ≤ 1. For all three values of a the current J divided by b shows a highly irregular behavior as a function of b which persists on finer and finer scales and which again points to a fractal structure. In particular, for small b the current appears to remain negative in all three cases. Hence, for these values of a there is no indication of the existence of a regime of linear response for infinitesimally small values of b. On the other hand, for extremely large values of b the current J obviously becomes linear in b. Comparing the plots of the current J as a function of b for a = 3.0001 and for a = 3.01, the sensitivity of J(a, b) as a function of a becomes apparent. Note in particular that J(3, b)/b = 1 exactly for all b, showing that the case a = 3 is in a sense “integrable”: in this case the response of J to b is linear on all scales. More details concerning the origin of these negative currents are shown in Fig. 3.9. The main plot depicts the probability on L coupled boxes starting from a uniform density on a unit interval after iteration of 40 time steps. The parameter pair (a, b) used in this figure yields a Markov partition for the map. Note that, on a coarse scale, the distribution resembles a

4:1

World Scientific Book - 9in x 6in

48

klages˙book

Deterministic drift-diffusion

0.18

(a)

0.16 0.14

J(a,b)

0.12 0.1 0.08 0.06 0.04 0.02 1

2

3

4

5

6

7

8

5

6

7

8

5

6

7

8

a 0.03

(b)

0.02

J(a,b)

0.01 0 -0.01 -0.02 -0.03 1

2

3

4 a

0.004

(c) 0.002 0

J(a,b)

May 8, 2007

-0.002 -0.004 -0.006 -0.008 1

2

3

4 a

Fig. 3.6 The current J(a, b) as a function of the slope a at a fixed bias b for (a) b = 0.1, (b) b = 0.01, (c) b = 0.001. One notices in particular that: 1. the dependence of the current J(a, b) on the slope a appears to have a fractal structure on the entire a-axis; 2. contrary to intuition, on considerable portions of the a-axis J(a, b) is negative; and 3. these two types of irregularities of the current J(a, b) become stronger the closer b → 0.

Gaussian, whereas on a fine scale a periodic fine structure exists. This is a very general feature of this kind of dynamical model not particular to the present one [Kla96b]. The center of the distribution is moving to the left indicating a current opposite to the bias. This can be seen more clearly in the magnification to the upper right which shows the invariant density for the circle map m. That the invariant density in such maps has a step-like structure has already been noticed by many authors, see, e.g., Refs. [D¨ or85;

4:1

World Scientific Book - 9in x 6in

klages˙book

49

3.4. Simple maps as deterministic ratchets

0.08 (a)

b

0.06 0.04 0.02 0

2

3

0.08

4

5 a

0.03

(b)

7

6

0.06

8

0.038

(c)

0.06 0.04

0.02 0.04 3.2

0.02 0

0.034

b

b

May 8, 2007

3

3.2

3.4

0.02

3.3

3.6

3.8

5.17 5.19 5.21 4

0

5

5.2

a

5.4

5.6

5.8

6

a

Fig. 3.7 Various close-ups of parts of the parameter plane where ‘macroscopic negative response’ occurs, i.e. where J(a, b)/b < 0. In part (a), regions with J(a, b)/b < 0 are coloured grey, and the boundaries thereof, which is where J changes sign, are in black. Parts (b) and (c) are further enlargements displaying now only these boundaries. Each contains an inset with a further enlargement, making the self-similar structures present in these boundary curves visible.

Klu90; Ers93]. As sketched in the figure, the fact that b > 0 would suggest that, on average, more particles would leave the box to the right. In contrast to that, the probability density has evolved in a way such that after a sufficiently long time, independently of the initial condition there is a higher probability for a particle to leave the box to the left than to the right. This result is typical for parameter values yielding negative currents. 3.4

Simple maps as deterministic ratchets

A ratchet may be defined as a system that is able to transport particles in a periodic structure with nonzero average velocity, although on

4:1

World Scientific Book - 9in x 6in

50

klages˙book

Deterministic drift-diffusion

0 -2 -4 J(a,b)/b

May 8, 2007

-6 -8 -10 slope = 3.0001 slope = 3.01 slope = 3.8

-12 -14 -16 -8

-7

-6

-5

-4

-3

-2

-1

0

log b Fig. 3.8 Plot of the current J(a, b) divided by the bias b as a function of log b for three different values of the slope a. These three graphs seem to indicate the following: 1. J(a, b)/b, as a function of b, is nowhere differentiable. The exact expressions of the twisted eigenstate method allow to go further and make the stronger statement, that this function “has fractal behavior”, which refers to a scaling behavior in the limit of infinitely small distances; 2. J(a, b)/b is sometimes, as, e.g., for these parameter values, negative; 3. for the three a-values chosen here, the limit limb→0 J(a, b)/b does not exist; in other words, that the “response J of the system to the bias b” is not linear in b for small b. This is confirmed as well by the exact results quoted.

average no macroscopic force is acting on it [H¨ an96]. This implies that there must exist a microscopic symmetry breaking at some point, instead of a force such as, e.g., an external electric field that induces a current. The name “ratchet” traces back to thought experiments by Smoluchowski and Feynman suggesting a mechanism of how to convert microscopic fluctuations of unbiased Brownian motion into work (see Ref. [Rei02] and further references therein). Synonymously one speaks of these systems as Brownian motors, Brownian rectifiers or Brownian machines [Ast97; Ast02]. Most often such devices are modeled by an overdamped Langevin equation6 of the type [Bar94] x˙ = −

d V (x) + ξ(t) + F (t) dx

.

(3.34)

6 For a more detailed introduction to the concept of the Langevin equation see Section 10.2.

4:1

World Scientific Book - 9in x 6in

klages˙book

51

3.4. Simple maps as deterministic ratchets

0.14 1.6 1.5

0.12 ∼ ρ (x)

1.4

0.1

1.3 1.2 1.1 1 0.9

ρt(x)

May 8, 2007

0.08

0.8 -0.4

-0.2

0

0.2

0.4

x

0.06 0.04 0.02 0 -20

-15

-10

-5

0

5

10

15

20

x

Fig. 3.9 Probability density ρn (x) at the Markov partition point a ' 3.112987 and b ' 0.021281 after n = 40 iterations, as computed by means of the renormalized iteration method described in Section 3.2. The order of the Markov partition is here 10 parts per unit interval. The inset in the upper right hand corner shows the invariant probability density on the circle of our projected map fa,b denoted as ρ˜(x). The corresponding inset to the left illustrates how it can be understood that the escape rate out of any given box on the infinite line is larger towards the left than towards the right, notwithstanding the fact that the bias b is positive. This follows from the fact that the equilibrium density on the left hand side of the basic unit interval is larger than that on the right hand side.

Here the dot holds for the time derivative and x is the position of a point particle on the real line. Note that there is no acceleration term, and that the friction coefficient is usually scaled away. The right hand side is composed of three fundamental ingredients defining a ratchet: (1) V (x + 1) = V (x) denotes an asymmetric potential that is periodically continued on the line; for an example see Fig. 3.10. (2) ξ(t) stands for thermal noise, which is usually modeled by Gaussian white noise. (3) F (t) is an external force that is periodic in time. Simple examples of ratchets defined by Eq. (3.34) are rocking ratchets with F (t) = A sin(ωt) and flashing ratchets where F (t) = ±F0 , F0 being a constant, over periodic intervals of time t; for other types of ratchets see, e.g., Ref. [Rei02]. What makes these systems so interesting is that they can convert random fluctuations into work. This ratchet effect yields no violation of the second law of thermodynamics, since Eq. (3.34) contains

May 8, 2007

4:1

World Scientific Book - 9in x 6in

52

klages˙book

Deterministic drift-diffusion

V(x)

x Fig. 3.10 Example of a ratchet potential V (x) as used in Eq. (3.34). Here we display a simple sawtooth function periodically continued on the real line.

an external force which, however, is zero on average. The rectification of the random fluctuations is thus due to a microscopic symmetry breaking. A typical property of ratchets is the occurrence of current reversals, that is, the current may become positive or negative with respect to the microscopic bias by varying control parameters [H¨ an96; Rei02]. This phenomenon reminds one very much of the existence of positive and negative currents of the model discussed in the previous sections. It turns out that stochasticity is not a crucial ingredient of the ratchet mechanism. This was shown by analyzing the inertia ratchet [Jun96] x¨ + αx˙ = −

d V (x) + A sin(ωt) dx

,

(3.35)

where the stochastic force has been eliminated, an acceleration term has been added, and α respresents a friction coefficient. According to the Poincar´e-Bendixson theorem, this dynamical system may exhibit chaotic behavior [Ott93; All97]. In a way, the stochastic forcing is thus replaced by deterministic chaos. As was shown in Refs. [Jun96; Mat00; Bar00a], this equation provides a suitable deterministic modeling of a ratchet mechanism. We proceed by extracting a time-discrete mapping from Eq. (3.35) capturing essential details of the dynamics. We first exploit the periodicity of the forcing by stroboscopically sampling the dynamics at discrete time steps ωt = π/(4n + 1)/2 , n ∈ N0 . Similarly to the time discretization of the kicked rotor [Ott93], this yields a two-dimensional map that bears some similarity to a dissipative standard map. We then need to assume that the velocity of the particle is approximately constant. This is a non-trivial assumption, which should be fulfilled particularly in the case of sufficiently

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

53

3.4. Simple maps as deterministic ratchets

strong damping. For the position variable we thus obtain the equation d ˜ xn+1 = xn − V (x) |x=xn + A˜ , (3.36) dx where n denotes the discrete time steps and α has been incorporated into the variables labeled with a tilde. This equation is obviously of the form of the time-discrete one-dimensional map xn+1 = xn + m(xn )

.

(3.37)

Matching this equation to our previous one-dimensional equation of motion Eq. (2.1) yields m(x) = M (x) − x and m(x + 1) = m(x). Working out the details for our specific example M (x) = ax + b , |x| < 0.5, see Eq. (3.4), we obtain m(x) = (a − 1)x + b and d ˜ V (xn ) + b2 (3.38) xn+1 = xn − dx corresponding to Eq. (3.36), where V˜ (x) := −(a − 1)x2 /2 − b1 x. The splitting b = b1 + b2 is somewhat arbitrary in this case. We thus conclude that the deterministic drift-diffusion map discussed in the previous sections may be understood as a time-discrete deterministic ratchet. This argument was referred to in Refs. [Kla98; Gro02; Kla04a]. It is analogous to the reduction of a driven nonlinear pendulum equation onto a sine circle map reported in Refs. [Kog83; Boh84; Bak85], see also Chapter 6. Recently related crosslinks between ratchets and time-discrete maps were suggested in Refs. [Art06; SG06]. Brownian motors constitute a very active field of research with applications in many different fields of the natural sciences. An important example are molecular motors in the cytoskeleton of biological cells [How01]. The cytoskeleton consists, among other things, of a network of biopolymers, which are long filaments equipped with periodic structures similar to ratchet potentials. Different motor proteins have been identified moving along specific filaments fuelled by chemical reactions. They can perform various tasks such as transporting chemical substances within the cell or linking different filaments with each other. It has been suggested to model the dynamics of such motor proteins by a two-state system similar to a flashing ratchet [J¨ ul97]. On a larger scale the action of molecular motors is responsible for a wide variety of biological processes such as muscular contraction, cell division, cell motility, hearing and many others [Ast97; How97; J¨ ul06]. Further well-known examples of ratchets are optical lattices [Fau95] and electronic SQUIDs [Wei00], which have both been studied experimentally.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

54

Deterministic drift-diffusion

In the field of granular matter ratchets have been proposed for particle separation methods exploiting the phenomenon of current reversals [Far99; Far02]. Plenty of further applications of the ratchet effect are described in reviews [H¨ an96; J¨ ul97; Ast97; How97; Rei02; Ast02]. We believe that simple maps with broken symmetry, such as the model studied above, yield a nice playground for learning more about the origin of current reversals and intricate parameter dependencies of the current, as they are typically observed in ratchets. 3.5

∗

Summary

(1) We have amended the simple map studied in the previous chapter by introducing a second parameter that breaks the symmetry of the system. We thus arrived at a simple deterministic model for a drift-diffusion process, which we defined in mathematical terms. (2) We outlined two basic methods enabling one to numerically exactly calculate the drift and diffusion coefficients of this model. The mathematically rigorous approach proceeds from the generating function of the system by eventually expressing the transport coefficients in form of simple recursion relations. The more physical approach generalizes the idea of using transition matrices leading to iterative solutions for the probability density as determined by the Frobenius-Perron equation. Advantages and drawbacks of both methods were briefly discussed. (3) We applied these methods in order to construct and analyze the phase diagram of the model in the parameter plane. Both the drift and the diffusion coefficient are typically fractal functions of the two control parameters. Depending on the choice of parameters the model exhibits different dynamical regimes such as phase locking and parameter regions with negative currents. The origin of the negative currents is related to strong microscopic backscattering in any unit cell of the periodically continued map. Typically there is no linear response. (4) Finally, we outlined a relation between this simple drift-diffusion model and deterministic ratchets. The negative currents observed in this map may thus be understood in relation to current reversals as they are wellknown for Brownian motors. We concluded with an outlook towards recent experimental applications of the concept of Brownian rectifiers.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Chapter 4

Deterministic reaction-diffusion

We proceed with our introduction to elementary deterministic transport processes by studying the consequences of deterministic chaos for a simple model of diffusion-controlled chemical reaction. In Section 4.1 we motivate the use of two-dimensional reactive-diffusive multibaker maps starting from a more physically realistic model, which is a reactive-diffusive particle billiard. In Section 4.2 we review the diffusive properties of multibakers. We then demonstrate that under parametric variation the diffusion coefficient of such a model exhibits a self similar-like structure analogous to the ones discussed in the previous two chapters. In Section 4.3 we focus on the reactive properties of our model. Particularly we derive the dispersion relation of the chemiodynamic modes, explicitly construct the phase space distribution of these modes and define the reaction rate by comparison with phenomenology. We then show that the parameter-dependent reaction rate behaves in a highly irregular manner displaying a complicated scenario of nonequilibrium phase transitions. The work presented in this chapter originated in collaboration with P.Gaspard [Gas98c]. 4.1

A reactive-diffusive multibaker map

Matter is most often the stage of reactions which evolve on a reactive time scale intermediate between the long time scale of hydrodynamic transport phenomena and the short time scale of microscopic chaos. This chaos is generated by the collisions between the atoms and molecules of the fluid beyond a temporal horizon caused by the Lyapunov instability of motion. Under nonequilibrium conditions, long-time trajectories organize themselves in phase space to form fractal structures and unusual invariant or condi55

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

56

klages˙book

Deterministic reaction-diffusion

(a)

(b)

Fig. 4.1 Reactive-diffusive Lorentz gas defined (a) on a regular and (b) on a random lattice. The disks represent fixed obstacles for a point particle moving between these scatterers and colliding elastically with them. The dark shaded disks are “catalysts” at which the particle changes its state.

tionally invariant measures, the consequences of which have just started to be explored. Within this perspective, in this chapter we study simple models of reaction-diffusion processes in order to confront the phenomenology with the new approach based on deterministic chaos. We shall focus on the simplest reaction-diffusion process with a linear chemical reaction law, A B,

(4.1)

which already provides a nontrivial dynamics. As a vehicle of our study we shall use multibaker models, which are deterministic versions of discrete Markov processes. The models we propose are spatially extended generalizations of a baker-type model of isomerization previously studied by Elskens and Kapral [Els85]. 4.1.1

Deterministic models of reaction-diffusion

In order to motivate the introduction of multibaker maps, let us first consider a reactive-diffusive particle billiard in which a point particle undergoes elastic collisions on hard disks which are fixed in the plane. Fig. 4.1 depicts the geometry of a Lorentz gas, where the disks form a random or a regular configuration.1 A fraction of the disks are supposed to be catalysts where 1A

more detailed introduction to the Lorentz gas model is provided in Section 10.4.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

57

4.1. A reactive-diffusive multibaker map

the point particle changes its state, or color, from A to B or vice versa at the instant of the collision. The mass of the particle is assumed to be the same in both states A and B. The phase space coordinates of each particle are given by its position, its velocity, and its color (x, y, vx , vy , c) with c ∈ {A, B}. Since energy is conserved at the elastic q collisions, the magnitude of the velocity is a constant of motion, v =

vx2 + vy2 , so that

the coordinates reduce to (x, y, ϕ, c) = (X, c), where ϕ = arctan(vy /vx ) is the angle between the velocity and the x-axis. The motion induces a time evolution of the phase space probability densities, or concentrations, for each color, f (X, A) f (X) = . (4.2) f (X, B)

The mean phase space density, defined by the average of the concentrations, does not distinguish between the colors. Thus, we may expect that 1 f˜(X) = [f (X, A) + f (X, B)] 2

(4.3)

evolves in time exactly as in the non-reactive Lorentz gas, which was studied elsewhere [Bun80; Bun81; Che94; Gas96]. The dynamics of reaction should appear in the difference between the concentrations g(X) = f (X, A) − f (X, B) ,

(4.4)

which is expected to follow a macroscopic relaxation towards zero if there is equipartition of particles between both colors. It has been explained elsewhere that the flow dynamics of the Lorentz gas can be reduced to a Birkhoff mapping from collision to collision [Gas96; Gas98a; T´el00]. Each collision can be represented by two variables, the angle θ giving the position of impact on the perimeter of the disk as (x = cos θ, y = sin θ) and the angle φ between the velocity after the collision and the normal at impact. The sine of the velocity angle $ = sin φ together with the position angle θ are the so-called Birkhoff coordinates in which the mapping is area-preserving.2 All the collision events with the disk of label l are thus represented by the rectangle Rl = (θ, $, l) : 0 ≤ θ < 2π , −1 ≤ $ ≤ +1 . (4.5) The dynamics of collisions can therefore be written as the Birkhoff mapping (θn+1 , $n+1 , ln+1 , cn+1 ) = Φ(θn , $n , ln , cn ) ,

(4.6)

May 8, 2007

4:1

World Scientific Book - 9in x 6in

58

klages˙book

Deterministic reaction-diffusion

l=

...

-1

0

1

2

3

4

5

6

...

L=3

...

...

L=4

...

...

L=5

...

...

L=6

...

... ...

Fig. 4.2 Caricature of a reactive-diffusive multibaker map mimicking the periodic Lorentz gas of Fig. 4.1 (a). The index l in the first row labels the single cells of a multibaker. The rows below display different configurations, where L indicates the position of a catalyst shaded in dark grey and periodically continued on a one-dimensional lattice.

which is known to be area-preserving, time-reversal symmetric and of hyperbolic character. A caricature of this mapping is provided by a multibaker model. In Fig. 4.2 we suppose that the rectangular domains Eq. (4.5) representing the disks are replaced by squares, Sl = (x, y, l) : 0 ≤ x ≤ 1 , 0 ≤ y ≤ 1 , (4.7)

where l ∈ Z is the label of the square. Each square of the multibaker model corresponds to a disk of the Lorentz gas. Now the collision dynamics is simplified in replacing the complicated Birkhoff map Eq. (4.6) by a baker-type map. Baker maps are standard models of chaotic dynamical systems with horizontal stretching of the square by a factor of two followed by cutting the elongated square into two [Sch89; Ott93; Gas98a; Dor99; 2 We remark that this fact is a mathematical reformulation of Lambert’s cosine law, which has been discovered earlier in physical experiments (J.H. Lambert, 1760). It states that the brightness of a diffusively radiating flat surface is proportional to the cosine of the angle formed by the line of sight and the normal to the surface [Car85].

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

59

4.1. A reactive-diffusive multibaker map

T´el06].3 The collisions from disk to disk are replaced by jumps of the particle from square to square according to the transition rule l → l − 1 if x ≤ 1/2 and l → l + 1 if x > 1/2 between next-neighbouring squares. The squares are arranged such that they form a one-dimensional chain. One out of L squares is assumed to be a catalyst where the color changes from c = A (resp. B) to its complement c¯ = B (resp. A). The map of the model is thus   2x, y2 , l − 1, c0 , 0 ≤ x ≤ 21 ,   (4.8) Φ(x, y, l, c) =    2x − 1, y+1 , l + 1, c0 , 1 < x ≤ 1 , 2

0

2

0

where c = c¯ if l = 0, ±L, ±2L, ... and c = c otherwise. This map is areapreserving, time-reversal symmetric and chaotic with a positive Lyapunov exponent λ+ = ln 2 and a negative one λ− = − ln 2 such as the purely diffusive multibaker map [Gas92a; Gas93; Gas95a; Tas94; Tas95; Gas98a; Dor99]. If the reactive sites are diluted in the system the reaction is controlled by diffusion, as applies to our model for L > 1. This important case of chemical reactions has been much studied in the literature since Smoluchowski’s seminal work [Wei86; Ber88]. It is known that the macroscopic reaction rate is determined by the time taken by particles to diffuse towards the reactive site. A crossover occurs at dimension two which is the Hausdorff dimension of a Brownian path. Accordingly, the flux of reactants toward a catalyst is sensitive to the presence of the next-neighbouring catalysts in less than two dimensions, but not in systems of dimensions higher than two. In particular, in a one-dimensional system the reaction rate should behave as D (4.9) κr ∼ 2 , L where D is the diffusion coefficient and L is the distance between the reactive sites or catalysts. The main goal of this chapter is to investigate the dynamical properties of our reaction-diffusion model in order to know whether this expected macroscopic behavior is confirmed from the microscopic dynamics or not. We shall also consider a parametric variation of this model exhibiting a more complicated dynamics. This is due to the extra dependency on a 3 For further details concerning baker maps plus some historical background see Section 13.1.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

60

klages˙book

Deterministic reaction-diffusion

shift parameter, which is introduced when the half squares are glued back into the chain. When this continuous parameter varies it induces topological changes in the trajectory dynamics, which reflects topological changes induced by varying the disk radius in the Lorentz gas as will be further discussed in Chapters 7 and 8. This parametric extension of the multibaker model shows that the diffusion coefficient as well as the reaction rate may vary in a highly irregular manner as a function of a control parameter. This is an important consequence of deterministic chaos which already appears on the level of one-dimensional maps, see the previous Chapters 2 and 3.

4.1.2

The Frobenius-Perron operator

Since Boltzmann’s work it is well known that transport and reaction-rate processes should be conceived in a statistical sense because the individual trajectories are affected by the famous Poincar´e recurrences. We therefore consider the time evolution of statistical ensembles of trajectories as represented by the probability densities Eq. (4.2). They evolve in time according to the Frobenius-Perron equation, see also Eqs. (2.7),(3.5), in general terms reading h i fn+1 (x, y, l, c) = fn Φ−1 (x, y, l, c) =: (Pˆ fn )(x, y, l, c) , n ∈ Z . (4.10) We choose quasiperiodic boundary conditions by assuming that the solution of the Frobenius-Perron equation is quasiperiodic on the chain with a wavenumber k. Moreover, we suppose that the solution decays exponentially with a decaying factor χ = exp s where |χ| ≤ 1 or Re s ≤ 0, fn (x, y, l, c) ∼ χn exp(ikl) .

(4.11)

The decay rate s is calculated by solving the eigenvalue problem of the Frobenius-Perron operator. We note that this operator is in general nonunitary so that root vectors associated with possible Jordan-block structures may exist beyond the eigenvectors. We shall focus here on the eigenvectors because they control the slowest decay on the longest time scales [Gas93; Gas95a]. For quasiperiodic solutions, the Frobenius-Perron operator reduces ˆ k which depends on the to the following Frobenius-Perron operator Q wavenumber k and acts on functions which are defined only in L successive

May 8, 2007

4:1

World Scientific Book - 9in x 6in

4.1. A reactive-diffusive multibaker map

klages˙book

61

squares of the chain,  1  f (x, y, 0, c) = θ − y fn x2 , 2y, 1, c +  n+1 2     1 −ikL  + e θ y − fn x+1  2 2 , 2y − 1, L − 1, c ,      fn+1 (x, y, 1, c) = θ 21 − y fn x2 , 2y, 2, c +      1  + θ y − ¯ , fn x+1  2 2 , 2y − 1, 0, c      fn+1 (x, y, 2, c) = θ 21 − y fn x2 , 2y, 3, c +    ˆ k := + θ y − 12 fn x+1 Q 2 , 2y − 1, 1, c ,    ..   .     1   − y fn x2 , 2y, L − 1, c + f (x, y, L − 2, c) = θ n+1  2      , 2y − 1, L − 3, c , + θ y − 21 fn x+1  2    1 ikL   − y fn x2 , 2y, 0, c¯ + f (x, y, L − 1, c) = e θ n+1  2     + θ y − 21 fn x+1 2 , 2y − 1, L − 2, c . (4.12) We notice that there is a reaction, i.e. a change of color, for particles passing the cell l = 0 so that a concentration with l = 0 and c¯ appears in the second and in the last line. We also notice that there is a factor exp(−ikL) in the first line for the particle coming from the previous segment of length L in the infinite chain, where the concentration functions are multiplied by the factor exp(−ikL). On the other hand, there is a factor exp(ikL) in the last line for the particle coming from the next segment where the concentration functions are multiplied by exp(ikL). Otherwise this Frobenius-Perron operator is the same as in the infinite dyadic multibaker model studied in Refs. [Gas92a; Gas93; Gas95a]. As we discussed in the previous section, the presence of two chemical ˆ k acts components c = A or B implies that the Frobenius-Perron operator Q on 2L functions which can be linearly combined to separate the functional space in two subspaces, on which two decoupled Frobenius-Perron operators would act. The first subspace is defined by Eq. (4.3) where the FrobeniusPerron operator reduces to the diffusive Frobenius-Perron operator of the multibaker map. The second subspace is defined by Eq. (4.4) which gives a different evolution operator of reactive type. Diffusive properties are studied in the next Section 4.2 while reactive properties will be discussed in Section 4.3.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

62

4.2 4.2.1

klages˙book

Deterministic reaction-diffusion

Diffusive dynamics +

Diffusive modes of the dyadic multibaker

In this subsection we consider the diffusive dynamics of the dyadic multibaker model Eq. (4.8) with quasiperiodic boundary conditions. The subspace of diffusion is defined by the mean density of Eq. (4.3). For hydrodynamic modes of wavenumber k we thus write " # ν 1 ˜ l ηn (x, y) , fn (x, y, l) = fn (x, y, l, A)+fn (x, y, l, B) =: exp i k+2π 2 L (4.13) and the new function obeys the simpler evolution equation 1 x (D) +i(k+2πν/L) ˆ − y ηn , 2y + ηn+1 (x, y) = (Qk ηn )(x, y) = e θ 2 2 1 x+1 + e−i(k+2πν/L) θ y − ηn , 2y − 1 , (4.14) 2 2

which is the Frobenius-Perron equation for the dyadic multibaker map based on quasiperiodic boundary conditions [Gas93]. The respective Frobenius-Perron operator has been analyzed in detail elsewhere [Gas93; Gas95a]. Its decay rates are 2πν (4.15) smν (k) = ln χmν (k) = − m ln 2 + ln cos k + L with m = 0, 1, 2, 3, ..., ν = 0, 1, 2, ..., L−1, and with a degeneracy of (m+1). The eigenvectors (D)

ˆ (Q k

ψmν )(x, y; k) = esmν (k) ψmν (x, y; k) ,

(4.16)

and some root vectors have been constructed in terms of the cumulative functions Z x Z y dx0 dy 0 ψ0ν (x0 , y 0 ; k) . (4.17) F0ν (x, y; k) = 0

0

For small enough wavenumbers k, these cumulative functions are continuous functions which are products of a monomial in x with a nondifferentiable de Rham function [dR57] in y. Accordingly, the eigenvectors ψ0ν are complex singular measures for small enough k. We observe that the decay rate with m = 0 and ν = 0 vanishes quadratically as k → 0 in agreement with the expected diffusive behavior, s00 (k) = ln cos k = −

k2 k4 − ··· , 2 12

(4.18)

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

63

4.2. Diffusive dynamics

which shows that the diffusion coefficient is D = 1/2. The nonequilibrium steady states corresponding to a uniform density gradient g for the mean density of Eq. (4.3) have also been constructed [Tas94; Tas95]. It has been shown that the nonequilibrium steady state of the infinite system corresponds to a singular invariant measure represented by a continuous but nondifferentiable cumulative function Z y Z x 0 dy 0 f˜st. st. (x0 , y 0 , l) = glxy + gxT (y) , (4.19) Fst. st. (x, y, l) = dx 0

0

where T (y) is the Takagi function [Tak03] obtained as a solution of the iteration [Hat84]  1  0 ≤ y ≤ 21 ,   2 T (2y) + y , (4.20) T (y) =   1  1 T (2y − 1) − y + 1 , 2 2
The Takagi function is nondifferentiable because its formal derivative is infinite almost everywhere. It is given by a Lebowitz-McLennan type of formula [Leb59; Mac59; Gas97a] ∞ X dT = j(mn (y)) , dy n=0

(4.21)

where m(y) = 2y mod 1, and j(y) = ±1 if y < 1/2 or y > 1/2, respectively is the jump velocity. The singular character of the diffusive steady state turns out to be a general feature in finite-dimensional deterministic chaos of large spatial extension [Tas94; Tas95; Gas96; Tas98; Gas98a; Dor99]. This singular character of the steady state measure plays a fundamental role in the explanation of entropy production in irreversible thermodynamics [Bre96; T´el96; Vol97; Gas97b; Bre98; Vol98; Gil99a; Tas99; Tas00; Gil00a; Gil00b; T´el00; Vol00; M´ at01; Dor02; Vol02; Vol04]. In the case of multibaker maps the expected entropy production can be derived from the Takagi function [Gas97b]. The presence of this singularity solves the famous paradox of the constancy of the Gibbs entropy, which can be set up when we do not recognize that the out-of-equilibrium invariant measure is very different from the equilibrium one on fine scales in phase space. The out-of-equilibrium invariant measure becomes singular if the nonequilibrium constraints are imposed at distances larger than several mean free paths. This is probably a paradoxical aspect of the local equilibrium hypothesis that a nonequilibrium system appears in local equilibrium on the largest scales of phase space although intrinsic correlations exist on finer

May 8, 2007

4:1

64

World Scientific Book - 9in x 6in

Deterministic reaction-diffusion

scales which are due to the chaotic dynamics. The singular character of the invariant measure explains that there is an entropy production in large nonequilibrium systems where the chaotic dynamics removes the signature of determinism down to extremely fine scales in phase space, see also Sections 13.2 and 13.4. 4.2.2

The parameter-dependent diffusion coefficient

In order to study transport under variation of control parameters in multibaker models like Eq. (4.8), again methods to cope with transport in one-dimensional maps are crucial. As we will show for a two-dimensional parameter-dependent multibaker which we introduce and discuss towards the end of this subsection, one-dimensional maps govern the dynamics of multibakers along the x-axis. We thus start our analysis by outlining yet another method of how to calculate and to understand fractal diffusion coefficients in one-dimensional maps. This method will become very important for most of the following chapters in Part 1. We will then show that both the diffusive and the reactive transport properties of our parameterdependent multibaker display fractal structures. We thus emphasize the striking analogies between the diffusive and the reactive properties in regard to their parametric sensitivity. 4.2.2.1 Fractal forms in the Taylor-Green-Kubo formula for diffusion For a moment let us turn back again to the model introduced in Section 2.1 denoted as map L, see Eqs. (2.3) to (2.6) and the illustration in Fig. 2.4. We briefly elaborate again on the diffusive properties of this map starting from the Taylor-Green-Kubo formula for diffusion4 [Fuj82; Gas92a; Kla96b; Gas98c; Gas98a; Dor99; Kla02d] * + ∞ X 1 2 n j (x) , (4.22) D(a) = ja (x) ja (ma (x)) − 2 a n=0

R1 where the average h. . .i := 0 dx %a (x) . . . has to be taken over the invariant probability density %a (x) on the unit interval. ja (xn ) := [xn+1 ] − [xn ] = [Ma (xn )], with the square brackets denoting the largest integer less than the argument, gives the integer number of boxes a particle has traversed

4 We remark that the time-continuous version of Eq. (4.22), which is often attributed to Green [Gre54] and Kubo [Kub57] only, was already derived by Taylor in Ref. [Tay21].

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

65

4.2. Diffusive dynamics

after one iteration starting at position x = x0 and is thus the parameterdependent extension of the jump velocity introduced in Eq. (4.21) above. ma (x) := Ma (x) mod 1 defines the map restricted onto the unit interval, that is, the associated circle map. Thus, the complete microscopic dynamics is divided into two parts in Eq. (4.22): the intra-cell dynamics, that is, the dynamics within a single box, which is represented by the invariant probability density, and the inter-cell dynamics given by the sum of the jump velocities, which contains the memory of the particles travelling between the single boxes of the chain. For computing the diffusion coefficient both parts can be treated separately. The invariant probability density is obtained by solving the Frobenius-Perron equation Eq. (4.10), respectively Eq. (2.7), for the map restricted to the unit interval. As was outlined particularly in Chapter 2, for this purpose Eq. (2.7) may be written as the matrix equation, see Eq. (2.12), by transforming the Frobenius-Perron operator into a transition matrix [Kla95; Gas92a; Kla96b; Bec93; Kla99a]. For piecewise linear maps exact transition matrices can be constructed whenever a Markov partition exists. For the present model this is the case for a dense set of parameter values a on the real line. The invariant probability density can then be calculated either by solving the eigenvalue problem of the transition matrix, which in simple cases can be performed analytically, or by solving the Frobenius-Perron equation by iterating the transition matrices numerically; see also the discussion in Section 3.2. In Fig. 4.3 (a) and (b), typical invariant probability densities are shown for two values of the slope. They are step functions on the unit interval, where the regions of the functions being piecewise constant correspond to the single cells of respective Markov partitions. For piecewise linear maps, the invariant probability densities should always be step functions, although for arbitrary parameter values they may consist of infinitely many steps [Kla96b; D¨ or85; Ers93; Klu90], cf. also Fig. 3.4 and the inset of Fig. 3.9. The sum of jump velocities, as the second ingredient of the Taylor-GreenKubo formula Eq. (4.22), gives the integer displacement of a particle after n iterations starting at initial position x. Since the system is chaotic, this function is highly irregular in x. Detailed properties of this function will be exploited in Section 7.1. In order to deal with this quantity, it is convenient to define functions Ta (x) via ∞ X dTa := ja [mna (x)] , dx n=0

(4.23)

4:1

World Scientific Book - 9in x 6in

66

klages˙book

Deterministic reaction-diffusion

1.25

0.25

(c)

1.15

0.2

1.05

0.15

Ta (x)

ρ (x) a

(a)

0.95

0.1

0.85

0.05

0.75 0

0.2

0.4

0.6

0.8

0

1

0

0.2

0.4

x

(b)

0.6

0.8

1

x

1.15

(d)

1.1

0.25 0.2

1.05 0.15

1

T a(x)

ρ (x) a

May 8, 2007

0.95

0.1

0.9 0.05 0.85 0

0.8 0

0.2

0.4

0.6 x

0.8

1

0

0.2

0.4

0.6

0.8

1

x

Fig. 4.3 (a), (b) Invariant probability densities %a (x) on the unit interval for the map of Fig. 2.3 mod 1. The slope is a ' 2.5004 for (a) and a ' 3.49997 for (b). (c), (d) Generalized Takagi functions Ta (x) for the same map at a = 3 in (c) and at a = 4 in (d).

which is a parametric extension of Eq. (4.21). Note that the cumulative functions Ta (x) give the integral of the displacement of particles which start in a certain subinterval. They behave much more regularly in x than the sums of jumps. Employing Ta (x) = limt→∞ Ta,n (x), it can be shown that these functions are obtained in terms of the recursion relation [Kla96b] 1 Ta,n (x) = ta (x) + Ta,n−1 (ma (x)) (4.24) a with ta (x) being determined by dta /dx := ja (x) and by requiring that Ta (0) = Ta (1) = 0. Ta (x) can be computed by iterating Eq. (4.24) numerically. For two special values of the slope the results are plotted in Fig. 4.3 (c) and (d). The functions Ta (x) are self-similar on the unit interval and scale with the slope a. Eq. (4.24) is a parameter-dependent generalization of Eq. (4.20). Therefore, functions like Ta (x) may be denoted generalized Takagi functions. Along these lines, the parameter-dependent diffusion coefficient can be calculated from Eq. (4.22). The numerically exact result was already shown in Fig. 2.5. Alternatively to the explanation provided in

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

67

4.2. Diffusive dynamics

Section 2.2 however, the fractal character of D(a) can now be understood by analyzing the Taylor-Green-Kubo formula Eq. (4.22) [Kla96b; Kor02]. From this point of view, two basic components are responsible for the fractality: On the one hand, the diffusion coefficient is given in terms of sums of jumps which, according to Eq. (4.23), are related to fractal generalized Takagi functions as shown in Fig. 4.3 (c) and (d). On the other hand, a second source of irregularity are the stepwise discontinuities in the density of the invariant measure %a (x) as shown in Fig. 4.3 (a) and (b). The irregular behavior of the diffusion coefficient results from a combination of these two ingredients, which are connected in the Taylor-Green-Kubo formula via integrating generalized Takagi functions over invariant densities [Kla96b; Kla02d]. This knowledge provides the key for developing systematic approximation procedures for fractal diffusion coefficients, as will be outlined particularly in Chapter 7. 4.2.2.2 An area-preserving multibaker with a fractal diffusion coefficient We now show that the same phenomenon of a fractal diffusion coefficient exists in a parameter-dependent generalization of the diffusive-reactive multibaker model introduced above. This two-dimensional area-preserving map is sketched in Fig. 4.4. Here the two rectangles of the left and of the right half of the square are “sliding” along the upper and the lower horizontal channel of the periodically continued map governed by a parameter h, as shown in the figure. It should be noted that for h = 0.5 and shifting the coordinate system by ∆x = 0.5 the model reduces to the simple dyadic multibaker of Eq. (4.8). The dynamics of the probability density f˜n (x, y, l) of the purely diffusive multibaker Φh (x, y, l) is determined by the Frobenius-Perron equation f˜n+1 (x, y, l) = f˜n [Φ−1 h (x, y, l)] ,

(4.25)

compare to Eq. (4.10), where Φ−1 h (x, y, l) is the inverse map. A projection of this two-dimensional Frobenius-Perron equation onto the unstable R x-direction by integrating over the stable y-direction via %n (x, l) := dy f˜n (x, y, l) [Dor99; Gas92a] shows that the dynamics of the probability density %n (x, l) is determined by the Frobenius-Perron equation of the simple one-dimensional map included in Fig. 4.4, which is a kind of generalized Bernoulli map shifted symmetrically by a height h as a control parameter. This one-dimensional map governs the dynamics of the multibaker map pro-

May 8, 2007

4:1

World Scientific Book - 9in x 6in

68

y

klages˙book

Deterministic reaction-diffusion

h

h

y

x

x

Fig. 4.4 Dynamics of one cell of an area-preserving multibaker with a non-trivial parameter-dependence h. Projection of the dynamics onto the horizontal axis reduces the system to the symmetric one-dimensional piecewise linear map shown in the figure to the left, which for h = 0 is the Bernoulli shift.

jected onto the x-axis. By extending the system periodically, we recover a chain of one-dimensional maps of the type as shown in Fig. 2.3. Concerning reversibility, we follow the definition [Rob92] that there must exist an involution G in phase space, G ◦ G = 1, which reverses the direction of time via5 G ◦ Φ ◦ G = Φ−1

.

(4.26)

For the special case of h taking multiples of 1/2 involutions G can be found which are related to a simple mirroring in phase space [Gas98c]. For general h it can be shown that the system has strong reversibility properties, however, the existence of an involution G remains an open question. To compute the parameter-dependent diffusion coefficient of this multibaker we use the fact that the projected dynamics is governed by a onedimensional map, which enables us to apply the same methods as outlined above. The result is shown in Fig. 4.5. The diffusion coefficient is again a non-trivial function of the parameter h sharing many characteristics with the curve presented in Fig. 2.5 as, for example, a random walk-like behavior on a coarse scale [Kla97]. But this curve also exhibits some new features, especially that the diffusion coefficient is constant in intervals 0.5 + m ≤ h ≤ 1 + m , m ∈ N0 . This is due to the fact that the topological transition matrices corresponding to respective Markov partitions, and 5 Note that reversibility in maps is more than the existence of an inverse. We remark that such a reversibility does not necessarily imply time-reversibility [Dol00] and that there is a controversy of how to define time-reversibility in maps [Rob92; Tas95; Hoo96b; Hoo98c; Bre98; Gil99b; Gas98a; Vol02].

4:1

World Scientific Book - 9in x 6in

klages˙book

69

4.2. Diffusive dynamics

4.5

0.5 0.4

3.5

0.3

D(h)

4

0.2

3

0.1

D(h)

May 8, 2007

2.5 0 0

0.1

0.2

2

0.3

0.4

0.5

h

1.5 1 0.5 0 0

0.5

1

1.5

2

2.5

3

h

Fig. 4.5 Parameter-dependent diffusion coefficient D(h) for the multibaker of Fig. 4.4 and blowup of the initial region. The main graph consists of 638 data points, the magnification of 514. In both cases the single points have been connected with lines, errorbars are too small to be visible.

thus the respective symbolic dynamics of the map, do not change in this parameter interval. It is worth mentioning that in contrast to the model of Chapter 2 the invariant probability density of the present projected one-dimensional map is always uniform for all parameter values of h. Therefore, according to the Taylor-Green-Kubo formula Eq. (4.22) the only contributions to the fractality of D(h) come from the inter-cell dynamics as described by the generalized Takagi functions Ta (x). A further interesting property of this map is that the slope is always equal to two, hence the Lyapunov exponent λ = ln 2 is constant under variation of the control parameter h. Motivated by a problem of electrical engineering, a piecewise linear one-dimensional map exhibiting the same properties has been studied in Refs. [Bar05; Bar06] by transition matrix methods, cf. Chapter 2. Here an autocorrelation function has been calculated exhibiting a sensitive dependence on parameters, which has been called parameter sensitivity or autocorrelation chaos. This property appears to be intimately related to the fractality of the diffusion coefficient, which via the Taylor-Green-Kubo formula Eq.(4.22) can be expressed in terms of velocity autocorrelation functions, see also Ref. [Kla96b]. We conclude this section by noting that the h-dependent one-dimensional map and its fractal diffusion coefficient have been reviewed in Ref. [Vol02].

May 8, 2007

4:1

World Scientific Book - 9in x 6in

70

4.3 4.3.1

Deterministic reaction-diffusion

Reactive dynamics +

Reactive modes of the dyadic multibaker

In this section we turn to the study of the chemiodynamic or reactive modes of our simple dyadic model Eq. (4.8) of diffusion-controlled reaction. Contrary to the total number of particles NA + NB , which is a constant of motion, the numbers of particles of each chemical species are not conserved. Accordingly, we should not expect that the reactive modes have a vanishing decay rate as k → 0. This is in contrast to the diffusive modes which are related to the conserved total number of particles and for which the decay rate (4.18) vanishes at k = 0. Here we consider the subspace defined by the difference between the particle concentrations in the multibaker, g(x, y, l) := f (x, y, l, A) − f (x, y, l, B) . (4.27) Thus we employ the fact that the dynamics of the concentration difference can be decoupled from the mean density for this model as has been mentioned before, see Eq. (4.4). With quasiperiodic boundary conditions, the difference of chemical concentration evolves in time according evolution operator to the reactive  x 1 gn+1 (x, y, 0) = θ − y gn , 2y, 1 +  2 2     1 −ikL  + e θ y − gn x+1  2 2 , 2y − 1, L − 1 ,      gn+1 (x, y, 1) = θ 21 − y gn x2 , 2y, 2 −      1 x+1  − θ y − g , 2y − 1, 0 , n  2 2      gn+1 (x, y, 2) = θ 21 − y gn x2 , 2y, 3 +    ˆk ≡ + θ y − 12 gn x+1 R 2 , 2y − 1, 1 ,    ..   .     1   − y gn x2 , 2y, L − 1 + g (x, y, L − 2) = θ n+1  2      , 2y − 1, L − 3 , + θ y − 12 gn x+1  2     gn+1 (x, y, L − 1) = − eikL θ 21 − y gn x2 , 2y, 0 +      + θ y − 12 gn x+1 2 , 2y − 1, L − 2 . (4.28) Our goal is to obtain the eigenvalues and eigenstates of this reactive evolution operator, ˆ k Ψ(x, y, l) L−1 = es(k) Ψ(x, y, l) L−1 , R (4.29) l=0 l=0

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

71

4.3. Reactive dynamics

with χ(k) = exp[s(k)]. We define the cumulative functions Gn (x, y, l) =

Z

x

dx

0

0

Z

y

dy 0 gn (x0 , y 0 , l) ,

(4.30)

0

which obey a set of equations that can be derived from Eq. (4.28). We suppose that the leading eigenstates are uniform along the unstable direction x, which is justified by the fact that the hyperbolic dynamics smoothes out any heterogeneities in this direction, Ψ(x, y, l) = D(y, l) ,

(4.31)

where D(y, l) is a Schwartz distribution. We note that the further eigenstates and root states do depend on x and require a more detailed analysis. The cumulative functions of the leading eigenstates are thus Geigenstate (x, y, l) := xC(y, l) with

C(y, l) =

Z

y 0

dy 0 D(y 0 , l) .

(4.32)

Replacing gn (x, y, l) by D(y, l) and gn+1 (x, y, l) by χD(y, l) in Eq. (4.28) and integrating over the interval [0, y], we obtain the following iterative equations for the new functions C(y, l), C(y, 0) =

(

1 2χ 1 2χ

C(2y, 1) ,

C(y, 1) =

(

1 2χ 1 2χ

C(2y, 2) ,

C(y, 2) =

(

1 2χ 1 2χ

C(2y, 3) ,

.. . (

1 2χ 1 2χ

C(2y, L − 1) ,

C(y, L − 2) =

C(y, L − 1) =

(

y < 1/2 ,

[C(1, 1) + exp(−ikL)C(2y − 1, L − 1)] ,

y > 1/2 ,

y < 1/2 ,

[C(1, 2) − C(2y − 1, 0)] ,

y > 1/2 ,

y < 1/2 ,

[C(1, 3) + C(2y − 1, 1)] ,

y > 1/2 ,

y < 1/2 ,

[C(1, L − 1) + C(2y − 1, L − 3)] ,

C(2y, 0) , y < 1/2 , − exp(ikL) 2χ 1 [− exp(ikL)C(1, 0) + C(2y − 2χ

y > 1/2 ,

(4.33) 1, L − 2)] , y > 1/2 .

The eigenvalue can be obtained by setting y = 1 in Eq. (4.33), which leads

May 8, 2007

4:1

World Scientific Book - 9in x 6in

72

klages˙book

Deterministic reaction-diffusion

to the eigenvalue equation  −2χ 1 0 0 ···  −1 −2χ 1 0 ···   0 1 −2χ 1 · · ·   0 0 1 −2χ · · ·   .. . . .. .. ..  .  . . . .   0 0 0 0 ··· − exp(ikL) 0 0 0 ···

  C(1, 0) 0 exp(−ikL)   C(1, 1)  0 0     C(1, 2)  0 0     0 0   C(1, 3)  = 0 .    .. .. ..     . . .    C(1, L − 2) 1 −2χ 1 C(1, L − 1) 0 1 −2χ (4.34) The characteristic determinant has been calculated for several values of the distance L between the reactive sites, L=3: L=4: L=5: L=6:

0 0 0 0 .. .

4χ3 + χ + cos(3k) = 0 , 8χ

4

16χ

5

32χ

6

− 2 + 2 cos(4k) = 0 , − 4χ

3

− 16χ .. .

4

− 3χ + cos(5k) = 0 , − 6χ

2

+ 1 + cos(6k) = 0 ,

(4.35) (4.36) (4.37) (4.38)

The corresponding dispersion relations of the reactive modes are depicted in Fig. 4.6 together with those of the diffusive modes. This figure shows that the slowest decay rate which gives the reaction rate appears at k = 0 for L odd and at k = ±π/L for L even. The cumulative functions {C(y, l)}3l=0 of the eigenstate corresponding to the reaction rate at k = 0 are depicted in Fig. 4.7 for the model with L = 3 by solving Eq. (4.33) iteratively. Near its maximum values, the dispersion relation behaves quadratically like L odd : s(r) (k, L) = −˜ κr (L) − D(r) (L)k 2 + O(k 4 ) at k = 0 , π 2 L even : s(r) (k, L) = −˜ κr (L) − D(r) (L) k ∓ + L π 4 π +O k∓ at k = ± . L L

(4.39)

(4.40)

An analytical calculation of the reaction rate κ ˜ r (L) and a numerical calculation of the reactive diffusion coefficient D (r) (L) versus L reveal that

κ ˜r (L) = − ln cos D(r) (L) ∼

1 . L

π π2 = + O(L−4 ) , L 2L2

(4.41) (4.42)

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

73

4.3. Reactive dynamics

Fig. 4.6 Dispersion relations of the diffusive modes (solid lines) and of the reactive modes (dotted lines) for the dyadic reactive multibaker with (a) L = 3; (b) L = 4; (c) L = 7; (d) L = 8.

The reaction rate thus behaves as expected for diffusion-controlled reaction in one dimension, cp. to Eq. (4.9). These results, combined with the results for the diffusive modes, show that on macroscopic scales the coarse-grained density Z 1 Z 1 ˜ y, l) ρ(l) = dx dy f(x, (4.43) 0

0

and the concentration Zdifference Z 1 1 σ(l) = dx dy g(x, y, l) = G(1, 1, l) 0

behave like

(4.44)

0

∂2ρ ∂ρ ' D 2 diffusive mode , (4.45) ∂t ∂l 2 ∂σ ∂ σ ' D(r) 2 − κ ˜ r σ reactive mode, L odd . (4.46) ∂t ∂l Corrections with higher-order spatial derivatives could also be taken into account in the dynamics of the reactive mode. For a model with L odd, this behavior corresponds to a macroscopic reaction-diffusion system with ∂ρA D + D(r) ∂ 2 ρA D − D(r) ∂ 2 ρB κ ˜r ' + − (ρA − ρB ) , (4.47) 2 2 ∂t 2 ∂l 2 ∂l 2 ∂ρB D − D(r) ∂ 2 ρA D + D(r) ∂ 2 ρB κ ˜r ' + + (ρA − ρB ) , (4.48) ∂t 2 ∂l2 2 ∂l2 2

May 8, 2007

4:1

World Scientific Book - 9in x 6in

74

Deterministic reaction-diffusion

Fig. 4.7 Cumulative functions {C(y, l)} with l = 0, 1, 2 of the reactive eigenstate at vanishing wavenumber k = 0 in the reactive multibaker model L = 3.

where ρA = ρ + σ/2 and ρB = ρ − σ/2. According to these macroscopic equations, the diffusion coefficient of each species is DA = DB = (D + D(r) )/2, the cross-diffusion coefficient is DAB = DBA = (D −D(r) )/2, while the reaction rate of Eq. (4.1) is given by the logarithm of the absolute value of the leading eigenvalue of the reactive evolution operator as 1 π 1 κ ˜r = − ln cos = − ln |χ(k = 0)| . (4.49) 2 2 L 2 We remark that, according to this microscopic analysis, the macroscopic equations of a reaction-diffusion system do not necessarily follow the simple assumption often made out that the cross-diffusion coefficients vanish, κr =

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

75

4.3. Reactive dynamics

DAB = DBA = 0. This particular case is only recovered if D = D (r) , which is not fulfilled here. The origin of this difference holds in the fact that the diffusion coefficient DA associated with the state A of a particle generally differs from the diffusion coefficient of the particle itself, which may be in two possible states A or B. In this regard, the cross-diffusion appears to be of importance in reacting systems. The models with L even follow more complicated reaction-diffusion equations, where the reactive diffusion coefficient D (r) has a different status because it is associated with a nonvanishing wavenumber k = ±π/L. Nevertheless, the part of the reaction-diffusion process which is responsible for the reactive exponential decay is recovered from the microscopic analysis. 4.3.2

The parameter-dependent reaction rate

We conclude this chapter by investigating the parameter-dependent reactivity of the multibaker. That is, in addition to the integer periodicity L of the catalysts of the multibaker, the reaction rate κr will also depend on h. One may thus raise the question how κr (h, L) changes with respect to varying h for fixed L. Moreover, we will give some illustrative features of the time-dependent dynamics of the reaction process for typical h parameters. Analogously to the previous section, we start with the difference of chemical concentrations g(x, y, l) as defined in Eq. (4.27). We again use the property that parallel to the x-axis the two-dimensional reactive multibaker can be projected onto a one-dimensional map as has been pointed out before,Rcf. Fig. 4.4. The time evolution of the projected reactive part ζn (x, l) := dyg(x, y, l) of the multibaker is then determined by the reactive evolution equation of the associated one-dimensional map, ˆ (1) (h, L) ζn−1 (x, l) . ζn (x, l) = R

(4.50)

ˆ (1) (h, L) represents the one-dimensional reactive evolution operator Here R and ζn (x, l) := %n (x, l, A) − %n (x, l, B) is the difference between A and Bparticle densities in the corresponding one-dimensional map. As for the purely diffusive case, see the previous two chapters, we write Eq. (4.50) ˆ (1) (h, L) a topological transition as a matrix equation, where instead of R matrix M (h, L) acts onto a particle density vector ζ n . The matrix M (h, L) is structured such that in the case of reactive scattering centers the elements in the corresponding columns of the matrix have a negative sign, and thus a particle changes color by leaving a reaction cell. Otherwise the matrix is the same as outlined for the diffusive case.

May 8, 2007

4:1

76

World Scientific Book - 9in x 6in

Deterministic reaction-diffusion

We first discuss some details of the time evolution of the reactive modes. By integrating over ζn (x, l) or its respective vector representation we obtain the difference between the total number of A and B-particles at discrete P R time n, which is ξn := l dxζn (x, l). From the corresponding phenomenological time-continuous reaction equation Eq. (4.46) one would expect that for the reactive multibaker ξn decays exponentially after a suitable coarse graining according to ξn = ξ0 exp(−˜ κr t). If this is the case, we can define the reaction rate of the reactive multibaker in analogy to the phenomenological equation as κr (h, L) = κ ˜ r /2. To compute κr (h, L) according to this definition, we solve the matrix formulation of Eq. (4.50) by iterating the transition matrices M (h, L) numerically. As an initial particle density we choose ζ0 (x, l) to be uniform in one reactive cell of the multibaker, which corresponds to having only A particles in this cell with the number of B particles being locally zero, and we make the chain long enough such that the evolving density is not affected by boundary conditions. Figs. 4.8 (a) and (b) give two typical examples of ζn (x, l) for certain parameter values of h after n = 40 iterations. They show how the “perturbation” ζ0 (x, l), which is a local initial deviation from the equilibrium state ζn (x, l) = 0 (n → ∞), spreads out along the x-axis by exhibiting an intricate fine structure with oscillations around zero. Figs. 4.8 (c) and (d) contain half-logarithmic plots of |ξn | with respect to the discrete time n. These plots exemplify different dynamical behavior of ξn for different magnitudes of the reaction rate. For κr (h, L) close to zero, see the upper two curves in Fig. 4.8 (c), ξn decays apparently non-exponentially for small times n. Only for larger times does it eventually reach exponential decay. Thus, the system shows that it is close to states of the h parameter where it is non-reacting. For the lowest curve in Fig. 4.8 (c), which corresponds to an intermediate reaction rate, ξn displays initially strong periodic fluctuations. They are partly due to the complex deterministic dynamics of the reactive baker in one cell of the chain, as has already been observed and explained for a one-dimensional purely diffusive case [Kla96b]. Apart from such strong periodic oscillations on a fine scale, in Fig. 4.8 (d) ξn exhibits an interesting crossover between a fast decay for smaller times and a slower decay for larger times, where again it approaches exponential behavior. This may reflect the fact that for larger reaction rates κr (h, L) 0 the reaction is controlled by the diffusive dynamics. These features of ξn should be compared to the characteristics of the respective probability densities ζn (x, l) in the figure.

klages˙book

4:1

World Scientific Book - 9in x 6in

klages˙book

77

4.3. Reactive dynamics

0

0.09

(a)

(c)

-0.5

0.03

-1 ln |ξ n|

ζ n(x,l)

0.06

0

-1.5

-0.03 -2

-0.06 -0.09 -10

-5

0

5

2

10

4

6

10

12

14

0

0.0002

(d)

0.0001 0

-1 -2

-0.0001

ln |ξ n|

(b)

8 n

x+l

ζ n(x,l)

May 8, 2007

-3

-0.0002

-4

-0.0003

-5

-0.0004 -0.0005

-6

-10

-7.5

-5

-2.5

0 x+l

2.5

5

7.5

10

5

10

15

20

25

30

35

n

Fig. 4.8 (a), (b) Densities corresponding to the difference in the number of A and B-particles per total number of particles after n = 40 iterations of the map. In (a) (L = 2, h ' 0.2429) the reaction rate is close to zero, whereas in (b) (L = 3, h ' 0.1496) it is locally maximal in h. (c), (d) Half-logarithmic plots of the total difference ξ n in the number of A and B-particles of the system as it varies in time n. In (c) the corresponding reaction rates κr (h, L) are close to zero for the upper two curves (κr < 0.006), for the lower curve the reaction rate has an intermediate value (κr ' 0.02), whereas in (d) it is locally maximal in h (κr ' 0.05). The parameters for the upper curve in (c) correspond to (a), the curve in the middle is at L = 3, h ' 0.247, and the lowest one is at L = 3, h ' 0.4472. Case (d) corresponds to (b).

To obtain quantitative values for the reaction rate, Eq. (4.50) has been analyzed by solving the eigenvalue problem of the corresponding transition matrix M (h, L), analogously to what has been done in the previous section for the dyadic reactive multibaker. In general, the spectra of M (h, L) are extremely complicated [Kla]. However, as has been argued in the previous section for special cases, and supported by our observation of long-time exponential decay of ξn in general, we expect that, in the limit of infinite time, the reaction in the multibaker will always be governed by the slowest eigenmodes and their respective eigenvalues. This motivates the definition of the parameter-dependent reaction rate κr (h, L) via the maximum of the absolute value of the eigenvalues of M (h, L), 1 (4.51) κr (h, L) := − ln |χmax (h, L)| , 2

4:1

World Scientific Book - 9in x 6in

78

klages˙book

Deterministic reaction-diffusion

0.2

(a) ψ (x,l)

0.1

0

-0.1

-0.2 -2

-1.5

-1

-0.5

0

0.5

1.0

1.5

2

0.5

1

1.5

2

x+l

(b)

0.15

ψ (x,l)

0.1 0.05 0 -0.05 -0.1 -0.15 -2

-1.5

-1

0.5

0 x+l

0.3

(c)

0.2 0.1 ψ (x,l)

May 8, 2007

0 -0.1 -0.2 -0.3 -4.5

-2.5

-0.5

1.5

3.5

5.5

x+l

Fig. 4.9 Examples of largest eigenmodes ψ(x, l) parallel to the x axis in the fundamental domain LF (see text) for the reactive multibaker corresponding to the largest eigenvalue χmax (h, L). For (a) (L = 2, h ' 0.1496) the reaction rate κr (h, L) is very large (κr ' 0.12), for (b) (L = 2, h ' 0.4947) it is very small (κr ' 0.003). In both cases there exist only two real largest eigenmodes, where the second ones are shifted by a phase. In (c) (L = 6, h = 1) both largest eigenmodes have been plotted (thick line for the one and thin line for the other, respectively).

analogously to Eq. (4.49). Numerically we find that for large regions of the h parameter a certain fundamental domain LF of the multibaker is sufficient to obtain the correct leading eigenvalue χmax (h, L). This domain must always include multiples of two reactive centers, and its length is defined by the number LF ≡ 2L Int(1 + h) of cells of the multibaker. In these regions, solutions for the eigenvalue problem of M (h, L) defined on the domain LF lead to a maximum eigenvalue χmax (h, L) as obtained by solving

May 8, 2007

4:1

4.3. Reactive dynamics

World Scientific Book - 9in x 6in

klages˙book

79

the corresponding eigenvalue problem for longer and longer chain lengths mL → ∞ , m ∈ N. However, especially for small L and large κr (h, L) this fundamental domain only provides an approximation to the exact result, which is then obtained by making the chain length mL large enough such that the error in χmax (h, L) with respect to mL is sufficiently small. Fig. 4.9 shows some typical largest eigenmodes ψ(x, l) on the fundamental domain in the cases where it yields the correct corresponding largest eigenvalue κr (h, L). For large reaction rates the largest eigenmodes behave, to a certain extent, like sine functions, see Fig. 4.9 (a), whereas for smaller reaction rates the largest eigenmodes approach two-periodic step-like functions as shown in Fig. 4.9 (b). Fig. 4.9 (c) depicts the largest eigenmodes for a parameter value of h which is just at the borderline of a non-reacting h region but where the system is nevertheless already highly reactive. Here the eigenmodes appear to be especially complicated [Kla]. In Fig. 4.10 (a) the reaction rate κr (h, L) defined by Eq. (4.51) has been computed for a series of reaction center periodicities L. For h = 0.25 there is no reaction rate in the system. In this case the iteration method confirms that the difference in the number of particles ξn oscillates periodically around zero instead of decaying exponentially. An analysis of the eigenvalue spectra of the corresponding transition matrices reveals that at this h parameter the respective reactive multibakers are not ergodic [Kla]. For all other h values shown in the figure, the reaction rate is welldefined and exhibits a complicated structure. By increasing L the reaction rate κr (h, L) decreases almost everywhere, as one can expect intuitively, except in certain small parameter regions of h. Fig. 4.10 (b) gives the full result for κr (h, L) at L = 2. This structure repeats itself with a periodicity of 2m ≤ h ≤ 2 + 2m , m ∈ N0 . In certain intervals of h the four peaks depicted in the figure are very similar, or even identical, however, there does not appear to be a simple scaling law by which the full peaks can be mapped onto each other. The plateau regions with zero reaction rate correspond to the respective regions observed in Fig. 4.5 for the diffusion coefficient of the system. They share the same characteristics as discussed above for the singular case of h = 0.25, except that at h = 1 the system is ergodic but not mixing. Topologically these regions are of the same origin as explained for the diffusive case. In Fig. 4.5 (c) the reaction rate has been computed for L = 3. In contrast to the two-periodic case, the change from a non-reactive region to a reactive region apparently occurs, for L = 3, discontinuously in the reaction rate by varying h around 1. This corresponds to the system at h = 1 and L = 3

4:1

World Scientific Book - 9in x 6in

80

klages˙book

Deterministic reaction-diffusion

0.12

(a) κ (h,L)

0.1 0.08 0.06 0.04 0.02 0 0

0.1

0.2

0.3

0.4

0.5

h

0.14

(b)

0.12

κ (h,L)

0.1 0.08 0.06 0.04 0.02 0 0

0.5

1

1.5

2

h

0.2

(c)

0.15

κ (h,L)

May 8, 2007

0.1

0.05

0 0

0.5

1

1.5

2

2.5

3

h

Fig. 4.10 (a) Reaction rate κr = κ(h, L) at different integer values of the catalyst periodicity: L = 2 (upper curve), L = 3 (middle), L = 4 (lower curve). (b) Reaction rate for 0 ≤ h ≤ 2 at L = 2. (c) Reaction rate for 0 ≤ h ≤ 3 at L = 3. In all cases, error bars are too small to be visible.

being mixing, whereas for the same h and L = 2 it was ergodic but not mixing. We note that for L = 4 there are even two such discontinuous transitions. The detailed irregular structure of the curves, as well as the phase transition like behavior shown in Fig. 4.10 (c), can be understood in more detail by analyzing the eigenvalue spectra of the reactive evolution operator and how they change under parameter variation [Kla]. Apart from varying h, other parameter dependencies can be studied in this reactive multibaker as well. For example, the distance between the single reaction centers could be changed by allowing L to be continuous, the positions of the reaction centers could be shifted by keeping L fixed,

May 8, 2007

4:1

4.4.

World Scientific Book - 9in x 6in

∗ Summary

klages˙book

81

and the size of the reaction centers could be increased or decreased. In all these cases we expect non-trivial parameter dependences to be typical, which are similar to the one depicted in Fig. 4.10. We finally remark that in parallel to Ref. [Gas98c], a chemically reactive Lorentz gas has been introduced and studied in Ref. [Nie98]. These publications triggered a series of works in which the chaotic and fractal properties of various types of reactive Lorentz gases have been investigated in more detail [Cla00; Cla01; Cla02; Gas02b; Cla04; M´ at05]. Recently such models attracted the attention of mathematicians, who formulated a rigorous mathematical framework for studying simple deterministic reactiondiffusion processes [Bun05a].

4.4

∗

Summary

(1) Motivated by a simple particle billiard, the Lorentz gas, we introduced two-dimensional multibaker maps, which form another widely studied class of models for deterministic transport. We then defined a parameter-dependent reactive-diffusive multibaker and outlined in which way the reactive-diffusive dynamics is governed by the FrobeniusPerron (continuity) equation determining the probability density of this dynamical system. (2) We first focused on the purely diffusive dynamics of this model. Choosing for the control parameter of the map the most trivial value, we constructed the diffusive eigenmodes of the system and established a relation of the associated nonequilibrium steady states to fractal Takagi functions. We continued by explaining another basic method to calculate and to understand fractal diffusion coefficients, which starts from the TaylorGreen-Kubo formula for diffusion. Along these lines, the fractal diffusion coefficient could be linked again to fractal, in this case generalized, Takagi functions. This set the stage for computing the parameterdiffusion coefficient of the reactive-diffusive multibaker. We obtained again a fractal function, however, the result exhibited some new features compared to the case discussed previously. (3) Finally, we studied the reactive properties of our model. We started by constructing the reactive modes, again first for a specific parameter value, and by computing the dispersion relation for both the diffusive and the reactive modes. We also outlined the relation of our model to

May 8, 2007

4:1

82

World Scientific Book - 9in x 6in

Deterministic reaction-diffusion

a set of coupled macroscopic reaction-diffusion differential equations. In the last section we presented results for the parameter-dependent reaction rate of the model. For this purpose the equations governing the multibaker dynamics projected upon the spatially elongated axis have been solved numerically by suitably adapting the set of methods for one-dimensional maps discussed previously. We detected very irregular structures in the time-dependent spreading of the particle densities and in the reactive eigenmodes culminating in fractal parameter dependences of the chemical reaction rate. All these quantities exhibited furthermore non-trivial scenarios of dynamical phase transitions. Their origin could be traced back to changes in the topology of the underlying chaotic dynamical system subject to parameter variations.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Chapter 5

Deterministic diffusion and random perturbations

So far we have focused on purely deterministic systems, where for a given equation of motion the initial condition reproducibly determines the outcome of the dynamics. In this chapter we turn our attention to the interplay between deterministic chaos and random perturbations. As an example we choose again our standard model whose deterministic transport properties have been analyzed in Chapters 2 and 3, that is, a piecewise linear map acting on the real line. In Section 5.1 we study the impact of spatial disorder onto deterministic diffusion in this map [Kla02b] while Section 5.2 deals with the case of perturbing the system by time-dependent noise [Kla02c]. 5.1

Disordered dynamical systems

The field of disordered dynamical systems brings together two at first view very different directions of research [Rad96; Rad99; Rad04; Fic05; Bun04]: Diffusion on disordered lattices with quenched (static) randomness is a traditional problem of statistical physics, which can be studied by probabilistic methods developed in the theory of stochastic processes [Hau87; Wei94; Bou90; Keh98]. However, as we learned in the previous chapters, diffusion may also be generated by deterministic chaos in nonlinear equations of motion [Gei82; Sch82; Gro82; Fuj82] making it possible to assess chaotic and fractal properties of diffusion by methods of dynamical systems theory. Understanding diffusion in disordered dynamical systems hence poses the challenge of suitably combining these different concepts and ideas. It appears that only very few cases of such models have been studied so far. Examples include random Lorentz gases for which Lyapunov exponents have been calculated by means of kinetic theory and by computer sim83

klages˙book

May 8, 2007

4:1

84

World Scientific Book - 9in x 6in

Deterministic diffusion and random perturbations

ulations [Dor99; vB95; Del95c; vB98; dW04b; M¨ ul04], numerical studies of diffusion on disordered rough surfaces and in disordered deterministic ratchets [Pop98; Pop00; Ari01] as well as numerical and analytical studies of chaotic maps on the line with quenched disorder [Rad96; Rad99; Tse99; Tse00; Rad04; Fic05]. In this section we focus again on the most simple example in the latter class of models. As we have seen in Chapters 2 and 3, piecewise linear maps periodically continued on the line may exhibit normal diffusion. However, as was shown in Refs. [Rad96; Rad99; Tse99; Tse00], adding quenched disorder in form of a local bias with globally vanishing drift profoundly changes the diffusive dynamics of these systems. In this situation one finds a mean square displacement that grows less than linear in time. This phenomenon of subdiffusion in a complicated potential landscape is well-know from the stochastic theory of random walks in random environments [Sin82], where one speaks of Golosov localization [Gol84]. Indeed, there is a crosslink between this effect and the problem of Anderson localization in quantum mechanical lattices [Rad04]. In this case the fractal diffusion coefficient discussed previously would immediately disappear unless one introduces a suitably generalized diffusion coefficient as will be done in the following chapter, see Section 6.2. In the following we illustrate the topic of disordered dynamical systems by considering a somewhat simpler situation. We do so by studying static randomness that is multiplicative, preserves the local symmetry of the model and is not related to L´evy distributions [Bar00b] thus not resulting in anomalous diffusion.1 Particularly, we wish to assess the stability of the fractal diffusion coefficient investigated previously under this type of perturbation. Let us consider again the unperturbed model defined as map L in Chapter 2, see Eqs. (2.3) to (2.6). We now modify this system by adding a random variable ∆a(i), i ∈ Z, to the slope a on each interval [i − 1/2, i + 1/2) (i) yielding Ma (x) = (a + ∆a(i))x. We assume that the random variables ∆a(i) are independent and identically distributed according to a distribution χα (∆a), where α is another control parameter. In the following we will consider two different types of such distributions, namely random variables distributed uniformly over an interval of size [−α, α], χα (∆a) = Θ(α + ∆a)Θ(α − ∆a)/(2α) , (5.1) and dichotomous or δ-distributed random variables, χα (∆a) = (δ(α + ∆a) + δ(α − ∆a))/2 . (5.2) 1 More detailed explanations about subdiffusion and anomalous diffusion can be found in the following chapter.

klages˙book

4:1

World Scientific Book - 9in x 6in

klages˙book

85

5.1. Disordered dynamical systems

2

1.5 D(a,α)

May 8, 2007

a

1

0.5

0

3

4

5 a

6

7

8

Fig. 5.1 Diffusion coefficient D(a, α) for the piecewise linear map shown in the figure. The slope a is perturbed by static disorder of maximum strength α as defined in Eq. (5.1). The bold black line depicts D(a, 0). The symbols represent computer simulation results for α > 0, the corresponding lines are obtained from the approximation Eq. (5.4) The parameter values are: α = 0.1 (circles), α = 0.4 (squares), α = 1.0 (diamonds).

Since |∆a| ≤ α, we denote α as the perturbation strength. As an example, we sketch in Fig. 5.1 the map resulting from the disorder of Eq. (5.1) as applied to the slope a = 3. In the absence of any bias the diffusion coefficient is defined by D(a, α) := limn→∞ < x2n > /(2n), see also Eq. (2.2), where the angular brackets denote an ensemble average over moving particles for a given configuration of disorder. An additional disorder average is not necessary because of self-averaging [Bou90]. Note that for locally symmetric quenched disorder and (a − α) > 2 there is no physical mechanism leading to infinitely high reflecting barriers as they are responsible for Golosov localization [Rad96; Rad99; Rad04; Bou90]. Thus diffusion must be normal, as is confirmed by computer simulations. The central question is now how the parameter-dependent diffusion coefficient D(a, α) is affected by variation of the two control parameters a and α. We know from Chapters 2 to 4 that in the unperturbed case α = 0 the diffusion coefficient D(a, 0) is a fractal funtion of the slope a as a control parameter. This function is depicted again in Fig. 5.1 together with results

May 8, 2007

4:1

86

World Scientific Book - 9in x 6in

klages˙book

Deterministic diffusion and random perturbations

from computer simulations for different values of the perturbation strength α in the case of uniform disorder.2 As expected, this irregular structure gradually diasppears by increasing α. Remarkably, even for large perturbation strength α oscillations are still visible as a function of a indicating that the underlying dynamical correlations are very robust against this type of perturbation. Note furthermore the non-analytical behavior of D(a, α) for α > 0 and small a indicating the emergence of a dynamical phase transition, which was not present in the unperturbed case. Recently, related phase transitions in the random Lorentz gas have been studied within the context of percolation theory [H¨ of06]. Before we proceed to more detailed simulation results we briefly repeat what is known for diffusion in lattice models with random barriers [Hau87; Wei94; Bou90; Keh98; Zwa82; Der83; Lyo85]. In the most simple version, the quenched disorder is defined on a one-dimensional periodic lattice with transition rates between neighbouring sites i and i+1 having the symmetry Γi,i+1 = Γi+1,i =: Γk for a given random distribution of Γk . In this situation an exact expression for the stochastic diffusion coefficient has been derived reading [Keh98; Zwa82; Der83; Lyo85] d = {1/Γ}−1l2

(5.3) PN

with the brackets defining the disorder average {1/Γ} = 1/N k=0 1/Γk at chain length N and for a distance l between sites. The double-inverse demonstrates that the highest barriers dominate diffusion in one dimension, thus Γk → 0 naturally leads to a vanishing diffusion coefficient. This scenario is translated to the map under consideration as follows: The equation of motion xn+1 = Ma (xn ) can be understood as a time-discrete Langevin equation, xn+1 = xn − dV (x)/dx|xn [Rad96; Rad99; Rad04; (i) Gro82; Fuj82]. By integrating Ma (x) the corresponding potential V (x) yields a random barrier model with the perturbation strength α determining the highest barriers. Hence, simple random walk theory predicts a suppression of diffusion for the chaotic map viz. D(a, α) being a monotonously decreasing function of α. To check this hypothesis, we choose fixed values of a corresponding to the two extreme situations of starting from a local maximum or minimum, respectively, of the unperturbed D(a, 0) in Fig. 5.1. We first focus on the local minimum at a = 6 for α ≤ 0.5 with both uniform and dichotomous 2 In

the unperturbed case the numerical error is less than visible. In the case of perturbations, for each point a maximum of 1,000,000 particles have been iterated up to 50,000 time steps each; representative error bars are included in all figures.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

87

5.1. Disordered dynamical systems

0.92 (a)

(b) 1

D(a,α)

D(a,α)

0.9

0.88

0.9

0.86

0.84 0

0.95

0.85 0.1

0.2

α

0.3

0.4

0.5

0

0.1

0.2

α

0.3

0.4

0.5

Fig. 5.2 Diffusion coefficient D(a, α) as a function of the perturbation strength α at slope a = 6: (a) disorder distribution Eq. (5.1), (b) disorder distribution Eq. (5.2). The circles represent results from computer simulations, the lines are obtained from the approximation Eq. (5.4).

disorder, see Fig. 5.2. In sharp contrast to the prediction of the theory outlined above, in both cases we observe enhancement of diffusion as a function of α. Moreover, this enhancement does not show up in form of a simple functional dependence on α: In (a), smoothed-out oscillations are visible on smaller scales while in (b) the resulting function is clearly non-monotonous and wildly fluctuating exhibiting multiple suppression and enhancement in different parameter regions. Results on larger scales of α are depicted in Fig. 5.3 for dichotomous disorder. Fig. 5.3 (b) shows that choosing a at a local maximum of D(a, 0) leads to suppression of diffusion for α < 1.0 while a local minimum generates enhancement in the same parameter region of α. We emphasize that in both cases the diffusion coefficient decreases on a very coarse scale by increasing α thus recovering qualitative agreement with the simple random walk prediction Eq. (5.3).3 Indeed, for (a − α) → 2 barriers are formed which a particle cannot cross anymore implying the existence of trivial localization. In order to theoretically explain the simulation results, we adapt Eq. (5.3) in a straightforward way such that it can be applied to our disordered deterministic map under consideration. We first note that for uniform transition rates Γk = const. it is d(Γk , l) = Γk l2 . Using this familiar expression for the random walk diffusion coefficient on the unperturbed lattice we rewrite Eq. (5.3) as d = {1/d(Γk , l)}−1 . In the case of our map, the transition rates and the distance between sites are both somewhat combined in 3 Simulations have also been carried out for uniform disorder Eq. (5.1) essentially yielding smoothed-out versions of the curves shown in Fig. 5.3 [Kla].

4:1

World Scientific Book - 9in x 6in

88

klages˙book

Deterministic diffusion and random perturbations

2

1.5

1.5 D(a,α)

1 D(a,α)

May 8, 2007

1

0.5 0.5 (a) 0 0

(b) 1

2 α

3

4

0 0

1

2

α

3

4

5

Fig. 5.3 Diffusion coefficient D(a, α) as a function of the perturbation strength α for dichotomous disorder Eq. (5.2) at two different slopes a: (a) a = 6, (b) a = 7. The circles represent results from computer simulations, the lines are obtained from the approximation Eq. (5.4).

the action of the slope a as a control parameter. Therefore, the unperturbed diffusion coefficient is correctly rewritten by replacing d(Γk , l) = d(a). The key question is now what function shall be used for the parameterdependent diffusion coefficient d(a) in the case of deterministic dynamics. Here we propose to identify the function d(a) with the exact, unperturbed deterministic diffusion coefficient previously defined as D(a, 0). Using this information, the exact Eq. (5.3) for stochastic dynamics becomes an approximation which can straightforwardly be applied to deterministic dynamics in disordered systems. If the disorder distributions χα (∆a) are bounded by the perturbation strength α, and by taking the continuum limit for the random variable, our final formula reads Z α −1 χα (∆a) Dapp (a, α) = d(∆a) . (5.4) D(a + ∆a, 0) −α The results obtained from this equation are depicted in Figs. 5.1 to 5.3 in the form of lines. For small enough α the agreement between theory and simulations is excellent thus confirming the validity of this equation. For larger α our theory still correctly predicts the oscillations generated from dichotomous disorder, however, since Eq. (5.4) is approximate it should not be surprising to detect quantitative deviations. We remark that a proof of this equation, perhaps under simplifying assumptions, as well as the calculation of higher order corrections to it for explaining the deviations between simulation and theory, as particularly being visible in Fig. 5.3, are currently open questions. We now show that this formula provides a physical explanation for

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

89

5.2. Noisy dynamical systems

the irregular dependence of the diffusion coefficient on the perturbation strength. For α → 0 Taylor expansion leads to Z α Dapp (a, α) = d(∆a) χα (∆a)D(a + ∆a, 0) . (5.5) −α

As will be shown in the following section, this equation can be proven without using Eq. (5.4) by starting from the definition of the diffusion coefficient. In this limit the perturbed diffusion coefficient reduces to an average of the exact diffusion coefficient over the neighbourhood [a−α, a+α] weighted by the respective disorder distribution χα (∆a). Consequently, if a is chosen at a local minimum the result must be enhancement of diffusion by increasing α and suppression at a local maximum, respectively.4 On these grounds it is clear that the fractal parameter dependence of D(a, 0) must reappear in the perturbed diffusion coefficient hence leading to multiple suppression and enhancement on fine scales. The results presented in this section could be applied to diffusion on stepped surfaces with a disordered arrangement of Ehrlich-Schwoebel barriers. In Ref. [Mus98] this problem has been analyzed by suggesting a stochastic random trap/random barrier model. Our map provides a deterministic generalization of such a model in terms of correlated random walks. It thus enables one to study the impact of memory effects on surface diffusion. Furthermore, our approach could be worked out for disordered deterministic ratchets [Pop98; Pop00; Ari01] or periodic Lorentz gases [Gas98a; Dor99] under the impact of random perturbations. Knowing the densitydependent diffusion coefficient for some periodic particle billiards [Kla00a; Har02] leads us to predict local and global suppression of the diffusion coefficient for these models in the case of static density fluctuations. 5.2

Noisy dynamical systems

Understanding diffusion in noisy dynamical systems, where deterministic equations of motion are randomly perturbed in time, represents a second fundamental problem in the field of dynamical systems with random perturbations. The most simple examples are again one-dimensional chaotic maps on the line. In seminal contributions by Geisel, Nierwetberg [Gei82] and Reimann et al. [Rei94; Rei96a; Rei96b], scaling laws have been derived yielding suppression and enhancement of normal diffusion under variation 4 Based on this heuristic argument, these phenomena had earlier been conjectured in Ref. [Kla96b], see p. 65.

May 8, 2007

4:1

90

World Scientific Book - 9in x 6in

klages˙book

Deterministic diffusion and random perturbations

of the noise strength. Related results have been obtained in Refs. [Fra91; Wac99; Eck02]. However, all these studies apply only to the onset of diffusion where the scaling laws are related to a dynamical phase transition while not much appears to be known far away from this transition point. In such more general situations, only deterministic perturbations by a nonzero average bias have been studied [Bar97; Bar98]. Related models are deterministic Langevin equations, in which the interplay between deterministic and stochastic chaos has been analyzed [Bec87; Bec95; Bec96; Kan98], however, without focusing on diffusion coefficients. Non-diffusive noisy maps have furthermore been investigated by refinements of cycle expansion methods [Cvi00]. Deterministic diffusion as discussed in the previous chapters refers to the asymptotically linear growth of the mean square displacement in a purely deterministic, typically chaotic dynamical system [Gei82; Rei94; Rei96a; Rei96b; Wac99; Kla95; Kla99a; Kla96b]. By stochastic diffusion we denote in this section the respective behavior of the same quantity in a system driven by uncorrelated random variables, vulgo noise. In the following we describe the transition scenario from deterministic to stochastic diffusion for our standard model, the piecewise linear map on the line. Particularly, we are looking for signatures of deterministic and stochastic dynamics in the diffusion coefficient as a function of the noise strength. In this aspect our work touches upon a recent dispute on a possible distinction between chaotic and stochastic diffusion in experiments [Gas98b; Bri01; Gra99; Det99b; Det00b; Cen00]. Indeed, some of the theoretical models studied in this context are very similar to the one to be introduced below. We start from the same unperturbed system as considered in the previous section, that is, map L as defined in Chapter 2, see Eqs. (2.3) to (2.6). However, here we apply two types of annealed disorder [Bou90] to the map, namely (i) noisy slopes [Rei94; Rei96a; Rei96b; Wac99]: we add the random variable ∆an , n ∈ N, to all slopes a making them time-dependent in form of Ma+∆an (x) = (a + ∆an )x ,

(5.6)

or (ii) noisy shifts [Gei82; Rei94; Rei96a; Rei96b; Fra91]: we add the random variable ∆bn , n ∈ N, as a time-dependent uniform bias yielding Ma,∆b (x) = ax + ∆bn .

(5.7)

In both cases we assume that the random variable ∆n ∈ {∆an , ∆bn } is independent and identically distributed according to a distribution χd (∆n ),

4:1

World Scientific Book - 9in x 6in

klages˙book

91

5.2. Noisy dynamical systems

2

1.5 D(a,da)

May 8, 2007

a

1

0.5

0

3

4

5 a

6

7

8

Fig. 5.4 Diffusion coefficient D(a, da) for the piecewise linear map shown in the figure. The slope a is perturbed by the uniform noise Eq. (5.8) of maximum strength da. The bold black line depicts numerically exact results for the unperturbed diffusion coefficient at da = 0. Computer simulation results for da 6= 0 are marked with symbols, the corresponding lines are obtained from the approximation Eq. (5.12). The parameter values are: da = 0.1 (circles), da = 0.4 (squares), da = 1.0 (diamonds).

where d ∈ {da, db} is again a control parameter. We consider two different types of such distributions, namely random variables distributed uniformly over an interval of size [−d, d] [Fra91; Wac99], 1 χd (∆n ) = Θ(d + ∆n )Θ(d − ∆n ) , (5.8) 2d and dichotomous or δ-distributed random variables [Rei94; Rei96a; Rei96b; Wac99], 1 χd (∆n ) = (δ(d − ∆n ) + δ(d + ∆n )) . (5.9) 2 Since |∆n | ≤ d, we denote the perturbation strength by d. As an example, we sketch in Fig. 5.4 our model for noisy slopes. We now define the diffusion coefficient as 1 (5.10) D(a, d) := lim (< x2n >ρ0 − < xn >2ρ0 ) , n→∞ 2n with Z Z < xkn >ρ0 =

dx

d(∆0 ) . . . d(∆n−1 )ρ0 (x)χ(∆0 ) . . . χ(∆n−1 )xkn , (5.11)

May 8, 2007

4:1

92

World Scientific Book - 9in x 6in

Deterministic diffusion and random perturbations

where ρ0 (x) denotes the initial distribution of an ensemble of moving particles, x0 = x, k ∈ N, and ∆j , j ∈ {1, . . . , n − 1}, is the random variable. D(a, d) has been computed by iterating Eqs.(5.6), (5.7) numerically for an ensemble of moving particles. Because of self-averaging [Bou90], it suffices to generate single series of random variables from Eqs.(5.8), (5.9) instead of evaluating all the integrals in Eq. (5.11). To obtain better numerical convergence, for noisy shifts the first moment squared in Eq. (5.10) was subtracted while Eqs. (5.8), (5.9) imply that the long-time average over the random variable ∆n does not yield any bias. In Chapter 2 we have shown that the unperturbed map L exhibits normal diffusion if a > 2, in Chapter 3 this result was generalized to the case of adding a bias b. Correspondingly, for the types of perturbations defined above diffusion should always be normal if (a − da) > 2, as was confirmed in simulations. Hence, in analogy to the previous section the central question is what happens to the parameter-dependent diffusion coefficient D(a, d) under variation of the two control parameters a and d in the case of the above two types of noise. We furthermore know from Chapters 2 and 3 that D(a, 0) , da = 0 is a fractal function of the slope a as a control parameter. This curve is displayed again in Fig. 5.4, together with results from computer simulations for uniformly distributed noisy slopes at different values of the perturbation strength da.5 As expected, the irregular structure gradually disappears by increasing da. Qualitatively the same result is obtained by applying noisy shifts [Kla]. Fig. 5.4 may be compared to the corresponding result for quenched slopes Fig. 5.1. Apart from numerical uncertainties, there are clear differences in the critical behavior close to the onset of diffusion. However, for small enough perturbation strength and large enough a the results look qualitatively similar indicating that in this limit quenched and annealed diffusion may be treated on the same footing. This statement is corroborated by a trivial approximation for the perturbed diffusion coefficient, which we motivate starting from dichotomous noisy slopes. Naive reasoning suggests that, at arbitrary fixed parameters a and da, the perturbed diffusion coefficient D(a, da) can be approximated by simply averaging over the unperturbed diffusion coefficient D(a, 0) at respective values of the slopes a − da and a + da yielding Dapp (a, da) = (D(a − da, 0) + D(a + da, 0))/2. It is straightforward to generalize this heuristic argument to any other type of uncorrelated noise 5 A maximum of 1,000,000 particles has been iterated up to 50,000 time steps each. In figures without error bars, the numerical error is less than the size of the symbols.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

93

5.2. Noisy dynamical systems

yielding the expression Dapp (p, d) =

Z

d(∆) χd (∆)D(p + ∆, 0) .

(5.12)

Here p is a vector of control parameters such as p = {a, b} in the case of the map above, d is the corresponding vector of perturbation strengths, and ∆ is the vector of perturbations such as ∆ = {∆a, ∆b} for noisy shifts and slopes. Further generalizations of this equation, for example, to arbitrary moments as defined in Eq. (5.11), are straightforward. Applying this formula to the case of quenched slopes discussed in the previous section [Kla02b] reproduces the diffusion coefficient approximation Eq. (5.5). The corresponding approximations for uniform noisy slopes are depicted in Fig. 5.4 as lines. They show that even for da = 1 the agreement between theory and simulations is excellent. This confirms that, in the limit described above, quenched and annealed disorder generating normal diffusion can indeed approximately be treated in the same way. Let us now look at the diffusion coefficient for a given value of a as a function of da. Fig. 5.4 shows that approximately at integer slopes the fractal diffusion coefficient D(a, 0) exhibits local extrema. Since Eq. (5.12) represents an average over the unperturbed solution in a local environment [a − da, a + da] it predicts local suppression and enhancement of diffusion at odd and even integer slopes respectively, under variation of the perturbation strength da. This has already been conjectured in Ref. [Kla96b] and has been verified in the previous section [Kla02b] for quenched slopes. We first check this hypothesis for noisy slopes around the local maximum of D(a, 0) at a = 7 distributed according to Eqs. (5.8), (5.9). Figs. 5.5 (a), (b) depict again results obtained from computer simulations in comparison to Eq. (5.12). As predicted, in both cases there is suppression of diffusion for small enough da. For dichotomous noise the perturbed diffusion coefficient increases on a coarse scale by exhibiting multiple, fractal-like suppression and enhancement on finer scales. For uniform perturbations there is a pronounced crossover from suppression to enhancement on a coarse scale, by again exhibiting oscillations on a fine scale. In both cases the agreement between simple theory and simulations is excellent for small enough da, whereas systematic deviations particularly in the case of dichotomous noise are visible for larger da. Related non-monotonicities for a deterministic diffusion coefficient perturbed by noise have been reported by Eckhardt et al. [Eck02]. Note that if a − ∆an < 2 particles get trapped within a box at a respective time step n, and that for a − ∆an < 1 the map is non-chaotic. In the

4:1

World Scientific Book - 9in x 6in

94

klages˙book

Deterministic diffusion and random perturbations

2

4 3.5

1.9

(a)

3

D(a,da)

D(a,da)

May 8, 2007

2.5 2

1.8 1.7 1.6

1.5 0

(b)

1.5 1

2

3 da

4

5

6

0

1

2

3 da

4

5

6

Fig. 5.5 Diffusion coefficient D(a, da) as a function of the perturbation strength da at slope a = 7 for noisy slopes distributed according to: (a) dichotomous noise Eq. (5.9), (b) uniform noise Eq. (5.8). The circles represent results from computer simulations, the bold lines are obtained from the approximation Eq. (5.12), the dashed lines represent the stochastic limit for the diffusion coefficient Eq. (5.17).

first case simulations and simple reasoning suggest that the perturbed map still exhibits normal diffusion. However, as soon as a−da < 1 numerical results indicate that there is no normal diffusion anymore [Kla]. This appears to be due to the contracting behavior of the non-chaotic map resulting in localization of particles. The oscillatory behavior of the diffusion coefficient in Fig. 5.5 (a) just below this transition point is not yet understood. Employing Eq. (5.12) we now analyze noisy shifts. The unperturbed two-parameter diffusion coefficient D(a, b, 0) has been calculated numerically exactly for the map under consideration in Chapter 3. Results for the perturbed diffusion coefficient D(a, db) = D(a, 0, db) are presented in Fig. 5.6 (a) for dichotomous noise and in (b) for uniform perturbations, both starting from D(a, 0) at a = 6. In both cases the perturbed diffusion coefficient exhibits strong enhancement of diffusion for da → 0 due to the fact that the unperturbed diffusion coefficient at a = 6 is approximately identical with a local maximum in the (a, b) parameter plane [Gro02]. For dichotmous perturbations it suffices to show results for 0 < db < 0.5 only. Translation and reflection symmetry of the map imply that this function is mirrored in the interval from 0.5 < db < 1 and that the full sequence in 0 < db < 1 is periodically repeated for higher values of db, cf. Chapter 3. As in the corresponding case of noisy slopes, the perturbed diffusion coefficient increases on a coarse scale by exhibiting multiple fractal-like suppression and enhancement on a fine scale. In the case of uniform perturbations there is a pronounced crossover to an approximately constant diffusion coefficient

4:1

World Scientific Book - 9in x 6in

klages˙book

95

5.2. Noisy dynamical systems

1.1 1.4 1.05 1

1.2

D(a,db)

D(a,db)

May 8, 2007

(a)

1

0.95 0.9

(b)

0.85 0

0.1

0.2

0.3 db

0.4

0.5

0

0.5

1

1.5 db

2

3

2.5

Fig. 5.6 Diffusion coefficient D(a, db) as a function of the perturbation strength db at slope a = 6 for noisy shifts distributed according to: (a) dichotomous noise Eq. (5.9), (b) uniform noise Eq. (5.8). The circles represent results from computer simulations, the bold lines are obtained from the approximation Eq. (5.12), the dashed line represents the stochastic limit for the diffusion coefficient Eq. (5.16).

for larger db. Before calculating the stochastic limit of the diffusion coefficient we provide an analytical justification for the heuristic approximation Eq. (5.12). For sake of simplicity, we demonstrate it only for noisy slopes, ∆n = ∆an . Noisy shifts as well as quenched disorder can be treated along the same lines. Let us start from the definition of the diffusion coefficient Eq. (5.10) where < xn >= 0. Let ∆an be uniformly distributed in [−da, da], ∆a0 = ∆a. In the case of da → 0 all random variables are bounded by ∆an = ∆a + , −2da ≤ ≤ 2da. We now put this expression into the perturbed equation of motion Eqs. (5.6), (5.7), as contained in Eq. (5.10), which we write as xn+1,a+∆an = Ma+∆an (xn ). As a first step we now take the limit → 0 resulting in the expression for the mean square displacement

<

x2n

>=

Z

=

Z

dx

Z

d(∆a)d(∆a1 ) . . . d(∆an−1 )ρ0 (x)χ(∆a)χ(∆a1 ) . . .

dx

Z

d(∆a)ρ0 (x)χ(∆a)x2n,a+∆a ( → 0) .

. . . χ(∆an−1 )x2n,a+∆an−1 (5.13)

As a second step we exchange the time limit contained in Eq. (5.10) with

May 8, 2007

4:1

96

World Scientific Book - 9in x 6in

klages˙book

Deterministic diffusion and random perturbations

the integration over d(∆a) yielding < x2n > Dapp (a, da) = lim n→∞ 2n Z Z x2n,a+∆a = d(∆a)χda (∆a) lim dxρ0 (x) n→∞ 2n Z = d(∆a)χda (∆a)D(a + ∆a, 0) ,

(5.14)

where we have used that the unperturbed diffusion coefficient was defined as Z D(a, 0) = lim dxρ0 (x)x2n,a . (5.15) n→∞

We have thus verified our previous approximation Eq. (5.12) for noisy slopes in the limit of small perturbation strength. A similar derivation can be carried out for noisy shifts arriving again at Eq. (5.12) in the case of very small perturbation strength. For quenched shifts it is known that a normal diffusion coefficient does not exist [Rad96], hence any approximation by Eq. (5.12) must fail. Indeed, it turns out that in this case taking the limit → 0 fundamentally changes the properties of the dynamical system and is thus not a valid operation [Kla]. Finally, we calculate the parameter-dependent stochastic diffusion coefficient related to the map with noisy slopes. Starting from the definition Eq. (5.15) the complete loss of memory in the unperturbed map is modeled by [Kla96b; Kla97] (i) replacing the distance xn a particle travels by n times the distance a particle travels at any single time step, n∆x = n(Ma (x) − x), and (ii) neglecting any memory effects in the probability density on the unit interval by assuming ρ0 (x) = 1. Then Eq. (5.15) yields

(a − 1)2 . (5.16) 24 As was shown in Refs. [Kla96b; Kla97], this equation correctly describes the asymptotic parameter dependence of the deterministic diffusion coefficient for a → ∞ thus explaining the increase of D(a, 0) in Fig. 5.4 on a coarse scale. On this basis, the corresponding result for noisy slopes is easily calculated by using Eq. (5.16) as the functional form for D(a + ∆a, 0) in the approximation Eq. (5.14) yielding Drw (a) =

Drw (a, da) = Drw (a, 0) + ∆a2 /c ,

(5.17)

where c = 24 for dichotomous noise Eq. (5.9) and c = 72 for uniform noise Eq. (5.8). Eq. (5.17) thus confirms the common sense expectation

May 8, 2007

4:1

5.2. Noisy dynamical systems

World Scientific Book - 9in x 6in

klages˙book

97

that noise should typically enhance diffusion and represents the stochastic limit of the perturbed deterministic diffusion coefficient. This equation is depicted in Fig. 5.5 (a), (b) in form of dashed lines. In case of dichotomous noise the correlations are obviously so large that even for large perturbation strength da there is no transition to the stochastic limit, whereas in the case of uniform noisy slopes the diffusion coefficient approaches the stochastic solution asymptotically in da thus verifying the existence of a transition from deterministic to stochastic diffusion. That such a distinct transition behavior exists in these models was already conjectured in Ref. [Kla96b]. Analogous calculations for noisy shifts yield Eq. (5.16) for all values of db reflecting the fact that for large enough a the stochastic diffusion coefficient should not depend on the bias. This result is shown in Fig. 5.6 (b) and again confirms an asymptotic approach of the diffusion coefficient to the stochastic limit under variation of db. Based on the known result of the existence of a fractal diffusion coefficient for the unperturbed D(a, b, 0) we conjecture that the typical transition scenario in this type of system consists of (multiple) suppression and enhancement of diffusion. We also note that Eqs. (5.16), (5.17) are closely related to the approximation outlined in Ref. [Gei82], and to the simple heuristic argument given by Reimann [Rei94; Rei96a; Rei96b] by which he explains the suppression of deterministic diffusion by noise in the climbing sine map near a crisis. We conclude this section with an outlook towards further applications of this simple theory. First of all, it would be interesting to study the problem of noisy maps with non-zero average bias along the same lines. In Chapter 3 it was shown that the unperturbed map does not exhibit linear response for b → 0, thus we conjecture that adding noise generates a transition to Ohm’s law. We secondly emphasize that our approach could be straightforwardly applied to diffusion in more complex models such as the standard map [Ott93; Lic92], particle billiards [Gas98a; Dor99] or the inertia ratchets discussed in Section 3.4. Finally, in physical experiments on classical diffusive transport in low-dimensional periodic arrays of scatterers such as antidot lattices [Wei91], a control parameter like the temperature may mimic the strength of random perturbations. Our results suggest that measuring parameter-dependent transport coefficients in such systems may reveal transition scenarios that are analogous to the ones described in this section.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

98

5.3

Deterministic diffusion and random perturbations ∗

Summary

(1) We started this chapter by studying the impact of spatial, locally symmetric disorder on deterministic diffusion in a model exhibiting a fractal diffusion coefficient. For two different types of independent and identically distributed random perturbations computer simulations revealed, as expected, that the fractal diffusion coefficient gradually “smoothes out” as a function of the perturbation strength. Interestingly, we observed a kind of phase transition for small enough control parameters in the perturbed system. Looking at diffusion for a fixed control parameter as a function of the perturbation strength we found non-trivial scenarios of suppression and enhancement of normal diffusion, which are again due to strong dynamical correlations. All these numerical results are well explained by a simple heuristic approximation that employs a well-known random walk formula for diffusion on disordered lattices, which we suitably transferred to our case of deterministic diffusion in a random environment. We mentioned that there exist crosslinks between our approach and a stochastic theory for surface diffusion. (2) Alternatively, we put time-dependent random perturbations, or “noise”, onto the same unperturbed map. The distributions of random variables were the same as used before, however, here we investigated both the impact of random slopes and random shifts, the latter breaking the symmetry of the system at a fixed time step. As before, computer simulations showed that the fractal diffusion coefficient changed into a smooth oscillatory structure by applying these types of noise, by eventually crossing over to a monotonically increasing function that is in full agreement with ordinary random walk theory. This crossover from deterministic to stochastic diffusion was displayed more clearly in the diffusion coefficient as a function of the perturbation strength. However, as in the case of quenched disorder the noise-dependent diffusion coefficient is no monotonous function but exhibits suppression and enhancement on small and large scales depending on the specific choice of control parameters. Again, we provided a simple analytical approximation for the randomly perturbed diffusion coefficient explaining these findings in relation to the unperturbed fractal diffusion coefficient. This approximation could be derived exactly and could straightforwardly be applied to diffusion in more realistic systems.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Chapter 6

From normal to anomalous diffusion

After having explored the most elementary types of dynamical systems generating deterministic transport, which are piecewise linear chaotic maps defined on a periodic lattice, we now investigate nonlinear generalizations of such models. These systems may exhibit intermittency and complicated bifurcation scenarios under parameter variation. In this chapter we show that such mechanisms lead to transitions from normal to anomalous diffusion, where the mean square displacement of an ensemble of moving particles is not linear in time anymore. In Section 6.1 we introduce the climbing sine map, which can be extracted from a driven nonlinear pendulum equation via discretization of time. We find that this model generates fractal hierarchies of normal and anomalous diffusive regions as functions of the control parameter. These structures are analyzed by applying the Taylor-Green-Kubo formula for diffusion discussed in the previous chapter. The final part of this section provides a more detailed analysis of the bifurcation scenario exhibited by this system. In Section 6.2 we study a paradigmatic model for intermittent transport, the Pomeau-Manneville map lifted onto the real line. This model exhibits subdiffusive behavior, which is analyzed within the framework of continuous time random walk theory, fractional diffusion equations and a generalized Taylor-Green-Kubo formula. This chapter reflects the author’s collaboration particularly with N.Korabel and A.V.Chechkin [Kor02; Kor04b; Kor04a; Kor07]. 6.1

Deterministic diffusion and bifurcations

In the previous chapters we have restricted our analysis to transport in dynamical systems that are piecewise hyperbolic. For one-dimensional 99

klages˙book

May 8, 2007

4:1

100

World Scientific Book - 9in x 6in

klages˙book

From normal to anomalous diffusion

maps the hyberbolicity follows from the fact that all these maps are piecewise linearly expanding [Dev89; dM93], for two-dimensional baker maps a more general definition of hyperbolic sets applies [Guc90; Dor99; Gas98a]. Hyperbolicity, and even more the related but stronger property of being Anosov [Eck85], usually indicates that the corresponding dynamics is somewhat “well-behaved” and very amenable to mathematical analysis. In the case of the maps discussed previously, the invariant probability densities of the associated circle maps, for example, were always piecewise constant functions along the unstable manifold, which to a large extent could be calculated analytically. This is not the case anymore for nonlinear maps losing the property of hyperbolicity. In the case of nonhyperbolic dynamics one often encounters singular invariant measures along unstable manifolds accompanied by complicated bifurcation scenarios under parameter variation [Ott93; All97; Sch82; Gro82; Fuj82]. Related to the chaotic hypothesis mentioned in the introduction Section 1.2, there is some discussion of whether hyperbolic or nonhyperbolic dynamical systems are more relevant in order to properly describe statistical physical systems in nonequilibrium situations [Gas98a; Gal99; Rue99b; Hoo99; Eva02b]. Here we introduce a typical example of a nonhyperbolic dynamical system exhibiting diffusive transport, the one-dimensional climbing sine map1 [Sch82; Gro82; Fuj82; Gei82; Gro83; Rei94]. This map can be derived from various types of driven nonlinear pendulum equations [Chi79; Hub80; Cir82; Kog83; Mir85; Bla96] by a discretization procedure [Boh84; Bak85] in complete analogy to the one outlined in Section 3.4 linking onedimensional maps with deterministic ratchets. For the hyperbolic maps analyzed in the previous chapters the transport coefficients usually existed and were fractal functions of control parameters. An important question is thus whether parameter-dependent diffusion in nonhyperbolic dynamical systems may as well exist and be of a fractal nature. The climbing sine map is defined by xn+1 = Ma (xn ) , Ma (x) := x + a sin(2πx) ,

(6.1)

where a ∈ R is a control parameter, x ∈ R, and xn is the position of a point particle at discrete time n. As in the case of map L, cf. Eqs. (2.4) and (2.5), Ma (x) possesses translation and reflection symmetry, Ma (x + p) = 1 If

restricted onto the circle, this map reduces to the sine circle map [Ott93]. However, here we are interested in the spatially continued lifted circle map that was coined climbing sine in the literature cited above.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

101

6.1. Deterministic diffusion and bifurcations

1

0.8

1

1.2

1.4

1.6

0.8

1

1.2

1.4

1.6

x 0.5

0 4 3

D(a)

2 1 0

a Fig. 6.1 Upper panel: bifurcation diagram for the climbing sine map restricted onto the circle. Lower panel: diffusion coefficient from simulations as a function of the con1 (a) trol parameter a in comparison with the correlated random walk approximation D 10 (dots), see Eq.(6.2) and below. The dashed vertical lines connect regions of anomalous diffusion, D(a) → ∞ or D(a) → 0, with ballistic and localized dynamics in respective windows of the bifurcation diagram.

Ma (x)+p, Ma (−x) = −Ma (x), p ∈ Z. The periodicity of the map naturally splits the real line into different cells (p, p + 1]. For what follows we will focus on parameters a > 0.732644 for which the extrema of the map exceed the boundaries of each cell for the first time indicating the onset of spatially extended motion. Fig. 6.1 shows the parameter-dependent diffusion coefficient D(a), which we computed from simulations by using the Taylor-Green-Kubo formula for maps Eq. (4.22) rewritten as 1 (6.2) Dn (a) = hja (x)Jan (x)i − hja2 (x)i . 2 We recall that the angular brackets denote an ensemble average over the invariant density on the unit interval and that ja (xn ) is the jump velocity giving the integer number of boxes a particle has traversed after one iteration

May 8, 2007

4:1

102

World Scientific Book - 9in x 6in

From normal to anomalous diffusion

P starting at position x. The jump velocity function Jan (x) := nk=0 ja (xk ) gives the integer value of the displacement of a particle after n time steps, which started at some initial position x = x0 . Eq. (6.2) defines a timedependent diffusion coefficient which, in the case of normal diffusion, converges to D(a) = limn→∞ Dn (a). In our simulations we truncated Jan (x) after having obtained enough convergence for D(a), that is, after 20 time steps. The invariant density was computed by solving the Frobenius-Perron equation for ρ(x) with the histogram method of Ref. [Lic92]. Fig. 6.1 includes the bifurcation diagram of the associated circle map, ma (x) := Ma (x) mod 1, which consists of infinitely many periodic windows. One can see that whenever there is a periodic window the diffusion coefficient of the map Eq. (6.1) either diverges or tends to zero. [Sch82; Gro82; Fuj82]. This demonstrates that the bifurcation scenario has a strong impact on the diffusion coefficient. D(a) = 0 corresponds to localized dynamics, that is, in that case all orbits are confined within some finite interval on the line. D(a) → ∞ means ballistic motion, where particles propagate superdiffusively with the diffusion coefficient being proportional to n. For the remaining “chaotic” parameter regions we find numerically that D(a) is nonzero and finite, thus the map is normal diffusive. At the boundaries of each periodic window there is transient intermittent-like behavior eventually resulting in normal diffusion with D(a) ∼ a±1/2 [Gei82; Gro83; Rei94; Sch82; Gro82; Fuj82]. We remark that a similarly complicated interplay between normal and anomalous diffusion under parameter variation has been numerically observed in a non-chaotic particle billiard [Alo04], where it was suggested to describe this irregularity by different notions of complexity [Jep06]. We now analyze the highly non-trivial parameter dependence of D(a) more quantitatively. As we explained in Section 4.2.2, the Taylor-GreenKubo formula Eq. (6.2) splits the dynamics into an inter-cell contribution, in terms of integer jumps, and into an intra-cell part, as represented by the invariant density. We first approximate the invariant density in Eq. (6.2) to ρ(x) ' 1 irrespective of the fact that it is a complicated function of x and a [Sch82; Gro82; Fuj82]. This approximate diffusion coefficient we denote with a superscript in Eq. (6.2), Dn1 (a). The term for n = 0 is well-known as the stochastic random walk approximation for maps, which excludes any higher-order dynamical correlations [Gei82; Gro83; Rei94; Sch82; Gro82; Fuj82; Kla96b; Kla97; Kla02a]. The generalization Dn1 (a) , n > 0, was called the correlated random walk approximation [Kla02d], and for hyperbolic maps it will be discussed in detail in Chapter 7. We now use this

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

103

6.1. Deterministic diffusion and bifurcations

systematic expansion to analyze the diffusion coefficient of the climbing sine map in terms of higher-order dynamical correlations. In Fig.6.2 (a) we depict results for Dn1 (a) at n = 1, . . . , 10. One clearly observes convergence of this approximation in parameter regions with nor1 mal diffusion. Indeed, a comparison of D10 (a) with D(a), as shown in Fig.6.1, demonstrates that there is qualitative agreement on large scales. Furthermore, for parameters corresponding to ballistic motion the sequence of Dn1 (a) diverges, in agreement with D(a) → ∞, whereas for localized dynamics it oscillates between two solutions. This oscillation reflects the dynamical origin of the localization generated by period-two orbits [Kor04b]. That the approximate solutions are non-zero is due to the fact that the invariant density was assumed to be uniform. In regions of normal diffusion this approximation matches quantitatively quite well and nicely reproduces many irregularities in the diffusion coefficient. Even more importantly, the magnifications in Fig. 6.2 give clear evidence for a fractal structure of the normal diffusion coefficient. We now further analyze the dynamical origin of the different types of diffusive behavior. According to its definition, the time-dependent jump velocity function Jan (x) fulfills the recursion relation [Kla96b] Jan (x) = ja (x) + Jan−1 (ma (x)) .

(6.3)

Jan (x) becomes extremely complicated after some time steps, thus, in line with Eq. (4.23), we introduce the more well-behaved function Z x Tan (x) := Jan (z) dz, Tan (0) = Tan (1) = 0 . (6.4) 0

For the nonlinear climbing sine map, integration of Eq. (6.3) yields instead of Eq. (4.24) the more complicated recursive functional equation Tan (x) = ta (x) +

1 m0a (x)

Tan−1 (ma (x)) − I(x)

(6.5)

containing the integral term I(x) :=

Z

ma (x) 0

dz g 00 (z)Tan−1 (z) ,

(6.6)

R 00 where ta (x) := dz ja (z), m0a (x) := dma (x)/dx, and g (z) is the second derivative of the inverse function of ma (x).2 For piecewise linear hyperbolic maps I(x) simply disappears and the derivative in front of the second term 2 In order to define an inverse of m (x), we split the unit interval into subintervals on a which this function is piecewise invertible; for further details see Ref. [Kor04b].

May 8, 2007

4:1

World Scientific Book - 9in x 6in

104

klages˙book

From normal to anomalous diffusion

4

(a)

(c)

0.04 0.02

3

Dn(a)

0 0.733 0.734 0.735 0.736 0.15 (b) 0.1 0.05 0 0.735 0.74 0.745 0.75 0.755

2

a

1

0

0.8

1

1.2

1.4

1.6

a 1 (a), see Eq. (6.2) Fig. 6.2 (a) Sequence of correlated random walk approximations Dn and below, for n = 1, · · · , 10. The dashed lines define the same periodic windows as in 1 (a) in the initial region of (a). Fig. 6.1. The insets (b) and (c) contain blowups of D10 They show self-similar behavior on smaller and smaller scales.

reduces to the local slope of the map thus recovering generalized Takagi-, respectively ordinary de Rham-type equations such as Eq. (4.24), see also Refs. [Gas98a; Dor99; Kla96b; Gas98c]. Instead of solving the generalized de Rham-equation for the climbing sine map directly, which is an open problem, solutions can alternatively be constructed from Eq. (6.4) on the basis of simulations. Results are shown in Fig. 6.3. For normal diffusive parameters the limit Ta (x) = limn→∞ Tan (x) exists and the respective curve is fractal over the whole unit interval somewhat resembling the generalized fractal Takagi functions shown in Fig. 4.3. In the case of periodic windows, Tan (x) either diverges due to ballistic flights or it oscillates indicating localization. Interestingly, in these functions the corresponding attracting sets appear in the form of smooth non-fractal regions on fine scales, whereas the other regions look fractal. The diffusion coefficient can now be formulated in terms of these fractal functions by integrating Eq. (6.2). For a ∈ (0.732644, 1.742726] we get D(a) = 2 [Ta (x2 )ρ(x2 ) − Ta (x1 )ρ(x1 )] − D0ρ (a) ,

(6.7)

where xi ∈ [0, 1/2] , i = 1, 2 , Ma (xi ) := 1 defines an escape region and

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

105

6.1. Deterministic diffusion and bifurcations

x 0 0.4

Ta

0.5

1

(a)

0.2 0 0.6

(c)

(b)

0.3

n

Ta

0.2 0.3

n

Ta

0.1 0 0

0.5

x

1

0.5

0 1

x

Fig. 6.3 Functions Tan (x) for the climbing sine map as defined by Eqs. (6.3) to (6.6): (a) diffusive dynamics at a = 1.2397 (upper curve) and at a = 1.7427 (lower curve), (b) ballistic dynamics at a = 1.0, and (c) localized dynamics at a = 1.5. In (a) the limiting case n → ∞ is shown, in (b) and (c) it was n = 5, 6, 7.

Rx D0ρ (a) := x12 dxρ(x). Our previous approximation Dn1 (a) is recovered from this equation as a special case. The intimate relation between periodic windows and the irregular behavior of the diffusion coefficient suggests investigation of the structure of the periodic windows in the climbing sine map in more detail. For nondiffusive unimodal maps periodic windows were already analyzed quite extensively [Met73; Jac78; Fei78; Fei79; Fei83], whereas little work exists for diffusive maps on the line [Sch82; Gro82; Fuj82]. The windows are generated by specific periodic orbits, there are infinitely many of them, and they are believed to be dense in the parameter set [Ott93]. Windows with ballistic dynamics are born through tangent bifurcations, further undergo Feigenbaum-type scenarios and eventually terminate at crises points. Windows with localized orbits only occur at even periods. They start with tangent bifurcations and exhibit a symmetry breaking at slope-type bifurcation points. In order to analyze the structure of the regions of anomalous diffusion

May 8, 2007

4:1

World Scientific Book - 9in x 6in

106

klages˙book

From normal to anomalous diffusion

25000 500 400

20000

300

15000

200

N

100

10000 0.75

0.8

0.9

0.85

N

0

a

5000

0.8

1

1.2

1.4

1.6

a Fig. 6.4 Devil’s staircase-like structure formed by the cumulative distribution of periodic windows as a function of the control parameter. N is the integrated number of period six-windows. The inset shows a blowup of the initial region.

we sum up the number of period six-windows as a function of the parameter, that is, the total number is increased by one for any parameter value at which a new period six-window is born. This sum forms a Devil’s staircaselike structure in parameter space indicating an underlying Cantor set-like distribution for the corresponding anomalously diffusive region, see Fig. 6.4. The Lebesque measure of periodic windows is obviously positive, hence this set must be a fat fractal [Yor85]. Its self-similar structure can quantitatively be assessed by computing the so-called fatness exponent [Far85]. Being interested in the parameter dependence of this fractal structure, we divide the parameter space into subsets labeled by the integer value of the map maximum on the unit interval, [Ma (xmax )] = j, j ∈ Z. For j = 1, 2, 3 we obtain a fatness exponent of 0.45 with errors of 0.03, 0.04, and 0.05 for the different j. This value was conjectured to be universal and was also obtained for non-diffusive unimodal maps [Far85]. We now study the measure of the windows as a function of the parameter. For this purpose we computed all windows up to period six for the first subset, up to period five for j = 2, 3, and we summed up their measure in the respective subsets. We find that the total measure decays exponentially as a function of j while oscillating with odd and even values of j on a finer

May 8, 2007

4:1

World Scientific Book - 9in x 6in

6.2. Anomalous diffusion in intermittent maps

klages˙book

107

scale [Kor04b]. This oscillation can be traced back to windows generated by localized dynamics. They appear at even periods only, thus contributing periodically to the total measure. However, different measures of ‘ballistic’ and ‘localized’ windows decay with the same rate. We have furthermore computed the complementary measure Cj of diffusive dynamics in the jth subset of parameters. We find that C1 = 0.783, C2 = 0.898, and C3 = 0.932 with an error of ±0.002, so the measure of the diffusive regions is always non-zero and seems to approach one with increasing parameter values. We conclude our analysis of the climbing sine map by suggesting relations to experiments. In line with Refs. [Kog83; Boh84; Bak85] we may argue that whenever a dynamical systems is governed by some driven nonlinear pendulum equation, in a suitable limit of parameter values where the velocity is approximately constant, which particularly is the case for strong damping, the stroboscopic dynamics of the onedimensional climbing sine map should be recovered. The climbing sine model may thus provide an access road for understanding parameterdependent diffusion in dissipative dynamical systems driven by periodic forces such as Josephson junctions in the presence of microwave radiation [BJ82; Cir82; Mir85; Mar89; Wei00], superionic conductors [Ful75; Bey76; Mar86] and systems exhibiting charge-density waves [Bro84] in which features of deterministic diffusion have already been observed experimentally. We also highlight the recent theoretical studies Refs. [Fes02; Tan02d], which focused on irregular parameter dependencies of diffusion in nonhyperbolic systems similar to the ones discussed in this chapter. We finally mention that the climbing sine map is of the same functional form as the nonlinear equation in the two-dimensional standard map, which serves as a standard model for many Hamiltonian dynamical systems [Lic92; Mei92; All97]. This map is derived via time-discretizing the kicked rotor, which is also a driven nonlinear pendulum, it is also strongly nonhyperbolic, and although the standard map is area-preserving it too exhibits a highly irregular parameter-dependent diffusion coefficient [Rec80; Rec81; Ott93]. A suitably adapted version of the approach outlined above may enable one to uncover the supposedly fractal nature of these irregularities.

6.2

Anomalous diffusion in intermittent maps

Anomalous diffusion derives from the fact that the mean square displacement of an ensemble of particles may not follow the law of normal diffusion,

May 8, 2007

4:1

108

World Scientific Book - 9in x 6in

From normal to anomalous diffusion

< x2 >∼ nα with α = 1. If α > 1 one says the dynamics is superdiffusive, if α < 1 it is subdiffusive. In the previous two sections we have encountered two extreme cases of anomalous diffusion: localization with α = 0 and ballistic flights corresponding to α = 2. We now show that there exists a specific dynamical mechanism for generating non-integer exponents α 6= 1. The study of anomalous diffusion is currently a very active field of research. This dynamics was observed in a wide range of theoretical models and experiments such as, for example, in turbulence, amorphous semiconductors, porous media, surface diffusion, glasses, granular matter, reaction-diffusion processes, plasmas, biological cell motility and econophysics, see Refs. [Bou90; Shl93; Shl94; Kla96a; Met00; Sok02; Zas02; Met04a; Bal05; Ebe05] for reviews. Parallel to this development, low dimensional nonhyperbolic maps attracted much attention as simple models of anomalous dynamics [Pom80; Man80; Gei84; Gei85; Shl85; Gas88; Wan89b; Wan89a; Wan93; Art93; Art97a; Cvi07; Zum93b; Zum93a; Zum95; Sto95b; Det97a; Det98; Dr¨ a00; Tas02; Tas04; Bar03a; Art03; Art04; Rad04]. A characteristic feature of the stochastic processes generated by these maps is provided by the probability density functions of the dynamical variables. In the case of normal diffusion these probability densities exhibit Gaussian forms, whereas for subdiffusion they yield tails with stretched exponential decay and for superdiffusion L´evy power laws [Zum93b]. Several theoretical approaches have been worked out over the past few decades in order to explain anomalous diffusion. Perhaps the most famous one is continuous time random walk (CTRW) theory, pioneered by the work of Montroll, Weiss and Scher [Mon65; Mon73; Sch75]. Their stochastic approach was later on adapted to sub- and superdiffusive deterministic maps [Gei84; Gei85; Shl85; Zum93b; Zum93a; Zum95; Dr¨ a00; Bar03a]. A classification of their dynamics was provided particularly by Gaspard and Wang combining stochastic with dynamical systems theory [Gas88]. Exploiting the deterministic properties of these maps, anomalous diffusion was further studied in the framework of the thermodynamic formalism [Wan89b; Wan89a; Wan93; Sto95b], by means of periodic orbit theory [Art93; Art97a; Cvi07; Det97a; Det98; Art03; Art04] and by spectral decomposition techniques [Tas02; Tas04]. Currently aging phenomena [Bar03a; Rad04], non-ergodic behavior [Lut04; Bel05; Bel06] and infinite invariant measures [Hu95; Zwe98; Aar97] are in the focus of investigations demonstrating that these models provide ongoing inspiration for important new research. As an example of a model which exhibits fully anomalous dynamics

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

6.2. Anomalous diffusion in intermittent maps

klages˙book

109

without displaying a complicated bifurcation scenario we study subdiffusion in a map whose functional form on the unit interval was introduced by Pomeau and Manneville for describing intermittency [Pom80; Man80], 1 , (6.8) 2 where the parameter z ≥ 1 stands for the degree of nonlinearity and a ≥ 1 is a second control parameter. As usual, this map is continued onto the real line by the translation Ma,z (x + m) = Ma,z (x) + m, m ∈ Z, and by reflection symmetry, Ma,z (−x) = −Ma,z (x). Note that for z = 1 this model reduces to the piecewise linear map studied in Chapter 2. The model and its dynamics are illustrated in Fig. 6.5. As was shown in Refs. [Gei84; Gas88; Zum93b], for 1 ≤ z < 2 the diffusion process is normal whereas for z ≥ 2 it is anomalous. The anomaly is due to the existence of marginally stable fixed points of slope |M 0 (x)| = 1 located at all integer values of xn around which a particle gets trapped for a long time [Wan89b; Wan89a; Wan93]. However, in regions with larger local derivative, which is where the particle can jump to different unit intervals, the dynamics becomes more irregular, see Fig. 6.5. Thus, a typical trajectory of the spatially continued map Eq. (6.8) consists of long laminar pieces interrupted by short chaotic bursts. Such a behavior is the hallmark of what is called intermittency [Pom80; Man80; Sch89]. An example of a typical intermittent trajectory generated by this map is shown at the bottom of Fig. 6.5. In the groundbreaking works of Geisel, Thomae [Gei84], Zumofen and Klafter [Zum93b], main attention was paid to the time dependence of the mean square displacement for specific fixed values of the parameter a. In line with the main theme of Part 1 of this book, here we study how a suitably defined generalized diffusion coefficient behaves under variation of the two control parameters a and z. We define this quantity by [Met00; Kos05a; Kos05b]

(xn − x0 )2 . (6.9) K(a, z) := lim n→∞ nα As in Eq. (2.2), the angular brackets denote an ensemble average.3 We first focus on numerical simulations4 of the map Eq. (6.8). CTRW theory predicts that α = 1 for 1 ≤ z < 2 and α = 1/(z − 1) for 2 < z irrespective of the parameter a of the map [Gei84; Zum93b]. Indeed, for all a we find xn+1 = Ma,z (xn ) = xn + axzn , 0 ≤ xn <

3 Note that for convenience the factor two in the denominator has been absorbed into K [Kor05; Kor07]. 4 All simulations were performed starting from a uniform, random distribution of 10 6 initial conditions on the unit interval by iterating for n = 104 time steps.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

110

klages˙book

From normal to anomalous diffusion

M3

2

1

0 -1

0 -1

1

2 x

x0

-2

10

x 5 0 0

2000

n

4000

Fig. 6.5 top: Illustration of the spatially continued Pomeau-Manneville map M Eq. (6.8) for the parameters z = 3 and a = 8 with a typical trajectory generated by the map (cobweb plot with dashed horizontal and vertical lines) starting at initial condition x0 . bottom: The demonstration of intermittency in a time series plot of a typical trajectory. Note the long laminar periods in the trajectory which are interrupted by chaotic bursts.

excellent agreement between these solutions and the values for α obtained from simulations. Consequently, the analytical expressions for α are used in the following. Results for K as a function of a for fixed z are presented in Fig. 6.6. Magnifications of Fig. 6.6 (a) shown in parts (b) and (c) reveal self similarlike irregularities indicating a fractal parameter dependence of K. One may wish to compare these numerical results with Fig. 2.5 representing the normal diffusion coefficient for the same map at z = 1. With respect to this relation, the irregular parameter dependence of K as a function of a

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

111

6.2. Anomalous diffusion in intermittent maps

(a) 20

K 10

0 10

20

a 30

4

1

(b) K

50

(c)

K 0.5

2

0

40

10

a

20

0

5

a

10

Fig. 6.6 The generalized diffusion coefficient K Eq. (6.9) as a function of a for z = 3. The curve in (a) consists of 1200 points, the dashed-dotted line displays the CTRW result K1 , Eqs. (6.19), (6.21). (b) (600 points) and (c) (200 points) show magnifications of (a) close to the onset of diffusion. The dotted line in (b) is the CTRW approximation K2 , Eqs. (6.20), (6.21). The dashed line represents the first term of the TGK formula Eq. (6.22). The triangles mark a specific structure appearing on finer and finer scales. The inset in (a) depicts again the model Eq. (6.8).

for nonlinear values of z is not too surprising. Note particularly the structure marked by triangles, which is repeated on finer and finer scales. The parameter values for these symbols correspond to specific series of Markov partitions, just in the spirit of the turnstile dynamics analysis performed in Section 2.2 for the normal diffusive case, cp. Fig. 2.6. In Fig. 6.7 K is depicted as a function of z for different values of a again displaying highly non-monotonic parameter dependences. Further details of this figure will be discussed later on. It is interesting to interprete these numerical results within the framework of stochastic theory. Let us start with a brief outline of the idea of a CTRW approach to maps; see Refs. [Bou90; Wei94; Ebe05] for introductions to CTRW theory. The basic assumption is that diffusion can be decomposed into two stochastic processes characterized by waiting times

4:1

World Scientific Book - 9in x 6in

112

klages˙book

From normal to anomalous diffusion

0.6

30

γ

2

n /<x >

May 8, 2007

15

0.4 0 0 10

K

2

4

10

10

n 0.2

0

2

2.5

3

z Fig. 6.7 K as a function of z for a = 3.14, 4, 5 consisting of 200, 250 and 300 points, respectively (bottom to top). Bold lines represent simulation results for n = 10 4 iterations, the dashed line is the CTRW solution K2 , Eqs. (6.20), (6.21), for n → ∞. The thin line displays the first term of the TGK formula Eq. (6.22). At z = 2 there is a transition from normal to anomalous diffusion. The inset describes the time-dependence of < x 2 > around this value with a = 3.14. Here dashed lines correspond to z = 1.5, 1.9, 1.95 (bottom to top), dashed-dotted lines to 2.5, 2.1, 2.05 (top to bottom) while the bold line depicts z = 2. The dotted line is proportional to ln n.

and jumps, respectively. Thus one has two sequences of independent identically distributed random variables, namely a sequence of positive random waiting times T1 , T2 , T3 , . . . with probability density function w(t) and a sequence of random jumps ζ1 , ζ2 , ζ3 , . . . with a probability density function λ(x). For example, if a particle starts at point x = 0 at time t0 = 0 and makes a jump of length ζn at time tn = T1 + T2 + ... + Tn , its position is x = 0 for 0 ≤ t < T1 = t1 and x = ζ1 + ζ2 + ... + ζn for tn ≤ t < tn+1 . The probability that R t at least one jump is performed within the time interval [0, t) is then 0 dt0 w(t0 ) while the probability for no jump during this time Rt interval reads Ψ(t) = 1 − 0 dt0 w(t0 ). The master equation for the probability density function P (x, t) to find a particle at position x and time t is then Z ∞ Z t P (x, t) = dx0 λ(x − x0 ) dt0 w(t − t0 ) P (x0 , t0 ) + Ψ(t)δ(x) . (6.10) −∞

0

It has the following probabilistic meaning: The probability density function

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

113

6.2. Anomalous diffusion in intermittent maps

to find a particle at position x at time t is equal to the probability density function to find it at point x0 at some previous time t0 multiplied with the transition probability to get from (x0 , t0 ) to (x, t) integrated over all possible values of x0 and t0 . The second term accounts for the probability of remaining at the initial position x = 0. The most convenient representation of this equation is obtained in terms of the Fourier-Laplace transform of the probability density function, Z ∞ Z ∞ ˜ (k, s) = Pˆ dx eikx dt e−st P (x, t) , (6.11) −∞

0

where the hat stands for the Fourier transform and the tilde for the Laplace transform. Respectively, this function obeys the Fourier-Laplace transform of Eq. (6.10), which is called the Montroll-Weiss equation [Mon65; Mon73; Sch75], 1 ˜ ˜ (k, s) = 1 − w(s) Pˆ ˆ s 1 − λ(k)w(s) ˜

.

(6.12)

The Laplace transform of the mean square displacement can be readily obtained by differentiating the Fourier-Laplace transform of the probability density function, Z ∞ ˜ (k, s) ∂ 2 Pˆ 2˜ ˜ 2 . (6.13) hx (s)i = dx x P (x, s) = − ∂k 2 −∞ k=0

In order to calculate the mean square displacement within this theory, it thus suffices to know λ(x) and w(t) generating the stochastic process. For one-dimensional maps of the type of Eq. (6.8), exploiting the symmetry of the map the waiting time distribution can be calculated from the approximation dxt = axzt , x 1 , (6.14) xn+1 − xn ' dt where we have introduced the continuous time t ≥ 0. This equation can easily be solved for xt with respect to an initial condition x0 . Now one needs to define when a particle makes a “jump”, as will be discussed below. By inverting the solution for xt , one can then calculate the time t a particle has to wait before it makes a jump as a function of the initial condition x0 . This information determines the relation between the waiting time probability density w(t) and the as yet unknown probability density of injection points, dx0 . w(t) ' Pin (x0 ) (6.15) dt

May 8, 2007

4:1

114

World Scientific Book - 9in x 6in

klages˙book

From normal to anomalous diffusion

Making the assumption that the probability density of injection points is uniform, Pin ' 1, the waiting time probability density is straightforwardly calculated from the knowledge of t(x0 ). The second ingredient that is needed for the CTRW approach is the jump probability density. Standard CTRW theory takes jumps between neighbouring cells only into account leading to the ansatz [Gei84; Zum93b] λ(x) = δ(|x| − 1) .

(6.16)

It turns out that a correct application of this theory to our results for K(z) requires a modification of the standard theory [Gei84; Zum93b] at three points: Firstly, the waiting time probability density function must be calculated according to the grid of elementary cells indicated in Fig. 6.6 [Kla96b; Kla97] yielding w(t) = a (1 + a(z − 1)t)

z − z−1

.

(6.17)

However, this probability density function also accounts for attempted jumps to another cell, since after a step the particle may stay in the same cell with a probability of (1 − p). The latter quantity is roughly determined by the size of the escape region p = (1 − 2xc ) with xc as a solution of the equation xc + axzc = 1. We thus model this fact, secondly, by a jump length distribution in the form of p (6.18) λ(x) = δ(|x| − l) + (1 − p)δ(x) . 2 Thirdly, we introduce two definitions of a typical jump length li , i ∈ {1, 2}. l1 = {|Ma,z (x) − x|}

(6.19)

corresponds to the actual mean displacement while l2 = {|[Ma,z (x)]|}

(6.20)

gives the coarse-grained displacement in units of elementary cells, as it is often assumed in CTRW approaches. In these definitions {. . .} denotes both a time and ensemble average over particles leaving a box. Working out the modified CTRW approximation sketched above by taking these three changes into account we obtain for the generalized diffusion coefficient γ a sin(πγ)/πγ 1+γ , 0 < γ < 1 2 , (6.21) Ki = pli a(1 − 1/γ) , 1≤γ<∞ where γ := 1/(z − 1), which for z ≥ 2 is identical with α defined before. Fig. 6.6 (a) shows that K1 well describes the coarse functional form of K for large a. K2 is depicted in Fig. 6.6 (b) by the dotted line and is

May 8, 2007

4:1

World Scientific Book - 9in x 6in

6.2. Anomalous diffusion in intermittent maps

klages˙book

115

asymptotically exact in the limit of very small a. Hence, the generalized diffusion coefficient exhibits a dynamical crossover between two different coarse grained functional forms for small and large a, respectively. An analogous crossover has been reported earlier for normal diffusion [Kla96b; Kla97] and was also found in Hamiltonian particle billiards [Kla00a]. Let us now focus on the generalized diffusion coefficient at a = 12, 20, 28, ..., which corresponds to integer values of the height h = [Ma,z (1/2)] of the map. Simulations reproduce, within numerical accuracy, the results for K2 by indicating that K is discontinuous at these parameter values. Due to the self similar-like structure of the generalized diffusion coefficient we arrive at the conjecture that K exhibits infinitely many discontinuities on fine scales as a function of a, in contrast to the continuity of the CTRW solution [Kor07]. We now switch back to the z-dependence of K, where CTRW theory predicts a dynamical phase transition [Bec93] from normal to anomalous diffusion at z = 2 [Gei84; Zum93b]. At the transition point one has < x2 >∼ n/ log(n). According to Eq. (6.9), Eq. (6.21) yields a continuous transition in z, see K2 in Fig. 6.7. However, this coarse functional form is obscured not only by fractal irregularities but also by an ultra-slow convergence of the mean square displacement in our simulations. Using CTRW theory, it can be verified [Kor07] that around z = 2 the mean square displacement is determined by a series of logarithmic corrections in time, where the lowest-order term is proportional to nγ / log n and dominates at time scales controlled by n n ˜ ∼ exp(1/(1−γ)). Hence, for γ 6= 1 and n n ˜ simple power laws are recovered while for γ = 1 the logarithmic term survives in the infinite time limit. The inset of fig. 6.7 exemplifies this behavior explaining the deviations between CTRW theory and the simulation results in the main figure. In other words, around z = 2 both normal and anomalous diffusion are suppressed, due to logarithmic corrections in time leading to a vanishing K. We strongly suspect that such a behavior of the generalized diffusion coefficient is typical for dynamical phase transitions in anomalous dynamics altogether [Zum93b]. We also remark that the ultra-slow convergence of the numerical results is connected with the aging phenomenon for z > 2 studied in Ref. [Bar03a], which indicates some dependence on initial conditions for large transient times while disappearing in the limit of t → ∞. The stochastic CTRW theory presented above describes only the coarse parameter dependence of the generalized diffusion coefficient. This motivates us to develop an approach that incorporates dynamical correlations.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

116

From normal to anomalous diffusion

For this purpose we generalize the Taylor-Green-Kubo formula, which was introduced in Section 4.2.2 and used in Section 6.1, to anomalous diffusion. We start by expressing Eq. (6.9) via sums over the integer velocities jm = [xm+1 ] − [xm ] [Kla96b; Dor99; Kla02d]. However, in contrast to normal diffusion the anomalous dynamics generated by Eq. (6.8) is not stationary for z ≥ 2 [Gas88; Zum93a], that is, there exists no non-trivial invariant density for the Pomeau-Manneville map defined on the circle [Hu95; Zwe98; Aar97]. Consequently, the resulting expression cannot be simplified using time-translational invariance, and we get " n−1 # n−1 X X n−1 X 1 2 jk > +2 < jk jk+l > . (6.22) K = lim α < n→∞ n k=0

k=0 l=1

Focusing on the first term only, simulations confirm that it is proportional to nα . For z < 2 the dynamics is ergodic, hence this term is equal to n < j02 >, and we recover the random walk formula for normal diffusion [Kla96b; Kla97]. For z ≥ 2 only generalized ergodic theorems may hold [Aar97], and whether the CTRW result Eq. (6.21) can be derived from the first term is an open question. Considering this term as an approximation of K for all z, the numerical results are depicted in Figs. 6.6 (b) and 6.7. The comparison of this data with the simulation values based on Eq. (6.9) indeed shows that this term already provides a first step beyond the modified CTRW theory: In the a dependence it reproduces the major irregularities of K and even appears to follow the discontinuities discussed above. However, we cannot say whether it yields exact values for K at integer heights. Eq. (6.22) thus provides a suitable starting point for a systematic understanding of the fractal generalized diffusion coefficient beyond CTRW theory. We remark that similarly to our analysis of normal diffusion outlined in Section 4.2.2, via the anomalous Taylor-Green-Kubo formula Eq. (6.22) the generalized diffusion coefficient can be related to de Rhamtype fractal functions [Kor07]. As discussed before, this yields another microscopic interpretation of the origin of the fractality of diffusion coefficients. One could also compute higher-order terms in the series expansion of Eq. (6.22), as was demonstrated for the climbing sine map in Section 6.1, thus systematically recovering more and more structure in the generalized diffusion coefficient. Another interesting detail is that the correlation function in the second term of Eq. (6.22) depends on two time scales, which enables one to express K in terms of aged [Bar03a; Rad04] de Rham-type fractal functions. We now turn to the probability density functions generated by the map

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

6.2. Anomalous diffusion in intermittent maps

klages˙book

117

Eq. (6.8). The great success of CTRW theory derives from the fact that it correctly predicts both the power α and the form of the coarse grained probability density function P (x, t) of displacements for a large class of models [Zum93b]. Correspondingly, the diffusion process generated by Eq. (6.8) is not described by an ordinary diffusion equation but by a fractional generalization of it. Starting from the CTRW model for the map Eq. (6.8) discussed above, one can derive the time-fractional diffusion equation [Kor07] Z t 0 0 ∂ γ P (x, t) ∂P 1 ∂2P dt (t − t )−γ 0 = D 2 := (6.23) γ ∂t Γ(1 − γ) 0 ∂t ∂x

describing the long-time limit of the probability density function of the coarse grained dynamics with initial condition P (x, 0) = δ(x) and D = KΓ(1 + γ)/2, 0 < γ < 1. The fractional derivative is understood in the Caputo sense [Mai97]. Time-fractional equations of such a form have already been extensively studied by mathematicians [Gor00]. We remark that two other fractional diffusion equations proposed in Refs. [Met00; Bar03a], which are based on a Riemann-Liouville fractional derivative, are equivalent to Eq. (6.23) under rather weak assumptions [Che03a]. The solution of Eq. (6.23) expressed in terms of an M-function of Wright type [Mai97; Gor00] reads γ 1 (6.24) M ξ, P (x, t) = √ 2 2 Dtγ/2

giving exactly the same asymptotics √ that was obtained in Ref. [Zum93b] for small and large values of ξ := |x|/ Dtγ/2 , ( 1 − a1 ξ + a2 ξ 2 , ξ 1 −γ/2 γ−1 , (6.25) P (x, t) ' t 2 , ξ1 ξ 2−γ e−b1 ξ 2−γ

where a1 , a2 , b1 are some constants. By using a series representation of M it can be demonstrated [Che03a] that Eq. (6.24) is equivalent to those expressed via H-functions [Met00] or one-sided extremal L´evy stable distributions [Bar03a]. Fig. 6.8 demonstrates an excellent agreement between the analytical solution Eq. (6.24) and the probability density function obtained from simulations of the map Eq. (6.8) if the probability density is coarse grained over unit intervals. However, it also shows that the coarse graining eliminates a periodic fine structure that is not captured by Eq. (6.24). As in the case of Fig. 3.9, see also Refs. [Kla96b; Kla02a; San05], this fine structure derives from the “microscopic” probability density function of an elementary cell (with periodic boundaries) as represented

4:1

World Scientific Book - 9in x 6in

118

klages˙book

From normal to anomalous diffusion

0

10

2

r 1 10

-1

0

0

0.5

x

1

Log P

May 8, 2007

10

-2

-3

10

-20

0 x

20

Fig. 6.8 Comparison of the probability density obtained from simulations of the map Eq. (6.8) (oscillatory structure) with the analytical solution Eq. (6.24) of the fractional diffusion equation Eq. (6.23) (continuous line in the middle) for z = 3 and a = 8. The probability density was computed from 107 particles after n = 103 iterations. For the generalized diffusion coefficient in Eq. (6.24) the simulation result was used. The crosses (x) represent the numerical results coarse grained over unit intervals. The upper and the lower curves correspond to fits with a stretched exponential and a Gaussian distribution, respectively. The inset depicts the probability density function for the map on the unit interval with periodic boundaries.

in the inset of Fig. 6.8. The singularities are due to the marginal fixed points of the map, where particles are trapped for long times. Remarkably, that way the microscopic origin of the intermittent dynamics is reflected in the shape of the probability density function on the whole real line: From Fig. 6.8 it is seen that the oscillations in the probability density are bounded by two functions, the upper curve being of a stretched exponential type while the lower is Gaussian. These two envelopes correspond to the laminar and chaotic parts of the motion, respectively.5 In this section we have only studied a one-dimensional subdiffusive map, however, we expect these findings to be typical for spatially extended, lowdimensional, anomalous deterministic dynamical systems altogether. The appearance of dynamical phase transitions is an interesting detail that may 5 The

two envelopes shown in Fig. 6.8 represent fit functions with the Gaussian a0 exp(−x2 /a1 ) and with the M-function b0 M (|x| /b1 , γ2 ), where a0 = 0.0036, a1 = 55.0183 and b0 = 3.05, b1 = 0.37.

May 8, 2007

4:1

6.3.

World Scientific Book - 9in x 6in

∗ Summary

klages˙book

119

deserve further study in terms of the generalized diffusion coefficient. An important point is to further investigate the origin of anomalous diffusion in terms of dynamical systems theory, with the long-term goal of arriving at a microscopic theory of anomalous deterministic transport. Further studies along these lines may focus on the relation between the anomalous Taylor-Green-Kubo formula and CTRW theory, on a spectral analysis of the Frobenius-Perron operator of this model and on analyzing a superdiffusive map by similar methods.

6.3

∗

Summary

(1) The climbing sine map defines a simple one-dimensional nonhyperbolic dynamical system exhibiting both normal and anomalous diffusive behavior. We calculated the parameter-dependent diffusion coefficient of this model showing that it exhibits a highly non-trivial interplay between normal and anomalous diffusive parameter regions. The normal diffusion coefficient again turned out to be fractal. Using a TaylorGreen-Kubo formula, both the normal and the anomalous dynamics could be linked to generalized fractal Takagi functions for which a functional recursion relation was derived. Thus, nonhyperbolic behavior not only amplifies fractal structures but generates even more complex fractal characteristics in deterministic diffusion under parameter variation. (2) In order to better understand the structure of the anomalous diffusive regions as functions of the control parameter we studied the emergence of periodic windows in this map. We found two different series of windows, one related to localization of particles and another one displaying ballistic flights leading to superdiffusion. Both types of windows form fractal sets in the normal diffusive regions as functions of the control parameter. However, the Lebesque measure of the normal diffusive areas eventually appears to approach one in the limit of large parameter values. (3) We then broadened our analysis of anomalous deterministic diffusion by introducing a subdiffusive version of the Pomeau-Manneville map, a standard model of intermittency. Studying a suitably generalized diffusion coefficient as a function of two control parameters we encountered fractal structures, whose dynamical origin is of the same type as the ones in the normal diffusion coefficient. We described in detail a dynamical phase transition from normal to anomalous diffusion,

May 8, 2007

4:1

120

World Scientific Book - 9in x 6in

From normal to anomalous diffusion

which is due to logarithmic corrections in the mean square displacement leading to a complete suppression of diffusion right at the transition point. Both this suppression and the coarse functional form of the generalized diffusion coefficient could be explained by a suitably adapted version of stochastic continuous time random walk theory. A refined analysis capturing part of the fractal structure of the generalized diffusion coefficient was suggested by generalizing the Taylor-Green-Kubo formula discussed previously to anomalous diffusion. We finally analyzed the probability density function of the position variables whose coarse-grained non-Gaussian form matched nicely to the solution of a fractional diffusion equation.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

Chapter 7

From diffusive maps to Hamiltonian particle billiards

In the previous chapters we have shown that low-dimensional periodic arrays of scatterers with moving point particles, modeled by time-discrete chaotic maps, typically exhibit transport coefficients that are fractal functions under variation of control parameters. This chapter demonstrates how to directly apply the knowledge gained by the analysis of simple maps to physically more realistic systems, which in this case are Hamiltonian particle billiards. We exemplify this connection by proposing a systematic scheme of how to approximate fractal diffusion coefficients in terms of a hierarchy of correlated random walks. In Section 7.1 we develop the theory for our standard model, the diffusive piecewise linear map L introduced in Chapter 2. In Section 7.2 we apply a suitably adapted version of this approach to diffusion in the periodic Lorentz gas [Kla02d]. 7.1

Correlated random walks in maps

Let us recall that one-dimensional diffusive maps of the type of map L are defined by the equation of motion xn+1 = Ma (xn ), see Chapter 2. Let ma (x) := Ma (x)

mod 1

(7.1)

be the reduced map related to Ma (x). This map governs the dynamics on the unit interval according to xn = mna (x). Let ρn (x) be the probability density on the unit interval of an ensemble of moving particles starting at initial conditions x = x0 , which evolves according to the Frobenius-Perron equation, see Eq. (2.7). Let the deterministic diffusion coefficient be defined by the Einstein formula Eq. (2.2). As was already mentioned in Chapters 4 and 6, this definition can be transformed exactly onto the Taylor-Green-Kubo formula 121

4:1

World Scientific Book - 9in x 6in

122

(a)

(b) 0.3

a

D(a)

0.2

precise D(a) random walk approx. 1st order approximation 2nd order approximation 3rd order approximation

0.1

0 2

klages˙book

From diffusive maps to Hamiltonian particle billiards

0.3

D(a)

May 8, 2007

2.5

3 a

3.5

0.2

precise D(a) 0th order approximation 1st order approximation 2nd order approximation

0.1

4

0 2

2.5

3 a

3.5

4

Fig. 7.1 Diffusion coefficient D(a) for the one-dimensional map shown in the upper left part of (a), where a is the slope of the map. The dots are obtained from the method described in Chapter 2. The different lines correspond to different levels of approximations based on the Taylor-Green-Kubo formula Eq. (7.2). In (a) the approximations were computed from Eq. (7.7) assuming a constant invariant density, in (b) they are from Eq. (7.8) which includes the exact invariant density. Any error bars are smaller than visible.

for maps, see Eqs. (4.22),(6.2), which for convenience here we rewrite as Dn (a) =

n X k=0

ck hj(x0 )j(xk )i

with D(a) = limn→∞ Dn (a) and 1 , k=0 ck = 2 1 , k≥1

.

(7.2)

(7.3)

As in Chapters 4 and 6, we need the jump velocity j(xn ) := [xn+1 ] − [xn ] ,

(7.4)

[x] being the largest integer less than x, which takes only integer values and denotes how many unit intervals a particle has traversed after one iteration starting at xn . From now on we specialize again onto map L defined by Eqs. (2.3) to (2.6), which is sketched in Fig. 7.1 (a). This figure furthermore depicts a magnification of the parameter-dependent diffusion coefficient of the map in the regime 2 ≤ a ≤ 4, cp. to Fig. 2.5, as calculated in Chapters 2 to 4. As explained in Chapters 4 and 6, Eq.(7.2) forms a suitable starting point for an analysis of the fractal diffusion coefficient, because it distinguishes between two crucial contributions of the dynamics to the diffusion process. These are, firstly, the motion of the particle on the unit interval xn mod 1 generating the invariant density ρ∗a (x), and, secondly, the integer jumps

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

123

7.1. Correlated random walks in maps

from one unit interval to another one as related to j(xn ). In Chapter 4 it was argued that both parts provide independent sources of fractality for the diffusion coefficient. However, there Eq. (7.2) was only discussed in the limit of infinite time. Chapter 6, in turn, provided an analysis of this formula for intermediate time steps, but this application concerned the complicated case of intertwined normal and anomalous diffusion as it exists in the climbing sine map and was essentially a numerical one. In this chapter we introduce the Taylor-Green-Kubo expansion in a more pedagogical way by working it out for the simple example of map L. We elucidate the specific contributions of all single terms to the fractality of the normal diffusion coefficient and investigate the convergence of the Taylor-Green-Kubo formula towards the fractal diffusion coefficient. For this purpose we build up hierarchies of approximate diffusion coefficients. These approximations are defined such that the different dynamical contributions to the diffusion process are properly filtered out. Another issue is how to evaluate the single terms of the Taylor-Green-Kubo expansion on an analytical basis. In this section we show in detail how to do this for map L. In the next section we apply the same idea to particle billiards such as the periodic Lorentz gas. In the following two chapters the same idea will be used for studying diffusion in more complicated billiards, which are partially motivated by physical experiments. We start by looking at the first term in Eq.(7.2). For a < 4 the absolute value of the jump velocity j(xn ) is either zero or one. Assuming that ρ∗a (x) ' 1 for a → 2 and cutting off all higher order-terms in Eq. (7.2), the first term leads to the well-known random walk approximation of the diffusion coefficient [Kla96b; Fuj82; Gei82; Sch82; Kla97; Kla02a] 1 D0 (a) = 2

Z

1

dx j 2 (x)

,

(7.5)

0

which in the case of the map Eq. (2.6) reads D0 (a) =

a−2 2a

.

(7.6)

This solution is asymptotically correct in the limit of a → 2. More generally speaking, the reduction of the Taylor-Green-Kubo formula to the first term only is an exact solution for arbitrary parameter values only if all higherorder contributions from the velocity autocorrelation function C(n) :=< j(x0 )j(xn ) > are strictly zero. This is only true for systems of Bernoulli type [Fuj82]. Conversely, the series expansion in the form of Eq. (7.2)

May 8, 2007

4:1

World Scientific Book - 9in x 6in

124

From diffusive maps to Hamiltonian particle billiards

systematically gives access to higher-order corrections by including higherorder correlations, or memory effects. This leads us to the definition of two hierarchies of correlated random walk diffusion coefficients: (1) Again, we make the approximation that ρ∗a (x) ' 1. We then define Z 1 n X 1 Dn (a) := ck dx j(x)j(xk ) , n > 0 , (7.7) 0

k=0

D01 (a)

with = D0 (a) given by Eq. (7.5). Obviously, this series cannot converge to the exact D(a). (2) By using the exact invariant density in the averages of Eq.(7.2) we define Z 1 n X Dnρ (a) := ck dx ρ∗a (x)j(x)j(xk ) , n > 0 , (7.8) k=0

D0ρ (a)

R 1 1

0

here with = 2 0 dx ρ∗a (x)j 2 (x), which must converge exactly. The approximations Dn1 (a) and Dnρ (a) may be understood as time-dependent diffusion coefficients according to the Taylor-Green-Kubo formula Eq. (7.2). According to their definitions, the Dn1 (a) enables us to look at contributions coming from j(xn ) only, whereas the Dnρ (a) assess the importance of contributions resulting from ρ∗a (x). The rates of convergence of both approximations give an estimate of how important higher-order correlations are in the different parameter regions of the diffusion coefficient D(a). Similarly to Eq. (7.6), Eq. (7.7) can easily be calculated analytically to first order reading √ √  (a − 2)/(2a) , 2 < √ a ≤ 1 + 3 and 2 + 2 < a ≤ 4  D11 (a) = 3/2 − 3/a − 2/a2 , 1 + 3 < a ≤ √3  −1/2 + 3/a − 2/a2 , 3 < a ≤ 2 + 2 . (7.9) Further corrections up to order n = 3 were obtained from computer simulations, that is, an ensemble of point particles was iterated numerically by using the equations of motion. All results are displayed in Fig. 7.1 showing that this hierarchy of correlated random walks generates a self-affine structure which resembles, to some extent, the one of the well-known Koch curve [Man82]. Fig. 7.1 (a) illustrates that this structure forms an important ingredient of the exact diffusion coefficient D(a) thus explaining basic features of its fractality. Indeed, a suitable generalization of this approach in the limit of time going to infinity leads to the formulation of D(a) in terms of fractal generalized Takagi functions as discussed in Chapters 4 and 6.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

125

7.1. Correlated random walks in maps

Fig. 7.1 (b) depicts the results for the series of Dnρ (a) up to order n = 2 as obtained from computer simulations. This figure illustrates that there exists a second source for an irregular structure related to the integration over the invariant density, as explained above. In the Taylor-Green-Kubo formula Eq.(7.2) both contributions are intertwined via the integration over phase space. We now perform a more detailed analysis to reveal the precise origin of the hierarchy of peaks in Fig. 7.1. For this purpose we redefine Eq. (7.7) as Z 1 Z 1 1 Dn1 (a) = dx j(x)Jn (x) − dx j 2 (x) , n > 0 , (7.10) 2 0 0 where we have introduced the jump velocity function Jn (x) :=

n X

j(xk )

,

(7.11)

k=0

again with x = x0 . From Eq. (7.4) it follows Jn (x) = [xn+1 ], that is, this function gives the integer value of the displacement of a particle starting at some initial position x. In Fig. 7.2 we depict some representative results for J1 (x) under variation of the control parameter a. Because of the symmetry of the map we restrict our considerations to 0 < x < 0.5. Eq. (7.10) tells us that the product of this function with j(x) determines the diffusion coefficient D(a). The shaded bar in Fig. 7.2 marks the subinterval in which j(x) = 1, whereas j(x) = 0 otherwise, thus an integration over J1 (x) on this subinterval yields the respective part of the diffusion coefficient. One can now relate the four diagrams (a) to (d) in Fig. 7.2 to the functional form of D11 (a) in Fig. 7.1 (a) thus understanding where the large peak in D11 (a) for 2.732 < a < 3.414 comes from: For a < 2.732, J1 (x) does not change its structure and the interval where particles escape to other unit intervals increases monotonically, therefore D(a) increases smoothly. However, starting from a = 2.732 particles can jump for the first time to next nearest neighbours within two time steps, as is visible in J1 (x) taking values of 2 for x close to 0.5. Consequently, the slope of D(a) increases drastically leading to the first large peak around a = 3. Precisely at a = 3, backscattering sets in meaning that particles starting around x = 0.5 jump back to the original unit interval within two time steps, as is visible in J1 (x) in the form of the region J1 (x) = 0 for x close to 0.5. This leads to the negative slope in D(a) above a = 3. Finally, particles starting around x = 0.5 jump to nearest neighbour unit intervals by staying there during the second time

4:1

World Scientific Book - 9in x 6in

126

2

(a)

(b) 1

J1(x)

J1(x)

klages˙book

From diffusive maps to Hamiltonian particle billiards

1

0

0 -1 0

0.25 x

0.5

-1 0

(c)

0.5

0.25 x

0.5

(d) J1(x)

1

1 0

0 -1 0

0.25 x

2

2

J1(x)

May 8, 2007

0.25 x

0.5

-1 0

Fig. 7.2 Jump velocity function J1 (x) as defined by Eq. (7.11) giving the integer value of the displacement of a particle starting at some initial position x. Shown are results for different values of the slope a. It is (a) a = 2.7, (b) a = 3.0, (c) a = 3.3, (d) a = 3.8. The shaded bar marks the subinterval where the jump velocity j(xn ) defined in Eq. (7.4) is equal to one.

step instead of jumping back. This yields again a monotonically increasing D(a) for a > 3.414. Any higher-order peak for Dn1 (a), n > 1 , follows from analogous arguments. Thus, the source of this type of fractality in the diffusion coefficient is clearly identified in terms of the topological instability of the function J1 (x) under variation of the control parameter a. Indeed, this argument not only amends the heuristic turnstile dynamics analysis of D(a) outlined in Chapter 2, it also explains why, on a very fine scale, there are still deviations between these results and the precise location of the extrema in D(a), compare to the “overhang” in Fig. 7.1 around a = 3 as an example. The obvious reason is that contributions from the invariant density slightly modifying this structure have not been taken into account. Looking at the quantities Jn (x) furthermore helps us to learn about the rates of convergence of the approximations Dn1 (a) to D(a) at fixed values of a, as is illustrated in Fig. 7.3 (a) to (d). Here we have numerically calculated Jn (x) at a = 3.8 for n = 0, 1, 2, 3. Again, the shaded bar indicates the region where j(x) = 1 enabling Jn (x) to contribute to the value

4:1

World Scientific Book - 9in x 6in

klages˙book

127

7.1. Correlated random walks in maps

2 (b)

(a) J1(x)

J0(x)

1

0.5

1 0

0 0

0.25 x

0.5

-1 0 4

(c)

2

0.25 x

0.5

0.25 x

0.5

(d) 2

J3(x)

J2(x)

May 8, 2007

0

0

-2 -2

0

0.25 x

0.5

0

Fig. 7.3 Jump velocity function Jn (x) as defined by Eq. (7.11) at the fixed parameter value a = 3.8 for the different number of time steps (a) n = 0, (b) n = 1, (c) n = 2, (d) n = 3. Again the shaded bar marks the subinterval where the jump velocity j(x n ) defined in Eq. (7.4) is equal to one.

of the diffusion coefficient according to Eq. (7.10). In fact, Jn (x) may be interpreted as the scattering function of an ensemble of particles starting from the unit interval, since it sensitively measures the final position to which a particle moves within n time steps under variation of its initial position x. One can clearly see that, with larger n, Jn (x) develops more and more discontinuities eventually leading to a highly singular and irregular function of x. Integration over further and further refinements of Jn (x) 1 determines the convergence of the series of Dn1 (a) to a fixed value D∞ (a), cp. to Fig. 7.1 (a). To obtain quantitative values for the associated rates of convergence with respect to parameter variation is an open question. As was pointed out in Chapters 4 and 6, a suitable integration over the functions Jn (x) leads to the definition of fractal generalized Takagi functions, which can be calculated in terms of de Rham-type functional recursion relations. In a way, the integration over jump velocity functions such as the one shown in Fig. 7.3 (d) is similar to the integration over Cantor set structures leading to Devils’ staircase-type functions [Man82]. Our results presented so far thus bridge the gap between understanding the coarse functional behavior of D(a) on the basis of simple random walk approxi-

May 8, 2007

4:1

128

World Scientific Book - 9in x 6in

From diffusive maps to Hamiltonian particle billiards

mations only and analyzing its full fractal structure in terms of Takagi-like fractal forms, in combination with an integration over a complicated nonuniform invariant density. We now show that essentially the same line of argument can be successfully applied to more physical dynamical systems such as Hamiltonian particle billiards. 7.2

Correlated random walks in billiards

The class of two-dimensional billiards that we discuss in this section is described as follows: A point particle of mass m moving with velocity v undergoes elastic collisions with obstacles which are all of the same size and shape, whose centers are fixed on a triangular lattice. The equations of motion are defined by the Hamiltonian H = mv 2 /2 supplemented by geometric boundary conditions as imposed by the scatterers. A standard example is the periodic Lorentz gas [Lor05; Gas98a; Dor99], which has already briefly been mentioned in Section 4.1, cf. also Section 10.4 for further details, for which the scatterers consist of hard disks of radius R, see Fig. 7.4. In the following we choose m = 1, v = 1, R = 1, and as a control parameter we introduce the smallest interdisk distance w such that the lattice spacing of the disks√is 2 + w. w is related to the number density n of the disks by n(w) = 2/[ 3(2 + w)2 ]. At close packing w = 0 the moving particle is trapped in a single triangular region √formed between three disks, see Fig. 7.4, part (1). For 0 < w < w∞ = 4/ 3 − 2 = 0.3094, the particle can move across the entire lattice, but it cannot move collision-free for an infinite time. For w > w∞ the particle can move ballistic-like in the form of arbitrarily far jumps between two collisions. The diffusion coefficient for this particle billiard can be defined by the two-dimensional time-continuous analogue of the Einstein formula Eq. (2.2) reading < (x(t) − x(0))2 > , (7.12) t→∞ 4t where again the average is taken over the equilibrium distribution of particles with position coordinates x(t). It can be proven that in the regime of 0 < w < w∞ the parameter-dependent diffusion coefficient D(w) exists [Bun80]. The parameter dependence of this function was discussed in Ref. [Kla00a] showing that, on a fine scale, D(w) is an irregular function of the parameter w similar to the diffusion coefficient D(a) of the D(w) = lim

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

129

7.2. Correlated random walks in billiards

lr

(1)

rl

(3) zz

w

ll

lz

rz

rr

r

l

zr

zl

(2)

z Fig. 7.4 Geometry of the periodic Lorentz gas with the gap size w as a control parameter. The hatched area related to region (1) marks a so-called trapping region, the arrow gives the lattice vector connecting the center of this trap to the neighbouring one above. In region (2) three lattice vectors are introduced and labeled with the symbols l, r, z. They indicate the positions where particles move along the hexagonal lattice of Wigner-Seitz cells, starting from the trap z, within two time steps of length τ . Region (3) depicts the situation after three time steps τ with the different lattice vectors associated with symbol sequences of length two.

one-dimensional map discussed in the previous section. Whether D(w) is indeed fractal, or maybe C 1 but not C 2 , in contrast to simple mappings, is currently an open question. Hence, in the following we have to be more careful with our terminology and may speak of irregular rather than fractal transport coefficients if it is not clear what type of non-monotonicity is present. This applies particularly to the case of particle billiards, where usually only computer simulations and no exact analytical results are available for deterministic transport coefficients. We now wish to develop quantitative approximations for the full parameter dependence of D(w) by checking for the importance of memory effects. A simple analytical approximation for the diffusion coefficient was derived by Machta and Zwanzig [Mac83]. This solution was based on the assumption that diffusion can be treated as a Markovian hopping process between the triangular trapping regions indicated in Fig. 7.4. For random walks on two-dimensional isotropic lattices the diffusion coefficient then reads `2 D= , (7.13) 4τ √ where ` = (2 + w)/ 3 is the smallest distance between two centers of the traps, and τ −1 is the average rate at which a particle leaves a trap. This

4:1

World Scientific Book - 9in x 6in

130

klages˙book

From diffusive maps to Hamiltonian particle billiards

0.2

D(w)

May 8, 2007

simulation results for D(w) random walk approximation 1st order approximation 2nd order approximation 3rd order approximation 1st order and coll.less flights

0.1

0 0

0.1

0.2

0.3

w Fig. 7.5 Diffusion coefficient D(w) for the periodic Lorentz gas as a function of the gap size w as a control parameter. The computer simulation results for D(w) are from Ref. [Kla00a], error bars are much smaller than the size of the symbols. The other lines correspond to different levels of the approximation Eq. (7.16), the last approximation (again with symbols) is from Eq. (7.20).

rate can be calculated by the fraction of phase space volume being available for leaving the trap divided by the total phase space volume of the trap thus leading to the Machta-Zwanzig random walk approximation of the diffusion coefficient DMZ (w) =

w(2 + w)2 π[ 3(2 + w)2 − 2π] √

.

(7.14)

This approximation is the billiard analogue to the one-dimensional random walk diffusion coefficient for maps Eqs.(7.5),(7.6). Correspondingly, this approximation is asymptotically exact only for w → 0, as is shown in Fig. 7.5 in comparison to computer simulation results [Kla00a; Mac83]. In Ref. [Kla00a], Eq. (7.14) was systematically improved by including higher-order correlations. Two basic approaches were presented both starting from the idea of Machta and Zwanzig of looking at diffusion in the Lorentz gas as a hopping process on a hexagonal lattice of “traps” with frequency τ −1 . This picture was quantified by introducing a simple symbolic dynamics for a particle moving from trap to trap as indicated in Fig. 7.4.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

7.2. Correlated random walks in billiards

klages˙book

131

Let us follow a long trajectory of a particle starting with velocity v parallel to the y-axis, cf. part (1) in Fig. 7.4. For each visited trap we label the entrance through which the particle entered with z, the exit to the left of this entrance with l, and the one to the right with r, see part (2) in Fig. 7.4. Thus, a trajectory in the Lorentz gas is mapped onto words composed of the alphabet {z, l, r}. One can now associate transition probabilities to these symbol sequences reading, for a time interval of 2τ , p(z), p(l), p(r). p(z) corresponds to the probability of backscattering, whereas p(l) = p(r) = (1 − p(z))/2 holds for forward scattering. For a time interval of 3τ we have nine symbol sequences each consisting of two symbols leading to the probabilities p(zz), p(zl), p(zr), p(lz), p(ll), p(lr), p(rz), p(rl), p(rr), and so on. Beside this hierarchy of conditional probabilities defined with respect to symbolic dynamics there is a different type of probability that a particle leaving a trap jumps without any collision directly to the next nearest neighbour trap. As was shown in Ref. [Kla00a], Eq. (7.14) can be corrected by analytically including all these probabilities. Alternatively, lattice gas computer simulations were performed by using the probabilities as associated with all symbol sequences. These heuristic corrections of Eq. (7.14) led to a satisfactory explanation of the overall behavior of D(w) on a coarse scale, however, the convergence was not exact. The lattice gas simulations, on the other hand, were converging exactly, however, here a proper analytical expression for the diffusion coefficient approximations in terms of the associated probabilities was not available. In analogy to the procedure outlined for the one-dimensional map, that is, starting from a suitable Taylor-Green-Kubo formula, we will now define a third approximation scheme that we expect to be generally applicable to diffusion in particle billiards. Compared with the two existing approaches mentioned above the advantage of the new method is two-fold, namely (1) that by using the set of symbolic probabilities the respective TaylorGreen-Kubo formula can be evaluated according to an analytical scheme, and (2) that the resulting approximations converge exactly to the computer simulation results. In the appendix of Ref. [Kla02d] it is proven that, starting from the Einstein formula Eq. (7.12), quite in analogy with the case of one-dimensional maps, a Taylor-Green-Kubo formula can be derived that is defined for an ensemble of particles moving on the hexagonal lattice of traps depicted in

May 8, 2007

4:1

132

World Scientific Book - 9in x 6in

klages˙book

From diffusive maps to Hamiltonian particle billiards

Fig. 7.4. The final formula reads ∞ 1 X D(w) = ck hj(x0 ) · j(xk )i 2τ

,

(7.15)

k=0

again with ck as defined in Eq. (7.3). Here j(xk ) defines jumps at the kth time step of length τ in terms of the lattice vectors `αβγ... associated with the respective symbol sequence of the full trajectory on the hexagonal lattice, cf. Fig. 7.4. Let us start with j(x0 ) = `/τ , ` := (0, `)∗ . The next jumps are then defined by j(x1 ) = `α /τ , α ∈ {l, r, z}, and so on. The averages indicated in Eq. (7.15) by the brackets are calculated by weighting the respective scalar products of lattice vectors with the corresponding conditional probablities p(αβγ . . .). Eq. (7.15) is the honeycomb lattice analogue of the Taylor-Green-Kubo formula derived by Gaspard for the Poincar´e-Birkhoff map of the periodic Lorentz gas [Gas98a; Gas96]. The Poincar´e-Birkhoff version is very efficient for numerical computations, however, according to its construction it fails to reproduce the Machta-Zwanzig approximation Eq. (7.14). Consequently, it does not appear to be suitable for diffusion coefficient approximations of low order. We remark that, in terms of using a symbolic dynamics, there is some link between Eq. (7.15) and diffusion coefficient formulas obtained from periodic orbit theory [Cvi95; Cvi92]. We now demonstrate how Eq. (7.15) can be used for systematic evaluations of the diffusion coefficient on the lattice of traps by building up dynamical correlations: As in the case of one-dimensional maps we start by looking at the first term in Eq. (7.15) and cut off all higher order contributions. Obviously, the first term is again the random walk expression for the diffusion coefficient on the hexagonal lattice of traps which, by including the solution for the jump frequency τ −1 , boils down to Eq. (7.14). For calculating higher-order corrections we now define the hierarchy of approximations 1 X l2 p(αβγ . . .)` · `(αβγ . . .) , n > 0 , (7.16) + Dn (w) = 4τ 2τ αβγ...

with D0 (w) given by Eq. (7.14). To our knowledge there is no method yet available to analytically calculate the conditional probabilities p(αβγ . . .). Our following evaluations are therefore based on the data presented in Ref. [Kla00a] obtained from computer simulations. In terms of the formal probabilities it is now easy to calculate the solution for the first order approximation at time step 2τ to D1 (w) = D0 (w) + D0 (w)(1 − 3p(z)) .

(7.17)

May 8, 2007

4:1

World Scientific Book - 9in x 6in

7.2. Correlated random walks in billiards

klages˙book

133

For a comparison of this formula with the simulation data D(w) see Fig. 7.5. We remark that the corresponding solution in Ref. [Kla00a] obtained from a heuristic correction of Eq. (7.14) reads D1,MZ (w) = D0 (w)3(1 − p(z))/2. Indeed, one can show that the new formula Eq.(7.17) is closer to D(w) for large w, whereas for small w the previous approximation is somewhat better. It is straightforward to calculate the two approximations of next higher order to D2 (w) = D1 (w) + D0 (w) (2p(zz) + 4p(lr) − 2p(ll) − 4p(lz))

(7.18)

and to D3 (w) = D2 (w) + D0 (w) [p(llr) + p(llz) + p(lrl) + p(lrr) + p(lzl)+ +p(lzz) + p(rll) + p(rlr) + p(rrl) + p(rrz) + p(rzr) + p(rzz) + +p(zll) + p(zlz) + p(zrr) + p(zrz) + p(zzl) + p(zzr) − 2(p(lll) + +p(lrz) + p(lzr) + p(rlz) + p(rrr) + p(rzl) + p(zlr) + p(zrl) + +p(zzz))]

.

(7.19)

All results are shown in Fig. 7.5 demonstrating that the series of approximations defined by Eq. (7.16) converges quickly and everywhere to the simulation results. Our new scheme thus eliminates the deficiency of the semi-analytical approximation proposed in Ref. [Kla00a] that was based on heuristically correcting the Machta-Zwanzig approximation Eq. (7.14). By comparing this new scheme with the lattice gas simulations of the same reference on the other hand, it turns out that the rate of convergence of the lattice gas approach is still a bit better. In any case, all these three methods unambiguously demonstrate that for achieving a complete understanding of the density-dependent diffusion coefficient in the periodic Lorentz gas it is unavoidable to take higher-order correlations, or the impact of memory effects, properly into account. We finally remark that a very good low-order approximation for the diffusion coefficient can already be obtained by combining Eq. (7.17) with the probability of collisionless flights pcf (w) mentioned above, that is, by taking into account the possibility of next nearest neighbour jumps. The correction of D0 (w) as given by Eq. (7.14) according to collisionless flights only was already calculated in Ref. [Kla00a] and read D0,cf (w) = D0 (w)(1+ 2pcf (w)). Adding now the second term of Eq. (7.17) to this expression by just following the Taylor-Green-Kubo scheme yields D1,cf (w) = D0 (w)(2 + 2pcf (w) − 3p(z)) .

(7.20)

May 8, 2007

4:1

World Scientific Book - 9in x 6in

134

From diffusive maps to Hamiltonian particle billiards

This solution is also depicted in Fig. 7.5 and shows that this approximation indeed significantly improves Eq. (7.17) for large w yielding a function that is qualitatively and quantitatively very close to D(w). We know of no better approximation for D(w) based on information such as p(z) and pcf (w) only. The succesful application of Eq. (7.20) suggests that collisionless flights form an important mechanism for understanding the full diffusive dynamics of this billiard. However, somewhat surprisingly they are explicitly contained neither in the Taylor-Green-Kubo expansion Eq. (7.16) nor in the lattice gas simulations of Ref. [Kla00a]. In both cases exact convergence is achieved by following the hierarchy of symbol sequence probabilities only in which collisionless flights are not apparent. The theory outlined above should have interesting applications to electronic transport in antidot lattices, see also Section 10.4. These are twodimensional semiconductors furnished with circular holes, which can be burned into the conduction band by electron lithography. Here periodic lattices of holes have been manufactured in which noninteracting electrons do not only exhibit quantum transport but also move quasiclassically under application of external electric and magnetic fields. In this limit, these two-dimensional nanostructures are thus very similar to classical periodic Lorentz gases. From experiments it is well-known that the classical magnetoresistance of such systems is a highly irregular function of the field strength [Wei91; Lor91; Wei95; Wei97]. The largest irregularities were theoretically explained in terms of certain periodic orbits, which enhance or suppress magneto-transport [Gei90; Fle92; Wei95; Fli96]. More recently, even a fractal magnetoresistance has been theoretically predicted for these systems in specific settings [Wie01]. These results thus fit very well to the main theme outlined in the previous chapters, which is the fractality of deterministic transport coefficients in low-dimensional periodic structures. However, this theory has not yet been applied to antidots, and so far experiments have not been carried out in order to search for such fractal structures.

7.3

∗

Summary

(1) In this chapter we have proposed a rigorous, systematic approach for understanding the irregular structures of parameter-dependent diffusion coefficients. For this purpose we employed, respectively derived, suitable Taylor-Green-Kubo formulas for diffusion on periodic lattices.

klages˙book

May 8, 2007

4:1

7.3.

World Scientific Book - 9in x 6in

∗ Summary

klages˙book

135

This theory reproduces simple random walk approximations for diffusion in lowest order but enables one to systematically include higherorder memory effects, hence we called the improved formulas correlated random walk approximations. We first outlined this approach for our standard model of a one-dimensional diffusive map on the line. For this map we discussed in full detail the origin of the fractal structure in the diffusion coefficient by evaluating analytically and numerically the single terms of the Taylor-Green-Kubo expansion. Particularly, we focused on the contribution eventually related to fractal generalized Takagi functions, which were already discussed in previous chapters. We showed that the convergence of the Taylor-Green-Kubo series expansion intimately depends on the choice of parameters reflecting the existence of intricate long-range dynamical correlations. (2) The big advantage of this theory is that it can straightforwardly be generalized to understand irregular diffusion in particle billiards such as the periodic Lorentz gas. Again one needs to start from a suitable Taylor-Green-Kubo formula for billiards, which recovers the well-known Machta-Zwanzig random walk approximation for diffusion in lowest order. We showed that such a formula can be evaluated along similar lines to the case of the one-dimensional maps discussed before. The convergence of this series expansion yielded evidence for an irregular structure of the Lorentz gas diffusion coefficient on fine scales. These irregularities can be physically understood in terms of sequences of forward- and backward scattering whose existence depends sensitively on parameter variations.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

This page intentionally left blank

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Chapter 8

Designing billiards with irregular transport coefficients

The periodic Lorentz gas exhibits an irregular diffusion coefficient, as we have seen in the previous chapter. However, the irregularities were so small that they were hard to detect without further scrutiny. Here we may turn the problem around: Having explored the origin of fractal transport coefficients enables us to use our understanding in order to design systems that exhibit irregular transport coefficients in an especially pronounced manner. This strategy may be particularly useful for practical applications of this phenomenon. A first example along these lines is provided by a suitable modification of the periodic Lorentz gas, the flower-shaped billiard, which we introduce in Section 8.1. Computer simulations indeed reveal a non-trivial parameter dependence of the diffusion coefficient in this model. In Section 8.2 we apply the approaches outlined in the previous chapter in order to analyze the irregularities of the diffusion coefficient. Section 8.3 concludes by establishing crosslinks to theory and experiments of molecular diffusion in porous crystalline solids. This chapter mainly emerged from a collaboration particularly with T.Harayama, who suggested studying flower shapes [Har02]; for related work see Refs. [Kla00a; Kla02d]. 8.1

Diffusion in the flower-shaped billiard

Parameter-dependent diffusion in the conventional periodic Lorentz gas suffers from the fact described in Section 7.2 that the parameter range of normal diffusion is very limited. In this small region the irregular behavior of the diffusion coefficient shows up on very fine scales only and appears to be rather smooth within the range of precision available from computer 137

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

138

Designing billiards with irregular transport coefficients

Fig. 8.1 A modified Lorentz gas composed of a point particle moving freely in the space between flower-shaped scatterers situated on a triangular lattice and colliding elastically with the obstacles.

simulations [Kla00a; Har]. Consequently, the question about the existence of an irregular or even truly fractal diffusion coefficient, such as the one existing in simple one-dimensional maps, turned out to be very difficult to answer for this model. However, as we have learned in the previous chapters the mechanism generating fractal transport coefficients is usually a topological instability of a dynamical system under parameter variation. Although the periodic Lorentz gas is dynamically unstable and thus chaotic it appears that its parameter-dependent topology is still stable enough such that it does not generate too pronounced irregularities in the parameter region of normal diffusion. The purpose of this chapter is therefore to modify the periodic Lorentz gas such that we arrive at a particle billiard with normal diffusion in a range of control parameters that features a stronger topological instability. This way, we intend to learn more about the emergence of possibly fractal transport coefficients in billiards. The key idea is to replace the circular scatterers in the Lorentz gas defined in Section 7.2 by a more complicated geometry for which we choose

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

8.1. Diffusion in the flower-shaped billiard

klages˙book

139

Fig. 8.2 Definition of a flower-shaped obstacle. The bigger hexagon (bold lines) is the elementary Wigner-Seitz cell. The arc always touches the smaller hexagon (dotted lines), which prohibits any infinite horizon.

flower-shaped obstacles. The resulting system, which we call the flowershaped billiard, is shown in Fig. 8.1. The purpose is to introduce more “curvature” by this change of geometry thus increasing the sensitivity with respect to initial conditions by hoping that this also increases the topological instability under variation of a suitable control parameter. According to our understanding of the origin of fractal transport coefficients, this should be reflected in an increase of the irregularity of the billiard’s diffusive properties under parameter variation. A single scatterer of our billiard is defined in Fig. 8.2: First we consider the inner hexagon whose vertices are on the middle points of the sides of the hexagon of the billiard’s elementary Wigner-Seitz cell, as depicted by the dotted lines. Next we join six arcs which have the same radii and touch the inner hexagon. Thus we obtain the flower-shaped obstacle shown in the figure. Note that in the following we keep the √ distance between the centers of two flower-shaped disks fixed to ` = 1/ 3. However, we may still vary the radius r of the arcs, which form the petals of the flower shape, from √ 1/(4 3) to infinity. The flower-shaped billiard belongs to the class of dispersing billiards whose hyperbolicity has been proven by Sinai [Sin70]. It is thus known that the motion of a point particle in the elementary Wigner-Seitz cell of this billiard is hyperbolic in the sense that all orbits are unstable of saddle type, and time averages are equal to averages over the Liouville equilibrium invariant measure. The spatially extended flower-shaped billiard is of the type of the periodic Lorentz gas whose normal diffusion has been proven by Bunimovich, Sinai and Chernov [Bun80; Bun81; Bun91; Che94; Che99]; for further details see Section 10.4.

4:1

World Scientific Book - 9in x 6in

140

Designing billiards with irregular transport coefficients

klages˙book

0.16 0.12 0.094

0.08

D

D

May 8, 2007

0.04 0.086 6.2

0.00

0

1

2

3

κ

κ 4

6.7

5

6

7

Fig. 8.3 Diffusion coefficient D versus the curvature κ = κc of a petal of a flower-shaped obstacle. The inset shows a magnification of D for large κc suggesting the existence of continuous irregularities on finer scales.

By using again the Einstein formula Eq. (7.12), the diffusion coefficient D of the flower-shaped billiard can be computed from simulations. Fig. 8.3 shows the numerical results for D as a function of the curvature κc = 1/r √ of the petals, which was varied from zero to its maximum 4 3. Indeed, we now observe a non-trivial structure of the diffusion coefficient. The existence of a global maximum in D(κc ) can qualitatively be understood as follows: When the curvature of a petal of the flower-shaped obstacle is zero, the inner hexagon shown by the dotted lines in Fig. 8.2 connects to the six hexagons surrounding it. In this case, the point particle remains trapped in compact domains bounded by the three neighbouring hexagons. When the curvature becomes positive, the point particle can run away from the compact domain and diffusion sets in. As already explained, at all positive curvatures of the petal, even if they are very small, the motion of the point particle is fully chaotic and the horizon is finite, hence diffusion is expected to be normal. The diffusion coefficient starts to increase from zero approximately linearly with the curvature related to the fact that the space between √ the petals also increases. When the radius of the petal is equal to RL := 3/4 ' 0.433, which is the distance between the center of the hexagon and the tangent point at it, the obstacle becomes a disk of radius RL , that is, for this parameter value our billiard is precisely the same as the conventional periodic Lorentz gas. This point corresponds

May 8, 2007

4:1

8.2.

World Scientific Book - 9in x 6in

+ Random

and correlated random walks

klages˙book

141

to the curvature κc ' 2.309 in Fig. 8.3. When the radius r of the curvature of the petal decreases below RL , the point particle is much more likely to be trapped in the space between two obstacles. This is due to the formation of wedges between any two petals of a flower-shaped obstacle. The inset of Fig. 8.3 suggests that, apart from a global maximum, there exists furthermore a non-trivial fine structure in D(κc ). However, we remark that the apparently continuous fluctuations shown therein are within numerical errors. We confirmed the convergence of our results within a precision of order 10−4 by taking an average over 1010 initial conditions. With the available computational power it was impossible to check whether this oscillatory behavior persists on even finer scales.

8.2

+

Random and correlated random walks

As was outlined in the previous chapter, there is a simple analytical approximation for the diffusion coefficient of the periodic Lorentz gas, which yields asymptotically exact results in the limit of small gaps between disks. In this limit the particle remains in the triangular regions between three adjacent scatterers for a long time. Hence, the particle is supposed to loose the memory of its past itinerary due to the multiple scattering in the trap region, and the transition probabilities to the neighbouring triangular cells are assumed to be equivalent. The average rate τ −1 of such transitions can be calculated from the fraction of phase space available for leaving the trap divided by the total phase space volume of the trap yielding τ = πA/(3w), where A is the trap area and w the width of the gap between the disks. The flower-shaped billiard has traps that are similar to the periodic Lorentz gas. Accordingly, this approximation can be applied to the flowershaped billiard as well, and the above equation for τ holds again for the average trapping time. Hence, we only need to calculate the area of the trap and the gap between the petals from simple geometrical considerations leading to √ h√ i p 3 3 − 3h 3h + r2 − h2 (8.1) A= 4 and i p 1 h√ 3h + r2 − h2 w= − , (8.2) 2 where ! √ 3 1 −r . (8.3) h= 2 4

4:1

World Scientific Book - 9in x 6in

142

Designing billiards with irregular transport coefficients

0.2

D

May 8, 2007

klages˙book

num. exact results Machta-Zwanzig

0.1

0.0

0

1

2

3

κ

4

5

6

7

Fig. 8.4 Diffusion coefficient D versus the curvature κ = κc of a petal of a flowershaped obstacle. The solid curve correspond to simulation results for D, whereas the dotted curve represents the Machta-Zwanzig random walk approximation Eq. (8.4).

Assuming that the gap size w is very narrow, we expect that the MachtaZwanzig random walk approximation for the diffusion coefficient [Mac83] `2 , (8.4) 4τ cf. Eq. (7.13), well describes the simulation results. This is confirmed in Fig. 8.4, which shows drastic deviations for larger values of the curvature. As we have outlined in Section 7.2, starting from the Taylor-GreenKubo formula for diffusion the Machta-Zwanzig approximation for the conventional periodic Lorentz gas can be generalized by taking memory effects into account. This generalization was based on the observation that, except in the asymptotic limit of narrow gap sizes, the diffusive dynamics is not a simple Markov process [Bec93], that is, there exist non-vanishing dynamical correlations. By mapping the orbit of a particle onto a suitable symbolic dynamics the probabilities to obtain certain symbol sequences of finite length were numerically calculated. Increasing the length of these symbol sequences yielded systematic corrections to the Machta-Zwanzig approximation. In Ref. [Kla00a] two further schemes related to a symbolic encoding were discussed, one suggesting straightforward heuristic corrections to the simple random walk model of diffusion Eq. (8.4) and another one based DM Z =

May 8, 2007

4:1

8.2.

World Scientific Book - 9in x 6in

+ Random

klages˙book

143

and correlated random walks

on lattice gas computer simulations defined by the associated conditional probabilities. We first explain these two alternative methods and apply them to the flower-shaped billiard in order to correct the Machta-Zwanzig approximation. The correlated random walk approach based on the TaylorGreen-Kubo formula is worked out below. The Machta-Zwanzig approximation assumes that a particle jumps from one trap to a neighbouring trap situated on the hexagonal lattice of traps. However, there exist non-vanishing probabilities that a particle can jump to next nearest neighbours, or even farther, without collisions. Accordingly, we should correct the Machta-Zwanzig approximation for the flower-shaped billiard by using the probabilities pcf 1 and pcf 2 of those collisionless flights which lead from one cell directly to its second nearest neighbours, or to the third nearest neighbours, respectively. The distances between √ the centers of √ a trap to second and third neighbours are l1 = 3 and l2 = 7, respectively. The diffusion coefficient Dcf with corrections due to these events then reads Dcf = (1 − pcf 1 − pcf 2 )DM Z + pcf 1 = (1 + 2pcf 1 + 6pcf 2 )DM Z .

l12 l2 + pcf 2 2 4τ 4τ (8.5)

Next we take memory effects of orbits due to backscattering into account. As in Section 7.2, orbits are coded by labeling the entrance through which a particle enters a trap with z, the exit to the left of this entrance with l, and the one to the right with r for this purpose. Thus, an orbit can be mapped onto a sequence of symbols z, l, and r. Particularly, p(z) = pbs is the backscattering probability, which is the probability for the moving particle to leave the trap through the same gate that it entered by. The Machta-Zwanzig approximation assumes that p(z) = p(l) = p(r) = 1/3. However, in general p(z) is not close to 1/3 as shown in Fig. 8.5, because the actual orbits do not loose their memory during their itineraries. A more profound explanation for the complicated form of p(z) is provided by the theory of chaotic scattering: Chaotic systems with multiple exit modes typically have fractal phase space boundaries separating the sets of initial conditions, or basins, going to the various exits. Open systems such as a three-disk scatterer of the periodic Lorentz gas possess the even stronger property of being Wada, that is, any initial condition which is on the boundary of one exit basin is also simultaneously on the boundary of all the other basins [Poo96; Agu01]. Changing the curvature κc sensitively affects the highly irregular structure of these basin boundaries. Fig. 8.5 thus reflects the topological instability of Wada basins under parameter

4:1

World Scientific Book - 9in x 6in

klages˙book

144

Designing billiards with irregular transport coefficients

0.6

Pbs

May 8, 2007

0.4

0.2

0

1

2

3

κ

4

5

6

7

Fig. 8.5 Backscattering probability pbs = p(z) as a function of the curvature κ = κc of the petal of the flower-shaped obstacle. In the case of a Markov process it is p bs = 1/3.

variation, and as we will now show, this has an impact on the parameter dependence of the diffusion coefficient.1 Modifying the Machta-Zwanzig random walk by including the backscattering probability p(z) we obtain the diffusion coefficient (1 − p(z))l22 3 = (1 − p(z)) DM Z . (8.6) DBS = 4(2τ ) 2 Combining the effects of collisionless flights and backscattering yields as a first order approximation 3 D1 = (1 − p(z))(1 + 2pcf 1 + 6pcf 2 )DM Z . (8.7) 2 Higher-order approximations for the diffusion coefficient related to longer symbol sequences and respective probabilities such as p(lrz . . .) can be derived in the same way [Kla00a]. For the flower-shaped billiard respective results are shown in Fig. 8.6. The above correction methods assume that all orbits follow higher-order Markov processes, where correlations are present in the form of initial transient times before the variance becomes linear in time. This dynamics appears to be more suitably represented in the form of lattice gas simulations 1A

respective analysis of the chaotic scattering process for the conventional periodic Lorentz gas was contained in the draft version (available online) of Ref. [Kla00a], see a note therein. This part was largely deleted for the printed version upon editorial request related to page limitations for a special issue. Later on quite the same analysis formed the starting point of Ref. [Agu03].

4:1

8.2.

World Scientific Book - 9in x 6in

+ Random

klages˙book

145

and correlated random walks

0.2

D

May 8, 2007

0.1 num. exact results Machta-Zwanzig 1st order 2nd order 3rd order 4th order 5th order

0.0

0

1

2

3

κ

4

5

6

7

Fig. 8.6 Random walk diffusion coefficient and higher-order corrections based on higherorder backscattering probabilities. The solid curve corresponds to numerically exact results, the other curves represent the approximate solutions.

on a honeycomb lattice, where the sites of the lattice represent the traps. For the periodic Lorentz gas such lattice gas simulations were performed in Ref. [Kla00a] confirming fast convergence to the numerically exact results. Compared to that scheme, the convergence of the intuitive correction method described above is firstly slower, and secondly it does not convere everywhere to the numerically exact results, which is due to the fact that this approach is of a purely heuristic nature. It is thus interesting to carry out lattice gas simulations for the flowershaped billiard, which can be done according to the following prescription: Particles hop from site to site with a frequency τ −1 , which is identical with the hopping frequency used in the Machta-Zwanzig approximation. The hopping probabilities are given by the backscattering probability p(z) and by those corresponding to respectively longer symbol sequences. The diffusion coefficient is then obtained from the Einstein formula Eq. (7.12) in the limit when the variance is becoming proportional to time. The correlations in the actual orbits are thus systematically and exactly filtered out according to the length of the symbol sequences. In Fig. 8.7 the results of such higher-order approximations according to lattice gas simulations are shown. One can see that the convergence to the numerically exact results is not only much better than in Fig. 8.6 but even exact. Strong memory effects are clearly visible especially after

4:1

World Scientific Book - 9in x 6in

146

Designing billiards with irregular transport coefficients

klages˙book

0.2

D

May 8, 2007

0.1 num. exact results 1st order 2nd order 3rd order 4th order 5th order

0.0

0

1

2

3

κ

4

5

6

7

Fig. 8.7 Diffusion coefficient as obtained from lattice gas simulations based on higherorder backscattering probabilities. The solid curve corresponds to numerically exact results while the other curves yield higher-order approximations.

the diffusion coefficient curve takes its maximum. In the previous heuristic modifications to the simple random walk model, the dynamics was modeled as a Markov process for a limited number of time steps only. Fig. 8.6 suggests that correlations as contained in the symbol sequences are more suitably represented by higher-order iterations in the form of lattice gas simulations. However, a disadvantage is that the lattice gas scheme requires a second round of computations which is put on top of the previous simulations by again looking at the time evolution of an initial ensemble of points. In summary, the above two methods of obtaining random walk corrections both showed major deficiencies: The heuristic amendments of the random walk formula were not converging exactly to the numerically exact results, whereas the lattice gas simulations were merely a numerical scheme without being represented in the form of analytical approximations. Let us thus compare these results with the correlated random walk approximation outlined in Section 7.2, which is based on the Taylor-Green-Kubo formula Eq. (7.15) that employs the symbolic dynamics on the hexagonal lattice of traps introduced in the previous section and in Section 7.2. In the case of the flower-shaped billiard the first term of Eq. (7.15) yields Eq. (8.4) amended by Eqs. (8.1)-(8.3). Higher-order corrections can then be calculated by using the hierarchy of approximations Dn defined by Eq. (7.16),

4:1

8.2.

World Scientific Book - 9in x 6in

+ Random

klages˙book

147

and correlated random walks

0.10 0.05 -0.00

Cn

May 8, 2007

-0.05 C1 C2 C3 C4 C5

-0.10 -0.15

0

1

2

3

κ

4

5

6

7

Fig. 8.8 Parameter dependence of the time-dependent correlation function C n Eq. (8.8) defined with respect to the symbolic dynamics on the hexagonal lattice of traps. At any parameter Cn decays exponentially related to the fact that the Taylor-Green-Kubo formula Eq. (7.15) is a convergent series. However, the speed of the convergence depends on the curvature κ = κc . For large κc the correlation function decays much more slowly than for small κc .

again with D0 given by Eq. (8.4) plus supplements. Within this scheme, the impact of dynamical correlations on the diffusion coefficient can be understood in terms of the velocity correlation function Cn := hj(x0 ) · j(xn )i

(8.8)

representing the single terms in the Taylor-Green-Kubo formula Eq. (7.15). In fully chaotic systems such as the periodic Lorentz gas and the flowershaped billiard this function decays exponentially, which is in agreement with the results depicted in Fig. 8.8. By comparing this figure to Fig. 8.9 one can see how the irregularities of the correlation function determine the diffusion coefficient depending on parameters. Let us start with the first-order approximation of Eq. (7.15) reading D1 = D0 + D0 (1 − 3p(z)), cf. Eq. (7.17). The functional form of p(z) in Fig. 8.5 then qualitatively explains the position of the global maximum of the diffusion coefficient, because at this value of the curvature the probability of backscattering is minimal. Adding up the three-jump contributions contained in C2 yields the most important quantitative corrections at this point. In the region of large curvature the diffusion coefficient decays monotonically according to the effect of two-hop correlations covered by

4:1

World Scientific Book - 9in x 6in

148

Designing billiards with irregular transport coefficients

klages˙book

0.2

D

May 8, 2007

0.1 num. exact results Machta-Zwanzig 1st order 2nd order 3rd order 4th order 5th order

0.0

0

1

2

3

κ

4

5

6

7

Fig. 8.9 Diffusion coefficient approximations obtained from the Taylor-Green-Kubo formula Eq. (7.15) as applied to the flower-shaped billiard. The solid curve corresponds to the asymptotic, numerically exact results while the other curves yield a hierarchy of correlated random walk approximations.

C1 . However, note the large fluctuations of the correlation function Cn as well as of the diffusion coefficient approximations Dn in this regime, which both indicate the dominant effect of long-range higher-order correlations. The detailed convergence of the approximations depicted in Fig. 8.9 shows that correlations due to orbits with longer symbol sequences yield irregularities in the parameter dependence of the diffusion coefficients on finer and finer scales. Note that in contrast to the two methods outlined in the previous section, the Taylor-Green-Kubo approach provides a scheme that can partially be worked out analytically and converges exactly.

8.3

Diffusion in porous solids

As usual, we conclude with relations to experiments. Here we think particularly of molecular diffusion in meso-, micro- and nanoporous crystalline solids [K¨ ar92]. These materials have wide technical applications as molecular sieves, adsorbants and catalysts. They can be found, for example, in everyday-life detergents, where they serve as ion exchangers, and they are important catalysts for cracking long molecules in fuel production [Sch03a]. The flower-shaped billiard reminds one of the mesoporous structure syn-

May 8, 2007

4:1

8.3. Diffusion in porous solids

World Scientific Book - 9in x 6in

klages˙book

149

thesized in Refs. [Hoa05], which consists of hexagonally ordered primary channels thus exhibiting the same symmetry as the periodic Lorentz gas. The walls of these channels are furnished with micropores of which the petals of the flower-shaped billiard may be considered as a first-order approximation. Experimental results on the diffusion of gas molecules through porous solids are often well-described in terms of Knudsen diffusion [Ste86], that is, the interaction of the molecules with the pore walls play the crucial role and intermolecular collisions can be neglected. Thus, we have a similar type of problem to diffusion in a Hamiltonian particle billiard. Interestingly, measurements of the diffusion coefficient in the bi-porous material mentioned above yielded an unexpected strong suppression of diffusion, which could not be explained by ordinary random walk Knudsen theory. What makes it difficult to relate these results more closely to diffusion in billiards of flower-shaped type is that in the experiments the micropore distribution is not periodic but random. This perhaps suggests studying a model with quenched randomness imposed onto the petals by cross-checking with the theory of Chapter 5. Motivated by experimental results on diffusion in porous solids, the theoretical study of diffusion in billiard channels became an active field. Non-chaotic polygonal billiard channels are at the abstract side of this type of model [Jep06] while chaotic Lorentz gas-like channel geometries are closer to reality [Alo99]. In other variants [Rus05] the wall porosity was mimicked by fractal Koch curves, and the scattering of point particles at these walls was chosen to be random and not specular. We remark that in such models anomalous diffusion has been observed pointing back to the discussion of Chapter 6. A related type of nanoporous channels can be found in zeolitic structures. Here single file diffusion became a prominent topic, which describes the transport of molecules in channels where mutual passages are excluded. We note in passing that this type of diffusion may also exhibit anomalous properties [Sch05]. In simulations these channels are usually modeled by specific smooth potentials. The moving particles furthermore interact with a thermal reservoir, a topic that will be covered in Part 2 of this book. However, it turns out that the methods developed in this and in the previous chapter are still straightforwardly applicable in order to explain simulation results of, for example, temperature-dependent diffusion in such channels [Sch07]. Carbon nanotubes represent a related class of systems that may be worth explorations along similar lines. Zeolitic channels form a subset of problems within the wide field of

May 8, 2007

4:1

World Scientific Book - 9in x 6in

150

Designing billiards with irregular transport coefficients

molecular diffusion in zeolites [K¨ ar92; Sch03a]. The latter are a chemically well-defined class of strictly periodic porous solids, which are particularly amenable to theoretical studies such as molecular dynamics computer simulations. In Ref. [Sch02] a non-monotonic temperature dependence of diffusion was predicted for such a structure, and Ref. [Sar00] describes a “bottleneck” phenomenon leading to a non-monotonic suppression of diffusion with respect to varying the diameter of the bottleneck. In both cases these phenomena were explained by analytical arguments which are very similar to the correlated random walk approach outlined before, that is, simple random walk (transition state) theory was refined by taking sequences of jumps from lattice point to lattice point into account [K¨ ar03]. Within this context first-order random walk corrections yield what is called the correlation factor [K¨ ar92; K¨ ar03; Sch03a], which is basically what we have calculated in Eqs. (7.17), (8.5) and (8.6) for our particle billiards. This suggests interesting crosslinks for applying the theory presented in Part 1 of this book to diffusion in porous solids, particularly in zeolites. Indeed, in Ref. [K¨ ar03] it was pointed out that “evaluating Eq. (7.15) analytically remains challenging” for understanding diffusion in zeolites. These structures thus promise to be an excellent testing ground in order to check for the existence of higher-order dynamical correlations in transport coefficients by experiments. We finally remark that correlation factors have also been introduced in order to describe experiments on atomic diffusion in intermetallic alloys [Kai01].

8.4

∗

Summary

(1) The idea of constructing a diffusive Hamiltonian particle billiard with flower-shaped scatterers emerged from the fact that the irregularities in the parameter-dependent diffusion coefficient of the periodic Lorentz gas were not too pronounced. Identifying the topological instability of the dynamical system as the origin of irregular parameter dependences suggested designing scatterers exhibiting a stronger curvature dependence under parameter variation thus leading to a flower shape. (2) We then obtained the diffusion coefficient of this model as a function of the curvature of the scatterers from computer simulations. The result showed a highly non-monotonic curve with further irregularities on finer scales. Whether this structure is eventually fractal could not

klages˙book

May 8, 2007

4:1

8.4.

World Scientific Book - 9in x 6in

∗ Summary

klages˙book

151

be resolved numerically. The coarsest functional form of the diffusion coefficient could roughly be understood by means of simple physical arguments. (3) More fine-tuned explanations were provided by different types of approximation: Firstly, we worked out the random-walk formula for diffusion, known as the Machta-Zwanzig approximation, for the flowershaped billiard. However, as in the case of the periodic Lorentz gas this formula is only asymptotically correct in the limit of small diffusion coefficients. Further corrections to this result could be obtained by simple heuristic improvements, which required the use of non-Markovian forward and backward scattering probabilities. Correction terms were also computed from simulations using higher-order transition probabilities in diffusion on a honeycomb lattice. The most suitable approximation scheme was provided by the Taylor-Green-Kubo expansion already outlined in the previous chapter, which here we worked out for the flowershaped geometry. (4) We finally outlined relations between amended particle billiards and diffusion in porous solids. Billiard channel geometries appear to be suitably toy models for studying diffusion phenomena in nanoporous channels. First experimental and theoretical results yield irregular parameter dependencies of diffusion coefficients and explanations which are both very similar to the findings described in the previous chapters.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

This page intentionally left blank

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Chapter 9

Deterministic diffusion of granular particles

We conclude our discussion of diffusion in particle billiards and Part 1 of this book by applying the theory outlined before to granular particles bouncing inelastically on a vibrating corrugated floor. We call this non-Hamiltonian system the bouncing ball billiard. In Section 9.1 we motivate our model and briefly review the role of resonances for a ball bouncing on a flat vibrating plate. We then show computer simulation results for the diffusion coefficient as a function of a suitable parameter and explain its largest irregularities. Section 9.2 contains a more refined analysis of parameter-dependent diffusion in terms of the Taylor-Green-Kubo approach, which corroborates the existence of irregularities on fine scales. In Section 9.3 we discuss relations between the bouncing ball billiard and experiments on vibratory conveyors as they are frequently encountered in industrial applications. The research outlined in this chapter reflects collaborations with L.M´ aty´ as and I.F.Barna [M´ at04a; Kla04b]. 9.1

Resonances and diffusion in the bouncing ball billiard

The billiard to be introduced below is designed in order to bring the theory of fractal transport coefficients and experiments together. From the theoretical side, this model is motivated by work of Harayama and Gaspard (HG) [Har01], who analyzed a Hamiltonian billiard where a particle subject to a constant vertical force is reflected elastically from a periodically corrugated floor. Again, the diffusion coefficient of the HG billiard turned out to be fractal, here as a function of the particle energy. Interestingly, the geometry of this model is the same as the one used in a recent experiment on granular material [Pre02], where a granular gas was driven by a vibrating surface coated with a periodic grid of small steel balls. A very 153

klages˙book

May 8, 2007

4:1

154

World Scientific Book - 9in x 6in

Deterministic diffusion of granular particles

related setup was studied theoretically and experimentally in Refs. [Far99; Far02], where a one-dimensional channel of asymmetric scatterers was restricted onto the circumference of a vibrating plate thus forming a granular ratchet.1 These crosslinks to granular material motivate a modification of the HG system by including some friction at the collisions while compensating for the loss of energy by vibrating the corrugated floor periodically. The resulting system we call the bouncing ball billiard, see the inset of Fig. 9.1. It thus mimicks the experimental situation in Ref. [Pre02], however, because of dissipation the dynamics of this model is now inherently different from the Hamiltonian one of the HG system. This becomes obvious for the limiting case of a flat vibrating surface. Here our model reduces to the famous bouncing ball problem, where a particle with a restitution coefficient α bounces vertically on a vibrating floor in the gravitational field g [Pie83; Pie88; Kov88; Tuf92; Meh90; Luc93; Lic92; Guc90]. Focusing first onto this non-diffusing bouncing ball, we assign the position y = A sin(ωt) to the floor at time t with amplitude A and angular velocity ω = 2πf , where f is the vibration frequency. The parameters A, ω and g can be grouped together into Γ := Aω 2 /g so that there are only two control parameters Γ and α. A fundamental property of this system is the existence of phase locking [Pie83; Pie88; Kov88; Tuf92; Meh90; Luc93; Lic92; Guc90], that is, if the time of flight between two collisions is k times the period of the vibration the bouncing ball exhibits a k/1-resonance. Linear stability analysis yields that the corresponding regions in the parameter space are bounded by [Hon89a; Hon89b; Meh90; Luc93] ( 2 2 )1/2 2(1 + α2 ) 1−α 1−α <Γ<π k + . (9.1) πk 1+α 1+α π(1 + α)2 In the following we study the impact of these resonances on diffusion in the bouncing ball billiard. The parameters of the system are chosen in order to be close to a possible experimental setup [Pre02]. The corrugated floor is constructed by circular scatterers with curvature K = 0.04mm−1 . This value was chosen for the surface to be sufficiently flat such that the bouncing ball resonances Eq. (9.1) survive the dynamical instability generated by the geometry. The intersection of any two arcs form wedges, whose centers are at a distance of ` = 2mm on a line. We applied periodic boundary conditions after one arclength. For the amplitude of the vibrations we chose

1 For

a definition of ratchets see Section 3.4.

klages˙book

4:1

World Scientific Book - 9in x 6in

klages˙book

155

9.1. Resonances and diffusion in the bouncing ball billiard

2/1

1/1

3/1

(b)(i) (b)(ii)

200

(b)(ii)

(c)

(d)

(b)(i) (b)(i)

150 y g

D

May 8, 2007

vy

100

vx A

ω

0

2

4

6

70

75 f

80

85

x

50

0 55

60

65

90

95

Fig. 9.1 Simulation results for the diffusion coefficient D of the bouncing ball billiard (inset) as a function of the vibration frequency f of the floor. The arrows on top of the frame show the simple bouncing ball resonances for α = 0.5 described by Eq. (9.1), which approximately yield an enhancement of diffusion. The two cross (×) symbols at f = 74.0 and 74.5 hold for the second resonance, which is produced by specific initial conditions. Other arrows and labels mark more subtle dynamical regimes as described in the text. In this and in the following figures D is given in units of mm2 /s, f in Hz. The standard deviation errors for D are smaller than the magnitude of the symbols.

A = 0.1mm. Between two inelastic collisions at the surface the motion of a particle is a free flight in the gravitational field g ' 9.81m/s2. The inelasticity is modeled by the perpendicular and tangential restitution coefficients α and β yielding the collision rules + − v⊥ n − vci⊥ n = α (vci⊥n − v⊥n )

(9.2)

vk+n − vcik n = β (vk−n − vcik n ) ,

(9.3)

and

where vci is the velocity of the vibrating floor and v⊥ (vk ), vci⊥ (vcik ) denote the perpendicular (tangential) component of the particle’s, respectively the floor’s velocity with respect to the surface at the scattering point. For the value of the normal restitution coefficient we chose α = 0.5. According to Eq. (9.1), at this value of α at most one k/1-resonance is possible for any Γ.

May 8, 2007

4:1

156

World Scientific Book - 9in x 6in

Deterministic diffusion of granular particles

By setting the horizontal restitution coefficient to β = 0.99 long horizontal creeping orbits, where a particle moves over many scatterers in the form of directed sequences of very tiny hops, are eventually eliminated. However, short creeps, being considerably smaller than one arclength, are still possible. Hence, for the set of parameters chosen above the dynamics of the bouncing ball billiard should be strongly determined by the resonances, which allows the study of how this phenomenon affects the diffusive properties of the model. The diffusion coefficient was computed numerically according to the Einstein formula Eq. (7.12) restricted to one dimension, that is, with a factor of two instead of four in the denominator, cf. Eq. (2.2). The simulated ensemble of points consisted of 103 initial conditions distributed uniformly over the phase space region 0 ≤ s ≤ 2.0004mm, |vx+ | ≤ 40mm/s and 0 < vy+ ≤ 40mm/s defined by the clockwise arclength at the impact s and the horizontal and vertical velocities vx+ and vy+ just after a collision. At resonant frequencies the dynamics may be nonergodic [Tuf92], nevertheless the average kinetic and total energy converged to constant values after transient times indicating the existence of a nonequilibrium steady state. Fig. 9.1 shows a complex scenario of different dynamical regimes under variation of the vibration frequency: (a) If Γ < 1 the maximal acceleration of the surface is smaller than the gravitational force, i.e., Aω 2 < g. Consequently, for f < 49.77Hz once a particle sticks to the surface it never leaves it again. The first frequency at which a particle moves out of a wedge for any initial condition is f = 58.0Hz, which denotes the onset of diffusion. (b) In the interval f ∈ [58.0, 65.6]Hz the 1/1-resonance strongly determines the dynamics and we identify two different dynamical regimes: (i) parameter regions where the system has two attractors corresponding to f ∈ [57.0, 57.8]Hz, f ∈ [60.3, 61.8]Hz and f ∈ [63.4, 65.6]Hz. In these regions some initial conditions lead to the 1/1-resonance, whereas other ones generate a time-to-time creeping motion on the surface. (ii) The other dynamical regime exhibits only one attractor with all inital conditions leading to the same resonant behavior, which occurrs for f ∈ [57.9, 60.2]Hz and f ∈ [61.9, 63.3]Hz. This resonant motion becomes regular only in the vertical direction where it is very similar to the 1/1-phase locking of the bouncing ball while in the horizontal direction the system shows a highly irregular normal diffusive dynamics. In this regime the interaction with the surface consists only of regular collisions without any creeps, see the attractor of Fig. 9.2.

klages˙book

May 8, 2007

4:1

9.2.

World Scientific Book - 9in x 6in

+ Diffusion

by correlated random walks

klages˙book

157

Fig. 9.2 Projections of the phase space in (s, vx+ ) coordinates, where s is the position of the colliding particle on the arc and vx+ the horizontal velocity after a collision. The left figure depicts the attractor at f = 58.0Hz for which Fig. 9.1 indicates a 1/1-resonance. The right half shows that in the case of non-resonant dynamics at f = 85.2Hz the particle may partly exhibit “creepy motion”. The position s is in units of mm, the velocity v x+ in mm/s.

(c) For f ∈ [67.2, 81.0]Hz the escape from the wedges takes long times due to long creeps. In the vicinity of 74.0Hz the 2/1-resonance shows up, however, it is only exhibited by a few initial conditions with vy+ > 100m/s. Hence, it is considerably weaker than the previous one. (d) The last dynamical regime we study here is in the interval f ∈ [81.1, 95.0]Hz demonstrating a significant increase of the diffusion coefficient over a larger frequency interval. The corresponding attractor reveals a from time-to-time “creepy motion”, see the almost vertical lines in Fig. 9.2 indicating collisions with the surface at almost any time step and with decreasing velocity. 9.2

+

Diffusion by correlated random walks

We now quantitatively analyze the irregular structure of the diffusion coefficient of the bouncing ball billiard by applying the theoretical approach developed in the previous two chapters. That is, we derive some simple analytical and numerical approximations starting with a prediction from random walk theory. For frequencies where diffusion is small the wedges of the billiard are like traps for the bouncing particle. In this case the diffusion coefficient is

4:1

World Scientific Book - 9in x 6in

158

klages˙book

Deterministic diffusion of granular particles

45 40 35 30 25 D

May 8, 2007

20 15 10 5 0 65

70

75

80 f

85

90

95

Fig. 9.3 Evaluations of the random walk formula Eq. (9.4) for the diffusion coefficient D with frequencies f > 65.6Hz. The solid line corresponds to the diffusion coefficient of Fig. 9.1. The dotted monotonically increasing curve represents the analytical formula Eq. (9.6), the dashed irregular graph close to zero depicts Eq. (9.4) with a numerically computed escape time.

determined by the average escape time τ needed to leave a trap leading to the Machta-Zwanzig random walk formula [Mac83] Drw (f ) =

`2 , 2τ (f )

(9.4)

cf. Eqs. (7.13), (8.4) and supplements, where ` is the distance between two trap centers. An analytical estimate of the average escape time is obtained by employing the average kinetic energy Ex , τ'

` ` , '√ < vx > 2Ex

(9.5)

where < vx > is the average velocity in the horizontal direction. Ex can be calculated from the energy balance Ex = E − Ey − mg y¯ in which Ey stands for the average kinetic energy related to vy and mg y¯ denotes the average potential energy of the particle. The total average energy E is approximated with respect to the harmonically oscillating surface to E ' A2 ω 2 /2, where the average height is set equal to the amplitude of the vibrations.2 Computer simulations indicate that only a small percentage 2A

simplified model with sawtooth vibrations supports the fact that the total average

May 8, 2007

4:1

9.2.

World Scientific Book - 9in x 6in

+ Diffusion

by correlated random walks

klages˙book

159

of the energy pumped vertically into the system is transferred into the horizontal direction, Ey ' 19Ex . With this input the energy balance yields 2Ex ' A2 ω 2 /20 − mgA/10. Combining Eq. (9.5) with Eq. (9.4) leads to the diffusion coefficient r `p ` A2 ω 2 mgA Dc (f ) ' 2Ex ' − . (9.6) 2 2 20 10

Fig. 9.3 shows that this result is close to the one for which τ in Eq. (9.4) was obtained from computer simulations. Note that Eq. (9.6) indicates the onset of a monotonically increasing diffusion coefficient, as one would expect from stochastic theory if the resonances were not present. The numerical approximation of Eq. (9.4) roughly follows this coarse functional form while qualitatively being more similar to the numerically exact values. However, quantitatively it is much too small indicating the existence of long-range dynamical correlations that go beyond a simple random walk picture. In Chapters 7 and 8 we have seen that such higher order memory effects can systematically be taken into account by using the Taylor-Green-Kubo formula for diffusion in billiards. As in the case of the Einstein formula, the one-dimensional version is obtained from the two-dimensional result by simply multiplying Eq. (7.15) by a factor of two. Correspondingly, for the bouncing ball billiard j(xk ) denotes jumps at the kth time step τ on a one-dimensional lattice with sites at distance `. Approximations Dn of order n are now defined by the respective one-dimensional version of Eq. (7.16). For this hierarchy of approximations Dn (f ) one can numerically compute the conditional probabilities p(αβγ . . .) contained in Eq. (7.16). As in Chapters 7 and 8, the random walk solution Eq. (9.4) is represented by the first term, and the higher order corrections yield information about the impact of dynamical correlations on the diffusion coefficient. Fig. 9.4 shows that the convergence of this sum is not monotonic in the frequency. This confirms the existence of irregularities on fine scales and identifies their origin in terms of long-time dynamical correlations. As was amply discussed in the previous chapters, this behavior is well-known from related systems exhibiting deterministic diffusion thus pointing towards a possibly fractal structure of the diffusion coefficient in the bouncing ball billiard. We remark that in Ref. [Kla04b] the same model was studied for larger curvature K and smaller perpendicular friction α. Again the diffusion coefenergy scales with ω 2 [War95]. That its prefactor is most conveniently chosen to be identical to the one of a particle oscillating harmonically in a quadratic potential appears to be, as far as we can tell, a mere coincidence.

4:1

World Scientific Book - 9in x 6in

160

klages˙book

Deterministic diffusion of granular particles

30 25 20 D

May 8, 2007

15 10 5 0 84.4

84.8

85.2

85.6

86

86.4

f Fig. 9.4 Blowup of the frequency interval f ∈ [84.6, 86.4]Hz in Fig. 9.1. The solid line on top shows the respective diffusion coefficient D, the lowest dotted line depicts the random walk approximation Eq. (9.4) with the escape time computed numerically. The other lines show the approximations Dn (f ) obtained from the one-dimensional version of the Taylor-Green-Kubo formula Eq. (7.16) for n = 0, 2, 4 and 7 (from bottom to top).

ficient turned out to be a highly irregular function of the frequency, however, surprisingly the origin of most of these irregularities required yet another explanation related to more subtle resonances called spiral modes. 9.3

Vibratory conveyors

Interestingly, bouncing ball-type models have been used for quite some time in order to understand transport on vibratory conveyors, also called vibro-transporters, which are common carriers of agricultural material and related matter. For a corrugated version that very much reminds one of the bouncing ball billiard and is widely used in industrial applications see Ref. [Per92]. In order to transport particles on such devices one generates an average drift velocity by means of some symmetry breaking. For tilted systems with a flat surface Hongler et al. [Hon89a; Hon89b] predicted an integer phase locking leading to local extrema in the transport rate as a function of a control parameter, as was recently verified in experiments [Han01]. Grochowski et al. constructed a circular vibratory conveyor, where

May 8, 2007

4:1

9.4.

World Scientific Book - 9in x 6in

∗ Summary

klages˙book

161

the trough was driven by asymmetric oscillations [Gro03; Rou04; Gro04]. Both for a tracer particle in a layer of granular matter and for a single highly inelastic test particle they observed a non-monotonic current reversal of the transport velocity under variation of control parameters. This ratchet-like effect was recently reproduced in simulations [hor05] again indicating some underlying phase locking. We remark that in contrast to single bouncing balls, the core of granular matter research deals with highly interacting many-particle systems, where phenomena like structure formation [Rou04] arise because of statistical physical mechanisms rather than being of an intricate nonlinear deterministic origin. The main message we wish to convey with this chapter is that one might observe surprising transport properties for even a single, or just a few, granular particles moving on corrugated surfaces. Looking for such counter-intuitive phenomena in experiments might thus open up new directions of research. Our conclusion is that in vibro-transporters not only the current, but also other transport property quantifiers such as the diffusion coefficient may be irregular functions of control parameters. For the bouncing ball billiard we have identified two mechanisms generating irregular parameter dependencies, which are (i) phase locking and (ii) higher-order dynamical correlations leading to fractal-like irregularities on finer and finer scales. A third one is the spiral modes discussed in Ref. [Kla04b]. Searching for irregular transport coefficients in experiments may thus pave the way for a very fine-tuned control of the dynamics of vibro-transporters. This may eventually provide a basis for practical applications such as, for example, sophisticated methods of particle separation [Far02].

9.4

∗

Summary

(1) In order to bring the phenomenon of fractal transport coefficients to physical reality, we have proposed a simple model for diffusion of a granular particle on a vibrating surface subject to an external gravitational field. In analogy to the dynamical systems studied previously, this surface was equipped with a periodic grid of circular scatterers with which the moving particle collides inelastically. Computing the frequency-dependent diffusion coefficient for this model we obtained again a highly irregular curve. We showed that the largest maxima of this function are due to phase locking as known for a ball bouncing

May 8, 2007

4:1

162

World Scientific Book - 9in x 6in

Deterministic diffusion of granular particles

inelastically on a flat vibrating plate. We furthermore numerically analyzed the phase space of the dynamical system showing that it exhibits different types of dynamics which change in a complicated fashion under parameter variation. (2) We finally applied the correlated random walk approach discussed in the previous two chapters to the frequency-dependent diffusion coefficient of the bouncing ball billiard. We first heuristically derived a simple random walk approximation for our model that qualitatively matched well to our results from computer simulations. By successively working out the first terms of the associated Taylor-Green-Kubo series expansion, we argued that, again, irregularities on fine scales detected in simulations are largely due to long-time dynamical correlations. With this model we hope to stimulate experimental work on closely related physical systems trying to verify the predicted existence of irregular structures in parameter-dependent diffusion coefficients. Applications of these findings may be particularly useful for transport on industrial vibratory conveyors.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

PART 2

Part 2: Thermostated dynamical systems

163

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

This page intentionally left blank

164

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Chapter 10

Motivation: coupling a system to a thermal reservoir

In this chapter we introduce some basic concepts, models and notations that are used throughout Part 2. In Section 10.1 we outline what thermostats are and explain why they are indispensable for studying steady states under typical nonequilibrium conditions. Section 10.2 briefly reminds the reader of the Langevin equation. Its derivation exemplifies two extreme cases of modeling thermal reservoirs, that is, either in terms of purely Hamiltonian dynamics or by using a dissipative, irreversible and stochastic equation of motion. In Section 10.3 we elaborate on the general functional form of the velocity distributions of a subsystem and thermal reservoir in thermal equilibrium. Section 10.4 contains a detailed definition of the periodic Lorentz gas, the main model of Part 2, which we will study under application of different thermostats. 10.1

∗

Why thermostats?

A common problem of thermostating is how to cool down a bottle of beer on a hot summer day [Rei65]. One option is to simply put the bottle into a swimming pool. Microscopically, there occurs a transfer of energy between the beer molecules composing a fluid at temperature TB and the water molecules in the pool at temperature TP < TB . Here we define temperature operationally by equipartitioning of energy onto all degrees of freedom [Rei65; Hua87] thus relating temperature to the average kinetic energy of a system. Under the assumption that the water molecules are suitably interacting with each other such that equipartitioning holds, the excess energy related to the temperature difference TB − TP flows on average across the surface of the beer bottle into the surrounding fluid. Under the very same assumption, it equally distributes onto all water molecules. However, since 165

klages˙book

May 8, 2007

4:1

166

World Scientific Book - 9in x 6in

Motivation: coupling a system to a thermal reservoir

the number of water molecules is extremely large the increase of the average temperature of the pool, in terms of the average kinetic energy per particle, will be negligibly small. What we have sketched here is just the well-known process of classical thermal equilibration between a sufficiently small subsystem and a surrounding thermal reservoir, where both systems are initially at two different temperatures. After equilibration both systems will be approximately at the same temperature, TB ' TP . However, so far we have only discussed the simple case of relaxation to thermal equilibrium when subsystem and reservoir together form a closed system. As a second, slightly more complicated example, think of a nonequilibrium situation in an open system such as a light bulb connected to a power source. The applied voltage generates a current, and the thread in the bulb, as an Ohmic resistance, exhibits Joule heating. In a naive microscopic picture this heating may be understood as follows: The applied electric field accelerates the electrons in the resistance on average. Consequently, if there were no mechanism for reducing their average kinetic energy the gas of electrons simply heats up, and no stationary current exists. On the other hand, the single electrons collide with the atoms constituting the resistance. During these collisions they exchange energy with the atoms, mainly by exciting lattice modes and eventually causing the whole resistance to heat up. However, across its surface the resistance allows an average flow of energy into the surrounding medium by collisions between the atoms of the thread and the atoms or molecules of the gas. In the case of the light bulb this flow of energy causes the gas around the thread to glow. The flow of energy into a thermal reservoir thus properly counterbalances the pumping of energy into the system by the external electric field and enables the system to evolve into a nonequilibrium steady state (NSS). This means that the statistical physical quantities describing the system on macroscopic scales are constant in time, despite the fact that the system is no longer in thermal equilibrium [dG62]. For our example, the existence of a NSS implies that under application of the voltage the electron gas eventually exhibits a stationary current with constant temperature. We note that the proper definition of a nonequilibrium temperature is a subtle problem in itself [Rug97; Mor99], although operationally the principle of equipartitioning of energy may be used [Pos88; Eva90b; Mor98; Hoo99]. If the existence of a NSS is due to the action of a thermal reservoir we say the system is properly thermostated. In other words, thermostats are mechanisms by which the internal en-

klages˙book

May 8, 2007

4:1

10.2.

World Scientific Book - 9in x 6in

∗ Modeling

thermal reservoirs: the Langevin equation

klages˙book

167

ergy of a many-particle system, and thus its temperature, can be tuned onto a specific value. Any thermostat may be thought of being associated with some thermal reservoir consisting of an infinite number of degrees of freedom, thus being large enough to absorb any amount of energy pumped into the system. Thermostats can be applied in order to achieve equilibration to a global thermal equilibrium or for sustaining a NSS in a nonequilibrium situation with a flux of energy through the system, such as induced by external fields or by imposing temperature or velocity gradients. In the same vein, other constraints instead of fixing temperature may be imposed onto a system leading, e.g., to constant pressure or constant stress ensembles [All87; Eva90b; Nos91; Hes96a; Tuc00]. The first question we must focus on before we elaborate on NSS is therefore: How can we suitably amend Newton’s equations of motion in order to model energy dissipation into a thermal reservoir? The basic problems involved here are exemplified by the Langevin equation. At this level of discussion we restrict ourselves to equilibrium situations, where the construction of thermal reservoirs appears to be a rather technical problem. However, our main objective is to eventually apply thermal reservoirs to nonequilibrium situations yielding NSS. In later chapters we will inquire to which extent the properties of NSS depend on the modeling of different thermal reservoirs and to which extent they might be universal.

10.2

∗

Modeling thermal reservoirs: the Langevin equation

The Langevin equation [Lan08] provides a simple theoretical framework for Brownian motion: A sufficiently small particle immersed in a fluid performs irregular motion resulting from the collisions between the atoms or molecules of the medium and the particle [Rei65; Wax54; vK92; Pat88; Gal99]. For coal dust particles on the surface of alcohol this phenomenon was observed by the Dutch physician J. Ingenhousz in 1785. However, it became known more widely only later on by the work of the Scottish botanist R. Brown in 1827, who reported similar irregular movements of pollen grains under a microscope. The name Brownian motion was coined by A. Einstein in his famous work from 1905 describing this irregular motion in terms of diffusion processes. A related approach was already proposed by the French mathematician L. Bachelier in 1900 in order to understand the dynamics of stock values. Einstein’s work led J.B. Perrin to the experimental measurement of the Avogadro number in 1908 and also motivated P.

May 8, 2007

4:1

168

World Scientific Book - 9in x 6in

klages˙book

Motivation: coupling a system to a thermal reservoir

Langevin’s modeling of Brownian motion published in the same year.1 We first outline some fundamental features of Langevin’s stochastic approach to Brownian motion. In the following chapters we will repeatedly come back to these characteristic properties by discussing to which extent they are reproduced in deterministic chaotic models of Brownian motion. According to Langevin, a Brownian particle is randomly driven by instantaneous collisions with the surrounding particles. On the other hand, energy is removed from the system by some bulk friction. In one dimension this dynamics is modeled by amending Newton’s equations of motion to r˙ = v v˙ = −αv + F(t) ,

(10.1)

where r and v denote the position and the velocity of the Brownian particle. For convenience, here and in the following we set constants like the mass m of the particle and the Boltzmann constant kB equal to one. F stands for white noise, that is, a δ-correlated stochastic force with zero mean modeling the random collisions with the fluid particles, whereas α is a Stokes friction coefficient. In a steady state the molecular stochastic forces and the friction balance each other leading to the fluctuation-dissipation theorem [Rei65; Wax54; Pat88; Zwa01] Z ∞ 1 dt < F(0)F(t) > , (10.2) α= 2T −∞

where < . . . > denotes an ensemble average over the noise acting on the moving particles. The expression CF (t) :=< F(0)F(t) > is an example of a correlation function. α is in turn related to the diffusion coefficient D of the Brownian particle according to the Einstein relation D = T /α ,

(10.3)

where T is the temperature obtained from equipartitioning of energy, T =< v 2 >. D is defined independently via the mean square displacement, 2

< [x(t) − x(0)] > . (10.4) t→∞ 2t cf. Eqs. (2.2), (7.12) for related formulas. Here the angular brackets denote an equilibrium ensemble average over moving particles. Alternatively, D can be obtained from the Taylor-Green-Kubo formula for diffusion Z ∞ dt < v(0)v(t) > , (10.5) D= D := lim

1 For

0

this short historical note on Brownian motion we largely followed the summary in Ref. [Met00]; for more details see, e.g., Refs. [Bru76; Sta89; Ebe05].

May 8, 2007

4:1

10.2.

World Scientific Book - 9in x 6in

∗ Modeling

klages˙book

169

thermal reservoirs: the Langevin equation

see Eqs. (4.22), (7.2), (7.15) for versions of this formula for lattices, which is an exact transformation of Eq. (10.4) [Zwa01]. The Taylor-Green-Kubo formula Eq. (10.5) and the fluctuationdissipation theorem Eq. (10.2) are formally analogous in relating a macroscopic quantity characterizing the fluid to an integral over some correlation function. Cv (t) :=< v(0)v(t) > is called the velocity autocorrelation function of the moving particle. As can easily be shown by solving the Langevin equation Eq. (10.1) for Cv , the velocity autocorrelation function decays exponentially in time thus ensuring that, in terms of the Taylor-Green-Kubo formula Eq. (10.5), the diffusion coefficient exists. If one considers the Brownian particle as a subsystem and the surrounding particles as an infinite dimensional thermal reservoir, the Langevin equation precisely models the situation where a subsystem suitably interacts with a thermal reservoir. However, so far we have motivated this equation purely heuristically. It is therefore interesting to discuss a simple derivation of the above Langevin equation that actually starts from a fully Hamiltonian modeling of subsystem plus thermal reservoir in which the heat bath consists of an infinite number of harmonic oscillators [Zwa73; For87; Kub92; Stu99; Zwa01].2 This nicely demonstrates the simplifying assumptions one has to make in order to arrive at a simple modeling of thermal reservoirs by starting from first principles. Let the Hamiltonian of the Brownian particle be Hs =

v2 2

(10.6)

and let the heat bath of harmonic oscillators be

HB =

X j



ω2 v2  j + j 2 2

γj xj − 2 x ωj

!2  

,

(10.7)

where the oscillators exhibit a special coupling to the subsystem. Here ωj is the frequency of the jth oscillator and γj the strength of the coupling between the Brownian particle and the jth oscillator. The Hamiltonian equations of motion for the combination of subsystem plus thermal reservoir 2 In

Ref. [Stu99] this model was attributed to Ford and Kac [For87], however, it already appears at least in the paper by Zwanzig [Zwa73]; see also further references in Ref. [For87].

May 8, 2007

4:1

170

World Scientific Book - 9in x 6in

klages˙book

Motivation: coupling a system to a thermal reservoir

then read x˙ = v X v˙ = γj j

γj xj − 2 x ωj

!

(10.8)

x˙j = vj

v˙j = −ωj2 xj + γj x .

(10.9)

The formal solution of the inhomogeneous differential equation of second order Eq. (10.9) reads Z t sin(ωj t) sin(ωj (t − s)) xj (t) = xj (0) cos(ωj t) + vj (0) + γj . ds x(s) ωj ωj 0 (10.10) Performing integration by parts for the latter integral and thereafter putting Eq. (10.10) into Eq. (10.8) yields the formal Langevin equation Z t v(t) ˙ =− ds K(s)v(t − s) + F (t) . (10.11) 0

By comparing this equation to the former stochastic Langevin equation Eq. (10.1), the first term on the right hand side is identified as some nonlinear, non-Markovian friction containing the memory function K(t) =

X γj2 cos(ωj t) ωj2 j

.

(10.12)

The second term on the right hand side of Eq. (10.11) must consequently yield the “noise”, which is given explicitly by ! X γj sin(ωj t) X + γj xj (0) − 2 x(0) cos(ωj t) . (10.13) F (t) = γj vj (0) ωj ωj j j Note that F (t) is fully deterministic in depending on the initial conditions of all heat bath variables. In order to recover Eq. (10.1) both expressions are now simplified as follows: If the spectrum of frequencies ωj is thought to be continuous, the memory function Eq. (10.12) may be written as a Fourier integral. By assuming convenient functional forms for the density of states of ωj and for the coupling coefficients γj the memory function can be replaced by a delta function, K(t) ∼ δ(t). In this special case the Langevin equation Eq. (10.11) is Markovian and the ordinary Stokes friction of Eq. (10.1) is recovered.

May 8, 2007

4:1

10.2.

World Scientific Book - 9in x 6in

∗ Modeling

thermal reservoirs: the Langevin equation

klages˙book

171

As far as the “noise” is concerned, according to Eq. (10.13) F (t) is a linear function of the initial positions and velocities of all bath oscillators, which are independent degrees of freedom. Hence, there are no correlations between these variables. If all initial conditions are sampled randomly from a canonical distribution one can show that the fluctuation-dissipation theorem takes the form [Zwa01] < F (t)F (t0 ) >= T K(t − t0 ) .

(10.14)

Under the above assumption that the memory function is Markovian this equation boils down to Eq. (10.2), or in other words, the fluctuations induced by F (t) turn out to be white noise. We see that, in this case, the noise is put in “by hand” into Eq. (10.11) by assuming a suitable distribution of initial conditions for the heat bath variables. In summary, we have discussed the connection between two extreme versions of modeling the interaction of a subsystem consisting of a single particle with a thermal reservoir: One employs the fully Hamiltonian, i.e., deterministic, time-reversible and phase space preserving equations of motion Eqs. (10.6) and (10.7). To start this way is satisfactory from a microscopic point of view. However, note that the statistical properties of the harmonic oscillator heat bath are pre-determined by the choice of initial conditions supplemented by the specific form of the memory kernel in Eq. (10.14). The latter yields the decay of the force correlations, hence the problem of introducing randomness is shifted towards choosing a proper initial distribution for the oscillator variables. A realistic heat bath, on the other hand, should generate a (canonical) equilibrium distribution of position and velocity variables in a self-contained way, that is, due to the action of the dynamical system defining the bath and irrespective of any specific initial conditions. That this is not the case for the harmonic oscillator heat bath one may consider as a deficiency of the model. Another disadvantage becomes clear in nonequilibrium situations, where the sum over harmonic oscillators in Eq. (10.7) must be infinite in order to allow for the existence of a NSS. Though taking this limit does not pose any crucial difficulty for analytical investigations, the Hamiltonian scheme thus becomes very inconvenient for computer simulations. An alternative type of modeling consists of the very intuitive linear and Markovian Langevin equation Eq. (10.1). However, according to its derivation from the harmonic oscillator heat bath these nice properties emerge only because of profound approximations. Even more, the dynamics eventually turns out to be non-Hamiltonian, that is, stochastic, irreversible and,

May 8, 2007

4:1

172

World Scientific Book - 9in x 6in

Motivation: coupling a system to a thermal reservoir

as a consequence of Stoke’s friction, on average phase space contracting; see Ref. [Hoo99] for related arguments. Hence, both types of heat bath model present somewhat limiting cases of the whole problem. In-between there is plenty of room for generalizations of modeling thermal reservoirs, in two directions: Starting from some Hamiltonian equations of motion, one may wish to put the derivation of equations modeling the action of a heat bath onto a small subsystem onto more rigorous grounds. This approach was pursued by Beck et al. [Bec87; Bec95; Bec96]. Here the analysis starts from a time-discrete version of the Langevin equation Eq. (10.1) in which randomness is generated by a deterministic chaotic map thus trying to understand to which extent deterministic chaos modeling reservoir degrees of freedom can mimick stochastic noise. In a similar vein, Just and Kantz et al. considered chaotic dynamical systems with time scale separation, where the fast degrees of freedom may be related to the action of a thermal reservoir [Jus01; Jus03; Kan04; Rie05]. Note that similarly to Beck’s approach they modeled thermal reservoirs by dynamical systems that consist of only very few degrees of freedom exhibiting a chaotic dynamics. By using projection operator techniques for the phase space densities of the dynamical system variables they studied to which extent the fast degrees of freedom can be replaced by stochastic noise acting on the slow variables of respective subsystems, and what the resulting equations of motion for the subsystems look like. A related problem was investigated by Eckmann et al., who mathematically analyzed the nonequilibrium properties of a chain of anharmonic oscillators coupled to two heat baths at different temperatures modeled by Hamiltonian wave equations [Eck99b; Eck99a]. Deriving rigorous results for the dynamical properties of thermal reservoirs and the associated subsystems by starting from first principles is hard work, even in equilibrium situations. We are not aware that there is yet a general theory for this problem. A major difficulty appears to be that the precise characteristics of a thermal reservoir depend on the detailed nonlinear and non-Markovian properties of the underlying equations of motion. Thus, typically one has to employ profound approximations in order to come up with some convenient statistical description, as indicated above. In nonequilibrium situations the problem of characterizing steady states with respect to their chaotic and dynamical properties becomes even more peculiar, which connects to our respective discussion in the introduction Chapter 1.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

10.3. Equilibrium velocity distributions for thermostated systems

klages˙book

173

An alternative to such a rigorous bottom-up approach is provided by studying successively more sophisticated models of thermal reservoirs by aiming to learn about the detailed statistical and dynamical properties of the associated equilibrium and nonequilibrium steady states generated by these models. One may then inquire about general characteristics that are valid for all these models. We emphasize that this top-to-bottom approach naturally bears other risks. In particular, one may too quickly draw conclusions from the fascinating properties of a specific toy model concerning universal features. But this can be cured by trying to construct counterexamples. Of course, top-to-bottom approaches towards the physics of steady states are of a more heuristic nature. Still, in order to qualify as a thermal reservoir all models have to fulfill certain basic criteria. In the following we will require that, like ordinary equations of motion of moving particles, they must be deterministic, time-reversible, and that they generate a NSS under nonequilibrium conditions. However, some details are still missing. So far we have focused on the equations of motion of a subsystem plus thermal reservoir only. We did not explicitly discuss the detailed properties of the associated probability densities of a subsystem and thermal reservoir and what they should look like, at least in equilibrium situations. This will be performed in the next section.

10.3

Equilibrium velocity distributions for thermostated systems

We shall now look at a subsystem coupled to a thermal reservoir from a purely statistical point of view. That is, we do not assume any detailed knowledge about the underlying equations of motion. We are only interested in the probability densities for the velocities of subsystem and thermal reservoir, which we call velocity distribution functions, in the general case of thermal equilibrium. We do not discuss the probability densities for the position coordinates associated with these degrees of freedom, since the action of the thermal reservoir primarily concerns an exchange of energy related to the velocities. Our goal is to assess how different dimensionalities of a subsystem and thermal reservoir, if the reservoir is first thought to be finite dimensional, affect the functional forms of the corresponding velocity distribution functions. In order to illustrate the general case we start by discussing the most

May 8, 2007

4:1

174

World Scientific Book - 9in x 6in

Motivation: coupling a system to a thermal reservoir

trivial example, where subsystem and thermal reservoir each consist of only one degree of freedom. Afterwards we switch to the general situation in which there is a subsystem with ds degrees of freedom interacting with a thermal reservoir of dr degrees of freedom, where typically ds < dr and where the total number of degrees of freedom is d = ds + dr .3 Let us consider a large collection of points in the full phase space of the system, which in general may be restricted by some boundary conditions. The equilibrium velocity distribution function ρ = ρ(v1 , v2 ) for d = 2 is then defined with respect to this ensemble of points moving according to the same equations of motion. It is the number of points one can find in the phase space volume element dv1 dv2 centered around the velocity vector (v1 , v2 ) at time t divided by the size of this volume element and by the total number of points N of the ensemble. By integrating over the whole accessible phase space the velocity distribution is properly normalized to one. Let us now assume that we are in an equilibrium situation where equipartitioning of energy holds between all degrees of freedom and that the total energy of a subsystem plus reservoir is conserved. For sake of simplicity we may assume that there is only kinetic energy as, e.g., in a gas of hard spheres. Hence, all points inp velocity space should be uniformly distributed on a circle of radius v := v12 + v22 , where for convenience we set v = 1, see Fig. 10.1 for an illustration. Subsystem plus reservoir thus obey the microcanonical distribution ρ(v1 , v2 ) ∼ δ(1 − v12 − v22 ) . (10.15) It is now straightforward to calculate the velocity distribution functions of the subsystem, respectively of the thermal reservoir, which are obviously identical, by projecting out one of the two degrees of freedom from this microcanonical distribution. One way of how to do this is by using the polar coordinates v1 = cos φ and v2 = sin φ, where φ denotes the polar angle in velocity space depicted in Fig. 10.1. The probability density corresponding to φ is then simply ρ(φ) = const. On the other hand, v1 is a function of φ. According to conservation of probability the probability densities for v1 and φ are related to each other by ρ(φ)|dφ| = ρ(v1 )|dv1 | . (10.16) Hence, ρ(v1 ) is easily calculated in terms of ρ(φ) by differentiating v1 leading to 1 ρ(v1 ) = p , (10.17) π 1 − v12 3 This

problem has been posed as an exercise in the book of van Kampen on p.11 [vK92].

klages˙book

4:1

World Scientific Book - 9in x 6in

klages˙book

10.3. Equilibrium velocity distributions for thermostated systems

175

v 2

v φ

v 1

2 1.5

ρ(v1)

May 8, 2007

1

0.5 0-1

-0.5

0 v1

0.5

1

Fig. 10.1 Projection of the two-dimensional microcanonical velocity distribution Eq. (10.15) onto one component. The density of points is uniform on a circle of radius v = 1. The velocity vector is composed of the two velocity components v 1 and v2 , and its direction is determined by the polar angle φ. The lower figure represents the one-component velocity distribution ρ(v1 ) Eq. (10.17) calculated from projecting out v2 from Eq. (10.15).

where the prefactor is obtained from normalization. As indicated in Fig. 10.1, this result matches physical intuition: By projecting the uniform distribution of points on the circle onto the v1 -axis it is clear that there must be a higher density of points nearby the values of v2 = 0 on the circle, whereas around v1 = 0 the density of points must be minimal. The main message of this trivial example is that non-uniform probability densities, which might otherwise be considered as an indicator of some nonHamiltonian average phase space contraction, may very well coexist with a uniform, microcanonical one as it characterizes ordinary Hamiltonian dynamical systems. The obvious reason is that one degree of freedom got projected out respectively, so we neglected some information by not looking

May 8, 2007

4:1

176

World Scientific Book - 9in x 6in

klages˙book

Motivation: coupling a system to a thermal reservoir

at the full combination of a subsystem plus thermal reservoir. Along the same lines and under the same conditions as described above, i.e., by assuming thermal equilibrium, equipartitioning of energy and a constant total kinetic energy E, the general case of a ds -dimensional subsystem interacting with a dr -dimensional thermal reservoir can now be studied. In this case the full velocity distribution function for subsystem plus thermal reservoir obeys the microcanonical distribution ρ(v1 , . . . , vd ) ∼ δ(2E − v12 − . . . − vd2 )

(10.18)

with d = ds + dr and vk , k = 1, . . . , d, as all the velocities. That is, all points √ in velocity space are uniformly distributed on a hypersphere with radius 2E. How do the corresponding velocity distributions of subsystem and reservoir now look for arbitrary numbers of degrees of freedom d, ds and dr ? Again, in order to calculate these distribution functions one has to project out the respective number of degrees of freedom from Eq. (10.18). Here we wish to focus on eliminating d − 1 or d − 2 degrees of freedom only. The solutions may be obtained along the same lines as in the twodimensional case by using generalized spherical coordinates. However, we skip the calculations, which are technically a bit more involved [Rat00b]. The results are as follows: By projecting out d−1 dimensions the probability density for one velocity componet v1 is calculated to ρd (v1 ) = √

Γ( d2 ) πΓ( d−1 2 )

(2E − v12 ) (2E)

d−3 2

,

d−2 2

(10.19)

where Γ(x) stands for the gamma function. This solution was already known to Maxwell and Boltzmann [Max79; Has09], however, in their calculation they took a different starting point; for an alternative derivation see Ref. [Mil98b]. Projecting out d − 2 dimensions yields the probability density for the absolute value of the velocity vector of two components v 1 p and v2 with v = v12 + v22 reading ρd (v) = (d − 2)v

(2E − v 2 ) (2E)

d−4 2

d−2 2

.

(10.20)

These solutions are depicted in Fig. 10.2 for d = 3, 4, 6. As can be seen, for d 1 both types of velocity distributions start to converge towards their canonical counterparts. This can be made precise by taking the limit of d → ∞ in Eqs. (10.19) and (10.20). Using equipartitioning of energy, Pd E = k=1 vk2 /2 = T d/2, where T is the temperature and kB = 1, we arrive at the well-known results

4:1

World Scientific Book - 9in x 6in

klages˙book

177

10.3. Equilibrium velocity distributions for thermostated systems

1 ρd(vx),ρd(vy)

2

ρd(v)

May 8, 2007

0

0

1

0

−1

1 vx,vy

v

Fig. 10.2 Left: equilibrium velocity distribution function ρd (v) for the absolute value v of the velocity vector composed of the two components (vx , vy ), which may be thought of representing a subsystem with two degrees of freedom coupled to a (d−2)-dimensional thermal reservoir. Shown are d = 3 (solid curve), 4 (dashed curve) and 6 (long dashed curve) at a kinetic energy of E = 0.5 as calculated from Eq. (10.20). Right: analogous results for the single velocity components vx , respectively vy , calculated from Eq. (10.19).

lim ρd (v1 ) = √

d→∞

2 1 e−v1 /2T 2πT

,

(10.21)

respectively lim ρd (v) =

d→∞

1 − v2 ve 2T T

.

(10.22)

We have thus rederived the canonical distributions for v1 and for v starting from the microcanonical one of a subsystem plus thermal reservoir by projecting out an infinite number of reservoir degrees of freedom. Altogether the above equations yield the full information of how, according to elementary equilibrium statistical mechanics, the velocity distribution functions of a subsystem consisting of ds degrees of freedom shall look if it experiences ideal interactions with a dr -dimensional thermal reservoir. We remark that the velocity distribution Eq. (10.21) is reproduced by an ensemble of particles obeying the stochastic Langevin equation Eq. (10.1) in the limit of t → ∞. The corresponding Markov process modeling the stochastic behavior of the velocity of a Brownian particle, which is characterized by a stationary Gaussian velocity distribution, is well-known as the Ornstein-Uhlenbeck process [vK92; Wax54]. An interesting question is now to what extent the above ideal functional forms of equilibrium velocity distributions are obtained for deterministic

May 8, 2007

4:1

178

World Scientific Book - 9in x 6in

Motivation: coupling a system to a thermal reservoir

dynamical systems of d degrees of freedom when one looks at a respective smaller number of degrees of freedom ds < d. A very popular example of this type of problem is the one- or two-dimensional harmonic oscillator coupled to a suitable deterministic thermal reservoir. If the reservoir mimicks an infinite number of degrees of freedom d → ∞, the above derivations suggest that the harmonic oscillator velocities should simply approach canonical distributions. However, for such a behavior it is necessary that the thermostated subsystem exhibits at least an ergodic dynamics, because only in this case does a single subsystem trajectory sample the whole phase space appropriately so that it might generate a canonical distribution. If the system is non-ergodic, the resulting velocity distributions will strongly depend on initial conditions and are typically not canonical. This discussion indicates again how sensitively the interplay between subsystem and thermal reservoir may depend on the detailed properties of the involved dynamical systems; for further details see Section 12.4. As we already mentioned, the emphasis of this work is on nonequilibrium situations, where thermal reservoirs are used as a tool to generate NSS. Unfortunately, there is yet no generally accepted analogue of microcanonical or canonical ensembles for nonequilibrium processes. In other words, nothing is known about how velocity distribution functions should generally look in nonequilibrium situations. The standard approach is therefore to define a thermal reservoir in an equilibrium situation by “gauging” it according to the requirement that it generates a microcanonical or a canonical velocity distribution for a respective subsystem connected to it. If this works, a suitable nonequilibrium situation may be created. For an arbitrary deterministic dynamical subsystem it is by no means obvious or guaranteed that a thermal reservoir still “acts properly” under these constraints. This intimately depends on the ergodic and chaotic properties of a subsystem and thermal reservoir. If it does, the reservoir should generate a NSS which is characterized by specific nonequilibrium velocity distribution functions. This way, one can learn something about the variety of different velocity distributions that may exist in different nonequilibrium systems under application of specific models of thermal reservoirs, rather than the other way around. Unless there is a general theory of NSS starting from first principles, which at the moment is not the case, no other approach appears to be feasible here. Of course, for modeling nonequilibrium situations the associated thermal reservoirs must always mimick an infinite number of degrees of freedom in order to prevent a subsystem from heating up.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

10.4. Applying thermostats: the periodic Lorentz gas

klages˙book

179

We finally remark that the above results for equilibrium velocity distributions of a subsystem plus finite-dimensional thermal reservoir will be used in Chapter 15 to systematically construct a specific class of deterministic, time-reversible thermal reservoirs, which dissipate energy according to inelastic collisions at the boundaries of a subsystem. We will furthermore use these results in Chapter 16 by discussing the origin of bimodal velocity distribution functions in models of active Brownian particles.

10.4

Applying thermostats: the periodic Lorentz gas

As we just discussed, whether or not a thermal reservoir is able to generate a specific velocity distribution for moving particles depends to quite some extent on the detailed dynamical properties of the subsystem to which it is applied. As an alternative to the problematic harmonic oscillator one may wish to choose a subsystem that is by default chaotic and ergodic thus providing a more suitable starting point for the action of a thermal reservoir. One may furthermore wish that this system is well understood in the mathematical literature. Such a paradigmatic toy model is the Lorentz gas, which we already discussed in Sections 4.1 and 7.2. Since it is the main model of Part 2 of this book, here we introduce it in full detail. The two-dimensional periodic version of it is sketched again in Fig. 10.3. It consists of hard disks of radius R at number density n whose centers are fixed on a triangular lattice in a plane. Between the scatterers a point particle moves with velocity v and mass m by performing a free flight, whereas at collisions it exhibits specular reflections with the disks. If no further external constraints are applied, the kinetic energy of the moving particle is constant, whereas the direction of the velocity changes according to the collisions with the scatterers. The phase space thus consists of the four variables (x, y, vx , vy ). However, because of energy conservation there are only three independent variables, where one may replace vx and vy by the angle of flight of the particle with the x-axis. For sake of simplicity we set R = m = v = 1. In the original work by H.A. Lorentz [Lor05], to whom the model is commonly attributed, the Lorentz gas consists of randomly distributed hard spheres. It was invented by him in 1905 in order to model the motion of classical electrons in metals, where the electrons do not interact with each other but scatter elastically only with hard spheres mimicking the atoms of the metal. As a laboratory device the two-dimensional periodic version

May 8, 2007

4:1

World Scientific Book - 9in x 6in

180

Motivation: coupling a system to a thermal reservoir

(a)

w

w

(b)

(c)

Fig. 10.3 Sketch of the two-dimensional periodic Lorentz gas, where hard disks of unit radius are situated on a triangular lattice. Part (a) contains the hexagonal fundamental cell of this system. It also shows the trajectory of a point particle moving with a constant velocity and scattering elastically with the disks. Part (b) represents the limiting case of the highest density of scatterers where the minimal distance w between two scatterers is zero. That is, the scatterers touch each other and there is no diffusion. Part (c) depicts the other limiting case at a lower density where, for a specific value of w, for the first time trajectories exist along which a particle can move collision-free for an infinite time and thus sees an “infinite horizon”. In this case the dynamics is ballistic and the diffusion coefficient is infinite.

of this system was, in a way, already introduced about 30 years earlier by Sir Francis Galton in 1877 [Gal77]. Hence, this model is also sometimes referred to as the Galton board. Galton used his device, which he called the quincunx, in lectures for demonstrating the existence of a binomial distribution, which results when particles falling under the influence of a gravitational field move left or right through a periodic board of scatterers. He actually claimed that the same probabilistic laws existed in the form of typical laws of heredity within populations of animate beings [Gal77]. His theme is therefore very different from Lorentz’ intention, who used his model to verify Drude’s theory of classical electronic transport in metals [Dru00] by adding nonequilibrium constraints such as an electric field and

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

10.4. Applying thermostats: the periodic Lorentz gas

klages˙book

181

a gradient of temperatures. Although Drude’s and Lorentz’ work provide some heuristic microscopic derivation of Ohm’s law and of the law by Wiedemann and Frantz, it is well-known that this classical approach is not suitable for yielding explanations for other fundamental properties of metals such as, e.g., their specific heat, which can only be understood quantum mechanically [Iba96]. A soft Lorentz gas, on the other hand, where the scatterers are equipped with smooth potentials, turned out to be very useful for understanding electronic transport in antidot lattices under the action of electric and magnetic fields. This crosslink has already been discussed in Section 7.2 related to fractal transport coefficients. Antidot lattices are low-dimensional semiconductor devices that can be manufactured with scatterers distributed both randomly and periodically in a plane. For small enough field strength the Fermi wavelength is smaller than the lattice constant. Hence, an electron exhibits an essentially classical transport process, where in good approximation the single electrons do not interact with each other; see Refs. [Wei91; Lor91; Wei95; Wei97] for some experimental as well as Refs. [Gei90; Fle92; Wei95; Fli96] for corresponding theoretical work. We thus emphasize that much care has to be taken regarding the physical significance of the Lorentz gas concerning a general understanding of transport in real matter. The usefulness of this model for physically realistic situations should best be assessed from case to case. In the following chapters we focus on the two-dimensional version of the Lorentz gas with circular scattereres on a triangular lattice. So unless stated otherwise, with “Lorentz gas” we denote precisely this configuration of scatterers. The unit cell, which is called the Wigner-Seitz cell in solid state physics, is a hexagon, see Fig. 10.3 (a). Note that alternatively the scatterers may also be put on a square lattice, in which case the unit cell is sometimes called a Sinai billiard [Sin70]. If no further external constraints are added there is only one control parameter, which is the number density of scatterers n or alternatively the smallest distance between two scatterers, the gap size w, as depicted in Fig. 10.3 (b) and (c). In the triangular case three different dynamical regimes are identified: (1) Particles are localized in space, i.e., they cannot leave a unit cell, since all disks touch each other, see Fig. 10.3 (b). This is the high density regime with a minimal gap size w between two adjacent disks of w = 0, and the diffusion coefficient is trivially zero. (2) There exist trajectories along which particles can move ballistically

May 8, 2007

4:1

182

World Scientific Book - 9in x 6in

Motivation: coupling a system to a thermal reservoir

and collision-free for an infinite time, i.e., under certain initial conditions particles see “infinite horizons” in the √ Lorentz gas as indicated by Fig. 10.3 (c). This happens for w > 4/ 3 − 2 ' 0.3094 and marks the onset of a low-density regime in which the diffusion coefficient is infinite. (3) In-between these two regimes the horizon for the moving particles is finite. As has been proven in Refs. [Bun80; Bun81; Bun91; Che99], in this case a central limit theorem for the velocities exists and the diffusion coefficient is finite. This is related to an exponential decay of the velocity autocorrelation function of the moving particle [Mac83; Mat97] from which the existence of a diffusion coefficient follows according to the Taylor-Green-Kubo formula for diffusion Eq. (10.5). On this occasion we may refer to a further simple physical interpretation of the periodic Lorentz gas reflecting these three different dynamical regimes: The scattering of a point particle with a Lorentz gas disk may be interpreted as a two-body problem, where a particle of reduced mass µ moves with a relative velocity v and scatters with a hard core potential V (R). By suitably rescaling the radii of the moving particle and of the fixed scatterer, whose sum should be constant, and by periodically continuing the unit cell including the moving particle, the dynamics may be considered as the collective motion of some lattice modes in a periodic medium such as a crystal. In this case the periodic Lorentz gas was coined the correlated cell model [Ald63; Del96a]. The three different regimes discussed above are characterized by different lattice modes: In case 1 the modes are again localized, in case 2 dislocations are possible, whereas case 3 may be interpreted as some kind of “melting”. The correlated cell model was actually used to prove the existence of a shear and a bulk viscosity in what was called a two-particle fluid [Bun96]. Note also that in kinetic theory the random Lorentz gas is well-known as a model for a gas mixture of heavy and light particles, where the difference of the masses is large enough such that the heavy particles, in good approximation, can be considered to be immobile with respect to the light particles [Cha70]. The periodic Lorentz gas possesses a number of important dynamical properties:4 First of all, it is Hamiltonian and thus deterministic, area-preserving and time-reversible. Secondly, it has been proven that 4 For

readers that are not too familiar with dynamical systems theory this paragraph may be skipped; alternatively, for respective definitions of dynamical systems quantities we refer to Refs. [Sch89; Ott93; Gas98a; Dor99].

klages˙book

May 8, 2007

4:1

10.5.

World Scientific Book - 9in x 6in

∗ Summary

klages˙book

183

the Lorentz gas is a K-system, which implies that it is mixing and ergodic [Bun80; Bun81; Bun91; Che99]. Furthermore, due to the defocusing character of the hard disks5 the Lorentz gas is a hyperbolic [Sin70], chaotic dynamical system. However, because of the hard walls it is not differentiable, as is exemplified by the existence of tangent collisions, and therefore strictly speaking it is not Anosov or Axiom A [Gas96; Gas98a]. Finally, corresponding to the three independent variables of the periodic Lorentz gas there are three different Lyapunov exponents reflecting its hyperbolic behavior: One is zero, which is the one parallel to the flow, the other two are smaller, respectively larger, than zero, and because the system is phase space preserving all Lyapunov exponents trivially sum up to zero. We remark that for random configurations of scatterers in two and three dimensions the Lyapunov exponents have been calculated analytically by means of kinetic theory, partly in comparison with results from computer simulations [vB95; Lat97; vB97; Del97b; vB98; M¨ ul04]. Respective results for higher-dimensional billiards and comparisons with manyparticle systems have been reported in Refs. [dW04b; vB05; dW05]. For the two-dimensional periodic Lorentz gas both on a quadratic and on a triangular lattice computer simulation results for the Lyapunov exponents were presented in Refs. [Gas95b; Gas98a]. As was outlined in Chapters 7 to 9, random and periodic Lorentz gases are prototypical models of Hamiltonian particle billiards, which have been widely studied in the field of chaos and transport both from the physical as well as from the mathematical side of dynamical systems theory, see Refs. [Gas98a; Dor99; Sza00] and further references therein.

10.5

∗

Summary

(1) The physical meaning of a thermal reservoir was discussed on a very elementary heuristic level. For thermal equilibrium, the action of a thermal reservoir on a subsystem was described. By means of a simple example we demonstrated that in nonequilibrium situations thermal reservoirs are usually indispensable for generating nonequilibrium steady states (NSS). (2) As a well-known example for modeling thermal reservoirs we have heuristically introduced the simplest form of a stochastic Langevin 5 This

type of scatterer is said to be convex, or dispersing.

May 8, 2007

4:1

184

World Scientific Book - 9in x 6in

Motivation: coupling a system to a thermal reservoir

equation by summarizing its most important properties. We then showed how this Langevin equation can be derived starting from Hamiltonian equations of motion, where the heat bath is modeled by a collection of non-interacting harmonic oscillators. On this basis we discussed deficiencies of both schemes as far as the modeling of thermal reservoirs is concerned. We also described some more recent approaches trying to go beyond this simple modeling. (3) On a more statistical level, we have outlined how to calculate the equilibrium velocity distribution functions for a single velocity component as well as for the absolute value of two velocity components for a subsystem that is coupled to a dr -dimensional thermal reservoir. In the limit of dr → ∞ the canonical distribution is recovered. We indicated that these “ideal” equilibrium velocity distribution functions may not by default be properly reproduced by an arbitrary deterministic subsystem such as the harmonic oscillator coupled to some model of a thermal reservoir. Related problems for modeling NSS were briefly discussed. (4) We defined in detail the periodic Lorentz gas, which exhibits convenient dynamical properties in an equilibrium situation by being, among others, Hamiltonian, mixing and fully chaotic. We started with a brief history of the (periodic) Lorentz gas, discussed its physical significance and some possible interpretations as far as physical reality is concerned. We classified its three different dynamical regimes and very briefly summarized its most important mathematical properties.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Chapter 11 ∗

The Gaussian thermostat

We are now prepared to introduce a very popular deterministic and timereversible modeling of a thermal reservoir, which became known as the Gaussian thermostat. In Section 11.1 we define this thermostating scheme for the periodic Lorentz gas driven by an external electric field. In Section 11.2 we summarize what we consider to be the most important features of the resulting model from a dynamical systems point of view. The properties we report are to a large extent typical for Gaussian thermostated systems altogether. Specifically, we review a connection between thermodynamic entropy production and the average phase space contraction rate as well as a simple functional relationship between Lyapunov exponents and transport coefficients. Furthermore, there exists a fractal attractor in the Gaussian thermostated Lorentz gas, which changes its topology under variation of the electric field strength. Correspondingly, the electrical conductivity is an irregular function of the field strength as a control parameter, which takes up the thread on fractal transport coefficients of Part 1 in this book. We also briefly elaborate on the existence of linear response in this model. 11.1

Construction of the Gaussian thermostat

In the following chapters the periodic Lorentz gas serves as a standard model to which different thermal reservoirs are applied. As discussed in the previous section, a big advantage of the Lorentz gas, in comparison to the harmonic oscillator, say, is that it exhibits deterministic chaos , which is a consequence of the defocusing geometry of the scatterers. Hence, in contrast to the Langevin dynamics of Eq. (10.1) we do not need to impose a stochastic force onto this model in order to enforce Brownian motion185

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

∗ The

186

Gaussian thermostat

like spatial fluctuations for a moving particle.1 Instead, some randomlooking trajectories in position space are generated due to the intrinsic Lyapunov instability of the system, see Fig. 10.3. This is in turn intimately related to the mixing and ergodic properties of the Lorentz gas. We thus replace stochasticity by deterministic chaos in order to generate a Brownian motion-like dynamics, cf. also our discussion in Section 10.2. As we know from the previous section and from Section 7.2, in a certain regime of densities of scatterers the periodic Lorentz gas exhibits normal diffusion. In the following we keep this density fixed by choosing for the minimal distance between two scatterers w ' 0.2361, as it is standard in the literature to ensure that a diffusion coefficient exists [Mor87a; Llo94; Llo95; Del95a; Mor96a; Det96a]. In order to drive the system out of equilibrium we apply an external electric field that, if not said otherwise, is parallel to the x-axis. However, as explained before, if there were no interaction with a thermal reservoir, any moving particle would on average be accelerated by the external field leading to an ongoing increase of energy in the system, and there would be no NSS. Therefore the model must be connected to some thermal reservoir, i.e., it must be thermostated. The periodic Lorentz gas amended by an external field and coupled to a thermal reservoir is called the driven periodic Lorentz gas. This model has been widely studied in the literature over the past fifteen years [Mor87a; Hoo88a; Hoo89b; Hoo92; Van92; Che93a; Che93b; Che01; Bar93; Lue93; Llo94; Llo95; Det95; Del95a; Mor96a; Det96a; Det97c; Det97d; Bon97; Coh98; Mor98; Hoo98d; Hoo99; Dor99; T´el00; Bon00a; Det00a; Mej01; Bon02; Vol02; Lar03 ]. Following our discussion of the Langevin equation one may remove energy from the field-driven Lorentz gas according to r˙ = v v˙ = ε − α(v)v

,

(11.1)

supplemented by the geometric boundary conditions imposed by the Lorentz gas scatterers. The choice α = const. yields ordinary Stokes friction, and the equations represent a deterministic variant of the stochastic Langevin equation Eq. (10.1) amended by an external electric field ε. Here and in the following the electric charge q of the particle is set equal to one. This viscous Lorentz gas has been studied by computer simulations in Ref. [Hoo92]. 1 On the other hand, without applying an external field the absolute value of the velocity for the moving particle is yet constant in the Lorentz gas, which is at variance to ordinary Langevin dynamics. However, see Chapter 12 for respective further modifications of the Lorentz gas.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

187

11.1. Construction of the Gaussian thermostat

For numerical solutions one has to discretize Eqs. (11.1) in time. The second of the above two equations then reads v(t + ∆t) = ε∆t + α ˜ v(t)

,

0 < ∆t 1 ,

(11.2)

with α ˜ := 1 − α∆t. For small enough ∆t there is 0 < α ˜ < 1, and by keeping the time interval fixed the viscous force amounts to a rescaling of the velocity periodically in time with a constant scaling factor α. ˜ This procedure can be made more efficient by replacing α ˜ according to s E , (11.3) α(v) ˜ := E(t) where E is the target kinetic energy of the system and E(t) the measured kinetic energy at time t. The scaling factor now depends on the velocity of the moving particle and may fluctuate in time. This elementary but rather convenient numerical method to keep the energy fixed became wellknown for molecular dynamics computer simulations as velocity rescaling [Eva83a; All87; Hol87; Hol95; Hes96a; Hes96b; Hoo97; Det00a ]. Note that the Lorentz gas with a Stokes friction coefficient and with rescaled velocities is both deterministic but not time-reversible. In order to further improve the efficiency of these thermostating mechanisms one may start directly from a velocity-dependent friction coefficient in Eq. (11.1). Requiring energy conservation at any time step implies dv2 /dt = 0 as a constraint on the functional form of α(v) and yields α(v) = ε · v/v 2

.

(11.4)

This method of thermostating was proposed simultaneously and independently by Hoover and coworkers [Hoo82] and by Evans [Eva83a] in 1982. Performed in the limit of ∆t → 0, the velocity rescaling Eqs. (11.2), (11.3) is identical to the combination Eqs. (11.1), (11.4) [Eva83a; Hol87; Hol95; Hes96a; Hoo97; Det00a]. In contrast to ordinary Stokes friction, which only reduces the kinetic energy, the velocity-dependent version Eq. (11.4) pumps energy into the system whenever the particle is moving opposite to the field. The friction is maximal when the velocity is parallel to the field, and it is zero when velocity and electric field are perpendicular to each other. Trivially, it is zero when the electric field is zero due to the fact that in the equilibrium periodic Lorentz gas a particle is moving with a constant absolute value of the velocity anyway. Still, in comparison to the friction of the deterministic, generalized Langevin equation Eq. (10.11), which was obtained starting

May 8, 2007

4:1

188

World Scientific Book - 9in x 6in

klages˙book

∗ The

Gaussian thermostat

from the heat bath of harmonic oscillators, such a generalization of ordinary Stokes friction may not be considered physically unreasonable. Alternatively, Eq. (11.4) can be derived from Gauss’ principle of least constraints [Eva83b; Eva90b; Mor98; Ron02a], which is a fundamental principle of classical mechanics.2 For this reason the velocity-dependent friction force Eq. (11.4) which keeps the energy of a particle constant at any time step was coined the Gaussian thermostat [Eva83b]. Correspondingly, the Gaussian thermostat applied to the driven periodic Lorentz gas generates a microcanonical-like distribution in velocity space [Mor98].3 In fact, here we are constraining only the kinetic energy of the moving particle, therefore this version is sometimes further classified as the Gaussian isokinetic thermostat. For soft particles one may fix alternatively the total internal energy yielding the Gaussian isoenergetic thermostat [Eva90b; Mor98; Det00a; Ron02a]. Suitably amended versions of the Gaussian thermostat can also be applied, for example, to many-particle systems under shear, see Chapter 15 for further discussions. Similarly to the case of Stoke’s friction, which models the loss of energy of a moving particle into a surrounding viscous fluid, one may associate with the Gaussian friction variable Eq. (11.4) some imaginary thermal reservoir that acts at any instant of time. That way the Gaussian thermostat is another example of a bulk thermostat. We remark that this physical picture contradicts the specific interpretation of the driven periodic Lorentz gas as a model for noninteracting electrons moving in a periodic crystal. Here one may expect that energy is only dissipated at the collisions with a scatterer and not during a free flight. However, the action of the Gaussian thermostat is conceptually fine if the Lorentz gas is considered as a two-particle fluid, which is immersed into another fluid serving as a thermal reservoir. Let us furthermore remark that in Chapter 15 we will introduce boundary thermostats for the Lorentz gas, where a particle exchanges energy with a thermal reservoir only at the collisions with a scatterer. On that occasion we will discuss similarities and differences to what we find for the Gaussian thermostated case summarized below. Surprisingly, the equations of motion Eqs. (11.1) supplemented by Eq. (11.4) are time-reversible modeling the, at first view, contradictory situation of a deterministic, time-reversible, dissipative dynamical system 2 In contrast to d’Alembert’s principle, for the Gaussian version the virtual displacements are acceleration terms instead of the positions of a moving particle. 3 Note that in nonequilibrium the distribution of points on the energy shell is not uniform anymore reflecting the existence of an average current.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

189

11.2. Chaos and transport in Gaussian thermostated systems

[Hoo96b]. As will be outlined in the following, under certain conditions the Gaussian thermostated driven Lorentz gas furthermore exhibits an ergodic and chaotic dynamics in well-defined NSS. In contrast to standard Stoke’s fricition and to stochastic thermostats the Gaussian scheme thus enables one to elaborate on the fundamental problem of how time-reversible, deterministic, microscopic equations of motion may generate macroscopic irreversible transport in nonequilibrium situations associated with energy dissipation. 11.2

Chaos and transport in Gaussian thermostated systems

We now analyse the Gaussian thermostated driven periodic Lorentz gas from a dynamical systems point of view and summarize what we consider to be characteristic chaos and transport properties of this class of dynamical systems. 11.2.1

Phase space contraction and entropy production

A first fundamental property of Gaussian thermostated systems is obtained from computing their average phase space contraction rate defined by κp :=< ∇ · F(r, v) > ,

(11.5)

where F(r, v) stands for the equations of motion of the system Eqs. (11.1) and the brackets denote an ensemble average. A straightforward calculation, by taking into account that α is a function of v, yields κp = − < α >

(11.6)

with α defined by Eq. (11.4), where κp < 0. Let us now rewrite this equation by introducing a nonequilibrium temperature T . If we want to do so by using equipartitioning of energy [Eva90b; Mor98; Hoo99] we need to know about the number of degrees of freedom of the system. The velocity space is two-dimensional, but one may argue that the energy constraint eliminates one degree of freedom of the system. However, this argument is debatable, and actually there is quite an ambiguity in defining a temperature for the driven Lorentz gas [Che93a; Che93b].4 By assuming that there is one degree 4 If

we switch back to the Lorentz gas without an external field we also, trivially, have energy conservation. However, here nobody would claim that the number of degrees of freedom is reduced respectively.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

∗ The

190

Gaussian thermostat

of freedom only we arrive at v2 = T

,

(11.7)

with kB = 1. We can now write −κp = ε· < v > /T

,

(11.8)

where the right hand side is the familiar expression related to the Joule heating derived from irreversible thermodynamics. Alternatively, one could have started from Clausius’ definition of entropy production in terms R of the heat transfer dQ between subsystem and thermal reservoir, ∆S = dQ/T [dG62; Hua87; Det00a; Gal03b], yielding the same expression.5 We thus arrive at the fundamental result that, for Gaussian thermostated systems, there is an identity between the absolute value of the average rate of phase space contraction and the rate of thermodynamic entropy production [Hol87; Pos87; Pos88; Eva90b; Che95b; Che97; Che93a; Che93b; Vol97; Bre98; Gil99a; Hoo99; Det00a; Vol04]. A third way to obtain this identity starts from Gibbs’ definition of entropy production, as will be discussed in Chapter 13. 11.2.2

Lyapunov exponents and transport coefficients

Eq. (11.8) provides a crucial link between dynamical systems properties, in terms of phase space contraction, and thermodynamics, in terms of entropy production. By employing this link a second fundamental property for Gaussian thermostated systems is easily derived as follows: First of all, there is an identity between the average phase space contraction of a dynamical system and the sum of Lyapunov exponents [Eck85; Gas98a]. For the driven Lorentz gas under consideration two Lyapunov exponents are zero, due to the energy constraint and assessing the neutral direction parallel to the trajectory of a moving particle. From the remaining two exponents one is positive and one is negative, at least for small enough field strength [Llo95; Del95b; Det95; Det96a; Mor96a; Mor98]. Consequently we have κp = λ + + λ − 5 In

.

(11.9)

the d-dimensional case and for N particles there is a prefactor of N d − 1 on the right hand side of Eq. (11.6), but this factor cancels with a respective prefactor on the right hand side of Eq. (11.7) such that the identity is preserved [Coh98]. Note also that the Clausius form is only valid for quasistatic processes in, at least, local thermodynamic equilibrium.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

11.2. Chaos and transport in Gaussian thermostated systems

191

Let us now replace the average current by introducing the field-dependent electrical conductivity σ(ε) according to6 < v >=: σ(ε)ε

,

(11.10)

where ε stands for the absolute value of the field strength. For sake of simplicity here we consider only the case of the field being parallel to the x-axis. Because of the symmetry of the two-dimensional model σ(ε) is then a scalar quantity instead of a tensor [Llo95]. Replacing the average current < v > on the right hand side of Eq. (11.8) by Eq. (11.10) and the average phase space contraction rate κp on the other side by Eq. (11.9) we arrive at a simple relation between the conductivity σ(ε) and the sum of Lyapunov exponents of the dynamical system, σ(ε) = −

T (λ+ (ε) + λ− (ε)) ε2

.

(11.11)

This important connection between transport coefficients and dynamical systems quantities was first reported by Posch and Hoover for a manyparticle system with an external field [Pos88] and was called the Lyapunov sum rule later on [Eva90a]. For the driven periodic Lorentz gas its existence was rigorously proven in Refs. [Che93a; Che93b]. Eq. (11.11) can also be related to periodic orbits [Van92] and was studied in much detail numerically [Bar93; Llo95; Del95b]. Functional forms that are completely analogous to Eq. (11.11) also hold for many-particle systems and for other transport coefficients such as viscosity [Eva90a; Eva90b; Coh95; Sar98] and thermal conductivity [Aok03] under the necessary condition that heat is dissipated by Gaussian and related thermostats [Mor98]. The Lyapunov sum rule moved particularly into the center of interest starting from the work by Evans, Cohen and Morriss [Eva90a], who introduced a further simplification of it for systems with many degrees of freedom. They observed that for dissipative many-particle systems the spectrum of Lyapunov exponents may exhibit a remarkable symmetry property [Pos88; Mor89a; Eva90a]. This was first reported for the case of a uniform viscous damping by Dressler [Dre88] and was later on called the conjugate pairing rule: If all non-zero Lyapunov exponents are ordered according to their values they can be grouped into pairs such that each pair sums up to precisely the same value [Dre88; Pos88; Eva90a; Sar92; Gup94; 6 Strictly speaking this quantity may rather be called the mobility of a particle, in distinction from the conductivity that involves the charge and the density per unit volume of the mobile electrons of a conductor [Wan66; Rei65]. However, in our case there is no difference.

May 8, 2007

4:1

192

World Scientific Book - 9in x 6in

klages˙book

∗ The

Gaussian thermostat

Coh95; Det95; Del96b; Det96c; Lat97; Del97b; Del97a; Det97b; Mor98; Sea98; Bon98b; vB98; Woj98; Rue99b; vB00b; Dol00; Pan02c; Pan02b; Pan02a; Mor02; Tan02b]. This pairing rule is well-known for Hamiltonian dynamical systems, however, here the Lyapunov exponents appear in pairs that always sum up to zero. Indeed, in phase space preserving dynamical systems it can be shown that conjugate pairing is a direct consequence of the symplecticity of the Hamiltonian [Eck85; Mei92; Abr94; Gas98a]. If conjugate pairing holds for the very different class of dissipative, thermostated systems it enables one to replace the sum over the full spectrum of Lyapunov exponents, which appears on the right hand side of Eq. (11.11), by the sum over one pair of exponents only. For higherdimensional dynamical systems this drastically simplifies the computation of transport coefficients according to Eq. (11.11), since the total number of different exponents is here extremely large and the full spectrum is very hard to extract numerically. Instead, for a conjugate pair one may just choose the largest and the smallest Lyapunov exponent, where the largest exponent is typically not too difficult to get [Ben80b; Wol85; Ott93]. The smallest one, in turn, is best computed by reversing the direction of time [Hoo88b; Eva90a; Coh95; vB96; Lat97; Sar98; Sea98; Tan02b]. However, note that the reversed trajectories still have to be constrained to the invariant set associated with the forward dynamics. As we will further explain in the following section, in dissipative systems this set is usually a fractal attractor that becomes a fractal repeller under time reversal, and it requires some additional numerical efforts to keep the trajectories in this now unstable region. From a computational point of view the Lyapunov sum rule Eq. (11.11) supplemented by conjugate pairing thus does not appear to provide a more efficient mean for computing transport coefficients than other existing standard procedures. For example, along the lines of Eq. (11.11) it would be easier to compute the average value of the thermostating variable instead of any Lyapunov exponent, see Eqs. (11.6) and (11.9). Conjugate pairing was analytically proven to hold under conditions that concern the form of the interparticle potential and the type of deterministic thermostat used [Det96c; Det97b; Mor98; Woj98; Rue99b; Pan02a]; see also the kinetic theory calculations for the three-dimensional Gaussian isokinetically thermostated driven random Lorentz gas in Refs. [Lat97; vB98; vB00b] and a respective comparison to computer simulations [Del97b].

May 8, 2007

4:1

World Scientific Book - 9in x 6in

11.2. Chaos and transport in Gaussian thermostated systems

klages˙book

193

Conjugate pairing in the three-dimensional driven periodic Lorentz gas, again connected to a Gaussian isokinetic thermostat, was numerically corroborated in Refs. [Det95; Mor96a]. For interacting many-particle systems under shear the situation concerning the validity of conjugate pairing appears to be considerably less clear, see Refs. [Eva90a; Coh95; Gup94; Mor98] and more recent discussions [Mor02; Tan02b; Pan02b; Pan02c]; see also Refs. [Del96b; Sar92; Sea98] for numerical studies of various thermostated systems in which conjugate pairing appears to hold, or respectively does not appear to hold. Generally it seems to be easier to give sufficient conditions under which conjugate pairing does not hold: one example is a three-dimensional periodic Lorentz gas under electric and magnetic fields for which there is, or is not, conjugate pairing depending on the specific direction of the two fields [Dol00]. Another generic class of counterexamples to conjugate pairing is provided by systems that are thermostated at the boundaries [Pos88; Del97a; Wag00]. The latter models indicate that conjugate pairing is intimately related to the thermostat acting homogeneously and ‘democratically’ on all particles in the bulk of a system [Mor98; Eva00], whereas inhomogeneous thermostating at the boundaries prevents the Lyapunov exponents from coming in conjugate pairs. In Chapter 15 we will more explicitly discuss counterexamples of the latter type. 11.2.3

Nonequilibrium fractal attractors

How does it fit together that time-reversible microscopic equations of motion yield irreversible macroscopic transport in a NSS? As was first discussed by Hoover et al. [Hol87; Mor87a] and by Morriss [Mor87b], for this type of driven thermostated system, reversibility and irreversibility are intimately linked to each other by the existence of a fractal attractor in phase space; see Refs. [Mor87b; Mor89b; Eva90b; Pet94] for a sheared two-particle system, Refs. [Pos87; Pos88; Mor89a; Pos89; Hoo94; Del97a; Wag00] for many-particle systems under various nonequilibrium conditions, Refs. [Mor87a; Hoo88a; Hoo89b; Van92; Hoo92; Che93a; Che93b; Che01; Lue93; Llo94; Llo95; Del95a; Det96a; Det97d; Mor98; Hoo98d; Kla00b; Rat00b; Rat00a; Rat02] for the driven periodic Lorentz gas and Refs. [Hol87; Kus90; Hoo91; T´el96; Pos97; Pos98; Gil99a; Gil99b; Hoo99; T´el00; Vol02] for related models.7 7 Here we use the term attractor according to the operational definition of Eckmann and Ruelle [Eck85]: Let there be points in phase space generated by the equations of motion

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

∗ The

194

Gaussian thermostat

γ β

Fig. 11.1 Phase space coordinates describing the collision of a point particle at a disk for a hexagonal unit cell of the two-dimensional periodic Lorentz gas. β denotes the position of the colliding particle on the circumference of the disk, whereas the impact angle γ is related to the direction of the velocity of the moving particle before the collision.

We now outline characteristic features for the attractor of the driven periodic Lorentz gas. For convenience the temperature of the thermostat is fixed at T = 1. In order to simplify the analysis a Poincar´e surface of section is considered. That is, we introduce a further constraint in position space by representing the dynamics in terms of the phase space coordinates at a collision with a disk only. These collisions are defined by the position on the circumference of the disk according to the angle β and by the impact angle γ related to the direction of the velocity of the moving particle, see Fig. 11.1. The set (β, sin γ) forms the Birkhoff coordinates of a particle billiard [Gas96; Gas98a], see also Section 4.1. A convenient property of the Birkhoff coordinates is that in the Hamiltonian case at field strength ε = 0 the invariant probability density ρ(β, sin γ) associated with these coordinates is simply uniform in the whole accessible phase space, which is related to Lambert’s cosine law [Car85]. In contrast to that, Fig. 11.2 shows that in the case of dissipation generated by an electric field and counterbalanced by a Gaussian thermostat the phase space density contracts onto a complicated fractal-looking object. Numerical analysis provides evidence that the attractor shown in Fig. 11.2 is multifractal [Ott93; Bec93]: Although the two-dimensional phase space of a dynamical system. The attractor is the set on which these points accumulate for large times. For more rigorous mathematical definitions see, e.g., Refs. [Eck85; Gas98a; Guc90; Dev89]. Sometimes the additional term strange is employed to further characterize attractors [Rue71]. However, its use appears to be ambiguous in the literature, partly indicating the existence of a chaotic dynamics on the attracting set [Bec93; Sch89; Eck85], partly being synonymous for the attractor exhibiting a fractal structure [Ott93; Gre84]. Hence, here we avoid this denotation.

4:1

World Scientific Book - 9in x 6in

11.2. Chaos and transport in Gaussian thermostated systems

klages˙book

195

sin γ

May 8, 2007

β Fig. 11.2 Attractor of the Gaussian thermostated periodic Lorentz gas driven by an electric field of strength ε = 1.5 which, here, is parallel to the y-axis. The temperature is T = 1, β and sin γ are defined in Fig. 11.1. The figure, originally being a 1msquare, is from Ref. [Hoo89b]. For related results see Refs. [Mor87a; Hoo88a; Del95a; Hoo99].

spanned by the Birkhoff coordinates is fully covered for small enough field strength yielding a box-counting dimension of two in computer simulations, the higher-order Renyi dimensions such as the information dimension and the correlation dimension are non-integer and slightly smaller than two indicating an inhomogeneous fractal folding of the phase space [Hoo89b; Det96a; Det00a]. However, note that here we are discussing the attractor in the reduced phase space defined by the Poincar´e surface of section only. As far as the whole accessible phase space corresponding to the three independent phase space variables of the model is concerned there is a proof that the Hausdorff dimension of this fractal set is less than three and identical to the information dimension for non-zero but small enough field strength [Che93a; Che93b], which appears to be consistent with computer simulation results [Del95a]. The basic structure of the attractor shown in Fig. 11.2 can roughly

May 8, 2007

4:1

196

World Scientific Book - 9in x 6in

klages˙book

∗ The

Gaussian thermostat

be understood as follows: points in phase space with an impact angle of γ = ±π/2 signify tangent collisions, where a moving particle just passes a disk. These points are due to the hard disk geometry and yield discontinuities in the equations of motion, which separate regions in phase space corresponding to different types of trajectories.8 Calculating higher iterates of these lines of discontinuity from the equations of motion yields complicated boundaries in phase space, which match to the structures depicted in Fig. 11.2, see Refs. [Gas96; Gas98a] for the field-free Lorentz gas and Refs. [Mor87a; Hoo88a; Bon00a] for the driven case. We remark that this procedure is intimately related to the construction of Markov partitions for the periodic Lorentz gas [Bun80; Bun81; Bun91; Che99]. Note that the topology of these structures changes under parameter variation, which has direct consequences for physical quantities such as the electrical conductivity of the system, as we will discuss in the following section. The bifurcation diagram Fig. 11.3 indicates how the attractor changes by increasing the field strength ε [Mor87a; Hoo89b; Llo94; Llo95; Det96a; Mor98]. Here we show results for the angle θ, which is the angle between the horizontal x-axis and the velocity of the moving particle after the collision. Consequently, θ is a simple function of the coordinate γ. Results for β and γ are analogous. This figure reveals a complicated bifurcation scenario: By increasing the field strength the attractor first experiences a crisis [Ott93] at which it suddenly collapses onto a subset in phase space [Llo94; Llo95; Det96a; Mor98]. For higher values of the field strength there are Feigenbaum-like bifurcations showing a complicated interplay between periodic windows and chaotic parameter regions. Around ε = 2.4 there is even an example of an elliptic fixed point in phase space. For small enough field strength, that is, before the first crisis occurs, the system is ergodic, see Refs. [Mor87a; Hoo89b; Llo94; Llo95; Det96a; Mor98] for numerical explorations and Refs. [Che93a; Che93b; Che01] for a proof. In the course of the bifurcations the system exhibits a complex transition scenario from ergodic and chaotic to non-ergodic and non-chaotic behavior. For high enough field strength the attractor eventually collapses onto a single periodic orbit. This orbit is such that a particle moves parallel to the field during a free flight, whereas at a collision it “creeps” along the 8 By different types of trajectories we mean that trajectories with initial conditions in different regions hit different scatterers in the spatially extended periodic Lorentz gas, as can be quantitatively assessed in terms of a symbolic dynamics by assigning different symbols to different scatterers [Llo94; Llo95; Gas96; Gas98a; Mor98].

4:1

World Scientific Book - 9in x 6in

11.2. Chaos and transport in Gaussian thermostated systems

klages˙book

197

θ

May 8, 2007

ε Fig. 11.3 Bifurcation diagram for the Gaussian thermostated periodic Lorentz gas driven by an electric field of strength ε which, here, is again parallel to the x-axis. θ is the angle between the x-axis and the velocity of the moving particle after a collision and is a simple function of γ as defined in Fig. 11.1. The temperature is set to T = 1. The figure is from Refs. [Llo94; Llo95; Mor98].

circumference of a disk [Mor87a; Llo94; Llo95]. For Gaussian-type thermostated dynamical systems the existence of multifractal attractors in Poincar´e sections appears to be typical. This applies not only to a single-particle dynamics but also to interacting manyparticle systems under general nonequilibrium conditions, see the long list of references cited at the beginning of this section for some examples. The fractality also survives in chaotic dynamical systems where the single particles interact with each other via smooth potentials [Pos87; Pos88; Pos89; Mor89a; Pos97; Hoo97] and in systems that are thermostated by a Stokes friction coefficient [Hoo92] or by some constant restitution coefficient at the collisions with a scatterer [Lue93]. Important questions are whether the existence of fractal attractors is a universal property of NSS in chaotic dynamical systems connected to thermal reservoirs, irrespective of the specific thermostat used, and to what extent the structure of the attractor depends on the specific type of thermostat.

4:1

World Scientific Book - 9in x 6in

klages˙book

∗ The

198

Gaussian thermostat

σ (ε)

May 8, 2007

ε Fig. 11.4 Electrical conductivity σ(ε) for the Gaussian thermostated periodic Lorentz gas driven by an electric field of strength ε, which is parallel to the x-axis. The temperature is again T = 1. The small diamonds are data points computed from higher precision simulations, see Ref. [Llo95] for further details, which also contains the original figure. Note the nonlinear response of this quantity and the irregular structure on fine scales.

11.2.4

Electrical conductivity

In the final part of this chapter we focus on the transport properties of the Gaussian thermostated driven periodic Lorentz gas by studying the electrical conductivity σ(ε) introduced in Eq. (11.10). Here we only discuss the ergodic regime of the model at field strengths ε 2 for which σ(ε) is uniquely defined. In this case it can be obtained from computer simulations in the form of the time average over the velocity of a single particle sampled from a long trajectory. If the system is non-ergodic the situation is more complicated, since the values for the conductivity then depend on the choice of initial conditions. Results for σ(ε) in the ergodic regime calculated from computer simulations [Llo95] are shown in Fig. 11.4. Related results have been obtained, also from simulations, in Refs. [Mor87a; Del95a; Bon00a; Det00a]; for calculations based on periodic orbits see Refs. [Van92; Det97c], for the conductivity of a driven Lorentz gas thermostated by a Stokes friction coefficient see Ref. [Hoo92], and for conduction in a modified many-particle Lorentz gas see Refs. [Bar93; Bon97; Bon02].

May 8, 2007

4:1

World Scientific Book - 9in x 6in

11.2. Chaos and transport in Gaussian thermostated systems

klages˙book

199

Two aspects are remarkable in this figure: First of all, the validity of Ohm’s law implies that σ(ε) = const + O(ε)

,

(11.12)

where O(ε) are higher-order corrections. However, here σ(ε) clearly decreases on average indicating that for the range of field strengths shown the response is already in the nonlinear regime. Indeed, even for the smallest values of the field strength that are accessible to simulations no linear response behavior is visible. These results [Llo95] are in agreement with simulations performed by other authors [Mor87a; Del95a; Bon00a; Det00a]. On the other hand, in the limit of ε → 0 the conductivity appears to converge to the correct value of the diffusion coefficient obtained from the Einstein formula9 D/T = σ(ε)

(ε → 0) ,

(11.13)

which is D ' 0.196 for T = 1 [Mac83; Bar93; Gas95b; Mor94; Del95a; Bon00a; Kla00a]. This numerically corroborates the analytical proof of the Einstein relation presented in Refs. [Che93a; Che93b] and is consistent with the special case of Ohm’s law d < v > . (11.14) σ= dε ε=0

For sake of completeness let us remark that the conductivity also depends on the density of scatterers. As expected, for higher densities the response appears to become more linear in the regime of field strengths shown in Fig. 11.4 [Del95a]. However, to decide about linearity or nonlinearity is still very ambiguous, since this depends on the scale according to which the results are plotted; see also Fig. 4 in Ref. [Det00a]. A second fact makes the situation even more complicated, namely the existence of irregularities in the conductivity as a function of the field strength in the form of non-monotonocities on fine scales, as was first reported in Ref. [Mor87a]. The existence of such irregularities has been confirmed by cycle expansion calculations for higher values of the field strength [Det97c] and by other computer simulations [Llo95; Del95a; 9 This

equation is easily obtained from the Einstein relation Eq. (10.3) if an external electric field ε is added to the (one-dimensional) Langevin equation Eq. (10.1). In a NSS the average over the time derivative of the velocity on the left hand side of Eq. (10.1) must be zero and the average over the noise on the right hand side is by definition also zero, hence ε − α < v >= 0. By using Eq. (11.10) one arrives at σ = 1/α [Rei65; Pat88]. Note that for the Gaussian thermostated driven Lorentz gas Eqs. (11.1), (11.4) the validity of these relations is not obvious.

May 8, 2007

4:1

200

World Scientific Book - 9in x 6in

klages˙book

∗ The

Gaussian thermostat

Bon00a], consequently they are no artifacts representing numerical imprecisions. This finding demonstrates that irregular, respectively fractal transport coefficients as discussed inf Part 1 also play a role in thermostated systems. Indeed, by using Birkhoff coordinates the periodic Lorentz gas can be linked to maps having the same properties as the ones exhibiting fractal transport coefficients discussed in Part 1 [Gas96; Gas98a; Gil99b; Gas98c; T´el00; Vol02]. It was thus early conjectured that the field-dependent conductivity of the driven Lorentz gas also exhibits fractal properties [Kla95; Kla96b; Kla99a]. For a one-dimensional map where a broken symmetry generates a drift-dependent current it was shown in Chapter 3 that the conductivity is typically10 a nonlinear fractal function of control parameters [Gro02; Gil04]. Similar results were obtained numerically for the parameterdependent diffusion coefficient in the field-free periodic Lorentz gas, see Chapter 7, and in related billiards of Lorentz-gas type, see Chapter 8. However, to rigorously prove the existence of irregular structures in transport coefficients is generally a hard task, even for the most simple systems [Gro02], as is exemplified by a recent attempt of a proof for the driven Lorentz gas [Bon00a]. Apart from these numerical assessments of the Lorentz gas conductivity there are seminal mathematical results proving Ohm’s law in the form of Eq. (11.12) for this model [Che93a; Che93b]. However, this proof does not yield any precise upper bound for the validity of the range of linear response and may thus rather be considered as an existence proof. Indeed, simple heuristic arguments quantifying such bounds [Dor99; Che93b] lead to wrong values for the driven periodic Lorentz gas in comparison to simulation results [Rat00b]. According to these mathematical and numerical analyses one may conclude that a regime of linear response exists in the driven periodic Lorentz gas which, however, is so small in the field strength that, up to now, it could not be properly detected in computer simulations. A related open problem is the existence of fractal-like irregularities on fine scales, which, as discussed for a simple map in Section 3.3, for the driven Lorentz gas as well one may expect to persist in the limit of ε → 0. It would thus be important to estimate the magnitude of these irregularities for small enough field strength in order to learn to which extent they are negligible or dominate any linear response behavior. These problems strongly remind one of the famous van Kampen ob10 i.e.,

for all control parameters except of Lebesgue measure zero

May 8, 2007

4:1

World Scientific Book - 9in x 6in

11.2. Chaos and transport in Gaussian thermostated systems

klages˙book

201

jections against linear response theory [vK71]. It is not our goal to give a full account of the debate around van Kampen’s arguments. For that we may refer, e.g., to Refs. [Mor89c; Eva90b; Kub92; Che93b; Fal95], to the outline in Ref. [Dor99] and to further references therein. Instead, we will highlight certain points that we believe are important for an understanding of the interplay between chaos and transport in the driven periodic Lorentz gas. Van Kampen criticizes derivations of the laws of linear response that assume microscopic linearity in the equations of motion. By taking this assumption seriously he arrives at a simple estimate for the range of validity of linear response which is 10−18 V /cm for electrons in a conductor.11 Van Kampen uses this unrealistically small value to illustrate his point that linearity of the microscopic motion is entirely different from macroscopic linearity. In other words, the question is how to reconcile the ubiquitous microscopic nonlinearity in the equations of motion of a dynamical system with the macroscopic linearity in the response of the same system. According to van Kampen it is the randomization of the microscopic variables that enables one to eliminate them from the macroscopic picture, whereas in linear response theory the previous history is not forgotten through randomization, but ignored through linearization. Curiously enough, van Kampen discusses just the Galton board under external forces as an example. By formally solving the Boltzmann equation, which is based on Boltzmann’s molecular chaos assumption, he concludes that the nonlinear deviations of the microscopic motions somehow combine to produce a linear macroscopic response; see also Refs. [Che93b; Coh98; Kla00a] for discussions of that point. To us, the existence proof of linear response for the driven periodic Lorentz gas given in Refs. [Che93a; Che93b] exemplifies how to reconcile macroscopic linearity with microscopic nonlinearity. On the other hand, results from computer simulations and related work [Mor87a; Llo95; Del95a; Det97c; Bon00a] still point back to van Kampen’s first objection of how to quantitatively identify the regime of validity of linear response. Thus, the driven periodic Lorentz gas appears to provide a good illustration for the difficulty formulated by van Kampen that both the existence and the size of a regime of linear response may not be trivially guaranteed in a 11 We remark that in Ref. [Bon00a] a number for the range of validity has been given for the driven periodic Lorentz gas, “possibly as small as of 10−20 ”, whose order of magnitude is close to the original estimate by van Kampen [vK71], however, without indicating how this value was calculated for this specific system.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

∗ The

202

Gaussian thermostat

nonlinear dynamical system; see also Chapter 3 for another example which is even more drastic but also more abstract than the Lorentz gas. A proper theoretical approach thus needs to take into account that, for microscopic chaotic dynamics, the macroscopic response of the system may very sensitively depend on the special character of the system. Let us emphasize that the periodic Lorentz gas exhibits specific properties such as being low-dimensional and spatially periodic, which may not be considered to be typical for gases or fluids that are composed of many interacting particles. Indeed, in random Lorentz gases, or in systems with more degrees of freedom, there is no evidence that the response exhibits problems of the type as described by van Kampen. However, as discussed in Part 1, there are plenty of other physical systems, particularly structures being defined on small scales, for which one may obtain a very different answer.

11.3

Summary

(1) Using the mixing and chaotic periodic Lorentz gas as a standard model saves us from using noise in order to generate Brownian motion-like trajectories of a moving particle. This model thus provides an alternative to stochastic Langevin equations for analyzing the interplay between a subsystem and thermal reservoir. By making the friction coefficient velocity-dependent, under the constraint that the total kinetic energy of the system is constant at any time step, we arrive at the formulation of the Gaussian thermostat for the periodic Lorentz gas driven by an external electric field. Surprisingly, this combination of subsystem plus abstract thermal reservoir is deterministic and time-reversible. It is also non-Hamiltonian reflecting that the resulting dynamical system is dissipative. (2) For Gaussian thermostated dynamical systems in nonequilibrium situations there is an identity between the absolute value of the rate of phase space contraction and the rate of thermodynamic entropy production. (3) As a consequence of this identity the electrical conductivity is a simple function of the sum of Lyapunov exponents of the dynamical system, which is known as the Lyapunov sum rule. Similar formulas hold for other transport coefficients. Gaussian thermostated dynamical systems with more degrees of freedom, such as interacting many-particle systems, may furthermore exhibit a specific symmetry in the spectrum of Lyapunov exponents, which enables one to split the spectrum into

May 8, 2007

4:1

11.3. Summary

World Scientific Book - 9in x 6in

klages˙book

203

conjugate pairs of exponents all summing up to the same value. If conjugate pairs exist the Lyapunov sum rule drastically simplifies. The precise conditions under which conjugate pairing holds are under active discussion. (4) Gaussian thermostated dynamical systems are characterized by the existence of multifractal attractors in the Poincar´e sections of the phase space. For the Gaussian thermostated driven periodic Lorentz gas the fractal structure of the attractor is determined by sequences of tangent collisions reflecting strong dynamical correlations. Under variation of the external electric field strength the topology of the attractor changes exhibiting a complicated bifurcation scenario. (5) Computer simulations show that the electrical conductivity of this model is a nonlinear function of the electric field strength. It furthermore displays irregularities on fine scales that might be of a fractal origin. Ohms law is proven to hold for sufficiently small field strengths, which do not seem to be accessible in computer simulations. This problem of detecting a regime of linear response reminds one to some of van Kampen’s objections against the validity of linear response in chaotic dynamical systems.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

This page intentionally left blank

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Chapter 12

The Nos´ e-Hoover thermostat

We construct and analyse a second fundamental form of deterministic and time-reversible thermal reservoir known as the Nos´e-Hoover thermostat. To motivate this scheme we need a Liouville equation that also holds for dissipative dynamical systems, which we introduce in Section 12.1. After defining this thermostat in Section 12.2 we focus on the chaos and transport properties of the Nos´e-Hoover thermostated driven periodic Lorentz gas by characterizing the associated NSS, see Section 12.3. For this purpose we first check whether there holds an identity between the average phase space contraction rate and thermodynamic entropy production as was obtained for the Gaussian thermostat. We then discuss computer simulation results for the attractor and for the electrical conductivity of the Nos´eHoover driven periodic Lorentz gas by comparing them to their Gaussian counterparts. In Section 12.4 we briefly elaborate on subtleties resulting from the construction of the Nos´e-Hoover thermostat, and we outline how this scheme can be generalized leading to a variety of additional, similar thermal reservoirs. 12.1

The dissipative Liouville equation

In this section we particularly follow the presentations of Refs. [Eva90b; Dor99]. For the sake of simplicity we shall formally restrict ourselves to the case of a one-particle system such as the Lorentz gas. Let the position r and the velocity v of a point particle be defined by some equations of motion. We wish to consider a Gibbs ensemble of particles moving in the same dynamical system. That is, we take a large collection of points in the phase space of the system, which in general may be restricted by (geometric) boundary conditions, cf. also Section 10.3 for related definitions. Let ρ = 205

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

206

klages˙book

The Nos´ e-Hoover thermostat

ρ(t, r, v) be the distribution function, or probability density, associated with the ensemble of moving particles. It is the number of points of the ensemble that one can find in the phase space volume element drdv centered around the position (r, v) at time t divided by the size of this volume element and by the total number of points N of the ensemble. By integrating over the whole accessible phase space this function is properly normalized to one. Let us consider the fraction N (t)/N , where N = N (0) is the number of ensemble members at time zero while N (t) is the number of points at time t. The total change of this fraction of ensemble points in a volume element V with surface A in the phase space obeys the balance equation Z Z ∂ρ d N (t) drdv dA · (Fρ) , (12.1) = + dt N ∂t V A

where (˙r, v) ˙ ∗ = F(r, v) stands for the flux given by the equations of motion of a dynamical system such as, e.g., Eqs. (11.1), and ∗ denotes the transpose. The first integral stands for the source and the second one for the flux term. If we assume that once an ensemble of particles is chosen the number of points is conserved by the equations of motion, i.e., that there is no vanishing of points in phase space due to chemical reactions or related mechanisms, the left hand side must be zero at any instant of time. Applying Gauss’ divergence theorem to the flux term enables us to combine both integrals of Eq. (12.1) leading to the differential formulation [Pat88; Dor99; Eva90b] ∂ρ + ∇ · (Fρ) = 0 ∂t

(12.2)

dρ + ρ(∇ · F) = 0 . dt

(12.3)

or alternatively

Note that the continuity equation Eq. (12.1) formulates only the conservation of the number of points in phase space. Hence, Eqs. (12.2), (12.3) are valid even if the dynamical system is not phase space volume preserving. They are also time reversible. For Hamiltonian dynamical systems ∇·F = 0 and the well-known Liouville equation for volume-preserving dynamics is recovered [Rei65; Hua87; Pat88; Dor99], dρ =0 . (12.4) dt This equation expresses the fact that according to Hamiltonian dynamics an ensemble of points moves like an incompressible fluid in phase space,

May 8, 2007

4:1

World Scientific Book - 9in x 6in

12.1. The dissipative Liouville equation

klages˙book

207

which is often referred to as the Liouville theorem in the literature [Rei65; Hua87]. On this occasion some historical remarks are in order. In his original work [Lio38] Liouville did not care at all about Hamiltonian dynamics. He just started from the equations of motion for an arbitrary dynamical system and derived the evolution equation for its Jacobian determinant, from which Eq. (12.3) follows in a straightforward way. Hence, as pointed out by Andrey [And86],1 Eq. (12.3) should rather simply be called “the Liouville equation”. However, in textbooks usually only Eq. (12.4) is derived and called “the Liouville equation” [Rei65; Hua87], probably because Liouville’s work was mainly applied to Hamiltonian dynamics. Respectively, in the more specialized literature Eq. (12.3) is often denoted as “the generalized Liouville equation” [Ger73; Ste79; Ste80; And85; Ram86; Hoo99; Ram02]. In this book we will stick to the historically correct denotation and call Eq. (12.3) “the Liouville equation”. Alternative derivations of the Liouville equation in its most general form can be found in Refs. [Ger73; Ste79; Ste80; And86; Ram02; Ser03], where we should highlight the seminal work of Gerlich [Ger73]. Recently Tuckerman et al. [Tuc97; MT99] derived a seemingly different (generalized) Liouville equation by claiming that Eq. (12.3) is not correct, see Refs. [Mun00; Tuc00] for summaries of this approach. The resulting controversy about the correct form of the Liouville equation [Hoo98b; Rei98; Tuc98b; Eva98; Tuc98a; MT99] was eventually resolved by Ramshaw [Ram02] starting from the observation that Eq. (12.3) is not covariant [Ram86]. However, covariance can be achieved by incorporating a respective coordinate transformation into the Liouville equation, which is generally a function of the metric. Along these lines the connection between Eq. (12.3) and the form of the Liouville equation considered in Refs. [Tuc97; Tuc98b; MT99] could be established. Applying this transformation both formulations turned out to be equivalent, however, for many purposes Eq. (12.3) appears to be the more convenient choice [Ram02; Ser03]. The Gaussian thermostated driven Lorentz gas provides a simple example where the divergence in Eq. (12.3) does not disappear. It may thus be replaced according to Eqs. (11.5), (11.6) leading to ∂ρ = αρ . (12.5) ∂t By employing this equation the Gibbs entropy production for the Gaussian thermostated driven Lorentz gas can be computed right away, which 1 See

also the footnote on p.69 of Ref. [Gas98a] for a related brief historical note.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

208

klages˙book

The Nos´ e-Hoover thermostat

furnishes an alternative derivation of the identity Eq. (11.8) between phase space contraction and thermodynamic entropy production [Dor99; Det00a]. This and further important implications of the Liouville equation concerning irreversibility and entropy production will be discussed in Section 13.2. We are now set up to introduce a second fundamental type of deterministic thermal reservoir generalizing the Gaussian scheme. For this construction the Liouville equation Eq. (12.3) will be used as a convenient starting point. 12.2 12.2.1

Construction of the Nos´ e-Hoover thermostat Heuristic derivation

For the Gaussian thermostated driven Lorentz gas the velocity distribution of the moving particle is essentially a microcanonical one, that is, the system is constrained onto an energy shell in phase space. Here we introduce a generalized thermostating scheme that, under certain conditions, enables one to transform the velocities of a dynamical system onto a canonical distribution in thermal equilibrium. Our heuristic derivation follows the presentations in Refs. [Hoo85; Hoo91]; see also a respective short note in Ref. [Hol86] and a formal generalization of this argument in Refs. [Kus90; Bul90a]. We remark that quantum mechanical formulations of deterministic and time-reversible thermal reservoirs can be obtained along the same lines [Kus93; Men01; Kus02; Men03]. Let us start again from the driven Lorentz gas equations of motion Eqs. (11.1), which already exhibit a coupling with a thermal reservoir in the form of the friction variable α. Let us now assume that α is not necessarily a simple analytical function of v like the Gaussian thermostat Eq. (11.4), but that it is determined by some unknown differential equation instead. Formally, the equations of motion then read (˙r, v, ˙ α) ˙ ∗ = F(r, v, α)

,

(12.6)

and the generalized Liouville equation Eq. (12.3) can be written as ∂ρ ∂ρ ∂ρ ∂ρ ∂ v˙ ∂ α˙ ∂ r˙ =0 . (12.7) + r˙ + v˙ + α˙ +ρ + + ∂t ∂r ∂v ∂α ∂r ∂v ∂α In contrast to the Gaussian thermostat, which was designed to keep the kinetic energy constant at any time step specifically in nonequilibrium, here

May 8, 2007

4:1

World Scientific Book - 9in x 6in

12.2. Construction of the Nos´ e-Hoover thermostat

klages˙book

209

our strategy is to define the thermostating mechanism first in thermal equilibrium. The goal is to come up with a dissipation term that transforms the velocity distribution of the subsystem onto the canonical one. In a second step we then create a nonequilibrium situation. Here we will check whether our thermostat is able to generate a NSS, and if so we will study what its precise properties are. In other words, our goal is to find a differential equation for α that is consistent with the existence of a canonical velocity distribution for an ensemble of subsystem particles. For this purpose we make the ansatz [Nos84a; Nos84b; Pos88; Pos89; Eva90b; Mor98; Hoo99] v2 2 , (12.8) − (τ α) ρ(t, r, v, α) = const. exp − 2T where both the distribution for the velocities of the subsystem and for the thermal reservoir variable α are assumed to be canonical, cf. Eq. (10.21). 2 According to equipartitioning of energy T is identified as the temperature of the desired canonical distribution, and in thermal equilibrium this temperature should be identical to the temperature of the associated thermal reservoir. Furthermore, since α has a dimension of 1/time we needed to introduce the new quantity τ > 0 with [τ ] = time in order to make the exponent of the second term dimensionless. We now combine Eq. (12.8) with the equations of motion for a single free particle, see Eqs. (11.1) with ε = 0. Feeding these equations as well as the ansatz for ρ into Eq. (12.3) yields for α the differential equation ∂ α˙ v2 − αα2τ ˙ 2 − 2α + =0 . (12.9) T ∂α This equation can be further simplified by ad hoc restricting to the case ∂ α/∂α ˙ = 0. We then arrive at the simple solution α

v 2 − 2T . (12.10) τ 2 2T This differential equation for the thermostating variable α, together with the equations of motion Eqs. (11.1), defines the so-called Nos´e-Hoover thermostat for the driven periodic Lorentz gas. The first version of such a thermostat transforming onto canonical distributions was developed by Nos´e in 1984 [Nos84a; Nos84b]. His original derivation proceeds along different lines by starting from a generalized Hamiltonian formalism, see Section 12.3.2. α˙ =

2 For

an N -particle system with d degrees of freedom the exponent compiling Pthe total energy of subsystem and thermal reservoir reads respectively Etotal /T = ( i vi2 /2 + N dT τ 2 α2 /2)/T [Nos84a; Nos84b; Pos88; Pos89; Eva90b; Mor98; Hoo99 ].

May 8, 2007

4:1

World Scientific Book - 9in x 6in

210

The Nos´ e-Hoover thermostat

Furthermore, it features an at least for practical purposes spurious differential equation related to another reservoir variable, which was eliminated by Hoover [Hoo85]. This simplified version of the Nos´e thermostat was coined the Nos´e-Hoover thermostat [Eva85b]. In first explorations it was particularly applied to the harmonic oscillator, and the combined system was respectively called the Nos´e-Hoover oscillator [Hoo85; Pos86; Hol86; Mar92b; Nos93; Hoo96a; Pos97; Hoo97; Lai03]. As for the Gaussian thermostat, the Nos´e-Hoover scheme can be used isokinetically by only constraining the kinetic energy of particles, see above, or isoenergetically by constraining the total internal energy of soft particles [Mor98; Det00a; Ron02a].3 It can also be adapted to drive interacting many-particle systems under nonequilibrium conditions into NSS [Nos84a; Nos84b; Pos88; Pos89; Eva90b; Hoo91; Hes96a; Mor98; Hoo99; Det00a; Mun00; Ron02a]. The results obtained from such simulations are typically well in agreement with predictions from irreversible thermodynamics and linear response theory [Eva90b; Sar98]. Hence, like the Gaussian modeling of a thermal reservoir, the Nos´e-Hoover scheme became a widely accepted useful tool for performing nonequilibrium molecular dynamics computer simulations not only of simple but also of more complex fluids such as polymer melts, liquid crystals and ferrofluids [Hes96a; Hes96b; Hes97], of proteins in water and of chemical processes in the condensed matter phase [Tuc00]. 12.2.2

Physics of this thermostat

Like the equations of motion Eqs. (11.1), (11.4) of the Gaussian thermostated Lorentz gas, the Nos´e-Hoover counterparts Eqs. (11.1), (12.10) are also deterministic and time-reversible, though non-Hamiltonian. However, in contrast to the microcanonical Gaussian thermostat Nos´e-Hoover was constructed for generating a canonical velocity distribution. More than Gaussian constraint dynamics, Nos´e-Hoover thermostated systems may thus be considered as deterministic and time-reversible variants of Langevin dynamics, cf. Eqs. (11.1), (12.10) with Eq. (10.1). 3 This

denotation is not unique in the literature in that some authors make an even finer distinction [Eva05]: They call a thermal reservoir that constrains the kinetic energy of a system a thermostat and a thermal reservoir that constrains the internal energy an ergostat. Only if the kinetic energy is constrained such that it does not fluctuate, as in the case a Gaussian scheme, they speak of isokinetic dynamics, whereas a non-fluctuating internal energy, such as generated by the Nos´e-Hoover method, is called isoenergetic dynamics. However, for this book it is not necessary to make this very fine distinction.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

211

12.2. Construction of the Nos´ e-Hoover thermostat

For interacting many-particle systems the Nos´e-Hoover thermostat conveniently works in thermal equilibrium [Nos84a; Eva85b; Pos89]. However, similarly to the Gaussian thermostat this is not the case for the Lorentz gas without external fields, which is due to a malfunctioning of the friction coefficient related to the hard disk collisions [Rat00a]. Nevertheless we are safe in studying the driven Lorentz gas because, in analogy to the Gaussian thermostat, as soon as an electric field is added this problem disappears. Following this comparison with Langevin dynamics one may inquire about the existence of a fluctuation-dissipation theorem such as Eq. (10.2) for equilibrium Nos´e-Hoover dynamics. But as we already discussed in Section 11.1, for dissipative deterministic dynamical systems such as the Gaussian or the Nos´e-Hoover thermostated Lorentz gas there is no separation between stochastic forces and dissipation matching to the respective terms in the Langevin equation Eq. (10.1). Actually, for both deterministic thermostats the velocity-dependent friction coefficient is itself a fluctuating variable, in addition to the fact that the position space Lorentz gas dynamics is also nonlinear and fluctuating. Hence, a balance equation like Eq. (10.2) between a constant friction coefficient and stochastic forces cannot exist; however, for more general fluctuation-dissipation relations in the case of deterministic dynamics see Ref. [Rue99b]. In order to better understand the action of the Nos´e-Hoover fricition force let us look at the ensemble averaged version of Eq. (12.10), < α˙ >=

< v 2 > −2T τ 2 2T

.

(12.11)

If the thermostat acts properly in equilibrium, the average of the fluctuations of the thermostating variable α must be zero and consequently < v 2 >= 2T on the right hand side, which corresponds to equipartitioning of energy for a system with two degrees of freedom.4 T thus serves as a control parameter representing the temperature of the thermal reservoir. Choosing some T , the thermostat variable α acts accordingly in order to thermalize the subsystem onto the same temperature. Consequently, in contrast to the Gaussian thermostat the proper temperature is not attained instantaneously but with respect to an ensemble average over a canonical velocity distribution, as we stipulated with our ansatz Eq. (12.8). That is, for a Lorentz gas particle the kinetic energy is allowed to fluctuate in time around a mean value of < v 2 > /2. 4 Since here we do not have Gaussian constraints we may safely assume the existence of two degrees of freedom for the subsystem.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

212

The Nos´ e-Hoover thermostat

As in the case of the Gaussian thermostat, α supplemented by the respective dynamics for this variable represents both the coupling of the subsystem to the thermal reservoir as well as the action of the thermal reservoir itself. In a way, one may think of α as a dynamical variable upon which all hypothetical degrees of freedom of the thermal reservoir are projected just for the purpose of dissipating energy from the subsystem. Along this line of reasoning τ may then be interpreted as the reservoir response time with respect to exchanging energy between the subsystem and the thermal reservoir. τ thus serves as another control parameter determining the efficiency of the interaction. Note that this control parameter does not exist for the Gaussian thermostat. Let us look at the two limiting cases of Eq. (12.8) with respect to a variation of τ : For τ → ∞ obviously α → const., and one recovers the familiar case of Newton’s irreversible equations of motion with constant Stokes friction. On the other hand, making τ smaller implies that there is a more immediate response of the reservoir to fluctuations of the kinetic energy in the subsystem. Consequently, in the limit of τ → 0 there must eventually be instantaneous control leading back to the constraint of keeping the kinetic energy constant at any time step as formulated by the Gaussian thermostat. This can be seen by moving τ 2 from the denominator on the right hand side of Eq. (12.10) to the left hand side and performing the limit τ → 0 enforcing v 2 = 2T .5 For more detailed studies of how Nos´e-Hoover dynamics depends on the value of the coupling constant τ see Refs. [Nos84b; Hoo85; Eva85b; Hol86; Nos91; Nos93; Hol95; Kus90]. The Nos´e-Hoover thermostat may furthermore be classified as an integral feedback mechanism, according to the formal solution of the differential equation for the friction variable Z t α(t) = α(0) + ds α(s) ˙ (12.12) 0

with α˙ being determined by Eq. (12.10) [Hol86; Hol87; Hol95]. Respectively, the Gaussian version is differential in time, due to the explicit functional form for the velocity-dependent friction coefficient. Interestingly, the existence of these two limiting cases shows up in the specific functional form of the velocity distribution under variation of τ : For τ → 0 and small enough 5 However,

note that for a vanishing external field ε = 0 the friction coefficient of the Gaussian thermostat Eq. (11.4) is by default zero, at least for the driven periodic Lorentz gas. Without external forces the respective solution for Nos´e-Hoover should therefore also go to zero for τ → 0. Thus, in detail the relation between the Gaussian and the Nos´e-Hoover thermostat is more intricate, see also the discussion in Chapter 16.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

213

12.3. Properties of the Nos´ e-Hoover thermostat

field strength ε this function must evidently approach a microcanonical velocity distribution ρ(v) ∼ δ(2E − v 2 ), whereas for τ → ∞ it yields a canonical one. Consequently, between these two limiting cases there must exist a superposition of these two different shapes in the form of a bimodal velocity distribution with a dip at the place of the former maximum of the canonical distribution. This transition is depicted in Fig. 12.1, where ρ(vx ) for a Nos´e-Hoover thermostated driven periodic Lorentz gas is represented for three different values of τ [Rat00a]; for related results concerning other dynamical systems thermostated by Nos´e-Hoover (chain) thermostats see Refs. [Nos91; RB06]. Note the symmetry breaking in Fig. 12.1 because of the external electric field, however, one clearly recognizes the approach towards a microcanonical distribution for τ → 0.6 We will come back to a more explicit discussion of this interesting point in Chapter 16, where it will be shown that bimodal velocity distribution functions that are very similar to the one discussed here play an important role for so-called active Brownian particles, which are thought to model, among others, biological cell motility. 12.3 12.3.1

Properties of the Nos´ e-Hoover thermostat Chaos and transport

For the Gaussian thermostated driven Lorentz gas the crucial relation linking thermodynamics to dynamical systems theory was the identity Eq. (11.8) between the average phase space contraction rate and the thermodynamic entropy production in the form of Joule’s heat. Let us now check whether the Nos´e-Hoover thermostat fulfills this identity as well. For this purpose we calculate the energy balance between subsystem and thermal reservoir [Pos88; Rat00a] by starting from the total energy Etot for the combined system (without incorporating any external field), see the exponent in Eq. (12.8), Etot = v 2 /2 + T τ 2 α2

.

(12.13)

If the system is properly thermostated yielding a NSS, the average of the time derivative of the total energy must be zero, < dEtot /dt >= 0. Differentiating Eq. (12.13) and replacing the derivatives by the equations of 6 Curiously, in the case of non-ergodic Nos´ e-Hoover dynamics similar microcanonicallike and canonical-like distributions can be obtained with respect to different choices of initial conditions for ensemble of particles, see the preprint of Ref. [Rat00a] for an example.

4:1

World Scientific Book - 9in x 6in

214

klages˙book

The Nos´ e-Hoover thermostat

0.4 ρ(vx)

May 8, 2007

0.2 0.0

−2

0 vx

2

Fig. 12.1 Velocity distributions %(vx ) for an ensemble of moving point particles in the thermostated driven periodic Lorentz gas for an electric field of strength ε = 0.5, which is parallel to the x-axis, at temperature T ' 0.6: Nos´e-Hoover thermostat with a reponse time of the thermal reservoir of τ 2 = 0.01 (dotted curve), τ 2 = 1 (dashed curve), τ 2 = 1000 (long dashed curve) in comparison to the application of the deterministic boundary thermostat discussed in Chapter 15 (solid curve). Note the transition from a microcanonical-like to a canonical distribution under variation of τ . The figure is from Ref. [Rat00a].

motion Eqs. (11.1), (12.10) yields <ε·v > =2<α> (12.14) T The right hand side equals minus the average phase space contraction rate κp defined by Eq. (11.5), whereas the left hand side is obviously Joule’s heat. We have thus corroborated the identity Eq. (11.8) for Nos´e-Hoover dynamics. Alternatively, in order to check for the identity one may start directly from the Nos´e-Hoover equations of motion Eqs. (11.1), (12.10). Multiplying the differential equation for the velocity with v and replacing the velocity squared on the right hand side by Eq. (12.10) one arrives at v˙ · v = ε · v − α(ατ ˙ 2 2T + 2T ) .

(12.15)

One may now collect all terms related to fluctuations on the left hand side by considering their statistical averages, d 1 2 2 2 v + τ T α >=< ε · v − α2T > . < (12.16) dt 2

May 8, 2007

4:1

World Scientific Book - 9in x 6in

12.3. Properties of the Nos´ e-Hoover thermostat

klages˙book

215

The left hand side is just identical to the average of the total energy Eq. (12.13) and must be zero in a NSS, see above, hence the right hand side yields again the identity. This derivation is useful if the expression for the energy term of the thermostat is not known in advance, see Section 14.3 for some examples. Based on the identity Eq. (12.14), the Lyapunov sum rule can be derived in the same way as explained in Section 11.2.2. For four nontrivial phase space variables it reads σ(ε) = −

4 T X λi (ε) . ε2 i=1

(12.17)

Hence, with respect to relations between chaos and transport the Nos´eHoover thermostat belongs to precisely the same class of thermal reservoirs as the Gaussian one. 12.3.2

+

Generalized Hamiltonian formalism

A crucial feature of Hamiltonian systems is that the dynamics is phase space volume preserving, which follows from their symplectic structure [Mei92; Abr94]. On the other hand, as was shown for the Gaussian thermostat in Section 11.2.1 and as we just verified for the Nos´e-Hoover thermostat, the application of such deterministic thermostating mechanisms leads to an average phase space contraction in nonequilibrium. From that point of view it is surprising that for Gaussian and Nos´e-Hoover dynamics a generalized Lagrangian and Hamiltonian formalism exists in which this dynamics is phase space volume preserving. Here we outline some main ideas of the Hamiltonian approach for the example of the isokinetic Nos´e-Hoover thermostat. Let us consider the general case of a subsystem where a particle of mass m moves under the influence of a potential u(r). Let the microscopic dynamics of this particle be described by the configuration variables and conjugate momenta (q, p) with q = r and p = mv. The dynamics of the thermal reservoir is modeled by only one degree of freedom in terms of the generalized phase space variables (q0 , p0 ) with q0 = r0 and p0 = M v0 . In formal analogy, the new quantity M is sometimes denoted the mass of the reservoir. All these representations may be called physical variables, since (q, p) represents the actual position and momentum of some particle. For the decoupled system the formal Hamiltonian is H0 (q, p, q0 , p0 ) = E(p, p0 ) + U (q, q0 ) ,

(12.18)

May 8, 2007

4:1

World Scientific Book - 9in x 6in

216

klages˙book

The Nos´ e-Hoover thermostat

where E(p, p0 ) = p2 /(2m) + p20 /(2M ) stands for the kinetic energy of particle plus reservoir and U (q, q0 ) = u(q) + 2T q0 for the potential energy of both systems. Here T is identified with the reservoir temperature.7 Hamilton’s equations of motion for the decoupled system then read p m ∂u p˙ = − ∂q p0 q˙0 = M p˙0 = −2T . q˙ =

(12.19)

The problem is now two-fold: firstly, to find a Hamiltonian that generalizes Eq. (12.18) in suitably representing the coupled system, and secondly, to derive the Nos´e-Hoover equations of motion, such as the special case Eqs. (11.1), (12.10) for the driven Lorentz gas, from this new Hamiltonian. In a slightly modified setting, this problem first appeared in Nos´e’s groundbreaking work [Nos84a; Nos84b] leading to the formulation of the so-called Nos´e Hamiltonian. Here we follow a simplified approach as it was proposed later on in Refs. [Det97b; Cho98]. Since the Nos´e-Hoover equations of motion are not Hamiltonian it is clear that a Hamiltonian formulation of this dynamics can only be achieved by choosing a new set of generalized coordinates, for which we may write (Q, P) in the case of the subsystem and (Q0 , P0 ) in the case of the reservoir. Now we implement the so-called exponential coupling ansatz for the Hamiltonian in these generalized coordinates, H(Q, P, Q0 , P0 ) := e−Q0 E(P, P0 ) + eQ0 U (Q, Q0 ) .

(12.20)

These exponential prefactors define a nonlinear, multiplicative coupling between subsystem and thermal reservoir in generalized coordinates. Physically speaking, they suitably rescale the total energy of the system thus serving to keep the energy of the full system constant on average, whereas in nonequilibrium it would increase to infinity otherwise. The different signs of the exponents are adjusted to the different types of kinetic and potential energy. We can now derive Hamilton’s equations of motion from this new 7 Strictly speaking the choice of the potential term for the reservoir, and fixing the number of degrees of freedom associated with it, is a bit more subtle [Cho98; Mor98].

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

217

12.3. Properties of the Nos´ e-Hoover thermostat

Hamiltonian to ˙ = e−Q0 P Q m Q 0 ˙ = −e ∂u P ∂Q P 0 Q˙ 0 = e−Q0 M P˙0 = 2(e−Q0 E(P, P0 ) − eQ0 T ) ,

(12.21)

where we have imposed that H(Q, P, Q0 , P0 ) = 0. Matching the first equation to the corresponding one in physical coordinates Eqs. (12.19) suggests the following transformation between physical and generalized coordinates of the subsystem, Q=q P = e Q0 p .

(12.22)

Making the same choice for the reservoir coordinates, Q0 = q 0 P0 = e Q 0 p 0

,

(12.23)

we obtain a complete set of equations for transforming between physical and generalized coordinates. Identifying M = 2T τ 2 and α = p0 /M , the Nos´e-Hoover equations Eqs. (11.1), (12.10) for the driven Lorentz gas are precisely recovered from Eqs. (12.21) for m = 1. Let us emphasize at this point that the transformation between the different coordinates given by Eqs. (12.22), (12.23) is strictly noncanonical by not preserving the phase space volume. Therefore, one should carefully distinguish between traditional Hamiltonian formulations of classical mechanics and the generalized Hamiltonian formalism outlined above. The idea of using a generalized Hamiltonian formalism in order to define the action of a thermal reservoir was pioneered by Nos´e [Nos84a; Nos84b] who constructed his thermostating mechanism along these lines. In Refs. [Nos91; Nos93; Hoo85; Eva85b; Pos86; Eva90b] this scheme was applied to different types of subsystem. Later on Dettmann and Morriss showed that there also exist generalized Hamiltonian formulations for the Gaussian isokinetic [Det96b] and isoenergetic [Det99a] thermostat, see Ref. [Hoo88a] for a pre-version of this approach. They also came up with a modified Nos´e Hamiltonian leading more straightforwardly to the Nos´eHoover equations than starting from Nos´e’s original formulation [Det97b;

May 8, 2007

4:1

World Scientific Book - 9in x 6in

218

The Nos´ e-Hoover thermostat

Hoo99], as is sketched in the derivation above. The generalized Lagrangian approach was outlined by Choquard in Ref. [Cho98], see also this reference as well as Refs. [Mor98; Det00a] for elucidating summaries concerning the subject of this section. There are furthermore interesting studies to formulate non-Hamiltonian dynamics by using Lie-Poisson [Kus92], generalized Poisson [Ser01; Ser03], or Dirac brackets [Ron02a; Ser05]. We remark that the symplectic properties of these generalized Hamiltonians appear to be intimately related to the validity of the conjugate pairing rule discussed in Section 11.2.2. 12.3.3

Fractals and transport

As we have shown in Sections 11.2.1 and 12.3.1, both the Gaussian and the Nos´e-Hoover thermostated driven Lorentz gas exhibit an identity between the average phase space contraction rate and thermodynamic entropy production. For the Nos´e-Hoover case we now discuss further chaos and transport properties, which are the fractality of the corresponding attractor, the structure of the associated bifurcation diagram and the response to an electric field in terms of the field-dependent conductivity. Particularly, we elaborate on similarities and differences of these properties in comparison to the respective figures shown for the Gaussian thermostat. The computer simulation results for the Nos´e-Hoover thermostated driven periodic Lorentz gas presented here are from Ref. [Rat00a]. Fig. 12.2 depicts a projection of the attractor of the Nos´e-Hoover thermostated driven periodic Lorentz gas onto the phase space in Birkhoff coordinates.8 Some multiple folding in phase space is clearly visible indicating that, as in the case of the Gaussian thermostat, this attractor might again be multifractal. For Nos´e-Hoover no values for a (fractal) dimension quantitatively assessing this structure have yet been computed.9 By comparing this attractor to the one obtained from Gaussian thermostating Fig. 11.2 one realizes that the topology of both sets is essentially the same. This is not too surprising, since both systems share the same geometry, and as we discussed in Section 11.2.3 the main features of this structure are induced by the geometry of the scatterers. That the Nos´e-Hoover thermostat attempts 8 Note that, in contrast to the Gaussian thermostated driven periodic Lorentz gas, Fig. 12.2 does not represent a Poincar´e surface of section: For Nos´e-Hoover the kinetic energy is not kept constant but fluctuates, hence this plot is composed of points with different absolute values of the velocity of the colliding particle. 9 But see Section 15.4.2 for a very similar attractor for which the Kaplan-Yorke dimension was found to be fractal.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

12.3. Properties of the Nos´ e-Hoover thermostat

klages˙book

219

Fig. 12.2 Attractor for the Nos´e-Hoover thermostated periodic Lorentz gas driven by an electric field of strength ε = 1, which is parallel to the x-axis. β and sin γ are the same as in Fig. 11.2 and are defined in Fig. 11.1. The temperature is T ' 0.774, and the reservoir coupling parameter, see Eq. (12.10), is τ = 0.1. The results are from Ref. [Rat00a].

to transform the system onto a canonical distribution explains why the attractor related to this dynamics appears to be a smoothed-out version of the Gaussian counterpart. These results suggest that the appearance of fractal attractors is also typical for Nos´e-Hoover thermostated systems. Numerical computations of fractal dimensionality losses in many-particle systems thermostated by Nos´e-Hoover confirm this statement [Hol87; Pos88; Pos89; Aok03; Kus02; Hoo02b; Pos04b]. Fig. 12.3 contains bifurcation diagrams for the Nos´e-Hoover thermostated driven periodic Lorentz gas at three different values of the reservoir coupling constant τ , see Eq. (12.10). For quick response of the reservoir, τ 1, the attractor is phase space filling up to ε < 2, which is reminiscent of the results for the Gaussian thermostated Lorentz gas presented in Fig. 11.3. Note that Fig. 11.3 depicts the angle θ as a function of the field strength ε, whereas Fig. 12.3 displays the angle β, respectively. However, both bifurcation diagrams exhibit qualitatively the same behavior. This is

4:1

World Scientific Book - 9in x 6in

220

klages˙book

The Nos´ e-Hoover thermostat

β

(a) 3

0

1

2

3

1

2

3

β

(b) 3

0 (c) 3 β

May 8, 2007

0

0

2

4

6

8

10

ε Fig. 12.3 Bifurcation diagrams for the Nos´e-Hoover thermostated periodic Lorentz gas driven by an electric field of strength ε, which is parallel to the x-axis. β is defined in Fig. 11.1. The temperature is fixed to T = 0.5 and the values of the reservoir coupling parameter τ , see Eq. (12.10), are: (a) τ = 0.1, (b) τ ' 31.6 (τ 2 = 1000) and (c) the dissipative limit of τ → ∞ corresponding to a constant friction coefficient α whose value was chosen to be 1. The figure is from Ref. [Rat00a].

in agreement with the discussion of Section 11.2.2 where we argued that the Nos´e-Hoover thermostat yields Gaussian constraint dynamics in the limit of τ → 0. By making the response time of the reservoir larger, the attractor collapses at smaller field strengths onto periodic orbits than for smaller response times, see Fig. 12.3 (b), which matches to naive physical reasoning. In detail, however, the situation is much more intricate [Rat00a], as is indicated by Fig. 12.3 (c), which displays the dissipative limit of τ → ∞. Here the bifurcation diagram depends on the value for the constant friction coefficient α. For all values studied numerically, one observes a kind of inverted bifurcation scenario compared to the Gaussian thermostated model and to Figs. 12.3 (a) and (b) in that the attractor starts with periodic orbits and covers the whole accessible phase space only for higher field

4:1

World Scientific Book - 9in x 6in

klages˙book

221

12.3. Properties of the Nos´ e-Hoover thermostat

(a)

σ

0.21

0.19

0.17 0.1

0.5

0.9

1.3

0.9

1.3

ε (b) 0.32 σ

May 8, 2007

0.3

0.28 0.1

0.5 ε

Fig. 12.4 Electrical conductivity σ(ε) for the Nos´e-Hoover thermostated periodic Lorentz gas driven by an electric field of strength ε, which is parallel to the x-axis. The temperature is T = 0.5 and the values of the reservoir coupling parameter τ , see Eq. (12.10), are: (a) τ = 0.1 and (b) τ ' 31.6 (τ 2 = 1000). The numerical uncertainty of each point is less than the symbol size. The figure is from Ref. [Rat00a].

strengths [Rat00a]. Fig. 12.3 thus shows that for the Nos´e-Hoover thermostated driven periodic Lorentz gas the specific structure of the fractal attractor sensitively depends on the choice of the reservoir coupling parameter τ ; see also Refs. [Nos84b; Hoo85; Eva85b; Hol86; Nos91; Nos93; Hol95; Kus90] for studies concerning the variation of τ in other systems and for the impact on dynamical systems properties. As one might suspect with respect to these bifurcation diagrams, there is no unique result for the field-dependent electrical conductivity σ(ε) at different values of the response time τ either, not even for small field strengths ε 2. Fig. 12.4 presents computer simulation results for the two parameter values of τ studied in Figs. 12.3 (a) and (b). The conductivity for τ = 0.1 looks like a smoothed-out version of the Gaussian thermostat conductivity depicted in Fig. 11.4, whereas the curve at τ ' 31.6 appears to be more stretched out along the ε-axis. In the latter case it is not even

May 8, 2007

4:1

World Scientific Book - 9in x 6in

222

The Nos´ e-Hoover thermostat

clear whether the conductivity is globally decreasing or increasing. There is little to say concerning the validity of the Einstein relation Eq. (11.13) in the limits of τ > 0 and ε → 0 as discussed for the Gaussian version in Section 11.2.4, since the Nos´e-Hoover thermostat does not work properly for the Lorentz gas in thermal equilibrium, see Section 12.2.2. Hence, the field-free diffusion coefficient cannot be computed independently. That both field-dependent Nos´e-Hoover conductivities are significantly less irregular than the Gaussian thermostat result of Fig. 11.4 may be understood with respect to the more “random” nature of the Nos´e-Hoover thermostat. According to its definition, it modifies the billiard dynamics more profoundly than in case of the Gaussian constraint by transforming onto canonical velocity distributions, see again Fig. 12.1 for the probability densities at these parameters. Nevertheless, both conductivities still exhibit pronounced irregularities in the form of non-monotonicities, and there are no indications for a regime of linear response, at least not in the numerically accessible region of ε > 0.1. In other words, even the Nos´e-Hoover thermostat featuring a canonical velocity distribution does not lead to a more profound appearance of an Ohmic regime. In summary, from a statistical mechanical point of view we have the rather unpleasant result that not only dynamical systems properties such as bifuraction diagrams but also transport properties such as the fielddependent electrical conductivity of the deterministically thermostated, driven periodic Lorentz gas strongly depend on details of the specific way of thermostating. As in the case of the discussion of van Kampen’s criticism in Section 11.2.4 one may speculate that this is a consequence of the small number of degrees of freedom and of the spatial periodicity of this model.

12.4 12.4.1

+

Subtleties of Nos´ e-Hoover dynamics Necessary conditions and generalizations

As we briefly discussed in Sections 11.1 and 12.2.2, both the Gaussian and the Nos´e-Hoover thermostat do not work for the periodic Lorentz gas without external fields: In this case the Gaussian thermostat is trivially non-existent, since the kinetic energy of the moving particle is conserved by definition anyway, whereas the Nos´e-Hoover thermostat does not act properly because of the hard disk collisions. Controlling the two velocity variables separately by Nos´e-Hoover eliminates this problem [Rat00a]. Alternatively, a small electric field resolves these intricacies for both ther-

klages˙book

May 8, 2007

4:1

12.4.

World Scientific Book - 9in x 6in

+ Subtleties

of Nos´ e-Hoover dynamics

klages˙book

223

mostats leading to well-defined NSS. For interacting many-particle systems it was confirmed numerically that both the Gaussian [Hoo82; Eva83b; Pos88] and the Nos´e-Hoover thermostat [Nos84a; Eva85b; Pos89] work correctly even in thermal equilibrium. These examples suggest that the thermostated subsystem must fulfill certain necessary conditions in order to attain canonical velocity distributions. In the literature this problem became well-known by studying both the Nos´e-Hoover oscillator, which is a harmonic oscillator subject to a Nos´eHoover thermostat, and a respectively thermalized particle moving in a double-well potential. In the first computer simulations of the Nos´e-Hoover oscillator it was already observed that this system did not reach a canonical velocity distribution in thermal equilibrium [Hoo85; Pos86; Hol86]. This is due to the fact that the unperturbed harmonic oscillator is non-chaotic and perfectly integrable. Hence, under application of some thermal reservoir its regular phase space structure may just be getting deformed, and there is no reason why the resulting subsystem should become ergodic. However, if a previously regular subsystem is not driven to ergodic behavior by the action of some thermostat, the whole phase space will not be sampled appropriately. As a consequence, the resulting velocity distribution may strongly depend on initial conditions and is typically not the canonical one [Hoo85; Pos86; Hol86; Kus90; Bul90a; Ham90; Hoo91; Win92; Mar92b; L’H93; Nos93; Win95; Hoo96a; Hoo97; Mun00; Liu00; Hoo01; Ser01; Lai03]. The same applies to the Nos´e-Hoover thermostated doublewell potential [Kus90; L’H93; Win95; Bar03b]. Jellinek and Berry [Jel88a; Jel89] and Kusnezov et al. [Kus90; Bul90a] thus emphasized that chaos and ergodic behavior are necessary conditions in order to thermalize a subsystem onto a canonical velocity distribution. This indicates that a thermostated subsystem should strictly speaking be mixing [Arn68; Cor82; Sch89; Gas98a; Dor99] implying ergodicity and, typically, also chaotic behavior [Dev89; Ban92; Gla93; Kol97]. These difficulties motivated the construction of a large variety of generalizations of the original Nos´e-Hoover thermostat [Hoo89a; Kus90; Bul90a; Bul90b; Ham90; Hoo91; Win92; Mar92b; L’H93; Win95; Hoo96a; Pos97; Hoo97; Bon99; Mun00; Liu00; Lai03; Bra00a; Bra00b; Hoo01; Ser01; Bar03b]. The formal basis for these efforts was laid out again by Jellinke and Berry [Jel88a; Jel88b] and by Kusnezov et al. [Kus90; Bul90a]: Jellinek and Berry showed that there is the additional freedom of changing the multiplicative coupling between subsystem and thermal reservoir in generalized coordinates while still being consistent with Nos´e’s

May 8, 2007

4:1

224

World Scientific Book - 9in x 6in

The Nos´ e-Hoover thermostat

basic idea of transforming onto canonical distributions.10 Kusnezov et al., in turn, pointed out that further additive contributions to Hamilton’s equations of motion in generalized variables, see Eqs. (12.21), are possible, again without contradicting the existence of canonical velocity distributions. Both approaches will typically lead to thermostated equations of motion that are still time-reversible but more nonlinear than the ones corresponding to the original Nos´e-Hoover scheme. Thus Nos´e-Hoover represents just one of infinitely many choices for modeling a deterministic thermal reservoir transforming onto canonical distributions. Starting from this important conclusion, alternatives to Gaussian and Nos´e-Hoover dynamics were explored by many authors. Their strategy was to improve the Nos´e-Hoover scheme by constructing a thermostat that is able to conveniently thermalize the harmonic oscillator, or the double-well problem, onto a canonical velocity distribution in equilibrium. A straightforward generalization along the lines of Nos´e-Hoover is the thermalization of higher even moments of the subsystem’s velocity distribution onto the canonical ones by means of additional thermostating variables [Hoo89a; Jel89; Hoo96a; Pos97; Hoo97; Liu00]; for a similar idea within the framework of the Gaussian thermostat see Refs. [Wil04; Bri05]. Another option is to thermalize the position coordinates of a moving particle leading to the cubic coupling scheme [Kus90; Bul90a]. Enforcing the virial theorem, Hamilton used a linear scaling of the position coordinates [Ham90], which later on was combined with some multiplicative coupling [L’H93]. A thermalization of odd and even particle velocities, partly supplemented by using higher powers of the corresponding friction coefficients, was designed as a deterministic, time-reversible analogue of the Langevin equation [Bul90b]. More recently, a directional thermostat called a twirler stirring the angular momentum of a moving particle was applied in combination with Gaussian and Nos´e-Hoover thermostats [Hes03]. Another scheme that became very popular is the Nos´e-Hoover chain thermostat [Mar92b; Tuc00], which couples the differential equation for the Nos´e-Hoover friction coefficient additively to another Nos´e-Hoover friction coefficient, and so on. However, this solution was assessed to be unstable out of equilibrium [Eva98; Bra00a] leading to another modification of it 10 In

more detail, starting from the Nos´e Hamiltonian they showed that changing the scaling rules between the physical and the generalized variables, cf. Eqs. (12.22), (12.23), plus possibly also changing a time scaling between physical and generalized time (that we did not discuss in Section 12.3.2), is not at variance with canonical velocity distributions being the solutions of the respective Liouville equation Eq. (12.3).

klages˙book

May 8, 2007

4:1

12.4.

World Scientific Book - 9in x 6in

+ Subtleties

of Nos´ e-Hoover dynamics

klages˙book

225

[Bra00a]. Some multiplicative extensions of Nos´e-Hoover along the lines of Jellinek’s formalism have been worked out by Winkler et al. [Win92; Win95] and became further amended in Ref. [Bra00b]. Another, in a way, multiplicative variant was discussed in Ref. [Ser01]. We finally mention the so-called Nos´e-Poincar´e method representing a scheme that is in-between the original approaches by Nos´e and Nos´eHoover. The resulting dynamics is identical to Nos´e-Hoover, however, it is generated by symplectic equations of motion that enable the application of symplectic integration algorithms, which considerably increases the numerical precision [Bon99; Bar03b]. A further extension of this approach allowing one to model generalized types of Nos´e-Hoover dynamics along these lines, which appears to share quite some similarities with Jellinek’s ideas [Jel88a; Jel88b], was recently proposed in Ref. [Lai03]. It is not the purpose of this work to give a full account of the specific physical and numerical advantages and disadvantages of all these different thermostating schemes. Here we may refer, e.g., to Refs. [Jel88a; Hoo96a; Pos97; Hoo97; Bra00a], which partly review and criticize these more recent methods. Motivated by Jellinek, Branka and Wojciechowski [Bra00b] particularly endeavored to systematically explore the numerical practicability and efficiency of different classes of thermostats. They arrived at conclusions essentially ruling out many of the existing solutions. Further, more physical constraints contradicting some of the methods mentioned above will be briefly summarized in the following section. In our view, one may even go so far as to question the philosophy underlying most of these novel constructions in that a ‘good’ thermostat should, by all means, enforce a canonical velocity distribution even onto regular dynamical systems such as the harmonic oscillator. The other way around, one may argue that the harmonic oscillator is rather unsuited to being properly thermalized at all, just because of its inherent regularity. Correspondingly, a successful transformation of this dynamics onto a canonical velocity distribution indicates that the original dynamics of this system is profoundly destroyed by the action of the thermal reservoir. This raises the question of to what extent the resulting dynamics merely represents the action of the thermostat and whether anything is left at all of the original characteristics of the previously regular subsystem. This discussion demonstrates that the world of deterministic and timereversible thermostats does not consist of Gaussian and Nos´e-Hoover thermostats only. Clearly, there exists a microcosmos of variants. This should

May 8, 2007

4:1

226

World Scientific Book - 9in x 6in

The Nos´ e-Hoover thermostat

be taken into account if one attempts to come to general conclusions concerning the second law of thermodynamics and the universality of chaotic and transport properties of NSS based on the analysis of Gaussian and Nos´e-Hoover thermostats only. 12.4.2

Thermal reservoirs in nonequilibrium

In Section 10.3 we have argued that thermal reservoirs should generally be constructed in equilibrium situations. In this case the statistical ensembles are well-defined and the corresponding velocity distribution functions for subsystem and thermal reservoir are known exactly. After their definition in thermal equilibrium one may apply these reservoirs to thermalizing a subsystem under nonequlibrium constraints, by expecting that the respective thermostat still works sufficiently well such that a proper NSS is created. One may now learn something new about the NSS resulting for a particular subsystem by analyzing the corresponding nonequilibrium distribution functions as well as the associated chaos and transport properties, which are generally not known in advance. However, first of all it is not guaranteed that a thermal reservoir which properly acts in thermal equilibrium also functions in nonequilibrium by generating a NSS. Actually, it appears that most of the generalizations of the Nos´e-Hoover thermostat listed above have only been tested in thermal equilibrium so far. Some of them have already been criticized for not working properly under nonequilibrium constraints [Eva98; Bra00a; Bra00b]. Apart from that, only a few studies of nonequilibrium situations are available for these generalized reservoirs [Pos97; Hoo97; Bra00a; Bra00b]. The same applies to the chaotic dynamical properties of subsystems connected to these different thermal reservoirs even in thermal equilibrium, see only Refs. [Kus90; Mar92b; Pos97; Hoo01] for results in some specific cases. Secondly, a thermal reservoir should always remain in thermal equilibrium, even if the subsystem is under nonequilibrium constraints, and it should only control the temperature of the subsystem, or possibly higher moments of it. Consequently, a thermal reservoir should only act on the even moments of the velocity distribution of a subsystem. If a thermal reservoir were defined to also control odd moments of the velocity, its action would go beyond simple temperature control and might change the dynamics of the subsystem profoundly. For example, a thermostat constraining the first moment of a velocity distribution would simply pre-

klages˙book

May 8, 2007

4:1

12.5.

World Scientific Book - 9in x 6in

∗ Summary

klages˙book

227

determine the current.11 In the same vein, a thermal reservoir acting onto position coordinates may pre-determine, or at least profoundly influence, the respective moments of the position coordinates hence affecting transport coefficients like diffusion and higher-order Burnett coefficients. According to this reasoning, one may wish to constrain the action of thermal reservoirs to even velocity moments only, however, this already rules out schemes such as the ones presented in Refs. [Kus90; Bul90b; Bul90a; Ham90; L’H93]. In conclusion, all that one can demand is that a thermal reservoir properly acts in a thermal equilibrium situation by thermalizing a subsystem onto a pre-determined velocity distribution. In nonequilibrium the general requirement should be the existence of a NSS at a certain temperature, whose properties as resulting from the action of this thermal reservoir may then be studied respectively. 12.5

∗

Summary

(1) Starting from the conservation of the number of points in phase space, we have derived the Liouville equation for dissipative dynamical systems. The Hamiltonian version of this Liouville equation, which typically appears in textbooks, is obtained from it under the additional assumption of conservation of phase space volume. (2) By employing this general form of the Liouville equation and by requiring that canonical distributions for the velocities of a subsystem and for a reservoir variable exist, we arrived at a simple differential equation determining a velocity-dependent friction coefficient. This equation defines the action of the Nos´e-Hoover thermostat, supplemented by a respective friction term in the original equations of motion of a subsystem. In complete analogy to the Gaussian thermostat, Nos´e-Hoover thermostated equations of motion are deterministic and time-reversible but non-Hamiltonian. However, in contrast to the Gaussian reservoir the Nos´e-Hoover scheme attempts to transform the subsystem velocities onto a canonical distribution in thermal equilibrium. Hence, it yields a dynamics that is more similar to Langevin’s theory than the one result11 Note, however, that a thermal reservoir may be defined such that it is comoving with the current, in order to thermalize the subsystem onto the proper temperature in the comoving frame; see Refs. [Pos88; Sea98] and also the discussion of a thermostated shear flow in Section 15.5, respectively Refs. [Wag99; Wag00].

May 8, 2007

4:1

228

World Scientific Book - 9in x 6in

The Nos´ e-Hoover thermostat

ing from the Gaussian constant energy approach. Another difference of the Nos´e-Hoover thermostat compared to the Gaussian one is the appearance of an additional control parameter, which may be interpreted as the response time of the thermal reservoir for its interaction with a subsystem. In the case of infinitely slow response the Stokes friction coefficient is recovered, whereas for infinitesimally quick response one approaches the Gaussian constraint. A variation of this control parameter changes the shapes of the corresponding velocity distributions, which may accordingly be composed of superpositions of microcanonical and canonical densities. (3) The Nos´e-Hoover thermostat belongs to the same class of deterministic and time-reversible thermal reservoirs as the Gaussian one as far as an identity between the average rate of phase space contraction and thermodynamic entropy production is concerned. Respectively, there is an analogous relation between transport coefficients and Lyapunov exponents for Nos´e-Hoover dynamics. For both thermal reservoirs there exist generalized Hamiltonian and Lagrangian formulations. For the isokinetic Nos´e-Hoover thermostat we have outlined how to derive the respective equations of motion from a Hamiltonian in generalized variables via non-canonical transformations. Like the Gaussian thermostated driven periodic Lorentz gas, the Nos´eHoover version is as well characterized by a fractal attractor of a similar type. The specific topology of this attractor depends on the value of the reservoir response time. For large enough response times the bifurcations displayed by the fractal attractor are typically very different from the ones observed for the Gaussian reservoir. The same applies to the field-dependent electrical conductivities in the case of Nos´e-Hoover, which still show irregularities on fine scales and no indication of linear response in the numerically accessible regime of the field strength. (4) We have summarized problems with Gaussian and particularly with Nos´e-Hoover thermostats in equilibrium and nonequlibrium situations. For regular dynamical systems such as the harmonic oscillator or a particle moving in a double-well potential it is well-known that the standard Nos´e-Hoover thermostat is not able to thermalize these systems onto canonical velocity distributions in thermal equilibrium. A necessary condition is that the thermostated system must be mixing, respectively chaotic and ergodic. In order to achieve this goal the Nos´eHoover scheme can be generalized yielding equations of motion that are

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

12.5.

∗ Summary

klages˙book

229

more nonlinear than the original method. However, such “stronger” thermostating forces pose the question of to what extent the original dynamics of the subsystem still plays a role at all. Our discussion shows that a great variety of deterministic and time-reversible thermal reservoirs exists. This fact one should take into account with respect to suspected universalities of NSS as concluded from applications of Gaussian and Nos´e-Hoover thermostats only.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

This page intentionally left blank

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Chapter 13

Universalities in Gaussian and Nos´ e-Hoover dynamics?

The approach to nonequilibrium transport outlined in the previous two chapters yields NSS on the basis of non-Hamiltonian equations of motion. This dynamics results from applying deterministic and time-reversible thermal reservoirs, which are designed to efficiently control the temperature of a subsystem under nonequilibrium conditions. The Gaussian and the Nos´eHoover thermostat constructed before provide two famous examples of such mechanisms. Coupled to the periodic Lorentz gas driven by an external electric field, the resulting NSS have been analyzed with respect to their chaos and transport properties. This chapter provides a shortlist of what we consider to be fundamental properties of Gaussian and Nos´e-Hoover dynamics. We put these results into a broader perspective and critically discuss whether these properties might be universal for NSS altogether, irrespective of the type of thermostat used. Our assessment sets the scene for the following two chapters. 13.1

Non-Hamiltonian nonequilibrium steady states

Constructing NSS requires one either to start either from a Hamiltonian that models a thermal reservoir of infinitely many degrees of freedom or to use non-Hamiltonian equations of motion, see Section 1.2. As an example, in Section 10.2 we have derived the stochastic Langevin equation from a Hamiltonian defining a thermal reservoir that consists of infinitely many harmonic oscillators. The resulting equation turned out to be dissipative, non-deterministic and non time-reversible thus providing a well-known example of non-Hamiltonian dynamics. This dynamics results from the fact that the equations of motion of the infinitely many reservoir degrees of freedom were eliminated in the course of the derivation, supplemented by some 231

klages˙book

May 8, 2007

4:1

232

World Scientific Book - 9in x 6in

klages˙book

Universalities in Gaussian and Nos´ e-Hoover dynamics?

further approximations. Section 10.3 demonstrated that projecting out spurious reservoir degrees of freedom typically1 yields a non-uniform probability density for the thermostated subsystem while the combined distribution of subsystem plus thermal reservoir is still uniform, respectively microcanonical. The non-uniformity of the projected equilibrium density may be taken as an indication of some phase space contraction in the associated variables, which is not present in the original phase space volume preserving Hamiltonian dynamics of the full system. Indeed, in case of the Langevin equation and for Gaussian and Nos´e-Hoover dynamics the equations of motion of the thermostated subsystem are always non-Hamiltonian. We thus argue that the non-Hamiltonian character results from conveniently simplifying the equations of motion of subsystem plus reservoir by considering the relevant degrees of freedom only while neglecting others. Hence, such a non-Hamiltonian formulation is very well compatible with a Hamiltonian description of the complete combination of subsystem plus reservoir. The use of non-Hamiltonian deterministic dynamics for modeling NSS is often criticized [Eyi92; Coh92; Lar03], therefore we illustrate our arguments by means of a very simple example. Let us consider the two-dimensional baker map acting onto the unit square, cf. Chapter 4, (2x, (y + 1)/2) , 0 ≤ x < 12 . (13.1) B(x, y) = (2x − 1, y/2) , 12 ≤ x ≤ 1 This map squeezes the unit square along the vertical direction while stretching it horizontally such that the area is preserved. The resulting rectangle is cut in the middle, and both parts are put on top of each other yielding the dynamics depicted in Fig. 13.1. The corresponding equations of motion read (xn+1 , yn+1 ) = B(xn , yn ) ,

(13.2)

where n ∈ N0 holds for the discrete time. It appears [Gas98a; Gas03] that the baker map was first defined and studied by Seidel in 1933 [Sei33] who, in turn, gives informal credit to Birkhoff for its invention. Later on this map was considered by Hopf [Hop37]. It became very popular starting with the book by Arnold 1 A counterexample was provided by subsystem and reservoir living altogether in three dimensions by projecting out two of them, see Fig. 10.2.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

233

13.1. Non-Hamiltonian nonequilibrium steady states

y

y B x

x

Fig. 13.1 Sketch of the baker map defined by Eq. (13.1): a unit square is stretched horizontally and squeezed vertically by preserving the unit area. The resulting rectangle is cut in the middle, and both parts are put on top of each other forming again a unit square. The two diagonal lines in the left part of the figure represent the one-dimensional Bernoulli shift acting onto the x variable only, which determines the dynamics of the baker map along the horizontal axis.

and Avez [Arn68], who coined the name.2 The baker map is one of the simplest chaotic dynamical systems, serving as a standard model in many textbooks on nonlinear dynamics [Sch89; Ott93; Gas98a; Dor99; T´el06]. A corresponding version with nonequilibrium transport is the spatially extended multibaker map introduced by Gaspard [Gas92a], which we already studied in Chapter 4 [Gas93; Tas95; T´el96; Vol97; Gas97b; Bre98; Vol98; Gas98c; Gil99a; Gil99b; Tas99; Tas00; Vol00; Ron00b; Gil00b; Gil00a; T´el00; Gil01; Vol02; Woj02; Gar02; Gil04; Gas03; Woj03; Vol04]. Dissipative multbaker maps have been constructed and analyzed as well mimicking the action of Gaussian and Nos´e-Hoover thermostats [T´el96; Vol97; Bre98; Vol98; Gil99a; Gil99b; Vol00; Ron00b; Gil00a; Vol04 ]. However, here we restrict ourselves to Gaussian and Nos´e-Hoover dynamics in terms of differential equations; for reviews of time-discrete versions see particularly Refs. [T´el00; Vol02; Dor99]. That the baker map Eq. (13.1) is area-preserving is easily verified by computing its Jacobian determinant. The map is furthermore invertible and shares a symmetry property that is usually referred to as reversibility in time-discrete maps, cf. Eq. (4.26) in Section 4.2 for the definition. 2 Both Seidel and Hopf used the baker map as an example for a system that is metrically transitive, or mixing, respectively. Arnold and Avez proved that this map is Bernoulli exhibiting the property of a K-system with positive Kolmogorov-Sinai entropy. Seidel originally discussed a baker with stretching/squeezing factors of 10, respectively 1/10.

May 8, 2007

4:1

234

World Scientific Book - 9in x 6in

klages˙book

Universalities in Gaussian and Nos´ e-Hoover dynamics?

These properties are analogous to respective features in time-continuous Hamiltonian dynamical systems.3 It is not our purpose to give a detailed introduction to dynamical systems properties of the baker map.4 We just wish to convey that it is a widely studied standard model mimicking, at least to some extent, a Hamiltonian chaotic dynamical system on the level of time-discrete equations of motion. In order to illustrate our above discussion about the origin of nonHamiltonian dynamics, we now project out one variable of the baker map Eq. (13.1). A detailed calculation starts from a time-discrete Liouville equation, the so-called Frobenius-Perron equation for the baker map [Ott93], by integrating over the y-variable [Dor99]. In Eq. (13.1) x and y are nicely decoupled with respect to the direction of the projection. One may thus immediately conclude that the Frobenius-Perron equation determining the evolution of the probability density along the x-axis only must be governed by the action of the one-dimensional Bernoulli shift xn+1 = Bx (xn ) = 2xn

mod 1 ,

(13.3)

which is included in Fig. 13.1. Its Jacobian determinant is equal to two, hence this map is uniformly expanding and does not preserve the phase space volume. The baker map thus provides a simple example of a system whose complete two-dimensional dynamics is area-preserving while, under projection, the associated one-dimensional equations of motion are dissipative. Curiously, the equilibrium density of the Bernoulli shift is still uniform as for the original baker map, whereas in Section 10.3 the probability densities obtained from projecting out reservoir degrees of freedom were typically non-uniform. In order to better mimick these results one may rotate the unit square of the baker map by 45 degrees [Hoo98c]. Now the density projected onto the horizontal line will be non-uniform while the full baker map is still area-preserving. Unfortunately, in this case the stable manifold of the map is not identical anymore with the direction of the projection,5 hence it is not straightforward to extract equations of motion governing the dynamics along the x-axis only. 3 However, note that the baker map is not symplectic, in contrast to, for example, the standard map [Ott93; Mei92]. 4 Apart from being mixing and a K-system implying chaotic behavior the baker map is also hyperbolic [Dor99]. Strictly speaking it is not Anosov because it is not differentiable at the point of discontinuity [Gas98a]. 5 That is, we are not projecting anymore onto the unstable manifold only, which is usually the appropriate choice for performing a projection.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

13.2. Phase space contraction and entropy production

klages˙book

235

This simple example demonstrates again that non-Hamiltonian dynamics goes very well together with Hamiltonian equations of motion for a subsystem plus thermal reservoir. In such a case non-Hamiltonian dynamics just results from conveniently projecting out superfluous (reservoir) degrees of freedom, such as the passive y-variable in the baker map [Ron02a; Eva02a]. In our view a non-Hamiltonian description thus emerges naturally in equations designed to model NSS, and in this respect Gaussian and Nos´e-Hoover thermostats are not particularly artificial or unusual. On the other hand, one should keep in mind that Gaussian and Nos´eHoover reservoirs have not been derived starting from a purely Hamiltonian dynamics, cf. the derivation of the Langevin equation in Section 10.2. Moreover, in contrast to the projected baker dynamics and to the Langevin equation the former two types of dynamics are even time-reversible. The generalized Hamiltonian approach for Gauss and Nos´e-Hoover described in Section 12.3.2 represents a formal analogy only and not any Hamiltonian derivation. We thus emphasize again that Gaussian and Nos´e-Hoover dynamics yield only a heuristic modeling of NSS rather than providing a theory starting from first principles. Hence, one cannot rule out in advance that the properties of dynamical systems thermostated that way may depend on specific modeling features of these thermal reservoirs. This should be taken into account when searching for universal properties of NSS along the lines of this approach.

13.2

Phase space contraction and entropy production

The periodic Lorentz gas driven by an external electric field yielded NSS after connecting this model to either the Gaussian or to the Nos´e-Hoover thermostat. Both types of thermostated systems were characterized by an identity between the absolute value of the average rate of phase space contraction and the thermodynamic entropy production in terms of Joule’s heat, see Sections 11.2.1 and 12.3.1. This identity represents a fundamental property of ordinary Gaussian and Nos´e-Hoover dynamics and furnishes a crucial link between dynamical systems quantities and transport properties. Interestingly, even the stochastic Langevin equation Eq. (10.1) supplemented by an external electric field exhibits this identity, as one can easily verify. However, in contrast to Gaussian and to Nos´e-Hoover dynamics this equation is less amenable to an analysis in terms of dynamical systems theory because of the stochastic forces involved.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

236

klages˙book

Universalities in Gaussian and Nos´ e-Hoover dynamics?

We now show that this identity is linked to a more general relationship between phase space contraction and entropy production, which holds irrespective of the specific type of dynamical system considered. In our presentation we follow Ref. [And85], for other derivations see particularly Refs. [Eva85a; Eva90b; Gas97b; Dor99; Gal99; Rue99b; Det00a; Eva02a]. Let us start from the general dynamical system x˙ = F(x)

,

(13.4)

where x and F denote vectors in a k-dimensional phase space Γ ⊂ Rk , k ∈ N. For convenience here we shall not distinguish between position and velocity components. One may now inquire about the average rate of entropy production in this system by using Gibbs definition of entropy [Pen79; Eva90b; Rue99b; Gas98a; Det00a] Z (13.5) SG := − dΓ ρ ln ρ . Γ

Here ρ = ρ(x1 , x2 , . . . , xk ) denotes the distribution function of the complete dynamical system Eq. (13.4) determined by the Liouville equation Eq. (12.2) and dΓ := dx1 dx2 . . . dxk represents a volume element in the phase space Γ. In order to calculate the Gibbs entropy production we differentiate Eq. (13.5) according to Leibnitz’ rule yielding Z Z ∂ρ X dSG = − dΓ (1 + ln ρ) − dΓ0 ρ ln ρ Fi |∂Γi (13.6) dt ∂t 0 Γ Γ i with dΓ0 := dx1 . . . dxi−1 dxi+1 . . . dxk and Fi defined at the boundary ∂Γi . The derivative ∂ρ/∂t can be substituted by using the Liouville equation Eq. (12.2) leading to Z Z dSG = dΓ ρ ∇ · F + dΓ ρ ln ρ ∇ · F + dt Γ Γ Z XZ + dΓ (1 + ln ρ) F · ∇ρ − dΓ0 ρ ln ρ Fi |∂Γi . (13.7) Γ

i

Γ0

Performing integration by parts, the second term on the right hand side of the above equation precisely cancels with the last two terms of the same equation. The final result thus reads Z dSG = dΓ ρ∇ · F , (13.8) dt Γ that is, the rate of Gibbs entropy production is identical to the average rate of phase space contraction defined by Eq. (11.5). This important identity was noted by Gerlich [Ger73] and was subsequently discussed in various settings by other authors [Dob76a; Dob76b; Ste79;

May 8, 2007

4:1

World Scientific Book - 9in x 6in

13.2. Phase space contraction and entropy production

klages˙book

237

And82]; for a very lucid statement of it see Andrey [And85]. In the context of Gaussian and Nos´e-Hoover dynamics the importance of the relationship Eq. (13.8) was particularly emphasized in Refs. [Eva85a; Bre96; Hol87; Hoo87; Pos88; Rue96; Rue97b; Gal99; Rue99b; Hoo99; Dor99; Rue03; Rue04]. As we have mentioned at the beginning of Chapter 1, it appears that there is no generally accepted definition of a thermodynamic entropy in nonequilibrium situations. Some authors consider the Gibbs entropy Eq. (13.5) to be a suitable candidate for such a nonequilibrium entropy [Eva90b; Gal99; Rue99b; Hoo99]. In our view, such an approach involves a number of problems. First of all, Eq. (13.8) tells us that for phase space volume preserving dynamics the Gibbs entropy production is strictly zero. This fact is problematic [Nic98; Det00a], for example, in the case of diffusion in Hamiltonian dynamical systems under concentration gradients imposed by flux boundary conditions [Gas97b; Gas98a], see also Ref. [Eck99a] for a Hamiltonian modeling of a heat flow under temperature gradients. In both nonequilibrium situations there is a well-defined irreversible entropy production while the Gibbs entropy production is strictly zero, since there is no non-Hamiltonian thermostating. Secondly, dissipative dynamical systems usually exhibit a negative average rate of phase space contraction. Specific examples are provided by the Gaussian and the Nos´e-Hoover thermostated driven Lorentz gas, see Eqs. (11.6) and (12.14). According to Eq. (13.8) the Gibbs entropy production is also negative reflecting a contraction of the system onto a subset in phase space such as a fractal attractor. The decrease of the Gibbs entropy thus measures the fact that the dynamical system is getting “more ordered” in phase space. However, a negative thermodynamic entropy production in NSS appears to be in contradiction with the second law of thermodynamics. In our earlier derivations of Sections 11.2.1 and 12.3.1, which specifically concerned the Gausssian and the Nos´e-Hoover thermostated Lorentz gas, we circumvented this “sign problem”, since we did not start from the Gibbs entropy production. Instead, we directly computed the average rates of phase space contraction by realizing afterwards that the absolute values yielded the thermodynamic entropy production in the form of Joule’s heat. This enabled us to “invert the sign” and to establish the identity without entering into a discussion concerning a negative Gibbs entropy production. A more physical interpretation of the sign reversal may be obtained as

May 8, 2007

4:1

238

World Scientific Book - 9in x 6in

Universalities in Gaussian and Nos´ e-Hoover dynamics?

follows [Che93a; Rue96; Dor99; Hoo99]: Let us consider a balance equation for the total entropy production of a subsystem under nonequilibrium conditions, in analogy to Eq. (12.1). The source term stands for the entropy production within the subsystem, whereas the flow term holds for the average heat transfer from the subsystem to the thermal reservoir. The reservoir fully absorbs the entropy produced by the subsystem such that in a NSS the average entropy production of subsystem plus thermal reservoir is zero. Hence, from the side of the thermal reservoir the entropy flux must be positive causing a positive entropy production in the reservoir. This entropy production may be identified as the relevant one in the sense of irreversible thermodynamics thus yielding the sign reversal. However, in detail this explanation is debatable [Nic96; Gas97b; Gas98a; Nic98; Gil99a; Dae99], because we did not identify a positive entropy production within the subsystem itself. Refined approaches resolve the sign problem by using methods of coarse graining for the phase space densities as they enter into the definition of the Gibbs entropy Eq. (13.5) [Bre96; Nic96; Vol97; Gas97b; Vol98; Bre98; Nic98; Gil99a; Tas99; Dae99; Gas98a; Gil00b; Gil00a; Tas00; T´el00; Dor02; Vol02; Mae03; Vol04; M´ at04b]. Such a coarse grained formulation may be motivated by a second basic deficiency of the Gibbs entropy [Nic96; Gas97b; Dae99; Gil99a; Gas98a]: As we have shown in Sections 11.2.3 and 12.3.3, in deterministically thermostated systems the phase space density typically contracts onto fractal attractors. However, the fractal structure implies that the probability densities characterizing the associated NSS must be singular and non-differentiable. Consequently, they are not well-defined mathematical objects anymore and the integration in Eq. (13.5) is also not well-defined. Nevertheless, the corresponding probability measures may still be well-defined. Typically, these are SRB measures which, very loosely speaking, are characterized as being smooth along unstable but fractal along stable manifolds [Gas98a; Dor99; Gal99; Rue99b; You02; Gal03b]. In this case the integral should be replaced by a Lebesgue-Stieltjes integral over the invariant measure [Dor99; Gas98a]. As a consequence of the formation of these fractal structures it was proposed by many authors to use coarse-grained Gibbs entropies in order to arrive at a suitable concept of a nonequilibrium entropy, see the long list of references cited above. Up to now these methods have mostly been worked out for time-discrete dynamical systems such as multibakermaps for which different types of coarse graining have been explored. Con-

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

13.2. Phase space contraction and entropy production

klages˙book

239

cerning further details we may refer, e.g., to Refs. [Gil99a; Gas98a; T´el00; Vol02] providing recent summaries of this issue. In the light of this criticism it may appear surprising that for the Gaussian and for the Nos´e-Hoover thermostated driven Lorentz gas the ordinary Gibbs entropy production can nevertheless be fully identified with the thermodynamic entropy production in terms of Joule’s heat subject to a sign reversal. A related fact, which further contributes to this surprise, is that Gibbs’ formulation assesses the entropy of a system in terms of all phase space variables and with respect to their complete deterministic dynamics in time. This is at variance with coarse-grained Gibbs entropies or Boltzmann entropies. Here there is already a loss of information, because these quantities do not employ the complete phase space densities or respective probability measures as determined by the Liouville equation. In a similar vein, a heat flux from a subsystem to a thermal reservoir will usually involve only a specific fraction of all phase space variables. From the point of view of the Clausius definition of entropy production, see Section 11.2.1, the Gibbs entropy may thus measure spurious contributions in comparison to a thermodynamic entropy production by assessing phase space coordinates that are not involved in any heat transfer to a thermal reservoir. All these physical arguments support the use of a coarse grained entropy that only assesses the physically relevant contributions to a thermodynamic entropy production. We believe that such a formulation of a nonequilibrium entropy should be compatible with the one of Clausius in terms of a heat transfer. From that point of view there is no reason why phase space contraction, or the respective Gibbs entropy production, should always be identical to the thermodynamic one. One may suspect that such an identity may rather be characteristic of the specific family of thermostats investigated so far. This hypothesis will be verified by presenting generic families of counterexamples, where the identity between phase space contraction and thermodynamic entropy production, respectively between Gibbs and thermodynamic entropy production, is not fulfilled. In order to qualify as counterexamples, alternative models of thermal reservoirs must share some basic physical properties with Gaussian and Nos´e-Hoover thermostats, such as being dissipative, deterministic and time-reversible. Furthermore, there should be an adequate physical interpretation of the thermostating mechanism. Such counterexamples will be introduced and analyzed in the following two chapters.

May 8, 2007

4:1

240

13.3

World Scientific Book - 9in x 6in

klages˙book

Universalities in Gaussian and Nos´ e-Hoover dynamics?

Transport coefficients and dynamical systems quantities

If there exists an identity between the average phase space contraction rate and thermodynamic entropy production, it is straightforward to link transport coefficients with dynamical systems quantities. Interestingly, for the field-dependent conductivity σ(ε) of both the Gaussian and the Nos´eHoover thermostated driven periodic Lorentz gas one obtains the same equation, σ(ε) = −

N T X λi (ε) , ε2 i=1

(13.9)

cf. Eqs. (11.11) and (12.17), where N denotes the total number of fielddependent Lyapunov exponents λi (ε) and T stands for the temperature. The existence of such a simple functional relationship between chaos quantities and transport coefficients is typical for Gaussian and Nos´e-Hoover thermostated dynamical systems, see Sections 11.2.2 and 12.3.1 for further details. Surprisingly, by using the escape rate approach to chaotic transport outlined in Section 1.1, very similar formulas were derived for the very different class of open Hamiltonian dynamical systems, where there is an escape of phase space points due to absorbing boundaries. This second fundamental approach to NSS does not require any modeling of thermal reservoirs, hence by default it does not involve any phase space contraction. We will now provide a brief summary concerning formulas linking transport coefficients to dynamical systems quantities as they emerge from both the escape rate and the thermostated systems approach. The resulting different but related formulas will be compared in detail, and possible crosslinks between them will be critically assessed. For sake of simplicity we mainly restrict ourselves again to the periodic Lorentz gas. We start with the driven Lorentz gas under application of a Gaussian thermostat for which N = 2 in Eq. (13.9). In order to connect to the escape rate formalism we reformulate this equation as follows [Gas98a; Gil01; Rue96]: For the Gaussian thermostated Lorentz gas the sum of Lyapunov exponents on the right hand side consists only of a positive one, λ+ (ε), and a negative one, λ− (ε), see Eq. (11.11). Both Lyapunov exponents are defined with respect to the invariant measure of the dynamical system, which is concentrated on the field-dependent fractal attractor Aε . We make this dependence explicit by writing λ± (Aε ) = λ± (ε).

May 8, 2007

4:1

World Scientific Book - 9in x 6in

13.3. Transport coefficients and dynamical systems quantities

klages˙book

241

As was proven in Refs. [Che93a; Che93b], under certain conditions there holds the Pesin identity [Eck85; Ott93; Bec93; Gas98a; Dor99] between the Kolmogorov-Sinai entropy hKS (Aε ) and the positive Lyapunov exponent on the attractor, hKS (Aε ) = λ+ (Aε ), for the Gaussian thermostated driven Lorentz gas. Having this in mind we rewrite the right hand side of Eq. (13.9) as −

2 X i=1

λi (Aε ) = −λ+ (Aε ) − λ− (Aε ) = −hKS (Aε ) − λ− (Aε )

= |λ− (Aε )| − hKS (Aε )

(13.10)

The escape rate formalism is particularly suited to describe diffusion processes, see Section 1.1. Accordingly, since the Einstein relation Eq. (11.13) holds for the Gaussian thermostated driven Lorentz gas for small enough fields [Che93a; Che93b], the electrical conductivity σ(ε) on the left hand side of Eq. (13.9) may be replaced by the diffusion coefficient D, D = T σ(ε) (ε → 0). This yields the final result [Gas98a; Gil01; Rue96; Bre96] 2 T (|λ− (Aε )| − hKS (Aε )) . (13.11) D = lim ε→0 ε A corresponding formula from the escape rate approach to chaotic diffusion is obtained as follows [Gas90; Gas92c; Gas93; Gas95b; Dor95; Gas95c; Kla96b; Kla99a; Gas98a; Dor99]: We consider a slab of the periodic Lorentz gas with reflecting or periodic boundaries, say, in the vertical direction while in the horizontal direction we use absorbing boundaries a distance L apart from each other. We do not apply any external field but just look at the diffusion of an initial ensemble of point particles along the x-axis. Choosing the density of scatterers such that the periodic Lorentz gas is normal diffusive, see Section 10.4, we expect an exponential decrease of the number N of particles in time t according to N (t) = N (0) exp(−γesc t) ,

(13.12)

where N (0) stands for the initial number of particles at time zero. Solving the one-dimensional diffusion equation with absorbing boundary conditions yields 2 L γesc , (13.13) D = lim L→∞ π

May 8, 2007

4:1

242

World Scientific Book - 9in x 6in

klages˙book

Universalities in Gaussian and Nos´ e-Hoover dynamics?

that is, the diffusion coefficient is obtained in terms of the escape rate, which strictly speaking depends on L, γesc = γesc (L). Let us now assume that a generalization of Pesin’s identity to open systems, the escape rate formula, holds for the open periodic Lorentz gas,6 γesc = λ+ (RL ) − hKS (RL ) .

(13.14)

Here γesc = γesc (RL ) denotes the escape rate of particles with respect to the fractal repeller RL . Combining this equation with Eq. (13.13) leads to the fundamental result 2 L (λ+ (RL ) − hKS (RL )) . (13.15) D = lim L→∞ π Obviously, the functional forms of Eqs. (13.11) and (13.15) are just the same. However, we emphasize again that both the types of dynamical systems considered and the approaches by which these equations have been derived are very different. On the other hand, for both derivations the key was to identify a quantity providing a link between chaos and transport properties. For thermostated systems this role is played by the average phase space contraction rate κp defined in Eq. (11.5). In the case of open Hamiltonian systems the escape rate γesc serves the same purpose. This analogy was first observed by Breymann, T´el and Vollmer [Bre96] and by Ruelle [Rue96]. These authors suggested defining a Gibbs entropy for open dynamical systems, in parallel to the Gibbs entropy of closed but thermostated, phase space contracting dynamical systems, related to the concept of conditionally invariant measures. Here the probability measures of particles remaining in the system, respectively the corresponding probability densities, are renormalized in time in order to make up for the loss of absorbed particles. Along these lines the Gibbs entropy production for open systems can be identified with the escape rate, dSG /dt = γesc . On the other hand, for closed systems we have shown in the previous section that the Gibbs entropy production is equal to the average rate of phase space contraction, see Eq. (13.8). Hence, if the rates of Gibbs entropy production for closed and open systems were the same both Eqs. (13.11) and (13.15), which apply to very different dynamical systems, would yield the same transport coefficients in terms of dynamical systems quantities. This was indeed shown to be the case for some simple model systems, that is, (multi)baker maps, under periodic, 6 So

far this identity is only proven for Anosov diffeomorphisms, see, e.g., Refs. [Eck85; Bec93; Dor99; Gas98a] for outlines concerning this generalization of Pesin’s identity and further references therein.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

13.3. Transport coefficients and dynamical systems quantities

klages˙book

243

absorbing and flux boundary conditions thus suggesting an equivalence of these very different nonequilibrium ensembles [Mor96b; Vol97; Vol98; Gil99a; Gil99b; Gil00a; T´el00; Vol02; Vol04]. T´el et al. have furthermore considered the situation of a hybrid system that is both open and dissipative [T´el96; T´el00; Vol02]. By solving the corresponding Fokker-Planck equation with absorbing boundaries and employing again the Pesin identity for open systems they arrived at the generalized formula (see also Ref. [Gas98a]) π 2 σ 2 ε 2 X λi (RL ) − hKS (RL ) (L → ∞, ε → 0) . (13.16) + D = L 4D λi >0

Carrying out the limit of ε → 0 eliminates the second term on the left hand side yielding precisely Eq. (13.15) of the escape rate approach if applied to the Lorentz gas. On the other hand, performing L → ∞ eliminates the first term on the left hand side. The conductivity may then be replaced again by Einstein’s formula Eq. (11.13). Taking into account that for the Gaussian thermostat the temperature T should be replaced by T = T /2, see Section 11.2.1, and assuming that in the limit of ε → 0 the spectrum P P of Lyapunov exponents exhibits conjugate pairing, λi >0 λi = λi <0 |λi |, one recovers the result for the thermostated Lorentz gas Eq. (13.11). We finally outline a third basic approach linking transport coefficients to dynamical systems quantities as recently formulated by Gaspard et al. [Gil01; Gas01; Cla02]. Here particularly diffusion in closed volumepreserving dynamical systems has been considered. Starting from the diffusion equation with periodic boundary conditions the solution for the diffusion coefficient reads 2 L γdec . (13.17) D = lim L→∞ 2π This result is in formal analogy to Eq. (13.13), however, here γdec denotes the decay rate by which an equilibrium state is approached. By solving the Liouville equation of the respective dynamical system γdec can be related to the second largest eigenvalue of the Liouville operator associated with the corresponding hydrodynamic mode of diffusion of the dynamical system [Kla96b; Kla99a]. The crucial observation is now that this eigenvalue is linked to the Hausdorff dimension of the respective mode. Thus, the diffusion coefficient in Eq. (13.17) can be written as a function of the fractal dimension of the second largest eigenmode of the Liouville equation [Gil01; Gas01]. For systems with two degrees of freedom, which are periodically

May 8, 2007

4:1

244

World Scientific Book - 9in x 6in

klages˙book

Universalities in Gaussian and Nos´ e-Hoover dynamics?

continued in one direction over a length L, one gets 2 L D = lim λ+ (dH (2π/L) − 1) L→∞ 2π

,

(13.18)

where the Hausdorff dimension dH is a function of the length of the system, or respectively of the wavenumber k = 2π/L of the corresponding hydrodynamic mode. λ+ denotes again the positive Lyapunov exponent of the system. This approach has been worked out in detail for a multibaker map [Gil01] as well as for diffusive and reactive-diffusive billiards of Lorentz gas type [Gas01; Cla02]. In order to compare Eq. (13.18) with the previous two formulas Eqs. (13.11) and (13.15) one may employ the Kaplan-Yorke formula, respectively Young’s formula, which in the case of the two-dimensional periodic Lorentz gas with a constrained kinetic energy reads [Eck85; Ott93; Gil01; Che93a]7 dI = 2 + hKS /|λ− | .

(13.19)

This formula links the information dimension dI of a fractal set to the Kolmogorov-Sinai entropy hKS and to the negative Lyapunov exponent λ− of the corresponding dynamical system. By using this equation Eq. (13.11) can be rewritten as 2 T λ+ (Aε ) (3 − dI (Aε )) . (13.20) D = lim ε→0 ε It appears that this formula was first derived in Ref. [Eva00]. Analogously, Eq. (13.15) reads 2 L D = lim λ+ (RL ) (3 − dI (RL )) . (13.21) L→∞ π In both cases the information dimension dI is furthermore identical to the Hausdorff dimension dH of the respective fractal set and may be replaced respectively [Che93a; Che93b; Gas95b]. In conclusion, all three equations Eqs. (13.18), (13.20) and (13.21) relate the diffusion coefficient to the largest positive Lyapunov exponent of the corresponding dynamical system times a term containing the information dimension of the associated fractal structure, which in Eq. 7 We

remark that so far this formula has only been proven for two-dimensional Anosov diffeomorphisms with an ergodic measure on compact manifolds, see Refs. [Eck85; Ott93] for further literature.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

13.3. Transport coefficients and dynamical systems quantities

klages˙book

245

(13.18) is a fractal hydrodynamic mode, in Eq. (13.20) a fractal attractor and in Eq. (13.21) a fractal repeller.8 Hence, on this level there is quite a formal analogy even between all three approaches. One may thus indeed wonder whether these formulas form a backbone of nonequilibrium transport in terms of dynamical systems theory [Bre96; T´el96; Rue96]. Additionally, there exist very interesting, simple formulas by which transport coefficients can be calculated in terms of periodic orbits, see, e.g., Ref. [Van92] for the electrical conductivity and Refs. [Cvi95; Cvi92; Gas98a; Cvi07] for the diffusion coefficient. However, these formulas are conceptually rather different from the ones discussed above. Therefore we do not review them here but refer instead the literature, see, e.g., Ref. [Cvi07] for an overview. Despite their striking formal analogy one may not overlook that all three equations concern very different physical settings [Coh92; Det00a; Gil01]: Eq. (13.21) applies to diffusion in open Hamiltonian dynamical systems without external fields, where the link between chaos and transport is formed by the rate assessing the escape from the fractal repeller. Eq. (13.18) concerns diffusion in closed Hamiltonian dynamical systems, again without using external fields, however, here there is no repeller. Instead, the fractality of the hydrodynamic mode of diffusion is assessed. In contrast to these two relations, Eq. (13.20) originally started from the current generated in a non-Hamiltonian dynamical system under application of external fields in combination with a specific thermostat. Here the link between chaos and transport is provided by the average phase space contraction rate onto the fractal attractor. Particularly in the case of Eqs. (13.20) and (13.21) the physical situations involved remain inherently different. That both equations can nevertheless be derived from a ‘master formula’ is nicely demonstrated by the hybrid equation Eq. (13.16). However, the left hand side of this equation represents obviously just the sum of the left hand sides of Eqs. (13.11) and (13.15) supplemented by the prefactors from the respective right hand sides. If we add the prefactor-free right hand sides of these two equations as well we obtain, by using the original formulation in the first line of Eq. (13.10) instead of the right hand side of Eq. (13.11), −λ+ (Aε ) − λ− (Aε ) + λ+ (RL ) − hKS (RL ) (L → ∞, ε → ∞). By assuming that, in the above limits, λ± (Aε ) = λ± (RL ) and that, as before, conjugate pairing holds, λ+ = −λ− , we recover the right hand side of Eq. (13.16) for 8 In

order to make this analogy even closer one may further introduce partial codimensions for the fractal structures in Eqs. (13.20) and (13.21), however, this does not appear to be possible for the hydrodynamic mode in Eq. (13.18) [Gil01].

May 8, 2007

4:1

246

World Scientific Book - 9in x 6in

Universalities in Gaussian and Nos´ e-Hoover dynamics?

the periodic Lorentz gas. Hence, one may argue that Eq. (13.16) represents an additive combination of both formulas. That a more intricate, common root of these two formulas going beyond this equation exists, possibly even combining these two relations with the third one in terms of hydrodynamic modes, appears to be rather unlikely. Still, all these equations are of the same type. We may furthermore emphasize again that the validity of Eq. (13.11), respectively Eq. (13.20), stands and falls with the existence of the identity between phase space contraction and thermodynamic entropy production. In the previous section we have already cast doubt on the general validity of this identity for dissipative systems. Indeed, the two key quantities linking thermodynamics to chaotic dynamics discussed above measure quite different physical processes: The escape rate merely assesses the absorption of the number of phase space points, or particles, at some boundary, whereas the phase space contraction rate is defined with respect to the full details of the dynamical system in all variables. As we have outlined towards the end of the previous section it is thus conceivable that the phase space contraction rate contains spurious information as far as thermodynamic entropy production and transport processes are concerned. Indeed, a number of models exemplifying problems of this type will be discussed in the remaining chapters. In contrast to the thermostated systems approach the escape rate theory, as has been presented by Gaspard and coworkers [Gas98a], is generically an equilibrium approach based on diffusion in chaotic dynamical systems. Within this framework is not clear how to deal with diffusion in non-chaotic systems [Det99b; Gra99; Det00b; Cec03; Alo02; Jep06; Sch06; San06]. Applying this approach to electrical conduction, above we simply presupposed the – for a driven Lorentz gas non-trivial [Che93a; Che93b] – existence of an Einstein formula. For more general nonequilibrium situations one may express escape rates in terms of Helfand moments [Dor95; Gas98a]. However, then the validity relies on Helfand’s theory for calculating nonequilibrium properties by evaluating equilibrium properties. From that point of view it might be interesting to apply the Helfand escape rate formalism to electrical conduction in the driven periodic Lorentz gas, in order to check whether the results discussed above can be recovered. Respective results for the case of shear viscosity have been reported recently [Vis03a; Vis03b], but not with respect to variations of the shear rate. The application of the third approach, in terms of the fractality of hydrodynamic modes, to nonequilibrium situations forms another important open

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

13.4. Fractal attractors for nonequilibrium steady states

klages˙book

247

question. Concerning practical applications for the calculation of transport coefficients it seems that these three formulas do not provide more efficient computational schemes than, say, Einstein formulas or Taylor-Green-Kubo relations. As far as we can tell, in most cases dynamical systems quantities such as Lyapunov exponents and Kolmogorov-Sinai entropies are more difficult to compute than statistical averages defined within the framework of common nonequilibrium statistical mechanics only. With respect to a simplification of the Lyapunov sum rule Eq. (13.9) for higher-dimensional dynamical systems we may recall that the conjugate pairing rule of Lyapunov exponents is not universal for NSS in dissipative dynamical systems. Boundary thermostats and systems under electric and magnetic fields provide counterexamples, see Section 11.2.2 for more details. Irrespective of these rather practical concerns, we emphasize that the existence of these three formulas linking transport coefficients to dynamical systems quantities provides a fundamental result from a physical and dynamical systems point of view. The formal similarity of these different equations is another remarkable fact.

13.4

Fractal attractors for nonequilibrium steady states

Another crucial property of Gaussian and Nos´e-Hoover thermostated systems is the existence of fractal attractors underlying NSS. These objects form a fundamental link between the microscopic non-Hamiltonian equations of motion that are, for this type of system, still deterministic and timereversible while irreversible transport is exhibited on macroscopic scales. Quantitatively, the existence of these attractors manifests itself in the average phase space contraction rate serving as a link between chaos and transport in dissipative systems, see our discussion in the previous sections. In Sections 11.2 and 12.3 we have studied the Gaussian and the Nos´eHoover thermostated driven periodic Lorentz gas as paradigmatic examples for this class of thermostated systems. For the Gaussian type numerical and analytical results provided evidence that in the full accessible phase space the Hausdorff dimension of the attractor is non-integer. In Poincar´e surfaces of section, or in respective projections of the phase space, the attractor even turned out to be multifractal; see Figs. 11.2 and 12.2 for plots depicting the associated fractal folding. Supported by results for many other Gaussian and Nos´e-Hoover thermostated systems it was thus conjectured that the

May 8, 2007

4:1

248

World Scientific Book - 9in x 6in

Universalities in Gaussian and Nos´ e-Hoover dynamics?

existence of fractal attractors, and of the associated singular probability measures, is typical for non-Hamiltonian dynamical systems in NSS subject to deterministic and time-reversible thermal reservoirs. How the topology of these attractors changes under variation of the electric field strength was assessed by means of bifurcation diagrams. For this purpose the positions of the colliding particles at the Lorentz gas disk were plotted as functions of the electric field strength, see Figs. 11.3 and 12.3. For both the Gaussian and the Nos´e-Hoover thermostated driven periodic Lorentz gas these diagrams revealed intricate bifurcation scenarios. On the other hand, the specific form of these scenarios depends intimately on the type of thermal reservoir used and is already very different for Gaussian and Nos´e-Hoover dynamics. Even more, in the case of Nos´e-Hoover thermostating different bifurcation diagrams were obtained for different values of the coupling strength between subsystem and thermal reservoir. Hence, there is no universality concerning the specific change of the topology of these attractors under variation of the field strength. Whether the mere existence of bifurcations, at least, is typical for the thermostated driven periodic Lorentz gas will be clarified in the following two chapters. Instead of applying Gaussian and Nos´e-Hoover schemes, thermal reservoirs can be mimicked by introducing stochastic interactions between subsystem and reservoir particles. Here we briefly sketch various existing numerical and analytical methods of how to model this second fundamental class of thermostats. We then connect to our discussion concerning the possible universality of fractal attractors in NSS. One way to introduce a stochastic reservoir is by using a Langevin equation, as was discussed in detail in Section 10.2. It appears that for molecular dynamics computer simulations this approach was first implemented by Schneider and Stoll [Sch78], see also later work in Refs. [Nos91; Pos00b]. Andersen [And80] proposed a variant of this stochastic bulk thermostat by considering an interacting many-particle system, where at certain time intervals the velocity of a randomly selected particle is chosen randomly from a canonical velocity distribution [All87; Jel88a]. Alternatively, a stochastic sampling of velocities from canonical velocity distributions can be performed at fixed boundaries of a subsystem. It seems that such stochastic boundary conditions have been proposed independently by many different authors and are somewhat “common sense” in the literature. The earliest works we could find are all on modeling heat conduction, cf. Section 17.3, by applying stochastic thermal reservoirs at some boundaries. Jackson et al. [Jac68] used stochastic boundaries for numerical

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

13.4. Fractal attractors for nonequilibrium steady states

klages˙book

249

simulations of heat conduction in one-dimensional lattices while Lebowitz et al. mathematically analyzed harmonic lattices and a three-dimensional random Lorentz gas under temperature gradients generated by stochastic boundaries [Rie67; Leb78]. For molecular dynamics computer simulations of heat conduction in fluids this thermostating scheme was implemented by Ciccotti and Tenenbaum [Cic80], see also Refs. [Ten82; Che95b; Che97; Pos98; Wag99] for later use of this method in the context of computer simulations. Stochastic boundary conditions will be introduced in full detail in Section 15.1. For the following discussion it is not necessary to know about such technical details. A particularly important result for the class of stochastic boundary thermostated models was reported by Goldstein, Kipnis and Ianiro [Gol85]. They studied a system of Newtonian particles maintaining a heat flux due to a temperature field that is modeled by stochastic boundaries and varies with the position at the boundary. A mathematical analysis of this system yielded that there exists a unique invariant probability measure that is absolutely continuous [Eck85; Dor99] with respect to the Lebesgue measure. In other words, in this case the measure is not singular and hence not fractal. This result was claimed to be typical for stochastically thermostated systems [Eyi92; Coh92] thus contradicting the universality of fractal attractors as conjectured from studying Gaussian and Nos´e-Hoover thermostated systems [Hol87; Hoo99]. The claim that the physically relevant probability measures are generally smooth in stochastically thermostated systems has been questioned by Hoover et al. on the basis of numerical explorations: In Ref. [Hoo98d] these authors studied a driven periodic Lorentz gas thermalized by a ‘hybrid’ thermostat consisting of deterministic and stochastic boundaries. Computations of the information dimension of the fractal attractor by evaluating phase space projections such as Fig. 12.2 indicated a slight deviation from an integer value. Furthermore, they considered a one-dimensional Hamiltonian model for heat conduction (a variant of the ding-a-ling model, cf. Section 17.3) with reservoirs defined by stochastic boundaries [Pos98]. They also investigated an interacting many-particle system under an external field [Pos00b] (the color conductivity model [Eva90b]). In the latter two cases they computed the information dimension according to the Kaplan-Yorke conjecture, respectively Young’s formula, cf. Eq. (13.19) of the previous section. The numerical results yielded again deviations from integer values. However, this dimension formula relies on the computation of Lyapunov exponents that in turn necessitate the definition of a

May 8, 2007

4:1

250

World Scientific Book - 9in x 6in

Universalities in Gaussian and Nos´ e-Hoover dynamics?

Jacobian quantifying the interaction of the subsystem with the stochastic boundaries. To us it appears that the definition which these authors used for the Jacobian only gives correct results under very specific conditions, hence we cannot consider the numerical results of Refs. [Pos98; Pos00b] to be conclusive.9 Thus, whether attractors in stochastically thermostated systems are generally smooth or fractal, and to which extent this property depends on the specific type of subsystem considered, still remains a very open question to us.10 Apart from these considerations we may emphasize again that both types of thermal reservoirs define very different classes of dynamical systems: Gaussian and Nos´e-Hoover thermostats keep the dynamics deterministic and time-reversible, whereas stochastically thermostated systems render the equations of motion non-deterministic and non-time reversible. The controversy [Gol85; Eyi92; Coh92; Hoo98d; Pos98; Pos00b; Gal03b] concerning the type of invariant measure that emerges under application of these two very different types of thermal reservoirs thus boils down to the question which type of thermostat one considers to be more ‘physical’. We furtmermore remark that, connected to our discussion in Section 13.1, Nicolis and Daems [Nic96; Nic98; Dae99] and Ruelle [Rue97c] argued for the positivity of the thermodynamic entropy production in stochastically thermostated systems.

9 In Ref. [Hoo98d] the phase space contraction at a stochastic boundary is defined by the second equation below Fig. 2. For this purpose an equality between phase space contraction and entropy production has been stipulated for this type of system. On the other hand, a natural assumption is just the conservation of phase space probability (or points in phase space) at the boundary, drdvρ(r, v) = dr 0 dv 0 ρ(r 0 , v 0 ), where (r, v) represent position and velocity of a particle before a collision, (r 0 , v 0 ) the same variables after a collision, and ρ is the respective probability distribution. A Jacobian at the collision is then straightforwardly defined by |J| = |dr 0 dv 0 /drdv| = ρ(r, v)/ρ(r 0 , v 0 ). If one follows this argument one arrives at the result of Ref. [Hoo98d] only (i) if both distributions before and after the collisions are identified with canonical distributions which is natural after a collision, but which is not so clear before a collision, cf., e.g., Ref. [Wag99] ; and (ii) if both distributions before and after a collision have precisely the same variance in terms of a temperature, T = T 0 . However, both distributions are only expected to be the same, with the same temperature, in the hydrodynamic limit, whereas systems of finite length as handled on the computer one may expect to exhibit temperature jumps [Kus02]. For this reason we consider the validity of the results reported in Refs. [Pos98; Pos00b] concerning the information dimension to be debatable. 10 We remark that, by computing Lyapunov exponents from trajectories perturbed by random numbers drawn from a generator with the same seed, new numerical results seem to exist which, again, indicate a phase space contraction in stochastically thermostated systems [Pos04a].

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

13.5. Nonlinear response in the driven periodic Lorentz gas

13.5

klages˙book

251

Nonlinear response in the driven periodic Lorentz gas

According to the numerical results for the thermostated driven periodic Lorentz gas, the topology of the attractors does not only depend on the specific type of thermal reservoir that has been applied, but it is also very sensitive to variations of the external field. From that point of view it is not too surprising that the field-dependent electrical conductivities of the Gaussian and the Nos´e-Hoover thermostated Lorentz gas are also not the same, as we have discussed in Sections 11.2.4 and 12.3.3, see Figs. 11.4 and 12.4. Furthermore, even for Nos´e-Hoover only the Lorentz gas conductivities still depend on the value for the coupling parameter between subsystem and reservoir, in the same way as the bifurcation diagrams of the different attractors. On the other hand, all conductivities jointly exhibit a profoundly nonlinear response for the numerically accessible values of the field strength ε > 0. For the Gaussian thermostated Lorentz gas there must exist a regime of linear response in the limit of very small fields according to the mathematical proof by Chernov et al. However, the respective range of field strengths close to zero appears to be so small that up to now computer simulations could not really verify its existence. This partly reminds one of van Kampen’s objections against the validity of linear response, who argued that a linear response is not trivially guaranteed for nonlinear chaotic dynamical systems. He emphasized that there might be a nontrivial interplay between microscopic nonlinearity and macroscopic linearity yielding a quantitatively negligibly small regime of linear response. In addition to this debate about coarse functional forms, all fielddependent conductivities exhibit irregularities on fine scales, which are not due to numerical errors. As was shown in Part 1 of this book, such irregular transport coefficients are in fact well-known from very related classes of dynamical systems, which share the properties of the periodic Lorentz gas of being deterministically chaotic, low-dimensional and spatially periodic. For very simple types of such systems it could be shown that these transport coefficients are even of a fractal nature. For thermostats of Nos´eHoover type the irregularities appear to be smoothed-out, which may be understood with respect to the fact that there the kinetic energy fluctuates according to a canonical velocity distribution. Imposing a constant kinetic energy onto the system by means of a Gaussian thermostat, these irregularities look considerably more “fractal-like”. In summary, to us it still remains an open question to quantitatively

May 8, 2007

4:1

252

World Scientific Book - 9in x 6in

Universalities in Gaussian and Nos´ e-Hoover dynamics?

identify a regime of linear response in the Gaussian thermostated driven periodic Lorentz gas, either analytically or numerically. One may furthermore ask whether such a regime of linear response may also be expected for applications of the Nos´e-Hoover and possibly of other types of thermostat. On the other hand, one may suspect that the existence of irregularities on finer scales of the field-dependent conductivity is rather typical for deterministically thermostated driven periodic Lorentz gases. To what extent this holds true, and particularly how this goes together with the expected linear response for very small fields, are important open questions. This points back to the question concerning the equivalence of nonequilibrium ensembles. Unfortunately, in low-dimensional deterministic and periodic dynamical systems like the Lorentz gas, the conductivity obviously reflects the specific type of thermal reservoir used. This appears to be at variance with an equivalence, which is expected to yield the same conductivities. Related findings have been reported in Ref. [Bon02]. One thus needs to apply further types of thermal reservoirs to the Lorentz gas dynamics in order to check for similarities and differences, which we will do in the following two chapters. Finally, as for the hotly debated fractality of attractors in NSS one may expect that the irregular or even fractal structure of parameter-dependent transport coefficients may become more regular by imposing stochasticity onto a system. This can be performed by either distributing the scatterers randomly in space, by using a stochastic thermal reservoir as discussed above, or by imposing additional noise onto a deterministic system. For thermostated driven random Lorentz gases, field and density-dependent Lyapunov exponents and diffusion coefficients have indeed already been calculated analytically and numerically [vB95; vB96; Lat97; Del97b; vB97; vB98; vB00b; dW04b; vB05; dW05] yielding a smooth dependence on parameters. Additionally, for simple low-dimensional maps it is known how the fractal structure of a parameter-dependent diffusion coefficient is affected while imposing different types of perturbations onto the maps in time or in space, see Chapter 5. As expected, the fractality on arbitrarily fine scales disappears, however, irregularities still survive in the form of smoothed-out oscillations. This irregular structure turns out to be very persistent against random perturbations, that is, typically rather strong random perturbations are needed to make the transport completely random walk-like.11 11 An interesting exception is provided by a specific type of quenched disorder leading to a Golosov random walk. In this case an originally normal diffusive process may

klages˙book

May 8, 2007

4:1

13.6.

World Scientific Book - 9in x 6in

∗ Summary

klages˙book

253

Hence, there may be complicated scenarios between the two limiting cases of completely deterministic and completely stochastic diffusion, and the application of external randomness significantly affects fractal properties of the type discussed above. Generally one may expect that problems with linear response and the irregularity of parameter-dependent transport coefficients specifically show up in low-dimensional dynamical systems. For interacting many-particle systems, for example, it is natural to assume that all transport coefficients are smooth in their physical parameters and that there are broad regimes of linear response, as predicted by standard nonequilibrium thermodynamics. Examples of such systems will be discussed in Chapter 15.

13.6

∗

Summary

(1) Gaussian and Nos´e-Hoover dynamics provide prominent examples for a non-Hamiltonian modeling of NSS. By means of a simple baker map we demonstrated how volume-preserving “Hamiltonian” dynamics goes together with dissipative equations of motion and non-uniform probability densities. The key is the elimination of reservoir degrees of freedom by projecting out variables from the full equations of motion. This illustrates the usefulness of a non-Hamiltonian description of NSS. (2) Ordinary Gaussian and Nos´e-Hoover thermostats feature an identity between thermodynamic entropy production and the average rate of phase space contraction. However, there exists an even more general identity relating phase space contraction to the Gibbs entropy production, which is completely independent of any type of modeling. In the case of a Hamiltonian approach to NSS the Gibbs entropy production is simply zero. This and other reasons cast doubts on whether the Gibbs entropy provides a useful definition of a nonequilibrium entropy being compatible with nonequilibrium irreversible thermodynamics. Dissipative dynamical systems in NSS exhibit a negative average rate of phase space contraction. Due to the abovementioned identity the Gibbs entropy production is negative as well. This appears to be at variance with the second law of thermodynamics. However, here the Gibbs entropy can be saved by applying methods of coarse graining leading to an inversion of the sign. immediately become anomalous, that is, a diffusion coefficient does not exist anymore [Rad96].

May 8, 2007

4:1

254

World Scientific Book - 9in x 6in

Universalities in Gaussian and Nos´ e-Hoover dynamics?

To consider a coarse grained entropy production is furthermore suggested by the fact that the invariant measures of deterministically thermostated systems are typically singular. The decisive question is whether the Gibbs entropy always yields the correct thermodynamic entropy production in nonequilibrium as, for example, compared to the Clausius entropy. For Gaussian and Nos´e-Hoover thermostats this holds true, but there is no reason why this should apply to other types of thermal reservoirs. (3) Having an identity between phase space contraction and entropy production at hand enables one to link transport coefficients to dynamical systems quantities. The escape rate approach to Hamiltonian dynamical systems yields a formula that is strongly analogous to the one obtained for dissipative dynamics: In both cases a transport coefficient is related to a difference between sums of Lyapunov exponents and the Kolmogorov-Sinai entropy on some fractal structure. In the case of thermostated systems this structure consists of a fractal attractor, whereas for open Hamiltonian systems it is a fractal repeller. There is even a third formula of a similar type derived for closed Hamiltonian dynamical systems, which connects the Hausdorff dimension of the hydrodynamic mode governing a transport process and a respective Lyapunov exponent to the corresponding transport coefficient. By performing suitable transformations the previous two formulas can be compared to the latter one. All resulting equations involve fractal dimensions of respective fractal sets and Lyapunov exponents thus revealing a striking formal analogy. On the other hand, these equations are obtained from three very different approaches to NSS, and correspondingly the physical meaning of the involved quantities is rather different. We described a ‘master formula’ linking two of the three equations, however, its additivity suggests that there is no further reduction onto a common ‘root’. The formula derived for thermostated systems stands and falls with the validity of the identity between phase space contraction and entropy production, which in turn appears to be doubtful. In this respect the formulas emerging from the two Hamiltonian approaches seem to be on safer grounds. (4) We commented upon the suspected typicality of fractal attractors in dissipative NSS. For Gaussian and Nos´e-Hoover thermostated systems there is a wealth of work corroborating the fractality of attractors. However, at least for the driven periodic Lorentz gas the topology of

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

13.6.

∗ Summary

klages˙book

255

these fractal structures depends not only on the type of thermal reservoir used, but also on the variation of control parameters such as the coupling strength between subsystem and reservoir and the electric field strength. This is exemplified by bifurcation diagrams exhibiting complicated bifurcation scenarios for the attractor under variation of these parameters. Alternatives to deterministic and time-reversible thermal reservoirs are provided by the stochastic Langevin equation and by stochastic boundary conditions. For a specific stochastically thermostated system it has been proven that the associated invariant measure is smooth hence contradicting the ubiquituous existence of fractal attractors in NSS. This was contrasted by numerical computations for other stochastically thermostated systems yielding again fractal objects. Whether attractors in stochastically thermostated systems are typically smooth or fractal, or whether this actually depends on the specific type of system, remains an important open question. (5) A similar reasoning applies to the field-dependent conductivities of the thermostated driven periodic Lorentz gas. As in the case of attractors, their functional forms reflects the specific type of thermal reservoir that has been applied. Numerically, no regimes of linear response could be detected for the driven periodic Lorentz gas, despite a proof for the existence of such a regime in the Gaussian thermostated case. In addition, all conductivities showed pronounced irregularities on finer scales as functions of the field strength, which partly may be of a fractal origin. How the latter property goes together with a suspected linear response and to what extent all these properties hold for arbitrary thermal reservoirs remains to be clarified. From a thermodynamic point of view these intricate dependencies of transport properties on parameter variations and on the type of reservoir are cumbersome and question the equivalence of nonequilibrium ensembles. One may thus attribute such nontrivial characteristics to the simplicity and low dimensionality of the model. This does not imply that the model itself is unphysical, since systems with these properties very well exist, as discussed in Part 1 of this book.. Imposing randomness onto such dynamics indeed smoothes out these irregularities. Considering a higher-dimensional system composed of interacting many particles is expected to yield similar consequences. In summary, a major question concerning the theory of transport in

May 8, 2007

4:1

256

World Scientific Book - 9in x 6in

Universalities in Gaussian and Nos´ e-Hoover dynamics?

dissipative systems might be formulated as follows: How typical are the results obtained from Gaussian and Nos´e-Hoover thermostats for NSS of dissipative dynamical systems? That is, to which extent are these properties universal in that they hold for other thermostating schemes as well, or do they depend on the specific way of thermostating? The remaining chapters of Part 2 will be devoted to sheding further light onto this fundamental question. Our approach will be carried out in two steps: A first step is to invent alternative, possibly more general thermostating schemes, which are not fully identical to the previous class of methods but share certain characteristics with them, such as being deterministic and time-reversible and leading to NSS. In the following we present two generic examples of such methods. In a second step we study whether, by using these schemes, the same results for quantities characterizing chaos and transport are obtained as exemplified by Gaussian and Nos´e-Hoover thermostats.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

Chapter 14

Gaussian and Nos´ e-Hoover thermostats revisited

So far we have identified important crosslinks between chaos and transport in Gaussian and Nos´e-Hoover thermostated models. Some authors conjectured these relations to be universal for dissipative dynamical systems altogether. It is thus illuminating to look at simple generalizations of ordinary Gaussian and Nos´e-Hoover dynamics, which we define again for the driven periodic Lorentz gas. We then analyze these thermostating schemes along the same lines as discussed before, that is, by analytically and numerically studying the transport and dynamical systems properties of the respectively thermostated driven periodic Lorentz gas. Finally, we compare the results obtained from our analysis to the ones characterizing conventional Gaussian and Nos´e-Hoover schemes as discussed in Chapters 11 to 13.

14.1

Non-ideal Gaussian thermostat

Consider again the Gaussian thermostated driven periodic Lorentz gas defined by Eqs. (11.1), (11.4) with an external field that is parallel to the x direction, ε = (εx , 0)∗ . We now distinguish between the interactions of subsystem and thermal reservoir parallel and perpendicular to the field direction by making the parallel coupling field-dependent, r˙x = vx v˙ x = εx − αvx − αεx vx r˙y = vy

v˙ y = −αvy 257

,

(14.1)

May 8, 2007

4:1

258

World Scientific Book - 9in x 6in

klages˙book

Gaussian and Nos´ e-Hoover thermostats revisited

where α = α(vx , vy ).1 The difference to the Gaussian case is reflected in the third term on the right hand side of the second equation. This additional term conveniently adjusts the action of the thermostat to the anisotropy of the external field thus providing a physical justification for its introduction. Correspondingly, one may expect that this version thermalizes the driven Lorentz gas more efficiently than the conventional, as we may now call it, Gaussian thermostat Eqs. (11.1), (11.4). Formally, this generalization just amounts to making the conventional coupling constant field-dependent and then expanding it in the field strength εx in the form of a power series, α = (1 + εx )α + O(ε2x ). The conventional Gaussian case is recovered in zeroth order while Eqs. (14.1) include the first order. The explicit functional form for α can be calculated in complete analogy to Section 11.1 from the requirement of energy conservation yielding2 εx vx . (14.2) α(vx , vy ) = 2 v + εx vx2 Comparing this result to the corresponding one for the conventional Gaussian thermostat Eq. (11.4) identifies the second term in the denominator as a new contribution. By using the definition of Eq. (11.5) the average phase space contraction rate is calculated to κp = − < α > − <

2ε2x vx vy2 > (v 2 + εx vx2 )2

.

(14.3)

Again, a comparison of κp with the result for the conventional Gaussian thermostat Eq. (11.6) is instructive: apart from the correction already contained in α, Eq. (14.3) features a second term for the phase space contraction rate. All these additional terms become negligible in the limit of εx → 0 thus approximately recovering the conventional Gaussian thermostat. But the point is that according to Eq. (14.3) there is no identity between phase space contraction and thermodynamic entropy production. Indeed, some entropy production related to Joule’s heat, dS = εx < vx > /T , see Eq. (11.8), is still present in Eqs. (14.2), (14.3), however, it does not show up in the form of a simple functional relationship to the average phase space contraction rate. 1 The careful reader may suspect some mismatch of units in the second of the four equations listed above. However, we replaced the Stokes friction coefficient by α(1 + εx /|ε0 |), where ε0 is some arbitrary field strength by which we make the field term dimensionless. As before with the masses, the charge, and the Boltzmann constant, we set ε0 = 1. 2 The same result is obtained from the Gaussian principle of least constraints [Mor98] by applying it to an anisotropic thermal reservoir, where the energy dissipation parallel to the field is not equal but proportional to the one perpendicular to it.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

14.1. Non-ideal Gaussian thermostat

klages˙book

259

We may thus refer to the coupling in terms of the conventional Gaussian thermostat defined by Eqs. (11.1), (11.4) as an ideal coupling, in the sense that it yields this simple identity, whereas we call the coupling Eqs. (14.1), (14.2) non-ideal in that it does not preserve the identity. Consequently, we call the thermostating scheme defined by Eqs. (14.1), (14.2) the non-ideal Gaussian thermostat.3 For another simple variant of the conventional Gaussian thermostat, called the µ-thermostat, and a discussion of its properties we refer to Refs. [Wil04; Bri05]. We now discuss computer simulation results for the driven periodic Lorentz gas thermostated that way [Kla99b]. For technical reasons here the field of strength ε was chosen to be parallel to the y-axis, ε = εy ,4 by adjusting the thermostating scheme in Eqs. (14.1) respectively. Fig. 14.1 (a) shows results for the conductivity σ(ε) of the non-ideal Gaussian thermostated driven Lorentz gas. The fluctuations on small scales are less irregular than for the conductivity of the conventional Gaussian thermostat Fig. 11.4, which might reflect the more efficient coupling with respect to the direction of the field. However, the oscillations seem to be less smooth than the ones for the Nos´e-Hoover case shown in Fig. 12.4. Again there is no visible regime of linear response for ε → 0. Curiously, the nonlinearities conspire [Coh98] to form an approximately linear response regime at intermediate values of the field strength. Reassuringly, for small enough field strength the conductivity approaches the value indicated by a star, which was calculated from the respective diffusion coefficient in the field-free case according to the Einstein formula Eq. (11.13), see Section 11.2.4. Fig. 14.1 (b) depicts numerical results for the average phase space contraction rate Eq. (14.3) and for the thermodynamic entropy production according to dS(ε) =< εvy /v 2 >= σ(ε)ε2 /T . These results quantitatively confirm the non-identity between both quantities as already suggested by Eq. (14.3). Note that in analogy to the irregularity of σ(ε), the entropy production dS(ε) particularly should yield irregularities on fine scales as well. However, this structure might be largely suppressed by the quadratic factor in ε. It would be interesting to investigate whether κp and dS exhibit different types of such irregularities corresponding to more intricate higher-order deviations between them. Since there is no identity between phase space contraction and entropy 3 This scheme was orally presented on occasion of the conference on Microscopic chaos and transport in many-particle systems in Dresden, August 2002 [Kla]. 4 Numerical results indicated that for ε||x the dynamics suffered from more creeping orbits than for ε||y.

4:1

World Scientific Book - 9in x 6in

260

klages˙book

Gaussian and Nos´ e-Hoover thermostats revisited

0.2 (a) σ(ε)

0.18 0.16 0.14

0

0.2

0.4

ε

0.6

0.8

1

0.15 κ(ε), dS(ε)

May 8, 2007

0.1 −κ dS

0.05

(b) 0

0

0.2

0.4

ε

0.6

0.8

1

Fig. 14.1 Computer simulation results for the driven periodic Lorentz gas equipped with the non-ideal Gaussian thermostat Eqs. (14.1) at the temperature T = 1: (a) Field-dependent conductivity σ(ε) as a function of the field strength ε = εy . The star denotes the result according to the Einstein formula Eq. (11.13). (b) Comparison between the average phase space contraction rate κ = κp (ε) and the thermodynamic entropy production dS(ε).

production there is trivially no simple relation between the conductivity and the Lyapunov exponents such as the Lyapunov sum rule Eq. (11.11), cf. to its derivation for the conventional Gaussian thermostat in Section 11.2.2. Indeed, for the average phase space contraction rate Eq. (14.3) it is not even obvious how to single out the conductivity without performing suitable approximations. For the two-dimensional Lorentz gas a conjugate pairing of Lyapunov exponents is guaranteed by default. However, one may conjecture that this property also holds in the case of higher-dimensional non-ideal Gaussian thermostated systems. It is furthermore natural to assume that a fractal attractor exists for the respectively thermostated driven periodic Lorentz gas, which is qualitatively of the same type as the one for the conventionally Gaussian thermostated model Fig. 11.2. These questions, as well as how the respective non-ideal bifurcation scenario looks, remain to be investigated.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

261

14.2. Non-ideal Nos´ e-Hoover thermostat

14.2

klages˙book

Non-ideal Nos´ e-Hoover thermostat

Precisely the same reasoning as for the non-ideal Gaussian thermostat can be applied to constructing a non-ideal Nos´e-Hoover scheme. We illustrate this by summarizing results from Ref. [Rat00a]. Without loss of generality, the coordinate system is chosen such that the direction of the field with strength ε is parallel to the x-axis. Indeed, replacing α by (1 + εx )α it is easy to see that the heuristic derivation for the conventional case provided in Section 12.2.1 must not be repeated, since all other functional forms remain precisely the same. This implies that even with a field-dependent α the respectively thermostated system is still transformed onto the same canonical distribution Eq. (12.8) as in the case of the conventional Nos´eHoover thermostat. For our definition of the non-ideal Nos´e-Hoover thermostat we thus simply combine Eqs. (14.1) with α defined by Eq. (12.10) yielding r˙x = vx v˙ x = εx − αvx − αεx vx r˙y = vy

v˙ y = −αvy

v 2 − 2T . (14.4) τ 2 2T The conventional Nos´e-Hoover thermostat Eqs. (11.1), (12.10) is recovered from these equations as a special case in thermal equilibrium, εx → 0. That the strength of the coupling between particle and reservoir is indeed properly adjusted to the anisotropy induced by the field is made more explicit as follows: Eqs. (14.4) can be rewritten by defining two fielddependent friction coefficients, αx := (1 + εx )α and αy := α, governed by α˙ =

α˙ x = (v 2 /2T − 1)(1 + εx )/τ 2

(14.5)

and α˙ y = (v 2 /2T − 1)/τ 2

,

(14.6)

respectively. For each velocity component there is consequently a separate √ reservoir response time according to τx := τ / 1 + εx and τy := τ . Correspondingly, with increasing field strength the response time parallel to the field decreases thus making the action of the thermostat parallel to the field more efficient compared to the coupling perpendicular to the field.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

262

klages˙book

Gaussian and Nos´ e-Hoover thermostats revisited Table 14.1 Average phase space contraction rate κp , see Eq. (14.7), and thermodynamic entropy production dS = ε < vx > /T , ε = εx , for the non-ideal Nos´e-Hoover thermostated driven periodic Lorentz gas Eqs. (14.4). As a value for the temperature we have T = 0.5. The numerical error is about ±0.001. The data are from Ref. [Rat00a]. ε 0.5 1.0 1.5

τ =1 −κp

dS

τ ' 31.6 −κp

dS

0.152 0.547 1.240

0.145 0.561 1.366

0.145 0.567 1.256

0.147 0.592 1.391

In analogy to the previous section and to Sections 11.2, 12.3, we shall now discuss the chaos and transport properties of the non-ideal Nos´eHoover thermostated driven periodic Lorentz gas. Starting from Eq. (11.5) the phase space contraction rate of this thermostated system is obtained to κp = −(2 + εx ) < α >

.

(14.7)

Performing an analogous calculation to that in Section 12.3.1, i.e., starting from the energy balance Eq. (12.13) and requiring that the average time derivative of the total energy is zero, by using Eqs. (14.4) we arrive at the equation εx < vx2 α > εx < v x > =2<α>+ , (14.8) T T which should again be compared to the result of the conventional case Eq. (12.14). Obviously, as in the case of the non-ideal Gaussian thermostat Eq. (14.3) there is a new second term that does not allow for an identity between the average phase space contraction rate and thermodynamic entropy production. In fact, if vx2 and α were independent quantities and if equipartition of energy were fulfilled, < vx2 >= T , then the identity would be recovered from Eq. (14.8). However, first of all, according to computer simulations vx2 and α are not independent quantities. Secondly, < vx2 >= T is only strictly fulfilled in thermal equlibrium, thus the identity cannot hold. Some numerical results for the phase space contraction rate κp and for the thermodynamic entropy production dS are presented in Table 14.1 at different values of τ and ε = εx yielding quantitative evidence for these deviations. An interesting aspect is that both functions apparently cross each other, which is at variance with the numerical results for the non-ideal Gaussian thermostat shown in Fig. 14.1, where −κp appears to provide a lower bound for dS. The reason for this phenomenon is not yet understood and may deserve further investigations.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

14.2. Non-ideal Nos´ e-Hoover thermostat

klages˙book

263

In analogy to the non-ideal Gaussian thermostat, Eq. (14.8) does not allow one to recover the Lyapunov sum rule Eq. (12.17) as it holds for the conventional Nos´e-Hoover thermostat: The electrical conductivity may be suitably introduced on the left hand side of Eq. (14.8), and by using Eq. (14.7) one may also replace the average friction coefficient in the first term on the right hand side by the average phase space contraction rate, respectively by the sum of Lyapunov exponents. Still, there remains the friction coefficient in the second term on the right hand side, which cannot be eliminated without making further assumptions. In any case, the resulting expression is significantly different from any ordinary Lyapunov sum rule as it holds for conventional Gaussian and Nos´e-Hoover thermostats. Thus, the non-ideal Nos´e-Hoover thermostat provides another counterexample against the universality of simple relations between chaos and transport in dissipative dynamical systems as discussed specifically in Section 13.3. No Lyapunov exponents have yet been computed for the non-ideal Nos´eHoover thermostated driven periodic Lorentz gas. However, since this scheme defines also a bulk thermostat we would expect that conjugate pairing holds for the four Lyapunov exponents of this system. In Section 12.3.2 we have outlined a generalized Hamiltonian formalism for the conventional Nos´e-Hoover thermostat. Along similar lines a derivation for the non-ideal Nos´e-Hoover thermostat can be performed [Kla]. The same holds for the non-ideal Gaussian thermostat in comparison to the conventional one. But these appear to be rather technical aspects such that we do not go into further detail here. The structure of the attractor for the non-ideal Nos´e-Hoover thermostated Lorentz gas, if projected onto Birkhoff coordinates, see Section 11.2.3, is qualitatively more “smoothed out” than the one of the conventionally Gaussian thermostated version Fig. 11.2. On the other hand, it is still a bit more detailed than the one corresponding to the attractor of the conventional Nos´e-Hoover thermostat Fig. 12.2. Here we do not show the attractor generated by Eqs. (14.4) but refer to Ref. [Rat00a] for a respective figure and for further details. Instead, we present a bifurcation diagram for the non-ideal Nos´e-Hoover thermostated driven periodic Lorentz gas. Fig. 14.2 may be compared to the one of the conventionally Nos´e-Hoover thermostated case Fig. 12.3 (a). Although the response time τ is one order of magnitude smaller for the conventional case than for the non-ideal version thus indicating a more efficient coupling to the thermal reservoir, the action of the non-ideal version

4:1

World Scientific Book - 9in x 6in

264

klages˙book

Gaussian and Nos´ e-Hoover thermostats revisited

3

β

May 8, 2007

0

0

2

4

6

8

10

ε Fig. 14.2 Bifurcation diagram for the non-ideal Nos´e-Hoover thermostated periodic Lorentz gas driven by an electric field of strength ε, which is parallel to the x-axis. β is defined in Fig. 11.1. The temperature is T = 0.5, the reservoir coupling parameter τ = 1. The figure is from Ref. [Rat00a].

appears to be similarly efficient concerning a full covering of the phase space, which in both cases breaks down approximately at ε ' 2. This is in full agreement with the physical motivation for introducing the field-dependent coupling as discussed in the previous section in that it should enhance the efficiency of the thermostat’s action compared to conventional schemes. A bifurcation diagram for the conventional Nos´e-Hoover case at τ = 1 can be found in the material accompanying Ref. [Rat00a] and confirms our above assessment. The parameter-dependent electrical conductivity for the non-ideal version has not yet been computed, however, based on our discussion of the previous cases one may expect that the results will just yield another highly nonlinear, irregular curve. 14.3

+

Further alternative thermostats

In order to conclude this chapter we summarize whether various mechanisms generating NSS fulfill the identity between the average rate of phase space contraction and thermodynamic entropy production discussed before. To a large extent, these mechanisms have already been encountered on previous occasions, see in particular Sections 12.4 and 13.2. Cohen and Rondoni showed that even the Gaussian isoenergetic thermostat applied to a many-particle system provides an identity between phase space contraction and entropy production only in the thermodynamic limit of a large number of particles [Coh98; Ron02a].

May 8, 2007

4:1

14.3.

World Scientific Book - 9in x 6in

+ Further

alternative thermostats

klages˙book

265

Benettin and Rondoni introduced a specific time-reversible rescaling of the single velocity components of a moving particle at the boundary of a system while keeping the total kinetic energy fixed [Ben01; Ron02a]. This mechanism may be considered as an intermediate case between a straightforward velocity rescaling, see Section 11.1, and an ordinary Gaussian isokinetic thermostat. The complete dynamical system is defined such that it exhibits a symmetry breaking thus generating a current in a NSS. Similarly to the non-ideal thermostats discussed in the previous two sections, for small enough bias the average phase space contraction rate becomes proportional to the thermodynamic entropy production while for larger bias this relationship becomes inherently nonlinear. Related dissipation mechanisms have been introduced earlier by Zhang and Zhang [Zha92] and in particular by Chernov and Lebowitz [Che95b; Che97], see also Section 15.5.1 where we will discuss the example of a shear flow in a NSS. However, Chernov and Lebowitz argued for an identity according to their method. As far as the generalized Nos´e-Hoover thermostats sketched in Section 12.4 are concerned, whether or not they exhibit an identity can easily be checked by following the second approach outlined in Section 12.3.1 for the example of the conventional Nos´e-Hoover thermostat. It turns out that controlling higher even moments along the lines of Nos´e-Hoover [Hoo89a; Jel89; Hoo96a; Pos97; Hoo97; Liu00], the Nos´e-Hoover chain thermostat [Mar92b; Hoo97; Tuc00] and the deterministic formulation of a Langevin-like equation according to Bulgac and Kusnezov [Bul90b] all preserve the identity. All these schemes are furthermore time-reversible thus sharing fundamental properties with the class of conventional Gaussian and Nos´e-Hoover thermostats. The cubic coupling scheme proposed by Kusnezov et al. [Kus90; Bul90a], on the other hand, is time-reversible but does not show the identity. This is due to a thermalization of the position coordinates creating an additional phase space contraction that does not contribute to a Joule heating. In a way, this method exemplifies our assessment formulated in Section 13.2 that the phase space contraction of a dynamical system may involve contributions that, from the point of view of a thermodynamic entropy production, do not matter. Unfortunately, as we discussed already in Section 12.4.2 the physical interpretation of thermalizing position coordinates in the cubic coupling scheme remains unclear. Hence, this modeling may not be considered as an unambiguous counterexample concerning the identity.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

266

Gaussian and Nos´ e-Hoover thermostats revisited

The generalized Nos´e-Hoover thermostats introduced by Hamilton et al. [Ham90; L’H93] and by Winkler et al. [Win92; Win95] are all timereversible but do not fulfill the identity either. However, here it is not yet clear whether these schemes work conveniently in nonequilibrium situations. The methods proposed by Branka et al. [Bra00a; Bra00b] and by Sergi et al. [Ser01] are already not time-reversible and hence do not qualify as counterexamples for Gaussian and Nos´e-Hoover dynamics. Apart from considering the action of thermal reservoirs leading to dissipative dynamical systems we may recall that there are other ways to create NSS which do not involve any phase space contraction thus also providing counterexamples concerning an identity. For example, Gaspard studied diffusion under concentration gradients, respectively dynamical systems under flux boundary conditions[Gas97b; Gas98a]. Eckmann et al. considered a model for a heat flow by using a fully Hamiltonian thermal reservoir consisting of infinitely many planar waves [Eck99a]. All these studies feature a Hamiltonian approach to NSS. In the latter case the system actually reminds one of the infinite Hamiltonian modeling of the Langevin equation, see Section 10.2, rather than the thermostating approach. In any case, in both situations there is by default no phase space contraction and consequently no identity. Similarly, van Beijeren and Dorfman argued for a non-existence of this identity by using a Lorentz gas model for a heat flow [vB00a], however, their argument appears to be debatable. Finally, Nicolis and Daems analyzed the general setting of dynamical systems perturbed by noise by deriving a balance equation for the entropy production [Nic96; Nic98; Dae99]. They found correction terms to an identity between phase space contraction and entropy production, which they identified as functions of the noise strength. But in this case the dynamical systems considered are again not time-reversible anymore because of the noise and thus belong to a different class to conventional Gaussian and Nos´e-Hoover thermostats.

14.4

∗

Summary

(1) We introduced a straightforward generalization of the conventional Gaussian thermostat, again for the example of the driven periodic Lorentz gas. This thermostating scheme is by default deterministic and time-reversible but is constructed such that it does not yield an identity between the average rate of phase space contraction and ther-

klages˙book

May 8, 2007

4:1

14.4.

World Scientific Book - 9in x 6in

∗ Summary

klages˙book

267

modynamic entropy production. We coined this method the non-ideal Gaussian thermostat. Since there is no identity, there is also no Lyapunov sum rule relating transport coefficients to dynamical systems quantities. The field-dependent conductivity of this model is again highly nonlinear and different from the one found in other thermostated driven periodic Lorentz gases. (2) In the same vein, we introduced a non-ideal Nos´e-Hoover thermostat that also does not exhibit the identity and hence no Lyapunov sum rule. In both cases, the non-identity resulted from modifying the coupling between subsystem and thermal reservoir by suitably adjusting it to the symmetry breaking caused by the external electric field. This increased the efficiency of the non-ideal thermostats compared to their conventional colleagues as could be seen, for example, by means of matching bifurcation diagrams. (3) We finally summarized various other existing approaches towards the construction of NSS by inquiring whether or not they feature an identity between phase space contraction and entropy production. Apart from Hamiltonian modelings of NSS, it appears that there are many other examples of thermostating schemes not providing such an identity. However, in most cases they have not yet been studied respectively in nonequilibrium situations.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

This page intentionally left blank

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Chapter 15

Stochastic and deterministic boundary thermostats

Gaussian and Nos´e-Hoover schemes belong to the class of bulk thermostats, where energy is removed from a subsystem during the originally free flight of a particle. Boundary thermostats, on the other hand, change the energy of a particle only at a collision with the boundaries of the subsystem.1 In this chapter we construct and analyze fundamental types of the latter class of thermostats. Section 15.1 focuses on stochastic boundary conditions as have already been mentioned in Section 13.4. In Section 15.2 we show how to make stochastic boundaries deterministic and time-reversible by using a simple chaotic map. This modeling of a thermal reservoir was called thermostating by deterministic scattering. The equations of motion resulting from this method are still deterministic and time-reversible hence sharing fundamental properties with Gaussian and Nos´e-Hoover schemes. However, as explained in detail in Section 15.3, both the construction of this deterministic boundary thermostat and the dynamics associated with it are generally very different in comparison to conventional bulk thermostats. An analysis of models thermostated by deterministic scattering thus enables us to learn more about universalities of nonequilibrium steady states in dissipative dynamical systems. In Section 15.4 we demonstrate this by studying again the transport and dynamical systems properties of a respectively thermostated driven periodic Lorentz gas. Our findings are compared to the ones of Gaussian and Nos´e-Hoover thermostats. Section 15.5 briefly reviews results for a hard-disk fluid under nonequilibrium conditions, which is also thermostated at the boundaries.

1 The

action of Gaussian and Nos´e-Hoover thermostats can be restricted to boundary layers of a subsystem thus bridging the gap between bulk and boundary thermostats [Pos89]. 269

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

270

Stochastic and deterministic boundary thermostats

reservoir vy vx

v’ v

Fig. 15.1 Setup for defining stochastic boundary conditions: A point particle of mass m = 1 moving with velocity v collides inelastically with a flat wall. The velocity after the collision is denoted by v0 . The difference in the kinetic energy (v 02 − v 2 )/2 is transfered to, or from, a thermal reservoir associated with the wall. v 0 is obtained from Eq. (15.3).

15.1

Stochastic boundary thermostats

Let us consider a point particle moving in the plane, which interacts with a thermal reservoir through collisions at a flat wall, see Fig. 15.1. If the reservoir is in thermal equilibrium at a given temperature, the reservoir degrees of freedom will be distributed according to a canonical distribution. Consequently, the subsystem particle will be thermalized at the collisions with respect to a Gaussian velocity distribution. If we assume that the memory of the particle is completely lost at a collision, the velocity after a collision can be randomly sampled from the respective velocity distribution of the thermal reservoir. This is the physical essence of what are called stochastic boundary conditions, see also our brief account on them in Section 13.4. For casting this physical picture into an equation it must be taken into account that an impenetrable wall breaks the symmetry of the subsystem. That is, there are different velocity distributions parallel and perpendicular to the wall: Parallel to the wall the particle’s outgoing velocity component vx0 , see Fig. 15.1, should be sampled from the Gaussian distribution for reservoir velocities parallel to the wall. However, in the perpendicular direction an observer at the wall will measure a non-zero average flux of reservoir, respectively subsystem particles coming to the wall. This flux is due to the fact that such particles can approach the impenetrable wall from one side only. Consequently, the symmetry of the distribution of velocity components perpendicular to the wall is broken. This is properly modeled by multiplying the bulk velocity distribution of reservoir or subsystem with the absolute value of the perpendicular velocity component, |vy |.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

271

15.2. Deterministic boundary thermostats

In summary, if the velocity distribution of the thermal reservoir is canonical at temperature T , an observer at the boundary measures for the two velocity components vx ∈ [−∞, ∞), vy ∈ [0, ∞) of a reservoir particle colliding with the boundary [Rie67; Jac68; Leb78; Cic80; Ten82; Che95b; Che97; Wag99] ! vx2 + vy2 3 −1/2 ρ(vx , vy ) = (2πT ) |vy | exp − . (15.1) 2T The subsystem particle must then exhibit precisely the same distribution of outgoing velocities, under the conditions that its velocities before and after the collision are not correlated and that the reservoir distribution is not modified. In practice, these scattering rules are implemented as follows [Wag99]: Draw two independently and identically distributed random numbers ζ, ξ ∈ [0, 1] from a uniform density ρ(ζ, ξ) = 1 by using a random number generator. The transformation rules of how to get from these uniform distributions to the Gaussian ones Eq. (15.1) can be calculated from the requirement of phase space conservation.2 For the outgoing velocities vx0 , vy0 ∈ [0, ∞) this condition reads ! r 2 2 dζdξ v0 x + v0 y 2 0 |v | exp − ρ(ζ)ρ(ξ) 0 0 = dvx dvy πT 3 y 2T ∂T (vx0 , vy0 ) . = (15.2) ∂vx0 ∂vy0 Integration yields for the transformation T the functional form (vx0 , vy0 ) = T −1 (ζ, ξ) √ p = 2T erf−1 (ζ), − ln(ξ)

,

(15.3)

which completes the algorithm. 15.2

Deterministic boundary thermostats

Not being deterministic and time-reversible, stochastic boundaries do not enable an analysis of transport and dynamical systems properties as outlined in the previous chapters. This problem can be resolved as follows [Kla00b; Rat00b; Wag99]: Let M : [0.1] → [0, 1] , (ζ, ξ) → M(ζ, ξ) be a 2 For a two-dimensional Gaussian without a flux term the same reasoning in polar coordinates leads to the well-known Box-M¨ uller algorithm [Pre92].

May 8, 2007

4:1

World Scientific Book - 9in x 6in

272

klages˙book

Stochastic and deterministic boundary thermostats

two-dimensional deterministic map with an inverse M−1 . Loosely speaking, the random number generator in Eq. (15.3) is then replaced by the action of this map in the form of the scattering rules (|vx0 |, |vy0 |)

=

T −1 ◦ M ◦ T (|vx |, |vy |) ; vx ≥ 0 T −1 ◦ M−1 ◦ T (|vx |, |vy |) ; vx < 0

(15.4)

with T −1 : [0, 1] × [0, 1] → [0, ∞) × [0, ∞). This prescription is not yet complete. First of all, we have to further specify the sign of the outgoing velocities. We do this for the geometry presented in Fig. 15.1 by requiring that particles going in with positive (negative) tangential velocity go out with positive (negative) tangential velocity. For the perpendicular component we require that vy always changes its sign. This distinction between positive and negative velocities, supplemented by the respective use of M and M−1 in the equations, renders the collision process time-reversible. This is easily verified by using the collision rules Eq. (15.4) and reversing the direction of time after a collision. Finally, to avoid any artificial symmetry breaking in numerical experiments, the application of M and M−1 in Eq. (15.4) should be alternated with respect to the position of the scattering process at the wall. It remains to further specify the map M of this equation. In order to compare with Gaussian and Nos´e-Hoover thermostats we require that this map is deterministic and reversible.3 Furthermore, M should be mixing [Arn68; Sch89; Gas98a; Dor99], preferably also uniformly hyperbolic [Eck85; Ott93; Gas98a; Dor99], and for convenience it should generate a uniform probability density in phase space. All these properties, except the last one, are necessary conditions for enabling a proper thermostating. That is, the whole phase space of the map should be sampled such that, in combination with the transformation T according to Eq. (15.4), the canonical distribution can be reproduced completely on the basis of these scattering events; see Section 12.4.1 for a related discussion of necessary conditions regarding a proper functioning of deterministic thermostats. Apart from these constraints the precise functional form of the map appears to be arbitrary. By choosing different maps one may model different particle-wall interactions. In this respect the map M together with the tranformation T plays the role of a scattering function containing detailed information about the microscopic scattering process of a particle. Fixing the functional form of the map completes our set of scattering rules, which 3 For

a definition of the latter property see Eq. (4.26) in Section 4.2.

May 8, 2007

4:1

15.3.

World Scientific Book - 9in x 6in

+ Boundary

thermostats from first principles

klages˙book

273

has been called thermostating by deterministic scattering (TDS) [Kla00b; Rat00b; Wag99; Rat02]. If not indicated otherwise, in the following we will choose for M the most simple example of a map fulfilling all of the above requirements. This is the two-dimensional baker map as explicitly discussed in Section 13.1, see Eq. (13.1) and Fig. 13.1. We remark that other simple maps such as the cat map, respectively for general values of the control parameter the sawtooth map, as well as the standard map [Sch89; Ott93; Mei92; All97], both for large enough values of their control parameters, have been used within this scheme as well [Wag99; Kla99b; Wag00].

15.3

+

Boundary thermostats from first principles

There are two crucial ingredients in the approach outlined above leading to stochastic and deterministic boundary thermostats: Firstly, the knowledge of the equilibrium velocity distributions of a thermal reservoir with infinitely many degrees of freedom is required. Secondly, the energy transfer at the boundaries must be defined by suitable microscopic collision rules. In both respects specifically our derivation of TDS is quite heuristic. That is, we started from pre-defined stochastic boundaries and conveniently modified them by choosing new collision rules. A systematic construction from first principles should proceed the other way around by first leading to the general solution, which is the deterministic case, and then recovering the stochastic version by eliminating any correlations at the collisions with the wall. Such an approach should provide more insight into the detailed mechanism and into the conditions for a proper functioning of this thermostat. Here we outline in three steps how such a systematic construction is performed [Kla00b; Rat00b]: 1. Equilibrium velocity distribution functions Our first goal is to construct a model for a thermal reservoir consisting of an arbitrary but finite number of degrees of freedom. For this purpose we need to know the velocity distribution functions of a subsystem consisting of ds degrees of freedom, which interacts with a thermal reservoir of dr < ∞ degrees of freedom in thermal equilibrium. With d = ds + dr we denote the total number of degrees of freedom, where ds < dr , and we may assume that there is only kinetic energy. In Section 10.3 we have summarized how

May 8, 2007

4:1

274

World Scientific Book - 9in x 6in

Stochastic and deterministic boundary thermostats

to calculate the finite-dimensional velocity distributions of the subsystem. The results were Eq. (10.19) for the equilibrium distribution of one velocity component v1 and Eq. (10.20) for the distribution of the absolute value of p 2 the velocity vector of two components v1 and v2 with v = v1 + v22 . Note that the corresponding distribution functions are bulk distributions only, whereas, for defining a boundary thermostat, we need to know the functional forms of these distributions at the boundaries of the thermal reservoir, respectively of the subsystem. This bulk-boundary transformation may be performed heuristically such as outlined in the previous section. Ref. [Rat00b] provides a microscopic derivation of this relation by calculating the average time of flight between two collisions and the average collision length. All these arguments eventually lead to a multiplication of the finite-dimensional bulk densities Eqs. (10.19), (10.20) with a “flux” factor of v1 , respectively of v, supplemented by a suitable renormalization of the associated velocity distributions. 2. Defining a proper coupling Knowing the functional forms of the subsystem’s boundary equilibrium velocity distributions, the next step is to define a suitable microscopic coupling between subsystem and thermal reservoir. The coupling should be such that it establishes an energy transfer between subsystem and reservoir degrees of freedom that generates the desired velocity distribution. In order to preserve the Hamiltonian character of the microscopic equations of motion of the subsystem to be thermostated, we demand that this coupling is deterministic and time-reversible. In some sense, there should furthermore be conservation of phase space volume, as will be explained later on. A first idea of how such a coupling may look can be obtained from the rotating disk model: Let us consider a unit cell of the field-free periodic Lorentz gas, see the geometry in Fig. 10.3, where the fixed disk rotates with an angular velocity ω. Altogether we have thus three degrees of freedom, two for the moving point particle and one for the rotating disk. Assuming that the particle is elastically reflected perpendicular to the disk reduces the problem of energy transfer between particle and disk to the problem of two elastically colliding masses on a line. The corresponding equations of motion can easily be derived and are of the form of a nonlinear two-dimensional map governing the exchange between translational and rotational energy. This rotating disk model was first proposed in Ref. [Rat00b] but became more popular only later on by applying it to a periodic Lorentz gas channel under a temperature gradient [Mej01; Lar03], see Section 17.3.3 for more

klages˙book

May 8, 2007

4:1

15.3.

World Scientific Book - 9in x 6in

+ Boundary

thermostats from first principles

klages˙book

275

more recent studies and further details. Here all disks of the spatially extended Lorentz gas rotate with different angular velocities thus mimicking a thermal reservoir of, in principle, arbitrary dimensionality. Interactions between the different rotational degrees of freedom are provided by a moving particle carrying energy from one disk to another. To some extent, this Hamiltonian modeling of a thermal reservoir reminds of the one employing an arbitrary number of harmonic oscillators, which was used for deriving the Langevin equation, see Section 10.2. It thus suffers from the same deficiency that, at least numerically, it cannot conveniently be applied to nonequilibrium situations involving external fields, because in this case an infinite number of reservoir degrees of freedom is needed in order to continuously remove energy from the subsystem. Hence we follow a different path leading to a more abstract and more general model of a thermal reservoir. This approach is further motivated by a second problem with the rotating disk model: Computer simulations for a single rotating disk under periodic boundary conditions show that only for specific values of the associated control parameters does the coupling reproduce the equilibrium velocity distributions Eqs. (10.19), (10.20) at d = 3 [Rat00b; Lar03]. This is due to non-ergodic behavior in a mixed phase space consisting of islands of stability and chaotic layers. As a consequence, there is not equipartitioning of energy for all values of the control parameters, which is needed for a microcanonical distribution for a subsystem plus thermal reservoir [Rei65]. In turn, only the presence of equipartitioning appears to guarantee a proper energy transfer between subsystem and reservoir degrees of freedom providing a proper thermalization. In other words, a map M defining the collision rules of a moving particle with a disk, such as provided by the rotating disk model, must generally be enforced to yield equipartitioning of energy, respectively the desired equilibrium velocity distributions for reservoir and subsystem.4 This can be achieved by either suitably tuning the control parameters of a given map or by choosing a new map that is sufficiently ‘nice’, in the sense of fulfilling the list of necessary conditions provided in the previous section. Map M must furthermore be combined with a second map T according to Eq. (15.4) in order to ensure that the desired functional forms of the equilibrium velocity distributions are attained. Map T is defined in complete formal analogy to Eqs. (15.2), (15.3) by replacing the Gaussian of Eq. (15.2) through the 4 In

Ref. [Lar03] it was argued that relating different rotational degrees of freedom to different disks also resolves this problem. However, in a way this hides the problem in a higher-dimensional phase space.

May 8, 2007

4:1

276

World Scientific Book - 9in x 6in

Stochastic and deterministic boundary thermostats

finite-dimensional “target” velocity distribution function of Eq. (10.19). As an example, let us consider a single Lorentz gas scatterer under periodic boundary conditions. We modify the collision rules according to the above reasoning such that this system mimicks a rotating disk-like situation with ds = 2 for the moving particle and dr = 1 associated with the disk, thus d = 3 altogether. As map M we choose again the baker map Eq. (13.1). The target equilibrium velocity distributions are determined by Eqs. (10.19), (10.20) at d = 3. The transformation T is then calculated from the prescription used in Eq. (15.2) if the Gaussian of this equation is replaced by the Eqs. (10.19), (10.20) at d = 3. Here one needs to make a choice about the coordinate system in which the velocity variables shall be defined. For the Lorentz gas disk one may choose, e.g., a local coordinate system with respect to the point at the collision. Further details are explained in the following section. The resulting expression for T is finally combined with map M according to Eq. (15.4). Computer simulations of this inelastic Lorentz gas scatterer provide evidence that, indeed, the desired velocity distributions are generated by this model [Rat00b]. Obviously, this method is not restricted to d = 3. According to the transformation T , all that we need to know is the proper equilibrium distribution function of a thermal reservoir associated with dr degrees of freedom while interacting with a subsystem of ds degrees of freedom. In other words, we do not need to care about any explicit equations of motion defining dr degrees of freedom in order to model a thermal reservoir. From now on we therefore carefully distinguish between the number of degrees of freedom of the thermal reservoir that we mimick, and the number of dynamical variables which are actually involved in modeling this reservoir. For the collision rules that we are currently discussing the action of the thermal reservoir may be represented by one dynamical variable only: Let k represent the absolute value of the velocity vector of all reservoir degrees of freedom. The value of k is then defined via energy conservation for subsystem plus reservoir, E = v 2 /2 + k 2 /2, where v is the absolute value of the velocity of a subsystem particle before a collision. Accordingly, the value v 0 after a collision is generated by the collision rules Eq. (15.4), with a suitable choice for T , and the corresponding value k 0 of the reservoir variable is again obtained from energy preservation. Thus, ∆k := k 0 − k yields the energy transfer between subsystem and reservoir at a collision. We add that this formalism could be used to obtain deterministic and time-reversible collision rules for two colliding granular particles instead of using normal and tangential restitution coefficients according to

klages˙book

May 8, 2007

4:1

15.3.

World Scientific Book - 9in x 6in

+ Boundary

thermostats from first principles

klages˙book

277

which the microscopic dynamics becomes dissipative, non-Hamiltonian, and not reversible, see Chapter 9. A finite-dimensional thermal reservoir would then mimick dr internal degrees of freedom of some granular material, storing energy from which kinetic energy can be restored at a collision, compare, e.g., to microscopic models of inelastic collisions [Asp98; Bri96]. However, note that we did not yet arrive at the transformation Eq. (15.3), since our collision rules still concern a finite dimensional thermal reservoir only. 3. Modeling an infinite dimensional reservoir: In a typical nonequilibrium situation, only a thermal reservoir consisting of an infinite number of degrees of freedom is able to continuously absorb energy and to possibly generate a NSS. For the equilibrium velocity distributions Eqs. (10.19), (10.20) this implies that we need to take the limit of d → ∞. As was shown in Section 10.3, employing equipartitioning of energy we arrive at the canonical distributions Eqs. (10.21), (10.22). In this limit the total energy of the thermal reservoir is not well-defined anymore, consequently it cannot be represented by a dynamical variable k as introduced for the finite-dimensional case discussed above. However, this variable can easily be dropped, since it does not appear explicitly in the collision rules. Using the canonical velocity distribution Eq. (10.21) we arrive at Eq. (15.2) in order to define the transformation T . Neglecting any correlations in the map M before and after a collision, the stochastic boundary conditions Eqs. (15.3) are recovered from their deterministic counterpart Eq. (15.4) as a special case. Hence, we may say that this approach provides a derivation of stochastic boundaries from first principles. In fact, TDS was first constructed by following this bottom-up approach [Kla00b; Rat00b]. Only subsequently was it realized that the final outcome may be considered as a deterministic generalization of stochastic boundaries along the top-to-bottom approach of Section 15.1 [Wag99]. This completes our derivation of a modeling of stochastic and deterministic boundaries “from first principles”. We conclude this section by taking up the thread from Sections 10.3 and 13.1, in which we provided a simple explanation for the origin of an average phase space contraction in thermostated nonequilibrium systems. In fact, the previous bottom-up approach towards deterministic boundary thermostats nicely illustrates this reasoning as follows: Let us start in an

May 8, 2007

4:1

278

World Scientific Book - 9in x 6in

Stochastic and deterministic boundary thermostats

equilibrium situation with constant total energy and let the number d = ds + dr of degrees of freedom of subsystem plus thermal reservoir for a moment again be finite. As we explained in Section 10.3, in this case an ensemble of points representing the combination of subsystem plus reservoir is uniformly distributed on a hypersphere in velocity space indicating that there is no average phase space contraction. However, the projected-out velocity distributions of the subsystem are typically not uniform anymore, cf. Eqs. (10.19), (10.20). Correspondingly, by calculating the Jacobian determinant for the collision rules Eqs. (15.4) only on the basis of the functional forms of T for given d, one finds that the subsystem dynamics only is not phase space preserving at a single collision [Kla]. This should not come as a surprise, since at any collision there is an exchange of kinetic energy with the reservoir. Nevertheless, in the case of an equilibrium situation the average phase space contraction rate for the subsystem is still zero reflecting the fact that on average there is no energy transfer between subsystem and reservoir. This was also confirmed numerically for a respectively modified Lorentz gas. The apparent phase space contraction at a collision simply results from the fact that so far we looked at the dynamics of the subsystem only. A complete analysis must take the dynamics of the thermal reservoir into account as well. This dynamics is adequately represented by the variable k that, here, is well-defined again because d < ∞ and can be obtained via energy preservation. The complete dynamics thus consists of the set of variables (v, k). By looking at the Jacobian determinant of the combined system (v, k) one can verify that the complete dynamics is locally phase space preserving. Consequently, there is no contradiction between a locally dissipative dynamics of the subsystem only and a phase space preserving Hamiltonian character of the full equations of motion for subsystem plus thermal reservoir. Let us now apply these arguments to a nonequilibrium situation, e.g., by imposing an external field onto the subsystem. The field will continuously increase the kinetic energy of a particle moving in the subsystem, and the thermal reservoir will continuously absorb this energy through collisions by storing it onto infinitely many degrees of freedom according to equipartitioning of energy. During the free flight of the particle there is no action of the thermal reservoir, and the subsystem is trivially Hamiltonian. At a collision we use precisely the same collision rules as defined above for which we have argued, in the case of d < ∞, that there is no loss of phase space volume either. Consequently, by looking at the subsystem plus thermal

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

15.4. Deterministic boundary thermostats for the driven periodic Lorentz gas

klages˙book

279

reservoir we conclude that the whole dynamics is phase space preserving, and in this respect Hamiltonian-like, in nonequilibrium as well. This is not in contradiction to the fact that there is an average flux of energy from the subsystem to the thermal reservoir, which counterbalances the generation of an average current parallel to the external field. These arguments should hold by making the number of reservoir degrees of freedom arbitrarily large, that is, in the limit of d → ∞. We finally remark that, although for TDS the equations of motion appear to be well compatible with Hamiltonian dynamics, as argued above, a Hamiltonian formulation of this scheme is currently not known.

15.4

Deterministic boundary thermostats for the driven periodic Lorentz gas

Let us now apply the TDS scheme to the driven periodic Lorentz gas. That is, the originally elastic collisions of the moving particle with the hard disk are made inelastic allowing for an energy transfer at a collision without changing the geometry of the system. Note that the unit cell with a single scatterer is periodically continued by applying periodic boundary conditions. In physical terms one may thus think of the arbitrarily many degrees of freedom that we are going to associate with the disk as mimicking, e.g., different lattice modes in a crystal, which remove energy from a colliding particle. This modification of the Lorentz gas is performed by adapting the collision rules Eqs. (13.1), (15.3) and (15.4) to the circular geometry of a disk. For this purpose we replace the usual Cartesian coordinates by the tangential, respectively the normal component of the velocity of the colliding particle in a local coordinate system at the scattering point. That is, we write vk = vx and v⊥ = vy in Eqs. (15.3), (15.4). Alternatively, one could choose local polar coordinates [Kla00b; Rat00b]. We are now prepared to study the chaos and transport properties of the driven periodic Lorentz gas thermostated by deterministic scattering and to compare the results with the ones obtained for ideal and non-ideal Gaussian and Nos´e-Hoover thermostats. Note that TDS was constructed in order to generate a canonical velocity distribution in thermal equilibrium. In this respect this boundary thermostat forms a counterpart to the Nos´eHoover thermostat acting in the bulk. Our previous Fig. 12.1 depicts some

May 8, 2007

4:1

280

World Scientific Book - 9in x 6in

Stochastic and deterministic boundary thermostats

representative nonequilibrium velocity distribution functions for the driven periodic Lorentz gas thermostated both by Nos´e-Hoover and by deterministic scattering. For moderate to large response times the Nos´e-Hoover distributions are indeed very similar to the one obtained from TDS. Concerning other chaos and transport properties a comparison is more non-trivial, as we will show in the following. First we check analytically for an identity between phase space contraction and entropy contraction. We then present numerical results for the attractor both resulting from deterministic and from stochastic boundary thermostats. In the deterministic case we further discuss the associated bifurcation diagram as well as the electrical conductivity. Finally, we elaborate on the spectrum of Lyapunov exponents. This section particularly summarizes results from Refs. [Kla00b; Rat00b; Rat02].

15.4.1

Phase space contraction and entropy production

In contrast to bulk thermostats such as Gauss or Nos´e-Hoover, for a driven periodic Lorentz gas thermostated at the boundaries the phase space volume is conserved during free flights between collisions. As was briefly mentioned in the previous section, any phase space contraction can thus only be generated by the collision rules defined in terms of the composition of maps Eqs. (13.1), (15.3) and (15.4). For maps any change of the phase space volume is assessed by the Jacobi determinant, see Section 13.1. In the case of TDS adapted to the Lorentz gas geometry this quantity may 0 |/|dvk dv⊥ |. Here the numerator is composed of the be denoted by |dvk0 dv⊥ two velocity components after a collision and in the denominator there are the corresponding velocity components before a collision with the disk. If a particle collides n ∈ N times the product of Jacobian determinants evolves in time according to ! n 0 0 n 0 0 X Y |dvi,k dvi,⊥ | | |dvi,k dvi,⊥ = exp ln |dvi,k dvi,⊥ | |dvi,k dvi,⊥ | i=1 i=1 ! 0 |dvk0 dv⊥ | > . (15.5) = exp n < ln |dvk dv⊥ | For the last step it was assumed that the system is ergodic, which enables one to replace the time average by an ensemble average over colliding particles. In this expression 0 |dvk0 dv⊥ | κb :=< ln > (15.6) |dvk dv⊥ |

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

15.4. Deterministic boundary thermostats for the driven periodic Lorentz gas

klages˙book

281

yields the average exponential rate of phase space contraction per unit time, where the index b indicates the definition at a boundary. This quantity is obviously zero if the map is volume preserving and deviates from zero otherwise. Hence, for time-discrete dynamics such as the collision map Eq. (15.4), κb replaces the average phase space contraction rate κp of timecontinuous dynamics defined by Eq. (11.5).5 For the periodic Lorentz gas thermostated by deterministic scattering, κb can be calculated from the collision rules Eqs. (13.1), (15.3) and (15.4) to [Rat02] < v 02 > − < v 2 > , (15.7) 2T where v 0 and v are the absolute values of the velocities after and before a collision, respectively. κb is thus identical to minus the average outward flux of kinetic energy to the reservoir dQ := (< v 2 > − < v 02 >)/2 divided by the temperature T , where T derives from Eq. (15.3). This result holds in equilibrium as well as in nonequilibrium. As in the case of Gaussian and Nos´e-Hoover thermostats, Eq. (15.7) may now be compared to the entropy production dS = dQ/Tr in terms of the heat transfer between subsystem and thermal reservoir, cf. Section 11.2.1, where Tr denotes the temperature of the thermal reservoir, leading to κb =

Tr dS . (15.8) T In this case the question concerning an identity between dS and κb boils down to the problem of whether the temperature T coming from Eq. (15.3) can always be identified with the actual temperature of the thermal reservoir Tr . In equilibrium, that is, without applying an external field ε, T = Tr for the periodic Lorentz gas, which follows from the definition of this thermostating scheme outlined in Section 15.1 to 15.3. However, in this situation Eq. (15.8) is trivially zero on both sides anyway. On the other hand, to define a proper temperature in a nonequilibrium situation is very problematic, see also our brief discussion in Section 10.1. For our model one can argue along two different lines in order to define a proper reservoir temperature: One reasoning is that the temperature of a thermal reservoir consisting of an infinite number of degrees of freedom should never change under whatsoever nonequilibrium conditions and, κb = −

5 More explicitly, κ naturally shows up in the formal solution for the probability disb tribution function of the Frobenius-Perron equation in dissipative maps. This is in full analogy to solving the Liouville equation Eq. (12.3), where the average of the divergence plays the same role.

May 8, 2007

4:1

282

World Scientific Book - 9in x 6in

klages˙book

Stochastic and deterministic boundary thermostats

hence, should always be the same as in thermal equilibrium [Hoo02a]. This implies that Tr = T , and according to Eq. (15.8) the identity between phase space contraction and entropy production, as discussed for ideal Gaussian and Nos´e-Hoover thermostats, holds as well. However, for all thermostats analyzed in Part 2 it is well-known that generally the nonequilibrium temperature of the subsystem, if defined via the kinetic energy of the moving particle on the basis of equipartitioning of energy, is not necessarily equal to the equilibrium reservoir temperature T that is contained in the equations of motion [Pos88; Pos89; Eva90b; Sea98; Rat00a]. In the case of boundary thermostats this implies that identifying Tr with T typically yields temperature jumps at the boundaries [Kus02]. Such discontinuities in the temperature profile are at variance with a local equipartitioning of energy between the reservoir degrees of freedom and the degrees of freedom of the colliding particle. In contrast to the first definition outlined above, this motivates simply assuming such an existence of a local equipartitioning of energy in a NSS. Under this assumption the reservoir temperature can be defined by computing the average kinetic energy of a particle at the moment of the collision [Wag99; Kla00b; Rat00b; Rat02]. This definition of a nonequilibrium temperature thus amounts to smoothly extrapolating the temperature profile from the bulk of the subsystem to the boundary of the reservoir. An application of this method to the boundary thermostated driven periodic Lorentz gas yields Tr > T for an electric field with strength ε > 0 [Rat02]. Consequently, by following a temperature definition that is based on a no temperature-slip assumption at the boundary, the identity is not recovered. In our view, both definitions of a reservoir temperature can be defended. Therefore, the issue of the existence or non-existence of the identity may be considered as undecided in this case. As far as the Lyapunov sum rule introduced in Section 11.2.2 is concerned the situation is simpler. Replacing κb on the left hand side of Eq. (15.8) by the sum of Lyapunov exponents we may employ dS = ε < vx > /Tr on the right hand side, which is the Joule heat that the particle gains during a free flight. Using furthermore σ(ε) =< v > /ε we obtain σ(ε) = −

4 T X λ(ε) ε2 i=1

,

(15.9)

which, formally, is precisely the Lyapunov sum rule derived for the ideal Gaussian and Nos´e-Hoover thermostats, cf. Eqs. (11.11), (12.17). The only ambiguity is, as above, whether T should be identified with the temperature

4:1

World Scientific Book - 9in x 6in

klages˙book

15.4. Deterministic boundary thermostats for the driven periodic Lorentz gas

1

283

1 (b)

sin(γ)

(a)

sin(γ)

May 8, 2007

0

−1

0

1

2

3

0

−1

0

1

β

2

3

β

Fig. 15.2 Attractor for the periodic Lorentz gas driven by an electric field of strength ε = 1, which is parallel to the x-axis. The reservoir equilibrium temperature is T = 0.5, β and sin γ are defined in Fig. 11.1. In (a) the system is thermostated by deterministic scattering, see Eqs. (13.1), (15.3) and (15.4), in (b) it is thermostated by stochastic boundaries, see Eq. (15.3) and explanations. The results are from Refs. [Kla00b; Rat00b].

of the thermal reservoir in nonequilibrium or not. In any case, whether or not there holds an identity between phase space contraction and entropy production, here there exists a simple relation between the conductivity and the Lyapunov exponents of the system. 15.4.2

Attractors, bifurcations and conductivity

Fig. 15.2 shows two attractors for the driven periodic Lorentz gas thermostated at the boundaries, where all results presented for this thermostat are at T = 0.5. Part (a) displays simulation results for the deterministic boundaries Eqs. (13.1), (15.3) and (15.4) adapted to the Lorentz gas geometry. Part (b) shows analogous results for stochastic boundaries, where the baker map Eq. (13.1) has been replaced by a random number generator. Both figures may be compared to the previous attractors Fig. 11.2 for the Gaussian thermostated Lorentz gas and Fig. 12.2 for Nos´e-Hoover. Clearly, TDS produces a fractal-like folding, in analogy to that displayed for the Gaussian and for the Nos´e-Hoover thermostat. In Fig. 15.2 (a) this structure just appears to be a bit more smoothed-out than for the Nos´eHoover case. On the other hand, apart from a few non-uniformities on a coarse scale no fine structure is visible anymore for the attractor obtained from stochastic boundaries suggesting that Fig. 15.2 (b) does not represent a fractal set.

4:1

World Scientific Book - 9in x 6in

284

klages˙book

Stochastic and deterministic boundary thermostats

3

2

β

May 8, 2007

1

0

0

1

2

3

4

5

ε

Fig. 15.3 Bifurcation diagram for the periodic Lorentz gas thermostated by deterministic scattering and driven by an electric field of strength ε, which is parallel to the x-axis. The reservoir equilibrium temperature is T = 0.5, β is defined in Fig. 11.1. The figure is from Refs. [Kla00b; Rat00b].

The dimensionality of the TDS attractor in the four-dimensional phase space was computed numerically by means of the field-dependent KaplanYorke dimension DKY (ε), which was briefly mentioned in Section 13.3 [Rat02]. The results show that the dimensionality monotonically decreases from DKY (0) = 4 to DKY (1) ' 3.7 quantitatively confirming the fractality of the attractor in Fig. 15.2 (a). For Fig. 15.2 (b) respective computations are more subtle6 and remain to be done. Such computations would be interesting, since according to our discussion in Section 13.4 it is not yet clear whether or not stochastically thermostated systems typically exhibit fractal attractors. The bifurcation diagram for the driven periodic Lorentz gas thermostated by deterministic boundaries is depicted in Fig. 15.3. Again, it should be compared to the previous counterparts Fig. 11.3 for the ideal Gaussian thermostat, Fig. 12.3 for the ideal Nos´e-Hoover and Fig. 14.2 for the non-ideal Nos´e-Hoover thermostat. Remarkably, for TDS there are no bifurcations at all. That is, Fig. 15.3 neither shows a contraction of the attractor onto periodic orbits nor a breakdown of ergodicity for higher field strengths as occurs for Gauss and Nos´e-Hoover. Analogous results have been obtained for other choices of projections in phase space [Rat00b]. These results indicate that TDS more strongly randomizes the dynamics 6 The Kaplan-Yorke dimension requires the computation of the full spectrum of Lyapunov exponents, and in the case of stochastic boundaries one has to compute Lyapunov exponents for a stochastically perturbed dynamics; see also our discussion in Section 13.4 for respective difficulties.

4:1

World Scientific Book - 9in x 6in

klages˙book

15.4. Deterministic boundary thermostats for the driven periodic Lorentz gas

285

0.40

σ

May 8, 2007

0.36

0.32 0.1

0.5

0.9 ε

Fig. 15.4 Electrical conductivity σ(ε) for the periodic Lorentz gas thermostated by deterministic scattering and driven by an electric field of strength ε, which is parallel to the x-axis. The reservoir equilibrium temperature is T = 0.5. Each point has a numerical uncertainty that is less than the symbol size. The results are from Refs. [Kla00b; Rat00b].

of the subsystem than Gauss and Nos´e-Hoover. The different bifurcation diagrams clearly show that for simple systems such as the periodic Lorentz gas the deformations of the attractor under variation of the field strength intimately depend on the specific type of thermostating. Fig. 15.4 displays the field-dependent electrical conductivity for the driven periodic Lorentz gas deterministically thermostated at the boundaries. In comparison to the conductivities corresponding to the ideal Gaussian, Nos´e-Hoover and to the non-ideal Nos´e-Hoover thermostats shown in Figs. 11.4, 12.4 and 14.1 (a), respectively, this curve looks rather smooth. As in the other cases, there is no indication of a regime of linear response in the numerically accessible regime of field strengths above ε ≥ 0.1. On the other hand, yet again some wiggles on fine scales are visible suggesting that even for a bifurcation diagram such as Fig. 15.3 the corresponding conductivity might be, to some extent, an irregular function on fine scales. Note that the diffusion coefficient of the field-free periodic Lorentz gas thermostated by deterministic scattering may be different from the one for fully elastic scatterers and has not yet been computed. Hence, we do not know the prospective limiting value of σ(ε) for ε → 0 according to the Einstein formula Eq. (11.13).

4:1

World Scientific Book - 9in x 6in

286

klages˙book

Stochastic and deterministic boundary thermostats

0.10

λ1

λ2

1.86 0.05

1.84

1.82 0.0

0.5 ε

0.00 0.0

1.0

0.0

−1.85

−0.1

−1.90

−0.2

λ4

λ3

May 8, 2007

0.5 ε

1.0

0.5 ε

1.0

−1.95

−0.3 −2.00

−0.4 −0.5 0.0

0.5 ε

1.0

−2.05 0.0

Fig. 15.5 The four Lyapunov exponents λi (ε) for the periodic Lorentz gas thermostated by deterministic scattering and driven by an electric field of strength ε, which is parallel to the x-axis. The reservoir equilibrium temperature is T = 0.5. The numerical uncertainty of each point is less than 5 · 10−4 . The results are from Ref. [Rat02].

15.4.3

Lyapunov exponents

By applying TDS the kinetic energy is not kept constant in time, in analogy to Nos´e-Hoover, but in contrast to the Gaussian thermostat. Consequently, the two-dimensional TDS Lorentz gas is characterized by four Lyapunov exponents. Numerical computations of the Lyapunov spectrum are delicate due to the inelasticity at the collisions and require some more intricate adjustments compared to standard methods [Rat02]. For zero field, numerical results corroborate that conjugate pairing holds with two zero, one negative and one positive Lyapunov exponent. However, in nonequilibrium there is no pairing of Lyapunov exponents anymore as shown in Fig. 15.5. This is in sharp contrast to the Lyapunov spectrum associated with bulk thermostats such as Gauss and Nos´e-Hoover but appears to be typical for boundary thermostats, see our discussion in Section 11.2.2. More precisely, one Lyapunov exponent is zero reflecting phase space

May 8, 2007

4:1

World Scientific Book - 9in x 6in

15.5. Hard disk fluid under shear and heat flow

klages˙book

287

conservation parallel to the flow. Two become negative and decrease monotonically in the field strength, similar to power laws. Their functional forms remind one of the non-zero exponents associated with the Gaussian thermostated driven Lorentz gas [Mor96a]. Indeed, as we have shown, for both thermostats the Lyapunov sum rule holds, see Eqs. (11.11), (15.9). This equation, in turn, requires that the Lyapunov exponents are quadratic at least for small enough field strength. The single positive Lyapunov exponent, on the other hand, displays a more intricate non-monotonic behavior under variation of the field strength. According to Pesin’s theorem, see Section 13.3, it must be equal to the Kolmogorov-Sinai entropy suggesting the following heuristic understanding of its functional form [Rat02]: For small field strength the particle attempts to move parallel to the field, hence the Kolmogorov-Sinai entropy decreases. However, by increasing the field strength the system starts to heat up in the bulk, consequently the Kolmogorov-Sinai entropy increases again. Such a behavior was not observed for the Gaussian thermostated Lorentz gas, where the Kolmogorov-Sinai entropy always decreases monotonically [Del95a; Det95], which is probably due to constraining the energy in the bulk.

15.5

Hard disk fluid under shear and heat flow

Up to now we have applied different schemes of thermostating to the driven periodic Lorentz gas only, which is a one-particle system that derives its statistical properties from the chaotic scattering of a moving point particle with a fixed hard disk. We deliberately chose a very simple model in order to focus on the impact that different thermostats have on this system. An important question is therefore to what extent our findings about relations between chaos and transport still hold for systems that include interactions between moving particles. Suitably amended periodic Lorentz gases of this type were already investigated in Refs. [Bar93; Bon97; Bon02; Mej01; Lar03]. Here we skip this intermediate step and immediately consider interacting many-particle systems for which hard disk fluids under nonequilibrium conditions are typical examples. By studying these systems we are back to the historical roots of thermostating: As explained in Section 1.2 and Chapters 10 to 11, deterministic thermostats were originally designed as efficient tools within the framework of molecular dynamics computer simulations for driving large, interacting

May 8, 2007

4:1

288

World Scientific Book - 9in x 6in

Stochastic and deterministic boundary thermostats

many-particle systems into well-defined NSS that are consistent with the laws of thermodynamics. That deterministically thermostated systems furthermore exhibit interesting chaos properties was only realized later on and subsequently motivated more specific studies of simple models such as the Nos´e-Hoover oscillator and the thermostated driven Lorentz gas. Here we move the other way around, that is, from simple to more complex systems. We start by sketching various methods for modeling a shear flow. A well-known example are the SLLOD equations of motion representing the application of the conventional Gaussian thermostat to obtaining a sheared NSS. We then outline the construction of Maxwell daemon boundaries for modeling a hard disk fluid under shear. Finally, we discuss in more detail hard disk heat and shear flows generated by TDS [Wag99; Wag00]. Interesting questions are whether these systems yield an identity between phase space contraction and thermodynamic entropy production, whether there are Lyapunov sum rules for the respective transport coefficients, and whether there exist fractal attractors. We furthermore summarize numerical results for the Lyapunov spectrum and for the KolmogorovSinai entropy of this system. 15.5.1

Homogeneously and inhomogeneously driven shear and heat flows

Modeling and understanding shear and heat flows is a central problem in nonequilibrium statistical mechanics. It is not our intention to review the huge literature on these subjects, cf. Refs. [All87; Eva90b; Hes96a; Hes96b; Hoo99; Mun00] instead. By focusing on shearing we outline some important approaches to this problem and give a short account of investigations related to chaos and transport. Probably the most naive way to shear a many-particle system is by using walls moving with constant velocities [Hes96a; Mun00]. Early attempts along these lines trace back to work by Hoover and Ashurst, see Ref. [Hoo75] and Refs. [Lie92; San92] for later investigations. Posch and Hoover used such sliding boundaries in combination with a Nos´e-Hoover thermostat restricted to these boundaries [Pos89]. In contrast to this inhomogeneous driving from the boundaries, LeesEdwards or sliding brick boundary conditions act homogeneously in the bulk of the system [All87; Eva90b; Hes96a; Hes96b; Hoo99; Mun00; Coh95]. The idea is to suitably displace copies of a unit cell of particles in space such that a linear shear profile is imposed onto the system. Lees-Edwards

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

15.5. Hard disk fluid under shear and heat flow

klages˙book

289

boundaries alone already yield a planar Couette flow for not too high shear rates. However, it turned out that this method could be improved by combining it with a Gaussian thermostat. In fact, the need to conveniently simulate sheared fluids and solids originally motivated the invention of the Gaussian scheme, cf. also Section 11.1 [Hoo82; Eva83a]. Further refinements of the approach consisting of homogeneous shearing and Gaussian bulk thermostating eventually led to the SLLOD7 equations of motion [Eva90b; Eva90a; Sar92; Coh95; Hes96a; Mor89c; Sar98; Sea98; Mun00]. Curiously, by applying these shear flow algorithms to fluids consisting of hard spheres, shear-induced microstructures emerged in simulations for high enough shear rates. That is, the particles aligned themselves parallel to the direction of the flow in the form of strings, or blocks (“plugs”). These orderings became known as the string phase, [Eva86; Eva90b; Mun00], respectively the plug phase [Loo90; Hes90; Hes96b; Hes97]. Partly it was argued that these phases represent artifacts of the homogeneous thermostating algorithms described above, since by construction these methods are adapted to the shear flow profile in a self-contained way [Eva86; Eva90b]. This led to the invention of more refined profile unbiased thermostating schemes (PUT) for which, partly, such structures were not observed anymore. Further refinements of PUT thermostats, on the other hand, reproduced these flow-induced positional orderings again [Loo90; Hes90], whereas more recent simulations using configurational thermostats that make no assumptions on flow profiles yielded again different conclusions [Del03b; Del03a; Del04]. To some extent, this ongoing controversy is reminiscent of our previous discussion starting from Chapter 11, where we have shown that the application of different thermostats to the driven periodic Lorentz gas may yield different NSS. Here we appear to have a similar situation in the nonlinear regime of a sheared and thermostated many-particle system. More recent work focused onto the connection between momentum transport and chaos properties in shear flows. It may not come as a surprise that, again, the SLLOD equations furnish an identity between phase space contraction and entropy production [Coh97]. Consequently, there is a Lyapunov sum rule as already discussed in Section 11.2.2 [Eva90a]. The most simple shear flows are two-particle systems [Lad85] for which the existence of a shear viscosity could be proven [Bun96]. Morriss et al. [Mor87b; Mor89a; Mor89b; Pet94] computed the complete spectra of Lyapunov ex7 Concerning the origin of this name, which is not an acronym, see, e.g., Ref. [Mun00] and further references therein.

May 8, 2007

4:1

290

World Scientific Book - 9in x 6in

Stochastic and deterministic boundary thermostats

ponents for such systems by showing that the phase space contracts onto fractal attractors. More recently, the escape rate formalism outlined in Section 13.3 has been applied to a two-particle shear flow. Using the concept of Helfand moments [Gas95c; Vis03a], it led to a formula that is formally analogous to Eq. (13.15), which was verified numerically [Vis03b]. Numerical computations of Lyapunov spectra for more than two-particle systems under application of conventional thermostating algorithms and under shear have been performed in Refs. [Pos89; Eva90a; Sea98; Sar92; Eva00]. The existence of a conjugate pairing rule in shear flows, even if homogeneously thermostated, turned out to be a more delicate topic thus connecting to our discussion in Section 11.2.2. We also mention that even a multibaker map modeling a shear flow has been constructed and analyzed in Refs. [T´el01; M´ at01]. Chernov and Lebowitz [Che95b; Che97] proposed a many-particle shear flow model exhibiting a well-defined NSS without using a conventional thermostat. They considered a two-dimensional system of hard disks confined in a square box of length L with periodic boundary conditions along the x-axis, i.e., the left and right sides at x = ±L/2 are identified. At the top and bottom sides of the box, y = ±L/2, they introduced rigid walls where the disks are reflected according to certain scattering rules. The disks interact among themselves by impulsive hard collisions such that the bulk dynamics is purely conservative. Shear is now introduced by imposing collision rules at the boundaries turning the angle θ, which the velocity vector of a particle moving towards the wall forms with the wall, into an outgoing angle θ 0 := cθ, 0 < c < 1. These “Maxwell daemon-like” boundaries mimic the impact of a shear force at the particle-wall interaction and enforce that particles move favorably parallel to the wall. Since energy is conserved by this operation, the total energy of the particles is strictly kept constant thus reminding one of the action of a Gaussian thermostat. Similar constraints have also been studied by other authors in different settings [Zha92; Ben01]. Chernov and Lebowitz used both time-reversible and irreversible formulations of such scattering rules. In both cases the resulting NSS of the hard disk-fluid were in full agreement with hydrodynamics. As a consequence of the specific scattering rules at the boundaries, again the total phase space volume was not conserved. However, in this case the authors argued for an identity between phase space contraction and entropy production. Subsequent numerical studies of the dynamical instability of this system performed by other authors [Del97b]

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

15.5. Hard disk fluid under shear and heat flow

klages˙book

291

clearly showed that, as usual for boundary thermostats, the model exhibits no conjugate pairing rule. By using the full spectrum of Lyapunov exponents, on the other hand, evidence was provided for the existence of a fractal attractor in nonequilibrium. In further work the fluctuations of the entropy production of this model were investigated [Bon98a; Bon01]. Modeling heat flows turns out to be even more sensitive with respect to chaos and transport properties, as we will discuss in detail in Section 17.3. There we will focus on the validity of Fourier’s law in Hamiltonian chains of oscillators and in billiard channels driven by boundary thermostats. For heat flows in many-particle fluids, basic molecular dynamics approaches have been described in Refs. [Hoo75; Hoo99; All87]. The application of stochastic boundaries for modeling such heat flows was proposed in Refs. [Ten82; Cic80]. A very efficient method is the Evans heat flow algorithm, which is the heat conduction analogue of the SLLOD shear flow algorithm discussed above [Eva90b; Sar98; Lep03]: By assuming the validity of linear response theory and starting from Green-Kubo formulas, a synthetic algorithm is constructed in which a fictitious field replaces the temperature gradient as the force driving the heat flux. The resulting system is homogeneous and, in order to stabilize the dynamics, coupled to a Gaussian thermostat. In the following we will mostly focus onto shear flows, heat flows will be touched upon only briefly. Our results were obtained from a system consisting of a large number of hard disks that collide elastically with each other. This hard disk fluid is sheared and thermostated at the boundaries. The shear is modeled by moving walls, or sliding boundaries, imposing a local shear force onto the particles. In order to remove energy from the system we apply TDS at the walls. 15.5.2

Shear and heat flows thermostated by deterministic scattering

We consider N hard disks confined between two flat walls, as described above for the Maxwell daemon model. However, here we change the scattering rules at both walls according to Eqs. (13.1), (15.3) and (15.4), cf. Fig. 15.1 and the respective discussion. Associating different temperatures T u and T d with the upper and the lower wall, respectively, enables us to study the case of a heat flow from a hot to a cold reservoir [Wag99]. In simulations the length of the box was chosen to be L = 28, hence

May 8, 2007

4:1

292

World Scientific Book - 9in x 6in

klages˙book

Stochastic and deterministic boundary thermostats

the volume fraction occupied by N = 100 hard disks of radius r = 1/2 is ρ = 0.1. The temperature gradient was set according to T d = 1 with T u = 1.5 or T u = 2. In both cases the temperature profile was found to be approximately linear, apart from boundary effects, and the kinetic energy of the particles was equipartitioned between the two degrees of freedom. As discussed in Section 15.4.1, assigning T u and T d of Eq. (15.3) to be the reservoir temperatures yields temperature jumps at the walls. This can be avoided by redefining the two reservoir temperatures from a smooth extrapolation of the bulk temperature to the boundaries under the assumption of equipartitioning of energy at the walls, which is our choice here. Computing the thermal conductivity from simulations, one finds good agreement with Enskog’s kinetic theory, particularly in the hydrodynamic limit [Ris96; Gas71]. This shows that TDS is able to generate a heat flow in a linear response regime, in full agreement with hydrodynamics. Next, the identity between phase space contraction and thermodynamic entropy production was checked for this model. We do not go into too much detail at this point, see Ref. [Wag99] for an explicit discussion. However, simulations showed good agreement between both quantities, especially in the hydrodynamic limit, so in this case the identity was confirmed. The in and outgoing fluxes at the walls were furthermore approaching local thermodynamic equilibrium in the hydrodynamic limit. Choosing T u = T d = 1, we now model the case of a shear flow. For this purpose we let the walls move with constant velocities opposite to each other. At a collision with a wall a hard disk then experiences the shear force Sd (vx , vy ) = (vx + d, vy )

(15.10)

with d being the positive (negative) velocity of the upper (lower) wall. The main problem is to suitably combine Sd with the thermostating rules Eqs. (13.1), (15.3) and (15.4). In Ref. [Wag99] three options were explored. We first discuss Sd ◦ T −1 ◦ M ◦ T ◦ Sd (vx , vy ) ; vx ≥ −d 0 0 . (15.11) (vx , vy ) = Sd ◦ T −1 ◦ M−1 ◦ T ◦ Sd (vx , vy ) ; vx < −d Requiring time-reversibility enforces us to apply Sd symmetrically before and after the thermalization. Performing simulations for d = 0.05 and d = 0.1, the analysis proceeded along the same lines as for the heat flow: It was found that the system exhibits a linear shear profile for the average velocity parallel to the x-direction and a quadratic temperature profile between the

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

293

15.5. Hard disk fluid under shear and heat flow Table 15.1 Results for the fraction ηexp /ηth in a sheared fluid consisting of N hard disks. The shear viscosity ηexp was obtained from computer simulations, whereas ηth was calculated from Enskog’s kinetic theory. The system is driven by boundaries moving with velocities ±d and thermostated by deterministic scattering at these walls, see Eq. (15.11). The values are from Ref. [Wag99] .

d=0.05 d=0.1

N=100

N=200

N=400

N=800

0.9616 0.9702

0.9904 1.001

1.0081 1.0226

1.0382 1.0232

walls, both as expected from hydrodynamics. Again, there were temperature jumps at the walls if the reservoir temperature Tr in nonequilibrium was identified with the equilibrium reservoir temperature T . A comparison of the viscosity computed from simulations with Enskog’s theory [Gas71; Che95b; Che97] yielded good agreement in the hydrodynamic limit, see Table 15.1. Thus, the combined mechanism of boundary thermostating and shear Eq. (15.11) generated a shear flow exhibiting a linear response regime in agreement with hydrodynamics. It is now interesting to study the relation between phase space contraction and entropy production for this model. The average phase space contraction rate κb at the walls is defined in complete analogy to Eq. (15.6) as κb :=< ln

|dvx0 dvy0 dx0 dy 0 | > |dvx dvy dxdy|

.

(15.12)

Calculating the Jacobian determinant from the collision rules Eq. (15.11) yields u/d

κb

=

1 02 v + vy02 − vx2 − vy2 − < vx0 >2 + < vx >2 2T u/d x 1

+ u/d < vx0 >2 − < vx >2 −2d(vx0 + vx ) , (15.13) 2T

where the angular brackets denote ensemble averages over moving particles. The temperatures T u/d are the reservoir equilibrium temperatures at the upper (lower) walls defined by the transformation T in Eq. (15.11). Eq. (15.13) was already decomposed such that it can be compared to the Clausius form of entropy production in terms of the average heat transfer at the walls, dS = −

1 u/d

2Tr

vx02 + vy02 − vx2 − vy2 − < vx0 >2 + < vx >2

.

(15.14)

May 8, 2007

4:1

World Scientific Book - 9in x 6in

294

Stochastic and deterministic boundary thermostats u/d

Here Tr stands for the reservoir temperature in nonequilibrium computed from the assumption that there is no temperature slip at the walls. Obviously, apart from the minus sign dS is formally identical to the first bracketed term in Eq. (15.13). As in Section 15.4.1, the issue is whether one may identify the reservoir equilibrium temperature T u/d with the reservoir u/d temperature Tr under nonequilibrium conditions. Irrespective of this problem the second term in Eq. (15.13) yields a contribution to the phase space contraction at the walls that has nothing to do with the thermodynamic entropy production of Eq. (15.14). One may inquire whether this term has any physical interpretation. For the scattering rules Eq. (15.11) the average over in and outgoing velocities parallel to the walls uw := (< vx0 > + < vx >)/2 is not necessarily identical to d. By assuming that it is at least some linear function of d, the second line in Eq. (15.13) yields some correction of order d2 . This contribution might be interpreted as a phase space contraction due to a friction parallel to the walls and may be thought of representing properties of a wall like roughness, or an anisotropy of the wall scatterers. In this respect our shear flow collision rules are very close again to the modifications that led to the non-ideal Gaussian and Nos´e-Hoover thermostats discussed in Sections 14.1 and 14.2. Note that for the latter types of thermostats the identity was destroyed due to what we called a non-ideal coupling between subsystem and reservoir. That is, we generalized the way subsystem and reservoir interact with each other by taking into account additional terms going beyond standard Gaussian and Nos´e-Hoover frictional forces. Similarly, the above collision rules allow one to incorporate further distinctive features into a microscopic subsystem-reservoir coupling. We conclude that such generalizations may lead to phase space contraction rates which, for general nonequilibrium situations, are indiscriminately more complicated than the (Clausius) entropy production expected from thermodynamics. Only close to equilibrium is an identity between phase space contraction and entropy production approximately recovered. That generally the identity between phase space contraction and entropy production does not hold is quantitatively assessed by Table 15.2. Here computer simulation results are presented for the fraction −dS/κb of the entropy production dS, Eq. (15.14), divided by the average phase space contraction rate κb , Eq. (15.13). Clearly, the absolute values of both quantities are not the same. The mismatch becomes even worse in the hydrodynamic limit.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

295

15.5. Hard disk fluid under shear and heat flow Table 15.2 Computer simulation results for the fraction −dS/κb in a sheared fluid consisting of N hard disks. The system is driven by boundaries moving with velocities ±d and thermostated by deterministic scattering at these walls, see Eq. (15.11). The entropy production dS was numerically computed from Eq. (15.14), the average phase space contraction rate κb from Eq. (15.13). The values are from Ref. [Wag99].

d=0.05 d=0.1

N=100

N=200

N=400

N=800

0.6761 0.6457

0.5882 0.5761

0.5023 0.4934

0.4230 0.4275

The hydrodynamic entropy production in the bulk [Che95b; Che97] was also computed from the simulations. The result was again compared to the entropy flux across the walls in terms of the Clausius entropy yielding good agreement by approaching the hydrodynamic limit. In order to sort out the possibility that the non-identity specifically depends on the specific choice of the baker map in the collision rules Eqs. (15.11), the same simulations were carried out by replacing the baker with the standard map mentioned in Section 15.2, which was tuned such that its dynamics was approximately hyperbolic. However, instead of recovering the identity the mismatch between phase space contraction and entropy production got even worse [Wag99]. A more microscopic understanding about the origin of this inequality can be obtained by looking at the velocity distributions of the in and outgoing particles at the walls. Fig. 15.6 shows that the two velocity components coming in from the bulk are nice Gaussian distributions at the wall. Indeed, the outgoing velocity distribution parallel to y is essentially indistiguishable from its ingoing counterpart. However, in sharp contrast to that the outgoing velocity distribution parallel to x exhibits an irregular structure with some discontinuities. The explanation is that, according to the collision rules Eqs. (15.11), there will always be more out than ingoing particles with vx ≥ d at the upper wall (and with vx ≤ d at the lower wall). Thus, due to normalization the two Gaussian halves in Fig. 15.6 (b) will never match to a full Gaussian even in the hydrodynamic limit. Correspondingly, the associated in and outgoing fluxes at the walls will never come close to a local thermodynamic equilibrium. This motivates checking for the importance of local thermodynamic equilibrium at the walls in order to obtain an indentity between phase space contraction and entropy production. For this purpose a second set of time-reversible collision rules was constructed [Wag99] yielding local ther-

4:1

World Scientific Book - 9in x 6in

296

klages˙book

Stochastic and deterministic boundary thermostats

0.45

0.6

(a)

0.4

(b) 0.5

0.35 0.4

ρ

ρ

0.3 0.25

0.3

0.2 0.15

0.2

0.1 0.1 0.05 0 −5

−4

−3

−2

−1

0

vx

1

2

3

4

0 −5

5

0.9

−4

−3

−2

−1

0

vx

1

2

3

4

5

0.9

(c)

0.8

(d)

0.8

0.7

0.7

0.6

0.6

0.5

0.5

ρ

ρ

May 8, 2007

0.4

0.4

0.3

0.3

0.2

0.2

0.1 0 0

0.1

0.5

1

1.5

2

2.5 vy

3

3.5

4

4.5

5

0 0

0.5

1

1.5

2

2.5

vy

3

3.5

4

4.5

5

Fig. 15.6 Velocity distributions ρ at the walls for a sheared fluid consisting of N hard disks. The system is driven by boundaries moving with velocities ±d and thermostated by deterministic scattering at these walls, see Eq. (15.11). Shown are results for the velocity components vx parallel and vy perpendicular to the upper wall of the sheared system. (a) and (c) depict the distributions of velocities before the collisions with the wall, (b) and (d) the same after the collisions. The results are from Ref. [Wag99].

modynamic equilibrium at the walls in the hydrodynamic limit. However, already the equation for the average phase space contraction rate was, as before, not identical to the entropy production Eq. (15.14), not even in the hydrodynamic limit. Computer simulations confirmed that, quantitatively, there is no identity either. Hence, these collision rules establish no identity, even in the case of local thermodynamic equilibrium at the walls. A third set of collision rules was thus invented, again generating local thermodynamic equilibrium at the walls, but now by breaking timereversal symmetry [Wag99]. In this case both the analytical expressions and computer simulations showed an identity between phase space contraction and entropy production in the hydrodynamic limit. Consequently, time-reversibility cannot be a necessary condition for an identity between phase space contraction and entropy production. This was already stated in Refs. [Che95b; Che97; Nic98; Dae99] but seems to be at variance with conclusions drawn from the analysis of multibaker maps [Vol97; Bre98; Vol98;

May 8, 2007

4:1

World Scientific Book - 9in x 6in

15.5. Hard disk fluid under shear and heat flow

klages˙book

297

T´el00; Vol02]; see also Ref. [Mae03] for a discussion of the relation between time-reversibility and entropy production on the basis of Gibbs states. Typically, the existence or non-existence of the identity will have direct consequences for the validity of the Lyapunov sum rule as it was discussed in Sections 11.2.2, 12.3.1, 14.1 and 14.2. An exception to the rule was the driven periodic Lorentz gas amended by TDS. Here the identity was ambiguous, however, formally a Lyapunov sum rule was recovered, see Section 15.4.1. In other words, the identity is a sufficient condition for the validity of the Lyapunov sum rule but not a necessary one. For the heat flow case outlined above the identity holds, hence there is a Lyapunov sum rule. On the other hand, a numerical computation of the Lyapunov spectrum shows [Wag00] that there is no conjugate pairing of Lyapunov exponents, as expected for a system that is thermostated at the boundaries. A computation of the Kaplan-Yorke dimension corroborates the existence of a fractal attractor. It might be interesting to construct collision rules that do not yield an identity for a heat flow, which should be possible along the lines as performed for the shear flow. As far as the latter is concerned, for sake of comparison we first state the Lyapunov sum rule for the SLLOD equations. Here the identity holds and the shear viscosity η can be computed to [Eva90a; Coh95] T X λi (γ) , (15.15) η(γ) = − 2 2 L γ i

where T is the temperature of the thermal reservoir, respectively the temperature pre-determined in the Gaussian thermostat, γ := dvx (y)/dy stands for the shear rate and L is again the system size. For TDS this relation may or may not be different, depending on whether or not the identity holds. Let us consider the model for a boundarydriven shear flow Eqs. (15.11) discussed above, for which there is no identity. By using the hydrodynamic relation dS = L2 γ 2 η/Tr and Eqs. (15.13), (15.14) one can nevertheless relate the viscosity of this model to the Lyapunov exponents arriving at ! X T 2 η(γ) = − 2 2 λi (γ) + O(d ) . (15.16) L γ i

Here we have abbreviated the second line in Eq. (15.13) by O(d2 ). Note that, in contrast to the non-ideal Gaussian and Nos´e-Hoover thermostats of Sections 14.1 and 14.2, in this case we could at least establish an explicit relation between viscosity and Lyapunov exponents. Still, the viscosity is

4:1

World Scientific Book - 9in x 6in

298

klages˙book

Stochastic and deterministic boundary thermostats

7.5 d=0.0 d=0.25 d=0.5

5.0

2.5 λ

May 8, 2007

0.0

−2.5

−5.0

−7.5

0

10

20

30

40

50

60

70

i

Fig. 15.7 Spectra of Lyapunov exponents λ for a sheared fluid consisting of 36 hard disks at a volume fraction of ρ ' 0.47. The system is driven by boundaries moving with velocities ±d and thermostated by deterministic scattering at these walls, see Eq. (15.11). Shown are results for three different driving forces as indicated in the figure. The lines in the middle represent the sums of conjugated pairs of Lyapunov exponents. The figure is from Ref. [Wag00] .

no simple function of Lyapunov exponents anymore thus providing a counterexample against a universality of the Lyapunov sum rule in thermostated interacting many-particle systems far from equilibrium. Fig. 15.7 shows Lyapunov spectra for the shear flow defined by Eq. (15.11) in the case of a system with 36 hard disks at a volume fraction of ρ ' 0.47, where the particles are in a quadratic box of length L ' 7.746. Depicted are results from simulations at three different shear forces d [Wag00]. The overall shape of these spectra is quite typical for sheared many-particle fluids [Mor89a; Mor89b; Pos89; Eva90a; Sar92; Del97a; Sea98]. However, note the asymmetry particularly between the largest positive and negative exponents. The sum of pairs of respectively ordered exponents is also included in the figure showing that, as in the case of the heat flow, there is no conjugate pairing rule for the Lyapunov exponents. Finally, we present results for the Kolmogorov-Sinai entropy of this system, which, according to Pesin’s theorem, is identical to the sum of positive Lyapunov exponents. In Fig. 15.8 this quantity is plotted as a

4:1

World Scientific Book - 9in x 6in

klages˙book

299

15.5. Hard disk fluid under shear and heat flow

5.2

5.1

hKS(γ)/Ν

May 8, 2007

5

4.9

4.8

0

0.1

0.2

0.3

0.4

0.5

γ

Fig. 15.8 The Kolmogorov-Sinai entropy hKS per particle as a function of the shear rate γ for a fluid consisting of N = 36 hard disks at a volume fraction of ρ ' 0.47. The system is driven by boundaries moving with velocities ±d and thermostated by deterministic scattering at these walls, see Eq. (15.11). The figure is from Ref. [Wag00].

function of the shear rate γ for the same model parameters as in Fig. 15.7. The figure may be compared to the respective result for the driven periodic Lorentz gas thermostated by deterministic scattering, see λ1 in Fig. 15.5. There is quite an analogy in that hKS exhibits a global minimum by increasing the driving force. Curiously, for the non time-reversible collision rules of the third model discussed above there is no such local minimum. Furthermore, respective results for the heat flow of both models do not exhibit a local minimum either, see Ref. [Wag00] for further details. Computations of the Kaplan-Yorke dimension for the first and the third shear flow model yield again evidence for the existence of fractal attractors. For a total of 144 phase space variables the associated loss of phase space dimensionality went up to about 6 at shear rates around γ = 0.5. Assessing the magnitude of such dimensionality losses for Gaussian and Nos´e-Hoover thermostats restricted to boundary layers became an active topic recently, after it was realized that the dimensionality loss may significantly exceed the number of constrained variables [Hoo02b; Aok03; Kus02; Pos04b].

May 8, 2007

4:1

World Scientific Book - 9in x 6in

300

15.6

Stochastic and deterministic boundary thermostats ∗

Summary

(1) Modeling a stochastic thermal reservoir at the boundaries of a system is well-known in the form of stochastic boundary conditions. The precise collision rules were briefly reviewed at the beginning of this chapter. (2) These stochastic collision rules can easily be made deterministic and time-reversible by using a time-discrete map. The map should exhibit certain properties in order to enable a proper thermalization. The resulting deterministic and time-reversible boundary thermostat was called thermostating by deterministic scattering. (3) Both types of boundary thermostats can be derived from first principles in three steps: Firstly, the precise functional forms of the equilibrium velocity distribution functions for arbitrarily many reservoir degrees of freedom are required, as they have been calculated earlier according to elementary equilibrium statistical mechanics. Secondly, one must define some coupling between subsystem and thermal reservoir. One then has to check whether the chosen coupling reproduces the correct velocity distributions. For arbitrary couplings this will typically not hold. In this case one should enforce that the correct target velocity distributions are attained by suitably modifying the collision rules. For a nonequilibrium situation one needs to consider the limit of infinitely many reservoir degrees of freedom. Following this construction step by step sheds light onto the origin of phase space contraction in thermostated systems and corroborates a heuristic argument outlined previously. (4) As an example for the application of a deterministic boundary thermostat, again the driven periodic Lorentz gas has been analyzed. Here the identity between phase space contraction and entropy production is ambiguous and depends on how one defines the temperature of the thermal reservoir. In any case, there is a Lyapunov sum rule for the electrical conductivity. The attractor for the driven periodic Lorentz gas deterministically thermostated at the boundaries exhibits a fractal structure that is quite analogous to the previous ones resulting from ideal and non-ideal Gaussian and Nos´e-Hoover thermostats. For stochastic boundaries this structure is not present suggesting that the corresponding attractor is possibly not fractal. In contrast to previous thermostated driven Lorentz gases the bifurcation diagram for the deterministic boundary thermostat is phase space covering for all values of the electric field

klages˙book

May 8, 2007

4:1

15.6.

World Scientific Book - 9in x 6in

∗ Summary

klages˙book

301

strength. The electrical conductivity shows a nonlinear field dependence that is also different from the one obtained for other thermostats, being smoother than in previous cases. Still, there are irregularities on fine scales. As is typical for boundary thermostated systems, there is no conjugate pairing rule of Lyapunov exponents for this model. The KolmogorovSinai entropy related to the single positive Lyapunov exponents shows an interesting, non-monotonic behavior as a function of the field strength, which has not been observed for bulk thermostats. (5) We have summarized important schemes that model shear flows for interacting many-particle systems, such as sliding boundaries, LessEdwards ‘sliding brick’ boundary conditions, an application of the Gaussian thermostat to shear in the form of SLLOD equations of motion and refinements leading to PUT thermostats. For ideal Gaussian thermostats applied to shear and heat flows there hold the usual relations between chaos and transport such as the identity between phase space contraction and entropy production and the Lyapunov sum rule. An alternative scheme was invented in the form of Maxwell daemon boundaries and was argued to share the same properties. We also briefly outlined some conventional models of heat flows. We then focused onto a hard disk fluid under shear and heat flow thermostated by deterministic scattering. For a heat flow this system exhibited linear response and yielded again an identity between phase space contraction and entropy production. For a shear flow three different versions of a deterministic boundary thermostat were discussed. All of them generated a linear response regime. Two of them did not yield an identity and consequently no Lyapunov sum rule. One of them was not time reversible but reproduced the identity. Two of the shear flow models were analyzed in further detail showing that there was no conjugate pairing rule, as expected for thermostats acting at the boundaries. Both models exhibited a fractal attractor. The latter properties are shared by the heat flow model. As in the case of the respectively thermostated driven periodic Lorentz gas, the Kolmogorov-Sinai entropy displayed an interesting non-monotonic behavior as a function of the shear rate.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

This page intentionally left blank

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Chapter 16

Active Brownian particles and Nos´ e-Hoover dynamics

With this chapter we return to the problem of Brownian motion reviewed in Section 10.2, which we used to introduce the concept of thermal reservoirs. In Section 16.1 we give Brownian motion a different touch by relating it to observations in biology. Our motivation derives from experiments on migrating biological cells, which show Brownian motion-like trajectories. However, the fact that biological entities like living cells are self-propelled, in contrast to passively driven tracer particles in a fluid, suggests that amendments to ordinary Langevin equations are necessary in order to describe such living systems. This motivates the formulation of active Brownian particles in terms of generalized Langevin equations, which we review in Section 16.2. We argue that limiting cases of such equations contain ingredients that trace back to the conventional Nos´e-Hoover dynamics discussed in Chapter 12. This may not come too much as a surprise if one reinterprets the thermal reservoir modeled by Nos´e-Hoover dynamics as the internal energy of a migrating cell. That is, the thermal reservoir is now part of a biological entity providing energy for its migration process. In Section 16.3 we focus on essential features of the velocity distribution functions of active Brownian particles. We argue that the appearance of bimodal profiles may be understood on the basis of Nos´e-Hoover dynamics. By employing the general functional forms of the equilibrium velocity distribution functions calculated in Section 10.3 we finally state some necessary conditions for the emergence of such bimodal structures.

303

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

304

klages˙book

Active Brownian particles and Nos´ e-Hoover dynamics 100

y [µm] 50

0

-50

-100 -100

-50

0

50

x [µm]

100

Fig. 16.1 Experimental results for the migration of renal epithelial MDCK-F (MadinDarby canine kidney) cells on a substrate. 20 single living cells were isolated, with their initial positions being shifted to the same origin in the figure. The positions x and y were extracted from microscopic phase contrast images taken at time intervals of one minute acquired over a period of 100 minutes [Die05; Die07].

16.1

Brownian motion of migrating cells?

Langevin’s theory of Brownian motion presupposes that a Brownian particle is passively driven by collisions from the surrounding particles. This input of energy is removed from the system by Stokes friction in the bulk leading to a balance between molecular stochastic forces and friction as represented by the fluctuation-dissipation theorem. Surprisingly, the Langevin equation was also used to describe the migration of biological cells. For theory and experiments on single living cells such as granulocytes1 and other types of crawling cells see Refs. [Dun87; Fra90; Sto91; Sch93; Har94; Die05; Die07]. In Refs. [Gru94; Rie00; Upa01; Dia06] migrating cells interact with more complex environments. In order to illustrate that, at first view, cells indeed seem to move like Brownian particles Fig. 16.1 depicts experimentally measured trajectories of isolated epithelial2 cells migrating on a substrate. However, careful analysis of this data shows that in detail the dynamics is much more complicated than simple Brownian motion; see Refs. [Har94; 1 These cells are attracted to sites of inflammation to destroy microorganisms and invaded cells. 2 An epithelial cell is one of the closely packed cells forming the epithelium, which is a thin layer of tissue that covers organs, glands, and other structures within the body.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

16.1. Brownian motion of migrating cells?

klages˙book

305

Rie00; Upa01; Dia06] for related results. It turns out that these cells exhibit anomalous dynamics, cf. Chapter 6, on different time scales [Die05; Die07]. Understanding these experimental results thus requires generalizing ordinary Langevin equations. A promising approach employs friction coefficients with power-law memory kernels, which can be defined in terms of fractional derivatives related to the ones discussed in Section 6.2. Such a modeling does not yet reveal the microscopic origin of the observed generalization of Brownian motion. This leads to another theory, which starts from the fact that biological entities such as cells, bacteria or nematodes3 do not comply with the physical assumption of being particles that move only because of stochastic environmental forces. In order to sustain their motion these organisms need some external supply of energy that may be stored internally, and one may think of some metabolic activity converting this energy into motion. This rough picture motivated amending conventional Langevin equations by using velocity-dependent friction coefficients arriving at active Brownian particles, sometimes also called Brownian agents, which are able to take up energy from the environment, to store it onto internal degrees of freedom, and to convert it into kinetic energy [Sch98; Ebe99; Til99; Erd00]. The earliest models of active Brownian particles featured a response of these entities to environmental changes yielding physicochemical structure formation on a macroscopic scale [Sch94; SG95]. Other variants were used to describe, for example, self-organization in networks, the trunk trail formation of ants and the movement of pedestrians, see Refs. [Sch03b; Ebe05] for introductions and overviews. More recent studies focused on confined systems with rotational excitations [Erd02; Ebe04]. Interacting active Brownian particles yielded explanations for experimental results of swarming behavior in daphnia [Erd04; Mac04]. Similar collective behavior has been observed in biological systems like schools of fish, bacterial colonies, slime molds or flocks of birds suggesting further applications of this concept. In the following section we introduce the formalism of active Brownian particles and discuss crosslinks with Nos´e-Hoover dynamics.

3 Nematodes are elongated, microscopic cylindrical worms parasitic in animals or plants, which are free-living in soil or water.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

306

16.2

klages˙book

Active Brownian particles and Nos´ e-Hoover dynamics +

Moving biological entities as active Brownian particles

A common ansatz for modeling active Brownian particles starts from the generalized two-dimensional Langevin equation [Sch98; Ebe99; Til99; Erd00] r˙ = v v˙ = −αv + F(t)

,

(16.1)

where F is white noise, that is, a δ-correlated stochastic force of strength S with zero mean value, < F(0)F(t) >= Sδ(t) .

(16.2)

The noise strength S can be straightforwardly associated with the friction coefficient α according to the fluctuation-dissipation theorem Eq. (10.2) and with the diffusion coefficient D by using the Einstein relation Eq. (10.3), T S =α= . (16.3) 2T D Let us now assume that there is a generalized friction coefficient α in form of α = α0 − de .

(16.4)

Here α0 stands for ordinary Stokes friction, the variable e denotes the internal energy of the moving entity, and the constant d > 0 yields the rate of conversion from internal into kinetic energy. e is in turn obtained from a suitable balance equation for which different choices have been discussed. A general ansatz, denoted in the following as case 1, reads [Sch98; Ebe99; Til99] 1 (16.5) e˙ = (q − e(c + dv 2 )) . µ In this equation q ≥ 0 describes a constant take-up of energy, which generally may be space-dependent, and c models some constant internal loss of energy. Note that d could be velocity-dependent, however, in Eq. (16.5) we stick to the most simple assumption that the prefactor of e is only quadratic in v. We remark that in Eqs. (16.4), (16.5) we use the same d, which may be justified by an additional balance equation between generalized friction and internal energy [Sch98; Ebe99]. µ determines the rate of change of the internal energy e and thus plays a role similar to τ 2 in the Nos´e-Hoover thermostat, see Eq. (12.10).

May 8, 2007

4:1

16.2.

World Scientific Book - 9in x 6in

+ Moving

biological entities as active Brownian particles

klages˙book

307

In the case of very fast feedback µ → 0 Eq. (16.5) boils down to the fixed point case 2, q e= . (16.6) c + dv 2 Combining this equation with Eq. (16.4) yields the velocity-dependent friction coefficient v 2 − v02 (16.7) α = α0 q 2 2 α0 + v − v 0 with v02 := q/α0 − c/d. A special case of this type of friction is obtained by assuming v 2 v02 , or alternatively c/d >> v 2 , in the denominator only leading to case 3 α = α1

v 2 − v02 v02

(16.8)

with α1 := −α0 +qd/c. This equation is sometimes called the Rayleigh-type model, because a similar ansatz for a velocity-dependent friction coefficient has been introduced by Rayleigh in the context of the theory of sound [Erd00]. Eq. (16.8) looks strikingly similar to Eq. (12.10) defining the Nos´eHoover thermostat. However, note that the former represents the explicit functional form for α, whereas the latter is a differential equation with a time-derivative for α on the left hand side. Motivated by this formal similarity, one may ask to what extent models of active Brownian particles can exhibit dynamical systems properties as outlined in Chapter 12 for the Nos´e-Hoover thermostat. A fundamental difference between active Brownian particles and Nos´e-Hoover thermostats is that the former have been introduced for modeling the energetic aspects of moving biological entities on a microscopic level, without knowing in advance about the velocity distributions generated by these models. In contrast to that, Nos´e-Hoover thermostats have been designed for generating, under suitable conditions, canonical velocity distributions, see Section 12.2.1. A further difference already resulting from the Stokes friction coefficient is that the equations of motion for active Brownian particles are irreversible. But even more, according to ordinary Langevin dynamics these equations include noise, which is in contrast to the concept of Nos´e-Hoover and other deterministic thermostats. In order to bring active Brownian particles and deterministic thermostats closer together, thus making the three models introduced above more amenable to methods of dynamical systems theory, one may replace the white noise term by deterministic chaos, as we already briefly described

May 8, 2007

4:1

World Scientific Book - 9in x 6in

308

Active Brownian particles and Nos´ e-Hoover dynamics

in Section 10.2. Another option is to simply eliminate the noise term and to generate deterministic chaos by choosing a suitable potential such as the Lorentz gas geometry, see the beginning of Section 11.1. Yet we do not know anything about the velocity distributions for deterministic active Brownian particles, and there is still the lack of time-reversibility. Hence, there is no reason to expect an identity between phase space contraction and entropy production for these models to hold. Similarly, there is no reason why by default equipartitioning of energy should be fulfilled. Whether fractal attractors exist for this class of dissipative dynamical systems and whether Lyapunov sum rules and conjugate pairing applies, as they typically do for ideal Nos´e-Hoover thermostats in nonequilibrium, are further interesting questions to be studied. However, we remark that for so-called canonicaldissipative systems containing active Brownian particles as a special case generalized Hamiltonian equations have been constructed, which remind one of the generalized Hamiltonian formalism developed for Gaussian and Nos´e-Hoover thermostats as outlined in Section 12.3.2 [Ebe00].

16.3

+

Bimodal velocity distributions and Nos´ e-Hoover dynamics

Let us now summarize what is known about the velocity distribution functions for the three different cases of active Brownian particles discussed above. In this framework we shall also discuss connections with the Nos´eHoover thermostat. In all cases, the most obvious solution for the velocity distribution is obtained at e = const. This is the Stokes limit of constant friction for which the velocity distribution is well-known to be purely canonical, see Section 10.3. For case 1, Eqs. (16.1), (16.4) and (16.5), not much appears to be known regarding general combinations of parameters in Eq. (16.5). For the special case 2 the situation is much better: Here the analytical solution for the velocity distribution function can be obtained from solving the corresponding Fokker-Planck equation and reads [Til99; Erd00] q/S α dv 2 0 exp − v 2 , (16.9) ρ(v) = C 1 + c S where C is a normalization constant. Interestingly, as shown in Fig. 16.2 this velocity distribution displays a transition from canonical to microcanonical-like behavior under variation of d or related parameters.

klages˙book

May 8, 2007

4:1

16.3.

World Scientific Book - 9in x 6in

+ Bimodal

velocity distributions and Nos´ e-Hoover dynamics

klages˙book

309

Fig. 16.2 Transition from canonical to bimodal velocity distributions in the model of active Brownian particles called case 2, see Eq. (16.6) in combination with Eqs. (16.1), (16.4). The figure displays the analytical solution ρ(v) = P 0 (v) of Eq. (16.9). The parameters are c = 1, q = 10, S = 4, α0 = 2 and (a) d = 0.07, (b) d = 0.2, (c) d = 0.7. The figure is from Ref. [Erd00].

Such bimodal velocity distributions exhibiting a dip at the place of the former maximum of the canonical distribution appear to be quite typical for active Brownian particles [Erd00; Sch01; Ebe01; Erd02; Ebe04]. Fig. 16.2 should be compared to our previous Fig. 12.1 displaying velocity distributions for the Nos´e-Hoover thermostated driven periodic Lorentz gas. Apart from the asymmetry due to the external electric field the scenario is completely the same as in Fig. 16.2: By tuning the reservoir coupling parameter τ the Nos´e-Hoover velocity distribution clearly shows a transition from canonical to microcanonical-like. The advantage of the Nos´e-Hoover thermostat is that due to its construction the origin of these

May 8, 2007

4:1

310

World Scientific Book - 9in x 6in

Active Brownian particles and Nos´ e-Hoover dynamics

different functional forms is rather well-understood, as already outlined in Section 12.2.2: In the limiting case of τ → 0 the Nos´e-Hoover thermostat generates a microcanonical velocity distribution, whereas in the limit of Stokes friction for τ → ∞ it yields a canonical one. Hence, there must be a transition between canonical and microcanonical for intermediate τ parameters. This transition is represented by a superposition of these two different velocity distributions. For Fig. 16.2 and case 2 a very similar explanation might thus be employed: The canonical velocity distribution is probably related to passive motion of the Brownian particles driven by the noise term in Eq. (16.1). The appearance of microcanonical-like features must then be due to the deterministic, velocity-dependent friction coefficient appearing in the equations of motion. This interplay may result in a transition from a canonical to a microcanonical(-like) distribution which is quite analogous to the one displayed by the Nos´e-Hoover system. In contrast to Nos´e-Hoover, however, for active Brownian particles this transition may rather reflect a change from noise-driven, passive motion to more active dynamics as determined by the deterministic, nonlinear part of the equations of motion. On the other hand, we also have this formal analogy between the Nos´eHoover equations of motion and at least case 3 of active Brownian particles as noted in the previous section. If this link could be made more explicit one might argue that, irrespective of stochastic contributions in the equations governing active Brownian particles, there should also be a Nos´e-Hoover like transition in the velocity distributions, which is purely generated by a tuning of the deterministic, velocity-dependent friction coefficient. This leads us to the conclusion that in the case of active Brownian particles there may actually be two rather analogous transitions between canonical and microcanonical distributions, one that is of a purely deterministic origin and another one that is due to the interplay between deterministic and stochastic forces. To what extent the one or the other is exhibited by the dynamics may then depend on the specific choice of the control parameters. We remark that in addition to the driven periodic Lorentz gas, the deterministic transition described above has recently been observed in a FermiPasta-Ulam chain of anharmonically coupled oscillators thermostated at the boundaries with Nos´e-Hoover chain thermostats [RB06]. An at first view analogous transition from unimodal to bimodal distributions has been reported for the stationary states of a nonlinear oscillator driven by L´evy noise [Che02; Che03b]. However, the detailed dynamical origin of this transition appears to be very different from the two scenarios discussed above

klages˙book

May 8, 2007

4:1

16.3.

World Scientific Book - 9in x 6in

+ Bimodal

velocity distributions and Nos´ e-Hoover dynamics

klages˙book

311

thus demonstrating that there may actually be a large number of different dynamical mechanisms generating such bifurcations in velocity distributions. In order to underline these arguments, and before discussing the velocity distribution function of case 3 in detail, let us analyze the transition between microcanonical and canonical distributions for the Nos´e-Hoover thermostat more explicitly. Note that in the following we consider a purely deterministic equlibrium dynamics without any external field. We recall that the equation governing the velocity-dependent Nos´e-Hoover friction coefficient α, Eq. (12.10), read v 2 − 2T . (16.10) τ 2 2T In the limit of τ → 0 the Nos´e-Hoover thermostat approaches the Gaussian one for which the kinetic energy is strictly kept constant at any time step, cf. Section 12.2.2. In equilibrium this Gaussian limit thus corresponds to v 2 = 2T and α = 0, see Section 11.1. We are now interested in the fluctuations of α for τ 1. For this purpose one may expand Eq. (16.10) linearly in time yielding α˙ =

v 2 − 2T ∆t + α(t) (τ → 0 , ∆t 1) . (16.11) τ 2 2T According to this equation, fluctuations of α around the fixed point α = 0 are roughly approximated by the first term on the right hand side, that way recovering Rayleigh friction Eq. (16.8). Note that in the other limit of τ → ∞ the friction coefficient α can be an arbitrary constant thus recovering Stokes friction, which in equilibrium corresponds to zero velocity. We thus have a Hopf-like bifurcation scenario [Ott93] from v = 0 for τ → ∞ to a limit cycle behavior with v = const. for τ → 0. On the basis of this approximation, let us study what happens in the limit of small but finite τ 1 for the equilibrium velocity distribution function Eq. (12.8) of Nos´e-Hoover reading v2 . (16.12) ρ(t, r, v, α) = const. exp − − (τ α)2 2T α(t + ∆t) =

The approximation Eq. (16.11) enables us to eliminate α(t) from the above equation yielding (v 2 − 2T )2 (∆t)2 v2 − ρ(v) = C exp − (16.13) 2T 4τ 2 T 2 2 v2 v − 2T = C δτ˜ exp − (˜ τ = τ /∆t 1) (16.14) 2T 2T

May 8, 2007

4:1

312

World Scientific Book - 9in x 6in

Active Brownian particles and Nos´ e-Hoover dynamics

For the second line we have used the definition of the δ-function as a series of exponentials in τ˜ [Rei65], where C is a normalization constant. We thus indeed arrive at the limiting case of a bimodal Nos´e-Hoover velocity distribution composed of a canonical and a microcanonical contribution. This solution helps to finally analyze the velocity distribution that was obtained for the active Brownian particle case 3 [Erd00; Erd02], α α2 4 1 2 (16.15) ρ(v) = C exp v − v S 2S with α2 := α0 d/c. The functional form matches nicely to Eq. (16.13), and there should be a respective decomposition into a canonical and a microcanonical component in analogy to Eq. (16.14). However, there is the subtlety that Eq. (16.15) yields the solution for a stochastic system as is represented by the strength S of the stochastic force. We therefore have to inquire whether this stochasticity is a crucial ingredient fundamentally distinguishing this distribution from the formally completely analogous but deterministic Nos´e-Hoover solution Eq. (16.13). Let us assume that in a suitable case 3 system both the fluctuationdissipation theorem and the Einstein relation are valid. According to Eq. (16.3), S is then a trival function of the temperature T and the diffusion coefficient D. The temperature T may be defined by assuming equipartitioning of energy, T =< v 2 > /2, where the right hand side is the average kinetic energy of the active Brownian particle. The diffusion coefficient D might be thought of as being generated deterministically. We therefore argue that Eq. (16.15) also holds in a purely deterministic setting of case 3 thus establishing the connection to the Nos´e-Hoover dynamics analyzed before. Under these conditions, case 3 must exhibit the very same transition in the velocity distributions as Nos´e-Hoover. For sake of completeness we remark that in the limit of small velocities and by expanding the prefactor of the velocity distribution Eq. (16.9) of case 2, Eq. (16.15) of case 3 is recovered again. Along these lines one may understand a transition such as the one depicted in Fig. 16.2 also on the basis of Nos´e-Hoover dynamics. To summarize, in all three cases of active Brownian particles discussed above there exists Stokes limit of constant friction yielding canonical velocity distributions irrespective of any stochastic contributions. We therefore conjecture that, in analogy to Nos´e-Hoover dynamics, for purely deterministic active Brownian particles there exists a completely analogous transition from canonical to microcanonical velocity distributions under suitable parameter variation yielding intermediate bimodal velocity distributions.

klages˙book

May 8, 2007

4:1

16.3.

World Scientific Book - 9in x 6in

+ Bimodal

velocity distributions and Nos´ e-Hoover dynamics

klages˙book

313

One choice for such a parameter variation should be tuning µ in case 1, but other combinations of parameters, in analogy to varying τ in Nos´e-Hoover dynamics, should be possible as well. For stochastic active Brownian particles defined in the framework of ordinary Langevin dynamics, in addition there appears to be a transition from a canonical velocity distribution function generated by the stochastic forcing to a microcanonical-like counterpart that, again, is due to a suitably tuned deterministic friction coefficient, see Fig. 16.2. This theoretical predicition still needs to be verified by computer simulations. Finally, we comment on two important necessary conditions limiting the range of existence of bimodal-like velocity distributions. For this purpose we refer again to the derivation of velocity distributions via projection from a microcanonical one outlined in Section 10.3. Let us assume that the velocity space of a moving particle has a dimension of d = ds , in order to be consistent with our previous notation. Let us furthermore assume that the full ds -dimensional velocity distribution of the moving particle is microcanonical as, for example, resulting from the application of a Gaussian thermostat, or in the limit of Nos´e-Hoover dynamics for τ → 0. The central formula is then Eq. (10.19) telling us that not for any dimensionality ds does a one-dimensional projection ρ(vx ) = ρ(v1 ) exhibit a minimum in its functional form. More precisely, Fig. 10.2 shows that for ds ≥ 3 the particle’s onecomponent equilibrium velocity distribution does not display a global minimum anymore. This eliminates any possibility for the generation of bimodal velocity distributions. That is, creating a transition from a one-dimensional projection of a microcanonical distrbution in ds ≥ 3 to a one-dimensional canonical distribution (by noise, or by tuning τ in the Nos´e-Hoover thermostat) cannot generate any non-monotonicities in such projected distributions. In other words, for moving particles with more than two degrees of freedom there cannot anymore be any transition between a projected microcanonical and a canonical one-component velocity distribution featuring bimodal intermediate functional forms. An analogous argument holds if additionally the dimensionality dr of an associated thermal reservoir can be changed as discussed, e.g., for thermostating by deterministic scattering in Section 15.3. For such a combination of subsystem and reservoir one can straightforwardly check [Kla], again by using Eq. (10.19) with d = ds , respectively with d = ds + dr , that nonmonotonic velocity distributions only appear if ds + dr ≥ 3. These two bounds put therefore quite a limit on the existence of bimodal velocity

May 8, 2007

4:1

World Scientific Book - 9in x 6in

314

Active Brownian particles and Nos´ e-Hoover dynamics

distributions. Our discussion thus shows how deterministic thermostats such as Gauss, Nos´e-Hoover and thermostating by deterministic scattering may help to understand the origin of velocity distributions in, at first view, quite different systems related to biological applications. We remark that bimodal velocity distributions generated by a nonlinear, velocity-dependent friction coefficient turned out to be crucial for understanding the rotational motion of swarms in biological systems [Ebe01; Sch03b; Erd04; Mac04]. They are also an important ingredient for predicting a new form of electric conduction in anharmonic Toda lattices mediated by dissipative solitons [Vel05].

16.4

∗

Summary

(1) Experiments on the migration of isolated biological cells show trajectories suggesting Brownian motion-like behavior. However, a closer look reveals profound deviations asking for generalizations of ordinary Langevin equations. One approach modeling the storage of internal energy of biological entities leads to the concept of active Browian particles. This theory has been widely applied to the biological and the social sciences. (2) Three fundamental cases of active Brownian particles were derived by replacing the ordinary Stokes friction of stochastic Langevin dynamics by a nonlinear velocity-dependent friction coefficient. The simplest version showed a striking similarity with the Nos´e-Hoover dynamics discussed before, irrespective of the fact that active Brownian particles and deterministic thermostats belong to the different classes of stochastic, respectively deterministic systems. (3) We then discussed the origin of bimodal velocity distributions both for deterministic Nos´e-Hoover dynamics and for stochastic active Brownian particles in thermal equilibrium. In the case of Nos´e-Hoover this transition is triggered by the tuning of the reservoir response time yielding a superposition between a microcanonical and a canonical distribution, as already mentioned in Section 12.2.2. Here we analyzed this scenario in more detail starting from the Nos´e-Hoover equations of motion. Based on a formal analogy of the underlying equations of motion, we concluded that the same scenario must be present for active Brownian particles.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

16.4.

∗ Summary

klages˙book

315

Due to their stochasticity, active Brownian particles appear to feature a second mechanism generating bimodal velocity distribution functions. This results from a superposition between a stochastically generated canonical distribution and a microcanonical one related to the action of the deterministic friction coefficient. Using results for equilibrium velocity distributions derived in Section 10.3, we finally stated two necessary conditions about the numbers of degrees of freedom for subsystem and reservoir, which must both be fulfilled in order to generate transitions under variation of control parameters that are characterized by bimodal velocity distributions.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

This page intentionally left blank

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

PART 3

Part 3: Outlook and conclusions

317

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

This page intentionally left blank

318

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Chapter 17

Further topics in chaotic transport theory

This chapter establishes crosslinks between the previous two parts of this book and other topics in the field of chaos and transport. Four research areas have been selected because of their importance and rapid development within the past few years: Fluctuation relations, under which we subsume both fluctuation theorems and work relations, are described in Section 17.1. Fluctuation theorems grew out of the thermostated dynamical systems approach to NSS [Eva93; Gal95a], and details of the thermostating mechanism play a crucial role for their validity. Work relations, on the other hand, generalize a relation between work and free energy, known from equilibrium thermodynamics, to nonequilibrium situations. A number of fluctuation relations have been verified in experiments thus bridging the gap between abstract models of thermal reservoirs and physical reality. Lyapunov modes are specific eigenmodes corresponding to the spectrum of Lyapunov exponents of many-particle fluids. They have first been discovered in equilibrium molecular dynamics computer simulations [Del96b]. As we summarize in Section 17.2, recent theoretical work yields quite a detailed understanding of these dynamical structures. Whether they exist in nonequilibrium still needs to be explored. Deriving Fourier’s law from first principles still forms a very open problem in the field of nonequilibrium transport and dynamical systems theory [Bon00b]. Section 17.3 focuses on heat conduction in classical onedimensional lattices, where recent work tries to reveal the necessary and sufficient conditions for the existence of Fourier’s law [Lep03]. Especially we discuss the rotating disk model, which became quite popular among mathematicians in order to rigorously analyze the foundations of heat conduction. 319

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

320

Further topics in chaotic transport theory

Normal deterministic diffusion is typically attributed to chaos in the sense of nonlinear dynamical instabilities with positive Lyapunov exponents. However, a controversial theoretical analysis of an experiment on microscopic chaos and diffusion [Gas98b] triggered a new line of research on non-Lyapunov chaotic diffusion [Det99b; Gra99]. At the heart of these studies, which we outline in Section 17.4, are polygonal billiards in which complicated topologies generate what may be called pseudochaotic diffusion [Zas03]. Applying a temperature gradient to billiard channels of this type sheds further light onto the origin of Fourier’s law. All these themes form large research areas in themselves. It is thus not our intention to provide full reviews of these topics. We rather give a flavor of some main ideas and outline perspectives of how to relate the knowledge covered by the previous two parts of this book to these active new research areas.

17.1

Fluctuation relations

What we denote by fluctuation relations is a collection of many different theorems, which are subtly related to each other. They emerged along two separate lines of research as fluctuation theorems and nonequilibrium work relations, respectively. We first outline the basic idea underlying fluctuation theorems and give a mathematical formulation of one of them. As an example, we apply this theory to the driven, thermostated periodic Lorentz gas. We proceed with a critical assessment of the theorem, point out generalizations of this concept and then elaborate on a second fundamental class of such theorems, which have been tested in experiments. We conclude this section by explaining fundamentals of work relations and discuss similarities and differences with respect to fluctuation theorems.

17.1.1

Entropy fluctuation in nonequilibrium steady states

In the pioneering work Ref. [Eva93], Evans, Cohen and Morriss studied a two-dimensional fluid of soft particles sheared by the SLLOD algorithm, cf. Section 15.5.1, in nonequilibrium molecular dynamics computer simulations. The focus was to compute the probability distribution ρ for the entropy production rate ξ(t) in a viscous NSS. Let us recall that the SLLOD equations define a dynamical system that is thermostated isoenergetically, cf. Section 11.1, by a Gaussian constraint, hence there is an identity be-

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

17.1. Fluctuation relations

klages˙book

321

tween the rates of phase space contraction and thermodynamic entropy production (subject to a minus sign). Ref. [Eva93] featured two crucial ideas: First of all, the authors investigated the ratio ρ(ξ)/ρ(−ξ) between the probability of positive and negative entropy production rates. Secondly, they introduced a coarse graining by computing time averages of ξ(t) over trajectory segments of length τ , yielding the family of coarse-grained probability distributions ρτ (ξ). Note that ρτ (ξ) converges to the average rate of entropy production for τ → ∞ which, according to the second law of thermodynamics, is always positive. Quantifying the asymmetry between positive and negative entropy production rates in ρτ (ξ) for different τ thus yields a refined view about the origin of the seond law in nonequilibrium situations. Motivated by a construction of the SRB measure, cf. Section 13.2, for SLLOD dynamics, the authors predicted the fundamental symmetry relation ρτ (ξ) = exp(τ ξ) (τ → ∞) (17.1) ρτ (−ξ) for the entropy production rate in a NSS, which they verified by computer simulations. The right hand side derives from the existence of an exponential rate of phase space contraction onto the underlying SRB measure, the sign change of ξ on the left hand side is related to time-reversal symmetry. Both are essential features of Gaussian and Nos´e-Hoover thermostated systems, see Chapter 13. Eq. (17.1) predicts that in the limit of large τ the probability to observe a positive entropy production rate ξ must be exponentially larger than that of the corresponding negative one. This formula may be considered as a generalization of the second law of thermodynamics, which particularly applies to the study of small, finite systems in a NSS at large enough times. From a practical point of view it may be useful for assessing the stability of systems on nanoscales working under nonequilibrium conditions, where fluctuations of negative entropy production may destroy their functionality [Eva02b]. We emphasize, however, that this analysis and the possibility of entropy production fluctuations that are smaller than zero are perfectly consistent with the second law, which holds only in the thermodynamic limit and for τ → ∞. 17.1.2

The Gallavotti-Cohen fluctuation theorem

Eq. (17.1) was put on mathematical grounds by Gallavotti and Cohen [Gal95a; Gal95b]. They started from the chaotic hypothesis (CH) which, in

May 8, 2007

4:1

World Scientific Book - 9in x 6in

322

klages˙book

Further topics in chaotic transport theory

its original formulation [Gal95b], states: A reversible many-particle system in a stationary state can be regarded as a transitive Anosov system for the purpose of computing the macroscopic properties of the system. This fundamental assumption generalizes Boltzmann’s ergodic hypothesis in summarizing general expectations on the chaotic nature of interacting manyparticle systems [Gal98]. The Anosov property implies that there exists an SRB measure, which can be constructed via Markov partitions. Along the lines of this construction and by using time-reversal symmetry, Eq. (17.1) could be proven by choosing for ξ the negative phase space contraction rate, ξ := −∇ · F(r, v), where F(r, v) holds for the equations of motion of a thermostated dynamical system in a NSS. The proof was based on the important assumption that the fluctuations of the phase space contraction rate ξ are bounded [Bon06a], i.e. |ξ| < ξ ∗ κ+

,

1 ≤ ξ∗ < ∞ ,

(17.2)

where ξ ∗ is a constant and κ+ := − < κp > the positive-valued average phase space contraction rate defined by Eq. (11.5). Ref. [Gal95b] coined the name fluctuation theorem for Eqs. (17.1), (17.2). Later on this result became known as the Gallavotti-Cohen fluctuation theorem (GCFT); see also Ref. [Rue99b] for a proof. Some further remarks are in order: The GCFT is a large deviation principle that is supposed to be valid even very far from equilibrium. Close to equilibrium it allows the derivation of Green-Kubo and Onsager reciprocity relations as well as of fluctuation-dissipation theorems. This requires one to make the additional assumption that the probability distribution ρ evolves into a Gaussian, respectively to employ the central limit theorem [Gal96; Gal98; Zam04; Giu05]. Secondly, the GCFT is based on the concept of modeling NSS by non-Hamiltonian, deterministically thermostated dynamical systems, which share the properties of being dissipative, time-reversible and chaotic. It thus presents an important application of the theory outlined in Part 2 of this book. The existence of a phase space contraction is crucial, since for the GCFT the variable ξ is defined by it. The existence of an SRB measure for a NSS is at the heart of the proof. Since the GCFT only applies to NSS, it is sometimes called a steady state fluctuation theorem. Thirdly, the CH implies the GCFT. It follows that a non-validity of the GCFT implies the non-validity of the CH. Hence, the GCFT was suggested as a practical means to test the validity of the CH [Gal95a; Gal95b; Gal98]. For a very elucidating, simple example of how to verify the GCFT ana-

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

323

17.1. Fluctuation relations

lytically in case of a dissipative baker map we refer to Ref. [Dor99]. As an example of a standard numerical analysis, here we consider the application of the GCFT to the driven periodic Lorentz gas with a Gaussian thermostat, see Chapter 11. As was shown by Chernov et al. [Che93a; Che93b; Che01], this model is compatible with the CH: Although the scatterers are hard disks defined by a singular potential and the system is hence not Anosov, for small enough field strength it exhibits Anosov-like properties in being hyperbolic and exhibiting an SRB measure. In the case of the Gaussian thermostated Lorentz gas the fluctuations of the phase space contraction rate are bounded and Eq. (17.2) is fulfilled. For practical tests it is convenient to define the function dτ (ξ) :=

ρτ (ξ) 1 ln τ ρτ (−ξ)

,

(17.3)

.

(17.4)

which enables rewriting of Eq. (17.1) as dτ (ξ) = ξ

(τ → ∞)

In order to verify the GCFT it thus remains to check for the linearity of Eq.(17.4) with slope one.1 In computer simulations one usually generates a long trajectory of a moving particle in a NSS and computes ρ via a histogram with respect to averages of the phase space contraction rate ξ over trajectory segements of size τ . In the following we show results from Ref. [Dol03], where a driven two-dimensional periodic Lorentz gas according to Eqs. (11.1), (11.4) but with disks on a quadratic lattice has been studied; for related results see Ref. [Bon97]. Fig. 17.1 depicts results for the (scaled) probability density ρτ (ξ) at different τ . One can see that for larger τ the curves seem to approach a Gaussian shape while for smaller τ there are strong deviations. We emphasize again that ρτ being a Gaussian is sufficient for the GCFT to be fulfilled, however, it is not a necessary condition. In fact, typically such distributions are not Gaussian [Bon97; Giu05]. The evaluation of ρτ (ξ) from Fig. 17.1 according to Eq. (17.3) is shown in Fig. 17.2, which clearly depicts the linearity with slope one as expected from Eq. (17.4). This is an example of a standard numerical analysis of a thermostated dynamical system in a NSS in order to check for the validity of the GCFT. 1 Note that for larger values of the field strength the Lorentz gas attractor contracts onto a submanifold in phase space, see Section 11.2.3. There is a discussion about the existence of a correction factor in this case such that the slope is not equal to one anymore [Bon97; Ron02a; Giu05; Wil06].

4:1

World Scientific Book - 9in x 6in

324

klages˙book

Further topics in chaotic transport theory

10

τ=5.9 τ=9.9 τ=16.9 τ=28.6 τ=48.4 τ=82.0 τ=138.9

8

6 Πτ(x)

May 8, 2007

4

2

0 -1

-0.8 -0.6 -0.4 -0.2

0 x

0.2

0.4

0.6

0.8

1

Fig. 17.1 The scaled probability density Πτ (x) := ερτ (εx) , x = ξ, for the driven Gaussian-thermostated periodic Lorentz gas with scatterers on a quadratic lattice. Disk radius, mass and velocity of the moving particle were set to one. The smallest distance between two scatterers is fixed to w = 0.1. The electric field strength is ε = (0.3, 0.48), the number of collisions 1.58 · 107 . The figure is from Ref. [Dol03].

The application of the GCFT looks fairly simple, however there are a number of subtleties involved in it: Firstly, as a mathematical theorem the GCFT requires a system to satisfy the CH, that is, the proof holds for transitive time-reversible dissipative Anosov(-like) systems only. Anosov systems exhibit many convenient dynamical properties and may be considered as a chaotic dynamical systems counterpart of the harmonic oscillator model for regular motion [Gal98]. However, one may inquire to which extent the Anosov property holds for realistic systems such as interacting many-particle fluids in experiments [Eva05]. Surprisingly, computer simulations suggest that the GCFT is valid for systems not being Anosov, some not even being Anosov-like [Bon97; Bon98a; Lep00; Bon01; Dol02; Zam04; Giu05]. This indicates that the theorem may hold under weaker assumptions. In fact, inspired by the GCFT a respective symmetry relation has been observed in real experiments on turbulent flows, however, here the fluctuating variable ξ has loosely been interpreted in terms of some (local) observable somewhat connected with entropy production [Cil98; Cil04]. This leads directly to a second fundamental issue related to the GCFT.

4:1

World Scientific Book - 9in x 6in

klages˙book

325

17.1. Fluctuation relations

1 1 0.5 0

0.5

-0.5 -1 Dτ(x)

May 8, 2007

τ=82.0 τ=138.9 -1 -0.5

0

0.5

1

0

-0.5

τ=5.9 τ=9.9 τ=16.9 τ=28.6 τ=48.4

-1 -1

-0.5

0 x

0.5

1

Fig. 17.2 The scaled function Dτ (x) := 1ε dτ (εx) , x = ξ, see Eq. (17.3). Dτ (x) = x according to Eq. (17.4) corresponds to the GCFT Eqs. (17.1), (17.2). The inset shows the same function for higher τ values. The figure is from Ref. [Dol03].

In the proof, ξ strictly represents the phase space contraction rate of the dynamical system, a quantity that is hardly measurable in any real experiment. A link to physics can only be established by identifying the phase space contraction with the thermodynamic entropy production of the system at hand. This establishes an important connection to the main theme of Part 2 of this book, in which we have shown that such an identification only holds for conventional Gaussian and Nos´e-Hoover dynamics. Actually, there we have focused on the relation between average rates, whereas the GCFT even requires the identification of the instantaneous rates. The latter holds true, for example, for the Gaussian thermostated driven periodic Lorentz gas, see Eq. (11.4) in Section 11, but it turns out to be wrong for the Nos´e-Hoover thermostated version, see Eq. (12.16) in Section 12. Indeed, computer simulation results for the Nos´e-Hoover thermostated Lorentz gas first suggested that the GCFT is valid by identifying ξ with the thermodynamic entropy production rate but not for the phase space contraction rate [Dol05].2 Similar conclusions were drawn from simulating a shear flow via isokinetic SLLOD dynamics where the rates are also 2 The same conclusion was obtained for simulating the driven periodic Lorentz gas thermostated by deterministic scattering as described in Chapter 15.4 [Dol06].

May 8, 2007

4:1

326

World Scientific Book - 9in x 6in

Further topics in chaotic transport theory

different, suggesting that a fluctuation formula like Eq. (17.1) may generally only hold for ξ being the thermodynamic entropy production and not for the phase space contraction rate, at variance to the proof [Sea00a; Eva05]. The solution to this problem is related to Eq. (17.2), which determines the range of validity of the GCFT [Bon06b]. Eq. (12.16) shows that, in contrast to the Gaussian constraint, for Nos´e-Hoover dynamics the phase space contraction fluctuations are singular thus violating Eq. (17.2). A detailed analysis shows that in such cases a generalized version of the GCFT holds for the phase space contraction rate implying a range of validity for Eq. (17.1) that rapidly shrinks to zero nearby equilibrium, which makes it considerably harder to detect numerically. For the Nos´e-Hoover thermostated Lorentz gas this conclusion was confirmed by further simulations [Gil06]. Ref. [Bon06b] also suggested a procedure for how to remove singularities from the phase space contraction rate by defining a fluctuating variable that is amenable to the standard GCFT analysis. These conclusions are in line with the numerical finding that checking Eq. (17.1) for ξ as the thermodynamic entropy production rate, the range of validity appeared to be much larger than for the phase space contraction in the same NSS [Dol05]. An analogous problem was encountered earlier by calculating a fluctuation symmetry for a Brownian particle dragged by a harmonic potential through water [vZ03a; vZ03b; vZ04], which was motivated by experiments [Wan02]. By solving the corresponding overdamped Langevin equation, van Zon and Cohen showed that there exists a Gallavotti-Cohen like symmetry for the work fluctuations of the particle, whereas for the heat a generalized version analogous to the one discussed above applies. This follows from singularities generated by the harmonic potential and inspired the analysis in Ref. [Bon06b]. In conclusion, the verification of a Gallavotti-Cohen type fluctuation symmetry in the form of Eq. (17.1) crucially depends on a proper choice of the fluctuating variable ξ. This aspect becomes nontrivial if there are singularities either in the thermostating mechanism or in the interaction potential. In the case of such singularities, extensions of the conventional GCFT appear to hold. In the light of this discussion it might be interesting to study the validity of the GCFT for further thermostating schemes such as, e.g., the non-ideal thermostats proposed in Chapter 14. Refs. [Eva93; Gal95a] triggered important work on the mathematical foundations of steady state fluctuation theorems within the framework of

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

327

17.1. Fluctuation relations

large deviation theory: Kurchan [Kur98] showed that a Gallavotti-Cohen type symmetry holds for stochastic systems obeying a Langevin-type dynamics. Lebowitz and Spohn [Leb99] extended Kurchan’s work to general Markov processes, where the GCFT was understood as a symmetry property of an associated large deviation functional. In both cases, time-reversal symmetry was replaced by a detailed balance condition. Maes [Mae99] argued that the GCFT can be understood as a Gibbs property, i.e., as a result from applying the theory of Gibbs measures to nonequilibrium situations. 17.1.3

The Evans-Searles fluctuation theorem

In parallel to the development of the GCFT, Evans and Searles derived a different version of a fluctuation symmetry (ESFT), which they denoted as a transient fluctuation theorem. It straightforwardly follows from solving the (dissipative) Liouville equation Eq. (12.3), cf. Section 12.1, plus using time-reversibility [Eva94; Eva02b] and reads ρτ (Ω) = exp(Ω) ρτ (−Ω)

,

(17.5)

see also Refs. [Coh99; Sea00b; Ron02a; Eva05] for derivations. Although formally this equation looks very similar to the GCFT Eq. (17.1), its meaning is in detail very different: Here the probability distributions ρτ are measured over a forward ensemble of trajectories generated for a system that starts with a time-invariant initial distribution in phase space. For the sake of simplicity let us restrict to an (ergodically consistent [Eva02b]) equilibrium distribution. At time t = 0 a dissipative external field is applied to the system. This field does work on the system, however, a deterministic thermostat enables it to relax to a NSS. Ω holds for an observable called the dissipation function. For the moment one may think of it as a generalized entropy production obtained by integrating a corresponding entropy production rate along each trajectory of the ensemble from time t = 0 up to time t = τ . This is the reason why on the right hand side, in contrast to Eq. (17.1), there is no τ in the exponent. A crucial difference to the GCFT is that Eq.(17.5) does not require the system to be in a NSS for the whole time but applies to a transient situation, where the system evolves from some given initial distribution into some NSS. Correspondingly, the probability densities ρτ are not computed with respect to one long trajectory in a NSS but from an ensemble of trajectories evolved over the time 0 ≤ t ≤ τ . This justifies the distinction

May 8, 2007

4:1

328

World Scientific Book - 9in x 6in

Further topics in chaotic transport theory

of the ESFT as a transient and of the GCFT as a steady state fluctuation theorem. Note also that Eq.(17.5) is valid for all times τ , that is, it does not require any long-time limit, and its derivation is not based on any chaos assumption like the CH. It is not based on identifying Ω with any phase space contraction or entropy production either. Recent work suggested to define Ω on the right hand side of Eq.(17.5) by inverting, in a way, the derivation of this equation: The probability distributions ρτ on the left hand side are then replaced by known solutions (forward and backward in time) of the Liouville equation for given nonequilibrium ensembles; the tables in Refs. [Sea00a; Eva02b; Rei05] list the different functional forms of Ω for given ensembles.3 For isoenergetic Gaussian thermostats Ω is identical to the phase space contraction, however, generally it can be very different. Note that the formalism also applies to Hamiltonian dynamical systems in some transient situation, hence, in contrast to the GCFT, it does not require the system to be dissipative [Eva02b]. The strategy of the ESFT thus seems to be to define a quantity called Ω for which one expects to obtain a fluctuation symmetry such as Eq.(17.5) in a given nonequilibrium situation. It then remains to check for the physical interpretation of Ω. It has furthermore been argued that in the limit of τ → ∞ and under specific conditions in a NSS, the ESFT converges to the GCFT [Ayt99; Eva02b; Eva05]. This relation and the detailed similarities and differences between the GCFT and the ESFT have been very much debated [Coh99; Ron02a; Eva05; Bon06b]. The ESFT has been tested in recent experiments, where a colloidal particle captured in an optical trap was dragged through water: An integrated version of Eq.(17.5) was confirmed in Ref. [Wan02] and the validity of the original formulation was reported in Ref. [Car04]. For a more detailed theoretical analysis of the experiment Ref. [Wan02] see Refs. [vZ03a; vZ03b; vZ04]. Particularly Ref. [Wan02] received wide attention and caused a lot of activity on further theoretical and experimental work trying to verify other types of fluctuation theorems. 17.1.4

Jarzynski work relation and Crooks relation

A third type of fluctuation relations emerged at first quite independently from fluctuation theorems. Let us consider again a finite classical system 3 This approach seems to yield a general expression for Ω that appears to be similar to what was called an action functional in Ref. [Leb99].

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

17.1. Fluctuation relations

klages˙book

329

in contact with a thermal reservoir. In Ref. [Jar97b] Jarzynski proposed to calculate the equilibrium free energy difference ∆F between two configurations of such a system in terms of the ensemble averaged nonequilibrium work W performed onto the system by parametrically switching from one configuration to the other. This Jarzynski work relation (JWR) states that < exp(−W/T ) >= exp(−∆F/T ) .

(17.6)

Here T holds for the temperature of the thermal reservoir, which is defined such that it generates a canonical equilibrium ensemble for the initial configuration of the system. As usual, the Boltzmann constant is set equal to one. The angular brackets denote an average over the probability distribution ρτ (W ) of all switching measurements starting from the initial canonical ensemble. The switching process is thought to be performed by varying a control parameter such as, e.g., the strength of an external field, from a given initial to a given final value. Over the total switching time τ the system evolves in contact with the thermal reservoir. The two limiting cases of τ → 0 and τ → ∞ lead to well-established thermodnamic identities. The important point is that Eq. (17.6) is claimed to hold for any time τ of the nonequilibrium switching process as well as arbitrarily far from equilibrium. It thus links equilibrium to nonequilibrium statistical mechanics by suggesting the extraction of equilibrium free energy differences from irreversible work performed in a nonequilibrium situation. In the original work Ref. [Jar97b] this result has been derived both for Hamiltonian dynamics under the assumption of a weak coupling between subsystem and reservoir as well as for non-Hamiltonian Nos´e-Hoover dynamics, where this assumption was not needed. Ref. [Jar97a] reported a stochastic derivation employing Markovian dynamics and detailed balance, again without a weak coupling ansatz. It also verifies Eq. (17.6) for many examples such as a form of the Nos´e-Hoover oscillator discussed in Sections 12.2.1 and 12.4.1. Ref. [Hum01] suggested applications of the JWR to single-molecule pulling experiments, which have later on been performed in Ref. [Lip02]. An experimental test of the JWR on thermal fluctuation of a mechanical torsion pendulum was performed in Ref. [Dou05]. Hatano and Sasa further generalized the JWR to arbitrary transitions between NSS [Hat01]. Their prediction was tested experimentally by measuring the dissipation and fluctuations of microspheres optically driven through water [Tre04], which reminds one of the experiments by which the ESFT was verified [Wan02; Car04]. For a review about theory and experiments related to the JWR see

May 8, 2007

4:1

World Scientific Book - 9in x 6in

330

klages˙book

Further topics in chaotic transport theory

Ref. [Rit03]. It took a while before a link between the JWR and fluctation theorems was realized in the form of the Crooks relation (CR) [Cro99; Cro00], ρF τ (ζ) = exp(ζ) ρR τ (−ζ)

.

(17.7)

As in the case of the JWR, this equation applies to a system that is initially in thermal equilibrium but then driven out of equilibrium by a parametric switching process. ζ(t) denotes the entropy production of the driven system measured over the switching time interval τ , ρF τ is the R probability distribution of this entropy production, and ρτ is the corresponding probability distribution when the system is driven in a timereversed manner, after having relaxed to equilibrium. In Refs. [Cro99; Cro00], Eq. (17.7) was derived for stochastic Markov processes fulfilling a detailed balance assumption. Eq. (17.7) is formally very similar to the ESFT. As we have discussed before, the dissipation function Ω in Eq. (17.5) may be considered as a generalization of the entropy production ζ of the CR. However, the ESFT does not contain any forward and time-reversed probability densities, and the precise connection is currently explored [Rei05]. On the other hand, R the CR also well applies to time-reversible NSS, where ρτ := ρF τ = ρτ . In this case the Eq. (17.7) reduces to the steady state fluctuation theorem ρτ (ζ) = exp(ζ) ρτ (−ζ)

.

(17.8)

With ζ˙ = ξ this equation is formally identical to Eq. (17.1). However, it is valid for any τ and already in this respect very different from the GCFT; for a more detailed discussion of the connection between the CR and steady state fluctuation theorems see Ref. [Jar00]. The JWR can be derived from Eq. (17.7) by exploiting the fact that Z ∞ Z ∞ < exp(−ζ) >= dζ ρF (ζ) exp(−ζ) = dζ ρR (17.9) τ τ (−ζ) = 1 −∞

−∞

and identifying the entropy production with the dissipative work [Cro99], −∆F + W . T Using the latter equation, the CR can be reformulated as ρF −∆F + W τ (W/T ) = exp , ρR T τ (−W/T ) ζ=

(17.10)

(17.11)

May 8, 2007

4:1

World Scientific Book - 9in x 6in

17.2. Lyapunov modes

klages˙book

331

which has been checked in recent experiments on RNA folding [Col05] and on a mechanical torsion pendulum [Dou05]. For a brief review of theory and experiments on the JWR and the CR see Ref. [Bus05]. From a theoretical point of view it appears to be interesting to further check the assumptions made in the derivation of both the JWR and the CR in order to learn about their range of validity. For example, one could systematically go beyond simple Markov processes by using deterministic models yielding correlated random walks as discussed in Part 1. One may also try to verify these relations by coupling the subsystems to more nontrivial thermal reservoirs as suggested in Part 2. It seems that experiments have all verified the JWR and the CR so far, hence it might be interesting to look for counterexamples. Another important issue is to better understand the connections between the different fluctuation relations trying to arrive at a unifying picture. We emphasize again that the field of fluctuation relations outlined in this section is an extremely active field of research. Here it is impossible and not our intention to give credit to all important works that appeared on these topics over the past few years. For more details and further literature we refer to Ref. [Kur05] for a very concise review as well as to Refs. [Eva02b; Ron02a; Jar02; Rit03] and further references therein.

17.2

Lyapunov modes

A central problem in the field of chaos and transport is to assess the dynamical instability of a many-particle system by calculating its spectrum of Lyapunov exponents. Such an analysis is at the heart of fundamental relations expressing transport coefficients in terms of dynamical systems quantities, as was outlined in Section 13.3. Recent work went one step further by exploring specific features of the Lyapunov spectra of interacting many-particle fluids in terms of eigenmodes associated with the single Lyapunov exponents. As we will summarize in the following, the maximum (respectively the minimum) Lyapunov exponent is connected with a localization behavior in the corresponding eigenmode. The Lyapunov exponents that are closest to zero, on the other hand, form surprising step-like structures in the Lyapunov spectrum. This unexpected degeneracy can be understood in terms of the associated eigenmodes, which form coherent wave-like patterns in phase space that were coined Lyapunov modes. In this section we mainly proceed along the lines of Refs. [For04; Pos05];

May 8, 2007

4:1

World Scientific Book - 9in x 6in

332

klages˙book

Further topics in chaotic transport theory

see also the introductory chapter of Ref. [Gas98a] for some basic definitions. Let us consider a classical interacting many-particle fluid, whose state is characterized by a phase space vector Γ(t) in a phase space of dimension 2dN . Here N stands for the number of particles in d dimensions and Γ := {qi , pi ; i = 1, . . . , N }, where qi := (xi , yi ) and pi := (px,i , py,i ) denote the position and momentum vectors of particle i, respectively. Let us assume that Γ(t) evolves according to the time-reversible equations of motion ˙ = F(Γ) Γ

.

(17.12)

The stability of a trajectory obtained from this equation with respect to a small perturbation of the initial condition Γ(0) is characterized by following the evolution of an offset trajectory Γs (t), where lims→0 Γs (0) = Γ(0). Reference and perturbed trajectory are related to each other by the offset vector Γs (t) − Γ(t) , (17.13) δΓ := lim s→0 s which can also be written as δΓ = {δqi , δpi ; i = 1, . . . , N }. This vector lives in the tangent space of the dynamical system and evolves according to the linearized equations of motion ∂F · δΓ . (17.14) ∂Γ The partial derivatives stand for the Jacobian matrix, which depends on the phase space point and thus varies with time. For Lyapunov unstable systems there exist L orthonormal initial vectors {δΓl (0); l = 1, . . . , L} in tangent space, whose norm grows exponentially in time. This leads to the definition of the associated L Lyapunov exponents ˙ = δΓ

1 |δΓl (t)| ln , l = 1, . . . , L , t→∞ t |δΓl (0)|

λl := lim

(17.15)

which are independent of the inital state Γ(0) and of the metric. The precise conditions for the existence of λl in dynamical systems defined on Ldimensional manifolds are formulated by the multiplicative ergodic theorem of Oseledec [Eck85]. Geometrically, the Lyapunov exponents quantify the contraction and expansion along linearly independent directions in phase space of an infinitesimal hypersphere comoving and corotating with the flow. The ordered set {λ1 , λ2 , . . . , λL } , λ1 ≥ λ2 ≥ . . . ≥ λL of exponents is called the spectrum of Lyapunov exponents of a dynamical system. Usually, the whole spectrum can be calculated analytically only for very specific models. An example is the Lorentz gas, see, e.g., Ref. [Dor99] and further

May 8, 2007

4:1

17.2. Lyapunov modes

World Scientific Book - 9in x 6in

klages˙book

333

references therein. Numerically, the algorithm by Benettin et al. [Ben80a; Ben80b] enables a computation of the full Lyapunov spectrum on the basis of computer simulations. Since the number of Lyapunov exponents depends on the dimensionality of the phase space, one may first of all wonder whether for an interacting many-particle systems the whole spectrum is well-defined in the thermodynamic limit. There is indeed a discussion about the existence of the largest Lyapunov exponent in this limit. Numerical results for FermiPasta-Ulam coupled oscillators and Weeks-Chandler-Anderson fluids seem to indicate a logarithmic density dependence [Sea97], whereas simulations on other systems suggest the existence of the maximum exponent [Del96b; Mil02]. Currently this problems appears to be unsettled. In the following we will restrict ourselves to Hamiltonian dynamical systems in thermal equilibrium. In this case the conjugate pairing rule applies, which was discussed in Section 11.2.1 particularly for nonequilibrium situations. That is, for each positive exponent there exists a conjugate negative exponent such that, in equilibrium, the sum of the pairs vanishes. In other words, the equilibrium Lyapunov spectrum is fully symmetric, hence it suffices to study the positive branch of the spectrum only. As an example, we discuss results from molecular dynamics computer simulations for the Lyapunov instability of a two-dimensional system of hard disks in a square box of length Lx with periodic boundaries. The system is assumed to be in thermodynamic equilibrium at a number density of ρ = N/L2x = 0.7, which is slightly smaller than the fluid-to-solid phase transition density. The following results are all in reduced units, that is, the particle diameter, its mass, the kinetic energy per particle and Boltzmann’s constant are set to unity [For04]. Fig. 17.3 shows the (positive) Lyapunov spectrum of this system for N = 1024 hard disks. Note the peaked shape of the spectrum around the largest Lyapunov exponent. Fig. 17.4 reveals that the dynamics associated with the maximum exponent is indeed of a very particular type. It depicts the projection of the Lyapunov vector δΓ1 corresponding to λ1 onto the four-dimensional subspaces spanned by the components belonging to the individual particles. The squared norm |δqi |2 + |δpi |2 of this projection measures the contribution of the particle i to the growth process determining λ1 . In Fig. 17.4 this quantity is plotted as a function of the particle positions (xi , yi ). It shows that the largest expansion activity is confined to a few very active zones only. These active zones move diffusively in space and perform

4:1

World Scientific Book - 9in x 6in

334

klages˙book

Further topics in chaotic transport theory

6 2 λ

5 1

4 0 λ

May 8, 2007

1960 1980 2000 2020 2040 l

3 2 1 0 0

0.1

0.2

0.3

0.4

0.5 l/2N

0.6

0.7

0.8

0.9

1

Fig. 17.3 Lyapunov spectrum for a system of N = 1024 hard disks in a square box with periodic boundaries. The number density is ρ = 0.7, a reduced index is used on the abscissa. The Lyapunov exponents close to zero are magnified in the inset, where the nonnormalized index l is used. Clearly, a step structure in the spectrum is visible. Note that the whole spectrum is defined for integer values of l only. The figure is from Refs. [For04; Pos05].

occasional jumps. This behavior was first observed in a nonequilibrium twodimensional shear flow [Hoo98a] and was later on verified for equilibrium fluids consisting of hard dumbbells [Mil98a; Mil02] and hard disks [For04; Pos05]. Loosely speaking it can be understood as a competition of exponential growth factors, which favors particles with the largest perturbation components. Numerically certain localization measures have been introduced in order to further quantify this phenomenon [For04; Tan03b]. Theoretically the localization could be explained by using “clock models” in combination with front propagation techniques [vZ98] and kinetic theory [vZ02] as well as by related “brick models” [Tan06]. Let us now focus onto the smallest exponents in Fig. 17.3. The inset shows that, surprisingly, there exists a number of specific steps in this region of the Lyapunov spectrum. The lowest step, consisting of six paired exponents that are exactly equal to zero, has a quite straightforward explanation. Four zero exponents are due to energy and momentum conservation and that there is no expansion or contraction of phase space parallel to the direction of the flow in Hamiltonian dynamical systems. The remaining two zero exponents arise from a more intricate combination of conserved quantities and geometric constraints [Pos05]. All these properties confine

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

335

17.2. Lyapunov modes

|δq|2+|δp|2 0.020 0.015 0.010 0.005 0-20

20 10 0

-10 x

0

10

-10

y

20 -20

Fig. 17.4 Spatial localization in the Lyapunov vector δΓ1 corresponding to the maximum Lyapunov exponent λ1 of the hard disk fluid discussed in Fig. 17.3. The squared norm |δqi |2 + |δpi |2 of the individual particle contributions to δΓ1 are plotted at the positions (xi , yi ) of the particles. The ensuing surface is linearly interpolated over a periodic grid covering the simulation box. The figure is from Refs. [For04; Pos05].

the dynamics onto submanifolds in phase space with no growth of a perturbation component orthogonal to them. The degeneracy of the other exponents is more subtle. To give a flavor of its origin, Fig. 17.5 displays the two-dimensional position perturbations of the mode δΓ2045 as a two-dimensional vector field δq(x, y) over the area of the simulation box. The picture to the right represents a snapshot of the full vector field δq(x, y) showing a coherent pattern in phase space. In the left picture the components δx(x, y) and δy(x, y) are plotted separately revealing a sine-like pattern that is orthogonal to the x- and the y-axis, respectively. This characterizes the Lyapunov vector as a transversal mode. Related transversal patterns exist for the other modes corresponding to the four Lyapunov exponents of the first non-zero step in the Lyapunov spectrum. The modes associated with the next step are all longitudinal ones, the other groups of exponents correspond to transversal and longitudinal modes along diagonal directions in the simulation box and respectively. Such stable periodic patterns in the Lyapunov vectors, which are either stationary or oscillating, have been coined Lyapunov modes. Steps in the Lyapunov spectrum have first been observed in a hard disk fluid by Dellago et al. [Del96b]. Later on their existence was confirmed for hard sphere [Pos00a] and hard dumbbell fluids [Mil98a; Mil02]. A first theoretical approach in order to explain these hydrody-

May 8, 2007

4:1

336

World Scientific Book - 9in x 6in

Further topics in chaotic transport theory

Fig. 17.5 Lyapunov mode corresponding to the Lyapunov exponent λ2045 , which is within the first step of exponents in Fig. 17.3 that are close but not equal to zero. The left hand side shows a snapshot for the components δx(x, y) and δy(x, y) of the twodimensional vector field δq(x, y) over the area of the simulation box corresponding to the mode δΓ2045 . The picture to the right represents a snapshot of the full vector field δq(x, y). Both pictures clearly reveal the existence of a coherent pattern in phase space. The figure is from Refs. [For04; Pos05].

namic mode-like structures was provided by Eckmann et al. by using random matrices [Eck00], see also Ref. [Tan02c] for related work. Soon there followed approaches employing kinetic [McN01; Mar04; dW04a] and periodic orbit theory [Tan02a]. A general classification of different modes was provided in Ref. [Eck05]. A detailed analysis of Lyapunov modes in a one-dimensional channel of hard disks was performed in a series of papers by Taniguchi and Morriss [Tan03a; Tan03b; Tan05b; Tan05a; Tan06]. To detect steps in Lyapunov spectra and Lyapunov modes in fluids consisting of soft particles turned out to be more intricate. Eventually their existence was confirmed by using the Fourier transform of static and dynamic correlation functions for the spatial densities of Lyapunov vector fluctuations in Lennard Jones fluids [Yan05] and by applying related techniques to particles interacting via Weeks-Chandler-Anderson potentials [For05]. Altogether it seems that the existence of both unstable localized excitations and stable periodic patterns in Lyapunov vectors is well understood as far as many-particle systems in thermal equilibrium are concerned. However, although localization in Lyapunov vectors has been observed in a shear flow, up to now no Lyapunov modes have been reported for nonequilibrium systems. Another open question is whether there exists any relation between Lyapunov modes and transport. In recent work a connection could be estabilshed between the time oscillation of Lyapunov modes and re-

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

17.3. Fourier’s law

klages˙book

337

spective oscillations in time of a velocity autocorrelation function [Tan05b; Tan05a]. The latter quantity, on the other hand, is experimentally observable and, via a Taylor-Green-Kubo formula, closely related to diffusive transport. On the other hand, one would expect that transport is dominated by the largest modes, so it is not yet clear in what way the smallest ones could contribute. Finally, related Lyapunov modes were shown to exist in coupled map lattices and similar models. This suggests that they may be useful as a general means in order to quantify collective spatio-temporal chaotic behavior [Yan06].

17.3

Fourier’s law

A fundamental challenge for any microscopic theory of macroscopic transport is to derive phenomenological transport laws starting from first principles. So far we have encountered three examples of such derivations: For a class of Hamiltonian particle billiards exemplified by Lorentz gases it can be proven that the diffusion coefficient exists [Bun80; Bun81], see Section 10.4. We remark that a similar result can be obtained if in the two-dimensional periodic case the hard disk potential is replaced by a smooth one, such as a Yukawa or Coulomb potential, under the condition that the energy of the moving particle is large enough [Kna87; Nob95]. That Ohm’s law exists in a driven, Gaussian thermostated periodic Lorentz gas was mathematically established in Refs. [Che93a; Che93b], see Section 11.2.4. And the existence of shear and bulk viscosity was analytically verified for a two-particle fluid [Bun96], see Section 15.5.1. Understanding the microscopic foundations of heat conduction leading to the existence of Fourier’s law, however, is still [Pei79] an open problem. In this Section we highlight what one may call a new era in this field, which focuses on heat conduction in classical lattices often starting from numerical approaches. In our presentation we particularly follow the reviews Ref. [Bon00b] emphasizing mathematical physical foundations and Ref. [Lep03] for numerical studies of one-dimensional lattices. A short partial review related to the latter theme can be found in Ref. [Pro05], which is part of a small collection of recent papers on heat conduction, see the last five contributions to the Focus Issue Ref. [Cam05]. Molecular dynamics approaches to heat flows in fluids were briefly discussed in Section 15.5.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

338

17.3.1

klages˙book

Further topics in chaotic transport theory

The basic problem

Fourier’s law of heat conduction states that the heat flux J(r, t), that is, the amount of heat transported through a unit surface per unit time, is proportional to the gradient of temperature ∇T (r, t), J = −κ∇T

.

(17.16)

Here the heat conductivity κ is typically a function of temperature and density of the system. This law is assumed to be valid close to thermal equilibrium. It presupposes that both temperature and heat flux are well-defined, which usually requires the existence of local thermodynamic equilibrium. Eq. (17.16) has been confirmed experimentally for many fluids and solids under thermal gradients. The measured heat conductivity was found to be well in agreement with results obtained within the framework of linear response theory, here particularly the Boltzmann-Peierls approach for heat conduction in crystals, the Green-Kubo formula and mode-coupling theory, see Refs. [Pei79; Bon00b; Lep03] for short descriptions of these theories and further references therein. For theorists the challenge is to derive Fourier’s law starting from microscopic deterministic equations of motion. A rigorous mathematical approach requires one to first learn about the conditions for existence and uniqueness of the NSS generated by a given model exhibiting heat conduction [Bon00b]. However, the invariant measure characterizing NSS in nonlinear dynamical systems is not necessarily simple: As we have discussed in Section 13.4, depending on the thermal reservoirs by which the temperature gradient is created the measure may be smooth or fractal. Models that are most accessible to answering these questions are classical one-dimensional chains of coupled oscillators defined by the Hamiltonian H=

N X p2k + U (qk ) + V (qk+1 − qk ) . 2mk

(17.17)

k=1

Here U stands for an external on-site (or substrate) potential exerting a pinning force onto N atoms of mass mk while V describes the interaction potential between neighbouring particles. The relative coordinates qk denote the displacements of the particles from their equilibrium positions, qk = xk − ka, where a is the lattice spacing, and pk are the associated momenta. This Hamitonian has to be supplemented by (periodic, fixed, free) boundary conditions plus a coupling to thermal reservoirs. A typical choice is two thermal reservoirs at different temperatures, which act on the first, respectively the last particle of the

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

339

17.3. Fourier’s law

chain thus creating a temperature gradient. For numerical computations, these reservoirs are usually modeled either by stochastic Langevin equations, cf. Section 10.2, by randomly sampling the particle velocities from canonical distributions in the spirit of the stochastic boundaries discussed in Section 15.1 or by using deterministic Nos´e-Hoover thermostats, see Section 11.1. For analytical calculations, Hamiltonian reservoirs in the form of infinitely many planar waves have also been considered [Eck99b; Eck99a]. For such one-dimensional chains Fourier’s law Eq. (17.16) takes the form dT , (17.18) dx where the brackets denote an ensemble average for the one-dimensional heat flux through the lattice. In simulations the brackets are typically interpreted as a time average, which presupposes ergodicity. Depending on the model, the precise definitions of J and of κ can be subtle [Lep03; Pro05]. When we speak of κ in the following we usually refer to the global thermal conductivity calculated by κ(L) =< J > L/∆T for a chain of length L in a NSS, where ∆T is the total temperature drop between two thermal reservoirs and < J > is defined in terms of the average energy transfer to or from a reservoir [Cas84].4 The simplest model one can think of is a chain of harmonically coupled oscillators, U (q) = 0 and V (q) = mω 2 q 2 /2. For this ideal harmonic crystal it has been proven that the temperature profile is asymptotically flat, apart from regions close to the boundaries, and that the heat conductivity diverges according to κ ∼ L [Rie67]. The same result holds if a harmonic on-site potential is added [Pro05]. This non-existence of Fourier’s law is due to the existence of linearly stable (Goldstone) modes. Physically speaking, one may say that there is not enough “randomness” in the system to generate an efficient mode-mode coupling mechanism and thus to attenuate these modes. Accordingly, heat propagates ballistically and not diffusively through the lattice. Imposing disorder onto the harmonic chain in the form of random oscillator masses makes the problem similar to the one of Anderson localization: Heat conduction may now be totally suppressed, and κ depends on the system size L with a functional form that is determined by the type of boundary conditions and the specific form of the thermal reservoir; see Ref. [Lep03] and further references therein for details. < J >= −κ

4 Strictly

speaking one has to distinguish now between the existence of κ in the limit of L → ∞ and the existence of a linear temperature profile as required by Fourier’s law [Alo99; Eck04; Eck06c] .

May 8, 2007

4:1

World Scientific Book - 9in x 6in

340

klages˙book

Further topics in chaotic transport theory

Instead of using random masses one may now reconsider a homogeneous chain, but now equipped with a nonlinear, still integrable Toda interaction potential, V (q) = exp(−q) with U (q) = 0. However, for this model one recovers κ ∼ L, which is now due to the presence of nonlinearly stable (soliton) modes. Numerical results indicate that Fourier’s law does not hold either if a nonlinear but integrable chain with U (q) 6= 0 is chosen [Pro05]. For rigorous analytical results on existence and uniqueness of NSS in a broader class of strongly anharmonic lattices under temperature gradients see Refs. [Eck99b; Eck99a]. 17.3.2

Heat conduction in anharmonic chaotic chains

In the previous section we have argued that neither harmonic nor nonlinear but integrable lattices are sufficient to yield Fourier’s law. It thus seems that “more dynamical randomness” is needed in order to generate normal heat conduction. This motivates the study of nonintegrable models of anharmonic chains. The Hamiltonian Eq. (17.17) suggests to distinguish between two fundamental classes: (1) chains without on-site potentials, U = 0 (2) chains with on-site potential, U 6= 0. The most famous example in the first class of models is defined by the Fermi-Pasta-Ulam (FPU) potential [Lep03], V (q) =

α β 1 mω 2 q 2 + q 3 + q 4 2 3 4

,

(17.19)

which can be regarded as an expansion of V close to the equilibrium position of a particle. The special cases β = 0 or α = 0 are referred to as the FPUα and FPU-β model, respectively. Numerical studies of the approach to equilibrium about 50 years ago and the discovery that this model did not show the expected equipartitioning of energy have played a pioneering role in the field of chaos and complexity [Cam05]. Heat conduction in this model was simulated about a decade later. It took further decades before clear numerical evidence was provided that the heat conductivity of this model does not exist [Lep97]. A groundbreaking example for the second class of systems with U 6= 0 is the ding-a-ling model [Cas84]. Here harmonically bound oscillators with same mass and frequency alternate with free particles of the same mass. The latter are constrained to lie between two adjacent oscillators and collide

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

341

17.3. Fourier’s law

elastically with them. For this model it was shown numerically that for large enough ω and long enough chain lengths the heat conductivity exists. That in the above two examples of anharmonic chains the heat conductivity exists when U 6= 0 while it diverges for U = 0 is no coincidence: The Hamiltonian Eq. (17.17) with U = 0 is momentum conserving, but it is not in the case of a non-zero on-site potential [Hu98]. As was shown analytically in Ref. [Pro00],5 if in momentum conserving chains of the type of Eq. (17.17) the thermodynamic pressure, i.e., the average force between an arbitrary pair of particles, is non-vanishing the heat conductivity κ diverges with the chain length L in the thermodynamic limit. However, the converse statement is not true: A counterexample is the harmonic chain with harmonic on-site potential and κ ∼ L as mentioned in the previous section [Pro05]. These results establish the importance of on-site potential and momentum conservation. However, they still do not yield any sufficient conditions for the existence of Fourier’s law. This motivated the invention and numerical study of a large variety of models highlighting different aspects of this problem - one is tempted to say that there exist at least as many different models for heat conduction as researchers working in this field. In the following we first classify momentum-conserving models with respect to the type of divergence if the heat conductivity does not exist. We then describe further important models with on-site potentials for which κ exists. The role of different thermostats for all these models is briefly discussed, and in the case of normal heat conduction we comment on relations between transport, phase space contraction and entropy production. In the next two sections we then discuss the importance of microscopic chaos for Fourier’s law. Recent numerical studies provided sound evidence that for most momentum-conserving models the heat conductivity diverges as κ(L) ∼ Lγ

,

0<γ≤1 .

(17.20)

An exception is the coupled rotor model, V (q) = V0 cos q. This model yields a vanishing pressure and simulations indicate that, despite U = 0, it exhibits a finite heat conductivity in the strongly chaotic regime. This appears to be due to the existence of phase-slips, which act like localized random kicks whenever a particle jumps from one local minimum of the periodic potential to the next one [Gia00]. 5 See

Refs. [Bon00b; Nar02; Lep03; Pro05] for a critical discussion of this result.

May 8, 2007

4:1

342

World Scientific Book - 9in x 6in

Further topics in chaotic transport theory

Interestingly, extensive simulations indicate that all the other models studied fall into two universality classes [Lep03; Pro05; Li05]: For FPU-α and -β models as well as for diatomic Toda lattices [Hat99] it has been found that 0.35 ≤ γ ≤ 0.44. This is in agreement with analytical arguments based on mode coupling theory, which predict that γ = 2/5 [Lep98; Wan04a; Wan04b]. For fluid-like systems, on the other hand, a renormalization group approach yields γ = 1/3 [Nar02; Mai06]. The latter theory appears to match to more recent numerical findings on the divergence of heat conduction in a one-dimensional gas of elastically colliding particles with alternating masses, where 0.25 ≤ γ ≤ 0.35 [Gra02; Pro05]. However, we remark that this identification of two universality classes is controversial [Nar02; Mai06; Wan04a; Wan04b] For anharmonic chains that do not conserve momentum, the existence of Fourier’s law has also been confirmed for modifications of the ding-a-ling model such as the ding-dong model, where all harmonically bound particles directly collide elastically with each other, that is without mediating collisions via free particles [Pro92]. Normal heat conduction has furthermore been obtained for a ding-a-ling model with a gravitational instead of a harmonic potential [Pos98]. Another widely studied system with on-site potential is the Frenkel-Kontorova model, U (q) = U0 cos q and V (q) = mω 2 q 2 /2, which can be interpreted as a chain of harmonically coupled particles in an external periodic field [Hu98]. Here as well as for the Φ4 model, where harmonically interacting particles are bound by a quartic potential [Aok03], the heat conductivity again exists. Let us summarize our results on the existence normal heat conduction up to now: • Integrability (by using harmonic or nonlinear Toda potentials) implies a non-existing κ. The converse statement is not true, see, e.g., FPU models. • Momentum conservation (i.e., no substrate potential) implies a nonexisting κ if (cf. coupled rotors) the thermodynamic pressure of the chain is non-vanishing. The converse statement is again not true, see, e.g., harmonic chains with harmonic substrate potentials. • Microscopic chaos in the sense of a Lyapunov instability does not imply that κ exists, see, e.g., again FPU chains. In other words, a non-zero substrate potential or a vanishing pressure, nonintegrability and anharmonicity in either on-site or interparticle potential appear to be all necessary conditions for the existence of a heat conductivity

klages˙book

May 8, 2007

4:1

17.3. Fourier’s law

World Scientific Book - 9in x 6in

klages˙book

343

[Pro05]. In Section 17.4 we will further amend point three in the above list, after we have obtained more insight into the role that microscopic chaos plays for heat conduction. We conclude this section by linking our present discussion to Part 2 of this book, where we have introduced and analyzed different types of thermal reservoirs. The importance of the modeling of thermal reservoirs for the problem of heat conduction has been emphasized in the review by Lepri et al. [Lep03]. In Part 2 of this book it has been shown that, depending on the type of thermostat used, systems showing electrical conduction or under shear can exhibit very different chaotic dynamical properties. However, it seems that there are not too many studies available which investigate how different thermal reservoirs or changing the coupling between subsystem and reservoir affect dynamical systems properties under thermal gradients. Hoover et al. [Hoo04] considered a two-dimensional lattice of Φ4 oscillators by applying seven different time-reversible deterministic thermostats to the boundaries, all reservoirs being of conventional Gauss or Nos´e-Hoover type. All models exhibited Fourier’s law, and the authors reported only quantitative differences in their numerically computed heat conduction properties. Romero-Bastida and Aguilar [RB06] compared both one-dimensional FPU-β and Frenkel-Kontorova chains under application of conventional Nos´e-Hoover and Nos´e-Hoover chain thermostats. They again found only quantitative differences with respect to applying different thermostats. However, a simple example where heat conduction strongly depends on the properties of the reservoir has been discussed by Dhar [Dha01]. Moreover, a still controversial issue is the impact of deterministic and stochastic reservoirs on heat conducting dynamical systems. They seem to generate different types of measures, as we have outlined in Section 13.4. We believe that particularly the last point deserves further investigations. Another interesting observation is that in deterministically thermostated heat flows the loss of phase space dimensionality, due to the dissipative action of the thermostat, may significantly exceed the number of variables to which the thermostat is actually applied [Aok03; Hoo02b; Kus02; Pos04b]. In Section 15.5 we have demonstrated this phenomenon for a deterministically thermostated shear flow [Wag00]. Some authors have interpreted this fact as evidence that dimensionality loss is a real feature of nonequilibrium systems and not an artifact of a particular choice of thermostats [Hoo02b; Pos04b]

May 8, 2007

4:1

World Scientific Book - 9in x 6in

344

klages˙book

Further topics in chaotic transport theory

Within this context, a Lyapunov sum rule expressing the heat conductivity κ in terms of the sum of Lyapunov exponents λi for a chain of oscillators has been derived. For the simplified situation where there are no temperature jumps at the boundaries it reads κ(T ) = −

X T2 λi (T ) , 2 Vs (∇T ) i

(17.21)

where Vs is the volume of the system and T holds for the average temperature of both reservoirs; for the formula including boundary jumps see Eq. (4) of Ref. [Aok03]. Eq. (17.21) is the heat conduction analogue to the Lyapunov sum rules for electrical conduction Eq. (11.11), respectively Eq. (12.17), and for viscosity Eq. (15.15). It was claimed to be true independent of the type of thermostat used [Aok03]. However, like all other Lyapunov sum rules its derivation starts from assuming the existence of an identity between phase space contraction and entropy production, see Eq. (3) in Ref. [Aok03]. This condition must be proven to be fulfilled if the thermostat used is not a conventional one for which it is true by default. In Sections 14.1, 14.2, 15.4.1 and 15.5.2 we have discussed thermostats where this identity is not fulfilled. Hence Eq. (17.21) suffers from the same drawback as discussed for the other Lyapunov sum rules, that is, it is only exact for conventional thermostats. We finally remark that the Evans heat flow algorithm briefly described in Section 15.5, which was originally designed for interacting many-particle fluids, has also been applied to model heat conduction in FPU-β chains. Starting from a critical value of the heat field this approach generated solitonic solutions [Zha00]. For smaller values the probability to observe solitons decreased, which is due to the existence of multiple steady states [Zha01]. This suggests that in simulations of heat conduction in one-dimensional lattices one may not ignore the possibility that there exists strongly non-ergodic behavior. 17.3.3

Heat conduction in chaotic particle billiards

In the previous section we have outlined the relation between heat conduction and dynamical properties such as momentum conservation, anharmonicity and nonintegrability. On the other hand, we have not yet assessed the importance of microscopic chaos for the existence of Fourier’s law. This appears to be a very hard task in high-dimensional dynamical systems like chains of coupled oscillators. However, as we have discussed in the previous

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

345

17.3. Fourier’s law

T0

T1

Fig. 17.6 The Lorentz channel consists of a point particle moving between six semicircles, which are situated on a triangular lattice. The particle collides elastically with the scatterers and with the two horizontal lines of length L at distance h = 1 from each other. The radius R of the half disks is chosen such that, except for tangent collisions (see the dashed line), a particle cannot move vertically without hitting a disk. A purely diffusive channel is modeled by using absorbing, reflecting or periodic boundary conditions to the left and to the right. For a heat conducting channel stochastic boundary conditions with reservoir temperatures T0 6= T1 are applied to the left and to the right, respectively. The figure is adapted from Ref. [Alo99].

two parts of this book, phenomenological transport laws may also hold in low-dimensional chaotic dynamical systems for which Hamiltonian particle billiards are paradigmatic examples. A billiard model that was suggested within the framework of the escape rate formalism [Gas93], see Section 13.3, in order to analyze the interplay between microscopic chaos and diffusion is the Lorentz channel depicted in Fig. 17.6. It represents a reduced version of the two-dimensional periodic Lorentz gas by restricting the billiard dynamics to a characteristic fundamental domain. This elementary model has the same convenient mathematical properties as the periodic Lorentz gas and is thus diffusive, cf. Section 10.4. In Ref. [Leb78] Lebowitz and Spohn have shown that Fourier’s law exists for the random Lorentz gas, where a temperature gradient is generated by stochastic boundaries. This and Gaspard’s work inspired Alonso et al. to study a heat conducting Lorentz channel equipped with stochastic boundary reservoirs [Alo99]. Note that in contrast to oscillator chains, here heat conduction is due to particle diffusion, which in turn is generated by the nonlinear dynamical instability of the scatterers. Hence, one can study the importance of microscopic chaos for the existence of Fourier’s law. Since the Lorentz channel is normal diffusive it was found that the heat conductivity κ exists, however, for larger temperature gradients the temperature profile started to deviate from a linear one. As we have briefly mentioned in Section 1.1, there are controversial views as to which extent models like the Lorentz channel exhibit thermodynamic properties. Some authors have argued [Coh98; Ron00a; Coh02; Ron02b] that purely diffusive Lorentz gases, here to be understood as systems without any thermal gradients, do not exhibit local thermodynamic equilibrium in the sense of the existence of local Maxwellian velocity distri-

May 8, 2007

4:1

World Scientific Book - 9in x 6in

346

klages˙book

Further topics in chaotic transport theory

T0

T1

Fig. 17.7 The rotating disk channel is defined by a geometry similar to Fig. 17.6, where now two rows of scatterers are put on top of each other. Each disk rotates with a different angular velocity ωi , which changes under collisions with an ensemble of moving particles according to energy conserving scattering rules. Periodic boundary conditions are used in the vertical direction. The figure is adapted from Refs. [Mej01; Lar03].

butions, see also Section 15.5. This is due to the lack of interaction between moving point particles. For diffusion these arguments appear to be debatable [Vol02; Dor02; T´el02; Gas02a; Gas03]. For heat conduction the situation becomes worse, because the linear combination of two Maxwellians generated by the two stochastic thermal reservoirs is not Maxwellian [Dha99]. This problem motivated Meija-Monasterio et al. [Mej01; Lar03] to study a modified version of the Lorentz channel in which the hard disks were replaced by rotating scatterers, see Fig. 17.7. That is, with each disk one associates a rotational degree of freedom, and a moving particle interacts with the disks according to modified collision rules [Rat00b]; see Section 15.3 for further details. Note that the rotating disks thus establish an indirect transfer of energy between an ensemble of diffusing particles by mediating energy between them at any collision thus leading to local thermodynamic equilibrium, even under temperature gradients. This system reminds one a bit of the ding-a-ling model, where energy between fixed harmonic oscillators was transferred by means of free particles colliding elastically between them. However, note that in the ding-a-ling model particles could not diffusive freely, hence the Lorentz channel mechanism of heat conduction is in detail very different. For the rotating disk channel it was verified numerically that the heat conductivity κ exists, in agreement with Green-Kubo formulas, and that the temperature profiles are linear [Mej01; Lar03]. We may remark that these simulations were only performed for a specific value of the coupling parameter governing the energy transfer between rotating disks and diffusing particles. Furthermore, the double-channel geometry of Fig. 17.7 with two different disks in the vertical direction was

May 8, 2007

4:1

World Scientific Book - 9in x 6in

17.4. Pseudochaotic diffusion

klages˙book

347

selected on purpose, because there exists non-ergodic behavior in the case that a particle collides subsequently multiple times with the same rotating disk, see our discussion in Section 15.3. We believe that these problems, which were first reported for a single rotating disk [Rat00b], deserve further investigation. As we now indicate, it seems that the original rotating disk channel, though inspiring, does not always exhibit such nice mathematical properties as claimed when it was proposed as a model for heat conduction. Bunimovich and Khlabystova obtained exact solutions for tangent collisions in the equilibrium model by mapping respective orbits to the dynamics of deterministic Lorentz lattice gases [Bun03]. Balint and Troubetzkoy analyzed an abstraction of a single rotating disk in a unit cell, where the geometry was changed to that of an annulus billiard with rotating boundaries [Bal04]. They proved the existence of transitions between ergodic and periodic behavior depending on variation of a control parameter, similar to the ones mentioned in Refs. [Rat00b; Lar03]. Finally, there exists a series of mathematical works by Eckmann et al. From simplifications of the rotating disk channel a universal law for the stationary temperature profile of this class of systems was derived [Eck04]. In Ref. [Eck06c] a single chain of rotating disks was proposed, however, due to problems with non-ergodic behavior the hexagonal unit cells in Fig. 17.7 were replaced by diamond-shaped boxes with small lateral holes. Higherorder memory effects of this deterministic model were further analyzed in Ref. [Eck06b]. A reduced version of this model consisting of only two asymmetric cells eventually led to proposing a simple thermal rectifier. Finding mechanisms that could be used for the construction of thermal diodes appears to be a very recent topic in this field, see Ref. [Eck06a] and further references therein.

17.4

Pseudochaotic diffusion

The final section of this outlook highlights a topic that will turn out to be related to the problem of heat conduction described above. Similarly to Fourier’s law, one may inquire about the necessary and sufficient conditions for the existence of Fick’s law implying the existence of a (normal) diffusion coefficient. As we have discussed in detail in Part 1 of this book, over the past few decades it was realized that diffusive behavior does not only occur in interacting many-particle systems with many degrees of freedom such as fluids. It is also typical in low-dimensional toy models if the dynamical

May 8, 2007

4:1

348

World Scientific Book - 9in x 6in

Further topics in chaotic transport theory

system exhibits chaotic behavior in the sense of Lyapunov, that is, there is at least one positive Lyapunov exponent generated by some nonlinear dynamical instability. This raises the question whether Lyapunov chaos can be identified as the microscopic source of diffusion in realistic physical situations. In the following we describe an experiment that was designed to verify the existence of microscopic chaos in a fluid [Gas98b]. The interpretation of the results triggered a vivid discussion about the connection between microscopic chaos and diffusion, which started from the demonstration that inherently non-chaotic models display the very same type of diffusive behavior as observed experimentally [Det99b; Gra99]. The second part of this section focuses on the ergodic and diffusive properties of a particular class of such models, which are polygonal billiard channels. These models are not chaotic in the sense of Lyapunov, however, they display a form of dynamical randomness that has been called pseudochaos [Zas03]. Applying a temperature gradient to these systems yields a crosslink to our previous analysis of Fourier’s law and eventually enables us to learn more about the relevance of chaotic behavior for heat conduction. 17.4.1

Microscopic chaos and diffusion?

By measuring the motion of small tracer particles immersed in a fluid, Perrin was able to verify Einstein’s stochastic theory of Brownian motion [Per09]. In a similar spirit, Gaspard et al. conducted an experiment which set out to demonstrate that the diffusive motion of such particles is generated by deterministic chaos in the collisions between a tracer particle and the surrounding fluid [Gas98b; Bri01]. With a video camera they recorded long time series of the position of a small colloidal particle suspended in water, sampled at regular time intervals. This data was analyzed by using a technique of nonlinear time series analysis for which one computes the probability of recurrences in the trajectory of the moving particle [Eck85]. That is, one checks for instances where the system approximately retraces part of its previous trajectory in phase space over a certain length of time. The probability for such recurrences is expressed by computing a pattern entropy K(n, , τ ), where τ holds for the chosen discretization in time and for the spatial resolution, defined in terms of a suitable maximum distance according to which a trajectory of length n is to be identified with a given reference trajectory. If this pattern entropy increases linearly in time, the slope yields the (, τ )-entropy per unit time h(, τ ). The interpretation

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

349

17.4. Pseudochaotic diffusion

of the experimental data by Gaspard et al. then follows the fundamental relation X h(, τ ) ≤ hKS = λi , (17.22) λi >0

where hKS stands for the Kolmogorov-Sinai entropy [Eck85; Gas98a; Dor99] and the sum is over all positive Lyapunov exponents λi of the system. For Anosov systems, hKS is obtained as the supremum over all possible entropies h(, τ ). As is well-known [Ott93], a system is regular if hKS = 0, stochastic if hKS → ∞ and deterministically chaotic if 0 < hKS < ∞. According to Eq. (17.22), a positive h(, τ )-entropy thus yields a non-zero lower bound on hKS and, for a deterministic system, implies the existence of chaotic motion in the sense of Lyapunov. For a Wiener process it can be shown that h(, τ ) ∼ −2 (τ → 0) P [Gas98a] while in case of a deterministic origin h(, τ ) ∼ λi >0 λi = const. (τ → 0) for small enough . In the experimental data for h(, τ ) the Wiener regime could be clearly detected, which very accurately confirms the microscopic origin of diffusion in terms of Brownian motion, while only the beginning of a crossover to a constant value for smaller could be seen. In any case, by means of Eq. (17.22) the data for h(, τ ) yielded a positive lower bound for hKS , and it was thus concluded that the diffusion of the tracer particle is due to microscopic chaos. This interpretion of the experimental data became subject of two lines of criticism: Dettmann et al. argued that the temporal resolution provided by the measured time series was not sufficient to conclude anything about deterministic chaotic behavior, because the time scale for microscopic collisions between the particles is still many orders of magnitude smaller than the experimental sampling rate [Det99b]. A similar argument applies to the spatial resolution. Vulpiani et al. [Cen00; Cec05] have demonstrated for simple models, which partially are special cases of maps studied in Part 1 of this book, that in the regime of accessible to the experiment it is impossible to decide whether the data is generated from a deterministic or from a stochastic process. Particularly, in the latter case h(, τ ) may still diverge for small enough which are beyond experimental resolution. A second criticism was drawn from the fact that there exist a number of theoretical models which yield diffusive behavior, although their microscopic dynamics is clearly non-chaotic, in the sense specified above. Examples are the diffusion of a tracer particle in an ideal gas and the motion of an impurity in a harmonic crystal [D¨ ur98;

May 8, 2007

4:1

350

World Scientific Book - 9in x 6in

Further topics in chaotic transport theory

Fig. 17.8 The wind-tree model consists of a point (wind) particle moving along straight lines in the plane by elastically colliding with fixed square scatterers (trees), which are oriented with their diagonals along the x and y axes. The motion of the particle is restricted to the four possible directions along these axes, and its velocity is constant. The trees are located at random positions with a given number density and do not overlap. The figure is after Refs. [Det99b; Det00b].

Gra99]; see also the starting point for Zwanzig’s derivation of the Langevin equation outlined in Section 10.2, where the fluid surrounding the Brownian particle is modeled by a Hamiltonian with an infinite number of harmonic oscillators. Perhaps most conclusive for this debate was the numerical study of Ehrenfest’s wind-tree model [Ehr59] sketched in Fig. 17.8: In contrast to the random Lorentz gas, cf. Section 4.1, where chaos is due to the defocusing collisions of the moving particle with hard disks, in this model the convex scatterers are replaced by diamonds which do not create any nonlinear dynamical instability. Hence, the wind-tree model is deterministic but not Lyapunov chaotic. We remark that a simple one-dimensional map version of the wind-tree model, which is very similar to the maps discussed in Chapters 2 - 5, was studied in Refs. [Cec03; Cec05]. By applying the same analysis as Gaspard et al. to a time series numerically generated from the wind-tree model, Dettmann et al. obtained results for trajectories and h(, τ ) that are almost indistinguishable from the experimental data, although the system does not exhibit any positive Lyapunov exponent [Det99b; Det00b]. The

klages˙book

May 8, 2007

4:1

17.4. Pseudochaotic diffusion

World Scientific Book - 9in x 6in

klages˙book

351

model was thus presented as a counterexample to the claim by Gaspard et al. of having verified microscopic chaos in a fluid system. This point deserves further explanations: In the wind-tree model the diffusive behavior is not generated by any infinitesimal dynamical instability but by the randomness of the scatterer positions, which is sampled at any collision. As emphasized by Gaspard et al., the origin of the Brownian motion thus lies in the structural disorder of the model [Gas99; Bri01]. Similarly, for the ideal gas and the harmonic crystal the diffusivity of the tracer particle, respectively the impurities, results from the stochasticity in the distribution of the initial conditions of the surrounding non-chaotic fluid, cf. also our discussion in Section 10.2 on exactly this detail. Note that in sharp contrast to deterministically chaotic models, these systems must live in an infinite phase space in order to provide a non-vanishing source of randomness generating diffusive behavior. However, for systems on non-compact spaces little is known about the validity of mathematical relations such as Eq. (17.22), which for the reasons mentioned above cannot hold for this class of systems. If one interprets the collisions of a moving point particle in an infinite deterministic system with spatial randomness as an effective stochastic process [Cec05], one would expect that for infinitely fine resolution h(, τ ) = hKS → ∞, which would be compatible with finite P resolution numerical experiments on these systems, while λi >0 λi = 0. This suggests that Pesin’s identity, which so far is only proven for Anosov(like) systems defined on compact spaces, cf. Section 13.3, is not valid for these non-chaotic systems.6 An additional difficulty for infinite systems was pointed out by Grassberger and Schreiber in that the definitions of entropy and Lyapunov exponents become norm dependent [Gra99]. In summary, the interpretation of the experimental data defended by Gaspard et al. relies on the validity of the relation Eq. (17.22), which somewhat presupposes deterministic Anosov(-like) properties on compact spaces. Per se, it thus sorts out the non-Lyapunov chaotic systems considered by the opponents of their analysis. However, Gaspard et al. have argued that these are not plausible models for the Brownian motion of a particle in a fluid because of their artificial microscopic construction, which does not match to the apparent molecular structure of the fluid, and that these models have various other shortcomings as explanations for their experiment [Gas99; Bri01]. Ceccini et al. [Cen00; Cec05], have further highlighted the long-standing problem of a distinction 6 Indeed, there does not exist any mechanism for the creation of information in such systems related to an exponential dynamical instability.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

352

Further topics in chaotic transport theory

between chaos and noise within the framework of nonlinear time-series analysis by studying a variety of simple deterministic as well as stochastic models. If at all, an answer to this question seems to require one the experimental study of low- rather than high-dimensional dynamical systems [Cen00; Bri01] such as a few atoms in ion traps, which can be described in terms of more simple Hamiltonian particle billiards. From this research it already becomes clear that microscopic chaos cannot be a necessary condition for the existence of normal diffusion, as exemplified by the wind-tree model. However, chaos only is also not sufficient since, for example, the periodic Lorentz gas with infinite horizon is chaotic [Gas98a] but yields superdiffusive behavior.7 An interesting question in this context is whether there exists a non-trivial generalization of Pesin’s theorem for non-chaotic systems, which would require to redefine fundamental chaos indicators like Lyapunov exponents and Kolmogorov-Sinai entropies [Kor04a]. We may also mention Refs. [Got04; Got05], which in a way turn the problem discussed above upside-down by suggesting to define chaos on the basis of normal diffusive behavior of a suitably chosen random variable. These very open questions motivated, among others, the study of further non-chaotic models, as we will summarize in the following. 17.4.2

Polygonal billiard channels

Studying deterministic transport in polygonal billiard channels links three different directions of research. We start with mathematically oriented works on the topological and ergodic properties of bounded polygonal billiards. We then present examples of unbounded polygonal billiard channels by focusing on their diffusive properties. Finally, we discuss the relation between diffusion and heat conduction in these channels. Fig. 17.9 shows three typical examples of polygonal billiards defined on compact domains. The boundaries of these billiard tables consist of straight lines enclosing two-dimensional polygonal areas in which a point particle scatters elastically with the walls. Interestingly, the dynamics of a particle in the triangular billiard Fig. 17.9 (c) is equivalent to the motion of three point particles on a circle colliding elastically with each other, where the different masses are related to the three angles of the triangle under the condition that no angle is larger than π/2 [Cas99]. The right triangular billiard Fig. 17.9 (b) corresponds to the motion of two point particles on 7 All

provided that the system is ergodic and that the thermodynamic limit exists.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

353

17.4. Pseudochaotic diffusion

(a) (b) (c)

α

α

γ

β

Fig. 17.9 Three simple examples of two-dimensional polygonal billiards: A particle with unit velocity moves inside the depicted geometric domains by scattering elastically with their boundaries. Part (a) depicts the square ring billiard defined by two squares of different sizes with parallel sides having the same center [Ric81; Art00], (b) shows the right triangular billiard [Art97b] and (c) the triangular billiard [Cas99].

the unit interval, where the mass ratio of the particles defines the angle α of the triangle [Cor82; Art97b]. Fundamental quantities determining the dynamics in a polygonal billiard are the inner vertex angles. If all angles are rational multiples of π the billiard is called rational, otherwise it is called irrational [Gut96; Tab05]. Rational billiards are not ergodic: Their phase space splits into invariant manifolds determined by the initial angle of any trajectory. The corresponding directional billiard flows are ergodic in Lebesgue almost all directions but not mixing in any direction. To what extent they are weakly mixing appears to be an open question; see Ref. [Gut96] for a review of these results, Refs. [Art97b; Cas99; Art00; Alo02] for short accounts of them and Ref. [Arn68; Cor82] for general introductions to ergodic theory. For rational billiards the topological structure of the surface in configuration space underlying all directional flows can be extracted by the technique of “tiling the plane” through periodic reflections of the billiard table [Ric81; Veg93]. Focusing onto the fundamental cell of the tesselation and joining all its identical edges yields a compact invariant surface. The structure of this surface, which is topologically equivalent to the original billiard table geometry, is determined by the genus counting the number of holes in the surface. This is illustrated in Fig. 17.10 for three different types of square billiards: For the simple square in 17.10 (a) the corresponding edges can be sewn together yielding a torus. The barrier billiard Fig. 17.10 (b) introduced in Ref. [Zwa83; Han90], sometimes also called bar-in-square billiard as discussed in Refs. [Zas01a; Zas02; Zas03; Zas05], is a spe-

May 8, 2007

4:1

World Scientific Book - 9in x 6in

354

klages˙book

Further topics in chaotic transport theory

(a)

(b)

(c)

Fig. 17.10 Three billiard tables with their corresponding invariant surfaces: The square billiard in (a) can be mapped onto a torus. The barrier billiard in (b) [Zwa83; Han90] is topologically equivalent to a non-integrable two-torus. Expanding the bar in (b) onto a square yields a five-handled sphere for the corresponding square ring billiard [Ric81]. The figure is adapted from Ref. [Zas05].

cial case of the square ring billiard, which can be obtained from it by vertically contracting the square scatterer in the middle onto a onedimensional bar. Taking reflection at the bar into account, this geometry can be mapped onto a two-torus. The square ring billiard, on the other hand, is topologically equivalent to a five-handled “pretzel” [Ric81; Art00]. For rational billiards an equation can be derived by which the genus is calculated in terms of the n rational vertex angles αi = πpi /qi , i = 1, . . . n of a polygon. If N is the least common multiple of qi , then the genus g of the surface S is determined by the formula [Ric81; Gut96; Tab05] n N X pi − 1 . (17.23) g(S) = 1 + 2 i=1 qi

Note that only in the four special cases of rectangles, equilateral triangles, isosceles right triangles and triangles with angles π/2, π/3, π/6 does the associated surface have the topology of a torus, hence the polygonal billiard is integrable. In all the other cases the topology is the one of non-integrable multi-handed spheres with more than one hole. Correspondingly, such rational billiards were coined pseudointegrable [Ric81]. These topological constructions reveal that for genus larger than one there exist isolated saddles resembling hyperbolic fixed points, which impose a “chaotic character” onto the polygonal billiard flows [Ric81; Veg93; Gut96; Viv06]. Topologically they are found at the points where different

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

355

17.4. Pseudochaotic diffusion

tori meet each other, geometrically they correspond to the singular vertices at the polygonal boundaries. In finite precision numerical simulations these corners indeed lead to positive values for chaos quantities like Lyapunov exponents [Veg93]. Physically speaking this is due to the fact that two trajectories, which are initially displaced from each other by some small but finite value in phase space, typically separate linearly from each other. Eventually such an orbit pair thus strikes different sides at the vertex of a polygon leading to dispersion with respect to the singularity at the corner. However, these peculiarities disappear by analytically taking the continuum limit of infinitesimally small displacements leading to zero values for Lyapunov exponents, topological and Kolmogorov-Sinai entropies, as was proven mathematically [Gut96]. Polygonal billiards thus belong to a class of systems that is intermediate between purely Lyapunov chaotic and purely regular dynamics. Let us be more explicit on this point by suggesting a classification of chaotic behavior in terms of the dispersion of initially infinitesimally close trajectories. We may start from the general expression for the asymptotic growth of the displacement ∆(t) of two trajectories in the form of [Gas88] ln ∆(t) ∼ tν0 (ln t)ν1 , 0 ≤ ν0 , ν1 ∈ R .

(17.24)

If ν0 = 1, ν1 = 0 we have the usual form of exponential dynamical instability ∆(t) ∼ exp(λt)

,

(17.25)

referred to as Lyapunov chaos [Ott93], whose strength may be characterized by the maximum positive Lyapunov exponent λ. If ∆(t) grows weaker than exponential, one often speaks of weak chaos [Zas01b; Gal03a; vB04; Art05]. This type of dynamics was also denoted as strange kinetics within the framework of fractional calculus [Shl93]. The regime of Eq. (17.24) with 0 < ν0 < 1 or ν0 = 1 and ν1 < 0, which is typical for intermittent dynamics, was characterized as sporadic by Gaspard and Wang [Gas88]; an example for such a system was discussed in Section 6. Here the dynamical instability is obviously either of stretched exponential type or it is exponential with logarithmic corrections, ∆(t) ∼ exp(tν0 (ln t)ν1 )

.

(17.26)

Eq. (17.24) with ν0 = 0 and ν1 = 1, on the other hand, yields purely algebraic dispersion, ∆(t) ∼ tν2 , 0 < ν2

,

(17.27)

May 8, 2007

4:1

356

World Scientific Book - 9in x 6in

klages˙book

Further topics in chaotic transport theory

for which Zaslavsky and Edelman [Zas02; Zas03] suggested the term pseudochaos.8 Note that algebraic dispersion with logarithmic corrections may also exist, ∆(t) ∼ tν2 (ln t)ν3 , ν3 ∈ R ,

(17.28)

covering a slightly larger class of dynamical systems. Polygonal billiards represent the special case of pseudochaotic dynamics with ν2 = 1 for which the dispersion is strictly linear in time, ∆(t) ∼ t .

(17.29)

However, in contrast to Lyapunov chaos here the linear dispersion does not actually capture the essential mechanism leading to dynamical randomness. For example, according to this classification both free flights and polygonal billiards of genus one, which clearly exhibit regular dynamics, are also pseudochaotic. As we have discussed above for the example of rational billiards, in polygonal billiards complicated topologies yielding pseudohyperbolic fixed points and pseudointegrability provide the source for non-trivial dynamics. One is thus tempted to speak of topology-induced chaos9 as a subclass of pseudochaos if there is linear dispersion on surfaces that are not integrable. Pseudointegrable rational billiards then form a subclass of topology-induced chaotic dynamical systems. As we will show in the following, systems with linear dispersion generating non-trivial dynamics due to complicated topological structures may exhibit ergodic and transport properties as they are usually associated with Lyapunov unstable chaotic dynamical systems. We conclude this very preliminary attempt of a classification of chaotic dynamics on the basis of dispersion by mentioning the bottom line of the purely regular, or periodic, case ∆(t) = const. .

(17.30)

Let us now get back to the description of polygonal billiards. While a good amount of mathematical work exists on rational models, few rigorous results are available for irrational ones, which motivated numerical studies. There is a well-established machinery available to check for ergodic properties of dynamical systems at the heart of which is the calculation of correlation functions; examples of such functions can be found in 8 We remark that in Refs. [Zas02; Zas03; Zas05] one finds several slightly different definitions of pseudochaos. Here we refer to the first one stated in Ref. [Zas02]. 9 This should not be confused with topological chaos as defined in Ref. [Gas98a].

May 8, 2007

4:1

17.4. Pseudochaotic diffusion

World Scientific Book - 9in x 6in

klages˙book

357

Section 10.2. Comparing time averages with ensemble averages of correlation functions and checking whether they, respectively their mean, decay in time to zero gives a numerical indication of ergodic properties like mixing, weak mixing and plain ergodicity. The Fourier transform of these functions provides furthermore a crosslink to the spectral theory of dynamical systems; see particularly Refs. [Alo02; Cas99; Art00; Cor82] for brief summaries of such techniques. These concepts of ergodic analysis have been applied numerically to the three polygonal billiards defined in Fig. 17.9: The square ring billiard [Ric81], which is rational, was found not to be weakly mixing for irrational directions of the particle trajectory in the case that the ratio between the sides of the two squares defining the billiard takes rational values. However, for irrational directions and irrational ratios of these sides the system turned out to be weakly mixing [Art00]. This complies with a theorem stating that for rational billiards of this type having more than four vertices, the directional flows are typically weakly mixing [Gut96]. For the right triangular billiard both the global dynamics in the case of irrational acute angles and the directional dynamics in the case of rational acute angles yielded ergodic and weak mixing properties [Art97b]. Surprisingly, in triangles where all angles are irrational the dynamics was numerically found to be mixing [Cas99], which contradicted “a prevailing” (but unproven) “opinion in the mathematical community that polygonal billiards are never mixing” [Gut96]. This model seems to provide a rare example that is mixing but has no positive Kolmogorov-Sinai entropy, which illustrates why it is useful to characterize such polygonal billiard dynamics as being topology-induced chaotic rather than being non-chaotic. One may now ask to which extent polygonal billiards can exhibit normal transport properties. This problem was put forward by Zwanzig, who designed “a cautionary example” for questioning the validity of short memory approximations as they are employed in continuous time random walk theory [Zwa83]. Simulation results for the mean square displacement in what we may call the barrier channel Fig. 17.11 (a), performed for irrational directional dynamics of this rational polygonal billiard, suggested a crossover from simple random walk behavior for short times to a rather undetermined, highly oscillatory functional form at larger times. Note that by restricting the dynamics to a unit cell we arrive again at the geometry of the barrier billiard Fig. 17.10 (b). Unfolding the barrier channel onto the whole plane, on the other hand, yields a convenient setting for obtaining analytical solutions for the horizontal

May 8, 2007

4:1

World Scientific Book - 9in x 6in

358

Further topics in chaotic transport theory

(b)

(a)

(c)

(d) φ

ψ Fig. 17.11 Four examples of polygonal billiard channels: (a) the barrier channel, where a particle moves through a periodic array of square boxes linked by small windows centered on the connecting barriers [Zwa83; Han90; Zas01a]; (b) the triangle channel, where the connecting walls of (a) are replaced by isosceles right triangles, and one side with walls is shifted by half of a (not necessarily quadratic) box length [Li02]; (c) the double-triangle channel with two different types of triangles [Alo02]; (d) the zigzag channel [Jep06; San06]. (a) and (b) depict parts of respective channels while (c) and (d) represent fundamental cells of them, which are periodically continued along the x-axis. Except in (a), the scatterers are defined such that there is no infinite horizon for a moving particle. For a similar figure summarizing different types of polygonal billiard channels see Ref. [Li05].

displacement of a moving particle [Han90]. In Ref. [Zas01a] these different levels of description are discussed, ranging from filamented topological surfaces over the barrrier billiard depicted in Fig. 17.10 (b) to the unfolded barrier channel of Fig. 17.11 (a), which in turn can be mapped onto one-dimensional time-discrete interval exchange transformations [Han90]. More extensive simulations of the barrier channel yielded superdiffusion, which was modeled by a fractional kinetic equation [Zas01a]. Similar barrier billiards were introduced and studied in Refs. [Zas02; Zas03; Zas05]. Another early study of diffusion in a polygonal billiard was carried out by counting the number of revolutions around the center of the square ring billiard Fig. 17.9 (a). Here the mean square displacement was found to increase superdiffusively in time, and the growth exponent was analytically related to the scaling exponent of the spectral measure of the system [Art00]. Note that in sharp contrast to the wind-tree model discussed in the previous section all these models feature no structural disorder, hence their diffusive properties are purely generated by the topology-induced chaos

klages˙book

May 8, 2007

4:1

17.4. Pseudochaotic diffusion

World Scientific Book - 9in x 6in

klages˙book

359

in these systems. A first example of a pseudochaotic system exhibiting normal diffusion was provided by the two-dimensional Poincar´e map of Ref. [Cas00], extracted as an approximation for the elongated version of the triangle Fig. 17.9 (c). This map was furthermore numerically shown to be ergodic and mixing. Relating the discussion of the Gaspard experiment outlined in the previous section to the problem of Fourier’s law, Li et al. [Li02] studied diffusion and heat conduction in triangle channels for which the periodic version is shown in Fig. 17.11 (b). Simulations of this model yielded superdiffusion and a divergent thermal conductivity. Motivated by the wind-tree model, structural disorder was then imposed onto the triangles by randomly changing their heights or by randomly shifting their positions along the walls, whereby normal diffusion and Fourier’s law were recovered. Later on another variant of the periodic triangle channel was proposed by using asymmetric triangles, which were periodically continued such that there was no average drift in the channel. For the irrational case diffusion and heat conduction were found to be normal in simulations, whereas for the rational case the system behaved superdiffusively, and there was no Fourier’s law [Li03a]. These results were put into a broader perspective by Alonso et al., who studied diffusion, heat conduction and ergodic behavior in what we may call the double-triangle channel, see Fig. 17.11 (c) [Alo02]. The angle φ was kept fixed and chosen to be irrational while for the angle ψ different rational multiples of π were considered. Simulations led to super-, sub- and normal diffusion depending on the choice of ψ, however, only in the latter two cases the decay of correlation functions indicated that their system was mixing. Corresponding normal and anomalous transport behavior was obtained for the heat conductivity. In Ref. [Alo04] it was suggested that these different transport properties are intimately related to specific families of periodic orbits, which are either ballistic or localized. In Ref. [San05] the central limit theorm and fine structures in the corresponding probability distributions were studied in a fully irrational double-triangle channel exhibiting normal diffusion. This model was simplified in the works of Jepps, Rondoni [Jep06], Sanders and Larralde [San06] leading to the zigzag or sawtooth channel depicted in Fig. 17.11 (d). The simulations reported in Ref. [Jep06] yielded superdiffusion for parallel zigzag-walls, both for rational and irrational billiards. For walls that were not parallel to each other only irrational cases were investigated suggesting essentially normal diffusion under parameter

May 8, 2007

4:1

360

World Scientific Book - 9in x 6in

klages˙book

Further topics in chaotic transport theory

variation when the horizon is finite, in agreement with the results of Alonso et al. [Alo02; Alo04].10 However, both groups of researchers reported subdiffusion if one of the angles is rational with precisely 90 degrees. These findings were largely confirmed by simulations of Sanders and Larralde [San06], who conjectured that if all vertex angles are irrationally related to π and not rationally related to each other, the billiard has a finite horizon and the walls are not parallel to each other, zigzag channels exhibit normal diffusion. In agreement with Ref. [Jep06] they found superdiffusion whenever the walls are parallel.11 Interestingly, in the case of superdiffusion their numerical results suggest that the anomalous diffusion exponent is a highly irregular function of control parameters. Similar results are available for subdiffusive exponents in this model [Jep06] and for superdiffusion in the standard map [Whi98]. This could point towards an underlying fractal parameter dependence of these anomalous exponents which would be a striking property of polygonal billiard channels, perhaps to be explained again in terms of topological instabilities as were generating the fractal transport coefficients extensively discussed in Part 1 of this book. An issue that needs further investigation is the numerically observed existence of subdiffusion in the zigzag channel. In the literature different formulas have been derived from which it can partially be inferred that subdiffusion cannot exist in Hamiltonian dynamical systems. As far as we know, there exist three different types of such equations. They all establish explicit relations between the exponent of the probability distribution function of Poincar´e recurrence times, P (t) ∼ 1/tµ , µ > 2 ,

(17.31)

< x2 >∼ tα , 0 ≤ α ≤ 2 .

(17.32)

α =µ−1

(17.33)

and the exponent of the mean square displacement, Note that the lower bound in Eq. (17.31) derives from Kac lemma stating that in Hamiltonian systems which are ergodic with non-zero measure on compact spaces the Poincar´e recurrence time, which is the first moment of the distribution given by Eq. (17.31), is finite [Zas02]. In Ref. [Zas00] the relation 10 The

above statement holds if one is quite generous concerning the error bars given in Ref. [Jep06]. Particularly, there is one case of subdiffusion for non-parallel walls, however, here the horizon is infinite, and as stated in Ref. [Jep06] the existence of logarithmic corrections [San06] has not been taken into account when analyzing the numerical data. 11 We thank D.P. Sanders for very helpful private correspondence related to polygonal billiard diffusion.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

361

17.4. Pseudochaotic diffusion

was extracted from a fractional Fokker-Planck-Kolmogorov equation applied to Hamiltonian dynamics. The result was claimed to hold for Hamiltonian systems exhibiting no self-similar structures in phase space. It was argued that this equation correctly describes diffusion in a large class of polygonal billiard channels [Zas01a; Zas03]. Combined with Eq. (17.31), this equation trivially implies that subdiffusion cannot exist for these systems. The same conclusion is obtained from a very different relation between these two exponents, which was derived from continuous time random walk theory [Zum99], 4−µ , 2<µ<3 α= . (17.34) 1 , 3<µ A third relation [Zas97], resulting from a renormalization group argument applied to a self-similar hierarchy of islands in Hamiltonian phase space, reads α =µ−2 .

(17.35)

The relation between Eqs. (17.34) and (17.35) and their validity was discussed controversially in the literature [Zum99; Ben99].12 As we have sketched above, deterministic transport in pseudochaotic channels is intimately related to number-theoretic properties, which makes numerical explorations highly delicate and might cast doubt onto the reliability of simulation results that are based on limited numerical precision. On the other hand, as we will further discuss below, these billiard channels feature an extremely correlated dynamics, which may suggest to checking in detail whether all mathematical assumptions underlying the above consequence of Kac lemma that there is no subdiffusion in Hamiltonian dynamics really apply to polygonal billiard channels. We may also remark that special cases of subdiffusion have already been observed in other Hamiltonian systems [Zas02]. It is worth mentioning that further pseudochaotic systems exist that are somewhat intermediate between polygonal billiards and the anharmonic chaotic chains discussed in Section 17.3.2. Examples are nonintegrable bingbang models featuring hard core potentials both for interparticle and on-site potentials of moving scatterers [Pro05]. Another example is provided by the alternate mass hard-core potential chain, see Ref. [Li05] and further references therein. The latter article also includes a brief review of different polygonal billiard channels and their dynamical properties. 12 We

thank E.G.Altmann for very helpful private correspondence on this point.

May 8, 2007

4:1

362

World Scientific Book - 9in x 6in

Further topics in chaotic transport theory

That the study of abstract non-Lyapunov chaotic models is not only of theoretical and mathematical interest but also bears direct relations to experiments is demonstrated particularly by Ref. [Zas05], where Zaslavsky and coworkers relate the highly complicated intertwined structure of invariant surfaces in confined plasmas to the topology of a multibar-insquare model, which is essentially a bounded version of the barrier channel Fig. 17.11 (a). Another example for experimental applications of this theory is provided by the semiclassical approach to the pseudointegrable ray dynamics of hexagonal shaped dielectric resonators which form a novel class of microcrystal lasers, see Ref. [Wie03] and further references therein. Hexagonal structures define also the honeycomb billiard studied in Ref. [Sch06], which yields a two-dimensional version of the zigzag channel shown in Fig. 17.11 (d). The study of this model was originally motivated by the observation of photon channeling in foams. A billiard that is not polygonal but nevertheless not Lyapunov chaotic is the circle. Though being an integrable system, studying escape rates from such a geometry reveals interesting relations between number-theoretic properties and possible experiments, e.g., on the escape of cold atoms from laser traps with holes [Bun05b]. As another somewhat related model we mention billiard channels where surface roughness is modeled by replacing the flat parallel walls with Koch curves being self-similar down to a certain level of fractality [Rus05]. This idea reminds one of our discussion of the flower-shaped billiard, see Section 8, where the circular shape of the Lorentz gas scatterers was modified by adding smaller circular petals onto the circumference. The Koch curve channel was actually motivated by experiments on molecular diffusion in nanopores. However, in contrast to simple billiard systems where the collisions are specular here the scattering at the walls was modeled stochastically according to Lambert’s cosine law, see Footnote 2 in Section 4.1. Other non-Lyapunov chaotic billiards that attracted much interest recently, both theoretically and experimentally, are mushroom billiards [Bun01]. We also remark that there appears to be a relation between pseudochaotic polygonal billiards and the recently evolving mathematical field of piecewise isometries, see Ref. [Viv06] and further references therein. Let us finally comment on the relation between diffusion and heat conduction in polygonal billiard channels. In Refs. [Alo99; Li02; Alo02; Li03a; Alo04; Li05] it was exploited that, in contrast to the chains of oscillators discussed in Section 17.3.2, in billiards heat transport can only be performed by the diffusing particles. Hence, there is a direct physical relation

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

363

17.4. Pseudochaotic diffusion

between particle diffusion and heat conduction, and as was already reported in Refs. [Li02; Alo02; Alo04], anomalous diffusion trivially implies a nonexistence of Fourier’s law. This crosslink was further explored by Li et al. in Ref. [Li03b] by deriving an explicit formula between the anomalous exponents of diffusion and heat conduction. That is, if the mean square displacement of a one-dimensional particle billiard is given by Eq. (17.32) and the thermal conductivity by Eq. (17.20) then, based on a heuristic mean first passage time argument, it was claimed that the two exponents α for diffusion and γ for heat conduction are related by γ = 2 − 2/α .

(17.36)

Interestingly, a different relation between these exponents was derived in Ref. [Den03]: By modeling the dynamics of a billiard channel in terms of a continuous time random walk on a discrete lattice, in the case of superdiffusion, α > 1, the relation γ =α−1

(17.37)

was obtained. The validity of these two different scaling laws was discussed controversially [Met04b; Li04]. Within this context, the applicability of continuous time random walk theory to diffusion in polygonal billiards was questioned [Li05]. Indeed, other works seem to confirm [Zwa83; Zas02; San06; Sch06] that there is a problem in straightforwardly applying such concepts, which in some way depend on the assumption of memory loss between successive collisions, to highly correlated pseudochaotic diffusion; see Section 6.2 for a related discussion. In summary, microscopic chaos in terms of an exponential dynamical instability is neither necessary nor sufficient in order to obtain normal diffusion. The same applies to heat conduction thus completing the third point on the list of conditions for the existence of Fourier’s law as stated in Section 17.3.2. Whether weaker, at least necessary “chaos” indicators can be found characterizing normal transport, perhaps in terms of sensitivity, periodic orbits or different (topological, metric, weak) mixing properties plus associated decay rates of correlations [Dev89; Gla93; Rob95; Kol97; Cor82] leading to more refined definitions of chaos remains an important open question [vB04].

May 8, 2007

4:1

World Scientific Book - 9in x 6in

364

17.5

Further topics in chaotic transport theory ∗

Summary

(1) We have presented a very short outline of different fluctuation relations generalizing the second law of thermodynamics to small systems in nonequilibrium situations. One line of this research has been motivated by thermostated dynamical systems leading to fluctuation theorems, which predict large deviation symmetry relations between positive and negative entropy production in NSS. We have discussed two fundamental formulations of such theorems due to Gallavotti and Cohen, and to Evans and Searles. The first one could be proven starting from what is called the chaotic hypothesis for NSS and is based on phase space contraction, the second one applies to transient nonequilibrium situations and was directly verified in experiments. The range of validity of these theorems intimately depends on the choice of the fluctuating variable and details of the thermostating mechanism. A second line of research yielded different kinds of work relations. The Jarzynski work relation establishes a connection between equilibrium free energy differences of a thermostated system driven out of equilibrium and the nonequilibrium work performed onto the system. The Crooks relation applies to a similar situation but considers the entropy production for both the forward and the time-reversed process. The Jarzynski work relation and equations that are formally identical with steady state and transient fluctuation theorems are obtained from it as special cases. However, it appears that all crosslinks between these different relations and their ranges of validity are not yet fully understood. (2) We have summarized recent findings on the Lyapunov instability of many-particle fluids in thermal equilibrium. Looking at phase space projections of the Lyapunov vector corresponding to the largest Lyapunov exponent shows that the growth process of the excitations determining this exponent is strongly localized. This particular dynamics is reflected in a very peaked functional form of the Lyapunov spectrum around the largest exponent. The Lyapunov exponents with the smallest absolute values, on the other hand, show degeneracy yielding step-like structures in the Lyapunov spectrum. The steps in the nonzero exponents are related to stable periodic patterns in the associated Lyapunov vectors, which have been called Lyapunov modes. They also reflect the geometry of the system. All these simulation results have been explained analytically by different theoretical approaches based

klages˙book

May 8, 2007

4:1

17.5.

World Scientific Book - 9in x 6in

∗ Summary

klages˙book

365

on random matrices, kinetic theory and periodic orbits. The importance of these findings for nonequilibrium transport is currently under investigation. (3) A fundamental question is to identify the microscopic conditions under which Fourier’s law for heat conduction is valid. We first discussed heat conduction in Hamiltonian chains of oscillators. They can be separated into two classes depending on whether there exists a non-zero substrate potential or not. In the general case of integrable (e.g., harmonic) interaction and on-site potentials the dynamics can be decomposed into independent modes, whose coupling is not strong enough to yield Fourier’s law. The non-existence of an on-site potential leads to momentum conservation, and typically13 this implies that Fourier’s law does not exist. In both cases the heat conductivity diverges with a power law in the chain length. For anharmonic chains it is conjectured that there exist at least two different universality classes defined by different power law exponents. These chains also demonstrate that microscopic chaos is not sufficient for normal heat conduction. For systems with a non-integrable on-site potential the heat conductivity usually exist. In order to better understand the role that microscopic chaos plays for Fourier’s law, heat conduction was studied in chaotic Hamiltonian particle billiards like Lorentz channels. In nonequilibrium these systems lack local thermodynamic equilibrium. This motivated the introduction of a billiard channel composed of rotating disks, which establish local thermodynamic equilibrium. The existence of non-ergodic behavior in this model led to further abstractions of this setting when analyzed mathematically. (4) The theoretical analysis of an experiment that aimed at verifying the existence of microscopic chaos in a fluid triggered a series of works on diffusion in systems that do not exhibit any exponential dynamical instability in the sense of Lyapunov. A paradigmatic example is the wind-tree model, which consists of randomly distributed scatterers in the plane that do not produce any nonlinear dynamical instability for a moving point particle. Simulations of this model yielded numerical results that are indistinguishable from the experimental data of a real fluid. This highlights the problem of a distinction between deterministic chaos and stochastic randomness on a microscopic level. 13 i.e.,

if the thermodynamic pressure of the chain is non-vanishing

May 8, 2007

4:1

366

World Scientific Book - 9in x 6in

Further topics in chaotic transport theory

Related work on transport in polygonal billiard channels confirms that Lyapunov chaos is not necessary for the existence of diffusion or normal heat conduction: Despite featuring only linear dispersion of infinitesimally close trajectories, these systems exhibit ergodic, weak mixing and even mixing properties as normally being associated with Lyapunov chaos. This novel type of dynamical randomness is induced by complicated topological structures underlying polygonal billiard domains, which generate pseudohyperbolic fixed points and non-integrable dynamics, as one can clearly see for pseudointegrable rational billiards. Such systems, which in the case of linear dispersion exhibit what may be called topology-induced chaos, form a subclass of pseudochaotic dynamics associated with algebraic dispersion of trajectories. These studies suggest finding new “chaos indicators” to replace standard Lyapunov exponents. Since in billiard channels heat is transported by the diffusing particles only, there is an intimate relation between heat conduction and diffusion. It was argued that there exist functional relationships both between the exponents of divergence for diffusion and heat conduction and between diffusion and Poincar´e recurrence times, however, the precise form of these relations is under discussion.

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Chapter 18 ∗

Conclusions

The final chapter consist of three parts: In Section 18.1 we outline what we believe represents a general picture of a chaotic dynamical systems approach to nonequilibrium statistical mechanics. Following the two central themes of this book, we summarize our main results in Section 18.2. Section 18.3 concludes this work with a number of open questions.

18.1

Microscopic chaos and nonequilibrium statistical mechanics: the big picture

The backbone of this book is provided by two fundamental approaches aiming at a general theory of NSS starting from microscopic chaos in the equations of motions of simple model systems: One of them proceeds along the lines of Hamiltonian dynamics supplemented by suitable (nonequilibrium) boundary conditions. Focusing on the parameter dependence of transport coefficients, Part 1 was largely inspired by this theory. The other approach applies to nonequilibrium situations where a coupling to a thermal reservoir is needed in order to sustain stationary states associated with specific transport processes. Using deterministic modelings of thermal reservoirs, this approach renders dynamical systems non-Hamiltonian as was analyzed in detail in Part 2 of this book. For a brief outline of both theories see our general introduction Chapter 1. In Fig. 18.1 we identify fundamental ingredients of both these approaches by conceptually relating them to the well-estabilshed equilibrium theory of many-particle systems.1 Nonequilibrium theories crosslinking these different levels of description were particularly worked out 1 We thank M.Dolowschiak for inspiration as regards this figure, who presented a related summary during a seminar talk at the MPIPKS Dresden in 2004.

367

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

∗ Conclusions

368

external forces

microscopic

macroscopic

equilibrium

nonequilibrium steady states

dynamical systems theory

ergodic hypothesis

microscopic chaos

statistical mechanics

Gibbs ensembles

fractal SRB measures

thermodynamics

thermodynamic properties

transport properties

general theory of nonequilibrium statistical mechanics starting from microscopic chaos? Fig. 18.1 Conceptual foundations of a theory of nonequilibrium statistical mechanics that starts from microscopic deterministic chaos, in analogy to the equilibrium physics of many-particle systems.

by Gallavotti, Ruelle, Gaspard and Dorfman, see Refs. [Gal98; Rue04; Gas06] for shorter and Refs. [Gas98a; Dor99; Rue99b; Gal99] for detailed presentations. As we have mentioned at the very beginning of this book, the goal of statistical mechanics is to understand macroscopic properties of matter, both in and out of equilibrium, starting from the microscopic equations of motion of single atoms or molecules. These equations are typically highly nonlinear, hence their understanding requires methods of dynamical systems theory [Cvi07]. In equilibrium one usually starts from modeling a physical system in terms of Hamiltonian equations of motion. In nonequilibrium the situation is more difficult, since under application of external forces one typically has to couple the system to a thermal reservoir. The escape rate formalism and related theories still employ a purely Hamiltonian formalism [Gas98a], however, as we have shown in Part 2, a deterministic modeling of thermal reservoirs leads to non-Hamiltonian equations of motion [Eva90b]. A rigorous foundation of equilibrium statistical mechanics is unthinkable without an assumption like Boltzmann’s ergodic hypothesis, which enables one to combine the laws of mechanics with concepts from probability theory [Gal98; Gal99; Dor99]. By using Gibbs ensembles, fundamental principles of equilibrium statistical mechanics can then be derived, which in turn

May 8, 2007

4:1

World Scientific Book - 9in x 6in

18.1. Microscopic chaos and nonequilibrium statistical mechanics: the big picture 369

enable the description of a wide variety of systems in terms of equilibrium thermodynamical properties; see, e.g., Refs. [Pen79] and [Tod92] for short and long presentations of this theoretical framework. Ideally, one wishes to carry out a similar programme for the case of nonequilibrium systems. However, here the problem is more intricate: Up to now there does not seem to exist a universally accepted theory of nonequilibrium steady states, particularly as far as the modeling of thermal reservoirs is concerned [Gal04b; Gal04a]. The microscopic theories we have been touching upon in this book favor a starting point that not only requires ergodicity but also some kind of “microscopic chaos”, understood in terms of a Lyapunov dynamical instability. This fact is particularly captured by the chaotic hypothesis (CH) of Gallavotti and Cohen [Gal95a; Gal95b], see Section 17.1.2, which was proposed precisely in order to generalize Boltzmann’s ergodic hypothesis to nonequilibrium situations. However, the CH, formulating the assumption that many-particle systems in nonequilibrium are dynamically somewhat “well-behaved”,2 implies even more, namely the existence of a specific type of measure that is ideally suited for calculating transport properties. We have encountered examples of such SRB measures in Section 4 of Part 1 and all over Part 2 of this book: They are fractal along stable but smooth along unstable directions in phase space thus enabling one to perform calculations of statistical physical quantities in terms of projections onto the unstable manifolds. Applying different types of thermal reservoirs defines different nonequilibrium ensembles, which ideally yield the same transport properties [Eva90b; Rue99b; Gal99]. These intimate connections between microscopic chaos, fractal SRB measures and transport in nonequilibrium statistical mechanics, illustrated in Fig. 18.1, become apparent in the problem of proving fundamental laws of nonequilibrium thermodynamics like linear response from first principles. Starting such proofs from microscopic equations of motion inevitably requires dealing with SRB measures [Che93a; Che93b; Rue97a], an issue we have touched upon particularly in Section 11.2.4. Related to this point, a crucial development over the past 20 years was the discovery of formulas that express transport coefficients in terms of dynamical systems quantities. Originally being derived independently for the escape rate [Gas90; 2 The

CH actually requires these systems to be transitive and to have Anosov-like properties [Eck85], i.e., loosely speaking to be ergodic and smooth plus featuring a hyperbolic structure [Gal95b].

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

∗ Conclusions

370

Dor95] and the thermostated dynamical systems approach [Pos88; Eva90a], we illustrate their fundamental form for the case of diffusion in two dimensions, ! 2 X L D= λi (FL ) − hKS (FL ) (L → ∞) . (18.1) const. λi >0

Here λi denote the positive Lyapunov exponents of a dynamical system of extension L, hKS its Kolmogorov-Sinai entropy, and FL stands for the set with fractal SRB measure on which the system’s stationary state is defined. Analogous equations hold for other transport coefficients, see particularly Section 13.3 for details. Interestingly, the fluctuation theorems discussed in Section 17.3.2 yield formally very similar relationships by rewriting, e.g., Eq. (17.1) as 1 (18.2) ξ = (ln ρτ (ξ) − ln ρτ (−ξ)) (τ → ∞) , τ where ξ defines some nonequilibrium entropy production in a NSS and ρτ (ξ) a family of coarse-grained probability distributions for this quantity over trajectory segments of length τ . Both sets of formulas expressing macroscopic transport in terms of microscopic dynamical properties thus share the same structure in that they consist of a difference between two large deviation quantities. Further formulas of a similar nature have been obtained for transport coefficients in terms of hydrodynamic modes as mentioned in Section 13.3 and by a very recently derived relationship between different entropies incorporating a time-reversed entropy per unit time [Gas04]. Such “balance equations” establishing explicit relationships between microscopic reversible dynamics and macroscopic irreversible quantities form a fundamental ingredient of all theories of nonequilibrium statistical mechanics that start from microscopic deterministic chaos [Gas05; Gas06], at least close to thermodynamic equilibrium. This appears to present an ideal scenario as it currently emerges both from Hamiltonian and non-Hamiltonian approaches to nonequilibrium transport. However, such a picture is subject to several lines of criticism: First of all, as we have learned particularly in Section 17.4, microscopic chaos in the sense of Lyapunov is neither necessary nor sufficient for ordinary transport laws. For example, heat conduction and normal diffusion exist under much weaker assumptions on the underlying dynamical instability as exemplified by various types of polygonal billiards. This suggests that the above mentioned hypothesis on microscopic chaos underlying nonequilibrium statistical mechanics has to be widened in order to take such phe-

May 8, 2007

4:1

World Scientific Book - 9in x 6in

18.2. Assessment of the main results

klages˙book

371

nomena into account, perhaps by introducing other chaos identifiers than Lyapunov exponents. This argument particularly applies to the chaotic hypothesis, which seems to define a rather narrow frame for carrying out the programme of connecting macroscopic physical quantities with microscopic chaotic dynamical properties: It strictly requires chaos in the sense of Lyapunov, is built upon the existence of SRB measures and to a large extent relies on the identification of thermodynamic entropy production with phase space contraction. The chaotic hypothesis thus excludes a wide range of dynamical systems exhibiting interesting statistical physical behavior: It does not apply to transport in pseudochaotic systems or to Hamiltonian systems with a mixed phase space. One may indeed say that all the lowdimensional models discussed in Part 1 of this book, which exhibit irregular or fractal transport coefficients due to non-trivial dynamical correlations, are outside of the scope of the chaotic hypothesis [Kla99a]. The usefulness of mathematical SRB measures for building up a general theory of nonequilibrium transport has been questioned along similar lines [Hoo99; Eva05]. Furthermore, it does not seem to be clear which type of measure governs transport in polygonal billiards. And as we have shown in Part 2 of this book, there exists a wide class of thermal reservoirs for which thermodynamic entropy production cannot be identified with phase space contraction somewhat contradicting the theoretical frame in which the chaotic hypothesis has been worked out. This problem as well as the issue of fractal transport coefficients will be further discussed in the following. 18.2

Assessment of the main results

Corresponding to the first two parts of this book, we now summarize the essential themes and results of both parts one after the other. 18.2.1

Existence of fractal transport coefficients

Chapters 2 to 9 focused on the phenomenon of irregular, respectively fractal transport coefficients in certain types of chaotic dynamical models. The different facets of the analysis associated with this theme we recapitulate as follows: (1) We developed an arsenal of methods suitable for computing parameter-dependent transport coefficients of deterministic, chaotic dy-

May 8, 2007

4:1

372

World Scientific Book - 9in x 6in

klages˙book

∗ Conclusions

namical systems. One type of these methods starts from the fact that, under certain conditions, the Liouville (Frobenius-Perron) operator of dynamical systems can be cast into the form of topological transition matrices. One may then either solve the eigenvalue problems of these matrices, by relating transport coefficients to specific eigenvalues of the Liouville operator, or one may simply iterate these matrices numerically hence obtaining the complete time-dependent solution of the Liouville equation. A rigorous mathematical solution for calculating deterministic transport coefficients of a simple model started from the generating function of the system and expressed the transport coefficients in form of some coupled recursion relations. A third method employs TaylorGreen-Kubo formulas for diffusion and relates diffusion coefficients to a combination of invariant probability densities and fractal generalized Takagi functions. The latter can be computed by again solving functional recursion relations, whereas the former are again obtained by some matrix method. Of course, these are not the only methods being available but just the ones used by us, as outlined in Part 1. (2) These methods were applied to a variety of models exhibiting deterministic transport. Our models were constructed such that they featured an increasing degree of complexity by gradually approaching physical reality. We started with simple one-dimensional, piecewise linear maps on the line yielding rather abstract, but on the other hand exactly solvable models. We then studied a multibaker map representing a two-dimensional extension of such one-dimensional systems. For truly nonlinear one-dimensional maps no exact results could be obtained anymore, and instead we employed computer simulations. The same applies to randomly perturbed one-dimensional maps as well as to deterministic diffusion in Hamiltonian and non-Hamiltonian particle billiards. (3) Studying these models with the set of methods outlined above, we found that they all displayed irregular, respectively fractal transport coefficients. That is, we provided examples for conductivities, normal and anomalous diffusion coefficients as well as chemical reaction rates that are all irregular, respectively fractal functions of control parameters. In the case of one-dimensional maps it could be shown, qualitatively and quantitatively, that these transport coefficients are strictly fractal. The diffusion coefficients of the three different particle billiards we investigated also yielded irregularities on large and fine scales, however, whether they are of a fractal origin remains an open

May 8, 2007

4:1

World Scientific Book - 9in x 6in

18.2. Assessment of the main results

klages˙book

373

question. (4) We explored different ways of understanding these fractal structures. What we coined the turnstile approach relates parameterdependent irregularities in a transport coefficient to specific microscopic orbits of particles characterizing specific parameter values. These orbits change in a complicated way under parameter variation due to the topological instability of the chaotic dynamical system hence transferring the instability to the associated transport coefficients. A second approach used the fact that Taylor-Green-Kubo formulas provide convenient series expansions for, at least, deterministic diffusion coefficients. In these equations the first term represents random walk approximations well-known from stochastic theory, whereas higher-order terms feature dynamical correlations of increasingly high order. By looking at the convergence of such a series one can thus identify contributions to fractal diffusion coefficients in terms of higher-order dynamical correlations. (5) Part 1 of this book was structured such that it provides an access road from simple models towards experiments. Particularly, we ended up with a model for single granular particles on vibratory conveyors whose predictions, we believe, can be directly tested in experiments. Further crosslinks to experimental work have been described on many occasions throughout this first part: Here we just mention the connection between simple maps and current reversals in deterministic ratchets, irregular diffusion coefficients in driven pendulum models as they describe the phase diffusion in Josephson junctions, the experimental finding of irregular magnetoresistances in antidot lattices, which are very similar to periodic Lorentz gases, and a similarity between flowershaped Hamiltonian particle billiards and diffusion in nanopores. That fractal transport coefficients appear to be rather stable against random perturbations increases the chance to see such phenomena in real physical experiments. Understanding the origin of fractal transport coefficients may furthermore be used to design systems with such unusual properties for specific applications. In the light of the discussion presented in Part 2, one may raise the question of what extent one may consider the fractal transport coefficients of Part 1 to be “universal” for transport in nonequilibrium steady states. In Chapter 5 we demonstrated that strong random perturbations eventually smooth out these fractal structures, as expected from naive physical reason-

May 8, 2007

4:1

374

World Scientific Book - 9in x 6in

klages˙book

∗ Conclusions

ing. We believe that, similarly, in high-dimensional systems a large number of degrees of freedom eliminates any measurable fractal parameter dependence of transport coefficients. This would also reconcile a low-dimensional deterministic dynamical systems’ approach towards non-equilibrium transport with conventional results from nonequilibrium thermodynamics such as linear response theory. On the other hand, irregular, respectively fractal transport coefficients were shown to exist for almost all models studied in this book when they shared specific properties. We thus arrive at the conclusion that fractal transport coefficients are typical for a specific class of dynamical systems exhibiting the properties of being low-dimensional, spatially periodic and deterministically chaotic.

18.2.2

Universalities in thermostated dynamical systems?

The major theme of Part 2 was to explore to which extent the description of NSS generated by conventional Gaussian and Nos´e-Hoover dynamics is universal. In this endeavour Chapter 13 played a central role, where we summarized what we consider to be the most important links between chaos and transport resulting from Gaussian and Nos´e-Hoover dynamics. We now briefly go through this list again and, on the basis of the analysis provided in Chapters 14 to 16, come to a final assessment. (1) non-Hamiltonian dynamics for NSS: In our view, this type of mechanics yields a rather natural description of NSS generated by external forces. As we argued on many occasions, the non-Hamiltonian character of thermostated systems such as Langevin, Gaussian or Nos´e-Hoover dynamics can be understood with respect to projecting out spurious reservoir degrees of freedom. From this point of view we do not see any reason to insist on a Hamiltonian modeling of NSS. (2) phase space contraction and entropy production: Conventional Gaussian and Nos´e-Hoover thermostats display a default identity between these two quantities, which is at the heart of linking thermodynamics to nonlinear dynamics in dissipative dynamical systems. However, it is not too difficult to construct counterexamples of deterministic and time-reversible thermal reservoirs generating well-defined NSS but not exhibiting this identity. We just mention non-ideal Gaussian and Nos´e-Hoover thermostats and sheared hard-disk fluids thermostated by deterministic scattering. Care should therefore be taken for exploring

May 8, 2007

4:1

World Scientific Book - 9in x 6in

18.2. Assessment of the main results

klages˙book

375

the second law of thermodynamics on the basis of this identity. (3) transport coefficients and dynamical systems quantities: Conventional Gaussian and Nos´e-Hoover dynamics features the Lyapunov sum rule linking transport coefficients to Lyapunov exponents. However, the existence of such a simple functional relationship is intimately connected with the existence of the abovementioned identity. Hence, it is not surprising that this relation between chaos quantities and transport coefficients is not universal either. Non-ideal thermostats and thermostating by deterministic scattering yield again counterexamples. Only very close to equilibrium do such relations seem to boil down to formulas exhibiting the same structure, as discussed in the previous section, possibly providing a universal characterisation of a linear response regime nearby equilibrium. We furthermore emphasize that there are broad classes of thermostated dynamical systems not exhibiting a conjugate pairing of Lyapunov exponents. (4) fractal attractors characterizing NSS: In all deterministically thermostated dynamical systems analyzed in this book the resulting attracting sets were characterized by fractal structures. This fractality, which is typical for SRB measures often associated with these attractors, is thus a promising candidate for a universal property of NSS associated with deterministic and time-reversible thermal reservoirs, even very far from equilibrium. However, note that the detailed topology of the attracting sets still intimately depends on the type of thermal reservoir applied. This is represented by the different types of bifurcation diagrams for the driven periodic Lorentz gas connected to different thermal reservoirs. (5) nonlinear response in the thermostated driven periodic Lorentz gas: Despite a mathematical proof for the existence of linear response in the Gaussian thermostated driven periodic Lorentz gas, no clear signs of a regime of linear response could be detected in computer simulations for the Gaussian as well as for other thermostated Lorentz gases. Even worse, applying different thermostats yielded significantly different results for the field-dependent electrical conductivity of this model. We must therefore conclude that there is no equivalence of nonequilibrium ensembles for the driven periodic Lorentz gas in this regime of field strengths. In the case of a thermostated hard disk fluid the situation appears to be better in that, at least, there is an equivalence of nonequilibrium ensembles as far as thermodynamic properties are concerned. However, we emphasize that on the level of chaos quan-

May 8, 2007

4:1

World Scientific Book - 9in x 6in

klages˙book

∗ Conclusions

376

tities there is no equivalence either. We thus conclude that, from the point of view of nonequilibrium thermodynamics, the periodic Lorentz gas seems to be a rather delicate dynamical system. This appears to be due to the low dimensionality of the model featuring a single moving point particle only. In this case statistical properties arise from the chaotic collisions of the particle with the fixed scatterers and contain intricate dynamical correlations in time and space. The different results for the field-dependent electrical conductivity presented in this book demonstrate the limits of standard thermodynamic descriptions applied to the thermostated model. However, in many other respects both the periodic Lorentz gas with and without external fields still exhibits “nice” thermodynamic behavior. And even the existence of irregularities in the field-dependent electrical conductivity should not be deemed “unphysical”: Interacting many-particle systems display conventional thermodynamic behavior due to their many degrees of freedom, whereas chaotically moving single particles may exhibit unconventional dynamical phenomena reflecting characteristic nonlinearities in their equations of motion. To us it appears that the periodic Lorentz gas, as well as multibaker maps, are right in-between these two large classes of high and lowdimensional, interacting and non-interacting many-particle systems. Such intermediate dynamics may thus display both ordinary thermodynamic behavior as well as specific chaotic dynamical properties. This assessment points back to our discussion of fractal transport coefficients summarized above. An adequate theoretical description thus needs to develop a wellbalanced combination of both statistical and dynamical systems methods in order to adequately assess both the chaos and transport properties of this highly interesting type of dynamical systems. As we have argued frequently throughout this book, we believe that such a theory is ideally suited to describe transport in the currently emerging field of the nanosciences, which involves the study of the very same type of “intermediate” physical systems.

18.3

Important open questions

Here we highlight a number of central problems that we consider worth to be studied in future research. More detailed open questions were already mentioned in the course of this book on many occasions.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

18.3. Important open questions

18.3.1

klages˙book

377

Fractal transport coefficients

(1) calculating fractal transport coefficients: For more non-trivial models such as the climbing sine map and particle billiards, transport coefficients could only be computed on the basis of computer simulations or by extremely simple and insufficient analytical random walk approximations. It would thus be important to further improve the available theoretical methods in order to obtain exact results for such systems. Our favorite approach would be to proceed along Taylor-Green-Kubo formulas by evaluating both generalized Takagi functions in more general settings, combined with improved matrix methods for computing invariant densities. Such methods may eventually also be applied to non-trivial Hamiltonian dynamical systems systems such as sawtooth [Dan89b; Dan89a], standard [Rec80; Rec81] and Harper [Leb98] maps as well as to periodic Lorentz gases under electric and magnetic fields [Gei90; Fle92]. For all these types of models irregular transport coefficients are well-known to exist theoretically, however, they have not yet been analyzed with respect to a possible fractal origin. These models are furthermore close to various experiments, as we will explain below. (2) more fractal transport coefficients: We strongly believe that other transport coefficients such as, for example, viscosities and thermal conductivities will also turn out to be fractal for specific classes of physical dynamical systems. Concerning viscosities one may study the simple two-particle fluid of Refs. [Bun96; Vis03a; Vis03b], for thermal conduction one may consider the Lorentz channel of Section 17.3.2, perhaps both with and without rotating scatterers. (3) fractal transport coefficients in experiments? This issue still asks for some pioneering, particularly experimental work to be performed. First of all, we do not suggest that “true” fractal functions may really be observable in experiments. Evidently, random perturbations and limits of experimental precision will always smooth out such curves on fine scales, as discussed in Chapter 5. However, the same chapter suggests that fractal behavior may be at the origin of smoothed-out oscillations of transport coefficients on very fine scales, and we believe that these oscillations may very well be observable in experiments. Our favorite candidates of experimentally accessible systems in which irregular transport coefficients may be detected are ratchets exhibiting drift-diffusion processes [Jun96;

May 8, 2007

4:1

378

World Scientific Book - 9in x 6in

klages˙book

∗ Conclusions

Rei02], antidot lattices with electric and magnetotransport [Wei91; Wei95; Wei97; Wie01], Josephson junctions displaying phase diffusion [Gei85; Wei00; Tan02d], and granular systems consisting of a diffusing bouncing ball as discussed in Chapter 9 [Far99; Pre02; Gro03; Gro04]. (4) quantum mechanical fractal transport coefficients? An interesting question is whether, or to which extent, classical fractal transport coefficients survive in quantum mechanical dynamical systems of the type discussed in Part 1. An appropriate model to start with would be the purely diffusive multibaker map with a fractal diffusion coefficient presented in Chapter 4. This model has recently been quantized at a certain parameter value [Woj02; Woj03; Woj04]. Similarly, a very interesting issue is to investigate whether there is any connection between the fractality of transport coefficients discussed here and the fractal conductance fluctuations predicted theoretically and observed experimentally in the mesoscopic nonlinear Hamiltonian systems of Refs. [Ket96; Heg96; Sac98; Huf01]. Note that in the latter case the fractality is intimately related to the mixed phase space of the Hamiltonian systems under study. (5) anomalous fractal transport coefficients: In Section 6.2 we have shown that a simple intermittent map can display fractal parameterdependent subdiffusion in terms of a suitably defined generalized diffusion coefficient. It seems clear that the same result will hold if one studies superdiffusion in similar types of maps, however, this behavior may be supplemented by interesting dynamical phase transitions between different anomalous parameter regions [Zum93b; Zum95]. Another interesting question is whether the anomalous exponents involved in suband superdiffusive laws are also fractal functions of control parameters, as is indicated by simulation results [Whi98; San06; Jep06]. Whether there exist both normal and anomalous fractal transport coefficients in pseudochaotic systems such as polygonal billiard channels defines another important open problem. As suggested by Refs. [Alo02; San06; Jep06], the topological instability under parameter variation might even be stronger in systems without Lyapunov chaos leading to a perhaps even more peculiar fractality of transport coefficients. On a more fundamental level, the study of anomalous transport asks for the definition of new chaos indicators beyond Lyapunov exponents. This may shed more light onto the precise dynamical randomness assumptions needed

May 8, 2007

4:1

World Scientific Book - 9in x 6in

18.3. Important open questions

klages˙book

379

to develop an anomalous deterministic theory of nonequilibrium transport similar to the one sketched in Fig. 18.1.

18.3.2

Thermostated dynamical systems

(1) defining a nonequilibrium entropy: From our point of view it is not enough to construct a theory of NSS starting from the ordinary Gibbs entropy. A proper definition of a nonequilibrium entropy should yield results that are compatible with standard nonequilibrium thermodynamics. In the case of the driven Lorentz gas we think here of entropy production in terms of Clausius’ entropy combined with Joule’s heat. To us it appears that, apart from other problems, the Gibbs entropy may contain spurious contributions compared to a thermodynamic entropy production, since it samples the complete phase space, which is at variance to an entropy production in terms of a heat transfer. This problem may be eliminated by using a revised definition of the Gibbs entropy based on a suitable coarse-graining. It might be interesting to apply recent methods of coarse-grained entropies to our examples of thermostated dynamical systems not yielding an identity between phase space contraction and entropy production. One may hope that starting from such an amended Gibbs entropy ordinary thermodynamics can be recovered even for our more general examples of thermal reservoirs. (2) fluctuation relations: Apart from the fractality of attractors and formulas relating transport coefficients to chaos indicators nearby equilibrium, the fluctuation relations outlined in Section 17.1 yield another important candidate for universal results characterizing NSS in dynamical systems. However, it seems that fluctuation theorems have mainly been tested in thermostated dynamical systems furnishing an identity between phase space contraction and entropy production. It would thus be important to check for their existence in thermostated systems not featuring this identity, such as the ones described in Part 2. Interestingly, the existence of a certain class of fluctuation theorems is intimately linked to the validity of the chaotic hypothesis mentioned in the introduction. Hence, testing these fluctuation theorems one also studies, to some extent, the validity of the chaotic hypothesis. We furthermore suggest looking for fluctuation theorems in experiments on cell motility.

May 8, 2007

4:1

380

World Scientific Book - 9in x 6in

klages˙book

∗ Conclusions

(3) fractal attractors in stochastic dynamical systems? As far as we know, this important question is not really settled. In order to solve this problem one may study, for example, the driven periodic Lorentz gas or a hard disk fluid under shear supplemented by stochastic boundaries. It might then be elucidating to compare the respective results to the ones obtained from thermostating by deterministic scattering. (4) NSS for generalized Nos´ e-Hoover dynamics and for active Brownian particles: Following the philosophy of Nos´e-Hoover there emerged a large collection of generalized deterministic and timereversible thermostats for which it is still not clear to which extent they are functioning under nonequilibrium conditions. If they do, they may provide further access roads towards an analysis of chaos and transport properties of NSS pointing beyond ordinary Gaussian and Nos´e-Hoover dynamics. The same reasoning applies to active Brownian particles, which may be considered as another type of thermostated dynamical systems. We also mention the open problem of studying Lyapunov modes in nonequilibrium situations as described in Section 17.2. (5) quantum-mechanical formulations of thermal reservoirs: We did not say anything about extensions of classical deterministic and time-reversible thermostats towards the quantum regime. It seems that this field is currently evolving, see Refs. [Gri89; Kus93; Kus95; Men01; Kus02; Men03] and further references therein. Eventually, it would be interesting to study to which extent these approaches furnish links between chaos and transport being analogous to their classical counterparts. Note added in proof Research as the one covered by this book is a very active, ongoing process. After submission of this work already new literature appeared that is relevant to specific topics discussed in the book. Please see the book’s homepage under www.maths.qmul.ac.uk/∼klages/cftbook for updates on such references.

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Bibliography

[Aar97]

[Abr94] [Agu01]

[Agu03]

[Ald63]

[All87]

[All97]

[Alo99]

[Alo02]

[Alo04]

[And80]

J. Aaronson, An introduction to infinite ergodic theory, volume 50 of Mathematical Surveys and Monographs. (American Mathematical Society, Providence, 1997). (Cited on pages 108 and 116.) R. Abraham and J.E. Marsden, Foundations of mechanics. (AddisonWesley, Reading, 1994). (Cited on pages 192 and 215.) J. Aguirre, J.C. Vallejo and M.A.F. Sanjuan, Wada basins and chaotic invariant sets in the Hnon-Heiles system. Phys. Rev. E 64, 066208/1– 11 (2001). (Cited on page 143.) J. Aguirre and M.A.F. Sanjuan, Limit of small exits in open Hamiltonian systems. Phys. Rev. E 67, 056201/1–7 (2003). (Cited on page 144.) B.J. Alder, W.G. Hoover and T.E. Wainwright, Cooperative motion of hard disks leading to melting. Phys. Rev. Lett. 11, 241–243 (1963). (Cited on page 182.) M.D. Allen and D.J. Tildesley, Computer simulation of liquids. (Clarendon Press, Oxford, 1987). (Cited on pages 1; 8; 167; 187; 248; 288 and 291.) K.T. Alligood, T.S. Sauer and J.A. Yorke, Chaos - An introduction to dynamical systems. (Springer, New York, 1997). (Cited on pages 13; 20; 52; 100; 107 and 273.) D. Alonso, R. Artuso, G. Casati and I. Guarneri, Heat conductivity and dynamical instability. Phys. Rev. Lett. 82, 1859–1862 (1999). (Cited on pages 149; 339; 345 and 362.) D. Alonso, A. Ruiz and I. de Vega, Polygonal billiards and transport: Diffusion and heat conduction. Phys. Rev. E 66, 066131/1–15 (2002). (Cited on pages 6; 246; 353; 357; 358; 359; 360; 362; 363 and 378.) D. Alonso, A. Ruiz and I. de Vega, Transport in polygonal billiards. Physica D 187, 184–199 (2004). (Cited on pages 102; 359; 360; 362 and 363.) H.C. Andersen, Molecular dynamics simulations at constant pressure and/or temperature. J. Chem. Phys. 72, 2384–3173 (1980). (Cited on

381

klages˙book

May 8, 2007

4:1

382

[And82] [And85] [And86]

[Aok03]

[Ari01]

[Arn68]

[Art91] [Art93]

[Art94] [Art97a] [Art97b]

[Art00]

[Art03]

[Art04]

[Art05] [Art06]

[Asp98]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

pages 8 and 248.) L. Andrey, Ideal gas in empty space. Il Nuovo Cimento 69B, 136–144 (1982). (Cited on page 237.) L. Andrey, The rate of entropy change in non-Hamiltonian systems. Phys. Lett. 111A, 45–46 (1985). (Cited on pages 207; 236 and 237.) L. Andrey, Note concerning the paper “The rate of entropy change in non-Hamiltonian systems”. Phys. Lett. 114A, 183–184 (1986). (Cited on page 207.) K. Aoki and D. Kusnezov, Lyapunov exponents and the extensivity of dimensional loss for systems in thermal gradients. Phys. Rev. E 68, 056204/1–6 (2003). (Cited on pages 191; 219; 299; 342; 343 and 344.) C.M. Arizmendi, F. Family and A.L. Salas-Brito, Quenched disorder effects on deterministic inertia ratchets. Phys. Rev. E 63, 061104/1–5 (2001). (Cited on pages 84 and 89.) V.I. Arnold and A. Avez, Ergodic problems of classical mechanics. (W.A. Benjamin, New York, 1968). (Cited on pages 223; 233; 272 and 353.) R. Artuso, Diffusive dynamics and periodic orbits of dynamical systems. Phys. Lett. A 160, 528–530 (1991). (Cited on pages 4 and 21.) R. Artuso, G. Casati and R. Lombardi, Periodic orbit theory of anomalous diffusion. Phys. Rev. Lett. 71, 62–64 (1993). (Cited on pages 4; 5 and 108.) R. Artuso, Recycling deterministic diffusion. Physica D 76, 1–7 (1994). (Cited on page 4.) R. Artuso, Anomalous diffusion in classical dynamical systems. Phys. Rep. 290, 37–47 (1997). (Cited on pages 5 and 108.) R. Artuso, G. Casati and I. Guarneri, Numerical study on ergodic properties of triangular billiards. Phys. Rev. E 55, 6384–6390 (1997). (Cited on pages 353 and 357.) R. Artuso, I. Guarneri and L. Rebuzzini, Spectral properties and anomalous transport in a polygonal billiard. Chaos 10, 189–194 (2000). (Cited on pages 353; 354; 357 and 358.) R. Artuso and G. Cristadoro, Anomalous transport: A deterministic approach. Phys. Rev. Lett. 90, 244101/1–4 (2003). (Cited on pages 5 and 108.) R. Artuso and G. Cristadoro, Periodic orbit theory of strongly anomalous transport. J. Phys. A: Math. Gen. 37, 85–103 (2004). (Cited on pages 5 and 108.) R. Artuso and G. Cristadoro, Anomalous deterministic transport. Chaos 15, 015116/1–7 (2005). (Cited on page 355.) R. Artuso, L. Cavallasca and G. Cristadoro, Recycling Parrondo games. J. Phys. A: Math. Gen. 39, 1285–1295 (2006). (Cited on pages 30; 46 and 53.) T. Aspelmeier, G. Giese and A. Zippelius, Cooling dynamics of a dilute gas of inelastic rods: a many-particle simulation. Phys. Rev. E 57,

May 8, 2007

4:1

Bibliography

World Scientific Book - 9in x 6in

klages˙book

383

857–865 (1998). (Cited on page 277.) R. Dean Astumian, Thermodynamics and kinetics of a Brownian motor. Science 276, 917–922 (1997). (Cited on pages 50; 53 and 54.) [Ast02] R. Dean Astumian and P. H¨ anggi, Brownian motors. Phys. Today 55, 33–39 (2002). (Cited on pages 50 and 54.) [Ayt99] G. Ayton and D.J. Evans, On the asymptotic convergence of the transient and steady-state fluctuation theorems. J. Stat. Phys. 97, 811–815 (1999). (Cited on page 328.) [Bak85] P. Bak, T. Bohr and M.H. Jensen, Mode-locking and the transition to chaos in dissipative systems. Physica Scripta T9, 531–539 (1985). (Cited on pages 53; 100 and 107.) [Bal04] P. Balint and S. Troubetzkoy, Rotor interaction in the annulus billiard. J. Stat. Phys. 117, 681–702 (2004). (Cited on page 347.) [Bal05] R. Balescu, Aspects of anomalous transport in plasmas, volume 18 of Series in Plasma Physics. (CRC Press, London, 2005). (Cited on page 108.) [Ban92] J. Banks, J. Brooks, G. Cairns, G. Davis and P. Stacey, On Devaney’s definition of chaos. Am. Math. Monthly 99, 332–334 (1992). (Cited on page 223.) [Bar93] A. Baranyai, D.J. Evans and E.G.D. Cohen, Field-dependent conductivity and diffusion in a two-dimendional Lorentz Gas. J. Stat. Phys. 70, 1085–1098 (1993). (Cited on pages 186; 191; 198; 199 and 287.) [Bar94] R. Bartussek, P. H¨ anggi and J.G Kissner, Periodically rocked thermal ratchets. Europhys. Lett. 28, 459–464 (1994). (Cited on page 50.) [Bar97] E. Barkai and J. Klafter, Crossover from dispersive to regular transport in biased maps. Phys. Rev. Lett. 79, 2245–2248 (1997). (Cited on pages 30 and 90.) [Bar98] E. Barkai and J. Klafter, Deterministic transport in biased maps: Crossover from dispersive to regular transport. Phys. Rev. E 57, 5237– 5246 (1998). (Cited on pages 30 and 90.) [Bar00a] M. Barbi and M. Salerno, Phase locking effect and current reversals in deterministic underdamped ratchets. Phys. Rev. E 62, 1988–1994 (2000). (Cited on page 52.) [Bar00b] E. Barkai, V. Fleurov and J. Klafter, One-dimensional stochastic L´evyLorentz gas. Phys. Rev. E 61, 1164–1169 (2000). (Cited on page 84.) [Bar03a] E. Barkai, Aging in subdiffusion generated by a deterministic dynamical system. Phys. Rev. Lett. 90, 104101/1–4 (2003). (Cited on pages 108; 115; 116 and 117.) [Bar03b] E.J. Barth, B.B. Laird and B.J. Leimkuhler, Generating generalized distributions from dynamical simulation. J. Chem. Phys. 118, 5759– 5768 (2003). (Cited on pages 223 and 225.) [Bar05] A.L. Baranovski, On generation of chaotic M-sequences. in: Proceedings of the international symposium on nonlinear theory and its applications, pages 1–4, Bruges, 2005. (Cited on pages 4; 39 and 69.) [Bar06] A.L. Baranovski and A.J. Lawrance, Autocorrelation chaos from piece-

[Ast97]

May 8, 2007

4:1

384

World Scientific Book - 9in x 6in

klages˙book

Bibliography

wise linear maps. preprint, 2006. (Cited on pages 4; 39 and 69.) C. Beck and G. R¨ opstorff, From dynamical systems to the Langevin equation. Physica A 145, 1–14 (1987). (Cited on pages 90 and 172.) [Bec93] C. Beck and F. Schl¨ ogl, Thermodynamics of Chaotic Systems, volume 4 of Cambridge nonlinear science series. (Cambridge University Press, Cambridge, 1993). (Cited on pages 13; 20; 22; 23; 65; 115; 142; 194; 241 and 242.) [Bec95] C. Beck, From the Perron-Frobenius equation to the Fokker-Planck equation. J. Stat. Phys. 79, 875–894 (1995). (Cited on pages 90 and 172.) [Bec96] C. Beck, Dynamical systems of Langevin type. Physica A 233, 419–440 (1996). (Cited on pages 90 and 172.) [Bel05] G. Bel and E. Barkai, Weak ergodicity breaking in the continuous-time random walk. Phys. Rev. Lett. 94, 240602/1–4 (2005). (Cited on page 108.) [Bel06] G. Bel and E. Barkai, Weak ergodicity breaking with deterministic dynamics. Europhys. Lett. 74, 15–21 (2006). (Cited on page 108.) [Ben80a] G. Benettin, L. Galgani, A. Giorgili and J.-M. Strelcyn, Lyapunov characteristic exponents for smooth dynamical systems: A method for computing all of them. Part 1: Theory. Meccanica 15, 9–19 (1980). (Cited on page 333.) [Ben80b] G. Benettin, L. Galgani, A. Giorgili and J.-M. Strelcyn, Lyapunov characteristic exponents for smooth dynamical systems: A method for computing all of them. Part 2: Numerical application. Meccanica 15, 21–29 (1980). (Cited on pages 192 and 333.) [Ben99] S. Benkadda, S. Kassibrakis, R.B. White and G.M. Zaslavsky, Reply to “Comment on ‘Self-similarity and transport in the standard map’ ”. Phys. Rev. E 59, 3761–3764 (1999). (Cited on page 361.) [Ben01] G. Benettin and L. Rondoni, A new model for the transport of particles in a thermostatted system. Mathematical Physics Electronic Journal 7, 1–22 (2001). (Cited on pages 265 and 290.) [Ber80] M.V. Berry and Z.V. Lewis, On the Weierstraß-Mandelbrot fractal function. Proc. R. Soc. Lond. A 370, 459–484 (1980). (Cited on page 44.) [Ber88] B.J. Berne, M. Borkovec and J.E. Straub, Classical and modern methods in reaction rate theory. J. Phys. Chem. 92, 3711–3725 (1988). (Cited on page 59.) [Bey76] H.U. Beyeler, Cationic Short-Range Order in the Hollandite K1.54 M g0.77 T7.23 O16 : Evidence for the Importance of Ion-Ion Interactions in Superionic Conductors. Phys. Rev. Lett. 37, 1557–1560 (1976). (Cited on pages 4 and 107.) [BJ82] E. Ben-Jacob, I. Goldhirsch and Y. Imry, Intermittent chaos in Josephson junctions. Phys.Rev.Lett. 49, 1599–1602 (1982). (Cited on page 107.) [Bla96] J.A. Blackburn and N. Grønbech-Jensen, Phase diffusion in a chaotic pendulum. Phys. Rev. E 53, 3068–3072 (1996). (Cited on page 100.) [Bec87]

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Bibliography

[Boh84]

[Bol64] [Bon97]

[Bon98a]

[Bon98b]

[Bon99]

[Bon00a]

[Bon00b]

[Bon01]

[Bon02]

[Bon06a]

[Bon06b]

[Bou90]

[Bow75]

klages˙book

385

T. Bohr, P. Bak and M.H. Jensen, Transition to chaos by interaction of resonances in dissipative systems. II. Josephson junctions, chargedensity waves, and standard maps. Phys. Rev. A 30, 1970–1981 (1984). (Cited on pages 53; 100 and 107.) L. Boltzmann, Lectures on Gas Theory. (Univ. of California Press, Berkeley, 1964). (Cited on page 1.) F. Bonetto, G. Gallavotti and P.L. Garrido, Chaotic principle: An experimental test. Physica D 105, 226–252 (1997). (Cited on pages 186; 198; 287; 323 and 324.) F. Bonetto, N.I. Chernov and J.L. Lebowitz, (Global and local) Fluctuations of phase space contraction in deterministic stationary nonequilibrium. Chaos 8, 823–833 (1998). (Cited on pages 291 and 324.) F. Bonetto, E.G.D. Cohen and C. Pugh, On the validity of the conjugate pairing rule for Lyapunov exponents. J. Stat. Phys. 92, 587–627 (1998). (Cited on page 192.) S.D. Bond, B.J. Leimkuhler and B.B. Laird, The Nos´e-Poincar´e method for constant temperature molecular dynamics. J. Comput. Phys. 151, 114–134 (1999). (Cited on pages 223 and 225.) F. Bonetto, D. Daems and J.L. Lebowitz, Properties of stationary nonequilibrium states in the thermostatted periodic Lorentz gas. I. The one particle system. J. Stat. Phys. 101, 35–60 (2000). (Cited on pages 186; 196; 198; 199; 200 and 201.) F. Bonetto, J.L. Lebowitz and L. Rey-Bellet, Fourier law: A challenge to theorists. in: A. Fokas, A. Grigoryan, T. Kibble and B. Zegarlinski, Eds., Mathematical physics 2000, pages 128–150, Imperial College Press, London, 2000. (Cited on pages 6; 319; 337; 338 and 341.) F. Bonetto and J.L. Lebowitz, Thermodynamic entropy production fluctuation in a two-dimensional shear flow model. Phys. Rev. E 64, 056129/1–9 (2001). (Cited on pages 291 and 324.) F. Bonetto, D. Daems, J.L. Lebowitz and V. Ricci, Properties of stationary nonequilibrium states in the thermostatted periodic Lorentz gas: The multiparticle system. Phys. Rev. E 65, 051204/1–9 (2002). (Cited on pages 186; 198; 252 and 287.) F. Bonetto, G. Gallavotti, A. Giuliani and F. Zamponi, Chaotic hypothesis, fluctuation theorem and singularities. J. Stat. Phys. 123, 39–54 (2006). (Cited on page 322.) F. Bonetto, G. Gallavotti, A. Giuliani and F. Zamponi, Chaotic hypothesis, fluctuation theorem and singularities. J.Stat. Phys. 123, 39–54 (2006). (Cited on pages 326 and 328.) J.-P. Bouchaud and A. Georges, Anomalous diffusion in disordered media: Statistical mechanisms, models and physical applications. Phys. Rep. 195, 127–293 (1990). (Cited on pages 83; 85; 86; 90; 92; 108 and 111.) R. Bowen, Equilibrium states and the ergodic theory of Anosov diffeomorphisms, volume 470 of Lecture notes in mathematics. (SpringerVerlag, Berlin, 1975). (Cited on page 1.)

May 8, 2007

4:1

386

[Bra00a]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

A.C. Branka, Nos´e-Hoover chain method for nonequilibrium molecular dynamics simulation. Phys. Rev. E 61, 4769–4773 (2000). (Cited on pages 223; 224; 225; 226 and 266.) [Bra00b] A.C. Branka and K.W. Wojciechowski, Generalization of Nos´e and Nos´e-Hoover isothermal dynamics. Phys. Rev. E 62, 3281–3292 (2000). (Cited on pages 223; 225; 226 and 266.) [Bre96] W. Breymann, T. T´el and J. Vollmer, Entropy production for open dynamical systems. Phys. Rev. Lett. 77, 2945–2948 (1996). (Cited on pages 3; 63; 237; 238; 241; 242 and 245.) [Bre98] W. Breymann, T. T´el and J. Vollmer, Entropy balance, time reversibility, and mass transport in dynamical systems. Chaos 8, 396–408 (1998). (Cited on pages 3; 9; 63; 68; 190; 233; 238 and 297.) [Bri96] N.V. Brillantov, F. Spahn, J.-M. Hertzsch and Th. P¨ oschel, Model for collisions in granular gases. Phys. Rev. E 53, 5382–5392 (1996). (Cited on page 277.) [Bri01] M.E. Briggs, J.V. Sengers, M.K. Francis, P. Gaspard, R.W. Gammon, J.R. Dorfman and R.V. Calabrese, Tracking a colloidal particle for the measurement of dynamic entropies. Physica A 296, 42–59 (2001). (Cited on pages 6; 90; 348; 351 and 352.) [Bri05] J.N. Bright, D.J. Evans and D.J. Searles, New observations regarding deterministic, time-reversible thermostats and Gauss’s principle of least constraint. J. Chem. Phys. 122, 194106/1–8 (2005). (Cited on pages 224 and 259.) [Bro84] S.E. Brown, G. Mozurkewich and G. Gruner, Subharmonic Shapiro steps and Devil’s-staircase behavior in driven charge-density-wave systems. Phys. Rev. Lett. 52, 2277–2280 (1984). (Cited on pages 4 and 107.) [Bru76] St.G. Brush, The kind of motion we call heat, volume 1/2. (North Holland, Amsterdam, 1976). (Cited on pages 14 and 168.) [Bul90a] A. Bulgac and D. Kusnezov, Canonical ensemble averages from pseudomicrocanonical dynamics. Phys. Rev. A 42, 5045–5048 (1990). (Cited on pages 208; 223; 224; 227 and 265.) [Bul90b] A Bulgac and D. Kusnezov, Deterministic and time-reversal invariant description of Brownian motion. Phys. Lett. A 151, 122–128 (1990). (Cited on pages 223; 224; 227 and 265.) [Bun80] L.A. Bunimovich and Ya.G. Sinai, Markov partitions for dispersed billards. Commun. Math. Phys. 78, 247–280 (1980). (Cited on pages 57; 128; 139; 182; 183; 196 and 337.) [Bun81] L.A. Bunimovich and Ya.G. Sinai, Statistical properties of Lorentz gas with periodic configuration of scatterers. Commun. Math. Phys. 78, 479–497 (1981). (Cited on pages 57; 139; 182; 183; 196 and 337.) [Bun91] L.A. Bunimovich, Ya.G. Sinai and N.I. Chernov, Statistical properties of two-dimensional hyperbolic billiards. Russ. Math. Surveys 46, 47– 106 (1991). (Cited on pages 139; 182; 183 and 196.) [Bun96] L.A. Bunimovich and H. Spohn, Viscosity for a periodic two disk fluid: An existence proof. Commun. Math. Phys. 176, 661–680 (1996).

May 8, 2007

4:1

Bibliography

World Scientific Book - 9in x 6in

klages˙book

387

(Cited on pages 182; 289; 337 and 377.) L.A. Bunimovich, Mushrooms and other billiards with divided phase space. Chaos 11, 802–808 (2001). (Cited on page 362.) [Bun03] L.A. Bunimovich and M.A. Khlabystova, One-dimensional Lorentz gas with rotating scatterers: Exact solutions. J. Stat. Phys. 112, 1207–1218 (2003). (Cited on page 347.) [Bun04] L.A. Bunimovich, Deterministic walks in random environments. Physica D 187, 20–29 (2004). (Cited on page 83.) [Bun05a] L.A. Bunimovich and M.F. Demers, Deterministic models of the simplest chemical reactions. J. Stat. Phys. 120, 239–252 (2005). (Cited on page 81.) [Bun05b] L.A. Bunimovich and C.P. Dettmann, Open circular billiards and the Riemann hypothesis. Phys. Rev. Lett. 94, 100201/1–4 (2005). (Cited on page 362.) [Bus05] C. Bustamante, J. Liphardt and F. Ritort, The nonequilibrium thermodynamics of small systems. Phys. Today 58, 43–48 (2005). (Cited on pages 11 and 331.) [Cam05] D.K. Campbell, P. Rosenau and G.M. Zaslavsky, Eds. The FermiPasta-Ulam problem - The first fifty years, volume 15 of Chaos, American Institute of Physics, College Park, 2005. focus issue. (Cited on pages 337 and 340.) [Car85] M. Carrascosa, F. Cusso and F. Agullo-Lopez, Lambert emitters: A simple Monte-Carlo approach to optical diffusers. Eur. J. Phys. 6, 183–187 (1985). (Cited on pages 58 and 194.) [Car04] D.M. Carberry, J.C. Reid, G.M. Wang, E.M. Sevick, D.J. Searles and D.J. Evans, Fluctuations and irreversibility: An experimental demonstration of a second-law-like theorem using a colloidal particle held in an optical trap. Phys. Rev. Lett. 92, 140601/1–4 (2004). (Cited on pages 328 and 329.) [Cas84] G. Casati, F. Ford, J. Vivaldi and W.M. Visscher, One-dimensional classical many-body systems having a normal thermal conductivity. Phys. Rev. Lett. 52, 1861–1864 (1984). (Cited on pages 339 and 340.) [Cas99] G. Casati and T. Prosen, Mixing property of triangular billiards. Phys. Rev. Lett. 83, 4729–4732 (1999). (Cited on pages 352; 353 and 357.) [Cas00] G. Casati and T. Prosen, Triangle map: A model of quantum chaos. Phys. Rev. Lett. 85, 4261–4264 (2000). (Cited on page 359.) [Cec03] F. Cecconi, M. Falcioni, A. Vulpiani and D. del Castillo-Negrete, The origin of diffusion: The case of non chaotic systems. Physica D 180, 129–139 (2003). (Cited on pages 246 and 350.) [Cec05] F. Cecconi, M. Cencini, M. Falcioni and A. Vulpiani, Brownian motion and diffusion: From stochastic processes to chaos and beyond. Chaos 15, 026102/1–9 (2005). (Cited on pages 349; 350 and 351.) [Cen00] M. Cencini, M. Falcioni, E. Olbrich, H. Kantz and A. Vulpiani, Chaos or noise: Difficulties of a distinction. Phys. Rev. E 62, 427–437 (2000). (Cited on pages 90; 349; 351 and 352.) [Bun01]

May 8, 2007

4:1

388

[Cha70]

[Che89]

[Che90]

[Che93a]

[Che93b]

[Che94]

[Che95a] [Che95b]

[Che97]

[Che99] [Che01] [Che02]

[Che03a] [Che03b]

[Chi79] [Cho98] [Cic80]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

S. Chapman and T.G. Cowling, The mathematical theory of nonuniform gases. (Cambridge University Press, Cambridge, 1970). (Cited on page 182.) Q. Chen and J.D. Meiss, Flux, resonances and the Devil’s staircase for the sawtooth map. Nonlinearity 39, 347–356 (1989). (Cited on page 27.) Q. Chen, I. Dana, J.D. Meiss, N.W. Murray and I. C. Percival, Resonances and transport in the sawtooth map. Physica D 46, 217–240 (1990). (Cited on page 27.) N.I. Chernov, G.L. Eyink, J.L. Lebowitz and Ya.G. Sinai, Derivation of Ohm’s law in a deterministic mechanical model. Phys. Rev. Lett. 70, 2209–2212 (1993). (Cited on pages 9; 186; 189; 190; 191; 193; 195; 196; 199; 200; 201; 238; 241; 244; 246; 323; 337 and 369.) N.I. Chernov, G.L. Eyink, J.L. Lebowitz and Ya.G. Sinai, Steady-state electrical conduction in the periodic Lorentz gas. Commun. Math. Phys. 154, 569–601 (1993). (Cited on pages 186; 189; 190; 191; 193; 195; 196; 199; 200; 201; 241; 244; 246; 323; 337 and 369.) N.I. Chernov, Statistical properties of the periodic Lorentz gas. Multidimensional case. J. Stat. Phys. 74, 11–53 (1994). (Cited on pages 57 and 139.) Ch.-Ch. Chen, Diffusion coefficient of piecewise linear maps. Phys. Rev. E 51, 2815–2822 (1995). (Cited on page 4.) N.I. Chernov and J.L. Lebowitz, Stationary shear flow in boundary driven Hamiltonian systems. Phys. Rev. Lett. 75, 2831–2834 (1995). (Cited on pages 8; 9; 10; 190; 249; 265; 271; 290; 293; 295 and 296.) N.I. Chernov and J.L. Lebowitz, Stationary nonequilibrium states in boundary-driven Hamiltonian systems: Shear flow. J. Stat. Phys. 86, 953–990 (1997). (Cited on pages 9; 10; 190; 249; 265; 271; 290; 293; 295 and 296.) N.I. Chernov, Decay of correlations and dispersing billiards. J. Stat. Phys. 94, 513–556 (1999). (Cited on pages 139; 182; 183 and 196.) N.I. Chernov, Sinai billiards under small external forces. Ann. Henri Poincare 2, 197–236 (2001). (Cited on pages 186; 193; 196 and 323.) A.V. Chechkin, V.Yu. Gonchar, J. Klafter, R. Metzler and L.V. Tanatarov, Stationary states of non-linear oscillators driven by L´evy noise. Chem. Phys. 284, 233–251 (2002). (Cited on page 310.) A.V. Chechkin. unpublished, 2003. (Cited on page 117.) A.V. Chechkin, J. Klafter, V.Yu. Gonchar, R. Metzler and L.V. Tanatarov, Bifurcation, bimodality, and finite variance in confied L´evy flights. Phys. Rev. E 67, 010102(R)/1–4 (2003). (Cited on page 310.) B.V. Chirikov, A universal instability of many-dimensional oscillator systems. Phys. Rep. 52, 263–379 (1979). (Cited on page 100.) Ph. Choquard, Variational principles for thermostatted systems. Chaos 8, 350–357 (1998). (Cited on pages 9; 216 and 218.) G. Ciccotti and A. Tenenbaum, Canonical ensemble and nonequilibrium states by molecular dynamics. J. Stat. Phys. 23, 767–772 (1980).

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Bibliography

[Cil98]

[Cil04]

[Cir82]

[Cla93] [Cla00]

[Cla01]

[Cla02]

[Cla04]

[Coh92]

[Coh95]

[Coh97] [Coh98]

[Coh99]

[Coh02]

[Col05]

klages˙book

389

(Cited on pages 8; 249; 271 and 291.) S. Ciliberto and C. Laroche, An experimental test of the GallavottiCohen fluctuation theorem. J. Phys. IV France 8, 215–219 (1998). (Cited on pages 11 and 324.) S. Ciliberto, N. Garnier, S. Hernandez, C. Lacpatia, J.F. Pinton and G.R. Chavarria, Experimental test of the Gallavotti-Cohen fluctuation theorem in turbulent flows. Physica A 340, 240–250 (2004). (Cited on page 324.) M. Cirillo and N.F. Pedersen, On bifurcations and transition to chaos in a Josephson junction. Phys. Lett. 90A, 150–152 (1982). (Cited on pages 4; 100 and 107.) I. Claes and C. Van den Broeck, Dispersion of particles in periodic media. J. Stat. Phys. 70, 1215–1231 (1993). (Cited on page 4.) I. Claus and P. Gaspard, Microscopic chaos and reaction-diffusion processes in the periodic Lorentz gas. J. Stat. Phys. 101, 161–186 (2000). (Cited on page 81.) I. Claus and P. Gaspard, Fractals and dynamical chaos in a twodimensional Lorentz gas with sinks. Phys. Rev. E 63, 036227/1–10 (2001). (Cited on page 81.) I. Claus and P. Gaspard, The fractality of the relaxation modes in reaction-diffusion systems. Physica D 168–169, 266–291 (2002). (Cited on pages 2; 81; 243 and 244.) I. Claus, P. Gaspard and H. van Beijeren, Fractals and dynamical chaos in a random 2D Lorentz gas with sinks. Physica D 187, 146–164 (2004). (Cited on page 81.) E.G.D. Cohen et al., Round table discussion (II): Irreversibility and Lyapunov spectra. in: Microscopic simulations of complex hydrodynamic phenomena, pages 327–343, 1992. see Ref. [Mar92a]. (Cited on pages 232; 245; 249 and 250.) E.G.D. Cohen, Transport coefficients and Lyapunov exponents. Physica A 213, 293–314 (1995). (Cited on pages 191; 192; 193; 288; 289 and 297.) E.G.D. Cohen, Dynamical ensembles in statistical mechanics. Physica A 240, 43–53 (1997). (Cited on page 289.) E.G.D. Cohen and L. Rondoni, Note on phase space contraction and entropy production in thermostatted Hamiltonian systems. Chaos 8, 357–365 (1998). (Cited on pages 3; 186; 190; 201; 259; 264 and 345.) E.G.D. Cohen and G. Gallavotti, Note on two theorems in nonequilibrium statistical mechanics. J. Stat. Phys. 96, 1343–1349 (1999). (Cited on pages 327 and 328.) E.G.D. Cohen and L. Rondoni, Particles, maps and irreversible thermodynamics. Physica A 306, 117–128 (2002). (Cited on pages 3 and 345.) D. Collin, F. Ritort, C. Jarzynski, S.B. Smith, I. Tinoco and C. Bustamante, Verification of the Crooks fluctuation theorem and recovery of RNA folding free energies. Nature 437, 231–234 (2005). (Cited on

May 8, 2007

4:1

390

World Scientific Book - 9in x 6in

klages˙book

Bibliography

page 331.) I.P. Cornfeld, S.V. Fomin and Ya.G. Sinai, Ergodic Theory. (Springer, New York, 1982). (Cited on pages 223; 353; 357 and 363.) [Cri06] G. Cristadoro, Fractal diffusion coefficient from dynamical zeta functions. J. Phys. A: Math. Gen. 39, L151–L157 (2006). (Cited on pages 4 and 37.) [Cro99] G.E. Crooks, Entropy production fluctuation theorem and the nonequilibrium work relation for free energy differences. Phys. Rev. E 60, 2721–2726 (1999). (Cited on pages 11 and 330.) [Cro00] G.E. Crooks, Path-ensemble averages in systems driven far from equilibrium. Phys. Rev. E 61, 2361–2365 (2000). (Cited on pages 11 and 330.) [Cvi92] P. Cvitanovi´c, P. Gaspard and Th. Schreiber, Investigation of the Lorentz gas in terms of periodic orbits. Chaos 2, 85–90 (1992). (Cited on pages 132 and 245.) [Cvi95] P. Cvitanovi´c, J.-P. Eckmann and P. Gaspard, Transport properties of the Lorentz gas in terms of periodic orbits. Chaos, Solitons and Fractals 6, 113–120 (1995). (Cited on pages 132 and 245.) [Cvi00] P. Cvitanovi´c, Chaotic field theory: A sketch. Physica A 288, 61–80 (2000). (Cited on pages 5 and 90.) [Cvi07] P. Cvitanovi´c, R. Artuso, R. Mainieri, G. Tanner and G. Vattay, Chaos: Classical and quantum. (Niels Bohr Institute, Copenhagen, 2007). webbook under chaosbook.org. (Cited on pages 3; 4; 5; 13; 14; 37; 108; 245 and 368.) [Dae99] D. Daems and G. Nicolis, Entropy production and phase space volume contraction. Phys. Rev. E 59, 4000–4006 (1999). (Cited on pages 238; 250; 266 and 296.) [Dan89a] I. Dana, Hamiltonian transport on unstable periodic orbits. Physica D 39, 205–235 (1989). (Cited on page 377.) [Dan89b] I. Dana, N.W. Murray and I.C. Percival, Resonances and diffusion in periodic Hamiltonian maps. Phys. Rev. Lett. 65, 1693–1697 (1989). (Cited on page 377.) [Dav79] P.J. Davis, Circulant matrices. (Wiley Interscience Publ., New York, 1979). (Cited on page 25.) [Del95a] C. Dellago, L. Glatz and H.A. Posch, Lyapunov spectrum of the driven Lorentz gas. Phys. Rev. E 52, 4817–4826 (1995). (Cited on pages 186; 193; 195; 198; 199; 200; 201 and 287.) [Del95b] C. Dellago, L. Glatz and H.A. Posch, Lyapunov spectrum of the driven Lorentz gas. Phys. Rev. E 52, 4817–4826 (1995). (Cited on pages 190 and 191.) [Del95c] C. Dellago and H. A. Posch, Lyapunov exponents of systems with elastic hard collisions. Phys. Rev. E 52, 2401–2406 (1995). (Cited on page 84.) [Del96a] C. Dellago and H.A. Posch, Lyapunov instability, local curvature, and the fluid-solid phase transition in two-dimensional particle systems. Physica A 230, 364–387 (1996). (Cited on page 182.) [Del96b] C. Dellago, H.A. Posch and W.G. Hoover, Lyapunov instability in a [Cor82]

May 8, 2007

4:1

Bibliography

World Scientific Book - 9in x 6in

klages˙book

391

system of hard disks in equilibrium and nonequilibrium steady states. Phys. Rev. E 53, 1485–1501 (1996). (Cited on pages 192; 193; 319; 333 and 335.) [Del97a] C. Dellago and H.A. Posch, Lyapunov instability of the boundary-driven Chernov-Lebowitz model for stationary shear flow. J. Stat. Phys. 88, 825–842 (1997). (Cited on pages 192; 193 and 298.) [Del97b] C. Dellago and H.A. Posch, Lyapunov spectrum and the conjugate pairing rule for a thermostatted random Lorentz gas: Numerical simulations. Phys. Rev. Lett. 78, 211–214 (1997). (Cited on pages 10; 183; 192; 252 and 290.) [Del03a] J. Delhommelle, J. Petravic and D.J. Evans, On the effects of assuming flow profiles in nonequilibrium simulations. J. Chem. Phys. 119, 11005–11010 (2003). (Cited on page 289.) [Del03b] J. Delhommelle, J. Petravic and D.J. Evans, Reexamination of string phase and shear thickening in simple fluids. Phys. Rev. E 68, 031201/1–6 (2003). (Cited on page 289.) [Del04] J. Delhommelle, J. Petravic and D.J. Evans, Non-Newtonian behavior in simple fluids. J. Chem. Phys. 120, 6117–6123 (2004). (Cited on page 289.) [Den03] S. Denisov, J. Klafter and M. Urbakh, Dynamical heat channels. Phys. Rev. Lett. 91, 194301/1–4 (2003). (Cited on page 363.) [Der83] B. Derrida, Velocity and diffusion constant of a periodic onedimensional hopping model. J. Stat. Phys. 31, 433–449 (1983). (Cited on page 86.) [Det95] C.P. Dettmann, G.P. Morriss and L. Rondoni, Conjugate pairing in the three-dimensional periodic Lorentz gas. Phys. Rev. E 52, R5746–R5748 (1995). (Cited on pages 186; 190; 192; 193 and 287.) [Det96a] C.P. Dettmann and G.P. Morriss, Crisis in the periodic Lorentz gas. Phys. Rev. E 54, 4782–4790 (1996). (Cited on pages 186; 190; 193; 195 and 196.) [Det96b] C.P. Dettmann and G.P. Morriss, Hamiltonian formulation of the Gaussian isokinetic thermostat. Phys. Rev. E 54, 2495–2500 (1996). (Cited on pages 9 and 217.) [Det96c] C.P. Dettmann and G.P. Morriss, Proof of Lyapunov exponent pairing for systems at constant kinetic energy. Phys. Rev. E 53, R5545–R5548 (1996). (Cited on page 192.) [Det97a] C.P. Dettmann and P. Cvitanovic, Cycle expansions for intermittent diffusion. Phys. Rev. E 56, 6687–6692 (1997). (Cited on pages 5 and 108.) [Det97b] C.P. Dettmann and G.P. Morriss, Hamiltonian reformulation and pairing of Lyapunov exponents for Nos´e-Hoover dynamics. Phys. Rev. E 55, 3693–3696 (1997). (Cited on pages 9; 192; 216 and 218.) [Det97c] C.P. Dettmann and G.P. Morriss, Stability ordering of cycle expansions. Phys. Rev. Lett. 78, 4201–4204 (1997). (Cited on pages 186; 198; 199 and 201.)

May 8, 2007

4:1

392

World Scientific Book - 9in x 6in

klages˙book

Bibliography

[Det97d] C.P. Dettmann, G.P. Morriss and L. Rondoni, Irreversibility, diffusion and multifractal measures in thermostatted systems. Chaos, Solitons and Fractals 8, 783–792 (1997). (Cited on pages 186 and 193.) [Det98] C.P. Dettmann and P. Dahlqvist, Computing the diffusion coefficient for intermittent maps: Resummation of stability ordered cycle expansions. Phys. Rev. E 57, 5303–5310 (1998). (Cited on pages 5 and 108.) [Det99a] C.P. Dettmann, Hamiltonian for a restricted isoenergetic thermostat. Phys. Rev. E 60, 7576–7577 (1999). (Cited on page 217.) [Det99b] C.P. Dettmann and E.G.D. Cohen, Microscopic chaos from Brownian motion? Nature 401, 875–875 (1999). (Cited on pages 6; 90; 246; 320; 348; 349 and 350.) [Det00a] C.P. Dettmann, The Lorentz gas: A paradigm for nonequilibrium steady states. in: Hard-ball systems and the Lorentz gas, pages 315– 366, 2000. see Ref. [Sza00]. (Cited on pages 7; 8; 13; 186; 187; 188; 190; 195; 198; 199; 208; 210; 218; 236; 237 and 245.) [Det00b] C.P. Dettmann and E.G.D. Cohen, Microscopic chaos and diffusion. J. Stat. Phys. 101, 775–817 (2000). (Cited on pages 6; 90; 246 and 350.) [Dev89] R.L. Devaney, An introduction to chaotic dynamical systems. (Addison-Wesley, Reading, second edition, 1989). (Cited on pages 21; 100; 194; 223 and 363.) [dG62] S.R. de Groot and P. Mazur, Non-equilibrium thermodynamics. (North Holland, Amsterdam, 1962). reprinted by Dover, New York, 1984. (Cited on pages 166 and 190.) [Dha99] A. Dhar and D. Dhar, Absence of local thermodynamik equilibrium in two models of heat conduction. Phys. Rev. Lett. 82, 480–483 (1999). (Cited on page 346.) [Dha01] A. Dhar, Heat conduction in the disordered harmonic chain revisited. Phys. Rev. Lett. 86, 5882–5885 (2001). (Cited on page 343.) [Dia06] L. Diambra, L.C. Cintra, Q. Chen, D. Schubert and L. da F. Costa, Cell adhesion protein decreases cell motion: Statistical characterization of locomotion activity. Physica A 365, 481–490 (2006). (Cited on pages 304 and 305.) [Die05] P. Dieterich, R. Preuss, R. Klages, V. Dreval, C. Stock and A. Schwab, Brownian motion of migrating cells? in: E.Sch¨ oll and K.L¨ udge, Eds., XXV Dynamics Days Europe 2005, volume 29E of Europhysics Conference Series, page 155, TU Berlin, Berlin, 2005. (Cited on pages 304 and 305.) [Die07] P. Dieterich, R. Klages, R. Preuss and A. Schwab, Anomalous dynamics of migrating biological cells. preprint, 2007. (Cited on pages 304 and 305.) [Din87] E.J. Ding and P.C. Hemmer, Exact treatment of mode locking for a piecewise linear map. J. Stat. Phys. 46, 99–110 (1987). (Cited on page 44.) [dM93] W. de Melo and S. van Strien, One-dimensional dynamics, volume 25 of Ergebnisse der Mathematik und ihrer Grenzgebiete, 3.Folge. (Springer, Berlin, 1993). (Cited on pages 23 and 100.)

May 8, 2007

4:1

Bibliography

World Scientific Book - 9in x 6in

klages˙book

393

[Dob76a] R. Dobbertin, On functional relations between reduced distribution functions and entropy production by non-Hamiltonian perturbations. Physica Scripta 14, 85–88 (1976). (Cited on page 237.) [Dob76b] R. Dobbertin, Vlasov equation and entropy. Physica Scripta 14, 89–91 (1976). (Cited on page 237.) [Dol00] M. Dolowschiak and Z. Kovacs, Breaking conjugate pairing in thermostated billiards by a magnetic field. Phys. Rev. E 62, 7894–7897 (2000). (Cited on pages 68; 192 and 193.) [Dol02] M. Dolowschiak and Z. Kovacs, Fluctuation formula for nonreversible dynamics in the thermostated Lorents gas. Phys. Rev. E 66, 066217– 1/4 (2002). (Cited on page 324.) [Dol03] M. Dolowschiak and Z. Kovacs, Fluctuation formula and nonGaussian distribution in the thermostated Lorentz gas. preprint arXiv:nlin.CD/0205059, 2003. (Cited on pages 323; 324 and 325.) [Dol05] M. Dolowschiak and Z. Kovacs, Fluctuation formula in the Nos´eHoover thermostated Lorentz gas. Phys. Rev. E 71, 025202(R)/1–4 (2005). (Cited on pages 325 and 326.) [Dol06] M. Dolowschiak. private communication, 2006. (Cited on page 325.) [Dor77] J.R. Dorfman and H. van Beijeren, The kinetic theory of gases. in: B.J. Berne, Eds., Statistical mechanics, volume B, chapter 3, pages 65–179. (Plenum Press, New York, 1977). (Cited on page 32.) [D¨ or85] M. D¨ orfle, Spectrum and eigenfunctions of the Frobenius-Perron operator of the tent map. J. Stat. Phys. 40, 93–132 (1985). (Cited on pages 49 and 65.) [Dor95] J.R. Dorfman and P. Gaspard, Chaotic scattering theory of transport and reaction-rate coefficients. Phys. Rev. E 51, 28–35 (1995). (Cited on pages 2; 7; 22; 241; 246 and 370.) [Dor99] J.R. Dorfman, An introduction to chaos in nonequilibrium statistical mechanics. (Cambridge University Press, Cambridge, 1999). (Cited on pages 1; 2; 3; 4; 7; 13; 22; 23; 59; 63; 64; 67; 84; 89; 97; 100; 104; 116; 128; 182; 183; 186; 200; 201; 205; 206; 208; 223; 233; 234; 236; 237; 238; 241; 242; 249; 272; 323; 332; 349 and 368.) [Dor02] J.R. Dorfman, P. Gaspard and T. Gilbert, Entropy production of diffusion in spatially periodic deterministic systems. Phys. Rev. E 66, 026110/1–9 (2002). (Cited on pages 3; 63; 238 and 346.) [Dou05] F. Douarche, S. Ciliberto, A. Petrosyan and I. Rabbiosi, An experimental test of the Jarzynski equality in a mechanical experiment. Europhys. Lett. 70, 593–599 (2005). (Cited on pages 329 and 331.) [dR57] G. de Rham, Sur quelques courbes definies par des equations fonctionnelles. Rend. Sem. Mat. Torino 16, 101–113 (1957). (Cited on page 62.) [Dr¨ a00] J. Dr¨ ager and J. Klafter, Strong anomaly in diffusion generated by iterated maps. Phys. Rev. Lett. 84, 5998–6001 (2000). (Cited on page 108.) [Dre88] U. Dressler, Symmetry property of the Lyapunov spectra of a class of

May 8, 2007

4:1

394

World Scientific Book - 9in x 6in

klages˙book

Bibliography

dissipative dynamical systems with viscous damping. Phys. Rev. A 38, 2103–2109 (1988). (Cited on pages 10; 191 and 192.) [Dru00] P. Drude, Zur Elektronentheorie der Metalle. Ann. Phys. 1, 566–612 (1900). (Cited on page 180.) [Dun87] G.A. Dunn and A.F. Brown, A unified approach to analysing cell motility. J. Cell Sci. Suppl. 8, 81–102 (1987). (Cited on page 304.) [D¨ ur98] D. D¨ urr and H. Spohn, Brownian motion and microscopic chaos. Nature 394, 831–833 (1998). (Cited on page 350.) [dW04a] A.S. de Wijn and H. van Beijeren, Goldstone modes in Lyapunov spectra of hard sphere systems. Phys. Rev. E 70, 016207/1–11 (2004). (Cited on pages 10 and 336.) [dW04b] A.S. de Wijn and H. van Beijeren, Lyapunov spectrum of the manydimensional dilute random Lorentz gas. Phys. Rev. E 70, 036209/1–7 (2004). (Cited on pages 84; 183 and 252.) [dW05] A.S. de Wijn, Lyapunov spectra of billiards with cylindrical scatterers: Comparison with many-particle systems. Phys. Rev. E 72, 026216/1– 10 (2005). (Cited on pages 183 and 252.) [Ebe99] W. Ebeling, F. Schweitzer and B. Tilch, Active Brownian particles with energy depots modeling animal mobility. BioSystems 49, 17–29 (1999). (Cited on pages 305 and 306.) [Ebe00] W. Ebeling, Canonical nonequililbrium statistics and applications to Fermi-Bose systems. Cond. Matt. Phys. 3, 285–293 (2000). (Cited on page 308.) [Ebe01] W. Ebeling and F. Schweitzer, Swarms of particle agents with harmonic interactions. Theory in Biosciences 120, 207–224 (2001). (Cited on pages 309 and 314.) [Ebe04] W. Ebeling and G. R¨ opke, Statistical mechanics of confined systems with rotational excitations. Physica D 187, 268–280 (2004). (Cited on pages 305 and 309.) [Ebe05] W. Ebeling and I.M. Sokolov, Statistical thermodynamics and stochastic theory of nonequilibrium systems. (World Scientific, Singapore, 2005). (Cited on pages 108; 111; 168 and 305.) [Eck85] J.-P. Eckmann and D. Ruelle, Ergodic theory of chaos and strange attractors. Rev. Mod. Phys. 57, 617–656 (1985). (Cited on pages 13; 100; 190; 192; 193; 194; 241; 242; 244; 249; 272; 332; 348; 349 and 369.) [Eck99a] J.-P. Eckmann, C.-A. Pillet and L. Rey-Bellet, Entropy production in non-linear, thermally driven Hamilatonian systems. J. Stat. Phys. 95, 305–331 (1999). (Cited on pages 8; 172; 237; 266; 339 and 340.) [Eck99b] J.-P. Eckmann, C.-A. Pillet and L. Rey-Bellet, Non-equilibrium statistical mechanics of anharmonic chains coupled to two heat baths at different temperatures. Commun. Math. Phys. 201, 657–697 (1999). (Cited on pages 8; 172; 339 and 340.) [Eck00] J.-P. Eckmann and O. Gat, Hydrodynamic Lyapunov modes in translation-invariant systems. J. Stat. Phys. 98, 775–797 (2000). (Cited on pages 10 and 336.) [Eck02] B. Eckhardt, E. Hascoet and W. Braun, Passive fields and particles in

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Bibliography

[Eck04]

[Eck05]

[Eck06a]

[Eck06b]

[Eck06c]

[Ehr59]

[Ein05]

[Els85]

[Erd00]

[Erd02]

[Erd04]

[Ers93] [Eva83a]

[Eva83b]

klages˙book

395

chaotic flows. in: N.S. Namachchivaya and Y.K. Lin, Eds., IUTAM Symposium on Nonlinear Stochastic Dynamics, volume 2, pages 990– 999, Urbana Champaign, 2002. (Cited on pages 90 and 93.) J.P. Eckmann and L.-S. Young, Temperature profiles in Hamiltonian heat conduction. Europhys. Lett. 68, 790–796 (2004). (Cited on pages 339 and 347.) J.P. Eckmann, C. Forster, H.A. Posch and E. Zabey, Lyapunov modes in hard-disk systems. J. Stat. Phys. 118, 813–847 (2005). (Cited on page 336.) J.P. Eckmann and C. Meija-Monasterio, Thermal rectification in billiardlike systems. Phys. Rev. Lett. 97, 094301/1–4 (2006). (Cited on page 347.) J.P. Eckmann, C. Meija-Monasterio and E. Zabey, Memory effects in nonequilibrium transport for deterministic Hamiltonian systems. J. Stat. Phys. 123, 1339–1360 (2006). (Cited on page 347.) J.P. Eckmann and L.-S. Young, Nonequilibrium energy profiles for a class of 1-D models. Commun. Math. Phys. 262, 237–267 (2006). (Cited on pages 339 and 347.) P. Ehrenfest and T. Ehrenfest, The conceptual foundations of the statistical approach in mechanics. (Cornell Univ. Press, Ithaca, 1959). reprinted by Dover, New York, 2002. (Cited on pages 1 and 350.) ¨ A. Einstein, Uber die von der molekularkinetischen Theorie der W¨ arme geforderten Bewegung von in ruhenden Fl¨ ussigkeiten suspendierten Teilchen. Annalen der Physik 17, 549–560 (1905). (Cited on page 19.) Y. Elskens and R. Kapral, Reversible dynamics and the macroscopic rate law for a solvable Kolmogorov system: The three baker’s reaction. J. Stat. Phys. 38, 1027–1049 (1985). (Cited on page 56.) U. Erdmann, W. Ebeling, L. Schimansky-Geier and F. Schweitzer, Brownian particles far from equilibrium. Eur. Phys. J. B 15, 105–113 (2000). (Cited on pages 305; 306; 307; 308; 309 and 312.) U. Erdmann, W. Ebeling and V.S. Anishchenko, Excitation of rotational modes in two-dimensional systems of driven Brownian particles. Phys. Rev. E 65, 061106–1/9 (2002). (Cited on pages 305; 309 and 312.) U. Erdmann, W. Ebeling, L. Schimansky-Geier, A. Ordemann and F. Moss, Active Brownian particle and random walk theories of the motions of zooplankton: Application to experiments with swarms of daphnia. preprint arXiv:q-bio.PE/0404018, 2004. (Cited on pages 305 and 314.) S.V. Ershov, Is a perturbation theory for dynamical chaos possible? Phys. Lett. A 177, 180–185 (1993). (Cited on pages 49 and 65.) D.J. Evans, Computer “experiment” for nonlinear thermodynamics of Couette flow. J. Chem. Phys. 78, 3297–3302 (1983). (Cited on pages 8; 187 and 289.) D.J. Evans, W.G. Hoover, B.H. Failor, B. Moran and A.J.C. Ladd, Nonequilibrium molecular dynamcis via Gauss’s principle of least con-

May 8, 2007

4:1

396

[Eva85a] [Eva85b]

[Eva86] [Eva90a]

[Eva90b]

[Eva93]

[Eva94]

[Eva98]

[Eva00]

[Eva02a]

[Eva02b]

[Eva05]

[Eyi92]

[Fal90]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

straint. Phys. Rev. A 28, 1016–1021 (1983). (Cited on pages 8; 188 and 223.) D.J. Evans, Response theory as a free-energy extremum. Phys. Rev. A 32, 2923–2925 (1985). (Cited on pages 236 and 237.) D.J. Evans and B.L. Holian, The Nos´e-Hoover thermostat. J. Chem. Phys. 83, 4069–4074 (1985). (Cited on pages 210; 211; 212; 217; 221 and 223.) D.J. Evans and G.P. Morriss, Shear thickening and turbulence in simple fluids. Phys. Rev. Lett. 56, 2172–2175 (1986). (Cited on page 289.) D.J. Evans, E.G.D. Cohen and G.P. Morriss, Viscosity of a simple fluid from its maximal Lyapunov exponents. Phys. Rev. A 42, 5990–5997 (1990). (Cited on pages 9; 10; 191; 192; 193; 289; 290; 297; 298 and 370.) D.J. Evans and G.P. Morriss, Statistical mechanics of nonequilibrium liquids. (Academic Press, London, 1990). (Cited on pages 1; 7; 8; 9; 13; 166; 167; 188; 189; 190; 191; 193; 201; 205; 206; 209; 210; 217; 236; 237; 249; 282; 288; 289; 291; 368 and 369.) D.J. Evans, E.G.D. Cohen and G.P. Morriss, Probability of second law violations in shearing steady flows. Phys. Rev. Lett. 71, 2401–2404 (1993). (Cited on pages 10; 319; 320; 321 and 326.) D.J. Evans and D.J. Searles, Equilibrium microstates which generate second law violating steady states. Phys. Rev. E 50, 1645–1648 (1994). (Cited on pages 10 and 327.) D.J. Evans, D.J. Searles, W.G. Hoover, B.L. Holian, H.A. Posch and G.P. Morriss, Comment on “Modified nonequilibrium molecular dynamics for fluid flows with energy conservation”. J. Chem. Phys. 108, 4351–4352 (1998). (Cited on pages 207; 224 and 226.) D.J. Evans, E.G.D. Cohen, D.J. Searles and F. Bonetto, Note on the Kaplan-Yorke dimension and linear transport coefficients. J. Stat. Phys. 101, 17–34 (2000). (Cited on pages 193; 244 and 290.) D.J. Evans and L. Rondoni, Comments on the entropy of nonequilibrium steady states. J. Stat. Phys. 109, 895–920 (2002). (Cited on pages 235 and 236.) D.J. Evans and D.J. Searles, The fluctuation theorem. Advances in Physics 51, 1529–1585 (2002). (Cited on pages 10; 11; 100; 321; 327; 328 and 331.) D.J. Evans, D.J. Searles and L. Rondoni, Application of the GallavottiCohen fluctuation relation to thermostated steady states near equilibrium. Phys. Rev. E 71, 056120/1–12 (2005). (Cited on pages 210; 324; 326; 327; 328 and 371.) G.L. Eyink and J.L. Lebowitz, Generalized Gaussian dynamics, phasespace reduction and irreversibility: A comment. in: Microscopic simulations of complex hydrodynamic phenomena, pages 323–326, 1992. see Ref. [Mar92a]. (Cited on pages 232; 249 and 250.) K. Falconer, Fractal geometry. (Wiley, New York, 1990). (Cited on

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Bibliography

[Fal95] [Far85] [Far99]

[Far02]

[Fau95]

[Fei78] [Fei79] [Fei83] [Fes02] [Fic05]

[Fle92]

[Fli96]

[For87] [For04]

[For05] [Fra90]

[Fra91]

[Fuj82]

klages˙book

397

page 25.) M. Falcioni and A. Vulpiani, The relevance of chaos for the linear response theory. Physica A 215, 481–494 (1995). (Cited on page 201.) J.D. Farmer, Sensitive dependence on parameters in nonlinear dynamics. Phys. Rev. Lett. 55, 351–354 (1985). (Cited on page 106.) Z. Farkas, P. Tegzes, A. Vukics and T. Vicsek, Transitions in the horizontal transport of vertically vibrated granular layers. Phys. Rev. E 60, 7022–7031 (1999). (Cited on pages 6; 54; 154 and 378.) Z. Farkas, F. Szalai, D.E. Wolf and T. Vicsek, Segregation of granular binary mixtures by a ratchet mechanism. Phys. Rev. E 65, 022301/1–4 (2002). (Cited on pages 54; 154 and 161.) L.P. Faucheux, L.S Bourdieu, P.D. Kaplan and A.J. Libchaber, Optical thermal ratchet. Phys. Rev. Lett. 74, 1504–1507 (1995). (Cited on page 53.) M.J. Feigenbaum, Quantitative universality for a class of nonlinear transformations. J. Stat. Phys. 19, 25–52 (1978). (Cited on page 105.) M.J. Feigenbaum, The universal metric properties of nonlinear transformations. J. Stat. Phys. 21, 669–706 (1979). (Cited on page 105.) M.J. Feigenbaum, Universal behavior in non-linear systems. Physica D 7, 16–39 (1983). (Cited on page 105.) R. Festa, A. Mazzino and D. Vincenzi, Lorenz deterministic diffusion. Europhys. Lett. 60, 820–826 (2002). (Cited on page 107.) A. Fichtner and G. Radons, Disordered iterated maps: Spectral properties, escape rates and anomalous transport. New J. Phys. 7, 30/1–16 (2005). (Cited on pages 5; 83 and 84.) R. Fleischmann, T. Geisel and R. Ketzmerick, Magnetoresistance due to chaos and nonlinear resonance in lateral surface superlattices. Phys. Rev. Lett. 68, 1367–1370 (1992). (Cited on pages 134; 181 and 377.) M. Fließer, G.J.O. Schmidt and H. Spohn, Magnetotransport of the Sinai billard. Phys. Rev. E 53, 5690–5697 (1996). (Cited on pages 134 and 181.) G.W. Ford and M. Kac, On the quantum Langevin equation. J. Stat. Phys. 46, 803–810 (1987). (Cited on pages 7 and 169.) C. Forster, R. Hirschl, H.A. Posch and W.G. Hoover, Perturbed phasespace dynamics of hard-disk fluids. Physica D 187, 294–310 (2004). (Cited on pages 10; 331; 333; 334; 335 and 336.) C. Forster and H.A. Posch, Lyapunov modes in soft-disk fluids. New J. Phys. 7, 32/1–20 (2005). (Cited on page 336.) K. Franke and H. Gruler, Galvanotaxis of human granulocytes: Electric field jump studies. Europ. Biophys. J. 18, 335–346 (1990). (Cited on page 304.) M. Franaszek, Influence of noise on the mean lifetime of chaotic transients. Phys. Rev. A 44, 4065–4067 (1991). (Cited on pages 5; 90 and 91.) H. Fujisaka and S. Grossmann, Diffusion in discrete nonlinear dynamical systems. Z. Physik B 48, 261–275 (1982). (Cited on pages 19; 20;

May 8, 2007

4:1

398

[Ful75]

[Gal77] [Gal95a]

[Gal95b]

[Gal96]

[Gal98]

[Gal99]

[Gal03a]

[Gal03b] [Gal04a]

[Gal04b] [Gan71] [Gar02]

[Gas71] [Gas88]

[Gas90]

[Gas92a]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

21; 25; 26; 64; 83; 86; 100; 102; 105 and 123.) P. Fulde, L. Pietronero, W. R. Schneider and S. Str¨ assler, Problem of Brownian motion in a periodic potential. Phys. Rev. Lett. 35, 1776– 1779 (1975). (Cited on pages 4 and 107.) F. Galton, Typical laws of heredity. Nature 15, 492–495, 512–514, 532–533 (1877). (Cited on pages 3 and 180.) G. Gallavotti and E.G.D. Cohen, Dynamical ensembles in nonequilibrium statistical mechanics. Phys. Rev. Lett. 74, 2694–2697 (1995). (Cited on pages 10; 319; 321; 322; 326 and 369.) G. Gallavotti and E.G.D. Cohen, Dynamical ensembles in stationary states. J. Stat. Phys. 80, 931–970 (1995). (Cited on pages 10; 321; 322 and 369.) G. Gallavotti, Extension of Onsager’s reciprocity to large fields and the chaotic hypothesis. Phys. Rev. Lett. 77, 4334–4337 (1996). (Cited on page 322.) G. Gallavotti, Chaotic dynamics, fluctuations, nonequilibrium ensembles. Chaos 8, 384–393 (1998). (Cited on pages 9; 10; 322; 324 and 368.) G. Gallavotti, Statistical mechanics - a short treatise. (Springer, Berlin, 1999). Chapter 9. (Cited on pages 1; 7; 8; 9; 10; 13; 100; 167; 236; 237; 238; 368 and 369.) S. Galatolo, Complexity, initial condition sensitivity, dimension and weak chaos in dynamical systems. Nonlinearity 16, 1219–1238 (2003). (Cited on page 355.) G. Gallavotti, Nonequilibrium thermodynamics? preprint condmat/0301172, 2003. (Cited on pages 1; 190; 238 and 250.) G. Gallavotti, Entropy production and thermodynamics of nonequilibrium stationary states: A point of view. Chaos 14, 680–690 (2004). (Cited on page 369.) G. Gallavotti and E.G.D. Cohen, Nonequilibrium stationary states and entropy. Phys. Rev. E 69, 035104/1–4 (2004). (Cited on page 369.) F.R. Gantmacher, Matrizenrechnung, volume 2. (VEB Deutscher Verlag der Wissenschaften, Berlin, 3rd edition, 1971). (Cited on page 25.) P. Garbaczewski and R. Olkiewicz, Eds. Dynamics of dissipation, volume 597 of Lecture Notes in Physics, Springer, Berlin, 2002. (Cited on pages 3; 13; 233; 400; 407; 411 and 422.) D. Gass, Enskog theory for a rigid disk fluid. J. Chem. Phys. 54, 1898–1902 (1971). (Cited on pages 292 and 293.) P. Gaspard and X.-J. Wang, Sporadicity: Between periodic and chaotic dynamical behaviors. Proc. Nat. Acad. Sci. USA 85, 4591–4595 (1988). (Cited on pages 5; 108; 109; 116 and 355.) P. Gaspard and G. Nicolis, Transport properties, Lyapunov exponents, and entropy per unit time. Phys. Rev. Lett. 65, 1693–1696 (1990). (Cited on pages 2; 22; 241 and 370.) P. Gaspard, Diffusion, effusion, and chaotic scattering. J. Stat. Phys. 68, 673–747 (1992). (Cited on pages 3; 4; 22; 23; 25; 59; 61; 64; 65;

May 8, 2007

4:1

Bibliography

World Scientific Book - 9in x 6in

klages˙book

399

67 and 233.) [Gas92b] P. Gaspard, Diffusion in uniformly hyperbolic one-dimensional maps and Appell polynomials. Phys. Lett. A 168, 13–17 (1992). (Cited on pages 22; 23 and 25.) [Gas92c] P. Gaspard and F. Baras, Dynamical chaos underlying diffusion in the Lorentz gas. in: Microscopic simulations of complex hydrodynamic phenomena, pages 301–322, 1992. see Ref. [Mar92a]. (Cited on pages 2; 21; 22 and 241.) [Gas93] P. Gaspard, What is the role of chaotic scattering in irreversible processes? Chaos 3, 427–442 (1993). (Cited on pages 2; 22; 23; 25; 59; 60; 61; 62; 233; 241 and 345.) [Gas95a] P. Gaspard, Chaos, fractals, and transport properties: A new insight into nonequilibrium statistical mechanics and irreversibility. in: Y. Aizawa, S. Saito and K. Shiraiwa, Eds., Dynamical systems and chaos, volume 2, pages 55–68, World Scientific, Singapore, 1995. (Cited on pages 59; 60; 61 and 62.) [Gas95b] P. Gaspard and F. Baras, Chaotic scattering and diffusion in the Lorentz gas. Phys. Rev. E 51, 5332–5352 (1995). (Cited on pages 2; 183; 199; 241 and 244.) [Gas95c] P. Gaspard and J.R. Dorfman, Chaotic scattering theory, thermodynamic formalism, and transport coefficients. Phys. Rev. E 52, 3525– 3552 (1995). (Cited on pages 2; 22; 241 and 290.) [Gas96] P. Gaspard, Hydrodynamic modes as singular eigenstates of the Liouvillian dynamics: Deterministic diffusion. Phys. Rev. E 53, 4379–4401 (1996). (Cited on pages 57; 63; 132; 183; 194; 196 and 200.) [Gas97a] P. Gaspard, Chaos and hydrodynamics. Physica A 240, 54–67 (1997). (Cited on page 63.) [Gas97b] P. Gaspard, Entropy production in open volume-preserving systems. J. Stat. Phys. 88, 1215–1240 (1997). (Cited on pages 3; 63; 233; 236; 237; 238 and 266.) [Gas98a] P. Gaspard, Chaos, scattering, and statistical mechanics. (Cambridge University Press, Cambridge, 1998). (Cited on pages 1; 2; 3; 4; 7; 9; 13; 22; 23; 57; 59; 63; 64; 68; 89; 97; 100; 104; 128; 132; 182; 183; 190; 192; 194; 196; 200; 207; 223; 232; 233; 234; 236; 237; 238; 239; 240; 241; 242; 243; 245; 246; 266; 272; 332; 349; 352; 356 and 368.) [Gas98b] P. Gaspard, M.E. Briggs, M.K. Francis, J.V. Sengers, R.W. Gammons, J.R. Dorfman and R.V. Calabrese, Experimental evidence for microscopic chaos. Nature 394, 865–868 (1998). (Cited on pages 6; 90; 320 and 348.) [Gas98c] P. Gaspard and R. Klages, Chaotic and fractal properties of deterministic diffusion-reaction processes. Chaos 8, 409–423 (1998). (Cited on pages 4; 37; 40; 55; 64; 68; 81; 104; 200 and 233.) [Gas99] P. Gaspard, M.E. Briggs, M.K. Francis, J.V. Sengers, R.W. Gammons, J.R. Dorfman and R.V. Calabrese, Microscopic chaos from Brownian motion? – reply. Nature 401, 876 (1999). (Cited on page 351.) [Gas01] P. Gaspard, I. Claus, T. Gilbert and J.R. Dorfman, Fractality of the hy-

May 8, 2007

4:1

400

World Scientific Book - 9in x 6in

klages˙book

Bibliography

drodynamic modes of diffusion. Phys. Rev. Lett. 86, 1506–1509 (2001). (Cited on pages 2; 243 and 244.) [Gas02a] P. Gaspard, Dynamical theory of relaxation in clasical and quantum systems. in: Dynamics of dissipation, pages 111–163, 2002. see Ref. [Gar02]. (Cited on pages 3 and 346.) [Gas02b] P. Gaspard and I. Claus, Chaos and fractals in dynamical models of transport and reaction. Phil. Trans. Roy. Soc. London Series A 360, 303–315 (2002). (Cited on page 81.) [Gas03] P. Gaspard, G. Nicolis and J.R. Dorfman, Diffusive Lorentz gases and multibaker maps are compatible with irreversible thermodynamics. Physica A 323, 294–322 (2003). (Cited on pages 3; 232; 233 and 346.) [Gas04] P. Gaspard, Time-reversed dynamical entropy and irreversibility in Markovian random processes. J. Stat. Phys. 117, 599–615 (2004). (Cited on page 370.) [Gas05] P. Gaspard, Brownian motion, dynamical randomness and irreversibility. New J. Phys. 7, 77/1–19 (2005). (Cited on page 370.) [Gas06] P. Gaspard, Hamiltonian dynamics, nanosystems, and nonequilibrium statistical mechanics. Physica A 369, 201–246 (2006). (Cited on pages 368 and 370.) [Gei82] T. Geisel and J. Nierwetberg, Onset of diffusion and universal scaling in chaotic systems. Phys. Rev. Lett. 48, 7–10 (1982). (Cited on pages 5; 19; 83; 89; 90; 97; 100; 102 and 123.) [Gei84] T. Geisel and S. Thomae, Anomalous diffusion in intermittent chaotic systems. Phys. Rev. Lett. 52, 1936–1939 (1984). (Cited on pages 108; 109; 114 and 115.) [Gei85] T. Geisel, J. Nierwetberg and A. Zacherl, Accelerated diffusion in Josephson junctions and related chaotic systems. Phys. Rev. Lett. 54, 616–619 (1985). (Cited on pages 108 and 378.) [Gei90] T. Geisel, J. Wagenhuber, P. Niebauer and G. Obermair, Chaotic dynamics of ballistic electrons in lateral superlattices and magnetic fields. Phys. Rev. Lett. 64, 1581–1584 (1990). (Cited on pages 134; 181 and 377.) [Ger73] G. Gerlich, Die verallgemeinerte Liouville-Gleichung. Physica 69, 458– 466 (1973). (Cited on pages 207 and 236.) [Ger99] A. Gervois, D. Iagolnitzer, M. Moreau and Y. Pomeau, Eds. Statistical Physics - invited papers from STATPHYS 20, volume 263 of Physica A, North Holland, Amsterdam, 1999. (Cited on page 1.) [Gia00] C. Giardina, R. Livi, A. Politi and M. Vassalli, Finite thermal conductivity in 1d lattices. Phys. Rev. Lett. 84, 2144–2147 (2000). (Cited on page 341.) [Gib60] J.W. Gibbs, Elementary principles of statistical mechanics. (Dover, New York, 1960). (Cited on page 1.) [Gil99a] T. Gilbert and J.R. Dorfman, Entropy production: From open volumepreserving to dissipative systems. J. Stat. Phys. 96, 225–269 (1999). (Cited on pages 3; 63; 190; 193; 233; 238; 239 and 243.) [Gil99b] T. Gilbert, C.D. Ferguson and J.R. Dorfman, Field driven thermostated

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Bibliography

[Gil00a]

[Gil00b]

[Gil01]

[Gil04]

[Gil06]

[Giu05]

[Gla93] [Gle88] [Gol84]

[Gol85]

[Gor00]

[Got04]

[Got05]

[Gra99]

[Gra02]

[Gre54]

klages˙book

401

systems: A nonlinear multibaker map. Phys. Rev. E 59, 364–371 (1999). (Cited on pages 68; 193; 200; 233 and 243.) T. Gilbert and J.R. Dorfman, Entropy production in a persistent random walk. Physica A 282, 427–449 (2000). (Cited on pages 3; 63; 233; 238 and 243.) T. Gilbert, J.R. Dorfman and P. Gaspard, Entropy production, fractals, and relaxation to equilibrium. Phys. Rev. Lett. 85, 1606–1609 (2000). (Cited on pages 3; 63; 233 and 238.) T. Gilbert, J.R. Dorfman and P. Gaspard, Fractal dimensions of the hydrodynamic modes of diffusion. Nonlinearity 14, 339–358 (2001). (Cited on pages 2; 9; 233; 240; 241; 243; 244 and 245.) T. Gilbert and J.R. Dorfman, On the parametric dependences of a class of non-linear singular maps. Discr. Contin. Dyn. Syst. B 4, 391–406 (2004). (Cited on pages 46; 200 and 233.) T. Gilbert, Fluctuation theorem applied to the Nos´e-Hoover thermostated Lorentz gas. Phys. Rev. E 73, 035102(R)/1–4 (2006). (Cited on page 326.) A. Giuliani, F. Zamponi and G. Gallavotti, Fluctuation relation beyond linear response theory. J. Stat. Phys. 119, 909–944 (2005). (Cited on pages 322; 323 and 324.) E. Glasner and B. Weiss, Sensitive dependence on initial conditions. Nonlinearity 6, 1067–1075 (1993). (Cited on pages 223 and 363.) J. Gleick, Chaos - Making a new science. (Penguin, New York, 1988). (Cited on page 14.) A.O. Golosov, Localization of random walks in one-dimensional random environments. Commun. Math. Phys. 92, 491–506 (1984). (Cited on page 84.) S. Goldstein, C. Kipnis and N. Ianiro, Stationary states for a mechanical system with stochastic boundary conditions. J. Stat. Phys. 41, 915–939 (1985). (Cited on pages 249 and 250.) R. Gorenflo, Y. Luchko and F. Mainardi, Wright functions as scaleinvariant solutions of the diffusion-wave equation. J. Comput. Appl. Math. 118, 175–191 (2000). (Cited on page 117.) G.A. Gottwald and I. Melbourne, A new test for chaos in deterministic systems. Proc. Roy. Soc. London 460, 603–611 (2004). (Cited on page 352.) G.A. Gottwald and I. Melbourne, Testing for chaos in deterministic systems with noise. Physica D 212, 100–110 (2005). (Cited on page 352.) P. Grassberger and T. Schreiber, Microscopic chaos from Brownian motion? Nature 401, 875–876 (1999). (Cited on pages 6; 90; 246; 320; 348; 350 and 351.) P. Grassberger, W. Nadler and L. Yang, Heat conduction and entropy production in a one-dimensional hard-particle gas. Phys. Rev. Lett. 89, 180601/1–4 (2002). (Cited on page 342.) M.S. Green, Markoff random processes and the statistical mechanics

May 8, 2007

4:1

402

[Gre84] [Gri89]

[Gro] [Gro81]

[Gro82]

[Gro83]

[Gro02]

[Gro03]

[Gro04]

[Gru94]

[Guc90]

[Gup94]

[Gut96] [Ham90]

[Han90]

[H¨ an96]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

of time-dependent phenomena II: Irreversible processes in fluids. J. of Chem. Phys. 22, 398–413 (1954). (Cited on page 64.) C. Grebogi, E. Ott, S. Pelikan and J.A. Yorke, Strange attractors that are not chaotic. Physica D 13, 261–268 (1984). (Cited on page 194.) M. Grilli and E. Tosatti, Exact canonical averages from microcanonical dynamics for quantum systems. Phys. Rev. Lett. 62, 2889–2892 (1989). (Cited on page 380.) J. Groeneveld. unpublished. (Cited on pages 32; 37 and 39.) J. Groeneveld, J. Jurkiewics and C.P. Korthals Altes, Twist as a probe for phase structure. Physica Scripta 23, 1022–1031 (1981). (Cited on page 33.) S. Grossmann and H. Fujisaka, Chaos-induced diffusion in nonlinear discrete dynamics. Phys. Rev. A 26, 1779–1782 (1982). (Cited on pages 20; 21; 25; 83; 86; 100; 102 and 105.) S. Grossmann and S. Thomae, Shape dependence of correlation times in chaos-induced diffusion. Phys. Lett. A 97, 263–267 (1983). (Cited on pages 21; 100 and 102.) J. Groeneveld and R. Klages, Negative and nonlinear response in an exactly solved dynamical model of particle transport. J. Stat. Phys. 109, 821–861 (2002). (Cited on pages 4; 17; 25; 29; 32; 37; 44; 46; 53; 94 and 200.) R. Grochowski, S. Strugholtz, P. Walzel and C.A. Kr¨ ulle, Granular transport on vibratory conveyor. Chemie Ingenieur Technik 75, 1103 (2003). (Cited on pages 6; 161 and 378.) R. Grochowski, P. Walzel, M. Rouijaa, C.A. Kr¨ ulle and I. Rehberg, Reversing a granular flow on a vibratory conveyor. Appl. Phys. Lett. 84, 1019–1022 (2004). (Cited on pages 6; 161 and 378.) H. Gruler and A. de Boisfleury-Chevance, Directed cell movement and cluster formation: Physical principles. J. Phys. I: France 4, 1085–1105 (1994). (Cited on page 304.) J. Guckenheimer and P. Holmes, Nonlinear oscillations, dynamical systems, and bifurcations of vector fields, volume 42 of Applied mathematical sciences. (Springer, Berlin, 3rd edition, 1990). (Cited on pages 100; 154 and 194.) D. Gupalo, A.S. Kaganovich and E.G.D. Cohen, Symmetry of Lyapunov spectrum. J. Stat. Phys. 74, 1145–1159 (1994). (Cited on pages 192 and 193.) E. Gutkin, Billiards in polygons: Survey of recent results. J. Stat. Phys. 83, 7–26 (1996). (Cited on pages 353; 354; 355 and 357.) I.P. Hamilton, Modified Nos´e-Hoover equation for a one-dimensional oscillator: Enforcement of the virial theorem. Phys. Rev. A 42, 7467– 7470 (1990). (Cited on pages 223; 224; 227 and 266.) J.H. Hannay and R.J. McCraw, Barrier billiards - a simple pseudointegrable system. J. Phys. A: Math. Gen. 23, 887–900 (1990). (Cited on pages 353; 354 and 358.) P. H¨ anggi and R. Bartussek, Brownian rectifiers: How to convert Brow-

May 8, 2007

4:1

Bibliography

World Scientific Book - 9in x 6in

klages˙book

403

nian motion into directed transport. in: J. Parisi, S. C. M¨ uller and W. Zimmermann, Eds., Nonlinear physics of complex systems - current status and future trends, Lecture notes in physics, pages 294–308. (Springer, Berlin, 1996). (Cited on pages 4; 50; 52 and 54.) [Han01] I. Han and Y. Lee, Chaotic dynamics of repeated impacts in vibratory bowl feeders. J. Sound. Vibr. 249, 529–541 (2001). (Cited on pages 6 and 160.) [Hao89] B.-L. Hao, Elementary symbolic dynamics and chaos in dissipative systems. (World Scientific, Singapore, 1989). (Cited on page 44.) [Har] T. Harayama. unpublished. (Cited on page 138.) [Har94] R.S. Hartmann, K. Lau, W. Chou and T.D. Coates, The fundamental motor of the human neutrophil is not random: Evidence for local nonMarkov movement in neutrophils. Biophys. J. 67, 2535–2545 (1994). (Cited on pages 304 and 305.) [Har01] T. Harayama and P. Gaspard, Diffusion of particles bouncing on a onedimensional periodically corrugated floor. Phys. Rev. E 64, 036215/1– 16 (2001). (Cited on pages 5 and 153.) [Har02] T. Harayama, R. Klages and P. Gaspard, Deterministic diffusion in flower-shaped billiards. Phys. Rev. E 66, 026211/1–7 (2002). (Cited on pages 5; 89 and 137.) [Has09] F. Hasen¨ ohrl, Eds., Wissenschaftliche Abhandlungen von L. Boltzmann, volume 2. (J.A. Barth Verlag, Leipzig, 1909). (Cited on page 176.) [Hat84] M. Hata and M. Yamaguti, The Takagi function and its generalization. Japan J. Appl. Math. 1, 183–199 (1984). (Cited on page 63.) [Hat99] T. Hatano, Heat conduction in the diatomic Toda lattice revisited. Phys. Rev. E 59, R1–R4 (1999). (Cited on page 342.) [Hat01] T. Hatano and S. Sasa, Steady-state thermodynamics of Langevin systems. Phys. Rev. Lett. 86, 3463–3466 (2001). (Cited on page 329.) [Hau87] J.W. Haus and K. W. Kehr, Diffusion in regular and disordered lattices. Phys. Rep. 150, 264–406 (1987). (Cited on pages 83 and 86.) [Heg96] H. Hegger, B. Huckestein, K. Hecker, M. Janssen, A. Freimuth, G. Reckziegel and R. Tuzinski, Fractal conductance fluctuations in gold nanowires. Phys. Rev. Lett. 77, 3885–3888 (1996). (Cited on page 378.) [Hes90] S. Hess and W. Loose, Flow properties and shear-induced structural changes in fluids: A case study on the interplay between theory, simulation, experiment and application. Ber. Bunsenges. Phys. Chem. 94, 216–222 (1990). (Cited on page 289.) [Hes96a] S. Hess, Constraints in molecular dynamics, nonequilibrium processes in fluids via computer simulations. in: K.H. Hoffmann and M. Schreiber, Eds., Computational physics, pages 268–293, Springer, Berlin, 1996. (Cited on pages 7; 8; 13; 167; 187; 210; 288 and 289.) [Hes96b] S. Hess, M. Kr¨ oger, W. Loose, C. Pereira Borgmeyer, R. Schramek, H. Voigt and T. Weider, Simple and complex fluids under shear. in: K. Binder and G. Ciccotti, Eds., Monte carlo and molecular dynamics

May 8, 2007

4:1

404

World Scientific Book - 9in x 6in

klages˙book

Bibliography

of condensed matter systems, volume 49 of IPS Conf. Proc., pages 825–841, Bologna, 1996. (Cited on pages 8; 187; 210; 288 and 289.) [Hes97] S. Hess, C. Aust, L. Bennett, M. Kr¨ oger, C. Pereira-Borgmeyer and T. Weider, Rheology: From simple and to complex fluids. Physica A 240, 126–144 (1997). (Cited on pages 8; 210 and 289.) [Hes03] S. Hess, Construction and test of thermostats and twirlers for molecular rotations. Z. Naturforsch. 58a, 377–391 (2003). (Cited on page 224.) [Hoa05] V.T. Hoang, Q. Huang, M. Eic, T.O. Do and S. Kaliaguine, Structure and diffusion characterization of SBA-15 materials. Langmuir 21, 2051–2057 (2005). (Cited on page 149.) [H¨ of06] F. H¨ ofling, T. Franosch and E. Frey, Localization transition in the three-dimensional Lorentz model and continuum percolation. Phys. Rev. Lett. 96, 165901/1–4 (2006). (Cited on page 86.) [Hol86] B.L. Holian and W.G. Hoover, Numerical test of the Liouville equation. Phys. Rev. A, 4229–4237 (1986). (Cited on pages 208; 210; 212; 221 and 223.) [Hol87] B.L. Holian, W.G. Hoover and H.A. Posch, Resolution of Loschmidt’s paradox: The origin of irreversible behavior in reversible atomistic dynamics. Phys. Rev. Lett. 59, 10–13 (1987). (Cited on pages 9; 187; 190; 193; 212; 219; 237 and 249.) [Hol95] B.L. Holian, A.F. Voter and R. Ravelo, Thermostatted molecular dynamics: How to avoid the Toda demon hidden in Nos´e-Hoover dynamics. Phys. Rev. E 52, 2338–2347 (1995). (Cited on pages 187; 212 and 221.) [Hon89a] M.-O. Hongler, P. Carter and P. Flury, Numerical study of a model of vibro-transporter. Phys. Lett. A 135, 106–112 (1989). (Cited on pages 154 and 160.) [Hon89b] M.-O. Hongler and J. Figour, Periodic versus chaotic dynamics in vibratory feeders. Helv. Phys. Acta 62, 68–81 (1989). (Cited on pages 154 and 160.) [Hoo75] W.G. Hoover and W.T. Ashurst, Nonequilibrium Molecular Dynamics. in: H. Eyring and D. Henderson, Eds., Theoretical Chemistry, volume 1, Academic Press, New York, 1975. (Cited on pages 288 and 291.) [Hoo82] W.G. Hoover, A.J.C. Ladd and B. Moran, High-strain-rate plastic flow studied via nonequilibrium molecular dynamics. Phys. Rev. Lett. 48, 1818–1821 (1982). (Cited on pages 8; 187; 223 and 289.) [Hoo85] W.G. Hoover, Canonical dynamics: Equilibrium phase-space distributions. Phys. Rev. A 31, 1695–1697 (1985). (Cited on pages 8; 208; 210; 212; 217; 221 and 223.) [Hoo87] W.G. Hoover and H.A. Posch, Direct measurement of equilibrium and nonequilibrium Lyapunov spectra. Phys. Lett. A 123, 227–230 (1987). (Cited on pages 9 and 237.) [Hoo88a] W.G. Hoover, B. Moran, C.G. Hoover and D.J. Evans, Irreversibility in the Galton board via conservative and quantum Hamiltonian and

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Bibliography

[Hoo88b]

[Hoo89a]

[Hoo89b]

[Hoo91]

[Hoo92] [Hoo94]

[Hoo96a]

[Hoo96b]

[Hoo97]

[Hoo98a]

[Hoo98b]

[Hoo98c] [Hoo98d]

[Hoo99]

[Hoo01]

klages˙book

405

Gaussian dynamics. Phys. Lett. A 133, 114–120 (1988). (Cited on pages 186; 193; 195; 196 and 217.) W.G. Hoover, C.G. Tull and H.A. Posch, Negative Lyapunov exponents for dissipative systems. Phys. Lett. A 131, 211–215 (1988). (Cited on page 192.) W.G. Hoover, Generalization of Nos´e’s isothermal molecular dynamics: Non-Hamiltonian dynamics for the canonical ensemble. Phys. Rev. 40, 2814–2815 (1989). (Cited on pages 223; 224 and 265.) W.G. Hoover and B. Moran, Phase-space singularities in atomistic planar diffusive flow. Phys. Rev. A 40, 5319–5326 (1989). (Cited on pages 186; 193; 195 and 196.) W.G. Hoover, Computational statistical mechanics, volume 11 of Studies in modern thermodynamics. (Elsevier, Amsterdam, 1991). (Cited on pages 1; 7; 8; 9; 193; 208; 210 and 223.) W.G. Hoover and B. Moran, Viscous attractor for the Galton board. Chaos 2, 599–602 (1992). (Cited on pages 186; 193; 197 and 198.) W.G. Hoover and H.A. Posch, Second-law irreversibility and phasespace dimensionality loss from time-reversible nonequilibrium steadystate Lyapunov spectra. Phys. Rev. E 49, 1913–1920 (1994). (Cited on page 193.) W.G. Hoover and B.L. Holian, Kinetic moments method for the canonical ensemble distribution. Phys. Lett. 211, 253–257 (1996). (Cited on pages 210; 223; 224; 225 and 265.) W.G. Hoover, O. Kum and H.A. Posch, Time-reversible dissipative ergodic maps. Phys. Rev. E 53, 2123–2129 (1996). (Cited on pages 9; 68 and 189.) W.G. Hoover and O. Kum, Ergodicity, mixing, and time reversibility for atomistic nonequilibrium steady states. Phys. Rev. E 56, 5517–5523 (1997). (Cited on pages 187; 197; 210; 223; 224; 225; 226 and 265.) W.G. Hoover, K. Boercker and H.A. Posch, Large-system hydrodynamic limit for color conductivity in two dimensions. Phys. Rev. E 57, 3911–3916 (1998). (Cited on page 334.) W.G. Hoover, D.J. Evans, H.A. Posch, B.L. Holian and G.P. Morriss, Comment on ”Toward a statistical thermodynamics of steady states”. Phys. Rev. Lett. 80, 4103 (1998). (Cited on page 207.) W.G. Hoover and H.A. Posch, Chaos and irreversibility in simple model systems. Chaos 8, 366–373 (1998). (Cited on pages 68 and 234.) W.G. Hoover and H.A. Posch, Multifractals from stochastic many-body molecular dynamics. Phys. Lett. A 246, 247–251 (1998). (Cited on pages 186; 193; 249 and 250.) W.G. Hoover, Time reversibility, computer simulation, and chaos. (World Scientific, Singapore, 1999). (Cited on pages 1; 3; 7; 8; 9; 13; 100; 166; 172; 186; 189; 190; 193; 195; 207; 209; 210; 218; 237; 238; 249; 288; 291 and 371.) W.G. Hoover, C.G. Hoover and D.I. Isbister, Chaos, ergodic convergence, and fractal instability for a thermostated canonical harmonic

May 8, 2007

4:1

406

[Hoo02a] [Hoo02b]

[Hoo02c]

[Hoo04]

[Hop37] [hor05]

[How97] [How01] [Hu95]

[Hu98] [Hua87] [Hub80]

[Huf01]

[Hum01]

[Iba96] [Ito81] [Jac68]

[Jac78]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

oscillator. Phys. Rev. E 63, 026209–1/5 (2001). (Cited on pages 223 and 226.) W.G. Hoover, 2002. private communication. (Cited on page 282.) W.G. Hoover, H.A. Posch, K. Aoki and D. Kusnezov, Remarks on nonHamiltonian statistical mechanics: Lyapunov exponents and phasespace dimensionality loss. Europhys. Lett. 60, 337–341 (2002). (Cited on pages 219; 299 and 343.) W.G. Hoover, H.A. Posch, C. Forster, C. Dellago and M. Zhou, Lyapunov modes of two-dimensional many-body systems; soft disks, hard disks, and rotors. J. Stat. Phys. 109, 765–776 (2002). (Cited on page 10.) W.G. Hoover, K. Aoki, C.G. Hoover and S.V. De Groot, Timereversible deterministic thermostats. Physica D 187, 253–267 (2004). (Cited on page 343.) E. Hopf, Ergodentheorie. (Springer, Berlin, 1937). (Cited on page 232.) H. El hor and S.J. Linz, Model for transport of granular matter on an annular vibratory conveyor. J. Stat. Mech., L02005/1–8 (2005). (Cited on page 161.) J. Howard, Molecular motors: Structural adaptations to cellular functions. Nature 389, 561–567 (1997). (Cited on pages 53 and 54.) J. Howard, Mechanics of motor proteins and the cytoskeleton. (Sinauer Associates, Inc., Sunderland, 2001). (Cited on page 53.) H. Hu and L.-S. Young, Nonexistence of SRB measures for some diffeomorphisms that are “almost Anosov”. Ergod. Theory Dyn. Syst. 15, 67–76 (1995). (Cited on pages 108 and 116.) B. Hu, B. Li and H. Zhao, Heat conduction in one-dimensional chains. Phys. Rev. E 57, 2992–2995 (1998). (Cited on pages 341 and 342.) K. Huang, Statistical mechanics. (Wiley, New York, 1987). (Cited on pages 1; 13; 165; 190; 206 and 207.) B.A. Huberman, J.P. Crutchfield and N.H. Packard, Noise phenomena in Josephson junctions. Appl. Phys. Lett. 37, 750–752 (1980). (Cited on page 100.) L. Hufnagel, R. Ketzmerick and M. Weiss, Conductance fluctuations of generic billiards: Fractal or isolated? Europhys. Lett. 54, 703–708 (2001). (Cited on page 378.) G. Hummer and A. Szabo, Free energy reconstruction from nonequilibrium single-molecule pulling experiments. Proc. Nat. Acad. Sci. 98, 3658–3661 (2001). (Cited on page 329.) H. Ibach and H. L¨ uth, Solid state physics. (Springer, Berlin, 1996). (Cited on page 181.) R. Ito, Rotation sets are closed. Math. Proc. Cambr. Phil. Soc. 89, 107–110 (1981). (Cited on page 31.) E.A. Jackson, J.R. Pasta and J.F. Waters, Thermal conductivity in one-dimensional lattices. J. Comput. Phys. 2, 207–227 (1968). (Cited on pages 248 and 271.) M.V. Jacobson, Topological and metric properties of one-dimensional

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Bibliography

[Jac81]

[Jar97a]

[Jar97b] [Jar00] [Jar02]

[Jel88a]

[Jel88b]

[Jel89]

[Jep06]

[J¨ ul97] [J¨ ul06] [Jun96]

[Jus01]

[Jus03]

[Kai01]

klages˙book

407

endomorphisms. Soviet. Math. Dokl. 19, 1452–1456 (1978). (Cited on page 105.) E. Ben Jacob, Y. Braiman, R. Shainsky and Y. Imry, Microwave induced ”Devil’s staircase” structure and chaotic behavior in current- fed Josephson junctions. Appl. Phys. Lett. 38, 822–824 (1981). (Cited on page 4.) C. Jarzynski, Equilibrium free-energy differences from nonequilibrium measurements: A master-equation approach. Phys. Rev. E 56, 5018– 5035 (1997). (Cited on page 329.) C. Jarzynski, Nonequilibrium equality for free energy differences. Phys. Rev. Lett. 78, 2690–2693 (1997). (Cited on pages 11 and 329.) C. Jarzynski, Hamiltonian derivation of a detailed fluctuation theorem. J. Stat. Phys. 98, 77–102 (2000). (Cited on page 330.) C. Jarzynski, What is the microscopic response of a system driven far from equilibrium? in: Dynamics of dissipation, pages 63–82, 2002. see Ref. [Gar02]. (Cited on pages 11 and 331.) J. Jellinek, Dynamics for nonconservative systems: Ergodicity beyond the microcanonical ensemble. J. Phys. Chem. 92, 3163–3173 (1988). (Cited on pages 223; 225 and 248.) J. Jellinek and R.S. Berry, Generalization of Nos´e’s isothermal molecular dynamics. Phys. Rev. A 38, 3069–3072 (1988). (Cited on pages 223 and 225.) J. Jellinek and R.S. Berry, Generalization of Nos´e’s isothermal molecular dynamics: Necessary and sufficient conditions of dynamical simulations of statistical ensembles. Phys. Rev. A 40, 2816–2818 (1989). (Cited on pages 223; 224 and 265.) O.G. Jepps and L. Rondoni, Thermodynamics and complexity of simple transport phenomena. J. Phys. A: Math. Gen. 39, 1311–1338 (2006). (Cited on pages 4; 6; 102; 149; 246; 358; 359; 360 and 378.) F. J¨ ulicher, A. Ajdari and J. Prost, Modeling molecular motors. Rev. Mod. Phys 69, 1269–1281 (1997). (Cited on pages 53 and 54.) F. J¨ ulicher, Molekulare Motoren und zellul¨ are Dynamik. Physik Journal 5, 59–63 (2006). (Cited on page 53.) P. Jung, J. G. Kissner and P. H¨ anggi, Regular and chaotic transport in asymmetric periodic potentials: Inertia ratchets. Phys. Rev. Lett. 76, 3436–3439 (1996). (Cited on pages 4; 52 and 378.) W. Just, H. Kantz, C. R¨ odenbeck and M. Helm, Stochastic modelling: Replacing fast degrees of freedom by noise. J. Phys. A: Math. Gen. 34, 3199–3213 (2001). (Cited on page 172.) W. Just, K. Gelfert, N. Baba, A. Riegert and H. Kantz, Elimination of fast chaotic degrees of freedom: On the accuracy of the Born approximation. J. Stat. Phys. 112, 277–292 (2003). (Cited on page 172.) M. Kaisermayr, J. Combet, B. Sepiol and G. Vogl, Correlation factors in intermetallic alloys determined using neutron scattering: Ni3Sb, NiGa and CoGa. Defect and Diffusion Forum 194–199, 461–466 (2001). (Cited on page 150.)

May 8, 2007

4:1

408

[Kan85]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

H. Kantz and P. Grassberger, Repellers, semi-attractors, and long-lived chaotic transients. Physica D 17, 75–86 (1985). (Cited on page 23.) [Kan98] H. Kantz and E. Olbrich, The transition from deterministic chaos to a stochastic process. Physica A 253, 105–117 (1998). (Cited on page 90.) [Kan04] H. Kantz, W. Just, N. Baba, K. Gelfert and A. Riegert, Fast chaos versus white noise: Entropy analysis and a Fokker-Planck model for the slow dynamics. Physica D 187, 200–213 (2004). (Cited on page 172.) [K¨ ar92] J. K¨ arger and D.M. Ruthven, Diffusion in zeolites. (Wiley, 1992). (Cited on pages 5; 148 and 150.) [Kar00] J. Karkheck, Eds. Dynamics: Models and kinetic methods for nonequilibrium many body systems, volume 371 of NATO Science Series E: Applied Sciences, Kluwer, Dordrecht, 2000. (Cited on pages 3 and 13.) [K¨ ar03] J. K¨ arger, S. Vasenkov and S.M. Auerbach, Diffusion in zeolites. in: S.M. Auerbach, K.A. Carrado and P.K. Dutta, Eds., Handbook of zeolite science and technology, pages 341–423, Marcel-Dekker Inc., New York, 2003. (Cited on page 150.) [Kat95] A. Katok and B. Hasselblatt, Introduction to the modern theory of dynamical systems, volume 54 of Encyclopedia of Mathematics and its Applications. (Cambridge University Press, Cambridge, 1995). (Cited on page 21.) [Keh98] K.W. Kehr, Diffusion of particles on lattices. in: R. Haberlandt J. K¨ arger, P. Heitjans, Eds., Diffusion in condensed matter, pages 265– 305, Vieweg, Braunschweig, 1998. (Cited on pages 83 and 86.) [Kel07] G. Keller, R. Klages and P. Howard, Parameter dependence of deterministic transport coefficients: A mathematical approach. preprint, 2007. (Cited on pages 4; 17; 25; 26 and 37.) [Ket96] R. Ketzmerick, Fractal conductance fluctuations in generic chaotic cavities. Phys. Rev. B 54, 10841–10844 (1996). (Cited on page 378.) [Kla] R. Klages. unpublished. (Cited on pages 42; 77; 79; 80; 87; 92; 94; 96; 259; 263; 278 and 313.) [Kla95] R. Klages and J.R. Dorfman, Simple maps with fractal diffusion coefficients. Phys. Rev. Lett. 74, 387–390 (1995). (Cited on pages 4; 17; 37; 65; 90 and 200.) [Kla96a] J. Klafter, M. F. Shlesinger and G. Zumofen, Beyond Brownian motion. Phys. Today 49, 33–39 (1996). (Cited on page 108.) [Kla96b] R. Klages, Deterministic diffusion in one-dimensional chaotic dynamical systems. (Wissenschaft & Technik-Verlag, Berlin, 1996). (Cited on pages 3; 4; 13; 17; 21; 22; 23; 25; 28; 33; 37; 38; 40; 48; 64; 65; 66; 67; 69; 76; 89; 90; 93; 96; 97; 102; 103; 104; 114; 115; 116; 117; 123; 200; 241 and 243.) [Kla97] R. Klages and J.R. Dorfman, Dynamical crossover in deterministic diffusion. Phys. Rev. E 55, R1247–R1250 (1997). (Cited on pages 4; 21; 22; 26; 33; 37; 68; 96; 102; 114; 115; 116 and 123.) [Kla98] R. Klages and J. Groeneveld, Simple deterministic dynamical systems with drift: negative currents and suppression of diffusion. in: Verhandl. DPG, volume VI 33, page 646, 1998. (Cited on page 53.)

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Bibliography

[Kla99a]

[Kla99b] [Kla00a]

[Kla00b]

[Kla02a]

[Kla02b]

[Kla02c] [Kla02d]

[Kla03]

[Kla04a]

[Kla04b]

[Kla04c]

[Klu90]

[Kna87]

[Kog83]

klages˙book

409

R. Klages and J.R. Dorfman, Simple deterministic dynamical systems with fractal diffusion coefficients. Phys. Rev. E 59, 5361–5383 (1999). (Cited on pages 4; 17; 23; 25; 37; 38; 65; 90; 200; 241; 243 and 371.) R. Klages and K. Rateitschak. unpublished, 1999. (Cited on pages 259 and 273.) R. Klages and C. Dellago, Density-dependent diffusion in the periodic Lorentz gas. J. Stat. Phys. 101, 145–159 (2000). see also chaodyn/9910038 for further details. (Cited on pages 5; 89; 115; 128; 130; 131; 132; 133; 134; 137; 138; 142; 144; 145; 199 and 201.) R. Klages, K. Rateitschak and G. Nicolis, Thermostating by deterministic scattering: construction of nonequilibrium steady states. Phys. Rev. Lett. 84, 4268–4271 (2000). (Cited on pages 193; 271; 273; 277; 279; 280; 282; 283; 284 and 285.) R. Klages, Comment on “Analysis of chaotic motion and its shape dependence in a generalized piecewise linear map”. Phys. Rev. E 66, 018201/1–3 (2002). (Cited on pages 4; 26; 37; 102; 117 and 123.) R. Klages, Suppression and enhancement of deterministic diffusion in disordered dynamical systems. Phys. Rev. E 65, 055203(R)/1–4 (2002). (Cited on pages 5; 83 and 93.) R. Klages, Transitions from deterministic to stochastic diffusion. Europhys. Lett. 57, 796–802 (2002). (Cited on pages 5 and 83.) R. Klages and N. Korabel, Understanding deterministic diffusion by correlated random walks. J. Phys. A: Math. Gen. 35, 4823–4836 (2002). (Cited on pages 4; 5; 26; 64; 67; 102; 116; 121; 131 and 137.) R. Klages and T. Klauß, Fractal fractal dimensions of deterministic transport coefficients. J.Phys.A: Math. Gen. 36, 5747–5764 (2003). (Cited on pages 4; 25; 26 and 44.) R. Klages. Microscopic chaos, fractals, and transport in nonequilibrium steady states, 2004. Habilitation thesis, Technische Universit¨ at Dresden; available under http://nbn-resolving.de/urn:nbn:de:swb:141107426004665-22819. (Cited on pages ix and 53.) R. Klages, I.F. Barna and L. M´ aty´ as, Spiral modes in the diffusion of a single granular particle on a vibrating surface. Phys. Lett. A 333, 79–84 (2004). (Cited on pages 6; 153; 159 and 161.) R. Klages, H. van Beijeren, J.R. Dorfman and P. Gaspard, Eds. Microscopic chaos and transport in many-particle systems, volume 187 of Physica D, North Holland, Amsterdam, 2004. special issue. (Cited on page 13.) R. Kluiving, H.W. Capel and R.A. Pasmanter, Phase-transition-like phenomenon in a piecewise linear map. Physica A 164, 593–624 (1990). (Cited on pages 49 and 65.) A. Knauf, Ergodic and topological properties of Coulombic periodic potentials. Commun. Math. Phys. 110, 89–112 (1987). (Cited on page 337.) H. Koga, H. Fujisaka and M. Inoue, Anomalous enhancement of the

May 8, 2007

4:1

410

[Kol97]

[Kor02]

[Kor04a]

[Kor04b]

[Kor05]

[Kor07]

[Kos05a]

[Kos05b]

[Kov88]

[Koz99]

[Koz04]

[Kub57]

[Kub92]

[Kur98]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

diffusion coefficient near the intermittency transition. Phys. Rev. A 28, 2370–2373 (1983). (Cited on pages 53; 100 and 107.) S. Kolyada and L’Ubomir Snoha, Some aspects of topological transitivity - a survey. Grazer Mathematische Berichte 334, 3–35 (1997). (Cited on pages 21; 223 and 363.) N. Korabel and R. Klages, Fractal structures of normal and anomalous diffusion in nonlinear nonhyperbolic dynamical systems. Phys. Rev. Lett. 89, 214102/1–4 (2002). (Cited on pages 4; 67 and 99.) N. Korabel, Deterministic transport: From normal to anomalous diffusion. PhD thesis, Technische Universit¨ at Dresden, 2004. available under http://nbn-resolving.de/urn:nbn:de:swb:14-1101988907015-56546. (Cited on pages 99 and 352.) N. Korabel and R. Klages, Fractality of deterministic diffusion in the nonhyperbolic climbing sine map. Physica D 187, 66–88 (2004). (Cited on pages 4; 99; 103 and 107.) N. Korabel, A.V. Chechkin, R. Klages, I.M. Sokolov and V.Yu. Gonchar, Understanding anomalous transport in intermittent maps: From continuous time random walks to fractals. Europhys. Lett. 70, 63–69 (2005). (Cited on pages 5 and 109.) N. Korabel, R. Klages, A.V. Chechkin, I.M. Sokolov and V.Yu. Gonchar, Fractal properties of anomalous diffusion in intermittent maps. preprint nlin.CD/0607010, accepted for publication in Phys. Rev. E, 2007. (Cited on pages 99; 109; 115; 116 and 117.) T. Kosztolowicz, K. Dworecki and S. Mrowczynski, How to measure subdiffusion parameters. Phys. Rev. Lett. 94, 170602/1–4 (2005). (Cited on page 109.) T. Kosztolowicz, K. Dworecki and S. Mrowczynski, Measuring subdiffusion parameters. Phys. Rev. E 71, 041105/1–11 (2005). (Cited on page 109.) Z.J. Kovalik, M. Franaszek and P. Pieranski, Self-reanimating chaos in the bouncing-ball system. Phys. Rev. A 37, 4016–4022 (1988). (Cited on page 154.) Z. Koza, General technique of calculating the drift velocity and diffusion coefficient in arbitrary periodic systems. J. Phys. A: Math. Gen. 32, 7637–7651 (1999). (Cited on page 4.) Z. Koza, Fractal dimension of transport coefficients in a deterministic dynamical system. J. Phys. A: Math. Gen. 37, 10859–10877 (2004). (Cited on pages 4; 25; 26 and 37.) R. Kubo, Statistical-mechanical theory of irreversible processes I: General theory and simple applications to magnetic and conduction problems. J. Phys. Soc. Jap. 12, 570–586 (1957). (Cited on page 64.) R. Kubo, M. Toda and N. Hashitsume, Statistical Physics, volume 2 of Solid State Sciences. (Springer, Berlin, 2nd edition, 1992). (Cited on pages 7; 169 and 201.) J. Kurchan, Fluctuation theorem for stochastic dynamics. J. Phys. A: Math. Gen. 31, 3719–3729 (1998). (Cited on pages 10 and 327.)

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Bibliography

[Kur05] [Kus90]

[Kus92]

[Kus93]

[Kus95] [Kus02]

[Lad85]

[Lai03]

[Lan08]

[Lar03]

[Las94]

[Lat97]

[Leb59] [Leb78]

[Leb98] [Leb99]

klages˙book

411

J. Kurchan, Non-equilibrium work relations. preprint arXiv:condmat/0511073, 2005. (Cited on pages 11 and 331.) D. Kusnezov, A. Bulgac and W. Bauer, Canonical ensembles from chaos. Ann. Phys. 204, 155–185 (1990). (Cited on pages 193; 208; 212; 221; 223; 224; 226; 227 and 265.) D. Kusnezov and A. Bulgac, Canonical ensembles from chaos, 2.: Constrained dynamic-systems. Ann. Phys. 214, 180–218 (1992). (Cited on page 218.) D. Kusnezov, Quantum ergodic wavefunctions from a thermal nonlinear Schr¨ odinger equation. Phys. Lett. A 184, 50–56 (1993). (Cited on pages 208 and 380.) D. Kusnezov, Dimensional loss in nonequilibrium quantum systems. Phys. Rev. Lett. 74, 246–249 (1995). (Cited on page 380.) D. Kusnezov, E. Lutz and K. Aoki, Non-equilibrium statistical mechanics of classical and quantum systems. in: Dynamics of dissipation, pages 83–108, 2002. see Ref. [Gar02]. (Cited on pages 208; 219; 250; 282; 299; 343 and 380.) A.J.C. Ladd and W.G. Hoover, Lorentz gas shear viscosity via nonequilibrium molecular dynamics and Boltzmann’s equation. J. Stat. Phys. 38, 973–988 (1985). (Cited on page 289.) B.B. Laird and B.J. Leimkuhler, Generalized dynamical thermostating technique. Phys. Rev. E 68, 016704/1–6 (2003). (Cited on pages 210; 223 and 225.) P. Langevin, Sur la th´eorie du mouvement brownien. C.R. Acad. Sci. (Paris) 146, 530–533 (1908). see also the English translation in Am. J. Phys. 65, 1079 (1997). (Cited on pages 7 and 167.) H. Larralde, F. Leyvraz and C. Mejia-Monasterio, Transport properties of a modified Lorentz gas. J. Stat. Phys. 113, 197–231 (2003). (Cited on pages 186; 232; 274; 275; 287; 346 and 347.) A. Lasota and M.C. Mackey, Chaos, fractals, and noise, volume 97 of Applied Mathematical Sciences. (Springer, Berlin, 2nd edition, 1994). (Cited on page 22.) A. Latz, H. van Beijeren and J.R. Dorfman, Lyapunov spectrum and the conjugate pairing rule for a thermostatted random Lorentz gas: Kinetic theory. Phys. Rev. Lett. 78, 207–210 (1997). (Cited on pages 10; 183; 192 and 252.) J.L. Lebowitz, Stationary nonequilibrium Gibbsian ensembles. Phys. Rev. 114, 1192–1202 (1959). (Cited on page 63.) J.L. Lebowitz and H. Spohn, Transport properties of the Lorentz gas: Fourier’s law. J. Stat. Phys. 19, 633–654 (1978). (Cited on pages 249; 271 and 345.) P. Leboeuf, Normal and anomalous diffusion in a deterministic areapreserving map. Physica D 116, 8–20 (1998). (Cited on page 377.) J.L. Lebowitz and H. Spohn, A Gallavotti-Cohen-type symmetry in the large deviation functional for stochastic dynamics. J. Stat. Phys. 95, 333–365 (1999). (Cited on pages 10; 327 and 328.)

May 8, 2007

4:1

412

[Lep97]

[Lep98]

[Lep00]

[Lep03]

[L’H93]

[Li02]

[Li03a]

[Li03b]

[Li04]

[Li05]

[Lic92]

[Lie92]

[Lie99] [Lio38] [Lip02]

[Liu00]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

S. Lepri, R. Livi and A. Politi, Heat conduction in chains of nonlinear oscillators. Phys. Rev. Lett. 78, 1896–1989 (1997). (Cited on page 340.) S. Lepri, R. Livi and A. Politi, On the anomalous thermal conductivity of one-dimensional lattices. Europhys. Lett. 43, 271–276 (1998). (Cited on page 342.) S. Lepri, L. Rondoni and G. Benettin, The Gallavotti-Cohen fluctuation theorem for a nonchaotic model. J. Stat. Phys. 99, 857–872 (2000). (Cited on page 324.) S. Lepri, R. Livi and A. Politi, Thermal conduction in classical lowdimensional lattices. Phys. Rep. 377, 1–80 (2003). (Cited on pages 6; 291; 319; 337; 338; 339; 340; 341; 342 and 343.) I. L’Heureux and I. Hamilton, Canonically modified Nos´e-Hoover equation with explicit inclusion of the virial. Phys. Rev. E 47, 1411–1414 (1993). (Cited on pages 223; 224; 227 and 266.) B. Li, L. Wang and B. Hu, Finite thermal conductivity in 1d models having zero Lyapunov exponent. Phys. Rev. Lett. 88, 223901/1–4 (2002). (Cited on pages 6; 358; 359; 362 and 363.) B. Li, G. Casati and J. Wang, Heat conductivity in linear mixing systems. Phys. Rev. E 67, 021204/1–4 (2003). (Cited on pages 6; 359 and 362.) B. Li and J. Wang, Anomalous heat conduction and anomalous diffusion in one-dimensional systems. Phys. Rev. Lett. 91, 044301/1–4 (2003). (Cited on page 363.) B. Li and J. Wang, Reply to comment on “Anomalous heat conduction and anomalous diffusion in one-dimensional systems”. Phys. Rev. Lett. 92, 089402/1 (2004). (Cited on page 363.) B. Li, J. Wang, L. Wang and G. Zhang, Anomalous heat conduction and anomalous diffusion in nonlinear lattices, single walled nanotubes, and billiard gas channels. Chaos 15, 015121/1–13 (2005). (Cited on pages 342; 358; 361; 362 and 363.) A.J. Lichtenberg and M.A. Lieberman, Regular and chaotic dynamics, volume 38 of Applied Mathematical Sciences. (Springer, New York, 2nd edition, 1992). (Cited on pages 97; 102; 107 and 154.) S.Y. Liem, D. Brown and J.H.R. Clarke, Investigation of the homogeneous-shear nonequilibrium-molecular-dynamics method. Phys. Rev. A 45, 3706–3713 (1992). (Cited on page 288.) E. Lieb, Some problems in statistical mechanics that I would like to see solved. Physica A 263, 491–499 (1999). (Cited on page 8.) J. Liouville, Sur la th´eorie de la variation des constantes arbitraires. J. Math. Pures Appl. 3, 342–349 (1838). (Cited on page 207.) J. Liphardt, S. Dumont, S.B. Smith, I. Tinoco Jr. and C. Bustamante, Equilibrium information from nonequilibrium measurements in an experimental test of Jarzynski’s equality. Science 296, 1832–1835 (2002). (Cited on page 329.) Y. Liu and M.E. Tuckerman, Generalized Gaussian moment ther-

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Bibliography

[Llo94]

[Llo95]

[Loo90]

[Lor05] [Lor91]

[Luc93]

[Lue93]

[Lut04] [Lyo85]

[Mac59] [Mac83]

[Mac84] [Mac04]

[Mae99] [Mae03] [Mai97]

klages˙book

413

mostatting: A new continuous dynamical approach to the canonical ensemble. J. Chem. Phys. 112, 1685–1700 (2000). (Cited on pages 223; 224 and 265.) J. Lloyd, L. Rondoni and G.P. Morriss, Breakdown of ergodic behavior in the Lorentz gas. Phys. Rev. E 50, 3416–3421 (1994). (Cited on pages 186; 193; 196 and 197.) J. Lloyd, M. Niemeyer, L. Rondoni and G.P. Morriss, The nonequilibrium Lorentz gas. Chaos 5, 536–551 (1995). (Cited on pages 186; 190; 191; 193; 196; 197; 198; 199; 200 and 201.) W. Loose and S. Hess, Shear-induced ordering revisited. in: M. Mareschal, Eds., Microscopic simulations of complex flows, volume 236 of NATO Asi Series B, Plenum Press, New York, 1990. (Cited on page 289.) H.A. Lorentz, The motion of electrons in metallic bodies. Proc. Roy. Acad. Amst. 7, 438–453 (1905). (Cited on pages 3; 128 and 179.) A. Lorke, J.P. Kotthaus and K. Ploog, Magnetotransport in twodimensional superlattices. Phys. Rev. B 44, 3447–3450 (1991). (Cited on pages 5; 134 and 181.) J.M. Luck and A. Mehta, Bouncing ball with a finite restitution: Chattering, locking, and chaos. Phys. Rev. E 48, 3988–3997 (1993). (Cited on page 154.) A. Lue and H. Brenner, Phase flow and statistical structure of Galtonboard systems. Phys. Rev. E 47, 3128–3144 (1993). (Cited on pages 186; 193 and 197.) E. Lutz, Power-law tail distributions and nonergodicity. Phys. Rev. Lett. 93, 190602/1–4 (2004). (Cited on page 108.) S. K. Lyo and P. M. Richards, Diffusion in a linear lattice with correlated random rates satisfying detailed balance. Phys. Rev. B 32, 301– 304 (1985). (Cited on page 86.) J.A. MacLennan Jr., Statistical mechanics of the steady state. Phys. Rev. 115, 1405–1409 (1959). (Cited on page 63.) J. Machta and R. Zwanzig, Diffusion in a periodic Lorentz gas. Phys. Rev. Lett. 50, 1959–1962 (1983). (Cited on pages 129; 130; 142; 158; 182 and 199.) R.S. MacKay, J.D. Meiss and I.C. Percival, Transport in Hamiltonian systems. Physica D 13, 55–81 (1984). (Cited on page 27.) R. Mach, A. Ordemann and F. Schweitzer, Modeling vortex swarming in daphnia. preprint arXiv:q-bio.PE/0404028, 2004. (Cited on pages 305 and 314.) C. Maes, The fluctuation theorem as a Gibbs property. J. Stat. Phys. 95, 367–392 (1999). (Cited on pages 10 and 327.) C. Maes and K. Netocny, Time-reversal and entropy. J. Stat. Phys. 110, 269–310 (2003). (Cited on pages 238 and 297.) F. Mainardi, Fractional calculus, some basic problems in continuum and statistical mechanics. in: A. Carpinteri and F. Mainardi, Eds., Fractals and fractional calculus in continuum mechanics, volume 378

May 8, 2007

4:1

414

[Mai06] [Man80]

[Man82] [Mar86]

[Mar89]

[Mar92a]

[Mar92b]

[Mar97]

[Mar04] [Mat97]

[Mat00] [M´ at01]

[M´ at04a] [M´ at04b]

[M´ at05]

[Max79]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

of CISM Courses and Lecture Notes, pages 291–348, Springer, Berlin, 1997. (Cited on page 117.) T. Mai and O. Narayan, Universality of one-dimensional heat conductivity. Phys. Rev. E 73, 061202/1–7 (2006). (Cited on page 342.) P. Manneville, Intermittency, self-similarity and 1-F spectrum in dissipative dynamical systems. J. Physique 41, 1235–1243 (1980). (Cited on pages 108 and 109.) B.B. Mandelbrot, The fractal geometry of nature. (W.H. Freeman and Company, San Francisco, 1982). (Cited on pages 25; 44; 124 and 127.) S. Martin and W. Martienssen, Circle maps and mode locking in the driven electrical conductivity of Barium Sodium Niobate crystals. Phys. Rev. Lett. 52, 1522–1525 (1986). (Cited on pages 4 and 107.) J.M. Martinis and R.L. Kautz, Classical phase diffusion in small hysteretic Josephson junctions. Phys. Rev. Lett. 63, 1507–1510 (1989). (Cited on pages 4 and 107.) M. Mareschal and B.L. Holian, Eds. Microscopic simulations of complex hydrodynamic phenomena, volume 292 of NATO ASI Series B: Physics, Plenum Press, New York, 1992. (Cited on pages 9; 13; 389; 396 and 399.) G.J. Martyna, M.L. Klein and M. Tuckerman, Nos´e-Hoover chains: The canonical ensemble via continuos dynamics. J. Chem. Phys. 97, 2635–2643 (1992). (Cited on pages 210; 223; 224; 226 and 265.) M. Mareschal, Eds. The microscopic approach to complexity in nonequilibrium molecular simulations, volume 240 of Physica A, North Holland, Amsterdam, 1997. special issue. (Cited on pages 9 and 13.) M. Mareschal and S. McNamara, Lyapunov hydrodynamics in the dilute limit. Physica D 187, 311–325 (2004). (Cited on pages 10 and 336.) H. Matsuoka and R.F. Martin, Long-time tails of the velocity autocorrelation functions for the triangular periodic Lorentz gas. J. Stat. Phys 88, 81–103 (1997). (Cited on page 182.) J.L. Mateos, Chaotic transport and current reversal in deterministic ratchets. Phys. Rev. Lett. 84, 258–261 (2000). (Cited on page 52.) L. M´ aty´ as, T. T´el and J. Vollmer, Multibaker map for shear flow and viscous heating. Phys. Rev. E 64, 056106/1–11 (2001). (Cited on pages 3; 63 and 290.) L. M´ aty´ as and R. Klages, Irregular diffusion in the bouncing ball billiard. Physica D 187, 165–183 (2004). (Cited on pages 6 and 153.) L. M´ aty´ as, T. T´el and J. Vollmer, Coarse-grained entropy and information dimension of dynamical systems: The driven Lorentz gas. Phys. Rev. E 69, 016205/1–8 (2004). (Cited on pages 3 and 238.) L. M´ aty´ as and P. Gaspard, Entropy production in diffusion-reaction systems: The reactive random Lorentz gas. Phys. Rev. E 71, 036147/1– 10 (2005). (Cited on page 81.) J.C. Maxwell, On Boltzmann’s Theorem on the average distribution in a system of material points. Cam. Phil. Trans. 12, 547 (1879). (Cited on page 176.)

May 8, 2007

4:1

Bibliography

World Scientific Book - 9in x 6in

klages˙book

415

[Max65a] J.C. Maxwell, Illustrations of the dynamical theory of gases. in: S.G. Brush, Eds., Kinetic Theory, volume 1, pages 148–171, Pergamon Press, Oxford, 1965. (Cited on page 1.) [Max65b] J.C. Maxwell, On the dynamical theory of gases. in: S.G. Brush, Eds., Kinetic Theory, volume 2, pages 23–87, Pergamon Press, Oxford, 1965. (Cited on page 1.) [McN01] S. McNamara and M. Mareschal, Origin of the hydrodynamic Lyapunov modes. Phys. Rev. E 64, 051103/1–14 (2001). (Cited on pages 10 and 336.) [Meh90] A. Mehta and J.M. Luck, Novel temporal behavior of a nonlinear dynamical system: The completely inelastic bouncing ball. Phys. Rev. Lett. 65, 393–396 (1990). (Cited on page 154.) [Mei92] J.D. Meiss, Symplectic maps, variational principles, and transport. Rev. Mod. Phys. 64, 795–848 (1992). (Cited on pages 27; 107; 192; 215; 234 and 273.) [Mej01] C. Mejia-Monasterio, H. Larralde and F. Leyvraz, Coupled normal heat and matter transport in a simple model system. Phys. Rev. Lett. 86, 5417–5420 (2001). (Cited on pages 186; 274; 287 and 346.) [Men01] D. Mentrup and J. Schnack, Nos´e-Hoover dynamics for coherent states. Physica A 297, 337–347 (2001). (Cited on pages 208 and 380.) [Men03] D. Mentrup and J. Schnack, Nos´e-Hoover sampling of quantum entangled distribution functions. Physica A 326, 370–383 (2003). (Cited on pages 208 and 380.) [Met73] N. Metropolis, M.L. Stein and P.R. Stein, On finite limit sets for transformations on the unit interval. J. Combin. Theor. 15, 25–44 (1973). (Cited on page 105.) [Met00] R. Metzler and J. Klafter, The random walk’s guide to anomalous diffusion: A fractional dynamics approach. Phys. Rep. 339, 1–77 (2000). (Cited on pages 108; 109; 117 and 168.) [Met04a] R. Metzler and J. Klafter, The restaurant at the end of the random walk: Recent developments in the description of anomalous transport by fractional dynamics. J. Phys. A: Math. Gen. 37, R161–R208 (2004). (Cited on page 108.) [Met04b] R. Metzler and I.M. Sokolov, Comment on “Anomalous heat conduction and anomalous diffusion in one-dimensional systems”. Phys. Rev. Lett. 92, 089401/1 (2004). (Cited on page 363.) [Mil98a] L. Milanovic, H.A. Posch and W.G. Hoover, Lyapunov instability of two-dimensional fluids: Hard dumbbells. Chaos 8, 455–461 (1998). (Cited on pages 10; 334 and 335.) [Mil98b] L.J. Milanovic, H.A. Posch and W. Thirring, Statistical mechanics and computer simulation of systems with attractive positive power-law potentials. Phys. Rev. E 57, 2763–2775 (1998). (Cited on page 176.) [Mil02] L. Milanovic and H.A. Posch, Localized and delocalized modes in the tangent-space dynamics of planar hard dumbbell fluids. J. Molec. Liqu. 96-97, 221–244 (2002). (Cited on pages 10; 333; 334 and 335.) [Mir85] R.F. Miracky, M.H. Devoret and J. Clarke, Deterministic hopping in a

May 8, 2007

4:1

416

[Mon65] [Mon73]

[Mor87a]

[Mor87b]

[Mor89a]

[Mor89b] [Mor89c]

[Mor94] [Mor96a]

[Mor96b]

[Mor98]

[Mor99]

[Mor02]

[MT99]

[M¨ ul04]

[Mun00]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

Josephson circuit described by a one-dimensional mapping. Phys. Rev. A 31, 2509–2519 (1985). (Cited on pages 4; 100 and 107.) E.W. Montroll and G.H. Weiss, Random walks on lattices II. J. Math. Phys. 6, 167–179 (1965). (Cited on pages 108 and 113.) E.W. Montroll and H. Scher, Random walks on lattices IV: Continuoustime walks and influence of absorbing boundaries. J. Stat. Phys. 9, 101–133 (1973). (Cited on pages 108 and 113.) B. Moran and W.G. Hoover, Diffusion in a periodic Lorentz gas. J. Stat. Phys. 48, 709–726 (1987). (Cited on pages 9; 186; 193; 195; 196; 197; 198; 199 and 201.) G.P. Morriss, The information dimension of the nonequilibrium distribution function. Phys. Lett. A 122, 236–240 (1987). (Cited on pages 9; 193 and 289.) G.P. Morriss, Dimensional contraction in nonequilibrium systems. Phys. Lett. A 134, 307–313 (1989). (Cited on pages 9; 191; 193; 197; 289 and 298.) G.P. Morriss, Phase-space singularities in a planar Couette flow. Phys. Rev. A 39, 4811–4816 (1989). (Cited on pages 193; 289 and 298.) G.P. Morriss, D.J. Evans, E.G.D. Cohen and H. van Beijeren, Linear response of phase-space trajectories to shearing. Phys. Rev. Lett. 14, 1579–1582 (1989). (Cited on pages 201 and 289.) G.P. Morriss and L. Rondoni, Periodic orbit expansions for the Lorentz gas. J. Stat. Phys. 75, 553–584 (1994). (Cited on page 199.) G.P. Morriss, C.P. Dettmann and D.J. Isbister, Field dependence of Lyapunov exponents for nonequilibrium systems. Phys. Rev. E 54, 4748–4754 (1996). (Cited on pages 186; 190; 193 and 287.) G.P. Morriss and L. Rondoni, Equivalence of ”nonequilibrium” ensembles for simple maps. Physica A 233, 767–784 (1996). (Cited on page 243.) G.P. Morriss and C.P. Dettmann, Thermostats: Analysis and application. Chaos 8, 321–336 (1998). (Cited on pages 7; 8; 9; 10; 13; 166; 186; 188; 189; 190; 191; 192; 193; 196; 197; 209; 210; 216; 218 and 258.) G.P. Morriss and L. Rondoni, Definition of temperature in equilibrium and nonequilibrium systems. Phys. Rev. E 59, R5–R8 (1999). (Cited on page 166.) G.P. Morriss, Conjugate pairing of Lyapunov exponents for isokinetic shear flow algorithms. Phys. Rev. E 65, 017201/1–3 (2002). (Cited on pages 192 and 193.) C.J. Mundy M.E. Tuckerman and G.J. Martyna, On the classical statistical mechanics of non-Hamiltonian systems. Europhys. Lett. 45, 149–155 (1999). (Cited on page 207.) O. M¨ ulken and H. van Beijeren, Thermodynamic formalism for fielddriven Lorentz gases. Phys. Rev. E 69, 046203/1–9 (2004). (Cited on pages 10; 84 and 183.) C.J. Mundy, S. Balasubramanian, K. Bagchi, M.E. Tuckerman, G.J. Martyna and M.L. Klein, Nonequilibrium molecular dynamics, vol-

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Bibliography

[Mus98]

[Nar02]

[Nic96]

[Nic98]

[Nie98] [Nob95] [Nos84a]

[Nos84b]

[Nos91]

[Nos93]

[Ott93]

[Pan02a]

[Pan02b]

[Pan02c]

[Pat88]

klages˙book

417

ume 14 of Reviews in Computational Chemistry, chapter 5. (WileyVCH, New York, 2000). (Cited on pages 7; 8; 207; 210; 223; 288 and 289.) K. Mussawisade, T. Wichmann and K.W. Kehr, Single-particle diffusion coefficient on surfaces with Ehrlich-Schwoebel barriers. Surf. Sci. 412/413, 55–60 (1998). (Cited on page 89.) O. Narayan and S. Ramaswamy, Anomalous heat conduction in onedimensional momentum-conserving systems. Phys. Rev. Lett. 89, 200601/1–4 (2002). (Cited on pages 341 and 342.) G. Nicolis and D. Daems, Nonequilibrium thermodynamics of dynamical systems. J. Phys. Chem. 100, 19187–19191 (1996). (Cited on pages 238; 250 and 266.) G. Nicolis and D. Daems, Probabilistic and thermodynamic aspects of dynamical systems. Chaos 8, 311–320 (1998). (Cited on pages 13; 237; 238; 250; 266 and 296.) S. Nielsen and R. Kapral, Coloring a Lorentz gas. J. Chem. Phys. 109, 6460–6468 (1998). (Cited on page 81.) B. Nobbe, Classical motion in two-dimensional crystals. J. Stat. Phys. 78, 1591–1605 (1995). (Cited on page 337.) S. Nos´e, A molecular dynamics method for simulations in the canonical ensemble. Mol. Phys. 52, 255–268 (1984). (Cited on pages 8; 9; 209; 210; 211; 216; 217 and 223.) S. Nos´e, A unified formulation of the constant temperature molecular dynamics methods. J. Chem. Phys. 81, 511–519 (1984). (Cited on pages 8; 9; 209; 210; 212; 216; 217 and 221.) S. Nos´e, Molecular dynamics simulations at constant temperature and pressure. in: M. Meyer and V. Pontikis, Eds., Computer Simulation in material science, pages 21–41. (Kluwer Academic Publishers, Netherlands, 1991). (Cited on pages 8; 167; 212; 213; 217; 221 and 248.) S. Nos´e, Dynamical behavior of a thermostated isotropic harmonic oscillator. Phys. Rev. E 47, 164–177 (1993). (Cited on pages 210; 212; 217; 221 and 223.) E. Ott, Chaos in Dynamical Systems. (Cambridge University Press, Cambridge, 1993). (Cited on pages 13; 21; 22; 25; 52; 59; 97; 100; 105; 107; 182; 192; 194; 196; 233; 234; 241; 244; 272; 273; 311; 349 and 355.) D. Panja, An elementary proof of Lyapunov exponent pairing for hardsphere systems at constant kinetic energy. J. Stat. Phys. 109, 705–727 (2002). (Cited on page 192.) D. Panja and R. van Zon, Lyapunov exponent pairing for a thermostatted hard-sphere gas under shear in the thermodynamic limit. Phys. Rev. E 65, 060102(R)/1–4 (2002). (Cited on pages 192 and 193.) D. Panja and R. van Zon, Pairing of Lyapunov exponents for a hardsphere gas under shear in the thermodynamic limit. Phys. Rev. E 66, 021101/1–12 (2002). (Cited on pages 192 and 193.) R.K. Pathria, Statistical mechanics, volume 45 of International series

May 8, 2007

4:1

418

[Pea05] [Pei79]

[Pei92] [Pen79] [Per09] [Per92]

[Pet94]

[Pie83]

[Pie88]

[Pom80]

[Poo96]

[Pop98]

[Pop00]

[Pos86]

[Pos87]

[Pos88]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

in natural philosophy. (Pergamon press, Oxford, 1988). (Cited on pages 7; 167; 168; 199 and 206.) K. Pearson, The problem of the random walk. Nature 72, 294, 342 (1905). (Cited on page 18.) R. Peierls, Surprises in theoretical physics. Princeton Series in Physics. (Princeton University Press, Princeton, New Jersey, 1979). (Cited on pages 337 and 338.) H.-O. Peitgen, H. J¨ urgens and D. Saupe, Chaos and Fractals. (Springer-Verlag, Berlin, 1992). (Cited on pages 21 and 25.) O. Penrose, Foundations of statistical mechanics. Rep. Prog. Phys. 42, 1937–2006 (1979). (Cited on pages 1; 7; 8; 236 and 369.) J.-B. Perrin, Mouvement brownien et r´ealit´e mol´eculaire. Ann. Chim. Phys. 19, 5–104 (1909). (Cited on page 348.) S.P.E. Persson and M.K. Megnin, Factors affecting material movement on a corrugated vibrating conveyor (grainpan). Trans. ASAE 35, 395– 400 (1992). (Cited on pages 6 and 160.) J. Petravic, D.J. Isbister and G.P. Morriss, Correlation dimension of the sheared hard-disk Lorentz gas. J. Stat. Phys. 76, 1045–1063 (1994). (Cited on pages 193 and 289.) P. Pieranski, Jumping particle model. Period doubling cascade in an experimental system. J. Physique 44, 573–578 (1983). (Cited on page 154.) P. Pieranski, Direct evidence for the suppression of period doubling in the bouncing-ball model. Phys. Rev. A 37, 1782–1785 (1988). (Cited on page 154.) Y. Pomeau and P. Manneville, Intermittent transition to turbulence in dissipative dynamical systems. Commun. Math. Phys. 74, 189–197 (1980). (Cited on pages 108 and 109.) L. Poon, J. Campos, E. Ott and C. Grebogi, Wada basin boundaries in chaotic scattering. Int. J. Bif. Chaos 6, 251–265 (1996). (Cited on page 143.) M.N. Popescu, Y. Braiman, F. Family and H.G.E. Hentschel, Quenched disorder enhances chaotic diffusion. Phys. Rev. E 58, R4057–R4059 (1998). (Cited on pages 84 and 89.) M.N. Popescu, C.M. Arizmendi, A.L. Salas-Brito and F. Family, Disorder induced diffusive transport in ratchets. Phys. Rev. Lett. 85, 3321– 3324 (2000). (Cited on pages 84 and 89.) H.A. Posch, W.G. Hoover and F.J. Vesely, Canonical dynamics of the Nos´e oscillator: stability, order, and chaos. Phys. Rev. A 33, 4253– 4265 (1986). (Cited on pages 210; 217 and 223.) H.A. Posch and W.G. Hoover, Direct measurement of equilibrium and nonequilibrium Lyapunov spectra. Phys. Lett. A 123, 227–230 (1987). (Cited on pages 190; 193 and 197.) H.A. Posch and W.G. Hoover, Lyapunov instability of dense LennardJones fluids. Phys. Rev. A 38, 473–482 (1988). (Cited on pages 9;

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Bibliography

[Pos89]

[Pos97]

[Pos98]

[Pos00a]

[Pos00b]

[Pos04a]

[Pos04b]

[Pos05]

[Pre92]

[Pre02]

[Pro92]

[Pro00]

[Pro05]

klages˙book

419

10; 166; 190; 191; 192; 193; 197; 209; 210; 213; 219; 223; 227; 237; 282 and 370.) H.A. Posch and W.G. Hoover, Equilibrium and nonequilibrium Lyapunov spectra for dense fluids and solids. Phys. Rev. A 39, 2175–2188 (1989). (Cited on pages 9; 193; 197; 209; 210; 211; 219; 223; 269; 282; 288; 290 and 298.) H.A. Posch and W.G. Hoover, Time-reversible dissipative attractors in three and four phase-space dimensions. Phys. Rev. E 55, 6803–6810 (1997). (Cited on pages 193; 197; 210; 223; 224; 225; 226 and 265.) H.A. Posch and W.G. Hoover, Heat conduction in one-dimensional chains and nonequilibrium Lyapunov spectrum. Phys. Rev. E 58, 4344– 4350 (1998). (Cited on pages 193; 249; 250 and 342.) H.A. Posch and R. Hirschl, Simulation of billiards and of hard body fluids. in: Hard-ball systems and the Lorentz gas, pages 279–314, 2000. see Ref. [Sza00]. (Cited on pages 10 and 335.) H.A. Posch, R. Hirschl and W.G. Hoover, Multifractal phase-space distributions for stationary nonequilibrium systems. in: J. Karkheck, Eds., Dynamics: Models and kinetic methods for non-equilibrium many body systems, volume 371 of NATO Science Series E: Applied Sciences, pages 169–189, Kluwer, Dordrecht, 2000. (Cited on pages 248; 249 and 250.) H.A. Posch, talk at the workshop on “Stability, Fluctuations and Transient Behaviour in Nonequilibrium Systems”, Lyon. unpublished, 2004. (Cited on page 250.) H.A. Posch and W.G. Hoover, Large-system phase-space dimensionality loss in stationary heat flows. Physica D 187, 281–293 (2004). (Cited on pages 219; 299 and 343.) H.A. Posch and C. Forster, Lyapunov instability of fluids. in: G. Radons, W. Just and P. H¨ aussler, Eds., Collective dynamics of nonlinear and disordered systems, pages 301–338, Springer, Berlin, 2005. (Cited on pages 331; 334; 335 and 336.) W.H. Press, B.P. Flanery, S.A. Teukolsky and W.T. Vetterling, Numerical recipes in FORTRAN. (Cambridge University Press, 2nd edition, 1992). (Cited on page 271.) A. Prevost, D.A. Egolf and J. Urbach, Forcing and velocity correlations in a vibrated granular monolayer. Phys. Rev. Lett. 89, 084301/1–4 (2002). (Cited on pages 6; 153; 154 and 378.) T. Prosen and D.K. Campbell, Energy transport and detailed verification of Fourier’s heat law in a chain of colliding harmonic oscillators. J. Phys A: Math. Gen. 25, 3449–3472 (1992). (Cited on page 342.) T. Prosen and D.K. Campbell, Momentum conservation implies anomalous energy transport in 1d classical lattices. Phys. Rev. Lett. 84, 2857–2861 (2000). (Cited on page 341.) T. Prosen and D.K. Campbell, Normal and anomalous heat transport in one-dimensional lattices. Chaos 15, 015117/1–17 (2005). (Cited on pages 6; 337; 339; 340; 341; 342; 343 and 361.)

May 8, 2007

4:1

420

[Rad96]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

G. Radons, Suppression of chaotic diffusion by quenched disorder. Phys. Rev. Lett. 77, 4748–4751 (1996). (Cited on pages 5; 83; 84; 85; 86; 96 and 253.) [Rad97] G. Radons, The thermodynamic formalism of random walks: Relevance for chaotic diffusion and multifractal measures. Phys. Rep 290, 67–79 (1997). (Cited on page 4.) [Rad99] G. Radons, Disorder phenomena in chaotic systems. Adv. Solid State Physics 38, 439–451 (1999). (Cited on pages 5; 83; 84; 85 and 86.) [Rad04] G. Radons, Anomalous transport in disordered dynamical systems. Physica D 187, 3–19 (2004). (Cited on pages 5; 83; 84; 85; 86; 108 and 116.) [Ram86] J.D. Ramshaw, Remarks on entropy and irreversibility in nonHamiltonian systems. Phys. Lett. A 116, 110–114 (1986). (Cited on page 207.) [Ram02] J.D. Ramshaw, Remarks on non-Hamiltonian statistical mechanics. Europhys. Lett. 59, 319–323 (2002). (Cited on page 207.) [Rat00a] K. Rateitschak, R. Klages and WG. Hoover, The Nos´e-Hoover thermostated Lorentz gas. J. Stat. Phys. 101, 61–77 (2000). see also chaodyn/9912018 for further details. (Cited on pages 193; 211; 213; 214; 218; 219; 220; 221; 222; 261; 262; 263; 264 and 282.) [Rat00b] K. Rateitschak, R. Klages and G. Nicolis, Thermostating by deterministic scattering: The periodic Lorentz gas. J. Stat. Phys. 99, 1339–1364 (2000). (Cited on pages 176; 193; 200; 271; 273; 274; 275; 276; 277; 279; 280; 282; 283; 284; 285; 346 and 347.) [Rat02] K. Rateitschak and R. Klages, Lyapunov instability for a periodic Lorentz gas thermostated by deterministic scattering. Phys. Rev. E 65, 036209/1–11 (2002). (Cited on pages 193; 273; 280; 281; 282; 284; 286 and 287.) [Ray05] J.W. Strutt (Lord Rayleigh), The problem of the random walk. Nature 72, 318 (1905). (Cited on page 18.) [RB06] M. Romero-Bastida and J.F. Aguilar, Heat conduction in onedimensional oscillator lattices using Nos´e-Hoover chain thermostats. J. Phys. A: Math. Gen. 39, 11155–11170 (2006). (Cited on pages 213; 310 and 343.) [Rec80] A.B. Rechester and R.B. White, Calculation of turbulent diffusion for the Chirikov-Taylor model. Phys. Rev. Lett. 44, 1586–1589 (1980). (Cited on pages 107 and 377.) [Rec81] A.B. Rechester, M.N. Rosenbluth and R.B. White, Fourier-space paths applied to the calculation of diffusion for the Chirikov-Taylor model. Phys. Rev. A 23, 2664–2672 (1981). (Cited on pages 107 and 377.) [Rei65] F. Reif, Fundamentals of statistical and thermal physics. (McGraw-Hill, Auckland, 1965). see also the amended German translation: Statistische Physik und Theorie der W¨ arme, 3rd edition, de Gruyter, Berlin, 1987. (Cited on pages 1; 7; 13; 18; 165; 167; 168; 191; 199; 206; 207; 275 and 312.) [Rei94] P. Reimann, Suppression of deterministic diffusion by noise. Phys.

May 8, 2007

4:1

Bibliography

World Scientific Book - 9in x 6in

klages˙book

421

Rev. E 50, 727–735 (1994). (Cited on pages 5; 89; 90; 91; 97; 100 and 102.) [Rei96a] P. Reimann, Noisy one-dimensional maps near a crisis. 1. Weak Gaussian white and colored noise. J. Stat. Phys. 82, 1467–1501 (1996). (Cited on pages 5; 89; 90; 91 and 97.) [Rei96b] P. Reimann, Noisy one-dimensional maps near a crisis. 2. General uncorrelated weak noise. J. Stat. Phys. 85, 403–425 (1996). (Cited on pages 5; 89; 90; 91 and 97.) [Rei98] P. Reimann, Comment on ”Toward a statistical thermodynamics of steady states”. Phys. Rev. Lett. 80, 4104 (1998). (Cited on page 207.) [Rei02] P. Reimann, Brownian motors: Noisy transport far from equilibrium. Phys. Rep. 361, 57–265 (2002). (Cited on pages 4; 50; 51; 52; 54 and 378.) [Rei05] J.C. Reid, E.M. Sevick and D.J. Evans, A unified description of two theorems in non-equilibrium statistical mechanics: The fluctuation theorem and the work relation. Europhys. Lett. 72, 726–732 (2005). (Cited on pages 328 and 330.) [Ric81] P.J. Richens and M.V. Berry, Pseudointegrable systems in classical and quantum mechanics. Physica D 2, 495–512 (1981). (Cited on pages 353; 354 and 357.) [Rie67] Z. Rieder, J.L. Lebowitz and E. Lieb, Properties of a harmonic crystal in a stationary nonequilibrium state. J. Math. Phys. 8, 1073–1078 (1967). (Cited on pages 249; 271 and 339.) [Rie00] J.P. Rieu, A. Upadhyaya, J.A. Glazier, N.B. Ouchi and Y. Sawada, Diffusion and deformations of single Hydra cells in cellular aggregates. Biophys. J. 79, 1903–1914 (2000). (Cited on pages 304 and 305.) [Rie05] A. Riegert, N. Baba, K. Gelfert, W. Just and H. Kantz, Hamiltonian chaos acts like a finite energy reservoir: Accuracy of the Fokker-Planck approximation. Phys. Rev. Lett. 94, 054103/1–4 (2005). (Cited on page 172.) [Ris96] D. Risso and P. Cordero, Two-dimensional gas of disks: Thermal conductivity. J. Stat. Phys 82, 1453–1466 (1996). (Cited on page 292.) [Rit03] F. Ritort, Work fluctuations, transient violations of the second law and free-energy recovery methods: Perspectives in theory and experiments. Poincar´e Seminar 2, 195–229 (2003). (Cited on pages 11; 330 and 331.) [Rob92] J.A.G. Roberts and G.R.W. Quispel, Chaos and time-reversal symmetry: Order and chaos in reversible dynamical systems. Phys. Rep. 216, 63–177 (1992). (Cited on page 68.) [Rob95] C. Robinson, Dynamical Systems. (CRC Press, London, 1995). (Cited on pages 21 and 363.) [Ron00a] L. Rondoni and E.G.D. Cohen, Gibbs entropy and irreversible thermodynamics. Nonlinearity 13, 1905–1924 (2000). (Cited on pages 3 and 345.) [Ron00b] L. Rondoni, T. T´el and J. Vollmer, Fluctuation theorems for entropy production in open systems. Phys. Rev. E 61, R4679–R4682 (2000). (Cited on page 233.)

May 8, 2007

4:1

422

World Scientific Book - 9in x 6in

klages˙book

Bibliography

[Ron02a] L. Rondoni, Deterministic thermostats and fluctuation relations. in: Dynamics of dissipation, pages 35–61, 2002. see Ref. [Gar02]. (Cited on pages 7; 8; 11; 13; 188; 210; 218; 235; 264; 265; 323; 327; 328 and 331.) [Ron02b] L. Rondoni and E.G.D. Cohen, On some derivations of irreversible thermodynamics from dynamical systems theory. Physica D 168-169, 341–355 (2002). (Cited on pages 3 and 345.) [Rou04] M. Rouijaa, C.A. Kr¨ ulle, I. Rehberg, R. Grochowski and P. Walzel, Transportverhalten und Strukturbildung granularer Materie auf Schwingf¨ orderern. Chemie Ingenieur Technik 76, 62–65 (2004). (Cited on page 161.) [Rue71] D. Ruelle and F. Takens, On the nature of turbulence. Commun. Math. Phys. 20, 167–192 (1971). (Cited on page 194.) [Rue78] D. Ruelle, Thermodynamic Formalism, volume 5 of Encyclopedia of mathematics and its applications. (Addison-Wesley, Reading, Mass., 1978). (Cited on page 1.) [Rue89] D. Ruelle, The thermodynamic formalism for expanding maps. Commun. Math. Phys. 125, 239–262 (1989). (Cited on page 23.) [Rue96] D. Ruelle, Positivity of entropy production in nonequilibrium statistical mechanics. J. Stat. Phys. 85, 1–23 (1996). (Cited on pages 8; 9; 237; 238; 240; 241; 242 and 245.) [Rue97a] D. Ruelle, Differentiation of SRB states. Commun. Math. Phys. 187, 227–241 (1997). (Cited on page 369.) [Rue97b] D. Ruelle, Entropy production in nonequilibrium statistical mechanics. Commun. Math. Phys. 189, 365–371 (1997). (Cited on pages 9 and 237.) [Rue97c] D. Ruelle, Positivity of entropy production in the presence of a random thermostat. J. Stat. Phys. 86, 935–951 (1997). (Cited on pages 9 and 250.) [Rue99a] D. Ruelle, Gaps and new ideas in our understanding of nonequilibrium. Physica A 263, 540–544 (1999). (Cited on page 7.) [Rue99b] D. Ruelle, Smooth dynamics and new theoretical ideas in nonequilibrium statistical mechanics. J. Stat. Phys. 95, 393–468 (1999). (Cited on pages 1; 7; 10; 13; 100; 192; 211; 236; 237; 238; 322; 368 and 369.) [Rue03] D. Ruelle, Extending the definition of entropy to nonequilibrium steady states. Proc. Natl. Acad. Sci. 100, 3054–3058 (2003). (Cited on page 237.) [Rue04] D. Ruelle, Conversations on nonequilibrium physics with an extraterrestrial. Phys. Today 57, 48–53 (2004). (Cited on pages 237 and 368.) [Rug97] H.H. Rugh, Dynamical approach to temperature. Phys. Rev. Lett. 78, 772–775 (1997). (Cited on page 166.) [Rus05] S. Russ, S. Zschiegner, A. Bunde and J. K¨rger, Lambert diffusion in porous media in the Knudsen regime: Equivalence of self-diffusion and transport diffusion. Phys. Rev. E 72, 030101(R)/1–4 (2005). (Cited on pages 149 and 362.) [Sac98] A.S. Sachrajda, R. Ketzmerick, C. Gould, Y. Feng, P.J. Kelly, A. De-

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Bibliography

[San92]

[San05]

[San06]

[Sar92]

[Sar98]

[Sar00]

[Sch75]

[Sch78]

[Sch82]

[Sch89]

[Sch93]

[Sch94]

[Sch98]

[Sch01]

[Sch02]

klages˙book

423

lage and Z. Wasilewski, Fractal conductance fluctuations in a soft-wall stadium and a Sinai billiard. Phys. Rev. Lett. 80, 1948–1951 (1998). (Cited on page 378.) A. Santos, V. Garz´ o and J.J. Brey, Comparison between the homogeneous-shear and the sliding-boundary methods to produce shear flow. Phys. Rev. A 46, 8018–8020 (1992). (Cited on page 288.) D.P. Sanders, Fine structure of distributions and central limit theorem in diffusive billiards. Phys. Rev. E 71, 016220/1–16 (2005). (Cited on pages 117 and 359.) D.P. Sanders and H. Larralde, Occurrence of normal and anomalous diffusion in polygonal billiard channels. Phys. Rev. E 73, 026205/1–9 (2006). (Cited on pages 6; 246; 358; 359; 360; 363 and 378.) S.S. Sarman, D.J. Evans and G.P. Morriss, Conjugate-pairing rule and thermal-transport coefficients. Phys. Rev. E 45, 2233–2242 (1992). (Cited on pages 192; 193; 289; 290 and 298.) S.S. Sarman, D.J. Evans and P.T. Cummings, Recent developments in non-Newtonian molecular dynamics. Phys. Rep. 305, 1–92 (1998). (Cited on pages 8; 191; 192; 210; 289 and 291.) S. Sarkar, A.V. Anil Kumar and S. Yashonath, Diffusion anomaly from analytical formula. J. Chem. Phys. 112, 965–969 (2000). (Cited on page 150.) H. Scher and E.W. Montroll, Anomalous transit-time dispersion in amorphous solids. Phys. Rev. B 12, 2455–2477 (1975). (Cited on pages 108 and 113.) T. Schneider and E. Stoll, Molecular-dynamics study of a threedimensional one-component model for distortive phase transitions. Phys. Rev. B 17, 1302–1322 (1978). (Cited on pages 8 and 248.) M. Schell, S. Fraser and R. Kapral, Diffusive dynamics in systems with translational symmetry: A one-dimensional-map model. Phys. Rev. A 26, 504–521 (1982). (Cited on pages 19; 26; 83; 100; 102; 105 and 123.) H.G. Schuster, Deterministic chaos. (VCH Verlagsgesellschaft mbH, Weinheim, 2nd edition, 1989). (Cited on pages 3; 13; 59; 109; 182; 194; 223; 233; 272 and 273.) M. Schienbein and H. Gruler, Langevin equation, Fokker-Planck equation and cell migration. Bull. Math. Biol. 55, 585–608 (1993). (Cited on page 304.) F. Schweitzer and L. Schimansky-Geier, Clustering of “active” walkers in a two-component system. Physica A 206, 359–379 (1994). (Cited on page 305.) F. Schweitzer, W. Ebeling and B. Tilch, Complex motion of Brownian particles with energy depots. Phys. Rev. Lett. 80, 5044–5047 (1998). (Cited on pages 305 and 306.) F. Schweitzer, W. Ebeling and B. Tilch, Statistical mechanics of canonical-dissipative systems and applications to swarm dynamics. Phys. Rev. E 64, 021110/1–12 (2001). (Cited on page 309.) A. Sch¨ uring, S.M. Auerbach, S. Fritzsche and R. Haberlandt, On en-

May 8, 2007

4:1

424

[Sch03a]

[Sch03b] [Sch05]

[Sch06]

[Sch07] [Sea97]

[Sea98]

[Sea00a]

[Sea00b] [Sei33] [Ser01]

[Ser03] [Ser05] [SG95]

[SG06]

[Shl85]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

tropic barriers for diffusion in zeolites: A molecular dynamics study. J. Chem. Phys. 24, 10890–10894 (2002). (Cited on page 150.) A. Sch¨ uring, Molekulardynamik-Simulationen und Sprungmodelle zur Diffusion in Zeolithen. PhD thesis, Technische Universit¨ at Leipzig, 2003. (Cited on pages 5; 148 and 150.) F. Schweitzer, Brownian agents and active particles. (Springer, Berlin, 2003). (Cited on pages 305 and 314.) A. Sch¨ uring, S. Vasenkov and S. Fritzsche, Influence of boundaries of nanoporous crystals on molecular exchange under the conditions of single-file diffusion. J. Phys. Chem. B 109, 16711–16717 (2005). (Cited on page 149.) M. Schmiedeberg and H. Stark, Superdiffusion in a honeycomb billiard. Phys. Rev. E 73, 031113/1–9 (2006). (Cited on pages 246; 362 and 363.) A. Sch¨ uring, R. Klages and S. Fritsche, Temperature-dependent diffusion in a zeolithic channel. unpublished, 2007. (Cited on page 149.) D.J. Searles, D.J. Evans and D.J. Isbister, The number dependence of the maximum Lyapunov exponent. Physica A 240, 96–104 (1997). (Cited on page 333.) D.J. Searles, D.J. Evans and D.J. Isbister, The conjugate-pairing rule for non-Hamiltonian systems. Chaos 8, 337–349 (1998). (Cited on pages 10; 192; 193; 227; 282; 289; 290 and 298.) D.J. Searles and D.J. Evans, Ensemble dependence of the transient fluctuation theorem. J. Chem. Phys. 113, 3503–3509 (2000). (Cited on pages 326 and 328.) D.J. Searles and D.J. Evans, The fluctuation theorem and Green-Kubo relations. J. Chem. Phys. 112, 9727–9735 (2000). (Cited on page 327.) W. Seidel, Note on a metrically transitive system. Proc. Nat. Acad. Sci. 19, 453–456 (1933). (Cited on page 232.) A. Sergi and M. Ferrario, Non-Hamiltonian equations of motion with a conserved energy. Phys. Rev. E 64, 056125/1–9 (2001). (Cited on pages 218; 223; 225 and 266.) A. Sergi, Non-Hamiltonian equilibrium statistical mechanics. Phys. Rev. E 67, 021101/1–7 (2003). (Cited on pages 207 and 218.) A. Sergi, Phase space flows for non-Hamiltonian systems with constraints. Phys. Rev. E 72, 031104/1–5 (2005). (Cited on page 218.) L. Schimansky-Geier, M. Mieth, H. Rost and H. Malchow, Structure formation by active Brownian particles. Phys. Lett. A 207, 140–146 (1995). (Cited on page 305.) R. Sagado-Garcia, M.Aldana and G.Martinez-Mekler, Deterministic ratchets, circle maps, and current reversals. Phys. Rev. Lett. 96, 134101/1–4 (2006). (Cited on pages 46 and 53.) M.F. Shlesinger and J. Klafter, Accelerated diffusion in Josephsonjunctions and related chaotic systems - comment. Phys. Rev. Lett. 54, 2551 (1985). (Cited on page 108.)

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Bibliography

[Shl93] [Shl94] [Sin70]

[Sin82]

[Sin91] [Sin00]

[Sma80] [Sok02] [Sta89] [Ste79]

[Ste80] [Ste86]

[Sto91]

[Sto94] [Sto95a] [Sto95b]

[Sto95c]

[Stu99]

[Sza00]

klages˙book

425

M.F. Shlesinger, G.M. Zaslavsky and J. Klafter, Strange kinetics. Nature 363, 31–37 (1993). (Cited on pages 108 and 355.) M.F. Shlesinger, G.M. Zaslavsky and U. Frisch, Eds., Levy flights and related topics. (Springer, Berlin, 1994). (Cited on page 108.) Ya.G. Sinai, Dynamical systems with elastic reflections. Ergodic properties of dispersing billiards. Russ. Math. Surv. 25, 137–189 (1970). (Cited on pages 139; 181 and 183.) Ya.G. Sinai, The limiting behavior of a one-dimensional random walk in a random environment. Theor. Prob. Appl. 27, 256–268 (1982). (Cited on page 84.) Ya.G. Sinai, Eds., Dynamical systems. Number 1 in: Advanced series in nonlinear dynamics. (World Scientific, 1991). (Cited on page 1.) Ya.G. Sinai, Eds., Dynamical systems, ergodic theory and applications. Encyclopaedia of mathematical sciences. (Springer, Berlin, 2nd edition, 2000). (Cited on page 1.) S. Smale, The mathematics of time. (Springer, Berlin, 1980). (Cited on pages 7 and 8.) I.M. Sokolov, J. Klafter and A. Blumen, Fractional kinetics. Phys. Today 55, 48–54 (2002). (Cited on page 108.) J. Stachel, Eds., The collected papers of Albert Einstein, volume 2. (Princeton University Press, Princeton, 1989). (Cited on page 168.) W.-H. Steeb, Generalized Liouville equation, entropy, and dynamic systems containing limit cycles. Physica 95A, 181–190 (1979). (Cited on pages 207 and 237.) W.-H. Steeb, A comment on the generalized Liouville equation. Found. Phys. 10, 485–493 (1980). (Cited on page 207.) W. Steckelmacher, Knudsen flow 75 years on: The current state of the art for flow of rarefied gases in tubes and systems. Rep. Prog. Phys. 49, 1083–1107 (1986). (Cited on page 149.) C.L. Stokes and S.K. Williams D.A. Lauffenburger, Migration of individual microvessel endothelial cells: Stochastic model and parameter measurement. J. Cell Science 99, 419–430 (1991). (Cited on page 304.) R. Stoop, Bivariate free energy and anomalous diffusion. Europhys. Lett. 25, 99–104 (1994). (Cited on page 4.) R. Stoop, The diffusion-related entropy function: The enhanced case. Europhys. Lett. 29, 433–438 (1995). (Cited on page 4.) R. Stoop, Thermodynamic approach to deterministic diffusion of mixed enhanced-dispersive type. Phys. Rev. Lett. 52, 2216–2219 (1995). (Cited on pages 4; 5 and 108.) R. Stoop, W.-H. Steeb and G. Radons, A new characterization of deterministic diffusion: Diffusion-related entropy functions. Phys. Lett. A 202, 195–200 (1995). (Cited on page 4.) A.M. Stuart and J.O. Warren, Analysis and experiments for a computational model of a heat bath. J. Stat. Phys. 97, 687–723 (1999). (Cited on pages 7 and 169.) D. Szasz, Eds., Hard-ball systems and the Lorentz gas, volume 101

May 8, 2007

4:1

426

[Tab05]

[Tak03]

[Tan02a]

[Tan02b]

[Tan02c]

[Tan02d]

[Tan03a]

[Tan03b]

[Tan05a]

[Tan05b]

[Tan06]

[Tas94]

[Tas95]

[Tas98]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

of Encyclopedia of mathematical sciences. (Springer, Berlin, 2000). (Cited on pages 3; 13; 183; 392; 419 and 427.) S. Tabachnikov, Billiards and geometry, volume 30 of Student Mathematical Library. (American Mathematical Society, Providence, 2005). (Cited on pages 353 and 354.) T. Takagi, A simple example of the continuous function without derivative. Proc. Phys. Math. Soc. Japan Ser. II 1, 176–177 (1903). (Cited on page 63.) T. Taniguchi, C.P. Dettmann and G.P. Morriss, Lyapunov spectra of periodic orbits for a many-particle system. J. Stat. Phys. 109, 747–764 (2002). (Cited on page 336.) T. Taniguchi and G.M. Morriss, Master equation approach to the conjugate pairing rule of Lyapunov spectra for many-particle thermostated systems. Phys. Rev. E 66, 066203/1–11 (2002). (Cited on pages 192 and 193.) T. Taniguchi and G.P. Morriss, Stepwise structure of Lyapunov spectra for many-particle systems using a random matrix dynamics. Phys. Rev. E 65, 056202/1–15 (2002). (Cited on pages 10 and 336.) K.I. Tanimoto, T. Kato and K. Nakamura, Phase dynamics in SQUID’s: Anomalous diffusion and irregular energy dependence of diffusion coefficients. Phys. Rev. B 66, 012507/1–4 (2002). (Cited on pages 4; 107 and 378.) T. Taniguchi and G.P. Morriss, Boundary effects in the stepwise structure of the Lyapunov spectra for quasi-one-dimensional systems. Phys. Rev. E 68, 026218/1–18 (2003). (Cited on page 336.) T. Taniguchi and G.P. Morriss, Localized behavior in the Lyapunov vectors for quasi-one-dimensional many-hard-disk systems. Phys. Rev. E 68, 046203/1–22 (2003). (Cited on pages 334 and 336.) T. Taniguchi and G.P. Morriss, Time-dependent mode structure for Lyapunov vectors as a collective movement in quasi-one-dimensional systems. Phys. Rev. E 71, 016218/1–4 (2005). (Cited on pages 10; 336 and 337.) T. Taniguchi and G.P. Morriss, Time-oscillating Lyapunov modes and the momentum autocorrelation function. Phys. Rev. Lett. 94, 154101/1–4 (2005). (Cited on pages 10; 336 and 337.) T. Taniguchi and G.P. Morriss, Dynamics of strongly localized Lyapunov vectors in many-hard-disk systems. Phys. Rev. E 73, 036208/1– 16 (2006). (Cited on pages 334 and 336.) S. Tasaki and P. Gaspard, Fractal distribution and Fick’s law in a reversible chaotic system. in: M. Yamaguti, Eds., Towards the Harnessing of Chaos, pages 273–288. (Elsevier, Amsterdam, 1994). (Cited on pages 59 and 63.) S. Tasaki and P. Gaspard, Fick’s law and fractality of nonequilibrium stationary states in a reversible multibaker map. J. Stat. Phys. 81, 935–987 (1995). (Cited on pages 59; 63; 68 and 233.) S. Tasaki, T. Gilbert and J.R. Dorfman, An analytical construction

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Bibliography

[Tas99]

[Tas00]

[Tas02]

[Tas04]

[Tay21] [T´el96]

[T´el98]

[T´el00]

[T´el01]

[T´el02]

[T´el06] [Ten82]

[Tho83]

[Til99]

[Tod92]

klages˙book

427

of the SRB measures for Baker-type maps. Chaos 8, 424–443 (1998). (Cited on page 63.) S. Tasaki and P. Gaspard, Thermodynamic behavior of an areapreserving multi-baker map. Theoret. Chem. Acc. 102, 385–396 (1999). (Cited on pages 3; 63; 233 and 238.) S. Tasaki and P. Gaspard, Entropy production and transports in a conservative multibaker map with energy. J. Stat. Phys. 101, 125–144 (2000). (Cited on pages 3; 63; 233 and 238.) S. Tasaki and P. Gaspard, Spectral properties of a piecewise linear intermittent map. J. Stat. Phys. 109, 803–820 (2002). (Cited on pages 5 and 108.) S. Tasaki and P. Gaspard, Spectral characterization of anomalous diffusion of a periodic piecewise linear intermittent map. Physica D 187, 51–65 (2004). (Cited on pages 5 and 108.) G.I. Taylor, Diffusion by continuous movements. Proc. London Math Soc. 20, 196–212 (1921). (Cited on page 64.) T. T´el, J. Vollmer and W. Breymann, Transient chaos: The origin of transport in driven systems. Europhys. Lett. 35, 659–664 (1996). (Cited on pages 3; 9; 63; 193; 233; 243 and 245.) T. T´el, P. Gaspard and G. Nicolis, Eds. Chaos and irreversibility, volume 8 of Chaos, American Institute of Physics, College Park, 1998. focus issue. (Cited on pages 9 and 13.) T. T´el and J. Vollmer, Entropy balance, multibaker maps, and the dynamics of the Lorentz gas. in: Hard-ball systems and the Lorentz gas, pages 367–420, 2000. see Ref. [Sza00]. (Cited on pages 3; 57; 63; 186; 193; 200; 233; 238; 239; 243 and 297.) T. T´el, J. Vollmer and L. M´ aty´ as, Shear flow, viscous heating, and entropy balance from dynamical systems. Europhys. Lett. 53, 458–464 (2001). (Cited on page 290.) T. T´el, J. Vollmer and L. M´ aty´ as, Comments on the paper ‘Particles, maps and irreversible thermodynamics’ by E.G.D. Cohen and L.Rondoni, Physica A 306 (2002) 117. Physica A 323, 323–326 (2002). (Cited on pages 3 and 346.) T. T´el and M. Gruiz, Chaotic Dynamics. (Cambridge University Press, Cambridge, 2006). (Cited on pages 13; 59 and 233.) A. Tenenbaum, G. Ciccotti and R. Gallico, Stationary nonequilibrium states by molecular dynamics. Fourier’s law. Phys. Rev. A 25, 2778– 2787 (1982). (Cited on pages 8; 249; 271 and 291.) S. Thomae, Chaos-induced diffusion. in: C. Benedek at al., Eds., Statics and Dynamics of nonlinear systems, pages 204–210, Springer, Berlin, 1983. (Cited on page 21.) B. Tilch, F. Schweitzer and W. Ebeling, Directed motion of Brownian particles with internal energy depot. Physica A 273, 294–314 (1999). (Cited on pages 305; 306 and 308.) M. Toda, R. Kubo and N. Saitˆ o, Statistical Physics, volume 1 of Solid

May 8, 2007

4:1

428

World Scientific Book - 9in x 6in

klages˙book

Bibliography

State Sciences. (Springer, Berlin, 2nd edition, 1992). (Cited on pages 1 and 369.) [Tol38] R.C. Tolman, The principles of statistical mechanics. (Oxford University Press, New York, 1938). reprinted by Dover, New York, 1979. (Cited on page 1.) [Tre04] E.H. Trepagnier, C. Jarzynski, F. Ritort, G.E. Crooks, C.J. Bustamante and J. Liphardt, Experimental test of Hatano and Sasa’s nonequilibrium steady-state equality. Proc. Natl. Acad. Sci. 101, 15038–15041 (2004). (Cited on page 329.) [Tri95] C. Tricot, Curves and fractal dimension. (Springer, Berlin, 1995). (Cited on page 25.) [Tse94] H.C. Tseng, H.J. Chen, P.C.Li, W.Y.Lai, C. H.Chou and H. W.Chen, Some exact results for the diffusion coefficients of maps with pruning cycles. Phys. Lett. A 195, 74–80 (1994). (Cited on page 4.) [Tse99] H.-C. Tseng and H.-J. Chen, The effects of quenched disorder on the chaotic diffusion for simple maps. Int. J. Mod. Phys. 13, 83–95 (1999). (Cited on page 84.) [Tse00] H.-C. Tseng, P.-R. Huang, H.-J. Chen and C.-K. Hu, The diffusion behavior of a simple map with periodic quenched disorder. Physica A 281, 323–336 (2000). (Cited on page 84.) [Tuc97] M.E. Tuckerman, C.J. Mundy and M.L. Klein, Toward a statistical thermodynamics of steady states. Phys. Rev. Lett. 78, 2042–2045 (1997). (Cited on page 207.) [Tuc98a] M.E. Tuckerman, C.J. Mundy, S. Balasubramanian and M.L. Klein, Response to “Comment on ‘Modified nonequilibrium molecular dynamics for fluid flows with energy conservation’ ”. J. Chem. Phys. 108, 4353–4354 (1998). (Cited on page 207.) [Tuc98b] M.E. Tuckerman, C.J. Mundy and M.L. Klein, Reply to comments on ”Toward a statistical thermodynamics of steady states”. Phys. Rev. Lett. 80, 4105–4106 (1998). (Cited on page 207.) [Tuc00] M.E. Tuckerman and G.J. Martyna, Understanding modern molecular dynamics: Techniques and applications. J. Phys. Chem. B 104, 159– 178 (2000). (Cited on pages 7; 8; 167; 207; 210; 224 and 265.) [Tuf92] N.B. Tufillaro, T. Abbott and J. Reilly, An experimental approach to nonlinear dynamics and chaos. (Add. Wesley Publ., New York, 1992). (Cited on pages 154 and 156.) [Uff01] J. Uffink, Bluff your way in the Second Law of thermodynamics. Stud. Hist. Philos. M. P. 32B, 305–394 (2001). (Cited on page 14.) [Upa01] A. Upadhyaya, J.P. Rieu, J.A. Glazier and Y. Sawada, Anomalous diffusion and non-Gaussian velocity distributions of Hydra cells in cellular aggregates. Physica A 293, 549–558 (2001). (Cited on pages 304 and 305.) [Van92] W.N. Vance, Unstable periodic orbits and transport properties of nonequilibrium steady states. Phys. Rev. Lett. 69, 1356–1359 (1992). (Cited on pages 9; 186; 191; 193; 198 and 245.)

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Bibliography

[vB95]

[vB96]

[vB97]

[vB98]

[vB00a] [vB00b]

[vB04] [vB05]

[Veg93]

[Vel05]

[Vis03a]

[Vis03b]

[Viv06]

[vK71] [vK92]

[Vol97]

klages˙book

429

H. van Beijeren and J.R. Dorfman, Lyapunov exponents and KS entropy for the Lorentz gas at low densities. Phys. Rev. Lett. 74, 4412– 4415 (1995). (Cited on pages 10; 84; 183 and 252.) H. van Beijeren, J.R. Dorfman, E.G.D. Cohen, H.A. Posch and C. Dellago, Lyapunov exponents from kinetic theory for a dilute, field-driven Lorentz gas. Phys. Rev. Lett. 77, 1974–1977 (1996). (Cited on pages 192 and 252.) H. van Beijeren, J.R. Dorfman, H.A. Posch and C. Dellago, Kolmogorov-Sinai entropy for dilute gases in equilibrium. Phys. Rev. E 56, 5272–5277 (1997). (Cited on pages 10; 183 and 252.) H. van Beijeren, A. Latz and J.R. Dorfman, Chaotic properties of dilute two- and three-dimensional random Lorentz gases: Equilibrium systems. Phys. Rev. E 57, 4077–4094 (1998). (Cited on pages 10; 84; 183; 192 and 252.) H. van Beijeren and J.R. Dorfman, On thermostats and entropy production. Physica A 279, 21–29 (2000). (Cited on page 266.) H. van Beijeren, A. Latz and J.R. Dorfman, Chaotic properties of dilute two- and three-dimensional random Lorentz gases. II. Open systems. Phys. Rev. E 63, 016312/1–14 (2000). (Cited on pages 192 and 252.) H. van Beijeren, Generalized dynamical entropies in weakly chaotic systems. Physica D 193, 90–95 (2004). (Cited on pages 355 and 363.) H. van Beijeren and O. M¨ ulken, Thermodynamic formalism for the Lorentz gas with open boundaries in d dimensions. Phys. Rev. E 71, 046203/1–9 (2005). (Cited on pages 183 and 252.) J.L. Vega, T. Uzer and J. Ford, Chaotic billiards with neutral boundaries. Phys. Rev. E 48, 3414–3420 (1993). (Cited on pages 353; 354 and 355.) M.G. Velarde, W. Ebeling and A.P. Chetverikov, On the possibility of electric conduction mediated by dissipative solitons. Int. J. Bifurcation Chaos 15, 245–251 (2005). (Cited on page 314.) S. Viscardy and P. Gaspard, Viscosity in molecular dynamics with periodic boundary conditions. Phys. Rev. E 68, 041204/1–10 (2003). (Cited on pages 246; 290 and 377.) S. Viscardy and P. Gaspard, Viscosity in the escape-rate formalism. Phys. Rev. E 68, 041205/1–10 (2003). (Cited on pages 246; 290 and 377.) F. Vivaldi and J.H. Lowenstein, Arithmetical properties of a family of irrational piecewise rotations. J. Phys. A: Math. Gen. 19, 1069–1097 (2006). (Cited on pages 354 and 362.) N. van Kampen, The case against linear response theory. Physica Norvegica 5, 279–284 (1971). (Cited on page 201.) N. van Kampen, Stochastic processes in physics and chemistry. (North Holland, Amsterdam, 1992). (Cited on pages 21; 22; 31; 167; 174 and 177.) J. Vollmer, T. T´el and W. Breymann, Equivalence of irreversible en-

May 8, 2007

4:1

430

World Scientific Book - 9in x 6in

klages˙book

Bibliography

tropy production in driven systems: An elementary chaotic map approach. Phys. Rev. Lett. 79, 2759–2762 (1997). (Cited on pages 3; 63; 190; 233; 238; 243 and 297.) [Vol98] J. Vollmer, T. T´el and W. Breymann, Entropy balance in the presence of drift and diffusion currents: An elementary chaotic map approach. Phys. Rev. E 58, 1672–1684 (1998). (Cited on pages 3; 63; 233; 238; 243 and 297.) [Vol00] J. Vollmer, T. T´el and L. M´ aty´ as, Modeling thermostating, entropy currents, and cross effects by dynamical systems. J. Stat. Phys. 101, 79–105 (2000). (Cited on pages 3; 63 and 233.) [Vol02] J. Vollmer, Chaos, spatial extension, transport, and non-equilibrium thermodynamics. Phys. Rep. 372, 131–267 (2002). (Cited on pages 3; 63; 68; 69; 186; 193; 200; 233; 238; 239; 243; 297 and 346.) [Vol04] J. Vollmer, T. T´el and W. Breymann, Dynamical-system models of transport: Chaos characteristics, the macroscopic limit, and irreversibility. Physica D 187, 108–127 (2004). (Cited on pages 3; 63; 190; 233; 238 and 243.) [vZ98] R. van Zon, R. van Beijeren and C. Dellago, Largest Lyapunov Exponent for many particle systems at low densities. Phys. Rev. Lett. 80, 2035–2038 (1998). (Cited on page 334.) [vZ02] R. van Zon and R. van Beijeren, Front propagation techniques to calculate the largest Lyapunov exponent of dilute hard disk gases. J. Stat. Phys. 109, 641–669 (2002). (Cited on page 334.) [vZ03a] R. van Zon and E.G.D. Cohen, Extension of the fluctuation theorem. Phys. Rev. Lett. 91, 110601/1–4 (2003). (Cited on pages 326 and 328.) [vZ03b] R. van Zon and E.G.D. Cohen, Stationary and transient workfluctuation theorems for a dragged Brownian particle. Phys. Rev. E 67, 046102/1–10 (2003). (Cited on pages 326 and 328.) [vZ04] R. van Zon and E.G.D. Cohen, Extended heat-fluctuation theorems for a system with deterministic and stochastic forces. Phys. Rev. E 69, 056121/1–14 (2004). (Cited on pages 326 and 328.) [Wac99] R. Wackerbauer, When noise decreases deterministic diffusion. Phys. Rev. E 59, 2872–2879 (1999). (Cited on pages 5; 90 and 91.) [Wag99] C. Wagner, R. Klages and G. Nicolis, Thermostating by deterministic scattering: Heat and shear flow. Phys. Rev. E 60, 1401–1411 (1999). (Cited on pages 227; 249; 250; 271; 273; 277; 282; 288; 291; 292; 293; 295 and 296.) [Wag00] C. Wagner, Lyapunov instability for a hard-disk fluid in equilibrium and nonequilibrium thermostated by deterministic scattering. J. Stat. Phys. 98, 723–742 (2000). (Cited on pages 193; 227; 273; 288; 297; 298; 299 and 343.) [Wan66] G.H. Wannier, Statistical Physics. (Dover, New York, 1966). (Cited on page 191.) [Wan89a] X.J. Wang, Abnormal fluctuations and thermodynamic phase transitions In dynamical systems. Phys. Rev. A 39, 3214–3217 (1989).

May 8, 2007

4:1

Bibliography

World Scientific Book - 9in x 6in

klages˙book

431

(Cited on pages 5; 108 and 109.) [Wan89b] X.J. Wang, Statistical physics of temporal intermittency. Phys. Rev. A 40, 6647–6661 (1989). (Cited on pages 5; 108 and 109.) [Wan93] X.-J. Wang and Ch.-K. Hu, Anomalous diffusion in dynamical systems: Transport coefficients of all order. Phys. Rev. E 48, 728–733 (1993). (Cited on pages 5; 108 and 109.) [Wan02] G.M. Wang, E.M. Sevick, E. Mittag, D.J. Searles and D.J. Evans, Experimental demonstration of violations of the second law of thermodynamics for small systems and short time scales. Phys. Rev. Lett. 89, 050601/1–4 (2002). (Cited on pages 11; 326; 328 and 329.) [Wan04a] J.-S. Wang and B. Li, Intriguing heat conduction of a chain with transverse motions. Phys. Rev. Lett. 92, 074302/1–4 (2004). (Cited on page 342.) [Wan04b] J.-S. Wang and B. Li, Mode-coupling theory and molecular dynamics simulation for heat conduction in a chain with transverse motions. Phys. Rev. E 70, 021204/1–15 (2004). (Cited on page 342.) [War95] S. Warr and J.M. Huntley, Energy input and scaling laws for a single particle vibrating in one dimension. Phys. Rev. E 52, 5596–5601 (1995). (Cited on page 159.) [Wax54] N. Wax, Selected papers on noise and stochastic processes. (Dover, New York, 1954). (Cited on pages 7; 19; 21; 167; 168 and 177.) [Wei86] G.H. Weiss, Overview of theoretical models for reaction rates. J. Stat. Phys. 42, 3–36 (1986). (Cited on page 59.) [Wei91] D. Weiss, M.L. Roukes, A. Menschig, P. Grambow, K. von Klitzing and G. Weimann, Electron pinball and commensurate orbits in a periodic array of scatterers. Phys. Rev. Lett. 66, 2790–2793 (1991). (Cited on pages 5; 97; 134; 181 and 378.) [Wei94] G.H. Weiss, Aspects and applications of the random walk. (NorthHolland, Amsterdam, 1994). (Cited on pages 18; 83; 86 and 111.) ¨ [Wei95] D. Weiss and K. Richter, Antidot-Ubergitter: Flippern mit Elektronen. Physikalische Bl¨ atter 51, 171–176 (1995). (Cited on pages 5; 134; 181 and 378.) [Wei97] D. Weiss, G. Lutjering and K. Richter, Chaotic electron motion in macroscopic and mesoscopic antidot lattices. Chaos, Solitons and Fractals 8, 1337–1357 (1997). (Cited on pages 5; 134; 181 and 378.) [Wei00] S. Weiss, D. Koelle, J. M¨ uller, R. Gross and K. Barthel, Ratchet effect in dc SQUIDs. Europhys. Lett. 51, 499–505 (2000). (Cited on pages 4; 53; 107 and 378.) [Whi98] R.B. White, S. Benkadda, S. Kassibrakis and G.M. Zaslavsky, Near threshold anomalous transport in the standard map. Chaos 8, 757–767 (1998). (Cited on pages 360 and 378.) [Wie01] J. Wiersig and K.-H. Ahn, Devil’s staircase in the magnetoresistance of a periodic array of scatterers. Phys. Rev. Lett. 87, 026803/1–4 (2001). (Cited on pages 5; 134 and 378.) [Wie03] J. Wiersig, Hexagonal dielectric resonators and microcrystal lasers. Phys. Rev. A 67, 023807/1–12 (2003). (Cited on page 362.)

May 8, 2007

4:1

432

[Wig92]

[Wil04]

[Wil06]

[Win92]

[Win95]

[Woj98]

[Woj02]

[Woj03]

[Woj04]

[Wol85]

[Yan05] [Yan06]

[Yor85]

[Yos06]

[You02]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

S. Wiggins, Chaotic transport in dynamical systems, volume 2 of Interdisciplinary Applied Mathematics. (Springer, New York, 1992). (Cited on page 27.) S.R. Williams, D.J. Searles and D.J. Evans, Independence of the transient fluctuation theorem to thermostatting details. Phys. Rev. E 70, 066113/1–6 (2004). (Cited on pages 224 and 259.) S.R. Williams, D.J. Searles and D.J. Evans, Numerical study of the steady state fluctuation relations far from equilibrium. J. Chem. Phys. 124, 194102/1–9 (2006). (Cited on page 323.) R.G. Winkler, Extended-phase-space isothermal molecular dynamics: Canonical harmonic oscillator. Phys. Rev. A 45, 2250–2255 (1992). (Cited on pages 223; 225 and 266.) R.G. Winkler, V. Kraus and P. Reineker, Time reversible and phasespace conserving molecular dynamics at constant temperature. J. Chem. Phys. 102, 9018–9025 (1995). (Cited on pages 223; 225 and 266.) M.P. Wojtkowski and C. Liverani, Conformally symplectic dynamics and symmetry of the Lyapunov spectrum. Commun. Math. Phys. 194, 47–60 (1998). (Cited on page 192.) D.K. Wojcik and J.R. Dorfman, Quantum multibaker maps: Extreme quantum regime. Phys. Rev. E 66, 36110/1–16 (2002). (Cited on pages 233 and 378.) D.K. Wojcik and J.R. Dorfman, Diffusive-ballistic crossover in 1D quantum walks. Phys. Rev. Lett. 90, 230602/1–4 (2003). (Cited on pages 233 and 378.) D.K. Wojcik and J. R. Dorfman, Crossover from diffusive to ballistic transport in periodic quantum maps. Physica D 187, 223–243 (2004). (Cited on page 378.) A. Wolf, J.B. Swift, H.L. Swinney and J.A. Vastano, Determinig the Ljapunov exponents from a time series. Physica D 16, 285–317 (1985). (Cited on page 192.) H.L. Yang and G. Radons, Lyapunov instabilities of Lennard-Jones fluids. Phys. Rev. E 71, 036211/1–15 (2005). (Cited on page 336.) H.L. Yang and G. Radons, Universal features of hydrodynamic Lyapunov modes in extended systems with continuous symmetries. Phys. Rev. Lett. 96, 074101/1–4 (2006). (Cited on page 337.) J.A. Yorke, C. Grebogi, E. Ott and L. Tedeschini-Lalli, Scaling behavior of windows in dissipative dynamical systems. Phys. Rev. Lett. 54, 1095–1098 (1985). (Cited on page 106.) M. Yoshida, S. Miyazaki and H. Fujisaka, Irregular parameter dependence of generalized diffusion coefficients based on large deviation statistical analysis. Phys. Rev. E 74, 026204/1–11 (2006). (Cited on pages 4; 25 and 39.) L.-S. Young, What are SRB measures, and which dynamical systems have them? J. Stat. Phys. 108, 733–754 (2002). (Cited on pages 1 and 238.)

May 8, 2007

4:1

Bibliography

[Zam04]

World Scientific Book - 9in x 6in

klages˙book

433

F. Zamponi, G. Ruocco and L. Angelani, Fluctuations of entropy production in the isokinetic ensemble. J. Stat. Phys. 115, 1655–1668 (2004). (Cited on pages 322 and 324.) [Zas97] G. M. Zaslavsky, M. Edelman and B. A. Niyazov, Self-similarity, renormalization, and phase space nonuniformity of Hamiltonian chaotic dynamics. Chaos 7, 159–181 (1997). (Cited on page 361.) [Zas00] G.M. Zaslavsky and M. Edelman, Hierarchical structures in the phase space and fractional kinetics: I. Classical systems. Chaos 10, 135–146 (2000). (Cited on page 360.) [Zas01a] G.M. Zaslavsky and M. Edelman, Weak mixing and anomalous kinetics along filamented surfaces. Chaos 11, 295–305 (2001). (Cited on pages 353; 358 and 361.) [Zas01b] G.M. Zaslavsky and D.A. Usikov, Weak chaos and quasi-regular patterns. Cambridge Nonlinear Science Series. (Cambridge University Press, Cambridge, 2001). (Cited on page 355.) [Zas02] G.M. Zaslavsky, Chaos, fractional kinetics, and anomalous transport. Phys. Rep. 371, 461–580 (2002). (Cited on pages 108; 353; 356; 358; 360; 361 and 363.) [Zas03] G.M. Zaslavsky and M. Edelman, Pseudochaos. in: E. Kaplan and J.E. Marsden, Eds., Perspective and problems in nonlinear science, pages 421–443, Springer, New York, 2003. (Cited on pages 6; 320; 348; 353; 356; 358 and 361.) [Zas05] G.M. Zaslavsky, B.A. Carreras, V.E. Lynch, L. Garcia and M. Edelman, Topological instability along invariant surfaces and pseudochaotic transport. Phys. Rev. E 72, 026227/1–11 (2005). (Cited on pages 353; 354; 356; 358 and 362.) [Zha92] K. Zhang and K. Zhang, Mechanical models of Maxwell’s demon with noninvariant phase volume. Phys. Rev. A 46, 4598–4605 (1992). (Cited on pages 265 and 290.) [Zha00] F. Zhang, D.J. Isbister and D.J. Evans, Nonequilibrium molecular dynamics simulations of heat flow in one-dimensional lattices. Phys. Rev. E 61, 3541–3546 (2000). (Cited on page 344.) [Zha01] F. Zhang, D.J. Isbister and D.J. Evans, Multiple nonequilibrium steady states for one-dimensional heat flows. Phys. Rev. E 64, 021102/1–5 (2001). (Cited on page 344.) [Zum93a] G. Zumofen and J. Klafter, Power spectra and random walks in intermittent chaotic systems. Physica D 69, 436–446 (1993). (Cited on pages 108 and 116.) [Zum93b] G. Zumofen and J. Klafter, Scale-invariant motion in intermittent chaotic systems. Phys. Rev. E 47, 851–863 (1993). (Cited on pages 108; 109; 114; 115; 117 and 378.) [Zum95] G. Zumofen and J. Klafter, Laminar-localized-phase coexistence in dynamical systems. Phys. Rev. E 51, 1818–1821 (1995). (Cited on pages 108 and 378.) [Zum99] G. Zumofen and J. Klafter, Comment on “Self-similarity and transport in the standard map”. Phys. Rev. E 59, 3756–3760 (1999). (Cited on

May 8, 2007

4:1

434

[Zwa73] [Zwa82] [Zwa83]

[Zwa01] [Zwe98]

World Scientific Book - 9in x 6in

klages˙book

Bibliography

page 361.) R. Zwanzig, Nonlinear generalized Langevin equations. J. Stat. Phys. 9, 215–220 (1973). (Cited on pages 7 and 169.) R. Zwanzig, Non-Markoffian diffusion in a one-dimensional disordered lattice. J. Stat. Phys. 28, 127–133 (1982). (Cited on page 86.) R. Zwanzig, From classical dynamics to continuous time random walks. J. Stat. Phys. 30, 255–262 (1983). (Cited on pages 353; 354; 357; 358 and 363.) R. Zwanzig, Nonequilibrium statistical mechanics. (Oxford University Press, Oxford, 2001). (Cited on pages 7; 8; 168; 169 and 171.) R. Zweim¨ uller, Ergodic structure and invariant densities of nonMarkovian interval maps with indifferent fixed points. Nonlinearity 11, 1263–1276 (1998). (Cited on pages 108 and 116.)

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Index

active Brownian particle, see Brownian motion, active aging, 108, 115, 116 annealed disorder, see noise anomalous diffusion coefficient, see diffusion coefficient, generalized transport, 4, 378 Anosov system, 13, 100, 183, 234, 242, 244, 322–324, 349, 369 antidot lattice, 5, 97, 134, 181 Arnold tongue, 44–45 attractor, 193 fractal, 9, 156, 193–196, 218–220, 238, 240, 247–248, 263, 283–284, 290, 291, 297, 299, 375, 380 smooth, 249–250, 283–284 Axiom A system, 183

crisis, 196 diagram, 100, 196, 219–222, 248, 263, 284 billiard channel, 149, 357, 362 Birkhoff coordinates, 57, 132, 194, 200, 218, 263 bouncing ball, 154, 160 billiard, 6, 154, 378 box counting dimension, see fractal dimension, box counting Brownian motion, 6, 21, 50, 167–171, 177, 348, 351 active, 303, 305–308 and cell migration, 304 Brownian motor, see ratchet canonical-dissipative system, 308 cell, biological, 53, 304 chaos deterministic, 21, 55, 63, 90, 185, 307, 350, 355–356, 363, 365, 368, 370 Lyapunov, 6, 350, 355, 366, 371 microscopic, 2, 3, 6, 55, 202, 320, 342, 344–345, 348–352, 363, 365, 367–371 molecular, 201 topological, 356 topology-induced, 356, 359, 366 weak, 355 chaotic hypothesis, 10, 100, 322, 369, 371

baker map, 232–234, 273, 276, 283, 323 barrier billiard, 353 channel, 357, 358 Bernoulli, 123, 233 process, 44 shift, 67, 232, 234 bifurcation, 4, 311 and deterministic diffusion, 99–107 and periodic windows, 105 435

klages˙book

May 8, 2007

4:1

World Scientific Book - 9in x 6in

436

scattering, 143 climbing sine map, 97, 100, 105, 107 complexity, 4, 8, 39, 97, 102, 156, 196, 210, 340 conductivity electrical, 191, 197–200, 221, 239–240, 250–253, 259, 285, 375, 376 fractal, 199–200 heat, 248, 291, 338, 359, 362, 370 conjugate pairing rule, see Lyapunov, conjugate pairing rule continuous time random walk, see random walk, continuous time correlated cell model, 182 correlation factor, 150 function, 69, 116, 123, 147, 168, 169, 337, 357 coupled rotor model, 341 Crooks relation, 330–331 current, 31, 166, 191 parameter dependence, 42 reversal, 4, 46, 48, 52, 54, 161 cycle expansion, 4, 37, 90, 108, 199 de Rham function, 62, 127 generalized, 104, 116 diffusion, 2 stochastic, 253 anomalous, 107 coefficient, 19, 23, 31, 37, 59, 66, 86, 91, 104, 121, 128, 169 parameter dependence, 22 and escape rate, 23, 242 and hydrodynamic modes, 243 deterministic, 4 existence, 337, 347 fractal, 25, 39, 60 generalized, 109, 114 random walk approximation, 123, 129, 142, 158 reactive, 72 time-dependent, 102, 124 deterministic, 19, 21, 90, 319

klages˙book

Index

equation, 23, 241, 243 fractional, 117 Knudsen, 149 stochastic, 17, 86, 90, 96 surface, 89 ding-a-ling model, 249, 340, 342, 346 ding-dong model, 342 disorder, 83, 84–89 dichotomous, 84, 86, 88 uniform, 84, 86 drift-diffusion, 28–32, 40–54, 200 drunken sailor’s problem, or drunkard’s walk, 17–18 dyadic multibaker map, see multibaker map Einstein formula, 19, 128 relation, 168, 199 electrical conductivity, see conductivity, electrical ensemble and noise average, 92 canonical, see velocity distribution, canonical constant pressure, 167 stress, 167 temperature, see thermostat disorder average, 85 equilibrium, 226 equivalence in nonequilibrium, 243, 252 Gibbs, 1, 205, 368 initial condition, 19, 32, 213 microcanonical, see velocity distribution, microcanonical nonequilibrium, 1, 178, 328, 369 entropy (, τ ) per unit time, 348, 349 fluctuations, 320, 321 Gibbs, 236, 242 Kolmogorov-Sinai, 2, 23, 233, 241, 244, 287, 298, 349, 357, 370 nonequilibrium, 1, 379 production, 258, 330

May 8, 2007

4:1

World Scientific Book - 9in x 6in

Index

and Second Law, 9, 226, 237, 321 Clausius, 190, 293, 294 coarse grained, 3, 238, 321, 370 Gibbs, 190, 207, 236, 242 nonequilibrium, 3, 10, 63, 189–190, 235–239, 281, 289, 290, 293, 294, 321, 325, 370 thermodynamic, 9, 237, 246, 259, 262, 294 equilibrium, local thermodynamic, see local thermodynamic equilibrium equipartitioning of energy, 165, 166, 174, 176, 189, 209, 211, 262, 275, 282, 291, 340 ergodic, 9, 79, 156, 178, 183, 196, 213, 223, 344, 347, 353, 357 hypothesis, 7, 368, 369 theorem, multiplicative, 332 ergostat, 210 escape rate, 23, 242, 246, 362 formalism, 23, 241–242, 243, 246, 290, 370 formula, 242 Evans-Searles fluctuation theorem, see fluctuation theorem, Evans-Searles exponential coupling ansatz, 216 Fermi-Pasta-Ulam model, 310, 333, 340, 341–344 fluctuation dissipation theorem, 168, 169, 171, 211, 304, 306, 312, 322 relation, 10, 320–331 Crooks, see Crooks relation Jarzynski, see Jarzynski work relation theorem, 10, 320–328, 330, 370, 379 Evans-Searles, 327–328 Gallavotti-Cohen, 321–327 Fourier’s law, 338 fractal, 28, 106, 195, 244

klages˙book

437

attractor, see attractor, fractal diffusion coefficient, see diffusion coefficient, fractal dimension, 2, 25, 219, 244 box counting, 25 Hausdorff, 195, 243, 247 function, 4, 44 transport coefficient, see transport coefficient, fractal Frenkel-Kontorova model, 342 friction, 7, 212, 294, 305 bulk, 168 coefficient, 52 field-dependent, 258 non-Markovian, 170 Rayleigh, 307 Stokes, 7, 168, 186, 187, 198 velocity-dependent, 187, 208, 305–307, 314 Frobenius-Perron equation, 22, 23, 37, 65, 102, 234 operator, 31, 32, 60–61, 62, 65 Gallavotti-Cohen fluctuation theorem, see fluctuation theorem, Gallavotti-Cohen Galton board, 3, 180, 201 Gaussian thermostat, see thermostat, Gaussian Golosov phenomenon, 84, 85, 253 Green-Kubo formula for diffusion, see Taylor-Green-Kubo formula hard disk fluid, 287, 291 sphere fluid, 289, 335 Hausdorff dimension, see fractal dimension, Hausdorff heat conductivity, see conductivity, heat Helfand moments, 7, 246, 290 hydrodynamic mode, 2, 10, 62, 243, 247, 336, 370 hyperbolicity, 58, 71, 99, 100, 139, 183, 234, 323, 354, 369 uniform, 272

May 8, 2007

4:1

World Scientific Book - 9in x 6in

438

infinite horizon, 140, 182, 352, 360 intermittency, 5, 102, 109, 118, 355 invariant density, 48, 67, 101–103, 124, 125 measure, 67, 100, 139, 238, 240, 250 absolutely continuous, 249 conditionally, 242 infinite, 108 singular, 63 irreversibility, 3, 63, 165, 189, 193, 247 Jarzynski work relation, 328–331 Josephson junction, 4, 107 K-system, 183, 233, 234 Kac lemma, 360 Kaplan-Yorke formula, 218, 244, 249, 284, 297, 299 Kolmogorov-Sinai entropy, see entropy, Kolmogorov-Sinai Lambert’s cosine law, 58 Langevin equation, 7, 167–173, 177, 185, 186, 199, 211, 235, 248, 304, 326 deterministic, 90, 210, 265 generalized, 305, 306 overdamped, 21, 29, 50, 326 time-discrete, 86 large deviation principle, 322, 326 lift of degree one, 21 Liouville equation, 2, 207, 243 dissipative, 205–208, 236, 327, 328 time-discrete, see Frobenius-Perron equation local thermodynamic equilibrium, 63, 190, 292, 295, 338, 345 Lorentz gas, 3, 5, 10, 57, 179–183, 337, 345 channel, 274, 345, 346 driven, 10, 185–189, 190–193, 200, 208, 235, 240, 249, 251, 257, 259, 279, 323, 375 flower-shaped, 138

klages˙book

Index

periodic, 128–134, 137, 139, 376 random, 83, 86, 252 reactive-diffusive, 56, 81 viscous, 186 Lyapunov chaos, see chaos, Lyapunov conjugate pairing rule, 191–193, 260, 263, 286, 290, 298, 333 exponent, 2, 9, 10, 20, 23, 41, 59, 83, 183, 186, 190, 249, 284–287, 320, 331, 332, 348, 349, 351, 355, 370, 371 and phase space contraction, 191 instability, 10, 20, 55, 333, 342, 369 mode, 10, 331–334, 335 spectrum, 290, 298, 331, 332, 334 sum rule, 191, 215, 260, 263, 282, 289, 297–298, 343 and escape rate formula, 240–247 vector, 332, 333, 335 Manneville map, see Pomeau-Manneville map Markov partition, 23–25, 37–39, 47, 65, 111 process, 56, 142, 143, 144, 177, 327, 330 metric entropy, see entropy, Kolmogorov-Sinai mixing, 79, 183, 223, 233, 234, 272, 357 weak, 353, 357 mobility, 191 Montroll-Weiss equation, 113 multibaker map, 3, 56–60, 62, 63, 243, 244, 290, 297 dispersion relation of modes, 72 parameter-dependent, 67 multifractal, 194, 197, 218, 247 nanopores, see porous material negative current, see current reversal noise, 90, 91–97, 170, 266, 307

May 8, 2007

4:1

World Scientific Book - 9in x 6in

439

Index

and chaos, 352 dichotomous, 91, 92, 94 L´evy, 310 uniform, 91, 94 white, 168, 306 Gaussian, 51 noncanonical transformation, 217 nonequilibrium steady state, 8, 11–13, 63, 156, 165, 166, 167, 178, 231, 238, 247, 320, 340, 367, 369, 373, 374 nonhyperbolic, 100, 107, 108 Nos´e-Hoover oscillator, 223 thermostat, see thermostat, Nos´e-Hoover Ohm’s law, 10, 97, 181, 199, 200 Ornstein-Uhlenbeck process, 177 Oseledec theorem, see ergodic theorem, multiplicative pendulum driven nonlinear, 4, 53, 100, 107 torsion, 329, 331 periodic orbit theory, see cycle expansion Perron-Frobenius theorem, 25, 36 Pesin theorem, 241, 242, 243, 287, 298, 351, 352 phase locking, 4, 6, 154, 161 space contraction, 9, 175, 189, 208, 214, 235–239, 246, 250, 258, 262, 277, 280, 289, 292–297, 322, 324–327 Φ4 model, 342 plug phase, 289 Poincar´e Birkhoff map, see Birkhoff coordinates recurrence time, 360 surface of section, 193, 195, 197, 218 polygonal billiard, 6, 149, 352–363, 370, 378

klages˙book

Pomeau-Manneville map, 109, 116 porous material, 108, 148–150 projection, 8, 67, 68, 174, 175–177, 212, 218, 232, 234, 313, 333 pruning, 4 pseudochaos, 348, 356, 361, 362, 371 pseudohyperbolicity, 356 pseudointegrability, 356, 362 quenched disorder, see disorder random walk, 17, 27, 86–88, 116, 141, 150 continuous time, 108, 111–114, 115, 116, 357, 363 correlated, 89, 103, 121–134, 141, 157 deterministic, 3, 18 in random environments, 84 ratchet, 49–52, 53 deterministic, 52–53, 84, 89 granular, 154, 161 inertia, 52 time-discrete, 53 rational billiard, 353, 356, 357, 359 and genus, 354 reaction rate, 59, 60, 72, 74, 75–81 reaction-diffusion, 2, 54–60, 69–81 repeller, 23, 192, 242, 245 response linear, 8, 47, 200, 210, 291, 293, 338, 369, 374 van Kampen’s objections, 200–202, 251 negative, 4, 46 nonlinear, 4, 47, 97, 199, 222, 250–253, 259, 285, 375 rotating disk, 6, 274 channel, 346–347 rotation number, 31 shear flow, 7, 193, 288–291, 292–299, 320, 334, 336 viscosity, 182, 191, 246, 289, 293, 337

May 8, 2007

4:1

World Scientific Book - 9in x 6in

440

and Lyapunov sum rule, 297 Sinai billiard, 181 sporadicity, 355 square ring billiard, 352, 353, 357, 358 SRB measure, 1, 3, 13, 238, 321, 322, 369, 371 standard map, 52, 97, 107, 234, 273, 360 stochastic boundary condition, 248–250, 270–271, 277 string phase, 289 subdiffusion, 5, 84, 107, 108–119, 360, 361 superdiffusion, 102, 107, 352, 358–360, 363 Takagi function, 63, 127 generalized, 66, 103 Taylor-Green-Kubo formula, 4 for billiards, 131, 142 expansion, 132–134, 147, 159 for maps, 64–67, 101, 122 anomalous, 116 expansion, 123–128 time-continuous, 168 temperature, 165–167, 176, 189, 209, 211, 281 gradient, 291, 338, 339, 345 nonequilibrium, 166, 282 thermal conductivity, see conductivity, heat thermal reservoir, 7, 8, 12, 165–167, 183, 185, 205, 231, 269, 273, 303, 343, 367, 369 coupling, 169, 212, 223, 274 degrees of freedom, 212, 232, 276 deterministic, 172, 287 dimensionality, 173–179 non-Hamiltonian, 7, 8, 210, 231–235, 374 nonequilibrium, 226–227 response time, 212, 220, 261, 263, 292 stochastic, 7, 167–172, 248–250 temperature, 211, 281, 292–294

klages˙book

Index

thermodynamic formalism, 2, 4, 39, 108 thermostat, 8–10, 166, 210 boundary, 269, 273–279 deterministic, 271, 279–287 stochastic, see stochastic boundary condition bulk, 188, 248, 269, 286 coupling, 208, 216, 258 differential feedback, 212 Gaussian, 8, 185–189 coupling, 257 non-ideal, 257–260 properties, 189–202 integral feedback, 212 isoenergetic, 210, 264, 328 isokinetic, 210 Nos´e-Hoover, 8, 208–213 chain, 224, 265 coupling, 212, 219 generalized, 223, 265–266 non-ideal, 260–264 properties, 213–222 Nos´e-Poincar´e, 225 PUT, 289 SLLOD, 289, 321 twirler, 224 universalities, 229–253, 374–376 time-reversibility, 9, 58, 59, 68, 171, 182, 187, 188, 193, 210, 250, 264, 265, 271, 272, 290, 292, 296, 321, 322, 324, 327, 330 Toda lattice, 314, 342 topological instability, 4, 60, 126, 138, 143 transition matrix, 4, 24, 37–39, 65, 69, 75 transport chaotic, 3, 189–202, 213–215, 242, 261, 279, 291, 319 non-chaotic, 6, 93, 149, 246, 319, 351 pseudochaotic, 6, 320, 347–363 transport coefficient and dynamical systems quantities, 239–247, 370, 375

May 8, 2007

4:1

World Scientific Book - 9in x 6in

441

Index

and Lyapunov exponents, see Lyapunov sum rule fractal, 4, 138, 371–374, 376–379 and experiments, 4, 5, 107, 134, 148, 153, 373, 377 and random perturbations, 5, 251 irregular, 129, 137 parameter dependence, 3 triangle channel, 358, 359 triangular billiard, 352, 357 turnstile dynamics, 26–28, 111, 126 twisted boundary condition, 33 velocity autocorrelation function, see correlation function velocity distribution bimodal, 213, 308–314

klages˙book

canonical, 8, 171, 176, 208, 209, 213, 223, 225, 248 at boundary, 270, 277 microcanonical, 174, 176, 188, 213 velocity rescaling, 187, 265 vibratory conveyor, 6, 160–161 Wada basin, 143 waiting time distribution, 113, 114 Wigner-Seitz cell, 128, 139, 181 wind-tree model, 350, 358, 359 Young’s formula, see Kaplan-Yorke formula zeolite, 149 zigzag channel, 358, 359, 360, 362