Lecture Notes in Physics Editorial Board H. Araki, Kyoto, Japan E. Brézin, Paris, France J. Ehlers, Potsdam, Germany U. Frisch, Nice, France K. Hepp, Ziirich, Switzerland R.L. Jaffe, Cambridge, MA, USA R. Kippenhahn, Gottingen, Germany H. A. Weidenmiiller, Heidelberg, Germany J. Wess, Miinchen, Germany J. Zittartz, Kéln, Germany Managing Editor W. Beighback Assisted by Mrs. Sabine Lehr c/o Springer-Verlag, Physics Editorial Department II Tiergartenstrasse 17, D-69121 Heidelberg, Germany Springer Berlin Heidelberg New York Barcelona Budapest Hong Kong London Milan Paris Santa Clara Singapore TokyoThe Editorial Policy for Proceedings The series Lecture Notes in Physics reports new developments in physical research and teaching - quickly, informally, and ata high level. The proceedings to be considered for publication in thisseries should be limited to only a few areas of research, and these should be closely related to each other. The contributions should be of a high standard and should avoid lengthy redraftings of papers already published or about to be published elsewhere. As a whole, the proceedings should aim for a balanced presentation of the theme of the conference including @ description of the techniques used and enough motivation for a broad readership. It should not be assumed that the published proceedings must reflect the conference in its entirety. (A listing or abstracts of papers presented at the meeting but not included in the proceedings could be added as an appendix.) When applying for publication in the series Lecture Notes in Physics the volume’s editor(s) should submit sufficient material to enable the series editors and their referees to make a fairly accurate evaluation (e.g. complete list of speakers and titles of papers to be presented and abstracts). If, based on this information, the proceedings are (tentatively) accepted, the volume’s editor(s), whose name(s) will appear on the title pages, should select the papers suitable for publication and have them refereed (as fora journal) when appropriate. Asarule discussions will not be accepted. The series editors and Springer-Verlag will normally not interfere with the detailed editing except in fairly obvious cases or on technical matters. Final acceptance is expressed by the series editor in charge, in consultation with Springer-Verlag only after receiving the complete manuscript. It might help to send a copy of the authors’ manuscripts in advance to the editor in charge to discuss possible revisions with him. As a general rule, the series editor will confirm his tentative acceptance if the final manuscript corresponds to the original concept discussed, if the quality of the contribution meets the requirements of the series, and if the final size of the manuscript does not greatly exceed the number of pages originally agreed upon. The manuscript should be forwarded to Springer-Verlag shortly after the meeting, In cases of extreme delay (more than six months afier the conference) the series editors will check once more the timeliness of the papers. Therefore, the volume’s editor(s) should establish strict deadlines, or collect the articles during the conference and have them revised on the spot. Ifa delay is unavoidable, one should encourage the authors to update their contributions if appropriate. The editors of proceedings are strongly advised to inform contributors about these points at an early stage. The final manuscript should contain a table of contents and an informative introduction accessible also to readers not particularly familiar with the topic of the conference. The contributions should be in English. The volume’s editor(s) should check the contributions for the correct use of language. At Springer-Verlag only the prefaces will be checked by a copy-editor for language and style. Grave linguistic or technical shortcomings may lead to the rejection of contributions by the series editors. A conference report should not exceed a total of 500 pages. Keeping the size within this bound should be achieved by a stricter selection of articles and not by imposing an upper limit to the length of the individual papers. Editors receive jointly 30 complimentary copies of their book, They are entitled to purchase further copies of their book at a reduced rate.As a rule no reprints of individual contributions can be supplied, No royalty is paid on Lecture Notes in Physics volumes. ‘Commitment to publish is made by etter of interest rather than by signing a formal contract. Springer-Verlag secures the copyright for each volume. The Production Process ‘The books are hardbound,and the publisher will select quality paper appropriate to the needs of the author(s). Publication time is about ten weeks. More than twenty years of experience guarantee authors the best possible service. To reach the goal of rapid publication at alow price the technique of photographic reproduction from a camera-ready manuscript was chosen. This process shifts the main responsibility for the technical quality considerably from the publisher to the authors. We therefore urge all authors and editors of proceedings to observe very carefully the essentials for the preparation of camera-ready manuscripts, which we will supply on request. This applies especially to the quality of figures and halftones submitted for publication. In addition, it might be useful to look at some of the volumes already published. Asa special service, we offer free of charge IATpX and TX macro packages to format the text according to Springer-Verlag’s quality require- ments, We strongly recommend that you make use of this offer, since the result will be a book of considerably improved technical quality. To avoid mistakes and time-consuming correspondence during the production period the conference editors should request special instructions from the publisher well before the beginning of the conference. Manuscripts not meeting the technical standard of the series will have to be returned for improvement. For further information please contact Springer-Verlag, Physics Editorial Department Il, Tiergartenstrasse 17, D-69121 Heidelberg, GermanyLutz Schimansky-Geier Thorsten Péschel (Eds.) Stochastic Dynamics 6 SpringerEditors Lut, Schimansky-Geier Thorsten Péschel Institut fir Physik Humboldt-Universitat zu Berlin Invalidenstrasse 110 D-1ous Berlin, Germany Cataloging-in-Publication Data applied for. Die Deutsche Bibliothek - CIP-Einheitsaufnahme Stochastic dynamics / Lutz Schimansky-Geier ; Thorsten Péschel (ed.). - Berlin ; Heidelberg ; New York ; Barcelona ; Budapest ; Hong Kong ; London ; Milan ; Paris ; Santa Clara ; Singapore ; Tokyo : Springer, 1997 (Lecture notes in physics ; 484) ISBN 3-540-62893-2 Gb. ISSN 0075-8450 ISBN 3-540-62893-2 Springer-Verlag Berlin Heidelberg New York This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, re-use of illustra- tions, recitation, broadcasting, reproduction on microfilms or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer-Verlag, Violations are liable for prosecution under the German Copyright Law. © Springer-Verlag Berlin Heidelberg 1997 Printed in Germany The use of general descriptive names, registered names, trademarks, etc. in this publica- tion does not imply,even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. ‘Typesetting: Camera-ready by authors/editors Cover design: design &production GmbH, Heidelberg SPIN: 10550691 55/3144-5432 10 - Printed on acid-free paperDedicated to Werner Ebeling on the occasion of his 60th birthdayEditorial Preface Werner Ebeling’s 60th birthday gave the occasion for the editors to invite highly qualified specialists in stochastic dynamics to contribute chapters to a collection of lectures on stochastic dynamics. Most of the authors at- tended the workshop “Stochastic Dynamics in Mesoscopic Physical Systems” in Schmerwitz near Berlin in the past year where the idea to edit this book was born. The aim of this collection is to provide a survey of modern developments in the theory of this field of statistical physics, which developed several im- portant new topics in nonlinear nonequilibrium system in recent years. The authors have been asked to write their lectures at a level suitable for gradu- ate students. Most of the authors follow that request. Hence, we believe that students or lecturers will find rich material for learning or teaching purposes, or simply to joint into the field of a stochastic description of physical systems. On the other hand we claim that specialists will find some new material and results as well. ‘The elegance and the apparent simplicity of the stochastic formulation of physical problems led many lecturers to include stochastics as part of their courses on nonequilibrium statistical mechanics. Stochastic approaches, thereby, are bridging between phenomenological models and statistical me- chanics. They account for both fluctuational macroscopic and embedded mi- croscopic descriptions of physical reality. Some of the highlights of statistical physics have their origin in an earlier solved stochastic models. However, it would be misleading to reduce the importance of stochastic solutions to be precursor for the statistical physics of first principles. Nowa- days, we notice the growing interest in mesoscopic descriptions of nature ~ not limited to physics but also in interdisciplinary research. Hence, the necessity increases to take into account fluctuations in nonlinear models of physical and nonphysical effects. Moreover, as shown in several of the lectures, fluctuations in nonlinear systems may play the key role in nonlinear dynamics. Stochastic methods became an intrinsic part of physics in the time when Brownian motion found its explanation in theories by A. Einstein, M. von Smoluchowski, and P, Langevin. In pioneering articles this “troika” already developed important stochastic tools for the physical description of Brownian motion as a memoryless process. They derived kinetic equations for the prob- ability density and the diffusion coefficient, propounded the famous relation of temperature and dissipative processes, and founded the theory of stochas- tic differential equations. Their fundamental results are the basic ingredients of stochastic approaches in physics and in other fields.vill Later on, the theory of nucleation processes and chemical reactions was the first treatment of nonlinear conservative dynamics and provided the main prerequisite to calculate rates of escape processes. By introducing dissipative nonlinear systems it has been possible to model the dynamics of electric circuits and the intensity of light in optical active fibres. Here, nonequilibrium processes came into play and the fluctuation dissipation relation was no longer justified. Stochastic methods have found a wide field of applications in modern science. Several of them will be discussed in the lectures of this volume. Par- ticular attention is payed to nonlinear dynamics and to systems in situations far from equilibrium. One of the simplest examples for nonlinear behavior of a system is conser- vative bistability or multistability. Stochastic considerations of such systems led to the problem of escaping regions of attractors through energy barriers. More than 60 years after the mathematical formulation of this problem and its physical modelling there is still much interest in the calculation of rates for transitions between attractors of a system and lifetimes of stochastic tra- Jjectories close to an attractor. Those calculations are sometimes first steps towards the solution of more complex stochastic problems. Recently, quan- tum formulations as well as considerations of systems with high dimensional phase space with multiple attractors or fluctuating energy barriers have been of particular interest. Other modern fields of research, such as stochastic resonance and stochastic ratchets, are based on the escape problem too. Many current problems in the theory of stochastic processes originate from the discussion about the influence of multiplicative and colored noise on non- linear dynamical systems. In this time nonlinear dynamics was re-detected by physicists and inspired by the description of living systems. It began to in- fluence the theory of nonequilibrium systems. Noise-induced transitions have been investigated by applying a rich mathematical apparatus were non white and non-Gaussian noisy sources in nonlinear dynamics have been considered. Noise-induced states lead to the formation of spatially distributed structures. Several of the included lectures deal with these phenomena. When stationary nonlinear systems are driven by monofrequency and mul- tifrequency signals, surprisingly one finds that the addition of noise to the input signal or to the nonlinear system, which filters or amplifies the input itself, results in a higher order of the output signal. It has been discussed intensively whether this effect could be of importance for information pro- cessing of early biological species. Since even simple functional devices could be able to make use of noise (instead of energy, which is concentrated in few degrees of freedom only) scientists have speculated whether one could apply noise for technical information processing. Brownian motors and stochastic ratchets make use of noise for break of symmetry too. The directed motion of fluctuating particles in potentials with broken reflection symmetry transforms heat into directed motion. OneK should note that the second law is not violated since detailed balance on the mesoscopic level is broken. Hence, one deals with nonequilibrium processes. ‘This is a main feature of most modern problems: colored noise in nonlinear systems, stochastic resonance and stochastic ratchets. We believe that sta- tistical physics will benefit much from the investigation of these problems at the stochastic level. A highlight of the mentioned conference in Schmerwitz was the session on granular flows. The interest of the editors in this field was also a reason for including here a larger part of lectures on this topic. In granular systems driven by periodic or stochastic forces one observes a variety of possible appli- cations of stochastic methods. We claim that fluctuational phenomena that account for the occurring stress phenomena, which possibly go beyond the hydrodynamic description, will be essential for the understanding of the rich dynamics of granular systeins. One can assume that driven many-particle sys- tems with inelastic impacts of the mesoscopic elements of the systems could be described at the mesoscopic level by using techniques of stochastic dy- namics. The question whether the challenge of such a stochastic description of granular dynamics will be successful or not can be decided by future work. ‘We hope that this collection of lectures will be helpful for students, lectur- ers and scientists in statistical physics and related interdisciplinary research. ‘The editors wish to thank B. Lindner for technical assistance. Dedication ‘The lecturers and the editors dedicated this collection to Prof. Dr. Werner Ebeling on the occasion of his 60th birthday. During the past decades he worked permanently in the field of stochastic description of nature. Several of his interesting and sometimes unusual approaches led to unexpected results and had an important impact in this field. Many of the authors and particularly the editors benefited from fruitful and encouraging discussions with Werner Ebeling. We admire Werner Ebeling not only for his scientific work but also for his outstanding personality and his friendly and warm attitude to colleagues and students. Werner Ebeling, who teaches at the Humboldt-University in Berlin, al- ways considered lecturing and teaching as an important part of his work. Many physicists received their PhD under his supervision and several of his former students are professors in physics, mathematics, and medical sciences. Berlin, December 1996 Lutz Schimansky-Geier and Thorsten PéschelContents Foreword: Probability in Physics Nico G. Van Kampen Part I: Concepts of Stochastics and Kinetics From Stratonovich Calculus to Noise-Induced Phase Transitions Christian Van den Broeck Generalized Langevin Equations : A Useful Tool for the Perplexed Modeller of Nonequilibrium Fluctuations? Peter Hanggi Colored Noise in Dynamical Systems: Some Exact Solutions Peter Jung On the Operator Method of Variable Contraction for Stochastic Processes Jerzy Luczka Recurrence Time Statistics in Low-Dimensional Dynamical Systems Catherine Nicolis, Grégoire Nicolis, V. Balakrishnan, and Michael Theunissen The Boltzmann Equation for the Gas of Partly Inelastic Balls with Regard to Random Interaction Forces Rouslan L. Stratonovich Part II: Activation and Brownian Motors Ratchets Driven by Colored Gaussian Noise Roland Bartussek A Motor Protein Model and How It Relates Stochastic Resonace, Feynman’s Ratchets, and Maxwell’s Demon Martin Bier 42 55 69 81xi A Diffusion-Limited Reaction Martin A. Burschka Stochastic Localization in Soft Anharmonic Oscillators Katja Lindenberg, Lisa Bernstein, and David W. Brown Quantum Transition State Theory for Multidimensional Dissipative Systems Eli Pollak and Gidon Gershinsky Surmounting Fluctuating Barriers: Basic Concepts and Results Peter Reimann and Peter Hanggi Some Problems of Cluster Dynamics of Biological Macromolecules Yuri M. Romanovsky Part III: Synchronization and Stochastic Resonance Stochastic Synchronization Vadim §. Anishchenko and Alexander B. Neiman Stochastic Analysis of Limit Cycle Behavior Florence Baras Symbolic Dynamics Approach to Stochastic Processes Jan A, Freund Stochastic Resonance (for Beginners) Fabio Marchesoni Dynamics of Globally Coupled Noisy Oscillators Arkady Pikovsky Cluster Statistics and Traffic on a Lattice ‘Thorsten Péschel and Jan A. Freund Phase Synchronization in Noisy and Chaotic Oscillators Michael Rosenblum, Arkady Pikovsky, and Jiirgen Kurths A Glauber-Dynamics Approach to Coupled Stochastic Resonators Lutz Schimansky-Geier and Udo Siewert 88 101 116 127 140 155 167 179 193 210 220 232 245Part IV: Distributed Systems Stochastic Aspects of the Force Network in a Regular Granular Piling Eric R. Clément, Christophe Eloy, Jean Rajchenbach, and Jacques Duran Burgers’ Turbulence and Dynamical Scaling Sergei E. Esipov and Timothy J. Newman Quantum Mechanics Simulated by Diffusion and Branching Processes Thomas Fricke Markov Chain Models for Spatially Distributed Reacting Systems Raymond Kapral Macromechanical Modeling of the Surface Flow of Granular Systems Stefan J. Linz Novel Pattern Formation in Granular Matter Hernan A. Makse, Shlomo Havlin, Peter R. King, and H. Eugene Stanley Self-Motion in Physico-Chemical Systems Far from Thermal Equilibrium Alexander §. Mikhailov and Dirk Meinkéhn Synthetic Random Flows: Generation and Applications Francesco Sagués, Jose M. Sancho, and Arturo C. Marti Active Brownian Particles: Artificial Agents in Physics Frank Schweitzer Planetary Rings — Nonequilibrium Systems in Space Frank Spahn xitl 261 273 249 306 319 334List of Participants Vadim S$, Anishchenko wadim@chaos.ssu.runnet.ru Department of Physics, Saratov State University, Astrakhanskaya str. 83, 410071 Saratov, Russia, V.Balakrishnan Department of Physics, Indian Institute of Technology, Madras 600 036, India, Florence Baras fbaras@ulb.ac.be Centre for Nonlinear Phenomena and Complex Systems, Université Libre de Bruxelles, Campus Plaine, C.P. 231,Belgium, Roland Bartussek roland@phibm14c .Physik.Uni-Augsburg.DE Institut fiir Physik, Universitat Augsburg, Memminger StrBe 6, D-86135 Augsburg, Germany, Lisa Bernstein lisa@math.isu.edu Department of Mathematics, Idaho State University, Pocatello, ID 83209- 8085, USA, Martin Bier mbier@surgery .bsd.uchicago.edu Section of Plastic and Reconstructive Surgery, Department of Surgery MC 6035, University of Chicago, 5841 South Maryland Avenue, Chicago IL 60637, USA, Christian Van den Broeck chris@luc.ac.be Dept. WNI, Limburgs Universitair Centrum, Universitaire Campus, B- 3590 Diepenbeek, Belgium, David W. Brown aromero@hypatia.ucsd.edu Institute for Nonlinear Science, University of California, San Diego, La Jolla, CA 92093-0402, USA, Martin Burschka burschka@xerxes.thphy.uni-duesseldorf .de Institut fiir Theoretische Physik III, Heinrich-Heine-Universitat, D-37083 Diisseldorf, Germany, Eric Clément erc@ccr. jussieu.fr Laboratoire d’Acoustique et d’Optique de la Matiére Condensée - Boite 86, Université Pierre et. Marie Curie, 4, Place Jussieu, 75005 Paris, France, Jacques Duran jd@aomc. jussieu.fr Laboratoire d’Acoustique et d’Optique de la Matiére Condensée - Boite 86, Université Pierre et Marie Curie, 4, Place Jussieu, 75005 Paris, France,xvi Christophe Eloy Laboratoire d’Acoustique et d’Optique de la Matiére Condensée - Boite 86, Université Pierre et Marie Curie, 4, Place Jussieu, 75005 Paris, France, Sergei E. Esipov Sergei_Esipov{CENTRE-REQnotes . worldcom.com 30 Newport Parkway 1705, Jersey City, NJ 07310, USA, Jan A. Freund janf@itp02.physik.hu-berlin.de Humboldt-Universitat zu Berlin, Institut fiir Physik, Invalidenstrafe 110, 10115 Berlin, Germany, Thomas Fricke thomas@itp02. physik.hu~berlin.de Humboldt-Universitat zu Berlin, Institut fiir Physik, Invalidenstrafe 110, 10115 Berlin, Germany, Gidon Gershinsky gideon@wiscpa.weizmann.ac.il Chemical Physics Department, Weizmann Institute of Science, 76100, Rehovot, Israel,, Hanggi@Physik.Uni-Augsburg.de t Augsburg, Institut fiir Physik, MemmingerstraBe 6, D-86135 Germany, Shlomo Havlin havlin@phys9.ph.biu.ac.il Center for Polymer Studies and Dept. of Physics, Boston University, Boston, MA 02215 USA, and Department of Physics, Bar-Ilan University, Ramat Gan, Israel, Peter Jung ph287pj@prism.gatech.edu School of Physics, Georgia Institute of ‘Technology, Atlanta, GA 30332, USA, Nico van Kampen L.J.M.Silkens@fys.ruu.nl Inst. voor Theoretische Fysica, Rijksuniversiteit Utrecht, Princetonplein 5, P.B. 80.006, NL-3508 TA Utrecht, Netherlands, Raymond Kapral rkapralégatto.chem.utoronto.ca Chemical Physics Theory Group, Department of Chemistry, University of Toronto, Toronto, Ontario MSS 3H6, Canada, Peter R. King kingpOrcscli.dnet.bp.com BP Exploration Operating Company Ltd., Sunbury-on-Thames, Middx., TWI16 7LN, UK, Jiirgen Kurths juergen@agnid.uni-potsdam.de Max-Planck-Arbeitsgruppe “Nichtlineare Dynamik”, Universitat Pots- dam, Am Neuen Palais 10, Postfach 601553, D-14415 Potsdam, Germany, Katja Lindenberg k1@hypatia.ucsd.edu Department of Chemistry and Biochemistry, University of California, San Diego, La Jolla, CA 92093-0340,, Stefan Linz stefan@phibmi3m.physik.uni-augsburg.de Universitat, Augsburg, Institut fiir Physik, Memminger StraBe 6, D-86135 Augsburg, Germany, Jerzy Luczka luczka@usk. pl Department of Theoretical Physics, Silesian University, PL-40-007 Ka- towice, Poland,xvll Herndn A. Makse hmakse@miranda.bu.edu Center for Polymer Studies and Dept. of Physics, Boston University, Boston, MA 02215, USA, Fabio Marchesoni. marchesoni@perugia. infn. it Department of Physics, University of Illinois, Urbana, IL 61801, USA, and Instituto Nazionale di Fisica Nucleare, VIRGO Project, I-06100 Perugia, Italy, Arturo C. Martf arturo@ecm.ub.es Departament d’Estructura i Constituents de la Matéria, Universitat de Barcelona, Av. Diagonal 647, E-08028 Barcelona, Spain,, Dirk Meinkéhn dirk.meinkoehn@dlr.de Deutsche Forschungsanstalt fiir Luft- und Raumfahrt, D-74239 Hardt- hausen 3, Germany, Alexander S. Mikhailov mikhailovOfhi-berlin.mpg.de Fritz-Haber—Institut der Max~Planck~Gesellschaft, Faradayweg 4-6, D- 14195 Berlin, Germany, Aleksander Neiman neiman@chaos.ssu.runnet.ru Department of Physics, Saratov State University, Astrakhanskaya 83, Saratov 410071, Russia, Timothy J. Newman tn@thp.uni-koeln.de Department of Physics, University of Manchester, Manchester, M13 9PL, UK, Catherine Nicolis enicolis@oma.be Institut Royal Météorologique de Belgique, 1180 Brussels, Belgium, Grégoire Nicolis gnicolis@ulb.ac.be Center for Nonlinear Phenomena and Complex Systems, Université Libre de Bruxelles, C.P. 231, 1050 Brussels, Belgium, Arkady Pikovsky pikovsky@agnld.uni-potsdam.de Max-Planck-Arbeitsgruppe “Nichtlineare Dynamik”, Universitat Pots- dam, Am Neuen Palais 10, Postfach 601553, D-14415 Potsdam, Germany, Eli Pollak cfpollak@weizmann.weizmann.ac.il Chemical Physics Department, Weizmann Institute of Science, 76100, Rehovot, Israel,, ‘Thorsten Péschel thorsten@itp02.physik.hu-berlin.de Humboldt-Universitat zu Berlin, Institut fiir Physik, InvalidenstraBe 110, 10115 Berlin, Germany, Jean Rajchenbach jer@ccr. jussieu.fr Laboratoire d’Acoustique et d’Optique de la Matiére Condensée - Boite 86, Université Pierre et Marie Curie, 4, Place Jussieu, 75005 Paris, France, Peter Reimann preimann@phibmi3m. physik.uni-augsburg.de Universitat Augsburg, Institut fir Physik, Memminger StraBe 6, D-86135. Augsburg, Germany,XVill Yuri M. Romanovskii romanov@lbp.phys.msu.su Department of Physics, Lomonosov Moscow State University, Vorob’evy Gory, 119899 Moscow, Russia, Michael Rosenblum mros@agnld.uni-potsdam.de Max-Planck-Arbeitsgruppe “Nichtlineare Dynamik”, Universitat Pots- dam, Am Neuen Palais 10, Postfach 601553, 14415 Potsdam, Germany, Francesc Sagués sagues@daphne.qf.ub.es Departament de Quimica Fisica, Universitat de Barcelona, Av. Diagonal 647, E-08028 Barcelona, Spain, Jose M. Sancho jmsancho@ecm.ub.es Departament d’Estructura i Constituents de la Matéria, Universitat de Barcelona, Av. Diagonal 647, E-08028 Barcelona, Spain, Lutz Schimansky-Geier alsg@itp02.physik.hu-berlin.de Humboldt-Universitat 2u Berlin, Institut fiir Physik, Invalidenstra8e 110, 10115 Berlin, Germany, Frank Schweitzer frank@itp02.physik.hu-berlin.de Humboldt-Universitat zu Berlin, Institut fiir Physik, InvalidenstraBe 110, 10115 Berlin, Germany, Udo Siewert udos@itp02. physik.hu-berlin.de Humboldt-Universitat zu Berlin, Institut fiir Physik, Invalidenstrafe 110, 10115 Berlin, Germany, Frank Spahn frank@agnld.uni-potsdam.de Max-Planck-Arbeitsgruppe “Nichtlineare Dynamik”, Universitat Pots- dam, Am Neuen Palais 10, Postfach 601553, D-14415 Potsdam, Germany, H. Eugene Stanley hes@buphyk.bu.edu Center for Polymer Studies and Dept. of Physics, Boston University, Boston, MA 02215 USA, Ruslan L. Stratonovich r1s@comsim. phys .msu.su Physics Department, Lomonosov Moscow State University, Vorob’evy Gory, 119889 Moscow, Russia, Michael ‘Theunissen mtheunis@is1.ulb.ac.be Institut Royal Météorologique de Belgique, 1180 Brussels, Belgium and, Center for Nonlinear Phenomena and Complex Systems, Université Libre de Bruxelles, C.P. 231, 1050 Brussels, BelgiumProbability in Physics Nico G. van Kampen Inst. voor Theoretische Fysica, Rijksuniversiteit Utrecht, Princetonplein 5, P.B. 80.006, NL-3508 ‘TA Utrecht, Netherlands ‘The concept of probability has an inauspicious origin in gambling and the insurance business Laplace developed it into a mature mathematical di: pline. Unfortunately at the same time he introduced the germ of a hereditary disease, which has plagued the field even since, namely the confusion between objective and subjective probability. Objective probability deals with the fre- quency of occurrence of an object or events. Its assertions can therefore be confronted with reality. Subjective probability is a degree of belief and cannot be verified or falsified. Numerous disparate and desperate attempts to mould it into a mathematical framework have merely produced an extensive litera- ture (Keynes 1921). It is clear that this kind of probability has no bearing on physics. It is amazing that there is still a lively subculture based on the idea that the behaviour of physical systems is governed by the degree of belief of the observer (Jaynes 1989). Objective probability is the subject of probability calculus. However, all that this calculus can do is transforming one distribution of probabilities another. Hence it can be applied only if an underlying probability distribution is given a priori. (The science of statistics is an attempt to get away from this basis fact.) When two dice are thrown I can compute the probabilities of the total number of points, provided I know a priori that each die has an equal probability for each of sit faces. That this is actually true is a consequence of the mechanical motion of the die while falling; during that motion each face is up one sixth of the time. This is the ergodic theorem applied to the dynamics of a cube. ‘Thus the indispensable a priori probability is determined by the physics of the system are is dealing with - not by fairness or lack of informa- tion (Tolman 1938, Quarrie 1973, Balian 1982). Such anthropomorphic ideas are a result of taking Gibbs axiomatic treatment of sti cal mechanics too seriously, without the antidote of the more physical treatment of Boltzmann and the profound analysis of the Ehrenfests. An entirely new application of probability in physics entered through Ein- steins theory of Brownian motion. In particular the alternative formulation by Langevin is very popular. He described the influence of the surrounding fluid on the particle by means of a simple damping term, plus a term repre- senting a force that varies rapidly and unpredictably in time. This led to the idea that in any system, of which one knows that noise occurs, this noise can be described by merely adding to the deterministic macroscopic equations of motion a “random force”. To provide a mathematical description of a “rapidly and unpredictably i2 Nico G. van Kampen varying” force one nowadays replaces it with a stochastic function, i.e., a function taken from an ensemble of functions endowed with a probability. Of course there is a logical jump from an actual force (which is in principle uniquely defined) to such a mathematical object, borrowed from the calculus of probability. (This jump is often concealed by the loose definition of the word “random”). Moreover, one is now also faced with the task of determin- ing the probability distribution in the ensemble. It is usually chosen to be Gaussian because that is traditional, convenient in calculations, and some- times correct. In addition one stipulates that the average is zero and that the autocorrelation function is a delta peak. The si 5 the requirement that the stationary solution coincides with the one known from equilibrium statistical mechanics. Note, that the delta peak serves as an idealization of a short-lived correlation; a physi who pears this in mind will not be led astray by mathematical vagaries like the calculus. In nonlinear systems, however, this device led to a difficulty because the average of the equation obtained in this way no longer coincides with the macroscopic equation that served as starting point. Also the equation does not have the known equilibrium as a solution. ‘This difficulty gave rise to many discussions (MacDonald 1954, van Kampen 1965). The upshot is that the correct choice must not be guessed, but must be found by studying the physical mechanism that causes the noise. There now exist two schools of research, ‘The first school tries to derive the correct from of the noise term on the basis of a more detailed description of its physical origin. In some cases that is no problem. For instance, once one knows the random force for the Brownian particle it is dear that the noise will not be affected when the particle is subjected in an external force, linear or not, since the origin of the noise is in the fluid. Also the noise produced by a linear resistor remains the same when s placed in a nonlinear electric circuit, but the noise produced in anon Ohmic resistance cannot be found so easily. In other cases, in which the noise is internal, caused by the discrete nature of the constituents (shot noise, chemical reactions, population problems) a more detailed descriptions provided by the master equation, which describes the single collisions between two molecules, or single birth and death events. In cases where the noise is external, such as the Brownian particle, one introduces the Langevin force by attaching to the system considered, say 3, a large second system B, called bath, reservoir, or environment. Then S will perturb B (which otherwise is in equilibrium) and this perturbation works back on S. In this way it gives rise to both the friction and the fluctuations. It thus appears that all one has to do is to solve (at least formally) the equations of motion of the combined system $+ B and from the result project out the motion of S alone. One can not expect, however, to get a differential equation for the evolution of S by itself. In fact, one gets an integral equation involving the entire past back to the initial time at which the interaction was switchedProbability in Physics 3 on. This is the “generalized Langevin equation” of Zwanzig. It is an exact consequence of the basic equations of motion, but in order to turn it into a differential equation of the type of the actual Langevin equation one has to make the drastic “Markov assumption”. This consists in replacing the past values occurring in the integral by the instantaneous present values. All conceptual difficulties (including the breaking of the time symmetry) are hidden in this assumption. On the other hand, without it, one is stuck with an exact equation, which not only is unsolvable but also involves the initial time. This is unrealistic, it is hard to believe that the initial time is relevant for the Brownian motion observed in scrapings from the Sphinx. Actually the whole idea of a Langevin equation makes sense only to lowest (i.e. second) order of the coupling, because in higher orders the separation between the systems S and B is no longer well-defined. the prescription for deriving the Langevin equation is therefore the following. In the integral over the past express the variables at the earlier times in their values at the present time using the unperturbed macroscopic equation of $. The provides the friction term (and a modification of the potential). The remaining term, which involves the initial values of the both variables is the Langevin force. Its probability distribution is provided by the equilibrium distribution of the bath. As the precise nature of the bath ought not to matter one customarily chooses collection of harmonic oscillators. This prescription can be carried out explicitly. What is more, it can be done in quantum mechanics as well (Ullersma 1966). It then yields the quantum Langevin equation, about which there has been so much discussion in the literature. ‘The second school does not. worry about the origin of the noise but takes the Ganssian form for granted and aims to solve equations involving this Langevin force. The idea is that the outcome will not depend sensitively on the particular choice of the noise and that therefore their results will be at least qualitatively correct. In this way one can study numerous problems such as first-passage times, escape over a boundary, stochastic resonance, and ratched motors. However, it goes too far to replace typical shot noise by a Gaussian (Kadar et al. 1986). In a number of systems (think of chemical reactions) there is a macro- scopic deterministic equation that has a bifurcation, or some other point of instability. In the vicinity of such a point the fluctuations about the macro- scopic behaviour can grow almost unchecked (critical opalescence) and these fluctuations determine which of the several macroscopically possible branches the system will happen to choose. ‘The term “fluctuation induced phase tran- sition”, however, is misleading, because the phase transition is a property of the macroscopic equation; the fluctuations de not displace the point of the phase transition, they only make it fuzzy. This description of the state of affairs leads us to the following conclusion. ‘There is a very rich field of applications of stochastic methods in physics. Many of them are this subject of the contributions in the present volume.4 Nico G. van Kampen Many others have appeared from the pen of Werner Ebeling. Yet the field is not yet wide enough for his mind. He has also studied the evolution of biological systems, animal motion, self-organization ion complex systems, the discussion of entropy and information, and the entropy of language. We hope that for many more years his publications will be appearing to teach us about, these fascinating subject. References R. Balian, Du microscopique au macroscopique (Ecole Polytechnique, Palaiseau 1982) E. T. Jaynes, Papers on Probability, Statistics, and Statistical Physics (R. d. Rosenkranz, ed. Kluwer, Dordrecht 1989) N. G. van Kampen, in: Fluctuation Phenomena in Solids (R. E. Burgess ed., Acad. Press, New York 1965) M. Kadar, G. Parisi, and Y.-C. Zhang, Phys. Rev. Letters 56, 889(1986) J. M. Keynes, A Treatise on Probability (Macmillan, London 1921); H. Jeffreys, Science Inference (Cambridge University Press 1931) D. K. C. MacDonald, Phil. mag. 45, 63, 345 (1954) D. A. Mc Quarrie, Statistical ‘Thermodynamics (University Science Books, Mill Valley, California 1973) R. C. Tolman, The Principles of Statistical Mechanics (Clarendon, Oxford 1938) P. Ullersma, Physica 32, 27, 56, 74, 90 (1966)Concepts of Stochastics and KineticsFrom Stratonovich Calculus to Noise-Induced Phase Transitions C. Van den Broeck L.U.C. B-3590 Diepenbeek, Belgium Abstract. We illustrate the Stratonovich interpretation for a stochastic differential equation on the basis of simple examples. The short and long-time behavior of such equations are contrasted with each other. We formulate a mean field theory for spatially distributed models with multiplicative noise, and present a simple model that displays a noise-induced phase transition. 1 Introduction: Ito versus Stratonovich ‘The phenomenon which we will describe relies on rather basic but peculiar properties of Langevin equations or stochastic differential equations with mul- tiplicative noise (Gardiner 1983, Van Kampen 1992). We therefore choose, as a matter of introduction, to explain on the basis of examples the origin of the so-called Stratonovich interpretation of a stochastic differential equation. Consider the following evolution equation for a scalar variable x: £= g(z)&(t) (1) where €(t) is a complicated time-dependent function which represents the realization of an external noise. One typically expects that the time scale on which the noise relaxes is much smaller than the one describing the evolution of z. It is therefore tempting to approximate é(t) as being a “white noise”, that is uncorrelated at different times. This simplification however leads to an interpretation problem of Eq. (1). We first illustrate this by assuming that &(t) is a sum of delta functions, centered at random time points (also called white shot noise): a(t) = ttt) @) At the time points ¢ = ¢;, the variable x will jump from one value to another. However the amplitude of the jump, g(z), depends on x. Which z-value should one then take: the value before the jump, after the jump, or something else ? The Stratonovich interpretation corresponds to taking the average of the value prior to and after the jump. The argument in favour of this rule is that one can show that it is the correct interpretation if €(1) is in fact the limit of a shot noise with very small correlation time. A related feature is that the normal rules of calculus apply in the Stratonovich interpretation (since8& C. Van den Broeck they apply for real noise). For example, it is quite allright in the Stratonovich interpretation to divide in Eq. (1) both sides by g(z) to obtain an equation with additive noise (and hence no interpretation problem). Mathematicians however prefer to work with the Ito interpretation, where the value of x is taken before the jump. This interpretation has mathematical advantages (use of the martingale formalism), and is also desirable if one would like to per- form simulations. One of the prices to pay however is that the normal rules of calculus are not valid and are replaced by the unfamiliar Ito calculus, see e.g. Gardiner 1983, White shot noise is not the most common type of white noise. Indeed, based on the central limit theorem, one often can argue that the noise will possess Gaussian properties. Gaussian white noise then - even though conceptually a very awkward object - turns out to play a much more central role. Fortu- nately a powerful mathematical formalism can be constructed based on the fact that stochastic differential equations with Gaussian white noise can be replaced by an equivalent description in terms of Fokker Planck equations. To illustrate how such an equation arises and, at the same time, to demon- strate that it is obtained in its Stratonovich form, we consider for (t) a non-white noise, namely the Markovian dichotomic process, which converges to Gaussian white noise in a limit specified below (Van den Broeck 1983). The Markovian dichotomic pro is defined as follows: €(t) can take on the val- ues +A or —A. In any time interval of length dt there is a probability k dt for a jump between these states. We now define P(x, +,t)dz and P(x, —, t)dz as being the probability to be at time t, in the interval [z, e-+dz] with €(t) = +A and €(t) = —A respectively. In the preceding time interval dé, two things can have happened: either € did not change its state (probability 1 — kdt) and hence the system was be in the interval [x +(x) Adt, «+ dx ¥ g(z +dzx) Adt] at time t — dt, or it did change (probability Adt) and the system was in the interval [x + O(dt), 2 + dz + O(dt)]. Hence we can write: P(e, +,t)dx = P(x — (2) Adt, +t — dt) [dz — g(x + da) Adt + to(e) Adt| (1 — kit) + P(x + O(dt), 1 dtdx kat (3) P(x, —,t)dx = P(e + 9(2)Adt, -t — dd) [de + g(x + d2)Adt — —9(2)Adt} (1 — kdl) + P(x + O(a), +,t~ di)de kdt (4) Collecting the terms in dtdz, one obtains: OP (2, +,t) = —LtroleyaPte. 401+ k[P(x,—.t)- P(x, +0) (6) 2 alayAP(e,—0)+ MP +.9- Pl -0) (6) AP(e, I) =FFrom Stratonovich Calculus to Noise-Induced Phase Transitions 9 For the quantities: P(x,t) = P(x, +,t) + P(x,-,t) P(t) = P(2, 4,2) — P(e, 1) iw one easily obtains the following equations: a OP(2,t) = —F-[Ag(z)p(e, t)] ; @) ale.) = — 2 [9(e) P(e, 1)] — 2kpl2,) p(2,t) can be eliminated leading to the following exact equation for P: oO a (8+ 2m). P(e,t) = 2 fAg(2) 2 [Ag(e)Pl=,2)) ®) ‘The final step is to consider the Gaussian white-noise limit of the dichotomic process, which at the same time provides a more intuitive picture of what this noise looks like. We take the following limit: ko} i al with za” fixed (10) Note that A/& corresponds to the average surface under a pulse of the di- chotomic process. In the above limit such a pulse becomes short (k -> 00) and high (A -+ 00), while however the surface goes to zero (as 1/k). By combining Eqs. (9) and (10), we finally obtain the Fokker Planck equation, equivalent to Eq. (1), in Stratonovich interpretation: 2a a aped=F2 {oad ola)P(e.n}} an) Following an entirely similar but more lengthy calculation, one can show that the Stratonovich Fokker Planck equation, equivalent to: L(x) + 9(x)é (12) with € a Gaussian white noise of intensity 0, is given by: é aPent)=-Ziepeo+ SZ {eg werreol} (1 2 Ox The first term in the r.h.s. is called the drift term, the second the diffusion term. The Fokker Planck equivalent to Eq. (12) in the Ito representation reads: a oP oy Plat) =~ FUE)Pt+ Fos (Plt) 4)10 C. Van den Broeck Note that the Stratonovich form can be rewritten in this way, but with f(x) replaced by f(x) -+ 2-9(x)g'(x). The term $-9(z)g'(2) is sometimes called the spurious drift term because it will appear in the equation for the first moment , see below. From now on, we focus on the Stratonovich interpretation, ef. Eq. (13). We start, by investigating the short-time properties of Eq. (13). Multipli- cation with z and subsequent integration over x yields the following exact equation for < x >: 2 =< fle) > +> <9(e)g'(e) > (15) Note the additional term arising from the multiplicative nature of the noise (it is zero for g(x) = constant ). As usual in noisy nonlinear problems, the first moment < x > is coupled to higher order moments. Suppose however that one follows the behavior for very short times for a distribution starting as P(«,t = 0) = 6(2—ao). One expects that one can neglect the fluctuations of x around < 2 > and write: flen>)+ Sa 2>9i(<2>) (16) Of particular interest, to us is to find out how the multiplicative noise affects the stability of a reference state. As a typical test case, we focus on f(x) = —2+ higher order, and g(x) = 1+2?+ higher order. One finds from Eq. (16) o = -(1F 5) + higher order (17) 2 One concludes that a multiplicative noise of the form g(z) = 1 — x? has the effect to stabilize the reference state, while g(z) = 1+ 2? has an opposite effect. In words, the variable x has on average the tendency to drift towards the regions of higher noise intensity. A similar result is found for g(z’ (note that the noise intensity is in fact /g?(x) = |2| in this case) We now turn to the long-time behaviour. The steady state solution of Eq. (13) is easily found to be: 2 oe je ~ Faluds' wv) era 0 ry) dy (18) and N is a normalization constant. Of particular interest here are the extrema = of the probability. They obey the following equation: = s@) - Fazio! @) 0 (19)From Stratonovich Calculus to Noise-Induced Phase Transitions uu Note the difference in sign appearing in Eqs. (16) and (19). As a result, the effect of multiplicative noise on the stability of a reference state is opposite for short and long times. The results (18) and (19) have been discussed at length in the context of noise induced transitions (Horsthemke 1984). However, in spatially distributed and coupled systems, the long-time behavior turns out to be no longer the relevant factor, at least not in the limit of a sufficiently strong spatial coupling, which is needed to generate cooperative effects between the elements constituting the system. In fact, as we will illustrate below, it is the short time behavior that can influence dramatically the macroscopic behavior of the spatially distributed system. This probably explains why the generic noise induced phase transitions were never discovered in the context of noise induced transitions. 2 An exactly solvable model Consider the following particular form of the stochastic differential equation, ef. Eq. (12): b=-rte€ (20) interpreted in the Stratonovich sense (Graham 1982). By dividing through x, this equation can be rewritten as follows: O,(Inz +t) =€ (21) and the corresponding Fokker Planck equation for the variable y = Inz +t reads: &P( (22) ‘This equation can be solved exactly (it is identical to the plain diffusion equation). We conclude that _ Weettctn za)? xv 2no%t is the exact time dependent. probability starting from the initial condition P(z,t = 0) = 6(e— 29). In other words, » has a log-normal distribution with = Ingo —t. Furthermore, for t ~+ 00, one finds P(x, t) + 0 for every finite value of x. We conclude that Jim P(2,t) = 6(2), in other words © = 0 is an absorbing 00 state. This result is in agreement with the formal solution for the steady state probability, cf. Eq. (18) : P(z,t) (23) eT P"(c)va a? (24)12 C. Van den Broeck which has a non-normalizable divergence for x +0. Let us now turn to the behavior of the first moment. From Eq. (15), one finds: <@>=-(l-=)<#> (25) 4 hence: see 7/2) < (t= 0) > (26) ‘This result is exact, and can also be obtained directly from Eq. (23). Note however that < x(t) > diverges for ¢ -+ 00, whenever a? > 2. This result seems to contradict the fact that the probability distribution converges to 5(z). The reason is that the average, being an integral over z of the probability multiplied by x, is dominated by large but unlikely deviations. Alternatively, one can trace back this divergence to the occurrence of an opposite sign in the contribution of the multiplicative noise term in the equations for < x > and P**(x) respectively. In the present context, the observed dichotomy in behavior may seem to be a mathematical peculiarity. When considering spatially distributed and coupled systems however, the behavior of the first. moment explains the existence of noise induced phase transitions. 3 Mean field description for spatially distributed systems with multiplicative noise We would like to formulate a spatially distributed version of Eq. (12). One of the simplest possibilities is to consider a lattice model, with on each lattice point i a stochastic dynamics of the type given in Eq. (12), and to couple these elements by a diffusive or harmonic coupling: f= sen tale -2 Yo (e- 2) (27) jen(i) where n(i) is the set of the z sites neighbouring to i, D the strength of the spatial coupling, and €; independent Gaussian white noises with intensity o?. In general, one can not hope to find a solution of this set of equations, not even at the stationary state. For this reason, we focus here on a simple mean field approximation (Van den Broeck 1994), which consists in replacing the values x; at neighbouring sites by an average value < x >, which has to be determined self-consistently. With this ansatz, the evolution for different lattice sites decouple, and one can write the following mean field equation: & = f(z) + g(x)f - D(z— <2 >) (28)From Stratonovich Calculus to Noise-Induced Phase Transitions 13 For simplicity, we have dropped the subscript i. The steady state solution P*'(2) is easily found to be (cf. Eqs. (12) and (18)): 9(y)9' (y) - Diy- < = >) P* (2) = Nexp ss a dy — (29) g Fi 9° (uy) while the mean field value < z > is the solution of the following nonlinear equation: [oo [Peete =F(<2>) (30) In the following we will restrict ourselves to the simple example introduced in the previous section. For this particular case, Eys. (29) and (30) reduce to: ae acre (31) and D <@>=F(<2>) =< 2 > ——— (32) 14D- = One concludes that < x >= 0 is the unique solution for a? < 2, while diverging solutions < x >= +00 arise for o? > 2. The surprising conclusion is that the instability of the first moment, observed in the scalar model for ¢? > 2, gives rise to a genuine phase transition (from a quiescent state z; = 0 to an exploding state 2; > oo) for the spatially distributed version of the system, at least in the mean field approximation. Simulations of a 2-dimensional system. with nearest neighbour coupling confirm this result, while the threshold value of o? = 2 is also in agreement with an expansion in 1/D (Becker 1994). It is expected that the addition of nonlinear terms in Eq. (27) will temper” the transition leading to first. (Miiller) or second order phase transitions (Van den Broeck 1994, Van den Broeck 1994, Van den Broeck 1996, Ojalvo). 4 Conclusion ‘The model Eq. (20) was introduced here for the sake of simplicity. It serves as an introduction to more complicated cases studied in Van den Broeck 1994, Van den Broeck 1994, Van den Broeck 1996, Ojalvo, see also Schimansky- Geier 1991, Garcia-Ojalvo 1993, Becker 1994, Grinstein 1996, Li 1996, Miiller. The conclusion is that the so-called spurious drift term that arises from the Stratonovich interpretation has an undeniable and important impact on the physical properties of spatially distributed systems: it can shift existing phase- transitions and bifurcation points, or even generate new ones. We expect that, this new phenomenon will also arise in more complicated situations involving spatial (Park), temporal or spatio-temporal structures.14 C. Van den Broeck Acknowledgements We thank the Program on Inter-University Attraction Poles, Prime Minister’s Office, Belgian Government and the the NFWO Belgium for financial support. References Becker A. and Kramer L., Phys. Rev. Lett. 73, 955 (1994). Gardiner ©. W., Handbook of Stochastic Methods (Springer-Verlag, Berlin,1983). Garcta-Ojalvo J., Herndndez-Machado A., and Sancho J. M., Phys. Rev. Lett. 71, 1542 (1998). Graham R. and Schenzle A., Phys. Rev. A 25, 1731 (1982). Grinstein G., Munoz M. A. and Yuhai T., Phys. Rev. Lett. 76, 4376 (1996) Horsthemke W. and Lefever R., Noise-Induced Transitions (Springer-Verlag, Berlin, 1984). Li J. H. and Huang Z., Phys. Rev. E 53, 3315 (1996). Miiller R., Lippert K., Kuhnel A. and Behn V., First order phase transition in spatially extended ystem, preprint University Leipzig (1996). Ojalvo J., Parrondo J. M. R., Sancho J. M. and Van den Broeck C., Reentrant Noise-Induced Transition in the Landau-Ginzburg Model, submitted to Phys. Rev. E. Park S. H. and Kim S., Phys. Rev. E 53, 3425 (1996). Parrondo J. M. R., Van den Broeck C., Buceta J. and de la Rubia J., Physica A 224, 153 (1996) Schimansky-Geier L, and Zulicke Ch., Z. Phys. B 82 157 (1991). Van den Broeck C., J. Stat. Phys. 31, 467 (1983). Van den Broeck C., Parrondo J. M. R., Armero J. and Herndndez-Machado A., Phys. Rev. E 49, 2639 (1994). Van den Broeck C., Parrondo J. M. R., and Toral R., Phys. Rev. Lett. 73, 3395 (1994 Van ony Broeck C., Parrondo J. M. R., Toral R. and Kawai R., Nonequilibrium phase transitions induced by multiplicative noise, submitted to Phys. Rev. E. Van Kampen N. G., Stochastic Processes in Physics and Chemistry (North-Holland, ‘Amsterdam, 1992).Generalized Langevin Equations: A Useful Tool for the Perplexed Modeller of Nonequilibrium Fluctuations? Peter Hanggi Universitat Augsburg, Institut fiir Physik, Memmingerstr. 6, D-86135 Augsburg, Germany Abstract. The author identifies some subtle difficulties that one encounters within the framework of nonlinear generalized Langevin equations. These difficulties be- come even more pronounced when describing an open system that is in contact with more than one heat bath. A well known technique to describe the dynamics of a system which is governed by fluctuations is the method of generalized master equations and the methodology of generalized Langevin equations. ‘This strategy is by now well developed for thermal equilibrium systems. Here, the projector operator methodology (Mori 1965, Kawasaki 1973, Nordholm and Zwanzig 1975, Grabert et al. 1980, Grabert 1982) yields a clear-cut way to obtain the formal equations, either for the rate of change of the probability, i.e. the Generalized Master Equation (GME), or the generally nonlinear Generalized Langevin Equation (GLE). Take the case of relaxation of a system that is coupled to a thermal bath towards its unique equilibrium as specified by a single temperature T. Then, the equivalence between the two approaches is expected, but it is by no means transparent. Clearly, it are the statistical properties of the fluctuational force that determine this equivalence, such as its cumulant averages to an arbitrary high order, see in (Grabert et al. 1980). This fact is not widely appreciated because one often restricts the discussion of the statistical properties of the fluctuating force to the first two cumulants only; namely its average and its autocorrelation. Fact is that little is known between the connection of the GME and the corresponding GLE. One such relation which relates the generalized, generally nonlinear memory-diffusion matrix in the GME with the generally nonlinear memory-friction kernel in the GLE has been put forward with eqs. (55) and (56) in (Grabert et al. 1980). A popular model consists of coupling a nonlinear system S bilinearly to a bath of harmonic oscillators. Then, the total Hamiltonian of a particle with mass M moving in a potential U(z) in presence of a bath of harmonic oscillators reads (Zwanzig 1973, Ford and Kac 1987, Pollak 1986): : : : ; P PR, Mawa (Ca H Frve+y (gh : (0 z+) |: a16 Peter Hnggi Let us elucidate the reasoning which leads to the GLE. With (1) the equations of motion for the system degrees of freedom read Mi= poe Tee gt at) (2) and for the bath degrees of freedom, respectively Maga = Pa Ba =~ atid + Ca2 « @) Next we integrate the bath degrees of freedom, which obey first order ordi- nary, linear differential equations with an inhomogeneity cq:r(t). Considering the solution «(t) as given, we formally solve (3) in terms of the Green’s func- tion for the bath oscillator, @(t — s) sin(wa(t— s))/wa, i.e, a(t) = qa(to) cos(wa(t ~ to)) + Pall sin(wa(t — to)) ds sin(wa(t— s))2(s) « (4) to ms Mee Upon inserting (4) into (2) we find + oh i ds wa sin(wa(t — s))x(s) +De0 [sata cos(wa(t —to)) + ball sin(wa(t — ta) . 6) ‘The last term will be denoted by F(é). It does determine a fluctuating force. ‘The second and the third term can be combined to read le Uf ds x(s 1B cos(walt~ 5) - 20] + F(t). (6) With a partial integration we thus arrive at the GLE structure, i.e., 2 p=- a oy mat Uf ds #(s) cos(wa(t — s)) + alfa) on(wa(t~to)] + FIO. @)Generalized Langevin Equations. .. 17 Note the dependence on the initial value 2(to) in the second contribution. Hence, we have found already the exact result for the GLE, (Zwanzig 1973, Ford and Kac 1987): eS Me4M [as alt — 8)a(s) + ee = — M(t —to)2(to) + FO, ( Oe ) where the memory friction reads a at 9) = PY ey comleoalt~ 9) ®) and PI) =D ca [oat (to) cos(wa(t — to)) + Pali) sinfoo(t~ to) (10) is a colored (i.e. the noise has a finite correlation time) Gaussian fluctuating force, which obeys the fluctuation-dissipation theorem of the second kind, ie, (F(t))os = 0 F(t) F(s))px = MkT y(t - 5) (11) where the average is taken with respect to the unperturbed bath, i.e., 0 = #03 fe (eteta)||, (12) and B = (kT)-}. Here, we already notice the role of the initial slip, 7(t—to)x(to), see also in (Bez 1980, Canizares and Sols 1994), which usually imply brushed under the rug, i.e. simply ‘dropped’. For strict ohmic dissipation it reduces to a tribution, 27(t — to): Although being zero for finite times t it still affects, of course, the time evolution of the realization, as well as the relaxation of time-dependent averages of the coordinate x(t) and momentum &(t) processes, which indeed depend on all previous times t > to. The result in (8) is not yet the usual GLE, although it begins to look close. Next, we absorb the initial slip term into the stochastic force. Then, the noise C(t) = F(t)- M y(t 7 => ea [ (cal) - er 2(to)) cos(wa(t — to)) Pa(to Maa + sinfoo(t ta) ] (as)18 Peter Hanggi no longer has a stationary autocorrelation when averaged with the ‘sudden? bath ensemble pg. The combined stochastic force ¢() is again stationary and colored Gaussian, however, if conditionally averaged with respect to the Gaussian equilibrium ensemble P({Pas Ga} | (to) = 2) : 2 = 7'en{-8[ f+ tet.) |} (a4) With this conditional probability for the bath variables the statistical force ¢(t) obeys the identical fluctuation-dissipation relations as in eq. (11), i-e., COs =0 (C(t) C(s))p = C(t ~ 8 + to) C(to))p = MAT y(t - 8) . (15) Hence, care must be applied if one refers to the statistical properties of the two different stochastic forces F(t) and ¢(¢), respectively! Our exercise elucidates that this exact GLE does not pop out of Pandoras box, but can be obtained in a straightforward manner. Nevertheless, it needed a master like Zwanzig (1973), who paved the way for us. The above derivation is worth to be commented on further: Note that the steps in (2) - (7) carry through for the corresponding Heisenberg equations of motion as well. This proce- dure yields the exact (operator)-GLE for the quantum case, see in (Ford and Kac 1987). Interestingly enough, with U(z) a harmonic oscillator potential and without the ‘counter-term’, i.e. without the second term in (5) so that U(x) undergoes a potential renormalization in the GLE, the quantum-GLE together with a discussion of the corresponding quantum Nyquist formula has already been presented clearly — on one and a quarter page — by Mag- alinskij as early as (1959). In the quantum case the stochastic forces are no longer c-numbers, but become operators which act on the Hilbert space of all bath degrees of freedom (and composite Hilbert space of system plus bath for the combined stochastic force ¢(t)). The symmetrized autocorrelation of the stochastic force(s) obey a quantum generalization of the fluctuation- dissipation theorem of the second kind (quantum Nyquist theorem); i.e., the symmetric quantum noise autocorrelation can be expressed solely in terms of the macroscopic memory friction kernel 7(t — s); not withstanding recent incorrect claims to the contrary (Cortes et al. 1985). Next, let me return to classical statistical mechanics. As already demon- strated by Zwanzig (1973), the above GLE can be generalized to a nonlinear system-bath coupling, i.e. we set for the coupling, —Cagat —> —CadaG(z), in (1). The bath motions are then still linear; thus we again can formally integrate the bath degrees of freedom, thereby generalizing the result in (4). With 444) = g(x) the exact GLE for this nonlinear system-bath interactionGeneralized Langevin Equations. . 19 is then evaluated to give the following GLE Mi +Mate(0) [as a(t sle(s)) 416) + FE = —M y(t — to) g(«(to)) G(x(to)) + g(x (t)) FE) - (16) With G(x) = 2 it reduces to the previous result in (8). This nonlinear GLE exhibits multiplicative noise with a generalized, state-dependent memory fric- tion. The noise F(t) and the combined fluctuating force, F(t) — M(t — to)G(w(to)), then obey the corresponding relations in (11) and (15), respec- tively. An unsolved problem, to the best of my knowledge, is the detailed connection between the GLE in (8,16), with the stochastic force given either yy (11), or by (15), and the corresponding non-Markovian GME for the joint probability density of coordinate and momentum variable; i.e. Be, #51) = (17) Although the noise is Gaussian, the joint-process [(x(t), #(t)] is with a (non- quadratic) nonlinear potential no longer Gaussian. Then the difficulty in obtaining the GME is rooted in the fact that its derivation requires the knowledge of corresponding functional derivatives, (Hénggi 1978,1989), such as (62(t)/5 C(s)) at times s < t; hence, the GME is at best known formally only. With U(z) being at most a quadratic function of «, the process pair [e(t), #(t)], given x(to) = 29, #(to) = vo can be shown to be Gaussian, but non-Markovian. Then, a master equation for p(x, #;t) for a stable process, i.e. U(x) > 0, and an initial equilibrium preparation (Adelman 1976) or, more generally, with an initial non-equilibrium preparation (Hanggi 1978,1989), as well as for an unstable process with U(x) < 0, see in (Hainggi and Mo- jtabai 1982), can readily be constructed. It exhibits a time-convolutionless structure. Thus, it clearly does not have the retarded (memory) structure predicted by the (formal) projection operator methodology (Grabert et al. 1980, Grabert 1982). What is the corresponding memory-GME in this linear case? — I simply do not know. Sailing becomes much smoother in the Markovian limit, with ¢(t) ap- proaching stationary Gaussian white noise. Given an arbitrary potential U(z), the corresponding connection is the well-known relationship between the Langevin equation and the Fokker-Planck equation, i.e. the Klein-Kramers equation, (Klein 1922, Kramers 1940), for which, even for the case that a spatially-dependent friction coefficient governs the inertia motion, no Ito- Stratonovich dilemma occurs. Another challenge presents the task of a mi- croscopic modelling of fluctuations in a general nonequilibrium system such as e.g, a thermal ratchet system (for a recent review see Hanggi and Bar- tussek 1996). Now, matters become even worse. For example, given the fact that the detailed balance symmetry of the total system (i.e. system plus all degrees of freedom of all baths) survives a coarse graining operation, see eq. (4.3.8) in (Hanggi and Thomas 1982), it is, prima facie, not obvious how the20 Peter Hanggi violation of the detailed balance symmetry for a stationary nonequilibrium system does occur. For an elucidating discussion of this point the reader is referred to page 267 in (Hanggi and Thomas 1982). ‘To gain further insight, we consider a single particle S which is coupled simultaneously (!) in a bilinear way, cf. (1), to two harmonic baths at differ- ent temperatures, with the baths composed of a large number of oscillators. This situation distinctly differs from the Carnot machine case, where the two baths never influence the system at the same time. Related in this spirit are the theoretical models of heat conduction in harmonic chains whose ends are brought into contact with independent heat baths, which have been modelled phenomelogically by two independent Gaussian noise forces together with ohmic friction for the end points, as studied classically by Rieder, Lebowitz and Lieb (1967), and quantum mechanically by Ztircher and Talkner (1990). We next assume that two baths are initially prepared in a thermal product state at the individual temperature T, and T2, respectively. Then, the re- sulting exact GLE for this situation has the same form as in (8), but with two memory kernels, and two stochastic forces of the type in (10), obey- ing corresponding statistical properties as given in (11). Most important is the observation that two different initial slip terms of the form in (8) appear. Likewise, one can introduce two different stochastic forces of the type in (13). By forming cross-correlations, or by combining them into a single stochas- tic force we encounter, however, difficulties: What @-value should be chosen for the conditional, longtime equilibrium ensemble in (14)? Although the two baths are not coupled between each other, they are in fact in contact via their finite coupling to the system 5. Thus, it is the (relative) heat capacities of the two baths that ultimately determine the overall equilibrium tempera- ture. Moreover, the two initial slip terms couple, via the dependence on (to), the two stochastic forces ¢,(t) and ¢2(t). I am not aware that these subtle aspects have formerly been discussed in the literature. Previous attempts, a most recent can be found in (Millonas 1995), can be shown to be inconsistent (neglect. of initial slips, and mix up with different ensemble averages, etc.). Playing naive, let us discuss the simplest situation, namely we use the Markovian limits, i.e. y(t — s) —+ 274(t — s), for the stochastic forces Fi(t) and F(t) of the two baths at temperature T; and 7», respectively. Further, we simply neglect corresponding initial slips such that F,(t) = ¢1(t), Fo(t) = ¢2(t). Note that the equality does not hold when t = to. Then, the autocor- relation for the force ¢1(t) reads (G(O) Ca(s)) = 2My Tr 5(t — 5). (18) With 72 =i = 7, the stochastic force corresponding to the bath at temper- ature Tz obeys {Co(t) G(s) = 2M 7 Tr d(t— 5) - (19) Next, we intuitively set (61) G(s) =0- (20)Generalized Langevin Equations. . . aq Moreover, we use a vanishing potential field U(x) = 0. With v = #, the Langevin equation which governs the relaxation dynamics is then expected to be of the form, Mb =-27 04+ G(t) + Gt). (21) With the equilibrium temperature being T = 3(T1 +72), the correspond- ing Fokker-Planck equation for the nonequilibrium situation, with the two baths at different temperatures T, # T2 can then formally be recast by the stochastically equivalent Langevin equation of a single bath at temperature T; namely Mi =—-27 0 +C(t) (22) with ¢(t) being a single white Gaussian noise, obeying (C(t) ¢(s)) = 2M (24)TS(t — 5). (23) Both equations (21), and (22) with (23), indeed yield the identical Fokker- Planck equation B(v,t) = or Qyup(v, t)] + 2M7 tes eae. t). (24) What does this result imply: Is the nonequilibrium relaxation dynamics of v(t), given two harmonic baths at temperature T; and 72, at all later times t identical to the equilibrium relaxation dynamics of v(t) in contact with a single bath at T = 3(T, + Tz)? Clearly, the answer must be ‘no’. Thus, we must have made mistakes for this most simple situation already, such as e.g. mixing different averages in eqs. (18 — 23). ‘The Langevin equation in (21) and (22) should describe different physical situations at finite times t > to; in the longtime limit, ta —+ —oo, however, these differences vanish and both equations describe the same equilibrium Boltzmann distribution for the momentum variable v. In this sense, the two Langevin equations become equivalent in the asymptotic longtime limit; how- ever, (21) no longer, of course, can be used to describe the nonequilibrium relaxation of averages towards asymptotic equilibrium. The generalization with quantum mechanics taken fully. into account is even more subtle. Then, as has been indicated above, an exact (operator) GLE - which solely acts on the Hilbert space of the system $ alone — does simply not exist. References Adelman S. A., J.Chem. Phys. 64: 124 (1976). Bez W., Z. Phys. B 9: 319 (1980). Canizares J. $. and Sols P., Physica A 212: 181 (1994). Cortes E., West B. J., and Lindenberg K., J. Chem. Phys. 82: 2708 (1985).22 Peter Hanggi Ford G. W. and Kac M., J. Stat. Phys. 46: 803 (1987); note also the survey of models presented in: Ford G. W., Lewis J. T., and O’Connel R. F., Phys. Rev. A 37; 4419 (1988). Grabert H., Hanggi P., and Talkner P., J. Stat. Phys. 22: 537 (1980). Grabert H., Springer Tracts in Mod. Phys. 95: 1-162 (1982). Hanggi P., Z. Phys. B 31: 407 (1978); Hanggi P., Noise in nonlinear dynamical systems, Vol. 1, Moss F. and McClintock P. V. E., eds., (Cambridge University Press, Cambridge, 1989); pp: 307 - 328. Hanggi P. and Bartussek R., Brownian Rectifiers: How to Convert Brownian Mo- tion into Directed Transport, in: Nonlinear Physics of Complex Systems - Current Status and Future Trends, J. Parisi, $.C., Miiller and W. Zimmermann, eds., Springer Series “Lecture Notes in Physics’ (Springer,Berlin, 1996). Hanggi P. and F. Mojtabai, Phys. Rev. A 26: 1168 (1982). Hanggi P. and Thomas H., Phys. Rep. C 88: 207 (1982). Kawasaki K., J. Phys. A 6: 1289 (1973). Klein O., Ark. Mat. Astron. Fys. 16, No. 5: 1 (1922). Kramers H. A., Physica (Utrecht) 7: 284 (1940) Magalinskij V. B., Sov. Phys. JETP 9: 1381 (1959) [Zh. Eksp. Teor. Fiz. 36: 1942 (1959)]. Millonas M., Phys. Rev. Lett. 74: 10 (1995). Mori H., Progr. Theor. Phys. 33: 423 (1965). Nordholm K. $. J. and Zwanzig R., J. Stat. Phys. 13: 347 (1975) Pollak E., Phys. Rev. A 33: 4244 (1986). Rieder Z., Lebowitz J. L., and Lieb E., J. Math. Phys. 8: 1073 (1967) Zircher U. and Talkner P., Phys. Rev. A 42; 3267 (1990); ibid 42: 3278 (1990). Zwanzig R., J. Stat. Phys. 9: 215 (1973).Colored Noise in Dynamical Systems: Some Exact Solutions Peter Jung School of Physics, Georgia Institute of Technology, Atlanta GA 30332, USA Abstract. I consider one dimensional nonlinear systems driven by colored Gaus- sian noise. A framework is presented which allows for the construction of models that can be exactly solved in terms of single variable Fokker-Planck equations. I further discuss an exactly solvable linear model with multiplicative colored noise, where the noise color mediates a stochastic-resonance like behavior. 1. Introduction The general subject. of colored noise driven dynamical flows is rooted in the study of the motion of small particles suspended in a fluid and moving under the influence of random forces which result from collisions with molecules of the fluid induced by thermal fluctuations, or in short the phenomenon of Brownian motion (Einstein 1956). In the earliest studies of Brownian motion, the damping of the motion of the suspended particles was very large compared to that of the fluid molecules, so that inertial effects could be neglected. Moreover, the thermal fluctuations occur on a time scale which is very much shorter than that of the Brownian particle. It is then a good approximation to assume that the random forces are uncorrelated as perceived by the particle on its own, much slower time scale. This assumption considerably simplifies the problem, because it allows one to treat the stochastic dynamical motions as a Markov process for which many methods and approximation schemes are available. The fluctuations which can be treated under this assumption have often been termed ”white noise”. Thus, white noise fluctuations €(t), are those for which the autocorrelation function is given by, < €(t)E(s) >= 2D5(t — 5), (1) with the noise intensity D. A large body of literature on white noise appli- cations exists, and appropriate starting points are the now classic reviews of Chandrasekhar, Uhlenbeck and Ornstein, and others in the collection of Wax (1954), the texts by Stratonovich (1963), van Kampen (1992), Risken (1984) and Horsthemke and Lefever (1984) or the reports by Hanggi and Thomas (1982), and Fox (1978). In the physical world this idealization, however, is sometimes not justified. For example, the photon statistics of a dye-laser (Zhou et al. 1986) and the motional narrowing of magnetic resonance lineshapes (Kubo 1962) are con- siderably affected by the finite correlation time of fluctuations. It has become24 Peter Jung customary to study colored noise driven systems! in terms of a Langevin equation #= h(x) + o(z)E(t), (2) where « is the observable of interest, and h(x) and g(x) are continuous dif- ferentiable functions. The noise €(t) is considered stationary and Gaussian distributed with zero mean and correlation function < €(t)€(s) >= O(t - 5). (3) ‘The spectral density, given by the Fourier-transform of the correlation func- tion is frequency dependent, hence the term colored noise. In many studies (for a recent review, see (Hanggi and Jung 1995)), an exponential correlation function ®(s) = (D/7e) exp(—|s|/r2) has been used, with 7. describing the time-scale of the fluctuations and D/r. its variance. ‘The power spectrum is given by a Lorentzian with corner-frequency 1/7. For this Ornstein- Uhlenbeck noise, the white-noise limits recovered by decreasing the correlation-time 7, and leaving D constant. While Ornstein-Uhlenbeck noise is generated by applying white noise to a first order linear filter with one real-valued pole of the Greens function, a whole class of other Gaussian noise sources can be constructed by applying white noise to linear filters of all orders with complex-valued poles of the Greens functions. Another example is the so called harmonic noise (Schimansky-Geier and Ziilicke 1990) where the Greens function of the filter has two complex conjugate poles. In section 2 we derive exact equations of motions of the Fokker-Planck type for the probability density P(z, t) corresponding to the Langevin equation(2) for a specified class of functions g(x) and h(z). In section 3, we discuss some of these exactly solvable models explicitly. In section 4, we consider a linear model, driven by multiplicative colored noise and periodic forcing, which is exactly solvable and shows noise-color induced stochastic resonance. 2 Equation of motion for the probability density Starting from (1), the stochastic Liouville equation for the density p(v,t) = 5(x —x{r(4)]), where z[r(t)] is the solution of (1) with r(t) being a realization of the noise source £(t), can be obtained as Pr(x,t) = (A+ Br(t) pr(e,t), (4) * One has to distinguish systems, where the randomness of the observable of inter- est results from the interaction with a large number of other intrinsic degrees of freedom (internal noise) and systems which are driven by noise, e.g. by a noisy parameter (ezternal noise).Colored Noise in Dynamical Systems 25 with the operators a =- he) B= 242). 6) In the interaction-picture, ice. pr(2,t) = exp(—At)pr(z,t), (6) the equation of motion simplifies to Br(2,t) = r(t)Lu(t)pr(2, t) Lu(t) = exp(—At)B exp(At), (7) with the formal solution : pe(v,t) = T exp [ f CL pe(2,0), @) , where 7 is the backward chronological operator. In the next step we have to average over realizations of the noise r(/). Since p, (2,0) in general cannot be factorized out of the average, it is clear that a closed equation of motion for the average P(x,t) = (p,(z,t)), cannot be derived in general from Eq.(8). For the specific initial condition P(x,0) = 6(x — x0), however, the average factorizes. For a general initial condition, we can still factorize the average for times larger than the correlation time of the noise. Taking the average over all realizations r(t) and factorizing yields the result Plest) = (ole) = These [[ ‘reyeu(ttt])- P20). () ‘The expression in the bracket. can be evaluated exactly, if the time-evolution operator commutes at arbitrary times t and t’, i.e. (Lu (t), Lu (t’)} = Lo (t)Lu(t!) — Lu (t!)Lw (t) = 0. (10) Using the condition (10), Eq.(3), and the Gaussian properties of the noise, Eq.(9) can be written as pt Pia.t)=em{5 [ i Lets) Lo(tls —ta}atadts Pla). (11) ‘Taking the time derivative, we obtain the equation of motion in the interaction- picture P(a,t) = L(t) [ dt’ Ly, (t')@(t — ') P(x, t). (12)26 Peter Jung In the original representation p(x,t) = exp(At)P(z,t), one obtains the equation of motion (x,t) = Ap(2,t) + of dt! exp(At')Bexp(—At)¢(t)o(x,t). (13) Some comments on this Fokker-Planck equation are in order: . It is exact only for initial conditions of the type P(z,0) = 6(x — x0) . The second term depends via the lower integration limit explicitly on the initial time tp = 0, a hint that we are dealing with a non-Markovian process. For times ¢ large against the correlation time of the noise, this dependence will disappear. . In the white noise limit, ie. ¢(t) = 2D6(1), the standard Fokker-Planck equation is recovered. © To construct exactly solvable models with colored noise, we have to find functions h(x) and g(z) such, that the condition (10) is fulfilled. Here, the Baker-Haussdorf relation exp(At)Bexp(—At) = B+ [A B]t + st, [ABy24.. (14) will be particularly useful. 3 Exactly solvable models In this section we discuss how one can use the results of the preceding sec- tion to construct exactly solvable models. An exactly solvable model in my notation is a model where the time evolution of the variable x is described by a single variable Fokker-Planck equation. Although, one cannot find in gen- eral explicit time-dependent solutions for such an equation, the stationary probability density and moments can be computed. 3.1 Transformations to linear systems The most simple way to fulfill (10) is to require that the operators A and B are commuting, ie. [A,B] = 0. Then, the equation of motion (13) becomes Bla.) = Apla,t) +B i * a’ale!yole,t), (15) which is identical to the usual Fokker-Planck equation, but with a time de- pendent diffusion coefficient. In order that the operators A and B commute, the functions h(x) and g(x) have to fulfill the ordinary differential equationColored Noise in Dynamical Systems 27 A(z)g'(x) — g(x)h'(z) = 0. The solution h(x) = eg(z) with the arbitrary constant ¢ suggests the simple coordinate transform of 1/h(z")dz! — ct, (16) to the linear Langevin equation 7 = é(t), which describes a free particle driven by colored Gaussian noise. The requirement of commuting operators A and B has thus lead us to a class of exactly solvable models which can be transformed to a linear model. The Hongler-models (see e.g. (Horsthemke and Lefever 1984)) are a more general class of such systems. 3.2 Exactly solvable nonlinear systems More interesting exactly solvable models which cannot be transformed to a linear process can be found by requiring [A,B] = 0B, (17) where a is an arbitrary constant. Using the Baker-Hausdorff relation (14), the evolution operator in the interaction picture can be brought to the simpler form Lu (t) = exp(—At)B exp(At) = Bexp(—at). (18) which fulfills - as required - the commutation-relation [Cus(t), Lun(t’)] = 0 for all times t and t’, The equation of motion for the probability density becomes A2,t) = Aplx,t) +B? f at exp(at')9(t/)ol2,t), (19) which is again of the Fokker-Planck type with a time dependent diffusion coefficient. The condition (17) requires that the functions g(x) and A(z) obey the ordinary differential equation h(x)g/(2) — g()h'(z) = —ag(z). For the specific choice g(z) = 1, the solution of the differential equation for h(x) reads A(z) = —ax + B with an arbitrary constant 8. ‘The only with our scheme exactly solvable model with additive colored noise is the Ornstein-Uhlenbeck pro We can, however, construct a variety of exactly solvable multiplicative models. For a given function h(2), the correct function g(x) which leads to a solvable model reads g(x) = Kh(x) exp (aV(z)), (20) where V(z) is the integral of (1/h(z)) and a and K are arbitrary constants. For the special choice h(x) = —yz, we find g(x) = Kyx*/7*1, This special exactly solvable model has been identified by Hanggi (Hanggi 1981) by using functional calculus.28 Peter Jung 4 Colored noise mediated stochastic resonance: an exact solution In this section, we present a periodically driven system with multiplicative colored noise, for which we can calculate the moments of the distribution function exactly (Fulinsky 1995, Berdichevsky and Gitterman 1996). The system is given by &= (a +€(t))e + Asin( Qt) (21) where we assume €(t) to be Gaussian, colored noise with zero mean. Although the stochastic differential equation is linear, i.e. linear superposition applies for two solutions, the response to the colored noise and the periodic forcing is not decomposable as in the case of a periodically driven, additive stochastic differential equation (Jung and Hanggi 1990). This has the consequence, that although equation (21) is linear, the amplitude of the systems response to the periodic forcing actually depends on the intensity of the noise. Fulinsky (1995) and Berdichevsky and Gitterman (1996) have shown that the exact solution of (21) with colored dichotomous noise shows stochastic resonance, i.e. the amplitude of the response to the periodic forcing as a function of the noise intensity goes through a maximum (for a review on stochastic resonance, see (Jung 1993). ‘The solution of (21) is written as x(t) = 2(0)G(t) + 4 f smaryce —t)dt', (22) with the Greens-function : =exp(—at— f e(t'yat’). 23 G(t) oxo ( at fe ) (23) For Gaussian noise, £(t) the averages of exponentials can be carried through yielding (G(t) = exp (a + 30) : (24) with the variance of P(t) = fy €(t/)dt’, given by q ¢ (t= f an f dtz(ty — t2). (25) ‘The time dependent mean value (x(¢)) is then written as (x(t)) = ro(G(t)) + af’ sin(2t’(G(t —t'))dt’, (26) with the function (G(t — t')), describing the response to the periodic forcing. Note that. the construction of higher momentsis tedious, but straight forward.Colored Noise in Dynamical Systems 29 Supplementing the response function (G(t)) by a Heaviside step function (1), i.e. R(t) = O()(G(4)), the response is described by the complex-valued susceptibility R(w), given by the one-sided Fourier-transform of R(t). ‘Lo become more specific, we are choosing now exponentially correlated noise, ie. o(t) = (D/r-) exp(—lIt|/r2). Inserting this correlation function into (25) yields for the variance o2(¢) o*(t) = 2Dt+ 2D% [rx (-+) - i ; (27) and the Greens function (G(t)) = exp [-(a — Dt + Dre(exp(—t/te) — 1)}- (28) For D > a, the Greens function increases exponentially since the noise in- duced drift Dz becomes larger than the binding force —az. We restrict our- selves to D < a, where the system approaches a steady state for large times. ‘The susceptibility can be written as the Fourier integral Rw) = fe [(o-«- iw) t+ Dre (exp (-4) ~ | di. (29) As mentioned in the beginning of this section, although the system is linear, the response of the system to periodic forcing actually depends on the noise intensity D. 4.1 The white noise limit In the limit of white noise, i.e. 7 ++ 0, this above described mixing of exter- nal noise and periodic forcing is readily understood as the consequence of the renormalization of the drift due to the noise induced term Dz (Horsthemke and Lefever 1984). The response function and the susceptibility assume the simple forms (G(t)) = exp (-(a— D)t) and Ro(w) = 1/(a — D + iw), respec- tively. For the amplitude 5 of the response to the periodic forcing, given by the absolute value of the susceptibility at the driving frequency 2, we find in turn 8(D, 2) (30) (a— D)? +2? ‘The response amplitude increases monotonically for increasing noise strength D until it reaches at the critical value D, = a its maximum 1/w - the response amplitude of a free particle.