This action might not be possible to undo. Are you sure you want to continue?
ShangKeng M a
Translated by
M. K. Fung
World Scientific
Philadelphia 0 Singapore
Published by
World Scientific Publishing Co. Pte. Ltd.
P 0 Box 128, Farrer Road, Singapore 9128
USA ojfice: Suite lB, 1060 Main Street, River Edge, NJ 07661
UK office: 73 Lynton Mead, Totteridge, London N20 8DH
Library of Congress CataloginginPublication Data Main entry under title: Ma, ShangKeng S a i t c l Mechanics ttsia Includes Index
I. Title QC174.8.M24 1985 530.1'3 852299 ISBN 9971966069 ISBN 9971966077 @bk)
1. Statistical Mechanics.
First published 1985 First reprint 1993
Copyright 01985 by World Scientific Publishing Co. Pte. Ltd.
All rights reserved. This book, orparts thereoi may not be reproduced in any form or by any meany, electronic or mechanical, including photocopying, recording orany information storage and retrieval system now known or to be invented, without written pemiission from the Publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 27 Congress Street, Salem, MA 01970, USA.
Printed in Singapore.
To
Chiansan ( Till ) and Tienmu ( X €4 1
PREFACE
This book serves both as a graduate text and as a reference. Statistical mechanics is a fundamental branch of science which has made substantial progress in the past decade, opening up many new areas of research. The aim of this book is not to give a complete introduction to this field, but rather to give a clear exposition of the most basic principles, and to use relatively uptodate material as examples of application. Concepts as well as applications are thus emphasised . The statistical mechanics of today is so fragmented into subfields and so extensively applied to various phenomena that the reader cannot hope to become an expert by just reading one or two textbooks. This text merely aims at helping the reader establish a solid foundation on the principles, clarify some concepts, as well as learn some simple ‘tricks’ so that he can cope with new problems himself. At the time of writing, there are already many textbooks around including many classics. But they now look somewhat oldfashioned both conceptually as well as in content. On the other hand, many new books are too specialised. Scientists are now now heavily devoted to research, and they expend less effort in writing books. In addition, new knowledge accumulates rapidly, so much so that it is extremely difficult to write a textbook which can stand the test of time. This book is a short one, serving the need of students for the time being. I hope that it can of use to contemporary readers. Statistical mechanics is a branch of theoretical physics, distinguished by its extensive applications in physics, astronomy, chemistry, materials science as well as biology. It is a convenient tool to link the molecular structure and properties with the macroscopic thermodynamic and electromagnetic behavior. The application of statistical mechanics can be classified according to its degree
viii
STATISTICAL MECHANICS
of difficulty as “elementary” and “advanced”. The elementary part is essentially the ideal gas, including the quantum ideal gas. It includes all cases when the interaction between particles is unimportant; the free electron model is one of these cases. Its application is rather extensive. The advanced part discusses the interaction, which plays an important role in many problems, for example in the phenomena of phase transitions. Naturally, the advanced part is much more difficult. Except for some special cases, approximation and numerical calculation are the only methods of solution. The simple and effective approximations usually simplify the problem to certain models of an ideal gas. The mean field approximation is an important example. In my view, advanced statistical mechanics is a product of the emergence of solid state physics. Although the relation between statistical mechanics and phase transitions is forty years old, most of the applications and developments were made within the last twenty years. The field especially has been fertile in the last ten years. Most of the applications are on solid state, because the phenomena in solid state are numerous, the experimental techniques have become increasingly sophisticated and the physical systems to be investigated are not as complicated as biological systems. Hence the theoretical analysis can cope with such systems. Many new concepts, in fact, originate from solid state physics. The greater part of this book is on the discussion of examples. The elementary examples are used to clarify the concepts, while the advanced examples are used to discuss the phenomena and methods. In fact, the basic principles and the applications are inseparable. All the examples in this book are used, directly or indirectly, to illustrate some basic concepts. The examples are selected for their extreme simplicity. The material is somewhat biased towards phenomena in solid state physics and I have tried to include some modern, interesting and inspiring topics, in the hope of providing the reader with a glimpse of the modern developments in statistical mechanics. One motivation for writing this book is to air some of my personal views, My views on some of the basic concepts differ from the conventional ones, I think that statistical mechanics, as it is at present, is an illproportioned subject, with many successful applications but relatively little understanding of the basic principles. Yet most textbooks treat statistical mechanics as a complete science. Roughly speaking, statistical mechanics is based on some rules of calculation. We can proceed forward and discuss the application of the rules of calculation, or we can turn back to discuss their origin. (These rules or calculation can be said to be the Boltzmann’s formula for the entropy, which is equivalent to the
PREFACE
i x
statement that a state occurs with a probability proportional to H being the energy of the state and T the temperature). There are many examples of applications, but works on the quest for the origin are relatively rare, especially at the textbook level. One reason is that, over the years, the applications have been so successful that a substantial degree of confidence is now attached to these rules of calculation. Although the origin may not be clear, the problem seems almost irrelevant because the rules are so useful and successful. There is another reason. At present we still d o not know how to start from mechanics and proceed to understand these rules of calculation; these rules are still an assumption. Many books start from the rules of calculation, ignoring their origin, and treat statistical mechanics purely as a matter of mathematical technique. This viewpoint has its validity. However, some textbooks attempt to make up for the lack of a firm conceptual foundation by giving fancy arguments to set up these rules as established laws. If students d o not understand these fancy arguments, it does not matter, so long as they can still use the rules t o calculate. This was my own experience as a student. Of course, many successful applications d o not require a profound and exact understanding. This is characteristic of scientific development. Nevertheless, if the range of applications is extended, problems will sooner or later appear. A deeper understanding will become urgent. Today the development of statistical mechanics has reached the point of delving into its origin. The basic assumption in statistical mechanics is in fact an assumption of independence or randomness. If we can understand this assumption, then it would be helpful in understanding other stochastic phenomena, such as random number generation in the computer. The converse is also true. We think that the proper attitude to science and to knowledge in general is expressed by the Chinese saying, “Recognise what you know, and recognise what you do not know.” Hence this book pays much attention to all the weak points in statistical mechanics, the limitations on its rules of calculation, as well as its questionable and ambiguous points. Although this book does not provide the answer to the important questions, I have at least tried t o state the questions clearly. From the experience of learning mechanics and electromagnetism, the students are accustomed to precisely stated laws. Yet I constantly remind the reader of all those questionable points in statistical mechanics. It may engender a feeling of unease. I hope that this feeling of unease will make the reader become more cautious, cultivating the habit of not accepting too much on faith. This book is divided into seven parts totalling thiry chapters. These seven
x STATISTICAL MECHANICS parts are Part I Part I1 Part 111 Part IV Part V Part VI Part VII Equilibrium Hypothesis Probability Applications Dynamics Theoretical basis Condensation Chapters 1 to 4 Chapters 5 to 9 Chapters 10 to 13 Chapters 14 to 19 Chapters 20 to 22 Chapters 23 to 26 Chapters 27 to 30 Part I is a warmup. Complicated calculations are no doubt very important. Part V discusses some nonequilibrium phenomena. but only illustrated by simple examples. The discussion in this book are not too rigorous. as applications of the theory. many conclusions are not rigorously proved. that is to say. Statistical mechanics is established on a daring assumption. Some very important but specialised topics have not been covered in this book. and it is of no great use to include a lot of complicated calculations here in the book. My view is that the concept of ensembles is unnecessary and indeed not compatible with reality. avoiding abstract concepts such as “ensembles”. It reviews some common concepts and terms. 111 and VI emphasise the basic concepts. The mathematical prerequisite for this book is not high. with a view t o elucidate the maintenance of equilibrium. Part IV and VII analyse various phenomena. Included are some discussions on frozen impurities. The meaning of this assumption is by no means clear. I do not mean to say that rigorous proof is not important. . This book discusses this assumption from the point of view of the underlying molecular motion. but I think that the necessary techniques can b e acquired only through actual practice. Metastable states and numerical simulation are included in this part. The contents are not a report of the results in the literature. as well the theory of the ideal gas (including the quantum ideal gas). It is a tool. superfluidity and the two dimensional Coulomb gas. and the necessary calculational techniques are few. Probability is regarded here as a way of handling statistical data. but it is beyond the scope of my expertise. There are in fact no complicated calculations in the book at all. My view on these problems may not be the most appropriate and the treatment given here reflect my personal prejudices. Part 11. not a theory. but are my own understanding of these problems expressed for the reader in the most direct fashion I know. The reader must read these chapters with this in mind.
If the reader does not d o the problems.PREFACE xi These include critical phenomena and the renormalisation group (although the phenomena of phase transitions occupy many pages in this book). and there should be no problem in writing scientific books in one’s own language. and writing this book was a bold attempt. scientists have become accustomed to writing research reports in the languages of the West. I found the writing difficult. . I have another motive in writing this book. quantum mechanics. Many points of emphasis and approach differ from that of standard textbooks. Indeed science is not literature. thermodynamics and electromagnetism. Some materials in the book appeared in journals but not in current books. the importance of the metastable state and the intimate relation of superfluidity with the conservation of the winding number. The mathematics used in undergraduate quantum mechanics is sufficient. Your own understanding of statistical mechanics is acquired through your own thinking and work. the echo phenomena in Chapter 24 and the calculation of entropy from the trajectory in Chapter 25. There is also little on thermodynamics and in general. Scientific texts in Chinese are extremely scarce today. and he may even miss the basic concepts. No profound mathematical prerequisite is necessary for reading this book. Problems are an integral part of the contents. but I gained fluency as I went along. for example the relation between the virial coefficient and the collision time in Chapter 14. If this book is not easy t o read. the answer would often be that one is unable t o do so. abstract formal things are avoided. This book as well as others are for your reference. The reader should be acquainted with mechanics. Rather. Statistical mechanics is a difficult subject in basic physics and this book is not easy to study. Nonequilibrium phenomena are little talked about. and I hope this book may be of some help in alleviating this situation. If asked to write a book in Chinese. That the presentation in Chinese is short of fluency is but a secondary factor. You should not be led along blindly. The standard of my Chinese is humble. he cannot learn the techniques of application. we emphasise the relation of the observation time with equilibrium. this book requires the reader to have a cautious and independent mind. and textbooks should be written by scientists who are native speakers of their languages. I deeply feel that basic sciences should be taught in one’s own tongue. In the beginning. it is mainly due to the complexity of the content. In this century. The reader should remember that the contents of this book as well as those of other books are not what you understand. Examples are biased t o problems in solid state. provided the contents are there. whereas the truth is that one does not try. For example.
Y Shen ( u s. T T Chen (as) and graduate students C L Lin (M#@). J Martinez. This book has much room for improvement. J Chen. F K Yang (BE$!). C H Lai. K C Chang. (@a). . Spring 1982 SKMa Publisher’s Acknowledgement The publisher is deeply grateful to Professor M K Fung for translating the Chinese text and to K Young.and D G Lin (MSZE4). but I have seldom doen my filial duty. especially all the directors and Miss Y M Chang (%H@). S n) T L Lin H C Yen (%R%i). I record my thanks to them here. That this book could be finished at all owes much to their encouragement and support.E Yen (WSB). W T Ni (fR#4). Y M Shih (ZZTE). I hope the reader will not hesitate to give his views. I have to thank my parents. help and encouragement. I therefore dedicate this book (Chinese edition) to my parents.xu STATISTICAL MECHANICS Most parts of this book were written during my stay at the National Tsing Hua University in 1977 and 1981. Y Koh and JTay for checking the translation and galley proofs. They have always expressed their full support for my work. Many professors and the students have offered me advice. especially Professors Y Y Lee (3%&). Moreover. I express my gratitude to them here. We hope that the spirit and depth of Professor Ma’s views on Statistical Mechanics are well communicated in this book.J L Chen (R@&).H M Huang (H9#). Since Professor Ma passed away before the English edition could be completed their cooperation with us in this effort was indispensable. The administrative and secretarial staff in the Graduate Institute of the National Tsing Hua University have given me much help.
6 2.4 1.5 2.CONTENTS Preface Introduction vii xxi PART I EQUILIBRIUM 1 1.1 1.3 2.5 1 Equilibrium Equilibrium and Observation Time Equilibrium and Molecular Motion Large Scale Change and Local Equilibrium Structure of Matter Supplement and Conclusions Problems The Basic Concepts of Thermodynamics Equilibrium and Reversible Processes Work and Heat Adiabatic Process and Entropy Absolute Temperature and Entropy The Ideal Gas The Second Law of Thermodynamics The Efficiency and Speed of a Heat Engine Problems 10 10 11 13 16 19 24 24 2 2.2 1.4 2.1 2.3 1.7 28 .2 2.
1 4.3 6.8 5.9 65 66 66 67 70 72 76 78 80 81 85 86 The Basic Assumption of Entropy and Molecular Motion Region of Motion Assumption for the Calculation of the Entropy Fermions The Ideal Gas Some Special Features of High Dimensional Space Peak Integration and the Law of Large Numbers The Law of Equipartition of Energy Perspectives of the Region of Motion Trajectory and the Region of Motion Problems 6 6.3 5.1 6.6 5.2 4.5 4 Electrons in Metals Dense Ideal Gas Heat Capacity Influence of Periodic Structure The Structural Change Caused by Electrons Problems 4.7 5.2 6.2 5.2 3.3 3.3 4.1 3.4 48 51 55 60 63 PART I1 HYPOTHESIS 5 5.4 Some Elementary Applications of the Basic Assumption Systems and Subsystems Vibrations The Debye Model Phonons and Second Sound 88 88 94 98 102 .4 3.4 5.1 5. Chemical Potential and the Constant Term in the Entropy Distribution of Bosons and Black Body Radiation Chemical Reactions and the Law of Concentrations Problems 30 31 35 39 41 43 45 48 3.xiv STATISTICAL MECHANICS 3 The Law of Detailed Balance The Energy Distribution of the State The Energy Distribution of Fermions Conservation Laws.5 5.
2 8. Thermodynamic Limit) Generalisation of the Theorem Problems 9.3 164 166 174 178 180 11 11.1 11.2 11.1 10.Lee Theorem Macroscopic Limit (i.4 180 186 188 190 195 .1 7.3 133 139 145 150 153 9 Yang.1 8.2 154 159 161 PART I11 PROBABILITY 10 163 164 Probability and Statistics The Use and Misuse of Probability Distributions and the Statistics of the Average Value The General Definition of Probability Problems Independence and Chaos The Definition and Consequence of Independent Phenomena Test of Independence Random Sequences Scattering Experiments and the Measurement of the Correlation Functions Problems 10.2 10.e.3 11.5 General Conclusions.1 9.2 118 122 130 133 8 8.CONTENTS xv 6. Average Values and the Normal Distribution Problems Rules of Calculation Various Thermodynamic Potentials The Various Rules of Calculation Problems Illustrative Examples Time Scale and the Region of Motion Gas of Hydrogen Molecules OneDimensional Model Problems 111 116 118 7 7.
3 12.5 245 246 246 248 250 254 257 259 Corrections to the Ideal Gas Law Intermolecular Interactions Corrections to the Ideal Gas Law Time Delay in Collisions Quantum Mechanical Calculation Bound State and the Levinson Theorem Problems 15 15.4 15.6 12.9 Sum of Many Independent Variables Normal Distributions The Central Limit Theorem (Rudimentary Discussion) Higher Order Averages and the Cumulant The Cumulant Expansion Theorem Central Limit Theorem (General Exposition) Repeaied Trials and the Determination of Probability Random Motion and Diffusion Fluctuation of Macroscopic Variables Fluctuations and an Extension of the Basic Assumption Problems 198 198 200 203 207 209 21 1 214 217 219 222 13 13.4 Correlation Functions Response and Fluctuation Density Correlation Function The Fermion Gas The Response and Correlation Functions in Quantum Mechanics Problems 225 225 230 23 5 23 8 242 PART IV APPLICATIONS 14 14.7 12.5 12.2 14.1 13.xvi STATISTICAL MECHANICS 12 12.1 15.2 12.8 12.1 14.5 Phase Equilibrium Gaseous and Liquid Phases The Growth of Water Droplets Latent Heat and the Interaction Energy of Molecules Melting and the Lindemann Formula Definitions of pl and p2 Problems 26 1 26 1 26 5 268 270 27 1 274 .3 14.3 15.4 12.2 13.3 13.4 14.2 15.1 12.
2 17.1 20.2 20.1 18.3 16.CONTENTS xvii 16 16.3 20.1 17.4 18.1 19.3 19.2 19.4 Magnetism Paramagnetism Magnetic Susceptibility Diamagnetism of Charged Particles Spinspin Interaction Problems 277 277 28 1 283 29 1 295 17 17.3 18.2 16.4 345 346 346 349 351 3 54 The Equation of Motion For Gases Flow and Collision Case of No Collisions .1 16.3 Ising Model king Ferromagnetism in One Dimension Proof of the Existence of Ising Ferromagnetism in Two Dimensions Other king Models Problems 297 297 301 307 310 18 Impurities and Motion Solutions and Osmotic Pressure Effective Interaction Energy Low Density Case Mobile and Stationary Impurities The Amorphous State Problems 312 312 3 14 316 318 3 23 325 18.5 19 19.2 18.4 19.Plasma Oscillations Zero Sound Collisions and Diffusion .5 Electrostatic Interaction Short Distance and Long Distance are Both Important The Plasma Gas and Ionic Solutions Electrons in Metals Electron Crystals Two Dimensional Coulomb Gas Model Problems 3 29 329 330 333 335 337 341 PART V DYNAMICS 20 20.
5 20.4 388 3 89 39 1 391 394 397 PART VI THEORETICAL BASES 399 400 400 403 405 408 41 1 413 23 23.4 21.2 21.3 21.3 23.1 22.1 24.2 22.theorem Problems The Diffusion Equation Simple Examples The Metastable State Transformation to the Wave Equation Derivation of the Diffusion Equation TwoState Cluster Model Problems Numerical Simulation Numerical Solution of Molecular Motion Random Sequences Monte Carlo Simulation Conceptual Problems to be Noted Problems 357 363 364 21 21.1 23.2 23.3 24.5 Laws of Thermodynamics Adiabatic Processes Adiabatic Process and Entropy The Second Law of Thermodynamics The Third Law of Thermodynamics The Amorphous State and the Third Law Problems 24 24.6 Collisions and Sound Waves.4 Echo Phenomena Demonstration of the Viscous Liquid Bottle Analysis of the Demonstration Spin Echo Experiment The Plasma Echo Problems 415 416 417 41 9 422 424 .xviii STATISTICAL MECHANICS 20.4 23.3 22.2 24. Viscosity and Heat Conduction H .1 21.5 367 367 372 374 3 76 3 80 386 22 22.
2 28.5 26.5 25.5 27.1 28.6 26.1 26.4 25.1 25.3 26.the Coulomb Gas Model Problems 457 458 458 46 1 464 466 470 473 476 480 481 485 488 493 28 28.6 Entropy Calculation from the Trajectory of Motion Number of Coincidences and the Size of the Region Independence of Different Parts of a System Correlation Time Process of Computation Entropy of the Metastable State The Third Law of Thermodynamics Problems The Origin of the Basic Assumption The Basic Assumption Ergodicity and Ensembles Mixing and Independence Probability and Experiments Instability of the Trajectory The General Problem of Independence Difficulties Encountered in Applications Problems 425 426 429 43 1 43 2 434 43 5 43 7 440 44 1 442 445 448 450 454 455 455 26 26.3 25.3 27.1 27.2 26.CONTENTS xix 25 25.2 25.6 Mean Field Solutions Mean Field Theory Total Magnetic Moment Thermodynamical Potential and the Coexistence of Different Phases The king Lattice Gas The Van der Waals Equation Common Properties of Condensation Phenomena Problems Fluctuation of the Boundary Surface The Liquid Surface Model The Boundary Surface of the Ising Model Crystal Surface .4 27.2 27.7 PART VII CONDENSATION 27 27.3 .4 26.
1 29.2 30.5 Models with Continuous Symmetry The Planar Vector Model Density Fluctuations of a Crystal Quantum Vector Model Continuous Symmetry and Soft Modes Defects in the Condensation Problems 495 495 498 502 505 508 5 14 518 30 30.4 29.2 29.4 Theory of Superfluidity Quantum Lattice Gas The Ground State and Low Temperature Model State of Flow and Winding Number Stability of Superfluidity Problems 519 522 525 528 535 References Index 54 1 545 .1 30.3 29.xx STATISTICAL MECHANICS 29 29.3 30.
These include many exercises of a conventional variety. The choice of material to be covered in the course would have to depend on the need and level of the students. We have already avoided complicated calculation so that students should be fully capable of repeating every derivation in the book. materials science and chemistry. the first three parts are elementary. we start from the elementary. On the whole. A dogmatic pedagogical approach would be most unwise. My own view is that graduate courses should stress independent thinking about a problem before consulting references. It would not be appropriate to skip any material in Parts 1111. Part VI is also general. and should not be neglected. and should not be omitted. hypotheses and consequences. leading gradually to more advanced topics. (b) Mathematical derivations should not be neglected. The objective of the derivation is to instill an appreciation of the approximations.INTRODUCTION 1. (c) Each chapter ends with a set of problems. or merely identify the sources of the tables and figures. with . as well as many intended for discussions. They provide direction of further study. (d) The references listed are not strictly necessary as a part of the course. The latter may not have ‘standard’ answers and often I do not know the answers myself. and the opinion of the teacher. I suggest below some points to be noted in teaching and reading. (a) It is important to discuss and anlyse different points of view. These discussion problems are important. while the last four are more difficult. On Teaching and Reading This book is meant for twosemester course for graduate students of physics. In each chapter.
e e2 = (4. books and articles in English should present no great difficulty.) Therefore it is also the task of the teacher to introduce some extra material where possible. 2. especially in broadening one’s views. Boltzmann’s constant k = 1. The new material is seldom very difficult in itself. nor can the problems be blamed to the lack of reference books. I must emphasise that with a clear understanding of the concepts and a proper foundation. The temperature T is regarded as an energy.98 cal deg’ mole’ lo4 K.05 x lo’’ erg sec [Note: 1 eV = h = 2nh [Note: 1 eV/h = 2. 1 cal/mole = 0. Planck’s constant h is also set equal to 1. Without clear concepts.42 x loi4 sec‘1 gm &m Proton mass = 1. Of course.11 x Electron charge = . Units and Constants For simplicity. and energy can be measured in temperature units. The following units and constants may be used for conversion.38 x = 8.80x 10’o)2 ergcm . (e) Difficulties encountered in the course can often be traced to a poor foundation in elementary concepts. These constants are nevertheless restored wherever necessary. (Nevertheless. Unfortunately.02 x loz3 mole’ Gas constant R = N k = 1. references are useful.62 x erg/K eV/K Avogadro’s number N = 6. It is extremely important to review elementary concepts periodically. It is indeed necessary to have clear concepts in order to benefit from the literature. for students who are not native speakers of English. the best ability in English would be of little use. we set Boltzmann’s constant to be 1.67 x Electron mass = 9. in order to exhibit their roles.xxii STATISTICAL MECHANICS references serving only as a source of specific information.5 K/molecule] Planck’s constant h = 1. so k T becomes T . there is an unavoidable language barrier.
INTRODUCTION xxiii Bohr radius a = 0.1 . Mathematical Formulas erglgauss. The formulas listed below are not meant to take the place of a proper table..529 Binding energy of hydrogen = e / 2 a = 13. = 0. x=o x<o .27 x 3. . (b) Useful formulas 1 : dxxneX = F(n+ 1)= n! .6 eV Bohr magneton p~ = efiJ2mc = 9. sgnx = . . Ix I d sgnx dx = 26(x) X . (a) 6function e ( x ) = 1. They may nevertheless be useful when a table is not at hand.' 2 x>o .
.i l z k 2 a ’ .((3).202 .I  x’dx  . m C = 0.s dxex2/~a2 eikx = f i a e . ((3) 2 3 = 1.5772 =  I dx e’ lnx 0 .xxiv STATISTICAL MECHANICS InF(1 + x ) = CX + qXz) .. N ! = N N eN (271N)” [ l 1 1 + + + 12N 288N’ ..
. 12! = 4. 220 x lo6 . 1. 6! = 7 2 0 . In 10 =2. 215 = 32768 .30 J= .693 .77 . 9! =362880.INTRODUCTION xxv (c) Useful numbers In2 = 0.8 x lo8 . 2" = 1024 . .
.
pointing out the interaction between molecules as the origin of equilibrium. This part lays the preparatory ground work. with special emphasis on the facts that equilibrium is not an instantaneous state and the observation time is important. Various basic concepts will be analysed again in the following parts. Chapter 3 discusses the law of detailed balance. reviewing some basic knowledge in solid state physics and some elementary calculations in statistical mechanics. Chapter 4 discusses electrons in metals. The outstanding feature of thermodynamics is the appearance of entropy.PART I EQUILIBRIUM This part is divided into four chapters which introduces the basic concepts of equilibrium and provides some simple examples. . The first two chapters review the concept of equilibrium and the laws of thermodynamics.
an ancient painting hanging on the wall is an unchanging state. or two or three days. Suppose now we have an extremely long observation time. For example. l .g. Equilibrium and Observation Time Basic concepts always originate from simple phenomena. because of t h e chemial reaction vith air. Consequently. one second. to say this state is unchanging is only an approximation and is not absolute. when one speaks of equilibrium. equilibrium is a state characterizing the system over a long observation time and not an instantaneous state. Equilibrium refers to a state which is unchanging.. On the time scale of molecular motion.Chapter 1 EQUILIBRIUM This chapter discusses the fundamental concept of equilibrium with emphasis on the relation between equilibrium and the observation time. or even one year. 1 . We use a few simple examples to illustrate the meaning of equilibrium and to discuss some of its related problems. To discuss equilibrium we must consider the approximate length of the observation time. this painting can be said to be in an equilibrium state. If the observation time is short. From the date of its creation to the present. Equilibrium is the most basic concept of thermodynamics and statistical mechanics. one must take into consideration the problem of the observation time. e.g. some pigments have changed and the paper has become pale.. For example. because during the observation time its change is rather minute. half a second. However. this ancient painting has undergone many changes. one hundred years 01 two hundred . e.
For example. the cup of water can be regarded as being in equilibrium. The specifications of this observation time depend on the requirement and interest of the observer. but if the time is long enough. not to mention the change in the hanger and the attached wall. but we can use different methods t o prolong the period of equilibrium. because its change is rather obvious and cannot be neglected. However. . The temperature of the water can be measured by a thermometer. water molecules will continually evaporate from the water surface. then again this empty cup can be regarded as being in an equilibrium state. Therefore. otherwise equilibrium will be meaningless. But absolute isolation is impossible and there is always some transfer of energy. and a more precise definition and better understanding will emerge from an analysis of more examples. after a few seconds the water in the cup becomes an unchanging state at rest. If the time is too long. and the state of the system does not change. even this cup is not in an absolute unchanging state. After a few days. Within one or two seconds. After an hour. then the temperature of the water will accordingly remain constant. In several hours the volume of the water will not change significantly. If the room temperature is constant for several hours. and any substance under tension or shear is a metastable state. This definition is somewhat ambiguous. then this painting cannot be regarded as being in an equilibrium state. Consider another example. the temperature of the water will be equal to the room temperature. The volume of the water in the cup can be measured from the height of the water and the crosssection area of the cup.EQUILIBRIUM 3 years. all the water will evaporate. and is not absolutely stable. What change is of importance and what may be neglected depends on the aims of our analysis. However. These examples point out that the observation time cannot be too large. strictly speaking. If we pour some boiling water from a thermos flask into a tea cup. if the observation time is within this range of several hours. these measured quantities will not change significantly. this cup of water cannot be regarded as being in an equilibrium state. to discuss equilibrium we must specify the observation time. Generally speaking. because the molecules of the cup can evaporate. or pour the water into a thermos flask rather than a cup. and this is an equilibrium state. If the observation time is not too long. the temperature will obviously change and the water cannot be regarded as being in an equilibrium state. such as two or three days. There are no absolutely unchanging things in this world. and thus during this observation time the water is in equilibrium. So if the observation period is over two or three days. we can put the ancient painting in a glass case. although in a few years’ time there will be n o change observed. we call this an equilibrium state. when all the water has evaporated. Notice that there is no permanent container.
it must be recognised that the painting of the previous section. i. Equilibrium and Molecular Motion Equilibrium is a macroscopic phenomenon. we introduce the notion of an observation time right from the beginning. on the scale of the size of the molecules. equilibrium is a very special state of motion in which. blowing winds and falling rain drops are examples of nonequilibrium phenomena. . To understand nonequilibrium phenomena. any phenomena changing with time can be called nonequilibrium phenomena. However.. we have emphasised the intimate relationship between equilibrium and the observation time in order to develop a correct viewpoint about the range of applicability of statistical mechanics. Large Scale Change and Local Equilibrium In the previous section.e. the water molecules in the cup as discussed above are executing very complicated motion and none of the molecules is stationary. times much longer than the time of molecular motion. From the microscopic point of view. the macroscopic properties are unchanged. Such approaches have their merits in providing a simpler and clearer introductory discussion. i. To understand equilibrium. referring to the state of an absolutely isolated system and an infinitely long observation time.e. The time scales. “Macroscopic” refers to a length and time scale much larger than those characterizing the molecules.. we must also consider nonequilibrium problems. it is essential to understand the molecular motion. As equilibrium is an unchanging state. discussed in the previous section. is composed of many molecules. This may appear to be cumbersome at first. we have emphasised that equilibrium is a large scale phenomenon. we must analyse the factors maintaining equilibrium and also the factors causing changes. 1.4 STATISTICAL MECHANICS 1. The conduction of heat and electricity. There are also more abstract definitions of equilibrium. In the previous section. the reader is somewhat hampered. Similarly. which vibrate continually.3. Most textbooks start with the concept of absolute equilibrium. The aim of statistical mechanics is to understand the relation between equilibrium and molecular motion. are macroscopic times. On the scale of a molecule. The so called “absolute equilibrium” is an idealisation. within the observation time. The macroscopic phenomena and properties that are observed are the time average of these very complicated molecular motion during observation. Hence to understand equilibrium. but it will provide a greater flexibility in the range of applications. it is not an instantaneous property. In contrast.2. From the microscopic point of view. since many problems in statistical mechanics become nonequilibrium problems under this idealised definition. the propagation of sound.
p+p(vV)v=VP at av 3 where p(r. t ) is the pressure and pv is the gas flow. The differential volume d 3 r = dxdydz is likewise a rather large volume. d t >> 7 .. i. from the large scale considerations. larger than the length scale of molecular motion d x >> A .dap I .e. The first equation expresses conservation of mass and the second is Newton’s law relating the change of momentum to the applied force.. i. p(r. t ) is the flow velocity. We neglect viscosity for simplicity. the momentum density. The integral d3r is over all points in a certain region and d a is an integration over the surface of the region. Within an observation time dt. (1 2) where r is the time scale of the molecular motion which may be taken t o be the mean free time. v(r. aP + at V(pv)=O.EQUILIBRIUM 5 but a property associated with a long observation period. i. The following equations describe gas motion on a large scale.3 ) . The meaning of (11) can be seen more clearly by integrating the equations L1pd3r at ajpvd3r at + Jdovp + = 0 . local equilibrium. the observation time cannot be too large.e. Now we use the equations of fluid mechanics as an example to further emphasise the importance of the scale.e. for otherwise equilibrium cannot be maintained. But if the time is much longer  (1 . i. On the other hand. the gas in a volume d 3 r is approximately in equilibrium. Notice that the scales of time and length in (11) are understood to be macroscopic and the differential d t is regarded as a rather long macroscopic observation time.e. vpv = . do.. X rv where v is the average speed of the moluecules. the free flight time between two consecutive collisions.. where X is the mean free path. t) is the mass density at position r and time t.
as described by (11).the changes of p . The socalled equilibrium properties of matter denote the structural as well as the thermal. it is an amorphous magnet or spin glass. Different compositions and motions result in different structures and properties. but each individual entity does not combine with others. as illustrated by some examples which we shall discuss later. Equations (12) and (13) point out the range of application of (11) . If the nuclei cannot move freely. Many metals and alloys at low temperatures are superconductors with zero resistivity. then it is ferromagnetism. If the arrangement is orderly. then changes can occur and n o n equilibrium appears. then it is a conductor. Low temperature liquid helium (near absolute zero) is a superfluid without viscosity. atom or molecule).. If the alignment is alternately up and down.6 STATISTICAL MECHANICS than d t or the volume is much larger than d3r.e.4. If the arrangement of the atoms is random. we have a solid. and to a certain extent can be created and controlled in the laboratory. Materials science is an active field and synthetic . 1000h? In principle this can be answered from microscopic analysis. Now we briefly review some special features of the structure of matter. pressure and magnetic field.e. Otherwise it is an insulator. The external conditions are described by physical variables such as temperature. we have a fluid. Structure of Matter Under different external conditions. The main task of statistical mechanics is to understand how these properties emerge from the molecular motion. but in fact these problems are very difficult. to deduce the large scale equations of motion from the molecular motion. Each electron has a spin with a magnetic dipole moment. If one or several nuclei attract several electrons and form an entity (i. it is called antiferromagnetism. electrical. we have a crystal. Matter in the usual states can be regarded as a collection of nuclei and electrons. There are millions of different molecules. but moves independently. whose magnitudes depend on specific situations. i. In a magnetic field the spins may align with the field and this is paramagnetism. If there are free electrons in the solid. but what is the lower limit: 107. If this alignment does not require an external magnetic field. or elastic properties etc. If the arrangement is random. A point to be noted is that (1 2) and (1 3) are not very precise. 1.. The above are only some possible structures. crystals and plastics. It is no doubt better to have larger d t and d3r. matter has different equilibrium properties. 1Oh or IOOOT. we have amorphous substance or a glassy state. v and p must be rather smooth on a scale dictated by T and h.
Iron loses ferromagnetism 1 043 K 100K   Water boils 373 K Ice point 273 K C 0 2 becomes dry ice 94.17 K .EQUILIBRIUM 7 rable 11.5 K 90 K 12 liquefies 0 10K  Lead becomes a superconductor 4He becomes a superfluid 7.19 K 2. Some examples of the change of the properties of matter with temperature (at one atmosphere pressure). Temperature 10000 K  Tungsten melts 3 653 K iron melts 1 808 K l000K .
. These changes are numerous in variety. From the ice point 273 K upwards: water boils at 373 K under one atmospheric pressure. Away from the transition temperature. “equilibrium” does not refer to the instantaneous positions and velocities of the molecules. aSee books on materials science. the more fuzzy is the concept of equilibrium. e. but rather to the state of motion over a long period of time. This structure is an equilibrium property. then these average values change. Consider again the example of crystal. These are examples in which the properties of matter change abruptly with temperature. iron melts at 1 808 Kand tungsten melts at 3 653 K . but the fixed impurities or cracks in a solid are also equilibrium properties. the various average values do not change.5. The observation time of an equilibrium state is not instantaneous.g.19 K and helium becomes a superfluid at 2. The equilibrium properties denote the average value of physical quantities over this time. and the corresponding temperatures are called the transition temperatures. Not only are the average positions of all the molecules in the crystal equilibrium properties. The smaller the limit. 1.a The structure and properties of matter change with temperature. (3) The characteristic of equilibrium is that these average values do not vary within the observation time. Tang (1971).g.17 K. carbon dioxide becomes dry ice at 94. Such abrupt changes are called phase transitions. e. But if the observation time is allowed to exceed a certain limit.. but is relatively long. iron loses its ferromagnetism at 1 043 K. pressure and other parameters describing the physical condition. (4) We regard all such average values as equilibrium properties.8 STATISTICAL MECHANICS materials are constantly being developed and utilized. lead becomes a superconductor at 7. Supplement and Conclusions We now summarise the above discussions with a few further remarks: (1) From the point of view of the molecular motion. and not just the few large scale quantities. (2) The scale of the observation time is determined by the observer. the structure of a crystal is determined by the positions of all the nuclei. The state is then no longer regarded as equilibrium. The position means the average position of the nuclei within the observation time. a small change in temperature produces only a small change in the properties of matter. For example. Every nucleus is vibrating and its instantaneous position is continually changing.5 K. We list some of the better known examples in Table 11. if we double or halve the observation time. . From the ice point downwards: oxygen liquefies at 9 0 K.
can the liquid helium flow out of the cup? 5 . i. The reader should consult a suitable book on this subject. and the box is at room temperature. On everyday time scales the two have very welldefined equilibrium properties. the distinction between large and small scale. 4 and 5 lies in the different time scales. which is more stable? (Answer: Graphite is more stable than diamond. The reader can refer to solid state physics data and find out their energies. 4. If instead of helium. cannot always be well defined.17 K. If the temperature is lower than 2. In short. there is water in the cup.EQUILIBRIUM 9 (5) Some equilibrium properties are not related to the molecular motion in an obvious way. Carbon has two common crystal structures: graphite and diamond. (6) The above discussion of equilibrium is not complete. Temperature and entropy (which we shall discuss carefully later) are the prominent examples. A more precise definition and a better understanding of equilibrium require additional concepts such as detailed balance and the independence of the motions of parts. Question: Under an infinite time scale.. If the temperature is slightly higher than 2. 3. Problems 1.) 2.e. Moreover. 11 Hint: The main difference in Problems 3. They are often complicated and fuzzy. between the macroscopic and the microscopic. what happens? Fig. There are many crystalline forms of sulphur and the structures of its liquid and gas phases are very interesting. Figure 11 shows a cup filled with liquid helium in a box at low temperature. .!7 K the liquid helium in the cup will flow out until the liquid levels are the same inside and outside. the topics and concepts we shall analyse in equilibrium are not always well defined. Consult reference to understand this phenomenon.
Equilibrium and Reversible Processes The basic assumption in thermodynamics is that under a specific set of environmental conditions. For example. This assumption is valid under many situations. many properties of this piece of iron will change.Chapter 2 THE BASIC CONCEPTS OF THERMODYNAMICS Thermodynamics is mainly concerned with the exchange of energy (heat and work) and the equilibrium properties. It does not consider the structure of matter. To understand this irreversible process and . but analyses only the macroscopic behaviour. This assumption has an important implication.1. if we heat a piece of iron to very high temperatures and then suddenly cool it. so this process is irreversible. the equilibrium state is defined. a given system has definite equilibrium properties. On the other hand. then the system will return to the original state. etc. The environmental conditions are determined by the external factors such as temperature. Different rates of heating and cooling can result in different structures. this cup of water will return to its original state. Once these factors are fixed. Therefore this particular process is reversible. if we heat a cup of water originally at room temperature and then cool it. namely the existence of reversible processes. This implies that if the external conditions of a system are changed and then restored. If it does not return to the original state. pressure and magnetic field. 2. then it would imply that under the same conditions we could have different states and the process is irreversible. but there are exceptions. This chapter reviews the basic assumptions and concepts as a preparation for the discussion of statistical mechanics.
then the work done is dW = p d N ..2) If we fix one end of a rod and apply a torque 17 to the other end. we must study the detailed change of the structure. a beaker of water contains N dissolved oxygen molecules. e. Of course. smoothness or abruptness depends on the time scale. if the time scale of change is much longer than the time scale of the motion of the molecules. while abrupt changes are usually irreversible. is dW = f d X . the work done. smooth changes are usually reversible. 2.e. Work and Heat The energy of an object may be changed by external factors. For example.4) The negative sign in d V denotes the decrease in volume due to compression.THE BASIC CONCEPTS OF THERMODYNAMICS 11 the property of iron. compress it. dW. the energy increases by We give two more examples of work in thermodynamic systems: the pressure p is the “force” that causes a change in the volume V . then the process can be called smooth. change of coordinates). If dNoxygen molecules are pushed into the water. (25) . turning the rod through an angle d o . (2 . or add an external magnetic field. As a second example. We now discuss only reversible processes. The oxygen molecules above the water provide a “force” on the number N below.2. (21) The bar on d means that dW is only an infinitesimal quantity and not the exact differential of the function W. For example. Roughly speaking. These external factors can do work on the object and change its energy. The increase in energy of the object is given in terms of the increase in the magnetic moment d M dW = h d M .g. The chemical potential p is the “force” that causes a change in the number of particles N . The work done is given by the product of force and displacement (i. etc. and this is described by a chemical potential p. by a tension f pulling an iron rod by a distance d X . we can stretch an object. and the corresponding work is dW = . (2 . let consider the work done on an object by an external magnetic field h . Generally speaking..p d V .
:ad is very important in the discussion of chemical reactions. high birth rates and longevity increase the population pressure. The presence of C exerts a “force” to increase the population of A and B . These generalised coordinates are large scale coordinates or thermodynamic coordinates. the magnetic moment M . and C decomposes to A and B A + B . the torque 77. The length X . thus causing a decrease in the flow of population into the city. The chemical potentials of various molecules determine the direction of the reaction. The small scale coordinates expressing the detailed motion of the molecules do not appear in thermodynamics. and they are called generalised coordinates. the particle number N are all “coordinates” characterizing the state and/or the configuration of the system. But we need not analyse the coordinates of every molecule. each species of particles is characterised by its own chemical potential. so we add a few remarks here. B combine to form C. similarly A and B tend to increase the population of C .C . the small scale coordinates exist and are very important. The most outstanding feature of thermodynamics is to introduce a new generalised coordinate to express the . This shows that the small scale motion cannot be ignored completely. The chemical potential is a “force” increasing the number of the particles. Early marriages. Energy can flow out of or into a system without the above generalised coordinates showing any change. Once the urban population increases. In equilibrium the chemical potentials on the two sides must balance. and the chemical potential p are external “forces” capable of changing the coordinates and hence doing work. jobs are fewer and the population pressure in the city rises. The development of a city creates more jobs and its population pressure is lower than the surrounding country and people will rush into the city. the living condition deteriorates. the turning angle 0 . The notion of chemical potential is analogous to that of population pressure in social science. In general. We shall discuss this in detail when talking about chemical reactions.12 STATISTICAL MECHANICS Similarly nitrogen and water vapour in the air have their own chemical potentials p’ and p ” . The chemical potential is not a familiar concept. the magnetic field h . The tension f. Suppose A . Increase of employment rates and improvement of living conditions will decrease the population pressure. Such energy flow is heat and is the result of the change of the small scale motion. Of course.
then during the process of energy change. If the process is very quick. some materials.1) to ( 2 . then the heat flow is small. Each point in the ( X . from A . Absolute heat insulation is. reviewing some terms and definitions. The increase in energy is Why does T turn out to be the absolute temperature? This can be said to be a coincidence.THE BASIC CONCEPTS OF THERMODYNAMICS 13 small scale motion in a collective way. Others such as metals are very bad insulators. In addition to insulating materials. of course. such as stone or asbestos. or can be considered as a choice. due to the change of the small scale molecular motions. In other words. Starting from the concept of an adiabatic process. The change of total energy d E is the sum of the two: d E = dW + dQ . 21 denotes X and the vertical axis is the total energy E. (27) This is the first law of thermodynamics. impossible. we first consider a single coordinate X of a physical system. n o heat flows into or out of the system and this is called an adiabatic process. and the curves are adiabatic. According to experience. because this is the work done by external “forces” t o change the large scale coordinates. This law is a statement of the conservation of energy and tells us that there are two ways of changing the energy of work done on the body or heat flowing into the body.3. and the “force” responsible for increasing this coordinate is the absolute temperature T. This new generalised coordinate is entropy. Adiabatic Process and Entropy If dQ = 0. it is not easy to ensure negligible heat exchange. we can reach B or any other points on the same curve by an adiabatic process. The horizontal axis in Fig. are rather good insulators. E ) plane represents an equilibrium state. 2. The sum of the energy changes in (2. We now carefully discuss this problem. To be concise. . Similarly we can start from A’ and reach B’ or any other points on the curve A’B’ by an adiabatic process. To change from one curve to another requires heat exchange. denoted by the symbol S. we shall develop stepbystep the concept of entropy. But if time is prolonged.5) is called the work done dW. We call (26) the heat dQ because this is the heat flowing in. the amount of heat flow is also controlled by the time duration of the process.
(210) . E ) 3 = u is also possible. g(X. (29) determines the adiabatic process characterised by u . If X is constant. 21 Adiabatic curves in the X . e. Note that its definition is still to a certain degree arbitrary. we can use another function to represent these curves. If the entropy is a constant. Here g(X. Each u determines a curve.14 STATISTICAL MECHANICS E X Fig.E plane. then the change of energy is equal to the heat flowing in: dQ = d E = (Elx do . The only condition is that the value of this function is constant on the same line. where points with the same height are joined together.u) . E ) . E ) is the “height” at the point ( X . (28) Each curve is determined by a constant u. (29) That is to say. the energy E of the system is a function of the coordinate X a n d the entropy u. These curves are in fact the solution of the differential equation d E = f d X where u is the integration constant. The conclusion of the above discussion is that we have introduced a new quantity u which will be called the entropy. formally E=E(X. From (28) we can solve for E.. E ) : g(X. Of course. These adiabatic curves can be represented by the lines of constant values of a function g(X.E) = u . This situation is like the contour lines in maps.g.
5).. 22(a). i. x k denote the various “forces” and coordinates in examples (21) to (2 .e. For example. and is also the pressure inside the body acting against the external “force”. These physical considerations imply that entropy must be a monotonic increasing function of E. A line curving back like that in Fig.. because the value o f f would then be + OQ or . we get dE= fdX+rdo . But f = d E / d X is the slope of the curve and is physically the external “force” at equilibrium. If each equilibrium state has only one unique value off. Two questions remain to be discussed: (1) The curves in Fig. If there were an intersection such as point C in Fig. the “force” inside the body equal and opposite to the external “force”. . the pressure p is a “force” acting on the body. then the intersection in Fig. Hence entropy can be considered as a new coordinate like x k . then f can have two different values at C.11) The above analysis can be slightly generalised to include all the coordinate changes : (212) where fk. 22(a) is not possible. d o = . thus maintaining equilibrium. 7 dQ (2. 22(b) is also impossible. To say that under the same equilibrium condition we have two different values o f f is contradictory to experience.THE BASIC CONCEPTS OF THERMODYNAMICS 15 Combining this with the first law given by (27). and r is a new force like fk . 21 do not intersect. In fact we have already assumed this.
22(a) and (b) would be possible and the above definition of entropy is unreliable.4.. If there are other coordinates whose changes are unknown to us.(212). see (29) .e. the entropy u and the variable T = (g)x. Hence we have the freedom to choose a suitable u as the entropy. (213) where A ( u ) must increase with u but is otherwise arbitrary. 21) and we introduced a new coordinate and a new “force”. The reader may ask: why not simply use the temperature as a coordinate? Why introduce the concept of entropy? .e. we can still define another function :such that 0 = A(u) . the equilibrium state is completely specified by a small number of coordinates) ismost important. This condition is very weak.16 STATISTICAL MECHANICS E E Fig. then Figs. 22 (a) Intersection of two adiabatic curves. A hidden worry in thermodynamics is: How many coordinates are necessary to determine the equilibrium state? How do we know that we have not missed any? (2) The above definition leaves some room for arbitrariness.. d Q = ?do . E ) so that u is defined by (28). Notice that the basic assumption of thermodynamics (i. If we find g ( X . Absolute Temperature and Entropy In the last section we discussed an adiabatic process (see Fig. i. 2. The validity of this basic assumption depends on the problem under consideration. (b) An adiabatic line curving back.
We define u and r as above and similarly ul. But then the distinction between heat and work will not be so clear. To treat T as the “force” and entropy as the coordinate yields a convenient conceptual framework. then the temperature is the same throughout.) Therefore. if it is in equilibrium. Fahrenheit. the socalled physical variables describing the surrounding must also describe the various parts.. If we focus on one part of the system. r . The most important variable of the surrounding is the temperature. 2 . Let Xlk. if temperature appears as a coordinate. 3. ul. Then dQ = dQl + dQz.r l . Of course the .e. i. although u changes with Xlk. (Of course. i. and u2 . we can say u=u(u1. dQl flows into part 1 and dQz into part 2. a certain amount of heat dQ enters this system. be all the coordinates of part 1 . because no matter how complicated or how inhomogeneous the structure of the system is. (215) Note that there is no other variable within the parentheses. A system in equilibrium has an important property: each and every part of it is in equilibrium. it follows that u. and u becomes the entropy S as commonly understood. T~ are also functions of the temperature 8 and all the coordinates. .. In principle there is nothing wrong in using temperature as another coordinate.e. then the temperature may be different. etc.) What is the relevance of this to u and r? Consider a physical system as composed of two parts 1 and 2. Out of this. Therefore. Now we consider some special features of equilibrium to further specify the definition of entropy so that r becomes the absolute temperature T. X Z k and 8. (ul and uz are functions of X l k and X Z k respectively.THE BASIC CONCEPTS O F THERMODYNAMICS 17 Actually all the coordinates plus temperature can determine the equilibrium state. Moreover. This differential formula shows that u is a function of u1 and uz . then the remaining parts become the surrounding. Since an equilibrium state is specified by the temperature and the other coordinates. and also functions of 0 . Similarly X Z k are all the coordinates of part 2 . r2 for the two parts.r 1 and u2. . this change is determined through the changes of u1 and u2. k = 1 . We use 8 to denote temperature measured on any convenient scale: Celsius. 21. entropy will appear like a force. if some part is isolated so that heat exchange cannot occur. Suppose in a certain process. . We can then use an isothermal process to draw curves in place of those in Fig.u2) .
a function of 8 and X l k only. but T ~T~. and 7 contain the same T(6). ul. but T ~T . and S’ may be different. Note that T~ cannot be a function of u2 because Equation (218) also shows that 7 = T(u. For the same reason. The functions S: . it can be applied universally to any body. e) (2. once T(8)is determined. 71 =7i(0i7e) TZ 3 = 7 z ( ~ 20) . from the second equation of (216) we know that T is related to only through u1 . T ~ T~ . combined into a single one. Therefore.19) because the property of the parts is shared by the entire system. and is not related to X z k . (220) That is. Hence T ~ / T r2/7 will not depend on 8 as a variable.themselves are directly related to 8. T~ (2. The only possibility is T1 = T ( e )SI(O. . Therefore. ~7 . Hence 7 = 7(e.) Now we note that the ratios r1/7and r 2 / r in (216) are not functions of 8. the relation between T = T~ / U l ( u l . we discover that u2 and 0 . their relations with B is through a certain function T(8). Substituting this result into (216). Since any two systems can be . (217) are functions of ul.18 STATISTICAL MECHANICS differential of u with respect to u1 or uz can also be a function only of u1 and (22 : (216) where T ~ is.18) is only related to X l k . S. .uz) . uz) and x z k is only through u2. (Part 1 can be considered to represent almost the entire system.) .
.THE BASIC CONCEPTS OF THERMODYNAMICS 19 Now we modify the definition of entropy. it is convenient to require S to be sum of the S j : s = 1si /=1 111 . If T is replaced by T and u by S i n (214). We now treat (222c) as an imposed condition. The additivity of the entropy is not straightforward or obvious. It is easily satisfied in actual calculations. Once the unit of T is established.22 a) where T has been cancelled on both sides. 2. bT= S. and divide S. the unit of S is fixed. How do we choose the integration constants? As d S is the sum of the d S i . If the ice point is fixed at T = 273 K. dQ = r d u = TdS . The Ideal Gas Now wc apply the above analysis to the case of an ideal gas. Our aim is not to derive the properties of the ideal gas. We can also use other units as well. S(u)= jS’(u)du .22b) Integrating each d S j we get S j with an undetermined integration constant. the equations are unchanged. But S still has an undetermined integration constant. This result can immediately be generalised: divide the system into m parts. The unit of T ( 0 ) has yet to be chosen. S). S’ : and by C.5. We define S(u) as the entropy in the manner of (213). The above analysis of the entropy is rather abstract. then one obtains the usual absolute temperature scale. (2 2 1) and call T ( 0 ) the absolute temperature. Now let us consider a n example. but to illustrate the abstract ideas of the preceding section. (2 22c) ~ I f we impose (222c). then dS= 1dSi i1 m . then the integration constants of each dSj cannot be chosen arbitrarily.dA/du = S’(u). but there are some details which we shall illustrate in the examples in the next section. (2 . w h c h we shall discuss later. then dS=dS1 +dSz . (2 . because in (220) if we multiply T(B) by a constant C.
while the other half move away from the wall.') = ( $ ) v 2 . (223) Its origin is as follows.24) __ . The distribution of the directions of the velocities is isotropic and hence v z in the integral has been replaced by + v. Let v be the velocity of a molecule and vx be the component perpendicular to one wall of the container. (We have not yet introduced entropy. (227) where u is an integration constant. i.26) with the solution EV2I3 = u .2 E 3 V ' where d3vf(v) is the density of the molecules with velocity v and d3vf(v)vX is the flux in the x direction. Each collision results in a momentum change (impulse) of 2mvX and the pressure is the impulse per unit time per unit area on the wall: p = .[d3vf(v) 2 'S 3 my2 1 2 (2. (225) Substituting in (224). 2E P=w.20 STATISTICAL MECHANICS There is a simple relationship for the pressure p . Eaqh value of u determines a curve in the . only half of the molecules move towards the wall.. The factor is due to the fact that on the average. The validity of (224) does not depend on the distribution functionf(v).we get (2. the total energy E and the volume V of an ideal gas.d3vf(v)vX 2mvX 2 = jd3vf(v)mvi = . + Y.e.) The curves drawn from the adiabatic processes are determined by 3 ($)(vi dE=pdV . Suppose the number of molecules N is constant.
nonetheless. We can. However. Therefore. use experimental results. then we obtain from (229)and (220) PV kT=N S'(u) =  . but its relation with the motion of the molecules is not clear. from the T in (228) and the experimental result (229) of the ideal gas.we obtain (230) Hence. From (227) and (223). As pointed out in the last section.THE BASIC CONCEPTS OF THERMODYNAMICS 21 (V. T must be the absolute temperature T ( 0 ) multiplied by a function S'(u) of u (see (220)). The concept of temperature has not yet appeared. (228)can be written as (231) Let us define the absolute temperature as f ( 0 ) l k = T(B). However. According to experiments. This fact yields the relation (2 . we can derive the absolute temperature. Since u defines the entropy. the temperature is different. then the absolute temperature T(0) is fixed. (228) Now we have the coordinates V and u and their values specify the equilibrium state of this ideal gas. we have to determine 8 first. The discussion in Section 24 showed that given a measured temperature 0 . The total energy of the gas is E = uV'/j. The steps are as follows. we can calculate pVand together with (229). its determination is easy. for a dilute gas of N molecules. the combination p V / N is unchanged at a fixed temperature. The functionf(0) can be determined experimentally on the scale chosen.2 9 ) where 0 can be measured on any scale. (231') (232) 'T 2) ( * . The pressure and the energy are precise concepts in mechanics and indeed this enabled us to find their relationship (224).E ) plane (see (28)).
If we now connect the two volumes. then k=l . (2 . Suppose we have two containers (1 and 2) of gas at the same temperature and pressure. with N1 and N z molecules and occupying volumes V1 and V. The entropy of each gas . where E/N is the average energy of one molecule and N/V is the density. S must be proportional to N. Here k is a constant determined by the unit of T . N is thus a constant. we obtain 3 S’(o)do= . Now we mention the problem of the entropy of mixing.. The combination pV/N has the same unit as energy. (235) The constant in the equation can be a function of N. (2 33) This is the Boltzmann constant. Since (225) has not taken d N into account. Let V = V1 + V. The choice of k is a choice of units and is not unique. so under a fixed temperature and density.22 STATISTICAL MECHANICS The absolute temperature can be measured by p . If we substitute (227) into (235) and write the constant in (235) as NIn(C3” N ..34) In the following chapters. if T has the usual unit. V and N in this way. we shall use k = 1 for simplicity. From (232).38 x erg K . The relation of the existing temperature scale and the absolute scale is established by the experimental function f(0). N = N . If we simply use the unit of the energy to measure T . In fact we do not observe any change (assume that the two gases have the same colour). The pressure and temperature will remain the same after mixing.N I n o 2 + const. The constant C is independent of N because S must be additive. (236) is the entropy of an ideal gas. the gases will mix.j “ ) then .’ . E = E l + Ez . then k = 1. + N z . S = Nln[(V/N)(CE/N)3’2 ] .
This example of the entropy of mixing illustrates that if a gas is composed of the same kind of molecules. (238) After mixing. of total volume V . For this case of one kind of molecule. We have said before that any system can be divided into several parts and we can calculate the entropies separately and then add them up to obtain the entropy of the entire system.S = Nlln+ NzlnNZ . the energy is unchanged but each gas occupies the two containers. This point will be reconsidered when we discuss the independent property of the various parts of a system.N l n e C + Nln 3 2 V N .THE BASIC CONCEPTS OF THERMODYNAMICS 23 before mixing can be calculated from (236) 3 Vl S1 = N1 l n E C + N l l n  2 Nl . (237) The sum of the two entropies is S = S 1 +Sz = . and is called the entropy of mixing. so their entropy is S i S i = NllneC = NzlnEC 3 2 3 2 V + N l l n N1 V + N z l n N Z 9 (239) Their sum S’ = S + S is larger than S : i i N N (240) S ’ . then the answer is simply given by (238) because for molecules of the same kind there is no problem of mixing. . then we cannot divide the molecules into two parts and use (239) to calculate the entropies of each part and then add them. 3 VZ S2 = Nz l n E C + N z l n 2 N Z . N1 This excess comes from mixing.
but they all have additional side effects in wasting some energy and increasing the temperature (above that of the environment). The content is the same.24 STATISTICAL MECHANICS 2. If it decreases. Later we shall reexamine this law. If heat is to be moved from low temperature to high temperature. we can supply heat to increase the entropy and the temperature. Therefore. it has really never been proved. Suppose an ideal gas is contained in a heat engine.e. The Second Law of Thermodynamics This law is different from the general conservation laws.3 ) : (a) adiabatic expansion. 2 . It states that no engine can continually absorb the internal energy in matter and convert it into work. (b) isothermal compression. performing work. This law may be considered to be empirical. If it does no work. then we would be able to transfer the energy from the sea water to the steam engine so that the steam engine can do work. Newton’s laws or the laws in electricity and magnetism. In other words.6. such as the steam engine using the energy of the steam to do work. absorbing heat and performing work while in . Every conclusion above can be regarded as a restatement of the second law. work must be done. violating the second law. As another example. It is a very special law in physics. i. Hence we can continually convert heat into work. wouldn’t it be nice if an engine could be designed for a ship which extracts energy from the water molecules in the sea to propel the ship? It would then be unnecessary to use steam or oil or nuclear power! But the second law states that this is impossible. releasing heat and having work done on it while in contact with a low temperature reservoir of temperature T I . Then we perform work by an adiabatic process to use up the heat and let the entropy decrease again. an isolated system cannot become half cold and half hot. For example. This heat engine has four processes (Fig. From this law we can immediately draw the following conclusions. Up to now it has not been known to be violated. The Efficiency and Speed of a Heat Engine Let us review the ideal heat engine. Of course. engines do convert internal energy into work. there is no machine whose sole effect is to convert internal energy into work.. The most important conclusion is that the entropy change of an adiabatic process cannot decrease. The second law will then be violated. the Carnot heat engine.7. but on the other hand. (d) isothermal expansion. a body in equilibrium cannot perform external work and its entropy must be a maximum. (c) adiabatic compression with work done on the engine. 2.
. is the heat wasted.e.3 The Carnot cycle : curves (a) and (c) represent adiabatic processes. the temperature of the engine is assumed to be Ti and heat flows very slowly out of the engine. When in contact with the upper reservoir. For every cycle. Qz Qz ’ (242) i. the net work is w=QzQi . 23). the absorbed heat less the heat released. The definition of the efficiency of the heat engine is q== W 1 Q. curves (b) and (d) represent isothermal processes. Q.and heat flows very slowly into the engine. The increase of total entropy is (2 43) Hence (2 44) .THE BASIC CONCEPTS OF THERMODYNAMICS 25 I high temperature T2 0 lw o engine heat temperature 71 v . when in contact with the lower reservoir. and heat flow occurs very slowly. This engine is said to be ideal because all processes are assumed to be reversible. Fig. According to the second law. We now regard the high and low temperature reservoirs as a huge adiabatic body (the big box in Fig. contact with a high temperature reservoir of temperature T z . the temperature of the engine is assumed to be T 2 .. (241) i. 2 . Likewise. the above cyclic process must increase the total entropy of the reservoirs.e.
= T I A S . then the rate of heat exchange must be zero. Let a1 ET:/T. Therefore to attain this maximum efficiency. (1 1. 24) and A S = S”S’..Q.  Q. This is. = T. ( 2 45) The efficiency of the heat engine is (2 . If the heat engine has a temperature which is the same as the reservoir when absorbing and releasing heat. =. Of course Tz > T i > TI > T I is the necessary condition. depends on the practical considerations.T diagram for an ideal heat engine. Hence q = 1 . . In what follows we shall roughly estimate the relation between the efficiency and the speed. (see Fig. 1 ‘ LJ I I I I I I I I 5 Fig. 24 S . Qi T.IT? .46) The time spent in each cycle t is equal to the sum of the times spent absorbing and discarding heat: to:+T. T2 .26 STATISTICAL MECHANICS For an ideal heat engine. not practical. however.. of course.T I/T2 and thus the efficiency of an ideal heat engine is the highest efficiency possible. To increase the speed we have to sacrifice some efficiency.T: (247) . What is the best compromise? This.. the heat engine must be infinitely slow.T. Q . Let TI be the temperature of the heat engine when it is in contact with the low temperature reservoir and T i that of the heat engine when it is in contact with the high temperature reservoir..AS.
e. and Ql a T : . The above is just a rough estimate. the original ideal efficiency is not high.THE BASIC CONCEPTS OF THERMODYNAMICS 21 It is assumed in (247) that the rate of heat flow is proportional to the temperature difference. the original ideal efficiency is very high. :hen the maximum of q / t is approximately at a = ( T2 The efficiency at this value of (Y is then (249) Therefore the efficiency is greately reduced..and the value of q is (250) i. i. If T l / T 2 is close t o 1. Let us suppose we want to maximise the product q / t . oc T.e.. is very small.e. The value of q / t does not necessarily increase with the overall efficiency but this result does . = a. then the maximum of q/t occurs at about a % 1 t f ( I .T1/ T 2 ) . 25 Graphs of efficiency (9)and cycling frequency l / t versus a* . i. Figure 25 shows the efficiency q and the cycling frequency l / t versus a’.. half of the ideal value. If T1/ T . TI T~ TI (248) For simplicity we assume a1 = a. Since Q. to:+ Ti T. Fig.T.
3. Derive (248). Tension is a negative pressure. h is the external magnetic field and A is a . with B much smaller than A . and elaborated upon using a statistical mechanical interpretation. (b) If heat can flow directly from 1 t o 2. The reader may have learnt the theorem in thermodynamics stating that pressure must be positive. over a long period of time.. e. 2. The reader should review the problem of stability in thermodynamics. the tension can be positive and the thermodynamic properties are unambiguous and the various elasticity coefficients are also well defined. The discussion of thermodynamics should not stop here. T 2 . As mentioned in Chapter 1. the result will not be the same. (a) Calculate To and the work done by the Carnot engine. B are constants. respectively ( T I > T z ) . Let the heat flow be A ( A T ) + B ( A T ) 2 where A T is the temperature difference and A .(250). a body under tension is not stable.28 STATISTICAL MECHANICS show that the efficiency of a practical heat engine is far below that of an ideal one. Calculate the corrections to (248) . such as the “coordinates” and the “forces” in thermodynamics. The equation of state represents a relation among the thermodynamic variables. The contents of Sections 3 and 4 are the socalled “Caratheodory theorem” and its consequences. The above assumes C1 and Cz do not change with temperature. Problems 1.g. If the heat flow is not proportional to the temperature difference. during a short period of time. The equation of state of a magnetic body is M =A H / T where M is the magnetic moment. focusing on the concepts of “coordinate” and “force”. The above sections give a brief introduction to several basic concepts in thermodynamics. and including entropy and temperature into this group of concepts. Similar discussions are found in Wu (1 979). 4. p V = NT is the equation of state of an ideal gas. The reader should attempt to obtain a more mathematical understanding of Sections 3 and 4. prove that entropy must increase. Suppose a Carnot engine absorbs heat from 1 and dumps the heat into 2 until 1 and 2 reach the same temperature T o . what would be the temperature at equilibrium? What is the change of entropy? (c) If C1 and C2 are both positive. Two adiabatic fluids 1 and 2 with fixed volumes have heat capacities and temperatures C 1 . Further discussions are scattered in later chapters. T I and C z . 5 . However. and Wannier (1964). The equation of state can be directly determined by experiment.
Given a set of thermodynamic variable S. (Hint: from d E = TdS + hdM we can get a S p h = a M / a T ) . 6. etc. Consider the following example. Why? Answer: The larger sphere is hollow. This is not meant merely as a joke. the time scale. The reader should devote more thought on the coordinates. and in its applications we must consider many other details. One is ten times bigger than the other. there may not be a definite equilibrium state. This has been discussed in Section 21. Find the relation between entropy and h. The hidden worry of thermodynamics is: we do not know how many coordinates or forces are necessary to completely specify an equilibrium state. Two copper spheres with the same mass and at the same temperature are both under one atmospheric pressure.THE BASIC CONCEPTS OF THERMODYNAMICS 29 constant. Knowing the number of carbon atoms and that of iron does not specify the equilibrium state. . For example the property of steel depends on the arrangement of carbon atoms in the iron. Thermodynamics is not a complete science.
but also for each microscopic reaction. The collision of the molecules causing a change of momentum is a kind of reaction. we can determine some properties of the reactions from the known density distribution of the particles. Detailed balance can be understood from the symmetry of the arrow of time. i. the law of the molecular motion is unchanged.Chapter 3 THE LAW OF DETAILED BALANCE Molecular motion can be looked upon as a sequence of reactions and the reactions cause changes. Detailed balance can be applied in reverse. including a dense quantum gas. This . This is the main content of this chapter. we can calculate the density distribution of the particles from detailed balance and it can explain the relationship of temperature and chemical potential with the conserved quantities of the various reactions. the number of occurrence of each reaction in the forward direction is the same as that in the reverse direction. In principle it is capable of analysing any equilibrium state. If the structure of the molecules is changed by the reaction. The now commonly used Monte Carlo method is based on detailed balance for calculation. Detailed balance can be used as the starting point of statistical mechanics. If time is reversed. Equilibrium is an unchanged state and looks the same whether time flows forwards or backwards. we have equilibrium not only macroscopically. That is to say. The density in the socalled quantum gas is so high that the wave nature of the particles must be considered. The reaction in an ideal gas is very easy to analyse. Detailed balance implies that in equilibrium. Hence the number of reactions in the forward and backward directions must be the same. this is then a chemical reaction. If we add some plausible assumptions.e. Detailed balance can be applied directly to analyse the ideal gas.
it may appear in different forms. In using detailed balance.g. The quantum numbers I. m . Hence this is a three dimensional ringshaped container without boundaries. The statistics of the energy distribution L. However. the state of the electrons in an atomic model is the most familiar.1.6 eV . i. i.e. n y . x = 0 and x = L are the same side. The state of the particle is specified by three integers ( n x . e. (31) where 2 is the charge of the nucleus (in units of electron charge).e. (For the moment we do not consider the spin . We review the simplest situation and calculate the energy distribution of the states. Detailed balance cannot take the place of the basic assumption in Chapter 5. m are respectively the angular momentum and its projection in a certain direction. Each state can accommodate one electron. 3. two particles cannot occupy the same state.I. The Energy Distribution of the State Each state is a trajectory.) = n . z directions. In most problems. The energy of a state in a hydrogenlike atom is R = 13. A .a). and u is the projection of the spin of the electron in a certain direction. Each state is specified by a group of quantum numbers ( n . because we are as yet unable to make careful analysis on more complicated reactions. n. We shall discuss this in Chapter 25.et us first look at the states of a free particle. the state of the particles are determined by the potential energy. The above atomic model is to put Z electron into the Z lowest energy states which are determined by the electrostatic potential of the nucleus. The results of a quantum mechanical analysis are as follows. and similarly in the y .THE LAW OF DETAILED BALANCE 31 method utilises numerical calculations on the computer to simulate various reactions. This particle is confined in a cube of volume L3 and obeys periodic boundary conditions. Even for the ideal gas case. we have to assume the independence of all the reactions. we cannot avoid introducing the basic assumption. because electrons are fermions and must obey the exclusion principle.
32 STATISTICAL MECHANICS of the particle or its other internal states. and E not small..) The energy for n is EP= P2 2m .) Each configuration is a state. the configurations in (34) have very large n. The energy distribution of the states is If L is very large. ny and nz. . and Z is replaced by an integral: Finally the integration over n can be changed to that over p .nz = 0.. k 2 . .ny. n. (See (32)..) Hence (34) becomes . (3 2) The wavefunction in the n state is a plane wave (We take h = 1 to simplify the formula. . + 1 . so they can be treated as continuous variables.
# = . (Fig. p(x) is the momentum. The integration in (38) is along a trajectory in phase space. the wavefunction @ (r) is zero at the boundary. nv . where V(x) is the potential energy of the particle. =p(x)2 2m + U(X) . B.e. ( 3 7) Equations (37) and (32) are not the same. i. . but each trajectory must satisfy a quantum condition. ($) 312 sin xpx sinypy sin zp. The statistics of the energy in the short wavelength approximation In the short wavelength approximation (WKB approximation) the trajectory of the particle can be calculated from classical mechanics. The quantum condition is $p(x)dx = n h En .THE LAW O F DETAILED BALANCE 33 Notice that if we use the hard wall boundary condition. 31) P Fio ? 1 A traiertnrw in n h i a e  anire . h is Planck’s constant and n is an integer. However. the reader can prove that the g ( E ) calculated from (37) is the same as (36). Let us look at the situation of a particle in one dimension. . then the boundary condition does not affect gie). then n = ( nx . nZ). Provided L is very large.
over a distance of h / p . (311 ) can be generalised to L3g(e) = 1 ld3pd3r S(E h3  . Each integer specifies a trajectory and each trajectory a state. Therefore. 32 Two trajectories n and n’ in phase space. namely U(x) must be nearly constant within a distance h/p. Therefore. We only need U(x). In this region.8 ) is just the area bounded by the trajectory. 32 is specified by two trajectories n and n’. According to (38) nn’= 1 h p d x h . At the same time we can think of the state as points uniformly distributed in phase space with density l / h . but we must remember the limitation on its range of validity. to calculate the states we only need to find the area and then divide it by h ..V ( x ) P2 2m (312) . Fig. there are nn’ trajectories. i.~ Z (311) This integral (311) has the merit that it does not require solution of the trajectories or the value of each en of (38). Generally speaking.34 STATISTICAL MECHANICS The quantum condition (38) determines the various E . U ( x ) changes very little. Here h / p is the wavelength of the particle and the integral of ( 3 . Hence the short wavelength approximation is very convenient.e. this approximation is only true for states of high energy. The annulus region in Fig. This method is valid for short wavelength. (310) That is to say. If space is three dimensional. and the trajectories. the area of the region divided by h is the number of trajectories in the region. L 3 g ( ~ ) = ~ ~ p d xP S ( ~ .
certain motions of small amplitude can be described by special “particles”. (1 .) 3.2. the states 1 and 2 interact to change to states 3 and 4. but has a crystal structure.e. spin wave (or magnons. and the radius of contact (i. etc. This illustrates the content of the law and also emphasises collision as a basic ingredient of the ideal gas. e. we must add a spin quantum number to specify the state. but only in matter. We assume that each reaction is an independent event and there is no relation between successive reactions. The rate of this reaction is where fi and fz are the probability that there is one particle in the states 1 and 2 respectively and (1 . A gas may be said to be ideal if the collision time (the time the particles are in contact) is very short. the effective interaction distance of the particles) is very small. Because of collisions.while bosons include photons (spin 1) and 4He (spin 0). etc. making the analysis much simpler. without collisions there would be no thermodynamics.T H E LAW O F DETAILED BALANCE 35 If the spin of the particle is not zero. Consider a fermionic gas and let fk be the average number of particles in the k state. . if states 3. These particles are usually bosons.f4) are the probability that there are no particles in the states 3 and 4. In some solids or liquids. The Energy Distribution of Fermions We now use the law of detailed balance to calculate the energy distribution of the particles.g. (Notice that these particles do not exist in vacuum. the reaction cannot proceed. Each collision is a reaction. Bosons have integer spins (in units of h) while the spins of fermions are half integers (integer + +). The most usual fermions are electrons with spin+ . Ordinary matter is not homogeneous. phonons (for crystal vibration).e. i. 1+2++3+4 .f3). Hence the concept of ‘‘spin” must be replaced by the symmetry property of the structure of the matter. where k denotes the momentum and spin of the particles.g. 4 are occupied. (313) i. This rate must be equal to that in the reverse direction. This is a rather effective hypothesis. e.e. for oscillation of magnetic moment). each particle continually changes its state. Because the gas is fermionic.
(3 20) Hence. (321) . If time is reversed I1. namely (318) Equations (317) and (318) are very stringent conditions. To solve (317) and (318) we also have to consider the conserved quantities of the reaction. Before and after the reactions (31 7) and (31 8). Of course.e. (318): particle number = 3 . (316) Here.4) are interchanged. to be satisfied for all fk. (319) 1 fk then (317) and (318) are certainly satisfied.36 STATISTICAL MECHANICS According to quantum mechanics R = R’. there is another solution to (317) and (318). i. if all the fk satisfy  fk . 2 ) and 13. From (31 5) we get . which is unreasonable._. the energy and the particle number are conserved. and M becomes M * . detailed balance will give results similar to (31 7). But this means that each fk is the same independent of the energy of the states.2) and 13.constant . We can obtain fk from (317) and (318). R’=IM I = R . (317): particle number = 2 .f l f2  f3 f4 1 f1 1 f* (317) 1 f3 1 f4 If there is a reaction of three particles 1+2+3 + 4+5+6 . energy = el + eZ = e3 + e4 energy = € 1 + eZ + €3 = e4 + eS + €6 . because the equations of quantum mechanics are invariant under time reversal R =[MIZ * z . 4). M is the matrix element of the scattering operator between states 11.
i. Using (325) for f. The parameter a can be calculated from (324).e.we get (3 27) . the Boltzmann distribution. ( 3 22) Equation (321) of course means (3 23) This is the particle distribution satisfying the condition of detailed balance.e. i. Because g(e) a 6 (see (35)). then fk 2 1 e OIPEk (325) This is the familiar result of the ideal gas. The parameters a and /3 are constants shared by all the particles and should be related to the temperature and chemical potential. because in thermodynamics we know that E/N = T . The relations of the total energy E and the particle number N with fk are (3 24) If the gas is very dilute. and can be directly calculated from the energy of the states. and use (320) to get e3aP(E1 +EZ + E 3 ) = e3cxP(E*+Es+e6) . f k << 1..THE LAW O F DETAILED BALANCE 37 where a and pare constants. To see that this is a solution. substitute (321) into (317) and (318). we get (3 26) We discover that l//3 is the temperature T.
For general densities the entropy can be obtained by integrating the heat capacity: (333) .e.38 STATISTICAL MECHANICS i. To express this idea more clearly: We can make the potential energy U(r) very high in one corner of the container so that the density becomes very low in this corner. The ideal gas can be looked upon as a device to measure the temperature and the chemical potential of the gas. (3 28) where s is the spin of the particle. :[ (yLf)@ (2s+ l ) ] + :N . so the ideal gas law is applicable. then we know the p and T for the whole gas. But p .3 1) Substituting (330) into (323) we get the familiar Fermi distribution 1 (3 32) Notice that in the above we have used the known ideal gas results to determine the meaning of (Y and 0. We can define P . Hence once we measure p and T i n this corner. Equation (236) becomes (3 30) S=Nln . The chemical potential p of the ideal gas can be obtained from the results in thermodynamics (236) E= T (EiE In = aN iy V CE N N 5 2 (3 29) Equations (328) and (329) are identical except for an undetermined constant C.a T This fixes the value of C. Tare quantities shared by all parts of the body because the various reactions involve all the particles. (3 .
h = ' y h . (334) then this allows a more general solution of (317) and (31 8): (335) i. from detailed balance and the results of thermodynamics in Chapter 2. I f s = i . which we shall discuss in the next chapter.THE LAW O F DETAILED BALANCE 39 The integration constant can be determined by the value at T + m . we determine the particle distribution and all the equilibrium properties of a fermionic gas. We shall discuss this at the end of the next section. For example. P1 + Pz = P3 + P4 in (317). the ideal gas result (331). if momentum is conserved. The most important fermions are the electrons. If there are other conserved quantities. i. i. This is a convenient choice. Conservation Laws. If the spins are unchanged during collisions. . t h e n s . 3.e.3. in (318) P1 + PZ + P3 = P4 + P S t.P6 . +l (3 37) where the parameter h can be looked upon as the force maintaining the total 1 1 spin. and vmust beO. collisions of the particles with the wall makes momentum nonconserved. Equation (321) is derived using the conservation of particle number and the conservation of energy. then 1 fk = e (EkMS*h)/T . (336) where v is the velocity of the whole gas. Hence. then (317) and (318) can have more solutions. The constant C in the definition of entropy is determined by (330).e. Hence T and p are the forces fixing the total energy and the total particle number. Chemical Potential and the Constant Term i the Entropy n The above analysis clearly shows the power of the conservation laws.e. But if the container is stationary.
v/T and h/T are obvious examples. and we immediately get (341). More complicated examples are found in chemical reactions. The result (341) can be understood from another point of view. it can change because of reactions. but the a’s must satisfy aA + a B = a c From (330). Conversely. Each chemical reaction results in a relation between the chemical potentials. but are completely independent.dNc. we get PA + p B . then the generalisation of (317) is (339) Equation (321) is still applicable. then from (340) and (341) it can be seen that 77 . dNA = dNB = .g.p C = o . but instead write (3 43) with 7) being a constant. But according to (338). If we do not use (330). If the particle number is not conserved.40 STATISTICAL MECHANICS Each conservation law will introduce a “force”. * (340) (341) Because of the reaction (338). p must be zero. B and Cmixing together undergo the reaction A+BC . (338) The particle numbers of A . For example. with remaining one determined by (341). the walls of a container cause momentum not to be conserved. the three originally independent chemical potentials are reduced to two. The above “forces” 1/ T . three molecules A .p / T . Applying detailed balance t o (338). it will cause a force to become zero. In equilibrium the chemical reaction (338) cannot perform any work. i. . e. and v = 0. This is a result in thermodynamics and does not require (339) for its derivation. We can see from this that (330) is a very ideal choice. B and C are not conserved. i.e. if a conservation law is broken.e.
Distribution of Bosons and Black Body Radiation Bosons are not restricted by the exclusion principle and hence have different reaction rates. Each chemical reaction gives a relation similar to (344). (1 .23). Equation (346) says that if states 3 and 4 are already occupied. Planck’s constant h enters into a. we simply change the negative signs in (314) to positive ones.4.THE LAW OF DETAILED BALANCE 41 cannot be any arbitrary constant but must satisfy VA +VB =vc (3 44) We have to consider all other possible chemical reactions. h naturally enters into entropy. This result is not obvious and requires some calculations to obtain it. Hence qA must be chosen to be qA = constant x A . completing its definition. Equation (332) now becomes (3 47) The low density case is independent of the statisticsandfk isagain the Boltzmann distribution (348) . the reaction of (31 3) should now be flfZ(1 +f3)(1 + f 4 P > (3 46) i. The factors (1 f3). other states would be more inclined to go to them. Hence. We regard this as known. (345) This constant must be the same for all molecules. 3. (See (328).(318) to obtain the distribution function for the bosons.) Then from the consideration of conservation laws. from the principle of detailed balance and the concept of states in quantum mechanics we obtain a. Now we only need to change the negative signs in (314) . Of course. the only possible choice is to set all the q to zero.f 4 ) in (314) are the results of the exclusion principle and are not valid for bosons. and each q must satisfy all these relations. to choose such a q one must start from the conservation laws. From the statistics of the states. If we further extend the definition of chemical potential t o the case that the number of nucleons may not be conserved.e.p and the distribution function (3 . All chemical and nuclear reactions must conserve nucleon number A (the total number of protons and neutrons). According to quantum mechanics.
53) . If there are molecules in the cavity. (See Prob. Photons are bosons and the photon gas in equilibrium is called the black body radiation. where c is the speed of light.42 STATISTICAL MECHANICS This can be called the classical distribution.52) N This formula can be derived by the principle of detailed balance. N*  RN+) = R'N* (3 . Einstein used this together with the principle of detailed balance to derive some properties of the rates of absorption and emission of light as mentioned below. and hence its chemical potential is zero. and precedes the reaction rate considerations in (346). The discovery of this distribution is a prelude to the birth of quantum mechanics. The distribution of the photons is (3 49) where cdk is the frequency or the energy of the photon (h = I). and also include the two polarisations. The number of photons is not conserved. The energy distribution of the states of the photons can be calculated by the method in Sec.50) Hence the frequency distribution of the black body radiation is w = g(w) ew/T  w 3 ~ nc 3 2 1 (351) ew/T . and we only have to change eP =p2/2min(36)to eP = c p .) The absorption rate of light is proportional to N and E ( W ) and its emission rate is proportional to N *. 10. while (347) and (332) are the quantum distributions. Therefore. Photons are continually absorbed and emitted by the walls of the cavity.1. The energy of A* is higher than that of A by a. the ratio of the population N of A to the population N" of A* is eW/T (3 . They must be equal. 3. In equilibrium. Let molecule A absorb light of frequency w to become the excited state A * . All the results of the last section can be reproduced for the boson case. i. Thus (3 . these molecules can also absorb or emit photons. Ek = hok.e.1 This is the Planck distribution.
each distributed according t o the classical formual (348). so as to obtain (354) (and the value of R ) . They are rather unexpected. the emission by A* is faster and this is stimulated emission. Besides the kinetic energy. we have mentioned chemical reactions. This is a property special to bosons.THE LAW O F DETAILED BALANCE 43 where R' is the emission rate of one A* and R is the absorption rate of the interaction of one A with one photon of energy w.e. A can spontaneously emit a photon and become A . From (352) and (353) we get = R r2c3 w3 + RE(o) . (355) i. (2) If f # 0. the half life of A* is In 2/r . 3. B and C mix as an ideal gas. the total energy of each molecule also includes the energy of . which we now use to derive relations between the concentrations of the various molecules. in the case of n o photons. Equation (354) shows that: * (1) Even if E ( W ) = 0. We start from the reaction (338).3. i. (3 . Each chemical reaction involves one conservation law and results in a relation between the chemical potentials like (341). assuming that A .54) The factor 1 + f is naturally the same as the 1 + f in (346). Detailed quantum mechanical calculation must involve a complicated analysis of the interaction of A . with a rate r = R  w3 7r2 c3 .5. (3) The absorption and emission rates are determined by R . and are obtained by detailed balance plus the Planck distribution. which depends on the structure of the molecule and the details of its interaction with photons. These results are related to the interaction of molecules with light.e. Chemical Reactions and the Law of Concentrations In discussing the conservation laws in Sec. A* and the photon. 3.
(358) From (31 4) we get 0 = @A nA 'B + GB GC + T l n  (359) nC or This is a theorem on equilibrium concentrations. This is very useful for a chemical engineer. if we know the structure of the molecules we can obtain information on the concentrations. In other words. Let nA = N A / V be the concentration of of A . . > & = + In nB pc = @c T l n n c + . If the concentrations of the various molecules can be measured but the structure of the molecules is not clear. then pA = @A @fj +TlnnA .44 STATISTICAL MECHANICS internal motion such as rotation and vibration (3 56) The total population of molecule A is (357) where @A is a function of the temperature and is related to the mass rn and the internal energy aAol the molecules. The right hand side of (359) can be calculated from the properties of the molecules and is only a function of temperature. then (359) can be used to deduce the structure of the molecules. The reader can consult books on chemistry for a deeper understanding of this topic.
the income distribution or the grade distribution in an examination. O(x x(s) ) is 1 and otherwise it is zero. Figure 33 is an age distributionf(x) (say in units of ten thousands). A distribution can be represented by a curve. s. The total area under the curve is the total population. (361) (362) The meaning of this distribution function is: f(x) d x is the population between ages x and x + dx. If it is less than x'. for example the age distribution. is the area under the curve to the left of x = x' . we can search through the census to collect the names of the people whose age is less than x' and then count the number: (363) where s is the name of each person and x(s) is the age of the person s . 33 Population distribution plotted against age. A 0 C n " 4 9 l 19 24 29 34 39 44 49 54 59 64 69 74 79 4 Age Fig. f ( x ) dx = total population (360) The population with age less than x'. Equation (363) represents . The concept of a distribution is the most common in statistical data. denoted as N(x'). To count the population with age less than x'.THE LAW OF DETAILED BALANCE 45 Problems 1.
This symmetry is a special feature of the equilibrium state. 2. Calculate the energy distribution of the free particle (a) in one dimension. Some states. Sections 32 and 33 point out that entropy is intimately related to the concept of a state in quantum mechanics. 6. do not possess this symmetry.46 STATISTICAL MECHANICS this statistical process. 3 . the population is f(x')Ax' = \. and we call them nonequilibrium states. But a uniformly moving or rotating body can be regarded as in equilibrium. For example. Detailed balance comes from the symmetry property of thetime direction. The following reaction occurs inside a star y + y e+ + e where y is a photon and e' are the positron and the electron respectively. electrons and neutrons the following reaction occurs p+e+n+v . assuming (a) V(r) = + K r2 (b) U f r ) = . 5 . even though the heat current is unchanged in time. Calculate the population of e+ and ein equilibrium. In a dense gas of protons. The age distribution f ( x ' ) is formally the differential of N (see (362)): (364) dx' S In a small internal Ax'. Use (312) to calculate the energy distribution of the state. The reader can think about these situations. They also show the importance of conservation laws. a body with a heat current passing through it cannot be called an equilibrium state. (365) The definition of the distribution should be based on (365).Z e 2 / r . We can use (b) to estimate the energy of heavy atoms (large 2 ) . (b) in two dimensions. (c) in three dimensions. x'+ Ax' dx6(xx(s)) X' . 4 . though apparently steady and unchanging. Later when we see summations such as (364) we should immediately think of its integral or imagine that the 6function is a sharp peak with finite width. The reader may try to derive (331) from (336) and (330).
The chemical potential of electron satisfies pe >> m c 2 . (a) Find the relation between the flux and time. A detector at distance R measures the flux of electrons and protons. and this star is a mixed gas of protons and electrons.THE LAW OF DETAILED BALANCE 47 where v is the neutrino.e. p = 0 andfo (the population of the state k = 0) is proportional to Y.i. 8. This phenomenon is usually called BoseEinstein condensation. (e) Prove that this phenomenon of condensation does not occur in one or two dimensions. This relation is linked with the distribution of the velocities of the particles. Suppose a star explodes at times t = 0. Find the population of p . Attempt to generalise the discussion in Section 33 to include bosons. . (b) If the temperature is T after the explosion. with R >> the radius of the star. (c) Calculate the relation of fo with T . From the reaction A* + + A +photon . The discussion of (338) to (345) is not limited to fermions. (a) Find the relation between p and N / V .. This result is independent of the condition that A is a boson. . derive (352). Notice that when the temperature is lower than a certain critical temperature T. (d) Calculate the heat capacity. 9. eand n in equilibrium. calculate the flux. In fact the molecule C in (338) must be a boson (two fermions can combine to form a boson). which escapes after production. 10. where m is the mass of the electron. The ideal boson gas can be analysed by (347). there are macroscopically large numbers of particles occupying the lowest state. (b) Calculate the critical temperature T. 7 .
We first introduce the free electron model and then discuss the wellknown heat capacity of an electron gas.e. Hence we shall give a rather detailed discussion. it can illustrate some underlying principles. Now we treat these electrons as an ideal Fermi gas. 4.e. the density is high. Although the Peierls’ instability (i.Chapter 4 ELECTRONS IN METALS The topic of electrons in metal is surely a major field of solid state physics. Many properties of the metal are determined by the motion of these mobile electrons. and ek its energy. and this is the free electron model. In this chapter we shall pick some simple problems as examples of the theory in the last chapter. We shall use a one dimensional model to illustrate the interrelation of electrons and the crystal structure. We use k to specify the state of the electron. with the atoms arranged in a crystal structure. This simple view produces unexpectedly good results. the phenomenon that a one dimensional metal is unstable) is limited to one dimension. Let fk be the average population of state k (see (332) ): fk = f ( e k k p ) 3 (41) . The electrons in the inner shells are tightly bound. Each state can accommodate at most one electron. Dense Ideal Gas In most metals. the usual electric potential times the charge of the electron).1. Let us now see the results of this ideal gas model. The structure of the crystal is the most important factor affecting the behavior of free electrons. The electrons in the outermost shell can move freely among the atoms. Let p be the chemical potential of the electron (i.
The radius of the spherepF is called the Fermi momentum There are N states with energy less than eF : . Let g(e) po =k V f 1 zw (Ek .e.g. The calculation of g(E) has been presented in detail in Section 31 of Chapter 3. Above this surface all the states are empty.LO .P o ) > . which we simply call density of states. the chemical potential when T = 0. and gives The thermodynamic properties of this model can be largely expressed in terms o f f and g . V of the electrons and the energy density E/Vare (45) If T = 0. then all the low energy states are filled up to the Fermi surface. the density N . The function of g(e) is the energy distribution of the states per unit volume. The energy at the Fermi surface is /. and E is the energy with respect to p o .ELECTRONS IN METALS 49 (42) The energy distribution of the states is an important property. i. (43) /J(T=O) . and is always denoted by : E F P(T= 0) = (46) The Fermi surface can be thought of as a spherical surface in the momentum space of the electrons. e.
(49) where n = N / V .39 From Kittel(1966).19 1.5 5. its momentum would have to be of order h/a.4 Na K Rb cs cu Ag AU 0. and a = h/pF is approximately the average distance between the electrons: eF h2 ma2 (410) is approximately the zeropoint energy of each electron.1 1.85 0. In most metals.73 0. only electrons very close to the Fermi surface can be excited and most of the electrons remain inside the sphere.1 1. experiencing no changes. i.34 1.56 1.5 x 104 3.7 3.3 x 10’ 1. the distance between electrons is about cm and eF 1 eV lo4 K (see Table 41). Therefore.6 x 2.8 1.5 7. T << eF . p.5 1.68 0. (volume of sphere in momentum space) x (volume) x (spin state (= 2) ) + ( 2 ~ h )with h = 1.1 x lo8 0.90 5.63 1. Table 4 . Properties of the electron gas at the Fermi surface electron density N/ v cm3 ~ momentum kF cm’ 1.20 velocity uF cm/sec 1.38 1.1 .1 2.50 5.76 5.e.86 8.90 0.1 0.4 6.8 8.2 6.7 2. To fix an electron to within a space of size a . at ordinary temperatures.35 1.73 1. . 208 . the temperature is very low. This zeropoint energy is a result of the wave nature of the electron or a necessary result of the uncertainty principle. pF = (3T2n)”3 .4 2.0 5.08 0. Hence ~.5 TF = EF/k K Li 4.50 STATISTICAL MECHANICS i.79 energy eF eV 4.e.
The lower the temperature T .11) producing a pressure. (we shall call these outer electrons. This pressure makes the metal expand like blowing up a balloon. we discover that this p ( 0 ) corresponds to a temperature of and N/Vis usually much higher than the usual gas. the total energy of this electron gas is 3 E ( 0 ) = NeF  5 3N ( 3 7 T l 3 a 10m v . Notice that eF is high because the mass of the electron is very small. This attractive force comes naturally from the negative charge of the electrons and the positive charge of the ions. It will not blow up because of the attractive force between the atoms. The states below the Fermi surface are nearly filled. and empty states are rare. . When T+O. 4. This electron gas model is quite different from the usual molecular gases. Now this model becomes a new mixed gas of holes together with electrons above the Fermi surface.~ ' ~. Hence the concept of holes appears naturally. Heat Capacity Only a very small portion of the electrons is influenced by temperature. The high density is also another reason. this gas disappears. the more dilute the gas is. Hence it is an enormous pressure.ELECTRONS IN METALS 51 kF = (3n2N/V)'I3 = p F / h VF = . called the zeropoint pressure (412) If we compare this with the ideal gas pressure p = NT/V.) The momentum of a hole is less than p F . The reader can consult books on solid state physics. This is a very convenient way of thinking. while that of the outer electrons is larger than p F . We shall call an empty state a hole. pFlm 9 TF = € F / k . At T = 0. 3eF. (4.2. The calculation of this attractive force is a special topic.
i. In this interval of energy. i. 41 Electron (hole) energy versus momentum. (413) Now the origin of the energy is shifted to the energy at the Fermi surface. Hence the energy distribution is the same for the holes or the outer electrons. Hence.e. The average population of can the hole is (for each state): 1 . the holes or the outer electrons cannot exceed T by too much. e. a state of energy E’ produce a hole of energy e‘.g. p o = 0.e. i. the total energy is (415) . The energy of a hole is the energy required to take an electron from inside the Fermi surface to the outside (Fig. g(e) is essentially unchanged. As T<< eF. p 0. All the calculations can now be considerably simplified. the energy of energy of a hole I energy of an outer electron P I 0 t \ LI D  / ’F Fig.52 STATISTICAL MECHANICS At T = 0 the total energy is zero.f( €’ . there are no holes or outer electrons. 41).p) = f(€’+ p) . Therefore. pF and eF are the Fermi momentum and energy. The holes are also fermions because each state has at most one hole.e.
that of the holes. The population in .. and differentiating once.ELECTRONS IN METALS 53 where 2g(O) is the density of states of the holes plus the outer electrons. but eF >> T and so the upper limit of the integral in (415) can be taken to be 0 0 . Hence C .N ( T / e F ) . Equation (419) is the first term of an expansion in T .w ' ( 0 ) (4 . we get the heat capacity 1 aE n2 1 . + (418) Each active electron contribute approximately 1 to the heat capacity C. But experiments show that C is in fact much less than + N . The energy of a hole cannot exceed eF. Now only a small portion of the electrons is involved in motion and the number of active electrons is about N T / e F .. at T = 0 the chemical potential is not zero. (417) This result is completely different from that of the ideal gas in which C = + N . Hence we expand g(e) to O(e) to get the next term: g(e) = g(0) . In that case each gas molecule contribute a heat capacity of .=g(O)T vaT 3 v . Only when the fermionic nature was discovered did this problem of heat capacity have a satisfactory explanation. What is the next one? If g(E) = g(O).C = .T2 eX+l 12 X n2 (416) o Substituting in (4. From (417) we get (419) In the initial stage of studying metals it is usual to treat the electrons as a dilute ideal gas. we do not get the next term. is the density of states of the outer electrons and g.15).20) where g. This integration is easy: 1 0 00 d e f(E)E = T 2 I m d x . As g+ f g .
Because eF >> O D . Hence the results of (417) or (419) do not contain the T2 terms.e. at low temperatures.54 STATISTICAL MECHANICS each state of the outer electron and the hole is 1 . 400 K. Therefore. its contribution is about 3 N T . The heat capacity due to lattice vibration at T > 0. hence p . the vibration of the atoms in the metal also contributes to the heat capacity. T << 0 it is N ( T / 0 . we just use the results here. g'(0) may cause other direct corrections to the energy (415). ) 3 . Therefore. Equation (415) should be  Upon expanding. (4 .) At very low temperatures. Besides the electrons. Above room temperatures. all terms of g'(0) are cancelled. The Debye temperature 0 of a metal is around 100 K to .T z / e F . hence this T 3 . a dominant term. Besides p . i. the total energy capacity is (425) .P ) 0 = 1 0 m deg(e)f(e+p) Y (422) where both p and g'(0) are small numbers. C=NTL[l+O($)] 'F .d € g + ( € ) f ( € . (This will be discussed fully in Chapter 6 .f(ep) = f ( e + p ) . term is much larger than that correction term in (424). we can solve for p : (423) Because g'(O)/g(O) = + e F . is about N.24) The correction terms come from g"(0) and g'(0)2. Expanding (422). (421) The value of p can be easily calculated because the population of the holes and the outer electrons should be equal: j.
133A 1370 (1964). The crystal structure is naturally an important factor. and n is the electron density. 4. In the following we shall use the simplest one dimensional model to discuss the influence of a periodic arrangement of the atoms.5? T 2 /. then its metallic property is affected too.a* )/”.3. Influence of Periodic Structure The above discussion points out the importance of g(e). Of course.ELECTRONS IN METALS 55 The reader should notice that the analysis from (420) to (424) is very important.*. The experimental value of r is larger than that calculated by (426). the free electron model is oversimplified. E. In solids there are many factors affecting g(E). the density of states. From W.. From the experimental data we can determine the coefficients of Tand T 3 in (425). This experiment is a crucial test of this free electron model. but only Tand T3 terms. . * . 42 The low temperature heat capacity of potassium/T (lo4 erg K’ mole’ ). 3 T2 (K2) Fig.08+2. The coefficient of T is (426) We have put in k and h explicitly. The reason for this is not clear. in that it demonstrates that the heat capacity of the electron does not have a T 2 term.#* C/T=2. Figure 42 is a plot of C/T versus T2. The interaction of the electrons with the impurities and the lattice vibrations plus the periodic structure of the lattice have not been taken into account here. and it is surprising that the results are so good. Phillips. Phys. Rev. Lien and N. Potassium *. The metallic properties of a material denote those properties similar to the free electron gas. H. According to (425) it should be a straight line. y e  ( Cll 02 . If g(E) is affected. and depend crucially on the near constancy of g(E) at the Fermi surface..
J/ are the and P same.56 STATISTICAL MECHANICS The energy E of the electron states can be solved from the wave equation (427) where U(x) is a periodic function with a period a equal to the lattice spacing: U(X) = U(X + a) (428) and which represents the influence of the arrangement of the atoms on the electrons.e. and 1 0 L U(x)dx = 0 . with value p2/2rn. there is degeneracy. The perturbation theory in quantum mechanics tells us that the first order correction to eP is (431) But this is not valid. we note that U ( x ) can be written as 4 = 2n/a . because this formula is appropriate only for nondegenerate energies. In (430). given by the (433) n=rn c m . i.30) Now we calculate the correction due to U on e P . If we neglect the solution of (427) U.it is simple to obtain €p = Q212rn (4 . (432) The potential V ( x ) has interaction between J p )and Ip matrix element (PlUlP') = + nq). For simplicity. we assume U to be very weak. To deal with this. the energies of J/. (4 29) where L is the total length of the body.
2 1 pz=q+k 2 . The energy operator in the space of these two states is represented by a matrix H : H=[:lz T) .E= 2 A . k<<q . 44) when the energy is (439) . .ELECTRONS IN METALS 57 Therefore. (435) Quantum mechanics tells us to consider these two states together in order to obtain the first order correction term. . . i. . Let us now look at the situation n = 5 1 . . When k = 0. The situation is similar for n = + 2 . when p = k q / 2 . etc. E+ . (436) (437) The eigenvalues of H are A IUil (438) These results are plotted in Fig.e. The density of states therefore has a gap (Fig. Let 1 pi=q+k . (434) then the result of (43 1) is not valid. Hiz = ui . . 43. n = 51. . 5 2 . +3.e. P =2 n 4 . if i.
then what ispF? Eachp .3 & I [2A I I I \ I I I 3qr2 q qi2 q12 Fig. 44 q 3q12 *P This is the most important effect of the periodic potential. there is one electron per lattice site on the average. 4 . Suppose N = LJa . Let the total number of electrons be N .e.58 STATISTICAL MECHANICS & Fig. (440) i.
43) The above assumes e << eF. the Fermi surface consists of the two points p = pF and p = pF. then “ 4 * p F = a = 2 (4 . 45). . therefore. and we get (4 .e. If the electron is to be excited. if T / E F << 1 it would exhibit the metallic property discussed in the last section. Therefore p = 0. As the gap at p = ? q / 2 is far from p F . Holes are empty states below the gap (see Fig.44) The above assumes T << A .ELECTRONS IN METALS 59 has two states (the two directions of the spin).q / 2 are fully occupied. (441) i.4 2 ) The states between 4 / 2 and . If each lattice site has two electrons.e ) . We use the method of holes to calculate the total energy. The total energy and heat capacity are 1 [[E(T)E(O)] N = L j  1 d e 2g(e) eE/T O +1 (4 . we can calculate g(e) and the heat capacity. With this in mind. The number of holes and that of outer electrons are equal and in (443) g(e) = g ( . We choose the origin of the energy to be in the middle of the gap. it must overcome the gap and must have an energy larger than 2 A .
. In the above example. this population is small and the body is now an insulator. p i . If A is not very large. 4.60 STATISTICAL MECHANICS I .45) This result is calculated from the approximation (4 43). 45 This result is obviously different from the situation in (4.4. The Structural Change Caused by Electrons The crystal structure produces a gap in the energy of the electron.go(€)]E N % ‘ F = . We have already seen the influence of A on C.7r2 2m 2ma2 (4 . If A is very large. More precise calculation gives similar results. if N = 2L/a. . this population can be effectively controlled by the temperature.19). The population of holes and outer electrons is directly related to the conducting property of the body. This body becomes a semiconductor.[ E ( A ) . The energy gap makes the inner electron difficult to excite and eA/T is the most important factor in (444). The main conclusion here is that E(0) is lowered because of the gap. then the influence of this gap is very important.E ( A = O)] = a d e [ g ( e ) . Now let us see how it affects the zeropoint energy: 1 A .= ter electron \ I / Fig.
a to 2a. This new gap can lower the energy of the electrons. then the increase in energy of the lattice is proportional to b 2 . Therefore. producing an energy gap. the atoms in the crystal can move.b ) = U(x.e.48) This bUi has the same effect as the above U. (4 . However. n=m + O(b2) . (4 . In this case eF is still far away from the gap.47) We have a new b U: : IbUiI = A' .!. Suppose every other atom moves by b (Fig. but the energy gap occurs at (4 49) which is the present kpF. U ( X .46) where K is the elastic constant. The new potential energy has a period of 2 a .ELECTRONS IN METALS 1 61 Now we return to the situation N = L / a . then the period changes from I0 0 0 0 0 0 0 Fig. the potential U(x) seen by the electrons is also changed. We have mentioned that the result is not much different from the case without U(x). since b = 0 is the equilibrium condition: [[E(b)E(b=O)] N 1 1 = b2 2 . (4 . After displacement. . i. 46). pF = T q . If b is not very large. . b = 0) +b m einq'x U. 4 6 The vibration of atoms in a crystal.
l / A . The conclusion is that the structure of period a is unstable.. (See the last section of Chapter 13). a structure with period 2a is lower in energy. and at the same time there is an energy gap at p = ? p. i. there must be an energy gap at p = + p F . but not in the y or z direction.b2 ~ IU[12 In ( b z . In addition. q/2. . and these are awaiting a solution. and must be considered together. The reader can easily see that pF can be any value. Notice that the above calculation uses pF = 4/4. We then obtain 1 [E(b) . Hence if b is sufficiently small this energy change is negative. the interaction between the electrons has not been considered. but it clearly demonstrates that a one dimensional crystal cannot be a metal.e.~ m a ~ lU[12/K . this model has not accounted for the vibration of the lattice. But this kind of substance is more complicated than the above model and the above conclusion can only be a reference but cannot be directly applied. and even if pF is not a rational multiple of q. If bZ is very small. A onedimensional substance in the laboratory must be part of a whole physical body. This was instability first pointed out by Peierka The most important conclusion is that the structure of the lattice and the motion of the electrons are mutually related. The motion of one dimensional electrons involve many problems.E ( b N =0)] = 4ma' T . ) t iKb2 (451) (450) The last term is the result of (446). the above conclusion may be modified. the above conclusion remains unchanged: the period of the lattice structure must be '/ (2PF ' Of course. This analysis is very simple. For example the electrons of some crystals can conduct easily in a certain x direction. h = 8 ~ . . One dimensional vibration has a very large amplitude. This substance can be thought of as composed of a bundle of one dimensional crystals. The original metallic property is lost. aPeierls (1955). giving the equilibrium value of b as b lull EF e .62 STATISTICAL MECHANICS by an amount obtained by replacing A' for A and 2 a for a in (445). then ln(l/b) is very large. The minimum of E ( b ) can be calculated by setting a E ( b ) / a b= 0. Therefore. both theoretical and experimental. Because of the interaction of the electron with the lattice.
This is the photoelectric effect. 3. find I(w). 47 Idealization of a metal. . 230. Notice that when T # 0 there are 'ejected electrons even when there is no light irradiated on the surface. Let @J be the difference of the chemical potential 11 of the electron with the potential outside of the metal. 5. Let Z be the average number of ejected electrons produced by one photon. the difference 9 must be overcome to eject electrons. The velocity of the electrons is close to the speed of light. The reader should derive every result.E 3 v 5 V F This is the equation of state at low temperatures. otherwise he will not be able to fully understand the implications. especially those in the last two sections. Light irradiated on the metal surface can eject electrons out of the metal. what is the relationship of I with the frequency w of the light? (b) When T # 0. This is the effect of thermal electron radiation.= . 47). in the theory of the white dwarf.. If there is a crack in a metal. and is far from realistic. Prove that at low temperatures the pressure of the electron gas is 2 E 2 N p ( T ) = . EF * + inside the metal outside Fig. 47 is a highly simplified model surface. e.g. (a) At T = 0. 2 . This chapter is mainly calculational. and we must use a relativistic treatment. this is a cluster of helium nuclei and electrons at high temperatures and high pressures. Roughly speaking. Assume that T is rather low. This problem is fully discussed in Huang (1 963). (Fig. p. what is the density of electrons in this crack? Notice that Fig. Dense electron gas appears in astrophysics.ELECTRONS IN METALS 63 Problems 1. T << 4. 4. Express the pressure of the electron in terms of the radius of the star.
Section 4 discusses the one dimensional model. The situation is much more complicated in two o r three dimensions. . The interaction of the lattice with the electrons produces a new periodic density.64 STATISTICAL MECHANICS 6. although there is a formal similarity. o r “charge density wave”. This can be called the “density wave”.
We use the YangLee theorem as the centre of our discussion. the reaction rate and the region of motion. Chapters 7 and 8 contain further deductions and more examples where the discussion focuses on the implications of the assumption. The aim of this part is to point out the applications of the basic assumption of statistical mechanics. and its relation with the equilibrium observation time. This part is divided into five chapters. We first introduce the physical content and motivation of this basic assumption and use a few examples to illustrate its use (Chapters 5 and 6). Chapter 9 discusses the mathematical problem of the macroscopic limit or the thermodynamic limit.PART I1 HYPOTHESIS Statistical mechanics is established on a daring assumption: the Boltzmann formula for the entropy S=lnr . All the results of statistical mechanics can be derived from this basic assumption together with the law of large numbers. where r is the volume of the region of motion. as well as certain points of ambiguity. The reader should notice that this assumption has never been proved and our present understanding of it is still incomplete. . without any other new concepts. This region of motion represents the region within which the states of the molecules of a body are constrained to vary and r essentially measures the product of the amplitudes of the various motions. thus enabling the reader to apply the assumption properly without being led blindly by the principle.
entropy is a measure of the small scale molecular motion. and magnetic moment. E/N is the average energy of each molecule and V/N is the average volume occupied by each molecule. including its motivation. 5. . This answer is not derived step by step from any basic principles. Up to now the origin and definition of this assumption are still not completely clear. Let (Ax)3 V E . etc.Chapter 5 THE BASIC ASSUMPTION OF ENTROPY AND MOLECULAR MOTION In thermodynamics. We examine ( 2 . particle number. on the same level as volume. Region of Motion The discussion of the ideal gas in Chapter 2 has provided a clue about the relation of entropy to the molecular motion. etc. This is its most important achievement.1. This chapter discusses the various concepts and calculational problems involved in this assumption. Indeed how does entropy measure the molecular motion? Statistical mechanics provides the answer to this question. the law of large numbers and the special feature of high dimensions.3 6 ) : where N is the total number of molecules. In the preceding chapter we have emphasised this point and treated entropy as a coordinate. the statistics of the states. but it is based on a daring assumption.
and each direction has a length representing the magnitude of change of one variable of motion. . each molecule essentially does not move out of a range of V/N If it did move out and another molecule moves in. p y k . 5. At any instant. k = 1 .2. Hence. this trajectory traces out a very long and meandering curve scattered in a region of the phase space. y k . Because of the continual collisions. As each molecule can reach any point in the container. Now we generalise (5 3). the ideal gas is a very simple and special case. In a sufficiently long period of observation. N ) . region has 3 N directions of length A x . 67 where m is the mass of a molecule. i. and the space is called phase space. that would be as if no such motion had taken place. should not AX)^ be V ? Why do we use V/N instead? The reason is: each molecule is identical and indistinguishable. Assumption for the Calculation of the Entropy Of course. . . . and 3N direc) tions of A p . where w = 2m/C is a constant and r is the volume of a 6N dimensional region (in units of w ~ ~This. tracing out a trajectory in this space. the result of the ideal gas has clearly indicated that to understand the meaning of entropy we must start from the magnitude of the molecular motion.THE BASIC ASSUMPTION O F ENTROPY AND . Hence the relation of entropy and the region of motion is very clear in this model of the ideal gas. A p is the magnitude of the change of momentum of the molecules and A x is that of the change of position. z k . Its entropy is easily calculated in the last section. Now we can write (5 . The lengths of this region in the 6 N directions are the amplitude . Therefore the trajectory is a set of actual states. 2 . 3N coordinates and 3N momenta ( x k . Nevertheless. Roughly speaking. p x k . The reader may ask the following question.e.1) as S=Nln  = l n r . The state changes with time. . but most of the other problems are not simple. the positions and momenta of N moleculescan be represented by a point in the above 6N dimensional space. This region can be said to represent the extent of the motion of the N molecules. the momentum and position of each molecule is changing incessantly. This point is called the state of the system at that instant. p z k . There are 6N variables in all.
we defer the discussion to later chapters. producing collisions and changes of momentum. However. No matter how complicated the molecular motion is. y . then the room cannot represent its region of motion. We consider a simple example in order to understand the implication of this assumption. but it is not so intangible. At the moment we use the following assumption to patch up some explanation. We need not analyse the trajectory. .68 STATISTICAL MECHANICS of change of the 6 N variables of motion. The above definition of r is not complete. the ceiling and the four walls. and uttered some words. and assume that entropy is the logarithm of I? : S=lnr . We must emphasise the following . the region of motion in phase space is determined by these invariant quantities. how can we determine this region of motion from the molecular motion? This is a difficult question. y . We call this region the distribution region of the trajectory or the region of motion of the states. This is the implication of this assumption. z ) . Indeed. but we can simply say that the magnitude of the changes of x. The region of motion is the room and the volume of the room is the product of all the magnitude of the changes in the three directions. Here we regard it as a proposal motivated by (5 3). We have only given some names.we must have motion and change so as to define the region of motion. Each molecule must be moving continually. z is determined by the floor. we can ask: what is the magnitude of the change of position of the fly? The position of the fly is specified by three coordinates ( x . z ) space to the phase space determined by the six variables of motion. The validity of this assumption should be built upon the continual change.a It is not derivable from any known principle. If the observation time is much longer than that required by the fly to go from one side of the room to the other. A fly moving inside a room has a very complicated trajectory of motion and its position is changing continually. A supplement to the basic assumption is: The region of motion of the states is equal to the region allowed by the invariant quantities. These are invariant quantities. This indeed is the implication of the word “motion”. Let r be the volume of this region. aThis assumption is a masterpiece of Boltzmann (18441906). Whether the S in (54) is indeed the entropy in thermodynamics should be validated by its applications in thermodynamics. If the fly does not move. We now generalise the ( x . (5 4) This is the basic assumption of statistical mechanics. This indeed is a daring assumption. y . These positions of the walls and ceiling are generalised to the various invariant quantities.
r will be meaningless. the volume of this region is r=2JZm& . This supplementary assumption should be regarded as part of the basic assumption. The situation for solids is more complicated. Hence. Let a certain region be defined as 0 < E < E . Besides the energy and the total number of molecules. there can be other invariant quantities due to its structure. the invariant quantities include the total energy. We now define the rule: r is the total fiumber of states in the region of motion. (5 6) The unit or the constant 2i~ not important. .THE BASIC ASSUMPTION O F ENTROPY AND .e. Why do we not use the square of the positions and the energy as the variables? This is an important question because if we use different variables. this problem of the right variables is a problem of evaluating the weighting of each state in the region of motion. like the positions of the frozen impurity atoms. The question is what variables should we use to define the phase space? In the above discussion we have used the positions and momenta of all the particles as the variables. In calculating the volume. The important observation is r a but r' a E o . . Of course whether this assumption is correct should be determined by the result of analysing the trajectory. The only way out is to forego the continuous variables. Now we must clarify an important point. We now add a definite rule to this assumption to avoid any confusion later. the volume and the total number of molecules. The phase space can be regarded as a collection of discrete points and each point represents a state. the high energy parts are not so important in r as in I". Quantum mechanics provides the solution. All states allowed by the invariants must be included in the region of motion. We first look at the momentum p and energy E = p 2 / 2 m of a particle. This supplementary assumption is very nice. the volume is r ' = E. if the variables are not specified. . This rule states that: the problem of statistical weighting is very obscure in continuous variables. Hence. i. (55) If we use E as the variable. each state has a statistical weight of 1. It reduces the complicated problem of the trajectory to a relatively simple problem of calculating the volume of the region. We start from the entropy of an ideal gas in (53) to point out its relation with the magnitude of the . . Now the rules for calculating entropy are ready. or I p 1 < . 69 For a gas. we have a different volume. / = If we use p as the variable. We use an example to illustrate this point. . and instead use discrete variables.
the total number of s.6.70 STATISTICAL MECHANICS motion. Let L states have the same energy.N ( 2 ~ N ) ” 21 + 0  [I ) : ( .e.region of motion. Now we first discuss some problems of the techniques of calculation as a preparation for the actual calculation. 5. In fact. N .3. Here the invariants are N . Now we just apply the result. In Chapter 3 we used the law of detailed balance to discuss the ideal gas including the quantum gas. If N >> 1 then N! = N N e . which are of little use unless evaluated more explicitly. Fermions The implications of the basic assumption cannot be clearly explained in a few words or by a few examples. Hence r is the total number of ways of putting N particles into L states and each state accommodates just one particle. i.e. We start from the simplest example. then what is the entropy? To use the basic assumption in computing the entropy. The collisions or the reactions between the particles are the crucial point of analysis. L >> 1. the main aim of this book is to clarify the implications of this assumption. Thip total number is r= L! N ! ( L. But (57) involve factorials. In fact. Its implication and validity await further analysis. we must first find r.N ) ! ’ (5 7) According to the basic assumption. We must remember that this is just an assumption which has not yet been proved. If L . and then we introduce the concept of the. S = In r. the basic assumption is true only for system of many particles i.ates allowed by the invariant quantities. The derivation of this law is given in Sec. Entropy is an extensive quantity. of which N are occupied by the fermions. If there is some cause making these particles change their states continually. (5 7) can be simplified by the law of large numbers. L and E . 5. Now we shall start from the basic assumption to tackle the same problem. N are very large. Finally we use an assumption to change the problem of calculating the entropy to a statistical problem of calculating the number of states in the region of motion.
we can group the set of states with the same energy E together to form the system of the above .12) The above d E i s edN(see (510)).dN T T dN . the differential of the entropy is related to the temperature T and chemical potential p . Therefore In . (5.e. 1 dS= . i.e lJ .N hence LN eEI T 3 (5. .p . but the states in an actual system have different energies. that of Chapter 3 where f is the average population of the states..e.11) According to thermodynamics.13) This result is. i.THE BASIC ASSUMPTION OF ENTROPY AND . of course. The reader may ask: is this system too simple? The energy of all states are E . The last term of (59) is O(1n L ) and can be neglected. The answer is that this system can be regarded as part of a large system. We now differentiate the result (59) and get d S = l n (LN N d N ) (5. 71 Substituting this into (57).. NIL is 0 ( 1 ) and S = O(L).dET . (5. we obtain after some simplifications S = L 1n :[ f N L  LN L L LN (59) + 1 1n 2 L 2n(LN)N Suppose L and N are of the same order. the total energy is E=eN. Because each state has the same energy e. i. Equations(512) and (51l)are equivalent.10) .e.
pk . to consider all the particles of an ideal gas and use the basic assumption to calculate its entropy. . 5 4 The Ideal Gas . Let us now look at a more complicated example. . .e. the positions of all the molecules must be inside this volume and (2) the total energy is invariant with a definite value E . To calculate F we must consider the restriction of the various invariants. . d 3pN . . .e. whose coordinates and momenta are rk . . d3pN in 6N dimensional space. k = 1. . we can then prove that T and p are uniform throughout the system.1 we obtained the statistical formula of the state of one particle : j d r = [d3r 1 d3p h3 (514) i. . (1) Let the integration of ' k be restricted as (516) . Suppose the ideal gas consists of N molecules. i. i. dr. We now generalise (514) to the statistical formula of N particle states dr.e. . d3rN d3p. N . (2) Energy invariance can be achieved by imposing a delta function & ( E. The proof is left to the next chapter. 3. This requirement is indeed a most important achievement of the basic assumption. the number of states in a volume d3r. drN .72 STATISTICAL MECHANICS example. Now the important requirement is that each part of the large system should have the same temperature T and chemical potential E.1 . (515) h3N i. d3rN d3 p1 . We first carefully consider a case where the particles have different energies. ..2.C. .e.H ) where H is the total energy function (517) . . If the basic assumption is true. . In Sec. . .k3r1 The volume of the container is I/ = L 3 . . in a volume of d 3 p d 3 r there are d r states. Here the constraints are: (1) the volume is invariant.
. Now we use (5 .17) are not enough and we have to divide the statistical result by N ! . For the results see (539) and (531). We have already emphasised that if the basic assumption is to be true. We get VN W(E) = j d 3 p . (518) where W(E) can be regarded as the surface area of a sphere in 3 N dimensions with a radius determined by H = E: [9 k=i Pilli’ = 2mE (519) Points on the surface of this sphere satisfy (5 . How do the variables in this . If the particles are not the same. The calculation of the surface area of a sphere is an interesting mathematical problem. As (5 .THE BASIC ASSUMPTION O F ENTROPY A N D . .16) and (5 . 73 The restrictions (516) and (5.17). and we defer its discussion to the next section. SO (5 21) The other properties.15) to calculate r with the constraints (5 .H ) 6(E . d 3 p ~ . . the various variables must be changing continually.19).17) put no restriction on the labels so the number of states has been overcounted N ! times.+ In 2 5 With this result we can find the differentials of all the quantities. This point has been mentioned at the beginning of this chapter. . Now we write down the answer first: S= lnr = N In [ ~ 47rmE (3Nh” + . .16) and (5. because each particle is identical and interchanging the labels of a pair of particles will not change the state. such as the heat capacity or the expansion coefficient can all be calculated. because d E = pd V + TdS. should we again divide the statistical result by N ! times? This will be discussed at the end of this section.
At other temperatures and pressures h and r can be calculated by the formula v 0: n a n d with T l p replacing n . ( 5 24) in order for the basic assumption to be established. v = 5 x lo4 cm/sec and r lo’ sec. h (5 22) is the mean free time. If L << h. i.2 3 ) then the collisions between atoms are frequent and the observation time t must satisfy the condition t > > T . then the change of momenta is always occurring. (5 .. The average distance between air molecules is nlI3 3 0 8 . Therefore even though there is no collisions between the molecules. the magnitude and direction of the momentum of a particle is always changing. Then r = .. the collisions between the molecules are caused by the interaction. i. there is energy exchange. If there is no energy exchange at all.V. then the directions of the momenta are changing but their magnitudes are not and this basic assumption cannot be established. If L>>h. where v is the average velocity. Notice that  A . V L ( 5 25) is the necessary condition.e. collisions are infrequent. Of course. If the molecules collide with the wall of the container. but (517) does not include the interaction energy. Hence. we have utilised the   .e. the distance between two successive collisions of each molecule. Let h be the mean free path of the molecules. 1 nu where n is the density and u 10A2 is the effective crosssection area of the collisions of the molecules. At room temperatures and normal pressure the mean free distance ofair molecules is A .lo’ cm = 1 000 A. then t>>. If the molecules collide with each other.74 STATISTICAL MECHANICS ideal gas change? The change of position is because each molecule has nonzero velocity while the change of momenta of the molecules is due to collisions either with the wall or with another molecule.
75 interaction causing collisions in order to satisfy the condition of the basic assumption. This is also often the most unclear step and the most difficult one. If the density is high. the details of calculations may be different and more complicated. The entropy determined by the basic assumption does not involve the structure or property of any substance in the definition. We must emphasize again that the determination of the region of motion is the first and the most important step. . and collisions are frequent. There is a reason for discussing this problem. the subsequent mathematical calculations are usually very complicated. The above result (520) is the entropy calculated from the basic assumption and it is the same as (331) in Chapter 3. If the molecules are all different we should ask: in the observation period how far can a molecule go? If the density is very dilute and the container is very small then each molecule can travel between the walls many times. If we do not use the definition in Chapter 3 then the p in each equation must be rewritten as p c where c is a constant. the constant c in the p must be zero. We shall discuss these stepbystep and simultaneously introduce new concepts t o help us understand the problem. in the period of observation a molecule cannot go + . However. the region of motion is still defined by (516) and (517) and we only need to drop N ! in (515) and (518). The main effect of collision is to change the momenta after the collision. It is not a coincidence that (5 20) is exactly the same as (3 31). Therefore. and the interaction only acts at the time of contact. Now we return to discuss the N ! factor in (515). These problems are very interesting and each has its own method of solution. In Chapter 3 we pointed out that once the chemical potential p is defined. Therefore. This does not lead to a contradiction. It can be defined by the conservation laws and the analysis of the chemical reactions. . but on the other hand we have neglected the interaction energy in calculating the entropy. if the entropy calculated by this assumption is correct. this is a guess. There is little choice for the definition of the chemical potential. If all the molecules are different.THE BASIC ASSUMPTION OF ENTROPY AND . But each substance will then have a different c and this is rather inconvenient. Once the region of motion is fixed. can we just discard this N ! factor? Of course. because the time of contact of molecule with molecule is extremely short. the gas then may not be ideal. then the entropy is defined also. It would not be easy to put N different molecules together for N >> 1. From the definitions (516) and (517) of the region of motion to the calculation of the various properties is a sequence of mathematical manipulation which in principle does not involve any new concept. Even if this were possible. The usual application of this basic assumption is essentially like this.
(The total number n is fixed already. The entropy will be different then. the 6 ( E . The density of the points is l / h 3 N uniformly distributed in this 6 N dimensional space and the region of motion is determined by E = H . These formulas are not valid for frozen impurity atoms which are dispersed in a solid.15) will be limited to the neighbourhood of a fixed frozen position and N ! should be omitted. Notice that if the atoms are frozen. and the situation is different from the unpartitioned case. 5. Each integration of rk in (5. Of course. 5. If the restriction of this time scale cannot be definitely defined. during the observation time the distance travelled must be much larger than the average distance between the solute molecules. Each factor has its own meaning and cannot be inserted or dropped at will. The region of motion is part of this space and is determined by the total number of particles. There is nothing wrong that S is not proportional to N . there are L positions and each has 2 states. Each point is a state. The positions of the frozen atoms are unchanged. it is impossible to divide the body into parts with the same property. the region of motion must be determined by the molecular motion. Equations (515) and (518) are not only appropriate for gases. but they can be applied to the solute molecules in a dilute solution as well.76 STATISTICAL MECHANICS very far away. whether or not the atoms are the same. if N ! is dropped. i. In the example of Sec. Hence the restriction (516) is not enough and we have to add restrictions due to the finite time. i. To sum up.e. Some Special Features of High Dimensional Space In the simple example of Sec. This is the case of a small container with no collision between the molecules. These shall be analysed when we discuss impurities in detail in Chapter 18.3. However the following condition must be satisfied: the solute molecules must be moving fast.H ) in (518). We shall frequently discuss the time restriction and the problems involved in applying the basic assumption. then molecules at one side cannot go to the other side. that is to say. the phase space has 2L states.5. then the basic assumption is useless. If we partition the container into two parts. 5. In the last section the phase space is defined by points in a 6N dimensional space. then it does not matter whether the atoms are identical or not.e. As the molecules are different and travelling between the walls. S will not be proportional to N and is not additive.3 the phase space can be considered as the 2N vertices of a cube in .) The region of motion can be regarded as the surface of a sphere.
Now let us look at the volume of a sphere.THE BASIC ASSUMPTION O F ENTROPY AND . Let x = ( x . The length of x is defined as (5 .. geometrical concepts are always very useful. such as 10". The use of the 6 N dimensional space in the last section is a very good example.26) Suppose N is a very large number. The representation by the 6 N dimensional space is not necessary. hence the phase space is related to some properties of a high dimensional space. and the surface area of the shpere is Q ' ( R ) = N C R N . A x i lAxl . Conversely. .EN"^  E. then its volume is Q ( R ) = C R N . This is a large number. 77 the L dimensional space. The important property of a macroscopic physical system is the large number of variables of motion. . The constant Ccan be determined as follows. The phase space in the above section can also be represented by the population of the particle states. zero. However. a collection of a finite number of discrete points has no geometrical meaning.. . (5 28) Even if E is not large. the change of each xi can be very small. 1 A x I can be very large. then If at least some xi are not (527) 1x1  .xn) be a point in an N dimensional vector space. (5 29) . then . The most special feature is the volume in this high dimensional space. Let = (27r)N/2 . . . x 2 . In this section we introduce some geometrical properties of high dimensional space and discuss the law of large numbers. If each xi changes by a bit. Strictly speaking. but the geometrical meaning of the region of motion is not so clear. if the points are numerous. Let the radius of the sphere be R .' . even though x may traverse a large distance in this space.
The special properties above are due to the fact that this space has many directions. we get I = 1 ca ‘Ra dRQ’(R)e 2 0 = NC 5 m 3R’ dRRN‘e 0 (530) Comparing (529) and (530) we get C. Essentially. the length in a high dimensional space is a large number. then Q ( R ) will increase to Even though E is not large. Hence the volume of a sphere is concentrated on its surface shell. 5. but the volume is an “enormously large” number Length Volume  a m . .78 STATISTICAL MECHANICS If we use spherical coordinates. Hence the volume of a sphere is (531) If the radius is increased from R to (1 + E ) R . Peak Integration and the Law of Large Numbers The following uses the derivation of the law of large numbers to illustrate . (1 + E volume of a spherical shell of width ) E can be an enormous number.6. Hence the ~ is This is not much different from the volume of the sphere in (5 32).aN where a is a length scale. (5 34) .
of the x’ integration can be taken as oo.) + .N as ax x 1 = 0 (5 36) Its solution is X . Equation (538) is the socalled law oflarge numbers or Stirling’s formula. . .)’ + . S ( X ) = S(X... and its width is . This width is much smaller than N . The 0(1/N) term in the equation can be determined by a more carefully calculation. Near X . 19 some common techniques in mathematics. X N e . Hence the lower limit. .. The above method of integration is applicable to peaklike integrands. = N .f%. The maximum is easily obtained: =. X N increases rapidly with X while ex decreases rapidly and e s ( x ) is very small except at the neighbourhood of its maximum.( X . = N .” f dXeXXN S(X) NlnXX (5 35) If N is a large number. .THE BASIC ASSUMPTION O F ENTROPY AND .x is a peak with centre at X .X. We write N ! as an integral: N! = 1. azs)c 2 (5 37) (axz Hence (5 38) Roughly speaking.
The Law of Equipartition of Energy We next consider another example for demonstration.=T 3N 2 . This law states that if the total energy function is the sum of M square terms: H = 1 Aiq: i= 1 M .7..e..e. R (539) Hence. + p i z ) . the law of equipartition of energy. the volume of a sphere Q ( R ) and the surface area of the sphere Q’(R ) can be simplified as Q ’ ( R ) u exp [$(In 7 1lnnN 2nR2 + 1 N + 1nN ”)I.tal energy H (see ( 5 . Hence (541)shows that the average energy of each term is T / 2 . i. This is an example of the law of equipartition of energy. E 1 . (541) where qi are coordinates or momenta and A j are positive constants.T . then the relation of the total energy E and the temperature is E l .17)) is the sum of 3N terms (each p i is in fact three terms p i x + pi. (543) The to. 1 lnQ(R) N = lnQ’(R) N (540) 1 5.e. when N >> 1. i.. Now we go back to result (521)for the ideal gas. . each variable 4iacquires an average energy T / 2 .80 STATISTICAL MECHANICS Using (538). M 2 (542) i.
x i ) .. The integration of (546) is also the volume of a sphere. dx.A. In this section we use the point of view of projection in geometry to make this meaning more concrete. (546) xi 5 JATqi . dqm 6(E . . Hence the total energy is E = (:) T h " h " 2N=NT. . 81 Another example is a system of simple harmonic oscillators with total energy H = N 1 1 (. Using (540).THE BASIC ASSUMPTION O F ENTROPY AND . we get S(E) = lnr(E) ="(In + 2 . Perspectives of the Region of Motion From the beginning of this chapter. so H has 2N terms.rrE Mh2 1 E 1>'f lnAi. we have emphasised that the meaning of the region of motion is the magnitude of change of the various variables. . . What we see are the projection of the three .M T 2 2 i=l (547) T is only related to the ( M / 2 ) In E term. . .? + i= 1 (5 44) Here each pi is a variable. The above discussion of the complicated techniques may make the reader. )lI2 5 dx. 5. . The calculation of r ( E ) is similar to that of the ideal gas case: r(E) = Idq. forget this meaning. . . . . . and each term. dx. . (545) The deduction of this law is very simple. . 1 1 dq. 2 a a s E l.. contributes an energy T / 2 . We first digress on some basic facts about projection.8.H ) = (A. Projection is quite familiar to us in our daily life. according to the above law. A . S(Ex: xi .
(548) What is its projected density on the x y plane? The answer is In this equation r is the total number of points and the projection is a disc with high density of points near the boundary. . . d z d y 6 ( J x 2 + y 2 + Z’ . Suppose points are distributed uniformly on a threedimensional spherical surface x2+y2 +z2 = R2 .. because (552) .. If n >> n’ the density concentrates at the centre. We use an example to illustrate the method of calculating the density distribution in the projected space. a uniform distribution on a nsphere gives in n’ dimensions a projected density y(xI.x n t ) a 8(R . (550) The meaning of this integration is to sum all the points perpendicular to the projected surface. Generally speaking.+ x i . .+ . . we need to integrate once more. For the same reason. points in an ndimensional vector space can be projected onto an n’dimensional base space (of course n’ < n ) . Engineering plans are also the projections of the three dimensions on a twodimensional plane. and are perspectives in various directions.x2. It is a crosspoint. We have to examine the projection of a collection of points in the various directions and its density distribution. . The reader can find out the reason for this himself. To get the projected density on the x axis.R ) T.r ) ( R 2 r2)(nn’)121 (551) r2 = x:+xZ.82 STATISTICAL MECHANICS dimensions on our retina.
y(pI )d3p1 is the total number of states in d 3 p l for the first particle and H' is the energy after taking out the first particle. The region of motion of an ideal gas is inside the surface of a sphere in 6 N dimensions and the states are represented by discrete points with uniform distribution. Nh Nh3 0 : exp{+  } P2 2mT ' (555) This result is in fact (552) where 2 m E = R 2 . p 2 = r 2 and T = R Z / N . 83 and if M is a large number. Now let us project the points in the region of motion onto the p1 space. . 5.5 3 ) p1 is not integrated but is fixed. The density distribution on the projected line is (553) (5 54) In ( 5 . Equation (552) is a normal or Gaussian distribution.THE BASIC ASSUMPTION OF ENTROPY AND . . The .6 and will often reappear. this distribution is essentially concentrated within a radius of This is very different from the situations in (549) and (550). From the distribution of points on the projected plane we can get the answer. The magnitudes in the various directions of this region of motion are the magnitudes of change of the various variables of motion.18). E ) in ( 5 . Hence according to the definition of l? ( N . How can we measure these magnitudes? The most direct method is to project the region of motion in the various directions. This distribution has been discussed in the peak integration in Sec.
The above analysis of projection tells us what the region of motion looks like in different directions. Hence the projection of the region of motion on the p1 direction or any pi direction is a normal distribution.84 STATISTICAL MECHANICS meaning of the various symbols in (555) is very clear: 1 T P = WN ixl + as .1 and compare with the discussion of this section. (557) (We do not consider the other differentials. From here we can return to Sec. are results of the basic assumption. . e. projection is statisticsby classification. this result is established. . We can use a bigger space as the projection plane.g. Generally speaking. These. The magnitude of this region of distribution is about Ap= m.= T (5 56) The relation between the chemical potential p and S can be obtained from the discussion of work and heat in Chapter 2 : dE = pdN TdS . We group together states of the same class and count their number. 5. The above analysis can be slightly generalised to give the density distribution of the projected points: If n << N .) The second differentiations of S with respect to N and E are quantities of order 1 / N . (5 . the ( p1 pz) space or an even larger space. of course.58) This can be regarded as an estimate of the magnitude of change of the momentum of the molecule..
9. If two molecules collide (e. (Of course. The trajectory of the fly fills the room. But the phase space of N molecules is not quite like the room because N is a large number.. the molecules move also. The space of p is a 3 N dimensional space. the trajectory in the region of motion is a very tiny thread.N(T/r) . i. pN) = p be the momenta of N molecules. molecules 1 and 2 ) . Trajectory and the Region of Motion Now we have some idea of the geometrical property of the region of motion. How many states are visited during this time? The number is about n . 5. The length of each q is different. Each collision changes the state once. After n collisions there are n q vectors. Let ( p l . from the point of view of geometry. The region of motion must be defined by the trajectory. i. ) A number of order N is far less than e N .e. The volume of the phase space is a large number of order e N .g. then q12 and qii are orthogonal vectors. and n is larger than (560) but is still a large number of order N . 85 5. For simplicity we omit the position space. Here. . we give a crude treatment. that is the length of the sides of the region of motion. Their average is the magnitude of the change of momentum of each molecule. . Let the next collision be between i and j . and the state moves to p + ql2 + qii. then the momenta of these two molecules are changed from p l . How can this tiny thread fill up this enormous volume? This question is hard to answer. p z .2 we used a fly moving in a room to illustrate the meaning of the region of motion.THE BASIC ASSUMPTION O F ENTROPY AND . Let That is to say. pz to p i . only IJ . . We shall make a more detailed analysis in Chapter 25. not just 3 . The trajectory is the set of states during the observation time. this tiny thread can remain in an extremely small region. At first sight. p i . Notice that if 12 and ij in ql2 and qii are different molecules. in between collisions. (562) because q12 has only the nonzero components a = 1 and 2 and 4. and n is a large number of order N ... (5 60) where r is the mean free time and T / r is the number of collisions in the time T . In Sec. state p moves to p + q l z .e.
In Chapter 6 we shall analyse these problems further.3. like the pillars and rafts of a house. (b) Find the relation of the population with T and p . The states in the trajectory is very scarce compared to the total volume. In building a house. Problems   1. Discuss the relation between them. but it can. total magnetic moment M = i=l c N si is just . Compare their magnitudes. The trajectory is linked by the various q vectors. (5 63) where si can be + 1 or . Let a be the length of the side of an Ndimensional cube. (a) Calculate the entropy. how can we do this? The answer is that the rod should be mutually perpendicular so as to make up a frame. Its volume is a N . Now we are in a 3 N dimensional space with n 3 N q vectors as the rods. calculate the entropy. We now need a bit of imagination. the trajectory can make up a frame of the volume of a 3N dimensional space. n N . (c) This model has many points of similarity with the fermion model in Sec. (b) Find the relation between the energy E and the temperature 2 The ' . The above is a crude explanation for the relation between the trajectory and the region of motion.1. Because the q vectors are mutually perpendicular.E / h .86 STATISTICAL MECHANICS those OL = i and j . The orthogonality of the various q vectors will be explained later as the independence property. Apply the analysis of Sec. 5 .) . 3 . To make up a frame for the 3N dimensional space. In the n collisions. make up a frame for a finite volume. If the rods are distributed evenly in all the direction. The magnetic field h is a fixed constant. the q's should be mutually orthogonal. the n q vectors are almost all mutually orthogonal because the molecules in the different parts are colliding. The Hamiltonian of a paramagnetic substance is H = i=1 c N sih. If we have several rods and we want to make up a frame of a space. they cannot make up a frame for the space because there are eN directions requiring eN rods. The change of the spin si is determined by a certain rule of motion. the pillars and rafts must be perpendicular. We disregard this here. (a) Using the basic assumption. 2. ( N is a large number. Calculate the volumes of the circumscribed and inscribed spheres. 3 to bosons. 5.
. i = 1. . 5. . Here it is regarded as a projected distribution from a high dimensional sphere. . . 8.1) and there are 2" such vectors. (a) Considering the air resistance. . ( . Call these vectors R. If we include air resistance. (b) Calculate the energy. . .. (a) From the following equation calculate the frequencies of vibration. (c) Is the result of (b) in the last problem influenced by r? If t h e string is immersed in water. The vertices of an ndimensional cube can be represented by the coordinates (1. . . the entropy and the heat capacity as a function of temperature. 7 . (1. 1 . (b) The observation time should be longer than the fluctuation period T so that equilibrium can be defined. . . .e.1 . i. l ) . . .). . will the result be different? 6. (b) Draw the projection of the 4. . find 7. the Boltzmann distribution already discussed in Chapter 3. These pictures can give the reader some impression of higher dimensional spaces.THE BASIC ASSUMPTION O F ENTROPY A N D . . ' . n = 3 .and 5 dimensional cubes.. assuming that the highest frequency is wD . (1. (565) Then the projection of R i in the x y plane is (a) Draw the projection of the threedimensional cube. calculate t h e correction term t o the wave equation. Calculate the projection in the s1 direction of the region of motion in problem 2. of course. . l ) . . Projection of a cube. Equation (555) is. A uniform string of mass M and length L is fixed at the two ends with tension F . ey*ev  1. . Avoid the overlapping projections of the vertices. . . 1. 2 " . . . 1. 87 4. 2.1 .eX = 1. e. Derive (551) and (552) and clarify the relation between (554) and ( 5 52). 1 . . . . Assume that the string in the above problem is in a room of temperature 2 Let d be the diameter of the string and 77 the coefficient of viscosity of air. e x * e y = 0 . 1 . . Take any pair of mutually perpendicular vectors e x and e.
since the exchange of energy between the two parts is mediated by Hlz. But its main function is to maintain energy exchange between the two parts.e. etc. such as the problem of the vibration of the lattice.Chapter 6 SOME ELEMENTARY APPLICATIONS OF THE BASIC ASSUMPTION In this chapter we start from the basic assumption and derive some properties of the equilibrium states. Now we start from this basic assumption to see the relation of the entropy of a system with those of its subsystems. the fact that the temperature.1. We use these properties to analyse some examples. This is not to say that it is unimportant. 6. i. are the same throughout the system. Divide the system into two parts 1 and 2 . Systems and Subsystems A system can be regarded as being composed of its subsystems. the state of the system is fixed also. After fixing the states s1 and sz of the two parts. Now we neglect Hlz because it is only effective at the boundary and is very small. pressure and chemical potentials. Therefore The total energy H is the sum of HI + H2 plus the interaction energy HIz across the boundary. and its contribution to the total energy . including the Debye model and second sound.
Consider a particle whose potential energy is V(r) but whose position is constrained on a curved surface @(r) = 0 .SOME ELEMENTARY APPLICATIONS O F THE BASIC ASSUMPTION 89 is very small. (6 . hence we must find the maximum of Sl(El) Sz(Ez). Equation (65) shows that V + A@ . @(r) = 0 . the . E l and E2 are not independent variables and are constrained by El + Ez = E. + However. the equilibrium point and the value two equations in (65). This constraint force is normal to the surface (neglecting friction) and hence is proportional to V@(r) and can be written as hV@.5 ) measures the magnitude of the we should find the minimum of of X can then be solved from the restriction to the surface @(r)= 0 The parameter (or Lagrange multiplier) h constraint force. where is the minimum of V(r) on this surface? The particle is not only acted upon by a force VV(r) but also by the constraint force of the surface. (63) where We wish to apply peak integration on (64). So we drop it for the moment. We can use the method of Lagrange multipliers for this problem and we now introduce the principle of the method. The entropy of the system is then S(E) = l n r ( E ) . At equilibrium this force should cancel V(r) : vv  XV@ = V(V+ A@ ) = 0 . For the particle. Where is the equilibrium position? That is.
asz/ a E z = Tz are the temperatures of the two parts. Ez = S1 + Sz around the neighbourhood of this maximum is E. the integral of (64) is maximum. We use the method of Lagrange multipliers to find the maximum ofSI(El) + S z ( E z ) . be the solution of (66). El)' + (Ez 1 K2 &)' + . ( E .. Equation (66) shows that When the temperatures are the same.5 ) we write where is the strength of the constraint force. (68) . The meaning of (66) is obvious: asl/?)El = T 1 . . then the expansion of Let E l = El. According to ( 6 .90 STATISTICAL MECHANICS gives a constraint force with magnitude h and this force keeps the particle on the surface.. We now return to (64).  t[:.
because only these kinds of states are in the region of motion. The product form r l ( E l ) FZ(E2)is the result of statistics by classification. This statistics by classification groups together states of a special kind. 1 Hence the entropy of the system is assuming K I K z> 0 . The above calculation points out to us that r1(El) and r z ( E 2 )are maximum at ( E l .SOME ELEMENTARY APPLICATIONS O F THE BASIC ASSUMPTION 91 (El + E. and that the number of states in the neighbourhood of (El Ez) comprises nearly all the states. i. Substituting (68) into (64) we get exp[ =. The above derivation seems very simple. r ( E ) is the integral of Fl(El) F Z ( E z ) (see ( 6 .10) S(E) = Sl(E1) + SZ(E2) + h ( m ) . * Ei2 2 Kz K = K I K z / ( K+ Kz) . The word “class” refers to (E17 z ) .e.4 ) ) . E El + Ez = E . Equation (64) sums up the results of this statistics by classification. and obtains the volume r ( E ) of the region of motion. and S1.. The value of each (El.1 E.S2 are of order N . as it is of order InN. (611) The last term can be neglected. but we want to further explain some obscure points. The integration of (64) is termed statistics by summation. the energy of the first part is El and that of the second is E. “Neighbourhood” here means (612) . Of course. Ez).) The first order term in (68) is zero under the condition (El + Ez) = 0. adding up all the kinds of states. (6. Ez) is a class. N being the total number of particles.
energy Hi (si) and Sj(Ei) = In r j ( E i ) . .10) ).” This result is peculiar. Once a variable is changed the changes of these superlarge functions are enormous. This class with the largest number of states is just the class with T1 = Tz . m . We divide the system into m parts.e. the entropy obtained after summation is the maximum of the entropies of the classes. (614) The total entropy of the system is . This conclusion constitutes a bridge linking equilibrium phenomena with the basic assumption of statistical mechanics.)are few in number. That is to say . . The above discussion can be generalised as follows. Consequently only the maximum values are important. 2. A?.“the whole region of motion is nearly occupied by the states in this class.e. The functions F1(El) and Fz(Ez) are all superlarge numbers. (6.92 STATISTICAL MECHANICS This is because rl(El) rZ(Ez) 0 : exp (612’) (see Eq.) That is to say. Sz N . It can be said to be the property of a superlarge number of order e N . . the two parts with the same temperature. This summation is in fact the step required to find the maximum and yields the class with the most states. ri(~i) =   C a(EHj(si)) Si . the (Although AE but S1. Hence states far from integration of (64) gives (El.i. each specified by the state si with entropy Si(Ei). i. . The entropy of this class is very nearly the sum of the entropies of the parts. and they are far larger than the nonmaximum values. i = 1 . Hence the maxima of the functions are quite disparate. hence ln A E can be ignored. The other classes have a minute effect on S ( E ) .
. + E m . besides classifying by energy we need also to make a classification according to particle number.14) and the generalisation of the conclusions (6. .17) is rn S(E.11) and obtain (616) T and Ej are the solutions of the following equations E = E. + . we regarded the particle number as unchanged in each subsystem. In the above. + E. p are the solutions of the following set of equations: Ni . statistics by classification is not limited to energy classification.16) and (6. (6 . Assume that the total particle number is a constant: rn N = i= 1 Ni(si) . N i ) = In ri(Ei.N) = i= 1 Si(Ei. Of course. We now discuss an important example.Ni) .(s.Nf) . i. The above result can be generalised to any variable of motion.20) Ei. .) of each subsystem become the variables of motion.e.). then the particle numbersN.SOME ELEMENTARY APPLICATIONS OF THE BASIC ASSUMPTION 93 We can repeat the calculations of Eqs.N.Nrn(s. . . and T . (618) Let S j ( E j . (617) The derivation is left as an exercise for the reader. If there is exchange of particles between the subsystems. .(s. (64) to ( 6 . We only require this variable to be a number of order N .). This is a generalisation of (6.
. 4. 4 .Em . 6. In fact we assume that entropy is a maximum (statistics by classification is smaller than the summation. We now look at some applications. ( 2 ) the simple harmonic oscillator. We had dwelt with the statistics of states of the free particle. The above discussions show that from the basic assumption we can directly derive some properties of the equilibrium states. giving us confidence about this assumption. But the assumption is still an assumption and not a proof. + N2 + . . Each fundamental vibration has its own frequency. The small amplitude vibration of a solid can be regarded as a collection of simple harmonic oscillators.94 STATISTICAL MECHANICS E = El N = N. Later we shall encounter more of this. (3) the hydrogen atom and (4) spin particles. (622) That is to say. 1 . Now let us look at simple harmonic oscillators. Vibrations Quantum mechanics can be applied to many problems.. (621) Hence each subsystem has the same temperature T and the same chemical potential p . and the total entropy is the result of the statistics by summation). 2 .. According to quantum mechanics. +Nm . the state of vibration can be specified by the integer n.. 1 n = 0. or “statistical mechanics proves that entropy is a maximum”. + E. Further consideration leads to the conclusion that each part has the same pressure if we assume that the volumes of the different parts are also the variables of motion. If there are N fundamental vibrating elements. a vibration of frequency w has energy En(h’ 1): 3 En = ( n + ?)a . The reader should notice some incorrect motions such as “statistical mechanics proves the existence of equilibrium”. but there are four cases giving very simple results. t h e n the state is specified by N integers .2. These are the basic applications.. They are (1) the free particle. .
.. (6 .23) Now let us calculate the entropy: S(E) = lnl?(E) . with total energy H(s) = 1 (ni+ T ) W i i= 1 1 N . n N ) = s.27) is regarded as an integer.SOME ELEMENTARY APPLICATIONS O F THE BASIC ASSUMPTION 95 ( n . = f n =o n =o 5 .26) then the Sfunction in (625) isa (6 . . Let (6 . Then the entropy of this group is The summations in (625) are easier to do because there is only one frequency ma. n 2 . The variable in the 6 function (627) is also regarded as an integer. elements with frequency m. We calculate the entropy for each group and use the results of the last section to sum up the entropies. We define 6 ( M ) = 1 if M = 0.. Let the (Y group have N.. 6 ( M ) = 0 if M # 0 and M is likewise an integer. . nN=o 2 S(E  1 (ni + 7 ) m j 1 N i= 1 These sums are not easy to evaluate. . . . Let us divide these N elements into groups so that frequency is nearly the same in each group.
. balls into N. . .28) y&y) = The reader may derive (628) himself. (626) and (632).l)! . Ma (632) 1 . Hence. According to the result of the last section. and take ni as the number of balls in the ith box.l)! o. we obtain (633) .) = lnr. Using the law of large numbers lnN! N 1nN .In M. lnM. lnN.M. + N.) In (Ma+ N . is the total number of ways of putting M. (631)  1 .96 STATISTICAL MECHANICS Suppose we have N boxes. ) . This is not the same as the calculation of the surface area of a sphere in the last chapter. 46 . (628) can be simplified and we get: S.): 1 (M.(E.N. T From the definition of M a . .16) and (6.M.17) ): (6 .) = E.(E. x (Ma+ N. so as to satisfy (625).29) (630) and T must be solved from (630).N. r. but the aim and the principle are the same. the total entropy is the maximum of the sum (see (6. boxes (divided by a. . the total number of balls must be M.!(N.+N. 0 .
/T << 1 = T q 1 + $)+ *($) (637) Substituting into (633) we get (638) . Then the sums in the above two equations can be replaced by integrations: where g ( o ) is the frequency distribution of the vibrating elements and depends on the characteristics of the. the frequency distribution is very dense.) This is the socalled zeropoint energy. of course.SOME ELEMENTARY APPLICATIONS O F THE BASIC ASSUMPTION 97 The relation between the total energy and T i s determined by (630)./Na = y w a .1 Similarly from (631). (See Eq.e. the energy of each element is E.physical system. a very important term. (633). C = E INa( a 1 (634) ewa/T . 1 As T + 0. when o. i. (632) and (629) the entropy is found t o be (635) In many applications. we first point out the significance of the zeropoint energy. At high temperatures. Before giving an example of g ( w ) . i.e. This is.
Notice that this zeropoint energy is still very important. Were it not for the zero point energy. = 0. . whose frequency is inversely proportional to its wavelength. Hence the normal mode is specified by three integers. then w1 = c .e.3. . ? 2 . If we use cyclic boundary conditions. of course. (640) assuming the crystal to be a cube of length L . while I denotes transverse wave. The Debye Model Now we apply the above results to the lattice vibrations. . the result of the law of equipartition of energy discussed in the last chapter such that each element has an energy T . The calculation of g ( w ) is similar to calculating the energy distribution of the states of a particle in Chapter 3. We first solve the elastic wave equation for the plane wave (or stationary wave. It will be necessary to find the frequency distribution g(o).e. n y . i. because it cancels the term w. The vibration at low frequencies is the elastic wave. The structure of a crystal is very complicated and the analysis of its vibrating modes is a special topic in itself./2. 2n/( wavelength). + 1 . . then k must satisfy n x . Equation (639) is the law of elasticity and is valid for long wavelength vibrations (Ilk >> lattice spacing). Let k be the wave number. The term O ( w i / T ) can be neglected when T is very large. Here we discuss a very crude estimation. 6. k . n. and cII and cI are their respective speeds of propagation. (639) where II denotes longitudinal wave.98 STATISTICAL MECHANICS The first term T is. the energy at high temperature would have a constant term. i. the Debye model. depending on the boundary condition). The frequency distribution (according to the definition (636)) is (641) .
e. 6 . (642) is obviously not correct. (644) If w < wDwe use (642) for g ( o ) . then N = 3NA because each atom has three degrees of freedom. From (643) we get w:. (636)).1+ . The most important example is the heat capacity (Fig. and this g ( o ) fixes the Debye model. The frequency wD is called the Debye frequency as determined by g ( w ) .2 c: c: (645) Here wD can be determined from the elastic wave. i. We can change the summation in (641) into an integration to obtain (642) This is valid for low frequency vibrations. When w is large.. However. let us set wD as the maximum frequency. If the crystal has NA atoms. We also have to notice that s g(w) dw =  N v ’ (643) is a necessary condition (see Eq. and one kind of longitudinal wave. we are just making a rough estimate. so (642) can be used throughout. The various equilibrium properties can be derived by differentiating (634) to (636). and right polarised). = c3 V 2n2N 9 3 _ c3 = .SOME ELEMENTARY APPLICATIONS OF THE BASIC ASSUMPTION 99 Note that there are two kinds of transverse waves (left polarised. g(0) = 0 9 0 > W D . To satisfy (643).1): (646) (647) .
/T. the ratio Cv/NAof all solids should be the same function of 0. This conclusion agrees quite well with experiments. 179. . The reader can derive these for himself./kT f 0. (from elastic coefficients) K 3 20 NaCl KC1 Ag 230 225 246 215 Zn 308 305 bSee Kittel (1966). p. Notice that we have used units with h = k = 1 . 61 Plot of the heat capacity C as a function of y (see (646). (Notice that the various Debye temperatures are close to room temperature./T where 0. (647)). is the Debye temperature.1. then o D / T should be replaced by ho.1 Debye temperature Solids OD (from C . This is a universal curve for all solids.) b Table 6 . From Kittel (1966) p. as can be seen from Table 6. According to (646). .100 STATISTICAL MECHANICS Fig. 132. ) K 308 0. If we use the usual units.
) . (646) should be very precise.SOME ELEMENTARY APPLICATIONS OF THE BASIC ASSUMPTION 101 The righthand column in the table is the value of 8 obtained from the . the medium temperature case is not expected to be too bad. The agreement is actually not bad. Figures 6 .) The curve here is calculated from (642). (645)). 62 A more precise graph of the vibrational frequency distribution of a lattice. Now the question arises: Why is this crude model so successful? The main reasons are: (1) the low frequency part of g ( w ) is precise. only the integral of g ( w ) (the total number of modes) appears in C.. If the Debye model is completely correct. (Again see Kittel (1966).2 and 63 show more precise calculations of g (a). i. As the high temperature and low temperature cases are good approximations. Fig.e. 177. with a cubic crystal. for T >> O D .p. hence when T << 0. the two columns should be identical. hence the result of (646) should not be bad for high temperatures. (2) at high temperature. elastic coefficients (see Eq. but the integration of k is taken in the first Brillouin zone. (Diagram by Fang. The middle column is obtained by fitting (646) with the experimentally determined C..
Hence it can be regarded as the motion of a collection of particles. Phonons and Second Sound The most special feature of a simple harmonic vibration is that its state can be specified by a positive integer ni and the energy of this state is niw i+ wi. See books on solid state. 1 Fig. Hence it is natural enough to think of ni as denoting the number of a certain particle. 0 8 09 . If the solid is a crystal then the vibrating element is a plane 3 .102 STATISTICAL MECHANICS 1 . 6 3 A more precise graph of the vibrational frequency distribution of a lattice. p. These particles are bosons because ni can be larger than 1 (see Chapter 3 for the discussion of the particle state and its boson property). obtained by the statistics of 2791 wave vectors (see Kittel(1966). 07 . the vibration of a solid can be thought of as the motion of 3NA normal modes. where wi is the frequency of vibration. Each particle has an energy wi. 177).4.(low frequency) I transverse I I 0 Q1 0 2 0 3 0 4 0. 6. 06 . : .5 . each vibrating element being a single particle state and ni is the number of particles in this state. . The first Brillouin zone of the wave vector space is related to the symmetry of the crystal structure. As discussed in the last two sections. The vibrational frequency distribution in aluminium.
But now we have a phonon gas and there should be a sound wave in this gas and this sound wave is propagated in this phonon gas and is different from the phonons themselves. (11) in Chapter 1). Let us review the theory. Let us look at this interesting problem. we can immediately obtain the wave equation a2 p i = at2 C2V2P' 9 (6 49) (650) where c is the speed of propagation. namely the sound wave. Because we are familiar with the analysis of a gas. Hence ni is the number of particles with momentum k ( = hk. The vibration of the lattice can then be analysed from the point of view of a Bose gas. Assuming that the amplitude of the sound wave is very small. From the analysis of this phonon gas. and p is the density when there is no sound wave. then (1 . we can obtain some properties of a solid related to vibrations. at a P' (647) (648) Here p ' represents the change of the mass density. Phonons are elastic waves and can be regarded as sound waves. There is a very common phenomenon related to vibrations. described by the wave vector k . if a collection of particles can be represented by a gas we can immediately apply all the known techniques of analysing a gas to deduce the properties of this collection of particles. the density and pressure changes in the sound wave are adiabatic processes. From these two equations. The sound wave in common gases can be explained by the equations of fluid mechanics (see Eq. and these particles are called phonons. h = 1). Because the diffusion of heat is slower than the periodic change of the energy.SOME ELEMENTARY APPLICATIONS OF THE BASIC ASSUMPTION 103 wave.1) can be simplified as + p v * v = 0 . We call it second sound. If we assume that the period of vibration is much longer than the mean free time of the molecules and that the wavelength is much longer than the mean free path (see the discussion of local equilibrium .
But since each phonon has a definite energy and we can write down an equation for the conservation of energy: . (653) where ~ ( r t. We let q(r. Each phonon has its momentum. We now digress from the present problem and consider a body with momentum P and assume that this body is in free motion. our phonon is massless and. then (654) From the viewpoint of local equilibrium. E. t ) be the density of momentum. Hence a phonon gas has no continuity equation llke (647). To find these functions. q and J are functions of the temperature T and the flow velocity v . the change of momentum is equal to the applied force. i. . As the state changes. we have to do some preparatory work. ) is the energy density and J(r. so that P does not change.e. t ) is the energy flux. However. We must remember that the number of phonons is just a state of the normal mode. the number of phonons changes also.1 ) ) then c2 can be measured from the equilibrium quantities: cz = ( $ ) s (651) In the adiabatic process of an ideal gas (see Eqs.+ V J at ae = 0 . Hence the speed of sound in an ideal gas is c = 3m (652) We expect to be able to derive equations like (647) and (648) for the phonon gas.104 STATISTICAL MECHANICS after (1 . in addition. (224) and (227)). the total number of phonons is not a constant but changes continually. Equation (648) is just Newton’s second law.
The . S (655) where P (s) is the momentum of the state s.e. 2. P ) = 1 6 ( E . r ( E . . Following our previous discussions we regard the body as the sum of m parts with momenta P I . besides having the same temperature. IEm . P) = lnI'(E. (617) and (620).SOME ELEMENTARY APPLICATIONS OF THE BASIC ASSUMPTION 105 According to our assumption.P.60) This shows that v is just P/M. (656) &. The entropy remains the same because entropy is a statistical quantity and is independent of the coordinate system : (659) Therefore as ap as aE V T (6 . . . (621) is S(E.P) = i= 1 1 S i ( E i .. w h c h can be interpreted as the velocity of the body. P ) . . pi) m . .P(s)) . m . all the parts of the body should have the Same value of v. .H ( s ) ) 6 ( P . (658) Thus. (657) E = El P = P. . + P. the velocity of the body. pi and T . P. .. + P. + .where M is the total mass. . the entropy of this body is S ( E . + . + E. .PZ/2M. .. . The reason is as follows. . and the total momentum is zero. If we use the centreofmass coordinate system the total energy is E . . . generalisation of (616). v are the solutions of the following set of equations: = a si a pi  V T ' i = 1 . i. .
(661) where N. is the total number of states of this class of phonons. . ni is the number of phonons in the i . is fixed likewise. Let us take a phonon gas with total momentum P.) . v/T is a “force”. (664) ‘Notice that Q is an unimportant constant and can be neglected after taking the logarithm: .N.106 STATISTICAL MECHANICS In (657). the energy . l n M .. ) ln(M.t h state and P.) = lnr. is fixed. . We introduce the definition P. (Each class has the same frequency .) will be the same as (631) S.) The entropy of this class is S.) = 1 n =o m . 2 “Na = O &(pa  i= 1 c Na njk. As k. = M.) = (Ma N . We divide these phonons into classes. + .(P.63) S. because frequency is determined by the wave vector and polarisation.(P. is the total momentum of this class of phonons.) . and its polarisation is the same.+N. like 1/T. Notice that the momentum of each phonon is k.(P. Now we apply these results to the phonon gas.(P.. k.M . with the a class consisting of phonons with nearly the same momentum k and the same polarisation.> 2 rap. is fixed. (662) The 6 function in (661) may be replaced by‘ (6 . We need not use a 6 function such as to restrict the energy. a is also fixed and once P.lnN.
= dE. (666) This is nearly the same as (632) except that w. (668) The last two equations of (657) are (669) A A h Each phonon has an energy flux of w. .k. of course the total number of phonons in the a class. (665) we can rewrite (657) as d .k.*v)dM. i.*dP. .v. V T T . M is . is the velocity and k . because Ma. is replaced by w. and N are large numbers. We now sum over these classes using (656) and (657). Because P . . = 1 ew/T  (667) 1 Then from (666) we have for M. Let .k./V where c k .e. 1 T .SOME ELEMENTARY APPLICATIONS OF THE BASIC ASSUMPTION 107 Note that lna can be neglected. S 1 = (a. . = M.ck.(a) .
We did not take into account the difference between the frequencies of the longitudinal and the transverse waves in order to avoid unnecessary complications. E . J =I d3k . an aw 2: (672) . a by 7 k and from (669) and (670) we obtain the jd3 (2 energy density E / V .  k.v)k . J and q become functions of T a n d v. t ) and v = v(r.. We use these in (653) and (654) to discuss the motion of the phonon gas. we can recast (653) and (654) into equations invo!ving T and v. Now we regard T and v as functions of position r and time t . These scales have to be discussed further. Hence the total energy flux is h (670) Now replace 1N. Since v is a small quantity.v ) w . e . t ) . c * k k . the momentum density P/Vand the energy flux J : q = P v =3 3 3 jd3k n(o.(ak. Notice that we must not forget that r and t here are variables on macroscopic scales and d t and dv are quantities much larger than the scales of the motion of the phonons.k*v) n ( w ) . Because 2 is a function of T. T = T(r. Therefore. Using (671). The zeropoint energy has also been dropped because it does not affect the following discussion. q and J become functions of r and t .108 STATISTICAL MECHANICS is the direction of k .k * v . i ( w . (2 (671) The factor of 3 in the equations above comes from the three polarisations.
c2VT av at I (675) It follows that (676) .... In addition we have V ~ = ... The reader can verify that after T an a. . (674) where we used p = 1 E.V T E 3 3 aT 1 I a€ .TV*v aT at 1 3 . substituting (673). 3 = an ao  . (673) In the above we replaced w with c k and interpreted p as the “mass density” of the phonon gas.SOME ELEMENTARY APPLICATIONS OF THE BASIC ASSUMPTION 109 Substituting this into (671) we obtain 5: c2pv . (674) into (653) and (654) we obtain .V V= .
From the point of view of the wave motion the conservation of momentum is due to the homogeneity of the medium. We will assume that these interaction are strong enough to maintain equilibrium but weak enough so as not to effect the statistical calculations.e. (ii) The defect or impurity atoms in a crystal can produce similar effects. But if the number of short wave phonons is large. then the long wave phonons will be scarce too (the total number of phonons is a T 3 ) . The difficult part has not started yet. When the gas is too dilute. this result is the simple part of the analysis. totally disobeys the law of conservation of momentum. Hence we have the law of conservation of momentum. Now the speed of sound is large (around l o 5 cmlsec) and for . But the arrangement of atoms in a crystal is periodic with a fixed lattice spacing. The crystal is not a homogeneous medium and hence the total momentum of the phonons is not conserved. Hence the change due to (i) does not affect the total momentum of a collection of long wave phonons. If some assumption is invalid then (676) will not be true. resulting in an interaction between the various modes and causing the number of phonons in each mode to change continually.110 STATISTICAL MECHANICS We have finally obtained the equation for the sound wave which is called the second sound and its speed i s l / f i of that of the phonons.e. i. the analysis of the validity of the various assumptions. Thus only in perfect crystals is (654) valid. (677) so that short wave phonons will be few in number and (6.54) can then be valid. The cause of these changes are: (i) The vibration is not perfectly harmonic. Short wave phonons have high energy. Hence the second sound can only exist in a very perfect crystal at intermediate temperatures. our basic assumption would not be applicable. Therefore the temperature has to be very low T << eD . Vacuum is homogeneous. Let us now see what assumptions we have made in the above discussion: (1) The momenta of the phonons must change continually. the change caused by the defects and impurities. ( 2 ) Although the momenta of the phonons are changing. There is then a problem with Eq. (654). for the long wave phonons the medium is effectively homogeneous and this short distance periodic change does not have any effect. sound waves cannot propagate. Therefore. momentum will not be conserved. If there are no such changes. However if T is low. However. the total momenta must be conserved (if there is no pressure gradient). i. The change due to (ii).
the parts have the same temperature. General Conclusions. pressure with volume.scale changes are usually related to these invariant quantities. But if the frequency is too high. In equilibrium. although big perfect crystals are here to grow. But the analysis of it has high pedagogical value. If the whole body is moving. Slow .) This is the result of the basic assumption. Our discussion of vibration shows that we may classify according to the energy or momentum of the modes. The various parts of a body are not necessarily divided according to the volume occupied.and large . 6.SOME ELEMENTARY APPLICATIONS OF THE BASIC ASSUMPTION 111 wavelengths of 1 cm the frequency will be l o 5 . We will content ourselves by mentioning that. and that entropy can be calculated separately. Each class can be regarded as a part of the body. it is not generally regarded as important. The independence of the parts means that the motions of the various parts are not correlated. We should notice that each force is connected with an invariant quantity: temperature with energy. Average Values and the Normal Distribution The main point in this chapter is on the relation between a system and its subsystems. local equilibrium can only be maintained by a quick response of the phonons.d Because this second sound does not occur easily. there are reports that a crystal like CdS can maintain second sound at about 10 K. leading ultimately to wave motion in a phonon gas. This problem is rather complicated and is beyond the scope of our discussion here. this chapter discusses only the steps of classification and summation and this is the characteristics of functions of dSee Prohofsky and Krumhansl(l964) and references therein. i. The reader can deduce this from the basic assumption. When this force varies then there will be changes in the invariant quantities. Thus we also must consider the speed of response. The analysis of the sound wave is one such example. the method of calculating the entropy of the classes is meaningful only when the parts are essentially independent. . chemical potential with particle number.e. The uniformity of the forces is a property of equilibrium.5. then each part has the same angular velocity. flow velocity with momentum. (If it is rotating. Nevertheless. This is indeed a useful illustrative example. From the viewpoint of statistics. pressure and chemical potential. the forces in the various parts of a body have constant values. Our discussion started from the very abstract basic assumption. This property of independence is a very complicated problem and later we shall study this aspect more carefully. then every part should have the same velocity.
(683) . The function r ( E .A) (679) * Peak integration will be used to evaluate (679). etc.A ) : (6 80) Let its solution be A ( E ) . A ) r(E)= I dA r ( E . is a variable of motion of order N . A ) = SdA eS(E. In the states of energy E . and r ( E ) can be obtained by a summation of r ( E . Let Let us suppose that A^($) (678) This is the result of statistics by classification. If 2 is a maximum.A(E)) . The results obtained above are not restricted to invariant quantities like energy. particle number.15) we get (682) S(E) = lnr(E) = S(E. A ) . We now explain this point as follows. We can make classifications according to any macroscopic quantities.112 STATISTICAL MECHANICS large value. we collect those states w i t h 2 equal to A and count their number. is the distribution of states in ( E . Each value of A is a class. of course. We first find the maximum of S(E. then a necessary condition is (681) From (6 . A ) space.
H ( s ) ) & ( A .SOME ELEMENTARY APPLICATIONS OF THE BASIC ASSUMPTION 113 Example: A ( s ) = H l ( s l ) .H z ( s z ) . while 21. (see (61) and (62)). Therefore S(E. Ez are the solutions of (66) and (67).A ^ ( s ) ) S E+A EA (684) For the definition of r 1 ( E l )and r Z ( E 2 )see Eq. of course.A) = 1& ( E . The analysis just given indicates that from the basic assumption we can calculate the value of any macroscopic A”(s). (685) Equation (685). Then r(E. This involves the calculation of the entropy of the classes and then finding the maximum (see (680)). as discussed in the beginning of this chapter. The result of (680) is now cast in the form where Ez = (EA) 1 2 .A) = lnr(E. and H z are the energies of the first and second parts of the system. is the same as (66) and (67) and is the solution of (685). Here H 1 .A) = S( E + A L+ + Sz(Y) . A=& E2 A . (64’).
The usual experimentally determined quantities are time averages within an observation period: ( A " ) = 1 3 h 3 dtA^(s(t)) . As A ^ ( s ( t ) ) is nearly always equal to 2.our conclusion is that ( A )= A . besides calculating the entropy.e. The conclusion we arrived at is that in the region of motion.We only need to know the region ofmotion tocalculate A. 2 is the average value of A^(s) with respect to all the states in the region of motion. the value of A^(s) i s 2 nearly everywhere. (687) r(E) s i. The observation time is from t = 0 to t = 7 . * (689) Hence. The results derived above follow from the observation that in the region of motion A^(s) is very nearly A( 1 + O(Ki' )) and KA N is a very large number. it is not difficult to imagine that along the trajectory the values offf(s) are all nearly equal to A. (6 .88) where s ( t ) is the trajectory in phase space representing the detailed molecular movement. the basic assumption provides a definite rule for computing ( A ) . .H ( s ) ) A(S) . As the trajectory winds inside this region.114 STATISTICAL MECHANICS We see that the distribution of states in A is a normal one (686) Hence 2 can also be written as  1 c 6 ( E . The use of arguments of this type is typical in statistical mechanics. It is a region in phase space bounded by the trajectory such that the trajectory winds inside it. Let us now review the meaning of "region of motion".
why do we not just choose a state and calculate Why do we have to do the average? This problem involves many aspects. but in practice. Hence the above conclusion is just an extension of the assumption. i.e. if the trajectory is not too exceptional. If we require the trajectory to spend essentially the same time in each large region. A large number of the order N is a sum of many variables of molecular motion and these molecular variables are essentially independent because the system can be divided into many independent parts. Suppose we divide the region of motion into M large regions. Hence = 2 is not a surprising conclusion.n. Notice that we have not yet discussed a way of determining the region of motion from the trajectory.e. (2) This conclusion does not tell us that the trajectory spends equal time in each point of the region of motion. ( 2 ) can be comfortably established. Although this requirement is not stringent. i. M Let each region have the same volume. we know that this is because A is a large number of order N . In principle this is permissible. A? =A =A (A^) . The calculation by averaging is still necessary. Why do distributions like (686) repeatedly appear? From our derivation. for variables of order N . we can only admit that we misidentified the region of motion or that the basic assumption is wrong.. then the distribution of A is normal. i. any state allowable by the invariant quantities is included in the region of motion. which states that if A is the sum of many essentially independent variables. e. =A . This is a hard problem and will also be discussed later. Later we shall discuss the central limit theorem. The statement ( 2 ) can be expressed as follows: The time average of a macroscopic variable of motion is equal t o its average in the region of motion.e. the question of whether the trajectory actually satisfies this condition is still unanswered. We shall discuss this later. then (2) can be established. Hence the normal distribution of macroscopic variables can be said to be the result of the parts of the system being essentially independent. where M is not too large a number.. Because A^ is almost 2 nearly everywhere in the region of motion. If the trajectory prefers to stay in places where 2(s) f 2. This requirement is completely different from that of requiring the trajectory to pass each point. The reader should note that: (1) This conclusion is applicable for a macroscopic variable only.g. Now the determination of the region of motion is part of the assumption.SOME ELEMENTARY APPLICATIONS OF THE BASIC ASSUMPTION 115 The reader may ask: since the value of A^(s) in nearly every state is 2. It does not impose a stringent condition on the trajectory. not staying in those exceptional points for too long. Finally we want to mention the meaning of the normal distribution. choosing a state at random is not a simple matter.
3. Use the method of calculating entropy by classification and summation to find the relations of T . 64 Heat capacity Cv versus T. calculate the frequency distribution g( a). E for a system of (a) N bosons and (b) N fermions. Sec.. . The calculation in Sec. 2. This “order of magnitude” requirement is not simple. (690) i. Wilks (1 967). the low temperature T law (Einstein’s law). Can sound wave phenomena exist in this gas? Make a detailed analysis. Notice that the electronic heat capacity is proportional to T .4. r = O ( e ). If there is a large range interaction it may not be established. 5. I.5 we emphasised the fact that 5’ = O ( N ). see Chapter 4. (See Chapter 3. For a onedimensional crystal made up of N atoms. The second sound was first found in liquid helium. 6.( with N.e.) Make a comparison.4 is rather complicated. N 7. the reader may consult Woo (1 972). The diagram below shows the heat capacity of a certain crystal. Were this not to be so the basic assumption might not have applied. 3. paying attention to every small point. 6. but it is quite helpful for understanding the basic principle and technique of the calculation. Prove that the area of the shaded part is the zeropoint energy. 4. The reader should do the derivation himself.116 STATISTICAL MECHANICS Problems 1. t Fig. The theory of black body radiation is similar to the theory of lattice vibrations. 6 3 . Derive (646) and (647) and prove that at low and high temperatures T >> . Black body radiation is a photon gas. In Sec. 6.
Analyse t h e order of magnitude of its various quantities. . i . . = . Each spin si has interaction with all the other variables. . . . N . .$ J M 2 . 2 ) with the present one. where M is the sum of all si and r ( E ) can be immediately calculated out. j = 1 . ( 5 . Compare Prob. 2. The general discussion of the Ising model is in Chapter 17. (691) This is the socalled Ising model with long range interaction. Hint: Write H as .j s.SOME ELEMENTARY APPLICATIONS OF THE BASIC ASSUMPTION 117 Solve the following model: J H=Zsisi 2 i.
Example 1 : The iron rod in Fig. Let us consider a simple example. The inclusion of different external forces will result in different thermodynamic potentials. be the unstretched length of the iron rod.1.supplied by a weight (of mass M ).Chapter 7 RULES OF CALCULATION The basic assumption is a rule for calculating entropy. Now we can add to it the energy related to the interaction of the system with external forces and define various thermodynamic potentials. 7 . . In many situations this method of calculation is more convenient than that of calculating the entropy directly. the free energy is an example. These new rules of calculation in turn have the same implications and consequences as the basic assumption. where g is the acceleration due to gravity. and then derive the various rules for calculation from the basic assumption. Let X . 7. and hence f = M g . This rule can be written as many different rules for calculation. Various Thermodynamic Potentials The energy in the above discussion refers to the energy of the molecular motion in the system and its molecular interactions. One uses these rules to calculate some thermodynamic potentials and then derive the entropy and other equilibrium properties from these potentials. We first discuss the definitions of the various thermodynamic potentials and their implications.1 is elongated by an external force j”. except that the method of calculation is different. It may be called the internal energy.
K ( S ) is the adiabatic elastic coefficient of this iron rod. X ) is more convenient for the discussion of the other properties. If we do not hang the weight. then H ( S . This. X ) is more appropriate. of course.e. .RULES O F CALCULATION 119 Fig.3) That is. (7 . X ) = E ( S .fX . the external force f is balanced by the internal tension: f = K ( S )( X Xo) sI:( = (72) We may include the potential energy of the weight and we obtain H ( S . The internal energy of the iron rod is where S is the entropy and we have omitted the other coordinates.In equilibrium (z)s = 0 I (74) i. In equilibrium. X ) .1 An iron rod under an external force. Hence (74) looks nicer than (72). we have included the coordinate X and its interaction energy with the external force fX.MgX = E(S. then E ( S . If we have the weight. H is a minimum. In mechanics. we are accustomed to identifying the equilibrium state with the minimum energy.X) . is the same as (72). 7 .
The pressure comes from a weight on the piston: p = M g / A where M g is the weight. like H . H can also be called the thermodynamic potential at constant tension. i. F is the thermodynamic potential at constant temperature.C must be a minimum. denoted by a E / a S .fX . If the temperature of the air is constant. At a fixed T and f. (75) where T is the temperature of the air and its effect is like that of the weight f . is the same as the temperature T of the air. (77) which includes the energy due to the two external forces T and f acting on the coordinates S and X respectively. and G is the thermodynamic potential at constant tension and temperature. is a minimum. X ) .X) . vlre may define F ( S . (s)x = 0 9 so F .X) = E(S. F and G the thermodynamic potentials. The environment is another factor to consider. We call H . Following the above steps the reader can define the thermodynamic potential H =E + Mgy (78) = E + p V . . X ) = E ( S .120 STATISTICAL MECHANICS In the above we did not consider the air around the iron rod.TS .e. Here the coordinate S is similar to X in (73). In equilibrium.T S . 7 . Of course we can define still another thermodynamic function G(S. the temperature of the iron rod. Example 2: The gas in Fig.2 is confined in a vessel with crosssectional area A and height y .
fdX . r.RULES OF CALCULATION 121 Fig. If we regard f as the coordinate. then X becomes the force. T as a coordinate.S d T + fdX . (710) since d E = T d S + f d X . Hence if we regard. 7 2 A gas confined under constant pressure. This interpretation is followed by the above results.T S .Xdf = T d S . Likewise F is a function of T and X:F = F( X). then .S appears as the force.Xdf . We list the differentials of the various thermodynamic potentials in Table 71. (711) We see from (7. the differential of H in (7 3) is dH = dE . The other thermodynamic potentials are: F = E .T S . .p N . S2 = E . f). For instance. The differential of F in (7 5) is dF = .10) that H can be regarded as a function of S and f:H = H ( S . where V = A y . G = E +pV TS. (79) The differentials of these thermodynamic potentials have special forms.
e..=($) 7. .g.p dN dH=TdS+ Vdp+pdN dF=SdTpdV+pdN dG=SdT Vdp+pdN E E+pV ETS E H F G .p d V 4. The Various Rules of Calculation T . i.T S +pV a ETSpN d i l =SdT pdV N d p The partial derivatives can be directly read from the above table. A.e. EL (7 . and H ( s ) is the energy of the s state.12) These differentials will be very useful when we do calculations. This . (713) where s denotes the various states. The calculation of entropy can be reduced to S(E) = lnr(E) . Rule for constant energy and for constant temperature We assume that the motion is confined to a region in phase space with a definite energy. the energy at each point of the region is E .122 STATISTICAL MECHANICS Table 71 The differentials of the thermodynamic potentials Definition E Differential dE = T dS .2.
13) we collect all those states with energy E and assign to them the same statistical weight. of course.H ) in the equation restricts the energy and we only collect those states with energy E .P H ( s ) .16) has a sharp peak. proportional to e . As E’ is a large number. The factor 6 ( E . all the states are collected whatever their energies but are assigned different statistical weights. In practice this rule is not very convenient and we have to modify it.H ) in (7 . In this section we discuss the various ways of modifying it. and may seem to have no specific meaning. the determination of the region of motion in (7 . and count their total number. Notice that in (7. Now we replace the factor 6 ( E .P(E‘E)] = 0 . not the same as S ( E .@ decreases rapidly with E’.13) by (714) and define a new quantity (715) where 0> 0. The function A ( E . Hence and are not the same function.If) is. its maximum is determined from [S(E’) aE’ a . they are related by r  r(E) = 1 SdE‘ S A(EE’) &(E’H(s)) = 1 d E ’ exp { S ( E ’ ) . e . At first sight.E ) } (7. This statistical step is called the rule of constant energy. its value is to be determined later.RULES O F CALCULATION 123 summation is a statistical process. In (715). (717) .P ( E ’ . Thus the integrand of (7 .15) is somewhat ambiguous.16) Assuming that a S ( E ’ ) / a E ’ > 0 we see that r ( E ’ ) increases rapidly with E ’ .H ) .
if N is very large.19) (7 . we have then S(E) = l n r ( E ) . E and E’ are all proportional to N . a N . K is too.H ) cutsoff the high energy states.) Expanding S ( E ’ ) and retaining terms up to second order only.20) Because S. As S ( E ’ ) and E’ are large numbers we can use the law of large numbers of the last chapter to calculate this integral. (723) Although A ( E . but the number of low energy states is too few. Therefore. We put o = .16) occurs at E’ = E . (721) (722) lni.124 STATISTICAL MECHANICS i. Thus if N is very large. The integration with respect to E’ gives a 0.H ) are not the same. (7. Hence only the states of energy around E count in F(E).T=  1 aS(E) aE * (718) The maximum of the integrand of (7.H ) and 6 ( E . S(E) 2 1 l n r ( E ) .1n(27rK) 2 1nK . In (723) the relation . (68) to (613).e. they give completely identical results for S because A ( E .  InN . (See Eqs. It does not suppress the low energy states (instead it magnifies their statistical weight). 1nNmay be neglected. aS(E’)/aE’ = 0. we have E” E ElE.
H ) .RULES OF CALCULATION 125 between T and E is through (718). it is better to eliminate E in favour of T .H ) we can also change it to B(E . we do not count the states with energy larger than E .H ( s ) ) = SdE' 0 ( E . This way of calculating F and then S is logically called the rule for constant temperature.e. In calculations it is more convenient than the rule for constant energy. It gives the same result as the above rule for constant energy. B.E ' ) I'(E') S where X = E . The calculation in (724) takes T as a constant.E' and K are defined by (720). Rule for restricted energy As we can change 6 ( E . (724) S Then from (723) we obtain F = E . i.T S . (725) This of course.H ( s ) ) S . this is not quite convenient. F is the thermodynamic potential at constant temperature. Hence.H ) in (713) to A ( E . S ( E ) = l n x 8(E . Because the temperature T is easier to measure and energy is not. (727) . The proof is very simple : x O ( E . Let F = TlnZ . and S can be obtained by direct differentiation (see (712)). The function Z is called the partition function at constant temperature. is one of the thermodynamic potential discussed in the previous section. if N is very large.
V ) . The equations of calculation will be changed. and then we get the rule for constant temperature and pressure.H ( s ) ) 6(VS P(S)) . Hence we obtain the rule for constant energy and pressure. We have to add the word “for constant volume” to the different rules above. (728) S(E.TS + p V .g. The steps of the calculation are similar t o the above rules and the reader may derive them as an exercise. The definition of G is G = . C. The generalisation of (725) is G = E . (7 .T l n Z . We can also write 6 ( E . For example. V ) = lnF(E.H ) as A ( E . then P(s) becomes a function of the state s. e.V ) = I s ( E .13) becomes r(E. Here we only mention the thermodynamic potential G at constant temperature and pressure. If the phase space includes states with variable volume. the rule for constant volume and energy. just like H ( s ) .H ) at the same time. Similarly we can replace 6 ( V  p) by (729) Here p is just the pressure of the system and T is the temperature. Rule for constant pressure and temperature In the rules of calculation given above we had assumed that the system had a fixed volume. . (730) This is a generalisation of (724) and 2 is the partition function at constant temperature and pressure.126 STATISTICAL MECHANICS We call this the rule for restricted energy and it is quite useful in some cases as well as in general discussions.
The rule for open systems In the above rules. then the number of molecules $(s) becomes a function of the state.T l n C . . We may now change 6 ( N .34) where C. is the partition function at constant temperature for an open system. the thermodynamic potential at constant temperature for an open system: S2 = . (7 .N) . . Now if states with different numbers of molecules are included in the phase space.N) = lnF(E. The entropy can be calculated by the following equation S(E. We can similarly write down a collection of rules. We mention here the commonly used rules for constant temperature for an open system.RULES OF CALCULATION 127 (731) D. The above rules use 6 ( N . the total number of molecules in each state is assumed to be constant.fi(s)) to collect states with definite number N . The rule can directly use p . T as fixed values to calculate S2.i ? ( s ) ) to (732) where p is the chemical potential of the molecules. (733) This is a generalisation of (723).
so that A^(s) in each region is fixed. 2 (735) These are the generalisations of (724) and (725). In many calculations. (If it were not the maximum.128 STATISTICAL MECHANICS The entropy and particle number may be obtained by differentiating R: i = E . Using the familiar method of peak integration and we obtain (737) . E. We may divide phase space into regions according to its value. “Open” implies that the molecules are not strictly confined in the system. Hence the volume of this region of motion is a maximum. Let (736) This is the volume of a particular region of phase space. The above rules can be regarded as the result of finding the maximum.p N . this particular property is very useful. The rule of maximum value All the states allowed by the invariant quantities are included in the region of motion.TS . Let d(s) be a variable of large magnitude.) Hence the basic assumption is that the entropy is a maximum. then we could put more states into the region of motion. but the entropy of a closed system can still be calculated by this rule. We shall use the rule for constant temperature to illustrate this.
. . For example. Why is the value of &(s) S is maximum and ( H ) = E ? The definition of ( H ) is such that (738) S This ( H ) is the average value of the energy in this region. . This rule of maximum value has its own utility. .. (See the discussion of Lagrange multipliers in Chapter 6.. . i. . a2. The other rules can be obtained by a similar procedure. [S a ai a . so that the calculation of 2 by the method of constant temperature is too difficult. Then we use .) The result of the differentiation is (after rearrangement): a(&)  (A"))  P(H(s) .RULES OF CALCULATION 129 We may now ask the following question. . a . when we use the mean field method we employ this rule to derive it.@(HI)] = 0 . For S to be maximum. In the following.( H ) ) = 0 . . we can use a simpler H ' ( S ) for H ( s ) where H ' ( s ) has several adjustable parameters a l . i = 1.e. To choose d ( s ) means to choose the d ( s ) value of each s .39) where p is the Lagrange multiplier. Because H is a function of large order H ( s ) is nearly equal to ( H ) in all points of the region. m .2 . the best choice is aA^(s)= PH(s). if H ( s ) is very complicated. (740) to determine the values of ai so that S is a maximum and at the same time keeping ( H ' ) constant. we must solve (7 . . This answer is just the rule of constant temperature. Notice that if we let (741) . not for the derivation of the above rules but to derive approximate calculational schemes.
The various rules of calculation in Sec. Equation (7 42). 2.130 STATISTICAL MECHANICS then the S in (737) can be written as s =  C p ( s ) Inp(s) .1 we discussed some of them. We only wrote the basic assumption in different forms. he may do so but he should be aware that the various concepts in .E ) / T 6 ( V . 3 .? ) o r e . The various thermodynamic potentials are the special combination of the various generalised coordinates. 6. Sec.( V .e:S.H ) . then the various rules of calculation are contradictory and cannot be used. Of course. 7. N . The results are the same. The discussion here points out that it is the same as the rule for constant energy or that for restricted energy or all the above rules. although commonly employed. We must remember that the A^(s) above must be a variable of large number. Review the method of Lagrange multipliers in Chapter 6. there are no additional hypotheses and we do not need any other new concepts. Problems 1. This is of course just the rule of maximum value discussed here. The rules of this chapter are the result of the basic assumption that S = l n r . i. In Sec.2 are linked to the orders of magnitude of the various quantities (i. If the reader prefers to use the method of ensembles. looks mysterious as a basic equation. or . The determination of this region of motion is ingependent of the way we use 6 ( E . It is just a special way of expressing the basic assumption. All the above rules have the same consequences but with different details. The rule of constant temperature is the same as the canonical ensemble.1. In the next chapter we shall use some examples to illustrate the application of these rules. They all calculate the number of states in the region of motion.1 with that of the method of Lagrange multipliers. Prove that if the order of magnitude is not right. Compare the discussion of Sec. The derivation of these rules does not require the introduction of any new concept. so that the results will be meaningful. Besides this assumption. The rule of constant temperature for an open system is the grand canonical ensemble. these combinations have special implications in thermodynamics. Some people prefer to start from (742) and then use the condition that S is a maximum to determine p ( s ) . All the rules of calculation stem from this basic assumption. E . The rules of calculation in this chapter have the same result as the method of ensembles in the other books.e. or e .( H . V are O ( N ) and r is O ( e N ) ). 7. of order N . 7. S (7 42) This is the common way of expressing the entropy.V ) p / T etc.
= + 1 . If the pressure in the gas is p and the temperature is T . . Calculate the equation of state. what will the final equilibrium temperature of the wire be? Fig. 8. Use the rule of constant temperature to solve the onedimensional king model. with binding energy B . 1 . The volume of the container is L . I xi . calculate the total number of adsorbed molecules.e.x i + l ) = f . . He must also note that the time scale does matter. An elastic metallic wire at a temperature T is bent by a force f . each molecule is rigid. Suppose the interaction energy for the above problem is U(Xi . the interaction is limited to nearest neighbours. the relation between pressure. Apply all the rules of calculation to the ideal gas. 4." ~ y / ~ .x i + l 1 > where xi is the position of the ith particle. The temperature of water is T and gravity acts downwards. Answer: p ( y ) = ~ ( O ) e . 9. the equation of state (i.3 A wire bent by a force f . Calculate the entropy. Calculate the distribution of molecules as a function of depth. 73). and the heat capacity of a system composed of N molecules is in onedimensional space. 6 . If f is suddenly released. A number of particles of radius a and mass rn are dispersed in water. (Fig. the entropy and the heat capacity. The walls of a gas container have NA adsorbent states and each state can at most adsorb one gas molecule. = o . 5 .RULES O F CALCULATION 131 ensemble theory (such as infinitely many systems) are rather abstract and not realistic. I X i Xi+l I < a a . s. The molecules are nonpenetrable (hence they are ordered). Which rule of calculation is the most convenient? 10. volume and temperature). 7 . with size a .
e. which has only two independent states. (a) Calculate the entropy and energy as a function of temperature.Z4 . the ordering of physical variables is important. = k 4 . 13. N ions of positive charge 4 and N with negative charge . (b) Find the heat capacity when T << 1/(2I) and T>> 1 / ( 2 I ) .1 H =J 1 Si 9 Si+l . we may choose any axis. Write down the equation of state for this system and prove that when T < q 2 / 2 the partition function 2 at constant temperature does not exist.4 are constrained to move in a twodimensional square of side L . . . However. The interaction energy is {( Iri . .J In Iri . 1 14. As a matter of fact. . Readers should review the physics of spin 7 systems.4 .model.132 STATISTICAL MECHANICS 1 1. This model will reappear in Chapter 19. there are also other (not independent) states obtained by linear combination. s 2 . 4J. the number of states with energy equal to the zeropoint energy). and that the moment of inertia of each disc is I. Suppose the molecules in a threedimensional space are flat discs. (a) Calculate the zeropoint energy and the degeneracy of the ground state (i. However. .r/ I ) = . the basic assumption can be applied t o quantummechanical problems. in quantum applications. The quantum nature is best exhibited in terms of the spin $. Let s l . These linear combinations may represent the spin pointing in different directions. sN be N spin 2 operators with the Hamiltonian operator N. When we count a state.rI. The discussion in this chapter did not make use of quantum mechanics except for the assumption of the existence of (discrete) states as a basis for statistical counting. since dynamical variables are represented by operators. because all that the basic assumption does is to perform statistical counting on the states. and consider the two states in which the spin is pointing 1 parallel or antiparallel to this axis. (b) Calculate the energy of the lowest excitation. without the introduction of new concepts. 1 where q i . 12.
First we use the rule for constant energy to analyse an ideal gas model. Then we use the rule for constant temperature to analyse the same problem. ni = 0. Now there are . besides moving and colliding. Let us now discuss a gas of such molecules. and show that a system under tension must eventually break up. If the molecule is in the ground state. the number of s = 1 and s = 0 states is constant. and the number of s = 1 and s = 0 states will change with temperature and the heat capacity will show a different behavior. Under ordinary observation times and without the help of catalysts. If there is a catalyst. The nuclei of the hydrogen molecule can have total spin s = 1 or 0. Time Scale and the Region of Motion Suppose that a certain molecule has two internal states . The last example is a onedimensional model which will illustrate the method for constant temperature and pressure. The first step is to determine the region of motion in the phase space for this system. can change their internal state due to collision. unaffected by temperature changes. The molecules of this gas have two internal states. The molecules. Let n i be the internal coordinate of the ith molecule. s changes rapidly.1. The rule for constant temperature is again applied to study the heat capacity of H2 gas mainly to discuss the rotational degrees of freedom of the hydrogen molecules. 8.the ground state of energy 0 and an excited state of energy E .Chapter 8 ILLUSTRATIVE EXAMPLES The aim of this chapter is to use simple examples to illustrate the rules of calculation derived in Chapter 7 and point out the importance of the observation period. otherwise ni = 1.
d3pN 6 ( E . . besides the positions and momenta we must also consider changes of the ni. It means the approximate time required for every ni to repeat itself once.e..H ) N. Now besides integrating with respect to position and momentum. Each n is unchanged. . + .one is the ground state molecules and the other the excited state molecules.nN also: r(E) = n = O n =O $ c 1 . be the volume of the region of motion in the case n .) ( 2 ) The internal states change very slowly. Let ro(Eo) = . .. We can use any one of the rules discussed in the last chapter. Now that the region of motion is determined the crucial step is accomplished. . or the observation period is very long. i. . and the constraint that the positions of the molecules must be within the container. we have to sum over n . . (8 . the two are independent. + n N . much longer than the period of change. This is the same as (516) and (517) in Chapter 5. . The next step is to calculate the entropy and the other properties. almost in every collision. = n2 = . = n. Later we shall discuss this situation.3) . The Hamiltonian is The region of motion can be determined by E = H . = nN = 0. (Here. and are nearly unchanged during the observation period. . . it is not easy for the collision to cause changes of state and the observation period is not long. .e. nN=0 VN N!h3N J d3p1 . Let us use the rule for constant energy. In situation (1 ) we must consider the change of the internal states. but H is slightly more complicated here. this gas can be regarded as the mixture of two kinds of molecules . “period” does not imply a regular periodic change. In situation ( 2 ) .134 STATISTICAL MECHANICS two possibilities: (1) The internal states are changing very rapidly. i. + n . .
This is just The number N1 above is regarded as a large number and (85) is the result of simplifying (84) by the law of large numbers.e N 1 (87) is is the total energy of the molecules.N .5) N N The combinatorial factor in (84) expresses the number of ways of dividing N molecules into two parts: N1 and N . ) l n NNI (8 . The integral of (84) can be done by the method of peak integration of Chapter 6 . We first find the maximum of S: i. (8 .7 ) where Eo .e N 1 ) + N 1 h N1 + ( N .N1.ILLUSTRATIVE EXAMPLES 135 where N1 is the number of excited molecules.E . The above equation can be written as r(E)=  (84) S(E.e. The solution of .N) = S o ( E .
E / T ) . the heat capacity is Cv=T 3T as (810) (see Fig.e. The various equilibrium quantities can be obtained by differentiation. This is the result of situation (l).1.9) where So is the entropy due to the translational motion of the molecules and the rest is that due to internal motion. the rapid change of the internal states of the molecules causes N1 to become Nl(T). This rapid change extends the region of motion in phase space. The last term of (810) no longer appears. 8. and there is no hump like that in Fig.the entropy is a maximum.136 STATISTICAL MECHANICS 1_ K ___ 2E2 3NT2 + N N1 (NN1) The entropy of the system is s = S(E. 8. when the internal states of the molecules are changing very rapidly.N is the heat capacity due to the translation degrees of freedom. Note that the term Cvo = 3 . This gas is then a mixture composed of two different molecules with constant composition. (8 . . e. i.1. From the large scale viewpoint. 'In situation (2) each n i is unchanged.) The curve has a hump near T = 4 2 . When N1 = N 1 ( T ) . Hence N1 is unchanged. Its heat capacity is then Cvo = +N.) N E/T = SO(E0) + eelT + 1 + N l n (1 + e .N.g.
1 comes from a i l / a T. 3 E=NT+EN~ 2 . (810). We would like to calculate F=TlnZ I . The total energy is . If N 1 is a constant. We now use the rule of constant temperature to solve this problem. If N 1 is not changing.e. i. 8. . then the basic assumption determines N1.N v. Thus the hump in the curve of Fig. The above result for the heat capacity can be analysed with the aid of the equation C = a E / a T . That is to say. ( T ) and (812) We get (810) [see (88) for then 3 C . in situation (2). i.ILLUSTRATIVE EXAMPLES 137 Fig.. (8. 81 Plot of Eq. then the basic assumption can say nothing about the value of N1.11) If N1 is a variable.e. if the internal states of the molecules are changing continually.2 N1 1. in situation (l). then N1 = N .
As nl . . . . For situation ( 2 ) the ni's are unchanged and the gas has N .is an invariant. ! ( N . nN=O eH/T .N l ) ! . NNI (816) ] See (815) for fo. . the various integrations and summations can be done separately. d 3 p N N ! h3" C n =O 1 . The above results apply for situation (1) when each ni changes many times within the observation period. .. . excited molecules and N .+ eN1 . We get The reader may obtain the entropy and the other quantities by differentiation. The thermodynamic potential obtained in this way is N + (NNl)ln . . nN in (813) can be dropped.. (See (240).) The heat capacity calculated from (816) is just ( $ ) N .N l ground state molecules. . The second term on the right is caused by the entropy of mixing. and N ! becomes N .138 STATISTICAL MECHANICS  VN j d 3 p. (813) Because each term in H has only one variable. the summation with respect to n .
The vibrational energy is where n = 0. . and w is the vibrational frequency of about 6 000 K. . . that is to say. the electrons will be mainly in the ground state. and 1421) has a value of about 80 K. B. Motion of the electrons As the electrons are light and the nuclei are heavy. However they are enveloped and bound together by the electron cloud. Let the angular momentum be J . In the analysis. Gas of Hydrogen Molecules Let us now look at the simplest molecule in nature. the change of ni minimizes F. to minimize F. . Vibration of the nuclei The two nuclei in the molecule repel each other because they are both positively charged. The reader should remember that in principle they give the same results. 8. it uses temperature as the scale of energy and this is easier to handle than the energy E . If the temperature is not too high. There is a large energy gap between the ground state of the electron and its first excited state. as far as electrons are concerned the nuclei are essentially stationary. C Rotation of the whole molecule Because the amplitude of vibration is small. composed of two atoms.3. (. 3 . (818) 1. we get the results of (813). .ILLUSTRATIVE EXAMPLES 139 Notice that at constant temperature if we change N .l o 5 K).lOeV . . then its energy is 1 J(J+l) 21 where J = 0. .2. 1 . Obviously the rule of constant temperature is more convenient than the rule of constant energy. . We first review some results from quantum mechanics. the molecule can be regarded as a rigid body with mass concentrated on the two nuclei. 2 . .2. The motion of the molecule as a whole and internally can be divided into several parts: A .
4 HCl hulk h2/2Zk 6 100 85. F.) Hence. Most diatomic molecules.1 4140 15. over a large range. (819) The energy associated with nuclear spin is negligible. not just H z .2 vibrational state is in the ground state ( n = 0). the Table 81 Comparison of the vibrational and rotational energies (unit = K ) H2 NZ 3 340 2. Spin of the electron The two electrons of H2 have spins that cancel each other in the ground state.9 0 2 NO 2690 2. . Translational motion of the whole molecule The kinetic energy is p 2 / 2 m . but the interaction is very E. and these need not be considered. have vibrational energies much higher than their rotational energies. (See Table 81. $ each. We are left with the rotational energy and the nucIear spin energy. In the discussion following we ignore the vibrational energy as well as the translational energy which has been thoroughly discussed in the ideal gas case. The energy is H= 1 J(J+ 21 1) .4 2 230 2. Now we calculate the thermodynamic potential according to the rule for constant temperature fl = Tln J (2J+l)eJ(J+1)" . The nuclear spin The two nuclei of H2 have a spin of value small. while the rotational energy is usually excited.140 STATISTICAL MECHANICS D.
M = . 1 7 (822) while s = 0 is a singlet: (823) where the arrows denote the direction of the nuclear spin. This result above is applicable to the HD molecules ( D means deuterium) but is not suitable for H 2 molecules. Therefore although the spin of the nucleus and the rotation the molecules have no mutual interaction. . Recall that the wavefunction is a product of the rotational part and the spin part: where k is the direction of the relative position of the nuclei. xl. J . and we have A YJM(k) = (1)J YJM(i) . .J . (820) 9 where f i is the thermodynamic potential for rotation. This is because the two nuclei of H z are identical fermions. . If we interchange n the nuclei.J + 1. and s is the total spin of the nuclei. R is changed to R.J . (824) Hence to make cp change sign. s = 1 can only have odd J while s = 0 can only have even J . .T l n 4 1 21T . J and s are not independent and f l and fz should not be calculated separately . which has four states. The state s = 1 is a triplet : m= 1 . two for each nucleus. is unchanged and xoo changes sign. . and fi is that for nuclear spin. This is a result from quantum mechanics.1.ILLUSTRATIVE EXAMPLES 141 fz = . 0. and an exchange of position changes the sign of the wavefunction.
3 0 3. 0 0.5 4.142 STATISTICAL MECHANICS but together: fi + f2 = In z z =3 c odd J (2J+ l)eJ(J+')a (825) From this result we can calculate the entropy and the heat capacity C.0 .0 2. . the nuclear spin of H2 during usual observation times is unchanged.5 1 0 1 5 2. but ( 8 .2 5 ) is only true for situation (2). (826) The curve in Fig. 82 The rotational capacity of H 2 . Why? The reason is very simple. The case of H2 is very similar to that of situation ( 2 ) discussed in the last section.5 . that is to say.2 is the result of (826) and the dots are experimental results. Fig. The solid line is calculated from (825) and ( 8 .2 6 ) and the dots are experimentalresults.=T as aT . Obviously they do not agree. Calculation and experiment do not agree at low temperatures. . CV N = T  a2 a T' (Tlnz) . 8 .
Notice that when Tis very small ( T <G + T ) . N1/ N = . respectively. The region of motion is defined according to the invariant quantities. 82 is similar to that in Fig. 84 is the result of (827) assuming that No / N = +. because the smallest value of J ( J + 1 ) is 2: so . With this in mind. then the values of No and N1 must be determined by the temperature. Notice that the hump of the curve in Fig. The correct calculation must consider N 1 the number of molecules at s = 1 and N o the number of molecules at s = 0 as invariant quantities. The nuclei are surrounded by the electrons and the collisions of the molecules do not affect the nuclear spins. The curve in Fig.ILLUSTRATIVE EXAMPLES 143 However (825) assumes that it is changing continually. the energy of each molecule for s = 0 and s = 1 . 81. leading to the hump in Fig. The interaction between the nuclear spins of H2 molecules is very weak and the interaction with the electrons is also weak. 83. Hence the experimentally measured quantity is a shorttime result and (825) is not applicable (unless we use an effective catalyst to speed up the change of nuclear spins). At the same time the spins of the electrons cancel each other and it is not easy for them to interact with the nuclear spin. The numbersNo andN1 = change with T. If the nuclear spin is changing continually. 82. and (829) This is similar to (826). and agrees with experiment. we find that the entropy is (827) (828) where eo and e l are. N1( T ) will be very small. The calculated results of aeo/aT and ael / a T are shown in Fig.
5 2.0 . 211 Fig.5 4. The three figures above are taken from Wannier (1966).144 STATISTICAL MECHANICS 3 211 Fig.5 3 0 3.50 Eucker Giacomini A Brinkworth Scheel 8. Heuse 0 Partington 8. 8 . 219.5 1. nuclei with spin s = 0 and s = 1.220 and 221. p. 84 The rotational heat capacity of H . Howe 0 A 0 0. C/N 0.2 : (830) .0 1. 83 The curves of the heat capacity of the H. showing agreement with experiment. We can estimate the low temperature part of the curve in Fig.0 2. calculated from (826) and (827).
. Onedimensional models are the simplest and they will now be discussed. We now use the rule for constant temperature and pressure to calculate the thermodynamic potential G: G = . The positions of the molecules are xi. and the maximum is at T = $Z. 1 . it can nevertheless explain some effects o f the interaction. but its other effects are ignored. We fix xo at the origin. Though rare. OneDimensional Model The two examples considered above are models of various ideal gases.ILLUSTRATIVE EXAMPLES 145 aN1 / a T is a curve with a hump.. In fact the discussion of last section was meant as a preparation for this section. dxN [ XN 0 dxNl . Hamiltonian is (831) The interactions of neighbouring molecules include a short distance repulsion and a long distance attraction.10). as in Fig. 8. the interaction has many important effects. 2 = eTFOtT [. . This is similar to (8. . < xN. A dilute gas is very similar to the ideal gas. 2 . Suppose the . 85. N . The volume of thls system is x. The effect of the interaction between the molecules is t o make the variables of motion change continually. Though such a model may be very simple. the simple and soluble models are most valuable. But in the most substances. jxzd x l 0 (832) . . i = 0. Once the interaction is included.3. the mathematics becomes extremely complex.T I ~ Z. In the statistical calculation the interaction energy had not been included. and xo < x1 < x2 . We begin by considering a collection of molecules on a straight line. .
(834) These are the distances between neighbouring molecules and xN is their sum: xN = y 1 + y z + . eFoIT c . . The integral for 2 becomes very simple: 2 = p eFoIT (835) . + yN.1d3p1 . d3 p~ exp N i=l Pi" 2mT hN Fo =NTln(. 8 . . . Now change the variables of integration to Y1 = x 1 . .5 Intermolecular potential showing short distance repulsion and long distance attraction.146 STATISTICAL MECHANICS Fig. 1 .) 1 2 21rmT . (833) Here xN is the ?(s) of (730).
)2nm Tn j . The entropy is also easily calculated: (840) .<< and T. (838) The volume can be obtained by differentiation: v = ( z ) T = T +Nu0 i. p( > v . is the shortest distance between two molecules. (837) then G=NTln 2 ( ) . If the temperature is high.36) 5 is a function of p ua .e. T 1 . i.aoN where a.ILLUSTRATIVE EXAMPLES 147 JO G=N 1 2 T I n ( . (839) Hence at high temperatures this system is like an ideal gas with effective volume V .aoN) = NT .e.N T l n ( F ) + 2nhmT NT paoN . NTI (8 .
The minimum of U( y ) + p y occurs at aU/ay = . hence these results describe a crystal. y  21 a (842) Therefore. The statement y = a means that distance between neighbouring molecules is a . The . (843) Substituting this result into (836) and making some rearrangement. Suppose p is not too large. u a . The main contribution of f comes from the neighbourhood of the minimum of V( y ) + p y .e. then the minimum occurs near a . i.148 STATISTICAL MECHANICS Now consider the situation for low temperatures.2a 6 (844) (845) These are the low temperature results. Expanding U( y ) and retaining terms to third order term we obtain U(y) N u. we get G = NTln 1 2 (2nLT) fNTln(?)+ N T 7 a’P 2a +pa . U”(Y) = (Y + a’ p a . i. 1 (841) This is the approximation of U( y ) near a .p .e.>> 1 T .. + p ( y .a ) 2 1 6’Y’(Yd3 .
then the relation of I/' with temperature is lost.e. Generally speaking. a system with shortrange interaction (i. The situation for p > 0 is simple. there is no phase transition here. if the tension Fig. This is so no matter how strong the shortrange interaction is. When the temperature is decreasing. (Of course. The situation p < 0 is a valley. each molecule interacts with its few neighbouring molecules) does not exhibit phase transitions. This is because U ( y ) must vanish at very large y (i. it does not suddenly change from the gas to the solid or liquid state at a certain temperature. this becomes the potential energy of a simple harmonic motion. The lattice vibration discussed in Chapter 6 uses this approximate potential. differentiable function of p . in one dimension. If we retain only the second order term of the expansion of U ( y ) in (845). However there is the following important result. Because (835) is a continuous. the interaction is shortranged). If P G O 7 (846) i. 86 . This will be a topic for later discussion.e. and once the barrier is overcome. Although this model exhibits some properties of the crystal and the gas. then there is a problem in the integral of (835). 86(a) tells us that y should remain in the neighbouring of the minimum. The expansion of a solid with temperature is associated with the anharmonic terms.e.ILLUSTRATIVE EXAMPLES 149 high temperature results (838) and (839) describe a gas. Fig.V plane have continuous derivatives everywhere. so the convergence of the integral depends on the factor ewYplT. From (845) it can be seen that if we use the simple harmonic potential and ignore a'. there is no pressure or a tension. it is an endless downward slope. the isothermus in the p .
assuming that the vapor is a monoatomic ideal gas. Even for p < 0. The above crystal is cooled suddenly from a high temperature T to a low temperature T ’ .. You should also use the results for lattice vibration given in Chapter 6. even the valley will disappear).150 STATISTICAL MECHANICS is large enough. At T ’ . 87. the atom can no longer leave the lattice site for an empty space. 6.2. From this basic process derive the result of the above problem..a(n ++) w . a solid under tension is an equilibrium state. (848) . When the amplitude of vibration of a diatomic molecule is very large. this result is still applicable. this forms an imperfect crystal with n defects. in (844) and (845) we did it by expansion. The conclusion of this example is appropriate for any system under tension: such a system will eventually break up. If the energy required to kick a molecule out of a crystal is E . in calculating the region of motion we should exclude the states of the broken system. prove that (847) 2.. An atom absorbing a phonon of energy w will leave the lattice thereby creating a vacant site. If the time is short. calculate the vapor pressure of this crystal. The potential curve is like that in Fig. Thus. If we move n atoms from the original lattice sites to the space in between the atoms. This result can be explained as follows. n=0. Calculate the entropy and heat capacity due to the defects. provided we ignore the problem in the integral of (835). then these molecules are linked as a chain. The vibrational levels can be approximated by the following equation 2 En = ( n + t>w .1. If each yi stays at the bottom of the valley. How do we exclude them? This question still has no completely satisfactory answer. e. 3. Is the above result still applicable? 5 . N atoms are arranged regularly to form a perfect crystal. Of course. . it becomes an anharmonic vibration. If w is the energy to move an atom from a lattice site to the empty space.g. If one or more yi jump out of the valley. But in many cases this is not hard to manage. The above problem can be solved by the principle of detailed balance. then the chain is broken and the tension vanishes. the problem of tension is a question of the length of time. Problems 1. with very low pressure. The number of such possible spaces N ’ is the same order as the number of atoms N . 4.
Here w = w o / w l . then the thermodynamic potential of charged particles will not be affected by a magnetic field. Prove that if we neglect quantum effects. oxygen and nitric oxide). Calculate the heat capacity and indicate the effect due to the anharmonic term up to the first order in a . 8 . I . Some molecules have two nearby energy levels for the electronic motion (e. The high energy level is E and has w1 single particle states. with the energy difference being far less than T .= ~ n w + l n ( e/T + 2 ) e RT €IT2 (851) . 7. ( b ) Obtain approximations for C in the limits of high and low temperatures and prove that the maximum of C occurs at . . (850) The formula assumes that the low energy level is zero and that there are wo single particle states. Fig. where a represents the degree of anharmonicity. the higher level can have a considerable population and thus influence the thermodynamic properties of the substance. ( a ) Prove that the heat capacity due to the motion of the electron is C = wNe'tT [ eE €w T: .g. Hint: the Hamiltonian is 8.ILLUSTRATIVE EXAMPLES w 4 151 5 6 3 I I I I 2 HCI simple harmonic  > 6 0 1 2 3 4 5 Distance between the atomic nuclei . As a result.7 Vibrational potential of a diatomic molecule.
A cylindrical container of radius R and length L is rotating round its axis with angular velocity a. 10. ( b ) Calculate the pressure at the walls. ( b ) Find ( x N ) . Answer the questions of problem 9 for the cylindrical container of problem 10. Gravity acts downwards.< X N . Notice that in the frame rotating with the container this gas is in equilibrium with fixed boundary. . (852) where xo = 0 and a pressure p is exerted on t h e N . assume that m u 2 a 2 >> 1. . w 1 = 3). The mass of each molecule is m. A rectangular box of height h and base area A contains N molecules forming a gas at temperature T . what is the form of its heat capacity? ( d ) Take E / T in the range 1. The energy difference of the two levels in oxygen in 11 300 K . with E = 174 K. and w i= 1. . ( e ) Notice oxide has two energy levels which are very close.t h molecule. Compare the result with that of (c). . At high temperatures w o = 3 and w 1 = 2 . (853) ( c ) Calculate the coefficient of expansion of this system. In the above calculation.152 STATISTICAL MECHANICS ( c ) If each molecule has three equally spaced energy levels. ignore (Y exp { rnu2a2/2T} . ( a ) Calculate the entropy and heat capacity as a functions of temperature for this system. ( a ) Calculate the gas pressure at the top and at the bottom.0 and calculate the maximum of the heat capacity of the doublet energy molecules ( w o = 1. A onedimensional system with N molecules is described by the Hamiltonian XO<Xi<X2<.e. i. Estimate the temperature and the heat capacity when the heat capacity is maximum. There is a gas in the container of temperature T and the mass of each molecule is rn. . but higher order terms of &may be neglected. 11. Calculate the Hamiltonian in this frame. Note that acannot be completely ignored.0 4. 9.  12.
Following them. then the calculation of the various thermodynamic potentials does not pose any problem. under given conditions the equilibrium property is fixed.e. we can use different statistical weights to perform the calculation and obtain the same result. the number of molecules N is macroscopically large as pointed out in Chapter 7. i. it is nevertheless a very large number like 1OZo. One of the appendices of Huang (1963) gives quite a detailed introduction. The rules of calculation listed in Chapter 7 agree perfectly well with thermodynamics. many have aYangLee (1952). If the number of molecules N is finite. However. The examples are all oversimplified models. seep. . In this chapter we pay some attention to mathematical rigor. they do not reveal the difficulties that would be encountered in more general cases. 458.Chapter 9 YANGLEE THEOREM It can be seen from the previous analyses and examples that the application of the basic assumption to thermodynamics gives satisfactory results. Because a body is macroscopic. Although these examples provide a measure of confidence and understanding for the rules of calculation and the basic assumption. our analysis is not rigorous. that is to say. The pioneering work in this topic is the YangLee theorem of thirty years ago a The theorem was proposed for phase transitions. The rigorous mathematical analysis of this limit is a branch of statistical mechanics. the socalled problem of the macroscopic limit. i.e. In the process of analysis we have not considered the mathematical meaning of the various equations. hence the problem of the N + m limit becomes a very important problem for the application of the basic assumption in thermodynamics. Although N is not infinite. chosen for being readily soluble.
i. we want to point out the reason for the existence of this limit. From the proof of this theorem we can see that the most important feature of a macroscopic body in equilibrium is that it can be divided into many parts. However the application of the YangLee theorem is quite universal. u(r) = u(r) > m. (Phase transitions will be discussed in later chapters. E is finite. Care should be exercised in this respect.1. the distance between the molecules cannot be smaller than a . i. (9 .e.1) That is to say.e. The necessary properties of the potential u ( r ) are (a) (b) (c) u(r) = 0 . if r > b . with each part being in the equilibrium state.e. and the relation of the macroscopic limit with the interaction of the molecules and the statistical procedure.) However.e. 9.+ 00.1. if r < a . Rather we want to have some further understanding of the basic assumption and to remind the reader that rigorous mathematics does not guarantee correct physical meaning. b is finite. If all the conditions in (9. because the model problems of phase transitions are not easily solvable and less than rigorous analysis is not reliable. Thermodynamic Limit) We first set up a model. 9.E . . Here we do not emphasise a rigorous proof. Macroscopic Limit (i. Hence the number of molecules accommodated in this volume V is finite: The attraction is finite. the detailed form of u ( r ) is unimportant. Let the volume of the body be V and the interaction potential between the molecules be as shown in Fig.1) are satisfied. The aim of this chapter is not to discuss phase transitions. . The interaction is zero for molecules far apart.154 STATISTICAL MECHANICS applied rigorous mathematical analyses to describe phase transitions. In this chapter we shall discuss one of the YangLee theorems. the existence of the thermodynamic potential !2/V in the limit V . i.
9 . . The definition of the thermodynamic potential 52 is P = i2 v' + u. . Using the rule for constant temperature we analyse the system.1 A mode interaction potential between molecu es. . .YANGLEETHEOREM 155 & I H = N C . (9 .Pi' i=l Fig. d 3 pN . d3rN d 3 p 1 .4) We first integrate out all the pi and get N=0 . 2m d 3 r 1 .
then the result will be the same if we calculate a/V in each part separately. n/V will not change with I/. If we impose the restriction that there is no molecule in the boundary region. The existence of the limit (98) means that if V is sufficiently large.. the integral becomes less because the integrand is positive definite. We divide the body into n parts. The integration of the different parts can be performed separately and we get (99). i.156 STATISTICAL MECHANICS c d 3 r N =  3 d r . (Fig. The proof starts from this viewpoint.e. the surface does not curve in and out too much. monotonically increasing function of z ..d3rN . d 3r2. Hence if we . It is also a continuous. The first YangLee theorem is: 2 If (9.1) is satisfied z > 0.7) [. If we divide V into n parts and each part is very large. 5 becomes because the effective distance of the interaction between the molecules is b . 92). Let the volume of the boundary region be V ’ . i and Z all are functions of z . then the limit exists and is independent of the shape of the container. If we eliminate the boundary region. From (96) we can see that if we diminish the region of integration. there is no interaction between the different parts. and set a “boundary region” of width b between each part. The above assumes that the surface of the system is less than V2’3 times a fixed constant. (9 . then the remaining volume is n Vo .
. and simplify the interaction so as to get an upper limit for t . to be inside the boundary region and let the others be in the n parts of the volume Vo. We rewrite (9 6) as (911) i. (9. we restrict rl . . we make [ smaller. i.e. eliminate the boundary region. Each molecule interacts at most with ( b / ~other molecules (conditions (9.e. . r z . The interaction energy ) ~ of each pair of molecules is not less than E (condition (9. . value of the U becomes smaller. [t(~.la) and (9. i.lc) ).VO>I” E(Z3 V) . .r N .lb)). then ZN can only decrease and t decreases too.YANGLEETHEOREM 157 Fig.10) Thus we conclude that if we cut up a system into n parts and ignore the boundary regions. 9 2 Division of a body into parts with boundary regions between them. (9. If the interaction energy inside the boundary region is replaced by .12) . We put back the boundary region.e.~ ( b / a ) ~ N ’ .
14) becomes [ t ( z . Now V ' is the volume of the boundary region and is approximately equal to the boundary surface times the width b . Now the last summation is just [ ( ( z .13) into ( 9 . V0)I" G t ( z .9 .11) and integrating. . we get Z E .N' as N " and taken away the constraint N ' < N .N . We have included the lower limit in (9..14) In the above we have written N . is the statistical weight of N . (9. Equation (9. (916) where c is a constant.e. Substituting (9./3 .158 STATISTICAL MECHANICS The first summation on the right is the interaction energy ignoring the boundary region. V ) G e""' [ t ( z . i. Let (917) (918) . V0)l .10). we get t ( Z 9 V ) G N'= 0 c e(:)" N' z " e N' ! ("1" N"= 0 ZN" zz.12) into (9. V O ) ] and the first summation can also be done.N ' molecules after we drop the boundary region. v' = c V. Substituting (9.
YANGLEETHEOREM
159
That is to say, if V , is sufficiently large, the difference between @ ( z ,V o )and V' In C;(z,V ) is very small. The thermodynamic potential per unit volume can be calculated from a part of the body; it does not matter which part. The error in the calculation is inversely proportional to the size of the portion. This has clearly proved the thermodynamic limit in this model. In the above analysis, Vo need not be a cube; we only require that (916) be satisfied. Nor is it necessary that each part have the same volume. From the point of view of physics, (9 18) is the result we seek. If we want complete mathematical rigor, we must supplement our discussion with some mathematical explanations. We shall not get into this here. Because (919) we have (9  20)
Each term in (920) is positive, so p must increase with q whether V is large or small. As the limit V + m exists, p ( z ) must increase with z in this limit. In addition N  <  Nm

1 
V
V
a3
(921)
So a p / a z is finite, i.e. p is continuous. This completes the proof of the first YangLee theorem. The second YangLee theorem is very useful in the discussion of phase transitions. The reader is referred to the literature, as we will not discuss it in this book. 9.2. Generalisation of the Theorem
Now let us return to examine the three conditions in (9  I ) , i.e. (a) a finite effective interaction distance, (b) a short distance repulsion and (c) an overall attraction not too large in magnitude. If (a) is not satisfied then there is long range interaction and there is no way to determine the boundary region. If (b) and (c) are not satisfied there will be instability, i.e., all the particles will be concentrated in a small region. In the problem of the macroscopic limit, we
160 STATISTICAL MECHANICS
ask what we shall obtain if we relax the conditions above. We shall mention several results. Condition (a) can be slightly relaxed. If u(r) vanishes sufficiently rapidly, i.e. when r > b ,
the theorem still holds. Here d is the spatial dimension of the system and is usually d = 3 . Condition (b) can also be slightly relaxed. When r < a, if we have
u(r) > B r  ”
,
A’ > d ,
(923)
then the theorem is still intact. Condition (c) cannot be relaxed. Now the most important interaction does not fit into the above scheme. Matter is made up of the positively charged nuclei and the negatively charged electrons. There is also the dipole interaction. These electromagnetic interactions are longranged and the interaction energy of positive and negative charges when r + 0 is oo. None of the three conditions are met. Hence the YangLee theorem cannot be used directly to discuss the stability of matter and its macroscopic limit. In the following we shall mention the effects of the interaction between the charges, and the quantum nature of matter. From our experience with electromagnetism we know that uncancelled charges always go to the surface. Here the energy and the other quantities will be dependent on the shape of the system. Hence, only those systems with no net charges can have a thermodynamic limit. We must also consider the quantum effects, otherwise the separation between the positive and negative charges can be infinitesimally small. All the particles will be concentrated at a point. Quantum mechanics says that a particle restricted to a small region will have a large momentum. This theorem ensures that the particles do not collapse to a point. But, if the particles are bosons, we can well imagine that although the particles do not collapse to a point, they still prefer to cluster and the thermodynamic limit will encounter difficulties. It has been proven that if there were n o exclusion principle, the zeropoint energy of each particle will be lower than N2’5 , i.e. as N + 00 the energy of each particle will tend to oo. If all the negativelycharged particles (or positivelycharged particles, or both) are fermions (the types of particles are finite, otherwise the exclusion principle is still useless), then the zeropoint energy of each particle is finite. These facts are quite difficult to prove. Much progress has been achieved
YANGLEE THEOREM
161
in the proof of the macroscopic limit.b The validity of the basic assumption is quite well established, since the results of various rigorous analyses do not lead to contradiction. Lastly we must emphasise the need for caution in such rigorous analysis, and we should not be too easily taken in by its conclusion. We must first understand the assumption in the analysis and the meaning of the various limits. For example, in the YangLee theorem we can arrive at the conclusion that under a fixed temperature, and chemical potential there is one and only one equilibrium state. This excludes the existence of metastable states. In reality there are various types of metastable crystal structures and also supersaturated vapour configurations. This is because this theorem starts from the basic assumption without considering the time scale. This assumption is completely rigorous only under an infinitely long period of observation. A rigorous proof means that there is no approximation between the assumption and the conclusion. In an effort to simplify matters the assumption usually ignores some facts. The basic assumption of statistical mechanics is an oversimplified assumption. Its applications are extensive but it is not the absolute truth. Besides the assumption, some limits such as the volume V  t m must be taken with care. A real body is finite. Hence the implication of the limit must not contradict reality. A common situation is
v+ 1 n V
lim
 0 
1
(9  24)
The meaning of this 0 is far from clear. Even for the largest system, the logarithm of its volume cannot be too large. Although (924) is mathematically impeccable, physically (9 24) is not rigorous enough and is questionable. Hence to understand a “rigorous conclusion” we must be careful in every detail of the proof. In general the true meaning of the conclusion usually becomes apparent from an examination of the steps of the proof.
Problem
1. The proof in Sec. 9.1 shows that once the Hamiltonian is fixed, all the thermodynamic properties are fixed. This is a rigorous proof. The proof pinpoints the obscurity in the basic assumption because it ignores the existence of metastable states; yet we regard a metastable state as being in equilibrium if the observation time is not too long.
See Lebowitz and Lieb (1969), Dyson (1967). For a detailed introduction t o the rigorous Droof. see Griffiths (1972).
162 STATISTICAL MECHANICS
Consider the examples of graphite and diamond and show that in a collective model of carbon atoms, with a rigorous proof we can rule out the existence of diamond.
2. Derive all the results in Sec. 9.1. 3. For a ddimensional model what are the changes in details of the proof? The reader may encounter some d + 00 models in the literature. (Of
course this is just a theoretical discussion. Nevertheless there are occasions when models with d > 3 can be very useful.) When d + , the proof in Sec. 9.1 is not correct, that is to say, the usual d + statistical models are not dependable, since their thermodynamic limit is questionable.
PART I11 PROBABILITY
Probability is a useful but often abused concept. This part repeatedly points out the inseparability of probability from statistics and discusses various applications. Probability is a number obtained from the statistics of the data, and does not stand for a conjecture or the degree of ignorance. Statistics is a procedure for compiling and interpreting the information at hand. It is a tool, not a theory. In statistical mechanics we have t o compile the information about the various details of the molecular motion. Chapter 10 discusses the relation of probability t o statistics. Chapter 11 discusses the phenomenon of independence or chaos. This phenomenon is very important and not easy to understand. The discussion here is a description of the phenomenon and not a quest for its origin. Chapter 12 discusses the central limit theorem, delving further in@ the meaning of indepcndcnce and probability, and explains some results of the basic assumption. This assumption is inseparable from the phenomenon of independence. Chapter 13 presents some calculations of the correlation function and somc cxamples of the concepts of fluctuation and response.
Chapter 10
PROBABILITY AND STATISTICS
We have introduced the basic assumption of statistical mechanics in the last five chapters. Although the examples are oversimplified models, it is not difficult to observe the following: (1) The basic assumption can be used to analyse and explain all the equilibrium phenomena and derive all the thermodynamic results. (2) This assumption and its applications do not require explicitly theorems or concepts in statistics other than those mentioned in the above chapters. (3) Not much information on the trajectory of the molecular motion is required. The role of the molecular motion is to make the basic assumption valid. Once the region of motion is fixed, the molecular motion can be ignored. However, if we want to have a further understanding of the basic assumption, we have to look at the molecular motion more carefully, and for this we need more tools in statistics. Thus, although we have just started to discuss the applications of the basic assumption, we digress for a moment and discuss instead some topics in statistics. The main point in this chapter is the definition of probability, and we emphasise that statistical data are the fundamentals of the definition of probability.
10.1. The Use and Misuse of Probability
Statistics is the science of compiling information and probability is one of the basic concepts in statistics. In its mathematical formulation, probability is rigorously defined and its various properties are fixed by rules. In applications, the concept of probability is often misused, because its definition is not well understood. Now we discuss two different uses. Probability has different meanings, and if we are not careful we shall be led to incorrect conclusions and
PROBABILITY A N D STATISTICS
165
unnecessary trouble. This is especially true for students doing theoretical research. They are more liable to be confused by these two different uses of probability. In the first use, probability is the ratio of two statistical numbers. For example, a die has been thrown 600 times, and the ‘1’ turns up 98 times. Then we say that the probability of obtaining ‘1’ is 98/600, or nearly 1/6. This is a fact or a natural phenomenon. The second use of probability is to measure the degree of ignorance or uncertainty of a person about an event. For example, a person betting for money does not know which number on the die will appear. If you ask him what is the number on the die, he will say, “1 do not know, because each different number has probability 1/6 to appear.” In this use, the probability is not derived from statistical figures, but only indicates the complete ignorance of the result. The above two uses cannot be mixed. Consider this example. Suppose a die is thrown 600 times and we ask how many times ‘1’ will turn up. The reply might be, “As the die has six possibilities and I do not know anything else, hence ‘1’ will appear 100 times.” This result is a guess and it turns out to be correct, but the argument is not logical. Ignorance has been used to predict a natural phenomenon; this is surely impossible. Take another example. If I ask, “How many days in a year does it rain in Hsinchu?” One might reply, “As there are two possibilities, to rain or not to rain, and I am completely ignorant about Hsinchu, therefore it rains six months in a year.” This reply relies on the same reason above, using ignorance to predict a natural phenomenon, and is equally invalid. Instead, the past record of the rainfall can be obtained from the meteorological office. The rainfall in the future can be predicted on the basis of the past records plus other information. We cannot use ignorance to make predictions. The above emphasises the two different uses of probability. There is nothing wrong with the second use. Every experimental result should be accompanied by its error bars. This error does not say that the experiment has gone wrong, it only expresses the degree of accuracy of the experiment. This is an estimate by the experimenter. If he says that the error is 0.176, this expresses that he is not sure of the figure to within 0.1%. Let us take an example to illustrate this. A person uses a mercury barometer to measure the air pressure on the main street. His result reads, “At a certain year, month, day, hour and minute the air pressure is 1.02 ? 0.05 standard atmosphere.” His estimation is that at that time the air pressure should be between 0.97 and 1.07, but where exactly in this range, he does not know. To estimate this error he must consider the accuracy o f the barometer calibration and the accuracy of his eyesight in reading the scale.
166
STATISTICAL MECHANICS
Notice that this error is completely different from the fluctuation of the pressure and the two should not be confused. The pressure changes with time. This is fluctuation. The usual pressure measured by a mercury barometer is an average pressure, an average value over a second or longer. Long time pressure fluctuations can be measured by this barometer. Long time fluctuations are caused by the changes in weather while the short time rapid fluctuations are caused by the environmental disturbances such as traffic or road work. They cannot be measured by a mercury barometer, but we can use a recorder for the measurement. This pressure fluctuation caused by noise, etc. has a very small amplitude, about lo" atmosphere, and is totally unrelated to the 5% error in the barometer reading. In most of this book, probability is used only with the first meaning. Unless otherwise stated, we shall not consider the second use. Later when we mention probability we shall have to discuss how it is obtained from statistical information. Here we need to emphasise that probability is meaningful only in so far as its value is independent of the size of the data pool. (Of course we cannot have too few pieces of data.) For example, if we throw a die 1 2 0 0 times and count the number of appearances of 'l', we can use the first 600 throws to get the probability, or the next 600 throws to get the probability, or the whole 1 2 0 0 throws to get the probability. The probability obtained should all be approximately 1/6. If we get 112 for the first 600 throws and 1/4 for the next 600 throws we can still say that the statistical result of 1 2 0 0 throws is 1/6 but the meaning of this result is questionable. That is to say, this result of 1/6 must be repeatable in order to be meaningful.
10.2. Distributions and the Statistics of the Average Value
In the last few chapters we have discussed distributions, like the age distribution of the population, or the energy distribution of the states of the particles. The distribution is a result of doing statistics by classification. For example, we classify the population according to age, each age group forming a class, and then we count the number in each class, obtaining an age distribution curve like Fig. 3.1. This is information about the population. Its statistical counting procedure can be expressed by the abstract equation
f(X')=
c 6(X'X(s))
S
,
(10 1)
where s is the name of each person, X(s) is the age of person s and Xcounts S the number of the population. The 6 function perferms the classification, i.e. it
PROBABILITY AND STATISTICS
167
collects all the names of the persons whose age is X’. The number of persons of age between XI and X2 is (102) The reader might review Chapter 3 at this point. Equation (101) is the standard procedure of statistics for classification. The analysis is usually made on the characteristic property X(s) of a collection of “samples”, with each of the samples identified by a assigned number (or name) s. We collect all those with the same X(s) into a class. The number of samples in each class is called the statistical weight of the class. If we plot the statistical weight in a curve, we get a distribution of this special property X. This provides one group of information about the collection of samples. Information so obtained may sometimes be uncorrelated. We then use average values t o describe the essential properties of distribution. For example the average value of X is
(X>= d X ’ p ( X ‘ ) X ’
=
I
c
s
X(s)/N
,
(103)
(104) Here N is the total number of samples and p ( X ) d X i s the normalised percentage corresponding to f ( X ) d X . The average value of (X (X))2is
=I
=
S
d X ’ (X’  ( X ) ) 2p ( x ’ )
1 (X(s)(X))’/N
.
(105)
The average value (1) ‘centre of mass’ of the distribution, and ( (AX)2 ) is the expresses the dispersion of the distribution. The function p ( X ) is called the probability of the X distribution. It is the statistical weight of the class divided by the total number. This is the definition. We do not enquire into the relation
168
STATISTICAL MECHANICS
between p ( X ) and the concept of “chance” or “possibility”. This is often possible, depending on the application of p ( X ) . Any function S ( X ) of X ( S ) has its own average value, e.g. p ( X ’ ) is the average value of 6 ( X ’  X ( s ) ) , and the average value of 1 is 1 :
(S(X) = dX’ S ( X ’ )p ( X ’ ) )
(106)
I
=  1 6 ( X ’ X(s))
1 N S
,
(107)
(1)
=
I
p(X)dX
=
1 = 100%
.
(107’)
Equation (107’) is the normalisation condition, expressing the fact that the total percentage must be 100%. We can combine some classes for our purpose. For example, the class of age below 3.0 can be called the “youth” class and those between 70 and 80 as the class of “the beginning of life”, etc. The number in each class is its statistical weight. Generally speaking, the statistical weight of a group R of the samples (i.e. a subset in the sample space) can be defined as follows. Let the “counting function” O R (s) of R be
eR(s) =
1 ,
if if
SE
R
,
,
eR(s) = 0 ,
Therefore,
s# R
(108)
(109) is the statistical weight of the R class. Let
(1010)
PROBABILITY AND STATISTICS
169
be the probability of the R class. This is a definition. Notice that 6(X’ X(s))dX’ is an example of R , i.e. the set o f the samples with X(s) between X ’ and X‘ + d X ’ is R . If R is the set consisting of samples with X(s) < XI, then
P(R)=P(X<X’)= (e(X’X))=
1
X’
dXp(X)
,
(10 11)
m
is the probability that X < X’. Those samples to be counted may be a group of persons or objects, or the record of events, experimental data, states in the phase sapce, etc. Let us now look at some examples. The following are the last digits of the telephone numbers listed on p. 205, 206, 214 of the directory of Hsinchu, for subscribers with the surnames Ma and Liang: 53631836117711836080666871376898 57815291573155577898809703583555 65652793069225309039677800596755 (1012) There are N = 96 households, and the above members are one characteristic X(s) of each household s, s = 1 , 2 , , . . , 96. Define the probability distribution as
2 S(X’X(s))
s= 1
N
P(X’) =

NX’
N
N
(10 13)
where IVY is the number of times X appears, and N = 9 6 . Here the argument o f the 6 function takes discrete values, hence the 6 function is defined as
6(M) = 1 ,
=
if
M=O
,
0,
if M f O
.
(1014)
170
STATISTICAL MECHANICS
The normalized probability p ( X ) can be calculated from the above information:
X
0
9
1 9
2 4
3 11
4
5 17
6 12
7 13
8 12
9 9
N(X)
p(X)%
0
9.37 9.37 4.16
11.46 0 17.71 12.50 13.54 12.50 9.37 (1015)
From p ( X ) , the various average values can be calculated, e.g.
( X ) = 4.95
,
) = ( X 2 )  ( X > 2 = 7.94
c, = ((AX)’
.
(10 16)
Equations (1015) and (1016) come from the statistics of (1012). The meaning of these results has yet to be understood. For example, why does the digit ‘4’ not appear in any of the 96 households? Another equally interesting and important question is why do the other digits have more or less similar p ( X ) and what accounts for their differences? The answer to the first question is easily discovered but the second question is not easy to answer. Consider another example. From the data (1012) we define
Y ( s ) = X ( s ) X ( s Jr 1)
,
s = 1,2,...,N‘
,
(1017)
where N ’ = N  1 = 95. These are the product of the neighbouring numbers in the sequence of (1012). The average value of Y is
( Y )=
+c
N
S=l
X ( s ) X ( s + 1)
= 24.05
(1018)
Now define
C1
f
( Y )  (X)’ = 0.45
CI
,
(1019)
C1 
c o
=  0.057
((AX)’ )
PROBABILITY AND STATISTICS
171
C1 is the socalled correlation value of the neighbouring numbers. In the same way we can define
k = 1 , 2 , 3, . . . , N”=Nk
.
(1020)
Of course C = ((AX)’ ). Moreover N r ris not N , and hence k must be much , smaller than N so that the definitions of the average values are essentially the same. C, is the correlation function of the distance k . The statistical results show that C1 ismuchless than ((AX)? >.(See Eqs. (1018)and (1019).) This result means that neighbouring numbers are essentially uncorrelated or independent, According to our ‘intuition’, if these numbers are ‘random’, i.e. some X ( s ) will be larger than (X) some smaller than ( X ) ,but there is no regularity, the and correlations C1,C, . . . should all be zero. In the next chapter we shall introduce these concepts systematically. Notice that this correlation function is of course dependent on the order of arrangement of the numbers. The label takes the order of the directory and has no statistical meaning. This ordering is a newly introduced concept and the distance between different s’s is determined by this ordering. Our aim is to use statistics to deal with the information on the molecular motion, and many experiments are performed to measure the average value of the various variables of motion. Our statistical procedure is based on the molecular motion and the average values defined by various probabilities must be defined in terms of the molecular motion. The ordering of time and spatial position is a basic concept. Various correlation functions are most important tools in discussing the regularity of the motion. Let us now look at a slightly different example. Let us consider a vertical x y plane, with gravity acting downwards, the y axis pointing upwards and the x axis horizontal. A particle moves in this plane above the xaxis, which is a rigid floor from which the particle will rebound without loss of energy. The horizontal motion is restricted to 0 < x < L . For simplicity we identify the two points x = 0 and x = I , Hence the motion of this particle is on a cylindrical surface with period L. This is the periodic or cyclic boundary condition (Fig. 101). The motion of this particle obeys the equation of a motion of mechanics. The trajectory is very simple. The horizontal motion is a circle around the
172
STATISTICAL MECHANICS
Fig. 101 Trajectory of a particle in 2 dimensions under a constant gravitational field.
cylinder and the vertical motion is up and down incessantly. Any property can be calculated from this trajectory and the observation time is 0 < t < + , J We first ask: What is the distribution of height of the particle? This problem is not clear because a distribution must have a precise statistical meaning and we have not made this definition yet. In the usual statistics about the variables of motion, we use equal time intervals to be the statistical weight, because the results of experiments are usually the time average. We now calculate the time the particle spends in a certain height and define it to be the statistical weight of that height. Let
(1021) where y ( t ) is the height at time t and p ( y ’ ) is the probability distribution of the height. Equation (1021) is an example of the general definition in (107). Here the sample space is the trajectory. At each instant t there is one sample, i.e. the state of the particle at t. The probability p ( y ’ ) can be easily calculated ( p a I d t / d y 1, i.e. p is inversely proportional to the vertical velocity).
=o
,
ify>h
,
(1022)
where h is the maximum height attained by the particle. The reader can derive (1022) for hiinself. Here T is the period of the vertical motion:
(1023)
mg
and py, is the perpendicular momentum at y = 0 .
PROBABILITY A N D STATISTICS
173
If 3 >>
7 ,then
0(7/3) can be neglected. The integration of p ( y ) is
P(Y)d.Y
= 3 (Yl <Y<Y2)/3
= P(Y1 < Y < vz)
>
(1 0 24)
can be defined as the probability that the height is between y 1 and y 2 .
3 ( y l < y < y 2 ) is the time spent by the particle between these heights.
The correlation function of the height (similar to C in (1020)) is ,
C(t')
Y'(t)
=
1
1
3
3
dty'(t)y'(t+t')
0
,
=
v(t)(y)
1
t'<<
3
,
(1025)
which is not difficult to calculate (see Fig. 102):
t"
= tf/27
,
O<tf<r
(1025 ')
Fig. 102 Graph of C ( t ' ) . C ( t ' ) is aperiodic function, because y ( t ) is a periodic function.
174
STATISTICAL MECHANICS
Next consider the average force acting on the floor between the time 0 < t < the particle is in the air, the floor is not subject to any force. When the particle hits the floor, the momentum change is 2py0. If the particle hits the floor once in every time interval, then the nth time occurs at n ~ The force . acting on the floor is
3.When
W(t)
=
n
c
m
2py, 6 ( t  n 7 )
.
(1026)
=rn
The average force is
(W)=
1 3 1 dt W ( t )
3
0
(1027) where N is the number of times the particle hits the floor. If 3 >> can be neglected:
7 ,O ( T / J )
( W >= 2pr0/r
T
=
mg
,
(1028)
is given by (1023). The above examples give the statistical procedure. The concepts of probability distribution and average values are used to compile and express accurately the available data. They do not involve any explanation for the data or matters of physical principle. These examples are not yet adequate to explain the usual definition and use of probability. In the next section we shall turn to a general discussion.
10.3. The General Definition of Probability
The above examples point out how probability is obtained from the statistics. Now we give a definition of probability in the most general and widely used terms, which is, unfortunately, somewhat more abstract. Then we give some examples. To define probability, we require three things: (a) sample set; (b) classes and (c) the statistical weight of each class. These should satisfy some conditions:
PROBABILITY AND STATISTICS
175
(a) The sample set E must be nonempty. (We have at least one sample.) (b) Each class is a subset of E and each subset is a class. Classes should satisfy the following conditions: if R l is a class and R 2 is another then their union R 1 u R 2 is also a class. Their intersection R 1 n R 2 is also a class. In addition, if R is a class then its complement is also a class. (c) Each class is assigned a statistical weight P ( R ) , satisfying
The last condition P ( E ) = 1 is the normalisation of the statistical weights. If it is not satisfied, we only need to divide each P ( R ) by P(E). The above definition is very general. Any function P(R ) satisfying the above conditions is a probability. This is the mathematical definition of probability and is the starting point of modern probability theory.a Notice that this general definition does not mention statistics, although the probability defined by statistics always satisfies this general definition. In many applications of probability, the statistical meaning is always ignored and regarded as obvious. But sometimes problems may arise. Our main point here concerns the statistics. Hence in applying the theory of probability we should always bear in mind the statistical meaning of various definitions and conclusion.
Example 1 : Throwing a die Let the sample set be E = { 1 , 2 , 3 , 4 , 5 , 6 ) . The classes are all the subsets of E. The probability P ( R ) is the number of samples in R divided by 6. The definition of this probability does not involve statistics but we know its statistical meaning. Notice that the sample here is not the experimental recordings of the example in the last section; nor the trajectory of motion but an abstract set of six possibilities. We can think of throwing a die many times and classify the events into six classes. The six samples of E are thzse six classes. Example 2: Throwing a die Using the same E , we use another classification ( R , ,R I , E ) :
R
=
{1,2,3}
=
, ,
R’
=
{4,5,6}
,
P(R)
0.55
P ( R ’ ) = 0.45
.
(1029)
aThe mathematical analysis of probability is discussed in Yang (1979).
176
STATISTICAL MECHANICS
The probability thus defined also satisfies the conditions (a), (b) and (c). This probability can be used on a biased die. This example indicates that classification need not be too refined, and it only needs to satisfy (b). The subsets in this example are fewer than in the above, subset like { 1 , 3 } , { 2 , 5 , 6 } , { 2 ) being excluded.
Example 3: Take the set of all integers: what is the probability of obtaining an even integer? The answer obviously is f . What is the probability of getting a multiple of lo? Answer: $ . The answers seem obvious. The statistical meaning is that , from 0 to N ( N is a large number) about N / 2 are even integers and N/lO are multiples of 10. Example 4: We draw any chord on a circle of radius 1, what is the probability of the length of the chord being less than l ? This question seems simple but is meaningless, because there is no obvious statistical meaning in drawing any chord. If we draw ten thousand chords, how do we draw them? Are they parallel and uniformly distributed in the circle? Or are they uniformly distributed on the circle? Each situation will give a different probability. Generally speaking, a continuousvalued sample space does not have obvious statistical weights. Now let us look at two important examples. Example 5: The distribution of time in a trajectory. Let E be the region of the trajectory in the phase space of a system, i.e. the region of motion. Now we define the states in the region as the samples and classify the states by a certain variable of motion X ( s ) , s E E . If R is a class, then define the probability of R as
fYR)=
J,/3
7
(1 0  30)
where R is a subset of E i.e. a region inside the region of motion and time the trajectory spends in R , where 7 is the total observation time.
3R is the
Example 6: The situation is as above but we now define the probability as
(1030’)
.n (1032) be a small number. 2. . should be smaller than the minimum of X ( s ) and X .PROBABILITY AND STATISTICS 177 where r ( R ) is the volume of R .e.. Notice that (1030) is based directly on the trajectory of motion. (whose value is of order N ) . Then the average value ( X ) and distribution function p ( X ) of X are p(Xk)dX P(Rk) .l = dX . should be larger than its maximum.<X. < X ... Of course. r. (1033) Example 7: Use (1030) as the definition of probability. . (The second example of throwing a die demonstrates this point.. . Let X . Here we must emphasise that a classification is a collection of subsets of the sample space but does not necessarily include all the subsets. it only needs to satisfy (b). Remember that in Chapter 5 on the basic assumption we used quantum mechanics to make the region of motion into a discrete rather than a continuous sample space.. i. From the general definition of probability we can define the average values.3. } k=l.n . k = 1 . (1031) That is to say all the samples with X(s) between XkWland X k are collected in R. Let X(s) be a macroscopic variable. 1 { s .. the total number of states in R and r ( E ) is the total number of states in the region of motion. and r have definite meanings. R. = . Equation (1034) is just (686) withA = X and KA = K .) . X. X . Xkl < X(s) < X . Hence r ( E ) and r ( R )are not ambiguous.. while (1030’) does not mention the trajectory (of course E itself is determined by the trajectory). whether the states are continuous or discrete. Let X. Then (1034) The definitions of X and K are like those of A and KA in the last section of Chapter 6. < X 2 < X ..2.
The total number of states in a trajectory is determined by the motion. 3. If in each interval 7 . Thh must be taken into account. Let r = (x. i.e. r ( R ) and I‘(E) are of order e N so that In r ( R ) and In r ( E ) are of orderN. f 0. y ) = 1 . 0 < y < l ) .each molecule changes its state once. (1036) = jdx p(x) &(x’f(x)) Notice that each x’ may have several x’s satisfying x’ =f(x). and x’ = f ( x ) . This kind of probability is of no use for the analysis of equilibrium properties. y ) in the polar coordinates as well as p ( r ) and p ( 0 ) . what is the distribution of p (x’)? Answer: P(X’) = (&(XI f(x) 1) . Change of variables. then each class R is a large region within the whole region. Calculate the distribution p ( x . But if the volume of R is comparable with E. The trajectory is the set of states actually occuring in the observation period. . i. b and the mean free distance on the answer. then the total number of states in the trajectory is about NJ IT (1035) This is negligible when compared with r ( E ) e N . Calculate the distribution of the velocity of the leaking molecules. If it accidentally passes through R in a certain instant. 2. The reason is as follows. Find the distribution of the intensity of illumination on the ground. Hence the definition of (1030) for P ( R ) can have a clear equilibrium meaning. Therefore. p ( x . y ) be distributed uniformly in the square (0 < x < 1. is zero.e.  TR Problems 1. If we know the distribution p ( x ) . A street lamp is at a height h above the ground. The wall is of thickness b .178 STATISTICAL MECHANICS If we classify the region of motion according to the values of macroscopic variables. Hence the classification indicated in (1030) only includes this kind of large regions. and P ( R ) will not depend on changes of ’3 . then the time the trajectory stays in R is comparable with 3 . if R is a very small subset. There is a small hole of radius u on the wall of a gas container. thus becoming an equilibrium property. the trajectory may not pass through R and 3. Discuss the influence of a.
5. ( x 2 ). (a) Calculate ( x ). Calculate the distribution of these points. 4. (c) Discuss the randomness of this sequence. (b) Calculate C. 1. N = 100. Use a calculator to calculate the coordinates of the particle bouncing on the floor in the last example of Sec.. 10. and r is the period of the vertical motion..) . (See Eq. . vx is the horizontal velocity. (10231. . This distribution naturally is dependent on the relative magnitude of L and 1 where 1 = v X r . Take the last 4 digits from 100 numbers in a directory and call them . 2 . x N .xz. k = 0.PROBABILITY AND STATISTICS 179 xI.2.
Statistics provides a quantitative method of analysis for such phenomena. The correlation value is a very important concept and it indicates the degree of independence of the different variables. Independence is a very important phenomenon and it can simplify many problems. 11. as discussed in the last section in this chapter . however its origin is not clear.The Definition and Consequence of Independent Phenomena Let us first review two definitions: The joint probability of R 1 and R z is denoted by (111) Given condition R z . the probability of R1 is (112) . The correlation value of many variables of motion can be measured by scattering experiments. Many variables of motion have essentially mutually independent fluctuations but the test of independence is by no means simple.Chapter 1 1 INDEPENDENCE AND CHAOS Independence means an absence of interference between phenomena.l. a situation often described by the words chaos or randomness. This chapter discusses the definition of independence and gives examples to illustrate its effects. which therefore show no correlation of causal relation.
Y) [see (1031) and (1033)]. The definitions are similar to (1031) and (1033). We classify the sample set with values of ( X . The definition of independence for the variables X and Y is (117) That is.INDEPENDENCE AND CHAOS 181 The condition of given R z means that we take R . Let (113) (114) The reader can prove that In a similar manner we can define the joint distribution of more variables and their conditional distribution. then any functions f a n d g will satisfy . i. R . From these definitions we can determine the joint distribution p ( X . as the sample set and then classify within R z . does not affect R 1 . the statistical data on X does not depend on Y. If X and Yare mutually independent.e. The definition of independence is as follows: If (1 16) then R 1 and R.or vice versa. Y) and the conditional distribution p(XI Y) of two variables X and Y. are mutually independent.
N ) . the distribution of n . Now let us assume that (1) the ni's are mutually independent and (2) the probability of ni = 1 is p = (n. independent variables to obtain the distribution or average values for other more complicated variables.182 STATISTICAL MECHANICS Let the correlation function of X and Y be c= (XU). Therefore n = i= 1 ni . .( X )( Y ) = ((X. Example I : N gas molecules are in a container of volume V. Let v be the volume of a small region. the application of probability will be severely limited. (119) Hence if X and Y are mutually independent. and n the number of molecules in this region. V are both large and the density is . (1 1.1 1) From these two assumptions we get p(n) = ( :> Pn (1 . Independence is a characteristic phenomenon. then C = 0. Calculate p ( n ) . We shall discuss it in the next section. then ni=l . It is not a priori obvious.) If N . Many applications in probability merely manipulate the combinations or permutations of some simple. If the ith particle is in the small region. Independent phenomenon is chaotic or random. . (1 1. To test whether several variables are mutually independent is not easy. 2 . ." . We must emphasise that independence must be tested by statistical data. N (1110) Otherwise ni = 0 . It seems simple but its origin is not obvious. .12) (This is the binomial distribution. Define ni as follows: ( i = 1 .P I N . The correlation function is a measure of the degree of regularity between the variables.)= l V V . Now consider several applications of some independent phenomena.(X))(Y(Y))) . If there is no such phenomenon.
13 ') where a 4 n r 2 d r is the probability of having one molecule between r a n d r + d r while e . R = r l + r z + . ' Example 3: A longchain molecule (free to move in water) is composed of N links.then (1 112) can be simplified as (1 1 . each of length 1.INDEPENDENCE AND CHAOS 183 N / V = a.where R is the relative position vector of the first and the last molecules.13) (This is the Poisson distribution.e i e ikj'R JN (1116) . .4 n r 3 a / 3 is the probability of having no other molecules in r . what is the distribution of distance r between two molecules? Answer: p(r) = 4nr2 (ye4nr2a13 (1 1 . . Let us now look at a more complicated example. + rN IriI = 1 .e. . Calculate (u2 ). The displacement u is thelinear combination of the fundamental modes of vibrations g3j : 1 $j(R) = .) Example 2: In a gas of density a. (1114) The value = may be regarded as the size of this molecule. Suppose the direction of each link is an independent variable. ( u ) = 0 . Calculate ( R 2 >. d s Example 4: Let u + R be the position of an atom in a crystal where R is its equilibrium position. distributed uniformly. i.
17) The value A is a large number. and N is the total number of atoms. Then we can add the different a groups and get T = ldwg(w) ma ' v . and we get 1 N. The variable of motion is v j . The total energy of vibration is pi" I(+ 2m i H = n wfvf z 2 (1 1 . Let us assume that the vj's are mutually independent. we can directly use the result of the equipartition of energy (if we assume the temperature is high so that we can ignore quantum effects). (1 1.18) where rn is the mass of the atom and wi is the frequency of the ith element. kj is its wave vector. i=l I Trnwiqf (1 1. e j is the polarisation.19) Now assume that these N. (1 120) . modes have essentially the same frequency a . Then A = C T $ i . Because each frequency is shared by many modes.184 STATISTICAL MECHANICS The fundamental vibration is a plane wave. We can use the methods of Chapters 5 and 6 to calculate it.T 2 = N. .
If the reader is not clear about this point. (777) = T  . The meaning of (u2) is a time average of u2. There N m u s t be a larger number and this is a necessary condition for calculation. (1 122) ma. . then g ( o ) a w and we run into difficulty with this integral. we get (1 121) The reader may ask: Why do we bother to go through (1119) and introduce No? Why d o we not directly say that each mode satisfies (. (636).17). but we must be careful here about the meaning of the average value ( . see Chapter 6 . . . then g(w) a w2 [see (642)] and this just cancels 1 / w 2 . In the discussion of the equipartition of energy in Chapter 5 . Hence the meaning of the ( . . we only used the basic assumption to define entropy and then calculated the total energy of Nmodes and the result was NT.g. since it is a large number. If w is small. Generally speaking.mc.INDEPENDENCE AND CHAOS 185 where g(w) is the distribution of the vibrational frequencies. Would that not be simpler? There is nothing wrong with this procedure. Hence there is no point in taking the average ( . Equation (1 1.20) comes from the basic assumption. But if the crystal is in a two dimensional space (e. . d dimensional crystals have g(w) a ud' (1 121) is essentially and hr h 42nmke.$V?) 1 = T 2 ' (equipartition of energy) : . ) . ) on the righthand side of (1 1 . Equation (1 121) is a very important and interesting result. ) symbol in (1 122) is very unclear. a layer of atoms attached to a plane). Substituting into (1 117). (1 123) . . It shows that the low frequency vibrations have a significant influence on the displacement of the atoms (because of the factor of l / w 2 in the integrand). he should review the last section of Chapter 6.
When d < 2 . the above result shows that the displacement of the atoms is not very large because OD is about 300 K (see Table 61). then (suppose ( X I ) = (XZ) = 0 ) 9 = cosl$ 1 + o (1 ) .186 STATISTICAL MECHANICS where 8 is the Debye temperature. are independent and uncorrelated. 2 a3 (1 124) i. However.  11. This result will be discussed later (in Chapter 29). Different statistical procedures determine different probabilities and average values.e. XI and X . temperature. Mutually independent variables can be considered separately. lack of connectedness or the absence of interaction do not guarantee independence. In addition to the time average. The g(w) in (1123) is a generalisation of (642) (x = w/w. Let their displacements be X 1( t ) = cos w t . (u2) + i.2. and X is the wavelength of the atom at this . it will pose no problem if we clarify the statistical procedure.5 8.). The magnitude of 7 depends on the problem under study. Test of Independence The examples in the above section obviously demonstrate the importance of the independence property. But if they have the same frequency. Each problem is different and should be treated by its own statistical procedure appropriate to the situation. Many experiments require repeated observation. . For example. they can interfere. the low frequency vibrations cause the displacement of the atoms to increase without bound. according to usual terminology. are not independent variables (assume l$ # 7 ~ / 2 ) Hence. In modern terms. we have two unconnected simple pendula oscillating separately. and h is about 0. X z ( 1 ) = c o s ( o t + @). Notice that (1) we have customarily taken the observation time 3 as finite. if the frequencies of two laser beams are exactly the same.e. Now we discuss some fundamental problems of independent phenomena. tion. (2) Here we repeat that statistics is the basis for calculating the probability and average values. simplifying many complicated problems. The terms “independent” and “uncorrelated” do not mean spatial separation or disconnection. Perhaps the whole process is repeated on several different bodies. we have also to take a grand average. separa. These two pendula. Besides this. then we take an average again.
The probability for each subset is the number of balls in the subset divided by 4. therefore P(X) = P ( Y ) = P(Z) = 1 (1125) In addition. Y . + 1 for yellow. .1) soft (1) red (.1 for soft. + 1 for hard. The data are listed as follows: Property \ Ball 1 2 3 ~~ 4 ~ colour ( X ) weight ( Y ) hardness ( Z ) yellow (1) heavy ( 1) hard (1) yellow ( 1 ) light (. half are heavy and half are light. Y. Z are pairwise independent. The variables have two values + 1 or . even if a group of variables do not completely satisfy the definition of independence. half are hard and half are soft.1) heavy (1) soft (. P(XIY) = P ( X ) 1 (1126) . each quality is shared by two of the four balls.1 ) light (.e. The classes are all the subsets.INDEPENDENCE AND CHAOS 187 To satisfy the definition of independence in the above section is not a simple task. 3 . To test whether a group of variables are independent is very difficult. i.1 for red.1) red (. 4 . Example 5: A sample has 4 balls labelled by S = 1 . . In fact.1. Y to be the weight. many useful results are not affected too much. We define X to be the colour. That is to say. + 1 for heavy.1) hard (1) Obviously.1 for light and Z to be the hardness. . of the yellow balls. Z be three variables (three qualities of the sample). Let X . the usual practice does not require complete independence. The variables X . We first look at an example to illustrate that independence is not easy to satisfy. 2 .
we discussed a process of copying out a sequence of numbers from the directory. Therefore. Y . Each throw is uncorrelated with any other throw. The variables that we want to examine naturally are the variables related to the motion of the molecules. From this example it is not difficult to see that if there are many variables. we want to calculate ( R 2)and we only need (1128) Then the result ( R z) = NZ2 can be established. etc. For example. because (1127) A heavy and hard ball can only be yellow and a light and soft one must be yellow too. There is no need to require the independence of three variables. the central limit theorem. This sequence of numbers is an example of a random sequence. i. this does not mean that any three variables are mutually independent. it is a very difficult job to test their independence. For the same reason if we can prove the case for three variables. 10. . The sequence thus obtained is random. etc. we have no guarantee for the case of four. the three variables together will be no longer so. This theorem can be used to estimate the mutual independence in the case of many variables. the test of independence is a major problem. 2 are not mutually independent. the pairwise independence is sufficient. But this test is not always necessary. Independent phenomena possess a very important property. although any two variables among the three are mutually independent. Random Sequences In Sec.188 STATISTICAL MECHANICS But X .e. From the above example we can see that if we have too many variables. We shall discuss it in the next chapter. That is to say. The numbers obtained by throwing a die are a sequence of random numbers. Of course. in the case of the long chain molecule discussed above. In addition a light and yellow ball must be soft and a heavy and red ball must be hard. we only require the ri to be pairwise independent.3. 11. In many applications.2 (see Eq. (1012)). there are also many cases requiring more. We know roughly what is meant by randomness: each case of throwing the die is an independent event. If we have proved that any two variables are mutually independent.
then this sequence is random. . Here f. Let X j ( r ) = Xi+. (1 130) then each X ( r ) is independent and the sequence is random. However. X 3 . The sequence given by the die cannot be altered at will. Statistics is the basis of our definition. If i=O. This way of thinking is not correct. .. Here N is a large number. X 2 . any sequence written “at will” may not pass the test. For example if we simply write down a sequence of numbers “at will”. or “indefinite”. we only need to consider the correlation function C(r) Ei (X(r)X(0))  (x)2 . then it would probably not be completely random. 1. But we must notice that we need a die as our tool. In many applications. be a certain sequence.e any two functions. If each number is independent of every other number.INDEPENDENCE AND CHAOS 189 We always mix the term “random” with “at will”. we must consult a random number table or obtain it from a computer programme.. . . The sequence must first of all be very long. . To obtain a truly random sequence. Assume that the sequence is much longer than N . yet it is “random”. The sequence obtained from a die is random enough. determining its degree of randomness. The definition is very simple: Let X I . We throw a di? at will and do not know a definite result before hand. We use this definition to check a sequence.2. But note the following: A sequence generated by a table or by a programme is definite enough. ( 1 132) . Of course we need to have a precise definition of randomness.. The definition of mutual independence must be determined by statistics: Let I N+k ( 1 129) This average value must be essentially independent of N and k .g ar.
Although it is not a discrete sequence. (See Eq. On a time scale larger than T . if n ( t ) is the number of molecules in a certain small region of the gas. This will certainly pass the test (1 131) or (1 133). we can shoot in some particles (e. n ( t ) is random.( n Y dt’n(t’) The observation time r G r . . energy and the angular distribution of the spins of the particles scattered by the . We write a sequence of numbers a t will.e. .4. 11.) Because of the development of the computer. This kind of calculation shall be discussed in Chapter 22.190 STATISTICAL MECHANICS If r f O .g. (1 135) then we call 7 the correlation time. random sequences of enormous length (up to millions) can be obtained. The sequence in (1012) can more or less pass the test. We can make use of these sequences to simulate the various variables in the molecular motion in order to calculate the equilibrium properties. If C ( t ) is only nonzero for C(t)= 0 . Scattering Experiments and the Measurement of the. For example. thevalue of n ( t ) at different times are independent. (The origin of this property is still not clear. i. photons. (1019)) Sequences associated with statistical mechanics are often functions of time. its randomness can still be defined. the various variables of motion of the system. (1 1 33) then the sequence can be regarded as random.e. Correlation Functions If we want to observe some details of the motion of a system. neutrons or electrons) and measure the momentum. (1134) 7 must t >> T be suitably long. According to the results of many experiments the variables for an equilibrium system all have this property of independence between different times.e. i. we can define its time correlation functions: C(r) I ( n ( t )n ( 0 ) ). C(r)= 0 . i.
(1137) where m is the mass of the neutron. (1 139) Assume that U is very weak. Assume that the positions of the molecules ri(t) are known and we neglect the spin of the neutron.r).V2 at 2m a 1 $(x. t ) is the interaction of the neutron with all the molecules. t ) $(x. and the interaction of the neutron with one molecule is v(x . The wave equation for the neutron is i . we can deduce certain details of the structure of the system.$(x. t ) + U ( x . When t = 0. where x . t ) . t ) .r is the relative distance between the neutron and the molecule. t ) = .INDEPENDENCE AND CHAOS 191 system. The scattering experiments directly measure some correlation values. The influence of the molecules of the body on the neutron is represented by a potential energy U ( x.(t) be the coefficient of expansion of the plane wave $ : (1 138) Let the momentum of the incoming neutron be q . then (1 137) can be written as i uq. ri ( t )is the position of the ith molecule at time t . Then U ( x . Assume this interaction is (1 136) where x is the position of the neutron.(t) at a = (1 140) (1141) . From the result of the measurements. Suppose we use neutrons to irradiate a certain body. Let us now review some basic ideas. Let a.
( 1 1 42) where k is the momentum transferred from the neutron and o is the energy. Different values of q. Hence.E = = (q + 9’) k/2m  .q2)/2m E E‘ . the amplitude that the neutron changes its momentum to q’ is the solution of (1 140) (1143) Its square divided by 3 is the scattering probability per unit time: P ( k . the integral of (1 144) with respect to o is d o P(k. o. o) . ( t ) Uk(t) 3 (1 1 45) . (1 1 44) The probability P ( q + q ’ ) can be obtained by experiment. o) = m 2n j 03d t U . q ’ can lead to different k . A .From P(k. Equal time correlation value Because s. We first look at the result of integration with respect to o. ( q f 2. o)we can calculate some correlation functions.192 STATISTICAL MECHANICS k =q’q o . m d o eiWt = 2 n 6 ( t ) . in time 7 .
Different time correlation value If we do not integrate over o. can obtain the different time correlation we value of the density of the molecules from (1 144). is a microscopic scale and the body is homogeneous. e . I/ is the volume of the system. . The number u k is directly related to the density of the molecules. t ) = C 6(x .INDEPENDENCE AND CHAOS 193 Unless otherwise specified. The analysis is as follows: . B.r i ( t > > i (1 146) Therefore from ( 1 145) we obtain (1 1 47) then Hence. Next assume that 5 . Assume that ~ ( x and ~ ( x ’ ) ) are uncorrelated if Ix . the correlation function of the fluctuation of the density of the molecules can be obtained from scattering experiments. From ( 1 136) we get p(x.x‘ 1 >> $ (the correlation length). . .i k * r ( P ( r >P ( 0 ) )c 9 (1 150) assuming that k # 0. Then (1149) can be simplified as A = v s d 3 . the bracket < . ) means time average.
w) = 2 T 6 ( W ) I<uk)lz f d t ‘ eiwt’ <u. u .194 STATISTICAL MECHANICS let t = ( t l + t 2 ) / 2 . 111 If t’ is fixed. Now let us suppose that t’ is much longer than the correlation time 7. In addition assume that ( U k ) is approximately a constant. ta = 0. t ’ .* k ( t ‘ ) ) . Fig. If these assumptions are valid. (1 153) .t’ = t l . Substitute (1 152) into (115l)andwe get P(k. 111 for the integration region o f t .52) is the correlation value of u k with its value after a time t’. then tb = J . the integration of t along the 45” line such as a b in the figure. The last term of (1 1 . so that this correlation value tends to zero.then See Fig. Define The average value is the time average between ta and tb .t 2 .
. ( F o r further analysis. Let x 1 .INDEPENDENCE AND CHAOS 195 Using (1 146).x3. . 3) . . The definition is very clear. n = 1.. Calculate r1nI X . (1 155) assuming that the distribution of each xi is normal and (x?)= 1 . Problems 1. see Feigenbaum ( 1 9 78).. 5 . .] = 4xx.2.) . Let the sequence x x 2 . . If is close t o 1 . x 2 . 1 . ( 115 9 ) . Calculate the distribution of (1156) where xi are as in the above problem. . Use a calculator t o find this sequence for 0 < x < 1 and 0 < h < 1 . . The correlation value is an average with respect t o time. . . Equation (1 149) is an integral of this equation with respect t o o. Let a random variable x be uniformly distributed in 0 the distribution of t f i x < 1 . be generated by the formula x.) 3. .(ix. (1158) Discuss the randomness of this sequence. be independent variables. .x. 2. 2 . . (Yang 1979 . ( x i) = 0.. (1157) The values of x 1 and h are fixed in advance. These results connect the scattering amplitude with the density fluctuation. Let x = sin( 2. . . n = 1 . These results involve neither thermodynamic concepts nor the basic assumption of statistical mechanics. we can simplify these equations as + VJdt (1154) This is a generalisation of (1 149).) 4. .x. . this sequence seems to be random. Calculate the distribution of the product y = X1X2.
The spins of the molecules can be regarded as classical variables. The velocity distribution of gas molecules is normal: 2 % p(v. . We first make a preparation here.196 STATISTICAL MECHANICS where r is a constant. 0 < 6 < 77. From a uniformly distributed sequence to a normally distributed sequence. u are all 2 x 2 matrices and the wavefunction of the neutron has t w o components. ( 1 160) of a normal distribution p ( y ) . the spin u of the neutron interacts with the spin si of the molecules. 1 1. 9. 7. The steps are as follows: ( a ) Take the uniformly distributed random sequences 0 < r < 1. The reader can think of other methods. In the process of scattering some spins change while others do not.z / 2 and hence Z has a distribution like that of the velocity of the gas molecules in t w o dimensions. ( b ) Let Y 2 z =[21n+] . If ( 1 136) is changed to i.) = ( 2 7 ~ 0) e v:/20= I = T/m . 8. with ( t Z > a * . (Answer: ( l / ~ ) e . .) . This distribution should be familiar.e. we can obtain an exponentially distributed random sequence from a uniformly distributed random sequence. The normal distribution will be discussed in detail in Chapter 12. This is not so simple as the exponential distribution (see problem 5) but it is not too difficult either. Notice that r = e . ( t ) = 0 . 6." ~i.4 using quantum mechanics. then what is the relation of the scattering amplitude with the spin correlation function of the molecules? Notice that here U . an exponential distribution. Therefore. t is the projected velocity in a specified direction. Analyse the content of Sec.e. Derive the above. ( 1 161) t = uzcose = then t is a normally distributed random sequence.
There is a probability ( 1 .e. The different yii’s are mutually independent. in which case there is a probability I that the guess is correct.with equal probabilities. What is the probability that the student actually knows the answer? 12. and the solutions may be found therein. Prove that (b) Let A u A . A i s i . i. but simply makes a guess.p ) that he does not know the answer.u A . . The elements yii of an n x n matrix are either + 1 or 1.INDEPENDENCE AND CHAOS 197 10.. = E and all classes are disjoint. It is found that the answer he gives i s indeed correct. .) . A i n Ai = (1164) 11. Prove that . In an examination. there is a probability p that the student knows the answers. . Find the average value of the square of the determinant En = [Det(v)12 (1 165) (The last three problems are taken from Yang (1979).
with the width proportional to f i We shall carefully discuss the meaning of the word “essentially” later. 12. which is very useful in calculations by expansion.1. Then we give examples to illustrate the applications of this theorem. The first gives a general idea of the theorem while the second probes into the essence of the theorem. We use two methods to derive this theorem. 121 . the first being a direct proof and the second containing some subtleties. This theorem points out that if N i s a large number. We also present some material about the cumulant.e. Normal Distributions The definition of a normal distribution is (121) Some of its properties can be seen from Fig. In this chapter we use it to analyse the central limit theorem. and extend this assumption so that it is more effective both in principle and application. then the sum ofNessentially independent variables will have an “essentially” normal distribution. Then we use this theorem to discuss the basic assumption of statistical mechanics. i. the central limit theorem.Chapter 12 SUM OF MANY INDEPENDENT VARIABLES Independence leads to a very important result.
7 % of area 4 I Fig. ) n = 1. Its properties can be expressed by various average values: ( Y 2 ) = u2 . .. Expanding this formula in powers of k . 6BL3% ofp: '" ~ 5 .SUM OF MANY INDEPENDE~TVARIABLES 199 50 'I: of area . we get (122). ( Y 2 n + 1= 0 . They can be expressed by a very simple formula where f ( k ) is called the characteristic function of p ( Y ) . (Y 2n (2n)! )= u2n . (122) These results can be directly integrated from (121). 121 Normal distribution with u = 1.2. 6 % area of 99.
. (128) These can be directly obtained from (126). getting ’ q k / f i ) c= Ai(0) + A. X .2. = The quadratic term is AI’(0) =  (32) . (126) Suppose N is a large number and we expand A i ( k / f i ) . Substituting (127) into (12S). . The Central Limit Theorem (Rudimentary Discussion) Let X I . . Let = (124) Let us now derive the characteristic function of Y : N = exp j=1 Ai(k/fl) ikXil fi (125) Ai(k/fi) = ln(e > .S ) O 2N k2 k3 .’(O) + N ( N .. XN be N mutually independent variables.200 STATISTICAL MECHANICS 12. . (129) . because it is proportional to (Xi) 0 . (127) Notice that the first order term in k is 0 . we immediately obtain 1 ( e . .i k Y ) = exp{ .k 2 0 2 2 + O ( )) fi 1 . . Suppose (Xi) 0.
. if we add N independent variables and divide by N . since (129) is the same as (123) if we neglect terms of O ( l / f i ) . Y / f i i s also normal with width a / f i i n the variable Y/&: (121 1) This can be regarded as the mean of the Ndifferent Xi’s i f N i s large. the fluctuation is very small. Of course fiY = X . & Example I : Binomial distribution In Chapter 11 (see discussion around Eq. Y is normally distributed. Notice that k in (129) cannot be too large. + XN has a distribution width of fro. If <Xi) 0.e. the mean differs very little from the average value of this mean (1212) i. then # (1 2 . If k > then (129) cannot be established. + . u / f i and is small. Let us now consider some examples. . then the fluctuation of the Xi’s is largely cancelled. If we add up N independent variables. ni = 0 or 1 (1213) .SUM O F MANY INDEPENDENT VARIABLES 201 This is the central limit theorem.( Y ) is a normal distribution and we only need to change Y to Y . i. That is. In other words.( Y ) in (1 29).10) Y . (1 110>) we discussed the distribution N n = C n i i=l . Hence the socalled normal distribution of Y is limited to the calculation of the average values of lower powers of Y . As Y is a normal distribution with width u .e. the increase of fluctuation is only f i not N . + X .
us how close are tells ( n ) = 50 No2 = 49 .0 199 0.202 STATISTICAL MECHANICS The probability of ni = 1 is p and the ni are mutually independent.0000 0.0001 .0212 0.0424 0.0002 0.0205 0.0570 0. (1217) Table 121 Values of p ( n ) calculated using (1214) and (1215).0442 0.0010 0.0460 0.0442 0.0052 0. The resulting distribution of n is the binomial distribution 4 = 1 . Notice that _. (1214) Eq.0205 0.006 1 0.0569 0.p . (1214) and (1215).0013 0. n 25 30 35 40 45 50 55 60 65 70 75 Eq.0010 0.005 7 0. . (1214) Using the central limit theorem.0001 0. (1215) 0.005 7 0.02.0006 0. we get (1215) (1216) Table 121 shows p ( n ) with N = 2 500 and p = 0.
.. + < $ X k ) c < X i ) + <XiXjXk)c The last term i s the third cumulant. To generalise. (1218) where each ai is a positive constant. (1219) Because ( X . . + . = ( X 2 ) can be called the second correlation value.3. the central limit theorem is not applicable. then its correlation value is 0 or the cumulant is 0. the second cumulant. Higher Order Averages and the Cumulant We first generalise the definition of the correlation value. (1222) If Xi and Xi are mutually independent. Equation (1221) is a special case of (1222). of different variables is defined as ( X j X j ) = < X i > < X j > <XiXj>c + .e. ) is infinite. Find the distribution of the sum of the X i . the case i = j .SUM OF MANY INDEPENDENT VARIABLES 203 Example 2: Let the distribution of the variables X i .+ XN . We give it another name . i = 1 . (1 2 . 2.e. * (12 23) . . We have defined before ( X 2 )= i. . . + X . Let the sum be X = X .( x ) 2 . The second correlation value or cumulant ( X i x i ) . The reader can prove that the distribution of X is Nu T(N’ a2 + x’) .20) 12.N be a. let <Xi$Xk ) = C X i > < X j > < X k ) + < X i $ ) ( X k ) + ( X i X k ) <xi) . c E ( X 2 ). . (1221) (X’). + (X”. i.
xk are mutually independent. (XkXl). two are not mutually independent. . But if any two variables are independent.e. + (xjxjxkxl)c . . . These expansion can be represented by simple diagrams.we # . If xi.) These diagrams are helpful for memorising and analysis. then the three are not independent. we can regard the correlation as the combination of the products of the cumulants.204 STATISTICAL MECHANICS The case of three variables is much more complicated than that of two. + . The correlation of X i . The expansion of higher products is more lengthy. 112 of the last chapter). + (Xj$xk). 122. The special case when the variables in (1223) and (1224) are all the same gives ( X 3 )= ( W 3+ 3 ( X 2 ) ( X ) + (X”. . i. ( X 4 )= (X)4 + 6 ( X 2 ) ( X > 2 + 3 ( X 2: . (If (Xi)0 .xi.22) to (1 2.. ( X l ) + . Let us first look at the role of ( X 2 )c in the high order average value.. ( x k X i ) . ( X i Xi ) = ( X i x k ) = . . and besides these is the “pure” correlation among the three. Let ( X i ) = 0 . (See Fig. If among the three variables. Therefore in discussing the average values of the product of more than two variables. Xi . ) Equations (1 2. . xi and x comes from pairwise correlak tion. this does not guarantee that the three variables are mutually independent (see Sec. The fourth cumulant (Xi x k X l ) can be similarly defined as: + ( X j X i ) . (6 terms) ) (3 terms) (4 terms) (1224) + (Xixi). Each cumulant is represented by a group of connected points and isolated points represent separate average values. = 0 .25) are the cumulant expansion of the average value of products. ( x k ) ( x l+ . then any two among the three must be mutually independent i.. the third cumulant ( X j x j x k ) .e. In the next section we shall discuss the expansion theorem for any product.
.2 6 ) 2n n ! The number A . xj Xk x. then the average value of the new variable is 0.xj>= 205 @ @ ' I + t i n e e + <x. then the expansion of ( X Z )is n (2n)! (XI") = ~ (XZ$ +R . Therefore. if there is only one variable. where (2n)! A.>= em ek e I+ Fig.SUM OF MANY INDEPENDENT VARIABLES <x. (1 2 27) .) In the expansion of the 2 n = order average values. 122 The expansion of the cumulants. = ~ (1 2 . terms containing the second cumulant. Therefore it is always possible to put (Xi) 0. can subtract (Xi) from Xi. is just the number of ways of arranging 2 n variables in pairs. there are A.
Wejust invert (1221) to (1225). (XiXj).3(x2)2 (1231) .e. = (XiXj) . Y = X2 ( Y ' ) . The reader can prove this. = (X4) ( Y 2 ) # ( X4 ) . let then (X)= 0 . . = ( x 3 ) 3(xZ( x ) ) + 2 ( ~ ) ~ (12 29) (2) If Yi is a linear combination of Xi. ) is defined by the variable. = ( Y 2 ). the definition is also changed.><Xj> .(X2)* but Hence (X').( Y ) 2 = (X4). For example. (3) If Y = f ( X ) is not a linear function of X . ( x 3) .<X. .206 STATISTICAL MECHANICS The terms in R contain at least one cumulant of third or higher order. For example. The reader should notice that: (1) Each cumulant can be written as the expansion in terms of the average values. That is to say the c in the subscript of ( . then (1 2 30b) The same is true for higher products. . i. . then the result like (1230b) cannot be established. Only in the case of the linear combination (1230) is it unchanged. . Yi = c aijxj I (12 30a) where aij are constants. If the variable is changed. For the same reason (1228) Each term in R' contains at least one cumulant of third or higher order.
with each group being disconnected. n ( 1 2 .'s are fixed there are still many ways of grouping. So we have to divide by the n ! . The interchange of points inside a group does not produce any change. We draw m points and divide these into groups. The Cumulant Expansion Theorem This theorem states that where fc ( k )can be called the cumulant characteristic function of p ( X ) .) This result is easily understood.4. etc. The linked points of a group is a cumulant. Let m. then m l groups have 1 point. The m points have m ! permutations. and connect all the points within a graph. (Notice that if mi = 0. . namely m! (1235) ways.34) The m. 121. then O! is defined as 1 . The proof of this theorem is as follows: We first expand the characteristic function: ( 1 233) Then we write ( X m ) as the linear combination of the cumulants as in Fig. be the number of groups with n points.SUM OF MANY INDEPENDENT VARIABLES 207 12. If the m. mz groups have 2 points. The sum of the total groups is m = 1nm. can be zero or a positive integer.
. . .) m ... (1236) Sum over various m.nmn. .. . m. . . (1237) Hence (1232) is proved. . This is an extremely important result. ! . . .3m.. . . .208 STATISTICAL MECHANICS Interchange of the groups with the same number of points also produces no change. (yl(y" .2m2 ..m . m 2 . . . . . . So we divide by ?n. .1. = exp(f. Hence . m. . nm. . Substitute (1236) into (1233) and we get = I(&)" c m 6(mm. ! .. . . subject to the condition (1234). .) ....(k)} . 6(m . m ! m2 ! .. .
except for the second cumulant. (122)) can be compared with (1227) and (1228).X . This is the essence of the normal distribution. n = <w.SUM OF MANY INDEPENDENT VARIABLES 209 The above theorem can be generalised to the expansion of many variables: (1 2 38) This is left as an exercise for the readeI 12. (1 2 40) That is to say. we can use (122) and (1232) to get 1 f J k ) = .5.. . In addition. Central Limit Theorem (General Exposition) Now let us examine the cumulants of the normal distribution. suppose n > 4 . . ( Y 2 ) c = u2 . . 0 = 3. The following derivation is to clarify the conditions of the theorem and its flexibility. .e.5.4.k 2 u2 2 . (12 41) . Let XI . .. The various average values (see Eq. Now let us return to the central limit theorem. all the other cumulants are zero. (12 39) i. .X N be N variables and N is sufficiently large.
Now let us carefully examine its implications. u’ . . Except for terms of O( 1 I n ) .e. . . Therefore the distribution of Y can be regarded as normal if we only limit our calculation to average values of low powers of Y. hence the average values of Y can be calculated from the normal distribution. a. are all O( 1).. When the number of terms is close to then O ( l / n ) terms cannot be neglected.g. .4. = 3 . (b) and (c) are valid. (1) The implication of the conclusion (1243) is as follows. (12 45) Therefore. . to within (1 There is an obvious restriction. The theorem is thus proved. (iii) is the main part of this theorem. (1243) is established too.pn. We first consider ( Y 3 ) :From (1242) we get 1 (1 2 44) The cases with n > 3 are just generalisation of ( 1 2 . . X. then (1241) is immediately established. (1243) The first two results (i) and (ii) are obvious. n /n).(1240) is the same as (1243). i.4 9 e. . . then n = 3. . provided (1241) is valid.XN are mutually independent variables. 4 . .210 STATISTICAL MECHANICS Let (1 2 42) Theorem: If (a). . (2) If XI. . the high order average values cannot use this result because the cumulant expansion of ( Y ) has many terms.
. If N is very large. we usualIy need to calculate only the 4 low order average values of Y. Y?. To be more explicit. (b) and ( c ) of Eqs. According to the central limit theorem. Equation (1241) is established and the theorem can be used. Hence. to examine whether a group of variables are independent is a difficult task. otherwise. If N is large. . . for example the probability of getting ‘ I ! in throwing a die. The value l j 6 means that if we have a repeated trial o 1%’ throws. Hence if (a). What is the meaning of the term “approximately”? According to the central limit theorem. Yw. then it is indeed impossible. If we do not demand too much. 12. This theorem not only has practical value but it also has important impact on some fundamental concepts. . the theorem can be used. (1241) are established for a few n. Let us look at some examples. We do not need to consider whether X . This is why in the beginning of the chapter we said that the variables are “essentially” independent of the distribution is “essentially” normal. .SUM OF M A N Y INDEPENDENT VARIABLES 21 1 (3) I f they are not independent. This is the reason why the theorem is so useful. if the ith throw gives 1. say M times (each with N throws). then the conditions for the validity are likewise few. (1246) = 0 . . . then Y/N approaches 1/6. the . “approximately” means a percentage error o f ( l / f i ) . suppose we throw a die N times and let the number of times ‘ 1’ turns up to be Y : ni = 1 . then we get M values o f Y : I .. How does it approach the limit? We repeat the experiment on Y . we use the central limit theorem always with the practical viewpoint. ‘ 1 ’ will turn u p approximately f N / 6 times.6. and in (c) at most Nm“ terms. . then the sum in (b) involves at most Nmz terms. In fact. . but each variable is only correlated to at most rn other variables and m << N . Repeated Trials and the Determination of Probability Repeated experiments are very important in many applications ofprobability.XN are independent. . ( ) From a practical point of view.
The general definition of probability usually assumes many repeated independent experiments. Each throw of the die is regarded as an independent event. The probability of an event is defined up to an accuracy of ( l / f i ) . Y/N is closer to 1/6. . . we can deduce that the determination of probability requires this theorem. If it is not. In the absence of a large number of repetitions. The above discussion uses the central limit theorem. . this probability is just a mathematical terminology with little physical content. is the definition of probability questionable? We can use an example to illustrate this problem. Therefore. . In such cases. The central limit theorem is inapplicable in some cases.( ”1 ) Yi . . then the meaning of the value 1/6 is questionable. Let the distribution of a variable Y be 1 PiY) = T ( Y 2 + 1) (1 2 49) The mean square value ( Y2 ) of this distribution is infinite. Let ( 12 . getting Y .50) . J z (1 2 47) 36 This value can be determined by M tries of the experiment : (12 48) That is to say. where N is the number of repetitions of the trial. if N is larger. Y N . and the width of the distribution of Y/Nis u = . . Seeing this. Y . and the derivation of this theorem must have probability defined first.212 STATISTICAL MECHANICS distribution of these values is normal. this theorem and the concept of probability are intimately related. Suppose we repeat Nexperiments.
SUM OF MANY INDEPENDENT VARIABLES 213 According to the example of Sec. (1 2 52) = o . and (n: ) < 1 and is finite. or ( X 3 ) + 00. b and (1253) we can determine p ( Y ) .e. b ) that Y occurs between a and b : ni = 1 . Now n is the sum of N independent variables. it is not accurate enough to yield the average value (12 . 12. . Although p ( Y ) is determined. This is completely different from the error of order ( 1 J J N ) discussed above. i. We do N experiments and count the number of times n ( a . That is to say no matter how many trials we have performed the average value of Y cannot be determined more accurately. (1218) t o (1220)) the distribution o f Z is the same as (1249). This example tells us that if a variable X has ( X 2 ) + m . then what is the meaning of probability here? The above result (1251) tells us that we cannot use the result of repeated trials to calculate the average value of Y. ifa<q<b otherwise . However. then the meaning of ( X ) is questionable. (1251) and is surprisingly independent of N . If we cannot determine the average value of Y. repeated trials can still be used to determine p ( Y).5 4) because the probability of Y > N is O( 1/N) but its influence on (1254) is NO( 1/N) = O( 1).2 (see Eq. . . Hence the central limit theorem can be used: (12 53) From different values of a.
p. Random Motion and Diffusion Suppose a collection of particles. The values of f at different times (separated by at least 7 ) are independent.7. (1 2.3 / 2 e. such as pollens or big molecules. are moving in water and continually colliding with the water molecules.5 5 ) is the force experienced by the particle.t ’ ) . in a time interval of 1/a.> > T 1 a I (12 57) That is to say. assume it has a correlation time 7 satisfying . then C ( t ) = 0. f fluctuates many times. Let p be the momentum of a certain particle. If It1 > T . =  dt (1 2 . Because 1/a >> T . and p z .e.214 STATISTICAL MECHANICS 12. in addition. Let C(t) = 1 7( f ( t ) * f(0)) (1 2 . because f evaluated at different times is independent outside a time interval T . where a p is the frictional force. Hence we use the central limit theorem to define the distribution of p x . and f ( t ) is assumed to be random. The solution of (1255) is p(t) = 1 m d t ’ eOLt‘f ( t ..58) shows that p( t ) can be regarded as the sum of many independent variables.59) . then dP .p = / 2 o = (1 2 . i. This force naturally is rather complicated and (1255) is only an assumption. The correlation time has been explained in the last section of the preceding chapter. 7 >> l/a. where = ( 2 n 0 2 ) .58) 7 is a very long observation time. (12 56) 0 Now assume that f fluctuates rapidly so that ( f ) = 0.
Hence from (12 . (1 262') 0 Now v ( t f f is a random velocity and its time correlation function is ) (v(tr) ~ ( 0 ) = ) (v2)eat' . . It relates the coefficient of friction with the fluctuation caused by the molecular motion of the liquid.SUM OF MANY INDEPENDENT VARIABLES 215 (1260) JO The above equations are obtained from (1258) and the square of (1256). then its displacement from t to t + t is Ar = = r ( t + t ' ) . Because 1 / a >> T .dissipation theorem. Let v = p/m be the speed of a certain particle. (1261) we get 1 m ( p i ) = m T . We shall talk about it in detail later. we can also obtain the distribution of the displacement of the particle. Using the central limit theorem. the above integral is (pi)= I m d t e2at 0 1 2t d t " C(t") 2t (126 1) According to the theorem of equipartition of energy.62) dt" C(t") . (1 263) This result is from the solution of (1255): v(t' + t ) = v ( t ) eat m f(t'+ tt") . This is called the fluctuation . This is a very important result.r(t) 1 t' d t " V(t"+ t ) .
t ) is a solution of the diffusion equation: n(r.216 STATISTICAL MECHANICS Assume that v(t) is uncorrelated with f at later times. .66) where D is the coefficient of diffusion. t ) n(r’) . If in a small region d3r’ the number of particles is n ( r ’ ) d 3 r t . If t‘ in (1262’) is much larger than l/a. the distribution of Ar is normal: and u2 can be obtained from (1262) and (1263): 1 ((Ar)2) 3 = +lo 1 t‘ t” dtl dtz (v(tl)*v(tz)) 0 (1 2 65) This result can be written in the more familiar form of the diffusion equation: ( 12 . the distribution of the positions of these particles is n(r. t ) = s d3r’ p(r . If we observe many particles simultaneously. then Ar is the sum of many independent variables.r’. then after a time t.t) . The above averages are all time averages. (12 . then the average values may also be regarded as averages over these particles.67) This n(r. According to the central limit theorem. at a t ) = D V z n(r.
(686) ). f(r) = I = 1 . Fluctuation of Macroscopic Variables In the last section of Chapter 6 . Divide the system into two halves 1 and 2 . . .. where C = aE/aT is the heat capacity of the system. Is this kind of normal distribution related to the central limit theorem? Let us give an example to discuss this problem. if r liesin part 1 if r lies in part 2 . and H2 are the energies of the two parts. Here same size. . This distribution is the statistical result of the states in the region of motion. We first write A as an integral of the energy density. Let where H . The definition of KA is 2 = 0 because the two parts are of the (1271) Rearranging.SUM O F MANY INDEPENDENT VARIABLES 217 12. This variable has been analysed in Chapter 6 (see Eqs.8. we mentioned that the distribution of large value variables is always normal (see Eq. Its distribution is (1 2 . This conclusion is only applicable to large value variables (of order N ) . (1 2 .70) r(E) J2nKA where 2 is the solution of (685). Now we analyse the distribution of 2 directly from the point of view of the * molecular motion. ( 12 .72) . we get KA = T 2 C.74) . (684) and (685) and the discussion).
j 8 r( r i ( t ) + rj(t) 2 > ’ (1 2 75) where p:/2m is the kinetic energy of the particle. then < ( d r ) 4 r ’ ) ) . t ) is the energy density. .Xu(ri(t)rj(t)) 2 i. Each term in the sum is a function of the momenta and positions of the particles. For example...Pz. then s (ri.218 STATISTICAL MECHANICS and e(r.r’) do)). i..r2.rj) is the interaction energy between i and j . (e(r)e(r’)) = 0 . if this system is a collection of particles.. Each variable changes with 1.. we obtain (1 2 79) . U(ri . (1275 ’) The average here is that over the observation time: (1 276) Now assume that the energy density is uncorrelated for points very far apart. (1 2 78) The origin 0 can be taken to be in the centre of the system. Now let us calculate (2’ ) : .. If the system is homogeneous. d’r’ f(r) f ( r ‘ ) ( e ( r ) e(r’) ) . ri(t) and p i ( t ) are the position and momentum of the particle at time t . = ( 4 r . (2’ = Jd’r >.. Hence (1273) is in fact the sum of a large number of variables.e.. if l r .PN) 3 1 + .r ’ l >> . From these assumptions.rN*Pi. ( 1 277) where is the correlation length and is on a microscopic scale. .
With this assumption we do not need detailed information on the molecular motion in order to calculate thermodynamic properties. The last term is related to the surface of the system. It represents the amplitude of change of the various variables. (1280) The distribution is obtained from the viewpoint of the molecular motion and the independent assumption of [ << L . In the preceding section . It indicates that almost all the A^ (s) in the region of motion have a value equal to A with an error of 0 (l/m). (See the last section of Chapter 16. Fluctuations and an Extension of the Basic Assumption Now we can use the result of the last section to perform a further analysis of the basic assumption. Simplicity and convenience are the outstanding features of this assumption. With the entropy known. Now use the central limit theorem to obtain the distribution of A : 1 P(& = rn 0 (A^)= . The energy density correlation function can be measured from experiment. 7 and 8 we gave many examples and introduced some rules of calculation to illustrate the application of this basic assumption. This assumption is introduced for calculating the entropy. In Chapters 6. Indeed the distribution (1 270) is calculated statistically from the number of states in the region of motion. From the analysis of Chapter 5 to Chapter 7. but there is no definite quantitative conclusion. Because [ << L . 12. it indicates that the fluctuation of macroscopic variables is very small. We have repeatedly emphasised that the region of motion must be determined by the motion. the other thermodynamic quantities can be calculated. But the relation of the normal distribution function (1270) and the molecular motion is not clear. and the error is determined by the boundary of the two parts.SUM O F MANY INDEPENDENT VARIABLES 219 where V = L 3 is the volume of the system. Such methods of measurement are discussed in the last section of the preceding chapter. But the basic assumption does not specify how to calculate the magnitude of the fluctuation.) In the last section of the last chapter we have pointed out that the correlation function can be measured by scattering experiments.9. (1273) is the sum of a large number of essentially independent variables.
Let R = {s. We do not discuss in full the details of these experiments or calculations. (1272) and (12go)). and check whether the two results are the same. (1 2 . They are both normally distributed. We can also use the basic assumption to calculate the lefthand side and other methods to calculate the righthand side. } i t . But are they equal? The condition of equality is K.220 STATISTICAL MECHANICS we have discussed the relation of the fluctuation o f A with the correlation function. The distribution of the variable means a time distribution. d3r (E(r) ~ ( 0 ) ) ~ .e. all the states s with value of . be the time the trajectory stays in R . = N o Z (see Eqs. This leads us to the conclusion that fluctuation can be determined by scattering experiments. A”($) between A l and A z are included in R .80) is as follows. It does not involve the entropy or the basic assumption. The meaning of the distribution (1 2.8 1) Let 7. A ) / I ‘ ( E ) in (1270) obviously have different meanings. (1 2 . i. Al < A ( s ) < A . The p ( E ) in (1280) and the ratio r ( E . This can be regarded as a quantity determined purely from the time distribution of the trajectory. The result is that they are equal. These results indicate that (1285) . then (1283) This is purely a calculation of the volume of the region and does not involve the concept of the trajectory or the time. then the meaning of (1280) is dA p ( A ) = JR 1 : 3 ’ (1282) where 3 is the total observation time. The experiments measure an average value with respect to time. The distribution (1270) comes from the sizes of the regions in the region of motion.84) The lefthand side can be measured by thermodynamic experiments while the righthand side can be measured by scattering experiments. Let I‘(R ) be the volume of R .
than the fluctuation O(N ). We have mentioned that R must be a large region in order for gR to have meaning.The average value of A^ is O ( N ) and its fluctuation is O ( a ).Hz (see Eq.86) . we can avoid this problem. but we can also use it to calculate the fluctuation of 2. Equation (1285) can be regarded as an extension of the basic assumption. 3 has no meaning.” part while the latter is a fluctuation. This conclusion can be stated as. the part with nonzero frequency.The measurement of the average valye of A^ and the measurement of its fluctuation are by completely different methods. then the fluctuation of H is very small. because H1 . (1034) to (1035). If R is any large region. Here R is a large region. The two are completely different.H 2 . If it is open. . In spite of all these words. Notice that R is determined by classification according to A . the changing part. where A is a large variable. The fluctuation of the energy density is better analysed by H I H2. If R is too small. making a measuring inaccuracy of O ( N 2 1 3 ) . i.) Although the above discussion concerns a special example. (69) ). It essentially tells us that our trajectory is dispersed throughout the region of motion.e. Not only can we use the basic assumption to calculate entropy and the average values of the thermodynamic coordinates A^. From O ( N ) to O ( a ) is a very important step.SUM O F MANY INDEPENDENT VARIABLES 221 i. then the fluctuation is larger. “The trajectory is uniformly distributed in the region of motion” or “Any state in the region of motion has equal probability of occurring”. When we use H1 . Generally speaking. Why do we not use H1 + H z = H as the example? The reason is that H is a conserved quantity.C. We emphasise again that the fluctuation of A^ discussed here represents the change of A^ with time and has nothing to do with the inaccuracy in measuring 2. we must remember that the R in (1285) must be a large region. the general conclusion is similar. Eq. The former has an inaccuracy of O ( N 2 1 3 ) but does not indicate that the latter O ( N 1 1 2 ) not measurable.H2 is not a conserved quantity. (1285) is valid. the time that the trajectory stays in R is proportional to the volume of R . (See the discussion from Eqs. this inaccuracy in measurement comes from the surface of the system.e. The scattering experiments analyse the wavevector: ”’ Y (1 2. The former is an is average in the observation time and is an unchanged “D. The reader may already notice that the above example concerns A = HI .The number of molecules around the surface of the body is O(NZI3 This error is much larger ). The boundary between the interior and exterior of the system is not definite. . If the interface is adiabatic.
222
STATISTICAL MECHANICS
where HI  H z may be regarded as the nonzero limit as k + 0, i.e. k 1/L where L is the size of the system. If k = 0, then e0 = H . Notice that the fluctuation experiments tell us of the existence of molecules or atoms. That is t o say, matter is not continuous but is composed of discrete elements. If it is continuous, the N must be infinite, and the fluctuation is zero because O ( N ’ ” ) / N = O ( N  ’ ” ). There is fluctuation only because N i s finite. Because we have used units with k = 1 , it is easy to overlook the role of the Boltzmann constant in the various equations. The energy k T is a microscopic quantity. It is about the energy of a molecule. If matter is continuous a “molecule” would be infinitesimal, and k must be zero. The fact that k is nonzero indicates the existence of discrete elements. The equations in thermo. dynamics do not involve k . But the equations concerning fluctuation and involving T must also involve k . For example in (1 262) and (1 284)

ma =
I d f ’ C ( t ’ )
2kT
,
(1 2 87)
(1 2 88) where a,m and the correlation function are concepts not involving temperature. The combination T C , / V is a macroscopic quantity. The remaining T must be accompanied by k , otherwise the units are not consistent. The correlation function is of course a special feature of fluctuation. Problems 1. Calculate the cumulant characteristic function
of the binomial distribution
(1289) Ans.: N l n [ 1  p ( 1  e  j k ) ]
.
2. In the above problem, when N and n are very large, what is the difference between p ( n ) and the normal distribution? 3. Calculate the characteristic function of (1218) and then derive (1220).
SUM OF MANY INDEPENDENT VARIABLES
223
4, The cumulant expansion theorem .is an extremely important discovery for statistical mechanics. Let
H = Ho
+U
(a) Prove that
where ( )o is defined as an average over e  H o / T : ( 1 290)
and Fo is the thermodynamic potential for Ho . (b) From the cumulant expansion theorem prove that
F = F~  ~ ( e  u / T 
,
(1291)
i.e. the correction U on Fo is the various cumulant values of U / T . (c) Let H o be the kinetic energy of a certain gas and U t h e interaction energy, which is assumed to be short range. Prove that the expansion of F with respect t o U has terms all proportional t o the volume. This result would be hard t o prove without the cumulant expansion theorem. (Hint: Use the model in Sec. 9.1 .) Write U as
r’p(r) p(r’)~ ( r  r ’ )
(1 292)
(1293) The density of an ideal gas is uncorrelated in different places. First find ( U )oc, ( U 2 ) o c , then the cases of high order terms can be seen. 5 . Effective energy. Let the state of a system be s = (sl, s2 ) where s1 and s2 are respectively the states of the two parts of the system. Let the total energy be
224
STATISTICAL MECHANICS
If a variable A (sl ) is only dependent on sl, prove that its average value can be calculated by an effective energy H i (sl ), defined as
That is to say, H : is H1 plus a correction term. The correction is due to the interaction with the second part. If H12 is comparable to H 1 , then this correction cannot be neglected. The above steps for calculating the effective energy is again an example of classification and also an example of projection. (Review the last section of Chapter 5 and the first section of Chapter 6.) 6. Notice that in the above problem s1 in the phase space (set of s = (sl, s 2 ) ) represents a region R ( s l ) , i.e. ( s 1 fixed, all the set of s z ) . If the second part has N2 particles, where N2 is a large number, then this region is a large region with a volume of order eNz :
Let
where r is the volume of the region of motion of the whole system. Discuss the meaning of P(sl ). Can it represent the probability of occurrence of s1 ? Let N1 be the number of particles in part 1 . If N1 is not a large number, what is the order of magnitude of P ( s l ) ? If N1 is a large number, what is the order of magnitude of P(sl ) then? Notice that r = O(e N, +Nz ).
Chapter 13
CORRELATION FUNCTIONS
The previous chapters discussed the importance of the concept of independence and correlation in statistical mechanics. The main point of this chapter is on some examples of the correlation function. Hence there is a bit more mathematical manipulation. From these examples we can have a more concrete understanding of the concept of correlation. First, we briefly summarise the relation between response and fluctuation discussed in the last chapter, and define the response function. Then we give examples of calculating the correlation function and response function in some gas models, including the dilute gas, the dense gas (onedimensional model) and the quantum fermion gas. Some special properties arising from quantum mechanics are also discussed.
13.1. Response and Fluctuation
Using the central limit theorem, the analysis of the last chapter relates the distribution of the variable A in the region of motion with the correlation of its fluctuation. The relation of heat capacity with the fluctuation of the energy density is an important example. Heat capacity is a kind of susceptibility. It measures the response of the body with respect to the temperature change. Now we use a somewhat abstract procedure to derive the relation between the general response with the fluctuation. The following result further shows that the correlation value can in fact be regarded as a large value variable and its value is about the same throughout the region of motion. Suppose that the Hamiltonian of a system is H. Now we add an external force h so that the Hamiltonian becomes
A
HAi
.
(131)
226
STATISTICAL MECHANICS
For example h is the magnetic field and A^ is the total magnetic moment, or h is the tension and A^ is the length of the system. The molecules inside the system respond to this external force and changes occur. This response can be determined by the change of the various variables. For example, the value of a variable B can be calculated from the average
n
S
B = (B)
=
C e(HAA^)/T
S
(132)
B is of course a function of h . Differentiating the above equation, we get
(133) This is the socalled differential susceptibility.
Example 1 : Let A =
= H , i.e. the total energy itself. Then
(HhA)/T
H/((1 + h ) T ) )
(134)
and h represents the fractional increase of temperature (assume that h << 1). From (133) we get
aE 1 T= (If2), aT T
i.e.
,
T 2 C, =
I
d 3 r d 3 r 1(e(r) e(r')), d3r" ( e(rr') ~ ( 0 ) ) ~ (135)
This is just (1284), and E ( r ) is the energy density.
CORRELATION FUNCTIONS
227
Example 2: Let h be the magnetic field h , Then from (1 33) we get
6 =A^
be the total magnetic moment
ik.
where m ( r ) is the magnetic moment density and x i s the magnetic susceptibility. From these results it can be seen that the correlation value can be regarded as a large value variable because it can be obtained by differentiating B , which is the value of the large value variable Let us now return and ask: What is the influence of the external force X on the thermodynamic potential? The thermodynamic potential is
2.
F=TlnZ
~
= c~(HL~)/T
S
(137)
Let Zo be the value of Z when h = 0. Then
h
z
=
Z, (eAAIT )o
(138)
The average value ( . . . )o means the average when h = 0 : (139)
Now use the cumulant expansion theorem of the last chapter (see Eqs. (1232) to (1238)) to get
(1310) Substitute (1310) into l n Z , and we get
F = TlnZ,
 Tln(e”A”/T),
(1311)
228
STATISTICAL MECHANICS
Therefore,
aF
ah

(A")
,
(1312)
etc. There is no subscript 0 in ( . . . ) of (1312) because this is unnecessary as the above expansion can take any value of h as the origin. Hence, all the average values and correlation values can be related to the differential of the thermodynamic potential. From this point of view, as the basic assumption specifies the rules of calculating entropy and the thermodynamic potential, it also specifies the rules of calculating its differentials. Hence the basic assumption also specifies the rules of calculating all the average values and correlation values. The above discussion has not taken quantum mechanics into account. Hence, some results may need corrections. If H ] # 0, then (138) must be amended, because
[A,
This will be rediscussed in the last part of this chapter. At the moment we d o not consider any quantum mechanical effects.
Example 3: The response function and the correlation function.
These are only examples of (13 3). Let
(1313) Substitute in (133) and we get
=
V T
 j d 3 r " ( a ( r " ) a(O)),
8k.r"
(1 3 14)
~
CORRELATION FUNCTIONS
229
where c is the Fourier transform of the correlation function of the energy k fluctuation and Gk is called the Fourier transform of the response function. Its Fourier integral is
G(rr')
= (?.ny3 j d 3 k
G(k) ek(' i''')
(1315)
Its meaning is as follows. If h(r) represents a distribution of external force, then
(a(r)> = {d3r' G(rr') h(r')
,
(1316)
The above assumes X to be very small, and ( a ) is zero for X = 0. (This result is obtained from ( ak ) = Xk Gk.) The response function shows the relation of causality: X(r'> is the cause and ( a ( r ) ) is the effect and G(r  r ' ) shows how the cause at r' affects the response at r . It can also be called the propagation function.
Example 4: The above result naturally can be applied to the distribution in an open system. Let A^ = = N , i.e. the number of particles, and h is the chemical potential p . The result of (133) becomes
(1317) Here p ( r ) is the density of the molecules. Notice that
(1318) In the above we have made use of n = a p / a p , n
N / V = (p(r)). Therefore
(1319)
Hence the response of the density to pressure is related to the density fluctuation.
230
STATISTICAL MECHANICS
13.2. Density Correlation Function
Now let us consider the evaluation of the density correlation function. Let
This is the density distribution given that a particle is at r = 0 . Notice that (p(r) P ( 0 ) )
= (probability of a molecule at r = 0) X (the conditional probability
that there is a particle at r given that there is a molecule at 0).
=
n x n(r l 0 )
.
(1321)
Therefore, (1320) can be regarded as a conditional probability. If the system is an ideal gas, then n(r l 0) = 6(r)
+n
.
(13 22)
The first term on the right is the density of the molecules at r = 0 and the second term n is the average distribution of the other molecules. We can apply (1322) to (1319) to obtain the relation between pressure and density. From (1322) and (1320) we get ( d r ) P ( O ) ) , = n 6(r) Subsituting in (1319), we get
( 13  24)
.
(1323)
This of course is the ideal gas law p = n T. If there is an interaction u(rj  ri) between the molecules (ri, ri are the positions of the ith and jth molecules), then on putting a molecule at r = 0 the distribution of the molecules will no longer be (1322), but becomes
n(rl0) = ne'(')lT
+ 6(r)
.
(13 25)
This equation tends to n when r is very large, and (1 323) now becomes
CORRELATION FUNCTIONS
231
Equation (1 3  26) is valid only when the density of the gas is low, because (1 3  25) is the distribution of an ideal gas in a potential u(r). If the density is high, then we must take into account of the interaction with other molecules. Hence the last term of (1326) is the first correction term for low density case. Substitute (1326) into (1319) and we get
(1327) After integrating we get
p
=
nT
+ n2T
I
d 3 r [ 1  eU(T)’T 1
(1328)
This is the ideal gas law plus a correction. This correction term will be discussed in detail in the next chapter. This just offers an application of (1319). The correlation length { is defined as follows. The correlation function tends to zero at distances larger than i . From (1323) it can be seen that the correlation length of an ideal gas is zero. From (1326) we see that for a dilute gas is just the effective interaction distance of the molecular interaction u(r). Strictly speaking, different correlation functions can have different correlation lengths. For example, the density correlation function can have a correlation length different from the spin correlation function. But generally speaking, the various correlation lengths are more or less the same. The above example is a dilute gas. Let us consider an example in one dimension where the density can be high or low. This is the example in Sec. 8.3. Now let us calculate ( p ( x ) p ( 0 ) ) . Fix a particle at the origin and k t the position of the first particle on the right be y l , the position of the second be y , + y 2 , and the position of the nth particle be y1 + y 2 + . . . + y n . Therefore,
(1 3 29)
The definition of n ( x 10) is (1320). The yi are the distances between neighbouring molecules. Equation (1329) is correct for x > 0 . If x < 0 , a similar
232
STATISTICAL MECHANICS
equation can be written. From (1329)we get
C
=
k
J
dx ejkx ( p ( x ) p ( 0 ) )
( 1 3  3 0)
(1331)
where p is the pressure, { has been discussed in Sec. 8.2 (Eq. (835)) and u(y) is the interaction energy between neighbouring molecules. The above assumes that each side has N / 2 molecules. In (1330) is contributed by the molecules on the left. The above also assumes that k # 0. The sum in (1330) can be calculated if we regard N as infinite. We get
(1 3  3 2 )
Now we look at a special form of u(y), i.e. the hard sphere interaction
U(Y) =
,
Y < b
>
= o ,
Y > b ,
(1333)
where b is the diameter of the hard sphere.
k’ . because u ( y ) does not have an energy scale. From (1334) and (1332) we get ck = nk2 ( a . Therefore.  2nm a i 4 n 2 m2 ( a .3 7) In the above. (1 . ( 13 .N b = N(a . If the density is high. at low density.: k. (1335) is about (1 3 36) This equation is appropriate for k b << 1. . but the effective volume is L . This gas is just like the ideal gas.b small. no matter what the temperature is. If a >> b. k’a.b) 1’ (1335) Notice that this result is independent of the temperature. the denominator of (1335) is very small for k 2n/b where m is an integer:  N .b).e.3 4) Here we employed p ( a . then when k is not very large.ikb + ik(ab) + i 1 (k’b)’ 2 E k 2nm/b i.eikb + ik(a .a / b ) are regarded as small quantities. (1338) m = 51. . the density correlation function will not be affected. the denominator of (1335)is zero when k = k. . a . That is to say.b) l/n 9 . i. for a collection of moving hard spheres. (1 3 .e.b)’ ’ 1 1 . 52.b)’/a3 .CORRELATION FUNCTIONS 233 Substitute (1333) into (1331) and we get a F 1 eika 1 + ik(a. i.b ) = T . . + 3. .
234 STATISTICAL MECHANICS Fig. (See Fig. then (1338) is not valid.Plot of C. = 1 1 277   dk cke i k x . 131. (See . 13 . In liquids. (13 .3 9 ) is approximately a function of period a.b)’.2.) If m is very large. 1 3 . but gradually decreasing with I x I . Hence the correlation length is about since The aim of this example is to point out that the strong repulsion between the molecules.) . The imaginary part of (1338) is the rate of decrease. as a function of k . such as the hard sphere interaction can produce a roughly periodic density distribution. this phenomenon was observed experimentally long ago. 7’ Fig. The meaning of these peaks is: c(X) = ( p ( x ) p(O)).  Hence C has a group of peaks at 2 n m l a with width 4 7 m 2a3 ( a .
it changes the momentum of a particle from p to p + k.2 The density correlation function of liquid sodium at 400°C. The scattering process is as follows.P L ) / T + 1 (1 3 43) while 1 fp+k is the probability that the state p + k is empty.3. We first examine fermions. The Fermion Gas Next consider the density correlation function in a quantum gas. 13.f p + k ) (see Chapter 3) (1 3 42) The factor 2 on the right hand side comes from the two states of spin+ and fp is the average particle number in the p state.23). (1342) becomes the well known ideal gas result (1 3 . Hence. The proportionality constant in (1342) is in fact 1 because when f is very small (high temperature. low density case). given by fp = 1 e@ .CORRELATION FUNCTIONS 235 Fig. 1 3 . An external particle comes in and transfers a momentum k to the body and goes out. The simplest method in determining the density correlation function is to start from its definition: and is proportional to the scattering rate. We know that the scattering rate must be proportional to fp(l . .
then pl C(r) 2n4r2 = n&(r) .236 STATISTICAL MECHANICS (1 3 44) a p sin pr (1 3 45) The integral I ( r ) is knowna (1 3 46) Ifr<<[orT+O. aSee Landau and Lifshitz (1980) p. This is caused by the sudden cutoff of the particle distributiori at the Fermi surface. The exclusion principle causes the density to form a layered structure around the origin. [ cospFr  1 sinpFr (pFr>2 2 (1 3 47) ’ The special feature of this result is the appearance of periodic change of period 27r/2pF. 358 .
50) the factor of 2 on the right comes from the two spin states. we have assumed that >> l / p F . there is no correlation length. The meaning of G ( r ) is this: if we add a potential X6(r). I’ ( 13 49) 1 The reader can evaluate 6. This wavefunction is cpo(r1 =  1 sin(pr+60) Jv pr Pr 6o = . The change of density is hG(r) = 2 1 P fp [ Icpo(r)l2  ( 1 3 . er/€ C(r) = 2n462r2 sin’ pFr . then the change in the density is XG(r) t O(h’). p. In deriving (1348). If T > 0. Rather C(r) a l / r 4 . Quantum mechanically this is incorrect. because 6(r) only affects the wavefunction at the origin. i. (1346)). . Only the 1 = 0 part of the wavefunction is affected.h m p / 2 n . for himself. E is much larger than the distance between the molecules. This will be discussed a little later. Let the volume of the gas be a sphere centred at the region.CORRELATION FUNCTIONS 237 When T + 0 . there is a correlation length E (See Eq. Now we add this potential and calculate the density. Let us now look at the response function G ( r ) and its relation with C(r). (13 48) where eF = p i / 2 m . The effect of this potential is to change the wavefunction of the particle. which is still a rapid decrease when r is large. From (1314) we get G = C/T. At low temperatures.e. C(r) does not decrease as of e’lE. Let usdirectly examine G.. When r is very large. But this is valid only classically.
Now let us find the first term of the expansion of (132) with respect to A . although C ( r ) and C ( r ) have many properties in common.4. we get U(P) = 1 + A s P 0 dTJ(7) + O(A2) .5 Therefore. When r . and after rearrangement. Then its differential is (1355) Integrating. Therefore.5 2 ) =  2 t. the periodic change and the correlation length. pFm erJc G(r) y  . $ becomes very important. the relation between them is not trivial. We again start from (132) where A^ are operators which may not commute. (1 3 . (1351) 1 ( 2 r ) is just (1345) but we need to change r to 2 r . (1 3 53) 22r2. 1 cos(2pFr) When T > 0 .238 STATISTICAL MECHANICS Hence from (1350) and (1349). The Response and Correlation Functions in Quantum Mechanics Now let us mention the relation of response and fluctuation in quantum mechanics. (133) may not be correct. (1356) . we get G(r) = (2r) I 2 2 r2 m . When r << from (1346) we get G(r) c or T + O . e. Let i.g. 13.
. (1358) If [A". Substitute S (1356) into (1357) and we get x= = ah jyTd r [ ( A ^ ( T ) ~ ). i. an eigenstate of H . A is a conserved quantity.e. HI = 0. in classical mechanics they commute.. Let I s ) be a stationary state i. then (1358) is the same as (133). i.e. and A. = (s I A I s' >. Of course [ 2.e. then d7 ( A ( 7 ) B ) =  1 dw 2?r 0 S(w) 3 (1359) (1 3 . ) ] . is the eigenvalue of the energy.60) E.( 2 ) ( . The reader can prove ( A B ) = jd"S'(w) 2n (1361) The above result can be summarised as follows: S ( w ) is the frequency spectrum of the response function and S ' ( w ) is the frequency spectrum of the .H f T U ( l / T ) The statistical sum Z in the phase space is now written as a trace. and the conclusion following (133) is then valid. ] + 0 H a s h + 0.CORRELATION FUNCTIONS 239 Now (132) can be written as Tr e"IT B = U(l / T ) j (1357) Tr e .
Therefore. See Abrikosov er al. S ’ ( w ) is in fact the frequency distribution of the fluctuation: S’(w) = I d t eiwt ( A ” ( t ) i ) . For the same reason the differential of the thermodynamic potential cannot simply be written as the correlation value of the fluctuation as in (1312). i. Here w is the energy difference.S’(w) . i. Only at low frequencies. Let A =  A 1 JvPk’ (13 . is it true that 1 S(w) w 1 . but in practice it is not hard to perform the calculation. when w << T.e. (1362) The Hamiltonian H not only represents the total energy but also governs the motion. The action of p i is to increase the momentum of a particle by k. It looks complicated.240 STATISTICAL MECHANICS correlation function. . the relation between the response function and the correlation function is relatively complicated in quantum mechanics. and on the other hand it also represents a time scale of the motion of 2. (1963). Hence from the analysis of the response function. Now let us look at a simple example. hw/T + 0 . (13 65) bThese techniques were developed by Matsubara (1955) and by the Russian School.64) Equation (1360) can be used to calculate S(w) and S ’ ( w ) .In classical mechanics h w can be neglected. (1 3 63) where A^(t) represents the behaviour o f 2 with time.e. we can at the same time analyse the motionb. This has been discussed in the analysis of the scattering experiments more explicitly. Consider again the density fluctuation of a fermion gas.
and notice that E. and a i + k ..ep . The sum of (1360) over s is z .e. Equation (1365) implies a propagation of the state (p. Hence the Is’) in (1360) is different from I s ) in that the momentum of a particle is increased by k .f p + k = fp(lfp+k)(leW’T) ‘ (1 3 69) The reader may verify this for himself. the particle numbers in (1368) have the following relation fp . The sum in (1366) is to calculate this probability.f p + k ) .fp+k (1370) Ep . we have to sum over p and a.+ko apOin (1365). because only the momentum of a particle is changed.€ p + k .CORRELATION FUNCTIONS 241 where a = ? 1 is the spin direction and a is the annihilation operator of a PO particle (of momentum p and spin a). Let us now consider the term a. Hence (1 366) is This is just the result of one term in (1365). If the state Is) has a particle in state (p. S’ 1 . a). a) then (s’ I $+ko apo 1 s ) is 1.. otherwise it is zero. E. o is the creation operator. To calculate (1366).1 eEs’T I ( S ’ I p i 1 s ) l2 6 ( ~ + E. the average over 1 s). The probability of being 1 is fp(l . Substituting in (1360) we get (1368) Because of the &function. a) and no particle in ( p + k. = Ep+k . ) E. a) to state ( p + k. (1 3 66) i. Substitute (1368) into (1361) and (1359) and we get (1342) and the response function d3p ‘ k z 2 j = fp .
. The reader should go over these again. 2 . a particle with charge e . The reader should derive every equation for himself.1 1) of the thermodynamic potential is a series in the cumulants. Such infinite integrals are usually regarded as a mathematical problems. is the thermodynamic potential proportional to the volume and particle number? If not. The expansion (1 3 . V E . why? Notice that in many cases of calculation. scattering experiments. In the last two chapters. various kinds of infinite integrals appear. (a) Calculate the first two terms of the expansion of the model (in powers of e 2 1. We use (1345) to get I ( 2 r ) . Let the interaction in that problem be u(r) = e2/r . we have talked about the independence of the parts of a system. We only consider the case when k is very small. .e. In this case Ep .E p + k kV . (b) Is the result of ( a ) reasonable? That is. the central limit theorem. This integration is easy and we immediately get (137 1). This expansion formula had been dicussed in problem 4 of the last chapter. Problems 1. and the correlation function. aP (137 1) This result can also be obtained by integrating (1 35 1) : lim G. = { d 3 r G ( r ) k+O .242 STATISTICAL MECHANICS This integral is not easy to do. This chapter emphasises the problems of calculation. i.
3. (a) Prove that the response function G.e. Expand the results of (1329) . Let the radius of the hard sphere be a . G.CORRELATION FUNCTIONS 243 The usual practice is to “regulate” these integrals or simply discard them. Let u(r) = eO(br) . Of course. with an interaction v(r) where r is the distance between two spheres. (1373) Calculate To = v o Coo and discuss the meaning of this temperature. and C are finite. 4. The response function describes the response of a body to an external force. Now assume that there is in addition an attractive force between the hard spheres. then (1372) shows that GO. However. Therefore. the reader should derive it for himself. But there is a reason for the appearance of these integrals. 6. This approximate result is similar to the mean field approximation in Chapter 27. when this system is very sensitive to the external force and undergoes large density fluctuations. Prove that onedimensional model (1 329) . . any molecule in the body can be regarded as an external agent on the other molecules. Let GO. 13. Suppose we have a hard sphere gas. see the example in Sec. and v > 0. the response function can also be used to analyse the structure of the body itself. i. The following is an example. and use this to dicuss the low density approximation. be the density response function COk = G o k / T . This is of course only a crude approximation. An infinite integral is not zero and cannot just be discarded./T and Cok do not change with temperature (a property of the hard sphere gas.2). Hence no phase transition can occur. Notice that Go. 13. Usually these seemingly unreasonable result will bring out new discoveries. The last section on the quantum mechanical calculation can be found in some books on solid state. (b) If v k C o k tends to 1.2. Hence. and COk is the density correlation function discussed in Sec.( 1332) does not exhibit instability. 5 . This can be regarded as the initial stage of a phase transition. unreasonable results must be understood. becomes very large. It is not a stable condition.(1332) in terms of the density. Equation (1 326) can be used to calculate Coo. = Co. . or it is due to the use of a certain approximation. Hence. where v k is the Fourier transform of ~ ( 7 ) . is approximately. Either the model is not reasonable or there is problem with the limiting process.
.
beyond the machinery of present day mathematics. In this chapter. which cannot be neglected even at large distances. We have to resort to some highly simplified models as well as various approximations. Most of the previous examples concern the ideal gas. that is. Chapter 18 discusses impurities. Chapter 15 gives a simple review of phase equilibrium. . leading to many interesting phenomena. Chapter 17 introduces the king model. Chapter 19 discusses the electrostatic interaction. we regard the interaction between molecules as a small quantity and make an expansion in this quantity to first order. we emphasise the influence of the interaction and the collision time on the thermodynamic potential. but the examples below focus upon the interaction between molecules and introduce the phenomena of magnetism. belongs to the realm of elementary statistical mechanics. possibly the most remarkable consequence of mutual interaction. The analysis of mutual interaction is the main theme of modern statistical mechanics. emphasising especially the difference between mobile and frozen impurities. this is the simplest model which can be used to analyse different phenomena. Chapter 16 reviews the different types of magnetism. and is a very important tool.PART IV APPLICATIONS These six chapters discuss some slightly more complicated examples. impurities and phase equilibrium. without interactions. while the ideal gas model. and provide a crude analysis of a number of wellknown phenomena. Chapter 14 discusses the corrections to the ideal gas law. The inclusion of interactions is a highly complicated problem.
Intermolecular Interactions The interaction between molecules naturally depends on the structure of the molecules. The interaction energy is proportional to r . but we neglect the duration of collision and other details. and describe in greater depth the role played by collisions. the effect of the interaction is more than this. the coefficient B of T ( N / V ) ' .e. averaging to zero.e.Chapter 14 CORRECTIONS TO THE IDEAL GAS LAW When we discuss the ideal gas.the dipole changes rapidly. the electrons will mutually interact. a strong repulsion is produced because of the fermionic character of the electrons. If there are two atoms. For simplicity. We only discuss the first correction term in the low density limit. This is mainly due to the electric dipole interaction. but this is only a time averaged property . the socalled second virial coefficient. 14. Then we introduce the concept of time delay in collisions.6 . We also touch upon some elementary concepts in the theory of scattering. In this chapter we shall discuss this and its influence on the ideal gas law p = TN/V. However. An atom alone has no electric dipole (i. the centreofmass of the electrons coincides with the nucleus). we regard the interaction between molecules as causing collisions. As a result. we shall only consider the interaction between simple molecules. Let us first review the interaction energy between molecules and point out its relation with B . i. This r6 interaction can be understood as follows: . If the two molecules are very far apart. collisions only lead to changes of the momenta of the molecules. there is a weak attractive force.1. where r is the distance between the atoms or molecules. If two molecules are so close that their electronic shells touch.
d/r3 a . 14. then E will distort its orbit and produce a dipole moment d’aE . with a minimum .1) If we place another atom at r. (14.d2/r6 .1. The shortdistance repulsion and the longdistance attraction result in an interaction potential as shown in Fig. Some examples are listed in Table 141. So these two atoms produce an interaction energy (142) U(r) = . E a d‘.CORRECTIONS TO THE IDEAL GAS LAW 241 Consider an atom at the origin.E at r = a . aor LenardJones potential. (143) Although d and d’ each has a zero average. The “61 2 potential”: (144) is commonly used as an approximation. and this is the reason for the r6 attractive potential. Fig. . 14. Even this crude interaction model has extensive applications. solids and liquids can be explained quite well by this model.1 LenardJones potential. Many properties of gases. The electric field at r is E a d/r3 . It possesses a fast rotating dipole d . the average of d 2 is nonzero.d ‘ .
r z I ) / T ) (148) . + z l ( z 2 . (146) z e''T ( 2 n m T)3'2 = ideal gas density . We only discuss the (N/V)' term.2. (1979) p.z:) 1 2 .2 1x 4 2.U ( I r l . (147) z1=v. 29 14.z: 2 v .7 4. Corrections to the Ideal Gas Law Now let us consider the corrections to the ideal gas law p = (N/V)T. (93) to (97)). In Chapter 9 we have the expansion (see Eqs. We can start from the method of open systems (grand canonical ensemble). we keep up to order z2 : p = 1nZ T V .248 STATISTICAL MECHANICS Table 141 Minimum interaction energy and its distance He H 2 2. Here.5 Ar 15 13 N2 COZ 40 Tabor.7 3.2 3. N = v 1 [.zl + 2z2(z2 . (145) 1nZ = zz. r2 e x p ( .I .
and after eliminating z 2 . i. As seen from above.we obtain (149) (1410) Therefore. the expansion is a series in (14. 14.11) The low density limit applies when there is seldom any particle within the interaction range. the interaction is seldom effective. B N 2T 7 a3 3 +. The value of the integral (1410) is approximately equal to b 3 . Because U ( r ) consists of a strongly repulsive core and a weakly attractive tail (see Fig.1 2T Irn a U(r)d3r (1412) Substituting into (149) we obtain P=T(1 N V +$)y() N Nv 2 (1413) .CORRECTIONS TO THE IDEAL GAS LAW 249 Rearranging the above equations.e. to get B we only need to know the pairwise interaction of the particles.where b is the effective range of the potential U(r).l).
Equation (149) can be directly obtained from the virial theorem. which reduces the available volume for the molecules to ( V .14) P2 P2 . (1415) .10) is directly proportional to the “time delay” in a collision. As in our earlier discussion on detailed balance. Density expansion is usually called virial expansion. The following analysis shows that B in Eq. To say that a gas is ideal does not imply that collisions never occur. Now we can look at the problem from a different viewpoint. d3rl d3pz d3r2 = d3R d 3 P d 3 r d 3 p . (See Problem 2. (14lo). it gives a good knowledge on the interaction energy of Fig. it is pulled back by intermolecular attraction. d’p. We shall encounter it later when we discuss phase transition. Time Delay in Collisions The above analysis is quite simple. 14. This is left as an exercise. we emphasise the importance of collisions. and (2) due to attraction. but that the duration of collision is extremely short.) We can also obtain B from measurements of the pressure of the gaseous state. (14. (14.Aho). the time during which each molecule can move freely is influenced by the interaction.3.. This is more complicated.250 STATISTICAL MECHANICS So the correction to the ideal gas law can be seen as (1) due to repulsion. So the corrections to the ideal gas law is caused by the finite collision time. 14. That is to say. The latter arises because as a molecule approaches the wall. which reduces the pressure. j 2h6 d3pz d3rl d3rz eHfT . but exhibits more clearly the role of the motion of the molecules.+ + 2M 2m’ V(r) . Together with Eq.1 . so that the collision with the wall is less violent. and B is the socalled second virial coefficient. Equation (1413) is the famous Van der Waals equation. We first rewrite z z z2as follows: e2P/T z2Z2 = d 3 p .
2 n ~ f T ) ~ ”X ( V h3 j e2P/T V l d e g ( e ) e‘IT . (14. p are the relative position and momentum in the centreofmass system respectively. and as the integrals over d 3 R and d 3 P do not involve U ( r ) . also include the angular momentum (see Fig. = . besides the energy E. . which. p = P l +P2 2 P = f(P1 . and m’ = m / 2 .15) into (1414).P z ) 9 3 R = L r1 ( fr2) r = r l r2 . r. and it can be directly computed from the knowledge of the trajectories. 142) 1 = ( r x p)/h . M = 2m is the total mass. Vg(e) = 1 P2 1 d 3 r d3p 6 (e h 2m’ (1417) In this equation. Substitute (14. 142 The trajectory of the particle in the centreofmass system. they can be performed immediately to give 222.CORRECTIONS TO THE IDEAL GAS LAW 251 where P is the total momentum of the two particles. g ( e ) is obviously the energy distribution of the particle in its trajectory. Trajectories can be classified by their constants of motion. (1418) Fig.16) We assume that U ( r ) is independent of the direction of r .
then 222. (1422) Therefore. t ’ ( e . However.e. . This is the time from i to fminus that from i to h as illustrated in Fig. = 2: 1 2 .19) becomes Vg(e) = J2T1 8 n 3h 1 d l 4m j d t (1421) If U(r) = 0. I ) as depicted in Fig.. the outgoing particle emerges earlier than the case of no collision. (1 423) ~ ( e I . = v. putting(l421) into (1417) we get X nh 1 1 :  I dl im 0 d e t’(e. 142.3 would be positive. = d r / d t . 1 4 . (1420) which is the time differential.. Because de/dp. i. and (14. The time delay of this trajectory in the figure is obviously negative.17) can now be written as (1419) where pr is the radial component of p and the integration d directions. we have dPr dr de dt is over all incoming = dr dr = dt . I ) e‘IT .) = J d t (trajectory) = time delay I d t (u=o trajectory) (1424) .252 STATISTICAL MECHANICS Equation (14.
CORRECTIONS TO THE IDEAL GAS LAW 253 Fig. while information during the interaction is not essential. . the above analysis shows that its calculation is not related to the shortdistance details. A bound state can then be treated as a free particle. i. The details of the trajectory within the circle are not required at all.e. where r is very large. (14 25) h = h/(2nmT)' This result shows that B is related to the trajectory only through the time delay. and the information outside the effective range is sufficient. This time delay can be defined from the free moving parts of the trajectory. l ) e'IT . (1410) it is directly linked to the interaction U ( r ) . In Fig. Ma and Bernstein (19697. From Sec. To conclude. 142 we only need to measure the times the particle enters and leaves the circle. What is important is the $ ( r ) in the free region. 142 except that the interaction is attractive and the trajectory of particle 2 is not shown. to calculate B we only need information on the incoming and outgoing particle. Dashen and Ma (1971). Although B is a concept in statistical mechanics. the above analysis shows that B = 23'2 nh 1 d l 1 {de t ' ( e . . and from Eq. It can be directly expressed in terms of quantities before and after the interaction. where details of the wavefunction $ ( r ) in the small r region is not essential.the coefficient B is not necessarily connected with the details of the interaction between the molecules. The above analysis can be further generalisedb Dashen. 143 Same as Fig. 143 to the present analysis we notice an essential point . This is more clearly demonstrated in the quantum mechanical case.
= nn . One is concerned with the bosonic or fermionic character of the particles. i. This is not very difficult and we shall not discuss it. (1 4 29) En[ = . Light . Its meaning can be simply explained as follows. The potential U ( r ) is nonzero only in the neighbourhood of the origin.e. as discussed below.254 STATISTICAL MECHANICS 14. 6. The most convenient analysis is to say that the mirror or the lens creates an image. 2m k2 In these equations.4. Quantum Mechanical Calculation To calculate B quantum mechanically we follow steps similar to those above. which is far away from the origin. (1 428) The value of k is determined by the condition @ ( L) = 0 . We restrict r to be within a sphere of radius L . is the quantum mechanical phase shift. In quantum mechanics. Mirrors and lenses cause light from a source to be reflected or refracted. We are familiar with the concept of “image” in elementary optics. There are two modifications. This is an interesting problem. Outside the effective range of U ( r ) . kL 177 ++ 2 6. n =integer .. vanishes on the boundary r = L . the distribution function g( E ) for the energy E can be written as (1426) where are the eigenvalues of the wave equation 1 (gv2 + with the boundary condition that &. The second is to calculate g ( e ) in (1417) quantum mechanically without using the short wavelength approximation in (1417).
If U ( r ) # O . the collision is to add a phase of 261 to the outgoing wave. then 6. t I 1 * r 1 Ik Fig. Once the position of the image is determined. ~ ( r = . The first term on the right is the incoming wave and the (. 144 Envelope a ( r ) as a function of r . the outgoing wave is altered and (1428) becomes +ln(') a (I) I e ikr + e z i S l eikr (1431) The second term in (1430) is now multiplied by a factor e Z i S 1So the result of . (1433) The envelope a ( r ) is much wider than l/k(see Fig. 144).(r) can be written as the sum of the incoming and outgoing waves and (1428) becomes (14 .30) The last term is the outgoing wave. the particle is not influenced by any interaction. To see how the wave is scattered. The phase shift in scattering is analogous.e. the whole problem is solved. i.CORRECTIONS TO THE IDEAL GAS LAW 255 then appears to radiate from the image.dq ) 2n 'J eiqr A(q) . which we need not worry about at present. We then consider how this wave moves outwards: (1432) In the above we consider only the outgoing wave and put back the time factor. If U ( r )= 0. In this case Gr. A ( 4 ) = l d r eiqr a ( r ) .1f factor is connected with the geometry ofthe3dimensional space. we superpose waves with momenta near k . = 0 . . resulting in a spherical wave packet of finite thickness.
We want to find g(e)go(c) where go(€)is the energy distribution when U ( r ) = 0 . (1 4 ./ak ro = . its Fourier transform A ( q ) is very narrow. we can write (1432) as v E k/m’ . Therefore from (1429) and (1437) we get (1438) From this we get (14 .(k+q) = 61 6 l ( k ) + 4T& 9 (1 4 . The introduction of U ( r )is to produce an “image” at ro .39) ‘This result appears in Bethe and Uhlenbeck (1937). Let us return to (1429) and (1426). . and the exponent in (1432) can be expanded: 6. the position of the “image” is at r = y o . From (1429) we get (1437) where d n / d c indicates the number of states with angular momentum I within an energy interval d e . That is t o say.36) The interpretation of (1435) is very clear: the outgoing wave moves outwards with the velocity v and its source is situated at y o .34) Using the approximation in (1434).256 STATISTICAL MECHANICS and because of this. 2a6. If 6. = 0 the source is at the origin.
ro /v = t’(e. Z must be odd and the whole expression must be multiplied by 3 . (1442) The pressure of this A 2 gas is h2 = hl(2vMT)‘ = h/fi . (1440) Since ro is the displacement of the image. this is a chemical reaction A + A . then 1 must be even.5. If the particles are bosons. P2 2M w .CORRELATIONS TO THE IDEAL GAS LAW 257 which is the second virial coefficient calculated quantum mechanically. From (1436) we get = . 14. Now we can compare (1439) with (1425).A . .l) . t’ is just the time delay. For s p i n 3 fermions. the outgoing wave is delayed by t’. If the two particles can combine to form a new molecule. when the spins are parallel. its mass is M = 2 m and the total energy is the kinetic energy minus the binding energy W : . we must consider the spin of the particles first. If the particles are fermions. The sum of the expression for parallel and antiparallel spins gives the coefficient B . (1425) and (1439) are exactly the same. then I must be even. When the two spins are antiparallel. Bound State and the Levinson Theorem The above analysis does not include the bound states of two particles. (1443) . (1441) The chemical potential of A 2 is 2 p . This formula has not yet taken into account the statistics of the particles. That is to say. Hence except for changing 1 to Z + + and replacing the integration by summation.
from the viewpoint of statistics the meaning of this theorem is very obvious. (1446) which is just Eq. we need two A’s. So the reaction (1441) reduces the total pressure by p 2 (1 4 44) Comparing with (149). (1438) with the bound state la added. Let g l ( e ) be the energy distribution of different states with angular momentum 1 then .go[(€) = (21+ 1) [ 1 6 ( ~ + W l a ) a + L3] 71 de . This theorem. then the delta function term 6 ( e + W l a ) does not exist. Equation (1425) also does not include the effect of the bound state. relating the results of scattering experiments. Of course. By the way we mention here an interesting theorem in the theory of scattering.258 STATISTICAL MECHANICS To make one A 2 . The explanation follows below. Levinson’s theoremd (1447) Here nl is the number of bound states of angular momentum I . and must be corrected. i.e. 1 21 .e. til. (1448) dSee books on quantum mechanics. . i. But. to the bound states. which is shown to be related to the difference between the low energy and high energy phase shifts.go(€) = I =O ( 2 1 + 1) + Wla) + . if a certain I does not possess a bound state. we see that the effect of the bound state A z gives (1 445) Combining the effects of collision and bound state we can write 00 g(e) . seems very profound.
. Fi is the force acting on it.50) From (1449) and (1450) we immediately get (1447).) Problems 1. K is the total kinetic energy. ( . r>a . F i ) + 2K N = 0 . (The reader should be aware that although (1449) is true in general. it may not hold when terms with different 1 values are summed over. (1454) Prove: if a + 0.e. So (1449) The integral on the right of (1448) is (1 4 . 2. even if U ( r ) is very large for r < a . But the volume of the phase space with total angular momentum I should not be affected by the interaction U ( r ) . The volume of a gas is V and the total energy is (1452) (a) Use the virial theorem t o prove: If U = 0 . . p V is still approximately equal to ( 2 / 3 ) K . then 2 p V = K 3 . which only shifts the energy of different states and modifies the wavefunctions. U(r) = 0 . ) is the longtime average and there are N particles. without altering the total number of states. Prove the virial theorem starting from Newton’s law 1 ( r i . . since the series is not guaranteed to converge. (1453) (b) Let a be the effective range of U .CORRECTIONS TO THE IDEAL GAS LAW 259 is the total number of states with angular momentum 1 . (145 1) where ri is the position of the ith particle. i.
g. (a) Obtain the change of specific heat due to the impurities. . and only the volume occupied by the impurities is needed.a >> 1 is very easy. A metal contains some fixed impurity atoms. which the reader can consult. Fi ). The case k. This aspect is discussed in many books. use the correlation function in the previous chapter to obtain T. The expansion can be started from Problem 4 of Chapter 12. Brout and Carruthers ( 1963). derive the second virial coefficient B : pV = NT(I I.260 STATISTICAL MECHANICS (c) If a = N a 3 / V << 1. (Hint: specific heat is only related to g(E). Consider the two special uses of k. evaluate the result of (a).. I 3. (1455) Starting from (145 l ) .B(:) I. ( a 2 ) ) . Let the scattering phase shift of the electron be 6. where r is the distance between the electron and an impurity atom.a >> 1 where k . ( r i . assuming that the density of the impurity atoms is very Iow. The interaction potential between each electron and the impurity atom is V ( r ) .. e. is the momentum of the electrons at the Fermi surface.a << 1 and k. 4. We need not consider 6. Chapter 1.) (b) If the impurity atoms are hard spheres with radius a . The calculation of higher virial coefficients is relatively complicated.
g. in some special situations.Chapter 15 PHASE EQUILIBRIUM Generally speaking.) However. To minimize E . or onehalf of the spins of the magnet can point upwards and onehalf downwards. the interatomic distance is close to a (Fig. if there is no external magnetic field. while on the other hand. For a magnet.g.. For example. its density. (We disregard the small thermal fluctuations. pressure and magnetic field) the properties of a substance (e. 151). This phenomenon is called phase equilibrium. When the temperature is low the minimum of F = ETS is determined by E .l). its magnetic moment is not definite and can point in any direction. and is also one of the most fascinating topics in modern statistical mechanics. At high temperatures. some properties may not have a unique value and may take on two o r more different values. the system assumes the crystalline form.. entropy and magnetic moment) have unique values. F is dominated by . 15 . and the system is in the gas phase because S is large for the gaseous state.T S . at fixed temperature. Liquid appears for intermediate temperatures. Gaseous and Liquid Phases The interaction between atoms is a strong repulsion at short distances and a weak attraction at large distances (Fig. at 100°C and one atmospheric pressure. the position is not fixed and . Thus each atom is like a hard sphere slightly attracting other atoms. water can be in the gas phase (low density) or the liquid phase (high density). 1 5. Under these special situations. but extremely difficult problem. This chapter merely introduces some basic concepts and wellknown facts about the topic.1 . under fixed conditions (e. This is a quite general. vapour and water can coexist. On one hand.
The details depend on the structure and other properties of individual atoms or molecules. water in a container with vapour above.there is crystalline arrangement on a short time scale. etc. e. the substance is not homogeneous because the densities of water and vapour are different. so is the vapour phase. further scrutiny reveals problems which require some explanation. rotation and vibration. Above this point there is no distinction between the gas and the liquid phases. Equation (151) defines a curve on the ( T .g. The end point of the line separating the gas phase and the liquid phase is called the critical point.. p ) must satisfy (152) . Only points on this curve satisfy condition (151). if there are three phases in equilibrium. 15 . The detailed motion of the atoms in a liquid is very complicated . This curve is the demarcation between phase 1 and phase 2 (see Fig. As molecules can move freely from one system to the other. 152). and yz seem obvious. Now we discuss some elementary facts about the coexistence of two phases.p ) plane. However. S is not small. which will not be discussed here. We shall discuss some simple conclusions from (151) and defer the complex ones t o the end of this chapter. But the water phase is homogeneous. then ( T . When two phases coexist. as well as sliding. the chemical potential of water and that of the vapour must be the same to be in equilibrium : (151) Initially. These two phases can then be treated as two separate bodies under fixed temperature and pressure. the meanings of y.262 STATISTICAL MECHANICS Fig.1 Interaction potential between atoms. For the same reason.
s 2 ) . Therefore Since p1 . z .p z = 0 and p1. we require not only the energy .e.z + p v l .PHASE EQUILIBRIUM 263 Fig. the triple point..e 2 . The gas phase is called the vapour. so i.ul(T. Differentiating. p ) = y 2 ( T .2= . Therefore the difference in entropy of the two phases is directly measurable. The equilibrium pressure is then called the saturated vapour pressure. but we also have to do work p ( v 1 . Now let us review some terms and basic knowledge.e. B. Let To niove a molecule from phase 2 to phase 1. p ) can be regarded as the intersection of the three transition lines. p ) . i. A. 4 = T(sl ..Tsl.v 2 ) against pressure. Equation of the transition line Along the transition line. The latent heat can be measured directly. Latent Heat The latent heat 4 is the energy required to move a molecule from one phase to another. we get . . 152 ( T .
This equation is very important._ . (159) can be immediately solved and we get the saturated vapour pressure of the solid.s d T + v d p . by rearranging we get dT . Hence if el .e. Let 1 be the gas phase and 2 the solid phase.264 STATISTICAL MECHANICS d p l = d p 2 . N . P e .v2) x €1 e2 +Tx €1  ~ 2 .€ I ) / T . v1 . the transition line is called the vapour pressure curve. d .v2 x v1 . If one of the two phases is a gas. the pressure is also low. t . (1510) where el . . Therefore.e2 does not change with T . because q .. So we can treat the vapour as an ideal gas. and v 2 are directly measurable and the whole transition line can be calculated by integration.4 P dT Now 4 = €1 T2 €2 (159) + P ( V 1 . This is the equation for the transition line of phases 1 and 2.P. (1511) This is as expected. The pressure outside the solid is very low and T is negligible. This serves as a simple application.a Now we use (157) to examine the transition line between solid and gas at extremely low temperatures. When the temperature is low. .~ 2 (156) (157) where q is introduced from (155). aEquation (157) is also called the ClausiusClapeyron equation. The density of the solid is much greater than that of the vapour. s1 Since + v1 d p = d T t v2 d p d p = . (158) Since pvl = T.( ~.e2 is the energy required to move one molecule out of the solid. i.
evaporation rate = a: P JGE . further heating produces boiling. 15. the external pressure. Of course. If the number of molecules in the boundary is much smaller than the number in the liquid and the gas. Some of the molecules will rebound from the surface. with 0 < a: < 1. Suppose the vapour is so dilute that it can be regarded as an ideal gas.. Rate of evaporation Let us estimate how many molecules per second evaporate from a unit area of the liquid surface. usually one atmosphere).e. and p = nT is the saturated vapour pressure. The pressure in the bubble is the saturated vapour pressure. Therefore when the saturated vapour pressure equals the liquid pressure (i. more molecules will leave the liquid.12). After a while all the liquid will evaporate. The Growth of Water Droplets The above discussion focuses on the coexistence of two phases and does not consider the boundary between the two phases. (1513) The value of a depends on the details of the motion and the structure of the liquid surface. then . then the number of molecules hitting the liquid surface per unit time per unit area is where n is the density of the gas and 7. Let the liquid be in equilibrium with its vapour. is the average velocity normal to the liquid surface. and thus the actual number of molecules entering the liquid is only a fraction a: of (1 5 . Evaporation is different from boiling. Then the evaporation rate is equal to the rate at which the gas molecules enter the liquid. (1513) is still correct because the evaporation rate is not strongly dependent on the number of molecules entering. We only need to take T as the temperature of the liquid. and the saturated vapour pressure may be regarded as a known function of temperature.PHASE EQUILIBRIUM 265 C.2. The parameter a! may be called the “sticking coefficient”. If a bubble appears in the liquid and can maintain itself from collapse. Now even if the vapour is not saturated. etc. this temperature is called the boiling point. if the vapour is not saturated. As the number entering is equal t o that leaving.
the molecules near the surface strongly influence the properties of the droplet. However. i. Therefore g = p ’ . the minimum of A occurs when aAjalv = o (1515) i. If N >> N213.I 6) where we have neglected terms smaller than N 2 ’ 3 .266 STATISTICAL MECHANICS it is indeed right to neglect them. p = a G j a N . The water droplet is not only subjected to constant temperature. Now G must be equal to y ‘ N . constant pressure.. where p’(T. ..13). we can neglect the N 2 I 3 term. (Note: this (Y is different from that in (1 5 . sometimes the number is not negligible. the chemical potential of the vapour.e. The growth of a water droplet is the first step of condensation. In equilibrium. These molecules have no neighboring molecules on one side and consequently experience less attractive force. (1517) where n’ is the density of the liquid and r is the radius of the water droplet. the boundary can be neglected. The surtace . and therefore have smaller negative energy. but also to an open environment. frost and fog. i.provided the surface is sufficiently flat. The number of molecules on the surface is proportional to N 2 1 3 .e. p is the chemical potential of the environment. The energy increases with surface area and 17 is proportional to the surface tension a . ( I 5 . dew. p ) is the chemical potential of the liquid. This is a major problem and is unavoidable in the discussion of the formation of rain.. and G is the Gibbs’ free energy of the water droplet. Therefore. The second term comes from the surface tension of the liquid.) 17N2I3 = a4nr’.e. We define the thermodynamical potential A as Here N is the number of water molecules inside the droplet. If N is a large number. We introduce the basic concepts of the growth of water droplets here. then G ( N ) should be G(N)=Ng+N213++. For a water droplet.. Notice that we have not yet given a definition of the water droplet. Meteorologists have done much research on this topic.
A ( p ) will drop (see Fig. A ( N ) first increases and then decreases. Recall that p is the chemical potential of the vapour around the water droplet. ( I S 21) PP’ = Tln(p/p’) TIns . and p’ is the chemical potential of the liquid at the same temperature and pressure. N is defined within t f i Note that fi< 3. The last term always increases with N .) Substituting (1516) in (1514) we get N2/ A ( N ) N (p’ . the saturated vapour pressure. 153). we call the vapour unsaturated. s < 1. but the behavior of ( p ’ . If s > 1 . p’ < p . Note that p’ is the vapour pressure when water and vapour are in equilibrium. (Generally speaking.e.p ) N is determined by whether p‘ or p is larger. where s = p i p ’ is called the degree of saturation. (1518) Figure 1513 shows A ( N ) . then p = eflp  Z x3 (1519) ( 1 s 20) The subscript Q in (1520) denotes the internal states of the water molecule. In supersaturation. In (1516) the neglected term has no great significance because the value of N has only an approximate meaning.PHASE EQUILIBRIUM 267 of the water droplet is not a definite geometric shape.e. N will increase and .p ) N + N2l377 . Ifthc pressure p around the water droplet is smaller than p’.. we call it supersaturated. If we treat the vapour as an ideal gas. . When N is large enough.~n(z/h’)] . i. then A ( N ) always increases with N . As N will change in direction to lower A ( N ) . From (1519) we get p = T[lnpIn(z/h3)] p’ = T[Inp’ . If p’ < p . If p‘ < 11. Which molecules are inside and which are outside the water droplet is not welldefined and thus the socalled surface of the water droplet is only an approximate boundary. i. Both p and p’ can be expressed in terms of the pressure p of the vapour and the saturated vapour pressure p ‘ .
refers only to pure water vapour. If there are impurities in the vapour (e. Many interesting problems such as the motion of fog. is related to the curve in Fig. If the vapour is unsaturated. Therefore. dew.E . N will not increase and the water droplet will disappear. the latent heat between solid and gas.268 STATISTICAL MECHANICS Fig. if there is no water droplet with volume greater than the critical volume. condensation can occur very quickly for s slightly greater than 1.     15.g.). dust. its effect on N. Experimentally.3. 153 Water droplet thermodynamical potential A as a function of number of molecules N . it is bound to disappear. etc. the heat of sublimation. Although this critical volume is not large. Latent Heat and the Interaction Energy of Molecules As mentioned in Sec. 300 the water droplet is then of radius 13 A. (1522) Since the degree of saturation s appears in the formula as Ins. 15. So. the vapour will not condense to water. The reader can consult books on meteorology for discussions on this topic. s 250% and N. Impurities form condensation nuclei and water droplets start to grow on the particles. the most important point is that in a supersaturated vapour. is slight. are all in the domain of meteorology. the water droplet will grow indefinitely.. and various charged particles such as ions. 151. if the volume of the water droplet is smaller than the critical volume N. much. of course. even supersaturated vapour with s 500% will not condense. The interaction between a pair of atoms is about . An increase of s by several times will not affect N. is approximately the same as the energy required to move an atom out of the solid (see (1510)). i. Of course. For T 300 K. of course. 153).1. This. . snow and raindrops and the techniques of inducing artificial rain. The value of N. frost.e. can be obtained by differentiation. This energy. in practice it is not easy to make 300 molecules come together by collision. In the solid phase the interatomic distance is approximately equal to a . water droplets. (Fig.
and there is a large zeropoint vibration. So the heat of sublimation is 1 q = nE 2 (1523) The factor of f arises because N atoms can have only fN n pairs of neighbors. The socalled experimental values in the table are actually the sum of the heat of melting and the heat of evaporation.5 Kr 10. should differ only slightly from the heat of sublimation. . n = 12. Its interatomic distance in a crystal is not near to that of the value of Fig. Helium is an exception because helium is very light. So one may expect that the heat of melting is much Table 151 Heat of sublimation q ( 1O3 J/mole) Cryst a1 Experimental value 0.7 5.9 9. The density of a crystal is very different from that of a gas. 151. q = 6 ~ Table 151 lists the heat of sublimation and the values of . 1952. but is shifted to a larger value. but is nearly the same as that of a liquid. 151 is the most reliable. being the energy required to change a solid to a liquid and then to a gas.PHASE EQUILIBRIUM 269 to move an atom out of the solid. some of these types of crystals.33 E ( J ) x 10” He Ne Ar 0.0 25 Tabor (1979). i. The numbers in Table 151 are approximate. p.3 1.0 4. The heat of sublimation is not easily obtained by direct measurement.e. The heat ofmelting plus the heat of evaporation.8 16.3 1. nevertheless they show the relation between the intermolecular interaction and the heat of sublimation. All the molecules in Table 151 are spherical for which the interaction energy in Fig.9 x lo” 1. For f c c (facecentered cube) and h c p (hexagonal closepacked) crystals. we must break up y1 pairs of atoms if we assume each atom has n neighbors.08 Calculation (1 523) 0. The value of E can be calculated or deduced from the properties of the gas (see Chapter 15).
The melting point is related to the pressure and can be determined by the transition curve (1 57). the interaction energy curve in Fig.270 STATISTICAL MECHANICS less than that of sublimation.. This is indeed the case. from very simple assumptions. and the heat of evaporation is comparable to the heat of sublimation. 151 is not very helpful for understanding melting. However. If the volume of the solid is less than that of the liquid. Melting and the Lindemann Formula Melting is a very complicated phenomenon. The values of r before and after melting are both close to a . The amplitude of vibration is related to the temperature by the law of equipartition K((r . Ice and water is the most important example of the latter. The relation of Young's modulus E with K in a cubic crystal is K = Ea. In general. then an increase in pressure will raise the melting point (i. one can obtain a surprisingly accurate relation between the melting point and the interaction energy.4. the crystal melts. We shall briefly mention it here. the crystal melts. If the volume of the solid is larger than that of the liquid. then Ka'a' = To (15 25) where To is the melting point. an increase in pressure will lower the melting point. The value of K can be directly determined from the elastic constants. the amplitude of vibration of the atoms increases and will produce defects in the structure. the whole structure collapses.a ) ' ) 2 1 = T 1 2 (1524) where K is the second derivative of the interaction potential at its minimum. the heat of melting is around 3 . The atoms in a solid are arranged in a definite manner. it does not melt so easily). Let us assume that when the molecules in a crystal vibrate to a certain extent. 15. As the temperature rises.10% of the heat sublimation. . When the temperature reaches the melting point. Assume that when the amplitude reaches aa.e. This is the Lindemann formulab. Therefore (1 5 26) bSee books on solid state physics. becoming a liquid. Since the density of the solid differs very little from that of the liquid.
Initially. The good agreement of (1526) is surprising. we have to understand first why (1 5 26) works so well. there Table 15 . and we must choose an (Y value to fit the measured melting points. 15.1. G ( T .01. p. The fitted value of (Y is 0.2 shows some results. 2 To from (1526) (K) To from experiment (K) 600 Lead Silver Iron Tungsten Common salt Fused quartz 400 1100 1800 4200 1200 1900 1270 1800 3650 1070 2000 Taken from Tabor (19791.e. which shows that the amplitude of vibration is not too large even when the crystal melts. Hence t o understand the process of melting. Definitionsof p1 and pz The preceding discussion mainly reviews some phenomena of the coexistence of liquid and gas. i. pz was the chemical potential of the gas and p1 that of the liquid. the definitions of p1 and pz were not clear. In (151). (1 5 27) From (1527) the calculation of G does not mention the gas phase or the liquid phase or its equilibrium. Both E and IZ can be measured directly. Table 1 5 . At a fixed temperature and pressure. . but the basic assumptions in statistical mechanics allow only one p.PHASE EQUILIBRIUM 271 where n = l / a 3 is the density of the crystal and k is Boltzmann’s constant (we had set k = 1 previously). We now want to clarify some basic conceptual problems posed in (151). p ) divided by N.2 Melting point calculated from Lindemann formula with a = 0. 246.5.
This implies that p z ( T ) is still defined as below T ( p ) (at least in the neighborhood of T ( p ) ) and p l ( T ) is defined as above T ( p ) . 155 Discontinuity of P ( T )at T ( p ) . 155. they are meaningless because (1527) clearly indicates that there is one and only one p ( T ) curve. where the unwanted parts have been removed and there is only one curve. I I I Tlpl 9 T Fig. But what is the meaning of p l ( T ) above T ( p ) and p z ( T ) below T ( p ) ? From the basic assumptions of statistical mechanics. Otherwise we do not know the intersection point. This intersection point fixes the equilibrium temperature. 154 as the situation depicted in Fig. At T ( p ) there is a discontinuity in the slope of this curve. Under a fixed pressure.272 STATISTICAL MECHANICS is only one G and only one p. Eq. (151) states that we can find the intersection point of two curves p l ( T ) and p z ( T ) . . 154 Plots of the chemical potentials for 2 phases. Then where d o the two chemical potentials p1 and pz come from? Now let us examine carefully the meanings of p1 and p z . We should really think of (151) and Fig. Fig.
e.e. we can see that the extrapolated portions of pl and pz in Fig. Since in Fig..g. 155. The cloud chamber and bubble chamber for detection of high energy charged particles utilise the principles of supercooled vapour and superheated liquid. i. i. Above T( p ) . transitions such as that from liquid to gas are called firstorder transitions. However the number of the molecules. the gas phase can still exist i. where we discussed the water droplet. and if N is sufficiently large. and cannot develop any discontinuity in its slope. supersaturated vapour).e. . Direct calculation from (15 27) often misses the properties of metastability.29) ‘Yang and Lee (1952). If the sum in (1527) contains a finite number of terms there would be no discontinuity. When N tends to infinity. the slope of the curve p( T ) appears to be discontinuous.28) is smooth in T and p and infinitely differentiable. However these supercooled and superheated states are metastable. we encountered no difficulty. A slight disturbance will vaporise or condense the particles. 155 are not entirely meaningless.2. If in a certain transition the second derivative. or liquid forms small bubbles. only the duration of time is an important problem. N . the thermodynamical potential is discontinuous in its first derivative. When a charged particle passes through.PHASE EQUILIBRIUM 273 Now the problem is: (1527) looks like a continuous function of T and p . Finally we mention some problems in terminology. is large.e. 155 and of T ( p ) is clarified.. Yang and Lee pointed out that when N + 00. vapour condenses to form tiny water droplets. as a supercooled gas (i. Each term in the sum. There is considerable literature on this subject. continuity becomes a problem. Griffiths (1972). a discontinuity can occur. which are often important characteristics of a phase transition. each statistical weight (1 5 .c Now that the meaning of the curve in Fig. Where does the discontinuity in the slope come from? This mathematical problem is of course extremely important. the specific heat (15 . e. In Sec. Below T ( p ) . 15.. the liquid phase can also exist.. as superheated liquid..
Experiments show that at 99OC and 10l°C the water vapour pressure is 733.0 mm Hg respectively. . a p / a V = 0. p. (The latent heat of evaporation of waters is 9717 cal per mole at 100°C. prove that (?)I(?)= 0. The boiling point of a certain liquid is 95OC at the top of a mountain and 105OC at the bottom.274 STATISTICAL MECHANICS is discontinuous. From the results of Table 15 3 and ( 15 9). Use (159) to obtain the latent heat evaporation of water.order”. then this transition is called secondorder. the specific heat approaches infinity.. should not be called secondorder.) 2. a p / a V < 0 . and d T is the change in the boiling point due t o dp. Also. where d p is the change in pressure. extend the discussion of Sec. 5. there is no distinction between gas and liquid... Its latent heat is 1 0 0 0 cal/mole. Only under certain approximations (the mean field calculation) is C finite and discontinuous. 15. Experimental results show that the entropy of evaporation (latent heat of evaporation divided by temperature) at the boiling point (at one atmosphere) is approximately the same for many liquids. . since at the transition temperature (i.. Above the critical temperature T. See Table 153 (taken from Zemansky (1957). The vapour pressure curve of a liquid also follows (15 9). Problems 1. 6. Under these conditions..e. Discuss the evaporation and growth of water droplets in twodimensional and in onedimensional space (i. what is the meaning of the formula (13. 322.a p / a V is very small. Discuss the scattering of neutrons or photons in the neighborhood of T . . the total volume can change but the pressure is constant.1 . Calculate the height of mountain. above T. In fact. At temperatures close to T.7 and 788. 4. i.. When a gas and liquid are in equilibrium. Trouton’s rule).e. some commonly known secondorder transitions such as the transition between the paramagnetic and ferromagnetic phases and the superfluid transition of liquid helium.19) in Chapter 13? 7. We should not attach too much significance to the terms “firstorder’’ and “second . the critical temperature).e. How do we explain these results? 3.2).
2 121 3 73 319 263 216 185 240 112 145 350 254 243 334 252 L Latent heat (cal/mole) 415 1560 2 158 3 021 3 920 1562 AS LIT 15.9 17.3 Experimental value for boiling points and latent heat for some liquids 'b (boiling point K ) 27.2 18.PHASE EQUILIBRIUM 275 Table 1 5 .3 120 165 21 1 85.5 20.1 26.9 18.6 18.4 20.0 21.4 23.3 17.1 6 970 5 576 .0 4 878 3 860 4 210 4 724 1330 1630 3 293 9 717 6 400 20.3 17.0 239 188 206 23Ei 77.5 20.8 22.0 18.0 5 955 4 463 3 956 5 581 1955 3 010 7 140 5 850 4 850 20.3 90.0 20.7 20.1 27.4 19.4 20.2 87.0 22.3 18.6 21.4 23.
1 19.1 17.9 c H6 2 C6H6 (C2H5 ) 2 0 Li Na K Rb cs Hg Ga In Cd .276 STATISTICAL MECHANICS Table 15 .4 22.4 15.4 24.1 20.3 18.8 18.9 22.8 21.7 22.3 Cont’d Tb L Latent heat (cal/mole) 3 517 7 350 6 470 30 800 21 800 18 600 17 100 15 700 14 100 62 000 51 600 23 900 (boiling point K ) 185 353 307 1599 1156 1030 985 986 630 2 540 2 273 1040 AS LIT 19.
the problem is simple. 16. Finally. Let the total energy be The product g p B is the ratio o f the magnetic moment to the spin. s be N spins. or a nucleus. magnetism has played a very important role. Of course. belong t o this second class of problems. . If the second type of interaction is very strong. Magnetic interactions can be classified into two types: the interaction of the external magnetic field with individual electrons or atoms. and the interaction among electrons or atoms. . we also have t o consider the external magnetic field. If it is the . etc. magnetism has always been a subject for research and application. Each spin comes from an electron or an atom . In modern statistical mechanics and many domains of natural science. s2 . the problem becomes very complicated because we have to consider the collective behavior of many spins: ferromagnetism and antiferromagnetism.1.Chapter 16 MAGNETISM From the time of our ancestors. etc. in this second class of problems. paramagnetism and diamagnetism belong to this class. If the second type of interaction is relatively weak. In general. Paramagnetism Let s l . we review the cause of the mutual interaction. This chapter discusses paramagnetism and diamagnetism caused by electron orbits.
i. The free energy can be calculated very easily: 3. (162) where m is the mass of the electron and kG = lo3 Gauss. . The term Ho in (161) includes all energy which does not involve the external field h. We shall discuss the meaning of F’ below. Eo and So are the energy and entropy. then g=2. (163) The term Fo is the free energy due t o Ho and 1 h’r gpBh 2 .N T l n ( 2 cosh h’lT) . . (167) (168) F’EEoTSrF+hM where M is the total magnetic moment along h . calculated from H o . Suppose s = Each spin can exist in two states. respectively.72 X K/kG . parallel or antiparallel to h.e.~ t h’n h a (165) = gpBNtanh 1  h’ T (166) E=EohM .27 X lo’’ erg/kG = 6.IT F = Fo .F. pB = e l ( 2 m c ) = 9.HIT S = ( eh’/T + eh’lT)h’ . (164) By differentiation.278 STATISTICAL MECHANICS electron spin. .FIT = 1. Now assume that Ho does not depend on si. we obtain the entropy and other quantities s= (s)h = S o + N [ l n ( 2 c o s h $ ) .
p is a constant and V is the variable whose equilibrium value is (161 1 ) In (161). . (163) and (167). In Chapter 7.h . ( . then E and F should be replaced by E+A=ESpV F+G=F+pV . (169) where p V is the potential energy due t o the external pressure p . d o not include the potential energy . when we discussed the thermodynamic potential. the free energy F’ of any magnetic substance can be regarded . i. (1610) T But in A and G . To understand this point we use a gas as an illustration. like E and F in (169). If we want to include this. including the potential energy of the spins in the magnetic field. A and G in (1610) are like E and F in (167) and (163).h M is like p V ) : dW = hdM (1612) On the other hand.h M . M = (dk)s aE = (g) T (1613) In general. . Therefore Eo and F’.M . The magnetic moment M is treated as a variable and not a constant.MAGNETISM 279 Note that here the free energy F is that under a fixed magnetic field.e. The work in the first law of thermodynamics is defined in terms of the internal energy: p = (g)s (%) = . we pointed out that the energy E is the external energy and does not include the potential energy due to the external pressure. we have included the potential energy due t o h .
(1614) This is not limited to paramagnetism. whose mutual interaction is extremely small. Then we decrease the magnetic field S/N In 2 t1 A Th’ I I I 73 72 I I 71 Fig. Vibrations essentially stop below K. (So isdiscarded. then S I N = In 2. At low temperatures. Figure 161 is the curve (165) of entropy versus T for different magnetic fields h . from A t o B). If h = 0. In the figure. 161. In low temperature experiments. .h M is a function of h and T: dF’ = SdT+ hdM d F = SdTMdh .). 161  T aThe magnetic interaction energy of two nuclear spins at a distance of mately lo’’ K . Now let us discuss cooling by adiabatic demagnetisation or “nuclear cooling”.a At 1K they can flip freely. To absorb heat requires freedom of molecular. and h 2 . all substances become solids. A low temperature environment is simply one which absorbs heat readily. cm is approxi .280 STATISTICAL MECHANICS as a function of M and T. we attempt t o absorb as much heat as possible from the substance under observation and thus we must lower the temperature of its environment. The only remaining mode of motion is that of the nuclear spins. while F = F ’ . The entropy S is lower for higher magnetic fields because the magnetic field restricts the motion of the spin. a major achievement in experimental low temperature physics. Adiabatic demagnetisation utilises this kind of motion to absorb heat. atomic or nuclear motion. h2 < h l . Cooling proceeds as follows: we first apply a strong magnetic field to align the spins (see Fig.
we want the interaction to be sufficiently weak so that this interaction will not hinder the free rotation of the spins. Equilibrium is defined only on sufficiently long time scales. it is nevertheless larger than the T 3 specific heat due to phonons. the number of mobile electrons is proportional to T . The process A + B + C in Fig. Although this is small. i. .b The use of metals is due to the fact that free electrons can still move at extremely low temperatures. the total magnetic moment M is unchanged. 16. At low temperatures. Therefore as the magnetic field is decreased the temperature is lowered along BC. the spins can flip more easily. On one hand. the specific heat is proportional to T . Dundon and Goodkind (1974). through a slight interaction with the electrons. These changes are brought about by the interaction of the spin with its surrounding environment. This is cooling by adiabatic demagnetisation. (1615) Since the entropy is not changed. we can use magnetic fields of different strengths. In adiabatic processes.. we require the interaction to be strong so that the spins can change continually to reach equilibrium.1 can be continued to D and then repeated. On the other hand. the observation time may require a week. is then able to reach equilibrium. entropy is unchanged. the experiment is much more complex.e.aM ah t (1616) bOsgood and Goodkind (1967). Some fifteen years ago. thus keeping the observation time reasonably short. This allows the absorption of more energy and hence a lowering of the temperature. every spin must undergo many changes.MAGNETISM 281 adiabatically. M is also unchanged. In small magnetic fields. the experiment was performed using nuclear spins in metals and a temperature below lou4K was reached. Magnetic Susceptibility The magnetic susceptibility is defined as x = . Note that when the magnetic field is lowered.2. Even so. Entropy is a function of M alone and'(165) can be written as rnrM/N . so that during the observation time. This poses quite a difficult task for the experimentalist. 16. Although the above theory is very simple. Of course. The nuclear spin.
. because the denominator of (1618) is also T. ‘Discovered by Mme. x = M / h . there are some particles outside the Fermi surface and some holes inside. ( M 2 ) = t ( M .are uncorrelated. contributing to the magnetic susceptibility. Therefore at T = 0. (1617) This inverse proportionality to T is a common phenomenon. the energy levels are filled up to the Fermi energy eF. At T > 0. all the spins are fixed .e.Curie.onehalf upwards.up spin and down spin. The number of these particles and holes is proportional to T. i. electrons in a metal. . or if we neglect quantum effects. or 3He. we found that (1618) If every spin is independent. e.g. We consider only the spin of the electrons and disregard magnetism due to orbital motion. Let N+ be the number of up and down spins respectively. Now let us examine the magnetic susceptibility of a collection of fermions..)= 0 because N+ and N . Note that (1618) assumes [ M .282 STATISTICAL MECHANICS usually under the condition h 0. Each spin has two states . which is valid if M is conserved.20) ( N + N . M ) . then (1619) The factor comes from the definition that M is the magnetic moment along the direction of h .e.‘ In Chapter 13 (see(136)). called Curie’s law. The magnetic susceptibility for the model of the preceding section can be calculated from (166): + x= $NkPB)*iT . (16 . The spins of these particles and holes are free t o orient. and onehalf downwards. i. at very low temperatures x is a constant. At T = 0. Then +. The calculation of this constant value of x is very simple. If s = we then get (1617). H ] = 0. Therefore.
.3. which. 16. From (1618). Equation (1623) is the Pauli susceptibility. in turn. and x reverts to the form ) (1617). this argument is not entirely correct.1 I I ‘0 ‘ 0 I I I Fig.d The experimental value of x for 3He is far greater than that given by (1623). = N . ( N : . (1620) and (1621) we finally obtain (1623) Note that at high temperatures. Repulsion causes (N.) The reason is that due to the interatomic interaction. resulting in diamagnetism. liquid 3He is far from being an ideal gas. 162 Orbits of charged particles in an external magnetic field. increases ( M ’ ) . Diamagnetism of Charged Particles Charged particles move in circular orbits in a magnetic field. considerably. (The magnetic moment gpB of 3He must be taken to be its nuclear moment.MAGNETISM 283 Now use the relation in Chapter 13 (see (1317)): (1621j At low temperatures a N / a p is the density of states (see (414)): Vg(0) = V mpF 7r2  (1622) and thus for an increase in d p of the Fermi energy p . . N)c < 0. However. the number of occupied states will increase by d N = Vg(0)dp. Figure 162 shows some orbits for positive charges moving in a magnetic field pointing into the r. who discovered the exclusion principle. This can produce a magnetic moment opposite to the external magnetic field. After Pauli.
r=R+r' . 162. Let R be the position of the centre of the circle. Let us suppose that the magnetic field points in the z direction and the orbit lies in the x y plane. If an orbit lies within the region in Fig. 163 . then the calculation is slightly more complicated. 162). Now let us choose any one region in the gas (inside the dotted line in Fig. The motion along z is neglected. then u is the area of the circle. If the orbit is circular. Orbits totally within the region contribute an upward magnetic moment but orbits near the boundary produce anticlockwise current. The magnetic moments of the circular orbits point upward. The orbital magnetic moment is proportional to (1 6 . If the orbit is partly outside the region. We can prove that the total magnetic moment within the region is zero. 162.24) the integral being along the orbit. (1625) The direction of o is along + Z and a is the radius of the circle. The first Y Fig.284 STATISTICAL MECHANICS paper. forming a downward magnetic moment. then u=na2 . (1 626) Figure 163 is an enlarged view of the orbit on the right of Fig.
All orbits within + a of the boundary lie partially within the region. its derivative with respect to h must vanish. the total magnetic moment of those orbits along the boundary is.n a 2 . So the magnetic moment of the whole region is zero. Integrating (1629) along the boundary. Indeed. (1631) where A is the area of the region pointing towards z . the total magnetic moment is zero: (1 632) . The magnitude of (163 1) is the same as that of the magnetic moment due to those orbits wholly within the region. a magnetic field can only change the direction of motion of the particles. (1 627) where y is the unit vector in the direction of the yaxis. Therefore. Since it is not a function of h . where n is the density of the particles (number per unit area).R x (2a(sin0))y 2 . The above analysis suffices to prove that orbital motion does not produce a net magnetic moment.. (1629) sin0 d ( c o s 0 ) = 4 n (1630) The differential d R is in the anticlockwise direction and we have neglected the last term in (1626) because its contribution to the magnetic moment is proportional to the perimeter while that of (1629) is proportional to the area. from (1 627) and (1 628) 1 2andR . i. but not the energy.MAGNETISM 285 integral on the righthand side of (1626) is s dr' = 2asinO y . but with opposite sign. The thermodynamic potential is related only to energy and hence is not a function of the magnetic field.e. There are n(2a d R ) (1628) such orbits along a segment dR of the boundary. we get nu2 E$R 2 x dR = na2nA .Rn d R x 2 .. This is a result of classical mechanics.
3) . The effect of the magnetic field is to into the same energy state. and E. The density of states of the particle becomes L2g(€)= c n W. (1633) where h is the magnetic field. Therefore. = ( n + +)a. n = 1 . so the magnetic moment is nonzero. 2 . + n =O With g ( e ) . . h = 1. we can calculate the thermodynamic potential and the other equilibrium quantities. . The result of the quantum mechanical calculation is: the energy density of states of the particles and hence the thermodynamic potential are changed by the magnetic field. is the statistical weight of the energy e n . ) wrehlrnc . so squeeze all the states between E. The above expression assumes that go( E ) does not change considerably over this energy interval. For simplicity. This result is extremely important. the original kinetic energy p 2 / 2 m now takes the form of energy characterising simple harmonic motion. In the absence of the magnetic field. and w is the rotational frequency of the particle. The thermodynamic potential of free electrons is 1  L2 a OD = T 0 d e g ( E ) l n ( l +e"'")'*) .) . the number of states between E. . we consider motion in the x y plane with the magnetic field in the z direction.) . (1637) eLandau's theory of diamagnetism. see any textbook o n solid state. (1634) L 2 is the area of the object and W. and is L2 j"" d e g o ( € ) = L2wgo(E.8(€€. c is the velocity of light. (1635) where go(€) is the density of states at zero magnetic field and E = 0 is the lowest energy state.e We shall only highlight the essential points here. we can solve for the energy of each particle E.286 STATISTICAL MECHANICS The situation will be slightly different if we take quantum mechanics into account. From the wave equation in quantum mechanics.
any integral can be written as 1 m R(E)dE = as long as R ( E ) is sufficiently smooth and the integral is finite. 164 Area approximated as a sum of elementary rectangular areas. nois the thermodynamic potential for I I I I I w 2w 31d 4w I 5w I I I Fig. The sum in (1638) is the area of the rectangular strips in Fig. + 0 1 n m =O R ( ( n + +)a) 24 w2 (  dR dE)e=o 0(w3) ’ (1638) = (16 . only the value at E = 0 is related to w and (1637) becomes 1 .M ) / )] T Note g o ( € ) = m / n is a constant and 0 = 0. Therefore.( e . If w is very small. the difference is negligible.MAGNETISM 287 Note that the only difference between the integral of g(E) and that of go(€) is that one is a discrete sum with step size w while the other is a continuous integral. each with height R ( ( n + $)a).a . As the reader can verify..39) & .1a w2 0 = L2 L2 24 T [ g o ( € )In (1 de d + e . 164. Figure 164 shows the difference between integration and summation.
44) . i. and (1639) is well approximated by 1 02m p I T . the orbit of the particle still produces a small effect but purely as a quantum effect. then e P l T is small. At low temperatures. (1639) and (1643) are obviously incorrect because the integrand cannot be regarded as a smooth function.e. This result is also valid in three dimensions because the motion of an ideal gas in the zdirection is independent of the field. So there is a periodic change. If the density of the gas is very high and p = eF >> T . 5) to move in to increase the area. Further movement will decrease the area again. in three dimensions. then (1639) becomes (1643) This result is valid only in two dimensions because. As w increases. (1640) . This change is more pronounced for lower temperatures.. The area of the strip between (5.288 STATISTICAL MECHANICS If the density of the gas is low..e. as illustrated by R ( E )in Fig. w / T >> 1. The area of the strips can be quite different from that under the curve. when (1 6 . the components of the motion of a dense fermion gas in different directions are not independent. i. Note that the area of the rectangle will undergo a sudden change whenever ( n + $ ) w moves across y . We have to wait for the rectangle between (4. (1 642) Therefore. 6 ) will decrease. we get the total magnetic moment o=2pgh .(aao)=L2 2 4n The density of an ideal gas is (1641) From (1640) and (1641). 165. At high densities and small T . 5 0 will shift to the right. abrupt changes must occur near E = p .
165 Abrupt changes can occur near E = u . Therefore.MAGNETISM 289 I w 2w 3w 4w 5w 6w Fig. Now write (1636) as (1 645) (1 646) The expression in (1646) is a 6function with period 1. but we shall not discuss the details here. As w increases. We consider E > 0.but the results of (1644) are disturbing. Note that the expansion in powers of w is not permissible: no matter how small w is. we have to reexamine (1638) which is a mathematical problem. this periodic change always exists. How much does the integral of a function R ( E ) differ from the area of the rectangles in Fig. No power series in w can explain this change. the change is a periodic function of l / w . 5w will shift to the right. 164? Equation (1638) is a direct expansion in terms of a. So the sum over equally spaced discrete arguments of any function can be . This is a most important area of research in solid state physics. 6) will decrease. Therefore. The area of the square between ( 5 .
290 STATISTICAL MECHANICS written as (16 47) (1648) JO The integral of R(E) isjust R . then (16 . This example demonstrates that expansion in a power series is not always reliable. Example I : R ( E ) = e‘IT. When w = 0 . If the k series converges rapidly. Let us look at two examples.50) The series is convergent and there is nothing peculiar about it. every derivative is zero. this formula is very useful. so w n=o 5 R((n+ m (1649) This is the socalled “Poisson sum rule”. so according to (1638) we would obtain zero. then This function is quite peculiar. I t cannot be expanded in terms of w . But the k series in (1649) converges rapidly. Example 2: R ( € ) = ee2/20’/JzT O . .
r l ) Antiparallel spins: @ ( r l . If the spins are antiparallel.x ( s z . s l ) . will 16. . In solids.e. e2nikplw E =p.MAGNETISM 291 Note that if there is any discontinuity or rapid change of R ( E ) near in Eq. Electrons are fermions. Short distance exchange interaction This is the strongest interaction among electron spins. so @ ( l . r 2 ) = @(rz. @ for parallel spins is zero. . because of the crystal structures and impurities. resulting in a variety of magnetic properties. This implies a lower average potential energy due to electrostatic repulsion. The wavefunction of two electrons can be written as where rl and i2 are the positions of the electrons and s 1 and s2 are their spins. but @ for antiparallel spins will not vanish. the most important being the short distance exchange force and the long distance dipole interaction. i. a periodic function of period l/o be obtained. but if they are antiparallel ~ ( s . Generally speaking. The wavefunction of ferniions must be antisymmetric. sl).s 2 ) = . SpinSpin Interaction Now let us review the mutual interactions between spins. electrons with parallel spins have lower energy. if the spins are parallel. (1648) there will appear terms of the type. Hence. etc.r l ) . (1654) Thus when rl = r 2 . then x ( s l . A . Therefore Parallel spins: @ ( r l . the electrons can be closer and the electric energy is higher. Here we discuss mainly the exchange interaction. s 2 ) = x ( s 2 . r z ) = @ ( r 2 . The condition rl = rz nieans that the two electrons are at the same position.2) satisfies Now we know that if the spins are parallel. it is caused by the electric repulsion and the exclusion principle of the electrons. the two electrons will seldom be close.4. Therefore. there are numerous types of interactions.
Under this approximation.are degenerate.is E + . with 9.is for parallel spins and @ + is for antiparallel spins. Let 91 (r) 92 = @(r) I (r) = 9 ( r .E . If the electrostatic interaction between the electrons is neglected. the energies @ + and 9. We shall make a crude estimate of J . Note that s1 * s2 = + I parallel for sl.292 STATISTICAL MECHANICS This can be summed up in the following formula effective interaction = Jsl s2 (1655) where J is a constant.e.3 (1 660) . then the wavefunction describing the two electrons is (1652).R) 5 (1 6 56) where @(r) is the 1 s wavefunction of the hydrogen atom.r2 I (1658) Using firstorder perturbation theory.( r 1 ) @ 2 ( r 2 ) &(91 * @1(r2)92(r1)) 3 (1 6 57) where 9. i.= J . Suppose two hydrogen nuclei are fixed at 0 and R .(r13rz) = . s2  .. Now we include the electrostatic interaction U(rl . we get (1 659) The difference in the energies of @ + and 4. electrons with parallel spins have an energy lower by the amount J . The meaning of (1655) is precisely this.r 2 ) = Itl e2 .
based on (1655).f We can only hope that (1662) is a reasonable approximation. not really so simple. Our aim is merely to discuss the dynamics of the spins and to summarise all the effects of the other degrees of freedom into the effective interaction strength J . the macroscopic magnetic property is not directly determined by the magnetic interaction among these dipoles. while J is very large if R 5 a .. i. then the method involving an effective interaction is still a convenient way of handling the problem. and the energy difference comes from the electrostatic interaction. There would be no need to discuss the effective interaction but this method is too cumbersome.MAGNETISM 293 Note also that J is a short distance interaction because the product Q l ( r l ) in the integrand of (1659) is nonzero only when both GI and G2 are nonzero. see Mattis (1965). can we write the effective interaction as The answer is no.e. the safest way is t o solve for the motion of all the electrons and the problem of the spins is obtained consequently. Even the basis of (1655) is not entirely clear. The calculation of the effective interaction between spins is. Of course. It is surprising that J is totally unrelated to the magnetism of the electrons: the spins only enter via the symmetry of the wavefunction. But G1 is centered at r = 0 while G2 is centred at r = R. it is not so simple. Although every electron is a magnetic dipole. This J carries a heavy burden. of course. is to fFor the theory of J . . 42(r2) (1661) where a is the Bohr radius of the hydrogen atom. The response of the whole system to an external magnetic field is now determined by J . Therefore J + 0 if R >> a . An interesting question is that if there are three spins. The expression (1655) is the energy difference between the two states of a pair of electrons. One of the most common models of magnetism. The above crude analysis shows how (1655) comes about. If we are interested in the spins only. only in the overlap of G1 and G2.
then this interaction cannot be neglected. In certain situations. so we must try to simplify the models. Jii B. even the simplest model presents a very difficult problem.g. with the purpose of obtaining a crude understanding of magnetism. We assume (1663) to be correct and calculate some results from it to compare with experiments. This is a quantum model: s j is an operator and so is H . If the spins exhibit collective alignment. Moreover. If we wish to discuss the correlation among many degrees of freedom. (1663) is a tool for analysis. be it the interaction between spins. the interaction energy of two magnetic dipoles m l and ml is m * . .. is longranged. However.r)(m2*r) (1665) r5 This interaction. In the next chapter we shall introduce the Ising model. In other cases.. may also be negative. corrections may be needed.. but may be the total spin of the atom. In any case. (!6 63) the exchange energy between spins i and j and (1663) is the socalled vector model or Heisenberg model. Long distance dipole interaction At a separation r . si need not be the spin of an electron. e.m 2 r3 3(m. terms like may be included in the Hamiltonian. (1663) can be very useful.294 STATISTICAL MECHANICS write the total energy of the spins as mi = gpClesi Jij is . The result of a mutual interaction is not easy to analyse.. or the attraction and repulsion between molecules. We can either fix J by experiment or we can calculate J and predict the result of experiments. then the usual mathematical analysis is inadequate. though much weaker than the exchange interaction.
. 3 . then what is the magnetic susceptibility? (c) Discuss the correlation of the spins among the electrons in metals. are always parallel and Si and s j + z are uncorrelated.g. (b) If the magnetic field is suddenly switched off. . 5 . . in a cubic lattice. 4. A container of volume I/‘ contains a dilute gas of N particles. If liquid 3He is pressurised adiabatically. each spin is opposite to those of its six nearest neighbors. This cooling method has been pointed out by Pomeranchuk and is named after him. The initial temperature is T . Assume that there is energy exchange between the gas molecules and the spins. Assume a model i = 1. Magnetic susceptibility is related to the correlation function of the spins. . of course. +. . and compare with antiferromagnetic materials. On the wall of the container there are N’ spins ( N ’ << N ) . using the fermion gas model with the Fermi energy approximately equal to 5 K.MAGNETISM 295 Problems 1. (a) If the magnetic field is turned off slowly. This problem is. s j + . explain the method of cooling by pressurisation. 2.. (b) If s j . calculated A T . i = 1. Each spin has two states of energy * h where h is proportional to the external magnetic field. Below frozen. 3He exhibits diamagnetism and the spins are from the spins. The spins in an antiferromagnetic system are arranged periodically. . with N spins s i of spin (a) If S i . are always antiparallel. Write the thermodynamic potential (1637) directly as (1666) Differentiate with respect to h to get M= aa ah . assuming that there is no heat loss. . (b) At low temperatures the entropy of solid 3He comes almost entirely K. s j + i . (c) From the diagram above. Draw the entropies of the liquid and the solid phases on the same diagram as a function of T . 5 . 3. e. 2. N . i = 1 . 3. calculate the magnetic susceptibility. it becomes a solid and the temperature drops. (a) Calculate the low temperature entropy of 3He. Cooling by pressurisation. related to problem 2 . Equation (1619) is the result if the spins are uncorrelated. . The following lists the steps of the theory. calculate the change of temperature A T .
The expression in ( 1670) is the energy operator for a simple harmonic oscillator with eigenvalues given by (1633).296 STATISTICAL MECHANICS (1667) The first term o n t h e right of (1667) can b e interpreted as t h e magnetic moment due to t h e circular orbit of t h e electrons. The vector potential is given by A = h x and points in the y direction and o = e h / r n c . We neglect the motion along the z direction. then (1668) can be written as (1670) Quantum mechanically.. prove that t h e orbit of the electron is a circle centered at ( x o . This problem is the main content of Landau theory of diamagnetism. Derive g ( E ) . .1 p $ + ( p y [ 2rn + r n w ~ ) ~ I.) What is the interpretation of t h e second term? First review the discussion from (1631) t o (1654). [ H . (a) From the equations of classical mechanics. p y 1 = 0 and thus p y is conserved. The total energy function of an electron in a magnetic field can be written as e 1 H= (p+A)* 2m C = .x o = x’. (b) Let x . 5 . y o ) and py =mwxo (1669) is an invariant. (1668) where the magnetic field h is normal t o the x y plane. (Treat each n as one orbit.
its solution in onedimensional space and the proof of its ferromagnetic nature in two dimensions.Chapter 17 ISING MODEL The simplest possible variable is one which takes on two values. no change is possible. Each si is an element of motion. s ~ . We shall then briefly mention some models akin to the Ising model. and we call this the Ising model.. The models mentioned in this chapter are related to phase transitions. Their values f 1 denote the two directions of the spins. Ising Ferromagnetism in One Dimension The Ising model in modern physics starts from the ferromagnetic model. (If there is only one value. We shall introduce the simplest king ferromagnetic model . (171) r=l This is a onedimensional ferromagnetic model. 17.1. N ) be N Ising variables.h + ~ 2 sr r=l N . Let sr ( r = 1 . There is interaction between neighbouring elements (or spins) and h is the external magnetic field. but its application is by no means limited to phase transitions. . 2 . The total energy is N1 H(s) = J x s. . . The Ising model has received the greatest attention in this respect. The simplest collective model is a set of Ising variables. the applications of this model are numerous.a aThis model was first discussed in Ising’s thesis. it nevertheless describes certain general features of many phenomena. Generalisations are likewise named after him. Although it is quite different from real physical systems. In modern statistical mechanics.) This kind of variable is called an Ising variable. . . The two values can be chosen to be + 1 and 1.
171). The above assumes N .1.e. that the basic assumption of statistical mechanics are still valid. (172) If T > 0. If the temperature T = 0. The reader can calculate the other quantities. so . then either all s are + 1 or all s. are 1. so that the partition points form a fermion gas. so its chemical potential is 0 . it is unlike the gas model and . so that . with the result 1 N'JN = e2J'T +1 . In this respect. (173) where N ' is the number of partition points.[ E ( T ) E(O)] = N +1 . other means. it does not say how each s varies with time. The equilibrium properties of this model can be derived by the method of fixed temperature. in which the Hamiltonian function or operator determines the equation of motion. The energy of each partition point is 2 J . then. region and the 1 region is called the partition point (Fig. We first look for the solution to the zerofield. This is a general result : onedimensional models d o not normally exhibit ferromagnetism. with a value E(T=O) = J(N1) .298 STATISTICAL MECHANICS This model has no timedependent dynamics. The boundary between + 1 . that is to say. . with at most one partition point per site. The number of partition points is not constant. N ' >> 1 . and the total magnetic moment is zero : = o . The possible sites are between any Y and r + 1. h = 0 case. . This model is then transformed to a model of a gas of partition points. At T = 0 there is no partition point and at low temperatures the partition points are few. Because N ' / N # 0 (unless T = 0). The evolution of s must be determined by . some s will be + 1. We shall only assume that each s can be changed at any time. and others . the s do not point in a unique direction. quantum mechanical models. H i s a minimum. (174) i. this model has no ferromagnetism (except at T = 0 ) . 1 25 e2JlT .
. The total length is the sum of the lengths of the sections: N = y1 + y 2 + . ) = 2J" + h(y1y2 +y3y4 . . be the lengths of the sections. if the total length of the system is large enough. t t 0 $ 4 0 + t etc partition points or + +. etc.1. .. Let G ( ~ . Let y l . the total energy (1) can be written as H ( y l j y 2 3 . i.+ + +.e. Therefore. y z . . those in the second are .e. This is similar to the last example in Chapter 10. Figure 171 shows how the partition points break up the body into sections with the spins in each section having the same value. +1 1 5r t t J 0 t t t 0 . At a certain temperature.. i. 1 7 . .YNt) . then every sr would be the same and this body can be regarded as a ferromagnet. and N' is assumed to be even. We can determine .y3 . (175) where is the average distance between two partition points. . T= TlnZ' ) (178) . . y i is the number of spins in the i . (177) This can be regarded as the volume of the gas of partition points.the various equilibrium quantities by the method of fixed temperature and fixed pressure.ISING MODEL 299 This conclusion holds for macroscopic bodies. The term 2JN' is the energy of the partition points.. (176) This energy is measured from the zero point and we assumed that the spins in the first section are + 1. If N is smaller than E.t h section. .1 +.+ + Now let us look at the situation with an external field h # O . .*. ferromagnetism will disappear due to the appearance of partition points. N>> E E eZJJT .+ Fig. +yNi .
increasing y . The magnetic field acts as a tension for spins of + 1 . From the second equation of (P712) we get (1713) In order for the pressure p to be meaningful. and as a pressure for spins of 1 . the force necessary to make they's decrease) and increases with temperature. .e. Putting (1713) into the first equation of (1712). (171 1) (1712) From these two equations we can solve for N' and p . the square root must be positive. Notice that p is the pressure of the partition points. we then get N ' .300 STATISTICAL MECHANICS Substituting (176) into (178) we get (179) The pressure p and the number of partition points N' can be determined by =o aN' The results are aG . (i.
Proof of the Existence of king Ferromagnetism in Two Dimensions The example in one dimension is simplest. .hN’ ePIT ( 1 . x 17. There exists a partition point between every pair of neighbouring spins with opposite signs. These points can be joined to form . ). and define a variable si for every lattice point i. abbreviated as +. (1714) x M == . The definition of the model in two dimensions i s as follows. = J if i. the total magnetic moment and the magnetic susceptibility are .2. Consider a square lattice. which is just the average distance over which the spins can maintain the same direction. Now this is modified by a factor t . We now give such a proof. We first give a simple argument to show why ferromagnetism can occur. the average value of the magnetic moment is nonzero.PIT)2 N * T .ISING MODEL 301 When h is small.2J/T N = h T T t (1715) If every spin is independent then =NIT. To prove the existence of ferromagnetism is to prove that in zero external magnetic field and at nonzero temperatures. Every point has four nearest neighbours. Then we present a rigorous derivation.. The concept of partition points in the previous section can be extended to two dimensions. The total energy is J11. j are nearest neighbours otherwise = 0 . Figure 172 shows some values of the spins ( + 1 or 1. But the proof that the king model exhibits ferromagnetism (in dimensions > 2 ) is not difficult. The Ising model is not easy to solve for two or more dimensions.
. and L is at least 4..e. there is ferromagnetism.+ + . Fig. At T = 0 all the spins are in the same direction (say positive).302 STATISTICAL MECHANICS + + + . dividing the whole lattice into regions. energy must be spent.+ + + + + + + + + + + + + + + + + + + + +I. negative regions will appear. The length of the partition line is determined by the number of points on the line.Fj: + + + .. 172 A twodimensional partition of spins for a square lattice. negative regions of large area are unlikely. Therefore most of the spins are positive and the total magnetic moment is positive. i. partition lines. Therefore if the temperature is sufficiently low. the energy then is 2 J L because the energy of each partition point is 2 J .I+ + 7. To create partition lines. If the perimeter of a certain negative region is L . The total magnetic moment is the area of positive regions minus that of negative regions.. When the temperature is increased to T . and there is no partition line. So the probability of having this negative region is e 2JLIT (1717) Suppose h = 0..
(1718) Now we have to prove that if T is sufficiently small. then we overestimateg(L). where partition lines become partition surfaces. Let the probabilities that s1 is + 1 or 1 be p + and p . So the probability of any configuration s1 = 1 can be written as e 2JIdJT x e 2JL'JT IZ 9 (1719) where L is the length of the innermost loop and L' that of the sum of the other loops. (1 7 20) This g ( L ) naturally depends on the positions of the other partition lines. At low temperatures. . except for some small regions.. all the spins will point to the same direction.) = p+ . differing only in some details. the negative regions cannot increase at low temperature because expansion requires longer boundary lines and hence more energy. . bThis is similar to the proof by Peierls (1936). Now let us repeat the above argument rigorously. In one dimension. .ISING MODEL 303 This is different from the situation in one dimension. (If it is surrounded by two loops or four or six . We cannot draw a loop intersecting the other loops. if the density of the partition points is nonzero. then it must be surrounded by one loop or three loops or five loops. In two dimensions. . Let all the spins at the boundary of the square lattice be + 1 . we first fix the outer loops and calculate the number of configurations with length L for the innermost loop : g(L) = number of configurations with length L for the innermost loop. loops. The result is similar in three dimensions. naturally it is + 1 . of closed partition lines.) If we change s1 and all the spins within the innermost loop to + 1.p . s1 must be + 1 . We shall prove that pis smaller than p + . (sl) # 0.. Then (s. The lengths of each region can be changed without expenditure of energy. The number of loops must be odd. Equation (1719) is only the probability for one configuration. If s1 is 1. we get one loop less and this is the configuration for s1 = + 1 . But if we neglect this restriction and count all the configurations looping s1 once. If there is no loop surrounding it. we have positive and negative regions.b Take any spin sl. So on the average the numbers of positive and negative spins are the same. although the density of partition lines is nonzero. To calculate p . and Z is the partition function. .
e. when all the four directions are permissible. it is obvious that g ( L ) must contain this area factor. the probability that s1 = 1 must be smaller than (1721) The factor g ( L ) in this formula is calculated neglecting the existence of the outer loops. the remaining 3 directions can be chosen.' .e.3L' 2L (g . Therefore we should divide 4 x 3L1 by 2L. except for going back on the previous step. (1723) The factor ( L / 4 ) 2 is just the largest area that can be enclosed by the partition line. The 4 comes from the first step. clockwise or anticlockwise. because getting rid of the innermost loop s1 becomes + 1 . there are 3 ways of drawing the next steps. i.we get (1 7 24) Notice that the shortest loop is L =4 and L must be even. In addition we can start from any point on the loop. So the total number is less than 4 x 3 L . The result is (17 22) where g ( L ) is the number of ways of drawing a loop around sl. Starting from any point. Because s1 can be any point within the loop. We can write g(L) < . To draw L steps we have 4 x 3L' ways. .304 STATISTICAL MECHANICS Therefore. Substituting (1723) into ( 1 722). The term outside the bracket in (1721) is the probability for a configuration of s1 = + 1 . Now we sum up the different configurations of the outer loops. We require L to return to the starting position. i. fixing the outer loops. So the sum is p + . a square with side L / 4 .
If T < J the above conclusion gives (17 . e" is less than .g.' but the mathematical manipulation is too complicated to be discussed here. i. The above proof can be extended to three or more dimensions. then ( sl ) must be zero because when h = 0 this model has a symmetry between + and ..27) In the above proof we assume that the spins at the boundary are + 1 . We can also use a very small magnetic field to define a preferential direction. (1725) (1726) When T = J .ISING MODEL 305 The factor p + at the end of (1722) is neglected because p + < 1 . If we do not make any such assumption. H(s) = H(s) .simplicity we add the odd terms and get p < 5 . The boundary is just such a tiny portion of the total volume.e.269 J . (YE  2J T In3 . This is left as an exercise. a force applied on a tiny portion of the system) will cause ( s1 ) # 0." ) 24 . (1729) 'Onsager (1944). The above proof can be improved by fixing fewer si to be + 1 . The model in two dimensions has been solved. Tc T. For. 173) 2J sinh= 1 . The following are some results (see Fig. (17 28) Ferromagnetism means that a very small applied force (e. . = 2.~ ~ " / (ie .. The series in (1724) is easily computed.
T. 173 (a) specific heat. (b) energy. The dotted lines are results from mean field approximation.306 STATISTICAL MECHANICS C/ N 0 r/ 0.2 E / JN o. 0.4 0 6. . (c) magnetic moment.81. (See Chapter 27.0 Tc is the critical point of ferromagnetism (C) Fig. is 4 J in the mean field approximation).
then H i s the same as before because the extra minus sign cancels that from J. . Near T. under slight modifications. eAB be the interaction energies between neighbouring atoms. if h = 0 . the total magnetic moment isd cosh’ 2 J / T sinh4 2 J I T M = N [ s i n h 2 ( 2 J / T ) . Other Ising Models The above ferromagnetic model. we know that in one dimension the Ising model does not exhibit antiferromagnetism but in two dimensions it does. From the results of the last two sections. Ising antiferromagnetic model If we replace J in (1716) by J ( J > 0) then energy is lowered for neighbouring spins with different signs. M/N is almost 1 . eBS. This is then a model of a binary alloy. In calculating the thermodynamic potential the positive and negative values of each si must be taken into account. the sign of J does not affect the thermodynamic potential. Therefore. In other words. B. (1730) At low temperatures. At low temperatures neighbouring spins are antiparallel. if we write every alternate si as .3. many different phenomena can be analysed by the same model. then the ferromagnetic and antiferromagnetic models are different and we need to reanalyse the problem. M a N (T.si. ( 17 . . Let e A A .ISING MODEL 307 When h = 0 and T < T.30’) 17. A .T)’” . This model can be transformed dYang (1952). This becomes the antiferromagnetic model.1]1’8 . When h = 0. . enabling us to understand the common features of these different phenomena. Model of binary alloy Let si = 1 represent an atom A at point i and si = 1 represents an atom B at point i . If h # 0. Let us give several examples. becomes models for other substances.
32) Similarly 2h J++ z K = . h . We assume that the same point cannot accommodate two or more molecules. If J < 0 and the temperature is low. Eq.34) Using (1734). C. j can be written as (1731) where z is the number of nearest neighbours for each spin. Lattice gas model Let si = 1 represent a molecule at point i. similar to the antiferromagnetic case. (17 33) J +K = EAB From (1732) and (1733) we can solve for J . Hence we have an equilibrium of different phases (assuming J > 0). and si = 1 represent no molecule at point i. then eij = eAA . At sufficiently low temperatures the atoms A will gather together while atoms B will form another collection. = 1. i. (1 7 .J . Let the interaction energy of neighbouring molecules be E . the atoms A and B will arrange alternately. first suppose both i and j are occupied by A . 2h z +K = ~ A A . si = s. h and K : ( 17 . The energy of a pair of neighbouring atoms i .e.308 STATISTICAL MECHANICS to the Ising ferromagnetic model. (1716) now describes a binary alloy. To determine the constants J . Then . J . K .
etc. ( s i >< 0 corresponds to the gas phase. The above examples show that many different phenomena can be analysed by the same model. We shall return to some of these later. So in using the Ising modele to analyse various phenomena. 2 = s. . Z 2h . longranged interaction. etc. = 1 . Eq.ISING MODEL 309 we may write the energy between two neighbouring points as (173 1) 2h J . impurities. e. The restriction of at most one molecule per site represents shortrange repulsion.3 5 ) 1 J = . Solving these. = 1 I . we must first understand its limitations.g. etc. 1 J + K = O . we get si = 1. the problems of spherical symmetry and spin waves. i. The literature is enormous. (17 .E 4 Z . Analysis may refer to phase transitions. chemists and mathematicians. (1736) h = E 4 Therefore. boundary.+ K = .e.e. (si) > 0 corresponds to the liquid phase while the low density state. The recent introduction by Fisher (1981) is a good fiarting point for references. Ising model is studied extensively by physicists. i.E . (17. There are of course many phenomena which cannot be analysed with this model. random interactions. s. where E represents the attraction between the gas molecules. Z s. Other applications include the inhomogeneous interaction.J + .16) is also a model of gas molecules. The high density state. Spin is a vector while the king variable does not contain the concept of angles.+ K = O .
The spin variable si need not be limited to only t w o values. i. 0.310 STATISTICAL MECHANICS Problems 1. i. F=TlnZ=Nh . s i = 1.1. (1740) where h is the smallest eigenvalue of K .1): First prove that (1741) 2.e. If si has n values. then Q is an n x n matrix.e.1 ). the energy being still given by (1 7 . Solve the n = 3 case. Many onedimensional models can be solved by the method of the transfer matrix if the interaction is limited to nearest neighbours. (1738) The definition of the transfer matrix Q is (1739) (b) Prove that for large N .’IT is the largest eigenvalue of Q . K can be any function of si and si+ (a) Prove that = Tr(QN) . . . Let N (1 737) In the above. (c) Use the method of the transfer matrix to solve model (1 7. .
and there is a small positive magnetic field h . If $(O) is the state vector at t = 0. Extend the proof in Sec.. then almost all the spins are + 1 . Consider any quantum mechanical model in which there are n states. then the state vector at time t is From this.2 to the Ising model in three dimensions.e.S I N G MODEL 31 1 3. 17. Then the Hamiltonian is an n x n matrix K . (Hint: refer to the section Growth of Water Droplets in Chapter 15. Discuss the subsequent motion. 5. i. T << T. a is a constant and u is the spin 9 operator. The matrix K in (1742) is like K in (1739). Thus the Ising model (1) is equivalent model in quantum mechanics: to the spin 3 K = a + b . 1 (1 743) where b is the “magnetic field”. h . it is equivalent to NIT and e V i K t is equivalent to Q N in (1738). At a certain moment the magnetic field is suddenly reversed to 4 . 4.o . If the temperature in a twodimensional Ising model is very low.) . b and J . we see that the onedimensional problem in statistical mechanics is similar to the problem of finding the zeropoint energy in quantum mechanics. Derive the relation between a .
The right half contains solute particles while the left half contains none. The solute molecules . There is no motion. Solutions and Osmotic Pressure Osmotic pressure is the pressure produced by the motion of solute molecules. and stationary impurities. Whether the arrangement is random or regular. hence no osmosis. The motion of molecules in an amorphous state often have different time scales. The impurity atoms are stationary and exhibit behaviour quite different from that of a gas. they can be classified into two types: mobile impurities.Chapter 18 IMPURITIES AND MOTION Impurities refer to foreign molecules mixed in a host substance. The pressures on the two sides are maintained by pistons. This chapter emphasises the difference between these two situations. This chapter gives a few examples to illustrate some characteristics of the amorphous state. According to the motion of the impurity atoms. We cannot discuss both in the same way. such as manganese in steel. Solute molecules in a solution have an osmotic pressure. which allows molecules of the solvent but not the solute molecules to pass through. such as oxygen dissolved in water. Figure 181 shows a semipermeable membrane dividing the liquid into two halves. A random but frozen state is called an amorphous state or simply a glass. and there is no sharp division between stationary and mobile. there is always a difference between stationary impurities and mobile ones. Examples include solutions and alloys. much like a gas. This pressure can be observed by using a semipermeable membrane.1. and the statistical mechanical analyses are not the same. The state of motion is strongly related to the equilibrium properties of the system. 18.
P' = Pgh . (181) where p is the density of the solution. 181 can also be maintained by suitable weights. 182 A semipermeable membrane separating right and left halves of a U tube. the semipermeable membrane is in the middle of the U tube. .IMPURITIES AND MOTION 313 Fig. Actually we can take away the pistons without causing any change.e. on the right cannot enter the left while the solvent molecules can move either way. The extra weight can be provided by a suitable height of the solution. As shown in Fig. The pressures in Fig. unaffected by the pressure of the membrane. From (181) the osmotic is easily determined. The righthandside requires more weight to maintain equilibrium. 181 A semipermeable membrane dividing B liquid into two parts. '1 lPtp' Fig. 182. the liquid on the right is pushed up. The excess pressure on the right is called the osmotic pressure. Therefore. i. The pistons on both sides have the same weight.
The volumes V and V’ are those occupied by the water molecules and solute molecules respectively. the solute molecules are like an ideal gas. r 2 . the result is not related to V . with a pressure given by p’ = n’T . . s’ E (Ti. (185) where s and s’ collectively denote the coordinates of the water molecules and the solute molecules respectively.. (183) where Ho includes the kinetic energy of the water molecultes. Suppose there are N water molecules and N ’ solute molecules. . Effective Interaction Energy We now examine carefully the action of the solute molecules in the solvent (referred to as “water” below for the sake of simplicity). The total energy is H=Ho+H1 . r N1 . .2. . rb) . the interaction among water molecules and the kinetic energy of the solute molecules and H1 is the interaction energy of the solute molecules among themselves and with the water molecules: (184) where ri is the position of the solute molecule i and rk the position of the water molecule k and u and v are the interaction energies of a pair of molecules. . V ’ . r. (182) where n’ is the density of the solute molecules. The total thermodynamic potential F ( N . . V . N ’ . T ) can be computed from the method of fixed temperature: s = ( r 1 . . 18. V’ being a part of V . . The kinetic energy can be immediately integrated out.314 STATISTICAL MECHANICS For a dilute solution.
in the electrostatic interaction in a medium. but we cannot derive the equation of the motion of the molecules by taking its derivative. Of course we can also define an effective interaction between the water molecules. . e. (186) (187) In these expressions Fo is the thermodynamic potential when H I is omitted. Using this concept of effective interaction energy. For example. the influence of the molecules of the medium can be represented by a dielectric constant. The sum in (185) can be done in two steps: first sum over s. Of course. and can be replaced by the effective interaction energy. the interaction between water molecules is influenced. This effective interaction energy is a very important concept. Notice that the effective interaction energy is defined for a specific use and cannot be used for other purposes. it may help in the analysis of the motion. Because of the introduction of solute molecules. while W ( s ’ . Then we can forget about the medium. the interaction among the solute molecules and the interaction between solute and water.g. This is an example of statistics by classification: we fix s’ and sum over all the configurations s of the water molecules.IMPURITIES AND MOTION 315 and V ’ . we can treat the surrounding as a medium. then over s’. T ) is the effective interaction energy of the solute molecules. When we isolate a certain part of the system (here the solute molecules) for examination. the interaction of this part with its surrounding may be very strong and must be taken into account. we can forget about the medium. and is the thermodynamic potential when every s’ is fixed. This method is very common. whose only effect is to make a suitable correction to the interaction energy. The function W represents an interaction energy taking into account the motion of the molecules. This is clearly very convenient. T ) defined here is used to analyse equilibrium phenomena and cannot be used directly to analyse the motion of the molecules. and will be ignored. the function W ( s ’ . After replacing the charge by the effective charge. The external charge divided by the dielectric constant is the effective charge.
Low Density Case Evaluation of the effective interaction energy is not easy. (184) ). we neglect the last term of (184) and each ri in the first term is independent. However. but is a more complicated object: W = Tln(e H. In addition. We now look at the low density case. The interaction between water molecules is shortranged. So the interaction v can be neglected. the water is homogeneous and the average value of (1810) does not change with r:.JT )O (189) 18. But the interaction u with the water molecules cannot be neglected (see Eq. MECHANICS Equation (187) can also be written as an average: (188) where the average value is taken over the motions of water molecules with the solute molecules fixed. The calculation (188) can be simplified as follows (1810) i.316 STATISTICAL.3. If the distance between solute molecules is large enough they become uncorrelated.e. Low density implies that the solute molecules are always far apart. So in this approximation . in some special cases approximation can be employed. Note that W is not the average of H I .
7V' av 9 (1813) because Fo is not a function of V ' and f' is a function of T only. the osmotic pressure is given by (1813)./ T ) A = h(27TmT)" . (1815) .IMPURITIES AND MOTION 317 where f does not depend on r'. and. its equilibrium properties are the same as those of an ideal gas. Putting (1811) into (186). we get ' F = Fo + N'f'  (18.. The solute molecules can freely enter or leave the water surface.12) This result shows that each solute molecule has a thermodynamic potential f'. therefore. the chemical potential is = T I ~ ( A ~ ~ .14) The chemical potential below the water surface can be obtained from (1812). So the chemical potential is the same above or below the water surface. The osmotic pressure is TN' aF P' = . Equations (1813) and (182) say the same thing. so long as the density of the solute is sufficiently low. Example 1: The pressure of the solute above the water surface is p . Except for this. We have thus reached the important conclusion: no matter how strong the interaction is between the water molecules and the solute molecules. Calculate the density N ' / V ' of the solute in the water. (1 8 . At low densities. We shall illustrate the importance o f f ' by an example. The gas above the water surface can be treated as an ideal gas. the effect of the interaction between solute and water is contained in f'.
318 STATISTICAL MECHANICS Notice that the Fo in (1812) includes all contributions due to the motion of the solute.02 cm2/sec. Let 3 be fhe observation time.. In time t the diffusion distance is 1  (Dt)' . Whether or not there is motion is a question of the time scale. If the impurity molecules do not move but remain in a fixed position. " V' P .e. the impurities are mobile if 7 > ' > Dn213 1 stationary if 3< . The result is that besides f we have the effective interaction energy between the ' solute molecules.. which is fitted by (1 819) with D o = G.) where D is the diffusion constant. 12. The impurities in solids are usually stationary.19) . 18.87 eV. From (1814) and (1815). then (1810) must be corrected. In highly viscous liquids. The exponential in (1 8.4. Mobile and Stationary Impurities The solute molecules mentioned above are mobile impurities. This point will become clearer in the next section. If the density of the solute is not very low. If n is the number of impurities. During the time of observation solute molecules move to and fro. then they are stationary or frozen impurities. < D n2I3 1 (1818) In solids. E = 0.'I3 is their typical separation. if we know f ' and p we can calculate the density of the solute in water. then . impurities also may not move. the diffusion constant of impurity atoms has the empirical dependence D = D o e E/T (1819) Figure 183 shows the diffusion constant of carbon atoms in iron. In fact the solute in water is the same as an ideal gas except for the energyf'. (1817) (see Sec. The motion of impurities can be described by diffusion.9.f ' / T T (1816) Therefore.
If N= 0. Figure 184 shows a square lattice fully occupied by two species of atoms. The impurities. The total number of atoms N is equal to the number of lattice sites. about lo4 K . there are no negative atoms. 183 Fit of experimental data on carbon atoms in iron against Eq. The interaction is limited to nearest neighbours. Suppose N<< N+ . i. the negative atoms are very few in number and can be treated as impurity atoms. say positive (black) and negative (white). are stationary. In many cases. mobile impurities and stationary impurities are totally different situations. they are constant positions. 569 Fig. with numbers. it must overcome a barrier of energy E . Here z = 4 is the number of nearest neighbours.J and that between different atoms be J . Let the interaction between similar atoms be . This is very important. If we change .e. i. N+ and N . The latter is like a solid for which the mean position of every impurity atom is fixed. In applying the basic assumptions. D is negligible.e. whether the positions of the impurity atoms are random or regular. the atoms are not free to move until the solid reaches the melting temperature. shows that in order for the impurity atom t o move.. Let us look at an example. Usually E is between 1 and 2 eV. the 1 total energy is . it is surrounded by other atoms and cannot move easily.e. The former is like a gas in which the coordinates of the impurity atoms change continually.r N z J . (1819). apart from small vibrations.IMPURITIES AND MOTION 319 1 3/ T ( K 0 Taken from Kittel (1966) p. From (1819) it can be seen that D is very sensitive to temperature. i. At low temperatures.
i = 1 . while the negative atoms are the solute.. (1820) must be modified.20) If there are nearest neighbour negative atoms. . Every pair of neighbouring negative atoms increases the energy by ... . The positive atoms are treated as a medium or the solvent. . However. . u(0) = u(1) = . . Fig. 2 . the energy is increased by 2 J z . (1821) R i .320 STATISTICAL MECHANICS . When an atom moves. . . the total energy is H0 =  1 NzJ 2 + 2JzN . In crystals this is a rather complicated . This is the Ising gas model at the end of Chapter 16. it must exchange its position with the nearest neighbour.is the position of the negative atom i . Exchange of position needs a “push”.... Therefore. . one positive atom to negative. 184 Two species of atoms occupying a square lattice. So the total energy is H = H o + H i .4 J U(Y) . The distance between neighbouring atoms is 1 . if the density of the negative atoms is very low.N.4 J . if r > 1 = 0 . we now want to discuss whether these atoms can move or not. ( 18 . m o o .
0 . From the discussion on the Ising model in the last chapter. (1 8 23) If we neglect J/T. O a . the term 2 J z in (1820) is just f' in (1812). In the above section we did not discuss the interaction between the impurity atoms. a Fig. 1 8 . then F = TlnZ . the probability that the atom in question will move to the left will be enhanced (see Fig. while the neighbour on the right gives a push to the left. (1 8 24) . Here we only suppose the existence of such processes and assume that such processes occur very quickly. . and the atom in question moves up a bit. a . The only difference is that the position of the atoms are restricted to the lattice sites. Let F be the thermodynamic potential at constant temperature. . a+ a . We only consider the situations before and after the process and do not ask what happens during the process. Every atom is continually making a small vibration. . 185). Now let us make a simple calculation. H I in (1821) has an important effect. . Hence. the result is that of an ideal gas F N Fo = HoTSo . a . . a . This process is a special topic in itself. C . segregating from the positive atoms in order to reduce the energy. 0 . If the neighbour on the left moves down a bit. we know that at low temperatures the negative atoms will coalesce together.IMPURITIES A N D MOTION 321 process. In this simple model. . this problem is similar to the problem of the solution.5 If the negative atoms move frequently.
mined by cooling from high temperatures. . the probability that its neighbour is a negative atom is (18. If the negative atoms d o not move. e. e. 14. at T = To .g. the situation is totally different. (1 827) The function H is defined from (1821) and (1820). R N _ )= H . the distribution of the negative atoms reflects the distribution before quenching.1).g. Equation (1 827) is quite different from (1 824) and (1 825). Different processes of cooling will result in different distributions of the positions. F’ in (1827) can be simplified. We simply write down the results. among the neighbours of each negative atoms there are on the average z p . Each R j is a fixed quantity and there is n o change in configuration. For any negative atom. . . Notice that no matter how randomly the negative atoms are distributed. .322 STATISTICAL MECHANICS If the concentration of the negative atoms is low. So the thermodynamic potential is just F ’ ( R . So the temperature does not influence the entropy and the specific heat. TN* v F . At that time. because there is no motion. when a body is cooled suddently (socalled quenching) from a high temperature To at which the negative atpms can move. leaving the details to the reader. Once the Riare frozen in. (1 825) In the above we assume that the temperature is not too low and the negative atoms do not coalesce. .26) Equations (1 823) to (1826) are the results of the motion o f the negative atoms. and the value of F’ is controlled by the predetermined positions of the negative atoms. showing that the influence of the impurities depends very much on the mobility o f the impurity atoms. the entropy is not affected.1 2N (e3J’T  1) T N* +2N . they will not be influenced by the changes of temperature (provided they are not heated to the temperature of motion again). But these positions are related to the process of cooling.( T o ) negative . Under special conditions. In this case. we can use the low temperature expansion (see Sec. predeter.
1=  1 Nz(3J) 2 Ne3J’T N .) . the value o f H . and we must call this model a “mixture”. But the basic concept is the same. If each atom is stationary but randomly distributed. but its actual value. we cannot apply the simple basic assumptions. This case is very difficult. In the next section we shall give an introduction to the amorphous state. There is no fluctuation in F’ and hence no average.5. We have to take into account the duration of observation as well as the details of the motion. The extent of motion is determined by the motion. We assume this average value to be zp. both long and short. the result will be more complicated. the term “impurity” will not be appropriate. Therefore. Rather.IMPURITIES AND MOTION 323 atoms. Glass is a good example. The concept of an amorphous state does not require the mixing of different molecules.( To ). (1826) for p . . If they are stationary. The above discusses two cases: mobile and stationary.e. but neither are they absolutely stationary. F ’ . 18. To summarise. (See Eq. This situation is the most troublesome and occurs in the amorphous state. This has already been emphasised in Chapter 8 when we discussed the hydrogen molecules. ice is another. That is to say the problem is not an equilibrium problem but that of nonequilibrium. In the above example. The Amorphous State The main point of the discussion in the above section is on the distinction between motion and immobility. Any random but frozen structure can be called amorphous. The average number of negative neighbours of a negative atom is statistical. (1828) Notice that (1828) is not just the average of F ’ . there is a range of time scales. If the range of R j is large. obtained from Nnegative atoms. then the trajectory is just a point. if the impurity atoms are allowed to move. The amount of impurity atoms is not important. we call this the amorphous state. i. is 3 R. What happens if the time scale falls between these two? The impurities d o not move very far. The same type of molecules can have random but frozen structure. if the numbers of the positive and negative atoms are comparable. the trajectory occupies a certain region in configuration space. In many substances there is no well defined unique time scale. the extent of this region becomes larger. No matter how long the observation time. If the numbers are comparable.
324 STATISTICAL MECHANICS Most solids in nature are amorphous. Yoshida (RKKY) interaction. The use of statistical mechanics to analyse amorphous materials started about ten years ago. a is the lattice distance and eF is the Fermi energy of the free electrons.a The most important characteristic of this interaction is that it changes sign with R . The theory is still in its infancy. A is a constant given by (1830) where K is the interaction energy between the spin and the spin of the electron. For detailed derivation see Mattis (1965) p.198. Kittel. and the electrons in turn influence spin 2 . the spins of two manganese ions (simply called the spin below) mutually interact. Some manganese atoms are dissolved in copper and then cooled to low temperature. This is the socalled RKKY interaction. which we shall not discuss. That is. This mutual interaction is approximately 4. the amorphous state is an important topic. This characteristic comes from the correlation function of the spins of the electrons. this is the socalled Kondo problem. which is approximately proportional to J ( R ) . Because of the interaction of the manganese ion with electrons in the metal. spin 1 influences the electrons. The fluctuation with period r / p F is due to the sharp cutoff of the free electrons distribution at the Fermi surface (see Chapter 13). because the interaction between s1 and s2 is mediated by the free electrons. This formula is correct for low densities of the manganese atoms and at low temperatures. . In this formula pF is the Fermi momentum of the free electrons of copper and R 1 2is the distance between spins 1 and 2 . The most widely studied material is manganesecopper alloy. a Ruderman. The following is an example that illustrates some concepts related to the amorphous state. In modern materials science. The problem of amorphous magnets (or spin glasses) has received much recent attention. either containing frozen impurities or possessing a random structure. 193. The electrons in the outermost shell of the manganese atom mix with the free electrons of copper and the remaining manganese ion is then fixed with spin This spin interacts with the spins of the electrons in the metal and peculiarly influences the motion of the electrons in the metal. Kasuya.
The above assumes that the vapour pressure of the solute is very low. Therefore. Experiments are usually performed for T . The experimenters are interested not in the random distribution of manganese. This amorphous phase has many interesting properties. . Such changes show the existence of different time scales of motion: some are fast. some spins will become frozen. but other quantities also follow this logarithmic decay. Problems 1.s 2 0 K and the concentration of manganese below 10%. Not only the magnetic moment. and the total magnetic moment is zero. In a certain solution. these spins form a paramagnetic body and each spin can freely rotate. the mutual interaction J ( R i i ) between the spins is positive for some pairs and negative for others. causing a random interaction.aTln(t/T) . This is the characteristic of not only the amorphous magnets but also of other amorphous systems. The frozen directions of the spins are random. the magnetic moment M does not vanish immediately when the magnetic field is removed. Prove that the vapour pressure of the solution p + A p is lower than that of the pure solvent p by an amount (1832) where p is the density of the solution and p is the density of the vapour of the g solvent. if we magnetise with a magnetic field. This is the amorphous phase of the spins. If the temperature decreases further.e. the most important being its “remanence”. more spins will be frozen out. When the temperature is lowered below a certain T o . i. Note that the position of every manganese atom is fixed and so every J ( R i i ) is fixed too. also called the spin glass phase. but decays gradually M ( t ) .M(0) . (1831) where a is a constant and T is a microscopic time scale. we shall analyse the problem again in the discussion of dynamics. but in the random interaction of the spins due to this distribution.IMPURITIES AND MOTION 325 Because pF is very large and the distribution of the positions of the manganese atoms is random. while some are in the intermediate case. the osmotic pressure of the solute is p ’ . At high temperatures. some change so slowly that they may be regarded as frozen during the time of observation. To understand the amorphous state we cannot avoid the problem of nonequilibrium.
7 . how high can we raise the sea water? The reader will find that this is quite a considerable number. (This idea was learnt from oceanographer J. Isaacs.326 STATISTICAL MECHANICS  Fig. 3. The fresh water generates electricity by flowing under the dam and then flows to sea via the semipermeable membrane. 1 8 . semipermeable membrane Iriver water Fig. If we use a semipermeable membrane. An estuary generator: Figure 1 8 . Calculate the osmotic pressure of salt in sea water.) Discuss the feasibility of this construction. 186 2.7 shows two dams built acrossanestuary.
J I = . and H have the same structure. Answer: h ’ = O .! s.e. . and h is a constant. i. . s. Derive F under the assumption that 2 J / T < 1 . .N 2 . The numbers ni can be regarded as the number of impurity atoms. .1 N si i=l . Assume that h = 0 . . s =s3. s3. (1833) can be transformed to another similar model. As in the above problem.e. (1835) (Assume N is even. i. The onedimensional Ising model H = J 1 sisi+l . if there is one impurity atom between si and . 1 N’ = . Calculate h ’ . h’ and Fo . .) Obtain J ’ . s6 . Let N = 2 m . The function H in the above model becomes a constant after m transformations. 5 .1T l n ( c o s h F ) . H . but take h # 0. (1834) Write HI as N’1 H i = J’ 1 s. 2 (1836) This model is very simple.h i=1 N. Assume a onedimensional Ising model of ferromagnetism to be described by: (1837) where n i = 0 or 1. be part (1) of the body and s 2 .=s 5 . . J ’ and Fo 6. the thermodynamic potential F. The steps are as follows: (a) Let sl. s5 .IMPURITIES AND MOTION 327 4. (b) Let s =sl.!+~  h i=1 i=l 1 s. 7 . . . .! N + F. s 4 . be part ( 2 ) of the body. Write an equation for H l where H i is the effective interaction energy of part (1).
e. calculate the same quantities.328 STATISTICAL MECHANICS N 'i+i the interaction is weakened. i. Let" = 2 i=l ni. they can move to and fro. (a) If the position of the impurity atoms are not fixed. specific heat and magnetic susceptibility. (b) If the impurity atoms are stationary and distributed randomly. . calculate the thermodynamic potential. entropy.
then the electrostatic potential of a collection of particles is (19. This gas is a plasma above a certain temperature T o . This model is quite important in current work on statistical mechanics.1. and we introduce in a simple way the experiment on electron crystals in two dimensions. Its main characteristics can be explained by quite simple analysis. Short Distance and Long Distance are Both Important The electrostatic interaction energy is inversely proportional to the distance between the charges and does not have a characteristic length scale.Chapter 19 ELECTROSTATIC INTERACTION The electrostatic interaction is very strong at short distances and is weaker at large distances. The method of analysis is to examine the response of the particles to a stationary charge and introduce the screening effect. therefore. Below To the positive and negative ions coalesce into pairs forming dipoles. 19.1) . the ionic solution. electrons in metals. cannot be analysed by a simple expansion. This chapter uses some examples to illustrate the characteristics of the electrostatic interaction. Let ei be the electric charge of the ith particle. However. in which the interaction energy between the charges is 4' In r. emphasising the long distance effect. The examples include the plasma gas. Lastly we discuss an abstract Coulomb gas model in two dimensions. where q is the charge and r is the distance between the two charges. it can interact with many particles at the same time and.
19. then its kinetic energy is increased. there are many problems directly related t o the electrostatic interaction.) Therefore the discussion of the electrostatic interaction must either be coupled with that of other effects. otherwise we must use approximations to avoid the instability mentioned above. . Thus condition (b) is not enough.range interaction. (192) will be larger than the potential energy e 2 / r . (b) kinetic energy is quantum mechanical. So. otherwise the electrostatic interaction alone cannot describe stable matter. condition (c). (192) where r = extent of the position. i. It has to be supplemented by the exclusion principle. We shall follow the method in Chapter 13 (see Sec. In Chapter 9 we mentioned that for the thermodynamic limit t o be established. the necessary conditions are (a) the total charge is zero. Also suppose that there is a repulsion at short distances preventing the charges from approaching each other indefinitely. the main characteristic of the electrostatic interaction is its longrange effect.2). In statistical mechanics. If U is positive then the charges will flyoff to infinity to lower the energy.2. The Plasma Gas and Ionic Solutions Suppose we have N particles with positive charge e and N with negative charge e. Because l / r is still significant even when r is large. Condition (a) means that uncompensated charges will be dispersed (to the surface of the body if possible). and (c) at least one type of charge (positive or negative) is fermionic. The discussion below stresses this long. we can enlarge or contract the body. The requirement of quantum mechanics is this: If the position of a particle is restricted. 13. (See Sec. Now consider the density correlation function of the particles and the thermodynamic potential. kinetic energy  h2/mr2 . Condition (b) avoids the coalescing of all the particles into one point. a particle can interact simultaneously with many other particles. First put a positive charge e at the origin and see how the other particles are distributed. i.330 STATISTICAL MECHANICS Since there is no length scale. i. Obviously other effects must exist.e. i ej = 0 .2. 19.e.e. Let rij = then U will acquire a factor l / h in the new scale. In these problems. if r is too small. If U is negative then all particles will coalesce into one point and U + w.
( r distance between the particles nl13 . Hence the charge at the origin is screened and its effect outside b tends to zero. l/b’ = 87re’ nfT . Around the origin there is a cloud of charges of density  e erlb 47rb’r (196) This is the screening layer. (194) Substitute (193) into (194). and we get V’$ = $/b’ .e . Let @(r)be the electric potential. Then rr T n e$(r)/T + n . (193) where n is the average density of the positive or negative charges The relation of $(r) with the density of the total charge is v’ $(r) = 47re(n+(r) . Notice that from (195) we have (87re2 n’13)/T = l/(b3n)’13 . (195) with the solution $(r) = . (197) The lefthandside is approximately the ratio of the interaction energy e 2 / rto the kinetic energy T. Now we have a new length scale b. The integral of (196) over 4nr’dr is e.) The denominator on the right is the number of particles in the screening layer raised   .n(r)) . An increase in temperature increases b and (195) is more accurate.ELECTROSTATIC INTERACTION 331 Let n+(r) be the densities of positive and negative charges. a charge at the origin produces a potential elr. the socalled screening length. Hence the effect of the particles is t o reduce the electric potential far away.r / b r If there are no other particles.
Originally the electrostatic interaction had no length scale. etc. The functions n. (195)). (1320)). Therefore. 2 F =Fa +U 3 I( = pa + UfN . and the new scale b is a function of density. but remains as a low density approximation. From n.1) : T =  V 7 87rb  e2 N b (199) From this result we can calculate the specific heat and thermodynamic potential. the above approximation is valid when b3n is large.( r l O + ) + (see Eq. Notice that Readers are invited to prove the following: S = So + 3U f T . As in Chapter 13. (r) are conditional distributions. (1911) The subscript 0 means quantities evaluated when e = 0. temperature and electric charge (see Eq. The condition is that there is a charge at the origin. (r) the density correlation function can be calculated (198) It can be seen that the correlation length of this gas is b . 1 . . From (1 9 8) we can derive the interaction energy (1 9 . it should be denoted by n .332 STATISTICAL MECHANICS to the $ power.
because E = 88 is quite large.8 x 10'2)2 erg cm. Electrons in Metals The analysis of the above section can be applied to electrons in a metal. (2) the density of electrons is very high.6 x lOZ0/cm3 . The major differences are: (1) the positive charges (nuclei) in a metal are more or less stationary. e2 = (4. we can ignore the temperature. because the water molecules have an electric dipole moment. the charge of the ions has already been screened partly. and the electrostatic interaction energy e 2 / r e2pF is much larger than T. The pressure p becomes the osmotic pressure (see Chapter 18). the above results can be directly applied. Therefore in discussing the electrostatic interaction. see books on physical chemistry. In water.ELECTROSTATIC INTERACTION 333 If these charged particles are the solute in a solution. If (197) is very small. E = 88.38 x x 300 erg. Na+ and C1in a saline solution. In water.E ) ~ ] (1912) In the above formula. the electrostatic interaction between the ions is reduced considerably. we get n << 1. The analysis in this section is the socalled DebyeHuckel theorya for ionic solutions. In water E = 88 and the quantities in (1911) now refer to the ions.n'I3 a .3. e. we have written out the Boltzmann constant k explicitly. 19. eF >> T . so e2 must be divided by the dielectric constant E . (1914) aFor a more detailed discussion on DebyeHuckel theory. The only correction is the electric charge. . then n << ( I c T ) ~ / [ ~ T ( ~ ~ / . (1913) This limitation is not too harsh.23 x molecules.  . The ratio of the interaction energy to the kinetic energy is (197).g. Putting in numerical values k T = 1. The ratio of the interaction energy to the kinetic energy is  a 2 2me2 ( e 2 / r ) / e F= PF a=h/me 2 2 . since each mole of solute contains 6.
i.53 A). the distance between the electrons ( n being the density).4 n e n ’ ( r ) . the density of states.e @ . = . the energy at the top of the Fermi sea is the sum of the potential energy plus the kinetic energy. Let @(r) be the electric potential. . Let eF be the kinetic energy of the electrons at the Fermi surface. and hence the electrostatic interaction of the electrons in a metal is not weak.e. i. (312) ) Here g o ( € )is the density of states when @ = 0 and go(€ + e @ )can be interpreted as the density of states at r. In the following analysis we still assume it to be small for the convenience of calculation. this ratio is approximately equal to n . divided by the Bohr radius a (i. The energy level is lowered by e@because the potential is now . and the remaining density n’(r) is the density distribution within the potential e@(r). The equation for the electrostatic potential is v’Q.334 STATISTICAL MECHANICS Therefore. The nuclei are stationary and hence do not respond.e. We place a charge e at the origin and see how the electrons respond. At T = 0 this is the chemical potential. the density of the electron minus N/V. The above assumes @(r) to be a slowly varying function of r .e. To calculate the energy distribution of the electronic states.e.l I 3 . the radius of the hydrogen atom is 0. Then i. We treat the stationary positive charges as homogeneously distributed. N being the total number of electrons and V the volume. This ratio is not small. so their charge just cancels the mean charge of the electrons. The result obtained is therefore not very accurate. (1915) where n’ is the excess electron density. we can use the short wavelength approximation (see Eq.
and the ratio a is defined in (1914).ELECTROSTATIC INTERACTION 335 Now assume e @ ( r ) to be very small. (If the distribution is not homogeneous. ( 19 20) ~ then we need not use quantum mechanics. This has been achieved in two dimensions.) This homogeneous arrangement of the positive charges with mobile electrons is not realisable in three dimensions. Now we present the relevant experiment.4. we get V 2 @ = @/b2 . then the density of electrons at r is n'(r) + n = J P dego(e+ e@) . the problem would be entirely different. in most metals (Y is not small. but regular as in a crystal. Substituting (1918) into(1915). in the laboratory we can use other effects to provide this homogeneous positive background. because this ensemble should become a crystal at low temperature. 2 libZ = 4e = 2 mpF/n = 2apF/n 2 6ne n/eF . (1919) This result is somewhat similar to (195). However. The above analysis assumes a to be very small. Notice that we have assumed that the positive charges form a homogeneous background. . However. 19. Electron Crystals If the density of the electrons is so small that eF << T << e 2 / r . Therefore the above result only tells us that the electrostatic screening is very strong and the effective distance of the electrostatic interaction is very short. 0 (1918) The factor g o ( p ) is just the g(0) in Chapter 4.
with a slight effect on the motion of the electrons.7 X position above the liquid surface is 100 A . there is some oscillation in the liquid surface. In fact the liquid surface attracts the electrons slightly. The density of the electrons can be controlled (from lo5 to 109/cm2). i.1 bThe binding energy of electron and the liquid surface is 0. However. Some electrons are sprayed onto the liquid surface. Positive charges are introduced into the bottom of the container. so that the electrons adhere to the surface. while the electrons on the liquid surface form a twodimensional system. The experimental procedure is to apply a weak oscillating electric field in the perpendicular direction. The electrons can move on the liquid surface and interact via their electrostatic repulsion. The vibrational frequencies of a crystal are related to its structure.b Because the temperature is very low. A flat container holds some low temperature liquid helium (about 0. because they are blocked by the helium atoms. If the temperature is sufficiently low. This is theoretically very close to a collective body.e. these electrons will be arranged into a crystal. eV. They serve two functions: on the one hand they pull the electrons downwards.336 STATISTICAL MECHANICS t ++++++++ Figure 191 is a schematic diagram of the experimental setup. Of course. This pulling can cause oscillation in the liquid surface and shake the electrons. There are no impurities.5 K). periodically pulling the electrons against the liquid surface. The average . the electrons are very light and their wave nature keeps them slightly above the liquid surface. the vapour pressure of helium can be neglected and the space above the liquid surface is more or less a vacuum. these electrons are the objectsunder observation. If the frequency and wavelength of the vibrating surface are the same as those of one of the characteristic vibrations of liquid He +tt++t+tt+ tt 1111 Fig. 19 . on the other hand they provide a uniform positive background. The electrons cannot enter the liquid. How can we prove experimentally that this crystal exists? The answer comes from the vibration of the crystal.
ELECTROSTATIC INTERACTION 337 the crystal. Kosterlitz (1973). Now we consider a collection of such charges. then @ ( r )is the velocity potential and V@ is the velocity.3 inapplicable. superfluidity and melting.5. The charges 4i can be + 4 or 4. The model (1923) has received much attention recently. and now these models are named after them.c 19. etc. to be greater than a. because it bears a dose relation to many twodimensional models involving magnetism. The analysis of these models was started by Kosterlitz and Thouless. resonance will occur and energy will be absorbed from the electric field. (1921) where @ is the potential due to a charge at the origin: (1 922) and a is a constant. In electrostatics we encounter the twodimensional formula v2@ = 2 n 4 q r ) @(r) = . rii. . 19. If we call (1921) the electrostatic equation. In fluid mechanics if we have a vortex of strength g a t the origin. Kosterlitz and Thouless (1973). then (1923) can rightly be called a collection of charges. 'Grimes and Adams (1979) is a good place to start. The reader can consult the literature. This collection is of course not the same as that in the last section. Two.2 and 19. The preceding section discussed an interaction e z / r .4 In ( r / a > . There is extensive experimental and theoretical work on this problem. The ln(r/a) interactionjs not only a long range interaction but increases with r. Such strong interaction makes the analysis of Secs.Dimensional Coulomb Gas Model Now let us discuss a collection of charges in two dimensions. From the frequency of the resonance we can deduce the existence of the crystal structure. or vortex lines. The interaction between them is (1923) and we assume that the distance between the ith and jth charges. .
when r is very large. 19. (19 . The probability distribution of the dipole separation r‘ in a molecule is (1925) Therefore.2 and 19.3.338 STATISTICAL MECHANICS Kosterlitz and Thouless were able to pinpoint the main characteristics of this model. then we have no free charges but only dipoles. Let us now estimate the number of these large molecules.2. If there are many such molecules. (1926) . We now ask: Will there be screening? (Screening means total screening. with their other halves too far away to have much effect. the positive charges ( 4 ) and the negative charges ( .4 ) will combine in pairs forming ‘‘molecules’’ or an “electric dipole”. then within the circle it will be like a collection of many single positive and negative charges. with one end within the circle and the other end outside. The dipoles near the boundary may cross the boundary.2 and 19. To create screening. these single charges will create screening. 19. no screening occurs because the dipole is totally in the circle and cannot change the charge within the circle. the molecules will be broken up. This positive charge produces an electric potential @(rj.24) and E (the dielectric constant) is unchanging. 19. This feature can be understood as follows. centred at the origin where 4 resides. From the experience gained in Secs. namely: if the temperature is low. the number density of molecules larger than r is jmdr’ 2713’ f ( r ’ ) r .) If. then we have screening. If l / e + 0 when r + 00. i. Starting from the method of Sec.3. we must have many molecules bigger than r. (Note that 4 / e is the unscreened charge. If the molecule is small (the dipole distance is small). we put a positive charge at the origin.) Figure 192a shows a circle of radius r. but this only causes a finite E .e. The total charge within the circle is 4 / e . the positive and negative charges all combine in pairs. If the temperature is higher than some T o . like that in Secs.
. Therefore when T < To = q2/4 . But when the temperature is not close to To this correction is not important.4q21T . 22‘ are more or less like single free charges. We shall calculate the . Let us now calculate the dielectric constant E for T < T o .1’ t2 I I Fig. Figure 192 shows a molecule with the negative charge inside and the positive charge outside of the circle. otherwise there will If this number tends to zero as r be. The number of large molecules with one end within the circle is approximately rr2 jwd r ’ 2711’ (rt/a)*2’T r .ELECTROSTATIC INTERACTION 339 I I . Suppose each molecule is much smaller than r . But there may be molecules between the dipoles. (1928) no screening occurs. + 00. but big molecules like 11’. The above analysis needs some correction.e. In (1925) we used q2 In r/a as $J. so this rj must be corrected. i. (19 27) there is no screening. 192 Small molecules can be looked upon as dipoles. the positive and negative charges combine in pairs.
(1 9 32) The average value of r l 2 can be calculated from ( 1 925) jm f(r') 0 r' d r ' This formula cannot be used when T is close to To . is assumed to be very weak. Averaging over r' and 6' and using (1929). the excess charge within the circle is n ( . (1930) If cos 8 is negative. Assume r' << r. then the probability distribution of the angle 8 is approximately given by e qEr' cos0IT I s. 192b will distribute uniformly between 0 and r'cos8. E = 1 + nn(rr2)q2/T .e. (1 9 . we know that 2nrE = 2 r r q / ~so . then the distance A B in Fig. it is distributed in a ring around the boundary with width r' cos 8 and area 2rrr x r' cos 6'." d B eqEr' cos0JT 1 .( 1 t q E r ' c o s B / T ) n 1 .340 STATISTICAL MECHANICS total charge within the circle for thesemolecules spanning the boundary.29) where E . the electric field at r. Suppose the molecules are homogeneously distributed with density n .q ) x 2nr x r ' c o s 8 . Therefore.we obtain the charge within the circle (1931) From Gauss' law. i. it indicates a positive charge in the circle and a negative charge outside.
n ( x ) is relatively complicated. n ( x ) + N / V . Consider the nuclei in a metal as a collection of homogeneously distributed positive charges. Near x = 0. (c) Impose the condition 2 = I d3r n(r) . n ( x ) + 0 . (b) From (1915) obtain the differential equation for @ .e. As x + m. (a) Express n ( r ) in terms of p . 19. The potential $ ( r ) and the electron density n ( r ) both have spherical symmetry. then we can use the method of Secs.2. eYoung (1979). A heavy atom has many electrons. [ n ( r ) a ( p . (1935) to derive p . where Z e is the charge of the nucleus. the density distribution of the electrons. and@(r). Model of the metal surface. The reader can prove that when Y is very large l/b2 = 4nnq2/T . . because as x + m. (1934) This is similar to (195). The nucleus resides at the origin.(1 91 7). Use the method of the preceding problem to solve for n ( x ) i. 2.e Problems 1. This is clearer than the paper of Kosterlitz and Thouless. These are very simple results and require no complicated analysis. a numerical procedure is necessary. The reader may refer to the literature.ELECTROSTATIC INTERACTION 341 If the temperature is high. then calculations of a slightly more complicated nature would be required. The statistical model of an atom (the FermiThomas model). If we want to study the properties for T close to T o . showing the property of a plasma gas.e$)3’2]. whose density can be derived from (1 91 5) . For x < 0 the density is N/V and for x > 0 the density is zero.1 and 19. The above question is not easily solved analytically.
if we require that the distance between the particles is not less than a . although the system is bound. then Z is still defined at low temperatures (see Eq. Note that q 2 / 2 is 2To (see Eq. Calculate the zeropoint energy of the electron and its mean position. 5. (b) Solve for the energy and the wavefunction of the electron.q at r . Hint: Suppose there is an image charge q ’ in the insulator q’ = . i. So the conclusion that r is finite is untenable. Z does not exist. between TO and 2 T 0 . i. (b) Prove that for To < T < 2T0 the dielectric constant E does not exist. (c) Substitute the E value of liquid He4. At low temperatures. Review Problem 7. assuming that the electron cannot enter the insulator. a pair of positive and negative charges are separated by quite a large distance. obtain the potential energy U ( x ) of the electron at x . Let T < 2 T 0 . the positive and negative charges combine together . So other charges can come in between. (a) Let the charge q be situated at the origin and . not T o . (1928)). Charge q is situated outside an insulator (see Fig. 193 4. = Therefore. This problem shows that when T < q 2 / 2 . 193). 7. In three dimensions the Coulomb gas does not have a critical temperature like To . The analysis for this range of temperatures is very complicated. The content of the preceding problem is that when T < 2T0 the positive and negative charges form a bound state. Obtain the force on the charge.1 1. (1923)). The insulator has a dielectric constant E . (a) Write down the wave equation for this potential. This is very obvious from (1 933). ( r2 ) + 0 0 . Nevertheless.e. these two charges are a bound state. 6.342 STATISTICAL MECHANICS 3.e. Prove that the probability for r to be finite is 1.4 E € + 1 1 (1936) Fig. From the result of the preceding problem.
Taking quantum mechanics into account. discuss the ionic solution and the conclusion of Sec. It is a continuous change from the combined molecules t o the plasma gas. 19.ELECTROSTATIC INTERACTION 343 while at high temperatures they dissociate into a plasma state. the negative charges the electrons. Under what conditions can these particles be considered as a plasma gas? Try to perform a rough analysis. . (b) Let the positive charges be the nuclei. (a) Let the interaction between q i and qi be qi = +e .e. how many atoms are ionised) in terms of the temperature and pressure. express the degree of ionisation (i.2. (1937) Let the average density of the particles be n and nu3 << 1. (c) From the above results.
.
Chapter 22 discusses methods for obtaining the equilibrium properties directly from the motion of the molecules. These methods rely on the use of computers. such as sound waves and heat conduction. This part focuses on the motion. according to the nature of the motion. The series of examples in these two chapters. and are a great achievement of modern physics. emphasising metastable states. molecular motion determines everything. But we have emphasised throughout that this assumption takes on different forms under different situations. We use it to discuss some phenomena of small fluctuations. the effect of which serves only to establish the validity of this assumption.PART V DYNAMICS The basic assumption of statistical mechanics (Chapter 5) is a very effective tool for analysis. but with a view to further the understanding of the equilibrium states. through phenomena displaying slow changes. The assumption says nothing directly about the motion of the molecules. . serving not only as computational tools but also as aids which permit a deeper understanding of the basic meaning of statistical mechanics. In the end. The Boltzmann equation (Chapter 20) is most important. enable us to understand the meaning of equilibrium. Chapter 2 1 discusses the diffusion equation.
But it is extremely difficult to derive it from Newton’s laws or the other laws of mechanics. The examples include the plasma oscillation.Chapter 20 THE EQUATION OF MOTION FOR GASES The equilibrium properties of a thermodynamical system can be explained quite satisfactorily by the basic assumption of statistical mechanics. while collisions are rapid. Therefore. The equation of Boltzmann and his hypothesis concerning entropy are unparalleled achievements in the history of physics. as demonstrated by the examples in this chapter. is the basic tool for discussing the motion of an ideal gas. Manybody motion is extremely complicated and the ideal gas is n o exception.e. scattering by impurities and the propagation of sound waves. zero sound. Regular motion is slowly varying. This chapter uses some examples to illustrate the meaning and application of the equation of motion. The dynamical equation of motion for gases. . Finally we mention the H .theorem. is one of the few useful tools in this respect. The Boltzmann equation averages out the fast. Flow and Collision The equation of motion of gases.1. irregular collisions. 20. The theory of nonequilibrium phenomena is but in its infancy. the scales of time and length of interest in this equation are much larger than the collision time and the molecular size. The following is a simple argument leading to this equation. the Boltzmann equation. called the Boltzmann equation. This equation makes a clear separation between the effects of irregular collisions and the effects of regular motion. This equation can be written down by intuition. nonequilibrium properties are beyond this simple tool. i. However.
p) and that scattered out of (r. Equation (201) is the generalisation of the conservation equation of particles in three dimensions: +v. (20. p. so that (v. This is a distribution in the sixdimensional (r. and a is the acceleration.THE EQUATION OF MOTION FOR GASES 347 Let f(r. Let the number of particles be a conserved quantity. (202) where p is the density of the particles. (204) where R is the rate of the twoparticle reaction p + p f + p" + p" containing '.vp ap at = 0. Equations (201) and (202) assume that particles d o not jump from one place to another. p) space. p) space. m a ) is the velocity in (r. then the momenta of the particles can suffer abrupt changes. t ) be the distribution function at time t of acollectionofparticles. then the equation of probability conservation is af a + at (vf) +a aP ar . If we take collisions into account. The correction term on the right is the difference between the number of particles scattered into (r. the positions of the particles still cannot change abruptly. . p : ) R a 6(p+p'p"p"') ~(E+E'E"E''') . but only flow continuously.(maf) = 0 . so (201) will not be correct.) To take account of these abrupt changes.1) where the velocity v is p / m . (However. we have to add a correction term to (201): (203) Here F = m a is the external force on every particle.
(206) (see Sec. . p)). E = p 2 / 2 r n + u(r) . hence (204) and (206) are zero. or 1 / ( e a + P e + 1) to make (206) vanish. (F = au/ar) .348 STATISTICAL MECHANICS 6 functions for the conservation of energy and momentum. the Boltzmann equation can be used to calculate any physical quantities. to make (204) vanish. Then we can get the equilibrium distribution of the particles. then the equilibrium properties will be those of the ideal gas. we pointed out that the value of R for the forward and reverse directions of the reaction are the same. e. stand for f”’ = f ( r . For the righthandside to vanish. t ) .) Notice that if f is a function of the energy E of the particles alone.ff’ must be changed to f”f”‘(1 f)(lf’) . and the functions f. However. f d3p d3r/h3 is the number of particles in d3p d 3 r . I f f < < 1 . . i. For fermions at high densities. 3.e. (205) In the discussion of detailed balance in Chapter 3. (20 7) i. . Notice that (af/a t )c is not related to equilibrium.2. the terms f”‘ f” . (See Chapter 3. p ” ’ . f(r.f’ .g. then the lefthandside of (203) is zero. collisions do not change f. If F is purely an external force. etc.f(r. Nevertheless we may regard f a s a function of F in order to incorporate part of the interaction.ff‘(1 f”)(lf”’) . then quantum effects can be ignored. p) = f ( E ( r . in d3p d 3 r there are d3p d3r/h3 states. and the factor 1 / h can be absorbed into the definition off. In equilibrium.e. p ) = f ( e ) . f ( ~ ) must take some special form. including the equilibrium ones. equilibrium quantities should not depend on r .) Each f is the number of particles in each state in phase space.
electrons in two dimensions. then e af + at v.0 ar af . ) 20. In a volume element d 3 r and a time interval d t . assuming F = 0 and no collisions. We first examine some cases with (a f / a t . Example 1: A gas at temperature T is concentrated at the origin at time t = 0 . Calculate its subsequent density distribution n(r. If one particle is at r at time t .THE EQUATION O F MOTION FOR GASES 349 The lefthandside of (203) describes slow and gradual motion. and the righthandside the effects of fast collisions. Every collision can be regarded as happening instantaneously and at a point.. its position is r .4) which are assumed to form a dilute gas. p ) = N(21rrnT)~’* . Now we consider a slightly more complicated example.Plasma Oscillations We start from the simplest example. (see Sec.v t . then at time t = 0. (a f l a t ) . Case of No Collisions . 19.1 1) The gas disperses because the velocities of the molecules are different. (208) The solution of this equation is where @ ( I .p ) is the distribution at t = 0. d t d 3 p d3r must be a large number. @ ( r . (2010) (20 . Example 2: Plasma oscillations (planar case).P 2 / 2 m T 6 ( r ) e Therefore. and then consider some cases with (a f / a t )c # 0 .2. = 0. Let . that is. . t ).
From (2012) and (2013) we get i(wk.14) Notice that Jdzr l e .12) We now make the following assumptions: f fo = fo + f ' eiwt+ik.= vfO/T af o aP From (20 . Then. (2016) . I r . n' are also plane waves with small amplitudes.2n k ' (2015) .i k . t ) (20 . @(r.t) = eV4 . F(r.v)f' + i 2ne2 n' k k*vfo/T = 0 .350 STATISTICAL MECHANICS n'(r.r ep212 m T > = n(2nmT)' * 9 f ' << f o (2013) The other quantities 4.r'l n' (r'.14) we get . t ) be the density distribution of the electrons minus the average NIL2 where L2 is the area of the plane. (20 . t ) =  I d'r' e J r.
(See problems 1 . but F in (2012) does not include these changes.. but also to electron crystals on a plane. the oscillation frequency is proportional to Jk. It only includes the slowly varying n’(r. and with imaginary terms O ( k 3 I 2 ) representing the rate of attenuation.THE EQUATION O F MOTION FOR GASES 351 Therefore. Notice that the F in (2012) is an averaged force. but it does not take into account the density correlation of the particles.19) is this: it comes from the electrostatic interaction between the electrons which produces a very strong restoring force similar to the elastic forces in solids.19) applies not only to a planar electron gas. Because of the motion of the molecules. ~ ) ~ k o2 fo/T +.. the forces acting on the molecules will change abruptly. The result of (20. only low temperature liquid 3He satisfies this condition. and the expansion in terms of k / o is feasible. Zero Sound Zero sound in a Fermi gas is similar to the plasma oscillation discussed in the preceding example. The above has not considered collisions but even if there are collisions. 1 = ( k . we can expand in terms of k v/ w on the right and get . k / o is indeed small. . When k is very small. Of course the purpose of (a f l a t ) . 3. Among neutral fermion systems easily produced in the laboratory. and occurs only for T i e . is to include the effect of collisions.) The characteristic of (20.19) Therefore. zero sound comes from short range interactions. the condition of oscillation is (20 . 2 .17) When k / o is very small. << 1 . t ) . (2018) (20 . 20. (2019) is still true. However.3. Equation (2019) can be directly derived from the equations of fluid mechanics. in which the scale of t is much larger than the collision time.
352 STATISTICAL MECHANICS (At standard atmospheric pressure.e p ) . fo = 6 ( .r fo eP E p 2 / 2 m . 5 K . (201 6) and (20. the Fermi momentum is p . t ) = fo + . (2023) . (20 20) where u(r) is the interaction energy between a pair of atoms at separation r . These results were first obtained by Landau. then (2014) and (2015) become (2021) f o  a  v6(ep) aP Using the same steps as before.) Now proceeding as in (2012). f’ eiwt+ik.eF . Nevertheless the properties still resemble those of free particles. Interatomic interactions are strong. Now assume that as T + 0 . we can write (2012) and (2013) as  9 = j d 3 r ’ u ( r .r’) n’(r’.17) become (20 22) 1 = 1 d a = 27rsine df3 .
but can also be called zero sound. s is close to 1 .11 . When the interaction is not strong. at low temperatures.THE EQUATION O F MOTION FOR GASES 353 where g(0) = rnpF/n2 is the density of states at the top of the Fermi sea (see Chapters 3 and 4). In other words. (2025) The wave velocity s vF is very close to uF. 4. they are elastic waves. The oscillation frequency can be obtained from this equation. This gas has a sound velocity vF/&.e.” What is the difference between this “zero sound” and ordinary sound waves? Sound waves in solids are transmitted via the interaction between the atoms. we get (see the zeropoint pressure formula in Sec. (20 .1 . uk g(0) is small. A careful analysis will show that this ordinary sound wave is the sound wave for the electrons outside the Ferrni surface and the holes inside. (2024) o=vFsk . Its velocity is c2 = co/& . similar to this zero sound. (10. 1 lnukg(o) 2 1 2 s1 . . The integration in (2023) is simple.e.In [ s+l s. The phonons in crystals are ordinary sound waves (see Chapter 6). i. This oscillation is the “zero sound”. a Fermi gas can be regarded as a dilute gas of electrons and holes.2 6 ) If we use the formula for ordinary sound to calculate the velocity of sound c for a Fermi gas. c = VF J3 (2027) This result is different from (2024) and (2025). i.1). and gives 1 1 = uk g ( 0 ) . The second sound is the sound wave of the phonon gas in a crystal. see Lifshitz and Pitaevskii (1980).28) aFor details. This sound wave is very similar to the “second sound” discussed in Chapter 6 .
P ’ = P > (2029) where p and p’ differ only in direction. dQ’ is the integration over the direction of p’.t ) . 20. so for the time being we do not discuss the collisions between phonons and particles but examine first of all a relatively simple problem: the collision of the particles with the fixed impurities.f ’ ( 1 f) = f .354 STATISTICAL MECHANICS where co is the velocity of the phonons. Zero sound is caused by the interaction u(r). The reader will understand this after reading Sec. (2031) . This results from the same effects as (2027). propagation by collisions. 20.f ’ ) .5. Even though there is no collision. i.f ’ . Not ice that f ( 1 . Suppose there are many impurity atoms fixed in a metal. We assume that y is not a function of the scattering angle but is a constant. Assume that f is not too much different from the equilibrium value of fo and that the difference is a plane wave of long wavelength + < < 1 . So we need not worry about the terminology “second” or “ordinary” (usually “ordinary” is called the “first”). Collisions and Diffusion Now let us look at the effect of collisions. Then how do the sound waves in (2027) and (2028) propagate by collisions? By this we mean that in the absence of collisions propagation is not possible. and the rate of scattering y is proportional to the density of the impurities. These terms themselves are of limited scientific value. This is a complicated problem. P’.4. and the collision term is f’ = f(r. The equation of motion of the free electrons is (203).e. this sound wave still propagates. It is better to understand the physical mechanism responsible for oscillation. (2030) Thus (2029) is much simpler than (204).
The solution of (2032) is then a&) = a. The linear space in question is the set of all functions of .oo = 1 . $Oa i. of K . is a solution and any eigenvalue 7. yoo = 0 . is defined as . so $ can be regarded as a function of the direction of p which is denoted by Now we use the techniques of linear spaces to solve (2032). We now want to obtain the eigenvalues 7 and eigenvectors J.e. K 1 may be regarded as a perturbation. We first obtain the eigenvalues of K .3 7) whose integral is zero is an eigenvector of K with . Obviously from (2033). J.(O)eTJ . i. (2036) Because we assume k to be very small. +K. KGo0 = 0 . only the direction of p is a variable while the magnitude of p is a constant. .THE EQUATION O F MOTION FOR GASES 355 Using (203) and (2029) we get (2032) Because we assume that energy is conserved in collisions.. The inner product of two functions @ and J. (20 .
i.356 STATISTICAL MECHANICS ~ c =l J4. ~ ~ . I+O . From elementary quantum mechanics the second order correction to yo can be immediately written down as (2040) The only nonzero term from this sum comes from where 0 is the angle between k and p . (2038) That is to say. The first order correction to the eigenvalues is zero because K 1 is proportional to v = p/m.rn) . cw=(r. Now expand in powers of K 1 . =dcosd =kv Substituting into (2040) we get i . (2041) (2042) . in powers of k . Go. . and $ has . The eigenvector of 0 is 1 while the eigenvalue y has infinitely many eigenvectors.e.rry1. K has only two eigenvalues. =G Y . : f p symmetry. 0 and y. So Next we look at the second order terms.
we can use it to look at slightly more complicated cases. the vibration of the crystals and the collision between electrons. From (2031) and (2043). the energy is not changed during the collision. But for small k the first term on the right vanishes very slowly. Because we assume that the impurities are fixed. we have not considered the problem of the oscillation of the impurities. considering only collisions but not mutual interaction. Therefore the energy of each particle is an invariant. Viscosity and Heat Conduction Now we use the equation of motion (203) and (204) to analyse the relation between collisions and sound waves. We write f as (2045) . Therefore (2036) is (20 . the O ( k 2 ) term is the whole of yo and cannot be neglected here. (a # 0) are nearly all equal t o y. y. (2044) where f . For the sake of simplicity. represents a collection of particles with the same energy. on a time scale 7 = l / y . The last term in (2043) vanishes very rapidly.5. however.43) The factor I / y is the mean free time T .THE EQUATION O F MOTION FOR GASES 351 If k is very small. This naturally is the result of the conservation of particle number: when k = 0 this term represents the change of total particle number. 20. which is not affected by collisions. We can then ignore the O ( k * ) terms. This is done only for an ideal gas. so D is proportional to 7. Collisions and Sound Waves. then f i will satisfy the diffusion equation fi at a = D V 2fi . if t >> T . The above analysis exhibits the relation between diffusion and collisions. Now that the steps of this analysis have been clearly explained.
. Substituting (2045) into (203) and K. .(t) = a. = ikv . (2046) (20 . then the solution of (2046) is a.47) (2048) Now we use the techniques of the last section to solve (2046).358 STATISTICAL MECHANICS For fo see (2013).e. + K . The inner product of two vectors $ and 6 is defined as If we can find the eigenvalues 7 and eigenvectors $ of K .(0)e7a‘ . The set of all functions of p is the function space. Assume (204). i. (2051) . we get 9 << 1 .
we only have to obtain $ and 7 . notice the conservation of energy and momentum (see 6 function in R of (204) )..54) . (Prove ($. T . but .5 2) This formula is obtained from (2049) and (2047). (20 . when k is very small. The main characteristic of K is that. is simple. + . . 7 (2053) T~~ = O ( k 2 ) . K $ ) = (4. . etc. K $ ) : ($.To. because of the conservation of number. We expand 7 and $a with respect to K . = To. + 71. (20 . K $).e. (20 . The operator K . .55) (2056) .) The factor R is the rate of collisions and hence the eigenvalues of K are comparable to 1 / ~where T is the mean free time.$”‘)* . K + o . The magnitude of the eigenvalue of K can be estimated from ($. energy and momentum (see the 6function in (204)).  $1. K $ ) = [d3p d3p’ d3p” d’p”’ fo ji R ( $ + $’ . From (2052) it can be seen that the eigenvalues of K can never be negative. it has five eigenvectors with zero eigenvalue.. K is now rather complicated. . We now consider the long wavelength motion. If each yOa were comparable to 1 / ~there would be no long time oscillation. i.$“ .T H E EQUATION O F MOTION FOR GASES 359 Therefore. .
3T 2 . E = ( e ) = . u.5 9) Obviously K . The other three Q o a . 4 .360 STATISTICAL MECHANICS The above and F are constants which satisfy (2056): A.)” 2 (20. . a = 0. . = n ( ( e . . . 0 = 1 . 1 . = ikv = ikp3/m . a = 0 . which are the eigenvalues of the 5 x 5 matrix K1. So y1 in these two directions is zero. .p = Kl % p ) 3 (20.5 8) Let k be along the third direction. (20. = ?(A. then K. ap = 0 . .E ) 2 ) = TnT 3 2 . . Now we examine yl. (2057) The average I . ) is with respect to fo. 1. for a .60) . p z and p 3 are mutually perpendicular. 4. = are the lowest eigenvalues./A. a = 0. .. 2 because p l . . These five yoor 0. 3 . 4 form a 3 x 3 matrix 1 uo = m A 3 / A o ) ’ ( = (T/m)’ = (2T/3m)’ . . .
3. are 0. but simply use (2063) as an approximation.e ik. the sound wave will attenuate . Therefore.Top cK:. f i k c c . = D .v2r rT/m (2064) rl/+ make periodic motion and are the sound waves in the ?r k directions:  fl then OC e ytt e i k . P> 4 . Dl. Because D.k2 9 . The next order term y2ffis a bit more involved : * Y2ff = P > 4 .2 . Because yoP 1/r. (2063) We do not attempt to calculate (2062) here.1/T . K . r  .D. k2 74 = D4 k 2 9 r p . 0 : (2065) so D.  k2v2r . ikv.4 Thus . = (5T/3m)' = . The results are as follows: 7. the other three yla are zero.T H E EQUATION O F MOTION FOR GASES 361 The eigenvalues of K .k2 is the attenuation.k2t 9 7 fl at a* = c2 V 2f . thus   y2.V2 a at fl .P (2062) This calculation requires knowledge of the other eigenvalues and eigenvectors of K . except for rl+= i k c .k2rT/m + D.r t icktD. (2061) Therefore. we need the details of R . = k i c k Y1. r .
2 t k .V2 fl . be calculated from (2062). t . and so describe a combination of the oscillations of particle density. The functions are linear combinations of J / o o . The above differential equations (2065) . If J/. If there are n o collisions then T + m and there is no sound wave. (20 . The function J/l.2 e ik. the particles will go in straight paths and there is no wave motion. I .b bFoster (1975) is a good place to start. That is to say.. J/03 and J/04. sound wave is propagated by collisions. This result is quite obvious.e ik. Equation (2064) summarises the properties of the sound wave. The coefficient D1 is proportional to the viscosity. energy and momentum (along k). viscosity and conduction of an ideal gas.2 expresses the change of the momentum of the particles (perpendicular to k) and is a kind of “shear” or “twist”. If there are no collisions. (2067) and this is the diffusion or conduction of heat..> > rv = I . fl a +1.(2067) can be regarded as another expression for (2064). . The different coefficients can.68) . (20 .(2067) have a time scale much larger than T and a length scale much larger than the mean free path I : kv << 1 1 7 . then a fl at = D l V 2 fl .6 6) The function If G4 is the change of energy. I Y. The above analysis shows that Eqs. together with a small density variation.362 STATISTICAL MECHANICS rapidly as T is increased. in principle. (2065) .D . k For further analysis one may consult the literature. then f1 at a = D.
! . . Therefore. So the probability for two particles to meet isproportional to T and their energy is about T . H is indeed the entropy. We would like to remark. In (2046) K .6. The reader may point out that H is just the entropy (apart from a constant). at low temperatures.69) As T + 0. 20. There is however a main difference. This is an expansion in terms of the small quantity kv/( I/T) = kE. (203) and if ( a f p t ) .theorem We mention here Boltzmann’s H . 20. Therefore. For fermions at low temperatures.THE EQUATION OF MOTION FOR GASES 363 Now let us compare the results of this section with those of the preceding two. The analysis of this section regarded collisions as the main part. was treated as a perturbation of K . there is still little chance for particles to collide. This already proves that f i = f . # 0. Although the final result was established mainly in the space K = 0. H . If there is a collision. The proof of this theorem is very simple.theorem. however. In Secs. 1 (20 . (This too is very easy to prove.) Although this proof is limited to very small f i . the number of holes and electrons outside the Fermi surface is proportional to T . Only zero sound can propagate. H will attain its maximum at f = f o . that if the collisions are frequent. (2070) must increase with time. where we have pointed out that eigenvalues of the operator K are all positive (see Eq. they are influenced by the exclusion principle: a particle must enter an empty hole.. and hence not constant. Indeed we have proved most of it in the last section. (2052)). then H E 1 d p d r f3l n ( . This section discussed the ideal gas but the same analysis can be applied to a gas of dense fermions. Even if they collide. 7 a T 2 . So the states after the collision are limited.fo will tend to zero and f will tend to f.2 and 20. with similar conclusions. the expansion in this section would not be possible without K . their energy after collision must be more or less evenly distributed. as it does not require collisions for propagation. otherwise it is not. the main points are revealed here. This theorem says that according to Eq.) 3 . there is no way to propagate a sound wave by collisions. because states much lower that T are forbidden.3 we did not consider collisions but rather the force arising from the variation of particle density.
H will not be the entropy because in momentum space. If there are no collisions. If there are no collisions each pi is an invariant and H naturally is not the entropy. Notice that the kind of collisions discussed in Sec. These are all very important problems. This is because the last term of (2071) is oscillating so rapidly that d H / d t vanishes. the integral of the last term over any interval of p is zero because its frequency is infinite. .364 STATISTICAL MECHANICS Notice that even if there are no collisions.e. if we substitute (2071) into (2070) and calculate d H / d t . . the trajectory is just a point.4 is not enough to allow H to be identified as the entropy because such collisions cannot change the energy of the particles. f can still tend to f o .e. 20. . For example at t = 0. the entropy does not increase. . we get zero (if there are no collisions). i. . These phenomena roughly show that even if the particles are dispersed. Problems 1. In Chapter 24 we shall carefully discuss how to calculate entropy from the trajectories of the motion. An initial irregular distribution will become a uniform distribution because the velocities of the particles are different. But we cannot say that H ( f ) tends to H ( f o ) . the 3Ndimensional space ( p1 . i. Only the collisions will disperse the trajectory. when t is very large. So it is not wrong to say that f tends to fo . if then i. According to our view. to identify H as the entropy we need to have collisions to change continually the momenta of the particles.e. In Chapter 23 we shall discuss the “echo phenomena” where the problem of collisions and entropy shall be further considered. Of course. pN). Suppose the positive charges are fixed while the electrons vibrate with a small amplitude. the entropy may not increase. Plasma oscillations are discussed in books on plasma physics and solid state physics. pz .
This assumes q is small. 4.THE EQUATION O F MOTION FOR GASES 365 (a) From fluid mechanics derive t h e following equations: an' + at nV*v= 0 . ( F o r details see an oscillation of frequency fiw. then it will produce an electric potential (47rq/k2e)eik* r . ('2075) The definition of E is: if the distribution of an external charge is q e i k * r . 20.2 t o rederive t h e results for three dimensions. The above problem is appropriate for three dimensions. A charge moves over a metal surface with a velocity parallel to the surface. Rudnick (1 967). 3. Change (2072) so that it is suitable for planar conditions and then derive (2019).i w r . Prove that the dielectric constant of a plasma gas is E = 1 / ( 1 w.) . m=eE at av . and at a distance x from it. (b) Apply (20. Compare the results with (2074). (2073) 2. (a) Prove that Then solve it. Prove that there is o n the metal surface. n' t h e change of electron density and v is the flow velocity. (b) Solve (2072) t o get t h e oscillation frequency 47re2n u p = ( )' . Add a pressure gradient V p t o the second equation of (2072). E =47ren' . taking into consideration t h e effects of collisions.17) t o three dimensions and expand it terms of k . Discuss t h e limits of validity of (2072). Discuss the response of the electrons inside the metal. Then use t h e methods of Sec. This problem can be considered together with Problem 3 of the last chapter. keeping the first two terms.72) where n is the average density.i w r . 5 . (20./w2) . V.
Also generalise the calculation of Sec.3. Prove that c 2 in (2061) is the adiabatic compressibility coefficient 1 a .366 STATISTICAL MECHANICS 6.4 and 20.5 and comparing with Sec. The reader can understand the true meaning of collisions by carefully performing the calculations in Secs. . 20. 7 . 20. 20. of an ideal gas.5 to fermions at low temperature.P ( ).
how crude are these equations? How can we improve on them? These questions are still unanswered. Therefore. We still cannot start from the more basic principles (Newton’s laws. The diffusion equation and its various generalisations are such simplifications. 21. but rather on some simple applications. Although crude and limited in applicability. we can simplify these details to avoid complicated calculations without affecting the outcome of the motion over long time scales. These equations are just convenient to use. the understanding of collective motion is still in its infancy. we know how to use them but d o not fully understand why they can be so useful. The cumulative result is usually independent of some of the details of the motion over short time scales. (1255) to (1268) ). we can have a better understanding. The diffusion equation can be . In some special cases. Simple Examples In Chapter 12 we started from the equations of random motion to derive the diffusion equation (see Eqs. But generally speaking.1. these equations are very useful if we require only rough answers. in the last chapter we used the Boltzmann equation to introduce diffusion. to which mathematicians have already made many contributions. quantum mechanics) to understand these equations. From the view point of physics. These equations themselves are interesting mathematical problems.g. In this chapter. with many examples related to the metastable state.Chapter 21 THE DIFFUSION EQUATION Changes over long time scales are the accumulation of changes over shortti time scales. emphasis is not on the theoretical basis. e. The problem of barrier penetration is the main point of analysis of the metastable state.
then (214) In equilibrium p should be proportional to e . then J=rFp . then 1 r = .U / T . the flux is proportional to F where rF is the flow velocity. (211) J=rFpDVp where p is the number density. They are temperaturedependent functions and reflect the conditions of the environment. 6nav (21 3) where v is the viscosity of fluid. The diffusion equation can be understood in terms of conservation of particle number: at  VJ ..4 will be devoted to a review of its relation with random motion. The parameter r is a kind of coefficient of friction. This section discusses some simple properties and examples of this equation. F is the external force on each particle and D is the diffusion coefficient. while the other is proportional to the external force. .one produced by diffusion proportional but antiparallel to the density gradient.368 STATISTICAL MECHANICS regarded as a consequence of the central limit theorem. The parameters D and r a r e regarded as known. Section 21. (2 12) i. The parameters r and D are closely related. Let F = VU.so r 1 __ D T (215) .e. with proportionality constant r . either measurable or calculable from other theories. If the particles are spheres with radius a immersed in a viscous fluid. J is the flux. The flow has two terms . In equilibrium J = 0. where U is the potential of the external force. If we ignore D .
Now let us consider some more examples.exE . . while those flowing to the left experience a large force. Fig. 2 1 .1 U(x) = xF . If we add an electric field E . and at b it suffers a very large force to the right.THE DIFFUSION EQUATION 369 This is a very important result and has been derived in Chapter 12 (see (1266)). The electron experiences a force F between 0 and b. otherwise =o . In equilibrium there is no current. (216) This is a simple “diode” model. 211. From (211) and (215) we get (217) (21 8) . what is the current? Adding the electric field we change U ( x ) to U’(x)= U(x) . Electrons flowing to the right experience a slight force. O < x < b . This is also the condition for detailed balance in Chapter 3. Example 1 : The effect of a certain electronic component is shown as U ( X )in Fig.
J is not a function of x while p is a constant outside ( 0 . Suppose p is nonzero to the left of the barrier and p = 0 to the right. 212 Example 2: Barrier penetration. = U ( c ) >> T. where E b is the potential difference across (0.370 STATISTICAL MECHANICS As the current becomes steady.e. i. assume that the barrier is sufficiently high. 212). Moreover. There is no current when E = 0. particles will penetrate the barrier to the right. Gradually. Using steps as in the last example we immediately get (2 1 . Therefore integrating (218) we get (2 19) The above assumes eE << F. The potential U ( x ) as shown in Fig. b). U. What is the flow rate? Because the flow rate is small. the situation on the left is roughly steady. b). 213 is a barrier. but when e E b >> T electrons can flow easily to the right but not to the left (see Fig.1 0) . Fig. So J a n d p ( 0 ) do not change. J and p will not change.
T . is seen to be a determining factor. the particle must overcome the barrier. . Notice that the factor ebF’T in (219) comes from the same reasoning. From the last example. The particle density is p inside the region. To get out. because the other routes require overcoming higher altitude. The highest point is the most difficult place to penetrate while the other places are insignificant by comparison. Place the xaxis through the mountain . Calculate the outward flux. Obviously the mountain pass is the easiest way out. Example 3: Figure 2 1 . 2 1 .4 shows a twodimensional barrier. A region is surrounded by a very high potential U(r).THE DIFFUSION EQUATION 371 L Fig. The parameter T / a = a is the width of the barrier at U. (2111) Substituting into (2110) we get e J N U.3 Expanding U ( x ) about x = c . At the northeast direction there is a place of lower altitude forming a mountain pass or “saddle point”. we know that the most important factor is the highest point on the way out./T p(0)D (2112) The height of the barrier U.
(21 . The flow rate of the particle out of the region is approximately given by J = pD [ m / ( 2 n T ) ] eUc’T . (we use the symbol A because it resembles the barrier.) The factor T~ is a time scale unrelated to A .12). then we can use (21 . being the height of the barrier to be overcome. (21 13) The subscript c above denotes the value at the pass and p is the density inside the region. pass.3 or those inside the region in Fig. p = (a2u/ay2). The exponential A/T is the most important factor. . The particles at the left in Fig. Equation (2114) can be used to estimate the lifetime of a metastable state. This time scale is of the order T = ~~e A / T . 21 4 A 2dimensional barrier with a saddle point.14) A = U. If the barrier is very high they can remain inside for a very long time. 21. The Metastable State The above are examples of the metastable state.2. 21 4 are in the metastable state. We can treat it as the relevant time scale when the barrier is low.372 STATISTICAL MECHANICS North mountain West South Fig. 21 . .
This formula assumes that the area of contact is a hard surface.15) we get the potential energy of the egg (Fig. A = mga2/2h . Now let us estimate the barrier height A of this metastable state. we need to raise the centre of mass of the egg by a bit.e. The first term a0 is the rise of the centre of the surface of contact and the second term .THE DIFFUSION EQUATION 373 Example 4: How long can an egg stand on its end? A standing egg is a metastable state. An egg stands because its centre of mass is on a vertical line passing through the surface of contact of the egg with the table. the smoothness of the shell surface and the flowing of the fluid inside the egg.1) 2 . From (2 1. When 6 is very small. Let a be the radius of the area of contact. the surface is created by the weight of the egg itself. the rise of the centreofmass is Ah N a8 + h(cos8 . i. etc. Acorrect analysis must take into account the elasticity of the egg shell. (2115) 1 = a 6 he2 . For the egg to fall. 215 Potential energy of an egg as a function of the angle e between the perpendicular and axis of the egg. h the height of the centreofmass and 8 the angle between the axis of the egg and the perpendicular. U ( 8 ) N mgAh . 2 15). (21 1 6) Its maximum occurs at 8 = a/h. This assumption is not too accurate.$heZ is the change of height of the centreofmasswith respect to the centre of the surface of contact. (21 1 7) e a/h \ Fig. .
At room temperatures. Transformation to the Wave Equation The above examples focus on the flow J. Therefore. The most convenient method is to transform (211) into forms similar to the wave equation in quantum mechanics and then use our knowledge of quantum mechanics. (21 22) . eA’T makes T very large. Assume that a A 10 erg. (2121) is similar to the quantum mechanical equation and K is like the Hamiltonian. then (2118) A/T .14) is limited to situations in which A / T is neither too large nor too small. the use of (2 1 . n o matter how small r0 is.374 STATISTICAL MECHANICS An egg weighs about 30 g. = O We can now use techniques developed in quantum mechanics. Let (21 20) Substituting into (211) and rearranging we get (2121) Other than some minor details. Generally speaking. The term V *J in (21 . the motion of molecules in equilibrium cannot by itself make the egg fall.  . to analyse the time that the egg can stand. h . a certain molecule must have a kinetic energy of at least 10erg.5 cm.10” . 21.3. Let us now examine the time derivative of (211). According to (2114). That is to say. If A / T is very large it indicates that we must also consider other more complicated processes. For the egg to topple. we must include other factors such as rotting of the egg. Now let us devise a method to eliminate it.1) contains the Vp term but the wave equation does not have this term. The first derivative is also not a symmetric operator. etc. wind.1.0. The ground state of K is $o with energy zero: K$.03 cm.
The boundary condition of the diffusion equation is that at the walls J . = 2 n D a 2 / T . the ground state wave function of the simple harmonic oscillator. (21 . O<x<L otherwise .$ a 2 x 2 / T ) .2 5 ) This example is given to remind the reader of the importance of the boundary conditions. the solution of this problem is where (21 . = O . (2123) This is. This is different from the condition that 9 = 0. depending on the density at t = 0. the solution of (211) is r. Example 6 : Let us consider a hardwall box U(x) = 0 . Needless to say. (21 .2 4 ) The coefficients A . The other eigenfunctions of K are then the other eigenfunctions of the simple harmonic oscillator. From (2120) we get $ o ( x ) = exp (. . Therefore. (21 26) where 1means the direction perpendicular to the wall. . of course. are constants.2 7 ) and N is the total number of particles.THE DIFFUSION EQUATION 375 Example 5: Let us solve the diffusion equation for U ( x ) = a 2 x z .  00 .
= A t E 1m d t ” ( v ( t f ’+ t’) v ( t ’ ) ) . (21 . . Derivation of the Diffusion Equation The abovementioned p(r) is the distribution of particlesin r space.4. i. x(t) = x(t’)+Ax . . . We now review the results of Chapter 12. . t = t’+At .( v . with (Ax)= (v)At. the sample set is (rl . a’ ((AX)’). (21 30) AX) = exp( . r2. then the central limit theorem shows that the distribution p ( A x ) of A x is normal. we can prove that p ( A x ) satisfies the following equation : aP aAt . rN ). The number N need not be the particle number. We shall consider the displacement due to the random distribution of velocity v ( t ) and derive the distribution of the displacement.( A x . The meaning of probability comes from the statistics of the information of N particles. A t ) 2 2 a’ dG 1 1 I (2131) By direct differentiation. Starting from t’..376 STATISTICAL MECHANICS 21. where ri is the result of the ith outcome. the displacement in A t is Ax = 1 At d t ” v(t’+ t”) . much shorter than A t .e.2 9 ) The distribution of x ( t ) can be deduced from that of x ( t ’ ) . it can be the number of times the experiment is repeated. (2 1 28) ~ 0 If the correlation time of v is very short.(v) aP a(Ax) a2 a2p + ~ 2 A t d(Ax)’ (21 32) . However r need not denote the position but can be any set of variables and p(r)/N is the probability distribution in the variable r.
from(2130)weget  5j d x ’ ( v ) ax AX) p(x’. If this theorem is not appropriate. written in a slightly more detailed form and pointing out how to obtain D and I F from the details of the random motion.. Here we consider ( v ) # 0.THE DIFFUSION EQUATION 377 BecauseAt=tt’. Ax=xx’. and ( v ) does not change with x’.p 2 ax2 1 a2 The above steps are more or less the same as those in Chapter 12 ( (1255) to (1268) ). t‘) (21 33) u2 at (21 34) . then x and X’ are nearly the same. and let ( v ) and u 2 / ( A t ) be slowly varying functions of x and t . .a( v ) p at ax + . This is the diffusion equation (21l). 2 .t’) a2p AX) p(x’.. + = jdx’ a(nx) [(’) a P rJ2 a(Ax)’ ]p(x’. and (21 32) becomes aP _ . (2135) . then J = (Ax.1) and (21 34). . These results come from the central limit theorem. (Y = 1 .) At The relation of (21 . Let x. t’) 0 ’ a2 ‘F! dx’ 2At Now assume that A t is not too large. Equation (2135) is the socalled FokkerPlanck equation. . (21 35) is now clear. then they are likewise invalid. be the variables. . If there is more than one variable. it is not difficult to generalise the above derivation.
+ is the probability in time d t for si to change from + 1 to 1 and d t r+. then the generalisation of (21 36) is c P (rpa P.  (2138) where a. . (21 35) and (21 35) are crude models of the molecular motion. then the form of the diffusion equation will be changed.1 respectively. These disturbances or changes must satisfy some conditions.378 STATISTICAL MECHANICS The x.l). . . then in (21 39) Therefore. the most important ones being independence and detailed balance. N . The Ising variable si is the simplest discrete variable. Equations like (2 1. They regard the cause of motion as an irregular disturbance or a sudden change at any moment. These . The change of si is discrete.is the probability from 1 to + 1. .+ are the reaction rates. the different rap must satisfy (21 40) This is the condition of detailed balance as discussed in Chapter 2. In equilibrium p + is unchanged. (21 36) where d t r .0 label the various states. 2 . so (21 37) If the variable takes on many values. Let si = k 1. in (2135) are continuous variables.. We give an example to illustrate this. If the variables are discrete. Then. If there is an external force so that the energy of state a is equilibrium E. r+and r . i = 1 . Let p+ and pbe the population or probability of the states + 1 and .
& dP* + P O > dt h* . 2 16 Energy levels for a hypothetical system..E. The time will . The energies are . Fig.e. while a more refined analysis will be too complicated. We use the simplest model: rap = 1 if E. 0. i. the reaction rate for a decrease of energy is 1. < eP and rap = exp[(e.THE DIFFUSION EQUATION 379 models are crude. . Let sj have three values + 1. 0.. r+.f.= r . but nevertheless valuable because they contain some of the main characteristics of molecular motion.e p ) / T ] if E. > e P .fT (21 41) The state 0 has a high energy while f 1 are states with low energy. 2 16). respectively (see Fig. i. Example 7: The elimination of high energy states. Suppose sj cannot change from + 1 to 1 directly. where E is the energy needed for the reaction. .+ = 0. Then . and e'fT for an increase.. which must satisfy the condition for detailed balance (2140). We first define the various reaction rates. 1.e.A . Let ey >> T .e + .
like (2112)..state. Generally speaking. Nevertheless. without direct reference to the 0 state. then (21 43) is also true. 21. This model assumes that most of the body is frozen and only a small part.5.380 STATISTICAL MECHANICS be short for a particle to stay in a high energy level. . caused by having to penetrate a barrier. separated by high energy states. so that the rate of change is slow. but long in a low energy one. Each cluster is characterised by e. . Notice that A. of which two are lowlying states. Each movable element is a small cluster composed of several molecules or spins. Assume that there is a barrier between the two states. in some special cases the low temperature properties of these complicated structures can be explained by the twostate cluster model.will be more complicated. although A. To change from the + state to the .p2 1  + TAT PT s (2 1 43) This is the same as (2136). TwoState Cluster Model Amorphous substances like glass and rock or amorphous magnetic substances mentioned in Chapter 18 have very complicated structure. can move freely. in discussing slow changes. The high energy state 0 is eliminated. the barrier height.= exp(e. the particle must. the energy difference between the two states. . the change in the probability of the two states + and . some special molecules or spins dispersed throughout the body.is determined by (21 43). e./T) are. Any reaction involving barrier penetration will have this factor in the reaction rates. and A.g. We can solve the equation for po assuming pi: are unchanged. Figure 21 7 plots the energy versus coordinates in the configuration space of the small clusters. if there are more than 3 states.overcome a barrier of height A . That is to say. (21 42) Substituting into the first equation of (21 36) we get dP+ 1 + d t ..A + . Each cluster has two lowlying states which are most important at low temperatures.
The interaction Jii can be positive or negative. From the solution of this equation and the distributions of E and A we can deduce the various properties of this model. large or small. . but does not change with time. 2 1 . . .. Suppose the number of clusters is sufficiently large.THE DIFFUSION EQUATION 381 Fig. The Jjj form a normal distribution +_ . essentially the reaction rate at high temperatures.7 A plot of energy versus coordinates for a cluster of molecules. . There are many clusters all with different A and E. How are A and E distributed? We now use a simple model to clarify these points.1 + + . (21 44) (21 45) where l / r o is a constant. Let H be the energy of an Ising magnetic model. i = 1 .e 1 p dt r €IT p . Assume dP+ . The reader may be impatient to know what these clusters are. then we can calculate the distribution of A and E and discuss the probabilities p* of the two states using equations like (2138). (21 46) si = 1..N . 2 .
Let us return to (21 44) and see the conclusion of this cluster model. If a change of sign will lower the energy. giving the reader a sense of “reality” for the abstract model. e < 0. To change si requires an energy 2\hi 1. Ma and Hu (1979). Nonetheless. We shall also study it carefully in the next chapter.382 STATISTICAL MECHANICS and the interaction is of the nearest neighbour type. where hi is the magnetic field produced by the neighbours of si : hi = 1Jii sj i . Jlzdoes not come in. If s1 and s2 change together. the reaction rate is e . This is a twostate cluster.a The low temperature properties can be explained to a large extent by the small cluster model. while those with large hi are frozen in. 217. spins si with small hi will move. the motion of the element is one at a time and not two at the same time. (21 47) At low temperatures. Therefore. Even if some hi is not small. the increase of energy is This can be obtained from (21 47). At low temperatures most of the si are frozen. as in Fig. and h z are very large. The analysis of a multistate cluster follows similarly from that of the twostate cluster. forming an irregular magnetic body or amorphous magnetic body.A E i T . (21 49) where e is defined by (2148). The distribution of A and e is calculated by statistics. is very strong. When the two change together. even though h . This is also the rule used in the example of the last section. If J . The barrier is A = 2Ih1I  E . others at 1. This model has been analysed carefully before. The motion of si is according to the following rule: If a change of sign of si increases the total energy. the simultaneous change of sign of si together with one neighbouring spin may not increase the energy very much. . E can be very small or negative. This example shows the origin of small clusters. Let I hl I < I h z I. then it is easier to first flip s1 and then sz . aFor details see Dasgupta. it may even decrease the energy. the reaction rate is 1. the energy must first increase by 2 I h l I or 21 h z I and then decrease.where AE is the increase of energy. Some of the si are frozen at + I . .
Let p + + p . The total energy is E(t) = s d e d A N ( e . Obviously the integration with respect to A gives a function depending on x. The times r' and T~ are nearly equal if e > 0. A ) p l ( 0 ) e e" + constant . A ) are the distributions of e and A for the various clusters. . Now let us estimate the change of energy as an application of this solution. and its value depends on the structure of the material. Here N ( e. If e < 0 and Iel / T is very large.= 1. Since T' appears in x as 1nT' we can replace it by T . then f> 9 (2150) (2151) Here p + ( 0 ) is the probability of being in the + state at t = 0. r is defined by (21 51) and r t can be written as x =T ln(t/T') . then r will not be small.T H E DIFFUSION EQUATION 383 The solution of (21 44) is very simple.
x ) . i. Notice that time appears as l n t . if the A of a small cluster is greater than x = T In t /T ~the small .5 6 ) That is to say. 0 (2154) The function err (see Fig. Substitute (21 56) into (21 54) and we get m E(t) z X dA g(A) + constant (21 57) . g( A) does not change much.e. (21 . If T is very small. The width of the change from 0 to 1 is about T . it has sufficient time to change.r t . then from (2152) E(t) = 1 m dA g ( A ) e . cluster has no time to change but is frozen in. then ert = 0(A . and be freed from the frozen state. and nearly 0 when A<<x: 'L 0 t x A Fig. 218 Plot of e" as a function of barrier height A .r t + constant . when A changes to A + T . If A < x .384 STATISTICAL MECHANICS E We can neglect e . 21 8) is nearly 1 when A >> x . Now we ignore the situation we get < 0.
. If the cluster is very large. it decreases slowly as T In t / T O . That is to say. So it does not matter whether we study the twostate. does not vanish immediately if the magnetic field disappears. If we have a distribution of barrier heights A. E decreases slowly with T l n t and this is a very slow change. For example. Similar quantum mechanical models have appeared in the theory of glassb ~ t Fig. But the above conclusion on the logarithmic change T In t / is ~ simple. it is not enough to study only the twostate model. This is familiar to those studying the magnetic properties of ancient rocks. and not sensitive to the details of the model. the magnetic moment. 219 bSee Anderson. The recent study of the amorphous magnetic state also depends heavily on such analysis. This small cluster model is extremely crude. E ( t ) will be a slowly decreasing function of x (see Fig.THE DIFFUSION EQUATION 385 If g(A) is a wellbehaved function. rather. once established by the magnetic field. 21 9). with many assumptions. Halperin and Varma (1975). threestate clusters or even the nocluster situation. we shall have a logarithmic rate of change. This is true not only for the energy but also for other quantities. The overlapping of the clusters and their interaction have not been considered. This collective model mainly tells us the possible origin of the distribution of A . Indeed this is the analogue of the result r = (l/To)eA'T.
4. 9)are the polar coordinates. . 9)is the distribution function on the sphere and ( 6 . calculate the retarding force on the ball. In (2158) we assume the direction of this force is distributed isotropically on the sphere. rn is the mass of the gas molecule and w is the angular velocity.. 3. The temperature is T . The radius of the ball is bigger than the radius of the molecules but smaller than the mean free path of the molecules. Suppose we have a mirror with diameter much smaller than the mean free path of the molecules. Suppose that each solute molecule has mass m and volume v . A ball of smooth surface moves with constant velocity in a dilute gas of temperature T . 2. The reader should notice that the righthandside of (21 58) cannot be written as The origin of diffusion comes from random forces. A circular tube of length L and crosssectional. (b) Calculate the entropy and other thermodynamic quantities.386 STATISTICAL MECHANICS Problems 1.area A is revolving about one end with angular velocity w . I is the moment of inertia of the mirror. 5..D _ . where where = J 8 T / n m . In the tube is a dilute solution with N solute molecules. (a) Calculate the density distribution of the solute n ( r ) where r is the distance of the molecules from the axis.a( s i n e G )a P D azp + sin'e at sine a 0 a4z 9 (21 58) where p ( 0 . Prove that the diffusion equation on the surface of a sphere takes the form ap . u is the mass per unit area of the mirror. Derive (2158) from (2135). Assuming that the collisions between the ball and the molecules are elastic. The axis of rotation is perpendicular to the tube. Prove that the torque acting on the mirror is 50. and water is the solvent. Notice especially the effect of the value of m Jv . It is suspended by a thread in a dilute gas at temperature T and pressure p . and the pressure at the axis is p .
Calculate the subsequent distribution of directions. 2 . . Every magnetic moment is continually influenced by disturbances in its environment. (Hint: use Eq. At the time t = 0. i = 1 . all the rods are pointing in the same direction. The parameters A i and hi describe the environment of each magnetic moment and Oi is the angle made b y m iwith a fixed direction.C ( h i c o s e i + ~ ~ c e o) s. . . Calculate the change of the total magnetic moment assuming T << Q .e. and h i are not the same but distributed as p(hi) = (2nb ) 2 ?h e hi/2b2 . Bi = 0 or n. 6. These molecules dissolve in a solvent and the viscosity of the solvent is 17. The energy is H = . . Suppose at t = 0 all the Bi = 0. A i causes m i to prefer two directions.THE DIFFUSION EQUATION 387 (c) If at t = 0 all the solute molecules are concentrated at the end of the axis. 3 .) 7 . ~ i i (21 59) Each A . . N . m3 = m 2 = constant. performing irregular motion. A rock has many small magnetic moments m i . (21 58). i. Certain large inolecules can be regarded as small rods of length I and radius Q . Analyse the subsequent change of the density distribution n ( r . The reader should first seek a clear understanding of the meaning of the quantities in the model and then simplify the model. t ) and estimate the time taken to reach equilibrium. with I >> (I. (2160) Assume Q >> b. This is a metastable model.
the result of the calculations can be compared with real experiments to determine the validity of the model. Not only it enables us to solve problems requiring complicated calculation. The second is “Monte Carlo simulation”.e. approximations. What are the advantages of these numerical experiments? First.Chapter 22 NUMERICAL SIMULATION The invention of the computer has had a profound influence on statistical mechanics. We can also compare the result with analytical solutions to justify the validity of various approximations or assumptions. the calculations are fixed. This chapter introduces two computational methods. the duration of the observation time and the size of the model must also be considered. and is used to calculate equilibrium properties. The Monte Carlo simulation also calculates a trajectory. not relying on assumptions. This chapter is mainly devoted to the second method. i. solving the equation of motion and calculating the positions and velocities of all the molecules. (Of course. the experimental steps must be correct. One is the numerical solution of molecular motion. Therefore.) This is more powerful than analytical treatments. The calculation of the molecular motion is indeed an numerical experiment: to calculate the trajectory of motion explicitly and then make all kinds of observations. All the results can be regarded as the experimental outcome of the model. This replaces the equation of motion by random sequences. Secondly. or the discarding of “small” terms. It can also be regarded as a numerical experiment. these calculations do not employ approximations. but it has also made an important contribution to the understanding of the basic concepts. these calculations give the details of motion of each molecule at each . Once the model is set up. but uses a model of random motion rather than Newton’s laws.
besides being useful tools. equilibrium states would emerge. being “experimental”. This infoimation is not available in real experiments. numerical experiments have become widely used. The positions and velocities of each molecule are calculated. This aspect will occupy much of this chapter. Before the invention of the computer. some scholars of statistical mechanics were tempted to abandom the notion of molecular motion and the physical meaning of the trajectory of motion. Even today traditional statistical mechanics still utilises the abstract concept of probability to define equilibrium states. 22. This kind of calculation is a type of experiment. because it seemed impossible to solve for these quantities. can only be fully appreciated if one has taken part in them. Therefore. Nowadays. those interested in molecular motion almost immediately used it to solve the equation of motion and t o see whether equilibrium would appear. Each equilibrium quantity can be . etc. The design of the various programmes rely heavily on experience. or heat loss at low temperature. calculating the trajectory of the system in a certain period by the computer. Numerical Solution of Molecular Motion The principle of this method is very simple: we just delegate the job of solving the equation of motion of the molecules (Newton’s laws) to the computer. With more elements. to give the reader some mental preparation for this topic. Fermi and Ulam solved such oscillation equations but failed to find equilibrium. not to mention the collective motion of many particles. This brought attention to the problem. Of course. and avoids the discussion of trajectories. Because of these advantages. making the concepts clearer. These kinds of calculations. provide a deeper understanding of statistical mechancs. These calculations. Gradually more people used more powerful machines to solve the equation of motion in order to understand equilibrium at a mare basic level. Each calculation involves thousands of particles and millions of collisions.1. and are seldom recorded in journals and indeed are difficult to explain. The difficulty encountered by Fermi and Ulam was that the oscillation elements were too few. using the computer to solve the collective equation of motion is a branch of statistical mechanics in itself. Any information about the system in this period can be obtained from the trajectory. to solve the equation of motion of more than 3 particles was an inconceivable job. This chapter discusses only some problems of principle. Soon after the advent of the computer. numerical experiments are easier to manage without having to worry about the problems of the purity of the sample.NUMERICAL SIMULATION 389 instant. This attitude tends to bias statistical mechanics towards abstraction and pure mathematics.
Point (b) is more serious. If [ is small enough we need not consider very many particles. and if T is short enough we do not need a long observation time. These two weaknesses seem insurmountable. Equilibrium is a special situation of motion. these computer calculations still have a wide range of applicability. These are quantities related to entropy. Many phenomena such as melting. etc.body quantum mechanical wave equations. The relation of entropy with the trajectory is a very important topic. But because of the independence of the various parts of a body. superfluidity.g. This numerical experiment is often easier to manage than real experiments and we can make observation on any scale. Solving the equation of motion is a complicated computation. Many important problems cannot avoid quantum mechanics. and discuss both equilibrium states and nonequilibrium ones. Temperature comes from averaging the kinetic energy of each particle. Of course. the success of this kind of calculation has had a profound significance since it reinforces the fundamental role of Newton’s law. There are many problems that do not require the motion to be represented realistically. and must obey the laws of mechanics. in the neighbourhood of the critical point. [ and T are very large and we need larger models. and any concept or any quantity can be expressed by this information and calculated from it.g. The time for millions of collisions is still too short compared to macroscopic times. (a) is not too serious. In some special cases. crystallisation and some collective properties of heavy atoms or molecules are not essentially quantum mechanical in nature. Some thermodynamic concepts have no obvious meaning in mechanics. The trajectory of motion carries all the information about the motion. and observation times much longer than the correlation time T . Every thermodynamic phenomenon and all concepts in statistical mechanics have their root in mechanics. Nevertheless. e. (a) A collection of thousands of particles is still much smaller than a macroscopic object. . Even today. To analyse the equilibrium properties we need to consider models much larger than the correlation length r. which usually go beyond the capacity of modern computers. (b) We cannot consider quantum mechanics. This greatly limits the use of computers. The magnitudes of and T depend on particular situations.390 STATISTICAL MECHANICS obtained by averaging. electrons in the metal. From the viewpoint of the basic concepts. In Chapter 25 we shall give a detailed discussion and point out a method to determine entropy from the trajectory. this calculation is still quite different from real experiments. Experience tells us that a model of a thousand particles is sufficient for most cases and the time is sufficient also. the computer still cannot solve many. e.
then the concept of probability will not be a useful computational tool. we then take 1 . If there is no such rapidly generated number sequence. At present there isno standard programme. Why are the numbers so generated random? This problem is very similar to our basic problem: Why can Newton’s laws produce random molecular motion? The answer is still unknown. Random Sequences Monte Carlo simulation uses a random sequence rather than Newton’s laws to determine motion. For example to choose between 1 . 22. This is a sequence of numbers satisfying certain conditions (see Sec.3). how do we proceed? The answer is: if x < p1 take l .) If we require probability p1 to choose 1. we choose 2 if x falls between f and 2 7 . p 2 to choose 2 and p 3 to choose 3. Today many scientific researchers rely heavily on the random sequence. t a k e 2 a n d i f p l + p z < x .2. 5. Monte Carlo simulation is such a simplified model of motion. Monte Carlo Simulation We can regard the rules of Monte Carlo simulation as a model of random motion. The random sequences mentioned in the following will be of this type unless otherwise specified. The random sequence is a defining tool for probability. This was unexpected thirty or forty years ago.NUMERICAL SIMULATION 391 Therefore. t a k e 3 . Now we use the k i n g model as an example to illustrate the principles. The use of the random sequences is to produce random forces. p l < x < p l + p 2 . we can use a simpler model of motion to replace Newton’s laws.3. The steps of calculation depend upon the model. If x is smaller than f . Notice that the programme is a fixed rule like Newton’s laws. 11. . Notice that this random choice is meaningless for a few trials. 22. the most common being a sequence of numbers between 0 and 1 . The programmes that generate a random sequence are like a fast die and are designed very much from experience. 2 and 3 we can first generate a random number x. The computers today are usually equipped with programmes that generate random number sequences. From Chapter 10 to Chapter 12 we have discussed random sequences. Probability is meaningful only after repeating many times. Having the random number program we can choose “at random”.3 if x is greater than (We have assumed that the probability distribution of x is uniform between 0 and 1.
. After repeating many times.s i .s j (under the same h i ) to that of si + si is e Therefore . This method satisfies the condition of detailed balance. Step (b) picks an element to flip. Step ( a ) determines the initial state and any configuration is possible. otherwise not). A E is the energy change due to the flip. If the energy increases.A E I T .e. then ( the average value of any A^(s) can be obtained as (224) . 2 . Step (c) determines whether to flip or not. the probability in (c) has a sufficiently large sample for itsjustification. . (b) Take one s i . . If AE > 0. (22.e. . then use the probability eVAEITto change si to si (i. (d) Repeat step (b). i. . the trajectory. . hi = i G Jii si 2sihj . and hi is the magnetic field on si created by the neighbours. and if x < eAE/T change s i . . If the energy decreases. take a random number x .1) where sj = 2 1. the change to .N . Calculate hi and A E .s2 ( t ).s i . the probability of si + . The simulation proceeds as follows: (a) First randomly take a set of values of s j .2 ~hilT i (223) P(si I h i ) is the conditional probability under a fixed hi. . Let s t ) E [ ( s1( t ) . then change si to . sN( t )] be the configuration at time t.s j occurs with a probability e . AE . time increases by one unit. This simulation process is quite simple. i. i = 1 . Every time we perform processes (b) and (c).2 ) (c) If A E G 0. Repeated calculation will produce a sequence of configurations. then change si to .e.392 STATISTICAL MECHANICS Let the total energy be (22. Here.
The configuration of t = 0 was chosen at random and it requires some time to “reach equilibrium”. The reason is that some configurations in the beginning are too far away from the typical equilibrium states. In the above example. the steps of simulation are: (a) From the configuration at time t . 2 . and . .e. If y falls in the j . The steps of simulation here are more complicated than the above one. Notice that (22 . each of length Wi/fl. (229) and let the new time be t + t’ . We discard the configurations before t o .5) (226) i. . Let c l a w i i . Now let us look at another model of motion. (227) and (228). = e si hi IT cosh(hi/T) This is also a model of motion. change sj to . Using this model for simulation is very simple.s j . Let Widt be the probability for si to change in a time interval d t . .t h section. calculate Wi and fl. (b) Take two random numbers x and y and let t’ = (lnx)/fl . This problem will be discussed below.NUMERICAL SIMULATION 393 where 7 is some large number. in each step we pick an element and then determine whether to flip it. . (228) According to (225). i = 1 . w. and then s‘ changing in d t ’ is e. Now return to (a). 1) into N sections. (c) Divide the interval ( 0 . (227) Starting from t and after an interval t’ the probability of having no change is So the probability of having no change up to t’. this model also satisfies the condition of detailed balance. This example determines the interval of rest at each step first. N .at’ W id t ’ .
Each element is acted upon by random but uncorrelated forces. so the situation is relatively complicated and requires further discussion.e. elements far apart are uncorrelated. Each element moves due to these forces and is also influenced by the other elements.4. the probability of change of si is determined by the environment of si. . The calculation of the average value is somewhat different. si11 . The only link with thermodynamics is through the probability of change eAE’T. I sj. These steps do not mention the basic assumption. there are n changes. P ( s i . . . . sl!.11) Here si. 22. Temperature is thus introduced. each element is not independent. The condition of detailed balance is that under the same environment the probabilities of the forward and reverse directions of si + . si . Therefore we get (223). . If the neighbours sj of sj are fixed (neighbours in the sense Jij # 0). Equation (223) can be slightly generalised. The change of si is due to a fixed hi. where ti is the interval of rest before the k . . ) is a joint probability . Therefore.394 STATISTICAL MECHANICS then which element to flip. Here we have interaction between the elements. The influence is through neighbouring elements whose interaction is not zero.s j are equal.t h change and s ( t k ) is the configuration at that moment. In Chapter 3 the discussion of detailed balance was limited to the ideal gas. these steps are a very reasonable model of motion. I si . In time 9 . So. . . ) P(sj. J=c k. i. . f (2210) k = k. From (2211) we can get (223). ) I1 = e 2sjhjlT (22. Conceptual Problems to be Noted The steps of simulation in the last section have very obvious meaning. hi is fixed: P ( S i . I1 . . Therefore. .+n t. include all spins having interaction with si and possibly other elements as well. k=k. . s Ii .
e.) . Likewise (sl. which is therefore quite large. The equilibrium under the basic assumption does not involve the single configuration probability in (22. The sample here is the configurations of the trajectory. . . s’ .e. i. The reader can review it. In carrying out the numerical simulation. we have said much about these. those simulated configurations. The above probabilities must be defined from these configurations. The probability of such a large region is meaningful because the trajectory remains in it for a sufficiently long time to define probability. But “infinitely long” has no physical meaning. . 2N configurations.1 0 ) points in this region. a conclusion from an infinitely long trajectory. This region is determined by the values of sl. s3). . . . . very obvious. s 2 . We have repeatedly emphasised that the definition of probability requires a sufficiently laige sample to do the statistics. Equation (221 1) is suitable for small numbers of elements like (sl. i.~ I=~ = C~H/T 3 S hence (22.e. s represents one point in configuration space and (sl. (2212) This is usually regarded as a proof of equilibrium. s 2 . Equation (2212) is only a mathematical conclusion. s3) represents a region of 2 ( N . . Can we use simulation to calculate the thermodynamic potential? The partition function can be written as an average value e . . sl0 with 2 ( N . It involves the volume of large regions and equilibrium properties are determined by large regions. the result of this abstract discussion becomes crystal clear. even if the trajectory includes millions of configurations. i. In the chapters on the basic assumption and probability. then the conclusion of (221 1) is . .12). . . . . . Nevertheless there are faults in this approach.13) . These have been discussed at the end of Chapter 10. . sf. slo) represents a region in this space. Notice that ( s l .e. . s 2 . that would still be far from 2 l o o . If N is not a small number.3 ) points. then the definition of P ( s l . s N ) requires many configurations i. .NUMERICAL SIMULATION 395 If si. The reader will benefit from doing some of these calculations. . . . s l o ) or (sl. If N = 100.. include all the elements. .
and eHJT is a superlarge number.1 we mentioned that in calculations we typically take Nto be 100 or 1000. But this l? is much smaller than the whole r 2N 20.14) Therefore. this ratio then is extremely small. of volume r=eS . it is not easy to take a point from the region of motion. (2213) is useless. The calculation of entropy from the trajectory is the main theme of Chapter 25. therefore. Why do we need to do the simulation? In fact. Is it too much to hope that thermodynamic properties could be deduced from this relatively small N ? The situation is actually not so bad. we can obtain Z and F. using (2212) as the probability it is still not possible to calculate the average value by random integration. The region of motion is very large.5N In 2 then configuration space  r  2 ° . Of course.16) If N = 100. then in the region of motion.TS we must directly calculate E = ( H ) and S. If we want to calculate F = E . but not 10”. which can then be regarded as part of the large body. (2215) If S 0. In Sec. Its change is also superlarge.5N (22. The reason is this: H is a large number of order N . If A (s) is a macroscopic variable. This “infinitely long” time indicates a time longer than the superlarge number eHIT.) The only way to enter the region of motion is to let the system evolve. 5 N . hopeless to choose one point out of the configuration space and hope that it comes close to the region of motion. it seems that we need only one configuration s and then we know A (s). But this is impractical unless N is very small. We only need to have a model much larger than the correlation length. It is. the majority of the s have nearly the same A (s) : (22. the boundary becomes an . Macroscopic bodies has the property of independence of parts. In Chapter 6 we mentioned that in the region sampled by the motion nearly all the configurations have the same macroscopic properties. (Because of this. Unless we use an infinitely long time.396 STATISTICAL MECHANICS Averaging eHJT with respect to time. 22. gradually bringing the configurations into the region of motion.
etc. Devise a random simulation of the motion of a gas. Discuss how to modify the various calculational rules in Chapter 7 for not too large N and V . many interesting problems involve long correlation time and length. 4. many conclusions (e. There are many technical problems in numerical simulation.5 and 1 1.3 discusses the motion under a constant temperature. Discuss the boundary problem of the models. e. Assume a constant temperature situation. that of (2214) ) must be handled with care. In addition to the metastable state. However. or doing the simulation ourselves. Many nonequilibrium phenomena. 5.) It is also ideal for dealing with metastable states. (Refer to Problem 3 . Devise a simulation for the constant energy and the restricted energy methods. then N'13 and Nl12 will not be very small numbers. Then calculate the difference in energy. 3 . Berne (1977) and Abraham (1979). models with n o boundary. The metastable state is an equilibrium state on short time scales. The main one is that the model is too small or the time of simulation too short. . for those situations when the basic assumption of statistical mechanics is not too useful. simulation is in the stage of accumulating experience and does not have a very solid theoretical basis yet. Because N is not a macroscopic number. we must be careful.e. Each step moves a molecule by a small distance. Use the Ising model as an example to discuss the problem of time for h = 0 and h # 0 at low temperature. Therefore in reading the literature. a Problems 1. Notice that we can ignore the kinetic energy of the molecules. This method is quite successful. crystallisation and melting.g. ) 6.g.7. changes appear. Usually a wrong interpretation of the result will lead to improper conclusions. simulation is an invaluable and the only dependable tool for analysis. 2. For longer time scales. (The Monte Carlo simulation on a larger time scale can be regarded as a very realistic model of motion.NUMERICAL SIMULATION 397 important problem. When N is not large. can be analysed by simulation and many interesting results have been obtained. Section 22. i. The usual practice is to use cyclic boundary conditions. Review Problems 1 1. At present. aFor bibliography we can start from Binder (1979). Simulation is a solution of the equation of motion so its application is not limited to equilibrium states.
the reader can devise some himself. 2 . r I . Devise these programmes. Kirkpatrick. we can simulate the zeropoint and obtain the minimum value of the total energy function H ( s ) . the result may be a metastable state but not a ground state. as we lower the temperature and do the simulation at the same time. and ri are limited within the volume L 3 . p is the pressure. . . Each method of Chapter 7 has its own steps for random simulation. Use H ( b r . So. . To determine the lowest cost we can use the method of random simulation. Remember the discussion of Problem 3. Therefore. Devise the programme of simulation and discuss the usefulness of this method. 8. There is a method of writing the positions of the molecules as b r i . the volume becomes a variable of motion. .e. For example. If the temperature is very low. . (This method of obtaining the minimum value is very useful in industry and was initiated by physicist S. If we use the cyclic boundary it is not easy to change the volume. Devise a step to change s and then we do the simulation. random simulation can be used as a tool for finding the minimum value. . Here L is a fixed volume. Starting from high temperatures. i = 1. .398 STATISTICAL MECHANICS 7 . . b r N ) k p b3 L 3 to do the simulation. 3 9. The tour is required to pass through certain places.N . Random simulation has many applications. . s can represent a tour around the world and H ( s ) the cost of this tour. The reader can discuss the problem of metastable states i.r N ) as its phase space. and is not limited to statistical mechanical problems. Then we take ( b . . . then zeropoint energy (the ground state) can appear.r 2 . The low temperature random simulation gives configurations of low energy. . . .) This method has many difficulties depending on the problem. H ( s ) can be any function and s any variable. How do we simulate constant pressure? Under a constant pressure. we then lower the temperature gradually.
Chapter 24 gives some examples of the “echo phenomena”.PART VI THEORETICAL BASES The preceding parts gave a general introduction to the concepts and applications of statistical mechanics. Chapter 25 introduces a method to calculate the entropy directly from the trajectory. clearing up some misconceptions about entropy. aiming at a further understanding of their meaning. Now we return and analyse some basic concepts in greater depth. These phenomena are also interesting by themselves. Chapter 2 3 discusses the three laws of thermodynamics. Chapter 26 discusses some problems of principles awaiting solution. thus linking all the basic concepts of statistical mechanics with molecular motion. .
1) . and further explains the role of the time scale. The distinction between heat and work is defined through adiabatic processes. It summarises the relation between the basic assumption of statistical mechanics and thermodynamics. Adiabatic Processes The first law of thermodynamics is. consider a gas in a container. In Chapter 2 we have discussed this. The usual violations of the third law are due to the occurrence of irreversible processes. Lastly we discuss the third law and the role of reversible processes. it divides the energy transferred into two kinds .Chapter 23 LAWS OF THERMODYNAMICS This chapter discusses again the basic concepts of thermodynamics first introduced in Chapter 2. where s is the configuration of the body and L is an invariant quantity. including the third law. L ) .heat and work. first of all. 23. We first discuss why the entropy is unchanged in an adiabatic process and then address the problem of the increase of entropy. For example. Then H = kinetic energy + interaction between molecules + N U(ri) i=1 (23 . a restatement of the law of conservation of energy. That the entropy must be a maximum is part of the basic assumption. Now we consider the adiabatic process. Let the total energy function of a body be H ( s . Secondly.1. The influence of the container walls on the gas molecules can be represented by a potential energy U.
moving one wall so that L is changed uniformly to L + A L . 231 If the environment of this gas is adiabatic. the potential energy is a function of the distance between the wall and the molecule.LAWS OF THERMODYNAMICS 401 The shape of U(r) is depicted in Fig. z directions. Assume that this takes time 3 . and is similar in t h e y . L ) is a conserved quantity: H(s(t). Fig. When x is near L . U(r) = v ( L . 231.x) . U increases.L) at a = 0 . Here L is regarded as an invariant quantity or a parameter. (233) Now we change L . . (232) i.e.= Jo d t dt dL dt aH (234) dt = A L ( y ) . then after the movement the change of energy is 3 dH 7 aH A E = Jo d t . representing the repulsion of the wall on the molecules.1) without exchanging energy with the outside and U is a fixed function. Near the walls. then the total energy is (23 . Under this situation H ( s .
In quantum mechanics the total energy operator H may depend on a parameter A . If the time for the process ’J is much longer than the correlation time of the molecular motion. i. (236) where ( a H / a L ) is the average with respect to time. L ) of (235) can be regarded as unchanged. then the L in H(s.402 STATISTICAL MECHANICS (23 . The above is purely the result of mechanics and is unrelated to thermodynamics or the basic assumption of statistical mechanics. If AL is very small. then the energy will not change. the change of energy is d E = ( TaHd L ) .a v / axi is the force of the wall acting on the ith particle. Let us explain. The conclusion is therefore: in an adiabatic process.5 ) Notice that dL/dt = A L / J and we have used (233). (237) where A is the crosssection of the container in t h e y . Do not confuse the adiabatic process here with the adiabatic process in quantum mechanics. where n . Hence on the one hand 3 must be longer than the correlation time. if L is unchanged. otherwise the process is not adiabatic. then ( a H / a L ) is an equilibrium property (assuming that the body is in equilibrium before and after the process). The change of energy is caused by the change of L alone.e. z directions and . Hence each En ( A ) is a function of A . Equations (236) and (237) describe of course one of the adiabatic processes as discussed in Chapter 2. The averages in (234) and (235) are over the duration of the process. However 7 cannot be too long. From (23 2) we get (z)dL = pdV . but on the other hand 7 must be short in a macroscopic sense.
In other words AE .e. each energy is En (A ( t )). The condition of an adiabatic process requires 3 to exceed the correlation time of fluctuation. (b) If the initial state is n . If we apply (238) to a macroscopic system. This theorem is valid for any isolated body. This is the theorem of the adiabatic process in quantum mechanics and the condition is that the time 7 over whch X changes must be very long 7 >> h / A E . Starting from the basic assumption. Therefore.eN .8) where A E is the energy difference between two neighbouring E n . of course. i. 23. But the condition(238)of thistheorem is not valid for a macroscopic system. 3 . including the adiabatic system discussed here. we can get (236). (727) ): (2310) . and this is a small time unrelated to (239). We can now define work as the change caused by changes on a macroscopic time scale and call heat the energy change caused by microscopic motion. an unrealistic condition.2.LAWS O F THERMODYNAMICS 403 is the quantum number of the stationary state. the time required for the trajectory to pass through each configuration in the region of motion).e. we now look at the relation between the change in entropy and adiabatic processes. From (a) and (b). this theorem in quantum mechanics cannot shed any light on the adiabatic process in thermodynamics. (23 . (239) This is. A E will be very small and 3 must be larger than eloZo(i. then it will stay in the state n. Adiabatic Process and Entropy The above analysis does tlot involve the concept of entropy. We use the method of “restricted energy” to calculate the entropy (see Eq. If we change X slowly we have the following results: (a) Each energy level keeps the same form and it is only necessary to substitute X ( t ) .E/I’(E) >> h I ‘ ( E ) / E .
This is a most important result directly derived from the basic assumption and the result (236) of mechanics. Therefore. Now from (236) we can see that if dL and d E are caused by an adiabatic process. dS=($). thermodynamics and the basic assumption of statistical mechanics agree perfectly. we get dS = where (z). [dE(g)dL] ..lo). the quantity (2312) calculated by the method of constant energy. i.404 STATISTICAL MECHANICS This method is used purely for the sake of convenience. The other factor is the average value of aH/ a L . then d S = 0.e. dE + ($)E dL > S S The last factor in the last line is just (aS/aE).. As seen from the above result. (2313) The quantity ( a H / a L ) is a macroscopic variable. Differentiating (23 . . If the basic assumption is established.12) should be the same as the time average (235). so the average value of (23. this result follows. This is the result from the discussion of the previous chapters. The basic assumption does not mention the process and (236) does not mention the entropy. In Chapter 2 the concept of entropy was derived from adiabatic process.
Therefore. . To take an example. The entropy is defined by the region of motion of the trajectory in phase space during this period of time.15) The second law of thermodynamics goes beyond equilibrium and points out the direction entropy changes. this d t must be greater than 3. .e. We have sufficient time to define entropy before and after the process.16) where T is the mean free time of the molecules. we must note that the definition of entropy requires some observation time 7 .13) can be written in a more familiar form +(z)dL . dE = T dS where (2314) Hence force is the isentropic derivative of energy with respect to displacement. If we want to discuss the rate of change of entropy d S / d t . The observation time 3 must satisfy (23. 1 T 23. Equation (23. Thus in time 3. 2 3 . Of course we can discuss the entropy before or after leakage and avoid discussing entropy during the process. the rate of leakage must be small: (2317) where N 1 is the number of molecules in the left. . changes are of course concerned with nonequilibrium states. let the gas in Fig. There is no instantaneous meaning for entropy. =(%>. First. N 1 can be regarded as unchanged and the entropy can be calculated. The Second Law of Thermodynamics (23 . If the leakage is too rapid.LAWS OF THERMODYNAMICS 405 Through (2313) we obtain (3=(E>.3. to speak of entropy at a certain time we must use a time scale larger than 3 . Initially the right half is empty and the gas slowly leaks into the right. i. entropy cannot be defined in this process.2 be separated into two parts by a wall with a small hole.
From the basic assumption. under adiabatic conditions. the entropy of a system cannot decrease. Those rapidly moving molecules in the left can . If the motion is very fast. where r is the total number of configurations in the region of motion. The previous section pointed out that entropy is unchanged before and after an adiabatic process. the adiabatic process discussed above is a slow process. 232 According to the second law. the molecules are restricted to only one part in a short time. the region is large and so is the entropy. This naturally gives the result that entropy cannot decrease. the region will grow. Within a short time. molecules can go to the other part. and use the basic assumption on the system plus the machine. At least no man or machine can achieve such rapid motion. entropy has already been assumed to be maximum because all possibilities are included. but once time is lengthened. (See the discussion on the various esembles in Chapter 6. There is a hole in each flywheel. Let us consider an example. will this change decrease the entropy? This problem is beyond the basic assumption. That is to say. much faster than the correlation time of the molecules. it may be true that the region of motion is restricted because time is not long enough for the system to go through all configurations. but to prove its infeasibility is also difficult. Because the hole is small. 2 3 . The leaking gas is a good example.2 (see Fig. Of course. When the numbers are equal on the two sides. These proofs assume that the system and the machine are in equilibrium. including all configurations permitted by conservation laws.406 STATISTICAL MECHANICS I Fig. When the time is long. the hole will no longer matter. the explanation of this law is quite obvious. There are proofs in the literature that this cannot be done. The flywheels are revolving rapidly. Hence to perform work on a system or to allow the system to do external work cannot decrease the entropy. Suppose a pair of flywheels are fitted to the hole in Fig.) So the fact that the entropy can only increase is an ingredient of the basic assumption. The basic assumption says: entropy is l n r . 233). Suppose initially the right chamber is empty.
. where a is the width of the small hole.. In fact this is not so. this pair of flywheels allow fast moving particles to go through while keeping slow moving particles to the left. the molecules after passing through hole 1 can pass through hole 2. * . .... If the velocity v is large enough.LAWS OF THERMODYNAMICS 407 ( a ) Flywheel with a hole . then the particle after passing through hole 1 will not pass through hole 2.e. a . ( b ) T h e flywheels fitted to the hole in Fig. At first sight. * . . .. and w is the angular velocity of the flywheels. : s 1 . . . . a . If v < aw. Of course there is still a chance to go to the right after many collisions. Hence the temperature on the right is higher than that on the left. violating the second law. i. 233 pass through the hole easily. 232 Fig.
this law is very easy to explain from the viewpoint of the motion of the molecules. Therefore. Therefore. it is not difficult to determine this constant. (23 . The entropy difference between two equilibrium states 1 and 2 is dQ C = . The choice of this constant was discussed in Chapter 2 to Chapter 5 . The entropy is S . From the viewpoint of thermodynamics. at very high temperatures a gas A B will dissociate into a mixture of gases A and B . Once we use this assumption.19) aProposed by Nernst in 1905 .18) where dQ is the heat added and the integration is along the reversible process from 1 to 2. More precisely. motions stop (except for the quantum mechanical zeropoint motion). In thermodynamics only the difference of entropy is defined. Because changes cease. the left and right chambers together with the flywheels will be at the same temperature. + Se + N l n 2 . 23. this law points out that as T + 0. after a long time. Because we are now quite familiar with statistical mechanics. The Third Law of Thermodynamics The content of this law is: the entropy tends to zero as the temperature T tends to zero. There is no exception to this rule. The following is a standard example illustrating the content of the third law.a From the concept of kinetic motion. = S. this law endows entropy with an absolute value.= heat capacity . no violation of the second law is possible.408 STATISTICAL MECHANICS Initially the molecules on the right indeed have higher velocities. the region of motion contracts and entropy also tends to zero. once we prove the basic assumption. But the flywheels are continually bombarded by the molecules and molecules accumulate in the space between them. the change of configuration also tends to stop. The above assumes the validity of the basic assumption (temperature is the same throughout).4. as T + 0. dT (23 . Hence entropy is defined up to an integration constant. If atoms A and B can combine to form molecules A B . we have proved the second law. So energy can be transferred by conduction.
then the entropy constant of all substances are fixed. A z B 3 . Now we can use experimental results to see whether (2321) .LAWS O F THERMODYNAMICS 409 The term Nln 2 is the mixing entropy. Since every element is composed of electrons and the nucleus. the entropy of the ideal gas plus some corrections.g. etc. (2321) together with the latent heat contributions in phase changes. These three heat capacities and latent heats can be determined experimentally. from T = 0 to high temperature. N being the number of A and also that of B. When the third law was proposed (1905). B and AB respectively. i.e. protons and electrons. This choice is identical to the choice made in Sec. the calculation of entropy is simple. The difference of the entropy at high temperature and that at T = 0 can be determined by (23. At high temperatures. then SA.(0). The third law puts all S ( 0 ) to be zero. According to the above conclusion the entropy of any body. their entropy constants can be determined by S’(0) and S B ( 0 ) . For the same reason. Once we have chosen the constant S A ( 0 ) and S B ( 0 ) . During the change of phase.3.is fixed. i. We have indeed no other choice. can be obtained by measuring the heat capacity.e. we have to add the latent heat terms. ’ ’ In these equations C C and CAB are the heat capacities of A . then the entropy constants of all the compounds are fixed. AB4. Equation (2319) is the result of thermodynamics (see Chapter 2). if we fix the entropy constants for neutrons. Obviously from the definition of entropy there should be a basic constant independent of the structure of the substances. Obviously if we fix the entropy constants of all the elements. statistical mechanics had not yet come into being.18): (2320) . and the nucleus is composed of neutrons and protons. S(T) = 5 T 0 C d T ’ 7 T . 3. for any possible compound e.
people had used the third law and (2321) to derive the entropy of an ideal gas. it is in the metastable “white state”. just like the coexistence of water and vapour. The grey state absorbs heat to change to the white state. the entropy of the white state is (2322) where C1 is the heat capacity of the white state. At To the grey and white states can coexist. Note that A + B and the combination A B are different phases of the same system. (2323) where L is the latent heat t o change from grey to white. When metallic tin is cooled below To = 292K. Now let us return to the discussion on (2319) and (2320). only one structure is stable while the other are metastable. At To.410 STATISTICAL MECHANICS agrees with the theory at high temperatures. even before the advent o f statistical mechanics. Equation (2320) indicates that there exist relations between the heat capacities of these different phases. bSackur (1911) and Tetrode (1912). (If time is infinitely long. The solid phase of many elements exhibits different structures. Put the entropy at T = 0 to be zero. This is a very stringent conclusion. and this is the relation between C1 and C. there is a relation between the heat capacities of the various structures. . In fact. To generalise.) Their heat capacities can be measured. Hence it suits our definition of equilibrium. The most common example is tin. .b identical to that determined later in statistical mechanics. if a molecule exhibits different structures. Then the three integrals of (2320) must agree with (2319) when T is very large. The stable structure is a semiconductor called the “grey state”. They discovered that the Planck‘s constant appears among the constants. Below To the white state is metastable. but changes very little within the usual observation time. The entropy S1 can also be obtained by integrating the heat capacity C of the grey state : . Many experiments can be performed to test this property. Equation (2322) and (2323) must be equal. Within ordinary observation time these structures are equilibrium states.
will be influenced by the nuclear motion. If the motion of the nuclear spins is unrelated to the other motions or the structure of the body. How low must T be in order to say that entropy becomes zero? This depends on the energy scale of the internal motion of the body. (If N l n ( 2 1 + 1) is added to both (2322) and (2323). When the temperature is lower than all the energy scales of motion.) If the temperature is so low as to approach the interaction energy of the nuclei. the motion with the lowest energy is the change of the nuclear spins. The meaning T + 0 indicates a temperature lower than this interaction energy. C1 and C. Experiments shows that S(0) = S(T)  0 C(T) dT T ’ (23 25) is not zero (in fact always greater than zero). it can be regarded as zero. . we mean (23 24) i. For example in the case of tin. because of the constancy of the the number of nuclei. Although this is not zero. ‘Wilks (1961). In solids. The Amorphous State and the Third Law About thirty years ago some chemists did experiments to determine the heat capacities of many relatively complex solids such as ice or solids made up of molecules (not monatomic solids). the relation between C1 and C. The results of all these experiments obviously violate the third law of thermodynamics. Notice that when we say entropy tends to zero. the entropy per molecule tends to zero. is unchanged. Later there were experiments to determine the heat capacity of glass. Let S ( T ) be known for the high temperatures gaseous state (the statistical mechanical result of a gas). If the temperature is not too low. It is no trivial task to produce low temperatures in the laboratory. the nuclear spins can still have entropy. each nucleus has an entropy of In ( 2 1 + 1) where 1 is the nuclear spin. At this time all the motions stop. For these experiments see references.e. 23. The energy scale therefore refers to the energy scale of each molecule. If the temperature is higher than this energy. it will not influence the above result. then it can be ignored.LAWS OF THERMODYNAMICS 411 In the above discussion we have not been careful about the limit T + 0.5. the nuclear spins are unrelated to the white or grey states.
some rapidly changing variables are frozen. those variables which previously could not change during the observation time will now start to move. Their entropy is N l n ( 2 1 + l). But substances in nature almost invariably interact. But during these experiments some elements of irreversibility are introduced. Therefore. d S = d Q / T . It is not hard to see that dQ d S > .2 6 ) . Notice that if the ground state of a system is degenerate and the logarithm of the degeneracy is proportional to N . See Chapter 25. Although different structures look the same on a large scale. . having many time scales.25) must be positive. As temperature increases.18) involves integration over reversible processes. some variables are unfrozen. Some variables of motion change rapidly and some slowly. According to ( 2 3 . entropy is determined by the volume of the region of motion in configuration space. At low temperatures. S(0) defined in (23 .2 6 ) If there is no time limit.412 STATISTICAL MECHANICS Actually these experiments do not violate the third law because the definition of entropy in ( 2 3 . T (23 .+ 0 motions stop and entropy also tends to zero. In many models there are ground states with nonzero entropy. the movable variables wl increase the amplitude il of change. and the third law itself is not violated.. as T . As temperature is lowered a little bit. These time scales are very sensitive to the temperature. The explanation of these experiments are as follows. From the viewpoint of kinetic motion. The process is irreversible. The third law cannot be violated. so how should we define entropy? What is the d S in (2326)? According to the basic assumption. From this viewpoint the experiments do not fulfill the conditions of the third law. In addition. The reader may now ask: if irreversible processes are unavoidable then entropy cannot be determined by heat capacities. can we define this region? These questions can be answered by analysing the trajectory of motion. When temperature rises. Cooling from high temperatures can produce different frozen structures. then S(O)/iVis not zero. This isunrelated to the amorphous states as discussed above. ice and glass are amorphous substances (see Chapter 17). no matter how low the temperature becomes. The simplest example is a system of N spins 1 without interaction. on a small scale they are different. freezing after unfreezing will not return to the original frozen structure. If the time scale is not clear.
(c) How small should A L / 3 be so as to avoid exciting oscillations in the fluid? (d) If the mean free distance of the molecules is larger than L . (b) These equations are differential equations. If ( a H / a L ) in (2311) t o (2313) is the same as the time average in (235) and (236).g. sound wave and diffusion can lengthen the correlation time considerably. (a) Write down the fluid equations for the motion in this process. e. 0. L is moved uniformly to L + A L in time 3. 3. then entropy is unchanged during an adiabatic process. and hi and different at different sites. long wavelength oscillations can persist for a long time. The reader can give several other examples.LAWS OF THERMODYNAMICS 413 Problems 1. The total energy is (2328) i i Each spin has three states. We only consider the rotation of the molecules. 1. The time during the process 3 must be much longer than the correlation time in order t o establish the above conclusion. The differential equations are like the single particle wave equation. The distribution is Ai are fixed. N . 231). 8.1 (see Fig. Review the analysis of the hydrogen molecule gas in Sec. . 2.2. (b) The ground state of the Hz molecules is that with spin 0. the correlation time can be taken to be the mean free time. i = 1. . what will be the conclusion then? 2. and the change L is the change of the boundary which can be regarded as a change of parameter. (a) Use the integration of the heat capacity to obtain the entropy when T + 0. (23 27) where N 1 is the number of molecules with spin 1. In a gas. . This problem is very similar to the adiabatic process in quantum mechanics. Prove S(0) = N 1 I n 3 . but are . si = + 1. 23. So the entropy obtained by integrating the heat capacity may not be related t o the ground state. A body is made up of N spins si. Now consider the process in Sec. But as pointed out by examples in Chapter 20. . This should be quite obvious.
the temperature is lowered by a / n . This model is nearly the same as Problem (21. Suppose the rules of motion are as follows: If A i > T In t . (c) Use the heat capacity d E / d t in (b) to perform the integration to get the entropy.6). then si will not change. then si can change. If Ai < T In t . heat capacity and entropy. Then discuss the third law of thermodynamics. Starting from T = 2 a . Here t is a constant and can be regarded as the observation time. and make a comparison with the entropy directly derived in (b). (a) If T >> a . . calculate the total energy. Notice that if si cannot move. calculate and sketch the relation of the various quantities with T.414 STATISTICAL MECHANICS (2329) Assumea >> h . then its entropy is zero. (b) Suppose this body is cooled gradually and at each period of time t . Let n = 100 and b = a/10.
which is a simple classroom demonstration. “spin echo”. In this chapter we shall discuss three dramatic experiments.Chapter 24 ECHO PHENOMENA Entropy comes from random motion. We have given many examples showing that the region of motion and entropy must be determined by motion. After that we shall mention aHahn (1950).a He reversed the directions of the precessing spins. causing the dispersed spins to restore their directions. This is a very simple experiment for classroom demonstration. (1968). but he must at least be credited for popularising it.) After Hahn’s experiment. The viscous liquid bottle makes the dispersed colours of the viscous liquid return to their original positions. the same is true for the other experiments. “viscous liquid bottle” experiment and “plasma echo”.e. We shall study carefully the demonstration of the viscous liquid bottle. many similar experiments were done including the “plasma echo”b Hahn’s aim was not to discuss entropy. bMalmberg ef al. . The first is the experiment of nuclear spin resonance by Hahn. i. Irregular arrangement or random distribution will not be related to entropy if they do not change with time. O’Neil and Could (1968). (Hahn claims that he is not the first to invent this demonstration. The techniques and results of these experiments are extremely interesting and invaluable. Then we analyse the principles involved. Hahn has used it to demonstrate his spin echo experiment. We have repeatedly emphasised the importance of motion. Our aim is to use them to further discuss entropy. These experiments are called the echo phenomena in the literature.
. 24.g. just enough to show the colour.1. The demonstration proceeds as follows: (a) Use a fine glass tube to suck up some coloured viscous liquid and transfer it to the bottle.. The construction of the bottle is as follows: two concentric cylinders (made of plastic or glass) are installed so that the outer cylinder is fixed and the inner cylinder can be turned by a handle (see Fig.1). The viscous liquid must be transparent and the more viscous the better. so that a coloured line is formed in the bottle. 241 Viscous liquid bottle demonstration. involving thought experiments. and involve more complicated theory. Fig. The dimensions are not crucial. is adequate. Use a beaker and mix some dye with the viscous liquid (we can use food dye. A small amount of dye. These cannot be demonstrated in the classroom. The molecules of the dye are distributed uniformly in the viscous liquid. Demonstration of the Viscous Liquid Bottle This demonstration is extremely simple. What we present are simplified versions. e. The distance between the two cylinders should be about 1 cm and the diameter of the outer cylinder about 10 cm. 24.416 STATISTICAL MECHANICS the spin echo and plasma echo experiments. In the space between the two cylinders is some viscous liquid. so that the reader can grasp the essence of these phenomena. ink or other dyes). corn starch paste.
(c) Turn the handle in the reverse direction for the same number of turns. however. unlike an ideal gas. Hence the coloured line will become dispersed. and the probability of distribution is l / V o . 24. Analysis of the Demonstration We first present an incorrect analysis and then give the right one. of the coloured line.1) where N is the number of dye molecules. The incorrect analysis goes as follows. Some people argue that when the dye molecules are restored. This kind of explanation is a misconception about entropy and is untenable. It is unbelievable unless one sees the experiment himself. When the hand pushes the handle. entropy flows through the handle of the bottle to the hand of the experimenter. The usual interpretation of entropy is the degree of randomness and the second law of thermodynamics expressed the inevitability of becoming more and more random. they are not moving. So it has an entropy (24. and this is unrelated to the arrangement of the dye molecules.2. .ECHO PHENOMENA 417 (b) Turn the handle in one direction several times. the experiment demonstrates that the dye molecules are distributed randomly like the molecules of an ideal gas. N / V o is the density of the dye molecules in the beaker. The molecules of the dye are restored to their original positions. In fact. the uncertainty in position of each dye molecule is the volume V. of course. there is no contradiction between this phenomenon and the second law. After one or two turns the colour disappears. and after ten or more turns the molecules of the dye are completely dispersed. Surprise! The coloured line reappears. the restoration of the dye molecules from the dispersed state. it does work no matter what the direction of motion is. Before the dye is put into the bottle. At this time. What is surprising in this experiment? It is. so that the molecules near the inner wall are moved along while those near the outer wall do not move. This demonstration seems to say that the second law is incorrect. and u is the entropy caused by the small amplitude vibration of the molecules.
This is unrelated to the distribution of molecules. since this heat current would be easily felt. Hence the entropy is N o . (It is not correct to say that entropy flows to the hand. This phenomenon of positional restoration is hardly surprising. and entropy increases slightly. Because the dye molecules are few (the dye is very dilute in the beaker). The volumes Vo and V are not related to the amplitude of motion of the dye molecules. its . the probability distribution of the position of the dye molecule is 1/ V where V is the volume of the viscous liquid bottle. the region of motion and the probability distribution are determined by the amplitude of motion. is stationary. This analysis seems quite reasonable. turning the handle several times just heats up the whole system slightly. unrelated to the distribution of the dye molecules in the liquid. Each dye molecule. violating the second law.418 STATISTICAL MECHANICS After the handle is turned several times. Now we look at the correct analysis. Within the observation time of several hours. In defining entropy.S o = Nln(V/Vo) This is the increase of entropy due to the dispersion. Because the dye molecule does not move. (242) = Nln(V/N) i.1) and (242) are not correct. but obviously there are faults because in turning the handle backwards the molecules of the dye are restored to Vo and entropy decreases. except for a slight vibration. entropy is that of small amplitude vibrations.1 second. which are moving incessantly. Hence the entropy is S = ln ( V N ) + ~o + N + No . Under this time scale. no matter in which direction the handle is turned the vibration of the viscous liquid and the dye molecules will be larger. about 0. If we use these concepts to define entropy we are naturally led to the wrong conclusion of (24. diffusion does not occur. S .) The fault of the above analysis lies in misconceptions about entropy. Hence. The time scale of this demonstration is the time of persistence of vision of our eyes. This is quite unlike the molecules in a gas. We emphasise once more that entropy is unrelated to the uncertainty of observation or the knowledge or ignorance of the observer.1) and (242). they are surrounded mostly by the molecules of the viscous liquid.e. . Hence (24. independent of Vo and V . Because of friction. We can also look at the problem this way.
we should use dyes with large molecules. so r can be about ten days or more. the dye molecules will be spread out and cannot be returned to the original positions.1 ) and (242) will then be correct..3. This point has been discussed under frozen impurities.1 second to several minutes. The important time scale is 1 = (v/N)1’3 (243) where D is the diffusion coefficient of the dye molecules. 0. the entropy of the dye molecules is the same in the dispersed state or the restored state (neglecting the heat produced by friction). If we use water instead of corn starch paste. Dispersed or restored. This does not contradict the second law. 24. In a fixed magnetic field. If the liquid in the bottle is not highly viscous. and 1 is the distance between the dye molecules. then the dye molecules can move rapidly. It is like a trapped impurity in a solid. After a few days. Hence the entropy is related t o the total number of dye molecules. For the above demonstration. Equations (24. Of course. Note that diffusion is related not only to the viscosity of the liquid but also to the size and shape of the dye molecules. Spin Echo Experiment Here we mention briefly the principle of Hahn’s experiment.ECHO PHENOMENA 419 coordinates and surroundings are unchanged (except for a small amplitude of vibration). To change the viscosity is to change D and the time scale correspondingly. they are equilibrium states on this time scale. i. The usual food dye is quite ideal. Therefore. The reader can find the details in the original paper. The dispersed state or the restored state have the same entropy. (24. far exceeding the observation time of the equilibrium states. Has entropy increased? The answer is no. The time scale after a few days is still the scale of human vision. a magnetic . The positions of the dye molecules are invariant quantities. The diffusion coefficient D of this demonstration viscous liquid is very small.1) and (242) would then be appropriate. during our demonstration.e. which must be large enough that the viscous fluid itself undergoes negligible diffusion. the most important factor is the viscosity. but not to their position. 7 will be smaller than the observation time and. The above conclusion is that entropy is unrelated to the distribution of positions of the dye molecules.
and they are randomly distributed with zero total magnetic moment. 242 Precession of a magnetic dipole m about the magnetic field h . 243(a). 243(c) represent the directions of the different spins. On the planar diagram two spins and their precession velocities are drawn. Modern techniques enable us to d o many experiments in this rotating frame. so that m precesses to the xdirection: Then immediately switch off the field. The arrows in Fig. The equation of precession is . then m is stationary in this frame.to the xdirection. (b) In a rotating frame with angular velocity Aho apply a magnetic field in the ydirection. The steps of Hahn's experiment are as follows: (a) Use a magnetic field h (along the zdirection) to align the magnetic dipoles of H nuclei in water along the h direction.420 STATISTICAL MECHANICS dipole m will precess. but lasting twice as long. This rapidly changing field flips m from the z . 243(b). This is like turning the whole set of the arrows upside down. Figure 243(d) shows the way that . The nuclear dipole moment of hydrogen in water molecules can be used in this experiment. The angular velocity of precession is h h . the spin directions at different places are different. so that all the spins precess by 180". After a period t. (244) where h is a constant and h is the magnetic field (Fig.= dm dt hhxm . (Fig. in the rotating frame the spins precess slightly differently at various places. Fig. If we use a rotating coordinate frame with angular velocity Ah. 242). (Fig. (d) Use the method of (b) to apply a rapid field in the ydirection.) The magnetic field may be slightly inhomogeneous and the magnetic field at different parts of the body may be slightly different. turning through 90".) (c) Because the magnetic field is slightly different at various places. the faster in front and the slower behind. Let ho be the average value.
ECHO PHENOMENA 421 J / /’ 1 Fig. This is the spin echo. The last step is like the reverse turning of the handle in the viscous liquid bottle experiment. the spins are restored to their original positions and the total magnetic moment reappears. all the relative angles of precession change sign. . but now the slower is in front while the faster is behind . That is to say. 2 4 .3 the two arrows in (c) are turned. Notice that after turning. the precession direction remains the same. and after a period t . The other steps are similar too. which is directly measurable.
i k * v t  . e . though the spin directions are randomly distributed. there is no random motion. any inhomogeneity in the density distribution will smooth out in a short time. Hence. it does not change with time. 24. it will be zero. the motion is prescribed. This phenomenon is more complicated. the momenta of the molecules will not change and thus cannot supply entropy. collisions will not occur if the observation time is shorter than the mean free time. if the distribution is where fo(v) is the uniform equilibrium distribution and a is a small constant. We have repeatedly pointed out that if there are no collisions. This is the socalled Landau damping. Although there are n o collisions because the velocities of the particles are different. Therefore. and we mention here only the principles.(l + &. The density of the particles can be obtained by integrating (246): n ( r . t ) = no + a e i k * rj d 3 v f o ( v ) e . As a consequence entropy does not increase or decrease.i k ' v t changes rapidly with v and if we integrate with respect to v. In an extremely dilute gas.422 STATISTICAL MECHANICS Notice that the dispersion of the spin directions in this experiment comes from the inhomogeneous magnetic field produced by the magnet. then the subsequent distribution is When t is very large.ik.n.r e+k2t2T/m 1 (247) . where the mean free time is very long. For example at t = 0. Although it is inhomogeneous.4. The Plasma Echo There are many other experiments similar to the above. the most interesting being the plasma echo. even though each spin precesses slightly differently.
If k' and k are very close. add an inhomogeneous electric field a ejk'*'. That is to say. creating an inhomogeneous density distribution like (24.ECHO PHENOMENA 423 i. it can be made to appear again.k ' ) ( r . This is a special echo phenomenon. only that the form of appearance is somewhat indirect. the inhomogeneous part will vanish after a time l / k 7 where V is the average speed. . This plasma echo experiment points out that even though the inhomogeneity vanishes. then t in (2410) can be much larger than l/k?. fo(v) aa' exp [ i ( k .the density is uniform again. the distribution becomes (246). even when t is very large ejk still has its effect and cannot be neglected. This effect is measured by the changes in the plasma.e. and then switcn it off. then at a time t = k' ' t k' . There are higher order terms. But this is not the complete solution. confirming that they cannot be neglected. their interference effect brings an inhomogeneous density distribution. even when the inhomogeneous terms of (246) and (248) seem to disappear.i k ' vt']  . the first order term.e. At this time. we add an inhomogeneous electric field 0: eik ' on the plasma.k 9 (2410) the exponential terms in (249) containing v cancel. The inhomogeneity in the distribution vanishes after a time l/kS. Among the second order terms there is one proportional to a d (obtained by replacing fo (v) in (248) (246) by ). ( r . (249) Therefore if k' and k point in the same direction.v ( t . and we get Its integral with respect to v is not zero. Hence the distribution has one more term  '"' f o ( v ) a' e ik'. The steps are as follows: (a) At t = 0. being only the terms proportional to the external electric field. That is to say. i. (b) At time t = t'.t ' ) ) (248) After the electric field has been switched off for l/k'V.When the electric 5). (2411)creates an inhomogeneous charge density which is measurable.v t ) . field is switched off.
In momentum space the configuration is unchanged. so this does not supply entropy. . i. r2. . what do the position. Problems 1. The plasma echo phenomenon is relatively complicated. . but it can still be analysed in configuration space.e. Try this. . On the time scale of this experiment there are no collisions between the particles. The second law is then not violated. . Let (rl . what is the region of motion of R ? (b) After the molecules are dispersed. there are no changes of the momenta of the particles. shape and size of the region of motion become? (c) What about these quantities when the dye is restored to the original position? 2. (a) Initially.424 STATISTICAL MECHANICS Hence the disappearance of inhomogeneous density cannot be regarded as an increase of entropy and its reappearance cannot be explained as a decrease of entropy. Try to analyse the viscous liquid bottle demonstration in configuration space. rN) = R be the positions of the dye molecules.
Chapter 25
ENTROPY CALCULATION FROM THE TRAJECTORY OF MOTION
If we know the details of motion of every molecule during the observation time, then any property of the system can be calculated. Many equilibrium properties, such as pressure, magnetisation, energy, etc., are averaged values during the observation time. Although the commonly seen macroscopic objects are too complicated, simple models with hundreds or thousands of particles or spins can be analysed on the computer (see Chapter 21). Now we want to discuss the following problem: Can entropy be calculated from information on the trajectory? The answer to this question should be yes. As the whole process of motion is known, all physical quantities should be calculable. But entropy is unlike quantities such as energy and pressure. It is not the averaged value over time of a dynamical variable. In mechanics and electromagnetism there is no such concept as entropy. Nevertheless, if the determination of entropy had to go beyond knowledge on the whole motion, then the concept of entropy would be outside the realm of science; that clearly is not the case. The problem is really one of finding a practical computational method. We shall devise a method of calculating the entropy step by step from knowledge about the whole motion. We introduce here a specially simple method called the “method of coincidence”. Better methods must certainly exist, waiting to be discovered. The reader may ask: Why must we discuss this problem of calculating entropy from the motion? We have two motives. One is a matter of principle. Starting from Chapter 5 we have discussed over and over again the relation of entropy with motion. Many examples were given to show that the definition of entropy is based on motion. But all these examples are very simple. Which quantities are invariant and which quantities are changing gradually are very clear. When
426
STATISTICAL MECHANICS
we use the basic assumption, we only need to treat the changing quantities of motion as the variables and those unchanged quantities as constants. But this distinction may not always be so clear, especially for amorphous states. Until the problem is solved, we do not know which are variables and which are constants. With these problems remaining unsolved, the region of motion must be determined by the trajectory. The second motive is practical. A method of calculating entropy from motion can be used to analyse various models, especially those exhibiting metastable states. The Boltzmann formula for entropy is
S = lnr(E)
,
where S is the entropy, E is the total energy of the system and r ( E ) is the volume of the region of motion 52, which includes all configurations with energy E. The discussion of this chapter starts from this formula. Our basic viewpoint is this: r ( E ) must be determined by the trajectory. The method of coincidence is simply a programme to calculate r ( E ) from the trajectory. We first explain the principles of the coincidence method and then use the king model as an example to illustrate the steps of the calculation. The reader should note that the method and techniques of this chapter are in the stage of development and the conclusions are only preliminary. The reason for including this relatively new method in this book is that the method is very important but not yet commonly known.a
25.1. Number of Coincidences and the Size of the Region
The basic principle for calculating entropy from the number of coincidences is very simple, as explained in Fig. 251. A collection of dots is scattered randomly in a certain region. The region 52 has the shape of a ring. The area r of this region 52 is n(a2  b 2 ) squares. In this area I?, only a small number of the squares are occupied by the dots and the majority are empty. If we do not look at the picture and only know the number of dots in all the squares, this area is not so easy to obtain. But if we know that these dots are randomly distributed in 52, then the area I' of the region 52 can be estimated as follows. Let n be the total number of dots, and r be the number of squares in 52. Because the distribution in 52 is random, it is still possible for two or more dots
aThis chapter comes from research reports, see Ma (1981a, b).
ENTROPY CALCULATION FROM T H E TRAJECTORY OF MOTION
427
t*26*
r
Fig. 25  1 A collection of dots randomly scattered in an annular region.
to occupy the same square, even though F is much larger than n . Take any pair of dots; the probability that these two are within the same square is I/r,which we call the coincidence probability R = I/r. Take Nt pairs of black spots, then the “coincidence number” Nc is N f R . The maximum value of N f is +n(n 1). Hence if the coincidence number is measured, then r can be calculated by
(25  1)
This method of estimating is called themethod of coincidence. Thismethod is very simple. It need not consider the shape of the distribution region or other geometrical properties. This method can be used to calculate the entropy. Each square can be thought of as a configuration, and each dot a configuration in the
428
STATISTICAL MECHANICS
trajectory. To calculate entropy by the coincidence method, we pick Nt configurations from the trajectory; each pair of identical configurations will count as one coincidence. If there are N, coincidences then the coincidence probability can be calculated and the entropy follows from the formula
S = ln(l/R)
.
(252)
This is the basic principle for calculating entropy by the coincidence method. Nevertheless, for practical applications we have to solve some problems and clarify some details first. We first review the important properties of the trajectory and the entropy and then try to set up a calculational algorithm. In the above, we have assumed that the distribution of points in is completely random, i.e. a uniform probability distribution. We now relax this condition to include cases of nonuniform distributions. We divide the configurations of the trajectory into groups and let 1 be the time occupied by each group h , so , that the probability is
PA
=
'J,/ J
9
(253)
where 7 is the total observation time. Now assume that each group h of configurations is distributed randomly in a region L , let ?; be the volume of i l k and R , the coincidence probability in this region. We can now define entropy as the average value of In ( 1/ R )
s = CP,l n ( l / R , )
A
.
(254)
This is a generalisation of (252). The factor R, is the coincidence probability for two configurations in the h group: R, =
P,K,
9
(25  5)
where Pk is the probability of any configuration in the h group, and the probability of choosing one configuration coinciding with the one already chosen is l/I',. Let N t h and NCh be the relative number and total number of coincidence in the group h , then
(256)
ENTROPY CALCULATION FROM THE TRAJECTORY OF MOTION
429
The choice of grouping depends on the actual situation. For example, if there is exchange of energy with the environment, the total energy of a system is not fixed and we can use h to label the different energies. Two points should be noted. (i)The method of coincidence is used to calculate r,. The probability P, can be directly calculated in the numerical simulation from J, and (253), or by other methods. (ii) The grouping cannot be too small. In the limit of a single configuration for each group, (254) becomes
s
=  I p s Inp,
s
.
(25 7)
Here s denotes the various configurations. But to measure p s we must measure the time 3, that the trajectory stays in the configuration s. This time is not measurable unless the total number of configurations is very small so as to make 3, relatively large. Usually it is said that statistical mechanics and information theory are the same. We want to clarify this here. In information theory, (25 7) is a basic formula where s denotes thk symbols and ps its probability of appearance in a message. The quantity S measures how much information is carried by these symbols. Each symbol is like a configuration in statistical mechanics. The appearance of the symbols in a message is like the trajectory. In information theory the number of symbols is very small and each symbol is used many times so that p s can be accurately determined. The situation is quite different in statistical mechanics. The total number of configurations s is very large and the trajectory is short. In a trajectory only a very small fraction of possible configurations actually appear. Hence, each group h must include many s in order to make P, a measurable quantity. Hence from the viewpoint of direct measurement, (257) is a practical definition in information theory but has no direct use in statistical mechanics, and is therefore not appropriate as a definition of entropy. If we use information theory on (257) t o discuss statistical mechanics we would be missing the point.
25.2. Independence of Different Parts of a System
This coincidence method uses a relatively few number of points to determine the volume of a very large region, i.e. n << r. But this method obviously has its limitations. We have said that Nt cannot exceed +n(n  1). Therefore
1 N , S nz/I' 2
.
(25  8)
430 STATISTICAL MECHANICS
The coincidence number N, must be quite large, i.e. N, >> 1 for meaningful. Therefore n cannot be too small:
to be
n 2 h  F
.
(259)
Hence the trajectory must be very long, because the r t o be determined is always very large. For a system consisting of N elements, increases exponentially with N :
r=eS
S a N
,
(25  10) (2511)
.
But the length of the trajectory n is proportional to N and the observation time 3 :

N(3/7)
,
(2512)
i.e. the configurations of the system change n times in time 7 . Here 7 is the time for each element to change once. If N is very large and 3 of reasonable value, (259) will not be easily satisfied. This is indeed a problem, but not a serious one. The volume r increases exponentially with N because the various parts of the body are to a large extent mutually independent (see Chapter 12). The molecular motions far apart are uncorrelated. If A ( x l ) is a variable of motion at x1 and B(x,) that at x2 and Ixl  x 2 1 >> [, then
( A ( x i ) B ( x z ) )  ( A ( X I ) ) ( B ( x z) ) = 0
.
(2513)
Here ( . . . ) denotes the averaged value within the observation time and is the correlation length. If the system is much larger than g, then it is composed of many largely independent parts, each larger than g, with the total entropy being the sum of the entropies of the parts. The total number of parts is proportional to N . Hence S a N and J? increases exponentially with N . So, if g is small enough, this coincidence method can be used to calculate the entropy of the various parts of the system. If a system has two independent parts A and B, then the coincidence probability R is RARE because the probability of coincidence in A and B simultaneously is the product of the separate probabilities. Therefore
=
s + s, ,
.
(25  14)
ENTROPY CALCULATION FROM THE TRAJECTORY O F MOTION
431
The product of independent coincidence probabilities and the additive property of entropy are parallel. If a system is divided mainly into independent parts, then (2515) Here SA is the entropy of part A . If A and B are neighbouring and not completely independent, we need, , S as a correction, which can be obtained from SA , S , and S A + B by
SAB = S A + B +
sA

sB
9
(2516)
where SA is the entropy for the combination of A and B. The independence of the various parts enables us to deduce the properties of the total system from those of the small parts (each larger than 5 ) . Because of this, the usual model of numerical simulation can be used to deduce many properties of the whole system, even though these models employ only hundreds or fewer elements of motion. Although these models are small, they are sufficient when they are larger than 5 . Using larger models merely shows clearly the independence property. Of course, the magnitude of depends on the problem. If 5 is very large (as in critical phenomena) numerical simulation is not very efficient. The coincidence method discussed here is only suitable for small 5 . 25.3. Correlation Time The coincidence method tests various uncorrelated configurations. Although the trajectory is not a sequence of completely random points, after a long time the configurations occurring will be independent and uncorrelated. If A is the variable of motion at an instant and B ( t ) is that after time t , then for t >> 7 ,
,
( A B ( t ) ) ( A ) ( B ) = 0
.
(2517)
Here T is the correlation time. If the observation time is much larger than T we can choose configurations separated by 7 for comparison. Equation (2517) points out the independence of motion over a long time, while (2513) expresses the independence over a long distance. This characteristic is a necessary condition for computation. We just assume that this condition is satisfied and do not look for its justification. The exposition of the basic principles ends here. There are, of course, many problems of principle or of calculational techniques, which we hope to clarify through the actual calculation in the next section.
432
STATISTICAL MECHANICS
25.4. Process of Computation
Now we demonstrate the above method of coincidence by calculating the entropy of the king model, listing the computational steps. The first stage is the usual numerical simulation whose results are used as input for the second stage. Our main discussion is on the second stage. To be precise, we first clear up the steps of the first stage. (For the principle and details of simulation, see Chapter 20.)
A . Ising model of motion
sl,
The configuration of this model is determined by the values of N spins . . . , sN. Each spin can be + 1 or 1, and they change according to the following rules. From the configuration at any time t we can calculate the rate of change of each spin (the probability of change per unit time)
(2518)
where T is the temperature of the system and hi is the force on si: hi = h
+ 1Jii si
i
(25  19)
in which h is the external magnetic field and the last term is the total force exerted by all the other spins on the ith spin. The coefficient Jij, the strength of interaction between i and j , is known and is constant over time. Now the rules of motion are completely defined. The trajectory is simulated by the above probabivty and the random numbers generated by the computer. For details see Sec. 22.3. The use of random numbers is convenient but not necessary. The principle of calculating entropy from the trajectory is suitable for any trajectory, the only condition being independence for long timescale and long distances. This independence property is the characteristic of the equilibrium state.
B. Programme of calculation
This programme is to record a number of configurations in the trajectory and then pick them out pairwise for comparison to observe the number of coincidences. The following describes a simple but nos very efficient programme to illustrate the principle of the coincidence method. This programme does not record all the configurations in the trajectory but it stores the labels of the changing
ENTROPY CALCULATION FROM T H E TRAJECTORY O F MOTION
433
elements as a sequence i , , i , , . . . , in. To compare two configurations we must consider the changes in between. For example to compare the configuration A before i, and B after i , , , we compute the number of elements which appear in the list i l , i , , . . . , ilo an odd number of times. An element is unchanged if it has changed an even number of times. The number of elements changing an odd number of times is called the “difference”. If there is no difference, then A and B are the same configuration and we have a coincidence. We discuss the details below.
C. Energy and the effective volume of a configuration
This model is a constant temperature system with fluctuation in energy. So we have to classify the configurations by their energy to calculate r(E ) and P( E ) , and then use (256) to get the entropy:
S
=
1P ( E ) In[ r ( E ) / P ( E ) ] E
.
(25 20)
The probability of the E group is given by the simulation programme:
where 7 ( E ) is the time that the energy E appears, and 3 is the total time of the trajectory. So we have to record the energy also. To increase the number of coincidences we can relax the rule of coincidence. For example, a pair of configuration with the same energy and a difference not exceeding rn can be regarded as coincident. Here rn i s some small integer such as 0, 1 or 2. If m = 0, then it is same as before. Let V, be the effective volume of the configuration s, i.e. the total number of configurations differing from s by not more than rn. That is to say, if s’ and s differ by not more than m and have the same energy, s’ will be in this volume (if m = 0, then V, = 1). To calculate the coincidence probability we have to make correction with V,. V, must be calculated in the simulation programme and recorded for use. No other quantities need to be stored.
D. Limitation of the programme
How large is the entropy that can be calculated from the programme? How many elements are analysed each time? The answer depends on the problem under investigation. But we can estimate roughly its general limitation. If the trajectory has n configurations in k groups, and with n/k configurations in each
434
STATISTICAL MECHANICS
group having rougly the same energy, then the number of coincidences in each group is approximately
N,(E)
 L(l)
2
k
2
r(E)
V
(25  2 2 )
In this formula v is a typical value of V,. The larger N , is, the more accurate r(E ) will be. If we require N, ( E ) > 25, the limit on the entropy is (25  2 3 )
To increase this limit we can increase v or decrease k , but it all depends on the actual problem. If v is sufficiently large the first term on the right of ( 2 5  2 3 ) is not important and the total number of configurations n appearing determines ( 2 5  2 3 ) . If n l o 6 , S 2 3 . Even if n is considerably larger, S will not be much larger. If n is 1000, S will not be much smaller. If each element has an entropy of 0.5, then this programme can test N 40 elements. So this programme can only be applied to very small systems, or to small independent parts of a system. Of course, this requirement limits the application of this programme, but not in a serious way. The range of application is still quite extensive. For example, it has been used to calculate the entropy of the amorphous magnetic state. b



The above programme can be improved in many places, the most important being the choice of the elements of motion and the grouping so that uncorrelated elements are in different groups. In this way, the method of coincidences is most effective.
25.5. Entropy of the Metastable State
Many metastable systems have time scales spanning an enormous range, , e.g. in H, gas, the spins (1 or 0) of H are usually unchanged by collisions, so their correlation time is very long. The correlation time of the velocities of the molecules, i.e. the mean free time, is on the other hand quite small. So if the observation time lies between these two time scales, equilibrium properties will be well defined. We simply disregard variables with change over the longer time scale, but consider those changing over the shorter one. (See Chapter 8.)
bMa and Payne (1980).
ENTROPY CALCULATION FROM THE TRAJECTORY O F MOTION
435
In many systems there is no sharp demarcation between the long and the short time scales; rather there are a number of time scales, long and short. This troublesome distribution of time scales usually appears in amorphous systems. The configuration space is divided into regions separated by barriers of different heights. Hence, the region in which the trajectory is distributed depends on the length of the observation time and so does the entropy. We can make an estimate of how the entropy changes with this time scale. To penetrate a barrier of height A , we need a time of
r
where t >> T
T~
T~ e A J T
,
(2524)
is a short time scale, i.e. the correlation time without barrier. When ~barriers higher than ,
are impenetrable and those lower than A, are easily overcome. So the value of the entropy changes logarithmically with the observation time. This change is very small, especially at low temperatures. Thus entropy is still unambiguously defined.
25.6. The Third Law of Thermodynamics
In thermodynamics, the entropy difference between two equilibrium states 1 and 2 is (25 26)
where 1 and 2 must be linked by a reversible process and d Q / d T is the heat capacity determined experimentally along this process. Apart from an integration constant, (2526) can be used as an experimental definition of entropy. This constant can be determined by the entropy of the high temperature gaseous state (by the theory of ideal gas) or from the fact that the entropy at T = 0 is zero (the third law of thermodynamics, see Chapter 23). The heat capacity of most models can be calculated by numerical simulation. Using (25 26), we can calculate the entropy. Therefore entropy and the trajectory are connected. But this calculation is different from the above discussion. It requires a series of trajectories for the intermediate equilibrium states and not only those under observation, so it does not fulfill the goal of calculating the
i. If between the high temperature and T = 0 some irreversible processes occur.e. This difference is a measure of irreversibility. the time scale is changed because of the restriction of the trajectory.2 6 ) .436 STATISTICAL MECHANICS entropy from the observed trajectory corresponding to a given equilibrium state. i. motion stops and this region will contract to a point. T (25 . This does not violate the second law. The entropy calculated from the trajectory cannot violate the third law because it directly measures the region of motion in phase space.2 6 ) . This is the so called “frozen entropy”. But the change d S of the entropy defined from the trajectory may be larger.e. it is impossible to transform heat into work by this process. Of course. d S < 0.2 6 ) changes by d S ‘ : (2527) This is the definition of S in ( 2 5 . If the entropy decreases upon cooling.2 6 ) will not be defined. The fact that motion stops is the basic meaning of the third law. the decrease of entropy is larger than d Q / T . Nevertheless these different structures exhibit the same macroscopic properties and are difficult to distinguish.2 8 ) because the process of heating may be irreversible. .2 6 ) as a definition of the entropy. dQ dS>. Otherwise ( 2 5 . the entropy ( 2 5 . the reverse (2528). we can disregard the question of reversibility. then at some temperatures the entropy defined from the trajectory may be different from that defined by the specific heat ( 2 5 . the process between I and 2 must be reversible. When T + 0. glass can have many different structures. violating the third law. Sometimes irreversible phenomena are not too obvious. Notice that if an amount of heat dQ is added. e. Notice that in cooling. This condition is quite stringent. An example of an irreversible process is one in which the trajectory does not return to the original region when the temperature is first raised and then lowered to the original value. In (2526). after heating and recooling. As temperature is lowered. We can determine the constant of integration by the high temperature entropy. but increases correspondingly upon heating.g. This definition is commonly used in discussing glasses. Hence if some irreversible processes occur. and use ( 2 5 . then the entropy obtained by integration may not be zero at T = 0.
.4). But if the size of the squares is chosen suitably so that the entropy is related to the size of the square in the manner of Problem 1.2..) (b) Use this method to test the randomness of a random sequence generated by a computer. within a certain scale. This type of programme requires a time proportional to n In n . It isintuitive that these remainders are distributed uniformly between 0 and 27r. .with 1) . m = 1.1) are 1 Nt = . in (25 .3. .3) and (1 1. If the distribution is completely random. Computer specialists have many methods for comparing numbers. . . Review Problems (1 1.there is the ordering programme t o arrange these numbers in order of their magnitude. For example. Identical numbers then appear in neighbouring positions. in (2529) is really modulo 27r. . ni points in square i. The degree of randomness of a distribution of points on a plane or any continuous space can be estimated by the coincidence method. Discuss this issue. = sin(zM) .4 are for demonstration only and are not efficient. then N t and N . (25 30) Note that 2. = 1. entropy must be defined by drawing squares. (25 29) Note that if there are n points and M squares labelled by i = 1. 2. what should the graph be? (Answer: A straight line with gradient 1. 25. l M N. phase space is continuous. . (c) Use this method to test the sequence.M .1) ) and inversely proportional t o the size v of each square. 25 . = n i ( n i . . So the x. So the value of the entropy is related to the size of the squares. Let S (v) = l n r (v) (a) Plot S(v) against ln(l/v).1 that r is proportional to the number of squares (see Eq.1) 2 i=l 1 . . Note that Newtonian mechanics must at least. 2. i given a sequence of numbers A i .n ( n 2 .ENTROPY CALCULATION FROM THE TRAJECTORY O F MOTION 437 Problems 1. (25 . Without quantum mechanics.n . The details in Sec. Purely from the viewpoint of Newtonian mechanics. 2. . the definition of entropy is unambiguous. X. We see from Fig. determine a sequence of independent trajectories so that a group of random points can be selected. (a) How does one represent the configuration of a system as a number? (b) How can one obtain the coincidence probability by making use of this programme? Discuss various details. 3. in (2529) will not have a uniform distribution.
. .. s N ) is the joint probability of (sz. . . . . Let si. . If s1 is related only to a small number of neighbouring elements s2.s ) and P ( s .sm ) be the joint probability of a small number of m elements. Finally each term of S contains only conditional probabilities. = 11 ‘1 . . > s N ) = p(s1 ISz. . fixed. . . There is another method to calculate entropy from the trajectory. .. . sN ) be the joint probability of these variables. i = 1. . then (b) Let p ( s . . 1PI~P . Apply the steps of (a) and (b) to the onedimensional lsing model.438 STATISTICAL MECHANICS 4. s N ) . . . If these conditional probabilities can be calculated from the trajectory then S can be calculated. . . s N ) p ( s z . The derivation is as follows. . . sm .. . . * . .sN) 9 (2532) where p ( s . . . . Prove that (2535) (c) In the above. but within the correlation distance. s z .sz. and the conditional probability will include more elements further away. (d) Analyse the twodimensional king model.N be the variables of motion and P(s.s . 2. ). each step separates an element. S N (2531) ‘a (a) Let p(s1. . then this method is quite ideal. .l is the entropy of ( s z . . . . related to a small number of neighbouring elements. . . ) is the conditional probability of s1 with ( s 2 . . But we have to include more neighbours. where S N . . Prove that . I sz . . .. . . .Sz) * . . If each conditional probability is the probability that the four neighbours are fixed. Write the entropy as S. .
(a) Discuss the ideal gas case. rn = O ( N ). The original derivation is somewhat different from that given above. Alexandrowicz (1 975) and Mierovitch (1 977). This method has an extensive range of applications and originally it was not used in numerical simulation. with each group being approximately independent. 6 . 5 . . We need only divide the N variables into m groups. But these parts need not be divided according to position. See Kikuchi (1 955). What is the trajectory of the dye molecules? How is the coincidence probability defined? Start from Problem 1 of the previous chapter. The various parts of a system being more o r less independent is a necessary condition for the definition of entropy.) (b) Discuss the vibrations of a crystal. Apply the definition of entropy in this chapter to the example of the viscous liquid bottle. (Hint: group according to momentum.ENTROPY CALCULATION FROM THE TRAJECTORY O F MOTION 439 (e) Discuss the advantages and disadvantages of this method.
Mathematicians make it elegant and abstract and use rigorous proofs to derive some results. Physicists are certainly interested in the origin of the basic assumption. From the experience of the above chapters. though certainly within the realm of physics. This difficulty is linked to that encountered in the basic assumption. The development downwards is less remarkable. whenever the interaction cannot be neglected. most calculations involve approximations that may not be reliable.Chapter 26 THE ORIGIN OF THE BASIC ASSUMPTION The development of statistical mechanics is very strange. Starting from the basic assumption the development is upwards. etc. Although the applications are quite successful. So the quest for the origin of the basic assumption. but are on the whole not enthusiastic in pursuing this problem because it is too difficult. tends to fall into the hands of some great mathematicians. Perhaps they are content with the success of the various applications and thus overlook the importance of understanding the origin. As mentioned in the preface. the problem becomes extremely difficult. How do the regular rules of motion cause random results? This is a very common problem. it is rather like an inverted pyramid. . but applications are not easy either. but also an understanding of irregular phenomena in general. successfully explaining many phenomena. Physicists do not have the habit of avoiding difficult problems. apply it to various models. most mathematicians like elegant mathematics and they sometimes neglect some of the requirements of physics. However. Specialists of physics. chemistry and materials science. Matters of principle are difficult. In fact the understanding of the basic assumption is not only a physics problem.
1) and (262) the entire theory of thermodynamics can be deduced and all the equilibrium properties can be calculated. Without it there would be no thermodynamics. Then we look into and criticise some conventional views. i. Hence (262) is the hypothesis of independence. our views on the bases of statistical mechanics. We then discuss the concepts of ergodicity. These ideas. (262) where N is the particle number or the total number of variables. The Basic Assumption We now review the basic assumption as discussed in the previous chapters. This is the conclusion of Chapter 2 and the content of Chapter 23. the fluctuations .e. Let r be the volume of the region of motion. The instability of the trajectory is also introduced here.e. After Chapter 4 we have come to regard this condition as that establishing the independence of the various variables of motion. 26. we pointed out that the time average of large value ( O ( N ) ) variables is the same as the average value in the region of motion. ensembles and mixing.1) i. We classify the variables of motion into two types: (1) the invariants. The basic assumption says that entropy is the logarithm ow: S=lnr . This assumption must be supplemented by the following condition S = O(N) . the total number of configurations of the region. but is the subjective opinion of the author. we shall summarise our views on the basic assumption. and (2) the rapidly changing variables. The discussion in this chapter cannot be said to be an objective view of statistical mechanics. r = O(eN) . scattered in previous chapters. i. the set of configurations satisfying the conditions of invariance. i. Then we define the region of motion as a set including all the configurations of the changing variable. The strict definition of the region of motion depends on the detailed trajectory of motion. We have emphasised repeatedly that invariance means unchanging quantity during the observation time. are summerised here. In discussing the central limit theorem (Chapter 12.e.e. and the end of Chapter 6).1. variables unchanged during the observation time. What variables are invariant depends on the time scale and in some cases it is not obvious how to classify the variables. (26.e.THE ORIGIN OF THE BASIC ASSUMPTION 441 In this chapter. Secondly. Notice that condition (262) is most important. From (26. i.
I .ik . Ergodicity and Ensembles The above basic assumption is not too harsh a condition because we discuss only large variables and large regions. Hence we see that (26. over various parts of the system. To understand statistical mechanics. R must be a large subset. the above basic assumption is sufficient to analyse all experiments of thermodynamics and scattering. When Boltzmann wrote down this assumption. This type of variable also includes variables like It is the sum. This conclusion can be written as (263) where 3 is the total observation time and R is a subset whose volume r ( R ) is comparable with : r ( R ) = O(eN) . 12.442 STATISTICAL MECHANICS of these variables are of O ( n ) and normally distributed. 26. (264) 7 ( R ) is the time that the trajectory stays in R . i. Otherwise (263) is meaningless. This argument is . e.e.) P ( R ) can be interpreted as the probability of the configurations being in R .) The fluctuation of a large variable is directly related to the differentials of the thermodynamical potentials. it is ergodic. we must first understand (262). The term “large variable” here denotes the sum of the variables describing various parts of the system. To sum up. Its average value may be zero.2. (See Chapter 12. (See Sec. two most important concepts in the development of statistical mechanics. (262) is the ‘ origin of this assumption. the total magnetic moment is the sum of the various magnetic moments. These also follow from (262).9. his thinking went like this: the trajectory will pass through every configuration in the region of motion. Notice that ’J is a finite time and cannot be regarded as a large number of O ( e N ) .1) is a daring assumption and it links entropy with the microscopic motion through I. The reader may notice that this basic assumption may appear in different forms in the literature.g. We now mention here ergodicity and ensembles. and hence the infinite time average is equal to the average in the region. From the deductive viewpoint. of the product of the density with .
This abstract way of thinking has many advantages. Others may say that whenever the basic assumption holds the motion is ergodic. However these results are not too helpful to statistical mechanics. it need not be truly ergodic. i. The term “ergodicity” may have different definitions in different papers. These points distributed in phase space can be thought of as an ideal gas (each point is a “molecule”). Each configuration of the system is a point in phase space.e.^ forms the main trend of modern statistical mechanics. a system with infinitely many similar structures. The ensemble theory has become the traditional basic concept.e. but is usually neglected nowadays. We need at least e N molecules to define the distribution.THE ORIGIN OF THE BASIC ASSUMPTION 443 wrong. i. The ergodicity of the trajectory is a major problem in mechanima Many mathematicians have devoted much effort in proving the ergodicity of some models. This is a mess. The ergodic theory of Boltzmann met with immediate criticism and doubt was cast on his theory including the Htheorem. because this infinite long time must be much longer than O ( e N ) . The theory of ensemble says that the equilibrium states of the system form an ensemble. In some places the definition is as above. The reader should be clear about the definition in any paper before delving into it. but we require that the time average of large variables equal the average over the region. Chap. The reader must have noticed that the definition of terms is not uniform in the literature. The ideas of Gibbs produce many useful formulas (equivalent to the calculational rules of Chapter 7) and applications. This condition is not too harsh because the values of large variables are nearly the same everywhere in the region of motion and the trajectory need not pass through every configuration. The theory of Boltzmann. not infinitely many. 6 of Yang (1978) and other books on Ergodic Theory.while the usual observation time is O(1). Ensemble is a mathematical concept and it cannot  aSee Arnold (1968). Nevertheless the ensemble is an abstract artifact. This was pointed out by Gibbs in his book. In reality there is one system. bGibbs (1960). reformulated by Gibbs in terms of ensemble^". We do not require this in this book. . Boltzmann perhaps died of the resulting frustration. In other places it means that the trajectory “nearly” passes through every point. because we have a rather strong intuition about the flow and distribution of a gas. To require the time average of every variable to equal the average in the region is a harsh condition. Properties of the equilibrium states are average properties of this ensemble.
=  I plnpdr . The content of the ensemble theory is identical to the assumption of Boltzmann. From this we can prove that the entropy defined by Gibbs in ( 2 6 . i. The ensemble is like a cloud in phase space and its density can be represented by a distribution function p . many scholars have tried to axiomatise this theory of ensembles or at least give it a more definite meaning. In the next section we shall compare and discuss this view with that in our book. After a long time.5 ) cannot be used to discuss phenomena of entropy increase. p is nonzero only in R (0). Many people tried to explain and patch up this unreasonable conclusion of constant entropy. It distributes the ensemble in the region of motion. the configurations of the system have a probability distribution.444 STATISTICAL MECHANICS solve the physical problem of Boltzmann’s assumption. The motion of each system in the ensemble obeys Newton’s law. Suppose R (0) is a region like a lump of dough.e.5 ) is over the whole of phase space. We have criticised this view in the above chapters (especially Chapter 24). before the observation. only the shape changes.c or “the degree of uncertainty”. Now phase space is a 6 N dimentional space o f N particles all of which obey Newton’s laws. that is to say. The integral in ( 2 6 . then p is nonzero ‘See Tolman (1962). This degree of uncertainty is regarded as an a priori probability.) The reader has probably come across Liouville’s theorem in mechanics. Obviously ( 2 6 . The ensemble average is the average over the region of motion. (265) where p is normalised s p d r = 1. tracing out a trajectory. Here we first mention some basic concepts of the ensemble theory. Gibbs discussed the change of p and defined entropy as S. If the ergodic property proposed by Boltzmann is said to be unrealistic then ensemble theory is perhaps more so. The trajectories of different systems are nonoverlapping. p is unchanged along the trajectory. After a time t they form another region R ( t ) . According to Liouville’s theorem the volumes of R ( 0 ) and R ( t ) are the same. Let us fix a region R ( 0 ) in phase space and let every point in this region move according to the laws of mechanics. If at t = 0. R ( t ) will be stretched out like intertwining noodles made from the dough. After Gibbs. . that would mean different trajectories can come from the same initial conditon.5 ) is an invariant. (If there is intersection. Today the most commonly held idea is to define the ensemble as the knowledge of a system by the experimenter.
THE ORIGIN OF THE BASIC ASSUMPTION 445 only in the noodles at time t . The expression in (265) is the volume of the dough or noodles. . then where R’ is any subset of the region of motion. The ensemble theory attempts to use the instantaneous properties of infinitely many systems to represent equilibrium.1 we pointed out that the independence property (262) is the root of the basic assumption.3. and it can be said to have descended from Gibbs’ theory. The view of this book is: the basic assumption is an assumption of independence. the volume of R ( t ) in R’ is proportional to the volume R’. That is to say. is the total volume of the region of motion.e. 26. This is an unreasonable approach. 26. and the independence property is a necessary property for defining entropy from the trajectory. Mixing and Independence Many scholars in the study of modern statistical mechanics devote themselves to the analysis of mixing. We will discuss it carefully in Sec. If t + m.d The theory of mixing is commonly accepted now. The analysis of mixing is a complicated mathematical problem and the author is no expert in this field. It requires that any region R ( 0 ) be transformed to a patch of fine threads spreading uniformly over the region of motion. The property of mixing is more reasonable than that of ergodicity. There are some recent books on this aspect. The calculation of entropy from the trajectory discussed in the last chapter shows that the region of motion must be defined and understood from the trajectory. Now we rewrite the view adopted in this book in a form similar to that of mixing and then make a general discussion. an assumption on the correlation time and the correlation length. and r ( R ’ ) is the volume of R’.5. This independence property dSee Sinai (1979) and Krylov (1979). The equilibrium state is not an instantaneous concept and entropy needs a time scale to be defined. but it does not represent the randomness of these noodles or threads. We do not discuss the correction to (265) because it is built upon an incorrect theory. i. 26. In Sec. Mixing implies that the noodles or threads R ( t )will distribute uniformly in the region of motion. This point is emphasised from the beginning of this book. It is more advanced than the theory of Boltzmann or Gibbs.
(262) is established. 26l(a) becomes the black threaded region in Fig. if we have correlation time r and correlation distance t and 1 t l . We can say that this model has the mixing property. Let us look at the property of mixing in these two examples. . This uniform distribution occurs when t >> r . the finer and denser the threaded region becomes. O < x < L . r = time interval to traverse length of box .446 STATISTICAL MECHANICS is the independence beyond long time scales and long length scales. we have not supplied the mathematical steps to obtain (262) from (267). What is the difference between mixing and independence? We look at a simple example here. then the region of motion can be determined. and the whole basic assumption is built up. (268) where A p = b . That is to say. Consider two onedimensional models: (A) a particle restricted in a rigid box between 0 < x < L and (B) a particle attached on a spring and executing simple harmonic motion with frequency w . or may require other conditions. The black disc region in Fig. how the independence property follows from mechanics. and “local” means related to a small number of variables of motion. A small change of 3 will not affect the average value. i.e. but the black disc is stretched to threads. The longer the time. The main points here is to understand (267). These steps are probably not easy. Obviously these black threads are distributed uniformly in the region a< IpI< b . “near” means distance smaller than t .rz I >> t . 26l(b) after a long time. The total area of the black region is unchanged. If we can start from the principles of mechanics to derive (267). In the above. then (267) The average value is the time average over the observation time 3 . The variablesA and B are local variables of motion near rl and rz .u is the diameter of the black disc.tz I >> 7 or Irl .
situations such as Fig. 26. No matter how long the time is. Both are periodic motion without chaos. .I(d). this model has no mixing. from the point of independence these two models have no great difference. The black disc in Fig.) Therefore. So.THE ORIGIN OF THE BASIC ASSUMPTION 447 P b a (A)  L P / / . 26l(b) cannot occur. 26l(c) goes in an elliptical path with slightly changing shape but never stretches into threads.e.1 Model (B) is completely different. i. these two models are completely different. neither shows the independence property. But. \ / \ I \ I I (B) I I I I X \ I I X I \ \ \ \ I I I I I I Fig. from the point of mixing. (See Fig. 26 .
The period of motion of model (A) is determined by 1 p I to be a constant.448 STATISTICAL MECHANICS The viewpoint of this book is not to consider the initial state as a group of points. In one point of view. In practise N Therefore N . The discussions of equilibrium and the increase of entropy necessarily involve the evolution of probability distributions in order to see how the probability evolves from an initial distribution (or a priori probability) and gradually disperse to fill up the region of motion. the probability distribution is more concentrated. since experiments give us a sense of objectivity. but at least these examples point out certain basic differences of viewpoints. the case of quantum mechanics would be hopeless. different persons observing the same system will have the same knowledge. 26. The above examples are obviously too simple. Correlation time refers to the time of dispersion for this probability distribution. probability expresses the knowledge of the observer. No experiments have been performed to such accuracy. This is unacceptable. The question is: What is the experiment? How many experiments must be performed? The answers of “macroscopic experiments” or “feasible experiments” are incorrect. The other large variables are also determined to O( l / f i ) . According to this view. Probability and Experiments In current literature probability usually denotes something like the Gibbs’ ensemble.4. In the author’s opinion. the position of each impurity atom is known. But this proposition is more or less the same as the previous one. If we cannot elucidate the situation in classical mechanics. The concepts of statistical mechanics are derived from classical mechanics. In addition if the probability   . the canonical ensemble represents a fixed temperature for which the experimenter should be able to measure energy to an accuracy of O ( 1 / f i ) . If he knows more about the system. In the discussion of fixed impurities and the analysis of echo phenomena in Chapter 24.” 10l’. Let us forget about quantum mechanics for the moment. We d o not discuss whether (266) and (267) are different in more complicated models. the meaning of this probability is very unclear. Is this a feasible experiment? The experiments discussed in the literature on statistical mechanics are usually unrelated to actual experiments. For example. The motion of a system is independent of the psychological condition of the observer. This seems to be a more reasonable view. This is obviously incorrect. Another way is to say that this probability represents the outcome of experiments. We consider only one initial configuration.
From the viewpoint of statistical mechanics. Probability is looked upon as a tool for arranging the information and not a physical concept. Yet the systems encountered in the laboratory are usually metastable. any metastable states such as diamond have a probability tending to zero. The total energy function H is fixed. Modern computers can calculate the trajectory of some simple models and can calculate the various thermodynamical properties from the trajectory. We must first have numbers and information before we can do statistics. But if we cannot even understand these models. we have a strong intuition about space and the flow and dispersion of a cloud in space. We cannot avoid assumptions of mathematical . But the above criticism is not meant to denigrate the achievements attained in the literature.THE ORIGIN OF THE BASIC ASSUMPTION 449 tends to zero. the understanding of these simple models does not form the basis of statistical mechanics. All the properties must come from the trajectoiy. then the event is explained to be not happening o r not observed in experiments. The author thinks that Gibbs’ ensemble and probability have no relation with mechanics. e.) This habit is a hindrance to the analysis of the logical roots of statistical mechanics. treating every configuration with the energy E as having the same probability. It is not a necessity but a convenience. make ensembles. These concepts are helpful for our abstract analysis. they are both unnecessary and unrealistic. This manner of analysis is a habit acquired from analysing the gaseous state. As mentioned above. The concept of ensemble gives us a new tool and quite an elegant one. (See Chapter 22 and Chapter 25.) So the problem of experiments can be separated from that of the logical roots.g. calculate the probability and discuss equilibrium or nonequilibrium. but they are unrelated to any discussion on the logical roots of statistical mechanics. we have to get rid of this concept of ensemble probabiltiy. classify the states. We have performed a critical analysis of the concepts of ensemble and probability as presented in the literature. the future does not look bright. These computer calculations are purely objective. The definition of probability given in this book is quite different. Of course. Our aim is to clarify the most basic concept. This too is untenable. One main weakness in the literature is to regard the region of motion or the equilibrium probability distribution as something known. If we want a correct analysis of these roots. and then the equilibrium probability can be simply determinded. (This book has repeatedly emphasised that the determination of the region of motion is not so simple and it depends on the actual distribution of the trajectories. The numerical solution requires Newton’s laws but not the concepts of ensemble or probability.
and 68 changes for every collision. but we have to clarify our concepts and make effective use of it.10) There is also instability.5. The meaning of instability is as follows. and after n collisions (2612) . Let s ( t ) be the trajectory. as in Fig. If 6 s ( t ) a efft . In model (A) of Sec. (269) with a > 0. Instability of the Trajectory To understand mixing or independence from the laws of mechanics is a very difficult task. Now let us look at a more complicated example. We now discuss instability and its details. 26. then 6 x ( t ) = t 6P m . The configuration of this particle is determined by its position and velocity. Starting from the origin. conforming to the requirement of mixing. then the change of the trajectory is 6 s ( t ) . We consider only a model with planar motion. 262. If there are many different 6s(O). The magnitude of the velocity is unchanged.11) ( 2 h / a ) 16 0 I. the trajectory is said to be unstable. if the direction makes a small change 6 8 ( 0 ) the trajectory will change. 26. Figure 262 shows the situation after two collisions. (26. If the initial state s 0 is () changed by a bit to s(0) + 6s(O).3. Assume that the density of the hard spheres is very low and the mean free distance X of the particle is much larger than the radius a of the spheres. so we only consider the direction of the velocity. Equations (266) and (267) seem to be rather remote from the details of the trajectory. forming a small cloud surrounding s(O). then this cloud will disperse. if initially momentum is increased by 6 p . (26. Suppose a particle collides with a number of fixed hard spheres.450 STATISTICAL MECHANICS concepts. As seen from the figure 68’ = 2 6 a = Therefore I 6 8’ I 2 ~XM/UCOS(Y . but not as serious as (269).
i. i. t > T l n ( l / I66(O)l) h _  In I 1 / 6 0 ( 0 ) In2h/a I v (2613) then ( 6 6 ( t )1 1. Let ri. ..2 Collision of a particle with fixed hard spheres.e.N be the positions of the spheres. . Suppose we have N small hard spheres of radius a and the density of the molecules is very low. . This result is the same as (269).. 2 6 . i = 1 . We mention it briefly here. i .rz. . 3 . 2 . the direction is randomised. 2 .e. ..THE ORIGIN OF THE BASIC ASSUMPTION 45 1 Fig. . . where v is the speed of the particle. If the time is sufficiently long.1 ) cylinders: lrirjl <a . Let r  = (rl.rN) 9 (2614) where r lies in a 3N dimensional space. .N . The above analysis can be extended to the gas model with similar results. In this space there are $ N ( N . (2615) . j = l . .
Now alter the direction of v(0) slightly to v(0) + 6v(O). 6v12 * 6 V 7 8 = 0 (2619) This is similar to the analysis in Sec.9. 262 as the cylinder in the 3Ndimensional space. 6V78 = (0. For example. then only 6 v 7 and 6v8 are nonzero. which goes in a straight line and is reflected whenever it hits a cylinder. 6 V s . 6 V 7 . So. O ) (2618) Therefore. we obtain a trajectory. If we follow r(t). where X is the mean free distance. so 6v12 = (SVl. This situation is similar t o that of the above example. 0.0.. after each collision I6vl will increase by a factor hla. there are changes of v1 and v2. So the motion of these molecules is represented by r ( t ) .452 STATISTICAL MECHANICS and r is excluded from the cylinders because they are hard spheres. (2620) 70 E hlv . ..0. after n collisions 6v( t ) becomes the sum of n vectors. 6v2. But collisions with different molecules will change the direction of 6v. . the analysis of the angle is similar to the above example.. O ) (2617) If molecules 7 and 8 collide. . . . 0. and see how 6 v ( t ) changes. It is the same as the above example except that 6v is more complicated than the above 6 8 . The infinitesimal velocity 6v has 3 N components. . 0. These n vectors are mutually perpendicular and t nN 70 . in a collision of molecules 1 and 2. The changes of 6v in collisions with different molecules are mutually perpendicular. . In the centreofmass coordinate system of two molecules. 5. We have to regard the black disc of Fig. . Notice that d r / d t = v is the velocity and (2616) is unchanged because v2 is proportional to the total energy. One good feature is that the collisions are binary. Approximately. . .
the final outcome will be completely different. The correlation ~ ~ time can be estimated as T~ / l n ( h / a ) similar to (2612). When 6 8 ( t ) becomes large. (26. 193. However. this way of analysing enabies us to have a deeper understanding of the characteristics of the trajectory. The above analysis shows how the volume expands. To use it to discuss mixing we must first assume that there is an initial probability distribution. In fact the above analysis does not say how 6 8 ( t )changes when it becomes very large. .21) Notice that t / is the number of collisions of each molecule. then the trajectory is not easily controlled and extremely sensitive. we are ignorant and naturally there is no reason to draw any conclusion. The above is only a crude analysise The above analysis explains approximately the meaning of instability.(t)  (y“ 6Vi(0) . p z . it will be “unrelated” to 6 8 (0) and hence we can disregard 6 8 ( 0 ) . there will be no problem in mixing. (2623) . see Krylov (1979). If the initial state differs a bit. .THE ORIGIN OF THE BASIC ASSUMPTION 453 where v is the average velocity of the moleculqs. Equation (26. As the vectors are mutually perpendicular. If the trajectory is unstable. Notice that (2623) is a large number of magnitude e N . In this kind of analysis it is very easy to allow the assumptions of mixing and independence to creep in unnoticed. .12) is appropriate only when 6 8 ( t ) is very small. The magnitude of each vector is about 6v.13) can only be regarded as an estimate. Notice that instability is a result of mechanics and does not involve the concept of probability. HNtlTo = exp [ 3 N ( t / 7 0 ) In ( h / a )] . in (2613) we assume that after 16 8 ( t ) I equals 1.p. . eFor details. they form a volume (2622) The increase of this volume is exponential: The space of v is momentum space p = ( p . assuming that 6 8 ( t ) will be unrelated to the situation at t = 0 or independent when 6 8 ( t ) grow large. Equation (26. . If the trajectory has the property of instability. For example. p N ) as p = m v . This analysis cannot be used as a proof of mixing or independence.12) or (26. It is detailed but not quite rigorous.
independent in time) motion. The cause of chaos seems to relate to the degree of complexity. When N is very small. The General Problem of Independence To summarise: From the point of view of physics. Large values of N 2 10 may produce random motion different from that of small N . Although to derive (267) from the principles of mechanics seems extremely difficult. . However. That is t o say. if we can understand random sequences we can understand more about the basic assumption. the research of mechanics in the direction of random trajectory has made great advances. the values of N investigated is still very small and is not sufficient to discuss the independence of the parts of a system. our understanding of the basic assumption is still very primitive. 26.454 STATISTICAL MECHANICS Part of the contents of this section came from inspiration while reading Krylov’s book. up to now. but of O(1) and not a large number of O ( e N ) . to understand statistical mechanics we must understand (a) the existence of a correlation time T = O(1). These works show that random motion is a result of Newton’s laws. the correlation time must not only be finite. The correlation distance 5 must also be of O(1) and cannot be a large number of order N . Indeed. These problems are waiting for an answer.e. The random sequence is generated by a computer programme and should be easier to understand. The property of independence is so common that it should not follow only from Newton’s laws.e. For example. the study of random motion is in the field of numerical simulation and has not reached a profound understanding of the problem. i. These two points are the contents of the independence condition (267). all these random phenomena are not understood. The discussion by Krylov is very penetrating but the views of the author do not quite agree with those of Krylov. this property of independence is actually not unexpected from our daily experience. The author thinks that the research should be directed to the influence of the value of N . especially when N is around 10. the existence of the correlation length [ . (b) the existence of a correlation distance 5 = O(1). From the molecular motion to the throwing of a die. This has been discussed in Chapter 3. to the random sequence and the stock market. no more than 3 or 4 variables) exhibit random (i. That is to say. lndependence is “randomness” or “chaos”. In recent years. Why a random sequence is random at all is still unknown.6. a system can essentially be regarded as composed of independent parts. but simple models can still give rise to chaos. Many simple models ( N very small.
8.1. i. As we said in Chapter 22. Correlation when mixed with independence makes the problem difficult to solve. we have to look for the relation between them and put the correlated elements into a bunch. assuming a’ to be small. If we cannot transform a model to one similar to the ideal gas. (841) as the interaction.e. In the various examples in this book.term solution.THE ORIGIN OF THE BASIC ASSUMPTION 455 26. Difficulties Encountered in Applications From the examples of the above sections we see that the application of statistical mechanics is an extremely difficult mathematical problem. But independence simplifies the application of the basic assumption. In the crystal model.3. (c) Find the correlation time. . etc.5 to the model of crystal vibration. no matter how complicated the solution was (e. Problems 1 . this is perhaps not a good long.g. At present the usual models are not tractable and have to be solved approximately or numerically. If there are N variables. and is the origin of correlation. normal mode) is an independent part. The mutual interaction sets up the independence property. the numerical method is very important. each elementary vibrating element (i. These special cases are usually overlooked. But from the point of understanding the origin of statistical mechanics. the exact solution of the twodimensional Ising model). each molecule is an independent part. (a) Use a onedimensional model. making the independence property obvious. This conclusion has some deficiencies.e. That is to say. (b) Add the a’ term and discuss its effect. At present. 26.5. 2. First neglect the a’ term and find the equation of motion. 23. we had to separate out the independence property and the correlation property of the model. This problem shows that in some special cases the correlation time is very long. 26. The author thinks that the difficulty of solving problems is related to our lack of understanding of the independence property. Apply the method of Sec. Use Eq. and then to solve it. From mechanics we cannot arrive at the independence property. reliance on numerical solution is gradually increasing. Try to find them from Prob.7. see Sec. As pointed out in Sec. In an ideal gas. we have to separate the variables of motion into mutually independent parts and then solve each part separately. the correlation time of a gas is approximately equal to the mean free time of the molecules. as in the discussion of this chapter our lack of understanding of the origin of the basic assumption. we cannot solve it.
5. This kind of definition has n o relation with the discussion here. (b) Review Secs. then the numbers following can be calculated. I n fact diffusion does not determine the arrow of time. We can test whether a sequence is random or not.2. 12. From the law of mechanics we cannot determine whether time is flowing forwards o r backwards. Some problems of a rather philosophical nature appear in the literature of statistical mechanics. 4. the arrow of time. But in diffusion we know the arrow of time. Discuss this problem in the viscous liquid bottle demonstration in Chapter 24 (the dispersion of the dye molecules by turning the handle in its subsequent restoration). between past and future. According to experience the more random the number. e.456 STATISTICAL MECHANICS 3 . the more difficult is the prediction.5 t o 12. (a) Review Sec. after a time they will disperse. Of course if we can discover the algorithm used by t h e computer. 11. if we calculate the correlation values from it. The problem is not the direction of time flow but the dispersion of the distribution with 1 t I.5 about the central limit theorem. If we have n numbers in a random sequence. What is the degree of independence necessary so that entropy has an unambiguous meaning? This question is not easy to answer. Therefore. It at t = 0 some solute molecules gather at a point in the solvent. Chaos and Prediction Chaos and independence are properties that can be measured. and the fact that the requirement of experimental observation is not strict independence but only some sort of approximate independence. the time direction has a symmetry.g. This is the problem of decoding. What is the relation between the degree of randomness and the degree of difficulty of prediction? Try to discuss this. how d o we calculate the next n numbers? This is a prediction problem. This section shows that it is not easy t o determine whether a group of variables is independent. (c) Discuss the definition of entropy and the requirement of independence. . But suppose we ask: according t o mechanics what is the distribution before t = O ? The answer is that the molecules are also dispersed. Notice that one definition of randomness is the impossibility of prediction.
i. leading to large fluctuations of the order parameter. One of the results is the appearance of “soft modes”. Displacement changes the order parameter and this is a continuous change. as it ignores the fluctuations of the order parameter and sometimes leads to erroneous conclusions. explaining it on the basis of condensation. states of motion with very low energy. The Ising model in Chapter 17 is one of the main models used to discuss condensation. The main point of this part is to introduce the order parameter. then different phases coexist. Chapter 30 discusses superfluidity. Chapter 29 discusses cases with continuous symmetry. Chapter 27 discusses the mean field approximation. The coexistence of different phases can be regarded as the order parameter taking on different values. if we displace all the molecules the thermodynamical potential is not changed. such as water changing to ice. the order parameter vanishes. Unfortunately the approximation is rather crude. and discuss its properties.e. the metastability of the superfluid state and the role played by the “defects” (vortices). expressing quite lucidly the common characteristics of various condensation phenomena. or paramagnetism changing to ferromagnetism. The coexistence of different phases discussed in Chapter 15 is the coexistence of different condensed states. Its advantages are simplicity and clarity. If one part of the system has a value of the order parameter different from that of the other part. The main points of this chapter are the appearance of the order parameter.PART VII CONDENSATION This part is divided into 4 chapters on topics related to the phenomena of condensation. (The order parameter here is the magnet moment. Chapter 28 considers the boundary between coexisting phases and analyses its fluctuation. for which the order parameter can have continuous values. etc. . For example a magnet may have half of its spins pointing up and the other half pointing down. For example. given the regular arrangement of molecules in a crystal.) If the temperature is high enough. Condensation is freezing.
However. The mean field theory is the simplest approximation and quite easy to understand. We use the mean field approximation to analyse the properties of ferromagnetism and the coexistence of different phases. The mean field method assumes every element to be independent and that every element moves in a fixed environment which is determined by the average value of these independent motions. The main model discussed here is the Ising model. and emphasises those points which are easily misunderstood. thus explaining the phenomenon of the coexistence of liquid and gas. The common features of various condensation phenomena are clearly exhibited in the mean field approximation. so approximations become very useful. and we also discuss it in detail. its simplicity gives it an important place in statistical mechanics.1. twodimensional or threedimensional Ising model are not easy to solve. To illustrate this method. The various conclusions can be directly applied to the Ising gas model.Chapter 27 MEAN FIELD SOLUTIONS The mean field method is extremely important. Let . 27. This chapter gives the method of calculation. this is the only possible tool for analysis. and even occasionally completely wrong. The vector model can also be solved by the mean field method. In solving many relatively complicated problems. Mean Field Theory In Chapter 17 we have discussed the solution of the onedimensiofial Ising ferromagnetic model. the Ising model is the best place to start. Although crude. Van der Waals theory is also a mean field approximation.
Jvm . (273) As the average value of every si is the same. hi is changing too. (272) Because every element is changing continually. 27.4) i.eh'/T m z (s) = eh'/T + eh'IT = tanh( h + Jvm ) . the first term on the righthandside of (27. (275) See Fig. The factor v in (273) is the total number of neighbours. i.1 for the ( h . = h + 1Jii sj i . (27 . j are nearest neighbours = 0 otherwise . so its average value is easily found to be eh'/T . (276) .1) The environment of each element si includes the external magnetic field h and the magnetic field due to the nearest elements. The mean field method replaces si by its average value.e.4) is larger than the second term. getting a mean field h' = h + Jv(s) . If T is sufficiently large.e. h = Ttanhlm . we have dropped the index i. m) curves. (27. Now every si is situated in a constant magnetic field h'. The total magnetic field is h.MEAN FIELD SOLUTIONS 459 the total energy be Jij = J if i.
But if the temperature is too low. m ) curves for three temperature values.e. then x is infinite when h = 0 and the curve touches the xaxis at the origin (curve 2). When T < T. indicating paramagnetism. the curve m o ( T )is (279) What is the meaning of these solutions? . 2 7 . and h = 0 there are two nonzero solutions of m (see Fig. / T ) together with an m = 0 solution. and when h is not too large.1 ( h . When mn ( T ) is near 0 or T. . T<T. 272) i. (278) m~ = tanh(mo T .460 STATISTICAL MECHANICS Fig. (277) one portion of the curve violates (276). = Jv . m can assume three values (curve 3). m = 'mo(T) .e. and m vanishes at h = 0 (curve 1). If T = T.. i.
2. Then maximise the entropy and under fixed E = ( H ) . (27. Total Magnetic Moment The mean field solution can also be derived by maximising the entropy. use H ’ to calculate the entropy and ( H ) .! = 2 5 [ ( m + 1 ) l n + ( 1 . calculate the minimum of the thermodynamical potential F = E . The entropy of this model is very easy to obtain: S = In N! N+! N.e. 2 ( E ) ) . T . From the minimum of the thermodynamical potential we can explain the situation when T < T.MEAN FIELD SOLUTIONS 461 Fig. regarding h’ as a variational parameter. so as to determine the best choice of H ’ (see Sec. 7 . Now we use H’(s) =  1 Sib' i= 1 N . i.m ) l n 2 l+m ~ 1 m F. Now we return and examine the thermodynamical potential under this approximation. Use a simple energy function H ’ ( s ) to replace H(s).10) to calculate the entropy.N m h 2 h’ ( s i > = tanh= m .T S 1 ( H ) = NvJm2 . 272 Average value of spin as a function of temperature. ( 2 7 . This method works as follows. .11) .(m) = ( H ) .TS at constant temperature. 27.
e. 272. its minimum value is however the thermodynamic potential under the present approximation. The solution of (275) is the horizontal point of the F o ( m ) curve. Figure 273 shows F o ( m ) for T > T. Now we adjust m to minimise the thermodynamical potential. 273 . ( m ) is not the thermodynamical potential yet. and solve (27. T>Tc T < Tc Fig. the curve of Fig.12) The result is (275).. so we may think of m as the parameter. and T < T. F. i.462 STATISTICAL MECHANICS Notice that m and h’ are monotonically related.
In other words. The other solutions are those outside A B . for which m and h are in the same direction.12). The point C i n Fig. 273 is a minimum.1 with unsuitable parts deleted. 214 Notice that Fo ( m )in Fig. but not an absolute minimum. ih Fig. 271. (275) has three solutions. 274. 27. (278)). but also fvring every (sir to m. We now show the result for the total magnetic moment M = N m in Fig. i.e. and that every si is independent.e. this is the curve between A B in Fig. but when h = 0 and T < T.MEAN FIELD SOLUTIONS 463 When T < T. This can be regarded as a characteristic of ferromagnetism. there are two minima. 273. there are two minima. when there is no magnetic field and m is nonzero the direction is not fixed. Now. which is just Fig. There is no question about the solution when h # 0. But there is a solution with maximum Fo . 273. If h = 0. . This is a metastable state and we shall discuss it at the end in the problems. not only fvring the total magnetic moment N m . i. If h # 0 then there is one absolute minimum. then these steps must be modified. the best approximation to the true thermodynamical potential is determined by (27. only if h # 0 does Fo(m)have an absolute minimum. and h is not too large. & mo(T) (see Eq. as can be seen from Fig. and m and h are opposite in direction. Eq. the steps from (2710) to (27112) are: assuming that every ( si ) is the same. If ( si ) is not everywhere the same. Equation (2710) is a very stringent condition. is not the thermodynamical potential for fixed total magnetic moment.
mo can coexist in equilibrium. F o ( m )is also a function of h and T (which we now display explicitly). This is a separate problem. and T < T. (2715) . T ) . 273 is not yet the thermodynamical potential. and it is much more complicated than the Ising model. = .e. one having m = mo . then m(0. m(0. m(0. (27. When h # 0 there is no problem for (2713). T ) can take values between f mo and F is unchanged.. O<a<l . Let the first part occupy a fraction a of the total volume. T ) = F o ( m( h . Our present conclusion is that when T < T. mo ( T ) and there are two solutions. and (2713) still has no problems. while the other m = mo . 274. So the energy of the boundary is only a small correction term. Let Fd(m. provided the boundary surface is not too large. h . m = mo and m = . T ) .14) i. In a real ferromagnet.464 STATISTICAL MECHANICS 27. T ) E F o ( m . Thermodynamical Potential and the Coexistence of Different Phases We have just emphasised that Fo ( m ) in Fig. When h = 0 . (2713) Besides being a function of m. T ) is still the same. or may be fixed by impurities. obtained by taking the minimum of F o ( m ) . i. and h = 0 . This can be regarded as the phenomenon of coexistence of phases. If the body is divided into two parts. where the spins in the same domain point to the same direction.(la)rno = (2a1)mo .e.. T ) < mo Of course if we divide the body into three or more parts. The function m = m ( h . then F ( h . The energy of the boundary is proportional to its area while F is proportional to its volume.T) = amo . but different domains point differently. . These regions are very much like the domains in ferromagnetism.1 v J m 2 . T ) is the curve in Fig. T ) = ?r. h .T S ( m ) N 2 . the thermodynamical potential is still essentially unchanged. mo < m(O.3. But Fo ( m ) does not depend on the sign of m . What then is the thermodynamical potential? The answer is very simple: the thermodynamical potential is F ( h . we have long range forces also.T ) + hNm . The boundary between different domains can move.
273. T ) = Fb(m. . F ’ ( ~ . F includes the potential energy due to the external force h . but the relation between F.S d T . with the bottom curve cut off by a horizontal line F’. 273 are quite different. F’ and Fa is not. m ( h . (2716) This thermodynamical potential can be regarded as a function of the total magnetic moment and temperature.13) plus h M be F + hM = F’(m(h. (2717) + hdM F ’ is the internal energy minus TS.m o . (See Chapter 16. T ) . aF‘ = aM h . We have not included the term hM.(WZ. The above discussion only cutsoff the convex portion of the upper right curve of Fig. Because different phases coexist and F is unchanged between ma and . Notice that the differentials of F and F’ are dF =SdTMdh dF’= . T). (2719) If h = 0. 273. i. If T > T. Figure 275 is just the upper right curve of Fig. then F’ = F must be the minimum. i.MEAN FIELD SOLUTIONS 465 F is a function of m and T only.) According to the above analysis. then F r ( m . 275 and Fig. = if I m l > mo . while F is the internal energy minus h M and TS. . Let the d thermodynamical potential (27 . T ) F.e. In equilibrium.e. T ) .T) .T). The reader must note that the meanings of Fig. if T < T then . The result is simple.. Hence m is the equilibrium value of M/N. so F’ = F must be horizontal and not convex.
(2720) i N = C ~. Because each ni has two values.466 STATISTICAL MECHANICS I I I t F’ I I I I Wrn I I Fig.PN = 1 E C ninj . 275 27. Let ai = 2ni . then (27 20) can be transformed to the Ising ferromagnetic model: (27 2 1) J e =4 ’ (27 22) .1 . with ni = 0 or 1.j 2 . The king Lattice Gas This model is defined by the total energy H : H . ~ i The variable ni is the number of particles at the site i.4. this model is an king model.p C n i i. The first term on the right of (2720) is limited to neighbouring i and j .
see (2716). the desired relationship is. (2726) n = ( n i >= (1 + m ) . (2726) then is our answer.T. The former number of elements N i s now represented by V . First consider the case T > T. (27 28) . Although F' has been solved. the pressure. however. and F' by (271 1). Simplifying. 275. we get p = T In() 1 1n . where p is the pressure. The definition of the thermodynamical potential C2 is (2723) and the definition of the thermodynamical potential F of the ferromagnetic model is . We want to know the relation between the temperature T. therefore (notice that the N in the ferromagnetic model is V here) F' + 1 2nh . The field h can be replaced by (27 S).FIT = C.MEAN FIELD SOLUTIONS 467 Notice that now we use V as the total number of lattice points and N as the number of particles.24) Therefore from (27 22) we know that (2725) Because C2 = . and the density n .p V . 2 . (2718). (2717). 2 (2727) where F is defined from (2716).2n2T. . (2719) and Fig.H/T S (27 . we can translate the results of the above sections into properties of the lattice gas. . With these relations between this model and the ferromagnetic model. T. = J V = 1€ v 4 1 . not yet very explicit.
the ideal gas law.(l 2 +rni(T)) + 1 Tln 2 4 lrn%(T) ’ (2 7 . and replace FA ( m o .e. T ) plane of h = 0 and T < T. T. 1 = . and m = 0. +. This curve is just the representation on the ( p .) is the endpoint of the equilibrium vapour pressure curve (see Fig. At high densities. then we get this curve. = T. ( l n 2 . i. This is because every site can accommodate only one particle. p tends to infinity. . T = T. the first and second derivatives of p are all 0. n + 1.$ ) = 0.T ) in (2718) by F’. The critical pressure is p .e. n << 1. . i.468 STATISTICAL MECHANICS At low densities. (2731) and ( p . T ) curve for an king lattice gas.. This is the “critical point”. 276 ( p .e.T. corresponding to h = 0.32) I T Fig. Let h in (2726) be zero. Equation (2728) can be rewritten as (2729) Hence (2730) It is easy to see that at T = T and n = . . i. 276).193 T . p = n T.
277.MEAN FIELD SOLUTIONS 469 When T < T. we can still use (2728). .e. and h # 0. The latent heat q is the heat required for one molecule to change from the liquid (I) to the gas (g) state. i. If we draw p as a function of 1/n = V / N . ) / V . but the density n = N / V is changed from+(1 + mo) to +(1 mo). The horizontal part is the pressure for the coexistence of the liquid and gas phases. In this model S ( m o ) / V is the same as S ( . the saturated vapour pressure. pc  l/n Fig.dp 4  dT TAv ’ (2734) This formula appears in Chapter 15.e. 27 7 Plot of pressure versus inverse density. then we get Fig. Therefore = TS(mo) 2 (1 . i.33) This result can be obtained by differentiating (27 32).m . .m o  2 1 (27 .
the total energy is 3 H = TNT + pV . bN2 V V . Let S be the entropy of the ideal gas. The interaction between gas molecules is repulsive at short distances and attractive at large distances.. (2735) This formula includes the potential energy p V due to pressure and the interaction energy b N Z / V due to the attraction between the molecules.(V) 1 T . is approximately the volume of one molecule. The Van der Waals Equation In the last section we had applied the mean field approximation to the gasliquid transition. NT P = .2. Now we maximise S at fixed H . i. and adjust the parameters in the model to maximise the entropy. The aim of this section is to provide further practice and introduces the law of corresponding states.e. 27.T or the minimum of 3 C o ( V ) = NT 2 . because of the repulsion between the molecules.bN2/V . with fixed total energy.e.Gc. i. this ideal gas model has a smaller volume. The repulsion reduces the volume available for the movement of the molecules. and v. Under a fixed temperature and pressure. but with a slightly different appearance. The van der Waals equation has the same content.e.N v o . (2736) + pV  V (2737) bN2 Now V is an adjustable parameter just like m in Sec. Now we replace this gas by an ideal gas model.470 STATISTICAL MECHANICS 27. we want to calculate the maximum of H S(VNuo) .5. In (2736) the volume for calculating the entropy is V .Nuo (27 39) . The maximum of (2736) or the minimum of (2737) can be derived from (2738) i.
the curve in Fig.p ( Y ) ] (27 . 278(a) the low temperature curve has three points with a G / a V = 0. 279(b). Obviously Go at A is the true thermodynamical potential. or from (2740) the area below the straight line A B . In this situation dV av = VA dV [ p .e. (See Fig. At low temperatures. the shaded portions in Fig. for the coexistence of the two phases. But in Fig. Notice that (2739) is obtained as the minimum of G at fixed p .MEAN FIELD SOLUTIONS 471 This is the van der Waals equation. This is not reasonable. i. So . 278. 279(b) must have equal areas. 218 If Go is the same at A and B . The last term in (2740) is the area under the curve A B . Therefore. volume increases with pressure. 278(b). We have to discard points B and C . These three points are indicated in Fig.e. then there are two solutions. Fig. a p / a V can be negative. i.40) where p ( V ) is the expression on the right of (2739).) We encounter no difficulty at high temperatures.
472 STATISTICAL MECHANICS L V Fig. The characteristic of the critical point is that the first and second derivatives of the p = p ( V ) curve are zero. (2743) = P/Pc . 277. The dotted line is the coexistence curve.e. to find the coexistence pressure and volume we need only to draw a straight line on the p = p ( V ) curve so that the two abovementioned areas are equal. 27 9 The Maxwell construction illustrated. (2742) Now we use these values as units of volume. pc = b/27vo . temperature and pressure andlet v' P' = V/V. . = 8 b / 2 7 ~ o . We get Vc = 3 v o N . (27 . the critical volume and temperature. Each temperature is given by one curve. . . T ' = TIT. We do not need to calculate the thermodynamical potential.4 1) From the van der Waals equation (2739) and (2741) we can solve for V. aP . This is the Maxwell construction of discarding the unreasonable portion of the curve. i. T. and the result is more or less like Fig.0 av 3 a2p = o a v2 . The highest point of the coexistence curve is the critical point. Above this point there will be no coexistence of two states.. and T .
Notice that the mean field approximation cannot accurately solve for T.6. Order parameter The concept of order parameter can now be naturally established. The values Tc . (2744) can be applied to all gases. there is no distinction between liquid and gas. Equation (2744) agrees surprisingly well with experiment. they have the same content. The 10% accuracy mentioned above refers only to (2744). We have to measure the values at the critical point and all the other properties can be derived from (2744). . there can be different equilibrium states. 173.1 (27 44) V'* This equation does not involve any parameters.MEAN FIELD SOLUTIONS 473 The van der Waals equation becomes 8 T' . Therefore. coexistence of different phases is not possible and there is a unique equilibrium state. Critical point Above the critical point. Many gases can be described by (2744)to better than 10%accuracy. This quantity is zero above the critical temperature. Below the critical temperature. In the ferromagnet. Coexistence of different phases Under certain fixed conditions. Equation (2744) can be called the law of corresponding states. the different states mean magnetic moments pointing to different directions. which we summarize as follows. 27. We can observe this in Fig. if the van der Waals equation is correct. Common Properties of Condensation Phenomena In the above we have used the mean field method to discuss the models of liquidgas phases and ferromagnetism. C. . all being the shape of Fig. Hence the many different phenomena thus discussed have many common points. In the liquidgas model. A . and not taken from the solutions of the mean field approximation. 141 . must be experimentally determined. B. In the liquidgas model the different states imply different densities.. In the model of a ferromagnet. The reason for the success is the similarity of the interaction between gas molecules. the magnetic moment disappears. pc and V. Although these models have formal differences.3 3v' .
B and C.) The force associated with the order parameter can be called the condensation force. Antiferromagnetism in solids shows a periodic arrangement of the direction of the spins. But quantum mechanics allows a linear combination of the empty and occupied states. When m is very small. The above common characteristics can be summarised by the simple mean field theory. Table 271 lists some common examples.e. If the reader thinks in terms of spin . then the “up” corresponds to the occupied state while “down” corresponds to the empty state.) + N(amz + bm4 . and b are essentially constant near T. (See Chapter 2. i. so spin represents the number population. the direction of any spin is opposite to that of its six neighbours. These different directions correspond to the amplitude of the particle. the various curve: in Fig. Notice that the discussion above is really limited to scalar order parameters only since the magnetic moment in the ferromagnet model is limited to two directions. it is associated with a generalised force. The state of a particle can be empty or occupied. and are the condensation forces of the above examples. In the ferromagnet. . The order parameter is a coordinate. But the spin can also point in other directions. (2745) where a’. there are other variables of motion. besides the particle population. and m is the order parameter. Vector order parameters have continually varying directions. This is the topic of Chapter 29.h m ) . it can be defined as the density minus the critical density. F o ( m ) = constant a = a’(TT. The state of a pair of electrons can be occupied or empty or it can be a combination of the two. Besides the ferromagnet and the coexistence of liquid and gas.. The order parameters for the superfluidity of 4He and for superconductivity are not obvious. The situation is similar in superconductivity. In the ferromagnet the order parameter is a vector while in the liquidgas model it is a scalar. For example in a cubic lattice. The magnetic field and the pressure (or the chemical potential) are the forces associated with the magnetic moment and the density. the order parameter can be defined as the average magnetic moment. This point will be discussed fully in Chapter 30. Hence. Like other coordinates. Near the critical temperature. there are many condensation phenomena possessing the characteristics of A .474 STATISTICAL MECHANICS it is nonzero and can have different values for different equilibrium states. In the liquidgas model. the amplitude of the particle. 273 can be represented by the following simple expressions.
More detailed experiments and theory show that when T is very close to T. The derivative a m p h is proportional to I T . When T < T. Let h = 0 . . 6 . 1 . . only the values of a’ and b are different.1 Condensation Phenomenaa Phenomena liquidgas ferromagnetism antiferromagnetism superfluid He superconductivity separation of mixture metallic alloy Order Parameter density magnetic moment periodic magnetic moment amplitude of particle amplitude of paired electrons difference of the two densities of the liquids periodic arrangement Example H2 0 Fe FeF2 He Pb CC14 . Its content is the same as the mean field theory. .l . . m a (T.g.8 Cu Zn 739 aTaken from Ma (1976). The heat capacity is discontinuous at T = T.17 7. Equation (2745) is the Landau theory.T ) ” .MEAN FIELD SOLUTIONS 475 Table 27. But its purpose is to discuss all the condensation phenomena within the same framework.T. the mean field approximation is not correct.T I P . (2746) . The mean field theory does not explain all the common points of the condensation phenomena. know that m = 0 when T > T..26 2. m is proportional we to ( T . These characteristics have been mentioned in the above sections. p .. The reader can rederive it from(2745).05 1044 78. from the condition for the minimum of Fo (m).CTF14 Critical Temp.19 30 1. K 647. Equation (2745) can be applied to any condensation phenomenon. e. This is a most important step.
5 0) . (2747) mi = ( s i ) .2. show that C(mjmi) i 1 : a* v2 m ( r ) . F o ( r n ) has an absolute minimum and also a local minimum. m is a vector) has continuous symmetry. (2748) where a is the distance between nearest neighbours and m is regarded as a continuous function of position. (c) Boundary Surface. Therefore h N Ttanh' m . Although this is a very interesting topic. Try to interpret the local minimum in Fig. x > > a . and h is not too large. the exponents are different also.J v m i Jx(mjmi) i . 273. If we suddenly reverse the direction of h . and the exponents in the two cases are different. 27.Ja2 V* m . The values of 0 and y are related to the symmetry of the order parameter. . x << a . and y is not 1. The absolute minimum is the equilibrium point.476 STATISTICAL MECHANICS The exponent p is not . Problems 1.m while the vector model (i. . The properties near the critical point are called critical phenomena. 2. The Ising model has a symmetry of + m and . (2749) The external magnetic field can also be a function of r . m + mo . (b) If mi is a slowly varying function of r i . Assume m = O m + m o . If the dimensionality of the space is different. x = o . If impurities are added. when T < T. The summation in the equation is limited to neighbouring j and i. what happens then? Review the section Growth of Water Droplets in Chapter 15 and Problem 5 of Chapter 17.e.1 and 27. (a) Derive (2747) using the methods in Secs. (2 7 . it is too specialised and we shall not discuss it any more. the results may change. then the generalisation of (275) is h = Ttanh' m i . 273 as ametastable state. As seen from Fig.J v m . If we allow the mean field to be different at different points. Let h = 0 and T < T.
is there any change to the above answer? Discuss the meaning of b and discuss the problem of the fluctuation of the boundary. except 1 that now si is the spin 1 operator.3).) (f) Thickness of the Boundary. z directions. 4.5 2) This result is approximate for situations when >> a . If we change this condition to m ( b ) = 0. (Detailed analysis will appear in the next chapter. v’) curves and the coexistence curve. Also assume that rn is unchanged in the y .T)/J]’ . where b is any constant. Prove that if x >> a .MEAN FIELD SOLUTIONS 477 where + m o are the two minimum points of F o ( m ) . (d) Surface Tension. Calculate and sketch m ( x 1. 5 . Prove that the thermodynamical potential per unit area due to the boundary is u is the surface of the boundary. 3. j are neighbouring sites on the lattice. Based on (27441. (see Secs. carefully draw the various ( p ’ . (2 7 . . Calculate the magnetic susceptibility and the heat capacity. Use the results of (c) to get U . In (2750) we imposed the condition m ( 0 ) = 0. (e) The Displacement of the Boundary. Apply the mean field approximation to the vector model (27 . 27. because the assumptions of (2748) and (2749) require m to vary very little over a distance a .5 3) where sj is a unit vector and i. Apply the mean field approximation to the above vector model.2 and 27. [ EZ a[(T.
5.2 can essentially be explained by this equation. and v. Notice that As is logarithmic. . and v i are respectively the volume of each molecule of the gas the liquid. Use (2749) to solve for rn(r). pc and vc are the temperature.478 STATISTICAL MECHANICS 6. The rate of scattering is particularly large near the critical point because it is proportional to the correlation length (see Sec. then they have the same ratio of critical values ~ J / ( r / u where the form of $ is fixed but ). 27. pressure and volume per molecule at the critical point respectively.4). (b) Prove that if this class of substances has a critical point. 9. this class of substances have the same equation of state. and is not sensitive to changes in v .. Assume T > T. 11. divided by the critical volume v. v i and As can be calculated.. 8. As the law of corresponding states can be applied extensively.. m is very small and h a 6 ( r ) . Prove that G ( r ) is the response function. Problem 15. 7.54) where v . Response and Correlation. derive the evaporation entropy for each molecule (27 . (2754) is also very useful. See Chapter 13 for background material. Starting from the analysis of Sec.. Therefore. once v i is measured. Try to analyze this phenomenon. The correlation function C ( r ) is obtained from G = C/T and 5 is the correlation length. Suppose a class of substances have the same form of interaction $ ( r ) = E and u vary from substance to substance. show that in suitable units for the thermodynamical variables. where T. This illustrates that the law of corresponding states is universal because the interaction between gas molecules has essentially the same shape. (a) From the method of fixed temperature.
) . and J > 0. Obtain the response function and correlation function for T < T. Find the heat capacity and magnetic susceptibility. and this simple analysis is not too reliable. Let the total energy be H =J i.. (a) Let h = 0 and solve the onedimensional model. (2756) This is like the ferromagnetic model with J replaced by . The last two problems show that the mean field method enables us to analyze fluctuation to a limited extent. The indices i and j still label neighbouring sites. the fluctuation is large and so is the correlation length.h z s i i . (b) Prove: if h = 0. it is independent of the sign of J . again starting from (2749).i 2 sisi .J . (c) Use the mean field approximation to solve (2756) for h # 0. 1 1. (d) What is the definition of the order parameter in this model? What is the condensation force? (It is not h . the thermodynamical potential of (2756) is like the ferromagnetic case. that is.MEAN FIELD SOLUTIONS 479 10. Antiferromagnetic king Model. Near the critical point.
Case (3) denotes temperatures around the critical point. we have not considered fluctuations. so is the correlation length. However. and the analysis becomes difficult. According to the mean field calculation. Case (2) denotes temperatures much lower than the critical point. and the interaction may be regarded as a relatively unimportant correction. and the correlation length is very short. The mean field theory is unable to account for this. the critical point. but fluctuations can still have very important effects. . there will be no condensation.Chapter 28 FLUCTUATION OF THE BOUNDARY SURFACE The mean field theory in the last chapter provided a general picture of condensation. Case (1) denotes situations with temperature much higher than the critical point. we do not discuss it here. condensation has been accomplished. The characteristics of fluctuations are intimately related to condensation. These situations can be easily analysed. The properties of the substance are not related to condensation. the fluctuation amplitude is rather large. such as the similarity of the phase coexistence curve. In Chapter 11 we had calculated the vibration amplitude of aSee Ma (1976). etc. If the fluctuation is too large. (2) low temperatures and (3) critical point. The analysis of fluctuations can be divided into three cases: (1) high temperatures. This situation is more complicated. This is a topic of considerable current interest. Stanley (1971). Fluctuation has a very intimate relation with the dimensionality of space and the symmetry of the order parameter. The reader may consult the referencesa Because it is too specialised. It tells us some common properties of these phenomena. Fluctuations have a strong influence on condensation. Near condensation.
We regard the boundary line as a continuous curve. causing the average magnetic moment to vanish. We do not use the Ising model yet. The reason for this is very simple. in discussing the coexistence of the positive and negative states.FLUCTUATION O F THE BOUNDARY SURFACE 481 atoms in a crystal and discovered that the amplitude was very large in one. but discuss a more simplified model. and hence the mean field approximation is not appropriate for this onedimensional model. Because the boundary points can move without change in energy. A negative region inside a positive region can increase by enlarging the boundary. This situation does not occur in the two. In this chapter we shall discuss the boundary between the coexisting phases at low temperatures. These types of problems will be discussed in the next chapter. The boundary points divide the body into positive and negative regions. The mean field calculation had not considered the fluctuations due to the boundary points. This is the content of the Peierls' proof in Chapter 17. The aim of the last section is to point out that similarity exists between the seemingly unrelated models. 28. We then use the Ising model and the Coulomb gas model to analyse the surface of a crystal. we have to consider also the boundary. We first use a simple model of surface tension to look at the fluctuation of the boundary between liquid and gas.or twodimensional space. At low temperatures the boundary is not easy to increase. Assume that the energy of the boundary line is proportional to its length. The Liquid Surface Model In Chapter 17 we solved the onedimensional Ising model.' 1 2 d x ( dz ) z 2 . The boundary points become boundary lines or boundary surfaces. the size of each region changes continually and randomly. Figure 28.or threedimensional Ising model. Now we look at the fluctuation of the boundary line in two dimensions. H = a L . The positive region lies below the line while the negative above. The conclusion is that there is no ferromagnetism.1) . (28. but rather small in three dimensions.1 shows a boundary line z(x). ignoring the discreteness of the lattice. So.1. L = j:' dx [ 1 + (z)] 2 % =Lo + 1. This is also a problem ignored by the mean field theory.
i.1) we get (283) From the law of equipartition of energy we obtain a k 2 1qkI2 = T . where a is the “surface tension”. The above assumes that the gradient is very small. 28 .482 STATISTICAL MECHANICS I ? . (284) .e. If this is the boundary between liquid and gas phases.1 A 2 dimensional boundary line z (x). then this model is rather accurate (if there is no gravity). Let (282) Substituting into (28. d z l d x << 1. Fig.
2 ( ~ ( x z(x’)) ) 1 L =I2 [ I . H = a A A = . pushing the boundary line up or down does not change the energy. The displacement k = 0 means that all the z(x) are the same. Let the energy of the boundary surface be proportional to its area. long wavelength oscillation requires small energy. The thickness of this boundary (i.e. Oscillations of small k.e.x ’ I L o .z(x’) . s d 2 r [ 1 + (Vz)’]’ (287) .   6. that is. that is. This result comes from the ) factor k 2 in (283).cos k ( x x’) ] = lxx’l a 2T (286) If 1x . Such a displacement expends no energy. then ( ( A z ) ~ ( T / a ) L o . Let us generalise this simple model to the boundary surface in three dimensions.FLUCTUATION O F THE BOUNDARY SURFACE 483 Let Az be the difference in height between the two points x and x’.cos k(xx’) Lo k a k ] J  2T dk m a 2nk2 [ 1 . a uniform displacement. i. AZ E Z(X) . can be regarded as the “thickness” of the boundary line.5) ( ( A z ) ~ = 2 (z2 ) ) . the square root of ((Az)’ ) is thus proportional to This boundary line is therefore “rough” and not “smooth”. the magnitude of Az ). and let us calculate ( ( A z ) ~ When x and x‘ are very far apart. (28 . This long wavelength motion makes the height of the boundary surface fluctuate substantially.
Lo in (289) should be changed to [ . ((Az)’ ) increases too.1) and (Y is the surface tension. this model is very accurate. When the liquid surface rises by z . The gravitational force is weak but is responsible for keeping the liquid surface horizontal. Hence it produces an energy 1 T p g z 2 (per unit area) and this should be added to (287).484 STATISTICAL MECHANICS This is the generalisation of (28. the thickness of the boundary surface is (289) where L i is the area of the (x. which then becomes 1 t 2 = alpg * (28 .cos k ( r .10) Hence.r ’ I >> a and Jo is the Bessel function. y ) crosssection. It is not correct to regard (289) as infinitely large. Equation (288) is correct only when I r . except that we have not considered the effect of gravity. The generalisation of (286) is ((Az) ) = 2 (?) 5 (zz)‘k2 [ 1 . No matter how low the temperature. 7T(Y U (28 . (289) cannot be very large.r ’ ) ]  T Irrr I . The fluctuation of the height can be expressed by ( ( A z ) ’ ). The above result (288) shows that when I r . For the boundary between liquidgas phases.8) kD We have limited the integral within kD and a is about the interatomic distance.In .r I increases. it produces a pressure p g z . the boundary is rough. where p is the density of the liquid minus that of the gas. This model treats matter as a continuous medium and regards the boundary as an elastic line or surface. . Notice that even if L o is very large ( e g 1 km). Therefore. Now we estimate the effect due to gravity..
then nearly all the elements will be 1.1. the elements above the boundary line are nearly all . If we take the si on the boundary as . If we consider the lattice effect. If T is not very close to T. and those at the bottom as t 1 . Let us calculate the thermodynamical potential of the boundary line.FLUCTUATION OF THE BOUNDARY SURFACE 485 The above is a discussion of the boundary between liquid and gas. 282). then the boundary surface at low temperatures will not be rough. If the discreteness of the lattice is considered.. The Boundary Surface of the Ising Model From the inequality (1727). Now we take the elements at the top of the square lattice as 1. The total length is (28 .2. and those below the line nearly all + 1.11) Fig. The boundary line in two dimensions can be directly analysed from the Ising model (1716). then the structure of the crystal must be taken into account. 282 (a) Boundary line for a square lattice. It is quite simple and can be analysed in considerable detail. The boundary surface in three dimensions is very complicated. The energy of the boundary line is its total length times 2 J . above. We first study the boundary line in the next section.1 . The lattice arrangement of the king model is the simplest description of the lattice. then we certainly get a boundary line in between (Fig. (b) An enlargement of a portion of (a) . we may understand better the surface of a crystal. 28. The results should be similar to those. If we have to analyse the boundary of the surface of a crystal. when T < J the average value of si is very close to 1 .
T In Z “ . (28 . T 5 J . The next question is the amplitude of fluctuations of the boundary line. etc. Therefore the thermodynamical potential F” of the boundary line is F” = . Equation (2811) is appropriate when T is very small. y 3 = 0. the boundary h e is of length Lo because we have ignored the fluctuations.e. y z E 1.t h horizontal position. .486 STATISTICAL MECHANICS Here.. i. In the k . y k is the length of the k . the 6 function can be ignored (with an error of O ( l n L o ) in F ” ) . .t h vertical boundary line counting from the left. Neglecting the 6 function. Z ” can easily be calculated: (2813) 6 = F“ = LO 25 + T htaA($) . the height of the boundary line i s (2815) . In Fig.12) The 6 function in the equation fixes the right end of the boundary line to be in the middle of the right edge. y4 = 2. Since Lo is a large number. y s = 1. but not much. 282(b) y 1 = 2. (2814) From the view point of a large length scale. The surface tension decreases with T. and 6 is the surface tension at temperature T.
each y . will be independent variables. the conclusion that the boundary line is rough i s not altered even if we include these effects. The amplitude of fluctuation in the middle is about 6 sinh ( J / T ) (2818) Although is much smaller than Lo . Therefore.FLUCTUATION O F THE BOUNDARY SURFACE 487 Hence. Using the central limit 1 theorem we get the distribution of Az : P(AZ) = 1 (A2)2/2~z  2n e2J/T ( 1 eZJ/T)’ = n/[2sinhZ( J / T ) ] . not a smooth and straight line. 283. the difference in height between two points on the boundary line is (2816) If n << L o . The above analysis did not consider the “overhanging” or isolated regions in Fig. it is still not a microscopic length scale. this boundary line is a wiggly and rough curve. (2817) Hence this boundary line fluctuates violently. Nevertheless. 283 . 6 Fig.
Cabrera and Frank (1951).. When the temperature exceeds a certain value. 284 A crosssection of a 3dimensional king model. and those below all + . (287) has no smooth boundary surface because it has no separated layers. This analysis also points out that the model of Sec.1 (see Eq. there will still be some fluctuation. But this boundary surface is smooth apart from this small fluctuation. fluctuation is not limited to the middle layer. Of course. bBurton. Therefore. the elements in the middle layer will condense into + or . and their effects tend to cancel each other. If T >> T. Crystal Surface .. 284 represents the boundary surface.488 STATISTICAL MECHANICS 28. the surface will be roughened. so that the change in energy of the elements in the middle layer is determined by that layer itself only.. Figure 284 is a crosssection of a threedimensional Ising model. Fig. because the upper neighbours are .values. this temperature should be approximately equal to the critical temperature of the twodimensional Ising ferromagnet. the boundary surface should be smooth. Of course. 28.b The argument goes as follows. If the temperature is lower than T. The elements above the middle layer are all .3. the layers next to the boundary line also have the above property of not being influenced by their upper or lower neighbours.the Coulomb Gas Model It was pointed out some thirty years ago that at low temperatues the surface of a crystal should be smooth. The elements in the middle layer can be positive or negative. the critical temperature of the twodimensional king model. . However. the elements in the middle layer form a twodimensional Ising model. The existence of separated layers is a necessary condition for a smooth boundary surface. the above analysis is essentially very reasonable. . Therefore. If we use the Ising model for analysis. The thick line in Fig.and the lower ones are + .
FLUCTUATION OF THE BOUNDARY SURFACE 489 The Ising model is too complicated for this analysis. The distance between each layer is 1 .1 with a Coulomb gas model (see Sec. if we add to H a term then 4(r) can be regarded as the charge distribution because z(r) is the electric potential. Weeks (1980). We now add some effects of the separated layers to the model described by Eq. The following analysis replaces the model of Sec.e. (287) has an intimate relationship with the electrostatic interaction. H a Jd2rE2 . see Chui and Weeks (1976).5). Now the question is how to expose the difference between continuous z and discrete z. This is not the ‘For the following method. .T l n Z . then Vz is the electric field E and (287) is just the energy of the electric field.z ( R + a ) I 2 R a . The thermodynamical potential is F = . If we regard z as the electric potential. Although the following steps are a bit more complicated. Now rewrite (28 7) as ‘ H = a L i +‘y 11 [ z ( R ) . (2822) and z ( R ) are now integers. i. (2819) and the minimum of the electrostatic energy is determined by V * E = 0. e. the introduction of the concept of “charges” is not unnatural. Equations (2821) and (2822) only put (287) on the discrete lattice. How do charges come in? The reason is that the (VZ)~ in Eq. 287. Any interaction involving z can be regarded as the interaction with charges. 28.g. (2821) where R denotes horizontal lattice sites and a the horizontal vectors. 19. which will then allow a crude analysis of the roughening transition. V2z = 0.
If we replace the summation in (2822) by integration over z. (See Eqs. as will become obvious later. (1649). . . see Chapter 16. We get where 2. The discreteness of the lattice in the horizontal direction has no great influence.) The value of z extends from . n =n R R jdz(R) . { d z ( R ) A eH’T 20 ’  . 2. (2823) m For detailed derivation. ) is the average evaluated with respect to this continuousvalued model. = (A) . Then This is the property of the normal distribution. This formula can be written as m dz .490 STATISTICAL MECHANICS Ising model but a new one. (2825) If z ( R ) is a continuous variable.m to 00. . then its distribution is nonndl because (2821) is quadratic in z . then this model would be no different from (287). (See Eq. Every summation in (2822) can be evaluated by the above formula.iznqz f(z) . and ( . is the value of 2 obtained by considering z ( R ) as continuous. The difference between summation and integration can be computed using the Poisson sum rule in Chapter 16. ( 1 232)?(1238) and .
FLUCTUATION O F THE BOUNDARY SURFACE 491 the discussion in Chapter 12. Here q ( R ) is the charge at R . Now ( z ( R ) z ( R ’ ) ) . 19. Therefore (2829) is the interaction energy between these two charges (divided by A X ’ ). Notice that because of the symmetry of + q and . then we are back at the model of Eq. except that the charge can be any integer. If 1 R .5. the total charge is zero. then (z(R) Z ( R ’ ) ) ~ .R‘ 1 is very large. Hence the increase in the thermodynamical potential can be computed from (2824).R ’ I 2 1101 a + constant . Now A and A’ can be regarded as charges added at R and R’.) The calculation of ( z ( R ) Z ( R ’ ) ) ~ .R ’ ) = In IRR’I 9 a T ’ SZ a 2nT (28 28) This model is just the twodimensional Coulomb gas model in Sec. (287). we get U ( R . and the temperature T’ is inversely proportional to T and proportional to the surface tension a. If all the 4 ( R ) are zero.  In I R . can be obtained by the following steps: Replace q ( R ) and q ( R ‘ ) by 4 ( R ) + A and 4 ( R ’ ) + A‘. where A and A’ are infinitesimal. To calculate ( ( A z ) ’ ) . (2827) Substituting into (2826) and (2824). Hence the influence of the separated layers can be ascribed to the influence of these charges. Hence the correlation value of z ( R ) and z ( R ‘ ) becomes the interaction energy between the two charges.is similar t o (288). .q . in addition to (288) we also have to consider the influence of these charges.
Krumbhaar and Binder (1979). but we omit it here.e. The above analysis utilises the Poisson summation formula to write the influence of the separated layers as the influence of a group of charges. the boundary surface is smooth and the correlation distance of the fluctuation is b . At low temperatures.30) The value $. This electrostatic model shall reappear in the next two chapters. We then use the model of electrostatic interaction to obtain the solution. T > 2 a / n . when T < 2a/rr.e. dSee Chui and Weeks (1980). i.d The reader should notice that the above model is not adequate to deal with the structure of the surface of a solid. The literature is plentiful on this aspect. (2831) = 41rn T’ bZ (2832) Here n is the density of the + 1 or 1 charges (see (1934)). The roughness of the surface of a crystal affects the rate of growth of the crystal very much.e. T’ < 4 . The interaction between the charges is still logarithmic. i. which is a major problem that can only be understood from the molecular interaction. producing “screening” phenomena. therefore T’ > 4 . positive and negative charges will pair up and the dielectric constant is finite. T > 2crl1r (28 . This way of transferring one model to another and the looking for a solution is a common technique. T ’ very small.492 STATISTICAL MECHANICS From the analysis of Chapter 19. (z(R) z(R‘)). The charge density n has yet to be calculated.e. Charges with 14 I > 1 contribute little to the process. Hence at high temperatures. T < 241r . 1 1 i. then single ions exist. . 1 a  In IRR’I 9 a i. at low temperatures. (1928) and the smallest 4 is 1 I f T ’ > + .is obtained from Eq. the conclusion of roughening is unchanged.
. b. v. and for ( y 2 ) see (2817). . take points at k = 0 . 3.~ V L i ) = . 1 m L' 1n7)(am) . Prove that 2" in (2812) can be written as L' = L o / b . The potential of each k element is also very easy to calculate. and estimate their difference using water as an example.2. . .T . . Discuss the meaning of the scale there for the model in Sec. 3 (2833) and let b >> 1. . If we join these points by straight lines. (2836) For the meaning of see (2813) and (2814). 3b . (c) Define the effective total energy as < H ' ( v ~ . (2834) k=l y k =m (b) Using the central limit theorem. 2. ... 28. This can be called the surface tension. they become a boundary. (a) Obtain the thermodynamical potential F of model (281) starting from (283). From the Ising boundary model derive the continuous boundary surface model. (a) Starting from model (281 l). 28. This is used to check the correctness of the above procedure. (2837) =1 Prove that the thermodynamical potential obtained from H ' is the same as that from H = 2JL in (281 1). This is macroscopic surface tension and is different from the microscopic one (see (281) and (283)). . . Review Problem 27. m b .FLUCTUATION O F THE BOUNDARY SURFACE 493 Problems 1. Of course the detailed change between m b and ( m + l ) b is lost.ln(2rb(y2))  s2/(2h(y2>) . 2b.1. (c) Compare this result with that in Sec. Discuss the meanings of 6 and a. The z in (281) and (287) are equivalent to b in that problem. .2. . prove that ln{(V) 21 1 bln< . (b) Let 6 = F / L o .
and also use (289) to calculate the thickness of the water surface at room temperature. the model in Sec. except that the scales are different. Prove that H ' = constant +2 (2838) T OLE (y2) . 28. Here. 4. . From (2838). Use (2810) t o calculate the value of { for water. The integral in the above equation is only an abbreviation for the summation of m. ( b ) Use the spherical harmonics t o analyse the shape of a water droplet. 5.1.494 STATISTICAL MECHANICS (d) In (2836). Try to discuss the importance. all are constants except the last term which contains 9. Notice that here b >> 1. The scale is (2838) or (281) is b . dx = b and q / b = d z / d x . this constant must be determined exactly. Discuss the fluctuation of the shape of a water droplet. The reader must understand the meaning and origin of H ' . If we want to calculate the exact 6 . Therefore. while that of (281 1) is 1. b does not seem to appear. The constant term in (2838) is very important. (a) Generalise (28 7) to discuss a surface which is nearly spherical. dz >> 1 and the constant term in (2838) is not cc L o . z is regarded as a smooth function of x = m b . (e) Notice that OL in (2838) is a complicated function of T (see (2814) ).2 is essentially like that in Sec. but without b the meaning of (2838) would be problematic. 28.
We shall use the twodimensional X Y model as an example to discuss the origin and effects of defects. long wavelength soft modes can appear.1.1) . “defects” are also an important fluctuations. The amplitude of fluctuation is related to the dimensionality of space as well as the symmetry of the system. s i v ) and sfx + . The Planar Vector Model We first look at the simplest vector model whose total energy is given by J 21. = 0 if i. Besides the smooth soft modes.e. using rather crude analysis to discuss the fluctuation of the order parameter. (29. low energy.i. these soft modes can soften and break the condensation. In one.the planar vector model (the XYmodel). Under this condition of condensation. This is the . . .Chapter 29 MODELS WITH CONTINUOUS SYMMETRY This chapter discusses three examples . si = ( s j x . These examples demonstrate such a relationship. = J . we .s = 1. i. condensation and continuous symmetry.e. where si is a unit planar vector. j are neighbours otherwise . 29.: socalled X Y model. As with the superfluid model in the next chapter. Lastly we discuss the common features of these models .or twodimensional models. the crystal model and the quantum vector model.
(292) Now assume that the neighbouring angles are nearly the same. The equipartition of energy gives ( l e k 1 2 ) = T/Jk .1) can be written as H = $ J z [B(R)B(R+ 1 R a a)]' + constant .4% STATISTICAL MECHANICS now look at the fluctuations of the direction of the vector. . Finally let (294) Substituting in (293) we get Jk r J C I l .1 ) a bit and let siv = sin Bi then si . Jk a k2 .i k * a 1 2 = 2 J z ( l .sj = cos(Bi . and therefore normally distributed. (296) . then cos(ei .Bi) . . All the average values are readily computed. then (29. First simplify (29.e j )2 2 Let R denote the position of the vectors. From (295).c o s k * a ) a a .e . we see that each Bk is independent.ei) 1 = 1 .(ei . (295) When k is very small. (293) where R + a denotes the nearest neighbours of R.
( 2 9 .8 )) . the solution of the electrostatic equation: V'I#J = S ( R ) .9 ) is the solution of (2910).8 ) This calculation has been performed for both one and two dimensions in the last chapter (see ( 2 8 . i.In na T a d=2 .MODELS WITH CONTINUOUS SYMMETRY 491 From this we can calculate n =  L rn N k Jk If R is very large. It can be easily seen from ( 2 9 . then 1 (I 1 cask.e. The correlation function of various si can be calculated from ( 2 9 . (2910) Except for some minor details. We only have to note that l / k ' is the solution of V'. t h e result being pi2 TR ' R ' d=l . Write Jk as the small k approximation Jk = ak'a' . (299) where d is the dimension of space.5 ) and ( 2 8 .9 ) : (291 1) . The integral for d = 3 is not difficult. The factor IlkD is approximately the size of the lattice. then the small k terms are important. ((Ae)* ) = . 28.1. R ) (29 7) (29 .9 ) that the fluctuation of the angle when d G 2 increases with R like the fluctuation of the liquid surface in Sec.
The above conclusion of no collective arrangement does not follow from a rigorous proof. then the above calculation is meaningless. which is proportional to (2913) . and hence the density is inhomogeneous. when d Q 2 there is no collective arrangement.12). The reader may attempt to solve it. From these results we can see that when d Q 2 no matter how small T is. . But why do we find no trace of the transition temperature in (2912)? The reason is that we have assume A0 to be a small quantity. . C(R) does not tend to zero and this has been mentioned in the last chapter.12) is correct. According to (29. 29. . and 0 and 0 + 2 n n . . The factor ln(R/a) in (299) cannot be very large. Therefore the conclusions of (299) and (2912) are valid for low temperature only. n = + 1 .498 STATISTICAL MECHANICS The last step uses the normal distribution of 0 (see the cumulant theorems (1232) and (1238)). collective arrangement occurs at any temperature. when d = 2 or 3. provided T f 0.2. The angles and . Therefore when d = 2 we cannot say that condensation does not occur. (The onedimensional model can be easily solved and the conclusion of (29.) We now look at another example. C(R) vanishes very slowly and on any reasonable scale. If 1 A0 1 2 71. C ( R ) decreases with R and tends to zero. From (299) we get d=l . (2912) C(R) = I (R/a)T’2”“ . Xray scattering measures the scattering intensity. + 2 . d=2 The above assumes that R >> a. are the same. Notice that when d = 2. Only when d = 3 does C ( R ) tend to a constant. At first sight this is quite different from the XY model. the crystal. but in fact they are very similar. but a rigorous proof exists and we shall mention it in Sec.n are the same direction.4. That is to say. 29. Density Fluctuations of a Crystal In a crystal the atoms are periodically arranged.
11. (29 . We first examine the relation between ( p k ) and the vibration displacement u : In the equation R denotes the average position of the atom and R + u ( R ) its actual coordinate. The displacements u ( R ) are the variable of motion. Hence if we use the cumulant theorems of Chapter 12 (see (1232) and (1238) ) w e get = .MODELS WITH CONTINUOUS SYMMETRY 499 The experimental analysis is like that in Sec. Now we ignore quantum mechanics. when k equals certain values. (See (1121).a ( p k ) is nonzero.e.W(k) (2917) and W has been computed in Chapter 11.) Using the Debye model we get k 2 To. . so u ( R ) are coordinates executing simple harmonic motion: u(R) = 1 i Aq f1 .15) The probability distribution of each qqAis normal. i. W(k) = 4n2nmc3 . Because of the periodic arrangement of the crystal.18) aThe reciprocal vector G represents the periodic structure and eiG" has the same period as the crystal. d = 3 . the reciprocal vector. and (2916) The cumulants of the normal distribution vanish after the second one.4.iq. R Q q A eAq (29 .
We now calculate (29 13) AU u(R) .2 W is the socalled DebyeWaller factor and represents the decay of the scattering due to vibration.~ ( 0 ) . When k = G . m is the mass of the atom.1 or that of ((At))' ) in this chapter. a reciprocal vector. d=2 . in a manner similar to (2917). this intensity decreases. (see (298) and(299)). 28. e .500 STATISTICAL MECHANICS where n is the density of the atoms N / V . (29 20) We use the cumulant theorem in the last step. (29 21) . Au)' ) is like that of (( Az)' ) in Sec. The calculation of ( ( k . As temperature rises. c is the speed of sound and wD is the Debye frequency. d=l .
1' ) does not exhibit periodic structure. (See (2918) and (2919). and e2w decreases with T. When q is very small the energy is small too. (2923) Therefore. Now substitute (2921) into (2920) and let k = G. For crystals in two dimensions. l n ( R / a ) cannot be too large and the periodicity is not totally lost. Crystals in onedimensional space have no periodic structure. long waves.e. Of course when d = 2. Because of the vibration of small q .MODELS WITH CONTINUOUS SYMMETRY 501 This is similar to (299). The factor 1/q2 in the integral comes from l / w 2 qh in (2916) as w i A = c 2 q 2 . The energy of an elastic body is proportional to the square of the gradient of the displacement. This q2 is Vz in another guise. . i. T > 4na . . then the energy is not changed. d=l.1). From (2918) and (2912) we know that when d G 2 the amplitude is so large that the periodic structure is lost. q = 0.) The proportionality to N2 represents the periodic structure of the crystal. a 5 n m c 2 / G2 (29 22) The case is special when d = 2: N (4 T a n . bFor a more detailed analysis. If the whole body undergoes a uniform displacement. plus a small correction of N 5 ' 3 . the amplitude u is large. because ( 1 p. see Imry and Guntler (1971). but when T < 4na it is intermediate between N 2 and N.e.N Z . We obtain I 2Nna/T . at high temperature ( I p. i. the intensity of Xray scattering by a threedimensional solid (scattering of k = G ) is . Iz ) is only proportional to N .
We call this state a0: H a o = E.. The ground state of H is easily seen. . while the dimension of the space d of the lattice can be chosen at will. . Now pick a particular direction. s= (. i. 2d is the number of nearest neighbours of each lattice site (assumed to be a cubic lattice) and N is the total number of lattice sites. as in the above examples. If all the spins point in the same direction. (2924) where R is the position of a lattice site. Notice that s has three components. 1) (2926) Each R is associated with a set of these operators. and s ( R ) is the spin operator. H is minimum. = .: s+ = sx f is. Now we look at the simplest quantum ferromagnet model. i.z .e. Quantum Vector Model The above examples do not involve quantum mechanics.502 STATISTICAL MECHANICS 29.3. . S ( R + ~ ).. the “downward” direction.J d N 4 where d is the dimensionality of space. . whose Hamiltonian is H =  1 J C 2 ~a C S ( R ) . (2927) 1 E. R + a denotes a neighbouring site. whose three components are (2925) In addition there are two convenient combinations s. Let all the spins point downwards. s+ = (. s is a vector in three dimensions. . the direction .) Q0 . and s.e.
a state of the same energy.e. not the stationary states. Hence. i.) is cdk = ~Jc(1coska) 1 a .3 1) This state i s an eigenstate of H a n d its energy cdk (over E. a linear combination of these states can give the stationary state. (2928) There are N such excited states. i. H has rotational symmetry: S E 1 s(R) R . i. (29 32) This result is obtained by directly substituting (29 31) into (29 24). Equation (2931) is the state of one .e.e. The simplest such combination is (2929) This combination is in fact another ground state. plane waves. Therefore. but they are not eigenvectors of H . and applying it again we get another and there is a total of N + 1 ground states. Now let (29 . (29 30) The operator S is the total angular momentum. (Assume cyclic boundary. However Hp ( R ) does not involve any other types of excited states (it only involves states with one up spin). and we get cp(R) = s + ( R ) @o . i. The simplest state is obtained by flipping one of the spins. not an excited state.e. it is invariant under a rotation of the coordinates. applying S+ on @ gives .) Hence the eigenstates of H must be eigenstates of the displacement operator. Since H is a scalar. no boundary. Notice that H is unchanged by a uniform displacement of the lattice.MODELS WITH CONTINUOUS SYMMETRY 503 We now look at the excited state of H .
and (2928) can be looked upon as a magnon at R . because there is only one spin turned up. The number of magnons is not conserved and hence its chemical potential is zero. The total energy is (29 33) When one spin is flipped. At low temperatures. if there is no overlapping). s(R’) ] = 0 if R + R‘ (i. This magnon can be treated as a particle with energy cdk and momentum k. If k is very small. (2935) (29 36) . the total angular momentum in the z direction is increased by 1. then 1 4 . (we neglect the situations when two or more magnons are at the same point. and the total number of magnons is N I(S>l = 2 ( 1  2” ) (29 34) As N ‘ increases the total magnetic moment decreases.e. The number distribution is like that of the phonons in a crystal. the number of magnons is very small. then (2932) becomes h r Ja2 If wk << T .) The magnons are bosons because [ s(R). From (2933) and (29 34) we can calculate the thermodynamical properties. and this model becomes an ideal gas.504 STATISTICAL MECHANICS magnon.
N'/N can exceed 1' If N'/N is too large then the total magnetic moment disappears. if L is sufficiently large. T L d=2 . When d d 2. that is.MODELS WITH CONTINUOUS SYMMETRY 505 We can estimate N ' : = . In this model the fluctuation of the spins is described by the magnons. 29. Obviously these examples reflect a common phenomenon. There is a common point in the models of these examples. Indeed we can ignore quantum mechanics and arrive at the conclusion ( 2 9 .4. For example if we rotate the coordinates. nh (29 37) where kD l / a is the upper limit of the k integration. the total energy function H has a certain continuous symmetry and in the condensed state this continuous symmetry is destroyed because of freezing. We can repeat the procedure of the above two sections. Continuous Symmetry and Soft Modes  From the fluctuation of the liquid surface to the fluctuation of the spins we have repeatedly encountered the same results. (See 2 (2934). ferromagnetism can be maintained.3 7 ) quickly. avoiding direct reference to the directions of the vectors. When d = 3 there is no problem of ferromagnetism at low temperatures. the energy of the vector model is . In (2936) we have already ignored quantum mechanics. no matter how T is. Therefore if T is sufficiently small. In ( L / a ) cannot be too large. L is the size of the system and n = N / V . The aim of this section is to show that conclusions such as (2937) have nothing to do with quantum mechanics. Symmetry means that H is invariant under a certain change of the variables.) If d = 2. When d = 1 ferromagnetism does not exist.l n ( ~ ) .
If we change all si to si. In amorphous magnets the frozen spins are random. and it cannot go beyond it. We can take any three mutually perpendicular directions of the spin as the coordinate axes attached inside the body. Thus. The change of this variable is not continuous. there cannot be a nonuniform density. if we translate the coordinates. the total magnetic moment) or a nonuniform density (e. the energy is unchanged.e. When the positions of the atoms are frozen in. it is not hard to perceive that there will be many excited states with very low energy. Due to the existence of so many ground states. If we rotate the coordinates the state will be different. This is because the determination of the region of motion in configuration space is related to the total energy only. For example the ferromagnet in the above example has N + 1 ground states. Coordinate translation is also a continuous change. there is a special direction (eg.506 STATISTICAL MECHANICS unchanged. this frozen state still has a preferred direction. These directions are frozen in and unchanged. i. if we move the liquid surface uniformly upwards or downwards. What is continuous symmetry? The opposite of continuous is discrete. condensation violates this homogeneity. Although the vector sum of all the spins is zero. Likewise the solid can stay in any position. Symmetry. whether it is a crystal or a random arrangement. Now rotate the individual spin . A system is made up of many parts. the energy is unchanged. The freezing of the continuous symmetry results in the existence of many ground states.g. the solidliquid interface). becomes the homogeneity of the equilibrium state. The trajectory of motion in phase space has been limited to the part of the region of motion determined by the basic assumption.g. All solids. through the basic assumption of statistical mechanics. the above ferromagnet model can be thought of as composed of many individual ferromagnets. each spin has its own average value pointing to a certain direction. We repeat once more that condensation is not limited to regular condensed states. In the ground state. e. We can pick the positions of four atoms not on the same plane to determine the position and orientation of this solid. The symmetry of the king ferromagnet model is discrete. the interaction between the atoms is unchanged. The explanation is as follows. Translational invariance results in homogeneity of the spatial distribution. The symmetry of the vector model is continuous: the direction of the coordinate axes can be continuously rotated. translational symmetry is destroyed. the magnetic moment can condense in any one direction. Hence. not only crystals. violate the homogeneity of space. According to our definition. all the spins point in the same direction.
Each term has only a few spin operators. and not just for the special form (2940).39) If each 8 is the same. i. This can be explained by writing down the equations from quantum mechanics. If they are originally parallel. U@. (29 . Let 8( R ) a ejk * . then this term will change.HR = O(ka) or 0 ( ( k a ) 2 ) .e. If all the rotation angles are the same.MODELS WITH CONTINUOUS SYMMETRY 507 by a small amount.i 1 O(R) s. If they are originally perpendicular. this term is unchanged. then UHRU' .(R) R . If these two spins are rotated by the same angle.40) in (2924). How much is the change? This depends on the original relative orientation of the two spins.i k . This is a low energy excited state. H is the sum of many terms which individually have their own symmetry. 8 ( R ) # 8 ( R + a). R (2942) . then it is proportional to the square of the difference between the two angles. But if the various rotation angles are slightly different. then the energy will be slightly increased. a . like HR . (29 .e.energy is unchanged. then the change of this term is proportional to the difference of 8 (R)and 0 ( R + a). (2941) Equation (2941) will hold so long as H R does not involve spins too far apart. 7 e. then the operator is U = 1 . is also the ground state. for small 8 . (2938) If each spin is rotated by a different angle. the operator for a rotation of 8 is. Therefore our conclusion is * 0(R) = (Ul)Cp. but if they are rotated differently. Let each spin be rotated about the xaxis.Js(R) s ( R + a) . then UHU' = H . changing smoothly from one part to another. the .. i.
However.5. this result is then unreliable. So is the surface tension wave in a liquid surface. This phenomenon of condensation softening can be treated rigorously. These are easy to read and are very important papers. but the rigorous analysis yield many other important details. If the distribution of the order parameter is random. they can break the condensation. There is as yet little work in this direction. In amorphous solids the soft modes may not be plane waves. The above analysis is not limited to spin models. Defects in the Condensation The above discusses the smooth fluctuation of the order parameter. soft modes appear. or quantum mechanics. . The amplitude of the fluctuation can be regarded as a smooth and continuous function of space. The change is smooth and the energy is low. Hence we have proved the existence of the low energy excited states. ‘See Hohenberg (1957). the role of long wavelength is quite important. Therefore. it is restricted to model with short range interaction. If a is too large. ‘ 29. Mermin and Wagner (1966). The conclusion is essentially the same as the above examples. We have deliberately displayed the factor of a in ka in (2941) and (2942). if there is a freezing in the continuous symmetry. We called these the “soft modes”.508 STATISTICAL MECHANICS where k can have many different small values. This is the characteristic of continuous symmetry. for both classical and quantum mechanical models. the result is not certain to be correct. If the amplitude of these soft modes is large. as their wavelength is long and their frequency low. The above result of the appearance of low energy excited states (energy + 0 as k + 0) when the continuous symmetry is frozen has the same meaning as the Goldstone theorem in quantum field theory. The above examples all reflect this phenomenon. and the soft modes can soften the order parameter. Both B(R) in the X Y model and u ( R ) in the crystal model can be regarded as continuous functions of R because their short distance change (of the order of a lattice spacing) is very small. In one and two dimensions. then the soft modes may not have a definite wavelength. Its form is determined by the distribution of the spins and the interaction energy. like the spins in an amorphous magnet. The phonons for the vibration of a crystal and the abovementioned magnons are this kind of low energy excitations. If we are dealing with models with electrostatic interaction.
. a ( R ) is shown in (c).n have the same meaning.M O D E L S WITH CONTINUOUS S Y M M E T R Y 509 / 0 \ / \ Fig. (b) Vector distributions B(R) on a plane. Notice that n and . 291 (a).
But near 0 it is not. (2944) is incorrect. 1049. Because when R 21 a .577 .d The reader may recognise that (2943) is the solution of vze = o . Fig. and R is measured from the origin 0 in the figure. a‘ will be slightly different from a . The vector distribution in Fig. .e. it is not infinitesimally small. Every black dot in the figure is a lattice point R . Discontinuous situations can occur.29l(b) (2943) In regions far from 0 i. Figure 291 shows the vector distribution on two planes. Now we use the twodimensional XY model as an illustration.1 is B(R) = a ( R ) 71 +7 . we get J H’ = a2 2 1(vqZ R (2945) where n L 2 is the area of the body. If R >> a . (2947) dSee Kosterlitz (1974). c=0. R >> a . then (2944) Ifwe substitute (2944) in (293). The four vectors near 0 have a difference of 7 ~ / 2 between neighbours. and these are very important too. B(R) is indeed a smooth function.= 2&eC a’ U . if we use . 29. (2946) then for a square lattice.510 STATISTICAL MECHANICS Although the lattice spacing is small. Let a ( R ) be the angle between R and the horizontal. p. (2945) is a very good approximation. According to a careful analysis.
were it not for the defect near the origin. (2944) can also be seen as the result of having a charge. (2949) These are expressions commonly seen in fluid mechanics and electrostatics. (2950) H I = 2nJln 7 (3 (2951) If there are many vortices. (2943) is quite “perfect”. representing the voriex strength or the charge. Now we summarise the meaning of (2952) as follows. If O(R) is regarded as the velocity potential in fluid mechanics. Hence the effect of this defect is easy to analyse. apart from the defect near the origin.dR=2n . Notice that going round this defect once. But if there is a pair of opposite charges or vortices. . (2943) would minimise the energy (2948) Therefore. (2945) points out that the energy of a vortex or charge increases like 1nL.MODELS WITH CONTINUOUS SYMMETRY 511 not too near the origin.R’) The energy of a pair of vortices is .a ( R . O(R) in Fig. Therefore. then energy is only related to their distance apart (see Fig. then V is O the velocity of flow and the configuration of O in Fig. then (2952) where R j are the position of the vortex centre or the position of the charge and (2953) is the integral around R i . Of course. 291 is a vortex. 292 is O(R) = a ( R ) . This is equivalent to saying that. 8 ( R ) in (2943) changes by 2 7 ~ : $VO. In two dimensional electrostatics a charge produces an electric field proportional to I / R . 292).
29. As B( R) is the electric potential.e.1.52) is the energy of a group of defects. i. The fluctuation and correlation of B(R) follow mainly the results of Sec. The solution of this equation is a smooth function. thus obtaining (2952). c q i = 0 . There is an additional condition that the total charge is zero. the lattice spacing is not zero. There is a condition for each small hole.e. we cut off a small hole of radius a‘ and on the plane punctured by these holes we solve V2 B = 0 . 292 The energy of two opposite vortices depends on their separation. i.512 STATISTICAL MECHANICS Fig. the integral (2953) must be an integer. . leading to a finite dielectric constant. positive and negative charges pair together. and this is substituted into (2948) to calculate the energy. At every Ri . the influence of the charges on the electric potential has been discussed in Chapter 19. i (2954) Equation (29 . This energy is not infinite because u’ is not zero. At low temperatures.
The paired charges or separated charges are built upon this assumption of near perfection. the answer is T. only with holes at a few places. it can be seen that defects can cause the condensation to collapse. When nowhere is “perfect”. = 2 J . Perhaps this rough estimate is not sufficient to dispel all doubts. and B(R) is an otherwise perfect and smooth function. defects involve not just a few lattice points but a large number of them. the defects are vortex lines. To eliminate a vortex we have eFor a more careful analysis. The term “charges” will not be appropriate and must be changed to line current. what is the point of talking about “defects”? The above charge model assumes that defects are few.e. and is left as an exercise for the reader. the charges will separate. Hence (2955) Therefore. see Kosterlitz (1974). the separation temperature To = r J / 2 for the paired charges is a rather high temperature compared to J . A vortex line can wind into a circle forming a vortex ring. if we want a l l 2 1 . But. the density of the charges at T = To is still very low. neighbouring vectors will not have directions which are nearly the same and the concept of defects will not be very useful. perhaps by luck. But. this vortex line is’metastable and once formed will not disappear easily. In three dimensions the planar holes would become tubes. .MODELS WITH CONTINUOUS SYMMETRY 513 If the temperature is high. Nevertheless. Generally speaking the energy of each charge is about e = q2 In Z/a and the average density is (all)’. where 1 is the average distance between charges. is the above assumption still valid? If we use the mean field method of the last chapter to calculate the critical temperature. the density of the charges should not be very large. then we must have T 2 q2 = 4T0 . So. The temperature where this takes place is To = q 2 / 4 = 71 J / 2 . and is somewhat worrying. hence condensation collapses. Notice that at high temperatures. At such a high temperature. e according to the rough estimate of Chapter 19.) This is close to To = 71 J / 2 . they should not appear in the equilibrium state. Later numerical simulation provided better evidence. and the correlation length for the fluctuation B(R) is finite. (The calculation is very simple. The above is for the twodimensional X Y model. i. and the energy is certainly very high. and VB denotes the direction of the magnetic field. or the ends of the line may be positioned on the boundary of the body. Therefore. In this way. Near T = T o .
Do not use (293). Use the mean field approximation for the X Y model (291). 293). There are many other types of defects and we shall not discuss these here. (a) Calculate the thermodynamic potential. Young (1979). [For definition see(291 l ) . With this definition of defect. calculate the defect density in this model.514 STATISTICAL MECHANICS to rearrange many elements.) These will be rediscussed in the next chapter on superfluidity. 293 A line dislocation in a crystal. The theory of defects also appears in other models.g. The reader :an look up the recent literature. (d) If the angle between two neighbouring vectors is greater than a right angle. entropy and heat capacity. (This is also true in two dimensions.f Fig. and sketch them versus 2 (b) Calculate the correlation function C ( R ) . In the above problem. ] (c) Compare the above result with the solution of (293). 3. Problems 1. Nelson and Halperin (1980). ' . . There is an extra line in the crystal shown in the figure. Solve the onedimensional X Y model directly from (291). fE. would we encounter difficulty when we define entropy? How can we resolve this? Notice that si is a continuous variable. and the simplest is the line dislocation (Fig. The defects in crystals are much more complicated. we say that there is a defect between these two elements. The reader can consult books on material science. Review Chapter 27. All continuous variables require quantum mechanics for the definition of entropy. 2. Calculate the critical temperature T c .
and space has dimension d .s i i . . Define the gauge transformation as q' = eiaN * . Compare it with the analysis of Sec. and for T near 2''. 29. . 2. si is a vector in three dimensions. eiaN eiaN (29 . . (a) Review Problem 3. . . .x h i . .e. Notice that when T < T.1.MODELS WITH CONTINUOUS SYMMETRY 515 4.60) N k is the number of particles in the state k.8.e. 6. Prove that the total energy is unchanged under the gauge transformation. = H . Apply the analysis of the above three problems t o t h e threedimensional vector model. . Bose gas and gauge transformation. The discussion in Chapter 3 neglected the interaction. assuming that hi is very small. and T near T. . N are the positions of the particles. the order parameter points t o a fixed direction. (b) Calculate the correlation function C ( R ) for T << T. . 5.1 ) : . (c) For the moment neglect H ' and consider the kinetic energy: (29 . (a) Generalise (2747). . Add a magnetic field t o (29. i. (2958) where (Y is the gauge rotation angle. (2956) Generalise the analysis of Problems 27. . This is t h e socalled classical Heisenberg model. (b) Let q ( r l . r N ) be a wavefunction of the system. Now we add an interaction term (295 7) where r i . i = 1.5 9) Here N is regarded as an operator and this equation says that particle number is conserved. Consider the two special cases of T << T. i.2 and 27 8 to the X Y model.
2. but condensation still occurs. No is very large. The reader can consult books on solid state physics. The expectation value (ao ) is a large number of order and this can be regarded as a condensation phenomenon. changing J in (2924) to . We shall rediscuss the bosons in the next chapter. Hence the state k = 0 can be treated classically. (d) Problem (3. Use the analysis in Sec. while a. Apply the analysis of soft modes in Sec.4 to discuss the low energy excited states.1) .6 2) We can write it more simply as (29 63) 1 (29 . (29 . 29. State k can be looked upon as a simple pendulum.516 STATISTICAL MECHANICS This can be regarded as a group of simple harmonic oscillators.e. (a) Compute W ( k ) for d = 1.64) The operator ak is the annihilation operator.8) shows that at low temperatures. the BoseEinstein condensation. p ) plane: = (qcosa+psinct. (e) Suppose H’ cannot be neglected. 7. (b) Find the position distribution of any one atom: a . i.4 to this model. because Nk is a positive integer. pcosctqsina) .e. p ) plane. No = O ( N ) . i. 29. The configuration is therefore represented by a point on the ( 4 . Prove that the gauge transformation is a coordinate transformation on the ( 4 .J . (see (2917) and (2918) and compare with (292 1). is the creation operator. This model is not so simple as the ferromagnetic one. 8. i.e. (2961) where p and 4 are the momentum and position of the simple pendulum. with energy f k N k = + f k ( p * + q2 . Discuss the quantum antiferromagnetic model.
11. but the mass m' of the impurity atoms is much larger than the mass rn of the other atoms. The total energy function is like problem (8. 9. paying special attention to the application of the complex variable theory.e. Suppose that in a certain neighbourhood. (b) Discuss the influence of the impurity atoms on the amplitude of atomic vibrations. . 10. (a) Let z = x + iy. i. f ( z ) is an analytic function of z . In a ddimensional crystal. and let f ( z ) be a function of z . (a) Calculate the influence of the impurity atoms on the thermodynamic properties. (b) Prove that the equipotential lines of @ and $ are orthogonal i. Review the fundamentals of fluid mechanics and electrostatics. (29 . Prove that in that neighbourhood (3)f + aY2 V$ *V@= 0. The last problem deals with d = 1.MODELS WITH CONTINUOUS SYMMETRY 51 7 (c) Suppose a copper film (with one layer of copper atoms) has a size L = 1 km.12). Then@and $ satisfy V2@ = 0 and V 2 $ = 0 . N' atoms are replaced by impurity atoms which are distributed randomly. Discuss how to use this function to analyse vortices and charges.2. Estimate the value of ( u2 >.6 5 ) Let f = @ +i $ .e. 29. Discuss the influence of the impurity atoms on the various conclusions of Sec. d f / d z exists. (c) Let f ( z ) = In z . a2 a2 = o . Suppose N atoms form a onedimensional chain with N ' impurity atoms distributed randomly in between. with N > > N ' .
On the other hand 4He are bosons. Our aim is to analyse the basic principles of superfluidity which is a rather mysterious property. Because 3He and electrons are both fermions. Superfluidity involves two phenomena: condensation and metastability. The thermodynamical potential of the flowing state is higher than that of the static state and i's not an absolute minimum. This chapter does not attempt to give a systematic introduction to the properties of liquid He and superconductors. e. Our aim for discussing the metastable state is to understand the process causing the flow velocity to decrease. continuous flow in the absence of external force. 4He liquid ( T < 2. the former being a necessary condition for the latter. All the phenomena of superfluidity share certain basic features.Chapter 30 THEORY OF SUPERFLUIDITY Superfluidity denotes frictionless. The superfluidity of 3He is rather complicated. This phenomenon occurs at low temperatures. and this superfluidity of electrons is called superconductivity. and over a short period of time (which may be years) it is effectively an equilibrium state and the thermodynamical potential is a minimum so that the flow is not stopped by perturbations (provided the perturbations are not too large). The state of flow is a metastable state.g. the theory of 3He is like the theory of superconductivity. Here we shall use the simplest model to explain it. The electrons of many metals exhibit this phenomenon at low temperatures.17K). But if the flowing state is to make a transition to the static state. (See Chapter 21 for the discussion of the metastable state. it must first overcome a potential barrier. and 3He liquid ( T lower than about K). with the transition temperature depending on the kind of metal. and the situation is simpler.) We must understand what this potential .
even though electrons are fermions and are charged.dimensional model (i. Indeed. Then we discuss the order parameter in the volume of a torus.e. We shall make a careful analysis of these two important points. It represents the change of the angle of the order parameter divided by 2 7 ~ when circling the torus once. We first simplify the model to a planar vector model.1. The winding number is then introduced.The defects are just the vortices. The content of this model is like the spin ferromagnetic model in the last chapter. which is nothing mysterious. discussing the metastability of the condensation and the flow. the stability of the flow will not occur. what is being condensed? We shall use a ferromagnetic model to explain this point.THEORY OF SUPERFLUIDITY 519 barrier is. Quantum Lattice Gas Let us set up a quantum lattice gas model now. Because we are quite familiar with the ferromagnetic model. The detailed analysis rests on the two. just as we analysed the coexistence of liquid and gas using the Ising ferromagnet in Chapter 27. The superfluid flow velocity is proportional to II. Hence the content of this chapter is quite minimal.e. There are many similarities with the basic principles of 4He. Both experiment and theory appeared only recently and the bibliography is rich. i. like the various sound waves. If there is no condensation. After establishing the order parameter. we can use it to analyse the property of the lattice gas. 30. Let n ( R ) be the particle . But the aim of this chapter is not to introduce the literature to the reader. We do not discuss this either. The theory is simpler in two dimensions but the experiment is more difficult. We first establish a lattice gas model. From this we can understand the essence of the theory of superfluidity. Another important topic is the behaviour of superfluidity and other properties near the critical point. the X Y model in the last chapter. have not been mentioned here. The order parameter is the magnetic moment. the model of 4He adsorbed on a smooth surface). only then will we have an initial understanding of superfluidity. Neither is superconductivity. an offspring of the quantum spin 1 model. even though it is an important and interesting science. the thermodynamic properties at low temperatures. The mystery of superfluidity lies in this order parameter. The experiences we gain in Chapter 19 and the last chapter will be very helpful here. This winding number n is a conserved quantity if we do not consider the condensed defects. we can discuss superfluidity. The content of the last part of this chapter is to investigate how vortices change a . Many other interesting problems. Here the fluid model is the quantum lattice gas.
The state with zero particle number is the vacuum. Now we construct an approximate form. in other words. Let (304) If ka << 1. But now the particle is restricted to a discrete lattice. We first ask: what is the kinetic energy? The kinetic energy of a particle should be k 2 / 2 m . i. S ( R + a) ~a 2 (306) aThis n is not the winding number which we later discuss. it adds a particle at R . (302) The state s (R) .e. . Equation (304) can also be written as K = .(R) . Now we have to write down a reasonable Hamiltonian operator H . operating with K on (303)will give (305) In (304)a is the neighbouring vectors of the site. a Let n(R)  = s.J 1 s ( R ) . and we take the lattice to be cubical. c n(R)cPo R = 0 . it is not easy to find a reasonable kinetic energy operator. then its wavefunction is (303) The operator s+ ( R ) increases the value of sz (R) by 1.520 STATISTICAL MECHANICS number on the site R . describes a single particle at R . with value 0 or l . If the particle is a plane wave. the state with all spins pointing to z and we call it . (301) operator. Hence the lattice gas model where sz is the z component of a spin has a content similar to the spin f model.
s(R)] a a . Without quantum mechanics. but is otherwise quite different. Lastly.J 1 2 11S ( R ) . + Therefore U is the second term on the right of (306) plus a constant.( R + a) U 1 . H = K + U = . However. of (303) gives. The variable of motion of the Ising model has two values. 4 3 3 . (308) This operator acting on the ' $ k ' 9. We consider the case when the particle number is unchanged. this is impossible. J = ( mu2 )' . 1. N' is a constant. it is still a vector.) Hence this model is exactly the same as the ferromagnetic model. i. we shall discuss some details of K and P. Hence K is very much like the exchange interaction in the ferromagnetic model. then the vector property of this ensemble can be analysed by the concepts of classical mechanics. Now we simply discuss the following model.THEORY OF SUPERFLUIDITY 521 where N ' is the total number of particles. this gas has kinetic energy K and potential energy U. The king ferromagnet is not a spin vector model. Let i. Notice that n = s. ( N ' is a constant.e. like the spin model. S ( R + a) + constant ~a . Although each spin has two states. we need a momentum operator. The aim of introducing this model is to point out several important features of quantum mechanics. if we have many spin together pointing in the same direction. Spin is a vector.[ s . Creating a threedimensional vector from two states is the result of quantum mechanics. '$k 21 . In the problems at the end of the chapter. for k u << (309) Therefore P can essentially be regarded as the momentum operator. (307) That is to say. which comes from the attraction between neighbouring particles.
The wavefunction is symmetric. and S is a necessary . We first look at the ground state. the quantum lattice gas model can have condensed states. in quantum mechanics. The Ground State and Low Temperature Model ferromagnetic model Now we analyse model (307). N 2 ’ (3010) is a constant. There is a restriction of at most one atom per lattice site. 3 + 3 i 30. because the total number of particles N ‘ is a constant. U is a special attractive force. and the quantum lattice gas model (307) is a disguise of the spin model. rigged to make up a simple form of H even though it may not be entirely realistic. have no equivalent in a classical gas.) These nonzero Sx and S . and S . = N ’ . see Fig. The Ising gas is a classical model and its state is determined by the distribution of the particles just like the value of each spin of the Ising ferromagnet. this is a straightforward conclusion. .522 STATISTICAL MECHANICS The Ising lattice gas model is a disguised form of the Ising ferromagnetic model. . then besides n . The size of the lattice a is the size of an atom. The condensation of S. Notice that this is a model for bosons. because of (30lo). S and S can all be nonzero. this is not due to the fermionic character but due to the hardsphere repulsion. If there are two states n = 1 and 0. is limited on a ring. For the . N = V/a3 . . there are also sx and s .11) is the number of lattice sites. which is the state with all the spins parallel.2.. which is just a spin and is like (2924) of the last chapter. Model (307) can be regarded as a prototype 4He system. there must be two other variables sx and s. The state of the quantum lattice gas model is specified by the distribution of the particles. The components Sx . same reason. I/ is the total volume and a3 is the volume of a lattice cell. The spin model can have a condensed state when the total spin S condenses in any direction. besides s. The only condition is that S = ” . But.. But from the ferromagnetic model. (30. . all operators are variables of motion. and have nonzero S. condition for superfluidity. N where N ’ is the total number of particles and N is the total number of lattice sites. The direction of the spins. That is to say. 301. Although each lattice site can accommodate at most one atom.. just like the value of each sz (R)in the spin model. representing the short distance repulsion between atoms. . (S.
1(b).e. This angle is determined by the values of S.13) In (30. then we get the direction of S in Fig. while I v l2 is that of finding one particle at R. is the ground state wavefunction. (30. S . The azimuthal angle $ can take any value. 301 (a) The ground state with all spins pointing downward. The total length of S is f N . (3012) We first rotate by 0 about the ydirection. and I S I. (b) shows the ground state obtained by rotating (a).e. The reader can prove that * v f ei@12 sin012 . The state of the spin pointing . i.2 direction.l(a) denotes Qo . I u l2 is the probability of having no particle at R . The vector So in Fig.13). the zero particle state with all the spins pointing downwards. The wavefunction of this ground state is very easy to calculate. 30. N spin linked together.THEORY OF SUPERFLUIDITY 523 $ S Fig. From the viewpoint of the spin model this combination is not surprising. and Sr are not zero and the angle $ is therefore defined. To obtain the ground state of N ’ particles. we only need to turn So by an angle 6 : \k=e i@Sz t e iBS Ya0 . i. 30. Because of a delicate combination of u and v . These rotations can be performed immediately because each spin can be rotated independently. The vector S makes an angle 0 with the . then by $ about the zdirection.
= N'/N J' = J n ( 1 .) Let (3014) This is the result of turning s (R) by @( R) . we get a low energy excited state. which is the reason for the factor sin2€' in (3015). Now s ( R ) has been turned by @ ( R ) and s ( R + a) by @ ( R+ a).1)' = 4n(ln) . so that the change of @ for neighbouring spins is small. If we change the angle @ of each spin slightly. this product is $ . then 9 is just another ground state.@ ( R + a ) ] 4 .n ) .( 2 N 1 / N . 29. only the components on the xy plane is influenced. and the change of this product is 1 . (3015) Because the rotation is around the zaxis. After each s ( R ) is rotated. (Another point on the ring of Fig. If each @( R) is the same. (3017) n Therefore. (See the discussion of Sec.) The energy of 9 can be estimated as follows. sin2 0 c o s [ @ ( R ) .524 STATISTICAL MECHANICS to a certain direction can always be expressed as a linear combination of the up and down states. Therefore the change of the total energy is 1 J ' = Jsin'8 4 .y N and I S 1 = N / 2 : .4. sin2€' = 1 . ' what is the change of s( R ) s ( R + a)? If all the spins are parallel. (3018) . 30l(b). (3016) 1 The angle 0 is determined by S = N ' .
16) is just the planar vector model or the X Y model discussed in the last chapter. The oscillation of the angle @ is a kind of the soft modes discussed. Because J = 1/ m u2 and cos[@(R) . we get VZ@=0 . Setting the derivative o f H ’ with respect to @(R) to zero. so @ is the ground state.V@) 2 . The volume is V = A L and the energy isH’. 30. Of course this model H ‘ is derived from the quantum model (307).16) is no longer a quantum model and the variables @ ( R ) are regarded as classical variables. The reader should be reminded that (3020) is the ferromagnetic model in a different guise.n(ln) U 1  “( V 1  ” 03) V (3021) If the lattice spacing is very small. If @ is a constant. We put our lattice gas in this torus. The arrangement of the spins is a condensation. (3019) (V@)’ + constant . (3022) . State of Flow and Winding Number Suppose we have a torus of length L and crosssection A . The use of (3020) is simpler for our purpose. Now we calculate the other minima of the energy H ’ .t(a. Notice that (30. then H ’ assumes the lowest value.@ ( R + a ) ] we have d3R 2m 2 : 1 . and this model has rotational symmetry with respect to the z axis. (3020) p = . The low temperature properties of this model can be analysed by the Hamiltonian H’ in (3016). H ’ can be written in a better form. It is the low temperature version of (307) and only considers the motion of angle @ of the various spins. then p is the density of the particles.THEORY O F SUPERFLUIDITY 525 Equation (30.3. with the z axis along the torus.
because @ + 2 n n is the same as $.2 3 )) along the torus. the z coordinate is along the torus and the length is L . y . a winding line on . sI sin@) is just (. since $ is an angle.8 )(4’ very small) we get En + a positive number. but (3023) are also solutions of (30201. s y ) .2 . L ) . Substituting (3023) into ( 3 0 . (3024) En is a minimum value o f H ’ and not a maximum because if we put $ = $’ + Gn into ( 3 0 . Therefore Fig. Hence $(x. whose meaning is obvious from Fig. 302 According to the present model the state represented by Gn is stable. 302. The winding number n is unchangeable. The zaxis is along the torus and z = 0 and z = L are the same point.2 0 ) . . the crosssection being a circle of radius 2 sin 6 E sL. But (sL cos@. We call n the winding number. 302 is the distribution of the order parameter s (represented by ( 3 0 .$(x. 0) must be a multiple of 2 n . and n is the winding number of this curve : (3025) Fig. 3 0 . The coordinates(sl cos$. 1 Figure 302 shows a torus.526 STATISTICAL MECHANICS Of course $ = constant is a solution. y . As seen from Fig. sl sin$) describe a curve winding the torus.we obtain H’= pV (I) 2m 2nn 2 = En .
(3030) where we have made use of sinZe 4 a3 1 = p . by #(R) in the u and v of ( 3 0 13).n represents a stable state. No matter how the soft modes oscillate. . is a continuous function.rrn/L) (2m/L)/m = . if Q. (3028) where u and v are obtained by replacing Q. This F . vz = pV(2. can be calculated from H’=E. (3031) see ( 3 0 17) and ( 3 0 19). Q. putting Gn into (3030). When the temperature is nonzero.8 ) t o calculate its total energy. 30. =En + Fo. = #n+Q. Now s + ( R ) = sin8 e+ i @ ( R ) . its winding number is unchanged. s = v*u .1 (b) ).’ . i. The state Qn is flowing. That is to say n is a conserved quantity. + I d 3 R (V$’)z 2m .e. The answer is obviously F. Notice that \k’ of ( 3 0 14) can bewritten as (3027) *‘ = n R [ u ( R ) + v ( R ) s + ( R )1 @o .= u * v .we get the momentum along the torus P . s. Equation (3027) represents the average value of s+ ( R ) with respect to \k’. is the thermodynamic potential for the ground state n = 0. Therefore. . (3029) Equation ( 3 0 .8 ) can be simplified as P p j d 3 R VQ. (30 . mpVv.32 ) = .THEORY O F SUPERFLUIDITY 527 on the torus cannot change its value of n except by breaking the line and reconnecting. 1 2 (see Fig. Although En > 0. . the thermodynamic potential F. We need ( 3 0 . (3026) Q.
regarding it as a continuous function of R . The probability of breaking is about (3033) i. these points arrange in a chain like a necklace and the curve in Fig. 30I).e. This kind of normal flow with vanishing viscosity is therefore not a metastable state but an unstable state. But R is not continuous. then n would not remain unchanged indefinitely. the fact that the curve in Fig. As this model is the X Y model.) The conservation of n depends on the continuity of $(R). 302 is continuous.(n/a)*a. we have a condensation of s(R). If we also consider discontinuous situations. Now we look at the influence of defects. 302 is really a chain of discrete points. the defects discussed in Chapter 5. slowing it down. then this curve can be regarded as broken. though tending to zero. we shall see how vortices change the above stable flow into a metastable or even an unstable state. (The above model is threedimensional but to change it to one or two dimensions is very easy. L There are about N . The viscosity. A superfluid is not the same as a fluid motion with the viscosity tending to zero. and vs is the socalled superfluid velocity.e. The constancy of the winding number n comes about because we only consider smooth @(R). starting from a onedimensional model.e. To destroy the conservatio’n of n is to destroy superfluidity . instead it represents a set of discrete points. In the next section. i. This is a macroscopic thermodynamical result. If the difference of 4 for neighbouring elements is close to 71. Stability of Superfluidity The last section concludes that superfluidity is maintained by the conservation of the winding number n .528 STATISTICAL MECHANICS This is a stable total momentum and represents superfluidity. i. is still nonzero and the addition of a few impurities will influence the flow velocity. = N~E/T = eE/T a (30 . (a@/?z)* dz . the defects are the vortices. and is completely different from the superfluid state. breaks where N. A slight increase of temperature or the addition of some impurities does not affect this stable equilibrium state.34) .e. The winding number n cannot change because sine is nonzero (see Fig. i. In one dimension. 30.4.
we know that these charges will pair up as dipoles at low temperatures. Now consider the twodimensional situation. T is very small. If there are no defects. Vqj(R) is regarded as the electric field. E = 1 . i. the energy of the electric field is proportional to L(AJ E . The time scale of breaking is r . (30 . n changes very slowly. the electric field is smaller by a factor E : Electric field a E A4 = 2nn . the breaks happen repeatedly. But the angle difference before and after the break may be zero or k 2 n . Superfluidity can be regarded as a metastable state. Equation (3024) can be rewritten as (3038) . If N. i. (3037) The difference between having and not having defects is in the value of E . then L/a will not be large. then e > 1 . This n is no longer a conserved quantity and superfluidity disappears correspondingly.3 6) The factor 1 / can ~ regarded as the reaction rate at high temperatures.e. is very small.THEORY O F SUPERFLUIDITY 529 Assume that N.e. If there are defects. n can be unchanged or it can increase or decrease. Therefore.e. When dipoles are formed. and can be regarded as charges. From Chapter 19 and the last chapter. i. >> 1. The defects in two dimensional space are vortices. that is to say. it will be reconnected. ~ be This is only a rough estimate. Of course after this curve is broken. and their interactions can be represented by a dielectric conStant E .
Notice that the relation between the unit of charge q and p / m is q = 2np/m (see (3020). If there are no vortices in the tyre. We can call p s the elasticity. at low temperatures.3 . The tyre is our twodimensional model with circumference L . 30. The superfluid state will no longer exist. the order parameter can withstand a twist. The total momentum can be obtained from (3032). i. (2948) and (2953) ). E + 00 and p3 +. this figure represents 4Jn.40) then the dipoles will separate and become free charges. Of course we can regard the picture as a big vortex with a big hole in the middle and superfluid flows in the tyre. 0. When the elasticity is not zero. n = 16. (3039) where A $ / L is the degree of twisting of s. The p in (3032) will now be replaced by p s . The above discussions can be illustrated by some diagrams.e. The strength of the big vortex is 16q. The parameter ps is the elasticity of the order parameter. 87 17 6 0 32 TT 2 4 Fig. The radial lines are the equipotential lines of 4. the flow velocity can be maintained. T > To where (30 . but with a slightly lower value.530 STATISTICAL MECHANICS Therefore. Figure 303 shows something like a bicycle wheel. If the temperature is too high.
30 . Some electric lines penetrate through. The strength of the electric field is (3041) Here L / 2 n is the radius of the wheel assuming that the width of the tyre is much smaller than L . Fig. On the average the electric field is reduced by a factor e. 304. with the electric field pointing radially outwards.THEORY OF SUPERFLUIDITY 531 The radial lines of Fig. We only perform a crude analysis here. T > T o . while some meet the dipoles and the electric field will be less between the dipoles. We shall not consider the detailed motion of the vortices. If T < T o . i. 303 are the equipotential lines and can also be regarded as the electric lines emanating from the central charge n q .5 we have to change the number of electric lines inside the system.e. then the free charges screen the electric lines as in Fig. This situation is much more complicated than the onedimensional case. but look at the problem from the point of view of the process of energy decrease. If free charges appear.) The velocity is also smaller by a factor E. We now examine whether the state of flow is stable when T < T o . the charges pair up to form dipoles. 304 Figure 305 is an enlarged portion of the tyre. . To change the flow velocity we must change the winding number n . and 1 6 / electric lines can ~ still penetrate. There will no longer be electric lines inside the tyre and there is no flow. (See the discussion at the end of Chapter 19. This is a rather convenient viewpoint. As seen from Fig.
532 STATISTICAL MECHANICS Fig. n is lowered by 1. This is just a barrier penetration problem. L 2nn  1 mvs (30 . This electric field is proportional to the flow velocity. In this way n will decrease in the system. otherwise they will not be paired. To remove 1 line we have to break up a dipole.e.43) where r is the distance between + q and . We have to expend quite a large energy to break them up. Of course there is an attraction between the charges in the dipole... Hence if the flow velocity is large.whose maximum A occurs at r = rc .4 . so that the positive charge goes up while the negative charge goes down.44) . the electric field represented by the electric lines will attract the negative charges to the inner ring of Fig. the dipoles are easier to break and n decreases rapidly. re = .q 2 InE a  (30 . 305 and push the positive charges to the outer ring to reduce the energy. The energy of the dipole is the interaction energy of + q and . Now let us look into the details. 305 As we decrease 1 line. 1 r u ( r ) = ..q plus the potential in the external electric field (3042) i. But once broken up. Figure 306 shows u ( r ) .
306 Plot of u ( r ) . and they must surmount the barrier A .rrPS 1 m m vs a (30 . We can estimate how many free charges flow out per unit time and unit length. The value of A is (1n1) 2. N.12). (See (21 . We use the results in Chapter 2 1. then the dipoles are bound very strongly.THEORY OF SUPERFLUIDITY 533 Fig.45) If r is much smaller than r c .) The flow rate is (3046) (Ar)2 = 1 ( a2 u / ar2 )c (3047) where D is the diffusion coefficient of the charges. V is the total area of the system and A r represents the width of the ./V is the density of the dipoles. rc is small. Therefore if the flow velocity is large. To separate one pair is extremely difficult.
The defects are vortex lines. 3 0 .J n . 2n L m dt m (3048) Substituting (3045) into (3046) and then into (3048). Superfluidity is quite stable in .534 STATISTICAL MECHANICS barrier peak in Fig. The most important conclusion of the above is that superfluidity in two dimensions is a metastable state.q 2 / T ) ( q Z / e ) = 2nps/m. 306.*IT. If mu. The diffusion coefficient must be calculated separately.6 will disappear and there is then no superfluidity. is very close to l / a . Hence it will be stable or metastable by itself and will not appear or disappear suddenly in the equilibrium state. This is related to the surface on which the 4He is adsorbed and the fluid mechanics of the vortices. The situation is even more complicated in three dimensions.. It involves a chain of atoms (because it is a line) and the energy is quite large. Each breaking up of a dipole decreases n by 1 . it is rather stable. 277 dn dS V dt = . / V ( l / a Z ) e x p ( . then the barrier in Fig. Hence dn _ dt =  LJ. The solution of (3049) for long times is (assume mvSa<< 1):  r 1 D aTa2 * (3050) At low temperatures. we get (3049) The above assumes N . vs disappears very slowly. This is due to the last factor in (3049) caused by .. The larger the flow velocity the faster it decreases. But at low temperatures and small velocities.
. etc. Let $(R) E s(R) .much like the ideal gas model. 3. If originally there are some vortex lines on the boundary. and also Putterman (1974). 2. .g. bSee the book by Wilks (1967). When N ' / N is very small ( N ' is the number of particles. Putterman (1974). 1. The properties of helium are discussed in considerable details in the reference books (e. Notice that this model is still not the same as the ideal gas model because each lattice site can only accommodate one particle.THEORY O F SUPERFLUIDITY 535 three dimensions. much earlier than the work in two dimensions. I s . The reader can read them for himself.l o ) ) . a large flow velocity will elongate them forming large circles. There was much work on this aspect some ten years ago. (3052) then Prove that $ * ( R ) $( R ) is approximately the number operator of the particles. Problems The requirements for this chapter are high and greater effort on the part of the reader is necessary. so there is still some interaction between the particles. The barrier for the metastable state is very high because the energy of the vortex lines is large. ' z( R ) .b The disappearance of superfluidity in three dimensions is usually due to the extremely high flow velocity. (a) Prove (305). The energy of the molecules in the equilibrium state is not sufficient to overcome this barrier. Wilks (1967). and Anderson (1966). or push them to the other side of the tube to decrease the winding number. These are very interesting fluid mechanics problems. Notice that (305 1) (b) Prove (309). Woo (1972).). see ( 3 0 . $* (R) f . prove that this model is very . The critical velocity is related to the details of the shape of the container.
Calculate 6 F / 6 @ ( x ) for the following: F = F = s I dx dx’ @’(x) $’(x‘) A(x.. sin .z . (3054) In (3012) a. he should review the quantum mechanics of spin and the transformation of axes: e iOs.( R ) ) = J/(R) . The derivation of (3020) from (3018) involves the computation of functional derivatives. If T is close to the critical temperature.536 STATISTICAL MECHANICS 4. Derive (3015) and (3016) from (3014).2 is not suitable. Condition (3010) must still be satisfied. the heat capacity and the order parameter (s). 30.2 is. 2 2 a=x. 5. If the reader finds difficulty following the derivations in Sec. the way out is to divide x into small sections and treat $(x) of each section as a variable. (b) Assuming that the order parameter is inhomogeneous: ( s ... . e e .6 0) .cos . (3055) If we run into difficulty. assuming that the order parameter is everywhere the same. 30.x ’ ) . To differentiate a function with respect to @ ( x ) we need to remember the formula 6 6 $(x 1 @(x’) = 6 ( x .2. (a) Use the mean field approximation to solve model (307). (3056) dx @ 2 ( x ) (3057) (3058) 7.. Calculate the critical temperature T.X’) . The discussion of this chapter is limited to the low temperature case. There fore the state of each spin can be represented by (:) . (30 5 9) Prove that J/ ( R ) is approximately the solution of V2J/AJ/+BJ/3 = 0 (30 .y. denotes the state with all the spins pointing downwards. 6. then the model in Sec.
(c) From $ to . From 6 F o / 6 $ = 0 we immediately get (3060). (3063) (d) From the discussion of Chapter 21 we can estimate the reaction rate of the process n to n . and the saddle points. The basic concept of this chapter can also be applied.1 3 . i. 8. 302 is not limited to the surface of the torus. Refer to Prob. That is to say. the curve must cut through the torus at a certain place. We assume that $ does not change withx o r y . But if n changes. Let the saddle point be $’. 29. Notice that I J/ ] in this model can change. The lowest point of the barrier is a saddle point.e. Discuss the change of the winding number n . Fortunately the solutions of (3060) includes the minima. 27.64) . 27.1 . . i. the above model can nevertheless be used. the maxima. during the process where n changes to n . Although we have not discussed the theory of superconductivity. calculate A and B . the curve of Fig. The above model has been used to analyse the superconductivity of fine metal wires. but the details are different.e.2 to obtain expressions like (271 1) and ( 2 7 .2 and Prob.T is very small).) The basic line of thinking is as follows: (a) The phase space is the set of $ . is a minimum of Fo and is the absolute minimum at $0 . Hence the height of the barrier is A = Fo($’> . 30. including Po($) = (H) TS (3061) (2712) is equivalent to (3060). This process has been carefully analysed. ( See Langer and Ambegoakar (1 966).3.A/T I0 (3 0 . Each $ is a function of z. which is the coordinate along the torus. (b) Equation (3060) determines the minimum of Fo in this space.THEORY OF SUPERFLUIDITY 537 (assume that T. we must overcome points with large F o ( $ ) . _ 1 I 1 . (c) Use (3060) to replace (3021) and discuss the various conclusions of Sec. The concept of winding number is still applicable. Starting from (307).Fo(Gn) . I $ I must decrease to zero at some point and then increase again. Use the method in Sec.1. overcome a barrier. The constants A and B should be computed from the model of the metal.3.
k) states. + constant . (d) Prove that the excited states can be regarded as an ideal phonon gas. but have not been discussed.538 STATISTICAL MECHANICS The factor T~ is not easily computed. The zeropoint wavefunction of each pair of (k. (a) Review Prob. Hence model (3066) can be easily solved. (3066) (c) Notice that state k is only correlated with k. is nearly the same as N . Phonon gas. At low temperatures No = a: a.6. 9. and the other ak can be treated as small quantities.e. Prove that the Hamiltonian can be written as u9 I d 3 r e i q a ru ( r ) . 29. Calculate the zeropoint energy and the wavefunction. Hence H i s approximately equal to H 2 : C (ek + 2 n u k ) a i a k k +n k u k ( a ~ k + ak a k ) e. the vacuum. Calculate A . The reader can attempt the problem first and then consult the reference material. . The reader can derive some results as follows. If the reader wants to do more extensive reading. The thermodynamic properties of bosons are very interesting too. Let ak 1 uk ak . i. The main emphasis of this chapter is on superfluidity. but the other factor eA’T comes from the barrier. n N/V .v k a*k  (3068) .) (b) Suppose the interaction u is very weak.k) states is (3067) @ is the zero particle state.. he can refer to Putterman (1974). (3065) (See (2957) to (2964). The phonons are defined by the excited states of (k. .
This view is incorrect. Here uk and uk are constants. (3069) H = C wk a. (e) Use (3070) to calculate the thermodynamical properties. . a calculate u k and uk and prove that ak \k.THEORY OF SUPERFLUIDITY 539 This is the annihilation operator of the phonon and a. and taking c > 0 as the origin of superfluidity. (3070) If k is very small. no phonons are emitted and energy cannot be dissipated. Notice that in older textbooks or literature it is usual to connect superfluidity with the energy of the phonon. c2 = nuo/m . (3071) This is the. is the creation operator. ak + constant k .] = 1 . The reader should take special note of this. From [ak. then wk = c k . but no superfluidity.) This chapter emphasises that the conservation of the winding number and the elasticity of the order parameter are the roots of superfluidity. (Solids have phonons too.motivation for the term phonon. The above were obtained some forty years ago by Bogoliubov. . When the flow velocity is lower than c. = 0 .
.
E. 53. Hu (1 979). J. Phys. New York). (1 979) Monte Carlo Methods in Statistical Physics (SpringerVerlag.. and P. B. Phys. Weeks (1976) Phys. Beth.. E. C. Z. K. New Jersey).. K. Alexandrowicz. (1979) Phys. T. Gorkov and I. N. Rev. Ma and C. A. New York). Rev. S. F. Arnold. ( 1 975) J. W. S. J. New York).. Avez ( 1 968) Ergodic Problems of Classical Mechanics (W.London 243A 299... I. Bernstein (1969) Phys. R o y . Abrikosov. Berlin). Mod. 38 298 Anderson. Stat. D. . Berne. No. ( 1 966) Rev. and J. W. Frank (1 95 1) Phil. and G. C. R. P. Uhlenbeck (1937) Physica 4 915. P. Trans. V. 2 5 1. 187 345. A. I. W. F. Varma (1 972) Phil. Rev. Carruthers (1963) Lectures in the Many Electron Problem (Gordon and Breach. E. P. B20 3837. Binder. 1 3 23 1. SOC. Dzyaloshinski (1963) Methods of Quantum Field Theory in Statistical Physics (PrenticeHall. (1 977). B14 4978 Dasgupta. ed. Brout. L. and A.. Chui... Cabrera and F. S. Dashen. Burton. Phys. K. B. Ma and H. Anderson. Halperin and C. Mag. Benjamin. 2. Rep. R.REFERENCES Abraham. Statistical Mechanics (Plenum Press. A.
R . Landau. 80 580. and E. J. M. London). S. J. Imry. D. Stat. Math. 32 13436. Lebowitz. Fisher. . 22 631. and S. Rev. Ma. C. M. B3 3939. Hohenberg. Lieb (1 969) Phys. K. (1979) Works on the Foundations of Statistical Physics (Princeton University Press. E. F. Kittel. Green. J . Lett. C. S. A. Phys. J. M. J. Pitaeviskii (1 980) Statistical Physics (Pergamon Press. M. eds. Rev. and E. Griffiths. Phys. Gibbs. Krylov. 19 25. (1963) Statistical Mechanics (J. Hahn. (1976) Modern Theory of Critical Phenomena (Benjamin. Wiley. and J . M. Wiley. (1 95 1) Phys. C. C. (1972) in Phase Transition and Critical Phenomena Vol. and L. B. Lett. Rev. New York). Rev. New York). J. C. 1. Goodkind (1974) Phys. 42 795.542 STATISTICAL MECHANICS Dashen. (1 967) Phys. Rev. Feigenbaum. C1 1046. 81 988. M. (1 966) Introduction to Solid State Physics. A. E. Princeton). A4 700. Thouless. and L. K. Dundon. S. L. New York). (1950) Phys. Gunther (1971) Phys. Forster. Lifshitz (1 980) Statistical Physics (Pergamon Press. J . New York). Kosterlitz. J. (1 973) J. (1974) J. (1 956). (1960) Elementary Principles of Statistical Mechanics (Dover Publications. and G. S. (Academic Press. (J. C6 1 18 1. 164 498. and D. P. Adams (1979) Phys. J.. W. Broken Symmetry and Correlation Functions (W. 8 1538. Lifshitz. Rev. Benjamin.(1967)J. 3rd ed. (1 975) Hydrodynamic Fluctuations. and V. Huang. Rev. London). L. Y. Dyson. New York). (1978) Physica 1 0 6 A 28. Rev. (198 1) J. H. Ambegaokar (1967) Phys. Phys. Domb and M. P. Grimes. N. Kosterlitz. M. Kikuchi. 158 383. E.F. New York). L. Ma (1 971) Phys. M. L e f t . D. Phys. Rev. Langer. R.
New York). (1968) Phys. Rev. H. R. (Cambridge University Press. Fluids 1 1 134. 45 389. Thesis. 32 477. Theor. D. G. Wiley. A. ( 1 974) Superfluid Hydrodynamics (NorthHolland. Fluids 1 1 1147. Prohofsky. Goodkind (1967)Phys. (1 962) Principles o f Statistical Mechanics (Oxford Unjversity Press. Wharton. Rev. Rev. (Oxford Clarendon Press. Gould. R. Weeks. (1936) Camb. M.. (1 970) Ph. W. Lett. Phys. J. Ma. C. (1971) Introduction to Phase Transition and Critical Phenomena (Oxford University Press. S . ( 1 944) Phys. K . 26 221. T. S. B24 3984. J. H. (1955) Quantum Theory ofSolid. (1965) The Theory o f Magnetism (Harper and Row. ed. T. (Kyoto) 14 35 1. J . Peierls. L. Princeton). W. K. Mermin. Riste. Onsager. C. Rev. Tolman. Rudnick. San Diego. London). and J. Amsterdam).(1981a)J. Mattis. C. (1 977) Introduction t o Ergodic Theory (Princeton University Press. B21 5312. T. Cambridge).. E. Peierls.REFERENCES 543 Ma. Stanley. New York). D. R. University of California. W. Krumhanvl(l964) Phys. Phys. and M. (1 955) Prog. . Tabor. (1 979) Surprises in TheoreticaZ Physics (Princeton University Press. Sac. Peierls. Sinai. M. and H. and B. (1 977) Chem. London). E. R. London). O’Neils. Princeton). Rev. and T. Nelson. (Plenum. Matsubara. D. 65 1 17. B. H. O’Neil. New York). D. and J . M. (1 979) Gases. G. I. N. R. S. Stat. ( 1980) in Ordering in Strongly Fluctuating Condensed Matter Systems. Malmberg. Wannier. B. Gould ( 1968) Phys. E. Phys. and R. D. Payne (1981) Phys. Wagner (1966) Phys. H. Lett. Phil. 18 894. Putterman. Osgood. Liquids and Solids 2nd ed. Ya. (1 966) Statistical Physics (J. A133 1403. 17 1133. Lett. J. Halperin (1980) Phys. R. Mairovitch. D. Rev.
D. T. Vol. (1967) “University Physics”. 4th ed. Young. London). 87 404. (1971). C. W. (McGrawHill.544 STATISTICAL MECHANICS Wilks. Wilks. N. (1961) Third Law of Thermodynamics (Oxford University Press. T. (1979) “Probability” %4%i ’ i : p g Ec z .. (1 98 1b) “Physics”. j. Rev. K. Lee (1952)Phys. Y. M. W. P. Rev. (1979)Phys. (1952)Phys. 85 809. Zemansky. New York). C. N. (1979) “Thermodynamics. J . References from Chinese Books Lee. C. “Materials”. Y. C. Yang. 2. and T. London). B19 1855. R?Jg ’ %%Z@Jifi ’ mtfi9 ’ m9Y3%wE Tang. S. (1957) Heat and Thermodynamics. C. E%Cl3%%@ ’ +SiftE Woo. (1972) “Helium and Many Body Physics”. mB ’ +plmBsc Yang. 2.smBY~%ssE Ma. ed. A. J . Yang. W. $4$$ #i@gHS*i I Wu. (1967) Liquid and SoZid HeZium (Oxford University Press. Y. Kinetic Theory and Statistical Mechanics”. Rev.
30. law of 43 Condensation 266 Condensation forces 474 Conditional distribution 181 Conductor 6 Correlation function 182. 430 Correlation time 190. 332. imperfect 150 DebyeHuckel theory 330 Debye model 9 8 DebyeWaller factor 500 Decoding 456 . 431 Correlation value 17 1 Corresponding states 470 Coulomb gas 342. 472 Cyrstal 6 Crystal.41 Boltzmann equation 346 BoseEinstein condensation 5 16 Boson 35. 343 Cumulant 203 Cumulant expansion theorem 207 Curie’s law 282 Critical point 262.456 Characteristic function 199 Charge density wave 64 Chemical reactions 12. 390.411 Amorphous substance 6 Antiferromagnetism 6.24.INDEX Adiabatic compressibility coefficient 366 Adiabatic demagnetisation 280 Adiabatic process 13.474.41 Bubble chamber 273 Coincidence.479 Anharmonic vibration 150 Average values 167 Barrier penetration 370. 103.94 Chemical potential for photon gas 42 Classical distribution 42 ClausiusClapeyron equation 264 Cloud chamber 273 Coefficient of friction 368 Collisions with impurities 354 Concentrations. 189 Fourier transform of 229 Correlation length 231 237.532 Binary alloy 307 Binomial distribution 112. method of 426 Canonical ensemble 130 Caratheodory theorem 28 Carnot heat engine 24 Central limit theorem 200 Chaos 185. atom in 183 Crystal.400 Amorphous magnet 6 . 3 2 4 Amorphous states 312. 323. 201 Black body radiation 41 Boiling point 265 Boltzmann distribution 37.43 Chemical potential 11. density fluctuation of 498 Crystal.
111 . 318 Impurities. 150 Diamagnetism 283 Diatomic molecule 150 Dielectric constant 333 Dielectric constant of plasma gas 365 Different time correlation 193. 236 Exponential distribution 196 Extensive quantity 69 Fermi distribution 38 Ferrni gas model.454 Independence. 236 FermiThomas model 341 Fermions 35 Fermion gas 240 Ferromagnetism 6 . 193 Equation of state 28 Equilibrium. 14. 13. local 5 Equilibrium state 2. 4 6 3 . mutual 185 Infinite integrals 232 Information theory 429 Instability 159. 312 Goldstone theorem 508 Grand canonical ensemble 130. 8.. 19. 13. 31 8 Independence 180. law of 8 0 . 8 . total 13 Ensembles 130 Entropy 9. energy 57 Gas dilute 231 Gas ideal 19. 9 8 Ergodicity 442 Evaporation 263 Exchange interaction 291 Exclusion principle 31.546 STATISTICAL MECHANICS Defects 508 Dense fermion gas 363 Densegas 46 Density correlation function 230. 194 Diffusion distance 318 Diffusion equation 216 Diode model 369 Dipole interaction 160 Dispersion 167 Distribution 166 Distribution function 20.5 3 .515 Holes 5 1 Homogeneity of space 506 Ice point 8 Ideal gas. 4 6 4 Ferromagnet model 502 First law of thermodynamics 13 Firstorder transition 273 Fluctuation 164.9 9 Heat of evaporation 269 Heat of melting 269 Heat of sublimation 268 Heisenberg model 294. 403 Heat capacity 3 8 . 60 Invariant quantities 68. stationary 312. 17 Equipartition of energy.403.450 Internal energy 118 Insulator 6.443 Entropy of mixing 2 gases 2 2 . 248 Htheorem 363 Hard sphere interaction 232. 444 Distribution region 66 Effective energy 223 Effective interaction energy 315 Efficiency 24 Einstein's law 116 Elasticity of order parameter 530 Electron cyrstals 35 1 Electronic dipole interaction 247 Energy distribution 32 Energy. correction to 231 Impurity atoms 260 Impurities. 67. 332 Density expansion 250 Density of states 49 Density wave 64 Detailed balance 30. 3. see free electron model €:errni momentum 49 Fermi surface 49. 233 Heat 12. mobile 312. 166 Fluctuationdissipation theorem 2 15 Fluid 6 FokkerPlanck equation 377 Free electron model 48 Free flight time 5 Free particle 31 Frequency spectrum 239 Frozen entropy 436 Gap. 2 3 Equal time correlation 192. 181.243.2 3 0 Gauge transformation 5 16 Generalised coordinates 12 Glassy slate 6.445. 3 5 . 128.
law of 69.255 Phase transitions 8. 66 Phase shift 254.460 Partition function 125 Partition line 302 Partition point 298 Partition surface 303 Pauli susceptibility 283 Peak integration 89 Peierl's instability 48. 227 Magnetic moment 11. 149. see ferromagnct model Quenching 322 Random motion 214 .73.227 Magnetic susceptibility 227. 134.243 Phonon 35. 62 Periodic structure of atoms 55 Phase equilibrium 261 Phase space 33. see thermodynamics limit Magnetic fluid 12. 342 Metastable state 161.436 Ising antiferromagnetic model 307 Ising lattice gas 466 Isingmodel 117.412.432 Ising variable 297 Isothermal process 17 Joint distribution 181 Kondo problem 324 KosterlitzThouless model 337 Lagrange multipliers.428 Onedimensional gas 145 Open system 229 Open system. 8 .102 Photongas 42 Planar vector model. model of 341.359 Metal surface.INDEX 547 Irreversible process 10. 198 Normalization 168 Nuclear cooling 280 Observation time 2 .214. method of 89 Landau damping 422 Landau diamagnetism 286. 103 Mean free time 5. 190.405. 364 Plasma gas 341 Poisson distribution 183 Poisson sum rule 290 Pomeranchuk cooling 295 Pressure 11 Pressurisation. 372. 78 Latent heat 263 Law of crystallisation 397 LeonardJones potential 247 Levinson theorem 258 Lindemann formula 270 Line dislocation 514 Long range interaction 117 Longitudinal wave 98 Macroscopic 4 Macroscopic limit.281 Magnon 35.296 Landau theory 475 Large numbers. 186. 175 Projection 81 Propagation function 229 Quantum distribution 42 Quantum lattice gas 520 Quantum numbcrs 31 Quantum vector model.504 Mean field approximation 243 Mean field theory 458 Mean free path 5. 434 Melting 397 Melting point 270 Mixing 445 Mixture 323 Molecular motion 4 Monte Carlo method 30 Monte Carlo simulation 391 Nonequilibrium phenomena 4 Nonequilibrium problem 3 13 Nonequilibrium states 41 Normal distribution 115. 114.131. see XY model Planck distribution 42 Plasma echo 422 Plasma oscillations 349.281.281. rule for 127 Order parameter 473 Osmosis 312 Osmotic pressure 313 Painvise independence 188 Paramagnetism 6. cooling by 295 Probability definition 167. 277.
435 Time correlation function 190 Time delay in collisions 250 Tin.470 Vapour pressure curve 264 Vector model 294 Vibration 94 Vibration of nuclei 139 Virial coefficient 246 Virial expansion 25 1 Virial theorem 259 Viscous liquid bottle 416 Viscosity 362.405 Second order transition 273 Second sound 102.4 0 3 XY model 495.219 Remanence 325 Response functions 228. see IiermiThomas model Statistical weight 168 Sticking Coefficient 265 Stimulated emission 43 Stirling’s formula 78 Superconductor 6 Superfluid 6 Superfluidity 5 2 8 SUperhYdtd liquid 273 Supersaturatcd vapour 267 Surface tension 266. differential 226 Symmetry 506 System.353 Second virial coefficient 250 Semiconductor 60 Soft modes 508 Solid 6 Sound waves 103 Smooth process 11 Spinecho 419 Spin glass 6. reaction occurring inside 47 Statistics by classification 91 Statistics by summation 9 1 Statistical model of the atom. 13 Temperature.548 STATISTICAL MECHANICS Random sequence 39 1 Randomness 180. subsystem 88 Temperature 9. see WKB approximation 334 Second law of thermodynamics 24.237 Reversible process 10.189.477 Susceptibility. absolute 19 Temperature critical 47 Thermal electron radiation 6 3 Thermodynamics 1 0 .437 Region of motion 114.436 RKKY interaction 324 Rotation of molecules 139 Rules of calculation 122 Sample set 174 Saturation. 12. 324 Spin glass phase 325 Spinwave 35 Star. states of 410 Transfer matrix.529 Whitedwarf 63 Winding number 526 WKB approximation 33 Work 1 1 .9 7 Zeropoint pressure 5 1 Zeropoint vibration 269 .1 2 Thermodynamic limit 154 Thermodynamic potentials 120 Thud law of thermodynamics 408. degree of 267 Saturated vapor pressure 263 Scattering cperator 36 Scattering probability 112 Screening 33 1 Screening length 33 1 Shortwavelength approximation. 368 Vortex 511. method 310 Transition temperatures 8 Transverse wave 98 Triple point 263 Twostate cluster model 380 van der Waals equation 250.513 YangLee theorem 156 Zerosound 35 1 Zeropoint energy 5 0 . 435.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.