Statistical Physics

Yuri Galperin and Jens Feder
FYS 203
The latest version at www.uio.no/ yurig
Contents
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Deterministic Chaos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Atomistic Philosophy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.1 Cooperative Phenomena . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 What will we do? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Historical Background 7
2.1 Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Steam Engines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Sadi Carnot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 The Maxwell’s Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3 Principles of statistical mechanics 21
3.1 Microscopic states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Statistical treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Equal weight, . . . microcanonical ensemble . . . . . . . . . . . . . . . . . . . . . 24
3.4 Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.5 Number of states and density of states . . . . . . . . . . . . . . . . . . . . . . . 27
3.6 Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.7 Normal systems in statistical thermodynamics . . . . . . . . . . . . . . . . . . . 30
4 Thermodynamic quantities 31
4.1 Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2 Adiabatic process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.3 Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.4 Equations of state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.5 Work and quantity of heat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.6 Thermodynamics potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.6.1 The heat function (enthalpy) . . . . . . . . . . . . . . . . . . . . . . . . 36
4.6.2 Free energy and thermodynamic potential . . . . . . . . . . . . . . . . . 37
4.6.3 Dependence on the number of particles . . . . . . . . . . . . . . . . . . 38
i
ii CONTENTS
4.6.4 Derivatives of thermodynamic functions . . . . . . . . . . . . . . . . . . 39
4.6.5 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.7 Principles of thermodynamic . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.7.1 Maximum work and Carnot cycle . . . . . . . . . . . . . . . . . . . . . 42
4.7.2 Thermodynamic inequalities. Thermodynamic stability . . . . . . . . . . 45
4.7.3 Nernst’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.7.4 On phase transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5 The Gibbs Distribution 49
5.1 Spin
1
2
in a magnetic field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.2 The Gibbs distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.2.1 The Gibbs distribution for a variable number of particles . . . . . . . . . 53
6 Ideal gas 55
6.1 Classical gas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
6.2 Ideal gases out of equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.3 Fermi and Bose gases of elementary particles . . . . . . . . . . . . . . . . . . . 65
6.4 Black-body radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
6.5 Lattice Vibrations. Phonons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
7 Statistical ensembles 81
7.1 Microcanonical ensemble . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
7.2 Canonical ensembles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
7.3 Ensembles in quantum statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 87
8 Fluctuations 91
8.1 The Gaussian distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
8.2 Fluctuations of thermodynamic quantities . . . . . . . . . . . . . . . . . . . . . 93
8.3 Correlation of fluctuations in time . . . . . . . . . . . . . . . . . . . . . . . . . 95
8.4 Fluctuation-dissipation theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 101
8.4.1 Classical systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
8.4.2 Quantum systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
8.4.3 The generalized susceptibility . . . . . . . . . . . . . . . . . . . . . . . 104
8.4.4 Fluctuation-dissipation theorem: Proof . . . . . . . . . . . . . . . . . . 107
9 Stochastic Processes 111
9.1 Random walks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
9.1.1 Relation to diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
9.2 Random pulses and shot noise . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
9.3 Markov processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
9.4 Discrete Markov processes. Master equation . . . . . . . . . . . . . . . . . . . . 117
9.5 Continuous Markov processes. Fokker-Planck equation . . . . . . . . . . . . . . 119
9.6 Fluctuations in a vacuum-tube generator . . . . . . . . . . . . . . . . . . . . . . 121
CONTENTS iii
10 Non-Ideal Gases 129
10.1 Deviation of gases from the ideal state . . . . . . . . . . . . . . . . . . . . . . . 129
11 Phase Equilibrium 135
11.1 Conditions for phase equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . 135
12 Continuous Phase Transitions 139
12.1 Landau Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
12.2 Scaling laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
12.3 Landau-Ginzburg theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
12.4 Ginzburg-Landau theory of superconductivity (SC) . . . . . . . . . . . . . . . . 153
12.5 Mean field theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
12.6 Spatially homogeneous ferromagnets . . . . . . . . . . . . . . . . . . . . . . . . 160
12.7 Renormalization framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
13 Transport phenomena 167
13.1 Classical transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
13.2 Ballistic transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
A Notations 175
B On the calculation of derivatives from the equation of state 179
iv CONTENTS
Chapter 1
Introduction
1.1 Motivation
Statistical physics is an unfinished and highly active part of physics. We do not have the the-
oretical framework in which to even attempt to describe highly irreversible processes such as
fracture. Many types of nonlinear systems that lead to complicated pattern formation processes,
the properties of granular media, earthquakes, friction and many other systems are beyond our
present understanding and theoretical tools.
In the study of these and other macroscopic systems we build on the established conceptual
framework of thermodynamics and statistical physics. Thermodynamics, put into its formal and
phenomenological form by Clausius, is a theory of impressive range of validity. Thermody-
namics describes all systems form classical gases and liquids, through quantum systems such as
superconductors and nuclear matter, to black holes and elementary particles in the early universe
in exactly the same form as they were originally formulated.
1
Statistical physics, on the other hand gives a rational understanding of Thermodynamics in
terms of microscopic particles and their interactions. Statistical physics allows not only the cal-
culation of the temperature dependence of thermodynamics quantities, such as the specific heats
of solids for instance, but also of transport properties, the conduction of heat and electricity for
example. Moreover, statistical physics in its modern form has given us a complete understand-
ing of second-order phase transitions, and with Wilson’s Renormalization Group theory we may
calculate the scaling exponents observed in experiments on phase transitions.
However, the field of statistical physics is in a phase of rapid change. New ideas and concepts
permit a fresh approach to old problems. With new concepts we look for features ignored in
previous experiments and find exciting results. Key words are deterministic chaos, fractals, self-
organized criticality (SOC), turbulence and intermitency. These words represent large fields of
study which have changed how we view Nature.
Disordered systems, percolation theory and fractals find applications not only in physics and
engineering but also in economics and other social sciences.
What are the processes that describe the evolution of complex systems? Again we have an
1
From the introduction of Ravndal’s lecture notes on Statistical Physics
1
2 CHAPTER 1. INTRODUCTION
active field with stochastic processes as a key word.
So what has been going on? Why have the mysteries and excitement of statistical physics not
been introduced to you before?
The point is that statistical physics is unfinished, subtle, intellectually and mathematically
demanding. The activity in statistical physics has philosophical implications that have not been
discussed by professional philosophers—and have had little influence on the thinking of the
general public.
DETERMINISTIC behavior rules the day. Privately, on the public arena, in industry, man-
agement and in teaching we almost invariably assume that any given action leads predictably
to the desired result—and we are surprised and discouraged when it does not.
2
The next level
is that some feed-back is considered—but again things get too simplistic. Engineers who study
road traffic reach the conclusion that more roads will make congestions worse, not better—a
statement that cannot be generally true. Similarly, the “Club of Rome” report predicted that we
would run out of resources very quickly. The human population would grow to an extent that
would outstrip food, metals, oil and other resources. However, this did not happen. Now, with
a much large population we have more food, metals, and oil per capita than we had when the
report was written. Of course, statistical physics cannot be applied to all areas. The point is that
in physics, statistical physics provides us with many examples of complex systems and complex
dynamics that we may understand in detail.
Statistical physics provides an intellectual framework and a systematic approach to the study
of real world systems with complicated interactions and feedback mechanisms. I expect that the
new concepts in statistical physics will eventually have a significant impact not only on other
sciences, but also on the public sector to an extent that we may speak of as a paradigm shift.
1.2 Deterministic Chaos
Deterministic chaos appears to be an oxymoron, i. e. a contradiction of terms. It is not. The
concept of deterministic chaos has changed the thinking of anyone who gets some insight into
what it means. Deterministic chaos may arise in any non-linear system with few (active) degrees
of freedom, e. g. electric oscillator circuits, driven pendulums, thermal convection, lasers, and
many other systems. The central theme is that the time evolution of the system is sensitive
so initial conditions. Consider a system that evolves in time as described by the deterministic
equations of motion, a pendulum driven with a small force oscillating at a frequency ω, for
example. Starting from given initial condition the pendulum will evolve in time as determined
by the equation of motion. If we start again with new initial condition, arbitrarily close to the
previous case, the pendulum will again evolve deterministically, but its state as a function of
time will have a distance (in some measure) that diverges exponentially from the first realization.
Since the initial condition cannot be specified with arbitrary precision, the orbit in fact becomes
unpredictable—even in principle.
It is this incredible sensitivity to initial conditions for deterministic systems that has deep
2
Discuss the telephone system at the University of Oslo
1.3. ATOMISTIC PHILOSOPHY 3
philosophical implications. The conclusion, at least how I see it, is that determinism and ran-
domness are just two aspects of same system.
The planetary system is known to be chaotic! That is, we cannot, even in principle, predict
lunar eclipse, the relative positions of planets, the rotation of the earth as a function of time.
What is going on? The point is that non-linear systems that exhibit deterministic chaos have
a time horizon beyond which we cannot predict the time evolution because of the exponential
sensitivity on initial conditions. For the weather we have a time horizon of two weeks in typical
situations. There exist initial conditions that have time horizons further in the future. for the
planetary system the time horizon is millions of years, but chaotic behavior has nevertheless been
observed for the motion of Io, one of Jupiter’s moons. Chaos is also required for an understanding
of the banding of the rings of Jupiter.
3
Very accurate numerical solutions of Newton’s equations
for the solar system also exhibit deterministic chaos.
1.3 Atomistic Philosophy
Now, for systems with many active degrees of freedom, we observe cooperative phenomena, i. e.
behavior that does not depend on the details of the system. We observe universality, or universal
behavior, that cannot be explained by an atomistic understanding alone.
Let me explain: The atomistic philosophy is strong and alive. We believe that with an atom-
istic understanding of the particles that constitute matter—and their interactions—it is merely
an exercise in applied mathematics to predict the macroscopic behavior of systems of practical
importance. This certainly is an arrogant physicists attitude. It must be said, however, that the
experiments and theories are entirely consistent with this view—in spite of deterministic chaos,
self-organized criticality and their ilk.
It is clear that the quantum mechanics required to describe the interactions of atoms in chem-
ical systems is fully understood. We have the exact solution for the Hydrogen atom, the approx-
imate solution for the Helium atom, and numerical solutions for more complex atoms and their
interactions in molecules. There is no evidence that there are open questions in the foundation of
quantum chemistry. Thus we know what is required for a rational understanding of chemistry—
the work has been done, we only need to apply what is known practical tasks in engineering and
industry so to say.
Since we understand chemistry, no new fundamental laws of Nature are required to under-
stand biochemistry, and finally by implication the brain—how we think and understand!
What folly! Clearly something is wrong with this line of reasoning. Molecular biology is
at present the field of science that has the highest rate of discovery, while physicist have been
waiting for almost thirty years for the observation of the Higgs Boson predicted by the standard
model. Elementary particle theory clearly lies at the foundation of physics—but the rate of
discovery has slowed to a glacial pace.
What is going on? I believe the point is that the evolution of science in the 20th century
has lead to a focus on the atomistic part of the atomistic philosophy. Quantum mechanics was a
3
The ring were first observed by Galileo in 1616 with a telescope he had built. By 1655 Huygens had resolved
the banding, not fully explained yet.
4 CHAPTER 1. INTRODUCTION
fantastic break-through, which completely dominated physics in the first half of previous century.
Discoveries, were made at an astounding rate. Suddenly we understood spectra, the black-body
radiation, X-rays, radioactivity, all kinds of new particles, electrons, protons, neutrons, anti-
particles and much more. Much was finished before the second World war, whereas nuclear
reactors, and weapons, were largely developed in 50’s and 60’s. The search for elementary
particles, the constituents of nuclear matter, and their interactions gained momentum in the 50’s
and truly fantastic research organizations of unprecedented size, such as CERN, dedicated to
basic research were set up.
This enthusiasm was appropriate and lead to a long string of discoveries (and Nobel prizes)
that have changed our society in fundamental ways.
Quantum mechanics is, of course, also at the basis of the electronics industry, computers,
communication, lasers and other engineering products that have changed our lives.
However, this huge effort on the atomistic side, left small resources—both human and otherwise—
to other parts of physics. The intellectual resources were largely concentrated on particle physics
research, the other aspects of the atomistic philosophy, namely the understanding of the natu-
ral world in terms of the (newly gained) atomistic physics, was left to an uncoordinated little
publicized and conceptually demanding effort.
1.3.1 Cooperative Phenomena
Cooperative phenomena are ubiquitous and well known from daily life. Take hydrodynamics
for example. Air, water, wine, molten steel, liquid argon — all share the same hydrodynamics.
They flow, fill vessels, they from drops and bubbles, have surface waves, vortices and exhibit
many other hydrodynamic phenomena—in spite of the fact that molten steel, water and argon
are definitely described by quite different Hamilton operators that encode their quantum me-
chanics. How can it be that important macroscopic properties do not depend on the details of the
Hamiltonian?
Phase transitions are also universal: Magnetism and the liquid–vapor transition belong to the
same universality class; in an abstract sense they are the same phase transition! Ferro-electricity,
spinodal decomposition, superconductivity, are all phenomena that do not depend on the details
of their atomistic components and their interactions—they are cooperative phenomena.
The understanding of cooperative phenomena is far from complete. We have no general
theory, except for second order phase transitions, and the phenomenological equations for hy-
drodynamics, to classify and simplify the general scaling and other fractal phenomena observed
in highly non-equilibrium systems. Here we have a very active branch of statistical physics.
1.4 Outlook
So what is the underlying conceptual framework on which physics attempts to fill the gap in un-
derstanding the macroscopic world? Howdo we connect the microscopic with the macroscopic?—
of course, by the scientific approach, by observation, experiment, theory, and now computational
modeling.
1.5. WHAT WILL WE DO? 5
First of all: For macroscopic systems, we have a fantastic phenomenological theory: THER-
MODYNAMICS valid at or near thermal equilibrium, and for systems that start in an equilibrium
state and wind up in an equilibrium state (explosions for example). Thermodynamics was devel-
oped largely in the 19th century, with significant developments in our century related to quantum
mechanics and phase transitions. Thermodynamics in the present form was really formulated as
an axiomatic system with the three laws of thermodynamics. The central concept is energy
4
, and
of course entropy, the only concept required in all the three laws of thermodynamics.
Secondly: Statistical physics started with Daniel Bernoulli (1700-1792), was given a new
start by Rudolf Clausius (1822–1888), James Clerk Maxwell (1831–1879) contributed the kinetic
theory of gases and his famous velocity distribution. Ludwig Boltzmann (1844–1906) made
the fundamental connection to kinetics and introduced the famous expression S = klnW the
statistical expression for the entropy, and explained why entropy must increase by his H-theorem.
Josiah Willard Gibbs (1839–1903), made fundamental contributions to the modern formulation
of statistical mechanics. In this century Lars Onsager (1903–1976) made several outstanding
contributions. By his exact solution of the Ising model, in two-spatial dimensions, he proved that
the framework of statistical physics could indeed tackle the problem of phase transitions. He
received the 1968 Nobel prize in chemistry for his work on irreversible thermodynamics. Claude
E. Shannon initiated information theory by his 1948 paper analyzing measures of information
transmitted through a communication channel. His measure of information is directly related to
Boltzmann’s statistical measure of entropy. The last break-trough contribution was by Kenneth
Geddes Wilson (1936–), who was awarded the 1982 Nobel prize in physics for Renormalization
group theory that allows the calculation of scaling exponents at second order phase transitions.
1.5 What will we do?
This is what now think we should do in this course. We cannot get into all the interesting things
at the frontier of research. On the contrary the purpose of this course is to acquaint you with
the central issues of statistical mechanics. I will be guided by Finn Ravndals notes, and my own
tastes, and hope I do not overload the course. It is better to understand the a number of central
issues thoroughly, they provide anchor points for further investigation if your studies, or later
work require them.
4
Energy, an old word first used by Thomas Young in 1807 to signify mv
2
, not [
1
2
mv
2
] from Greek en in +ergon
work (or rather: at work)
6 CHAPTER 1. INTRODUCTION
Chapter 2
Historical Background
The sensation of hot and cold; that is temperature, and the use of fire in cooking (heat) pre-
dates written history. The scientific understanding of temperature and heat is surprisingly recent.
Emilio Segr` e has written a wonderful account of the evolution of our understanding of heat,
temperature and other aspects of the foundation of modern physics.
The concepts of temperature, heat, work and entropy grew gradually from the time of Galileo
and culminated with the formulation of classical thermodynamics by Rudolf Clausius (1822–
1888). At this stage there were two principles of thermodynamics formulated by Clausius:
I The energy of the universe is constant
II The entropy of the universe tends to a maximum
Clausius introduce the word entropy in 1865,from the Greek τρoπ` η (Verwandlung; transforma-
tion). Entropy, a fundamental concept in thermodynamics, is involved in both principles. Now
entropy is most readily understood in terms of statistical physics. In classical physics there is no
reference point for entropy. Walther Hermann Nernst (1864–1941) formulated the third principle
of thermodynamics:
III The entropy vanishes for a system at zero absolute temperature,
a conclusion Nernst reached by experimental investigations of vapors at high temperature, solids
at low temperature and galvanic cells. He received the 1920 Nobel Prize for Chemistry for his
discovery
Classical mechanics is reversible. For particles that interact with forces that may be derived
from a potential the classical paths are strictly time-reversible, and the total energy of the system
is conserved. This situation is not changed by quantummechanics. The deep problemis therefore
to understand how entropy can always increase as equilibrium is approached.
Interestingly the principle of energy conservation, i. e. the so called first law of thermody-
namics I, was established after Carnot’s discovery of the second law of thermodynamics.
Entropy is a concept of great practical importance, but it is a concept that is difficult to
understand. The following quote from a nice little book: Steam Engine Principlesby Calvert
(1991) illustrates the difficulty:
7
8 CHAPTER 2. HISTORICAL BACKGROUND
A forth property is entropy.
1
This has no physical existence and it has an arbitrary
datum, but its changes can be precisely calculated. Such changes are listed in tables
and plotted on charts. It is unprofitable to find an analogue for entropy. There is
none. It must be accepted and used. It is as much a handicap to try to understand
what is going on in our steam engine without recourse to entropy tables or charts as
it is for a coastwise navigator to reject the tide tables. In both cases the tables are
prepared by the learned and they are intended to be used without question by the
practical man.
Of course, there are objections to Clausius’ formulation of the first law: The energy universe
is not conserved unless we take into account the equivalence of mass and energy as stated in
Einstein’s famous formula E = mc
2
. However, in classical systems that are ‘closed’ energy is
indeed conserved.
There are equivalent formulations of these two ‘laws’ of thermodynamics that give more
insight.
2.1 Temperature
Galileo (1564–1642) around 1592 built a fluid based thermoscope which could be used to show
changes in temperature. However the thermoscope had no fixed point and temperatures form
different thermoscopes could not be compared. Disciples of Galileo and others in the Academia
del Cimento (in Florence) made systematic studies using a thermometer. Isaac Newton (1642–
1727) proposed a temperature scale in which the freezing point of water was set to zero, and the
temperature of the human body to twelve.
Daniel Gabriel Fahrenheit (1686–1736) was born in Danzig but studied in Holland and
worked in Amsterdam. Fahrenheit proposed a temperature scale still used today. He used al-
cohol thermometers and was the first to make a mercury thermometer. Fahrenheit used three
fixed points: (1): 0 degrees for a mixture of ice, water and salt; (2): 32 degrees for a mixture
of ice and water; (3): 96 degrees at the body temperature of a ‘healthy’ person in the mouth or
under the arm. This temperature scale has the advantage that most temperatures experienced in
daily life are all positive.
Fahrenheit was the first to observe and describe super-cooling of water in small capillaries.
He used rain-water in a 1 inch bulb that he evacuated in the manner used by him to make ther-
mometers. Fahrenheit found that the water was liquid at 15 degrees, i. e. -9.4

C, and solidified
immediately if he broke the tip of the capillary attached to the bulb to admit air.
We now mostly use the Celsius temperature scale. Anders Celsius (1701–1744) was a pro-
fessor of astronomy at the university of Uppsala. In 1742 he described his thermometer to the
Swedish Academy of Sciences. The thermometer had the melting point of snow as one fixed
point at 100

C and the boiling point of water was set to 0

C. This was later inverted so that
0

is the melting point of snow, and 100

as the other, for the centigrade scale. The name was
changed in 1948 to honor Celsius.
1
the others are: pressure, temperature, and density
2.2. STEAM ENGINES 9
William Thomson (later Lord Kelvin) introduce the absolute temperature scale. (More later)
2.2 Steam Engines
The steam engine was developed and played an important role in the industrial revolution, start-
ing in England in the mid-18th century. The name we normally associate with the steam engine
is James Watt (1736–1819) who introduced the condenser, patented in 1768: “A New Invented
Method of Lessening the Consumption of Steam and Fuel in Fire Engines.”
2
Watts invention
greatly increased the efficiency
3
of the steam engine patented in 1705 by Thomas Newcomen
(1663–1729) used to drive pumps in coal mines (see Fig. 2.1.) The steam engine was further de-
Figure 2.1: The Newcomen steam engine worked
in a simple fashion. Steam generated in the boiler is
admitted to the cylinder by opening the valve (a) and
the weight of the other end of the arm pull the pis-
ton upward. The cylinder volume is filled and heated
by the steam at atmospheric pressure. By closing the
valve (a) and opening the valve (b) cold water is in-
troduced into the cylinder. The steam condenses and
cools thereby reducing the pressure in the cylinder
so that the piston moves down pushed by the atmo-
spheric pressure. The excess water runs out of the
tube (f). When the cylinder has cooled down the cy-
cle can start again. The maximumpressure difference
that can be obtained is then given by the length of the
tube (f) and the temperature of the surroundings.
veloped in the coming years by many engineers without a deeper understanding of the efficiency
of the engine. Watt’s invention alone significantly reduced the amount of coal needed to perform
a given amount of work, and thus made steam engines practical. There were dangers, of course.
For instance, in 1877 there were in Germany 20 explosions of steam boilers injuring 58 persons.
In 1890 there were 14 such explosions (of a vastly larger number of boilers) injuring 18 persons.
Stationary steam engines were used in industry, for railroads, in agriculture and later for electric
power generation. The table shows that there was a dramatic increase in the use of steam engines
only 120 years ago—now they are almost extinct. Steam turbines have replaced steam engines
completely for electric power generation.
2
‘Watt, James’ Britannica Online. <http://www.eb.com:180/cgi-bin/g?DocF=micro/632/91.html>
3
from 1% to 2%
10 CHAPTER 2. HISTORICAL BACKGROUND
Figure 2.2: The steam was generated at atmospheric
pressure and introduced into the cylinder just as in
the Newcomen engine. However, in contrast with
the Newcomen engine the steam was condensed in
the small cylinder (K) immersed in cold water. The
machine shown was double-acting,i. e. it worked
both on the down- and the up-stroke. various arrange-
ments opened and closed valves automatically. The
water in the condenser was pumped out. The central
improvement was that there was no need to cool the
cylinder between strokes.
1879 1879
Type of Steam engine Number Horse power Number Horse power
Free standing 29 895 887 780 45 192 1 508 195
Mobile 5 442 47 104 11 916 111 070
Ship 623 50 309 1 674 154 189
Table 2.1: Steam engines in Prussia. Globally there were about 1.9 million steam engines in the
world around 1890.
2.3 Sadi Carnot
Thermodynamics as a field of science started with Sadi Carnot (1796–1831). He published in
1824 remarkable bookR´ eflexions sur la Puissance Motrice du Feu, et sur les Machines Propres
a d´ evelopper cette Puissance—his only published scientific work. But hardly anyone bought the
book, and it became very difficult to obtain only a few years later. It was
´
Emile Clapeyron (1799–
1864), who understood Carnot’s work and expressed the results in a more mathematical form in
the paper published in 1834. Fourteen years later Clayperon’s work was expanded by William
Thomson
4
(1824–1907). Thomson studied Carnot’s paper in 1847 and realized that the Carnot
cycle could be used to define an absolute temperature scale. However, with Carnot, he as late as
in 1848 thought that heat was conserved, which is incorrect. This view was, however, changed
by James Prescott Joule (1818–1889), and together they established the correct interpretation and
thereby the principle of conservation of energy.
4
Thomson was knighted in 1866, raised to the peerage in 1892 as Baron Kelvin of Largs.
2.3. SADI CARNOT 11
The beginning of Sadi Carnot’s book gives a feel for the technological optimism at the be-
ginning of the 19th century, and the motivation of his work:
EVERY one knows that heat can produce motion. That it possesses vast motive-
power no one can doubt, in these days when the steam-engine is everywhere so well
known.
To heat also are due the vast movements which take place on the earth. It causes
the agitations of the atmosphere, the ascension of clouds, the fall of rain and of
meteors,
5
the currents of water which channel the surface of the globe, and of which
man has thus far employed but a small portion. Even earthquakes and volcanic
eruptions are the result of heat.
From this immense reservoir we may draw the moving force necessary for our
purposes. Nature, in providing us with combustibles on all sides, has given us the
power to produce, at all times and in all places, heat and the impelling power which
is the result of it. To develop this power, to appropriate it to our uses, is the object of
heat-engines.
The study of these engines is of the greatest interest, their importance is enor-
mous, their use is continually increasing, and they seem destined to produce a great
revolution in the civilized world.
Already the steam-engine works our mines, impels our ships, excavates our ports
and our rivers, forges iron, fashions wood, grinds grains, spins and weaves our cloths,
transports the heaviest burdens, etc. It appears that it must some day serve as a
universal motor, and be substituted for animal power, waterfalls, and air currents.
After a discussion of the importance of steam-engines for England, he notes:
The discovery of the steam-engine owed its birth, like most human inventions, to
rude attempts which have been attributed to different persons, while the real author
is not certainly known. It is, however, less in the first attempts that the principal
discovery consists, than in the successive improvements which have brought steam-
engines to the conditions in which we find them today. There is almost as great
a distance between the first apparatus in which the expansive force of steam was
displayed and the existing machine, as between the first raft that man ever made and
the modern vessel.
Here Carnot expresses, in an amicable fashion, the unreasonable effectiveness of engineering.
The path for the original invention to the final product of generations of engineers, through a
never ending series of improvements leads to qualitatively new products.
After this motivation of his work he approaches the more general scientific question:
Notwithstanding the work of all kinds done by steam-engines, notwithstanding the
satisfactory condition to which they have been brought today, their theory is very lit-
tle understood, and the attempts to improve them are still directed almost by chance.
5
This is a misunderstanding, the movement of meteors has nothing to do with heat. When they enter the atmo-
sphere, friction from the air cause them to heat and glow.
12 CHAPTER 2. HISTORICAL BACKGROUND
The question has often been raised whether the motive power of heat
6
is un-
bounded, whether the possible improvements in steam engines have an assignable
limit—a limit which the nature of things will not allow to be passed by any means
whatever; or whether, on the contrary, these improvements may be carried on indefi-
nitely. We have long sought, and are seeking today, to ascertain whether there are in
existence agents preferable to the vapor of water for developing the motive power of
heat; whether atmospheric air, for example, would not present in this respect great
advantages. We propose now to submit these questions to a deliberate examination.
Then he really becomes precise and defines the scientific question, which he the actually
answers:
The phenomenon of the production of motion by heat has not been considered from
a sufficiently general point of view. We have considered it only in machines the
nature and mode of action of which have not allowed us to take in the whole extent
of application of which it is susceptible. In such machines the phenomenon is, in a
way, incomplete. It becomes difficult to recognize its principles and study its laws.
In order to consider in the most general way the principle of the production
of motion by heat, it must be considered independently of any mechanism or any
particular agent. It is necessary to establish principles applicable not only to steam-
engines
7
but to all imaginable heat-engines, whatever the working substance and
whatever the method by which it is operated.
Carnot goes on to describe the workings of the steam-engine, and concludes:
The production of motive power is then due in steam-engines not to an actual con-
sumption of caloric, but to its transportation from a warm body to a cold body, that
is, to its re-establishment of equilibrium an equilibrium considered as destroyed by
any cause whatever, by chemical action such as combustion, or by any other. We
shall see shortly that this principle is applicable to any machine set in motion by
heat.
. . .
We have shown that in steam-engines the motive-power is due to a re-establish-
ment of equilibrium in the caloric; this takes place not only for steam-engines, but
also for every heat-engine—that is, for every machine of which caloric is the motor.
Heat can evidently be a cause of motion only by virtue of the changes of volume or
of form which it produces in bodies.
6
We use here the expression motive power to express the useful effect that a motor is capable of producing.
This effect can always be likened to the elevation of a weight to a certain height. It has, as we know, as a measure,
the product of the weight multiplied by the height to which it is raised. [Here, in a footnote, Carnot gives a clear
definition of what we today call free energy, even though his concept of heat (or caloric) was inadequate]
7
We distinguish here the steam-engine from the heat-engine in general. The latter may make use of any agent
whatever, of the vapor of water or of any other to develop the motive power of heat.
2.3. SADI CARNOT 13
Here, Carnot has unclear concepts about the role of heat, or caloric. in a footnote (see below) he
evades the question. However, he is clear in the sense that a temperature difference is required if
work is to be generated. In that process there is caloric or heat flowing from the hot to the cold,
trying to re-establish thermal equilibrium.
Then Carnot rephrases the problem he is considering:
It is natural to ask here this curious and important question: Is the motive power of
heat invariable in quantity, or does it vary with the agent employed to realize it as
the intermediary substance, selected as the subject of action of the heat?
It is clear that this question can be asked only in regard to a given quantity of
caloric,
8
the difference of the temperatures also being given. We take, for example,
one body A kept at a temperature of 100

and another body B kept at a temperature
of 0

, and ask what quantity of motive power can be produced by the passage of a
given portion of caloric (for example, as much as is necessary to melt a kilogram of
ice) from the first of these bodies to the second. We inquire whether this quantity of
motive power is necessarily limited, whether it varies with the substance employed
to realize it, whether the vapor of water offers in this respect more or less advantage
than the vapor of alcohol, of mercury, a permanent gas, or any other substance. We
will try to answer these questions, availing ourselves of ideas already established.
We have already remarked upon this self-evident fact, or fact which at least ap-
pears evident as soon as we reflect on the changes of volume occasioned by heat:
wherever there exists a difference of temperature, motive power can be produced.
Here he announced the basic principle: A temperature difference is required in order to
generate work. Then he presents the first sketch of the cycle:
Imagine two bodies A and B, kept each at a constant temperature, that of A being
higher than that of B. These two bodies, to which we can give or from which we can
remove the heat without causing their temperatures to vary, exercise the functions of
two unlimited reservoirs of caloric. We will call the first the furnace and the second
the refrigerator.
If we wish to produce motive power by carrying a certain quantity of heat from
the body A to the body B we shall proceed as follows:
(1) To borrow caloric from the body A to make steam with it—that is, to make
this body fulfill the function of a furnace, or rather of the metal composing the
boiler in ordinary engines—we here assume that the steam is produced at the
same temperature as the body A.
(2) The steam having been received in a space capable of expansion, such as a
cylinder furnished with a piston, to increase the volume of this space, and
8
It is considered unnecessary to explain here what is quantity of caloric or quantity of heat (for we employ these
two expressions indifferently), or to describe how we measure these quantities by the calorimeter. Nor will we
explain what is meant by latent heat, degree of temperature, specific heat, etc. The reader should be familiarized
with these terms through the study of the elementary treatises of physics or of chemistry.
14 CHAPTER 2. HISTORICAL BACKGROUND
consequently also that of the steam. Thus rarefied, the temperature will fall
spontaneously, as occurs with all elastic fluids; admit that the rarefaction may
be continued to the point where the temperature becomes precisely that of the
body B.
(3) To condense the steam by putting it in contact with the body B, and at the
same time exerting on it a constant pressure until it is entirely liquefied. The
body B fills here the place of the injectionwater in ordinary engines, with this
difference, that it condenses the vapor without mingling with it, and without
changing its own temperature.
Here, Carnot introduced several of the basic concepts of Thermodynamics: heat reservoir and
thermodynamics process. In (1) and (2) he describes an isothermal process, whereas in (2) he
describes and isentropic process. Reversibility reversibility is the next concept:
The operations which we have just described might have been performed in an in-
verse direction and order. There is nothing to prevent forming vapor with the caloric
of the body B, and at the temperature of that body, compressing it in such a way as
to make it acquire the temperature of the body A, finally condensing it by contact
with this latter body, and continuing the compression to complete liquefaction.
This then completes the Carnot cycle. However, Carnot continues in a fashion that is difficult to
understand, and in contradiction with what he writes later. First the continuation:
By our first operations there would have been at the same time production of motive
power and transfer of caloric from the body A to the body B. By the inverse opera-
tions there is at the same time expenditure of motive power and return of caloric from
the body B to the body A. But if we have acted in each case on the same quantity of
vapor, if there is produced no loss either of motive power or caloric, the quantity of
motive power produced in the first place will be equal to that which would have been
expended in the second, and the quantity of caloric passed in the first case from the
body A to the body B would be equal to the quantity which passes back again in the
second from the body B to the body A; so that an indefinite number of alternative
operations of this sort could be carried on without in the end having either produced
motive power or transferred caloric from one body to the other.
Here, by the assumption that caloric is conserved, he also gets no work done by his cycle! Nev-
ertheless he an-ounces his second principle:
Now if there existed any means of using heat preferable to those which we have
employed, that is, if it were possible by any method whatever to make the caloric
produce a quantity of motive power greater than we have made it produce by our
first series of operations, it would suffice to divert a portion of this power in order
by the method just indicated to make the caloric of the body B return to the body
A from the refrigerator to the furnace, to restore the initial conditions, and thus to
be ready to commence again an operation precisely similar to the former, and so
2.3. SADI CARNOT 15
on: this would be not only perpetual motion, but an unlimited creation of motive
power without consumption either of caloric or of any other agent whatever. Such
a creation is entirely contrary to ideas now accepted, to the laws of mechanics and
of sound physics. It is inadmissible. We should then conclude that the maximum of
motive power resulting from the employment of steam is also the maximum of motive
power realizable by any means whatever.
. . .
Since every re-establishment of equilibrium in the caloric may be the cause of
the production of motive power, every re-establishment of equilibrium which shall
be accomplished without production of this power should be considered as an actual
loss. Now, very little reflection would show that all change of temperature which is
not due to a change of volume of the bodies can be only a useless re-establishment
of equilibrium in the caloric. The necessary condition of the maximum is, then, that
in the bodies employed to realize the motive power of heat there should not occur
any change of temperature which may not be due to a change of volume.
Here he has two more fundamental concepts: There is am maximum work attainable that is in-
dependent of what type of heat-engine is employed. Secondly he notes that any heat conduction
results in a loss of work as compared with the work produced by an ideal heat-engine. The prob-
lem is to understands how he reaches these correct conclusions, based on a cycle that produces
neither work nor a transport of caloric (heat) from the hot to the cold reservoir.
Let us follow his argument for the second demonstration: First the basic assumption:
When a gaseous fluid is rapidly compressed its temperature rises. It falls, on the
contrary, when it is rapidly dilated. This is one of the facts best demonstrated by
experiment. We will take it for the basis of our demonstration.
Then he goes on to describe in detail what we now call the Carnot cycle:
This preliminary idea being established, let us imagine an elastic fluid, atmospheric
air for example, shut up in a cylindrical vessel, abcd (Fig. 1), provided with a mov-
able diaphragm or piston, cd. Let there be also two bodies, A and B, kept each
at a constant temperature, that of A being higher than that of B. Let us picture to
ourselves now the series of operations which are to be described, Fig. 2.3
This is the complete Carnot cycle. Carnot continues to explain that work is done:
In these various operations the piston is subject to an effort of greater or less magni-
tude, exerted by the air enclosed in the cylinder; the elastic force of this air varies as
much by reason of the changes in volume as of changes of temperature. But it should
be remarked that with equal volumes, that is, for the similar positions of the piston,
the temperature is higher during the movements of dilatation than during the move-
ments of compression. During the former the elastic force of the air is found to be
greater, and consequently the quantity of motive power produced by the movements
of dilatation is more considerable than that consumed to produce the movements of
16 CHAPTER 2. HISTORICAL BACKGROUND
Figure 2.3: (1) Contact of the body A with the air enclosed in
the space abcd or with the wall of this space—a wall that we will
suppose to transmit the caloric readily. The air becomes by such
contact of the same temperature as the body A; cd is the actual
position of the piston.
(2) The piston gradually rises and takes the position ef. The body
A is all the time in contact with the air, which is thus kept at
a constant temperature during the rarefaction. The body A fur-
nishes the caloric necessary to keep the temperature constant.
(3) The body Ais removed, and the air is then no longer in contact
with any body capable of furnishing it with caloric. The piston
meanwhile continues to move, and passes from the position ef to
the position gh. The air is rarefied without receiving caloric, and
its temperature falls. Let us imagine that it falls thus till it be-
comes equal to that of the body B; at this instant the piston stops,
remaining at the position gh.
(4) The air is placed in contact with the body B; it is compressed
by the return of the piston as it is moved from the position gh to
the position cd . This air remains, however, at a constant temper-
ature because of its contact with the body B, to which it yields its
caloric.
(5) The body B is removed, and the compression of the air is
continued, which being then isolated, its temperature rises. The
compression is continued till the air acquires the temperature of
the body A. The piston passes during this time from the position
cd to the position ik.(6) The air is again placed in contact with the
body A. The piston returns from the position ik to the position ef;
the temperature remains unchanged.
(7) The step described under number (3) is renewed, then succes-
sively the steps (4), (5), (6), (3), (4), (5), (6), (3), (4), (5); and so
on.
2.3. SADI CARNOT 17
compression. Thus we should obtain an excess of motive power—an excess which
we could employ for any purpose whatever. The air, then, has served as a heat-
engine; we have, in fact, employed it in the most advantageous manner possible, for
no useless re-establishment of equilibrium has been effected in the caloric.
Then he points out that the cycle may be run in reverse and thereby by the use of mechanical
work take heat from the lower temperature of B to the higher temperature of the Reservoir A,
and he continues:
The result of these first operations has been the production of a certain quantity
of motive power and the removal of caloric from the body A to the body B. The
result of the inverse operations is the consumption of the motive power produced
and the return of the caloric from the body B to the body A; so that these two series
of operations annul each other, after a fashion, one neutralizing the other.
The impossibility of making the caloric produce a greater quantity of motive
power than that which we obtained from it by our first series of operations, is now
easily proved. It is demonstrated by reasoning very similar to that employed at
page 8; the reasoning will here be even more exact. The air which we have used to
develop the motive power is restored at the end of each cycle of operations exactly
to the state in which it was at first found, while, as we have already remarked, this
would not be precisely the case with the vapor of water.
9
We have chosen atmospheric air as the instrument which should develop the
motive power of heat, but it is evident that the reasoning would have been the same
for all other gaseous substances, and even for all other bodies susceptible of change
of temperature through successive contractions and dilatations, which comprehends
all natural substances, or at least all those which are adapted to realize the motive
power of heat. Thus we are led to establish this general proposition:
The motive power of heat is independent of the agents employed to realize it; its
quantity is fixed solely by the temperatures of the bodies between which is ejected,
finally, the transfer of the caloric.
We must understand here that each of the methods of developing motive power
attains the perfection of which it is susceptible. This condition is found to be ful-
filled if, as we remarked above, there is produced in the body no other change of
temperature than that due to change of volume, or, what is the same thing in other
words, if there is no contact between bodies of sensibly different temperatures.
9
We tacitly assume in our demonstration, that when a body has experienced any changes, and when after a certain
number of transformations it returns to precisely its original state, that is, to that state considered in respect to density,
to temperature, to mode of aggregation-let us suppose, I say, that this body is found to contain the same quantity of
heat that it contained at first, or else that the quantities of heat absorbed or set free in these different transformations
are exactly compensated. This fact has never been called in question. It was first admitted without reflection, and
verified afterwards in many cases by experiments with the calorimeter. To deny it would be to overthrow the whole
theory of heat to which it serves as a basis. For the rest, we may say in passing, the main principles on which the
theory of heat rests require the most careful examination. Many experimental facts appear almost inexplicable in
the present state of this theory.
18 CHAPTER 2. HISTORICAL BACKGROUND
Here we are the final conclusion. The amount of work that may be extracted by a heat-engine,
of any design whatever, depends only on the temperatures of the hot and the cold reservoir, and
on the amount of heat transported. This conclusion is correct. However, the view that the caloric
is conserved is clearly incorrect. The point is that the conservation of energy was only established
much later. Carnot was clearly aware that his idea of heat was not clear as may be seen from his
remarks:
According to established principles at the present time, we can compare with suffi-
cient accuracy the motive power of heat to that of a waterfall. Each has a maximum
that we cannot exceed, whatever may be, on the one hand, the machine which is acted
upon by the water, and whatever, on the other hand, the substance acted upon by the
heat. The motive power of a waterfall depends on its height and on the quantity of
the liquid; the motive power of heat depends also on the quantity of caloric used,
and on what may be termed, on what in fact we will call, the height of its fall,
10
that
is to say, the difference of temperature of the bodies between which the exchange
of caloric is made. In the waterfall the motive power is exactly proportional to the
difference of level between the higher and lower reservoirs. In the fall of caloric
the motive power undoubtedly increases with the difference of temperature between
the warm and the cold bodies; but we do not know whether it is proportional to this
difference. We do not know, for example, whether the fall of caloric from 100 to 50
degrees furnishes more or less motive power than the fall of this same caloric from
50 to zero. It is a question which we propose to examine hereafter.
Carnot thought that caloric was a conserved quantity—an attitude consistent with the exper-
imental accuracy of his day. The correct interpretation was first given by Rudolf Clausius in
1850,with the formulation:
dass in allen F¨ allen, wo durch W¨ arme Arbeit entstehe, eine der erzeugten Arbeit
proportionale W¨ armemenge verbraucht werde, und dass umgekehrt duch Verbrauch
einer ebenso grossen Arbeit dieselbe W¨ armemenge erzeugt werden k¨ onne.
2.4 The Maxwell’s Distribution
Maxwell
11
read a paper at the meeting of the British Association of Aberdeen in September
1859. It was published in the Philosophical Magazine in 1860, and it contains the following
10
The matter here dealt with being entirely new, we are obliged to employ expressions not in use as yet, and which
perhaps are less clear than is desirable.
11
Maxwell, James Clerk; born June 13, 1831, Edinburgh, Scotland; died November 5, 1879, Cambridge, Cam-
bridgeshire, England; Scottish physicist best known for his formulation of electromagnetic theory. He is regarded by
most modern physicists as the scientist of the 19th century who had the greatest influence on 20th-century physics,
and he is ranked with Sir Isaac Newton and Albert Einstein for the fundamental nature of his contributions. In 1931,
on the 100th anniversary of Maxwell’s birth, Einstein described the change in the conception of reality in physics
that resulted from Maxwell’s work as ”the most profound and the most fruitful that physics has experienced since
the time of Newton.”
2.4. THE MAXWELL’S DISTRIBUTION 19
heuristic argument on the velocity distribution.
12
If a great many equal spherical particles were in motion in a perfectly elastic
vessel, collisions would take place among the particles, and their velocities would
be altered at every collision; so that after a certain time the vis viva will be divided
among the particles according to some regular law, the average number of particles
whose velocity lies between certain limits being ascertainable though the velocity of
each particle changes at every collision.
Prop. IV. To find the average number of particles whose velocities lie between
given limits, after a great number of collisions among a great number of equal parti-
cles.
Let N be the whole number of particles. Let x, y, z be the components of the ve-
locity of each particle in three rectangular directions, and let the number of particles
for which x lies between x and x +dx, be N f (x)dx, where f (x) is a function of x to
be determined.
The number of particles for which y lies between y and y +dy will be N f (y)dy;
and the number for which z lies between z and z +dz will be N f (z)dz where f
always stands for the same function.
Now the existence of the velocity x does not in any way affect that of the veloc-
ities y or z, since these are all at right angles to each other and independent, so that
the number of particles whose velocity lies between x and x +dx, and also between
y and y +dy, and also between z and z +dz, is
N f (x) f (y) f (z)dxdydz .
If we suppose the N particles to start from the origin at the same instant, then
this will be the number in the element of volume (dxdydz) after unit of time, and the
number referred to unit of volume will be
N f (x) f (y) f (z) .
But the directions of the coordinates are perfectly arbitrary, and therefore this
number must depend on the distance from the origin alone, that is
f (x) f (y) f (z) =φ(x
2
+y
2
+z
2
) .
Solving this functional equation, we find
f (x) =Ce
Ax
2
, φ(r
2
) =C
3
e
Ar
2
.
If we make A positive, the number of particles will increase with the velocity,
and we should find the whole number of particles infinite. We therefore make A
negative and equal to −
1
α
2
, so that the number between x and x +dx is
12
Appendix 10 from Emilio Segr` e’s book Maxwell’s Distribution of Velocities of Molecules in His Own Words
20 CHAPTER 2. HISTORICAL BACKGROUND
NCe
−x
2

2
dx
Integrating from x =−∞ to x = +∞, we find the whole number of particles,
NC

πα = N , therefore C =
1
α

π
,
f (x) is therefore
1
α

π
e
−x
2

2
Whence we may draw the following conclusions:—
1st. The number of particles whose velocity, resolved in a certain direction, lies
between x and x +dx is
N
1
α

π
e
−x
2

2
dx
2nd. The number whose actual velocity lies between v and v +dv is
N
1
α
3

π
v
2
e
−v
2

2
dv
3rd. To find the mean value of v, add the velocities of all the particles together
and divide by the number of particles; the result is
mean velocity =


π
4th. To find the mean value of v
2
, add all the values together and divide by N,
mean value of v
2
=
3
2
α
2
.
This is greater than the square of the mean velocity, as it ought to be.
Chapter 3
Basic principles of statistical mechanics
3.1 Microscopic states
Classical phase space: Classical phase space is 2 f -dimensional space
(q
1
, q
2
, . . . , q
f
, p
1
, p
2
, . . . , p
f
).
Here q
i
are positional coordinates, p
i
are momenta, f = 3N for n independent particles in 3-
dimensional space. Each point corresponds to a microscopic state of the system. The simplest
Hamiltonian of such a system is
H =
f

j
p
2
j
2m
+E(q
1
, q
2
, . . . , q
f
). (3.1)
Motion of the system is determined by the canonical equation of motion,
˙ p
j
=−
∂H
∂q
j
, ˙ q
j
=
∂H
∂p
j
, ( j = 1, 2, . . . , f ). (3.2)
The motion of the phase point P
t
defines the state at time t. The trajectory of the phase space is
called the phase trajectory. For a given initial conditions at some t −t
0
, q
0
≡¦q
j0
¦, p
0
≡¦p
j0
¦,
the phase trajectory is given by the functions q
i
(t, q
0
, p
0
), p
i
(t, q
0
, p
0
).
For a conservative system energy is conserved,
H(q, p) = E . (3.3)
Thus the orbit belongs to the surface of constant energy.
Exercise: Analyse phase space for a one-dimensional harmonic oscillator and find the de-
pendence of its energy on the amplitude..
Quantum states: Quantum states ψ
l
are defined by the Schr¨ odinder equation ,
H ψ
l
= E
l
ψ
l
(l = 1, 2, . . .). (3.4)
They are determined by the quantum numbers. We will discuss quantum systems later.
21
22 CHAPTER 3. PRINCIPLES OF STATISTICAL MECHANICS
3.2 Statistical treatment
Fundamental assumption: Let A be a physical quantity dependent on the microscopic state.
In classical mechanics,
A(q, p) = A(P).
In quantum mechanics,
A
l
=

ψ

l

l
(dq) ≡'l[A[l`.
Here (dq) ≡ ∏
f
j=1
dq
j
. The observed value (in a macroscopic sense), A
obs
must be a certain
average of microscopic A,
A
obs
=
¯
A.
Realization probability of a microscopic state: Let M be the set of microscopic states which
can be realized under a certain microscopic condition. The realization probability that one of the
microscopic states belongs to the element ∆Γ of the phase space is defined as

∆Γ
f (P)dΓ, (∆Γ ∈ M).
The probability that the quantum state l is realized is defined as
f (l), (l ∈ M).
f is usually called the distribution functions.
The observed quantity is given by the formulas
A
obs
=
¯
A =

M
A(P) f (P)dΓ,
¯
A =

M
A
l
f (l). (3.5)
Statistical ensembles: great number of systems each of which has the same structure as the
system under consideration. It is assumed that
A
obs
= ensemble average of A =
¯
A.
Example – ideal gas: The phase space consists of spatial coordinates r and moments p. In
thermal equilibrium, at high temperature, and for a dilute gas, the Maxwell distribution reads:
f (p) = (2πmk
B
T)
−1/2
exp
¸

p
2
2mk
B
T

, (3.6)
where T is the absolute temperatures, m the mass of a molecule, and k
B
is the Boltzmann con-
stant.
3.2. STATISTICAL TREATMENT 23
General properties of statistical distributions
Statistical independence: For two macroscopic and non-interacting subsystems,
f
12
dp
(12)
dq
(12)
= f
1
dp
(1)
dq
(1)
f
2
dp
(2)
dq
(2)
→ f
12
= f
1
f
2
.
Statistical independence means
¯
A
12
=
¯
A
1
¯
A
2
,
where A is any physical quantity. Thus, for any additive quantity,
¯
A =

j
¯
A
j
.
If one defines the fluctuation, ∆A = A−
¯
A, then
'(∆A)
2
` =

N

j
∆A
j

2
¸
=
N

j
'(∆A
j
)
2
`.
As a result,
'(∆A)
2
`
1/2
¯
A

1

N
<1.
Liouville theorem: Consider a set of instances, t
1
, t
2
, . . . ,. The states at that instances are rep-
resented by the points P
1
, P
2
, . . . ,. The Liouville theorem states that the distribution function
f (p, q) is time-independent. Indeed, in the absence of external forces, the state must be station-
ary, and for the phase-space points one has
div( f v) = 0 →
f

i=1
¸

∂q
i
( f ˙ q
i
) +

∂p
i
( f ˙ p
i
)

= 0.
Expanding derivatives we get
f

i=1
¸
˙ q
i
∂ f
∂q
i
+ ˙ p
i
∂ f
∂p
i

+ f
f

i=1
¸
∂ ˙ q
i
∂q
i
+
∂ ˙ p
i
∂p
i

= 0.
According to the Hamilton equation (3.2),
∂ ˙ q
i
∂q
i
+
∂ ˙ p
i
∂p
i
= 0.
Thus
d f
dt
=
f

i=1
¸
˙ q
i
∂ f
∂q
i
+ ˙ p
i
∂ f
∂p
i

= 0. (3.7)
According to the Hamilton equation (3.2) the sum can be rewritten as
f

i=1
¸
˙ q
i
∂ f
∂q
i
+ ˙ p
i
∂ f
∂p
i

=
f

i=1
¸
∂ f
∂q
i
∂H
∂p
i

∂ f
∂p
i
∂H
∂q
i

≡¦ f , H¦. (3.8)
This combination is called the Poisson bracket.
24 CHAPTER 3. PRINCIPLES OF STATISTICAL MECHANICS
Role of integrals of motion Since log f (p.q) must be (i) stationary, and (ii) additive for macro-
scopic subsystems of the system,
log f
12
= log f
1
+log f
2
,
it depends only on additive integrals of motion, E(p.q), total momentum P(p, q) and angular
momentum M(p.q). The only additive function is the linear combination,
log f (p, q) =α+βE(p, q) +γ P(p, q) +

δ M(p, q). (3.9)
Here in the equilibrium the coefficients β, γ and

δ must be the same for all subsystems in the
closed system. The coefficient α is defined from the normalization condition. Thus, the values
of the integrals of motion completely define statistical properties of the closed system. In the
following we assume that the system is enclosed in a rigid box, and P = 0, M = 0.
3.3 The principle of equal weight and the microcanonical en-
semble
An isolated system after some time reaches the equilibrium. The energy of the system, E, is fixed
with some allowance, δE. In a classical system the set M(E, δE) is the shell-like subspace of
the phase space between the constant-energy surfaces for H = E and H = E +δE.
For quantum mechanics, it is a set of quantum states having energies in the interval E < E
l
<
E +δE.
The principle of equal weight: In a thermal equilibrium state of an isolated system, each of
the microscopic states belonging to M(E, E +δE) is realized with equal probability:
f (P) = constant =
¸

E<H<E+δE

−1
, P ∈ M(E, δE),
f (l) = constant =
¸

E<E
l
<E+δE
1
¸
−1
, l ∈ M(E, δE). (3.10)
This distribution is called the microcanonical one, the corresponding ensemble being called the
microcanonical.
1
The microcanonical ensemble represents an isolated system which has reached thermal
equilibrium.
1
Some other additive integrals of motion besides the energy can be important.
3.3. EQUAL WEIGHT, . . . MICROCANONICAL ENSEMBLE 25
Classical limit (δE →0): One can take a set σ(E) on the surface of constant energy. Then All
the phase trajectories are concentrated at the surface
H(p, q) = E ,
and the distribution function can be written as
f (P) = const δ[H(p, q) −E] . (3.11)
To determine the constant we have to calculate the integral
D(E) =

dΓδ[H(p, q) −E] (3.12)
which is called the density of states, see below. It is very convenient to use the finite-width layer
in the phase space and to transform variables from ¦p, q¦ to the element of surface σ defined
by the equality H(p, q) −E, and to the coordinate n along the normal to this surface. Then the
energy is dependent only on the coordinate n, and
dE =
∂H
∂n
dn =[∇H[ dn.
Here
[∇H[ =
¸

j

∂H
∂p
j

2
+

∂H
∂q
j

2
¸¸
1/2
. (3.13)
Consequently,
dΓδ[H(p, q) −E] =
dσdn
[∇H[
δ(n).
Then one can perform integral over n and introduce the distribution over the constant energy
surface,
f (P)dσ =
1
D(E)

[∇H[
, D(E) =
¸

H=E

[∇H[

. (3.14)
An average is then given by the equation,
¯
A =
1
D(E)

H=E
A(P)dσ
[∇H[
. (3.15)
Ergodic theorem: Originally, A
obs
is assumed to be time average of A(p
t
). The ergodic theo-
rem can be formulated as
A
time average
= A
phase average
.
Actually one can deduce the principle of equivalent weight from this theorem.
26 CHAPTER 3. PRINCIPLES OF STATISTICAL MECHANICS
The finite allowance of the energy δE: In quantum mechanics, the energy of a system has
uncertainty,
(δE)
qu
∼ ¯ h/t ,
where t is the length of the observation. The statistical treatment is still possible only if the set
M

E, δE ≥(δE)
qu

contains many quantum states.
If the system occupies only a single quantum state, statistics is not applicable.
3.4 The thermodynamic weight of a macroscopic state and en-
tropy
Variables defining a macroscopic state: One can choose the energy E (with allowance δE),
the numbers N
A
, N
B
, . . . of particles of various kinds, the volume V of the box which contains the
system, and other parameters x, . . . which specify external forces. Thus the Hamiltonian depends
on those parameters, H =H[E(δE), N
A
, N
B
, . . . ,V, x, . . .]. In a classical system one can tend the
allowance δE to 0. It is effectively true also for a quantum system provided the measuring time
is long enough. Then the following definition do not depend on δE.
Thermodynamic weight: In quantummechanics, the total number of quantumstates, W(E, δE, N,V, x)
of the possible quantum states for the set of prescribed values
E(δE), N
A
, N
B
, . . . ,V, x, . . .
is called the thermodynamic weight of that particular macroscopic state. Namely
W(E, δE, N,V, x) ≡

E<E
l
(N,V,x)<E+δE
1. (3.16)
In classical mechanics it is defined as a limit of quantum statistical mechanics
W(E, δE, N,V, x) ≡

E<H(N,V,x)<E+δE

h
3(N
A
+N
B
+...)
N
A
!N
B
! . . .
. (3.17)
Here h is the Planck constant.
Statistical definition of entropy:
S(E, N
A
, N
B
, . . . ,V, x, . . .) = k
B
logW(E, δE, N,V, x). (3.18)
This relation is called the Boltzmann relation while the constant
k
B
= 1.3810
−23
J/K (3.19)
is called the Boltzmann constant. To convince yourself that the statistical definition of entropy
(3.18) is compatible with thermodynamics one has to check thermodynamic relations.
2
2
Note that allowance δE does not affect the value of the entropy
3.5. NUMBER OF STATES AND DENSITY OF STATES 27
3.5 Number of states and density of states
Let us chose the energies to be non-negative,
0 ≤E
1
≤E
2
≤. . . .
The number of states between 0 and E is called the number of states of the system:
Γ(E, N,V, x) ≡

0<E
l
≤E
1. (3.20)
In classical mechanics,
Γ(E, N,V, x) ≡

0<H<E

h
3(N
A
+N
B
+...)
N
A
!N
B
! . . .
. (3.21)
State density:
D(E, N,V, x) ≡

∂E
Γ(E, N,V, x). (3.22)
If allowance δE is small,
D(E, N,V)δE =W(E, δE, N,V, x).
Correspondence of classical and quantum mechanics: It is important to keep in mind
(i) Uncertainty condition,
∆p∆q ∼h.
Because of that the classical phase volume ∆Γ corresponds to ∆Γ/h
f
quantum states ( f is
the number of degrees of freedom;
(ii) Indistinguishability of identical particles: it is an important features of Bose and Fermi
statistics. In the classical limit h →0 it leads to a factor 1/∏
i
N
i
!.
We shall come back to that point later discussing quantum statistics in more detail. Though the
factor 1/N! arises only from quantum statistics, its necessity has been understood much earlier
to make the entropy extensive quantity.
After definition of number of states
Another definition of entropy: Another definition of entropy can be formulated through the
probability distribution. For a closed system, the distribution function can be written as a function
of the energy, f
l
= f (E
l
). Then the required number of states in the region E, E +dE can be
written as D(E)dE, while the energy probability distribution is
P(E) = f (E)D(E)dE .
28 CHAPTER 3. PRINCIPLES OF STATISTICAL MECHANICS
Since

P(E)dE = 1 and P(E) is centered near the average energy,
P(
¯
E)∆E ≈ f (
¯
E)D(
¯
E)∆E ≈ f (
¯
E)W = 1.
Thus
S =−klog f (
¯
E) =−k'log f (E
l
)`. (3.23)
The distribution function f is constrained by the normalization condition

i
f (E
i
) = 1. (3.24)
The latter inequality follows from statistical independence. Indeed, log f (E
l
) must be an additive
function of the energy (in general, also of other additive integrals of motion, P and M),
log f (E
l
) =α+βE
l
.
As a result
log f (
¯
E) ='log f (E
l
)` =−

l
f
l
log f (E
l
).
This expression is written for quantum statistics. In the classical statistics one has to remember
the normalization factor [h
f

j
N
j
!]
−1
for the thermodynamic weight. Thus one has to replace
f
l

h
f

j
N
j
!
¸
f (p, q)
in the argument of the logarithm to obtain
S =−

dΓ f (p, q) log
¸
h
f


j
N
j
!

f (p, q)
¸
. (3.25)
The entropy, defined in this way, is additive.
Physical significance of entropy: Let us start with the microcanonical distribution,
dP = const δ(E −E
0
)

a

a
= const δ(E −E
0
)

a
(dΓ
a
/dE
a
)dE
a
(after replacement dΓ
a
/dE
a
→∆Γ
a
/∆E
a
)
= const δ(E −E
0
)e
S

a
dE
a
/∆E
a
.
Here S =∑
a
S
a
, E =∑
a
E
a
.
Partial entropy S(E
1
, E
2
, . . .) must have maximum at E
a
=
¯
E
a
. Since these values correspond
to the equilibrium state the entropy of a closed system in a state of complete statistical equilib-
rium has its greatest possible value (for a given energy of the system).
3.6. INFORMATION 29
The law of increase of entropy: If a closed system is at some instant in a non-equilibrium
macroscopic state, the most probable consequence at later instants is a steady increase of the
entropy of the system.
This is the law of increase of entropy or the second law of thermodynamics. This law has been
formulated by R. Clausius (1865), its statistical explanation was given by L. Boltzmann (1870).
It has a character of assumption which cannot be rigorously proved because the equations of
mechanics are time-reversible.
If the entropy remains constant. the process is called reversible, if it increases the process is
called irreversible.
3.6 Information
The definition of entropy in Eq. (3.23), is directly related to the definition of the information
measure. As an example consider the probability distribution for the number of ‘eyes’ that come
up when throwing a fair die. The probabilities of the six possible states [1`, [2`, . . . , [6` are f
i
,
with i = 1, . . . , 6. Small variations δ f
i
, lead to a variation in S as defined in (3.23). The maximal
value of S may be found, taking into account the constraint (3.24), using a Lagrange parameter
k
B
λ:
0 = δ(S−k
B
λ 1) =δ


6

i=1
k
B
f
i
ln f
i
−k
B
λ
6

i=1
f
i

= −k
B
6

i=1

ln f
i
+
f
i
f
i

δ f
i
⇒ f
i
= exp(−1−λ) ⇒ f
i
=
1
6
. (3.26)
Here we have chosen λ so that f
i
satisfies the constraint (3.24). We see that requiring the infor-
mation entropy to have its maximum value leads to the result that f
i
= 1/6, independent of i—as
expected for fair dice.
When we throw two dice, we have many more possibilities. As before, the ‘first’ die has
the possible states ¦[i`¦, with i = 1, . . . , 6. The other die has the states ¦[ j`¦ with j = 1, . . . , 6,
of course. Counting the eyes for the pair of dice we have 36 possible outcomes ¦[α`¦, with
α = 1, . . . , 36 corresponding to the states [11`, [12`, . . . [66`. With the argument developed above
we will find that f
α
= 1/36. It is important to note, however, that the statistical independence of
the two dice allows another approach.
The states [α` =[i`[ j` are in a product space of the two independent states [i` and [ j`, much
in the same way that the position of a point in the plane r = (x, y), is given in terms of the
independent coordinates x and y; or with our notation [r` = [x`[y`. The process of throwing the
dice yields eyes that are statistically independent, and the probability, f
i j
, that the first die has i
eyes and the second has j eyes is the product of the probability f
i
for the first die to show i eyes
and f
j
, which is the probability that the second die shows j eyes:
f
α
= f
i j
= f
i
f
j
statistically independent (3.27)
30 CHAPTER 3. PRINCIPLES OF STATISTICAL MECHANICS
It follows that the entropy of the combined system may be written
S = −k
B ∑
i j
f
i j
ln f
i j
=−k
B ∑
i j
f
i
f
j
ln f
i
f
j
= −k
B ∑
i
f
i
ln f
i ∑
j
f
j
+−k
B ∑
j
f
j
ln f
j ∑
i
f
i
= S
1
+S
2
,
where we have used that probabilities are normalized: ∑
i
f
i
= ∑
j
f
j
= 1. We conclude that the
(information) entropy of a system consisting of statistically independent parts is the sum of the
entropies of the parts. Entropy is an extensive property, i.e. entropy is proportional to system
size.
3.7 Normal systems in statistical thermodynamics
Asymptotic forms for the number of states and state density:
(i) When number of particles N, or the volume V tends to infinity, the number of states in
normal systems approach the following asymptotic behavior:
Γ ∼exp[Nψ(E/N)] or exp[Vψ(E/V)] ,
Γ ∼exp[Nψ(E/N,V/N)] or exp[Vψ(E/V, N/V)], (3.28)
and
ψ > 0, ψ
/
> 0, ψ
//
> 0. (3.29)
(ii) Therefore
D(E) =
∂Γ
∂E

/
exp(Nψ) > 0,
∂D(E)
∂E
∼ ψ
/
2
exp(Nψ) > 0. (3.30)
Entropy of a normal thermodynamic system: For a normal system, the statistical entropy is
S = k log[D(E)∆E] ≈k logΓ(E) = kNψ. (3.31)
It is proportional to N, or to V.
Chapter 4
Thermodynamic quantities
4.1 Temperature
Consider 2 bodies in a thermal equilibrium which are in thermal contact with each other but are
isolated form the surrounding, i.e. form a closed system. Since both the energy and entropy are
additive, we have
E = E
1
+E
2
, S = S(E
1
) +S(E
2
).
In the equilibrium the entropy has a maximum with respect to the parameters of subsystems.
Consequently, dS/dE
1
= 0. Form this equality we get
dS
dE
1
=
dS
1
dE
1
+
dS
2
dE
2
dE
2
dE
1
=
dS
1
dE
1

dS
2
dE
2
= 0,
Here we have used the energy conservation, E
2
= E −E
1
. As a result,
dS
1
dE
1
=
dS
2
dE
2
.
The absolute temperature is defined as T = dE/dS, or, more rigorously,
T = (dE/dS)
V
(4.1)
which means that the volume V is kept constant. In the equilibrium, T
1
= T
2
.
The previous argument may be generalized to a system consisting of many parts. At equilib-
rium one concludes that ∂S/∂E has the same value everywhere. Therefore one finds that absolute
temperature T, defined as the change in the (equilibrium) entropy with energy is independent of
position for a system in thermal equilibrium; it is also the same for two systems in equilibrium
with each other, for example for a liquid in equilibrium with its vapor. In Eq. (4.1) we have ex-
plicitly written that the change is to take place at a fixed volume. Volume changes are discussed
in the section 4.3.
31
32 CHAPTER 4. THERMODYNAMIC QUANTITIES
Now let us discuss the case when the bodies and not in an equilibrium with each other,
T
1
= T
2
. During the equilibration the entropy S = S
1
+S
2
must increase,
dS
dt
=
dS
1
dt
+
dS
2
dt
=
dS
1
dE
1
dE
1
dt
+
dS
2
dE
2
dE
2
dt
> 0.
Since the energy is conserved,
dS
dt
=

dS
1
dE
1

dS
2
dE
2

dE
1
dt
=

1
T
1

1
T
2

dE
1
dt
> 0.
If T
2
> T
1
, then dE
1
/dt > 0, dE
2
/dt < 0 - energy passes from bodies at higher temperature to
bodies at lower temperature.
The temperature is measured in degrees Kelvin, the number of Joules per degree is called the
Boltzmann constant, see Eq. (3.19). Below we omit the Boltzmann constant in all the formulas.
Then the temperature is measured in the units of energy. To get custom formulas one has to
replace T →k
B
T, S →S/k
B
.
4.2 Adiabatic process
Consider a thermally isolated system under time-dependent force λ, the classical Hamiltonian
being H(p, q, t). Thermally isolated means that H is dependent on the coordinates and moments
of the molecules belonging only to the body in question. If the force is purely mechanical (we
ignore the feedback to the force source from the system under consideration), then one can apply
the law of increase of entropy to the body in question.
Let us then assume that λ varies in time sufficiently slow
1
and expand dS/dt in powers of
dλ/dt. Up to the lowest order,
dS
dt
= A


dt

2

dS

= A

dt
.
Indeed, the 1st power is absent because entropy must increase with time. We obtain that at
dλ/dt →0 the entropy tends to be λ-independent. Thus, the adiabatic process appears reversible.
Using adiabatic process one can calculate mean values. For example, let us define the ther-
modynamic energy as
E ≡H(p, q; λ).
Since dH(p, q; λ)/dt =∂H(p, q; λ)/∂t (that follows from the Hamilton equation),
dE
dt
=
dH(p, q; λ)
dt
=
∂H(p, q; λ)


dt
.
1
comparing to the relaxation time of the system
4.3. PRESSURE 33
On the other hand, E is a function of the entropy, S, and of external parameter, λ. Thus
dE
dt
=

∂E
∂λ

S

dt
.
As a result,
∂H(p, q; λ)
∂λ
=

∂E
∂λ

S
.
4.3 Pressure
We have already mentioned that both the entropy and the energy are additive quantities, i.e. they
are proportional to system size, and therefore do not depend on the shape for systems in thermal
equilibrium. We may consider the entropy to be a function of energy, S(E), or equivalently the
energy to be a function of the entropy, E(S), for the closed system, i.e. in the case when we have
no external system or medium. Let us consider below the case where the system does not take
part in a macroscopic motion. Then the energy E is equal to the internal energy which we will
denote as U. Thus, in our case, U =H(p, q).
Now, consider a process in which the wall of the container moves adiabatic, i.e. slowly
enough for the system to stay in internal thermal equilibrium. A force F acts from the contents
of the system on the container wall surface element, δs. If the container wall moves a small
distance, δr, then work is done on the surroundings given by δW = F δr. This work must be
equal to the change of the internal energy of the system. Consequently, the force acting on a
surface element ds is
F =−
∂H(p, q; r)
∂r
=−

∂U
∂r

S
=−

∂U
∂V

S
∂V
∂r
=−

∂U
∂V

S
ds.
The magnitude of the force per area ds is called the pressure,
P =−

∂U
∂V

S
. (4.2)
Combining definitions of the temperature and pressure, one can write
dU = T dS−PdV . (4.3)
This relation is often called the first law of thermodynamics. The pressures of the bodies in the
equilibrium are the same (to preserve mechanical stability).
2
It follows from Eq. (4.3) that
T =

∂U
∂S

V
, P =−

∂U
∂V

S
. (4.4)
2
This property can be violated in metastable states.
34 CHAPTER 4. THERMODYNAMIC QUANTITIES
The first law (4.3) can be rewritten as
dS =
1
T
dU +
P
T
dV , (4.5)
and it follows that
P
T
=

∂S
∂V

U
. (4.6)
It the microscopic expression for the entropy is known, this equation allows one to calculate the
equation of state connecting P, T and V.
4.4 Equations of state
As we have seen, the systems like liquid or gas in an equilibrium can be described by state
variables like temperature T, pressure P, or total volume V. They are not independent since the
variables in equilibrium related by an equation of state. Usually it is taken in the form,
P = P(V, T). (4.7)
Thus, only two variables can be assigned independently. The equation of state can be obtained
either from experiment, or from microscopic theory.
An ideal gas of N particles obeys the equation of state
P = NT/V =ρT (temperature is measured in energy units) (4.8)
Here ρ ≡N/V is the gas density.
For one mole of the gas N equals to the Avogadro’s number N
A
= 6.02310
23
mole
−1
. If T
is measured in Kelvin, then T →k
B
T, and the equation of state for 1 mole reads
P = RT/V , R ≡N
A
k
B
= 8.314 J/(K mole).
Including interaction between particles in a simple approximate form leads to more realistic van
der Waals equation,
P =
NT
V −bN
−a

N
V

2
=

1−bρ
−aρ
2
. (4.9)
The item bρ in the denominator take care of particle repulsion at small distances, while the term
∝ ρ
2
allows for particle attraction at relatively large distances.
The phase diagram of real gases can be obtained from the equation of state. In general, it has
the form shown in Fig. 4.1. We will come back to phase transitions later where various phase
diagrams will be discussed.
4.5. WORK AND QUANTITY OF HEAT 35
Solid
Gas
Liquid
Critical point
Triple
point
T
P
Figure 4.1: Phase diagram of a real liquid-gas
system.
4.5 Work and quantity of heat
Consider a closed system that consists of two parts – the ‘system’ and the surrounding medium
or ‘heat bath’, Fig. 4.2
First consider the situation in which the system is thermally insulated from the medium. If
the volume of the system changes by δV then the medium, which is has a pressure P (equal to
the pressure of the medium), performs work on the system.
Figure 4.2: A closed system consisting of a heat bath surrounding the system of interest. (a) –
The surroundings work on a thermally insulated system by changing its volume by δV. The work
done by the system is δW =−PδV. The increase in energy of the system is δU =δW =−PδV.
(b) – The walls of the system conduct heat and in the process of changing the volume of the
system the heat δQ enters the system, therefore the energy change is δU = δQ+δW. Together
the system and the surroundings form a closed system
We assume that W > 0 when external forces produce work on the body in question, while
W < 0 if the body itself produces work. One obtains
dW/dt =−PdV/dt
when the volume changes in time. For a reversible process the pressure must be the same
throughout the body.
36 CHAPTER 4. THERMODYNAMIC QUANTITIES
If the body is not isolated thermally, then
dE
dt
=
dW
dt
+
dQ
dt
,
where Q is the heat gained or lost by the body per unit time. Here E is understood as a whole
energy of the body including macroscopic motion. For the body at rest E equals to the internal
energy U. In this case,
dQ
dt
=
dU
dt
+P
dV
dt
.
Assuming that the body is in an equilibrium corresponding to its energy and the volume at
the instant, one can put U =U(S,V),
dU
dt
= T
dS
dt
−P
dV
dt
.
Thus
dQ
dt
= T
dS
dt
. (4.10)
Note that dQ and dW are not complete differentials, i.e. that they depend on the way in which
the states of the system change in time.
One can imagine the situation when the pressure and temperature are constant throughout
the body, while the state is non-equilibrium (say, chemical reaction in the mixture). Since this
process is irreversible, one gets the inequality
dQ
dt
< T
dS
dt
. (4.11)
instead of Eq. (4.10). If the systems goes from one equilibrium state to another equilibrium state
through some non-equilibrium states, than
δQ ≤TδS. (4.12)
This relation can be considered as a form of second law of thermodynamics.
4.6 Thermodynamics potentials
4.6.1 The heat function (enthalpy)
If the process occurs at constant volume, dQ = dU. If the pressure is kept constant,
dQ = d(U +PV) ≡dH, H(S, P) =U +PV . (4.13)
Since
dU = T dS−PdV , dH = dU +PdV +V dP,
4.6. THERMODYNAMICS POTENTIALS 37
one obtains
dH = T dS+V dP. (4.14)
Thus,
T = (∂H/∂S)
P
, V = (∂H/∂P)
S
. (4.15)
As a result, if the body is thermally isolated and the pressure is kept constant, then
H = const .
4.6.2 Helmholtz free energy and thermodynamic potential (Gibbs free en-
ergy)
Consider the work on a body in a reversible isothermal change of the state,
dW = dU −dQ = dU −T dS = d(U −TS) = dF , F(V, T) ≡U −TS. (4.16)
F is called the (Helmholtz) free energy. Since
dU = T dS−PdV , dF = dU −T dS−SdT
one obtains,
dF =−SdT −PdV . (4.17)
Thus,
S =−(∂F/∂T)
V
, P =−(∂F/∂V)
T
. (4.18)
Since U = F +TS
U = F −T(∂F/∂T)
V
=−T
2


∂T
F
T

V
.
Thus, if we know any of the quantities, U(S,V), F(T,V), or H(S, P), we can determine all the
remaining thermodynamic quantities. Thus U, F and H are called the thermodynamic potentials
with respect to their variables.
We still miss the thermodynamic potential with respect to P, T. Introducing
G =U −TS+PV = F +TS = H−TS
and using the equality
PdV = d(PV) −V dP
we obtain
dG =−SdT +V dP → S =−(∂G/∂T)
P
, V = (∂G/∂P)
T
. (4.19)
If there are other parameters, λ
i
the expression

i
Λ
i

i
38 CHAPTER 4. THERMODYNAMIC QUANTITIES
should be added to all thermodynamic potentials. In this way we can obtain the expressions for
the averages,
Λ
i
=
∂H(p, q; λ)
∂λ
i
=

∂F
∂λ
i

T,V,λ
j=i
=

∂H
∂λ
i

S,P,λ
j=i
= . . . .
If the parameters λ
i
change slightly, the changes in the thermodynamic potentials are equal if
each is considered for the appropriate pair of constant quantities,
(δU)
S,V
= (δF)
T,V
= (δH)
S,P
= (δG)
T,P
. (4.20)
This statement is called the theorem of small increments.
Since
dU
dt
+P
dV
dt
< T
dS
dt
at constant V and T we get dF/dt < 0, while at constant P and T dG/dt < 0. Thus at the
equilibrium the thermodynamic potentials reach minima with respect to the variations of state
with proper constant parameters.
4.6.3 Dependence on the number of particles
Since S and V are additive the internal energy can be written as
U = N f (S/N,V/N)
since this is the most general homogeneous function of the first order in N, S and V. In a similar
way,
F = N f (V/N, T), H = N f (S/N, P), G = N f (P, T).
If one considers N as a formal independent variable, then all the thermodynamic potentials ac-
quire additional terms µdN, where
µ =

∂U
∂N

S,V
=

∂H
∂N

S,P
=

∂F
∂N

T,V
=

∂G
∂N

P,T
.
µ is called the chemical potential. We easily find that G = Nµ(P, T). Thus
dµ =−(S/N)dT +(V/N)dP.
If one considers the volume is constant, but the number of particles as a variable, the independent
variables are, say, T, N. Then it is convenient to introduce the new thermodynamic potential,
Ω= F −µN = F −G =−PV .
Then
dΩ=−SdT −Ndµ.
4.6. THERMODYNAMICS POTENTIALS 39
Sometimes Ωis called the Landau free energy. The number of particles can be expressed through
Ω(T,V, µ) as
N =−

∂Ω
∂µ

T,V
=V

∂P
∂µ

T,V
.
One can show that if 2 bodies are brought together and they can exchange by particles, then
µ
1
= µ
2
We can summarize the thermodynamic potentials in the table 4.1.
Thermodynamic potential Notation Independent variables Differential
Internal energy U S, V, N T dS−PdV +µdN
Heat function (enthalpy) H S, P, N T dS+V dP+µdN
Helmholtz free energy F T, V, N −SdT −PdV +µdN
Gibbs free energy G T, P, N −SdT +V dP+µdN
Landau free energy Ω T, V, µ −SdT −PdV −Ndµ
Table 4.1: Thermodynamic potentials (summary)
4.6.4 Derivatives of thermodynamic functions
The quantity of heat which must be gained in order to raise the temperature of the body by one
unit is called the specific heat. We get for the cases of fixed volume and pressure, respectively:
C
V
= T(∂S/∂T)
V
, C
P
= T(∂S/∂T)
P
.
Another important response functions are the compressibilities,
K
X
=−
1
V

∂V
∂P

X
(x =V, P, or S),
and the thermal expansion coefficient
α =
1
V

∂V
∂T

P
.
Usually, pairs of thermodynamic variables T,V and T, P are mostly used. If V and T are used
as independent variables, the results can be expressed through the specific heat C
V
(V, T) and
through pressure P. If we know the equation of state, P(V, T), then we can calculate everything.
Similarly, if P and T are used as independent variables, the results can be expressed through
the specific heat C
P
(P, T) and through the equation of state V =V(P, T).
As an example of such derivation one can find the relation between C
P
and C
V
considering
the internal energy U as a function T and V. Since
T dS =

∂U
∂T

V
dT +
¸
P+

∂U
∂V

T

dV
40 CHAPTER 4. THERMODYNAMIC QUANTITIES
we obtain assuming the pressure to be constant
C
P
=C
V
+
¸
P+

∂U
∂V

T

∂V
∂T

P
. (4.21)
Let us also demonstrate how one can obtain relations between second derivatives of thermo-
dynamic quantities. Since
S =−

∂F
∂T

V
we have

∂C
V
∂V

V
= T

2
S
∂V∂T
=−T

3
F
∂V∂T
2
=−T

2
∂T
2
2

∂F
∂V

T
.
Now, since
−P =

∂F
∂V

T
we get finally

∂C
V
∂V

T
= T


2
P
∂T
2

V
.
Consequently, this derivative can be calculated from the equation of state. Similarly,

∂C
P
∂P

T
= T


2
V
∂T
2

P
,
that van be obtained using Gibbs free energy instead of the Helmholtz one (prove!).
Calculation of derivatives from the equation of state
The calculations are based on the so-called Maxwell relations. Consider a thermodynamic po-
tential, F (X,Y), which depends on the independent variables X,Y. Then, dF can be written
as
dF = R
x
dX +R
y
dY .
Since

2
F
∂X ∂Y
=

2
F
∂Y ∂X
one obtains

∂R
x
∂Y

X
=

∂R
y
∂X

Y
.
Example:

∂T
∂V

S
=−

∂P
∂S

V
.
Other important relations (see derivation in Appendix A) arise from the properties of partial
differentiations. Consider 3 quantities, X,Y, Z, related by the equation of state K(X,Y, Z) =
4.6. THERMODYNAMICS POTENTIALS 41
const. We can consider X,Y as independent variables, while Z =Z(X,Y). Another way is assume
Y, Z to be independent variables, while X = X(Y, Z). The relations between the derivatives of
theses functions read

∂X
∂Y

Z

∂Y
∂Z

X

∂Z
∂X

Y
=−1, (4.22)

∂X
∂Y

Z
=

∂Y
∂X

−1
Z
. (4.23)
Any quantity F (X,Y), the differential of which can be expressed as
dF =

∂F
∂X

Y
dX +

∂F
∂Y

X
dX .
From this expression we get

∂F
∂Z

Y
=

∂F
∂X

Y

∂X
∂Z

Y
, (4.24)
and

∂F
∂X

Z
=

∂F
∂X

Y
+

∂F
∂Y

X

∂Y
∂X

Z
. (4.25)
Equations (4.24), (4.25) together with Eqs. (4.22), (4.23) and the Maxwell relations are usually
used for transformations and computation of derivatives from the equation of state.
4.6.5 Examples
Entropy:

∂S
∂V

T
=−

∂V

∂F
∂T

V
=−

∂T

∂F
∂V

T
=

∂P(V, T)
∂T

V
.
Here we have used the realtion dF =−SdT −PdV. Similarly,

∂S
∂P

T
=−

∂P

∂G
∂T

P
=−

∂P

∂G
∂P

T
=−

∂V(P, T)
∂T

P
.
Internal energy:

∂U
∂V

T
= T

∂S(V, T)
∂V

T
−P = T

∂P(V, T)
∂T

V
−P.
Here we have used the realtion dU = T dS−PdV. Similarly,

∂U
∂P

T
=−T

∂V(P, T)
∂T

P
−P

∂V(P, T)
∂P

T
,
42 CHAPTER 4. THERMODYNAMIC QUANTITIES

∂U
∂T

P
=C
P
−P

∂V
∂T

P
.
One can also obtain the following relations (prove!)
C
P
−C
V
= (TV/K
T

2
; (4.26)
K
P
−K
V
= (TV/C
P

2
; (4.27)
C
P
/C
V
= K
T
/K
S
. (4.28)
4.7 Principles of thermodynamic
4.7.1 Maximum work and Carnot cycle
Consider a thermally isolated system consisting of several bodies not in thermal equilibrium, see
Fig. 4.3. The system may do during establishment of equilibrium. This work, as well as the final
equilibrium state, depend on the concrete way which the system follows.
Let the total original energy be E
0
while the energy in the final equilibrium state is E(S).
Since the system is thermally isolated,
[W[ = E
0
−E(S), ∂[W[/∂S = (∂E/∂S)
V
=−T .
Here T is the temperature in the final state. Since the derivative is negative, the work decreases
with increasing S. Since the entropy cannot decrease the greatest possible work is done when
the entropy retains constant throughout the process. That means the reversible way.
Now let us consider two bodies with different temperatures, T
2
> T
1
. If one brings the bodies
into thermal contact, no work will be done since the process is irreversible. The entropy of two
bodies will increase be (δE)(1/T
1
−1/T
2
) where δE is the energy transfer. To get a maximum
work one has to accomplish a reversible cyclic process in a way to keep the bodies among which
the direct energy transfer takes place at the same temperature. For that purpose one needs a fur-
ther body (working medium). The working medium at temperature T
2
is brought in contact with
the “heater” and receives some energy isothermal. It is then adiabatic cooled at the temperature
T
1
and then adiabatic returned to the original state. The cyclic process is called the Carnot cycle.
Carnot discussed a process that has four stages:
1. The heat engine, at temperature T
2
, is brought into contact with the reservoir at T
2
, and
receives the amount −δE
2
isothermally.
2. The heat-engine is removed from the T
2
-reservoir, thermally isolated and undergoes an
adiabatic expansion until its temperature is decreased to T
1
.
3. The heat-engine is put in thermal contact with the T
1
reservoir, and gives off isothermally
the energy δE
1
at temperature T
1
.
4. The heat-engine is removed from the T
1
-reservoir, thermally isolated and undergoes an
adiabatic compression until its temperature is increased to T
2
. The cycle is then, and the
four stages may be repeated.
4.7. PRINCIPLES OF THERMODYNAMIC 43
Figure 4.3: A thermally isolated system contains two parts at different temperatures T
2
>T
1
. The
two parts are connected via a heat engine that may perform the work [W[ on the surroundings
outside the system. as the system approaches equilibrium. (a) – Initially the energy of the system
is E
0
. (b) – Finally the two parts have the same temperature T, and an amount of work was done
on the surroundings. The energy of the equilibrium system is a unique function of entropy, E(S).
Both E and S depends on how the equilibrium state was reached.
The heat-engine performs work on the outside world during expansion, and receives work during
contraction.
To calculate the work one can ignore the working medium for which the initial and final
states are the same. Since
−δU
2
=−T
2
δS
2
, δU
1
= T
1
δS
1
, δS
2
+δS
1
= 0 (reversible process)
we obtain
[δW[
max
=−δU
1
−δU
2
=−T
1
δS
1
−T
2
δS
2
=−(T
2
−T
1
)δS
2
=
T
2
−T
1
T
2
[δU
2
[, .
Consequently, the maximum efficiency η ≡[W[/(δU
2
) is
η
max
= (T
2
−T
1
)/T
2
< 1.
Lord Kelvin realized that the Carnot cycle may be used to define an absolute scale of temperature
– the scale used here. Since the efficiency η
max
≤ 1 it follows that there is a lower limit of
(equilibrium) of zero. The Kelvin scale therefore has a natural fixed point, absolute zero. Since
η
max
depends only on the ratio T
1
/T
2
, there is an arbitrary scale factor that is fixed by setting
T = 273.16 K for the triple point of pure water. With this scale one degree on the Celsius scale
equals one degree on the Kelvin scale.
44 CHAPTER 4. THERMODYNAMIC QUANTITIES
Work in an external medium
Consider a body embedded into a large external medium having T
0
= T, P
0
= P where T, P are
the temperature and the pressure of the body. Since the medium is large T
0
and P
0
are assumed
to be fixed. We are interested in the maximal work [W[
max
done by the thermally isolated body.
This work is equal to the minimal work W
min
done by an external source on the body. The total
change of the internal
3
energy, ∆U consists of the three parts
(i) mechanical work W done by the external source;
(ii) the work P
0
∆V
0
done by the medium;
(iii) the heat −T
0
∆S
0
gained from the medium.
Thus
∆U =W +P
0
∆V
0
−T
0
∆S
0
.
Since
∆V
0
+∆V = 0, ∆S
0
+∆S ≥0
W ≥∆U −T
0
∆S+P
0
∆V .
For a reversible process,
W
min
=−[W[
max
=∆(U −T
0
S+P
0
V). (4.29)
We come to the following conclusions.
(i) If the volume and the temperature of the body remains constant and equal to T
0
, then
W
min
=∆F.
(iii) If the pressure and the temperature of the body remains constant and equal to P
0
and T
0
,
respectively, then W
min
=∆G.
(iii) If no mechanical work is done and the body is left itself, the irreversible processes lead to
the inequality
∆(U −T
0
S+P
0
V) ≤0.
One can relate the minimal work W
min
with the deviation of the total entropy S
t
from its equilib-
rium value S
t
(U
t
). Indeed,
∆S
t
=−
dS
t
(U
t
)
dU
t
W
min
=−
∆U −T
0
∆S+P
0
∆V
T
0
=−
(T −T
0
)∆S−(P−P
0
)∆V
T
0
.
To reach maximum entropy one should require T = T
0
, P = P
0
.
3
We assume that there is no macroscopic motions as a whole.
4.7. PRINCIPLES OF THERMODYNAMIC 45
4.7.2 Thermodynamic inequalities. Thermodynamic stability
The vanishing of the 1st derivatives of thermodynamic quantities, though provide conditions for
the equilibrium, like T
1
= T
2
, P
1
= P
2
, is only a necessary condition for an extremum and do
not ensure that the entropy reaches the maximum. The sufficient conditions require analysis the
second derivatives.
As we have stated, the quantity
∆U −T
0
∆S+P
0
∆V
has a minimum. Thus
δU −T
0
δS+P
0
δV > 0
for any small deviation from the equilibrium. Thus, the minimum work which must be done to
bring the selected part from equilibrium to some other state is positive.
Now we can expand the internal energy U(S,V) in powers of the variables,
δU =
∂U
∂S
δS+
∂U
∂V
δV +
1
2
¸

2
U
∂S
2
(δS)
2
+

2
U
∂S∂V
δSδV +

2
U
∂V
2
(δV)
2

.
The 1st order terms provide T δS−PδV, and we have

2
U
∂S
2
(δS)
2
+

2
U
∂S∂V
δSδV +

2
U
∂V
2
(δV)
2
> 0.
Since δS and δV are arbitrary, one has to assume

2
U
∂S
2
> 0, (4.30)

2
U
∂V
2
> 0, (4.31)

2
U
∂S
2

2
U
∂V
2


2
U
∂S∂V

2
> 0. (4.32)
Since

2
U
∂S
2
=

∂T
∂S

V
=
T
C
V
we obtain from Eq. (4.30)
C
V
> 0.
In a similar way,

2
U
∂V
2
−=−

∂P
∂V

S
=
1
VK
S
> 0.
From Eq. (4.32) one obtains

∂T
∂V

2
S
<
T
VK
S
C
V
.
46 CHAPTER 4. THERMODYNAMIC QUANTITIES
After some algebra (see the book by Landau and Lifshitz, '21) one can get from Eq. (4.32) the
inequality

∂P
∂V

T
< 0.
Here we have calculated the conditions for thermodynamic stability.
4.7.3 Nernst’s theorem
As we know, for normal systems S →0 at T →0. This statement is called the Nernst’s theorem.
If it is true,
C
P
=C
V
= 0 at T = 0.
One can obtain similar equalities for T = 0,
(∂V/∂T)
P
= −(∂S/∂P)
T
= 0 (Maxwell relation for G)
(∂P/∂T)
V
= 0
(C
P
−C
V
)/C
P
= 0.
The following useful formulas follow from the Nernst’s theorem:
S =

T
0
(C
V
/T)dT ,
H = H
0
+

T
0
C
P
dT ,
G = H
0
+

T
0
C
P
dT −T

T
0
(C
V
/T)dT .
4.7.4 On phase transitions
The equilibrium between the phases requires
T
1
= T
2
, P
1
= P
2
, µ
1
= µ
2
.
Along the the co-existence curve between gas and liquid,
µ
g
(T, P) = µ
l
(T, P), dµ
g
(T, P) = dµ
l
(T, P).
Introducing entropy per particle s ≡S/N and volume v ≡V/N we get
−s
g
dT +v
g
dP =−s
l
dT +v
l
dP.
Consequently,

dP
dT

coexist
=
s
g
−s
l
v
g
−v
l
=
λ
T(v
g
−v
l
)
,
4.7. PRINCIPLES OF THERMODYNAMIC 47
where λ ≡T(s
g
−s
l
) is the latent heat. This is the Clausius-Clapeyron equation. From the equa-
tions of state for gas and liquid phases one can find the dependence P(T) along the coexistence
line. The latent heat λ depends on the temperature and vanishes at some temperature T
c
. At
T = T
c
the two phases become identical. So the entropy changes continuously while its first
derivatives change abruptly. We shall come back to phase transitions later in much greater detail.
48 CHAPTER 4. THERMODYNAMIC QUANTITIES
Chapter 5
The Gibbs Distribution
5.1 Spin
1
2
in a magnetic field
Let us start with a simple example. Consider N noninteracting localized particles with spin S =
1
2
embedded in an external magnetic field H. The Hamiltonian is
ε =−m H =−m
z
H, m = 2µS,
µ = e/2mc is the Bohr magneton. Since S
z

1
2
,
ε
↑(↓)
=∓µH.
Denoting n
↑(↓)
as numbers of states with spin up(down), respectively, we get for the energy
E =−µ(n

−n

)H, (5.1)
with the normalization condition
N = n

+n

. (5.2)
The total magnetic moment is
M = µ(n

−n

).
Under the assumptions of the microcanonical ensemble (i. e. with fixed total moment M), the
thermodynamic weight for the system is
W =
N!
n

!n

!
.
The numbers n
↑(↓)
can be expressed in terms of E and M. Using the Stirling’s formula,
1
logn! ≈n logn−n+1 at n 1 (5.3)
1
When N 1 one can replace ∑
n
= 0
N
lnn as

N
0
lnxdx = Nln(N/e).
49
50 CHAPTER 5. THE GIBBS DISTRIBUTION
we obtain
S = N logN−n

logn

−n

logn

.
Now we can express the quantities n
↑(↓)
in terms of the energy E given by Eq. (5.1) as
n
↑(↓)
=
µHN ∓E
2µH
.
consequently,
1
T
=

∂S
∂E

N
=
1
2µH
log
µHN +H
µHN −H
=
1
2µH
log
n

n

. (5.4)
Thus,
M = Nµ tanh
µH
T
, χ ≡

∂M
∂H

T
=

2
T
1
cosh
2
(µH/T)
.
χ is called the magnetic susceptibility.
Now let us define the same relations from the canonical ensemble, i. e. assuming that the
temperature is fixed by a thermal contact with a “thermal bath”. Since the energy cost for one
spin flip
n

→n

−1, n

→n

+1
is ∆U = 2µH and the entropy change is
∆S = log
N!
(n

−1)!(n

+1)!
−log
N!
n

!n

!
= log
n

n

+1
we obtain
∆S
∆U

1
2µH
log
n

n

.
That must be equal to the inverse temperature, and we recover previous results.
The previous results can be summarized as a distribution function
n
σ
=
N
Z
1
e
−βε
σ
, Z
1
=

σ=↑,↓
e
−βε
σ
= 2 cosh
µH
T
. (5.5)
Here β = 1/T. We have
¯ ε =
1
Z
1

σ=↑,↓
ε
σ
e
−βε
σ
=−µH tanh
µH
T
, U = N¯ ε.
Now there exist fluctuations around the average energy which are absent in the microcanonical
ensemble.
Exercise: Calculate the fluctuation in energy,

ε
2
−E
2
and compare it with the internal energy
E = N¯ ε. Show that the difference is small for a macroscopic body.
5.2. THE GIBBS DISTRIBUTION 51
5.2 The Gibbs distribution
Eq. (5.5) is the particular case of the Gibbs distribution. If there is a body, embedded into a
medium, the microcanonical distribution can be written as
dw ∝ δ(E +E
/
−E
(0)
)dΓdΓ
/
where superscript “prime” relate the quantities to the medium. If one fixes the body in a given
state n, then the probability to find the body in this state is
f
n

δ(E
n
+E
/
−E
(0)
)dΓ
/
=

δ(E
n
+E
/
−E
(0)
)

/
dE
/
dE
/
=

e
S
/
(E
/
)
∆E
/
δ(E
n
+E
/
−E
(0)
)dE
/
.
As a result,
f
n

e
S
/
(E
/
)
∆E
/

E
/
=E
(0)
−E
n
.
Now let us consider the body as very small comparing to the medium. Then
E
n
<E
(0)
→ S
/
(E
(0)
−E
n
) ≈S
/
(E
(0)
) +E
n
dS
/
(E
(0)
)
dE
(0)
= S
/
(E
(0)
) −E
n
/T .
Finally,
f
n
=
1
Z
e
−βE
n
, Z =

s
e
−βE
s
, β ≡1/T .
In a classical system,
f (p, q) =
1
Z
e
−βE(p,q)
, Z =

e
−βE(p,q)
dΓ.
Note that in the derivation we use the Hamiltonian of the whole system.
The Gibbs distribution is relevant for a body in an equilibrium with a large medium - “thermal
bath”.
If the systems consists of non-interacting particles which have eigenenergies ε
s
, sth state
being degenerate by g
s
times, then we arrive at the single-particle distribution function,
f
n
=
g
n
Z
1
e
−βε
n
, Z
1
=

s
g
s
e
−βε
s
.
Here Z
1
is the single-particle partition function.
The Maxwell-Boltzmann distribution
Derivation from the Gibbs distribution
Consider a system of non-interacting particles. Dividing the total energy into potential and ki-
netic parts,
E(p, q) =V(q) +K(p)
52 CHAPTER 5. THE GIBBS DISTRIBUTION
we can write
f (p, q) = f
q
(q) f
p
(p)
f
q
= e
−βV(q)
/

dqe
−βV(q)

;
f
p
= e
−βK(p)
/

dpe
−βK(p)

.
For non-interacting particles the single-particle distribution is
f
p
(p) = (2πmT)
−3/2
e
−(p
2
x
+p
2
y
+p
2
z
)/2mT
.
Exercise: Show that the average kinetic energy per atom is
3
2
T.
Statistical derivation
Consider particles with energy levels ε
i
and degeneracy factor g
i
, N =∑
i
n
i
. Here n
i
is the number
of particle occupying ith level. If the particles are identical and distinguishable the total number
of microstates for a given state specified by the set ¦n
i
¦ is
W = N!

i
g
n
i
i
n
i
!
, E =

i
n
i
ε
i
.
The easiest way is to use canonical ensemble which specifies the temperature. Let us consider
the transition from jth to ith state,
n
i
→n
i
+1, n
j
→n
j
−1.
Then
∆U = T ∆S =ε
i
−ε
j
, (5.6)
while
∆S = log
g
n
i
+1
i
(n
i
+1)!
g
n
j
−1
j
(n
j
−1)!
−log
g
n
i
i
n
i
!
g
n
j
j
n
j
!
= log
g
i
n
j
g
j
(n
i
+1)
. (5.7)
Comparing Eqs. (5.6) and (5.7), we get
g
i
n
j
g
j
(n
i
+1)
= e
β(ε
i
−ε
j
)
.
Thus we arrive at the same result with Z = Z
1
= ∑
i
g
i
e
−βε
i
where ε
i
are single-particle energy
levels.
5.2. THE GIBBS DISTRIBUTION 53
Expressions for thermodynamic quantities
Average single-particle energy:
¯ ε =
1
Z
1

i
ε
i
g
i
e
−βε
i
=−
∂logZ
1
(β)
∂β
, U = N¯ ε.
Entropy: Since the entropy is given by the relation
S =−'log f
n
` =−

n
f
n
log f
n
we obtain in general
S = logZ +U/T .
The same result can be obtained from single-particle considerations. For N distinguishable
particles we have
S = logW ≈N logN−N+

i
(n
i
logg
i
−n
i
logn
i
)
≈ N logN−

i
n
i
log
n
i
g
i
= N logN−

i
n
i

log
N
Z
1
−βε
i

= N logZ
1
+U/T = logZ
N
1
+U/T = logZ +U/T .
Here we employ the fact that Z = Z
N
1
. If the particles are non-distinguishable, them one
has to divide the thermodynamic weight by N! to obtain
Z =
Z
N
1
N!
, S =−

s
g
s
( f
s
log f
s
− f
s
).
Helmholtz free energy:
F =U −TS =−T logZ =−T log
Z
N
1
N!
This formula is fundamental in thermodynamic applications of the Gibbs distribution. Us-
ing this formula one can express the Gibbs distribution in the form
f
n
= e
−β(F−E
n
)
.
5.2.1 The Gibbs distribution for a variable number of particles
.
If the body can exchange by particles with the medium, one has to specify the particle num-
bers, N and N
/
, respectively, so that N +N
/
= N
(0)
. Both the entropies and energies become
functions on N, N
/
. As a result, we get
f
nN
∝ exp

S
/
(E
(0)
−E
nN
, N
(0)
−N)

.
54 CHAPTER 5. THE GIBBS DISTRIBUTION
Next, we expand the entropy up to the 1st order in E
nN
and N. Since

∂S
∂U

V,N
=β,

∂S
∂N

U,V
=−βµ,
we get
S
/
(E
(0)
−E
nN
, N
(0)
−N) = S
/
(E
(0)
, N
(0)
) −βE
nN
+βµN
(due to the assumption of the equilibrium, the temperatures and chemical potentials of the body
and the medium are the same). Thus,
f
nN
= Ae
β(µN−E
nN
)
.
The normalization constant A can be expressed through the thermodynamic quantities. Since
S =−'log f
nN
` =−logA−βµ
¯
N+β
¯
E → T logA =U −TS−µ
¯
N =Ω,
we obtain
f
nN
= e
β(Ω+µN−E
nN
)
.
Since

N

s
f
sN
= e
βΩ

N

e
βµN

s
e
βE
sN

= 1,
one obtains
Ω=−T log

N

e
βµN

s
e
βE
sN

.
Remember that for classical system of N non-distinguishable particles one has to use the classical
Hamiltonian H(¦p, q; N¦) instead E
nN
, the phase volume element being
1
(2π¯ h)
dN
N!
N

i=1
dp
i
dq
i
,
while for distinguishable particles there is no factor 1/N!. Here d is the dimensionality of the
real space.
Chapter 6
Ideal gas
6.1 Classical gas
Having in mind quantization in a box,
ψ
n
(r) = sin(k
x
x) sin(k
y
y) sin(k
z
z), k
i
=πn
i
/L (n
i
= 1, 2, . . .)
we obtain
ε
n
=
¯ h
2
π
2
2mL
2
(n
2
x
+n
2
y
+n
2
z
).
Thus,
Z
1
=

n
x
,n
y
,n
z
e
−(β¯ h
2
π
2
/2mL
2
)(n
2
x
+n
2
y
+n
2
z
)

1
2


−∞
dn
x
e
−(β¯ h
2
π
2
/2mL
2
)n
2
x

3
=V/Λ
3
. (6.1)
Here
Λ =

2πβ¯ h
2
/m

1/2
= ¯ h

2π/mT ∝ T
−1/2
is the thermal de Broglie wave length. Another way to obtain this result is to use the continuous
limit one can write
Z
1
=V

d
3
p
(2π¯ h)
3
e
−βp
2
/2m
.
It is very convenient to introduce the density of states ,
D(ε) = g

d
3
p
(2π¯ h)
3
δ

ε −
p
2
2m

= g
(2m)
3/2

2
¯ h
3

ε ≡D
0

ε, D
0
≡g
(2m)
3/2

2
¯ h
3
. (6.2)
Here g is the degeneracy factor of the state with given momentum p Now, the integral is easily
done, to get
Z
1
=VD
0

T


0

xe
−x
dx =V/Λ
3
.
Introduction of the density of states helps in calculation for different systems.
55
56 CHAPTER 6. IDEAL GAS
Exercise: Calculate the partition function for two-dimensional gas.
Thermodynamic quantities
Now we proceed with thermodynamics:
F = −T log
Z
N
1
N!
=−NT log(V/Λ
3
) +T logN! ; (6.3)
P =

∂F
∂V

T,N
=
NT
V
=ρT (ρ is the gas density) ; (6.4)
µ =

∂F
∂N

T,V
≈−T log
Z
1
N
= T log

3
T
; (6.5)
U = −

∂β
logZ =
3
2
NT ; (6.6)
C
V
=
3
2
N; (6.7)
S = −

∂F
∂T

V,N
= log
Z
N
1
N1
+
U
T
= N
¸
5
2
−logρΛ
3

. (6.8)
Something is wrong: the entropy and the specific heat does not vanish at zero temperature.
Indeed, the classical statistics is applicable only at
ρΛ
3
<1. (6.9)
This inequality is violated at low temperatures. What was wrong in out derivation? We implicitly
assumed that one state can either be empty, or occupied only by one particle. Indeed, according
to Eq. (6.9), the number of states V/Λ
3
is much greater than the number of particles, N. Only
because of this assumption the equality
Z =
Z
N
1
N!
holds.
If the particles have internal degrees of freedom, k, with the energies ε
k
, then the free energy
acquires an additional item,
δF =−TNlog

k
e
−βε
k
.
Ideal gas with a constant specific heat
In many important cases the specific heat can be regarded as temperature-independent. Using the
Stirling’s formula for large number of particles , N 1, we can write the Helmholtz free energy
as
F =−NT log
¸
eV

3

k
e
−βε
k
¸
≡−NT log
eV
N
+N f (T), C
V
=−NT f
//
(T),
6.1. CLASSICAL GAS 57
the assumption C
V
= const(T) leads to the expression
f (T) =−c
V
T logT −ζ
c
T +ε
0
,
where c
V
≡C
V
/N, while ζ
c
and ε
0
are integration constants. ζ
c
is called the chemical constant.
We get,
U = Nε
0
+Nc
V
T ;
G = Nε
0
+NT logP−Nc
V
T logT −Nζ
c
T (since PV = NT);
H = U +PV = Nε
0
+Nc
P
T ;
S = N log(eV/N) +Nc
V
logT +(ζ
c
+c
V
)T
= −N logP+Nc
P
logT +(ζ
c
+c
P
)N.
As a consequence, we can derive the Poisson adiabatic, i. e. the relation between V, P, and T
for adiabatic expansion of compression of ideal gas with constant specific heat. Since
−N logP+Nc
P
logT = const
one obtains
T
c
P
/P = const → T
γ
P
1−γ
= const , γ = c
P
/c
V
.
Here we employed the relation c
P
−c
V
= 1 (prove!) which is valid for the ideal gas. Since
PV = NT
TV
γ−1
= const , PV
γ
= const .
Polyatomic gases
Vibrational degrees of freedom
If any internal motion of the molecules is present, then the potential energy of the molecule po-
tential energy can be expanded in powers of the mechanical displacements q fromthe equilibrium
positions,
V =ε
0
+
r
v

i,k=1
a
ik
q
i
q
k
.
Here r
v
is the number of vibrational degrees of freedom. If the molecule contains n atoms, then
the total number of degrees of freedom is 3n. If the atoms in the molecule are collinear, there
are 3 translational degrees of freedom (for a molecule as a whole) and 2 rotational degrees of
freedom, while 3n −5 modes are left for vibrational motion. If the molecule is not a collinear
one, then r
v
= 3n −6. The kinetic energy depends on all 3n momenta. Consequently, the total
energy and be written as
ε =ε
0
+ f
II
(p, q),
where f
II
(p, q) is a quadratic function of 3n momenta and 3n−6 (or 3n−5) vibrational coordi-
nates. Thus we have the total number of variables l = 6n−6, or l = 6n−5. For a monoatomic
58 CHAPTER 6. IDEAL GAS
gas the total number of variables is 3 because coordinates do not appear in the expression for the
energy. As a result, we obtain,
F =−NT log
¸
e e
−ε
0
/T
N

dΓe
−f
II
(p,q)
¸
.
Now we can transform the variables as
p = p
/

T, q = q
/

T → f
II
(p, q) = T f
II
(p
/
, q
/
), dΓ = T
l/2

/
.
In this way, we get
F =−NT log

AV
−βε
0
T
l/2
/N

.
Expanding the logarithm, we get
c
V
= l/2, c
P
= c
V
+1 = (l +2)/2.
Thus a classical gas has a constant specific heat. We can formulate a rule:
Each degree of freedom contributes T/2 (or kT/2 in ordinary units) to the average energy.
Monoatomic ideal gas
Let exclude the kinetic energy of translational motion and label atomic intrinsic levels as ε
/
k
.
Denoting
Z =

k
e
−βε
/
k
we obtain
F =−NT log
¸
eV

3
Z

=−NT log
¸
eV
N

mT
2π¯ h
2

3/2
Z
¸
.
In the simplest case, when the atoms are in the ground state with zero angular moment and spin,
L = S = 0, then Z = e
−βε
0
where ε
0
is the energy of non-degenerate ground state. Starting the
energy scale from ε
0
we obtain:
Z = 1, c
V
= 3/2, ζ
c
= (3/2) log(m/2π¯ h
2
).
These results were obtained by Sackur and Tetrode in 1912. If only one of the angular momenta
is non-zero in th ground state of the atom, then the situation is rather simple. Usually, in a ground
state L = 0. while the total spin can be finite (e. g. vapor of alkali metals). Then Z = (2S +1)
instead of 1, and ζ
S
= log(2S +1). If both momenta are non-zero than there is a fine structure
in the levels which can be compared with the temperature. If we label the energy by the total
angular momentum J, then
Z =

J
(2J +1)e
−βε
J
since each level with a given J is (2J +1) degenerate with respect to directionds of J. This
expression can be simplified at high temperatures, ε
J
<T, when e
−βε
J
≈1. Then ∑
J
(2J +1) =
(2L+1)(2S+1) and
ζ
SL
= log[(2S+1)(2L+1)] .
6.1. CLASSICAL GAS 59
Diatomic gases
In this cases additional rotational and vibrational degrees of freedom must be taken into account.
Under some approximation, the energy can be written as sum to electronic, rotational, and vibra-
tional contributions. In the simplest case of a singlet electronic state
ε
vK

0
+ ¯ hω
0

v +
1
2

+
¯ h
2
2I
K(K+1).
Here ¯ hω
0
is the vibrational quantum, v is the vibrational quantum number, K is the rotational
quantum number, I = r
2
0
m
1
m
2
/(m
1
+m
2
) is the inertia moment of the molecule. Thus,
Z = e
−βε
0
Z
rot
Z
vib
Z
rot
=


K=0
(2K+1)e
−β¯ h
2
K(K+1)/2I
,
Z
vib
=


v=0
e
−β¯ hω(v+1/2)
.
As a result,
F =−NT log
¸
eV
N

mT
2π¯ h
2

3/2
¸
+Nε
0
+F
rot
+F
vib
.
The 2 last terms are defined according to the definition δF
α
= −NT logZ
α
. Introducing the
translational contributions to the specific heat and chemical constant as
ζ
c
= ζ
tr

rot

vib
, ζ
tr
= (3/2) log(m/2π¯ h
2
),
c
V
= c
tr
+c
rot
+c
vib
, c
tr
= 3/2,
c
P
= c
tr
+c
rot
+c
vib
+1.
As an example let us consider the case of large temperatures, when T ¯ h
2
/2I. Then one can
replace summation by the integration to obtain
Z
rot
=


0
(2K+1)e
−β¯ h
2
K(K+1)/2
= 2TI/¯ h
2
, c
rot
= 1, ζ
rot
= log(2I/¯ h
2
).
In this case
c
V
= 5/2, c
P
= 7/2.
In the opposite limit it is sufficient to retain the first two terms,
Z
rot
= 1−3e
−β¯ h
2
/I
to obtain
F
rot
=−3NTe
−β¯ h
2
/I
, c
rot
= 3(¯ h
2
/IT)
2
3e
−¯ h
2
/IT
.
60 CHAPTER 6. IDEAL GAS
Note that above expressions are valid only for the molecules with the different atoms.
If the atoms are equivalent, then at large temperatures two opposite directions correspond
to the same physical state. As a result, the rotational partition function must be halved. The
quantum case is more difficult and we do not discuss it.
The vibrational degrees of freedom can be easily accounted by direct summation of geomet-
ric series. Using the lowest vibrational level as the origin for the energies, we obtain
Z
vib
=

1−e
−β¯ hω

,
F
vib
= NT log

1−e
−β¯ hω

,
S
vib
= −Nlog

1−e
−β¯ hω

+N
β¯ hω
e
β¯ hω
−1
,
U
vib
= N
¯ hω
e
β¯ hω
−1
,
c
vib
=
(β¯ hω)
2
e
β¯ hω

e
β¯ hω
−1

2
.
The graphs for the rotational and vibrational contributions to the specific heat are shown in
Fig. 6.1.
1 2
Reduced temperature
2 1.5 1 0.5
1
0.8
0.6
0.4
0.2
0
Page 1
Figure 6.1: Contributions to the
specific heat from rotational (1)
and vibrational (2) degrees of free-
dom. Temperature is measured in
units of ¯ h
2
/2I in the graph (1) and
in units of ¯ hω in the graph (2).
Exercise: Calculate approximate expressions for the case of low and high temperatures and
discuss their physical meaning.
6.1. CLASSICAL GAS 61
Quantum Gases
Statistics of identical particles
Consider N indistinguishable particles moving in a common potential,
ˆ
H =
N

i=1
ˆ
H
i
,
ˆ
H
i
=
ˆ p
2
i
2m
+V(ˆ r
i
).
Here ˆ p
i
=−i ¯ h∇
i
and ˆ r
i
ore operators for the momentum and coordinate of the i-th particle. Let
us denote single-particle eigenenergies as ε
k
and the states as ψ
k
,
ˆ
H
i
ψ
k
(r
i
) =ε
k
ψ
k
(r
i
).
Specifying the total wave function as Ψ
s
=∏
i
ψ
k
i
(r
i
) we obtain E
s
=∑
i
ε
k
i
since the Hamiltonian
is the sum of independent terms.
Now we have to use the concept that the particles are indistinguishable. Introducing the
permutation operator
ˆ
P which interchanges i-th and j-th particle, r
i
↔ r
j
. Then, one has to
require

ˆ
H,
ˆ
P

= 0, or
ˆ
P
ˆ
H
ˆ
P
−1
=
ˆ
H .
According to the quantum theory, the proper many-body wave function must be an eigenstate of
the permutation operator,
ˆ

s
= pΨ
s
.
Since double permutation is just the identity operator,
ˆ
P
2
=
ˆ
I, we get
p
2
= 1 → p =±1.
The particles with p = 1 called the bosons, while the particles with p = −1 are called the
fermions.
The antisymmetric wave function vanishes if two particles are in the same state. This property
is called the Pauli principle.
Example: Particles in a Harmonic Potential: Ignoring zero-point vibrations we have ε
n
=
n¯ hω, n = 0, 1, 2. . . . Then
Z
1
=


n=0
x
n
=
1
1−x
, x ≡e
−β¯ hω
.
For the Maxwell-Boltzmann (MB) statistics we have
Z
MB
2
=
Z
2
1
2!
=
1
2
1
(1−x)
2
.
Note: The product Z
2
1
contains 1 configuration where 2 particles occupy the same state. By
dividing this product by 2! we ascribe the weight 1/2 to this configuration. This procedure is
62 CHAPTER 6. IDEAL GAS
wrong from the point of view of quantum statistics. Indeed, bosons are allowed to occupy the
same single-particle state, so the the diagonal term
D =


n=0
x
2n
=
1
1−x
2
must be added with the weight 1/2.
Z
BE
2
= Z
MB
2
+
1
2
D, → Z
BE
2
=
1
(1−x)(1−x
2
)
.
In a similar way, the diagonal term must be subtracted for Fermi-Dirac (FD)statistics,
Z
FD
2
= Z
MB
2

1
2
D, → Z
FD
2
=
x
(1−x)(1−x
2
)
.
The difference is negligible for 1−x <1 (large temperature).
The Fermi-Dirac Statistics
This statistics is based on the Pauli principle:
Each quantum state cannot be simultaneously occupied by more than 1 particle.
The particles obeying the Fermi-Dirac statistics are called the fermions.
Derivation from a grand canonical ensemble
Consider the system of particles with single-particle levels ε
i
and degeneracy factors g
i
. If the
state is occupied, then there are only g
i
−1 states at the i-th level for the second particles, the
number of available states for two particles being g
i
(g
1
−1)/2. The factor 1/2 appears because
the particles are non-distinguishable. For n
i
particles we have
W =

i
g
i
!
n
i
!(g
i
−n
i
)!
.
Now we apply the conventional approach. Namely, consider a transition n
i
→n
i
+1. The energy
change is ε
i
, the change in the particle number is ∆N = 1, while the entropy change is (prove!)
∆S = log
g
i
−n
i
n
i
+1
≈log
g
i
−n
i
n
i
.
Since
T ∆S =∆U −µ∆N =ε
i
−µ
we obtain
g
i
n
i
−= e
β(ε
i
−µ)
→n
i
=
g
i
e
β(ε
i
−µ)
+1
.
6.1. CLASSICAL GAS 63
As a result, the probability for a state i to be occupied is
f
i
=
1
e
β(ε
i
−µ)
+1
, (6.10)
the normalization condition (which determines µ) being
N =

i
g
i
f
i
=

i
g
i
e
β(ε
i
−µ)
+1
.
Derivation from the Gibbs distribution
Consider the particles belonging to i-th state. We have the general expression

i
=−T log

n
i
e
β(n
i
µ−n
i
ε
i
)
=−T log

n
i

e
β(µ−ε
i
)

n
i
. (6.11)
Now, since n
i
= 0 or 1, we get

i
=−T log

1+e
(βµ−ε
i
)

. (6.12)
The following way is standard:
f
i
= ¯ n
i
=−
∂Ω
i
∂µ
=
1
e
β(ε
i
−µ)
+1
. (6.13)
The total Landau free energy is
Ω=−T log

i

1+e
β(µ−ε
i
)

. (6.14)
The Bose-Einstein distribution
For bosons the number of particles occupying any state can be arbitrary. A proper picture is
g
i
boxes separated by g
i
−1 walls. To count the number of configurations one can count the
different grouping of g
i
−1 walls and n
i
bosons. As a result,
W =

i
(n
i
+g
i
−1)!
n
i
!(g
i
−1)!
→∆S ≈log
g
i
+n
i
n
i
.
Consequently,
f
i
=
n
i
g
i
=
1
e
β(ε
i
−µ)
−1
. (6.15)
We can derive this expression also from the general formula (6.11) by performing summation
over n
i
from 0 to ∞. Using the properties of geometric series we obtain,

i
= T ln

1−e
β(µ−ε
i
)

. (6.16)
Again, we recover the expression (6.15) as f
i
=−∂Ω
i
/∂µ. It is worthwhile noting that Eq. (6.15)
is applicable only if
e
β(µ−ε
i
)
< 1 → µ < 0.
We immediately recover the Boltzmann distribution at n
i
<g
i
by replacing (n
i
+g
i
−1)! by g
n
i
i
.
64 CHAPTER 6. IDEAL GAS
6.2 Ideal gases out of equilibrium
Let us assume that we label the states of the system by the subscript i, the degeneracy factor and
the occupation number of ith state being g
i
and n
i
, respectively. Then the set ¦n
i
¦ describes the
state of the system if it is also out of equilibrium. Since the particles are independent,
W =

i
∆Γ
i
For the Boltzmann gas one can place each of n
i
particles in one of the possible g
i
state. Then the
number of different physical configurations is
∆Γ
i
= g
n
i
i
/n
i
! . (6.17)
As a result,
S = lnW =

ln(g
n
i
i
/n
i
!) =

(n
i
lng
i
−lnn
i
!) . (6.18)
Now we can use the Stirling’s formula to get
S =

i
n
i
ln(eg
i
/n
i
) ≡

i
g
i
f
i
ln(e/ f
i
). (6.19)
Here we have introduced the distribution function f
i
≡n
i
/g
i
as the probability for ith state to be
occupied.
Let us now require that the entropy must be a maximum for the equilibrium state. Then
the problem is to find f
i
such that the sum (6.19) has the maximum value possible under the
restrictions

i
n
i
=

i
g
i
f
i
= N (particle number conservation) , (6.20)

i
ε
i
n
i
=

i
g
i
ε
i
f
i
= E energy conservation) . (6.21)
Requiring that
∂(S+µβN−βE)/∂ f
i
= 0,
one obtains
g
i
(−ln f
i
+α+βε
i
) = 0 → f
i
= e
β(µ−ε
i
)
. (6.22)
For the Fermi gas, not more than one particle can be in each state. Thus one has to count the
number of way to fill n
i
states from g
i
available ones, and ∆Γ
i
= g
i
!/n
i
!(g
i
−n
i
)!. Thus
S =−

i
g
i
[ f
i
ln f
i
+(1− f
i
)ln(1− f
i
)] . (6.23)
Again (Check!), requiring this expression to have a maximum under conditions (6.20) and (6.21)
we arrive at the Fermi-Dirac distribution (6.13).
For the Bose-Einstein statistics one has to calculate the number of ways of distributing n
i
particles among g
i
states. This problem can be mapped on the problem of distributing n
i
identical
6.3. FERMI AND BOSE GASES OF ELEMENTARY PARTICLES 65
balls among g
i
urns. Let us imagine that the latter are separated by by g
i
−1 vertical strokes
placed at intervals along the points. For example, the diagram
.[...[[....[..
represents 10 balls among 5 urns as ¦1, 3, 0, 4, 2¦. The total number of places (occupied both by
strokes and points) is then (g
i
+n
1
−1). The number of configuration is just the number of ways
to choose g
1
−1 positions for the strokes. Thus, ∆Γ
i
= ((g
i
+n
1
−1)!/n
i
!(g
i
−1)!, and
S =

i
g
i
[(1+ f
1
)ln(1+ f
i
) − f
i
ln f
i
] . (6.24)
The expressions (6.23) and (6.24) tend to the Boltzmann formula (6.19) at n
i
<g
i
since
g
i
! ≈(g
i
−n
i
)!g
n
i
i
, (g
i
+n
i
−1)! ≈(g
i
−1)!g
n
i
i
at n
i
<g
i
.
Finally, let us specify the expression for the entropy of Bose gas at f
i
1 since we will need this
expression later to discuss the Bose condensation. We have ∆Γ
i
≈n
(g
i
−1)
i
/(g
i
−1)! and
S =

i
g
i
ln(e f
i
). (6.25)
6.3 Fermi and Bose gases of elementary particles
For elementary particles we have
ε =
p
2
2m
, g = (2s +1).
Here g is the spin degeneracy. Thus the equation for the chemical potential has the form, see
Eq. (6.2),
N
V
=

dεD(ε) f (ε) = gD
0


0

εdε
e
β(ε−µ)
±1
= D
0
T
3/2


0

zdz
e
z−βµ
±1
.
Here “+” stands for fermions while “-” stands for bosons. In a similar way,
Ω=∓VD
0
T


0


ε log

1±e
β(µ−ε)

=−
2
3
VD
0


0
dεε
3/2
f (ε).
The second inequality is obtained using integration by parts. We recognize that
Ω=−(2/3)

i
g
i
ε
i
f
i
=−(2/3)


0
D(ε)ε f (ε)dε =−(2/3)U . (6.26)
Since Ω=−PV we arrive at the following relationship:
PV =
2
3
U .
66 CHAPTER 6. IDEAL GAS
This result holds also for Boltzmann distribution.
Using the substitution of the variables z =βε we observe that
Ω=−PV =VT
5/2
η(µ/T),
where η is a function of a single variable. Since
S
V
=−
1
V

∂Ω
∂T

V,µ
and
N
V
=−
1
V

∂Ω
∂µ

T,V
the ratio S/N depends only on the ratio µ/T.
If the process is adiabatic, then S = const, and µ/T must be constant. Consequently,
VT
3/2
= const , PV
5/3
= const , T
5/2
P = const.
The exponents, however, have nothing to do with the ratio of specific heats, c
V
/c
P
, since the
conditions c
P
/c
V
= 5/3 and c
P
−c
V
= 1 do not hold any more.
Equation of state
The equation of state can be written in a parametric form:
P(V, T)
D
0
T
5/2
=


0
z
3/2
dz
e
z−βµ(T,V)
±1
,
N(V, T)
VD
0
T
3/2
=


0
z
1/2
dz
e
z−βµ(T,V)
±1
.
We have to know some tips to calculate the involved integrals.
In the limiting case of the Boltzmann gas, e
βµ
<1, one can expand
1
e
z−βµ
±1
≈e
βµ−z

1∓e
βµ−z
±. . .

.
The lowest approximation leads to the Boltzmann value of the chemical potential and to Clapey-
ron’s equation PV = NT (Check!). The next term provides proper corrections. From Eq. (6.26)
we obtain
Ω≈Ω
Bol

1∓
1
2
5/2
e
βµ

=≈Ω
Bol
±
gVm
3/2
T
5/2
16π
3/2
¯ h
3
e
2βµ
where Ω
Bol
is the Boltzmann value for the Landau free energy. Substituting the Boltzmann value
for the chemical potential in the second term and using the concept of small increments (4.20)
one can express the free energy as
F = F
Bol
±
π
3/2
2g

N
2
¯ h
3
VT
1/2
m
3/2
.
6.3. FERMI AND BOSE GASES OF ELEMENTARY PARTICLES 67
Finally, differentiating this expression with respect to volume we arrive at the equation of state:
PV = NT
¸

π
3/2
2g
N¯ h
3
V(mT)
3/2
¸
. (6.27)
The conditions for the correction to be shall is naturally the same as for applicability of the
Boltzmann statistics. Quantum effects behave and an additional attraction for bosons and as an
additional repulsion for fermions.
A degenerate electron gas
For electron spin equals to 1/2, and g = 2s +1 = 2. At zero temperature, one can approximate
the Fermi function
f (ε) =
1
e
β(ε−µ)
+1

1 at ε < µ
0 at ε > µ
Thus there is a maximum energy
µ[
T=0
≡ε
F
= p
2
F
/2m.
Since
2
4πp
2
dp V
(2π¯ h)
3
=V
p
2
dp
π
2
¯ h
3
we have
N
V
=
p
3
F

2
¯ h
3
, → p
F
= (3π
2
)
1/3
(N/V)
1/3
¯ h, ε
F
= (3π
2
)
2/3
(N/V)
2/3
(¯ h
2
/2m).
Limiting the integration of the absolute value of the electron moments by p
F
we obtain
U
V
=
p
5
F
10mπ
2
¯ h
3
=
3(3π
2
)
2/3
10
¯ h
2
m

N
V

5/3
,
P =
(3π
2
)
2/3
5
¯ h
2
m

N
V

5/3
.
To calculate specific heat at lowtemperatures one has to be able to find temperature-dependent
corrections. Let us demonstrate the way to calculate these corrections. Consider
I =


0
f (ε)dε
e
β(ε−µ)
+1
,
where f (ε) is an arbitrary function such that the integral converges. Introducing z =β(ε−µ) and
remembering that β ≡1/T we get
I = T


−βµ
f (µ+Tz)
e
z
+1
dz = T

βµ
0
f (µ−Tz)
e
−z
+1
dz +T


0
f (µ+Tz)
e
z
+1
dz.
68 CHAPTER 6. IDEAL GAS
Now we can replace in the 1st integral
1
e
−z
+1
= 1−
1
e
z
+1
to obtain
I = T

βµ
0
f (µ−Tz)dz +T


0
f (µ+Tz) − f (µ−Tz)
e
z
+1
dz +


βµ
f (µ−Tz)
e
z
+1
dz.
The last term is exponentially small since βµ 1. Returning to the initial variables in the first
term, we obtain
I =

µ
0
f (ε)dε +T


0
f (µ+Tz) − f (µ−Tz)
e
z
+1
dz.
At low temperature one can expand the second integrand as
f (µ+Tz) − f (µ−Tz) = 2T f
/
(µ)z +
2
3!
T
3
f
///
(µ)z
3
+ .
As a result,
I =

µ
0
f (ε)dε +2T
2
f
/
(µ)


0
zdz
e
z
+1
+
1
3
T
4
f
///
(µ)


0
z
3
dz
e
z
+1
+
=

µ
0
f (ε)dε +
π
2
6
T
2
f
/
(µ) +

4
360
T
4
f
///
(µ) + . (6.28)
Now we can calculate temperature-dependent corrections to the thermodynamic quantities. Sub-
stituting f (ε) =ε
3/2
we get
Ω=Ω
0
−VT
2
m
3/2


6¯ h
3
.
Since the second termis a small correction we can replace µ →ε
F
in that term. Using the theorem
of small increments we find
F = F
0

1
2
γNT
2

V
N

2/3
, γ ≡

π
3

2/3
m
¯ h
2
,
S = γNT

V
N

2/3
,
C = γNT

V
N

2/3

T
ε
F
.
A degenerate Bose gas
The chemical potential of bosons is negative. It is given by the equation
N
V
=
g(mT)
3/2
2
1/2
π
2
¯ h
3
I (β[µ[), I (z) =


0

xdx
e
z+x
−1
. (6.29)
6.3. FERMI AND BOSE GASES OF ELEMENTARY PARTICLES 69
It increases with the temperature decrease and reaches zero at the temperature T
0
given by the
equation
N
V
=
g(mT
0
)
3/2
2
1/2
π
2
¯ h
3
I (0).
Since I (0) = (

π/2)ζ(3/2) where ζ(x) is the Riemann’s ζ-function,
T
0

3.31
g
2/3
¯ h
2
m

N
V

2/3
.
From Eq. (6.29) we obtain
(βT
0
)
3/2
=I (β[µ[)/I (0). (6.30)
The ratio F (z) ≡I (z)/I (0) is plotted in Fig. 6.2. The dependence of the product β[µ[ versus βT
0
can be easily found graphically from this plot.
0
0.2
0.4
0.6
0.8
1
0.5 1 1.5 2 2.5 3
z~
Figure 6.2: Plot of the ratio F (z) ≡
I (z)/I (0). Chemical potential can
be found by graphical solution of
Eq. (6.30).
What happens at T <T
0
? The above equation is based on replacement of the summation over
discrete states by the integration. This is not true for bosons because they can condense in the
ground state with ε
k
= 0. We omit this term in our integration because

ε
k
= 0 for the ground
state.
As a result, at T < T
0
the particles with ε > 0 are distributed according to the BE distribution
with µ = 0,
dn
ε
=
gm
3/2
V
2
1/2
π
2
¯ h
3

εdε
e
βε
−1
.
The total number of such particles is
N
ε>0
=
gm
3/2
V
2
1/2
π
2
¯ h
3


0
=

εdε
e
βε
−1
= N

T
T
0

3/2
.
70 CHAPTER 6. IDEAL GAS
The remaining
N
ε=0
= N
¸
1−

T
T
0

3/2
¸
particles are in the lowest state with zero energy. The energy and other thermodynamic quantities
can be obtained from general expression for Bose gas by putting µ = 0. In this way we obtain
(check!)
U = 0.770NT(T/T
0
)
3/2
,
C
V
= 5U/3T ,
F = U −TS =−(2/3)U =Ω (since) µ = 0,
P = −(∂F/∂V)
T
= 0.0851gm
3/2
T
5/2
/¯ h
3
. (6.31)
Thus at T ≤ T
0
the pressure is ∝ T
5/2
and it is volume-independent. The reason is that the
particles in the condensate do not contribute to the pressure.
6.4 Black-body radiation
Electromagnetic radiation which in a thermal equilibrium is called the black-body radiation. It
can be expressed as gas of non-interacting photons. To achieve the equilibrium, photons must
interact with some amount of matter which is able to absorb and emit photons. Thus we arrive
at the situation with variable number of photons, and the chemical potential µ should be fixed.
Since the free energy must be minimum
µ =

∂F
∂N

T,V
and
∂F
∂N
= 0
we obtain
µ = 0.
The photons are specified by the wave vector k (the momentum being ¯ hk) and frequencies ω
k
=
ck (the energy being ¯ hω
k
= ¯ hck), as well as by polarization. Since photons are transverse, there
are 2 directions of polarization, and the one obtains for the density of states
2 V
d
3
k
(2π)
3
= 2 V
4πk
2
k
(2π)
3
=V
ω
2

π
2
c
3
≡VD(ω)dω.
Here D(ω) =ω
2

2
c
3
. In this way we obtain
dn
ω
=V
ω
2
π
2
c
3
N(ω)dω
where
N(ω) ≡
1
e
β¯ hω
−1
6.4. BLACK-BODY RADIATION 71
is called the Planck distribution. The energy distribution is obtained by multiplication by ¯ hω
(Planck, 1900),
dE
ω
=V
¯ hω
3
π
2
c
3
1
e
β¯ hω
−1
dω,
its low-frequency asymptotics is the Rayleigh-Jeans law,
dE
ω
=V
T ω
2
π
2
c
3
dω,
while the high frequency one,
dE
ω
=V
¯ hω
3
π
2
c
3
e
−β¯ hω
dω,
is the Wien’s law.
The thermodynamic quantities can be derived in a usual way, e. g.,
F = T


0
D(ω) log

1−e
−β¯ hω

=−4σVT
4
/3c.
Here σ ≡ π
2
/60¯ h
3
c
3
is the Stefan-Boltzmann constant. It temperature is measured in degrees,
then
σ ≡π
2
k
4
/60¯ h
3
c
3
= 5.6710
−5
g/sec
3
deg
4
.
For other quantities we obtain,
S = 16σT
3
/3c, U = 4σVT
4
/c =−3F , C
V
= 16σVT
3
/c, PV =U/3 = 4σT
4
/c.
A body in thermal equilibrium with black-body radiation: Let us denote
E
ω
=
1
4πV
dE
ω

=
¯ hω
3

3
c
2

e
β¯ hω
−1

for the spectral density of black-body radiation per unit volume and unit solid angle. Then the
energy flux density leaving each point of the body is cE
ω
dodω. The incident energy flux on unit
area of the surface at an angle θ to the normal is therefore cE
ω
cosθdo, do = 2πsinθdθ. Let us
not define the absorbing power, i. e. fraction of the incident energy absorbed by the body, as
A(ω, θ). Then the energy absorbed per unit time and surface area is
cE
ω
A(ω, θ)cosθdodω.
For simplicity, we can also assume that the body does not scatter radiation and is not fluorescent,
i. e. that reflection occurs without change in the angle θ and in frequency ω. If we also assume
that the radiation does no pass through the body we arrive at the balance equation between the
radiation density and the intensity of emission J(ω, θ)dodω in the same direction:
J(ω, θ)
A(ω, θ)
= cE
ω
cosθ → a universal function of ω and θ.
72 CHAPTER 6. IDEAL GAS
This is the Kirchhof ’s law.
1
If A(ω, θ) = 1, i. e. if the body absorbs all the radiation, it is called
the black body. Then
J(ω, θ) = cE
ω
cosθ, → J
0
=

dodωJ(ω, θ) = cU/4V =σT
4
.
6.5 Lattice Vibrations. Phonons
For small displacement on an atom from its equilibrium position one can expand the potential
energy near its minimal value
V(R) = V(R
0
) +

dV
dR

R
0
(R−R
0
) +
1
2

d
2
V
dR
2

R
0
(R−R
0
)
2
++
+
1
6

d
3
V
dR
3

R
0
(R−R
0
)
3
+ (6.32)
If we expand the energy near the equilibrium point and denote

d
2
V
dR
2

R
0
≡C > 0,

d
3
V
dR
3

R
0
≡−2γ > 0
we get the following expression for the restoring force for a given displacement x ≡R−R
0
F =−
dV
dx
=−Cx +γx
2
(6.33)
The force under the limit F =−Cx is called quasi elastic.
One-Atomic Linear Chain
Dispersion relation
We start with the simplest case of one-atomic linear chain with nearest neighbor interaction.
If one expands the energy near the equilibrium point for the nth atom and use quasi elastic
approximation (6.33) he comes to the Newton equation (see Fig. /reff-ph1)
Figure 6.3:
m¨ u
n
+C(2u
n
−u
n−1
−u
n+1
) = 0. (6.34)
1
If there is scattering, then the Kirchhof’s law can be formulated only for integrals over frequencies and angles.
6.5. LATTICE VIBRATIONS. PHONONS 73
To solve this infinite set of equations let us take into account that the equation does not change
if we shift the system as a whole by the quantity a times an integer. We can fulfill this condition
automatically by searching the solution as
u
n
= Ae
i(qan−ωt)
. (6.35)
Immediately we get
ω =ω
m
sin
qa
2
, ω
m
= 2

C/m. (6.36)
The expression (6.36) is called the dispersion law. It differs from the dispersion relation for an
homogeneous string, ω = sq. The spectrum is shown in Fig. 6.4. Another important feature is
Figure 6.4: Dispersion law for vibrations
in a one-atomic chain. Periodic character
is clearly seen.
that if we replace the wave number q as
q →q
/
= q+
2πg
a
,
where g is an integer, the solution (6.35) does not change (because exp(2πi integer) = 1).
Consequently, it is impossible to discriminate between q and q
/
and it is natural to choose the
region

π
a
≤q ≤
π
a
(6.37)
to represent the dispersion law in the whole q-space. Note that there is the maximal frequency
ω
m
that corresponds to the minimal wave length λ
min
= 2π/q
max
= 2a. The maximal frequency
is a typical feature of discrete systems vibrations.
Now we should recall that any crystal is finite and the translation symmetry we have used
fails. The usual way to overcome the problem is to take into account that actual number L of
sites is large and to introduce Born-von Karmann cyclic boundary conditions
u
n±L
= n
n
. (6.38)
74 CHAPTER 6. IDEAL GAS
This condition make a sort of ring of a very big radius that physically does not differ from the
long chain.
2
Immediately, we get that the wave number q should be discrete. Indeed, substituting
the condition (6.38) into the solution (6.35) we get exp(±iqaL) = 1, qaL = 2πg with an integer
g. Consequently,
q =

a
g
L
, −
L
2
< g <
L
2
(6.39)
(it is convenient to consider L as a large even number). So, for a linear chain, the wave number q
takes L discrete values in the interval (−π/a, π/a). Note that this interval is just the same as the
Wigner-Zeitz cell of the one-dimensional reciprocal lattice.
Density of States
Because of the discrete character of the vibration states one can calculate the number of states,
D(ω), with different q in the frequency interval ω, ω+dω. One easily obtains (check!)
D(ω) =
2L
π
1

ω
2
m
−ω
2
. (6.40)
It is divergent at ω →ω
m
which is a feature of 1D systems. The density of states D(ω) is shown
in Fig. 6.5.
Figure 6.5: Density of vibrational states in a one-atomic
chain. There is a singularity at the maximal frequency that
is a typical feature of 1D systems.
Phase and Group Velocity
Now we discuss the properties of long wave vibrations. At small q we get from Eq. (6.36)
ω = sq, (6.41)
2
Note that for small structures of modern electronics this assumption need revision. Violation of this assumption
leads to the specific interface modes.
6.5. LATTICE VIBRATIONS. PHONONS 75
where
s = a

C
m
(6.42)
is the sound velocity in a homogeneous elastic medium. In a general case, the sound velocity
becomes q-dependent, i. e. there is the dispersion of the waves. One can discriminate between
the phase (s
p
) and group (s
g
) velocities. The first is responsible for the propagation of the equal
phase planes while the last one describes the energy transfer. We have
s
p
=
ω
[q[
= s

sin(aq/2)
aq/2

,
s
g
=


dq

= s[ cos(aq/2)[ . (6.43)
At the boundaries of the interval we get s
p
= (2/π)s while s
g
= 0 (boundary modes cannot
transfer energy).
Diatomic Chain. Acoustic and Optical branches.
We use this case to discuss vibrations of compound lattices. The corresponding picture is shown
in Fig. 6.6
Figure 6.6: On a two-atomic chain.
One can see that the elementary cell contains 2 atoms. If we assume the elastic constants to
be C
1,2
we come to the following equations of motion:
m
1
¨ u
n
= −C
1
(u
n
−v
n
) −C
2
(u
n
−v
n−1
),
m
2
¨ v
n
= −C
1
(v
n
−u
n
) −C
2
(v
n
−u
n−1
). (6.44)
It is natural to use once more the translation symmetry condition and search the solution as
u
n
= A
u
e
i(qan−ωt)
, v
n
= A
v
e
i(qan−ωt)
. (6.45)
After substitution to Eqs. (6.44) we get the set of equations for the constants A
i
. To formulate
these equations it is convenient to express these equations in a matrix form introducing the vector
A ≡

A
u
A
v

and the so-called dynamic matrix
ˆ
D =

C
1
+C
2
m
1

C
1
+C
2
e
−iaq
m
1

C
1
+C
2
e
iaq
m
2
C
1
+C
2
m
1

(6.46)
76 CHAPTER 6. IDEAL GAS
The equation for A has the form
ω
2
A−
ˆ
DA =
ˆ
0. (6.47)
This is homogeneous equation; it has a solution only if
det(ω
2
ˆ
1−
ˆ
D) = 0. (6.48)
This is just the equation which determines the eigenfrequencies. We get
Figure 6.7: Dispersion laws for
acoustic and optical phonons at
C
1
=C
2
=C.
ω
2
1,2
=
ω
2
0
2
¸
1∓

1−γ
2
sin
2
aq
2

(6.49)
where
ω
2
=
(C
1
+C
2
)(m
1
+m
2
)
m
1
m
2
, γ
2
= 16
¸
C
1
C
2
(C
1
+C
2
)
2
¸
m
1
m
2
(m
1
+m
2
)
2

.
The frequencies ω
1,2
are real because [γ[ ≤1. The spectrum is shown in Fig. 6.7. We see a very
important difference with the case of monoatomic chain: there are 2 branches ω
1,2
for a given
value of q.
The lower branch is called the acoustic branch while the upper one is called the optical
branch. To understand the physical reason for these names let us consider the limits of zero and
maximal q. We get
ω
ac
(0) = 0, ω
ac
(π/a) =
ω
0

2

1−

1−γ
2
,
ω
opt
(0) =ω
0
, ω
opt
(π/a) =
ω
0

2

1+

1−γ
2
. (6.50)
So, we have the inequality chain
ω
opt
(0) =ω
0

opt
(π/a) >ω
ac
(π/a) >ω
ac
(0) = 0.
6.5. LATTICE VIBRATIONS. PHONONS 77
Now we can discuss the structure of vibrations in both modes. From the dispersion equation
(6.47) we get
P
ac,opt
=

u
n
v
n

ac,opt
=
A
u
A
v
=
C
1
+C
2
e
−iqa
(C
1
+C
2
) −m
1
ω
2
ac,opt
. (6.51)
At very long waves (q →0) we get
P
ac
= 1, P
opt
=−
m
2
m
1
(6.52)
So, we see that in the acoustic mode all the atoms move next to synchronously, like in an acoustic
wave in homogeneous medium. Contrary, in the optical mode; the gravity center remains unper-
turbed. In an ionic crystal such a vibration produce alternating dipole moment. Consequently, the
mode is optical active. The difference between acoustic and optical modes is shown in Fig. 6.8.
Figure 6.8: Acoustic and optical branches of
vibrational spectrum.
Vibration modes of 3D lattices
Now we are prepared to describe the general case of 3D lattice. Assume an elementary cell with
s different atoms having masses m
k
. Then there are 3 acoustic modes with frequencies vanishing
at q →0. At low frequencies they correspond to acoustic waves in an elastic media. One mode is
a longitudinal, while two others are transverse ones. The rest 3s −3 ones are the optical modes.
Specific Heat of Crystal Lattice
From the classical statistical physics, the average kinetic energy per degree of freedom is
¯ ε
k
= T/2,
the total energy being
¯ ε = ¯ ε
pot
+¯ ε
kin
= 2¯ ε
kin
.
As a result, the total energy is
U = 3N
0
¯ ε
kin
= 3RT,
78 CHAPTER 6. IDEAL GAS
where R is the Rydberg constant while N
0
is the Avogadro one. As a result, the specific heat is
c
V
= 3R = 5.96 cal/K mole.
This relation is violated at low temperatures as far as the temperature becomes less than the so-
called Debye temperature (which is for most of solids is within the interval 100-400 K), namely it
decreases with the temperature decrease. To understand this behavior one should apply quantum
mechanics.
First let us recall the average energy for photons calculated previously. The average energy
is
¯ ε =
¯ hω
2
+ ¯ hω N(ω), N(ω) =
1
e
β¯ hω
−1
where N(ω) is the Planck function. The first item is energy-independent while the second one is
just the average energy of bosons with zero chemical potential.
Nowwe have to recall that the proper excitations in a lattice are collective vibrations, phonons,
the average energy being
U =

jq
¯ hω
j
(q)N[ω
j
(q)] =

j


0
D
j
(ω) ¯ hωN(ω)dω.
Here j is the number of the vibration branch.
For acoustic modes we use the so-called Debye model. Namely, we introduce the average
sound velocity s
0
as
1
s
3
0
=
1
3

1
s
3
l
+
2
s
3
t

and do the calculation of the density of states similar to the one for photons. As a result, we get
VD(ω) =
3V ¯ hω
2

3
s
3
0
≡V D
0
ω
2
.
The contribution of acoustic modes to the internal energy is
U
ac
=V D
0

ω
D
0

ω
3
e
β¯ hω
−1
where the so-called Debye frequency is determined from the condition of the total number of
modes to be equal to 3N for all acoustic branches,
3N =V D
0

ω
D
0
ω
2
dω.
From this equation
ω
D
= s
0


2
V
0

1/3
, q
D
=
ω
D
s
0
6.5. LATTICE VIBRATIONS. PHONONS 79
where V
0
is the cell volume. The order-of-magnitude estimate for the maximal wave vector q
D
is π/a. So according to the so-called Debye model all the values of q are confined in a sphere
with the radius q
D
. Usually, the Debye energy is introduced as
Θ = ¯ hω
D
= ¯ hs
0


2
V
0

1/3
.
The typical value of this temperature can be obtained from the rough estimate a =10
−8
cm, s
0
=
10
5
cm/s. We get ω
D
= 10
13
s
−1
, Θ= 100 K. For optical modes, let us assume that ω
j
(q) =ω
j0
,
i. e. that frequencies of optical phonons are q-independent. Then all the procedure is similar
for harmonic oscillators with fixed frequencies. It is conventional also to introduce the energies
corresponding to optical branches as Θ
j
= ¯ hω
j0
. These energies (in degrees) are of the order
of 10
2
-10
3
K. The quality of the Debye model is demonstrated in Fig. 6.9, where experimental
DOS for NaCl is shown together with the Debye approximation.
Figure 6.9: On the quality of the Debye ap-
proximation for the density of states.
Finally, we get the following expression for the internal energy
U =U
0
+NT
¸
3D

Θ
T

+
3s

j=4
Θ
j
/T
e
Θ
j
/T
−1
¸
(6.53)
where the Debye function D(z) is
D(z) =
3
z
3

z
0
x
3
dx
e
x
−1
. (6.54)
At high temperatures, T Θ
j
(and, consequently, T Θ) we get z <1 in Eq. (6.54) and then
expand the integrand in powers of x. We see that D(0) = 1. The item under the sum sign in Eq
(6.53) are equal to 1, and we get for the sum 3s−3. consequently, we get the classical expression
E =E
0
+3sNT
80 CHAPTER 6. IDEAL GAS
that leads to the classical expression for the specific heat. At low temperatures we immediately
see that optical modes give exponentially small contributions and we can discard them. At the
same time, we can replace the upper limit of the integral in (6.54) by infinity. Taking into account
that


0
x
3
dx
e
x
−1
=
π
4
15
we get
U =U
0
+
π
2
VT
4
10¯ h
3
s
3
0
that leads to the following expression for the specific heat
c
V
=
12π
4
5

T
Θ

3
.
The Debye model is very good for low temperatures where only long wave acoustic modes are
excited. The acoustic contribution to the specific heat can be expressed through the derivative of
the Debye function. We have
c
ac
3N
= 3

T
Θ

3

Θ/T
0
x
4
dx
e
x
−1
.
Temperature dependence of the specific heat according to the Debye model is shown in Fig. 6.10.
One can see that really the border between the classical and quantum region corresponds to
Figure 6.10: Temperature dependence of the
specific heat according to the Debye model.
T ≤ Θ/3 rather that to T ≤ Θ. The physical reason is strong frequency dependence of phonon
DOS. In real life, DOS behavior is much more complicated because of the real band structure
effects.
Chapter 7
Statistical ensembles
We have already discussed the Liouville theorem which can be expressed as
d f
dt
=
∂ f
∂t
+

i
¸
˙ q
i
∂ f
∂q
i
+ ˙ p
i
∂ f
∂p
i

= 0.
Because of the Hamilton equations, this equation can be re-written as
d f
dt
=
∂ f
∂t
+

i
¸
∂H
∂p
i
∂ f
∂q
i

∂H
∂q
i
∂ f
∂p
i


∂ f
∂t
+
¸
f , H
¸
= 0.
Here the Poisson brackets for the quantities A(p, q) and B(p, q) are defined as
¦A, B¦ ≡

i
¸
∂A
∂q
i
∂B
∂p
i

∂B
∂q
i
∂A
∂p
i

. (7.1)
Thus, for a stationary distribution we have ¦ f , H¦ = 0. We observe that any function of the
Hamiltonian H, f [H(p, q)], satisfies the Liouville equation and can be a proper stationary dis-
tribution. Different distributions correspond to different statistical ensembles/
7.1 Microcanonical ensemble
As an example, we have already considered the microcanonical ensemble. Let as define Γ(E, N,V)
as number of states with energies ≤E for given N and V. In the classical mechanics we get
Γ(E, N,V) ≡

0<H<E

(2π¯ h)
3(N
A
+N
B
+...)
N
A
!N
B
! . . .
=


(2π¯ h)
3(N
A
+N
B
+...)
N
A
!N
B
! . . .
Θ[E −H(p, q)] (7.2)
where
Θ(x) =

0, x < 0
1/2, x = 0
1, x > 0
(7.3)
81
82 CHAPTER 7. STATISTICAL ENSEMBLES
is the Heaviside unit-step function. The many-particle density of states is
D(E, N,V) =
∂Γ
∂E
=


(2π¯ h)
3(N
A
+N
B
+...)
N
A
!N
B
! . . .
δ[E −H(p, q)] . (7.4)
Consequently, the normalized distribution function is
f (p, q) =
1
D(E, N,V)
δ[E −H(p, q)] . (7.5)
Since the entropy of a normal system is given by Eq. (3.31), S(E, N,V) ≈ logΓ(E, N,V), we
obtain
1
T
=

∂E
logΓ =
1
Γ
∂Γ
∂E
=
D(E, N,V)
Γ(E, N,V)
.
Now we can re-derive the equipartition principle for ideal gas as follows. Consider the auxiliary
relationship,

p
n
∂H
∂p
n

=
1
D


(2π¯ h)
3N
N!
p
n
∂H
∂p
n
δ(E −H)
= −
1
D


(2π¯ h)
3N
N!
p
n

∂p
n
Θ(E −H) → integration by parts →
=
1
D


(2π¯ h)
3N
N!
Θ(E −H) =
Γ
D
= T .
Using this relation, we obtain
K =

p
2
n
2m

=
T
2
.
Another quantity which we shall need later is the so-called virial,
θ =−

n
q
n
˙ p
n
=

n
q
n
∂H
∂q
n
.
For a 3D system with harmonic potential energy, V ∝ ∑
ik
α
ik
q
i
q
k
, one can easily prove that
'θ` = 3NT .
Free particles: Let us demonstrate the general principles using the gas of free particles. We
have to be able to calculate the volume of the hyper-sphere,
V
n

n

i=1
dp
i
Θ

E −
3N

i=1
p
2
i
2m

= (2mE)
3N/2
V
n
, (7.6)
where
V
n
=


−∞

i
dx
i
Θ

1−

i
x
2
i

.
7.2. CANONICAL ENSEMBLES 83
To calculate V
n
let us introduce an auxiliary function
˜
V
n
(R) =


−∞

i
dx
i
Θ

R
2


i
x
2
i

, V
n
=
˜
V
n
(1).
consider an auxiliary integral
I
n


−∞
n

i=1
dx
i
e
−x
2
i

n/2
. (7.7)
This integral can be also written as
I
n
=

d
˜
V
n
e
−R
2
, R
2
=

i
x
2
i
. (7.8)
Since
˜
V
n
=V
n
R
n
we have d
˜
V
n
= nV
n
R
n−1
. Comparing Eqs. (7.7) and (7.8) we get
π
n/2
= nV
n


0
dRR
n−1
e
−R
2
→ V
n
=
π
n/2
(n/2)!
.
Consequently, we recover old results,
Γ(E) =
V
N
(2π¯ h)
3N
π
3N/2
(3N/2)!
(2mE)
3N/2
;
D(E) =
∂Γ
∂E
=
V
N
(2π¯ h)
3N
π
3N/2
(3N/2−1)!
(2mE)
3N/2−1
;
S(E) = N

5
2
+log
¸
V
N

mE
3πN¯ h
2

3/2
¸¸
,
etc. Then, we recover the equation of state.
7.2 Canonical ensembles
Suppose that there are other than the energy E integrals of motion,
C
i
(q, p) =C
i
= const .
Since ¦C, H¦ = 0 any function f [H(q, p),C
i
(q, p)] can be a proper stationary distribution func-
tion. Different choice of the system gives rise to different ensembles.
84 CHAPTER 7. STATISTICAL ENSEMBLES
Canonical ensemble
In this ensemble, the temperature, volume, and number of particles are fixed. Moving along the
lines of Gibbs’ method we have obtained the canonical distribution (Sec. 5.2)
f
l
=
1
Z
e
−βE
l
, Z =

s
e
−βE
s
.
In classical mechanics,
f (p, q) =
1
Z
e
−βH(p,q)
,
where Z is defined by the normalization condition. Here the temperature is fixed, while the
energy fluctuate around the average energy U. The fluctuations are given by the relation
(∆E)
2
≡ 'E
2
` −'E`
2
=
1
Z

s
E
2
s
e
−βE
s

¸
1
Z

s
E
s
e
−βE
s

2
=
1
Z

2
Z
∂β
2

1
Z
∂Z
∂β

2
=

∂β

1
Z
∂Z
∂β

=−
∂U
∂β
= T
2
C
V
.
Grand canonical ensemble
Now let as allow an exchange by particles with the surrounding medium. Along the Gibbs
method, we have obtained (see Sec. 5.2.1)
f
Nl
=
1
Ξ
e
β(µN−E
Nl
)
, Ξ =


N=0
e
βµN

l
e
−βE
Nl
=


N=0
e
βµN
Z
N
.
Again,
U =
1
Ξ

Ns
E
Ns
e
β(µN−E
Ns
)
=−
∂logΞ
∂β
;
S = −

Ns
f
Ns
log f
Ns
=−β(µ
¯
N−U) −logΞ.
Here we have introduced the average particle number,
¯
N ≡

Ns
N f
Ns
=
1
Ξ


N=0
Ne
βµN
Z
N
=
T
Ξ
∂Ξ
∂µ
. (7.9)
Introducing
Ω≡−T logΞ
we get
¯
N =−

∂Ω
∂µ

T,V
.
7.2. CANONICAL ENSEMBLES 85
Then is becomes obvious that Ω coincides with the Landau free energy since
Ω=U −TS−µ
¯
N =−PV .
It is easy to derive the fluctuation in the particle number (check!),
(∆N)
2
='N
2
` −
¯
N
2
=−T


2

∂µ
2

T,V
.
The latter expression can be converted into a more convenient form introducing the gas density,
ρ = N/V, using the expression for
¯
N,
ρ ≡
¯
N
V
=

∂P
∂µ

T
.
Then we obtain,
(∆N)
2
= TV

∂ρ
∂µ

T
= TV

∂ρ
∂P

T

∂P
∂µ

T
= TNρK
T
.
Poisson distribution. An instructive exercise is to find the distribution of the particle number
in a given volume V of an ideal gas. Since f
N
∝ e
βµN

l
e
−βE
Nl
∝ e
βµN
Z
N
we get
f
N
=
Z
N
Ξ
e
βµN
, Ξ =


N=0
e
βµN
Z
N
.
For an ideal gas,
Z
N
=
Z
N
1
N!
=
1
N!

V
Λ
3

N
, Ξ =


N=0
e
βµN
Z
N
= exp

V
Λ
3
e
βµ

.
Now, from Eq. (7.9) we have
¯
N =
∂logΞ
∂βµ
= (V/Λ
3
)e
βµ
→ Ξ = e
¯
N
, Z
N
1
e
βµN
=
¯
N
N
.
Finally,
f
N
=
¯
N
N
N!
e

¯
N
. (7.10)
that is the famous Poisson distribution.
Exercise. Find ∆N ≡

'∆N
2
` from the Poisson distribution. Derive an approximate expres-
sion for Eq. (7.10) at
¯
N 1, [N−
¯
N[ <
¯
N.
86 CHAPTER 7. STATISTICAL ENSEMBLES
A
h
Figure 7.1: On canonical ensemble with a fixed pressure.
Other ensembles
Consider the canonical ensemble with a fixed pressure rather that volume, the so-called pressure
ensemble. The system is shown in Fig. 7.1. Here the integral of motion is the enthalpy rather
that the energy. Then we come to the distribution function
f (q, p) =
1
Z
P
e
−βH
P
, H
P
=H(p, q) +
k
2
2M
+PAh.
Here k is the piston momentum while M is its mass. Z
P
is again the normalization factor. Simi-
larly, we can define
H =U +P
¯
V =−

∂β
logZ
P
,
¯
V =−

∂β
logZ
P
.
One can easily check that
G =−T logZ
P
.
Now we can integrate out the kinetic energy of a piston an replace dh →dV/A to get
dh

dk
2π¯ h
e
−βk
2
/2M
=
dV
V
M
, V
M
= AΛ
M
= A

2π¯ h
2
MT
.
As a result,
Z
P
=

dV
V
M
e
−βPV
Z
N
(V, T),
where Z
N
(V, T) is the partition function calculated for fixed volume and fixed number of parti-
cles. We observe that the pressure ensemble can be considered as Laplace-transformed canonical
ensemble for fixed volume. Since for an ideal gas Z
N
∝ V
N
the distribution function for the
volume occupied by the gas under a given pressure is
f
V
=CV
N
e
−βPV
,
where C is the normalization constant (find it and express through
¯
V!)
We can apply the above considerations to microcanonical and canonical ensembles to obtain
the relation
Z
N
(V, T) =

dE e
−βE
D(E),
7.3. ENSEMBLES IN QUANTUM STATISTICS 87
where D(E) is the partition function for the microcanonical ensemble.
To find the energy distribution in the canonical ensemble let us expand the function
u(E) ≡−log[D(E)e
−βE
] =βE −S(E)
up to the second order in the difference E −U and using the relation

2
S
∂E
2
=
∂β
∂E
=−β
2
/2C
V
we obtain the energy probability distribution as
f
E
∝ exp
¸

β
2
(E −U)
2
2C
V

, (∆E)
2
= T
2
C
V
. (7.11)
Exercise. Calculate the normalization constant.
7.3 Ensembles in quantum statistics
Pure and mixed ensembles
Let us consider a particle with spin S = 1/2. Quantizing spin along z axis, we find 2 basic states
[ ↑` =

1
0

, [ ↓` =

0
1

.
Hence,
ˆ
S
z
[ ↑` =
1
2
[ ↑`,
ˆ
S
z
[ ↓` =−
1
2
[ ↓`.
Consequently, the operator S
z
in this representation can be expressed as
ˆ
S
z
=
1
2

1 0
0 −1

,
while the conjugated wave functions are just the strings
'↑ [ = (1 0), '↓ [ = (0 1).
Consider now the the ensemble where the particles are quantized along different direction. Still
the state can be expanded in eigenfunctions as
[θ` = cos
θ
2
[ ↑` +sin
θ
2
[ ↓`.
Then we easily obtain,
'
ˆ
S
z
` =
1
2

cos
2
θ
2
−sin
2
θ
2

=
1
2
cosθ.
88 CHAPTER 7. STATISTICAL ENSEMBLES
This state corresponds to the system polarized along the axis
n = cosθe
x
+sinθe
y
.
Here e is the unit vector. The operator
ˆ
S
x
in the representation used has the form
ˆ
S
x
=
1
2

0 1
1 0

.
We can easily show that
'
ˆ
S
x
` =
1
2
sinθ.
The above example where all the spins are polarized along a given axis is called a pure state. If
the spins occupy different states with a given probability, the ensemble is called mixed. To see
the difference with a pure ensemble let us consider a system there the probabilities for the states
[ ↑` and [ ↓` are equal. Then,
'
ˆ
S
α
` =

i=↑,↓
w
i
'i[
ˆ
S
α
[i` = 0,
for any direction, '
ˆ
S n` = 0. For different probabilities w
i
this average is finite, but ≤1/2. The
maximum is reached only for a pure state where one of the probabilities is zero.
Now we can describe the general picture for a mixed ensemble. The ensemble average is now
'
ˆ
A` =

i
w
i
'i[
ˆ
A[i`.
Expanding the states [i` is terms of any orthonormal basis,
[i` =

α
c
i
[α`
we rewrite the average as
'
ˆ
A` =

i
w
i ∑
αβ
c

βi
c
αi
'β[
ˆ
A[α` ≡

αβ
A
βα
ρ
αβ
.
Here A
βα
≡'β[
ˆ
A[α`is the matrix element of the operator
ˆ
A, while
ρ
αβ
=

i
w
i
c

βi
c
αi
=

i
w
i
'α[i`'i[β`
is called the density matrix. It can be expressed as a matrix element of the density operator,
ˆ ρ =

i
w
i
[i`'i[ .
Now the average can be expressed as
'
ˆ
A` = Tr
ˆ
A ˆ ρ. (7.12)
For a pure state,
ˆ ρ =[ψ`'ψ[ , ˆ ρ
2
= ˆ ρ,
thus the density operator can have the eigenvalues 0 or 1.
7.3. ENSEMBLES IN QUANTUM STATISTICS 89
Density operator for spin 1/2
Any 22 matrix can be generally expanded in the unit matrix,
ˆ
1 and the Pauli matrices
ˆ σ
x
=

0 1
1 0

, ˆ σ
y
=

0 −i
i 0

, ˆ σ
z
=

1 0
0 −1

.
In general, we can write,
ˆ ρ =
1
2

ˆ
1+ ˆ σ P

,
where
ˆ
1 is the unit matrix, ˆ σ = ( ˆ σ
x
, ˆ σ
y
, ˆ σ
z
), while P is called the polarization vector. We can
easily show that Tr ˆ ρ = 1 (since Tr
ˆ
1 = 2 and Tr ˆ σ
i
= 0). For a pure state one has to require
P = cosθe
x
+sinθe
y
,
while for a mixed ensemble P =[P[ < 1. P is called the degree of polarization.
Density matrix for thermodynamic ensembles
Since
i ¯ h

∂t
[i` =
ˆ
H[i`
we can show that
d ˆ ρ
dt
=
∂ˆ ρ
∂t
+
i
¯ h

ˆ ρ,
ˆ
H

= 0.
Here [
ˆ
A,
ˆ
B] =
ˆ
A
ˆ
B−
ˆ
B
ˆ
A is the commutator which in classical physics corresponds to the Poisson
brackets. Now we can construct statistics similarly to the classical one replacing distribution
functions by density operators. In this way, for canonical ensemble we obtain
ˆ ρ =
1
Z
e
−β
ˆ
H
, Z = Tr e
−β
ˆ
H
=

α
e
−βE
α
.
The grand canonical ensemble is defined in a similar way,
ˆ ρ =
1
Ξ
e
−β(
ˆ
H−µ
ˆ
N)
.
For a microcanonical ensemble, in the representation of eigenstates [α
i
` belonging to ith level
the Hamiltonian can be written as
ˆ
H =

i
E
i
ˆ
Π
i
,
ˆ
Π
i
=

α
i

i
`'α
i
[ .
Since
ˆ
Π
i
ˆ
Π
j
= δ
i j
ˆ
Π
i
the projection operator Π
i
commutes with the Hamiltonian. Thus, for the
system having the energy E
i
ˆ ρ =
1
W
i
ˆ
Π
i
, W
i
= Tr
ˆ
Π
i
.
As usual,
S =−Tr ˆ ρ log ˆ ρ.
90 CHAPTER 7. STATISTICAL ENSEMBLES
Examples
Spin 1/2 in a magnetic field:
ˆ
H =−µ B =−µBσ
z
,
where µ is the magnetic moment. Since
e
−β
ˆ
H
=

n
(−β
ˆ
H)
n
n!
=

n
(βµBσ
z
)
n
n!
= cosh(βµB) +σ
z
sinh(βµB)
(here we employ the fact that σ
2k
z
= 1, σ
2k+1
z

z
) we obtain
ˆ ρ =
1
Z
e
−β
ˆ
H
=
1
2

ˆ
1+ ˆ σ
z
tanh(βµB)

, 'σ
z
` = Tr ˆ ρˆ σ
z
= tanhβµB.
A free particle in a box with volume V:
ˆ
H = ˆ p
2
/2m → 'r[ ˆ ρ[r
/
` =
1
Z

p,p
/
'r[p
/
`'p
/
[e
−βˆ p
2
/2m
[p`'p[r
/
`.
Since
'r[p` =
1

V
e
(i/¯ h)pr
, 'p
/
[e
−βˆ p
2
/2m
[p` =δ
pp
/ e
−βp
2
/2m
we get
'r[ ˆ ρ[r
/
` =
1
Z

d
3
p
(2π¯ h)
3
e
−βp
2
/2m+(i/¯ h)p(r−r
/
)
=
1
Z

m
2πβ¯ h
2

3/2
e
−(m/2β¯ h
2
)[r−r
/
[
2
.
From the normalization conditions we obtain
Z = Tr e
−β
ˆ
H
=

d
3
r 'r[e
−β
ˆ
H
[r` =
V
Λ
3
, 'r[ ˆ ρ[r
/
` =
1
V
e
−π[r−r
/
[
2

2
.
Chapter 8
Fluctuations
In this section we discuss some additional aspect regarding fluctuations.
8.1 The Gaussian distribution
Let us consider a closed system described by a microcanonical ensemble.
Assume that x is some quantity of the subsystem, ¯ x =0. At this point entropy has a maximum,
∂S
∂x
= 0,

2
S
∂x
2
< 0.
Thus one can expand
S(x) = S( ¯ x) −γ
x
2
2
,
the normalized distribution being
w(x)dx =

γ

e
−γx
2
/2
dx. (8.1)
As
'x
2
` =γ
−1
, (8.2)
we can write the result (the so-called Gaussian distribution) in the final form
w(x)dx =
1

2π'x
2
`
exp


x
2
2'x
2
`

dx. (8.3)
As x is small, for any quantity φ
'(∆φ)
2
` =
¸
d
2
φ
dx
2

x=0
'x
2
`.
91
92 CHAPTER 8. FLUCTUATIONS
For several (n) variables we have
w = Ae
−∑
ik
γ
ik
x
i
x
k
/2
,

w

i
dx
i
= 1.
The last equality determines the normalization constant A. In this way we get
w(x) =

γ
(2π)
n/2
exp


1
2

ik
γ
ik
x
i
x
k

, γ ≡det¦γ
ik
¦.
Now let us define thermal conjugates
X
i
=−
∂S
∂x
i
=

k
γ
ik
x
k
.
The averages 'x
i
X
k
` can be easily calculated:
'x
i
X
k
` =

γ
(2π)
n/2

l

x
i
γ
kl
x
l
exp


1
2

pq
γ
pq
x
p
x
q

n

s=1
dx
s
.
One can prove that it is equal to δ
ik
,
1
'x
i
X
k
` =δ
ik
.
From that we get
'x
i
x
k
` = (ˆ γ
−1
)
ik
, 'X
i
X
k
` =γ
ik
.
Thus
'(∆φ)
2
` =
∂φ
∂x
i
∂φ
∂x
k
(ˆ γ
−1
)
ik
.
For statistically-independent fluctuations
(ˆ γ
−1
)
ik
= 0.
Thus the distributions are uncoupled.
1
Proof: Assume for a moment that the average ¯ x
i
= x
i0
= 0. Then
x
i0
=

γ
(2π)
n/2


i
dx
i
x
i
e
−(1/2)∑
st
γ
st
(x−x
s0
)(x−x
t0
)
.
After differentiation of this equation with respect to x
x0
and putting x
i0
= 0, x
k0
= 0 we obtain the above result.
8.2. FLUCTUATIONS OF THERMODYNAMIC QUANTITIES 93
8.2 Fluctuations of fundamental thermodynamic quantities
As the change in the entropy is proportional to minimal work to change the thermodynamic
properties reversible the probability for a given configuration can be written as, see Eq. (4.29),
w ∝ e
−W
min
/T
∝ exp


∆U −T ∆S+P∆V
T

.
Regarding that all the quantities are the functions of S and V and expanding U in small deviations
∆S and ∆V from the equilibrium values,
∆U = T∆S−P∆V +
1
2
¸

2
U
∂S
2
(∆S)
2
+2

2
U
∂S∂V
∆S∆V +

2
U
∂V
2
(∆V)
2

we get
W
min
=
1
2
¸
∆S∆

∂U
∂S

V
+∆V∆

∂U
∂V

S

=
1
2
(∆S∆T −∆P∆V)
Thus
w ∝ exp

∆P∆V −∆S∆T
2T

. (8.4)
This is the general formula to find fluctuations of various thermodynamic quantities, and below
we give several examples in which different parameters are considered as independent.
Examples
Independent V and T:
∆S =

∂S
∂T

V
∆T +

∂S
∂V

T
∆V =
C
V
T
∆T +

∂P
∂T

V
∆V ,
∆P =

∂P
∂T

V
∆T +

∂P
∂V

T
∆V .
Substituting these expressions in Eq. (8.4), we get
w ∝ exp


C
V
2T
2
(∆T)
2
+
1
2T

∂P
∂V

T
(∆V)
2

.
Thus we reproduce previous results:
'∆T ∆V` = 0, '(∆T)
2
` = T
2
/C
V
, '(∆V)
2
` =−T(∂V/∂P)
T
. (8.5)
94 CHAPTER 8. FLUCTUATIONS
Independent P and S:
∆V =

∂V
∂P

S
∆P+

∂V
∂S

P
∆S ,
∆T =

∂T
∂S

P
∆S+

∂T
∂P

S
∆P =
T
C
P
∆S+

∂T
∂P

S
∆P.
As
dH = T dS+V dP
we have

∂V
∂S

P
=

2
H
∂P∂S
=

∂T
∂P

S
and
∆V =

∂V
∂P

S
∆P+

∂T
∂P

S
∆S.
Thus
w ∝ exp

1
2T

∂V
∂P

S
(∆P)
2

1
2C
P
(∆S)
2

.
As a result,
'∆S∆P` = 0, '(∆S)
2
` =C
P
, '(∆P)
2
` =−T

∂P
∂V

S
.
Fluctuations per particle: From Eq. (8.5) we get:

V
N

2
¸
=−
T
N
2
(∂V/∂P)
T
.
Putting V = const we obtain
'(∆N)
2
` =−
TN
2
V
2
(∂V/∂P)
T
This formula can be re-written as

N
2
V
2

∂V
∂P

T,N
= N


∂P
N
V

T,N
.
Then
N


∂P
N
V

T,N
=
N
V

∂N
∂P

T,V
=

∂N
∂P

T,V

∂P
∂µ

T,V
=

∂N
∂µ

T,V
.
Here we have employed the fact that the ratio N/V depends only on P and T (additivity principle).
The second equality,
N
V
=

∂P
∂µ

T,V
8.3. CORRELATION OF FLUCTUATIONS IN TIME 95
follows from the formula
dΩ=−V dP =−SdT −Ndµ.
Finally we arrive at a very useful formula
'(∆N)
2
` = T

∂N
∂µ

T,V
.
8.3 Correlation of fluctuations in time.
Langevin method
Consider correlation between the values of the quantity x ( ¯ x = 0) at different instants. Such
correlation is characterized by the average 'x(t)x(t
/
)`. Since under stationary external conditions
the average depends only on the difference t
/
−t one can define
φ(t) ≡'x(0)x(t)`.
It is clear from obvious symmetry considerations that
φ(t) =φ(−t).
A proper definition for quantum mechanical case is
φ(t −t
/
) =
1
2
' ˆ x(t) ˆ x(t
/
) + ˆ x(t
/
) ˆ x(t)`.
Now let us consider a state of partial equilibrium characterized by a given value x. For simplicity
let us also assume that x is a classical quantity.
Now we are prepared to discuss the state far from the equilibrium, namely assume that x

'x
2
`. If we assume that the state under consideration is determined only by the quantity x we
can suggest that the rate ˙ x for restoring the equilibrium is some function of one variable x. The
simplest function valid for small enough x is the linear one,
dx/dt =−λx, λ > 0. (8.6)
It is natural to assume that this equation is the same as the “equation of motion” for a macroscopic
quantity ¯ x. To express this concept in a more mathematical form let us define the quantity ξ
x
(t) as
the mean value of x at time t > 0 under condition that at t = 0 its value was x. The quantity ξ
x
(t)
is defined through the conditional probability P(x
/
, t
/
[x, t) to find the the value x
/
of the variable
x under condition that at time t it was x. We have,
ξ
x
(t) =

dx
/
x
/
P(x
/
, t
/
[x, t). (8.7)
Consequently,
φ(t) ='xξ
x
(t)` =

dxw(x)x ξ
x
(t), t > 0.
96 CHAPTER 8. FLUCTUATIONS
For the values ξ
x

'x
2
` we obtain

x
(t)/dt =−λξ
x
(t), t > 0.
Consequently,
ξ
x
(t) = xe
−λt
→ φ(t) ='x
2
`e
−λt

−1
e
−λt
, t > 0.
Here γ is the constant which enters the Gaussian distribution, see Eq. (8.2). In general, at
φ(t) =γ
−1
e
−λ[t[
. (8.8)
It the random quantity x is comparable with the mean square fluctuation one has to add the
random force y to the “equation of motion” (8.6) for the fluctuation x,
˙ x =−λx +y(t).
The correlation function 'y(0)y(t)` must be defined in a way to provide correct expression (8.8)
for 'x(0)x(t)` at large x. To do that we find a formal solution of the above equation with some
initial condition,
x(t) = e
−λt

t
t
0
dτe
λτ
y(τ).
The average 'x(t)x(t
/
)` is then
'x(t)x(t
/
)` = e
−λ(t+t
/
)

t
t
0
dt
1

t
/
t
0
dt
2
e
λ(t
1
+t
2
)
'y(t
1
)y(t
2
)`.
This result must be independent of t
0
provided λ(t −t
0
) 1 and λ(t
/
−t
0
) 1. Thus we can put
t
0
=−∞. The only way to obtain 'x(t)x(t
/
)` dependent only on [t −t
/
[ is to suggest that
'y(t
1
)y(t
2
)` =Cδ(t
1
−t
2
).
Substituting this expression into Eq. (8.8) for 'x(t)x(t
/
)` for large t we get
'y(t
1
)y(t
2
)` = (2λ/γ)δ(t
1
−t
2
).
The above result can be easily generalized for the case of several fluctuating quantities, x
k
.
Similarly, we define
φ
ik
(t
/
−t) ='x
i
(t
/
)x
k
(t)`, φ
ik
(t) =φ
ki
(−t).
However, there is one more symmetry property which is a consequence of the invariance with
respect to time reversal.
2
Because of the time-reversal symmetry,
φ
ik
(t) =φ
ik
(−t) =φ
ki
(t).
2
The system is assumed to be not in a magnetic field and not rotating as a whole.
8.3. CORRELATION OF FLUCTUATIONS IN TIME 97
There is an important remark here. We have tacitly assumed that the quantities x
k
are not affected
directly by time reversal. However, these quantities can change sign under time reversal, like the
quantities proportional to velocities. If both x
i
and x
k
possess such a property the above equation
is valid, of only one of two changes its sign, then there is an additional “-” sign.
The following generalization is straightforward. We introduce the matrix λ
ik
instead of the
scalar λ. The fluctuations x
k
are to be determined from the set of differential equations
˙ x
i
=−

k
λ
ik
x
k
+y
i
(t) (8.9)
with random forces y
i
defined by the correlation properties
'y
i
(t
1
)y
k
(t
2
)` = 2η
ik
δ(t
1
−t
2
), ˆ η =
ˆ
λˆ γ
−1
. (8.10)
Matrix elements of ˆ η are called the kinetic coefficients. Their physical meaning will be discussed
in the next section. The scheme discussed above is called the Langevin method. Again,
φ
ik
(0) = (ˆ γ
−1
)
ik
. (8.11)
Simple example: Brownian motion
Consider a particle moving in a liquid. It is subjected to collisions with liquid molecules. Due to
such collisions the particle’s velocity v changes in time according to the equation
m ˙ v
i
=−νv
i
+ f
i
(t)
where ν is the viscous friction coefficient while f
i
are random forces. Consequently,
ˆ
λ = (ν/m)δ
ik
, y
i
= f
i
/m, ˆ γ
−1
='v
2
i
(0)`δ
ik
=
1
3
'v
2

ik
.
Let us now consider a displacement x(t) = r(t) −r(0) and calculate 'x
i
(t)x
k
(t)` for t > 0. We
have
'x
i
(t)x
k
(t)` =

t
0
dt
1

t
0
dt
2
'v
i
(t
1
)v
k
(t
2
)` =δ
ik

t
0
dt
1

t
0
dt
2
'v
i
(t
1
)v
i
(t
2
)` ='x
2
i
(t)`δ
ik
.
Here we employ the fact the different components of the velocity are uncorrelated. Now we
employ the Langevin method to find the velocity correlation function
'v
i
(t
1
)v
i
(t
2
)` ='v
2
i
`e
−λ[t
1
−t
2
[
. (8.12)
As a result,
'x
2
i
(t)` = 'v
2
i
`

t
0
dt
1
¸

t
1
0
dt
2
e
−λ(t
1
−t
2
)
+

t
t
2
dt
2
e
−λ(t
2
−t
1
)

= 2'v
2
i

−1

t −λ
−1
+e
−λt

.
At large times, t λ
−1
we obtain (in general)
'x
2
i
(t)` = 2D[t[ , D ≡'v
2
i

−1
. (8.13)
The kinetic coefficient D is called the diffusion constant. We will discuss its physical meaning
later. At small times, 'x
2
i
(t)` ='v
2
i
`t
2
that corresponds to ballistic motion.
98 CHAPTER 8. FLUCTUATIONS
Onsager relations
The “equations of motion”
˙ x
i
=−

k
λ
ik
x
k
have a deep internal symmetry. To reveal it let is express the r.h.s. through the thermodynamically
conjugated quantities,
X
i
=−
∂S
∂x
i
,
which were introduced earlier. We have
'x
i
X
k
` =δ
ik
, 'x
i
X
k
` = (ˆ γ
−1
)
ik
, 'X
i
X
k
` =γ
ik
.
If x
i
are comparatively small,
X
i
=

k
γ
ik
x
k
, γ
ik

ki
.
The latest equation is a consequence of the fact that the coefficients are second derivatives of the
entropy. Then we get,
˙ x
i
=−

k
η
ik
X
k
, η
ik
=

l
λ
il
(ˆ γ
−1
)
lk
= (
ˆ
λˆ γ
−1
)
ik
.
The quantities η
ik
are called the kinetic coefficients, they are just responses to generalized forces.
According to the Onsager principle,
η
ik

ki
.
The proof of the principle is based on accurate application of time-reversal symmetry and the
property 'X
i
x
k
` =δ
ik
.
If the system is embedded into an external magnetic field H, or it rotates as a whole with the
angular velocity Ω then the time-reversal operation must include reversal of H or Ω. Then the
Onsager principle reads
η
ik
(H) =η
ki
(−H), η
ik
(Ω) =η
ki
(−Ω).
For proper application of the Onsager principle it is important to find proper thermodynamic
forces X
i
conjugated to the variables x
i
. A convenient way to do that is to study entropy genera-
tion. Indeed,
˙
S =

i
∂S
∂x
i
˙ x
i
=−

i
X
i
˙ x
i
. (8.14)
Thus, one can express the entropy generation through the heat release at a given temperature,
˙
Q,
as
˙
S =T
−1
˙
Q and then express it through the quantities x
i
in the form (8.14). An example is given
below.
8.3. CORRELATION OF FLUCTUATIONS IN TIME 99
Determination of proper generalized forces
According to the general principle, one has to find the entropy production rate,
˙
S, and express is
as −∑
i
˙ x
i
X
i
. Thus,
X
i
=−

˙
S
∂˙ x
i
.
Let us consider an example. An external electric field E =−∇ϕ (where ϕ is the electrochemical
potential) with combination of the temperature gradient create electric current with density j, as
well as energy current w. The currents are conventionally expressed as
j = σE−η∇T ,
˜ w ≡w−ϕj = γE−β∇T . (8.15)
Our aim is to determine the corresponding thermodynamic forces, X
j
and X
w
. Since the heat
release per unit volume consists of the Joule heating j E and of absorption −divw and in the
linear response approximation divw = div(w−ϕj) = div ˜ w,
3
we have
˙
S =−

div ˜ w
T
dV +

j E
T
dV
Let us assume that the total heat flux through the surface is zero. Then the first integral can be
calculated by integration by parts, and we have
˙
S =

dV
¸
˜ w ∇

1
T

+
j E
T

≡−

dV

˜ w X
w
+j X
j

.
Thus, the generalized forced conjugated to the currents j and w are:
X
j
=−E/T , X
w
=−∇(1/T) = T
−2
∇T . (8.16)
Compare equations (8.16) and (8.15) we arrive at the following matrix for kinetic coefficients:
ˆ η =

−σT −ηT
2
−γT −βT
2

.
As a result, we have derived an important relation
γ =ηT .
It can be shown that Onsager relations are also valid for the inverse matrix of kinetic coefficients,
ˆ
ζ ≡(
ˆ
λˆ γ
−1
)
−1
= ˆ γ
ˆ
λ
−1
.
Exercise:
1. Analyze Onsager relations between the kinetic coefficients given by the equations
E = ρj −α∇T ,
w−ϕj = Πj −κ∇T . (8.17)
2. Express the coefficients ρ, α, Π, κ through the quantities σ, η, γ, β and check the Onsager
relations explicitly.
3
Indeed, divϕj =ϕdivj +j ∇ϕ ≈ϕdivj = 0.
100 CHAPTER 8. FLUCTUATIONS
Dissipative function
The problem of dissipation cannot be solved generally within statistical physics. However, a
formulation becomes possible if one assumes that the state of the system is fully determined by
macroscopic coordinates, Q
i
, and by the velocities,
˙
Q
i
. That means that high-order derivatives
can be neglected. One has also to assume that the velocities are relatively small and high powers
of
˙
Q
i
can be neglected, and that the motions consist of small oscillations around some equilibrium
positions.
Under that assumptions both kinetic energy, K(
˙
Q
i
), and the potential one, V(Q
i
), are quadratic
functions of their arguments. One can derive the momenta in a usual way,
P
i
=
∂K(
˙
Q
i
)

˙
Q
i
.
Then we can express
˙
Q
i
in terms of P
i
. In the absence of dissipation we come to the usual
Hamilton equations of motion,
˙
Q
i
=
∂K(P
i
)
∂P
i
,
˙
P
i
=−
∂V(Q
i
)
∂Q
i
.
Now let us consider the quantities Q
i
and P
i
as fluctuating quantities x
k
. Since the change in the
entropy is equal to the minimal work, W
min
= K(P
i
) +V(Q
i
), divided by the temperature of the
thermal bath, T, the generalized forces are
X
Q
i
=
1
T
∂W
min
∂Q
i
=
1
T
∂V
∂Q
i
=−
P
i
T
,
X
P
i
=
1
T
∂W
min
∂P
i
=
1
T
∂K
∂P
i
=
Q
i
T
,
As a result, η
Q
i
P
i
=−η
P
i
Q
i
=−T. The coefficients satisfy the Onsager relations because one of
the quantities (P
i
) changes its sign under the time reversal.
Now we can formally write the equations of motion which will allow for dissipative pro-
cesses. In general, we can add to the r.h.s. of the Hamilton equations any linear combination
of X
P
i
, X
Q
i
which possess the proper symmetry. However, the equation for
˙
Q
i
should be left un-
changed because it is equivalent to the definition of momenta. On the other hand, the additional
terms to the equation for
˙
P
i
can contain only X
P
i
, otherwise the symmetry of kinetic coefficients
will be violated. In this way we can write
˙
P
i
=−
∂V(Q
i
)
∂Q
i


k
η
ik
∂K(P
k
)
∂P
k
=−
∂V(Q
i
)
∂Q
i


k
η
ik
˙
Q
k
, η
ik

ki
.
Now we can construct a quadratic form.
D =
1
2

ik
η
ik
˙
Q
i
˙
Q
k
8.4. FLUCTUATION-DISSIPATION THEOREM 101
which is called the dissipative function. The dissipative forces lead to the equation of motion
˙
P
i
=−
∂V(Q
i
)
∂Q
i

∂D(
˙
Q
i
)

˙
Q
i
.
The energy conservation law has to modified as
d
dt
(K+V) =

i
¸
∂K
∂P
i
˙
P
i
+
∂V
∂Q
i
˙
Q
i

=

i
˙
Q
i
¸
˙
P
i
+
∂V
∂Q
i

= −

i
˙
Q
i
∂D

˙
Q
i
=−2D.
Note that we applied the symmetry principle which has to be modified in an external magnetic
field. For that case out approach is strictly speaking not valid.
8.4 Spectral properties of fluctuations and fluctuation-dissi-
pation theorem
8.4.1 Classical systems
Let us start with a classical system and discuss properties of Fourier transformed fluctuation,
x
ω


−∞
x(t)e
iωt
dt → x(t) =


−∞
x
ω
e
−iωt


. (8.18)
If x(t) is real, then
x
−ω
= x

ω
. (8.19)
Since
φ(t
/
−t) =


−∞
dωdω
/
(2π)
2
e
−i(ωt+ω
/
t
/
)
'x
ω
x
ω
/ `,
and it must be a function of t
/
−t, we have 'x
ω
x
ω
/ ` ∝ δ(ω+ω
/
). The proportionality coefficient
is usually denoted as 2π(x
2
)
ω
, so
'x
ω
x
ω
/ ` = 2π(x
2
)
ω
δ(ω+ω
/
). (8.20)
Obviously,
φ(t) =



(x
2
)
ω
e
−iωt
,
'x
2
` = φ(0) =



(x
2
)
ω
. (8.21)
102 CHAPTER 8. FLUCTUATIONS
The quantity (x
2
)
ω
is called the spectral density of fluctuations. From the known expression (8.8)
for φ(t) we obtain
4
(x
2
)
ω
=

γ(ω
2

2
)
.
We can introduce also transform of random forces, y
ω
,
(y
2
)
ω
= (ω
2

2
)(x
2
)
ω
= 2λ/γ.
Again, the results can be expressed in the form of Langevin approach. For the correlation
function of the Fourier components we obtain:
'y

y

/ ` = 2π(y
i
y
k
)
ω
δ(ω+ω
/
), (y
i
y
k
)
ω

ik

ki
. (8.22)
Here ˆ γ is the matrix of kinetic coefficients.
Example: Fluctuations in one-dimensional oscillator.
Consider one-dimensional oscillator which is at rest is in in the equilibrium position Q = 0, but
able to perform small oscillations in some macroscopic coordinate Q. We write the potential
energy as U = (1/2)mω
2
0
Q
2
where m is the “mass” while ω
0
is the oscillator eigenfrequency.
The generalized moment is then P = m
˙
Q. Since the minimal work against the “spring” is equal
to the potential energy, W
min
= (1/2)mω
2
0
Q
2
,
f (Q) ∝ e
−W
min
/T
= e
−mω
2
0
Q
2
/2T
, ⇒ 'Q
2
` = T/mω
2
0
. (8.23)
If friction is present, the equations of motion acquire the form
˙
Q = P/m, (8.24)
˙
P = −mω
2
0
Q−γP/m, (8.25)
where −γP/m = −γ
˙
Q is the friction force. Now let us take Q and P as random quantities and
determine the conjugated thermodynamic forces, X
Q
and X
P
. They are determined by the entropy
generation, i.e. by the minimum work to bring the system from the equilibrium to the state with
given Q and P. We have
W
min
= P
2
/2m+mω
2
0
Q
2
/2.
Consequently,
X
Q
= T
−1
∂W
min
/∂Q = T
−1
∂U/∂Q = mω
2
0
Q/T , (8.26)
X
P
= T
−1
∂W
min
/∂P = T
−1
∂K/∂Q = P/m. (8.27)
Combining Eqs. (8.25) and (8.27) we get the ˆ η-matrix has the form
ˆ η =

0 −T
T γT

. (8.28)
4
Note that according to general principles of quasistationary fluctuations this expression is valid for ω less that
the inverse time for partial equilibrium to be established.
8.4. FLUCTUATION-DISSIPATION THEOREM 103
In order to apply these equations for fluctuations we rewrite Eq. (8.25) as
˙
P =−mω
2
0
Q−γP/m+y. (8.29)
Equation (8.24) is just a definition of the momentum must remain unchanged. According to
Eq. (8.22),
(y
2
)
ω
= 2η
22
= 2γT . (8.30)
To calculate the required fluctuation spectrum (Q
2
)
ω
we substitute P = m
˙
Q in (8.22), obtaining
m
¨
Q+γ
˙
Q+mω
2
0
Q = y. (8.31)
Multiplying by e
−ωt
and integrating over time, we find
(−mω
2
−iωγ +mω
2
0
)Q
ω
= y
ω
,
and finally,
(Q
2
)
ω
=
2γT
m
2

2
−ω
2
0
)
2

2
ω
2
. (8.32)
8.4.2 Quantum systems
Now let us turn to a quantum system. First, we have to remind some basic concepts of quantum
mechanics to be able consider time-dependent quantities.
Revisiting quantum mechanics
According to quantum mechanics, the state of the system can be described by a time-dependent
wave function Ψ(t) satisfying the Schr¨ odinder equation,
i ¯ h(∂Ψ/∂t) =HΨ (8.33)
where H is the Hamiltonian. The states ψ
n
which satisfy the stationary Schr¨ odinder equation,

n
(q) = E
n
ψ
n
(q), (8.34)
i.e. eigenstates of the Hamiltonian are called the stationary states. We will also use Dirac
notations,
ψ
n
(q) ≡[n`, ψ

(q) ≡'n[ .
Since eigenfunctions of the Schr¨ odinder equation form a complete set,

dqψ

n
(q)ψ
n
(q) = 1. (8.35)
For a stationary state,
Ψ
n
(t) = e
−(i/¯ h)E
n
t
ψ
n
(q). (8.36)
104 CHAPTER 8. FLUCTUATIONS
In general,
Ψ(t) =

n
a
n
Ψ
n
(t) =

n
a
n
e
−(i/¯ h)E
n
t
ψ
n
(q). (8.37)
The mean value of any physical quantity,
¯
f , represented by the operator
ˆ
f is then
¯
f (t) =

nm
a

n
a
m
f
nm
(t) (8.38)
where
f
nm
(t) =

Ψ

n
(t)
ˆ
f Ψ
m
(t). (8.39)
Now, let us assume that the operator
ˆ
f does not contain time explicitly. Then using Eq. (8.36)
we have,
f
nm
(t) = f
nm
e

nm
t
, (8.40)
where
f
nm
≡'n[
ˆ
f [m` =

dqψ

n
ˆ
f ψ
m
(q), ω
nm

E
n
−E
m
¯ h
. (8.41)
8.4.3 The generalized susceptibility
Let us now assume that the body is subjected to some perturbation,
ˆ
V(t) =−f (t) ˆ x.
Then the average value
¯ x(t) =


0
α(τ) f (t −τ)dτ. (8.42)
Consequently,
¯ x(ω) =α(ω) f (ω), (8.43)
where
α(ω) =


0
dt α(t)e
iωt
. (8.44)
α(ω) ≡α
/
(ω) +iα
//
(ω) is called the linear response function. Since ¯ x is real
α(ω) =α

(−ω), → α
/
(ω) =α
/
(−ω), α
//
(ω) =−α
//
(−ω).
For a periodic perturbation,
ˆ
V =−
1
2

f e
−iωt
+ f

e
iωt

ˆ x,
and
¯ x(ω) =
1
2

α(ω) f e
−iωt
+α(−ω) f

e
iωt

. (8.45)
Then in general the dissipation is
dE
dt
=

ˆ
H
∂t
=−¯ x
d f
dt
. (8.46)
8.4. FLUCTUATION-DISSIPATION THEOREM 105
Substituting ¯ x from Eq. (8.45),
f (t) =
1
2

f e
−iωt
+ f

e
iωt

(8.47)
and averaging over time we obtain
Q =

dE
dt

t
=
1
4


−α)[ f [
2
=
1
2
[ f [
2
ωα
//
(ω). (8.48)
Analytical properties of generalized susceptibility. Kramers-Kronig relations
Let us regard ω as a complex variable, ω = ω
/
+iω
//
and consider the properties of α(ω) in the
upper half of the ω-plane. From the definition (8.42) it follows that α(t) is finite at any positive t.
Consequently, α(ω) is a one-valued regular function in the upper half of the ω-plane. Indeed, for
ω
//
> 0 the integrand of Eq. (8.44) contains the factor e
−ω
//
t
and since α(t) is finite the integral
converges. The function α(ω) has no singularities at the real axis (ω
//
= 0), except possible at
the origin. Note that the definition (8.44) is not valid for the lower half of the ω-plane since the
integral diverges at ω
//
< 0.
The conclusion that α(ω) is regular in the upper half-plane is the consequence of the causality
principle. It is because of this principle the integration over time in Eq. (8.42) is taken only over
times previous to t. It is also evident from the definition (8.44) that
α(−ω

) =α

(ω). (8.49)
Hence for a pure imaginary frequency, ω = iω
//
, we have α(iω
//
) =α

(iω
//
), i.e. α(ω) is real.
The following theoremcan be proved: The function α(ω) does not take real numbers at any fi-
nite point of the upper half-plane except on the imaginary axis, where it decreases monotonously
from a positive value α
0
> 0 at ω = i0 to zero at ω = i∞. In particular, it follows that α(ω) has
no zeros in the upper half-plane. The proof is given in the book [1], '123.
Let us now derive a formula relating the real and imaginary part of α(ω). To do so, we choose
some real positive value ω
0
of ω, integrate the expression α(ω)/(ω−ω
0
) round the contour C
shown in Fig. 8.1, and then tend the outer contour radius R to infinity and the radii ρ of the inner
circles to zero.
ω
ω
o
0
C
R
ρ
Figure 8.1: On the derivation of analytical properties
of the susceptibility.
Since at infinity α →0, and the ratio α(ω)/(ω−ω
0
) tends to zero more rapidly than ω
−1
.
Since the function α(ω) is regular and possible divergence points ω=0 and ω=ω
0
are excluded,
the result must vanish,

C

α(ω)
ω−ω
0
= 0.
106 CHAPTER 8. FLUCTUATIONS
The integral over the infinite semicircle is also zero, so we are left with the integral over the real
axis and the parts around the ω=0 and ω=ω
0
. The first part in not important since α(0) is finite.
The integration around ω
0
yields −iπα(ω
0
) since the direction is clockwise. Consequently,
lim
ρ→0

ω
0
−ρ
−∞

α(ω)
ω−ω
0
+


ω
0


α(ω)
ω−ω
0

−iπα(ω
0
) = 0.
The first item is just the principle value of the integral, so we obtain
iπα(ω
0
) = P


−∞

α(ω)
ω−ω
0
. (8.50)
Here the variable of integration, ω has only real values. For convenience we replace the notations
as
ω →ξ, ω
0
→ω,
to obtain the famous Kronig-Kramers relations (1927):
α
/
(ω) =
1
π
P


−∞

α
//
(ξ)
ξ−ω
, (8.51)
α
//
(ω) = −
1
π
P


−∞

α
/
(ξ)
ξ−ω
. (8.52)
Since α
//
(ω) is the odd function, Eq. (8.51) can be rewritten as
α
/
(ω) =
2
π
P


0

ξα
//
(ξ)
ξ
2
−ω
2
. (8.53)
If the function α(ω) has a pole at ω = 0, near which α(ω) = iA/ω, then we get instead of
Eq. (8.52):
α
//
(ω) =−
1
π
P


−∞

α
/
(ξ)
ξ−ω
+
A
ω
. (8.54)
The Kramers-Kronig relations are very important since they allow to reconstruct the whole re-
sponse function from its real or imaginary part. This is a consequence of the causality principle.
Quantum expression for susceptibility
On the other hand,the probability for the transition n →m according to the Fermi golden rule is
w
mn
=

¯ h
[ f [
2
4¯ h
[x
mn
[
2
[δ(ω+ω
nm
) +δ(ω+ω
mn
)] . (8.55)
The total absorbed power due to all transitions from the state n is then
Q
n
=

m
¯ hω
mn
w
mn
=
πω[ f [
2
2¯ h

m
[x
nm
[
2
[δ(ω+ω
nm
) −δ(ω+ω
mn
)] . (8.56)
8.4. FLUCTUATION-DISSIPATION THEOREM 107
Comparing Eqs. (8.48) and (8.56) we write down the response function
5
α
//
for n-th state as
α
//
n
(ω) =
π
¯ h

m
[x
nm
[
2
[δ(ω+ω
nm
) −δ(ω+ω
mn
)] . (8.57)
This quantity should be averaged over the Gibbs distribution of the initial states occupations,
f
n
= Z
−1
e
−βE
n
, as
α
//
(ω) = Tr ˆ ρˆ α =

n
f
n
α
//
n
(ω) (8.58)
where f
n
are diagonal elements of the density matrix ˆ ρ = Z
−1
e
βH
, see Eq. (7.12). We have
α
//
(ω) =
π
¯ h

nm
( f
n
− f
m
)[x
nm
[
2
δ(ω+ω
nm
). (8.59)
Immediately, we get
α
/
(ω) =
1
¯ h

nm
[x
nm
[
2
f
n
− f
m
ω
mn
−ω
, (8.60)
or, in the complex form,
α(ω) =
1
¯ h

nm
[x
nm
[
2
f
n
− f
m
ω
mn
−ω−i0
, (8.61)
8.4.4 Fluctuation-dissipation theorem: Proof
According to quantum mechanics, one has to define a real quantity which is an average of a
Hermitian operator,
1
2
¦ˆ x
ω
, ˆ x
ω
/ ¦ ≡
1
2
¦ˆ x
ω
ˆ x
ω
/ + ˆ x
ω
/ ˆ x
ω
¦ .
Obviously, the correlation function must depend only on the time difference, and one can write
down the spectral correlator as
1
2
' ˆ x
ω
ˆ x
ω
/ + ˆ x
ω
/ ˆ x
ω
` ≡2π(x
2
)
ω
δ(ω+ω
/
). (8.62)
This relation can be regarded as a definition for (x
2
)
ω
. Using the definition, as well as the fact
that (x
2
)
ω
is real and is an even function of ω we get
φ(t) =


−∞


x
2
)
ω
e
−iωt
→ (x
2
ω
=


−∞
dt φ(t)e
iωt
.
In particular,
φ(0) ='x
2
` =


−∞


(x
2
)
ω
.
5
α
//
is the imaginary part of the susceptibility.
108 CHAPTER 8. FLUCTUATIONS
The aim of the present section is to relate the spectral correlation function (x
2
)
ω
to the dissi-
pation in the system. Let us consider a quantum body in a given state, say n. Then, the average
value to the correlator is its diagonal matrix element,
1
2
' ˆ x
ω
ˆ x
ω
/ + ˆ x
ω
/ ˆ x
ω
`
nn
=
1
2

m
[(x
ω
)
nm
(x
ω
/ )
mn
+(x
ω
/ )
nm
(x
ω
)
mn
] .
Here (x
ω
)
nm
is the Fourier transform of the nm matrix element of the operator ˆ x(t),
(x
ω
)
nm
=


−∞
dt x
nm
(t)e
iωt
=


−∞
dt x
nm
e
i(ω
nm
+ω)t
= 2πx
nm
δ(ω
nm
−ω).
Since x is real x
nm
= x

mn
and
δ(ω
nm
+ω)δ(ω
mn

/
) =δ(ω
nm
+ω)δ(ω+ω
/
)
we can extract 2πδ(ω+ω
/
) at get:
(x
2
)
ω


m
[x
nm
[
2
[δ(ω+ω
nm
) +δ(ω+ω
mn
)] . (8.63)
Again, the expression for the fluctuations should be averaged over the Gibbs distribution of the
initial state occupations, f
n
= Z
−1
e
−βE
n
. Averaging, we obtain
(x
2
)
ω
= π

nm
f
n
[x
nm
[
2
[δ(ω+ω
nm
) +δ(ω+ω
mn
)]
= π

nm
( f
n
+ f
m
)[x
nm
[
2
δ(ω+ω
nm
) =π

nm
f
n

1+e
β¯ hω
nm

[x
nm
[
2
δ(ω+ω
nm
)
= π

1+e
−β¯ hω


nm
f
n
[x
nm
[
2
δ(ω+ω
nm
). (8.64)
Now, let us note that Eq. (8.59) can be rewritten in the form
α
//
(ω) =
π
¯ h

1−e
−β¯ hω


nm
f
n
[x
nm
[
2
δ(ω+ω
nm
). (8.65)
Comparing Eqs. (8.64) and (8.65), we obtain
(x
2
)
ω
= ¯ hα
//
(ω)coth(β¯ hω/2) → 'x
2
` =
¯ h
π

α
//
(ω)coth(β¯ hω/2)dω. (8.66)
This is the famous fluctuation-dissipation theorem (Callen & Welton, 1951).
For small frequencies, when ¯ hω <T, one can approximate coth(β¯ hω/2) as
coth(β¯ hω/2) ≈2T/¯ hω.
As a result, instead of (8.66) we get
(x
2
)
ω
= T
α
//
(ω)
ω
. (8.67)
8.4. FLUCTUATION-DISSIPATION THEOREM 109
Exercise: Check that fluctuations in the one-dimensional classical oscillator (example in p. 102)
given by Eq. (8.32) obey the classical limit (8.66) of the fluctuation-dissipation theorem.
An important example is fluctuations of current density in a resistor. In this case x
i
= j
i
, α
//
=
ωℜσ(ω), where σ(ω) is the complex conductivity. Thus
( j
2
)
ω
= ¯ hωℜσ(ω) coth

¯ hω
2T

.
Note that all the concept is valid only for small fluctuations near the equilibrium state.
Exercise: 1) Find spectrum of fluctuations of the current I through the circuit shown in Fig. 8.2,
(I
2
)
ω
, for a fixed voltage V
12
between the points 1 and 2) Calculate the integrated spectrum,

(I
2
)
ω
(dω/2π) and compare it with the square of the average current, 'I`
2
.
1
2
L
R
C
Figure 8.2: On the current fluctuations. L is the induc-
tance, C is the capacitance, while R is the resistance.
The circuit is kept under constant voltage V
12
.
110 CHAPTER 8. FLUCTUATIONS
Chapter 9
Stochastic Processes
Now we discuss other aspects of random processes which take place in course of evolution of
nonequilibrium states to equilibrium.
9.1 Random walks
The simplest stochastic process is random walk in one dimension. A particle can make a step to
the right with the probability p or to the left with probability q = 1−p.
We are interested to find the displacement S after N 1 steps. It can vary between +N and
−N.
The probability of the walk with S = +N is P
N
(N) = p
N
. If there occurs 1 step to the left the
displacement is S = N −2, the probability being P
N
(N −1) = Np
N−1
q because the step to the
left can occur at any of the N different times (see Fig. 9.1). In general, the probability to find a
Figure 9.1: On the 1D random walk.
walk with R steps to the right from total N steps is
P
N
(R) =
N!
R!(N−R)!
p
R
q
N−R
. (9.1)
This distribution is called binomial or Bernoulli. Certainly
N

R=0
P
N
(R) = (p+q)
N
= 1.
111
112 CHAPTER 9. STOCHASTIC PROCESSES
As a result, we can easily calculate the average displacement
'S` = 'R` −(N−'R`) = 2'R` −N
= −N+2

R
N!
R!(N−R)!
Rp
R
q
N−R
= −N+2p

∂p

R
N!
R!(N−R)!
p
R
q
N−R
= −N+2

∂p
(p+q)
N
= N(p−q).
The symmetric random walk can be generated by tossing a coin. One makes N attempts. To
obtain the statistical average one has to make M 1 runs. This set forms a statistical ensemble,
the average being
'S` =
1
M
M

m=1
S
m
.
In general case the way is more difficult, one needs a random number generator.
One can also obtain
'R
2
` =

p

∂p

2
(p+q)
N
= (Np)
2
+Npq
to obtain
'S
2
` = 4'R
2
` −4N'R` +N
2
= N
2
(p−q)
2
+4Npq.
Thus
'(∆S)
2
` ='S
2
` −'S`
2
= 4Npq → = N (for p = q = 1/2).
For very large N the Bernoulli distribution P
N
(R) = (N!/R!(N −R)!)p
R
(1 − p)
N−R
can be ap-
proximated by the Gaussian distribution. Using the Stirling’s formula for n! we obtain
P
N
(R) =
1

2πNpq
e
−(R−Np)
2
/2Npq
.
Since R = (N+S)/2 we have
P
N
(S) =
1

8πNpq
e
−[S−N(2p−1)]
2
/8Np(1−p)
=
1

2πσ
e
−(S−
¯
S)
2
/2σ
2
.
Here
¯
S = N(p−q) = N(2p−1), σ =

'(∆S)
2
` = 4Np(1−p).
If the number of steps is very large then we can think that each step is very small and come to
continuous description. Assuming x = Sa and regarding the time step t = Nτ. Then we obtain
for the symmetric case the probability distribution
P(x, t) =

τ
2πa
2
t
e
−a
2
τ/2a
2
t
=
1

4πDt
e
−x
2
/4Dt
, D ≡
a
2

.
9.1. RANDOM WALKS 113
3
1
0.1
0.2
0.3
0.4
0.5
–3 –2 –1 1 2 3
x
Figure 9.2: P(x, t) as a function of x at Dt = 1 and 3.
This function is shown in Fig. 9.2 We immediately obtain


−∞
P(x, t) = 1, 'x` = 0, 'x
2
` = 2Dt .
The above expressions can be easily generalized for multidimensional space keeping in mind
that the probability is multiplicative. For 3D case we obtain,
P(r, t) =
1
(4πDt)
3/2
e
−r
2
/4Dt
, D =
a
2

.
We have
'x
i
(t)x
j
(t)` = 2δ
i j
D[t[ , 'r
2
` = 3'x
2
` = 6DT .
Now we can map the above results on the results on Brownian motion obtained in Sec. 8.3 by
the Langevin method. Indeed, we can assume that during the time λ
−1
between collisions the
displacement is a ∼v/λ. Thus,
D ='v
2
i
`λ ∼a
2
/λ.
9.1.1 Relation to diffusion
Consider a drop of colored dye in the water. The concentration of dye obeys the continuity
equation,
∂C
∂t
+divJ = 0
where J is the current of colored particles. It is driven by gradients, J =−DgradC. In this way
we get the diffusion equation
∂C
∂t
−D∇
2
C = 0. (9.2)
Let us consider the initial condition
C(r, 0) = Nδ(r).
114 CHAPTER 9. STOCHASTIC PROCESSES
Here N is the total number of particles in the drop. The solution with the above initial condition
can be easily obtained by the Fourier method. Since
C(r, t) =

d
3
k
(2π)
3
c(k, t)e
ikr
, and c(k, t) = Ne
−Dk
2
t
we obtain
C(r, t) =

N
d
3
k
(2π)
3
e
−Dk
2
t+ikr
=
N
(4πDt)
3/2
e
−r
2
/4Dt
.
To derive this formula it is convenient to use spherical coordinates and to direct the polar axis
along r. The integral over the angles is

π
0
sinθdθ


0
dφe
−kr cosθ
= 2π 2

1
0
dµ coskrµ = 4π
sinkr
kr
.
Then,


3


0
k
2
dk
sinkr
kr
e
−Dk
2
t
=
1
(4πDt)
3/2
e
−r
2
/4Dt
.
The relation between diffusion and random walk has been realized by A. Einstein in 1905.
His reasoning is like that.
Consider a thin layer between x and x +∆x. Let the number of particles in this layer at time
t will be denoted as N(x, t). If we introduce the probability P(a, τ)∆a, to move a distance falling
within a small interval between a and a+∆a during a short time interval τ, we can write
∆N(x, t +τ) = P(a, τ)∆a ∆N(x −a).
Since C =∆N/∆x we get an integral equation:
C(x, t +τ) =


−∞
daP(a, τ)C(x −a, t). (9.3)
Since a and τ are small one can expand the l.h.s. in powers of τ, while the r.h.s. in powers of a.
Using the equalities

daP(a, τ) = 1,

daaP(a, τ) = 0 (we assume symmetric distribution)
and the definition
1

daa
2
P(a, τ) ='x
2
(τ)`/2τ ≡D,
we arrive at the same 1D diffusion equation as for the Brownian motion (with correct coeffi-
cients).
9.2. RANDOM PULSES AND SHOT NOISE 115
F(t)
t
Figure 9.3: On the random pulse process.
9.2 Random pulses and shot noise
Random pulses, or pulsed random processes are encountered very often. In the simplest case the
pulses have the same shape, and the only random quantity is the instant at which a pulse starts,
Fig. 9.3. One well known example is the current in the anode circuit of the thermionic tube. Each
electron escapes the cathode and moves in the space charge region toward the anode producing
anode current i(t),

dt i(t) = e.
There are many examples of this type of process, e. g. impacts of the gas molecules on the wall
of the container. Let us denote the number of pulse as k, t
k
being its starting time. Then the
random quantity can be expressed as
x(t) =

k
F(t −t
k
; a
k
).
It is usually assume that
1. The random quantities t
k
and a
k
are statistically independent for different k, and their
distribution functions are k-independent.
2. The probability of the time t
k
being in the range between t and t +dt is constant and equal
to νdt.
Belowwe shall use the above example to introduce the important concept - characteristic function
of the random quantity. For a quantity x it is defined as
φ
x
(u) ≡'e
iux
` =


−∞
dx p(x)e
iux
.
Here p(x) is the probability distribution for x while the average is performed over realizations of
the random process. For our case
φ
x
k
(u) =

daw
a
(a)

1
t
m

t
m
/2
−t
m
/2
dt exp[iuF(t −t
k
; a)]

.
Here w
a
(a) is the distribution of the parameters a, while t
m
is the measuring time interval. If n
pulses took place during the time t
m
we obtain
φ
x
(u[n) = [φ
x
k
(u)]
n
116 CHAPTER 9. STOCHASTIC PROCESSES
which then must be averaged over the Poisson distribution,
P
n
=
¯ n
n
n!
e
−¯ n
, ¯ n =νt
m
.
Since
e
−¯ n

n
[ ¯ nφ
x
k
(u)]
n
n!
= exp¦¯ n[φ
x
k
(u) −1]¦ = exp¦νt
m

x
k
(u) −1]¦
we obtain for the well separated pulses
φ
x
(u) = exp

ν

daw
a
(a)


−∞
dt [exp(iuF(t, a)) −1]

.
Having the characteristic function we can calculate any moment of x:
'x` =

∂φ
x
(u)
∂iu

u=0

daw
a
(a)


−∞
dt F(t, a), (9.4)
'(δx)
2
` ='x
2
` −'x`
2
=


2
lnφ
x
(u)
∂(iu)
2

u=0

daw
a
(a)


−∞
dt F
2
(t, a), (9.5)
'δx(t
1
)δx(0)` =ν

daw
a
(a)


−∞
dt F(t; a)F(t +t
1
; a). (9.6)
The noise is conventionally measured as doubled Fourier component of the correlation function
(9.6). We obtain (check!):
S
x
(ω) = 2ν

daw
a
(a) [F(ω; a)[
2
.
Since F(0, a) ≡q where q is the charge of the particle, we get a very universal formula
S
x
(0) = 2q
2
ν → 2eI
where I is the electric current in the case of thermionic tube (Schottky, 1918).
9.3 Markov processes
Many stochastic processes belong to the class of Markov, or Markovian processes. To introduce
a definition let us consider a random function of time, x(t) which is known at n instants t
1
<
t
2
< . . . < t
n
. Let the time intervals be much greater than the microscopic time scale. We can
characterize the random process by the probability
w
n
(x
1
, t
1
; x
2
, t
2
; . . . ; x
n
, t
n
)dx
1
dx
n
that the random quantity at t
i
lies within the range (x
i
, x
i
+dx
i
). We can now relate w
n
to w
n−1
by the conditional probability P
n
(x
n
, t
n
[x
1
, t
1
; x
2
, t
2
; . . . ; x
n−1
, t
n−1
) as
w
n
(x
1
, t
1
; x
2
, t
2
; . . . ; x
n
, t
n
) = w
n−1
(x
1
, t
1
; x
2
, t
2
; . . . ; x
n−1
, t
n−1
)
P
n
(x
n
, t
n
[x
1
, t
1
; x
2
, t
2
; . . . ; x
n−1
, t
n−1
).
9.4. DISCRETE MARKOV PROCESSES. MASTER EQUATION 117
The process is called the Markov one if the conditional probability depends only on the value
x
n−1
at the last instant t
n−1
which precedes t
n
,
P
n
(x
n
, t
n
[x
1
, t
1
; x
2
, t
2
; . . . ; x
n−1
, t
n−1
) = P
2
(x
n
, t
n
[x
n−1
, t
n−1
) ≡P(x
n
, t
n
[x
n−1
, t
n−1
). (9.7)
It means that the system forgets all but the last previous values of the random process. Both
random walks and pulse processes are the Markov ones.
Let us consider a Markov process where the random variable ξ(t) acquires discrete value
¦x
k
¦. Then the conditional probability can be uncoupled as
P(x
k
, t[x
i
, t
0
) =

j
P(x
k
, t[x
j
, θ)P(x
j
, θ[x
i
, t
0
), t
0
<θ <t . (9.8)
This is the Markov equation. For a homogeneous random quantity ξ(t) depending in discrete
times t
n
the above equation acquires the form
P(x
k
[n−l, 0) =

j
P(x
k
[x
j
, n−m)P(x
j
[m−l, x
i
), l < m < n. (9.9)
9.4 Discrete Markov processes. Master equation
There is an important class of the processes where the random quantity acquires only discrete
values x
i
, (i = 1, 2, . . . , N). An example is the random telegraph noise produced by a defect
hopping between two states in the lattice, see Fig. 9.4.
x
1
x
2
x
Time
x
2
x
1
Figure 9.4: Random
telegraph noise in the
defect position.
In the discrete case one can introduce the probability w
i
to find the system in the i-th state, as
well as transition probability p
ik
. For two-state system we can denote, w
1
=w, p
12
= p, p
21
=q.
Since w
2
= 1−w
1
= 1−w we can write the master equation for the probabilities in the form
dw
dt
=−pw+q(1−w). (9.10)
The stationary solution of this equation is
w
0
=
q
p+q
, w
0
2
= 1−w
0
=
q
p+q
.
118 CHAPTER 9. STOCHASTIC PROCESSES
Consequently,
'x` =
qx
1
+ px
2
p+q
, '(δx)
2
` =
pq(x
2
−x
1
)
2
(p+q)
2
.
We can also obtain the correlation function 'δx(t)δx(0)`. To calculate the correlation function
one has to find conditional probability, P(x, t[x
1
, t
1
). Obviously, at t ≥t
1
it has the form
P(x, t[x
1
, t
1
) = w(x) +[δ(x −x
1
) −w(x)]g(t, t
1
)
where w(x) is the stationary probability to find the system in the state x, while g(t, t
1
) is the
probability that the state remains unchanged during the time between t
1
and t ≥t
1
. The quantity
g(t, t
1
) satisfies Eq. (9.10) with the initial condition
g(t, t
1
)[
t=t
1
+0
= 1. (9.11)
The physical meaning of g(t, t
1
) is the probability It is natural to look for a non-stationary solution
in the form
g(t, t
1
) = g
0
e
−λ[t−t
1
[
.
Substituting that into the master equation (9.10) we get λ = p+q, and from the initial condition
(9.11) g
0
= 1.
g(t, t
1
) = e
−(p+q)[t−t
1
[
.
Since at t −t
1
(p+q)
−1
we have 'x(t)x(t
1
)` = 'x`
2
, or '(δx)
2
` = 0, we obtain
'δx(t)δx(0)` ='(δx)
2
`e
−(p+q)[t[
=
pq(x
1
−x
2
)
2
(p+q)
2
e
−(p+q)[t[
.
Consequently the noise (doubled Fourier transform) is
S(ω) = (x
1
−x
2
)
2
2pq
(p+q)
2

1
π
p+q
(p+q)
2

2
.
Thus noise spectrum is a Lorenzian. The integral noise is


−∞
dωS(ω) = (x
1
−x
2
)
2
2pq
(p+q)
2
= 2'(δx)
2
`.
Let us for simplicity assume that p = q =λ/2, than
S(ω) =
(x
1
−x
2
)
2

λ
λ
2

2
.
In many random systems the fluctuation are induced by structural defects with different transition
rates λ. Usually logarithm of λ is distributed almost uniformly in a wide region between λ
min
and λ
max
, and for realistic values of ω the inequalities
λ
−1
max
<ω <λ
−1
min
(9.12)
9.5. CONTINUOUS MARKOV PROCESSES. FOKKER-PLANCK EQUATION 119
hold. Then the distribution of λ is
w(λ)dλ =
n
d
L

λ
, L ≡ln(λ
max

min
). (9.13)
Here n
d
is the density of the defects. An example leading to the distribution (9.13) is so-called
tunneling defects formed by an atom which can tunnel between two adjacent states in the mate-
rial. Since logarithm the tunneling probability is proportional to the tunneling distance and the
latter is uniformly distributed we arrive at the distribution (9.13).
Averaging the noise over the distribution (9.13) we get
S(ω) =
n
d
(x
1
−x
2
)
2
2L
1
ω
(9.14)
provided ω satisfies inequalities (9.12). Noise with the spectrum (9.14) is usually observed at
low-frequencies, it is called the flicker noise. Crossovers from random telegraph noise and flicker
noise were observed in point contacts between metals.
9.5 Continuous Markov processes. Fokker-Planck equation
Let us return to the problem of random walks and note that Eq. (9.3) can be reformulated for
probabilities. We get
P(x, t[x
0
, t
0
) =

dyP(x, t[y, θ)P(y, θ[x
0
, t
0
), t
0
<θ <t . (9.15)
This is the Smoluchowski equation. The instant θ can be arbitrary, so let us make it close to t.
Namely, let us put θ =t −τ and then tend τ to zero. We get
P(x, t[x
0
, t
0
) =

dyP(x, t[y, t −τ)P(y, t −τ[x
0
, t
0
). (9.16)
Now let us multiply this equation by some arbitrary function q(x) which has the properties
q(x) →0, q
/
→0 at x →±∞
and then integrate over x. We get


−∞
dxq(x)P(x, t[x
0
, t
0
) =


−∞
dyP(y, t −τ[x
0
, t
0
)


−∞
dxq(x)P(x, t[y, t −τ)
=


−∞
dyP(y, t −τ[x
0
, t
0
)


−∞
dxq(x)P(x, t[y, t −τ)

¸
q(y) +q
/
(y)(x −y) +q
//
(y)
(x −y)
2
2
=

.
Since


−∞
dxP(x, t[y, t −τ) = 1
120 CHAPTER 9. STOCHASTIC PROCESSES
we can replace in this integral y →x move it into l.h.s. Then we notice that
1
τ
[P(x, t[x
0
, t
0
) −P(x, t −τ[x
0
, t
0
)] ≈
∂P(x, t[x
0
, t
0
)
∂τ
.
As a result, we have


−∞
dxq(x)
∂P(x, t[x
0
, t
0
)
∂τ
=


−∞
dyP(y, t[x
0
, t
0
)
¸
q
/
(y)A(y, t) +q
//
(y)
B(y, t)
2
+

. (9.17)
Here
A(y, t) = lim
τ=0
'x −y`
τ
= lim
τ=0
1
τ


−∞
dx(x −y)P(x, t[y, t −τ), (9.18)
B(y, t) = lim
τ=0
'(x −y)
2
`
τ
= lim
τ=0
1
τ


−∞
dx(x −y)
2
P(x, t[y, t −τ). (9.19)
Now we replace y →x in the r.h.s. of Eq. (9.17) and make integration by parts. Since q(±∞) =
q
/
(±∞) = 0 and q(x) is arbitrary we finally obtain
∂P(x, t[x
0
, t
0
)
∂t
=−

∂x
A(x, t)P(x, t[x
0
, t
0
) +
1
2

2
∂x
2
B(x, t)P(x, t[x
0
, t
0
). (9.20)
This equation was derived in several contexts, so it has different names: Einstein-Fokker equa-
tion, Fokker-Planck equation, 2nd Kolmogorov equation. The diffusion equation (9.2) is a special
case with constant A and B.
The solution of Eq. (9.20) must be non-negative and satisfy the initial condition
P(x, t
0
[x
0
, t
0
) =δ(x −x
0
). (9.21)
The physical interpretation of Eq. (9.20) is the following. Let a stream (ensemble) of parti-
cles emerge form the point x
0
at time t
0
. Assume that they move independently. Then their
concentration at the point x at time t is just P(x, t[x
0
, t
0
). The the flow (mass current) S con-
sists of the “convection” part AP, where A is the convection velocity, and of the diffusion part
−(1/2)(∂BP/∂x) where B/2 has a meaning of the diffusion coefficient,
S = AP−
1
2
∂BP
∂x
. (9.22)
We see that Eq. (9.20) is equivalent to the particle conservation,
∂P
∂t
+
∂S
∂x
= 0.
Equation (9.22) is conventionally written as
S =

A−
∂B
∂x

P(x, t[x
0
, t
0
) −
B
2
∂P(x, t[x
0
, t
0
)
∂x
. (9.23)
The combination A−∂B/∂x is called the drift velocity while B/2 is the diffusion constant.
9.6. FLUCTUATIONS IN A VACUUM-TUBE GENERATOR 121
9.6 Fluctuations in a vacuum-tube generator
Consider a simplest generator of electric oscillations based on 3-electrode vacuum tube, see
Fig. 9.5, the notations are clear from the picture.
I
V
M
C
L
g
a
I
R
Figure 9.5: On the vacuum tube generator.
The tube acts as an amplifier, and the nonstationary part of the current is given by the expres-
sion
I
a
= SV
g

1−
V
2
g
2V
2
a

(9.24)
where V
0
is the DC anode voltage. The transformer facilitates feedback, such that
V
g
= MI
/
(9.25)
where M is the mutual inductance while prime means here the time derivative. Since the charge
at the capacitor C is just
˜
Q =

t
dt
/
(I
a
−I),
we have the following equation for the current through inductor L,
LI
/
+RI = C
−1

t
dt
/
[I
a
(t
/
) −I(t
/
)] → (9.26)
I
//
+
R
L
I
//
+
1
LC
I =
SM
LC
I
/

1−
M
2
(I
/
)
2
3V
2
0

. (9.27)
Now,it is convenient to introduce dimensionless variables,
τ ≡ω
0
t, I
0
≡2V
0

0
M, x ≡I/I
0
, ω
0
MS ≡µ, ω
0
(MS−RC) ≡ p,
where ω
0
= (LC)
−1
is the eigenfrequency of the circuit. Then Eq. (9.27) can be rewritten as
¨ x +x = µ˙ x

p−
4
x
˙ x
2

. (9.28)
122 CHAPTER 9. STOCHASTIC PROCESSES
Let us consider the situation when the feedback parameter µ is small and use the so-called method
of slow perturbations. Namely, the solution of Eq. (9.28) at µ = 0 is
x = r cos(τ +ϕ), (9.29)
and we will search the general solution in the form
x = r(τ) cosu(τ), u(τ) ≡τ +ϕ(τ), (9.30)
assuming the both r(τ) and ϕ(τ) are “slow” functions. What does it mean will be clear a bit later.
Then
˙ x = ˙ r cosu−r ˙ usinu,
¨ x =
d

(˙ r cosu−r ˙ usinu) ≈−r ¨ usinu−2˙ r ˙ usinu−r ˙ u
2
cosu.
Here we neglected ¨ r since µ <1. Now,
˙ u = 1+ ˙ ϕ, ˙ u
2
≈1+2˙ ϕ, ¨ u ≈0.
Then the l.h.s. of Eq. (9.28) is
−2˙ r sinu−2r ˙ ϕcosu.
Since the r.h.s. of this equation is proportional to µ we have to substitute the derivatives up to the
lowest order, as ˙ x =−r sinu to get
˙ x

1−
4
3
˙ x
2

=−pr sinu+
4
3
r
3
sin
3
u =−r(p−r
2
)sinu−
r
2
3
sin3u.
As a result we get,
−2˙ r ˙ usinu−2r ˙ ϕcosu =−µr(p−r
2
)sinu−µ
r
2
3
sin3u. (9.31)
Now we can average this equation over “rapid motion”. Namely, we multiply these equation by
sinu(τ) or by cosu(τ) and then integrate over τ from 0 to 2π assuming ϕ(τ) = const. From such
a procedure we get
˙ r =
µ
2
r(p−r
2
), ˙ ϕ = 0. (9.32)
So up to our approximation, ϕ is constant. In Fig. 9.6 the function r(p −r
2
) is plotted for
p =−1, 0, 1. In is clear that at p ≤0 there is only one stable point, r = 0. It can be shown that at
p > 0 the only stable point r =

p.
Now we are prepared to write down the Fokker-Planck equation for conditional probability
P(r, ϕ, τ[r
0
, ϕ
0
, τ
0
). In general, when we have several variables, the conditional probability can
be expressed as a function of vectors x =¦x
α
¦ and x
0
=¦x

¦. In a straightforward way we get
vector A and tensor
ˆ
B as
9.6. FLUCTUATIONS IN A VACUUM-TUBE GENERATOR 123
0
1
–1
–0.6
–0.4
–0.2
0
0.2
0.4
0.6
0.2 0.4 0.6 0.8 1 1.2 1.4
Figure 9.6: On the sta-
tionary points. The
function r(p−r
2
) vs. r
is plotted.
A(y, t) = lim
τ=0
'x−y`
τ
= lim
τ=0
1
τ


−∞
(dx)(x−y)P(x, t[y, t −τ), (9.33)
ˆ
B(y, t) = lim
τ=0
'(x
i
−y
i
)(x
k
−y
k
)`
τ
= lim
τ=0
1
τ


−∞
(dx)(x
i
−y
i
)(x
k
−y
k
)P(x, t[y, t −τ). (9.34)
The Fokker-Planck equation acquires the form
∂P(x, t[x
0
, t
0
)
∂t
=−divA(x, t)P(x, t[x
0
, t
0
) +
1
2

ik

2
∂x
i
∂x
k
B
ik
(x, t)P(x, t[x
0
, t
0
). (9.35)
Below we shall use a simple model of isotropic random forces with
B
rr
= r
2
B
ϕϕ
= µB B

= 0.
In our case,
A
r
= ˙ r =
µ
2
r(p−r
2
), A
ϕ
= 0,
and we get
µ
−1
∂P
∂t
=−
µ
2
∂[r(p−r
2
)P]
∂r
+
B
2
¸

∂r

r

∂r
P
r

+
1
r
2

2
P
∂ϕ
2

. (9.36)
It should be solved with the initial conditions
P(r, ϕ, t
0
[r
0
, ϕ
0
, t
0
) =δ(r −r
0
)δ(ϕ−ϕ
0
).
124 CHAPTER 9. STOCHASTIC PROCESSES
Let us start from the stationary distribution
w(r, ϕ) = lim
t−t
0
→∞
P(r, ϕ, t[r
0
, ϕ
0
, t
0
)
for which the l.h.s. of previous equation is zero. If ϕ is chosen in the interval (0, 2π) all the
values of ϕ are equal, and the solution is ϕ-independent. Then the equation (9.36) acquires the
form
1
2

∂r

r(p−r
2
)w−Br
d
dr
w
r

= 0 or r(p−r
2
)w−Br
d
dr
w
r
= const . (9.37)
The only solution which is regular at r = 0 implies that the constant is equal to 0, and
w(r) ∝ r exp
¸
1
B

r
0
ρdρ(p−ρ
2
)

=Cr exp


(p−r
2
)
2
4B

. (9.38)
The proportionality coefficient is given by the normalization condition. Since w is ϕ-independent
C
−1
= 2π

rdr exp


(p−r
2
)
2
4B


3/2

B
¸
1+erf

p
2

B

.
Here
erf x =
2

π

x
0
e
−t
2
dt . (9.39)
The profile w(r) has a maximum at
r
m
=

1
2
+

p
2
4
+B ≈


p(1+B/2p
2
) for p 2

B,

B/[p[ for p <−2

B.
The function w(r) is plotted in Fig. 9.7.
At negative x,
erf (−x) =−1+
e
−x
2
[x[

π
−. . . .
One can also neglect r
4
in the exponent of Eq. (9.38. Then we obtain
w(r, ϕ)dr dϕ =
[p[
B
e
−r
2
[p[/2B
rdr


, 'r
2
` = 2B[p[ (9.40)
which is called the Rayleigh distribution.
If the generator is strongly excited, p 2

B, we can put r =

p +ρ and neglect ρ
2
in
Eq. (9.38). Then we obtain,
w(r, ϕ)dr dϕ =

p
πB
e
−ρ
2
p/B



, 'ρ
2
` = B/2p, (9.41)
i.e. the Gaussian distribution.
9.6. FLUCTUATIONS IN A VACUUM-TUBE GENERATOR 125
0
–2
2
0
0.2
0.4
0.5 1 1.5 2
Figure 9.7: Plots of the distribution w(r)
for different values of p/

B, shown near
the curves.
To find the generation spectrum let us consider the case of large amplitudes,
r
m
=

p


2
` =

B/2p.
In this case we can linearize the dynamic equations (9.32) with respect to ρ = r −

p to get
˙ ρ =−pρ, ˙ ϕ = 0. (9.42)
The simplified Fokker-Planck equation acquires the form
µ
−1
∂P
∂t
= p
∂ρP
∂ρ
+
B
2


2
P
∂ρ
2
+
1
p

2
P
∂ϕ
2

. (9.43)
This equation we can analyze for arbitrary initial conditions. Denoting P(ρ, φ, t[ρ
0
, φ
0
, t
0
) ≡
P(ρ, φ[ρ
0
, φ
0
, t −t
0
) we immediately get that fluctuations of ρ and ϕ are statistically-independent,
P(ρ, φ[ρ
0
, φ
0
, t −t
0
) = P
ρ
(ρ[ρ
0
, t −t
0
)P
ϕ
(ϕ[ϕ
0
, t −t
0
). (9.44)
The fluctuations in the amplitude are Gaussian,
P
ρ
(ρ[ρ
0
, t −t
0
) =
1

2πσ
e
−(ρ−¯ ρ)
2
/2σ
2
(9.45)
with
¯ ρ =ρ
0
e
−µp(t−t
0
)
, σ
2
=
B
2p

1−e
−2µp(t−t
0
)

. (9.46)
The ¯ ρ(t) is just the dependence due to mechanical equations of motion (9.32) linearized near
r =

p. The solution for P
ϕ
(ϕ[ϕ
0
, t −t
0
) has the form
P
ϕ
(ϕ[ϕ
0
, t −t
0
) =
1

¸
1+2


n=1
e
−n
2
(B/2p)µ(t−t
0
)
cosn(ϕ−ϕ
0
)
¸
. (9.47)
126 CHAPTER 9. STOCHASTIC PROCESSES
As t −t
0
increases, the average deviation from the stationary solution, ¯ ρ, decays while dispersion
σ grows. One can easily check that the t −t
0
the system tends to stationary distributions. At
t =t
0
+0 it gives proper initial conditions since
1+


n=1
cosn(ϕ−ϕ
0
) = 2πδ(ϕ−ϕ
0
).
Having the conditional probabilities (9.46) and (9.47) one can find all necessary correlation func-
tions. Indeed, the joint probability, w(ρ
τ
, ρ, τ, to find the values ρ and ρ
τ
after time delay τ is
w(ρ
τ
, ρ, τ) = w(ρ)P(ρ
τ
[ρ, τ) =
1
2πσ

σ
τ
exp


ρ
2

2



τ
− ¯ ρ
τ
)
2

2
τ

, (9.48)
w(ϕ
τ
, ϕ, τ) = w(ϕ)P(ϕ
τ
[ϕ, τ) =
1

2
¸
1+2


n−1
e
−n
2

cosn(ϕ
τ
−ϕ)
¸
. (9.49)
Here
¯ ρ
τ
=ρe
−pµτ
, σ
2
τ
=
B
2p

1−e
−2pµτ

, D = µB/2p.
using these equations let us calculate the correlation function
ψ
τ
≡'x(0)x(τ)`. (9.50)
Lets us for brevity denote x(0) = x, x(τ) = x
τ
. Since
x(t) = [

p+ρ(t)] cos[t +ϕ(t)]
and 'x(t)` = 0 due to the facts that 'ρ`=0, ρ and ϕ are statistically independent at any time, and
the phase distribution is uniform. Hence 'cosϕ` ='sinϕ` = 0. As a result,
ψ
τ
= '(

p+ρ)(

p+ρ
τ
)`'cos(t +ϕ)cos(t +τ +ϕ
τ
)`
=
1
2
(p+'ρρ
τ
`)['cos(τ +ϕ
τ
−ϕ)` +'cos(2t +τ +ϕ
τ
+ϕ)`] . (9.51)
Then, from Eq. (9.48) we get,
'ρρ
τ
` =


τ
dρρ
τ
ρw(ρ
τ
, ρ, τ) = De
−pµ[τ[
. (9.52)
Since the joint probability w(ϕ
τ
, ϕ, τ), Eq. (9.49) is an even function of ϕ
τ
−ϕ one can easily
show that
'cos(ϕ
τ
+ϕ)` ='sin(ϕ
τ
±ϕ)` = 0.
As a result, 'cos(τ +ϕ
τ
−ϕ)` ='cos(ϕ
τ
−ϕ)` cosτ. Then,
'cos(ϕ
τ
−ϕ)` =
1

2


0

τ


0
dϕ cos(ϕ
τ
−ϕ)
¸
1+2


n=1
e
−n
2
Dµ[τ[
cosn(ϕ
τ
−ϕ)
¸
= e
−Dµ[τ[
. (9.53)
9.6. FLUCTUATIONS IN A VACUUM-TUBE GENERATOR 127
Collecting all the factors we finally obtain:
ψ
τ
=
1
2

p+De
−pµ[τ[

e
−Dµ[τ[
cosτ. (9.54)
The first factor in given by the fluctuations in the amplitude while the second is due to the phase
fluctuations. The oscillation spectrum is given by Fourier transform of the above function. In the
dimensional variables, τ →ω
0
τ, we get
ψ
ω
=Φ(ω−ω
0
) +Φ(ω+ω
0
) ≈Φ(ω−ω
0
)
with
Φ(α) =
p

¸
D
α
2
+D
2
+
D
h

D+h
α
2
+(D+h)
2

. (9.55)
Here
D = µDω
0
, h = µpω
0
are dimensional phase diffusion constant and increment. That means a superposition of a narrow
line with the half-width D with a broad line with the half-width D+h.
128 CHAPTER 9. STOCHASTIC PROCESSES
Chapter 10
Non-Ideal Gases
10.1 Deviation of gases from the ideal state
We start from the case of a weakly non-ideal gas where the only pair interaction is important.
The simplest case is the monoatomic gas where the interaction potential depends only on the
distance between the atoms. Since
H(p, q) =

a
p
2
a
2m
+V
we can write the free energy as
F = F
id
+F
i
,
F
i
= −T log
¸
1
V
N

e
−βV(q)

a
dV
a

.
Here F
id
is the free energy of the ideal gas. The factor V
−N
in the argument of the logarithm is
just because for the ideal gas the integral over the volume is V
N
. It is convenient to rewrite the
above expression in the form
F = F
id
−T log
¸
1+
1
V
N

e
−βV(q)
−1

dV
1
. . . dV
N

.
Now let us make use of the assumption that the gas is rarefied and only two atoms can collide at
the same time. Then the integrand is not small only if some two atoms are close together. Such
a pair can be extracted in N(N−1)/2 ways. Thus the integral can be expresses as
N(N−1)
2V
N

e
−βV
12
−1

dV
1
. . . dV
N

N
2
2V
2

e
−βV
12
−1

dV
1
dV
2
.
Here V
12
is the pair interaction potential. Since log(1+x) ≈x at x <1 and the gas density N/V
is assumed to be small,
F
i
=−
TN
2
2V
2

e
−βV
12
−1

dV
1
dV
2
.
129
130 CHAPTER 10. NON-IDEAL GASES
Since the interaction depends only on the distance between the atoms we can integrate out the
center of mass and get V. In this way we are left with the expression
F
i
=
TN
2
V
B(T), B(T) =
1
2

1−e
−βV
12

dV ,
where dV stands for relative coordinates. Since P =−∂F/∂V we obtain the following correction
to the equation of state,
P =
NT
V
¸
1+
N
V
B(T)

.
The corrections to other thermodynamic potentials can be found using the principle of small
increments. For example, G
i
= NBP.
A typical dependence of the interaction potential energy is shown in Fig. 10.1
2 1.8 1.6 1.4 1.2 1 0.8
V
3
2
1
0
-1
2r
0
Figure 10.1: A typical dependence of the nor-
malized potential energy V/V
0
versus the nor-
malized interatomic distance r/r
0
.
Expansion in powers of density
In principle, the equation of state can be expressed in the form of expansion in the density,
P = T


n=1

N
V

n
B
n
(T), B
1
= 1
where coefficients B
n
(T) are called the virial coefficients. To determine them it is convenient to
use the Landau fee energy, Ω=−PV,
e
βΩ
=


N=0
1
N!
e
βµN


N
e
−βH
N
(p,q)
.
For the N-particle Hamiltonians we have,
H
1
=
p
2
2m
, H
2
=

a=1,2
p
2
a
2m
+V
12
, H
3
=
3

a=1
p
2
a
2m
+V
123
, . . .
where in general
V
123
=V
12
+V
13
+V
23
.
10.1. DEVIATION OF GASES FROM THE IDEAL STATE 131
If one denotes
ζ ≡
e
βµ
(2π¯ h)
3

d
3
pe
−βp
2
/2m
=

mT
2π¯ h
2

3/2
e
βµ
the expression for Ω can be written as
Ω=−PV =−T log
¸
1+ζV +
ζ
2
2!

e
−βV
12
dV
1
dV
2
+
ζ
3
3!

e
−βV
123
dV
1
dV
2
dV
3
+. . .

.
Again, we can integrate out the center-of-mass motion to get
Ω=−PV =−T log
¸
1+ζV +
ζ
2
V
2!

e
−βV
12
dV
2
+
ζ
3
V
3!

e
−βV
123
dV
2
dV
3
+. . .

.
We arrive at the expansion in powers of ζ
P = T


n=1
J
n
n!
ζ
n
with
J
1
= 1, J
2
=

e
−βV
12
−1

dV
2
J
3
=

e
−βV
123
−e
−βV
12
−e
−βV
23
−e
−βV
13
+2

dV
2
dV
3
, . . .
The structure of the integrals J
n
is clear, they are not small only if n particles are close together.
To obtain equation of state one has to relate the parameter ζ to the density. Since
N =−

∂Ω
∂µ

T,V
=V

∂P
∂µ

T,V
we have
N
V
=


n=1
J
n
(n−1)!
ζ
n
.
The set of equations P(ζ) and N(ζ) defines the equation of state.
Spatial correlation of density fluctuations
Let us denote n(r) the fluctuating number density, ndV being the particle number in the element
dV. The correlation in the positions of the particles can be characterized by the average
'∆n(r
1
)∆n(r
2
)` = n(r
1
)n(r
2
) − ¯ n
2
.
For a uniform system it can depend only on the difference r =[r
1
−r
2
[, it must decay at r →∞.
132 CHAPTER 10. NON-IDEAL GASES
The product 'n(r
1
)n(r
2
)` can be expressed through the probability density w
12
to find the
particle 2 at the distance r from the particle one. Since the total probability of such event is
¯ nw
12
(r)dV
2
we might write
'n(r
1
)n(r
2
)` = ¯ n
2
w
12
.
That would be a mistake because it ignores the fact that if the points 1 and 2 coincide, the particle
belongs both to dV
1
and dV
2
. Thus the correct formula is
G(r
1
−r
2
) ='n(r
1
)n(r
2
)` = ¯ n
2
w
12
+ ¯ nδ(r
1
−r
2
).
In a similar way,
'∆n(r
1
)∆n(r
2
)` = ¯ nν(r) + ¯ nδ(r
1
−r
2
), ν(r) ≡ ¯ n[w
12
(r) −1] .
The quantities '∆n(r
1
)∆n(r
2
)` and ν(r) are called the correlation functions. Integrating the
previous equation with respect to dV
1
dV
2
over a volume V we have to recover '(∆N)
2
`. From
the formula above we find the normalization condition,

νdV =
'(∆N)
2
`
N
−1 =−
TN
V
2

∂V
∂P

T
−1.
This is zero for an ideal gas, only interaction leading to a finite compressibility leads to a corre-
lation. Since for the pair interaction
ν(r) = ¯ n

e
−βV
12
(r)
−1

we get
J
2
=

e
−βV
12
(r)
−1

dV =
V
N

ν(r)d
3
r .
We see that the correlation function is an important property.
It can be directly determined from the experiments on the light scattering. Consider a wave
F(R) = F
0
e
ikR−iωt
which is scattered by the particles residing at points r
i
(see Fig. 10.2). Let the scattering ampli-
ρ
i
k
k’
Detector
R r
i
Figure 10.2: On the scattering problem.
tude f be the same for all the particles. Then the scattered wave at the detection point r is
F
sc
(r) = f
N

i=1
F(r
i
)
e
ikρ
i
ρ
i
= f F
0
N

i=1
e
i(kr
i
−ωt)
e
ikρ
i
ρ
i
.
10.1. DEVIATION OF GASES FROM THE IDEAL STATE 133
Here ρ
i
=[R−r
i
[ is the distance between the scatterer and the detector. We have ρ
i
=R−r
i
cosφ
where φ is the angle between r
i
and R Now let us take into account the fact that the detector is
usually place far from the scattering medium, R r
i
. Under this approximation we can replace
ρ
i
by R in the denominator of the previous formula, while the phase needs more exact treatment,
kr
i
+kρ
i
= kr
i
+kR−kr
i
cosφ = kR−r
i
(k−k
/
).
In this way we get,
F
sc
= f F
0
e
ikR−iωt
R

i
e
−ir
i
q
, q ≡k−k
/
.
Thus the ratio between the average intensity of the scattered wave and the intensity of the incident
wave is
[ f [
2
N

i=1
N

j=1
e
iq(r
i
−r
j
)
≡N[ f [
2
S(q).
S(q) is called the structure factor. Replacing the double sum by its ensemble average, we can
write it as
S(q) =
1
N

dV
1

dV
2
'n(r
1
)n(r
2
)`e
iq(r
1
−r
2
)
= 1+

d
3
r ν(r)e
iqr
.
This is the way to extract the correlation function from the experiment on light scattering. Note
that for small scattering angles q →0 the structure factor is proportional to the compressibility
of the system. Thus one can expect singularities of the scattering cross section at the phase
transition point.
Van der Waals’ formula
There is an interpolation formula which allows to follow the phase transition between gaseous
and liquid state. To derive such a formula let us assume that the interaction is small in a sense
that V
0
<T. To perform the integration for the virial coefficient let us assume that at r 2r
0
we
have V
12
> T.The quantity 2r
0
has a meaning of the atomic diameter. Let us split the integral for
B(T) into 2 parts,
B(T) = 2π

2r
0
0

1−e
−βV
12

r
2
dr +2π


2r
0

1−e
−βV
12

r
2
dr .
In the first integral we neglect the exponential and obtain for that b = 16πr
3
0
/3. In the second
integral we expand the exponential to get
−2πβ


2r
0
[V
12
[ r
2
dr ≡−2βa.
In this way,
B(T) = b−a/T → F
i
= N
2
(bT −a)/V , G
i
= NP(b−a/T).
134 CHAPTER 10. NON-IDEAL GASES
The total free energy is
F = N f (T) −NT log(e/N) −NT(logV −Nb/V) −N
2
a/V .
Now we do something approximate. Since we have assumed that the gas is rarefied,
V Nb → logV −Nb/V ≈log(V −Nb).
Let us assume that this expression holds also for large density! As a result,
F
i
=−NT log(1−Nb/V) −N
2
a/V .
This formula take account of the finite compressibility at large density. Now we get
P =−
∂F
∂V
=
NT
V −Nb

N
2
a
V
2

P+
N
2
a
V
2

(V −Nb) = NT .
This the van der Waals’ equation.
Chapter 11
Phase Equilibrium
11.1 Conditions for phase equilibrium
Consider an isolated body with the energy E and volume V. This state is non necessary ho-
mogeneous, it can consist of two (or more) phase. For the case of two phases, the equilibrium
conditions are
T
1
= T
2
, P
1
= P
2
, µ
1
= µ
2
.
Since the temperature and pressure remain constant, the equality
µ
1
(P, T) = µ
2
(P, T)
correspond to a line in the (P, T)-plane along which the phases coyexist. In the (T,V)-plane,
there are 2 lines because at a given pressure and temperature can have different volumes. This is
illustrated in Fig. 11.1
T
P
I
II
V
T
I II
Figure 11.1: The hatched area cor-
responds to a mixture of 2 phases.
The Clapeyron-Clausius formula
Differentiating the equilibrium condition µ
1
= µ
2
with respect to temperature we get
∂µ
1
∂T
+
∂µ
1
∂P
dP
dT
=
∂µ
2
∂T
+
∂µ
2
∂P
dP
dT
.
135
136 CHAPTER 11. PHASE EQUILIBRIUM
Since

∂µ
∂T

P
=−s,

∂µ
∂P

T
= v (s ≡S/N, v ≡V/N)
we obtain
dP
dT
=
s
1
−s
2
v
1
−v
2
=
q
T(v
1
−v
2
)
where q = T(s
1
−s
2
) is the heat of transition. This the Clapeyron-Clausius formula. This equa-
tion gives the temperature dependence of pressure along the phase equilibrium curve. Written
as
dT
dP
=
T(v
1
−v
2
)
q
it gives the change in the equilibrium temperature (freezing point, or boiling point) when the
pressure changes.
The law of corresponding states
Let us discuss as an example the van der Waals equation of state,
P =
NT
V −Nb

N
2
a
V
2
.
We can notice that there is a special temperature, T
c
, at which the isotherm has an inflection point.
We have:

∂P
∂V

T
c
= 0,


2
P
∂V
2

T
c
= 0 → T
c
=
8
27
a
b
, V
c
= 3Nb, P
c
=
1
27
a
b
2
.
Measuring the P, V, and T in the units P
c
, V
c
and T
c
, respectively, we can rewrite the equation of
state in the form

P
/
+
3
V
/2

(3V
/
−1) = 8T
/
.
The van der Waals’ isotherms determined by this equation are shown in Fig. 11.2. So we are
able to express the equation of state in a form which is independent of the particular properties
of the gas. We observe that at T < T
c
there are regions with positive ∂P/∂V. In that regions the
system cannot be stable. On the other hand, we know that the equilibrium can be achieved only
at P = const. As a result we arrive at the situation shown in Fig. 11.3 To the right from the point
e the system is in a gas state, while to the left from the pint b it is in a liquid state. The phase
equilibrium takes place along the straight line b −e which corresponds to P = const. This line
must be constructed in a way to keep µ
1
= µ
2
. This difference can be expressed as an integral
along the transition path,
µ
e
−µ
b
=

e
b
dµ = 0.
11.1. CONDITIONS FOR PHASE EQUILIBRIUM 137
1.05
1.0
0.95
0.9
v
2.4 2.2 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4
p
1.4
1.2
1
0.8
0.6
0.4
0.2
0
Figure 11.2: Van der Waals’
isotherms. The numbers near the
curves are the ratios T/T
c
.
K
a
b
c
d
e
f
Gas
L
i
q
u
i
d
V
P
A
B
Figure 11.3: On the gas-liquid transition.
Since dµ = (V/N)dP we come to the conclusion that the integral along the isotherm,

e
b
V dP = 0.
Geometrically it means that the areas between the isotherm and the line b −e below and above
the line must be equal. This rule is called the Maxwell construction.
Van der Waals theory of the critical point
The calculations can be easily performed near the critical point, T
c
. Introducing notations
P = P
c
(1+ p), T = T
c
(1+τ)
N
V
=
N
V
c
(1+n)
and expanding in powers of τ and n up to non-trivial order we arrive at the equation of state
p = 4τ +6τn+(3/2)n
3
, (∂p/∂n)
τ
= 6τ +(9/2)n
2
.
138 CHAPTER 11. PHASE EQUILIBRIUM
The Maxwell condition can be written as

e
b
V dP = 0 →

n
e
n
b
n

∂p
∂n

τ
dn = 0.
Since the integrand is an odd function of n we get n
e
= −n
b
. From the condition p
1
= p
2
we
obtain
n
e
=−n
b
= 2

−τ, p = 4τ.
The boundary of the metastable region are given by the equation
n
d
=−n
c
= 2

−τ/3.
According to the Clapeyron-Clausius formula, the heat of transition is proportional to

−τ.
Chapter 12
Continuous Phase Transitions
Introduction
In the following set of lectures we shall discuss so-called continuous phase transitions, or transi-
tions of the 2nd order. A generic system to discuss the transitions is a magnet. Thus we start in
recollection of thermodynamics of a magnet.
Thermodynamics of a magnet
Consider a long and thin solenoid with the length L and number of windings N. Then the mag-
netic field inside the solenoid is H = NI/L, where I is the electric current. When the magnetic
flux Φ
through the solenoid is changing an electro-motive force, E = −∂Φ/∂t, the work made by
the battery being
∆W
b
=EI ∆t = I ∆Φ= NAI ∆B =VH∆B.
Here A is the solenoid area, V is its volume, while B ≡Φ/A is the magnetic induction. Introduc-
ing magnetization
˜
M from the relationship B = µ
0
(H +
˜
M) we can express the work per volume
as
∆W
b
/V =∆(µ
0
H
2
/2) +µ
0
H∆
˜
M.
The first item is the energy of external magnetic field which is present also in the absence of
material. The second term is the work done by the field on the material. Consequently, the work
done by the material can be expressed through the total magnetic moment M ≡V
˜
M as
∆W =−H∆M.
As a result, the 1st law of thermodynamics reads as
dU = T dS−HdM
and
T =

∂U
∂S

M
, H =

∂U
∂M

S
.
139
140 CHAPTER 12. CONTINUOUS PHASE TRANSITIONS
Generalizing of other thermodynamic potentials is more or less obvious:
F(T, M) = U −TS →dF =−SdT +HdM, S =−

∂F
∂T

M
, H =

∂F
∂M

T
;
G(T, H) = F −HM → dF =−SdT −MdH, S =−

∂G
∂T

H
, M =−

∂G
∂H

T
.
General considerations
Uniaxial ferromagnet is a simplest example of phase transition. At zero external magnetic field
magnetization vanishes at certain temperature, T
c
. Below T
c
a spontaneous magnetization M
0
(T)
is observed, its direction can be “up” and “down” with equal likelihood. The dependence M
0
(T)
is schematically shown in Fig. 12.1. This figure can be explained in the framework the Ising
T
M (T)
o
c
T
Figure 12.1: The spontaneous magnetization as a function
of temperature.
model. According to that model, spins are located at lattice points in a regular lattice. They can
point up and down. Nearest neighbors interact in a way that is is favorable to point in the same
direction. As we know, the free energy G = U −TS −HM has to be minimized. At H = 0 it
consists of two terms, U and −T S(T). At low temperatures the energy U is more important and
the system is a ferromagnet. At high temperature the free energy is minimized by disordered
state in which the entropy is large. It is the fight between the order and disorder that makes the
critical state specifically interesting.
In real life, the above conclusion is true for the dimension d ≥ 2, while for d = 1 the mag-
netization never vanishes. That indicates that phase transition is a collective effect depending on
interaction of many spins. The higher the dimension the greater the number of neighbors, the
stronger the collective effects.
Coming back to the case of a uniaxial ferromagnet, we can define the spontaneous magneti-
zation M
0
(T) as the order parameter. In some cases it is difficult to specify the order parameter.
In general it can be a scalar, a vector, a tensor, etc.
When an order parameter is specified, the next question is if the transition is of 1st order,
or it is continuous. In the example of the ferromagnet, it is the question if the spontaneous
magnetization goes to zero at T → T

c
is a continuous way, or it is abrupt. This is a delicate
issue. Assume that we ramp the external magnetic field from a negative to a positive value. Than
the H = 0 the magnetic moment changes from−M
0
(T) to M
0
(T), i. e. the transition is of the 1st
order. Thus the character of transition cannot be specified without specifying the transition path.
141
At continuous phase transitions critical phenomena are observed. Usually a number of quan-
tities show power law behavior close to T
c
, e. g.
M
0
(T) ∼(T
c
−T)
β
.
Other examples of critical exponents for thermodynamic quantities are
C ∼[T
c
−T[
−α
, χ =

∂M
∂H

T
∼[T
c
−T[
−γ
, [M(H, T = T
c
)[ ∼H
1/δ
.
An important concept of the theory of continuous phase transitions is the universality class.
It became clear during 1960’s that a number of details in a given system are irrelevant as far as
critical exponents are concerned. Systems of the same dimension and with a phase transition
into an ordered state with the same symmetry belong to the same universality class. They have
essentially the same critical properties. The justification of that concept will be discussed later.
Examples of phase transitions
We shall classify the phase transitions by the spatial dimensionality and the number of indepen-
dent components of the order parameter, (d, n).
Magnetic systems
Uniaxial ferromagnets: described by the Ising model, the universality class being (d, 1).
Hard axis ferromagnets: m
0
is a vector in the xy-plane perpendicular to the hard axis, (d, 2).
This model is called the isotropic xy-model, the universality class being (d, 2).
Heisenberg model: the magnetization is isotropic in 3 dimensions, the universality class being
(d, 3).
So far we have assumed isotropy in spin space. This assumption can be relaxed, and one can
also expect, say, cubic anisotropy. In this case the corners of a (hyper)cube in spin space are
energetically favored (or suppressed). That gives rise to new universality classes.
Antiferromagnets: Neighboring spins prefer to point to different directions, and on the average
there are equally many spins pointing in opposite directions. The order parameter on a 3d
cubic lattice can be chosen as staggered magnetization,
m
s
=

klm
(−1)
k+l+m
'm
klm
`,
where 'm
klm
` is the magnetic moment of spin at the lattice point (k, l, m). Note that the
universality classes for antiferromagnets are the same as for ferromagnets.
142 CHAPTER 12. CONTINUOUS PHASE TRANSITIONS
Fluid systems
The universality class is the same as for the Ising model, (d, 1) because the order parameter is a
scalar. The situation is more complex for liquid mixtures.
Superfluids and Superconductors
At the temperature T = T
λ
≈ 2K, liquid He
4
undergoes a continuous phase transition from the
normal phase (HeI) to a (partially) superfluid phase (HeII). The relative fraction of superfluid
component, ρ
s
/ρ decreases continuously from 1 at T = 0 to 0 at T = T
λ
. It appears that the
proper order parameter is a two-component rotation invariant quantity, so the universality class
in (3, 2).
A similar situation is relevant to superconductors. However, the critical region in the ”old”
superconductors is so narrow that it is almost irrelevant form the experimental point of view.
1
The critical phenomena were, however, observed in high-T
c
superconductors.
Liquid crystals
Liquid crystals consist of long molecules. At the 1st approximation they can be considered as
inflexible rods. The simplest phase is nematic where the rods are oriented. Otherwise the nematic
is a disordered liquid.
There exist more complicated structure like cholesteric phase where a director rotates at a
constant pitch in a plane perpendicular to the direction in which one moves. The period is about
3000
˚
A. Smectic liquid crystals combine orientation order with a spatial one along one direction.
Structural phase transitions
These transitions are changes in the spatial symmetry of the crystalline lattice.
Other phase transitions
One can name many other phenomena which can be regarded as critical ones, e. g. change in the
symmetry of adsorbed monolayers, percolation, dynamical phase transitions, etc.
12.1 Landau Theory
Landau theory was designed to describe systems close to a phase transition where the order
parameter ψ is small. ψ can be a scalar, vector, etc. Near the transition point we can expand the
free energy as
˜
G(P, T, ψ) = G
0
+αψ+Aψ
2
+Cψ
3
+Bψ
4
. . . .
Having in mind the magnetization as an example, we can specify magnetization M as an order
parameter, then the free energy depends on T, H, M where H is the external magnetic field. Those
1
The situation is a bit different at low dimensions.
12.1. LANDAU THEORY 143
are the intensive parameters which characterize the state. Note that G is not a true Gibbs free
energy, it is a hypothetical free energy which would be the case if the system with the order
parameter ψ could exist as an equilibrium one. The equilibrium is reached at the minimum of G,
and in the minimum it coincides with the Gibbs free energy.
The total scheme of the Landau theory is the following.
• The free energy must be properly defined and then expanded in a Taylor series up to the
minimal relevant order (see below).
• The spontaneous value(s) of the order parameter is found by minimization of
˜
G with re-
spect to ψ keeping the thermodynamic parameters constant.
Uniaxial ferromagnets
In zero magnetic field there is up-down symmetry,
˜
G(T, M) =
˜
G(T, −M). Thus

˜
G(T, M) ≡
˜
G(T, M) −
˜
G(T, 0) =
r(T)
2!
M
2
+
u(T)
4!
M
4
+ .
Existence of stable minima requires u(T) to be positive. At r(T) > 0 the free energy has one
stable minimum, while at r(T) < 0 there are 2 equivalent minima ±M
O
, see Fig. 12.2. The
∆G
∆G
M
M
M
0
-M
0
t<0
t>0
Figure 12.2: The free energy above
(left) and below (right) the critical
temperature T
c
.
higher symmetry phase at r > 0 is replaced by a lower symmetry phase. This phenomenon is
called the spontaneous symmetry breaking. At r(T) = 0 the high symmetry of the disordered
phase in broken and one of the low temperature ordered phases is chosen.
Obviously, r(T
c
) = 0, and near T
c
we can assume
r(T) = at ; u(T) = u(T
c
) ≡u.
Here τ = (T −T
c
)/T
c
is the reduced temperature variable. Differentiation of
˜
G with respect to M
yields
0 =


˜
G
∂M

T
= aτ +
u
3!
M
3
.
This equation has 3 solutions,
M = 0, M =±M
0

−6aτ/u.
144 CHAPTER 12. CONTINUOUS PHASE TRANSITIONS
From Fig. 12.2 it is clear that the 1st solution represents a stable minimum above T
c
, while the
other represent degenerate minima below T
c
. If we present the solution as
M
0
∝ (−τ)
β
→ψ
0
∝ (−τ)
β
we obtain β = 1/2 for the critical exponent. The critical exponents following from the Landau
theory are called “classical”.
Jump in Specific Heat
Now let us consider the specific heat,
C = T

∂S
∂T

H=0
=−T


2
G
∂T
2

H=0
.
Here G is the usual free energy,
G(T, H) = min
M
˜
G(T, H, M).
At H = 0 we get
G(T, H) =

G(T, 0, 0) ; τ > 0
G(T, 0, M
0
(T)) ; τ < 0
.
As a result, at τ < 0 we obtain
G(T, 0) =
˜
G(T, 0, 0) +

2
6a(−τ)
u
+
u
4!

6a(−τ)
u

2
.
Now we easily obtain the specific heat by differentiation. Denoting C
+
= −T∂
2
˜
G(T, 0, 0)/∂T
2
we obtain
C =

C
+
; τ > 0
C
+
+(T/T
c
)
2
(3a
2
/u) ; τ < 0
.
Thus we observe a negative jump in the specific heat if one passes the critical point from below.
Magnetic Susceptibility
Now let us study a response to a small external magnetic field. Then we have

˜
G(T, M) =−HM+
r(T)
2!
M
2
+
u(T)
4!
M
4
+ .
The minimization with respect to M leads to the equation for the equilibrium value of the mag-
netization, M
e
,
H = rM
e
+
1
3!
M
3
e
+ .
12.1. LANDAU THEORY 145
The inverse susceptibility is defined as
χ
−1
=

∂H
∂M
e

T
= r +
u
2
M
2
e
.
Substituting the equilibrium magnetization at H = 0 we obtain
χ
−1
=

aτ ; τ > 0
2a(−τ) ; τ < 0
.
If we define the critical exponent of the quantity A(τ) as
s ≡− lim
τ→0
+
lnA(τ)
lnτ
we find the classical critical exponent γ =1. We get χ(τ) ∝[τ[
−γ
with the slope ratio A
+
/A

=2.
At the critical isotherm, τ = 0,
H =
u
3!
M
3
e
→M
e
∝ H
1/3
.
If we in general assume M
e
∝ H
1/δ
, then the classical value of δ is 3. Later on we shall compare
the values of critical exponents with more general theory and with experiment.
Role of External field
Let us now discuss the situation in a finite external field. The plot of the free energy vs. M is
shown in Fig. 12.3 for τ > 0 (solid line) and τ < 0 (dashed line). We observe that there is a stable
M
G
Figure 12.3: The free energy at a finite external field.
minimum in both cases – magnetic field removed the system from its critical point.
Conclusion: If we are interested in continuous phase transition the fields which linearly
coupled with the order parameter must be removed.
146 CHAPTER 12. CONTINUOUS PHASE TRANSITIONS
Role of Cubic Invariants
Now let us consider the case when symmetry of the system allows ψ
3
terms. Then

˜
G =
r(T)
2!
ψ
2

c
3!
ψ
3
+
u
4!
ψ
4
+ .
Let us put c > 0. The analysis of that free energy becomes clear from Fig. 12.4. At sufficiently
large positive τ the profile has a shape shown in the left panel. It has one stable minimum at
ψ = 0. At low temperatures an additional minimum develops, and at some τ
2
this minimum is
degenerate with ψ = 0 (middle panel). At lower temperatures the second minimum is the only
stable one (right panel). Thus there is a (weakly) first order transition fromψ=0 to some ψ=ψ
0
M
G
M
1
G
0.01
M
2
G
Figure 12.4: The free energy with a cubic invariant in-
cluded.
at τ =τ
2
.
Conclusion: Cubic terms results in a first-order transition rather in a continuous one. τ = 0
does not have the significance of the transition temperature in this case.
12.1. LANDAU THEORY 147
Tricriticality
Now let us return to the simplest case where the free energy is even in the order parameter. Let
us assume that r > 0, but u is negative. To keep the stability one has to add the higher order term
to get

˜
G =
r
2!
ψ
2
+
u
4!
ψ
4
+
w
6!
ψ
6
.
At w>0 the system undergoes a 1st order phase transition at some temperature T
0
corresponding
to r = r
0
> 0. Above T
0
the equilibrium order parameter is zero (see right panel of Fig. 12.5).
At T = T
0
the transition takes place to one of a symmetric set of states with lower symmetry
at which the order parameter is ±ψ
0
. The temperature dependence of the order parameter is
M
G
M
G
M
G
Figure 12.5: The free energy with a cubic invariant in-
cluded.
shown in Fig. 12.6. Suppose now that u is a function of some parameter, p, and it changes sign at
p = p
c
. At this point the character of phase transition will change from 1st order to continuous.
The point r = aτ = 0, u(p
c
) = 0 is called the tricritical point. It is easy to find classical critical
exponents at the tricritical point. Putting u = 0 and minimizing the free energy we obtain the
148 CHAPTER 12. CONTINUOUS PHASE TRANSITIONS
T T
o
ψ
o
Figure 12.6: The order parameter as a function of the
temperature at u < 0.
expression for the order parameter,
aτψ
0
+
w
5!
ψ
5
0
= 0 → ψ
0
=

5! a(−τ)
w

1/4
.
Thus β
t
= 1/4.
Summary of the Landau recipe
• Given that the system has a continuous or weakly 1st order phase transition form a disor-
dered phase to the phase of higher symmetry.
• Characterize the order of the low symmetry phase and define the order parameter (scalar,
vector, tensor, or . . .).
• Introduce the proper free energy
˜
G(T, ψ) for a given space-independent order parameter.
The given value of this parameter does not have to be the equilibrium value.
• Write down the Taylor expansion of
˜
G in powers of ψ, where all the terms compatible with
the symmetries of the system are included. Include only the minimum number of terms.
• Write the coefficients of the ψ-expansion as a Taylor expansion in τ = (T −T
c
)/T
c
and
keep only the leading ones.
2
• Determine the equilibrium value(s), ψ
0
, by minimizing
˜
G with respect to ψ.
• Substitute ψ
0
into the minimum, free energy and calculate equilibrium properties like crit-
ical exponents.
2
Sometimes a parameter different from the dimensionless temperature can play a role of τ.
12.2. SCALING LAWS 149
Critique
• The Landau theory assumes the existence of the phase transition.
• A general recipe for constructing the order parameter does not exist.
• Spatial dependence of the order parameter is ignored.
• Fluctuations of the order parameter are neglected.
12.2 Scaling laws
Consider a system close to continuous phase transition. Let us introduce reduced temperature,
τ, and external field h, the transition point corresponding to (τ, h) = 0. Since we are interested
in singular behavior near the critical point let us separate the singular part of the free energy,
G
s
(τ, h). The crucial point is to assume (following B. Widom, 1965) that
G
s

a
τ, λ
b
h) =λG
s
(τ, h).
This is the so-called homogeneity postulate. The physical reason for the postulate will be dis-
cussed later in connection with renormalization group arguments. The postulate allows one to
express all the critical exponents through the constants a and b, and to find relationship between
the exponents. These relations are called the scaling laws.
To derive scaling laws we choose λ
a
=[τ[
−1
. Then,
G
s
(τ, h) =[τ[
1/a
G
s

τ
[τ[
,
h
[τ[
b/a

.
Since τ/[τ[ =±1 we can express G
s
as
G
s
(τ, h) =[τ[
2−α
Φ
±
(h/[τ[

), α = 2−1/a, ∆ = b/a.
After such an assumption the specific heat is
C ∼

2
G
s
(τ, 0)
∂τ
2
∼[τ[
−α
while Φ
±
(0) = A
±
are constants. The generalized magnetization is
m ∼
∂G
s
∂h
=[τ[
2−α−∆
Φ
/
±
(h/[τ[

).
At τ > 0 spontaneous magnetization m
0
vanishes, thus Φ
/
+
(0) = 0. However, when τ < 0
Φ
/

(0) = B

= 0. As result,
β = 2−α−∆.
150 CHAPTER 12. CONTINUOUS PHASE TRANSITIONS
In a similar way, the isothermal susceptibility in zero field is given as
χ =

∂m
∂h

T


2
G
s
∂h
2
=[τ[
2−α−2∆
Φ
//
±
(0).
With Φ
//
±
(0) =C
±
= 0 we obtain
γ =−2+α+2∆.
Excluding the ”gap exponent” ∆ we have
∆ =β+γ → α+2β+γ = 2.
This is the first example of the scaling law. Note that classical exponents are (α=0, β =1/2, γ =
1) while the exact Ising one for a 2D system are (α = 0, β = 1/8, γ = 7/4), as has been derived
by L. Onsager. Both sets meet the scaling law.
What is left is to determine the exponent δ for the critical isotherm. The problem is the
argument of the function Φ
/
±
tends to infinity at [τ[ →0 for finite h. Assuming that Φ
/
±
(x) ∼x
κ
for large x we obtain
m([τ[) →0, h) ∼[τ[
2−α−∆−κ∆
h
κ
.
For m to have a finite value at [τ[ →0 we have to put
2−α−∆−κ∆ → δ =

2−α−∆
.
After a simple algebra we obtain
∆ =βδ =β+γ → α+η(1+δ) = 2,
that is the second scaling law.
Using the scaling laws we can express the behavior of the order parameter as a law of corre-
sponding states as
m
[τ[
β

/
±

h
[τ[
βδ

.
12.3 Landau-Ginzburg theory
The Landau-Ginzburg (LG) theory is an attempt to take into account fluctuations of the order
parameter in its simplest way. Taking the uniaxial ferromagnet as an example we consider now
the local magnetization m(r) as the order parameter. Then the free energy becomes a functional
of m(r) rather the function of a total magnetic moment, M. The recipe is to add only simplest
non-trivial terms. Since the free energy must be an even function of magnetization one can
imagine the combinations like m(r)∇m(r), m∇
2
m, (∇m(r))
2
, etc. The 1st combination can be
transferred into a surface integral, and the 2nd and the 3d ones can be combined using partial
integration. As a result, the LG functional is constructed as
G
LG
=

dr

g
2
(∇m)
2
+
r
2
m
2
(r) +
u
4!
m
4
(r) +
¸
.
12.3. LANDAU-GINZBURG THEORY 151
As before, r = aτ, u > 0, g > 0. The latter inequality implies that the homogeneous state is the
equilibrium one.
The LG theory allows for long-range fluctuations because it is an expansion in powers of the
magnetization gradient.
Classical fluctuations
In the classical liming discussed we keep only the terms quadratic in fluctuations. Then the
fluctuations are Gaussian and various fluctuation modes are statistically independent. At T >
T
c
, and with h = 0, the spontaneous magnetization 'm(r)` = 0. Consequently, m(r) is itself a
fluctuation. To order of m
2
we have
G
LG
=

dr

g
2
(∇m)
2
+
r
2
m
2
(r) +
¸
. (12.1)
Below T
c
the fluctuation is δm(r) = m(r) −m
0
(T) where m
0
(T) = 'm(r)` is the spontaneous
magnetization. The expansion up to (δm)
4
gives
G
LG
=

dr

g
2
(∇δm)
2
+
r
2
m
2
0
+
u
4!
m
4
0
+

r +
u
3!
m
3
0

(δm) +

r
2
+
u
4
m
2
0

(δm)
2
+
1
3!
um
0
(δm)
3
+
1
4!
(δm)
4
+

. (12.2)
According to the Gaussian approximation, the items proportional to (δm)
3
and (δm)
4
are ne-
glected. Furthermore, m
0
=−6r/u, and the item proportional to δm vanishes. As a result,
G
LG
=

dr

g
2
(∇δm)
2
+
(−2r)
2
m
2
0
+

. (12.3)
Since

drA(r)B(r) =
1
V

k
A
−k
B
k

dk
(2π)
d
A
−k
B
k
we obtain above T
c
G
LG
=
1
2V
/

k
(gk
2
+r)m
−k
m
k
. (12.4)

/
means the we restrict summation up to some k
0
we have to do that because we neglect higher
gradients.
Now we are able to calculate correlation functions like
Γ
m
qq
/ ≡'m
q
m
q
/ ` =


/
k
dm
k
m
q
m
q
/ exp(−G
LG
)


/
k
dm
k
exp(−G
LG
)
.
152 CHAPTER 12. CONTINUOUS PHASE TRANSITIONS
Here ∏
/
k
means that one has to take into account only [k[ ≤k
0
. Since m(r) is real, m
k
= m

−k
, it
is clear from(12.4) that
Γ
m
qq
/ =V δ
q
/
,−q
Γ
m
q
where
Γ
m
q
≡'m
−q
m
q
` =


/
k
dm
k
m
−q
m
q
exp(−G
LG
)


/
k
dm
k
exp(−G
LG
)
.
Let us count only those k for which k
x
> 0. Then
/

k
dm
k
=
//

k
dm
−k
dm
k
/

k
(gk
2
+r)m
−k
m
k
= 2
//

k
(gk
2
+r)m
−k
m
k
.
As a result, we are left with the integral
Γ
m
q
=
1
V

dm
−q
dm
q
m
−q
m
q
exp[−(gq
2
+r)m
−q
m
q
/V]

dm
−q
dm
q
exp[−(gq
2
+r)m
−q
m
q
/V]
.
Since m
−q
is a complex variable we can parameterize it as m
q
= me

and calculate the Jacobian
∂(m
−q
, m
q
)
∂(m, φ)
=

e
−iφ
e

−ime
−iφ
ime

= 2im.
Then we can introduce a new variable x = m
2
, dx = 2mdm and use the result for Gauss integral.
As a result,
Γ
m
q
=
1
gq
2
+aτ
.
This formula has been obtained previously by Ornstein and Zernike in 1914 using other consid-
erations. After Fourier transform to real coordinates we obtain for d > 1 and r ξ
Γ
m
(r) ∼

r
(1−d)/2
e
−r/ξ
for r ξ, d > 1
r
2−d
for r <ξ, d > 2
; ξ =

g/aτ
In this way we predict critical exponents for the correlation length, ν = 1/2.
The Ginzburg criterion
Let us estimate the validity of the approximation made. The measure of the total strength of the
long-wave fluctuations are given by the integral

drΓ
m
(r) ≡Γ
m
0
, Γ
m
0

m
q=0
= (aτ)
−1
.
12.4. GINZBURG-LANDAU THEORY OF SUPERCONDUCTIVITY (SC) 153
Let us denote a typical value of the fluctuation squared by (δm)
2
. Since the typical diameter of
correlated region is ξ we have
Γ
m
0
≈(δm)
2
ξ
d
.
Consequently,
(δm)
2
∼ξ
−d
(aτ)
−1

1
a

a
g

d/2
τ
(d/2)−1
.
Now we can estimate when we can neglect 4th order terms in the expansion in powers of fluctu-
ations. A typical value of m when the 4th and 2nd order terms are comparable is
r
2
¯ m
2
=
u
4!
¯ m
4
→ ¯ m
2
=
12aτ
u
.
The LG theory is valid at (δm)
2
< ¯ m
2
, or at
τ
(d−4)/2
<
a
2
u

g
a

d/2
.
This is the famous Ginzburg criterion. We see that at d > 4 this criterion is always satisfied
because the r.h.s. is a constant depending on the concrete problem under consideration. Thus
d
c
= 4 is called the upper critical dimensionality. Accordingly, all the critical exponents take
their classical values.
At d < d
c
= 4 the Ginzburg criterion is violated close to the critical point. For d = 3 the
criterion has the form
τ u
2
/sg
3
,
so the LG theory is reasonable outside that region. The upper critical dimensionality d
c
= 4 is
valid for all the universality classes (d, n). We postpone a discussion what is going on in a close
vicinity of the critical point.
12.4 Ginzburg-Landau theory of superconductivity (SC)
Let us formulate the theory of SC transition as the one for a general continuous transition. First,
one should define the order parameter. As the order parameter it is natural to chose the wave
function of the Bose condensate, Ψ. According to the principles of SC, it is natural to assume
that the ground state corresponds to the total momentum equal to 0, the wave function being the
same for all the Bose particles.
Near the critical temperature, the modulus of the order parameter is small, and one can expand
the thermodynamic potential as
G
s
=G
n
+a[Ψ[
2
+
b
2
[Ψ[
4
+ .
Because we are interested in the vicinity of T
c
we can expand the coefficients in power of
τ =
T −T
c
T
c
.
154 CHAPTER 12. CONTINUOUS PHASE TRANSITIONS
According the definition of T
c
, above T
c
the stable state corresponds to Ψ = 0, while below T
c
Ψ= 0. Thus the coefficient a should change the sign at the transition point:
a =ατ, α > 0.
As a result, we describe the equilibrium values of the order parameter as
Ψ= 0 at T > T
c
,
[Ψ[
2
=−(α/b)τ =[Ψ
0
[
2
at T < T
c
.
Inserting the equilibrium value if [Ψ
0
[
2
into the expansion and comparing the result with the
known relation for the critical field, we get
G
s
−G
n
=
(ατ)
2
2b
=
H
2
c

.
The last equality is a consequence of the definition of the critical field. Indeed H
2
c
/8π is just
the energy of magnetic field repelled from the sample in the Meißner state. The ratio α
2
/2b can
be related to the critical temperature T
c
from the BCS theory of superconductivity.
3
The most
interesting case is the case of external magnetic field. In this case Ψ is coordinate dependent.
Thus we should add both the energy of magnetic field H
2
/8π and the energy connected with
the inhomogeneity. Near the critical point it is enough to add [∇Ψ[
2
. Here we come to the most
important point: Cooper pairs are charged particles. Consequently, because of gauge invariance,
only combination
−i ¯ h∇+
2e
c
A
is possible. To make a proper dimension we write the corresponding term as
1
4m

−i ¯ h∇+
2e
c
A

Ψ

2
.
Finally, we get for δG =G
s
−G
0
n
δG =

dV

ατ[Ψ[
2
+
b
2
[Ψ[
4
+
1
4m

−i ¯ h∇+
2e
c
A

Ψ

2
+
H
2

¸
.
Here G
0
n
is the free energy of the normal state without magnetic field. To get the minimum we
calculate variation with respect to Ψ

:

dV

ατΨδΨ

+b[Ψ[
2
ΨδΨ

+
1
4m

−i ¯ h∇+
2e
c
A

Ψ

i ¯ h∇+
2e
c
A

δΨ

3
α
2
/2b = c1g(ε
F
)(k
B
T
c
)
2
. where g(ε
F
) is the density of states at the Fermi level while c
1
is a numerical
constant.
12.4. GINZBURG-LANDAU THEORY OF SUPERCONDUCTIVITY (SC) 155
The item with the ∇δΨ

can be integrated by parts. Using the Gauss theorem we get
i ¯ h
4m

S
δΨ

−i ¯ h∇+
2e
c
A

ΨdS+
1
4m

dV δΨ

−i ¯ h∇+
2e
c
A

2
Ψ.
Then we put δΨ

= 0 at the surface and arbitrary inside. As a result, we get the following
equation for the order parameter
1
4m

−i ¯ h∇+
2e
c
A

2
Ψ+ατΨ+b[Ψ[
2
Ψ= 0.
If now we put δΨ

at the surface to be arbitrary, we obtain the boundary condition
n

−i ¯ h∇+
2e
c
A

Ψ[
S
= 0.
The variation with respect to δΨ leads to the complex conjugate expressions.
Now it is important to get the equation for electric current. To obtain it one should calculate
the variation with respect to the vector potential A. The variation of H
2
leads to
δ(curl A)
2
= 2curl Acurl δA.
Then we can use the relation
div [ab] = bcurl a−acurl b
to get
δ(curl A)
2
= 2δAcurl curl A+2 div [δAcurl A] .
The second integral can be transformed into the surface one, thus it is not important. On the
other hand, from the Maxwell equations
curl curl A =curl H =

c
j → δ(curl A)
2
= 2jδA.
The rest of the free energy can be calculated in a straightforward way, and equating the total
variation to zero we obtain
j =
ie¯ h
2m


∇Ψ−Ψ∇Ψ

) −
2e
2
mc
[Ψ[
2
A
This is just the quantum mechanical expression for the current for a particle with the charge
(−2e) and the mass 2m. The set
(1/4m)[−i ¯ h∇+(2e/c)A]
2
Ψ+ατΨ+b[Ψ[
2
Ψ= 0,
j = (ie¯ h/2m)(Ψ

∇Ψ−Ψ∇Ψ

) −(2e
2
/mc)[Ψ[
2
A
forms the Ginzburg-Landau equations.
156 CHAPTER 12. CONTINUOUS PHASE TRANSITIONS
Dimensionless parameters and their meaning
To make the equations simples new dimensionless parameters are usually introduced:
Ψ
/
=
Ψ
Ψ
0
, H
/
=
H
H
c

2
, r
/
=
r
δ
, A
/
=
A
H
c
δ

2
with δ =

2mc
2
4π(2e)
2
Ψ
2
0
, H
c
= 2

π
ατ

b
, Ψ
2
0
=
α[τ[
b
.
The quantity Ψ
0
is a spatially-homogeneous solution of the GL equations in the absence of
external magnetic field. In the following we will omit primes and use these variables. As a
result, the set of equations can be expressed as

i
κ
∇−A

2
Ψ−Ψ+[Ψ[
2
Ψ = 0,
curl curlA−
i



∇Ψ−Ψ∇Ψ

) −[Ψ[
2
A = 0
with the boundary condition
n

i
κ
∇−A

Ψ
surface
= 0.
This set contains only one dimensionless parameter
κ = 2

2
eH
c
δ
2
¯ hc
which is called the Ginzburg-Landau(GL) parameter.
To understand the meaning of the parameters let us consider the simplest problem, namely
penetration of a weak field into the superconductor x > 0. Let H| z, A | y, (see Fig. 12.7). We
ξ
δ
x
z
H
H
Ψ
N S
Figure 12.7: On the magnetic field penetration into a super-
conductor. Solid line is the field profile while dashed one
shows spatial dependence of the order parameter.
have
H(x) =
dA
y
dx
.
It is natural to consider Ψ as a function only of x. We get

1
κ
2
d
2
Ψ
dx
2
−Ψ+A
2
[Ψ[ = 0,

dx
surface
= 0.
12.4. GINZBURG-LANDAU THEORY OF SUPERCONDUCTIVITY (SC) 157
At κ <1 one can assume Ψ=const, in the bulk of material it is natural to assume [Ψ[
2
= 1, we
chose Ψ real. Immediately we get from the second GL equation
d
2
A
dx
2
−A = 0 → H = H
0
e
−x
, in usual units H = H
0
e
−x/δ
.
Thus δ is just the London penetration depth near T
c
.
Application of the GL theory
Surface energy of N-S interface
First, let us derive an auxiliary expression for the free energy which will be used below. Using
dimensionless notations we get


s
−Ω
(0)
n

dV =
H
2
c

dV

−[Ψ[
2
+
[Ψ[
4
2
+

i∇
κ
−A

Ψ

2
+H
2
¸
.
Then we integrate the item with ∇Ψ

by parts, the surface integral being zero. In the rest integral
we take into account that Ψ obeys the GL equation. As a result,


s
−Ω
(0)
n

dV =
H
2
c

dV
¸
H
2

[Ψ[
4
2

.
If one keeps the external field H
0
fixed the corresponding thermodynamic potential can be ob-
tained by subtracting H
0
B/4π. In a normal phase, correspondingly Ω
nH
= Ω
(0)
n
−H
2
0
/8π. As a
result,

(Ω
sH
−Ω
nH
) dV =
H
2
c

dV
¸
(H−H
0
)
2

[Ψ[
4
2

. (12.5)
Consider the problem where all the quantities depend only upon one co-ordinate, say x, while
A ⊥x. The set of equations has the form (we’ll see that Ψ is real)
1
κ
2
d
2
Ψ
dx
2
+Ψ(1−A
2
) −Ψ
3
= 0,

dx
[
surface
= 0,
d
2
A
dx
2
−Ψ
2
A = 0.
It is easy to find the first integral of this set. Let us multiply the first equation by dΨ/dx, the last
one – by dA/dx and them add the equation and integrate over x:

x
0
dx
¸
1
κ
2
d
2
Ψ
dx
2

dx


dx
(1−A
2
) −

dx
Ψ
3
+
d
2
A
dx
2
dA
dx
−Ψ
2
A
dA
dx

= 0.
We obtain
1
κ
2


dx

2

2
(1−A
2
) −
Ψ
4
2
+

dA
dx

2
= const =
1
2
158 CHAPTER 12. CONTINUOUS PHASE TRANSITIONS
(the constant being determined from the boundary conditions).
To find the surface energy we formulate the boundary conditions
x →∞ (S): Ψ= 1, H = A = 0, dΨ/dx = 0,
x →−∞ (N): Ψ= 0, H = H
0
= 1/

2, dΨ/dx = 0.
At κ <1 the most important region where A and H are small. Thus, we get
1
κ
2


dx

2

2

Ψ
4
2
=
1
2


dx
=
κ

2
(1−Ψ
2
).
The solution is
Ψ= tanh
κx

2
,
it is wrong in the region where the field penetrates, but this region is small. We observe that
the typical length at which the order parameter is changed is (in dimensional units) ξ =κδ. ξ is
called the coherence length.
Now we can employ Eq. (12.5) and put H
0
= 1/

2, H = 0. We obtain
σ
ns
=
H
2
c

¸
1−tanh
2
κx

2

dx =
H
2
c

4

2

.
In dimensional units that means
σ
ns
=
4

2
3
H
2
c

δ
κ
.
The opposite limiting case can be treated only numerically. The result is that σ
ns
=0 at κ =1/

2
and at κ > 1/

2 it is negative. This property has very important implication on the magnetic
field structure in superconductors.
Quantization of magnetic flux
Consider a hollow cylinder placed into a longitudinal magnetic field. Assume that the cylinder is
Figure 12.8: On the flux quantization in a hollow cylinder
in the Meißner state. If the order parameter has the form
Ψ=[Ψ[e

12.5. MEAN FIELD THEORY 159
the current is
j =−
e¯ h
m
[Ψ[
2
¸
∇χ+
2e
¯ hc
A

.
We can divide the current density by [Ψ[ and integrate over a closed circuit in the bulk of super-
conductor:

j dl
[Ψ[
2
. .. .
=−
e¯ h
m
¸

dl ∇χ
. .. .
+
2e
¯ hc

Adl
. .. .
¸
.
0 −2πk flux
Thus
Φ= kΦ
0
, Φ
0
=π¯ hc/e = 2.07 10
−7
G cm
2
.
It is interesting that the effect has been predicted by F. London (1950) before the concept of
Cooper pairs. He assumed the quantum to be π¯ hc/e = 2Φ
0
.
In any case the quantized quantity is in fact fluxoid

∇χdl
which is equal to the total flux through the hole and the surface layer. As a result, the flux
quantization through the hole is not exact.
Quantization of magnetic flux through a hollow cylinder has been experimentally observed
and has many implications.
12.5 Mean field theory
The mean field approximation
This is a basic approximation which allows one to find main results regarding phase transitions
leaving alone some subtle problems about correct critical exponents. For simplicity, let us restrict
ourselves with the generic Ising model with the partition function
Z
N
=

s
exp

β

i<j
J
i j
s
i
s
j


i
h
i
s
i

.
The main idea is to look at the world from a point of view of a given spin. A given spin i “feels”
the external field + the fields created by the neighbors,
h
i
(¦s¦) =
/

i j
J
i j
s
j
+h
i
,
where prime means that i = j. The first step is to replace the actual field by the average one,
h
i
(¦s¦) →h
e
i
=
/

i j
J
i j
's
j
` +h
i
=
/

i j
J
i j
m
i
+h
i
.
160 CHAPTER 12. CONTINUOUS PHASE TRANSITIONS
Then
Z
N
→Z
e
=

s
e
β∑
i
h
e
i
s
i
=
N

i=1
2cosh(βh
e
i
).
Then we have to close the scheme, i. e. to calculate m
i
='s
i
`. Since
m
i
=
1
β
∂lnZ
e
N
∂h
e
i
= tanh(βh
e
i
)
we have the equation
m
i
= tanh
¸
β

/

j
J
i j
m
j
+h
i
¸
which should be solved for m
i
.
In principle, it is a set of many equations, and further simplifications are required. Before
doing that we shall list some comments.
• The mean field approximation neglects the fluctuations of the order parameter. So the
critical exponents are classical.
• Contrary to the Landau theory, the mean field theory can be used even when the order
parameter is not small. This is why it can be used to discuss global properties of the phase
diagram.
• The mean field theory works better if the interaction range is large.
12.6 Spatially homogeneous ferromagnets
Assume the spatial homogeneity, h
i
= h, and that surface effects are not important. Then J
i j
=
J(r
i
, r
j
) = J([r
i
−r
j
[). We obtain
m = tanh[β(
¯
Jm+h)] ,
¯
J =
/

j
J
i j
.
Let us discuss the case h = 0. The situation is illustrated in Fig. 12.9 where the l.h.s. and r.h.s.
are plotted At small β (high temperatures) there is a unique solution m = 0, while at low T (large
β) 3 solutions exist. Analysis of the free energy shows that the solution m = 0 is unstable, the
other ones are degenerate at h = 0. At the critical point all 3 solutions merge. That takes place at

∂tanh(β
¯
Jm)
∂m

m=0

¯
J = 1, → T
c
=
¯
J =
/

j
J
i j
.
With nearest neighbor interaction and for a (hyper)cubic lattice
¯
J = 2dJ where J is the nearest
neighbor interaction.
4
4
At d = 1 the predictions of the mean field theory are wrong even qualitatively.
12.7. RENORMALIZATION FRAMEWORK 161
-1
1
m
-m
o
o
T > T
T < T
T=T
c
c
c
tanh
β
Jm
m
Figure 12.9: Graphical solution of the mean field
equation.
Let us re-derive the same expression form another point of view which seems useful for
following discussions. From the mean field equation we we can calculate the susceptibility
χ =

∂m
∂h

T
= cosh
−1
[β(
¯
Jm+h)]
¸
β
¯
J

∂m
∂h

T

.
Putting h = 0 and having in mind that at T > T
c
m = 0 we obtain
χ =
β
1−β
¯
J
=
T
T −T
c
.
In this way we obtain that susceptibility id divergent at the critical point, and the critical exponent
γ = 1.
12.7 Renormalization framework
The main idea (L. Kadanoff, 1966) is that since ξ →∞ at the critical point there is no intrinsic
scale for the long range fluctuations. Thus the fluctuations must have the similar appearance
at all the scales. Geometrically speaking, the critical state should have properties of a fractal.
Thus, critical Hamiltonian should exist for the variables on any scale. The Kadanoff’s idea was
a starting point for K. Wilson who has developed a quantitative approach (Nobel Prize, 1983).
The method is a mapping, or the renormalization group transformation which transforms the
Hamiltonian from one scale into one describing the system on a larger length scale.
The proper way to perform such a scaling is to keep long-range (critical) fluctuations un-
changed. We shall briefly discuss only one way to perform such a scaling – the real space
renormalization constructed by Th. Niemeijer and J.M.K. van Leeuven.
The mapping
Let us consider a general class of Ising models. The N spins take the values s
i
=±1, the Hamil-
tonian can be written as
H(s) =

i
K
i
s
i
+

ik
K
ik
s
i
s
k
+

ikl
K
ikl
s
i
s
k
s
l
+ ≡

α
K
α
Π
α
(s). (12.6)
162 CHAPTER 12. CONTINUOUS PHASE TRANSITIONS
Let us fix the origin on the energy scale by the stipulation

s
H(s) = 0.
The equation (12.6) can be inverted to express the coupling constants K
β
as
5
K
β
=
1
2
N

s
Π
β
(s)H(s) (12.7)
The thermodynamics of the model is contained in the partition function,
Z
N
=

s
e
−βH(s)
, G =−T lim
N→∞
1
N
lnZ
N
.
Here G is the free energy per spin.
Our goal is to map the effective Hamiltonian (12.6) to another one relevant to a larger scale.
Consider spins on a 2D triangular lattice, as shown in Fig. 12.10. Let us consider a triangle as a
Figure 12.10: Cells in a triangular lattice.
new spin. The “coarse-grained” spins are defined according to the algorithm that it is the same
as the majority of spins in the covered triangle. This mapping can be analytically expressed as a
projection operator,
P(s
/
, s) =

i
P
i
, P
i
=

1 if s
/
i
= sgn(s
i
1
+s
i
2
+s
i
3
)
0 otherwise
.
Here i labels the cell while s
i
k
are original spins in the cell i. Then we can write,
Z
N
=

s
/

s
P(s
/
, s)e
−βH(s)
,

s
/
P(s
/
, s) = 1.
Now we can determine a new Hamiltonian which is valid on the new scale as

s
P(s
/
, s)e
−βH(s)
≡e
βΩ
1
−βH
/
(s
/
)
.
5
To prove this equation one can substitute (12.6) into (12.7). For every pair of sets α = β the spin s
k
summed
over all the states gives zero, ∑
k
s
k
= 0. When β =α each term gives unity, and there are 2
N
such terms.
12.7. RENORMALIZATION FRAMEWORK 163
We have to introduce a spin-independent exponent βΩ
1
to keep the property

s
/
H
/
(s
/
) = 0.
This is a unique definition, and using the inversion scheme (12.7) we can determine a new set k
/
α
of the coupling constants. So we know, at least in principle, the mapping K→K
/
(K).
The mapping H(s) →H(s
/
) is called the renormalization group transformation (RG). In this
way we map a triangular lattice on another triangular lattice increasing the scale by

3. So we
can expect that singularities are the same for both Hamiltonians.
Global properties of the mapping
As a result of RG transform we get
Z
N
=

s
/
,s
P(s
/
, s)e
−βH(s)
= e
βΩ
1
Z
/
N
/ .
Here Z
/
N
/
is the partition function for the transformed Hamiltonian with N
/
cells. Since N
/
/N =
l
−d
where l is the scaling factor,
G(K) =Ω
1
+
1
l
d
G(K
/
), Ω
1
=Ω
1
/N.
The spin-independent item is calculated from the initial configuration when the fluctuations are
frozen. Thus it is not responsible for the singular behavior. Keeping only singular items we
require
G(K) =
1
l
d
G(K
/
).
Before using the RG scheme let us recall that the physical correlation length remains the same.
However while rescaling we use different yard sticks, so ξ(K
/
) =
1
l
ξ(K). Since at the critical
point ξ(K
c
) = ∞ it follows that the critical Hamiltonian is mapped also on the another critical
Hamiltonian.
To demonstrate the procedure, let us choose the 2D set of the coupling constants, K =
(K
1
, K
2
). An example is a magnetic system in a zero magnetic field having nearest and next-
nearest neighbor interaction. To discuss dynamics let us recall that the quantities K
e
= −βK
rather than K enter the the theory. Thus changing the temperature we can move radially in the
plane of effective coupling constants K
e
. The origin of the (K
e
1
, K
e
2
)-plane corresponds to infinite
temperature, while T = 0 corresponds to the infinity in the direction given by the ratio K
2
/K
1
(dashed line in Fig. 12.11). Since for every such direction there is a critical temperature, the
subspace of the critical Hamiltonians is represented by a line of criticality in the (K
e
1
, K
e
2
)-plane.
Suppose that we start from a supercritical Hamiltonian represented by point 1. The repeated
RG transforms generate Hamiltonians 2, 3, 4, . The correlation length of the initial Hamil-
tonian was finite, so in shrinks by l after each iteration. Thus the Hamiltonian moves more and
more away from criticality. Eventually the point (K
e
1
, K
e
2
) = (0, 0) is reached. In other words, the
164 CHAPTER 12. CONTINUOUS PHASE TRANSITIONS
K
K
e
e
1
2
1
2
4
3
1’
2’
3’ 4’
Figure 12.11: Results of RG transformation. The full line
defines the critical subspace. Initial Hamiltonians in this
subspace converge to the fixed point. An initial sub-critical
Hamiltonian, 1, is, by subsequent transformations mapped
on the Hamiltonians 2, 3, 4, etc., converging to the origin.
An initial supercritical Hamiltonian, 1’, is, mapped on the
Hamiltonians 2’, 3’, 4’, etc., moving off to infinity. The
dashed line represents the locus of the laboratory system,
with fixed physical coupling constants, as the temperature
changed.
origin is a fixed point for the RG transformation, with all the supercritical Hamiltonians belong
to the basin of attraction associated with that fixed point. These considerations can be repeated
for supercritical Hamiltonians which end at some of the fixed points at the infinity.
The critical Hamiltonian stays at the critical line, so this line is an invariant space for RG
transformation. Following K. Wilson, let us take for granted that under repeated mapping the
critical Hamiltonian will converge to a fixed point Hamiltonian, H

. However, one can imagine
several fixed points with its own basin of attraction. This picture is usually represented as a
RG flow diagram, Fig. 12.12 The consideration above can be generalized for the case of multi-
K
K
e
e
1
2
Figure 12.12: RG flow lines.
dimensional parameter space which we shall not do here.
As we shall see, the critical exponents are determined by the properties of RG transformation
near fixed points. That gives a new understanding of the concept universality class.
Fixed point properties
The definition of fixed point is
K

= K
/
(K

).
Near the fixed point one can linearize the mapping as
δK
/
α
=

β
T
αβ
δK
β
, δK = K−K

.
12.7. RENORMALIZATION FRAMEWORK 165
Let us assume that eigenvalues of the matrix
ˆ
T are real and positive. Since
ˆ
T is asymmetric
one must discriminate between left and right eigenvectors. The left eigenvector belonging to the
eigenvalue λ
i
is defined as

Φ
i

ˆ
T =λ
i

Φ
i
.
One can introduce the scaling field as
u
i
=

Φ
i
δK.
Then the rescaled field will be
u
/
i
=

Φ
i
δK
/
=

Φ
i

ˆ
T δK =λ
i

Φ
i
δK =λ
i
u
i
.
Then one can assume that n repeated RG transforms with rescaling factor l are equivalent to a
single transform with a scaling factor l
n
,

i
(l)]
n

i
(l
n
).
The solution of this equation has the form
λ
i
(l) = l
y
i
where y
i
is the critical exponent.
Usually 3 classes of the eigenvalues are discussed.
• Irrelevant, with λ
i
< 1, i. e. y
i
< 0. The corresponding scaling fields decay exponentially
under RG transformations.
• Relevant, with λ
i
> 1, i. e. y
i
> 0. They grow exponentially and remove the system from
the critical point.
• Marginal, with λ
i
= 1, i. e. y
i
= 0.
Connection to Physics
Scaling fields are convenient variable to analyze the equation
G(u) =
1
l
d
G(u
/
).
For magnetic system there are only 2 relevant fields, u
τ
→τ and u
h
→h. Thus we have
G(τ, h) =
1
l
d
G(τ
/
, h
/
) =
1
l
d
G(l
y
τ
τ, l
y
h
h).
This is nothing else then the Widom’s homogeneity postulate with replacement
l
d
→λ, y
τ
/d →a, y
h
/d →b.
166 CHAPTER 12. CONTINUOUS PHASE TRANSITIONS
Summary
At present time, the RG theory is only an appealing picture, a seemingly fruitful way to think
about critical phenomena. The precise conditions must be met to make the transform accept-
able are less than clear. The main requirement is to leave long range fluctuations essentially
untouched.
According to K. Wilson, “There is no RG cookbook!”.
Chapter 13
Transport phenomena
13.1 Classical transport
Boltzmann equation
Let us consider electrons in a solid state as an example. We shall concentrate on the transport
under influence of electric and magnetic field.
According to classical physics, one can specify the coordinate r and the momentum p of the
particle. Thus one can calculate the non-equilibrium distribution function, f
p
(r, t), the current
density being
j(r, t) = e

(dp)v f
p
(r, t).
Here we denote (dp) ≡ d
d
p/(2π¯ h)
d
. The distribution function is defined by the Boltzmann
equation
d f
p
(r, t)
dt

∂ f
∂t
+
∂ f
∂r
dr
dt
+
∂ f
∂p
dp
dt
+I
coll
=
∂ f
∂t
+v
∂ f
∂r
+F
∂ f
∂p
+I
coll
= 0
Here
F = e

E+
1
c
[vH]

is the force acting upon the electrons, v ≡ ∂ε
p
/∂p is the (group) electron velocity, while I
coll
is
the collision operator. It expresses the changes in the state due to quantum collisions and can
be expressed through the the transition probability W
i f
between the initial state (i) and the final
state ( f ),
I
coll
( f
α
) =

α
/
[W
αα
/ f
α
(1− f
α
/ ) −W
α
/
α
f
α
/ (1− f
α
)] .
Here α denotes the electronic state (in our case α ≡¦p, s¦).
167
168 CHAPTER 13. TRANSPORT PHENOMENA
In a stationary situation and in the absence of external fields the solution of the Boltzmann
equation is the Fermi function. Substituting it into the collision operator we obtain a very impor-
tant relation between the probabilities of the direct and reverse processes,
W
αα
/ e
−ε
α
/ /kT
=W
α
/
α
e
−ε
α
/kT
.
Drude formula
The simplest problem is to calculate the linear response to a small stationary electric field, E.
Since we assume E to be small, it is reasonable to search solution as
f
p
(r) = f
0

p
) + f
(1)
, [ f
(1)
[ < f
0
.
Since I
coll
( f
0
) = 0 we come to the integral equation
I
coll
( f
(1)
) =−eE
∂ f
0

p
)
∂p
= eEv


d f
0

p
)

p

.
The linearized collision operator has the simplest form for elastic processes when W
pp
/ =W
p
/
p
.
Then,
I
coll
( f
(1)
) =

p
/
W
pp
/

f
(1)
p
− f
(1)
p
/

.
Since we are interested in the function odd in p (to create a current) and it must be a scalar it is
natural to assume that
f
(1)
p
∝ (p ν), ν ≡E/E .
Under such an assumption
I
coll
( f
(1)
) =
f
(1)
p
τ
tr
,
1
τ
tr
=

p
/
W
pp
/

p
ν
−p
/
ν
p
ν

.
The quantity τ
tr
is called the transport relaxation time. If the material is isotropic then W is depen-
dent only on the scattering angle θ between p and p
/
. From the simple geometrical construction
shown in Fig. 13.1 we observe that
p
/
ν
= p
/
cosθ cosφ, p
ν
= pcosφ.
Consequently, we can express the relaxation time in the form
1
τ
tr
= g(ε
p
)
1
2

π
0
dθ sinθ(1−cosθ)W(θ).
Using the expression for the relaxation time we can write
f
(1)
p

tr
e(E v)


d f
0

p
)

p

→ j = e
2

(dp)v(E v)τ
tr

p
)


d f
0

p
)

p

.
13.1. CLASSICAL TRANSPORT 169
p
p’
θ
φ
ν
Figure 13.1: On the calculation of the transport relaxation time.
As a result, we arrive at the Ohm’s law, j =σE with
σ =


0
dεg(ε)'v
2
E
τ
tr
`
ε


d f
0
(ε)

. (13.1)
Here v
E
means the projection of the velocity on the direction of the electric field, while '. . .`
ε
means the average over the surface of a constant energy ε,
'A(p)`
ε

(dp)A(p)δ(ε −ε
p
)

(dp)δ(ε −ε
p
)
.
The quantity
D ='v
2
E
τ
tr
`
ε
=
1
d
v
2
τ
tr
=
1
d
v
has an explicit physical meaning. This is just the diffusion constant for the electrons with a given
energy ε. Indeed, for low temperatures we get
σ = e
2
g(ε
F
)D(ε
F
). (13.2)
On the other hand, the phenomenological expression for the current density in the presence of
the density gradient reads as
j =σE−eD∇n,
where D is th diffusion constant. In the equilibrium the current is zero and
eD∇n = eDg(ε
F
)∇ζ =−σ∇ϕ.
At the same time, the electro-chemical potential ζ +ϕ/e, must be constant and ∇ϕ = −e∇ζ.
Thus we identify the quantity with the diffusion constant. Eq. (13.2) is known as the Einstein
relation.
170 CHAPTER 13. TRANSPORT PHENOMENA
13.2 Ballistic transport
Landauer formula
We start this chapter by a description of a very powerful method in physics of small systems -
so-called Landauer approach.
The main principle of this approach is the assumption that the system in question is coupled
to large reservoirs where all inelastic processes take place. Consequently, the transport through
the systems can be formulated as a quantum mechanical scattering problem. Thus one can reduce
the non-equilibrium transport problem to a quantum mechanical one.
Another important assumption is that the system is connected to reservoirs by ideal quan-
tum wires which behave as waveguides for the electron waves. We start our analysis from the
discussion of the properties of an ideal quantum wire.
Ideal quantum wire
Consider 2 large reservoirs of electron gas reservoirs having the difference δn in the electron
density and separated by a pure narrow channel. For small δn one can assume that there is a
difference in a chemical potential, δµ =δn/g(ε
F
). In the following we shall use the Fermi level
of non-biased system as the origin for the chemical potentials. So the difference between the
chemical potential in α-th reservoir will be denoted as µ
α
.
If the channel is long and uniform, then the total current carried by the state characterized by
a transverse mode n and a given direction of spin which propagates without scattering is
J
n
= e

dk
z
2π¯ h
∂ε
n
(k
z
)
∂k
z
=
2
2π¯ h

ε
F

β
ε
F

α

∂ε
n
(k
z
)/∂k
z
[∂ε
n
(k
z
)/∂k
z
[
=
2
h
δµ.
If we take into account electron spin and N transverse modes are open, then the conductance is
given by the expression G =
2e
2
h
N.
We come to a very important conclusion: an ideal quantum wire has finite resistance h/2e
2
N
which is independent of the length of the wire.
As we have seen, even an ideal quantum wire has a finite resistance. That means a finite heat
generation even in the absence of any inelastic processes inside the wire. Below we will discuss
the physical picture of heat release by a current-carrying nanostructure.
First of all let us specify what heat release is. It will be convenient to consider an isolated
system. Therefore we will have in mind the following physical situation. There is a capacitor
which is discharged through the conductor of interest. The product RC of the whole system,
R and C being the resistance and capacitance respectively, is much bigger than any relaxation
time characterizing the electron or phonon system of the conductor. This means that for all the
practical purposes the conduction process can be looked upon as a stationary one. The total
energy of the system, U, is conserved, while its total entropy,
ˆ
S, is growing. The rate of heat
generation is expressed through T∂
ˆ
S/∂t, where T is the temperature, i.e. through the applied
voltage and characteristics of the nanostructure itself. This means that the result is independent
of the assumption that the considered system is isolated, which is made only for the sake of
13.2. BALLISTIC TRANSPORT 171
derivation. This thermodynamically defined heat is generated in the classical reservoirs over the
length having a physical meaning of the electron mean free path. That is the same mean free path
that enters the Drude formula, which determines the conductivity of the reservoirs themselves,
the amount of heat generated per second in both reservoirs being the same.
It is interesting to indicate that even purely elastic collisions can result in a heat generation
although they of course cannot establish full equilibrium. This has a clear physical meaning.
The amount of order in the electron distribution resulting in electric current can bring about
mechanical work. For instance, one can let the current flow through a coil, and a magnetic rod
can be drawn into the coil. In such a way the electrons transferring the current can execute a work
on the rod. As a result of scattering, the amount of order in the electrons’ distribution diminishes,
and this means dissipation of mechanical energy into the heat. It has been shown that the heat
release is symmetric in both reservoirs even if the scatterers in the system are asymmetric.
All the above considerations do not mean that the collisions that give the main contribution
to the heat release, also establish full equilibrium. What equilibrium needs is inelastic colli-
sions which transfer the energy of electrons taking part in charge transfer to other degrees of
freedom, such as to other electrons and phonons. In particular, a local equilibrium electron dis-
tribution is established over the length scale determined by electron-electron interaction. Such a
distribution can be characterized by a local electro-chemical potential and sometimes an electron
temperature. The latter can in principle be measured by optical methods. On the other hand, the
equilibrium with respect to the lattice is established at the scales of electron-phonon and phonon-
phonon mean free paths. Only over those distances from the channel one can treat the results in
terms of the true local temperature.
Resistance of a quantum resistor
Consider a system shown in Fig. 13.2 consisting of a barrier connected to reservoirs by ideal
quantum wires. If there is some reflection only a part of the current is transmitted. In this
Figure 13.2: On the resistance of a quantum resistor.
case one can introduce the transmission probability of the mode n, T
n
, to obtain (including spin
degeneracy)
J =
2
h
δµ
N

n=1
T
n
.
As a result,
G =
2e
2
h
N

n=1
T
n
=
2e
2
h
Tr tt

. (13.3)
172 CHAPTER 13. TRANSPORT PHENOMENA
Here t is the matrix of scattering amplitudes while the expression is called two-terminal Landauer
formula.
This very important and looking simple formula was confusing during a long period. Indeed,
this is the conductance which is measured between two reservoirs. Having in mind that the
resistance of the connecting ideal wires (per one conducting mode) is h/2e
2
we can ascribe to
the scattering region the resistance
h
2e
2
¸
1
T
−1

=
h
2e
2
R
T
,
where R is the reflection coefficient. Consequently, in the original formulation the quantum
resistance was described as
G =
2e
2
h
N

n=1
T
n
1−T
n
. (13.4)
However, the quantity which is usually measured is given by Eq. (13.3).
Now we derive the Landauer formula for finite-temperature and so-called multichannel case
when the leads have several transverse modes. Consider ideal wires which lead to a general
elastic scattering system. Let each lead has the cross section A and have N

transverse channels
characterized by wave vectors k
i
so that,
E
i
+
¯ h
2
k
2
i
2m
= E
F
.
The incoming channels are fed from the electron baths with the same temperature and chemical
potentials µ
1
, µ
2
, . . .. The outgoing channels are fed up to thermal equilibrium population. We
shall assume that the particles are absorbed in the outgoing baths. The sources are assumed to be
incoherent, the differences µ
1
−µ
2
are also assume small to yield linear transport. We introduce
the scattering amplitudes t
i j
for the transmission from jth incoming to ith outgoing channel.
Reflection amplitudes r
i j
are introduces in a similar way for reflection into the ith incoming
channel. If we replace the incoming and outgoing channels, we denote the proper amplitudes by
primes. In this way it is convenient to introduce 2N

2N

scattering matrix as
S =

r t
/
t r
/

.
From the current conservation we must require unitarity while from time reversal symmetry
S =
˜
S. Thus we have also SS

= I where star stays for complex conjugation while tilde for
transposition. In a magnetic field the Onsager relation requires S(H) =
˜
S(−H).
It one defines the total transmission and reflection into ith channel as
T
i
=

j
[t
i j
[
2
, R
i
=

j
[r
i j
[
2
.
then from unitarity condition we get

i
T
i
=

i
(1−R
i
).
13.2. BALLISTIC TRANSPORT 173
Since the densities of states in each channel are 1D like, g
i
(E) = (π¯ hv
i
)
−1
we write the current
through outgoing channels as
I =
e
π¯ h

i

dE

f
1
(E)T
i
(E) + f
2
(E)R
/
(E) − f
2
(E)

=

1
−µ
2
)e
π¯ h

dE


∂ f
∂E


i
T
i
(E).
Thus the conductance becomes
G =
2e
2
h

dE


∂ f
∂E

Tr tt

.
This is the two-terminal conductance measured between the outside reservoirs which includes
contact resistances.
Point ballistic contact
The most clean system is the so-called quantum point contact (QPC) - short and narrow con-
strictions in 2d electron gas. A sketch of QPC is shown in Fig. 13.3 The conductance of QPC
Figure 13.3: A sketch of a QPC
formed by split gates.
is quantized in the units of 2e
2
/h. The quantization is not that accurate as in quantum Hall ef-
fect (about 1%) because of non-unit transparencies T
n
and finite temperature. It is interesting to
compare quantum and classical behavior of QPC. In a classical picture one can write
J =W(δn)v
F

π/2
−π/2


cosα =
1
π
Wv
F
(δn).
Thus the “diffusion constant” is
D
e f f
=
J
δn
=
1
π
Wv
F
→ G = e
2
g(ε
F
)D
e f f
=
2e
2
h
k
F
W
π
.
Note that the integer part of the quantity k
F
W/π is just the number of occupied modes according
to quantum mechanics.
174 CHAPTER 13. TRANSPORT PHENOMENA
Gate voltage
C
o
n
d
u
c
t
a
n
c
e

(
2
e


/
h
)
2
Gate
Figure 13.4: Quantization of conductance of a point
contact: Schematic picture (left) and experimental re-
sult (right).
F
E + eV E
F
α
W
Figure 13.5: On the classical conductance of a
point contact.
Appendix A
Notations
Microscopic variables
q →(q
1
, q
2
, . . . , q
f
) - generalized coordinates,
p →(p
1
, p
2
, . . . , p
f
) - conjugated momenta,
H(q, p) - classical Hamiltonian,
ˆ
H - quantum Hamiltonian,
dΓ ≡∏
f
j=1
dq
j
dp
j
- element of phase space,
m - mass of a molecule,
Γ(E, . . .) - number of states with the energy ≤E ,
D(E, . . .) =∂Γ/∂E = Tr δ

E −
ˆ
H

=
1
V

dΓδ[E −H(p, q)] - state density,
S - spin,
c - velocity of light,
g
i
- degeneracy of i-th state.
Macroscopic variables
V - volume of the box which contains the system,
P - pressure,
T - absolute temperature (we use also energy units),
β = 1/T - inverse temperature,
E - total energy,
U - internal energy, S - entropy,
S
t
- total entropy for the body and surrounding medium,
F - (Helmholtz) free energy,
H - thermal function, enthalpy,
G - Gibbs free energy, thermodynamic potential,
Ω - Landau free energy.
175
176 APPENDIX A. NOTATIONS
Electric and Magnetic Variables
B - magnetic field,
e - electronic charge,
µ = e/2mc - Bohr magneton.
Constants
k = 1.3810
−16
erg/K - Boltzmann constant,
h - Planck constant, ¯ h ≡h/2π ,
Useful mathematical formulas
The Stirling’s formula:
logn! ≈n logn−n+1 at n 1.
or more accurate
n! ≈

n+
1
6

n
n
e
−n
.
Thermodynamic relations
Thermodynamic potentials
Thermodynamic potential Notation Independent variables Differential
Internal energy U S, V, N T dS−PdV +µdN
Heat function (enthalpy) H S, P, N T dS+V dP+µdN
Helmholtz free energy F T, V, N −SdT −PdV +µdN
Gibbs free energy G T, P, N −SdT +V dP+µdN
Landau free energy Ω T, V, µ −SdT −PdV −Ndµ
Table A.1: Thermodynamic potentials (summary)
Basic distributions
The Gibbs distribution
w
n
=
1
Z
e
−βE
n
, Z =

s
e
−βE
s
.
177
The Maxwell-Boltzmann distribution
f (p, q) = f
p
(p) f
q
(q);
f
p
(p) = (2βπm)
−3/2
e
−β(p
2
x
+p
2
y
+p
2
z
)/2m
, β ≡1/T ;
f
q
(q) =
¸

dqe
−βV(q)

−1
e
−βV(q)
.
The Fermi-Dirac distribution
f
k
=

e
β(ε
k
−µ)
+1

−1
.
The Bose-Einstein distribution
f
k
=

e
β(ε
k
−µ)
−1

−1
.
The chemical potential µ is determined by the normalization
V

k
g
k
e
β(ε
k
−µ)
±1
= N.
Relations between thermodynamic and statistical quantities
Z =

n
e
−βE
n
;
F = −T logZ;
Ω = −T log

N

e
βµN

s
e
βE
sN

.
178 APPENDIX A. NOTATIONS
Appendix B
On the calculation of derivatives from the
equation of state
Important relations arise from the properties of partial differentiations. Consider 3 quantities,
X,Y, Z, related by the equation of state K(X,Y, Z) = const. Now let us consider X,Y as indepen-
dent variables, while Z = Z(X,Y). We have

∂Z
∂X

Y
dX +

∂Z
∂Y

X
dY −dZ = 0. (B.1)
If Y, Z are taken as independent variables, then
−dX +

∂X
∂Y

Z
dY +

∂X
∂Z

Y
dZ = 0 (B.2)
Now we multiply Eq. B.1 by

∂X
∂Y

Z
and Eq. B.2 by

∂Z
∂Y

X
and subtract the results. We obtain
¸
∂Z
∂X

Y

∂X
∂Y

Z
+

∂Z
∂Y

X

dX +
¸

∂X
∂Y

Z

∂X
∂Z

Y

∂Z
∂Y

X

dZ = 0.
Since dX and dY are independent, this equation is compatible if

∂Z
∂X

Y

∂X
∂Y

Z
+

∂Z
∂Y

X
= 0

∂X
∂Y

Z
+

∂X
∂Z

Y

∂Z
∂Y

X
= 0,
or

∂X
∂Y

Z

∂Y
∂Z

X

∂Z
∂X

Y
=−1, (B.3)

∂X
∂Y

Z
=

∂Y
∂X

−1
Z
. (B.4)
179
180APPENDIXB. ONTHE CALCULATIONOF DERIVATIVES FROMTHE EQUATIONOF STATE
General scheme for transformation: Consider any quantity F (X,Y), the differential of which
can be expressed as
dF =

∂F
∂X

Y
dX +

∂F
∂Y

X
dX .
Then we divide it by dZ and assume Y = const, (dY = 0). We get

∂F
∂Z

Y
=

∂F
∂X

Y

∂X
∂Z

Y
. (B.5)
Another important relation arises if one divides the expression for dF by xX,

∂F
∂X

Z
=

∂F
∂X

Y
+

∂F
∂Y

X

∂Y
∂X

Z
. (B.6)
Equations (B.5), (B.6) together with Eqs. (B.3), (B.4) and the Maxwell relations are usually
used for transformations and computation of derivatives from the equation of state.
Bibliography
[1] L. D. Landau and E. M. Lifshitz, Statistical Physics, 3rd edition, Part 1 (Pergamon Press,
Oxford - New York - Seoul - Tokyo), 1980.
181

Contents
1 Introduction 1.1 Motivation . . . . . . . . . . . . 1.2 Deterministic Chaos . . . . . . . 1.3 Atomistic Philosophy . . . . . . 1.3.1 Cooperative Phenomena 1.4 Outlook . . . . . . . . . . . . . 1.5 What will we do? . . . . . . . . 2 Historical Background 2.1 Temperature . . . . . . . . 2.2 Steam Engines . . . . . . 2.3 Sadi Carnot . . . . . . . . 2.4 The Maxwell’s Distribution 1 1 2 3 4 4 5

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

7 . 8 . 9 . 10 . 18 21 21 22 24 26 27 29 30 31 31 32 33 34 35 36 36 37 38

3 Principles of statistical mechanics 3.1 Microscopic states . . . . . . . . . . . . . . 3.2 Statistical treatment . . . . . . . . . . . . . . 3.3 Equal weight, . . . microcanonical ensemble . . 3.4 Entropy . . . . . . . . . . . . . . . . . . . . 3.5 Number of states and density of states . . . . 3.6 Information . . . . . . . . . . . . . . . . . . 3.7 Normal systems in statistical thermodynamics

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

4 Thermodynamic quantities 4.1 Temperature . . . . . . . . . . . . . . . . . . . . 4.2 Adiabatic process . . . . . . . . . . . . . . . . . 4.3 Pressure . . . . . . . . . . . . . . . . . . . . . . 4.4 Equations of state . . . . . . . . . . . . . . . . . 4.5 Work and quantity of heat . . . . . . . . . . . . . 4.6 Thermodynamics potentials . . . . . . . . . . . . 4.6.1 The heat function (enthalpy) . . . . . . . 4.6.2 Free energy and thermodynamic potential 4.6.3 Dependence on the number of particles . i

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . . .4. . . . . . . . .3 Markov processes . . . . . . . . . Phonons . . . . . . . . 9. Thermodynamic stability 4. . . . . . . . . . . . . 4. . . . . . . . . . . . . . . Principles of thermodynamic . . . . . . . . . . . . . . . . . .6. 87 8 Fluctuations 8. . . . . .4 Black-body radiation . . . . 4. . . . . . . . . . .ii 4. . . . . . . . . . . . . . .1 Spin 1 in a magnetic field . . .1 The Gaussian distribution . . . . . . . . . . . . . . . . . . . . . . . . . . .6 Fluctuations in a vacuum-tube generator .2 Canonical ensembles . . . . . . Fokker-Planck equation 9. . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . .1 Microcanonical ensemble . 4. . .3 Nernst’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8. . . . . . . . . . . . . . . . . . . . . . . . . . .1 Random walks . . . . . . . . . . . . . . .7. . . . 8. . . . . . . . . . . . . . . 6. . . . . . . . . . . . . . . 9. .4 Fluctuation-dissipation theorem: Proof 91 91 93 95 101 101 103 104 107 111 111 113 115 116 117 119 121 . .6. . . . . . . . . . .5 Lattice Vibrations. . . . . . . 2 5. 5.2 Ideal gases out of equilibrium . . . . . . . . . . . . . . . . . . . . . . 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . .7. . . . . . . . . . .1 Relation to diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Stochastic Processes 9. . . . . . . . . . . . . . . . .2 Quantum systems . . . . . .4 Discrete Markov processes. . . .1 The Gibbs distribution for a variable number of particles . . . . . . . . . . . .3 Fermi and Bose gases of elementary particles 6. . . . . . . . . . . . .7 5 The Gibbs Distribution 5. . 8. . . . . . . . . . . . . . Master equation . .2. . . . . . . . . . . . . . .4 Derivatives of thermodynamic functions . . . . . . . . . . . . . . . . . . . . . . . 6. . . . 6 Ideal gas 6. . .4 Fluctuation-dissipation theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Fluctuations of thermodynamic quantities . . . . . . . . . . . . . . . . . . . . . . . 9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8. . .2 Random pulses and shot noise . .7. . . . . 7 Statistical ensembles 81 7. . . . . . . . . . . . 39 41 42 42 45 46 46 49 49 51 53 55 55 64 65 70 72 4. . . . . . . . . . . 8. 81 7. . . . . . . . . . . . . . . . . . . 8. . . . . . . . . . . . . . . . . 83 7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Thermodynamic inequalities. . . . . . . . .1 Maximum work and Carnot cycle . . . .3 Correlation of fluctuations in time . 8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9. . . . . 9. . . CONTENTS . . . . . . . . . . . . . . . . . . . . .7. .5 Continuous Markov processes. . . . . .4. . . . 6. . .2 The Gibbs distribution . . . . . .3 The generalized susceptibility . . . . . .4. . . . . . . . . . . . . . . . . . .1 Classical gas . . . .4 On phase transitions . . . . . . . .1 Classical systems . . . . . . .3 Ensembles in quantum statistics . . . . . . . . . . . . . . . . . . .5 Examples . .4. . . . . . . .

. . 167 13. . . 12. . . . . . . . . .3 Landau-Ginzburg theory . . . . . . . . . . . . . . . . . .1 Conditions for phase equilibrium . 12. . . . . 135 12 Continuous Phase Transitions 12. . . . . . . . . . . . . . . . . . .1 Landau Theory . . . . . . . . . . 12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7 Renormalization framework . . . . . . . . . . . . . . . . . . . . . . .5 Mean field theory . . . . . . 139 142 149 150 153 159 160 161 . . . . . . . . . . . . . . . . . . . . . 12. . . . . . . . . . . . . . . . . . . . .2 Ballistic transport . . . . . . . . . . . . . . . . .CONTENTS iii 10 Non-Ideal Gases 129 10. . . . 13 Transport phenomena 167 13. . . . .2 Scaling laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 11 Phase Equilibrium 135 11. . . . . . . . . . . . . . . . . . . . . . .4 Ginzburg-Landau theory of superconductivity (SC) 12. . . . . . . . . . . . . . . . . 170 A Notations B On the calculation of derivatives from the equation of state 175 179 . . . . . . . . . 12. . . . . . . .1 Classical transport . . . . . . . . . . . . . . . . . .6 Spatially homogeneous ferromagnets . . . . . . . . . . . .1 Deviation of gases from the ideal state . . . . . . . . . .

iv CONTENTS .

selforganized criticality (SOC). is a theory of impressive range of validity. statistical physics in its modern form has given us a complete understanding of second-order phase transitions. the properties of granular media. earthquakes. to black holes and elementary particles in the early universe in exactly the same form as they were originally formulated. fractals. With new concepts we look for features ignored in previous experiments and find exciting results.1 Statistical physics. Disordered systems. Key words are deterministic chaos. percolation theory and fractals find applications not only in physics and engineering but also in economics and other social sciences. What are the processes that describe the evolution of complex systems? Again we have an 1 From the introduction of Ravndal’s lecture notes on Statistical Physics 1 . Moreover. Many types of nonlinear systems that lead to complicated pattern formation processes. the field of statistical physics is in a phase of rapid change. such as the specific heats of solids for instance. Statistical physics allows not only the calculation of the temperature dependence of thermodynamics quantities. We do not have the theoretical framework in which to even attempt to describe highly irreversible processes such as fracture. on the other hand gives a rational understanding of Thermodynamics in terms of microscopic particles and their interactions. These words represent large fields of study which have changed how we view Nature. the conduction of heat and electricity for example. In the study of these and other macroscopic systems we build on the established conceptual framework of thermodynamics and statistical physics.Chapter 1 Introduction 1.1 Motivation Statistical physics is an unfinished and highly active part of physics. through quantum systems such as superconductors and nuclear matter. and with Wilson’s Renormalization Group theory we may calculate the scaling exponents observed in experiments on phase transitions. Thermodynamics. turbulence and intermitency. but also of transport properties. New ideas and concepts permit a fresh approach to old problems. friction and many other systems are beyond our present understanding and theoretical tools. put into its formal and phenomenological form by Clausius. Thermodynamics describes all systems form classical gases and liquids. However.

and oil per capita than we had when the report was written. Engineers who study road traffic reach the conclusion that more roads will make congestions worse. D ETERMINISTIC behavior rules the day. Since the initial condition cannot be specified with arbitrary precision. The central theme is that the time evolution of the system is sensitive so initial conditions. i. statistical physics provides us with many examples of complex systems and complex dynamics that we may understand in detail. subtle. electric oscillator circuits. If we start again with new initial condition. the “Club of Rome” report predicted that we would run out of resources very quickly. lasers. management and in teaching we almost invariably assume that any given action leads predictably to the desired result—and we are surprised and discouraged when it does not. arbitrarily close to the previous case. It is this incredible sensitivity to initial conditions for deterministic systems that has deep 2 Discuss the telephone system at the University of Oslo . So what has been going on? Why have the mysteries and excitement of statistical physics not been introduced to you before? The point is that statistical physics is unfinished. The concept of deterministic chaos has changed the thinking of anyone who gets some insight into what it means. thermal convection. Now. Deterministic chaos may arise in any non-linear system with few (active) degrees of freedom. not better—a statement that cannot be generally true. but also on the public sector to an extent that we may speak of as a paradigm shift.2 CHAPTER 1. Statistical physics provides an intellectual framework and a systematic approach to the study of real world systems with complicated interactions and feedback mechanisms. the orbit in fact becomes unpredictable—even in principle. The point is that in physics. a pendulum driven with a small force oscillating at a frequency ω. Of course. this did not happen. e. on the public arena. statistical physics cannot be applied to all areas. a contradiction of terms. driven pendulums. in industry. metals. g. intellectually and mathematically demanding. The human population would grow to an extent that would outstrip food. but its state as a function of time will have a distance (in some measure) that diverges exponentially from the first realization. Privately. Similarly. metals.2 The next level is that some feed-back is considered—but again things get too simplistic.2 Deterministic Chaos Deterministic chaos appears to be an oxymoron. for example. The activity in statistical physics has philosophical implications that have not been discussed by professional philosophers—and have had little influence on the thinking of the general public. I expect that the new concepts in statistical physics will eventually have a significant impact not only on other sciences. Consider a system that evolves in time as described by the deterministic equations of motion. However. with a much large population we have more food. oil and other resources. INTRODUCTION active field with stochastic processes as a key word. It is not. 1. and many other systems. Starting from given initial condition the pendulum will evolve in time as determined by the equation of motion. the pendulum will again evolve deterministically. e.

we observe cooperative phenomena. we only need to apply what is known practical tasks in engineering and industry so to say. What is going on? I believe the point is that the evolution of science in the 20th century has lead to a focus on the atomistic part of the atomistic philosophy. while physicist have been waiting for almost thirty years for the observation of the Higgs Boson predicted by the standard model. however. There is no evidence that there are open questions in the foundation of quantum chemistry.3 Atomistic Philosophy Now. for systems with many active degrees of freedom. but chaotic behavior has nevertheless been observed for the motion of Io. We have the exact solution for the Hydrogen atom. for the planetary system the time horizon is millions of years. that the experiments and theories are entirely consistent with this view—in spite of deterministic chaos. It is clear that the quantum mechanics required to describe the interactions of atoms in chemical systems is fully understood. the approximate solution for the Helium atom. Elementary particle theory clearly lies at the foundation of physics—but the rate of discovery has slowed to a glacial pace.1. one of Jupiter’s moons. predict lunar eclipse. no new fundamental laws of Nature are required to understand biochemistry. is that determinism and randomness are just two aspects of same system. We believe that with an atomistic understanding of the particles that constitute matter—and their interactions—it is merely an exercise in applied mathematics to predict the macroscopic behavior of systems of practical importance. What is going on? The point is that non-linear systems that exhibit deterministic chaos have a time horizon beyond which we cannot predict the time evolution because of the exponential sensitivity on initial conditions. ATOMISTIC PHILOSOPHY 3 philosophical implications. 3 The . For the weather we have a time horizon of two weeks in typical situations. the rotation of the earth as a function of time. self-organized criticality and their ilk. By 1655 Huygens had resolved the banding.3. we cannot. that cannot be explained by an atomistic understanding alone. The conclusion. There exist initial conditions that have time horizons further in the future. The planetary system is known to be chaotic! That is. 1. i. and numerical solutions for more complex atoms and their interactions in molecules. the relative positions of planets.3 Very accurate numerical solutions of Newton’s equations for the solar system also exhibit deterministic chaos. or universal behavior. We observe universality. not fully explained yet. even in principle. Let me explain: The atomistic philosophy is strong and alive. Molecular biology is at present the field of science that has the highest rate of discovery. e. Chaos is also required for an understanding of the banding of the rings of Jupiter. and finally by implication the brain—how we think and understand! What folly! Clearly something is wrong with this line of reasoning. Thus we know what is required for a rational understanding of chemistry— the work has been done. Quantum mechanics was a ring were first observed by Galileo in 1616 with a telescope he had built. It must be said. at least how I see it. behavior that does not depend on the details of the system. This certainly is an arrogant physicists attitude. Since we understand chemistry.

in an abstract sense they are the same phase transition! Ferro-electricity. Here we have a very active branch of statistical physics. This enthusiasm was appropriate and lead to a long string of discoveries (and Nobel prizes) that have changed our society in fundamental ways. wine. namely the understanding of the natural world in terms of the (newly gained) atomistic physics. Much was finished before the second World war. How can it be that important macroscopic properties do not depend on the details of the Hamiltonian? Phase transitions are also universal: Magnetism and the liquid–vapor transition belong to the same universality class. fill vessels. The intellectual resources were largely concentrated on particle physics research. was left to an uncoordinated little publicized and conceptually demanding effort. Take hydrodynamics for example. They flow. The understanding of cooperative phenomena is far from complete. which completely dominated physics in the first half of previous century. Suddenly we understood spectra. were made at an astounding rate. have surface waves. the black-body radiation. the other aspects of the atomistic philosophy. dedicated to basic research were set up. 1. experiment. molten steel.4 Outlook So what is the underlying conceptual framework on which physics attempts to fill the gap in understanding the macroscopic world? How do we connect the microscopic with the macroscopic?— of course. Quantum mechanics is. and their interactions gained momentum in the 50’s and truly fantastic research organizations of unprecedented size. and now computational modeling. all kinds of new particles. lasers and other engineering products that have changed our lives. and weapons. are all phenomena that do not depend on the details of their atomistic components and their interactions—they are cooperative phenomena.1 Cooperative Phenomena Cooperative phenomena are ubiquitous and well known from daily life. neutrons. were largely developed in 50’s and 60’s. The search for elementary particles. superconductivity. theory. Discoveries. the constituents of nuclear matter. liquid argon — all share the same hydrodynamics. antiparticles and much more. spinodal decomposition. . of course. by observation. electrons. Air. except for second order phase transitions. water and argon are definitely described by quite different Hamilton operators that encode their quantum mechanics. radioactivity. computers. communication. vortices and exhibit many other hydrodynamic phenomena—in spite of the fact that molten steel. this huge effort on the atomistic side. We have no general theory. protons. X-rays. 1. water. by the scientific approach.4 CHAPTER 1. to classify and simplify the general scaling and other fractal phenomena observed in highly non-equilibrium systems.3. whereas nuclear reactors. and the phenomenological equations for hydrodynamics. they from drops and bubbles. However. also at the basis of the electronics industry. INTRODUCTION fantastic break-through. such as CERN. left small resources—both human and otherwise— to other parts of physics.

Claude E. His measure of information is directly related to Boltzmann’s statistical measure of entropy. Shannon initiated information theory by his 1948 paper analyzing measures of information transmitted through a communication channel. who was awarded the 1982 Nobel prize in physics for Renormalization group theory that allows the calculation of scaling exponents at second order phase transitions. he proved that the framework of statistical physics could indeed tackle the problem of phase transitions. Thermodynamics was developed largely in the 19th century. the only concept required in all the three laws of thermodynamics. they provide anchor points for further investigation if your studies. He received the 1968 Nobel prize in chemistry for his work on irreversible thermodynamics.5. not [ 1 mv2 ] from Greek en in +ergon 2 work (or rather: at work) 4 Energy. By his exact solution of the Ising model.1. and of course entropy. was given a new start by Rudolf Clausius (1822–1888).5 What will we do? This is what now think we should do in this course. and explained why entropy must increase by his H-theorem. an old word first used by Thomas Young in 1807 to signify mv2 . and my own tastes. with significant developments in our century related to quantum mechanics and phase transitions. Thermodynamics in the present form was really formulated as an axiomatic system with the three laws of thermodynamics. made fundamental contributions to the modern formulation of statistical mechanics. and hope I do not overload the course. The central concept is energy4 . In this century Lars Onsager (1903–1976) made several outstanding contributions. James Clerk Maxwell (1831–1879) contributed the kinetic theory of gases and his famous velocity distribution. On the contrary the purpose of this course is to acquaint you with the central issues of statistical mechanics. It is better to understand the a number of central issues thoroughly. . I will be guided by Finn Ravndals notes. and for systems that start in an equilibrium state and wind up in an equilibrium state (explosions for example). Secondly: Statistical physics started with Daniel Bernoulli (1700-1792). in two-spatial dimensions. or later work require them. Ludwig Boltzmann (1844–1906) made the fundamental connection to kinetics and introduced the famous expression S = k lnW the statistical expression for the entropy. 1. we have a fantastic phenomenological theory: THER MODYNAMICS valid at or near thermal equilibrium. We cannot get into all the interesting things at the frontier of research. WHAT WILL WE DO? 5 First of all: For macroscopic systems. Josiah Willard Gibbs (1839–1903). The last break-trough contribution was by Kenneth Geddes Wilson (1936–).

INTRODUCTION .6 CHAPTER 1.

was established after Carnot’s discovery of the second law of thermodynamics. work and entropy grew gradually from the time of Galileo and culminated with the formulation of classical thermodynamics by Rudolf Clausius (1822– 1888). and the use of fire in cooking (heat) predates written history. For particles that interact with forces that may be derived from a potential the classical paths are strictly time-reversible. The concepts of temperature. a conclusion Nernst reached by experimental investigations of vapors at high temperature. He received the 1920 Nobel Prize for Chemistry for his discovery Classical mechanics is reversible. is involved in both principles. the so called first law of thermodynamics I.Chapter 2 Historical Background The sensation of hot and cold. In classical physics there is no reference point for entropy.from the Greek τρoπη (Verwandlung. a fundamental concept in thermodynamics. Now entropy is most readily understood in terms of statistical physics. Entropy. but it is a concept that is difficult to understand. and the total energy of the system is conserved. Entropy is a concept of great practical importance. i. This situation is not changed by quantum mechanics. e. that is temperature. The following quote from a nice little book: Steam Engine Principlesby Calvert (1991) illustrates the difficulty: 7 . e temperature and other aspects of the foundation of modern physics. transformation). The scientific understanding of temperature and heat is surprisingly recent. The deep problem is therefore to understand how entropy can always increase as equilibrium is approached. At this stage there were two principles of thermodynamics formulated by Clausius: I The energy of the universe is constant II The entropy of the universe tends to a maximum ` Clausius introduce the word entropy in 1865. solids at low temperature and galvanic cells. Emilio Segr` has written a wonderful account of the evolution of our understanding of heat. heat. Walther Hermann Nernst (1864–1941) formulated the third principle of thermodynamics: III The entropy vanishes for a system at zero absolute temperature. Interestingly the principle of energy conservation.

in classical systems that are ‘closed’ energy is indeed conserved. However. It is unprofitable to find an analogue for entropy. (2): 32 degrees for a mixture of ice and water. There are equivalent formulations of these two ‘laws’ of thermodynamics that give more insight. 1 the others are: pressure. In both cases the tables are prepared by the learned and they are intended to be used without question by the practical man. This was later inverted so that 0◦ is the melting point of snow. 2. Fahrenheit was the first to observe and describe super-cooling of water in small capillaries. Fahrenheit proposed a temperature scale still used today. HISTORICAL BACKGROUND A forth property is entropy. -9. and the temperature of the human body to twelve. Isaac Newton (1642– 1727) proposed a temperature scale in which the freezing point of water was set to zero. However the thermoscope had no fixed point and temperatures form different thermoscopes could not be compared. Anders Celsius (1701–1744) was a professor of astronomy at the university of Uppsala. i. but its changes can be precisely calculated. In 1742 he described his thermometer to the Swedish Academy of Sciences. there are objections to Clausius’ formulation of the first law: The energy universe is not conserved unless we take into account the equivalence of mass and energy as stated in Einstein’s famous formula E = mc2 . Fahrenheit used three fixed points: (1): 0 degrees for a mixture of ice. It is as much a handicap to try to understand what is going on in our steam engine without recourse to entropy tables or charts as it is for a coastwise navigator to reject the tide tables. temperature. It must be accepted and used. He used alcohol thermometers and was the first to make a mercury thermometer.8 CHAPTER 2. We now mostly use the Celsius temperature scale. Of course.4 ◦ C. Daniel Gabriel Fahrenheit (1686–1736) was born in Danzig but studied in Holland and worked in Amsterdam. Fahrenheit found that the water was liquid at 15 degrees. He used rain-water in a 1 inch bulb that he evacuated in the manner used by him to make thermometers. and 100◦ as the other. This temperature scale has the advantage that most temperatures experienced in daily life are all positive. and solidified immediately if he broke the tip of the capillary attached to the bulb to admit air. (3): 96 degrees at the body temperature of a ‘healthy’ person in the mouth or under the arm. There is none. water and salt. and density .1 Temperature Galileo (1564–1642) around 1592 built a fluid based thermoscope which could be used to show changes in temperature. The name was changed in 1948 to honor Celsius. Disciples of Galileo and others in the Academia del Cimento (in Florence) made systematic studies using a thermometer. Such changes are listed in tables and plotted on charts.1 This has no physical existence and it has an arbitrary datum. The thermometer had the melting point of snow as one fixed point at 100 ◦ C and the boiling point of water was set to 0 ◦ C. for the centigrade scale. e.

2. and thus made steam engines practical. Watt’s invention alone significantly reduced the amount of coal needed to perform a given amount of work.html> 1% to 2% . 2.2.2 Steam Engines The steam engine was developed and played an important role in the industrial revolution. The cylinder volume is filled and heated by the steam at atmospheric pressure. patented in 1768: “A New Invented Method of Lessening the Consumption of Steam and Fuel in Fire Engines. For instance. starting in England in the mid-18th century. The table shows that there was a dramatic increase in the use of steam engines only 120 years ago—now they are almost extinct.) The steam engine was further de- Figure 2. The excess water runs out of the tube (f).eb.com:180/cgi-bin/g?DocF=micro/632/91. for railroads. veloped in the coming years by many engineers without a deeper understanding of the efficiency of the engine. Steam turbines have replaced steam engines completely for electric power generation. 2 ‘Watt. There were dangers. Steam generated in the boiler is admitted to the cylinder by opening the valve (a) and the weight of the other end of the arm pull the piston upward. By closing the valve (a) and opening the valve (b) cold water is introduced into the cylinder. of course. When the cylinder has cooled down the cycle can start again. in 1877 there were in Germany 20 explosions of steam boilers injuring 58 persons.1: The Newcomen steam engine worked in a simple fashion. STEAM ENGINES 9 William Thomson (later Lord Kelvin) introduce the absolute temperature scale. in agriculture and later for electric power generation. The steam condenses and cools thereby reducing the pressure in the cylinder so that the piston moves down pushed by the atmospheric pressure. (More later) 2. 3 from James’ Britannica Online. The name we normally associate with the steam engine is James Watt (1736–1819) who introduced the condenser. In 1890 there were 14 such explosions (of a vastly larger number of boilers) injuring 18 persons. The maximum pressure difference that can be obtained is then given by the length of the tube (f) and the temperature of the surroundings. <http://www. Stationary steam engines were used in industry.1.”2 Watts invention greatly increased the efficiency3 of the steam engine patented in 1705 by Thomas Newcomen (1663–1729) used to drive pumps in coal mines (see Fig.

9 million steam engines in the world around 1890. Type of Steam engine Free standing Mobile Ship Number 29 895 5 442 623 1879 Horse power 887 780 47 104 50 309 Number 45 192 11 916 1 674 1879 Horse power 1 508 195 111 070 154 189 Table 2. in contrast with the Newcomen engine the steam was condensed in the small cylinder (K) immersed in cold water. Fourteen years later Clayperon’s work was expanded by William Thomson4 (1824–1907). however. 2. which is incorrect. who understood Carnot’s work and expressed the results in a more mathematical form in the paper published in 1834. and it became very difficult to obtain only a few years later.1: Steam engines in Prussia. However. with Carnot.and the up-stroke. various arrangements opened and closed valves automatically. changed by James Prescott Joule (1818–1889). e. Thomson studied Carnot’s paper in 1847 and realized that the Carnot cycle could be used to define an absolute temperature scale. This view was. 4 Thomson was knighted in 1866. he as late as in 1848 thought that heat was conserved. et sur les Machines Propres e a d´ velopper cette Puissance—his only published scientific work.2: The steam was generated at atmospheric pressure and introduced into the cylinder just as in the Newcomen engine. raised to the peerage in 1892 as Baron Kelvin of Largs. and together they established the correct interpretation and thereby the principle of conservation of energy.3 Sadi Carnot Thermodynamics as a field of science started with Sadi Carnot (1796–1831). But hardly anyone bought the e ´ book. HISTORICAL BACKGROUND Figure 2. .i. it worked both on the down. He published in 1824 remarkable bookR´ flexions sur la Puissance Motrice du Feu. It was Emile Clapeyron (1799– 1864). The central improvement was that there was no need to cool the cylinder between strokes.10 CHAPTER 2. However. The water in the condenser was pumped out. Globally there were about 1. The machine shown was double-acting.

however. their theory is very little understood. It appears that it must some day serve as a universal motor. impels our ships. and the attempts to improve them are still directed almost by chance. has given us the power to produce. he notes: The discovery of the steam-engine owed its birth. SADI CARNOT 11 The beginning of Sadi Carnot’s book gives a feel for the technological optimism at the beginning of the 19th century. When they enter the atmosphere. while the real author is not certainly known.3. excavates our ports and our rivers. to appropriate it to our uses. From this immense reservoir we may draw the moving force necessary for our purposes.5 the currents of water which channel the surface of the globe. is the object of heat-engines. To heat also are due the vast movements which take place on the earth. less in the first attempts that the principal discovery consists. notwithstanding the satisfactory condition to which they have been brought today. to rude attempts which have been attributed to different persons. and they seem destined to produce a great revolution in the civilized world. fashions wood. the ascension of clouds. the fall of rain and of meteors. That it possesses vast motivepower no one can doubt. It is. spins and weaves our cloths. like most human inventions. The study of these engines is of the greatest interest. The path for the original invention to the final product of generations of engineers. transports the heaviest burdens. through a never ending series of improvements leads to qualitatively new products. To develop this power. is a misunderstanding. Even earthquakes and volcanic eruptions are the result of heat. and the motivation of his work: EVERY one knows that heat can produce motion. in providing us with combustibles on all sides. at all times and in all places. Here Carnot expresses. as between the first raft that man ever made and the modern vessel. in these days when the steam-engine is everywhere so well known. It causes the agitations of the atmosphere. After a discussion of the importance of steam-engines for England. forges iron. Nature. and of which man has thus far employed but a small portion. There is almost as great a distance between the first apparatus in which the expansive force of steam was displayed and the existing machine. waterfalls. and be substituted for animal power. Already the steam-engine works our mines. and air currents. their use is continually increasing. 5 This . friction from the air cause them to heat and glow. the movement of meteors has nothing to do with heat.2. After this motivation of his work he approaches the more general scientific question: Notwithstanding the work of all kinds done by steam-engines. their importance is enormous. heat and the impelling power which is the result of it. the unreasonable effectiveness of engineering. etc. than in the successive improvements which have brought steamengines to the conditions in which we find them today. in an amicable fashion. grinds grains.

whatever the working substance and whatever the method by which it is operated. and concludes: The production of motive power is then due in steam-engines not to an actual consumption of caloric. HISTORICAL BACKGROUND The question has often been raised whether the motive power of heat6 is unbounded. on the contrary. whether atmospheric air. It becomes difficult to recognize its principles and study its laws. The latter may make use of any agent whatever. In order to consider in the most general way the principle of the production of motion by heat. it must be considered independently of any mechanism or any particular agent. We have considered it only in machines the nature and mode of action of which have not allowed us to take in the whole extent of application of which it is susceptible. and are seeking today. by chemical action such as combustion. which he the actually answers: The phenomenon of the production of motion by heat has not been considered from a sufficiently general point of view. but to its transportation from a warm body to a cold body. We shall see shortly that this principle is applicable to any machine set in motion by heat. It has. We propose now to submit these questions to a deliberate examination. that is. We have long sought. to its re-establishment of equilibrium an equilibrium considered as destroyed by any cause whatever.12 CHAPTER 2. for example. for every machine of which caloric is the motor. would not present in this respect great advantages. Carnot goes on to describe the workings of the steam-engine. to ascertain whether there are in existence agents preferable to the vapor of water for developing the motive power of heat. the product of the weight multiplied by the height to which it is raised. [Here. even though his concept of heat (or caloric) was inadequate] 7 We distinguish here the steam-engine from the heat-engine in general. but also for every heat-engine—that is. Heat can evidently be a cause of motion only by virtue of the changes of volume or of form which it produces in bodies. incomplete. in a footnote. as we know. In such machines the phenomenon is. as a measure. We have shown that in steam-engines the motive-power is due to a re-establishment of equilibrium in the caloric. 6 We .. this takes place not only for steam-engines. use here the expression motive power to express the useful effect that a motor is capable of producing. or by any other. or whether. in a way.. It is necessary to establish principles applicable not only to steamengines7 but to all imaginable heat-engines. of the vapor of water or of any other to develop the motive power of heat. these improvements may be carried on indefinitely. Then he really becomes precise and defines the scientific question. . Carnot gives a clear definition of what we today call free energy. This effect can always be likened to the elevation of a weight to a certain height. whether the possible improvements in steam engines have an assignable limit—a limit which the nature of things will not allow to be passed by any means whatever.

or fact which at least appears evident as soon as we reflect on the changes of volume occasioned by heat: wherever there exists a difference of temperature. that of A being higher than that of B. and ask what quantity of motive power can be produced by the passage of a given portion of caloric (for example. such as a cylinder furnished with a piston. We will try to answer these questions. If we wish to produce motive power by carrying a certain quantity of heat from the body A to the body B we shall proceed as follows: (1) To borrow caloric from the body A to make steam with it—that is. a permanent gas. Then he presents the first sketch of the cycle: Imagine two bodies A and B. in a footnote (see below) he evades the question. SADI CARNOT 13 Here. However. or rather of the metal composing the boiler in ordinary engines—we here assume that the steam is produced at the same temperature as the body A. Carnot has unclear concepts about the role of heat. for example. We will call the first the furnace and the second the refrigerator. We have already remarked upon this self-evident fact. or to describe how we measure these quantities by the calorimeter.2. degree of temperature. We take. to make this body fulfill the function of a furnace. motive power can be produced. or does it vary with the agent employed to realize it as the intermediary substance. Nor will we explain what is meant by latent heat. of mercury. whether the vapor of water offers in this respect more or less advantage than the vapor of alcohol.3. etc. to increase the volume of this space. In that process there is caloric or heat flowing from the hot to the cold. or any other substance. (2) The steam having been received in a space capable of expansion. one body A kept at a temperature of 100◦ and another body B kept at a temperature of 0◦ . These two bodies. availing ourselves of ideas already established. as much as is necessary to melt a kilogram of ice) from the first of these bodies to the second. exercise the functions of two unlimited reservoirs of caloric. to which we can give or from which we can remove the heat without causing their temperatures to vary. We inquire whether this quantity of motive power is necessarily limited. kept each at a constant temperature. 8 It . whether it varies with the substance employed to realize it. The reader should be familiarized with these terms through the study of the elementary treatises of physics or of chemistry. selected as the subject of action of the heat? It is clear that this question can be asked only in regard to a given quantity of caloric. or caloric. Then Carnot rephrases the problem he is considering: It is natural to ask here this curious and important question: Is the motive power of heat invariable in quantity.8 the difference of the temperatures also being given. Here he announced the basic principle: A temperature difference is required in order to generate work. trying to re-establish thermal equilibrium. he is clear in the sense that a temperature difference is required if work is to be generated. specific heat. and is considered unnecessary to explain here what is quantity of caloric or quantity of heat (for we employ these two expressions indifferently).

Here. First the continuation: By our first operations there would have been at the same time production of motive power and transfer of caloric from the body A to the body B. the temperature will fall spontaneously. whereas in (2) he describes and isentropic process. HISTORICAL BACKGROUND consequently also that of the steam. and so . as occurs with all elastic fluids. to restore the initial conditions. By the inverse operations there is at the same time expenditure of motive power and return of caloric from the body B to the body A. But if we have acted in each case on the same quantity of vapor. The body B fills here the place of the injectionwater in ordinary engines. it would suffice to divert a portion of this power in order by the method just indicated to make the caloric of the body B return to the body A from the refrigerator to the furnace. Thus rarefied. However. that it condenses the vapor without mingling with it. so that an indefinite number of alternative operations of this sort could be carried on without in the end having either produced motive power or transferred caloric from one body to the other. and at the temperature of that body. and in contradiction with what he writes later. and continuing the compression to complete liquefaction. Carnot introduced several of the basic concepts of Thermodynamics: heat reservoir and thermodynamics process. (3) To condense the steam by putting it in contact with the body B. and without changing its own temperature. with this difference. compressing it in such a way as to make it acquire the temperature of the body A. Carnot continues in a fashion that is difficult to understand. if there is produced no loss either of motive power or caloric. Reversibility reversibility is the next concept: The operations which we have just described might have been performed in an inverse direction and order. the quantity of motive power produced in the first place will be equal to that which would have been expended in the second. There is nothing to prevent forming vapor with the caloric of the body B. In (1) and (2) he describes an isothermal process. This then completes the Carnot cycle. by the assumption that caloric is conserved.14 CHAPTER 2. and the quantity of caloric passed in the first case from the body A to the body B would be equal to the quantity which passes back again in the second from the body B to the body A. admit that the rarefaction may be continued to the point where the temperature becomes precisely that of the body B. he also gets no work done by his cycle! Nevertheless he an-ounces his second principle: Now if there existed any means of using heat preferable to those which we have employed. if it were possible by any method whatever to make the caloric produce a quantity of motive power greater than we have made it produce by our first series of operations. Here. and thus to be ready to commence again an operation precisely similar to the former. finally condensing it by contact with this latter body. that is. and at the same time exerting on it a constant pressure until it is entirely liquefied.

The problem is to understands how he reaches these correct conclusions. We will take it for the basis of our demonstration. to the laws of mechanics and of sound physics. the temperature is higher during the movements of dilatation than during the movements of compression. Since every re-establishment of equilibrium in the caloric may be the cause of the production of motive power. for the similar positions of the piston. We should then conclude that the maximum of motive power resulting from the employment of steam is also the maximum of motive power realizable by any means whatever. every re-establishment of equilibrium which shall be accomplished without production of this power should be considered as an actual loss. 15 Here he has two more fundamental concepts: There is am maximum work attainable that is independent of what type of heat-engine is employed. on the contrary. SADI CARNOT on: this would be not only perpetual motion. and consequently the quantity of motive power produced by the movements of dilatation is more considerable than that consumed to produce the movements of .3. but an unlimited creation of motive power without consumption either of caloric or of any other agent whatever. Let us picture to ourselves now the series of operations which are to be described. then. But it should be remarked that with equal volumes. the elastic force of this air varies as much by reason of the changes in volume as of changes of temperature. based on a cycle that produces neither work nor a transport of caloric (heat) from the hot to the cold reservoir.. shut up in a cylindrical vessel. Fig. It falls. Let there be also two bodies. very little reflection would show that all change of temperature which is not due to a change of volume of the bodies can be only a useless re-establishment of equilibrium in the caloric. It is inadmissible. Carnot continues to explain that work is done: In these various operations the piston is subject to an effort of greater or less magnitude. Let us follow his argument for the second demonstration: First the basic assumption: When a gaseous fluid is rapidly compressed its temperature rises. that is. Such a creation is entirely contrary to ideas now accepted. abcd (Fig. 1). This is one of the facts best demonstrated by experiment. that of A being higher than that of B. 2. A and B. The necessary condition of the maximum is. exerted by the air enclosed in the cylinder.. kept each at a constant temperature. atmospheric air for example.3 This is the complete Carnot cycle. .2. Secondly he notes that any heat conduction results in a loss of work as compared with the work produced by an ideal heat-engine. During the former the elastic force of the air is found to be greater. that in the bodies employed to realize the motive power of heat there should not occur any change of temperature which may not be due to a change of volume. cd. Then he goes on to describe in detail what we now call the Carnot cycle: This preliminary idea being established. Now. when it is rapidly dilated. provided with a movable diaphragm or piston. let us imagine an elastic fluid.

the temperature remains unchanged. The air is rarefied without receiving caloric. The piston returns from the position ik to the position ef.16 CHAPTER 2. The compression is continued till the air acquires the temperature of the body A. it is compressed by the return of the piston as it is moved from the position gh to the position cd . HISTORICAL BACKGROUND Figure 2. (3). (5). (4) The air is placed in contact with the body B. at this instant the piston stops. (5). (5). The piston meanwhile continues to move. The piston passes during this time from the position cd to the position ik. then successively the steps (4). . however. (3) The body A is removed. Let us imagine that it falls thus till it becomes equal to that of the body B. The body A is all the time in contact with the air. (4). and so on. (6). The air becomes by such contact of the same temperature as the body A. This air remains.3: (1) Contact of the body A with the air enclosed in the space abcd or with the wall of this space—a wall that we will suppose to transmit the caloric readily. cd is the actual position of the piston. to which it yields its caloric. The body A furnishes the caloric necessary to keep the temperature constant. which is thus kept at a constant temperature during the rarefaction. (6). and the air is then no longer in contact with any body capable of furnishing it with caloric. and the compression of the air is continued.(6) The air is again placed in contact with the body A. and its temperature falls. (3). remaining at the position gh. (7) The step described under number (3) is renewed. (5) The body B is removed. at a constant temperature because of its contact with the body B. its temperature rises. (4). which being then isolated. and passes from the position ef to the position gh. (2) The piston gradually rises and takes the position ef.

the main principles on which the theory of heat rests require the most careful examination.9 We have chosen atmospheric air as the instrument which should develop the motive power of heat. we may say in passing. 9 We tacitly assume in our demonstration. Thus we are led to establish this general proposition: The motive power of heat is independent of the agents employed to realize it. its quantity is fixed solely by the temperatures of the bodies between which is ejected. . which comprehends all natural substances. as we remarked above. and when after a certain number of transformations it returns to precisely its original state. in fact. or else that the quantities of heat absorbed or set free in these different transformations are exactly compensated. Many experimental facts appear almost inexplicable in the present state of this theory. We must understand here that each of the methods of developing motive power attains the perfection of which it is susceptible. one neutralizing the other. SADI CARNOT compression. finally. This condition is found to be fulfilled if. this would not be precisely the case with the vapor of water. the reasoning will here be even more exact. then. The air. It was first admitted without reflection. has served as a heatengine.2. while. as we have already remarked. to mode of aggregation-let us suppose.3. that when a body has experienced any changes. and he continues: The result of these first operations has been the production of a certain quantity of motive power and the removal of caloric from the body A to the body B. Thus we should obtain an excess of motive power—an excess which we could employ for any purpose whatever. I say. if there is no contact between bodies of sensibly different temperatures. This fact has never been called in question. The result of the inverse operations is the consumption of the motive power produced and the return of the caloric from the body B to the body A. or. 17 Then he points out that the cycle may be run in reverse and thereby by the use of mechanical work take heat from the lower temperature of B to the higher temperature of the Reservoir A. we have. and even for all other bodies susceptible of change of temperature through successive contractions and dilatations. the transfer of the caloric. but it is evident that the reasoning would have been the same for all other gaseous substances. what is the same thing in other words. or at least all those which are adapted to realize the motive power of heat. that is. is now easily proved. and verified afterwards in many cases by experiments with the calorimeter. The air which we have used to develop the motive power is restored at the end of each cycle of operations exactly to the state in which it was at first found. there is produced in the body no other change of temperature than that due to change of volume. to temperature. after a fashion. for no useless re-establishment of equilibrium has been effected in the caloric. employed it in the most advantageous manner possible. To deny it would be to overthrow the whole theory of heat to which it serves as a basis. For the rest. The impossibility of making the caloric produce a greater quantity of motive power than that which we obtained from it by our first series of operations. so that these two series of operations annul each other. to that state considered in respect to density. It is demonstrated by reasoning very similar to that employed at page 8. that this body is found to contain the same quantity of heat that it contained at first.

James Clerk. The point is that the conservation of energy was only established much later. Carnot was clearly aware that his idea of heat was not clear as may be seen from his remarks: According to established principles at the present time.” . England. However. Edinburgh. of any design whatever. whatever may be. the view that the caloric is conserved is clearly incorrect. It was published in the Philosophical Magazine in 1860. whether the fall of caloric from 100 to 50 degrees furnishes more or less motive power than the fall of this same caloric from 50 to zero. 1879. Cambridgeshire. We do not know. the motive power of heat depends also on the quantity of caloric used. It is a question which we propose to examine hereafter. the machine which is acted upon by the water. In the fall of caloric the motive power undoubtedly increases with the difference of temperature between the warm and the cold bodies. and whatever. This conclusion is correct. for example. 11 Maxwell.18 CHAPTER 2. In 1931. He is regarded by most modern physicists as the scientist of the 19th century who had the greatest influence on 20th-century physics. Cambridge. on the 100th anniversary of Maxwell’s birth. Scotland. Carnot thought that caloric was a conserved quantity—an attitude consistent with the experimental accuracy of his day. The amount of work that may be extracted by a heat-engine.4 The Maxwell’s Distribution Maxwell11 read a paper at the meeting of the British Association of Aberdeen in September 1859. and which perhaps are less clear than is desirable. and he is ranked with Sir Isaac Newton and Albert Einstein for the fundamental nature of his contributions. the difference of temperature of the bodies between which the exchange of caloric is made. Each has a maximum that we cannot exceed. HISTORICAL BACKGROUND Here we are the final conclusion. died November 5. but we do not know whether it is proportional to this difference.with the formulation: dass in allen F¨ llen. eine der erzeugten Arbeit a a proportionale W¨ rmemenge verbraucht werde. 1831. und dass umgekehrt duch Verbrauch a einer ebenso grossen Arbeit dieselbe W¨ rmemenge erzeugt werden k¨ nne. wo durch W¨ rme Arbeit entstehe. the substance acted upon by the heat. Scottish physicist best known for his formulation of electromagnetic theory. on the one hand. we can compare with sufficient accuracy the motive power of heat to that of a waterfall. The motive power of a waterfall depends on its height and on the quantity of the liquid. we are obliged to employ expressions not in use as yet. on what in fact we will call. a o 2. born June 13. Einstein described the change in the conception of reality in physics that resulted from Maxwell’s work as ”the most profound and the most fruitful that physics has experienced since the time of Newton. and it contains the following 10 The matter here dealt with being entirely new. depends only on the temperatures of the hot and the cold reservoir. In the waterfall the motive power is exactly proportional to the difference of level between the higher and lower reservoirs. the height of its fall. on the other hand. and on the amount of heat transported. The correct interpretation was first given by Rudolf Clausius in 1850. and on what may be termed.10 that is to say.

we find f (x) = CeAx . then this will be the number in the element of volume (dxdydz) after unit of time. and we should find the whole number of particles infinite. so that the number of particles whose velocity lies between x and x + dx. 2 If we make A positive. after a great number of collisions among a great number of equal particles. so that after a certain time the vis viva will be divided among the particles according to some regular law. z be the components of the velocity of each particle in three rectangular directions. collisions would take place among the particles. If we suppose the N particles to start from the origin at the same instant.2. Let x. that is f (x) f (y) f (z) = φ(x2 + y2 + z2 ) . so that the number between x and x + dx is 12 Appendix 10 from Emilio Segr` ’s book Maxwell’s Distribution of Velocities of Molecules in His Own Words e . Prop.12 If a great many equal spherical particles were in motion in a perfectly elastic vessel. Let N be the whole number of particles. The number of particles for which y lies between y and y + dy will be N f (y)dy. since these are all at right angles to each other and independent. and therefore this number must depend on the distance from the origin alone. and their velocities would be altered at every collision. Now the existence of the velocity x does not in any way affect that of the velocities y or z. and also between y and y + dy. and the number for which z lies between z and z + dz will be N f (z)dz where f always stands for the same function. IV. and let the number of particles for which x lies between x and x + dx. be N f (x)dx. the average number of particles whose velocity lies between certain limits being ascertainable though the velocity of each particle changes at every collision. and the number referred to unit of volume will be N f (x) f (y) f (z) . the number of particles will increase with the velocity. and also between z and z + dz. 2 19 φ(r2 ) = C3 eAr . y. But the directions of the coordinates are perfectly arbitrary. Solving this functional equation. We therefore make A 1 negative and equal to − α2 . To find the average number of particles whose velocities lie between given limits.4. THE MAXWELL’S DISTRIBUTION heuristic argument on the velocity distribution. is N f (x) f (y) f (z)dxdydz . where f (x) is a function of x to be determined.

therefore C = √ . 2 This is greater than the square of the mean velocity. α π f (x) is therefore 2 2 1 √ e−x /α α π Whence we may draw the following conclusions:— 1st. The number of particles whose velocity. To find the mean value of v. add all the values together and divide by N. . 3 mean value of v2 = α2 . we find the whole number of particles. lies between x and x + dx is 2 2 1 N √ e−x /α dx α π 2nd. √ 1 NC πα = N . The number whose actual velocity lies between v and v + dv is N 2 2 1 √ v2 e−v /α dv α3 π 3rd. the result is 2α mean velocity = √ π 4th. To find the mean value of v2 . as it ought to be. HISTORICAL BACKGROUND NCe−x 2 /α2 dx Integrating from x = −∞ to x = +∞. resolved in a certain direction. add the velocities of all the particles together and divide by the number of particles.20 CHAPTER 2.

q0 . p0 ≡ {p j0 }.. . (3. p0 ). the phase trajectory is given by the functions qi (t. . . 2. q0 . The trajectory of the phase space is called the phase trajectory. . 21 (3. p2 . pj = − ˙ ∂H .1 Microscopic states Classical phase space: Classical phase space is 2 f -dimensional space (q1 . . . ∂q j qj = ˙ ∂H . Here qi are positional coordinates. Quantum states: Quantum states ψl are defined by the Schr¨ dinder equation . q2 . 2. p0 ). They are determined by the quantum numbers. We will discuss quantum systems later. f = 3N for n independent particles in 3dimensional space. For a given initial conditions at some t − t0 . q2 . The simplest Hamiltonian of such a system is H =∑ j f p2 j 2m + E(q1 . . . pi (t. q f ) .) . f ) . o H ψl = El ψl (l = 1. . q f . p f ) .2) The motion of the phase point Pt defines the state at time t. . . ∂p j ( j = 1. Exercise: Analyse phase space for a one-dimensional harmonic oscillator and find the dependence of its energy on the amplitude. . . .4) . . . . H (q.Chapter 3 Basic principles of statistical mechanics 3. For a conservative system energy is conserved. Each point corresponds to a microscopic state of the system. (3. p) = E . q0 ≡ {q j0 }.3) Thus the orbit belongs to the surface of constant energy. . (3. pi are momenta. . p1 .1) Motion of the system is determined by the canonical equation of motion.

The realization probability that one of the microscopic states belongs to the element ∆Γ of the phase space is defined as ∆Γ f (P) dΓ .6) where T is the absolute temperatures.2 Statistical treatment Fundamental assumption: In classical mechanics.5) ∑ Al f (l) . Al = f Let A be a physical quantity dependent on the microscopic state. Example – ideal gas: The phase space consists of spatial coordinates r and moments p. In quantum mechanics. The observed value (in a macroscopic sense). . at high temperature. The probability that the quantum state l is realized is defined as f (l) . ψ∗ Aψl (dq) ≡ l|A|l . (3. p) = A(P) . f is usually called the distribution functions. (l ∈ M ) . 2mkB T (3. PRINCIPLES OF STATISTICAL MECHANICS 3. the Maxwell distribution reads: f (p) = (2πmkB T )−1/2 exp − p2 . l Here (dq) ≡ ∏ j=1 dq j . The observed quantity is given by the formulas ¯ Aobs = A = ¯ A = M A(P) f (P) dΓ . ¯ Aobs = A . m the mass of a molecule. M Statistical ensembles: great number of systems each of which has the same structure as the system under consideration. Realization probability of a microscopic state: Let M be the set of microscopic states which can be realized under a certain microscopic condition. In thermal equilibrium. A(q.22 CHAPTER 3. Aobs must be a certain average of microscopic A. (∆Γ ∈ M ) . It is assumed that ¯ Aobs = ensemble average of A = A . and for a dilute gas. and kB is the Boltzmann constant.

H }.8) This combination is called the Poisson bracket.3.2).. q) is time-independent. Liouville theorem: Consider a set of instances. j ¯ If one defines the fluctuation. ˙ ∂qi ∂pi i=1 ∑ f f ∂f ∂qi ∂ pi ˙ ∂f ˙ qi ˙ + pi ˙ +f∑ + = 0. ∂qi ∂pi ∂pi i=1 ∂qi According to the Hamilton equation (3. → f12 = f1 f2 . Thus. (∆A)2 ¯ A 1/2 2 = ∑ ∆A j j N 2 = ∑ (∆A j )2 . 12 where A is any physical quantity. . The Liouville theorem states that the distribution function f (p. . . in the absence of external forces. the state must be stationary. . t1 . .. . ∂qi ∂pi Thus f df ∂f ∂f = ∑ qi ˙ + pi ˙ = 0. Indeed.2. then (∆A) As a result. j N 1 ∼√ N 1.t2 .7) According to the Hamilton equation (3. ∆A = A − A. . .2) the sum can be rewritten as i=1 ∑ f f ∂f ∂f ∂ f ∂H ∂ f ∂H qi ˙ + pi ˙ =∑ − ∂qi ∂pi ∂pi ∂qi i=1 ∂qi ∂pi ≡ { f. dt ∂qi ∂pi i=1 (3. ¯ ¯ A = ∑Aj . f12 d p(12) dq(12) = f1 d p(1) dq(1) f2 d p(2) dq(2) Statistical independence means ¯ ¯ A¯ = A1 A2 . P2 . ∂qi ∂ pi ˙ ˙ + = 0. The states at that instances are represented by the points P1 . for any additive quantity. . STATISTICAL TREATMENT 23 General properties of statistical distributions Statistical independence: For two macroscopic and non-interacting subsystems. and for the phase-space points one has div ( f v) = 0 Expanding derivatives we get → i=1 ∑ f ∂ ∂ ( f qi ) + ˙ ( f pi ) = 0 . (3.

E + δE) is realized with equal probability: −1 f (P) = constant = E<H <E+δE dΓ −1 . q) + γ · P(p. δE) . M = 0. The only additive function is the linear combination. δE) is the shell-like subspace of the phase space between the constant-energy surfaces for H = E and H = E + δE. and P = 0. PRINCIPLES OF STATISTICAL MECHANICS Role of integrals of motion Since log f (p. The coefficient α is defined from the normalization condition. l ∈ M (E. The energy of the system.3 The principle of equal weight and the microcanonical ensemble An isolated system after some time reaches the equilibrium. In the following we assume that the system is enclosed in a rigid box. Thus. In a classical system the set M (E. .10) This distribution is called the microcanonical one. δE. q) and angular momentum M(p. and (ii) additive for macroscopic subsystems of the system. the values of the integrals of motion completely define statistical properties of the closed system. f (l) = constant = E<El <E+δE ∑ 1 . it depends only on additive integrals of motion.9) Here in the equilibrium the coefficients β. each of the microscopic states belonging to M (E.1 The microcanonical ensemble represents an isolated system which has reached thermal equilibrium. 3.24 CHAPTER 3.q) must be (i) stationary. P ∈ M (E. the corresponding ensemble being called the microcanonical. is fixed with some allowance. The principle of equal weight: In a thermal equilibrium state of an isolated system. E. log f12 = log f1 + log f2 . E(p. q) . For quantum mechanics. δE) . it is a set of quantum states having energies in the interval E < El < E + δE. 1 Some other additive integrals of motion besides the energy can be important. q) + δ · M(p. total momentum P(p. (3.q). (3. γ and δ must be the same for all subsystems in the closed system. log f (p. q) = α + βE(p.q).

q) − E] = dσ dn δ(n) .12) (3.15) Ergodic theorem: Originally. q) = E .11) which is called the density of states. dΓ δ[H (p. (3.14) D(E) |∇H | H =E |∇H | An average is then given by the equation. To determine the constant we have to calculate the integral D(E) = dΓ δ[H (p. . MICROCANONICAL ENSEMBLE 25 Classical limit (δE → 0): One can take a set σ(E) on the surface of constant energy.3. EQUAL WEIGHT. . |∇H | ∂H dn = |∇H | dn . . and the distribution function can be written as f (P) = const · δ[H (p. Then All the phase trajectories are concentrated at the surface H (p. . Aobs is assumed to be time average of A(pt ). ¯ A= 1 A(P) dσ . dσ dσ 1 f (P) dσ = . Then the energy is dependent only on the coordinate n. D(E) = . The ergodic theorem can be formulated as Atime average = Aphase average . and dE = Here |∇H | = Consequently. It is very convenient to use the finite-width layer in the phase space and to transform variables from {p. see below. (3. q} to the element of surface σ defined by the equality H (p.13) Then one can perform integral over n and introduce the distribution over the constant energy surface. q) − E] . q) − E.3. q) − E] (3. and to the coordinate n along the normal to this surface. D(E) H =E |∇H | (3. Actually one can deduce the principle of equivalent weight from this theorem. ∂n ∂H ∂p j 2 ∑ j ∂H + ∂q j 2 1/2 .

δE ≥ (δE)qu contains many quantum states. of particles of various kinds. (δE)qu ∼ h/t . .V. . . δE. . If the system occupies only a single quantum state. . . This relation is called the Boltzmann relation while the constant kB = 1. . Thermodynamic weight: In quantum mechanics. x) ≡ E<El (N. To convince yourself that the statistical definition of entropy (3.V. δE. δE. NB .. . . (3. . .26 CHAPTER 3. In a classical system one can tend the allowance δE to 0. . x.18) is compatible with thermodynamics one has to check thermodynamic relations.. . It is effectively true also for a quantum system provided the measuring time is long enough. and other parameters x. . .) = kB logW (E. . PRINCIPLES OF STATISTICAL MECHANICS The finite allowance of the energy δE: In quantum mechanics. (3. 3. NB . δE. .16) In classical mechanics it is defined as a limit of quantum statistical mechanics W (E. NB . . .x)<E+δE h3(NA +NB +.V. Statistical definition of entropy: S(E. the numbers NA .V. . The statistical treatment is still possible only if the set M E. the energy of a system has uncertainty. N. NA .2 2 Note that allowance δE does not affect the value of the entropy . NA . . Thus the Hamiltonian depends on those parameters. x) of the possible quantum states for the set of prescribed values E(δE).V. W (E. . . statistics is not applicable.V. x) . N.18) dΓ E<H (N. .38 × 10−23 J/K (3. Then the following definition do not depend on δE.17) is called the Boltzmann constant. . Namely W (E.V. x. which specify external forces. . NB . x. NA . .V. .]. is called the thermodynamic weight of that particular macroscopic state. ¯ where t is the length of the observation.V. . H = H [E(δE).19) (3.) NA !NB ! . N.4 The thermodynamic weight of a macroscopic state and entropy Variables defining a macroscopic state: One can choose the energy E (with allowance δE). x) ≡ Here h is the Planck constant. the total number of quantum states. N. the volume V of the box which contains the system.x)<E+δE ∑ 1. .

x) . (3. 0 ≤ E1 ≤ E2 ≤ .V ) δE = W (E. It is important to keep in mind Correspondence of classical and quantum mechanics: (i) Uncertainty condition.V. ∂E (3.5. We shall come back to that point later discussing quantum statistics in more detail. Then the required number of states in the region E.3. . NUMBER OF STATES AND DENSITY OF STATES 27 3. E + dE can be written as D(E) dE.V.V. x) ≡ If allowance δE is small. Though the factor 1/N! arises only from quantum statistics. For a closed system.21) ∂ Γ(E. (3. N. The number of states between 0 and E is called the number of states of the system: Γ(E. N. x) ≡ State density: D(E.20) . (ii) Indistinguishability of identical particles: it is an important features of Bose and Fermi statistics. N. while the energy probability distribution is P (E) = f (E) D(E) dE . Because of that the classical phase volume ∆Γ corresponds to ∆Γ/h f quantum states ( f is the number of degrees of freedom. dΓ 0<H <E h3(NA +NB +. its necessity has been understood much earlier to make the entropy extensive quantity.. After definition of number of states Another definition of entropy: Another definition of entropy can be formulated through the probability distribution. . fl = f (El ). the distribution function can be written as a function of the energy. N.V. . x) ≡ In classical mechanics. In the classical limit h → 0 it leads to a factor 1/ ∏i Ni !. . . ∆p ∆q ∼ h ..22) D(E. 0<El ≤E ∑ 1. N. δE.V.5 Number of states and density of states Let us chose the energies to be non-negative. Γ(E. N. x) .) NA !NB ! . .

28

CHAPTER 3. PRINCIPLES OF STATISTICAL MECHANICS

Since P (E) dE = 1 and P (E) is centered near the average energy, ¯ ¯ ¯ ¯ P (E) ∆E ≈ f (E) D(E) ∆E ≈ f (E)W = 1 . Thus ¯ S = −k log f (E) = −k log f (El ) . The distribution function f is constrained by the normalization condition (3.23)

∑ f (Ei) = 1 .
i

(3.24)

The latter inequality follows from statistical independence. Indeed, log f (El ) must be an additive function of the energy (in general, also of other additive integrals of motion, P and M), log f (El ) = α + βEl . As a result ¯ log f (E) = log f (El ) = − ∑ fl log f (El ) .
l

This expression is written for quantum statistics. In the classical statistics one has to remember the normalization factor [h f ∏ j N j !]−1 for the thermodynamic weight. Thus one has to replace fl → in the argument of the logarithm to obtain S=− dΓ f (p, q) log h f h f ∏ N j!
j

f (p, q)

∏ N j!
j

f (p, q) .

(3.25)

The entropy, defined in this way, is additive. Physical significance of entropy: Let us start with the microcanonical distribution,
a a

d P = const × δ(E − E0 ) · ∏ dΓa = const × δ(E − E0 ) · ∏(dΓa /dEa ) dEa (after replacement dΓa /dEa → ∆Γa /∆Ea ) = const × δ(E − E0 ) eS ∏ dEa /∆Ea .
a

Here S = ∑a Sa , E = ∑a Ea . ¯ Partial entropy S(E1 , E2 , . . .) must have maximum at Ea = Ea . Since these values correspond to the equilibrium state the entropy of a closed system in a state of complete statistical equilibrium has its greatest possible value (for a given energy of the system).

3.6. INFORMATION

29

The law of increase of entropy: If a closed system is at some instant in a non-equilibrium macroscopic state, the most probable consequence at later instants is a steady increase of the entropy of the system. This is the law of increase of entropy or the second law of thermodynamics. This law has been formulated by R. Clausius (1865), its statistical explanation was given by L. Boltzmann (1870). It has a character of assumption which cannot be rigorously proved because the equations of mechanics are time-reversible. If the entropy remains constant. the process is called reversible, if it increases the process is called irreversible.

3.6 Information
The definition of entropy in Eq. (3.23), is directly related to the definition of the information measure. As an example consider the probability distribution for the number of ‘eyes’ that come up when throwing a fair die. The probabilities of the six possible states |1 , |2 , . . . , |6 are fi , with i = 1, . . . , 6. Small variations δ fi , lead to a variation in S as defined in (3.23). The maximal value of S may be found, taking into account the constraint (3.24), using a Lagrange parameter kB λ: 0 = δ (S − kB λ · 1) = δ − ∑ kB fi ln fi − kB λ ∑ fi
i=1 i=1 6 6

= −kB ∑ ln fi +
i=1

6

fi + λ δ fi fi

1 . (3.26) 6 Here we have chosen λ so that fi satisfies the constraint (3.24). We see that requiring the information entropy to have its maximum value leads to the result that fi = 1/6, independent of i—as expected for fair dice. When we throw two dice, we have many more possibilities. As before, the ‘first’ die has the possible states {|i }, with i = 1, . . . , 6. The other die has the states {| j } with j = 1, . . . , 6, of course. Counting the eyes for the pair of dice we have 36 possible outcomes {|α }, with α = 1, . . . , 36 corresponding to the states |11 , |12 , . . . |66 . With the argument developed above we will find that fα = 1/36. It is important to note, however, that the statistical independence of the two dice allows another approach. The states |α = |i | j are in a product space of the two independent states |i and | j , much in the same way that the position of a point in the plane r = (x, y), is given in terms of the independent coordinates x and y; or with our notation |r = |x |y . The process of throwing the dice yields eyes that are statistically independent, and the probability, fi j , that the first die has i eyes and the second has j eyes is the product of the probability fi for the first die to show i eyes and f j , which is the probability that the second die shows j eyes: ⇒ fi = exp(−1 − λ) ⇒ fi = fα = fi j = fi f j statistically independent (3.27)

30

CHAPTER 3. PRINCIPLES OF STATISTICAL MECHANICS

It follows that the entropy of the combined system may be written S = −kB ∑ fi j ln fi j = −kB ∑ fi f j ln fi f j
ij i ij

= −kB ∑ fi ln fi ∑ f j + −kB ∑ f j ln f j ∑ fi
j j i

= S1 + S2 , where we have used that probabilities are normalized: ∑i fi = ∑ j f j = 1. We conclude that the (information) entropy of a system consisting of statistically independent parts is the sum of the entropies of the parts. Entropy is an extensive property, i.e. entropy is proportional to system size.

3.7 Normal systems in statistical thermodynamics
Asymptotic forms for the number of states and state density: (i) When number of particles N, or the volume V tends to infinity, the number of states in normal systems approach the following asymptotic behavior: Γ ∼ exp[Nψ(E/N)] Γ ∼ exp[Nψ(E/N,V /N)] and (ii) Therefore D(E) = ∂Γ = ψ exp(Nψ) > 0 , ∂E (3.30) ψ > 0, or or exp[V ψ(E/V )] , exp[V ψ(E/V, N/V )], ψ > 0.

(3.28)

ψ > 0,

(3.29)

∂D(E) ∼ ψ 2 exp(Nψ) > 0 . ∂E Entropy of a normal thermodynamic system:

For a normal system, the statistical entropy is (3.31)

S = k log[D(E)∆E] ≈ k log Γ(E) = kNψ . It is proportional to N, or to V .

dS/dE1 = 0. dE1 dE2 Here we have used the energy conservation. Volume changes are discussed in the section 4. The previous argument may be generalized to a system consisting of many parts. As a result. Consequently. T = (dE/dS)V (4. In Eq. or. we have E = E1 + E2 . T1 = T2 . dS1 dS2 = . (4. defined as the change in the (equilibrium) entropy with energy is independent of position for a system in thermal equilibrium. In the equilibrium the entropy has a maximum with respect to the parameters of subsystems. At equilibrium one concludes that ∂S/∂E has the same value everywhere.Chapter 4 Thermodynamic quantities 4. i.3. 31 .1 Temperature Consider 2 bodies in a thermal equilibrium which are in thermal contact with each other but are isolated form the surrounding. more rigorously.1) we have explicitly written that the change is to take place at a fixed volume. for example for a liquid in equilibrium with its vapor. dE1 dE2 The absolute temperature is defined as T = dE/dS. S = S(E1 ) + S(E2 ) . In the equilibrium. E2 = E − E1 . form a closed system. it is also the same for two systems in equilibrium with each other. Since both the energy and entropy are additive.e.1) which means that the volume V is kept constant. Form this equality we get dS dS1 dS2 dE2 = + dE1 dE1 dE2 dE1 dS1 dS2 = − = 0. Therefore one finds that absolute temperature T .

dS = dt dS1 dS2 − dE1 dE2 dE1 = dt 1 1 − T1 T2 dE1 > 0.32 CHAPTER 4. the 1st power is absent because entropy must increase with time. then dE1 /dt > 0. q. let us define the thermodynamic energy as E ≡ H (p. Up to the lowest order. Thus. If the force is purely mechanical (we ignore the feedback to the force source from the system under consideration). dE2 /dt < 0 . λ) dλ = = . Let us then assume that λ varies in time sufficiently slow1 and expand dS/dt in powers of dλ/dt. dS dλ =A dt dt 2 → dS dλ =A . q. For example. λ) . 4. Since d H (p. the classical Hamiltonian being H (p. T1 = T2 . dS dS1 dS2 = + dt dt dt dS1 dE1 dS2 dE2 = + > 0. Thermally isolated means that H is dependent on the coordinates and moments of the molecules belonging only to the body in question.energy passes from bodies at higher temperature to bodies at lower temperature. the number of Joules per degree is called the Boltzmann constant. dt dt dλ dt 1 comparing to the relaxation time of the system . The temperature is measured in degrees Kelvin. q. Below we omit the Boltzmann constant in all the formulas. q.19). S → S/kB . Then the temperature is measured in the units of energy. λ)/∂t (that follows from the Hamilton equation). then one can apply the law of increase of entropy to the body in question. To get custom formulas one has to replace T → kB T. λ) ∂H (p. q.2 Adiabatic process Consider a thermally isolated system under time-dependent force λ. dt If T2 > T1 . the adiabatic process appears reversible. dλ dt Indeed. see Eq. Using adiabatic process one can calculate mean values. λ)/dt = ∂H (p. dE d H (p. (3. We obtain that at dλ/dt → 0 the entropy tends to be λ-independent. During the equilibration the entropy S = S1 + S2 must increase. q.t). THERMODYNAMIC QUANTITIES Now let us discuss the case when the bodies and not in an equilibrium with each other. dE1 dt dE2 dt Since the energy is conserved.

e. A force F acts from the contents of the system on the container wall surface element. PRESSURE On the other hand.3. S. q. r) =− ∂r ∂r =− S ∂U ∂V S ∂V ∂U =− ∂r ∂V ds . U = H (p. δr. S (4. Thus dE = dt As a result. i. λ. S The magnitude of the force per area ds is called the pressure. If the container wall moves a small distance. or equivalently the energy to be a function of the entropy. q. Consequently. Thus. S(E).3) This relation is often called the first law of thermodynamics. slowly enough for the system to stay in internal thermal equilibrium. i. the force acting on a surface element ds is F=− ∂U ∂H (p. consider a process in which the wall of the container moves adiabatic. λ) = ∂λ ∂E ∂λ . and therefore do not depend on the shape for systems in thermal equilibrium. they are proportional to system size. The pressures of the bodies in the equilibrium are the same (to preserve mechanical stability).4) property can be violated in metastable states. S (4. 2 It follows from Eq.e. dt 4. and of external parameter. in the case when we have no external system or medium. V P=− ∂U ∂V . Then the energy E is equal to the internal energy which we will denote as U. Now. (4. This work must be equal to the change of the internal energy of the system. q).3 Pressure We have already mentioned that both the entropy and the energy are additive quantities. then work is done on the surroundings given by δW = F · δr. in our case. δs. S 33 ∂E ∂λ S dλ . (4. E(S). P=− ∂U ∂V . Let us consider below the case where the system does not take part in a macroscopic motion. . one can write dU = T dS − P dV . E is a function of the entropy. ∂H (p. i.3) that T= 2 This ∂U ∂S . for the closed system.2) Combining definitions of the temperature and pressure.e.4. We may consider the entropy to be a function of energy.

the systems like liquid or gas in an equilibrium can be described by state variables like temperature T . U (4. T T ∂S ∂V (4. P = P(V. N NT −a P= V − bN V 2 = Nρ − aρ2 . T and V . 4. .1.4 Equations of state As we have seen. The equation of state can be obtained either from experiment. T ) . THERMODYNAMIC QUANTITIES dS = and it follows that 1 P dU + dV .314 J/(K · mole) . only two variables can be assigned independently. or from microscopic theory. We will come back to phase transitions later where various phase diagrams will be discussed. 4. Usually it is taken in the form. while the term ∝ ρ2 allows for particle attraction at relatively large distances. An ideal gas of N particles obeys the equation of state P = NT /V = ρT (temperature is measured in energy units) (4. then T → kB T .023 × 1023 mole−1 . In general. (4. The phase diagram of real gases can be obtained from the equation of state. this equation allows one to calculate the equation of state connecting P.34 The first law (4. If T is measured in Kelvin.3) can be rewritten as CHAPTER 4. 1 − bρ (4. For one mole of the gas N equals to the Avogadro’s number NA = 6. They are not independent since the variables in equilibrium related by an equation of state.6) It the microscopic expression for the entropy is known. and the equation of state for 1 mole reads P = RT /V . pressure P.9) The item bρ in the denominator take care of particle repulsion at small distances.8) Here ρ ≡ N/V is the gas density.7) Thus. R ≡ NA kB = 8. it has the form shown in Fig.5) P = T . or total volume V . Including interaction between particles in a simple approximate form leads to more realistic van der Waals equation.

2 First consider the situation in which the system is thermally insulated from the medium. (b) – The walls of the system conduct heat and in the process of changing the volume of the system the heat δQ enters the system. (a) – The surroundings work on a thermally insulated system by changing its volume by δV . performs work on the system. Together the system and the surroundings form a closed system We assume that W > 0 when external forces produce work on the body in question.1: Phase diagram of a real liquid-gas system. If the volume of the system changes by δV then the medium.4. therefore the energy change is δU = δQ + δW . which is has a pressure P (equal to the pressure of the medium). WORK AND QUANTITY OF HEAT P Critical point 35 Solid Triple Liquid Gas point T Figure 4. 4. Figure 4. One obtains dW /dt = −P dV /dt when the volume changes in time. . The increase in energy of the system is δU = δW = −PδV . Fig. The work done by the system is δW = −PδV . 4. while W < 0 if the body itself produces work.5 Work and quantity of heat Consider a closed system that consists of two parts – the ‘system’ and the surrounding medium or ‘heat bath’.2: A closed system consisting of a heat bath surrounding the system of interest. For a reversible process the pressure must be the same throughout the body.5.

i. In this case. then dE dW dQ = + . If the pressure is kept constant. dQ = d(U + PV ) ≡ dH . dH = dU + P dV +V dP . one gets the inequality dQ dS <T . This relation can be considered as a form of second law of thermodynamics.6 Thermodynamics potentials 4. If the systems goes from one equilibrium state to another equilibrium state through some non-equilibrium states.12) 4. THERMODYNAMIC QUANTITIES If the body is not isolated thermally. H(S. Since this process is irreversible. (4. Since dU = T dS − P dV . while the state is non-equilibrium (say. (4. chemical reaction in the mixture). dQ = dU. dt dt dt Assuming that the body is in an equilibrium corresponding to its energy and the volume at the instant. than δQ ≤ T δS . P) = U + PV . dt dt dt Thus dQ dS =T .11) instead of Eq.e. that they depend on the way in which the states of the system change in time. one can put U = U(S. One can imagine the situation when the pressure and temperature are constant throughout the body. dQ dU dV = +P .10) dt dt Note that dQ and dW are not complete differentials. (4. dU dS dV =T −P . (4.6.10).36 CHAPTER 4.V ). dt dt (4. Here E is understood as a whole energy of the body including macroscopic motion.13) . dt dt dt where Q is the heat gained or lost by the body per unit time. For the body at rest E equals to the internal energy U.1 The heat function (enthalpy) If the process occurs at constant volume.

T ) ≡ U − T S . T = (∂H/∂S)P . Introducing G = U − T S + PV = F + T S = H − T S and using the equality P dV = d(PV ) −V dP we obtain dG = −S dT +V dP → S = −(∂G/∂T )P . F is called the (Helmholtz) free energy. 37 (4. one obtains. if we know any of the quantities. Thus.V ).6.16) dF = dU − T dS − S dT (4. we can determine all the remaining thermodynamic quantities. F(T. dW = dU − dQ = dU − T dS = d(U − T S) = dF .15) As a result. λi the expression ∑ Λi dλi i . S = −(∂F/∂T )V . 4. We still miss the thermodynamic potential with respect to P. Since U = F + T S U = F − T (∂F/∂T )V = −T 2 P = −(∂F/∂V )T .2 Helmholtz free energy and thermodynamic potential (Gibbs free energy) Consider the work on a body in a reversible isothermal change of the state. then H = const .4. dF = −S dT − P dV .18) Thus. V = (∂H/∂P)S .V ).19) If there are other parameters.17) (4. THERMODYNAMICS POTENTIALS one obtains dH = T dS +V dP . T .14) (4.6. if the body is thermally isolated and the pressure is kept constant. (4. (4. Thus. Thus U. V = (∂G/∂P)T . or H(S. U(S. Since dU = T dS − P dV . F and H are called the thermodynamic potentials with respect to their variables. V F(V. P). ∂ F ∂T T .

G = N f (P. Since dU dV dS +P <T dt dt dt at constant V and T we get dF/dt < 0. Then dΩ = −S dT − N dµ . Then it is convenient to introduce the new thermodynamic potential. T ) .V /N) since this is the most general homogeneous function of the first order in N. q. (δU)S. 4. but the number of particles as a variable. ∂λi ∂λi T. T ). If one considers N as a formal independent variable. In a similar way. THERMODYNAMIC QUANTITIES should be added to all thermodynamic potentials. . the changes in the thermodynamic potentials are equal if each is considered for the appropriate pair of constant quantities.38 CHAPTER 4.V ∂H ∂N = S. the independent variables are.V = (δH)S. where µ= ∂U ∂N = S. Thus dµ = −(S/N) dT + (V /N) dP . H = N f (S/N.P ∂F ∂N = T. S and V . then all the thermodynamic potentials acquire additional terms µ dN. We easily find that G = Nµ(P. N. (4.20) This statement is called the theorem of small increments.T µ is called the chemical potential. Ω = F − µN = F − G = −PV .3 Dependence on the number of particles Since S and V are additive the internal energy can be written as U = N f (S/N. P) . Thus at the equilibrium the thermodynamic potentials reach minima with respect to the variations of state with proper constant parameters.λ j=i ∂λi S. If one considers the volume is constant. In this way we can obtain the expressions for the averages.. F = N f (V /N. T. while at constant P and T dG/dt < 0.V = (δF)T.. ∂H (p.P = (δG)T.V ∂G ∂N .λ j=i If the parameters λi change slightly.P . say.V.P. P. λ) ∂F ∂H Λi = = = = . . T ) .6.

T ). or S) .V. We get for the cases of fixed volume and pressure. Similarly. V. N T. then we can calculate everything. X and the thermal expansion coefficient α= 1 V ∂V ∂T . If we know the equation of state. V. N T. P. respectively: CV = T (∂S/∂T )V . KX = − 1 V (x = V.V One can show that if 2 bodies are brought together and they can exchange by particles.1: Thermodynamic potentials (summary) 4. N=− ∂µ T. Since T dS = ∂U ∂T dT + P + V ∂U ∂V dV T . ∂V ∂P CP = T (∂S/∂T )P . the results can be expressed through the specific heat CP (P. If V and T are used as independent variables. the results can be expressed through the specific heat CV (V. T ) and through the equation of state V = V (P. P Usually. V.V and T. As an example of such derivation one can find the relation between CP and CV considering the internal energy U as a function T and V . µ Differential T dS − P dV + µ dN T dS +V dP + µ dN −S dT − P dV + µ dN −S dT +V dP + µ dN −S dT − P dV − N dµ Table 4. T ). if P and T are used as independent variables. Thermodynamic potential Internal energy Heat function (enthalpy) Helmholtz free energy Gibbs free energy Landau free energy Notation U H F G Ω Independent variables S. P are mostly used. N T. µ) as ∂Ω ∂P =V . T ) and through pressure P. N S. The number of particles can be expressed through Ω(T.6. P.4 Derivatives of thermodynamic functions The quantity of heat which must be gained in order to raise the temperature of the body by one unit is called the specific heat. THERMODYNAMICS POTENTIALS 39 Sometimes Ω is called the Landau free energy. pairs of thermodynamic variables T. P(V.4. Another important response functions are the compressibilities.1.V ∂µ T. then µ1 = µ2 We can summarize the thermodynamic potentials in the table 4.6. P.

Consider 3 quantities. P that van be obtained using Gibbs free energy instead of the Helmholtz one (prove!). related by the equation of state K (X. F (X. d F can be written as d F = Rx dX + Ry dY . Consider a thermodynamic potential. Z) = . T Now.Y ). which depends on the independent variables X. V Consequently. Y Example: =− S . X.40 CHAPTER 4. Since ∂F S=− ∂T V we have ∂CV ∂V =T V ∂2 S ∂3 F ∂2 = −T = −T 2 ∂V ∂T ∂V ∂T 2 ∂T 2 −P = ∂F ∂V ∂F ∂V . V Other important relations (see derivation in Appendix A) arise from the properties of partial differentiations. Then. P (4. Since ∂2 F ∂2 F = ∂X ∂Y ∂Y ∂X ∂Rx ∂Y ∂T ∂V = X one obtains ∂Ry ∂X ∂P ∂S .21) T Let us also demonstrate how one can obtain relations between second derivatives of thermodynamic quantities. THERMODYNAMIC QUANTITIES we obtain assuming the pressure to be constant CP = CV + P + ∂U ∂V ∂V ∂T . Calculation of derivatives from the equation of state The calculations are based on the so-called Maxwell relations. this derivative can be calculated from the equation of state. since T we get finally ∂CV ∂V ∂CP ∂P =T T ∂2 P ∂T 2 ∂2V ∂T 2 . =T T .Y .Y.Y. Z. Similarly.

P ∂S(V. We can consider X. Similarly. V Here we have used the realtion dU = T dS − P dV . (4. The relations between the derivatives of theses functions read ∂X ∂Y ∂Z = −1 . T ) ∂T −P P ∂V (P.5 Examples Entropy: ∂S ∂V =− T ∂ ∂V ∂F ∂T =− V ∂ ∂T ∂F ∂V = T ∂P(V.Y ). (4. T ) ∂T −P.24) = Z ∂F ∂X + Y ∂F ∂Y X ∂Y ∂X . while X = X(Y. V Here we have used the realtion dF = −S dT − P dV . T ) ∂P . ∂S ∂P Internal energy: ∂U ∂V =T T =− T ∂ ∂P ∂G ∂T =− P ∂ ∂P ∂G ∂P =− T ∂V (P. Z). Z (4.Y as independent variables. T ) ∂V −P = T T ∂P(V.23) and the Maxwell relations are usually used for transformations and computation of derivatives from the equation of state.22). 4. X ∂F ∂X Y ∂X ∂Z . THERMODYNAMICS POTENTIALS 41 const. the differential of which can be expressed as dF = From this expression we get ∂F ∂Z and ∂F ∂X = Y ∂F ∂X dX + Y ∂F ∂Y dX .23) Any quantity F (X.6. Similarly. (4. Z (4. Y (4.25) together with Eqs.24). Z to be independent variables. T ) ∂T .Y ). ∂U ∂P = −T T ∂V (P.6. while Z = Z(X. (4. Another way is assume Y. T .4.25) Equations (4.22) ∂Y Z ∂Z X ∂X Y ∂X ∂Y = Z ∂Y ∂X −1 . T ) ∂T .

It is then adiabatic cooled at the temperature T1 and then adiabatic returned to the original state. thermally isolated and undergoes an adiabatic compression until its temperature is increased to T2 . the work decreases with increasing S. is brought into contact with the reservoir at T2 . The working medium at temperature T2 is brought in contact with the “heater” and receives some energy isothermal. Now let us consider two bodies with different temperatures.42 CHAPTER 4.27) (4. The cycle is then. |W | = E0 − E(S) . KP − KV = (TV /CP )α2 . For that purpose one needs a further body (working medium). 2.1 Maximum work and Carnot cycle Consider a thermally isolated system consisting of several bodies not in thermal equilibrium. The heat engine. . If one brings the bodies into thermal contact. The heat-engine is removed from the T2 -reservoir.3.7. no work will be done since the process is irreversible. Carnot discussed a process that has four stages: 1. The heat-engine is removed from the T1 -reservoir. as well as the final equilibrium state. P CP −CV = (TV /KT )α2 . 4. thermally isolated and undergoes an adiabatic expansion until its temperature is decreased to T1 . The entropy of two bodies will increase be (δE)(1/T1 − 1/T2 ) where δE is the energy transfer. Since the system is thermally isolated. Let the total original energy be E0 while the energy in the final equilibrium state is E(S). and gives off isothermally the energy δE1 at temperature T1 . To get a maximum work one has to accomplish a reversible cyclic process in a way to keep the bodies among which the direct energy transfer takes place at the same temperature. 3. The system may do during establishment of equilibrium. That means the reversible way. Here T is the temperature in the final state. T2 > T1 . Since the entropy cannot decrease the greatest possible work is done when the entropy retains constant throughout the process. The cyclic process is called the Carnot cycle. depend on the concrete way which the system follows. The heat-engine is put in thermal contact with the T1 reservoir.7 Principles of thermodynamic 4.26) (4. and the four stages may be repeated.28) 4. (4. THERMODYNAMIC QUANTITIES ∂U ∂V = CP − P ∂T P ∂T One can also obtain the following relations (prove!) . ∂|W |/∂S = (∂E/∂S)V = −T . and receives the amount −δE2 isothermally. see Fig. at temperature T2 . 4. CP /CV = KT /KS . This work. Since the derivative is negative.

and receives work during contraction. absolute zero. The two parts are connected via a heat engine that may perform the work |W | on the surroundings outside the system.4. Lord Kelvin realized that the Carnot cycle may be used to define an absolute scale of temperature – the scale used here.16 K for the triple point of pure water. Both E and S depends on how the equilibrium state was reached. Since the efficiency ηmax ≤ 1 it follows that there is a lower limit of (equilibrium) of zero. E(S). there is an arbitrary scale factor that is fixed by setting T = 273. δS2 + δS1 = 0 (reversible process) .3: A thermally isolated system contains two parts at different temperatures T2 > T1 . With this scale one degree on the Celsius scale equals one degree on the Kelvin scale. we obtain |δW |max = −δU1 − δU2 = −T1 δS1 − T2 δS2 = −(T2 − T1 ) δS2 = Consequently. (a) – Initially the energy of the system is E0 . The energy of the equilibrium system is a unique function of entropy. T2 δU1 = T1 δS1 . The Kelvin scale therefore has a natural fixed point. PRINCIPLES OF THERMODYNAMIC 43 Figure 4. Since −δU2 = −T2 δS2 . T2 − T1 |δU2 |.7. The heat-engine performs work on the outside world during expansion. and an amount of work was done on the surroundings. To calculate the work one can ignore the working medium for which the initial and final states are the same. as the system approaches equilibrium. (b) – Finally the two parts have the same temperature T . the maximum efficiency η ≡ |W |/(δU2 ) is ηmax = (T2 − T1 )/T2 < 1 . . Since ηmax depends only on the ratio T1 /T2 .

3 We assume that there is no macroscopic motions as a whole. (ii) the work P0 ∆V0 done by the medium. (iii) If no mechanical work is done and the body is left itself. We are interested in the maximal work |W |max done by the thermally isolated body. Indeed.29) To reach maximum entropy one should require T = T0 . This work is equal to the minimal work Wmin done by an external source on the body. the irreversible processes lead to the inequality ∆(U − T0 S + P0V ) ≤ 0 . Since the medium is large T0 and P0 are assumed to be fixed. We come to the following conclusions. respectively. Wmin = −|W |max = ∆(U − T0 S + P0V ) . For a reversible process. Since ∆V0 + ∆V = 0. (iii) the heat −T0 ∆S0 gained from the medium. One can relate the minimal work Wmin with the deviation of the total entropy St from its equilibrium value St (Ut ). P0 = P where T. ∆U consists of the three parts (i) mechanical work W done by the external source. . ∆S0 + ∆S ≥ 0 W ≥ ∆U − T0 ∆S + P0 ∆V . Thus ∆U = W + P0 ∆V0 − T0 ∆S0 . (iii) If the pressure and the temperature of the body remains constant and equal to P0 and T0 . (i) If the volume and the temperature of the body remains constant and equal to T0 . then Wmin = ∆G.44 Work in an external medium CHAPTER 4. P are the temperature and the pressure of the body. ∆St = − ∆U − T0 ∆S + P0 ∆V (T − T0 )∆S − (P − P0 )∆V dSt (Ut ) Wmin = − =− . then Wmin = ∆F. dUt T0 T0 (4. THERMODYNAMIC QUANTITIES Consider a body embedded into a large external medium having T0 = T . P = P0 . The total change of the internal3 energy.

32) one obtains < S T . the quantity ∆U − T0 ∆S + P0 ∆V has a minimum. is only a necessary condition for an extremum and do not ensure that the entropy reaches the maximum.32) T CV we obtain from Eq. Thus δU − T0 δS + P0 δV > 0 for any small deviation from the equilibrium. δU = ∂U 1 ∂2U ∂2U ∂2U ∂U δS + δV + (δS)2 + δSδV + 2 (δV )2 .4.30) CV > 0 . Thus. Thermodynamic stability The vanishing of the 1st derivatives of thermodynamic quantities.2 Thermodynamic inequalities. the minimum work which must be done to bring the selected part from equilibrium to some other state is positive. ∂S ∂V 2 ∂S2 ∂S ∂V ∂V The 1st order terms provide T δS − P δV . one has to assume ∂2U > 0. (4. ∂S2 ∂V 2 ∂S ∂V Since ∂2U = ∂S2 ∂T ∂S = V (4.31) (4. PRINCIPLES OF THERMODYNAMIC 45 4. though provide conditions for the equilibrium. P1 = P2 . and we have ∂2U ∂2U ∂2U (δS)2 + δSδV + 2 (δV )2 > 0 . In a similar way. ∂S2 ∂2U > 0. (4. Now we can expand the internal energy U(S.7. As we have stated. ∂S2 ∂S ∂V ∂V Since δS and δV are arbitrary. ∂V 2 2 ∂2U ∂2U ∂2U − > 0. V KS From Eq. like T1 = T2 . ∂2U ∂P −=− 2 ∂V ∂V ∂T ∂V 2 = S 1 > 0. The sufficient conditions require analysis the second derivatives.7.30) (4. V KSCV .V ) in powers of the variables.

46 CHAPTER 4. (Maxwell relation for G) The following useful formulas follow from the Nernst’s theorem: S = 0 T (CV /T ) dT . (∂V /∂T )P = − (∂S/∂P)T = 0 (∂P/∂T )V = 0 (CP −CV )/CP = 0 . P) . dµg (T. THERMODYNAMIC QUANTITIES After some algebra (see the book by Landau and Lifshitz. T 0 T 0 H = H0 + G = H0 + CP dT . dP dT = coexist sg − sl λ .3 Nernst’s theorem As we know. P1 = P2 . for normal systems S → 0 at T → 0. Along the the co-existence curve between gas and liquid. µg (T. 4. §21) one can get from Eq. P) . µ1 = µ2 .7. 4. Introducing entropy per particle s ≡ S/N and volume v ≡ V /N we get −sg dT + vg dP = −sl dT + vl dP . CP dT − T T 0 (CV /T ) dT . This statement is called the Nernst’s theorem. CP = CV = 0 at T = 0 . If it is true.32) the inequality ∂P < 0. Consequently. P) = dµl (T. One can obtain similar equalities for T = 0. (4.7. P) = µl (T. ∂V T Here we have calculated the conditions for thermodynamic stability. = vg − vl T (vg − vl ) .4 On phase transitions The equilibrium between the phases requires T1 = T2 .

We shall come back to phase transitions later in much greater detail.4.7. So the entropy changes continuously while its first derivatives change abruptly. At T = Tc the two phases become identical. This is the Clausius-Clapeyron equation. PRINCIPLES OF THERMODYNAMIC 47 where λ ≡ T (sg − sl ) is the latent heat. The latent heat λ depends on the temperature and vanishes at some temperature Tc . From the equations of state for gas and liquid phases one can find the dependence P(T ) along the coexistence line. .

48 CHAPTER 4. THERMODYNAMIC QUANTITIES .

2) (5. respectively. we get for the energy E = −µ(n↑ − n↓ ) H . Using the Stirling’s formula. Consider N noninteracting localized particles with spin S = 1 2 embedded in an external magnetic field H. m = 2µS . with the normalization condition N = n↑ + n↓ . the thermodynamic weight for the system is W= N! . Under the assumptions of the microcanonical ensemble (i.3) N 1 one can replace ∑n = 0N ln n as N 0 ln x dx = N ln(N/e). e.1 Spin 1 in a magnetic field 2 Let us start with a simple example. 49 . with fixed total moment M). n↑ !n↓ ! (5. 1 µ = e/2mc is the Bohr magneton. Denoting n↑(↓) as numbers of states with spin up(down). Since Sz = ± 2 . The total magnetic moment is M = µ(n↑ − n↓ ) . The Hamiltonian is ε = −m · H = −mz H . 1 log n! ≈ n log n − n + 1 1 When at n 1 (5.1) The numbers n↑(↓) can be expressed in terms of E and M.Chapter 5 The Gibbs Distribution 5. ε↑(↓) = µH .

↓ ¯ U = Nε .1) as n↑(↓) = consequently. 2µH µHN − H 2µH n↓ ∂M ∂H = T (5. n↓ → n↓ + 1 is ∆U = 2µH and the entropy change is ∆S = log we obtain n↑ N! N! − log = log (n↑ − 1)!(n↓ + 1)! n↑ !n↓ ! n↓ + 1 n↑ ∆S 1 ≈ log . . Now we can express the quantities n↑(↓) in terms of the energy E given by Eq. The previous results can be summarized as a distribution function nσ = Here β = 1/T .5) Now there exist fluctuations around the average energy which are absent in the microcanonical ensemble.↓ ∑ e−βεσ = 2 cosh µH . Show that the difference is small for a macroscopic body. i. 1 = T Thus. T χ≡ ∂S ∂E µHN E .4) Nµ2 1 . THE GIBBS DISTRIBUTION S = N log N − n↑ log n↑ − n↓ log n↓ . M = Nµ tanh µH . ∆U 2µH n↓ That must be equal to the inverse temperature. assuming that the temperature is fixed by a thermal contact with a “thermal bath”. Z1 σ=↑. N −βεσ e . (5.50 we obtain CHAPTER 5. Z1 Z1 = σ=↑. T (5. ε2 − E 2 and compare it with the internal energy ¯ E = N ε. e. and we recover previous results. Now let us define the same relations from the canonical ensemble. We have ¯ ε= 1 µH ∑ εσ e−βεσ = −µH tanh T . Since the energy cost for one spin flip n↑ → n↑ − 1 . 2 T cosh (µH/T ) χ is called the magnetic susceptibility. 2µH = N n↑ 1 µHN + H 1 log = log . Exercise: Calculate the fluctuation in energy.

The Gibbs distribution is relevant for a body in an equilibrium with a large medium . The Maxwell-Boltzmann distribution Derivation from the Gibbs distribution Consider a system of non-interacting particles. s Here Z1 is the single-particle partition function. dE (0) β ≡ 1/T . THE GIBBS DISTRIBUTION 51 5. q) = V (q) + K(p) . fn = gn −βεn e .“thermal bath”. embedded into a medium. E =E (0) −En Now let us consider the body as very small comparing to the medium.5. fn ∝ .2 The Gibbs distribution Eq.q) dΓ .2. (5. Then En Finally. then we arrive at the single-particle distribution function. s E (0) → S (E (0) − En ) ≈ S (E (0) ) + En dS (E (0) ) = S (E (0) ) − En /T . If the systems consists of non-interacting particles which have eigenenergies εs . fn = In a classical system. 1 −βEn . f (p. Z1 Z1 = ∑ gs e−βεs . q) = 1 −βE(p. ∆E As a result. If there is a body.5) is the particular case of the Gibbs distribution.q) e . the microcanonical distribution can be written as dw ∝ δ(E + E − E (0) ) dΓ dΓ where superscript “prime” relate the quantities to the medium. Dividing the total energy into potential and kinetic parts. Z Z= e−βE(p. sth state being degenerate by gs times. E(p. e Z Z = ∑ e−βEs . then the probability to find the body in this state is fn ∝ δ(En + E − E (0) ) dΓ = δ(En + E − E (0) ) eS (E ) ∆E dΓ dE = dE eS (E ) δ(En + E − E (0) ) dE . Note that in the derivation we use the Hamiltonian of the whole system. If one fixes the body in a given state n.

d p e−βK(p) . 2 Statistical derivation Consider particles with energy levels εi and degeneracy factor gi . g j (ni + 1) Thus we arrive at the same result with Z = Z1 = ∑i gi e−βεi where εi are single-particle energy levels.6) and (5. while j gjj gi n j gni g j gni +1 i i − log = log . THE GIBBS DISTRIBUTION f (p. Here ni is the number of particle occupying ith level.52 we can write CHAPTER 5. . If the particles are identical and distinguishable the total number of microstates for a given state specified by the set {ni } is W = N! ∏ i 2 2 2 gni i .7) Comparing Eqs.6) n −1 n (5. N = ∑i ni . (5. For non-interacting particles the single-particle distribution is f p (p) = (2πmT )−3/2 e−(px +py +pz )/2mT . ∆S = log (ni + 1)! (n j − 1)! ni ! n j ! g j (ni + 1) nj → nj −1. i The easiest way is to use canonical ensemble which specifies the temperature. Then ∆U = T ∆S = εi − ε j .7). Exercise: Show that the average kinetic energy per atom is 3 T . ni ! E = ∑ ni εi . Let us consider the transition from jth to ith state. (5. q) = fq (q) · f p (p) fq = e−βV (q) / f p = e−βK(p) / dq e−βV (q) . ni → ni + 1 . we get gi n j = eβ(εi −ε j ) .

2. s N Z1 N! This formula is fundamental in thermodynamic applications of the Gibbs distribution.1 The Gibbs distribution for a variable number of particles . them one has to divide the thermodynamic weight by N! to obtain Z= Helmholtz free energy: N Z1 . As a result. N Here we employ the fact that Z = Z1 . N! S = − ∑ gs ( fs log fs − fs ) .2. The same result can be obtained from single-particle considerations. Using this formula one can express the Gibbs distribution in the form F = U − T S = −T log Z = −T log fn = e−β(F−En ) . Both the entropies and energies become functions on N. we get fnN ∝ exp S (E (0) − EnN . one has to specify the particle numbers. Z1 i ¯ U = Nε . For N distinguishable particles we have S = logW ≈ N log N − N + ∑(ni log gi − ni log ni ) i ni N ≈ N log N − ∑ ni log = N log N − ∑ ni log − βεi gi Z1 i i N = N log Z1 +U/T = log Z1 +U/T = log Z +U/T . N (0) − N) . so that N + N = N (0) . 5. THE GIBBS DISTRIBUTION 53 Expressions for thermodynamic quantities Average single-particle energy: ¯ ε= 1 ∂ log Z1 (β) ∑ εi gi e−βεi = − ∂β . N and N . Entropy: Since the entropy is given by the relation S = − log fn = − ∑ fn log fn n we obtain in general S = log Z +U/T . If the particles are non-distinguishable. . If the body can exchange by particles with the medium. N .5. respectively.

Thus. one obtains Ω = −T log ∑ eβµN ∑ eβEsN N s . Remember that for classical system of N non-distinguishable particles one has to use the classical Hamiltonian H ({p.N ∂S ∂N = −βµ . ∑ ∑ fsN = eβΩ ∑ N s N eβµN ∑ eβEsN s = 1. . THE GIBBS DISTRIBUTION Next. Here d is the dimensionality of the real space. V. (2π¯ )dN N! ∏ h i=1 while for distinguishable particles there is no factor 1/N!. q. fnN = A eβ(µN−EnN ) . N}) instead EnN . N (0) − N) = S (E (0) . N (0) ) − βEnN + βµN (due to the assumption of the equilibrium. we expand the entropy up to the 1st order in EnN and N.54 CHAPTER 5. the phase volume element being N 1 d pi dqi . The normalization constant A can be expressed through the thermodynamic quantities. Since ¯ ¯ S = − log fnN = − log A − βµN + βE we obtain Since → ¯ T log A = U − T S − µN = Ω . fnN = eβ(Ω+µN−EnN ) . Since ∂S ∂U we get = β. the temperatures and chemical potentials of the body and the medium are the same).V S (E (0) − EnN . U.

Another way to obtain this result is to use the continuous limit one can write d 3 p −βp2 /2m Z1 = V e .ny . y z 2mL2 x ∞ −∞ 3 nx . to get ∞√ √ Z1 = V D0 T x e−x dx = V /Λ3 .nz ∑ e −(β¯ 2 π2 /2mL2 )(n2 +n2 +n2 ) h x y z ≈ 1 2 dnx e −(β¯ 2 π2 /2mL2 )n2 h x = V /Λ3 . 0 Introduction of the density of states helps in calculation for different systems. we obtain εn = Thus.1 Classical gas Having in mind quantization in a box. the integral is easily done. 2. . (2π¯ )3 h It is very convenient to introduce the density of states . Z1 = Here ki = πni /L (ni = 1. (6.Chapter 6 Ideal gas 6. . . D(ε) = g p2 d3 p δ ε− (2π¯ )3 h 2m √ (2m)3/2 √ =g ε ≡ D0 ε .) h2 π2 2 ¯ (n + n2 + n2 ) . ψn (r) = sin(kx x) · sin(ky y) · sin(kz z) . 55 . 4π2 h3 ¯ (6.1) Λ = 2πβ¯ 2 /m h 1/2 = h 2π/mT ∝ T −1/2 ¯ is the thermal de Broglie wave length. 4π2 h3 ¯ (2m)3/2 D0 ≡ g .2) Here g is the degeneracy factor of the state with given momentum p Now.

Indeed. N T ∂ 3 log Z = NT . IDEAL GAS Exercise: Calculate the partition function for two-dimensional gas. ∂β 2 3 N. the classical statistics is applicable only at ρΛ3 1. with the energies εk . If the particles have internal degrees of freedom. N1 T 2 (6. N 1. N! NT ∂F = = ρT (ρ is the gas density) .4) (6.N V (6. ∂V T. Thermodynamic quantities Now we proceed with thermodynamics: F = −T log P = µ = U = − CV = N Z1 = −NT log(V /Λ3 ) + T log N! . k. the number of states V /Λ3 is much greater than the number of particles. then the free energy acquires an additional item. or occupied only by one particle.9).V Z1 PΛ3 = T log .N N Z1 U 5 + =N − log ρΛ3 . 2 ∂F S = − ∂T = log V. Using the Stirling’s formula for large number of particles . Only because of this assumption the equality Z= N Z1 N! holds.7) ∂F ∂N ≈ −T log T.9) This inequality is violated at low temperatures. F = −NT log 3∑ NΛ k N . Indeed.6) (6. k Ideal gas with a constant specific heat In many important cases the specific heat can be regarded as temperature-independent. What was wrong in out derivation? We implicitly assumed that one state can either be empty.56 CHAPTER 6. CV = −NT f (T ) . according to Eq. δF = −T N log ∑ e−βεk . we can write the Helmholtz free energy as eV eV e−βεk ≡ −NT log + N f (T ) .8) Something is wrong: the entropy and the specific heat does not vanish at zero temperature. N.3) (6. (6. (6.5) (6.

then the total number of degrees of freedom is 3n. Consequently. U + PV = Nε0 + NcP T . The kinetic energy depends on all 3n momenta. CLASSICAL GAS the assumption CV = const(T ) leads to the expression f (T ) = −cV T log T − ζc T + ε0 . As a consequence. We get. or l = 6n − 5.1. the relation between V .k=1 ∑ rv aik qi qk . q) is a quadratic function of 3n momenta and 3n − 6 (or 3n − 5) vibrational coordinates. U G H S = = = = = Nε0 + NcV T . P. Nε0 + NT log P − NcV T log T − Nζc T (since PV = NT ) . Here rv is the number of vibrational degrees of freedom. If the molecule contains n atoms. e. then the potential energy of the molecule potential energy can be expanded in powers of the mechanical displacements q from the equilibrium positions. Polyatomic gases Vibrational degrees of freedom If any internal motion of the molecules is present. PV γ = const . the total energy and be written as ε = ε0 + fII (p. we can derive the Poisson adiabatic. 57 where cV ≡ CV /N. Thus we have the total number of variables l = 6n − 6. If the atoms in the molecule are collinear. N log(eV /N) + NcV log T + (ζc + cV )T −N log P + NcP log T + (ζc + cP )N . while ζc and ε0 are integration constants. V = ε0 + i. i. there are 3 translational degrees of freedom (for a molecule as a whole) and 2 rotational degrees of freedom. For a monoatomic . Since −N log P + NcP log T = const one obtains T cP /P = const → T γ P1−γ = const . Here we employed the relation cP − cV = 1 (prove!) which is valid for the ideal gas. Since PV = NT TV γ−1 = const . ζc is called the chemical constant. and T for adiabatic expansion of compression of ideal gas with constant specific heat. while 3n − 5 modes are left for vibrational motion. If the molecule is not a collinear one. where fII (p.6. q) . γ = cP /cV . then rv = 3n − 6.

while the total spin can be finite (e. εJ T . then Z = e−βε0 where ε0 is the energy of non-degenerate ground state. e · e−ε0 /T N dΓ e− fII (p. q) = T fII (p . then the situation is rather simple. when e−βεJ ≈ 1. Usually. Expanding the logarithm. If we label the energy by the total angular momentum J. L = S = 0. dΓ = T l/2 dΓ . we get cV = l/2 . g. . ζc = (3/2) log(m/2π¯ 2 ) . q = q T → In this way. IDEAL GAS gas the total number of variables is 3 because coordinates do not appear in the expression for the energy. If only one of the angular momenta is non-zero in th ground state of the atom.58 CHAPTER 6. This expression can be simplified at high temperatures. Then Z = (2S + 1) instead of 1. Denoting Z = ∑ e−βεk k we obtain F = −NT log eV eV Z = −NT log 3 NΛ N mT 2π¯ 2 h 3/2 Z . Monoatomic ideal gas Let exclude the kinetic energy of translational motion and label atomic intrinsic levels as εk . Thus a classical gas has a constant specific heat. and ζS = log(2S + 1). vapor of alkali metals). cV = 3/2. q ). In the simplest case.q) . then Z = ∑(2J + 1)e−βεJ J since each level with a given J is (2J + 1) degenerate with respect to directionds of J. when the atoms are in the ground state with zero angular moment and spin. Then ∑J (2J + 1) = (2L + 1)(2S + 1) and ζSL = log[(2S + 1)(2L + 1)] . we get F = −NT log AV −βε0 T l/2 /N . As a result. Starting the energy scale from ε0 we obtain: Z = 1. we obtain. If both momenta are non-zero than there is a fine structure in the levels which can be compared with the temperature. F = −NT log Now we can transform the variables as √ √ p = p T. cP = cV + 1 = (l + 2)/2 . fII (p. We can formulate a rule: Each degree of freedom contributes T /2 (or kT /2 in ordinary units) to the average energy. in a ground state L = 0. h These results were obtained by Sackur and Tetrode in 1912.

ctr = 3/2 . cP = ctr + crot + cvib + 1 . h crot = 3(¯ 2 /IT )2 3e−¯ h 2 /IT . h Zrot = 1 − 3e−β¯ /I 2 to obtain h Frot = −3NTe−β¯ 2 /I . Introducing the translational contributions to the specific heat and chemical constant as ζc = ζtr + ζrot + ζvib . ζrot = log(2I/¯ 2 ) . I = r0 m1 m2 /(m1 + m2 ) is the inertia moment of the molecule. when T replace summation by the integration to obtain h2 /2I. cP = 7/2 . v is the vibrational quantum number. h In this case cV = 5/2 . CLASSICAL GAS Diatomic gases 59 In this cases additional rotational and vibrational degrees of freedom must be taken into account. h cV = ctr + crot + cvib . Then one can ¯ Zrot = 0 ∞ h (2K + 1) e−β¯ 2 K(K+1)/2 = 2T I/¯ 2 . and vibrational contributions. Z = e−βε0 · Zrot · Zvib Zrot = Zvib = As a result. The 2 last terms are defined according to the definition δFα = −NT log Zα . Under some approximation. . eV F = −NT log N mT 2π¯ 2 h 3/2 K=0 ∞ v=0 h ∑ (2K + 1) e−β¯ K(K+1)/2I . 2 ∞ h ∑ e−β¯ ω(v+1/2) . In the opposite limit it is sufficient to retain the first two terms. the energy can be written as sum to electronic. K is the rotational ¯ 2 quantum number. 2 2I Here hω0 is the vibrational quantum.1. As an example let us consider the case of large temperatures.6. rotational. h crot = 1 . ζtr = (3/2) log(m/2π¯ 2 ) . Thus. In the simplest case of a singlet electronic state εvK = ε0 + hω0 v + ¯ 1 h2 ¯ + K(K + 1) . + Nε0 + Frot + Fvib .

6. IDEAL GAS Note that above expressions are valid only for the molecules with the different atoms. Temperature is measured in units of h2 /2I in the graph (1) and ¯ in units of hω in the graph (2).4 0.5 Reduced temperature 2 Figure 6. Page 1 . 1 0. ¯ Exercise: Calculate approximate expressions for the case of low and high temperatures and discuss their physical meaning.1: Contributions to the specific heat from rotational (1) and vibrational (2) degrees of freedom. The graphs for the rotational and vibrational contributions to the specific heat are shown in Fig.5 1 1. The quantum case is more difficult and we do not discuss it. 2 h (β¯ ω)2 eβ¯ ω h h eβ¯ ω − 1 . As a result. Using the lowest vibrational level as the origin for the energies. The vibrational degrees of freedom can be easily accounted by direct summation of geometric series. h Svib = −N log 1 − e−β¯ ω + N Uvib = N cvib = hω ¯ h eβ¯ ω − 1 β¯ ω h . h Fvib = NT log 1 − e−β¯ ω . the rotational partition function must be halved.6 0.8 0.1. then at large temperatures two opposite directions correspond to the same physical state. h eβ¯ ω − 1 .2 0 2 1 0. If the atoms are equivalent.60 CHAPTER 6. we obtain Zvib = h 1 − e−β¯ ω .

Now we have to use the concept that the particles are indistinguishable. This procedure is . 1. Introducing the ˆ permutation operator P which interchanges i-th and j-th particle. ˆ PΨs = pΨs . Then h ∞ 1 h . Example: Particles in a Harmonic Potential: Ignoring zero-point vibrations we have εn = n¯ ω. 2 . ˆ ˆ ˆ H = ∑ Hi . we get p2 = 1 → p = ±1 . The particles with p = 1 called the bosons. Hi = i=1 N ˆi p2 +V (ˆ i ) . ri ↔ r j . ˆ Hi ψk (ri ) = εk ψk (ri ) . CLASSICAL GAS 61 Quantum Gases Statistics of identical particles Consider N indistinguishable particles moving in a common potential. 2! 2 (1 − x)2 2 Note: The product Z1 contains 1 configuration where 2 particles occupy the same state. ˆ ˆ Since double permutation is just the identity operator. Specifying the total wave function as Ψs = ∏i ψki (ri ) we obtain Es = ∑i εki since the Hamiltonian is the sum of independent terms.1. P = 0 . n = 0.6. The antisymmetric wave function vanishes if two particles are in the same state. This property is called the Pauli principle. Z1 = ∑ x n = 1−x n=0 For the Maxwell-Boltzmann (MB) statistics we have MB Z2 = 2 Z1 1 1 = . the proper many-body wave function must be an eigenstate of the permutation operator. one has to require ˆ ˆ ˆ ˆˆ ˆ H . . Then. P2 = I. or PH P−1 = H . Let h us denote single-particle eigenenergies as εk and the states as ψk . r 2m ˆ ˆ Here pi = −i¯ ∇i and ri ore operators for the momentum and coordinate of the i-th particle. while the particles with p = −1 are called the fermions. By dividing this product by 2! we ascribe the weight 1/2 to this configuration. x ≡ e−β¯ ω . . . According to the quantum theory.

(1 − x)(1 − x2 ) ∞ 1 In a similar way. (1 − x)(1 − x2 ) 1 (large temperature). Derivation from a grand canonical ensemble Consider the system of particles with single-particle levels εi and degeneracy factors gi . 1 FD MB Z2 = Z2 − D . The energy change is εi . IDEAL GAS wrong from the point of view of quantum statistics. the number of available states for two particles being gi (g1 − 1)/2. 1 BE MB Z2 = Z2 + D . while the entropy change is (prove!) ∆S = log Since T ∆S = ∆U − µ ∆N = εi − µ we obtain gi − = eβ(εi −µ) ni → ni = gi . then there are only gi − 1 states at the i-th level for the second particles. 2 → BE Z2 = n=0 ∑ x2n = 1 − x2 1 . consider a transition ni → ni + 1. so the the diagonal term D= must be added with the weight 1/2. For ni particles we have W =∏ i gi ! . bosons are allowed to occupy the same single-particle state. Namely. the change in the particle number is ∆N = 1. ni + 1 ni . The particles obeying the Fermi-Dirac statistics are called the fermions. Indeed. 2 The difference is negligible for 1 − x → FD Z2 = x . β(εi −µ) + 1 e gi − ni gi − ni ≈ log . If the state is occupied. ni !(gi − ni )! Now we apply the conventional approach.62 CHAPTER 6. The factor 1/2 appears because the particles are non-distinguishable. The Fermi-Dirac Statistics This statistics is based on the Pauli principle: Each quantum state cannot be simultaneously occupied by more than 1 particle. the diagonal term must be subtracted for Fermi-Dirac (FD)statistics.

i (6. ni Consequently.13) (6. e i +1 the normalization condition (which determines µ) being gi N = ∑ gi fi = ∑ β(ε −µ) .15) is applicable only if eβ(µ−εi ) < 1 → µ < 0 . It is worthwhile noting that Eq. ni 1 = β(ε −µ) .16) Again. we recover the expression (6. (6. W =∏ i (ni + gi − 1)! ni !(gi − 1)! fi = → ∆S ≈ log gi + ni .1.11) by performing summation over ni from 0 to ∞. The following way is standard: fi = ni = − ¯ The total Landau free energy is Ω = −T log ∑ 1 + eβ(µ−εi ) . CLASSICAL GAS As a result. ∂µ e i +1 (6. (6. the probability for a state i to be occupied is 1 fi = β(ε −µ) .12) ∂Ωi 1 = β(ε −µ) . (6.10) .11) Now. since ni = 0 or 1. Using the properties of geometric series we obtain.15) gi e i − 1 We can derive this expression also from the general formula (6. We have the general expression Ωi = −T log ∑ eβ(ni µ−ni εi ) = −T log ∑ eβ(µ−εi ) ni ni ni 63 (6.6. i . Ωi = T ln 1 − eβ(µ−εi ) . A proper picture is gi boxes separated by gi − 1 walls. We immediately recover the Boltzmann distribution at ni gi by replacing (ni + gi − 1)! by gni . we get Ωi = −T log 1 + e(βµ−εi ) . (6.14) The Bose-Einstein distribution For bosons the number of particles occupying any state can be arbitrary. To count the number of configurations one can count the different grouping of gi − 1 walls and ni bosons.15) as fi = −∂Ωi /∂µ. i +1 i i e Derivation from the Gibbs distribution Consider the particles belonging to i-th state. As a result.

21) we arrive at the Fermi-Dirac distribution (6. not more than one particle can be in each state. IDEAL GAS 6.20) and (6. gi (− ln fi + α + βεi ) = 0 → fi = eβ(µ−εi ) . (6. This problem can be mapped on the problem of distributing ni identical . respectively. i S = ∑ ni ln(egi /ni ) ≡ ∑ gi fi ln(e/ fi ) . i i (6. requiring this expression to have a maximum under conditions (6.2 Ideal gases out of equilibrium Let us assume that we label the states of the system by the subscript i. Let us now require that the entropy must be a maximum for the equilibrium state. Thus one has to count the number of way to fill ni states from gi available ones. i As a result.64 CHAPTER 6.17) (6. Then the problem is to find fi such that the sum (6.13).19) has the maximum value possible under the restrictions ∑ ni = ∑ gi fi = N i i i i (particle number conservation) . i (6. S = lnW = ∑ ln(gni /ni !) = ∑ (ni ln gi − ln ni !) . the degeneracy factor and the occupation number of ith state being gi and ni . Since the particles are independent. energy conservation) .18) Now we can use the Stirling’s formula to get (6.23) Again (Check!).20) (6.19) Here we have introduced the distribution function fi ≡ ni /gi as the probability for ith state to be occupied. Then the number of different physical configurations is ∆Γi = gni /ni ! . W = ∏ ∆Γi i For the Boltzmann gas one can place each of ni particles in one of the possible gi state. For the Bose-Einstein statistics one has to calculate the number of ways of distributing ni particles among gi states.22) For the Fermi gas.21) ∑ εi ni = ∑ gi εi f i = E Requiring that one obtains ∂(S + µβN − βE)/∂ fi = 0 . Thus S = − ∑ gi [ fi ln fi + (1 − fi ) ln(1 − fi )] . Then the set {ni } describes the state of the system if it is also out of equilibrium. and ∆Γi = gi !/ni !(gi − ni )!. (6.

(6. The second inequality is obtained using integration by parts. z−βµ ± 1 β(ε−µ) ± 1 V 0 e 0 e Here “+” stands for fermions while “-” stands for bosons. i (6.26) Since Ω = −PV we arrive at the following relationship: 2 PV = U . Ω = V D0 T ∞ 0 √ 2 dε ε log 1 ± eβ(µ−ε) = − V D0 3 ∞ 0 ∞ 0 dε ε3/2 f (ε) .. Finally. The number of configuration is just the number of ways to choose g1 − 1 positions for the strokes.2). Thus the equation for the chemical potential has the form. i (6. the diagram .19) at ni gi ! ≈ (gi − ni )!gni .. √ √ ∞ ∞ N ε dε z dz 3/2 = dε D(ε) f (ε) = gD0 = D0 T . We have ∆Γi ≈ ni i /(gi − 1)! and S = ∑ gi ln(e fi ) .25) 6.. Here g is the spin degeneracy. We recognize that Ω = −(2/3) ∑ gi εi fi = −(2/3) i D(ε)ε f (ε) dε = −(2/3)U . 0.3. (gi + ni − 1)! ≈ (gi − 1)!gni i i at ni gi . FERMI AND BOSE GASES OF ELEMENTARY PARTICLES 65 balls among gi urns. 3 . In a similar way.. For example.||. (6.6.3 Fermi and Bose gases of elementary particles For elementary particles we have ε= p2 . and S = ∑ gi [(1 + f1 ) ln(1 + fi ) − fi ln fi ] . 4. see Eq. let us specify the expression for the entropy of Bose gas at fi 1 since we will need this (g −1) expression later to discuss the Bose condensation.24) gi since The expressions (6..|. 2}. 3.23) and (6. ∆Γi = ((gi + n1 − 1)!/ni !(gi − 1)!. Let us imagine that the latter are separated by by gi − 1 vertical strokes placed at intervals along the points.24) tend to the Boltzmann formula (6. Thus. The total number of places (occupied both by strokes and points) is then (gi + n1 − 1).. represents 10 balls among 5 urns as {1. 2m g = (2s + 1) .|.

20) one can express the free energy as F = FBol ± π3/2 N 2 h3 ¯ · . . one can expand 1 ez−βµ ± 1 ≈ eβµ−z 1 eβµ−z ± . In the limiting case of the Boltzmann gas.26) we obtain 1 βµ gV m3/2 T 5/2 2βµ Ω ≈ ΩBol 1 e =≈ ΩBol ± e 25/2 16π3/2 h3 ¯ where ΩBol is the Boltzmann value for the Landau free energy. . IDEAL GAS N 1 =− V V ∂Ω ∂µ T. have nothing to do with the ratio of specific heats. . (6. T 5/2 P = const.66 This result holds also for Boltzmann distribution. and µ/T must be constant.V ) ± 1 ∞ z1/2 dz . T ) = D0 T 5/2 N(V. The next term provides proper corrections.V the ratio S/N depends only on the ratio µ/T . Since S 1 =− V V ∂Ω ∂T and V. 0 ez−βµ(T. cV /cP . If the process is adiabatic. 1/2 m3/2 2g V T .µ CHAPTER 6. however. since the conditions cP /cV = 5/3 and cP − cV = 1 do not hold any more.V ) ± 1 0 ∞ z3/2 dz We have to know some tips to calculate the involved integrals. Equation of state The equation of state can be written in a parametric form: P(V. T ) = V D0 T 3/2 . V T 3/2 = const . ez−βµ(T. then S = const. where η is a function of a single variable. Consequently. Substituting the Boltzmann value for the chemical potential in the second term and using the concept of small increments (4. The lowest approximation leads to the Boltzmann value of the chemical potential and to Clapeyron’s equation PV = NT (Check!). From Eq. PV 5/3 = const . eβµ 1. Using the substitution of the variables z = βε we observe that Ω = −PV = V T 5/2 η(µ/T ) . The exponents.

F Since 2· we have p3 N F = . eβ(ε−µ) + 1 where f (ε) is an arbitrary function such that the integral converges. and g = 2s + 1 = 2. Consider I= 0 ∞ f (ε) dε . ¯ (3π2 )2/3 h2 P = 5 m .6. Introducing z = β(ε − µ) and remembering that β ≡ 1/T we get I=T ∞ −βµ f (µ + T z) dz = T ez + 1 βµ 0 f (µ − T z) dz + T e−z + 1 ∞ 0 f (µ + T z) dz .3. 2 h3 V 3π ¯ → pF = (3π2 )1/3 (N/V )1/3 h.27) The conditions for the correction to be shall is naturally the same as for applicability of the Boltzmann statistics. FERMI AND BOSE GASES OF ELEMENTARY PARTICLES 67 Finally. h 4πp2 d p ·V p2 d p =V (2π¯ )3 h π2 h3 ¯ Limiting the integration of the absolute value of the electron moments by pF we obtain U V p5 3(3π2 )2/3 h2 ¯ F = = 2 h3 10 m 10mπ ¯ N V 5/3 N V 5/3 . To calculate specific heat at low temperatures one has to be able to find temperature-dependent corrections. differentiating this expression with respect to volume we arrive at the equation of state: PV = NT 1 ± π3/2 N h3 ¯ . ez + 1 . Quantum effects behave and an additional attraction for bosons and as an additional repulsion for fermions. ¯ εF = (3π2 )2/3 (N/V )2/3 (¯ 2 /2m) . At zero temperature. A degenerate electron gas For electron spin equals to 1/2. one can approximate the Fermi function 1 1 at ε < µ f (ε) = β(ε−µ) ≈ 0 at ε > µ e +1 Thus there is a maximum energy µ|T =0 ≡ εF = p2 /2m . 2g V (mT )3/2 (6. Let us demonstrate the way to calculate these corrections.

ez + 1 The last term is exponentially small since βµ term. 2/3 ∼ T . ez + 1 0 0 At low temperature one can expand the second integrand as f (ε) dε + T f (µ + T z) − f (µ − T z) = 2T f (µ) z + As a result. Returning to the initial variables in the first f (µ + T z) − f (µ − T z) dz . I = 0 µ ∞ 2 3 T f (µ) z3 + · · · . we obtain I= µ 1. IDEAL GAS = 1− 1 ez + 1 ∞ βµ f (µ − T z) dz + T f (µ + T z) − f (µ − T z) dz + ez + 1 f (µ − T z) dz . εF A degenerate Bose gas The chemical potential of bosons is negative.28) Now we can calculate temperature-dependent corrections to the thermodynamic quantities. 3! ∞ 0 = 0 µ z dz 1 + T 4 f (µ) z +1 3 0 e 2 4 π 7π 4 f (ε) dε + T 2 f (µ) + T f (µ) + · · · . γ≡ π 3 2/3 m . h2 ¯ . 6¯ 3 h Since the second term is a small correction we can replace µ → εF in that term. I (z) = . 3 z+x − 1 V 0 e 2 π2 h ¯ (6. It is given by the equation √ ∞ N g(mT )3/2 x dx = 1/2 I (β|µ|) . Substituting f (ε) = ε3/2 we get 3/2 √2µ 2m Ω = Ω0 −V T .29) . Using the theorem of small increments we find 1 V F = F0 − γNT 2 2 N S = γNT C = γNT V N V N 2/3 2/3 . 6 360 f (ε) dε + 2T 2 f (µ) ∞ z3 dz +··· ez + 1 (6.68 Now we can replace in the 1st integral 1 e−z + 1 to obtain I=T 0 βµ ∞ 0 CHAPTER 6.

(6. FERMI AND BOSE GASES OF ELEMENTARY PARTICLES 69 It increases with the temperature decrease and reaches zero at the temperature T0 given by the equation N g(mT0 )3/2 = 1/2 I (0) .5 1 1. at T < T0 the particles with ε > 0 are distributed according to the BE distribution with µ = 0.8 0. This is not true for bosons because they can condense in the √ ground state with εk = 0.2: Plot of the ratio F (z) ≡ I (z)/I (0).5 z~ 2 2. V 2 π2 h3 ¯ √ Since I (0) = ( π/2)ζ(3/2) where ζ(x) is the Riemann’s ζ-function. =N T0 e −1 2 π2 h3 0 ¯ . 0. 6. 2 h3 eβε − 1 2 π ¯ The total number of such particles is √ gm3/2V ∞ ε dε T 3/2 Nε>0 = 1/2 = βε .4 Figure 6. Chemical potential can be found by graphical solution of Eq.29) we obtain (βT0 )3/2 = I (β|µ|)/I (0) .2.6.5 3 What happens at T < T0 ? The above equation is based on replacement of the summation over discrete states by the integration. (6.3. T0 ≈ From Eq.2 0 0. As a result.30). We omit this term in our integration because εk = 0 for the ground state.30) The ratio F (z) ≡ I (z)/I (0) is plotted in Fig. 0. 1 3.31 h2 ¯ 2/3 m g N V 2/3 . √ gm3/2V ε dε dnε = 1/2 . (6.6 0. The dependence of the product β|µ| versus βT0 can be easily found graphically from this plot.

770NT (T /T0 )3/2 . photons must interact with some amount of matter which is able to absorb and emit photons.70 The remaining T Nε=0 = N 1 − T0 3/2 CHAPTER 6.V ∂F =0 ∂N Here D(ω) = ω2 /π2 c3 . (6. In this way we obtain (check!) U = 0. In this way we obtain dnω = V where N(ω) ≡ ω2 N(ω) dω π2 c3 1 h eβ¯ ω − 1 .0851gm3/2 T 5/2 /¯ 3 . Since the free energy must be minimum µ= we obtain µ = 0. and the one obtains for the density of states 2 ·V 4πk2 k ω2 dω d3k = 2 ·V = V 2 3 ≡ V D(ω) dω . It can be expressed as gas of non-interacting photons. CV = 5U/3T . The energy and other thermodynamic quantities can be obtained from general expression for Bose gas by putting µ = 0. h Thus at T ≤ T0 the pressure is ∝ T 5/2 and it is volume-independent. The reason is that the particles in the condensate do not contribute to the pressure. and the chemical potential µ should be fixed. (2π)3 (2π)3 π c ∂F ∂N and T. as well as by polarization. Since photons are transverse. IDEAL GAS particles are in the lowest state with zero energy.4 Black-body radiation Electromagnetic radiation which in a thermal equilibrium is called the black-body radiation. The photons are specified by the wave vector k (the momentum being hk) and frequencies ωk = ¯ ck (the energy being hωk = hck). Thus we arrive at the situation with variable number of photons. 6. To achieve the equilibrium. F = U − T S = −(2/3)U = Ω (since) µ = 0 .31) P = −(∂F/∂V )T = 0. there ¯ ¯ are 2 directions of polarization.

67 × 10−5 g/sec3 deg4 . BLACK-BODY RADIATION 71 is called the Planck distribution. . θ) = cEω cos θ A(ω. dEω = V T ω2 dω . U = 4σV T 4 /c = −3F .4.6. h For other quantities we obtain. Let us not define the absorbing power. e. If we also assume that the radiation does no pass through the body we arrive at the balance equation between the radiation density and the intensity of emission J(ω. It temperature is measured in degrees. Then the energy flux density leaving each point of the body is cEω do dω. π c e h −1 its low-frequency asymptotics is the Rayleigh-Jeans law. Here σ ≡ π2 /60¯ 3 c3 is the Stefan-Boltzmann constant. g. we can also assume that the body does not scatter radiation and is not fluorescent. e. θ) cos θ do dω . h then σ ≡ π2 k4 /60¯ 3 c3 = 5. e. CV = 16σV T 3 /c . 2 c3 π is the Wien’s law. dEω = V while the high frequency one. S = 16σT 3 /3c . i. For simplicity. The incident energy flux on unit area of the surface at an angle θ to the normal is therefore cEω cos θ do. F =T 0 ∞ h D(ω) log 1 − e−β¯ ω = −4σV T 4 /3c . θ) do dω in the same direction: J(ω. as A(ω. A body in thermal equilibrium with black-body radiation: Let us denote 1 dEω h ω3 ¯ Eω = = 3 2 β¯ ω 4πV dω 4π c e h − 1 for the spectral density of black-body radiation per unit volume and unit solid angle. 1900).. PV = U/3 = 4σT 4 /c . i. that reflection occurs without change in the angle θ and in frequency ω. do = 2π sin θ dθ. π2 c3 h ω3 −β¯ ω ¯ e h dω . θ). Then the energy absorbed per unit time and surface area is cEω A(ω. The thermodynamic quantities can be derived in a usual way. fraction of the incident energy absorbed by the body. The energy distribution is obtained by multiplication by hω ¯ (Planck. h ω3 ¯ 1 dEω = V 2 3 β¯ ω dω . θ) → a universal function of ω and θ .

Phonons For small displacement on an atom from its equilibrium position one can expand the potential energy near its minimal value V (R) = V (R0 ) + + 1 6 d 3V dR3 R0 dV dR (R − R0 ) + R0 1 2 d 2V dR2 (R − R0 )2 + + R0 (R − R0 )3 + · · · (6. R0 d 3V dR3 ≡ −2γ > 0 R0 we get the following expression for the restoring force for a given displacement x ≡ R − R0 F =− dV = −Cx + γx2 dx (6. → J0 = do dω J(ω.32) If we expand the energy near the equilibrium point and denote d 2V dR2 ≡ C > 0. One-Atomic Linear Chain Dispersion relation We start with the simplest case of one-atomic linear chain with nearest neighbor interaction.3: mun +C(2un − un−1 − un+1 ) = 0.5 Lattice Vibrations.72 CHAPTER 6. θ) = cEω cos θ . then the Kirchhof’s law can be formulated only for integrals over frequencies and angles. IDEAL GAS This is the Kirchhof’s law. if the body absorbs all the radiation. Then J(ω. θ) = cU/4V = σT 4 . /reff-ph1) Figure 6. 6. θ) = 1. ¨ 1 If (6.33) he comes to the Newton equation (see Fig.34) there is scattering.33) The force under the limit F = −Cx is called quasi elastic. If one expands the energy near the equilibrium point for the nth atom and use quasi elastic approximation (6.1 If A(ω. i. it is called the black body. e. .

Consequently. (6.6. The usual way to overcome the problem is to take into account that actual number L of sites is large and to introduce Born-von Karmann cyclic boundary conditions un±L = nn . The maximal frequency is a typical feature of discrete systems vibrations. Another important feature is Figure 6. We can fulfill this condition automatically by searching the solution as un = Aei(qan−ωt) . It differs from the dispersion relation for an homogeneous string. The spectrum is shown in Fig.4: Dispersion law for vibrations in a one-atomic chain. it is impossible to discriminate between q and q and it is natural to choose the region π π − ≤q≤ (6. 6. LATTICE VIBRATIONS.4. 2 ωm = 2 C/m. (6.36) is called the dispersion law. a where g is an integer. Note that there is the maximal frequency ωm that corresponds to the minimal wave length λmin = 2π/qmax = 2a.5.35) The expression (6.37) a a to represent the dispersion law in the whole q-space. PHONONS 73 To solve this infinite set of equations let us take into account that the equation does not change if we shift the system as a whole by the quantity a times an integer. Periodic character is clearly seen. ω = sq.35) does not change (because exp(2πi × integer) = 1). that if we replace the wave number q as q → q = q+ 2πg . Immediately we get ω = ωm sin qa .36) (6. the solution (6. Now we should recall that any crystal is finite and the translation symmetry we have used fails.38) .

5. − <g< (6. The density of states D(ω) is shown in Fig. we get that the wave number q should be discrete. substituting the condition (6.41) that for small structures of modern electronics this assumption need revision. At small q we get from Eq.74 CHAPTER 6. IDEAL GAS This condition make a sort of ring of a very big radius that physically does not differ from the long chain.5: Density of vibrational states in a one-atomic chain. . 6. Phase and Group Velocity Now we discuss the properties of long wave vibrations. (6. π/a). So. the wave number q takes L discrete values in the interval (−π/a. ω + dω. qaL = 2πg with an integer g. Violation of this assumption leads to the specific interface modes.40) It is divergent at ω → ωm which is a feature of 1D systems. with different q in the frequency interval ω. Density of States Because of the discrete character of the vibration states one can calculate the number of states.2 Immediately.35) we get exp(±iqaL) = 1. L L 2π g . Consequently.39) q= a L 2 2 (it is convenient to consider L as a large even number). Indeed. There is a singularity at the maximal frequency that is a typical feature of 1D systems.36) ω = sq . 2 Note (6. One easily obtains (check!) D(ω) = 2L π 1 ω2 − ω2 m . Note that this interval is just the same as the Wigner-Zeitz cell of the one-dimensional reciprocal lattice.38) into the solution (6. (6. Figure 6. D(ω). for a linear chain.

6. LATTICE VIBRATIONS. i. The corresponding picture is shown in Fig. 6. We use this case to discuss vibrations of compound lattices. vn = Av ei(qan−ωt) . We have s=a sp = sg sin(aq/2) ω . PHONONS where 75 C (6.5. there is the dispersion of the waves.6: On a two-atomic chain. Diatomic Chain.46) . =s |q| aq/2 dω = = s| cos(aq/2)| . ¨ (6. the sound velocity becomes q-dependent. If we assume the elastic constants to be C1. e. (6.44) It is natural to use once more the translation symmetry condition and search the solution as un = Au ei(qan−ωt) . The first is responsible for the propagation of the equal phase planes while the last one describes the energy transfer.6 Figure 6.2 we come to the following equations of motion: m1 un = −C1 (un − vn ) −C2 (un − vn−1 ) . One can see that the elementary cell contains 2 atoms. ¨ m2 vn = −C1 (vn − un ) −C2 (vn − un−1 ) . One can discriminate between the phase (s p ) and group (sg ) velocities.45) After substitution to Eqs. Acoustic and Optical branches. To formulate these equations it is convenient to express these equations in a matrix form introducing the vector A ≡ Au Av and the so-called dynamic matrix ˆ D= C1 +C2 m1 iaq 2 − C1 +C2 e m − C1 +C21e m −iaq C1 +C2 m1 (6. (6. dq (6.42) m is the sound velocity in a homogeneous elastic medium.44) we get the set of equations for the constants Ai .43) At the boundaries of the interval we get s p = (2/π)s while sg = 0 (boundary modes cannot transfer energy). In a general case.

it has a solution only if ˆ ˆ det(ω2 1 − D) = 0 .7: Dispersion laws for acoustic and optical phonons at C1 = C2 = C. 1 − γ2 .2 for a given value of q. ωopt (0) = ω0 . We see a very important difference with the case of monoatomic chain: there are 2 branches ω1. ω0 ωac (π/a) = √ 2 ω0 ωopt (π/a) = √ 2 1− 1+ 1 − γ2 . We get Figure 6. (6.2 where ω2 = ω2 0 1 2 1 − γ2 sin2 aq 2 m1 m2 . we have the inequality chain ωopt (0) = ω0 > ωopt (π/a) > ωac (π/a) > ωac (0) = 0 .2 are real because |γ| ≤ 1.48) This is just the equation which determines the eigenfrequencies. So. m1 m2 γ2 = 16 C1C2 (C1 +C2 )2 The frequencies ω1. CHAPTER 6.47) (6.7. 6. To understand the physical reason for these names let us consider the limits of zero and maximal q. We get ωac (0) = 0 .76 The equation for A has the form ˆ ˆ ω2 A − DA = 0 . ω2 = 1. The spectrum is shown in Fig.49) (C1 +C2 )(m1 + m2 ) . This is homogeneous equation.50) . (m1 + m2 )2 (6. The lower branch is called the acoustic branch while the upper one is called the optical branch. IDEAL GAS (6.

6.5. LATTICE VIBRATIONS. PHONONS

77

Now we can discuss the structure of vibrations in both modes. From the dispersion equation (6.47) we get u Au C1 +C2 e−iqa . (6.51) Pac,opt = n = = vn ac,opt Av (C1 +C2 ) − m1 ω2 ac,opt At very long waves (q → 0) we get

Pac = 1 , Popt = −

m2 m1

(6.52)

So, we see that in the acoustic mode all the atoms move next to synchronously, like in an acoustic wave in homogeneous medium. Contrary, in the optical mode; the gravity center remains unperturbed. In an ionic crystal such a vibration produce alternating dipole moment. Consequently, the mode is optical active. The difference between acoustic and optical modes is shown in Fig. 6.8.

Figure 6.8: Acoustic and optical branches of vibrational spectrum.

Vibration modes of 3D lattices
Now we are prepared to describe the general case of 3D lattice. Assume an elementary cell with s different atoms having masses mk . Then there are 3 acoustic modes with frequencies vanishing at q → 0. At low frequencies they correspond to acoustic waves in an elastic media. One mode is a longitudinal, while two others are transverse ones. The rest 3s − 3 ones are the optical modes.

Specific Heat of Crystal Lattice
From the classical statistical physics, the average kinetic energy per degree of freedom is ¯ εk = T /2, the total energy being ¯ ¯ ¯ ¯ ε = εpot + εkin = 2εkin . As a result, the total energy is ¯ U = 3N0 · εkin = 3RT,

78

CHAPTER 6. IDEAL GAS

where R is the Rydberg constant while N0 is the Avogadro one. As a result, the specific heat is cV = 3R = 5.96 cal/K · mole. This relation is violated at low temperatures as far as the temperature becomes less than the socalled Debye temperature (which is for most of solids is within the interval 100-400 K), namely it decreases with the temperature decrease. To understand this behavior one should apply quantum mechanics. First let us recall the average energy for photons calculated previously. The average energy is hω ¯ 1 ¯ ε= + hω N(ω), ¯ N(ω) = β¯ ω 2 e h −1 where N(ω) is the Planck function. The first item is energy-independent while the second one is just the average energy of bosons with zero chemical potential. Now we have to recall that the proper excitations in a lattice are collective vibrations, phonons, the average energy being U = ∑ hω j (q) N[ω j (q)] = ∑ ¯
jq j ∞ 0

D j (ω) hωN(ω) dω . ¯

Here j is the number of the vibration branch. For acoustic modes we use the so-called Debye model. Namely, we introduce the average sound velocity s0 as 1 1 1 2 = + 3 3 3 s0 3 sl st and do the calculation of the density of states similar to the one for photons. As a result, we get V D(ω) = 3V hω2 ¯ ≡ V D 0 ω2 . 3 s3 2π 0

The contribution of acoustic modes to the internal energy is Uac = V D0
ωD


0

ω3 h eβ¯ ω − 1

where the so-called Debye frequency is determined from the condition of the total number of modes to be equal to 3N for all acoustic branches, 3N = V D0 From this equation ωD = s0 6π2
1/3 ωD 0

ω2 dω.

V0

,

qD =

ωD s0

6.5. LATTICE VIBRATIONS. PHONONS

79

where V0 is the cell volume. The order-of-magnitude estimate for the maximal wave vector qD is π/a. So according to the so-called Debye model all the values of q are confined in a sphere with the radius qD . Usually, the Debye energy is introduced as Θ = hωD = hs0 ¯ ¯ 6π2
1/3

V0

.

The typical value of this temperature can be obtained from the rough estimate a = 10−8 cm, s0 = 105 cm/s. We get ωD = 1013 s−1 , Θ = 100 K. For optical modes, let us assume that ω j (q) = ω j0 , i. e. that frequencies of optical phonons are q-independent. Then all the procedure is similar for harmonic oscillators with fixed frequencies. It is conventional also to introduce the energies corresponding to optical branches as Θ j = hω j0 . These energies (in degrees) are of the order ¯ 2 -103 K. The quality of the Debye model is demonstrated in Fig. 6.9, where experimental of 10 DOS for NaCl is shown together with the Debye approximation.

Figure 6.9: On the quality of the Debye approximation for the density of states.

Finally, we get the following expression for the internal energy U = U0 + NT 3D where the Debye function D (z) is 3 D (z) = 3 z
z 0 3s Θ j /T Θ + ∑ Θ /T j T −1 j=4 e

(6.53)

x3 dx . ex − 1

(6.54)

At high temperatures, T Θ j (and, consequently, T Θ) we get z 1 in Eq. (6.54) and then expand the integrand in powers of x. We see that D (0) = 1. The item under the sum sign in Eq (6.53) are equal to 1, and we get for the sum 3s − 3. consequently, we get the classical expression

E = E0 + 3sNT

10: Temperature dependence of the specific heat according to the Debye model. One can see that really the border between the classical and quantum region corresponds to Figure 6. DOS behavior is much more complicated because of the real band structure effects. We have T cac =3 3N Θ 3 0 Θ/T x4 dx . T ≤ Θ/3 rather that to T ≤ Θ.54) by infinity. . The acoustic contribution to the specific heat can be expressed through the derivative of the Debye function.10. At the same time. IDEAL GAS that leads to the classical expression for the specific heat. The physical reason is strong frequency dependence of phonon DOS.80 CHAPTER 6. At low temperatures we immediately see that optical modes give exponentially small contributions and we can discard them. ex − 1 Temperature dependence of the specific heat according to the Debye model is shown in Fig. 6. The Debye model is very good for low temperatures where only long wave acoustic modes are excited. Taking into account that ∞ x3 dx π4 = x 15 0 e −1 we get π2V T 4 U = U0 + 10¯ 3 s3 h 0 that leads to the following expression for the specific heat 12π4 cV = 5 T Θ 3 . we can replace the upper limit of the integral in (6. In real life.

∂qi ∂pi ∂qi ∂pi = 0.2) (7. q) and B(p.V ) ≡ = where dΓ h 0<H <E (2π¯ )3(NA +NB +. In the classical mechanics we get Γ(E. for a stationary distribution we have { f . . dΓ (2π¯ )3(NA +NB +. N. ˙ dt ∂t ∂qi ∂pi i Because of the Hamilton equations. . x = 0 Θ(x) =  1. f [H (p. H } = 0. satisfies the Liouville equation and can be a proper stationary distribution. .. Here the Poisson brackets for the quantities A(p. h  x<0  0.. We observe that any function of the Hamiltonian H .) NA !NB ! .V ) as number of states with energies ≤ E for given N and V . this equation can be re-written as df ∂f ∂H ∂ f ∂H ∂ f ∂f = +∑ − ≡ + f.Chapter 7 Statistical ensembles We have already discussed the Liouville theorem which can be expressed as df ∂f ∂f ∂f = + ∑ qi + pi ˙ = 0. q)] (7..1 Microcanonical ensemble As an example. x>0 81 Θ[E − H (p.H dt ∂t ∂pi ∂qi ∂qi ∂pi ∂t i ∂A ∂B ∂B ∂A − . q) are defined as {A.1) Thus. B} ≡ ∑ i (7. . 1/2. Different distributions correspond to different statistical ensembles/ 7.3) . we have already considered the microcanonical ensemble. q)].. N. Let as define Γ(E.) NA !NB ! .

N. N.V ) = ∂Γ = ∂E dΓ (2π¯ )3(NA +NB +. . STATISTICAL ENSEMBLES is the Heaviside unit-step function.V ) Now we can re-derive the equipartition principle for ideal gas as follows. N. T ∂E Γ ∂E Γ(E.4) Consequently. S(E.V ) ≈ log Γ(E.31). (7. Vn ≡ where ∏ d pi Θ i=1 n E−∑ p2 i 2m i=1 3N = (2mE)3N/2 Vn . Consider the auxiliary relationship. 2 ∂H . one can easily prove that θ = 3NT .6) Vn = ∞ −∞ i ∏ dxi Θ 2 1 − ∑ xi i . . V ∝ ∑ik αik qi qk . (3. h δ[E − H (p. q)] .) NA !NB ! . we obtain 1 ∂ 1 ∂Γ D(E. D(E. ∂qn integration by parts → Using this relation. 3N N! D (2π¯ ) h D = 1 D p2 n 2m T .82 CHAPTER 7. q) = 1 δ[E − H (p. the normalized distribution function is f (p.. (7. we obtain K= = Another quantity which we shall need later is the so-called virial.. . We have to be able to calculate the volume of the hyper-sphere.5) Since the entropy of a normal system is given by Eq. The many-particle density of states is D(E.V ) (7.V ). N. N.V ) = log Γ = = . pn ∂H ∂pn dΓ ∂H p δ(E − H ) 3N N! n ∂p (2π¯ ) h n dΓ 1 ∂ = − p Θ(E − H ) → 3N N! n ∂p D (2π¯ ) h n dΓ 1 Γ = Θ(E − H ) = = T . q)] . N. θ = − ∑ qn pn = ∑ qn ˙ n n For a 3D system with harmonic potential energy. Free particles: Let us demonstrate the general principles using the gas of free particles.

etc. Ci (q.7) 2 R2 = ∑ xi .Ci (q. Comparing Eqs. (n/2)! Consequently. CANONICAL ENSEMBLES To calculate Vn let us introduce an auxiliary function ˜ Vn (R) = consider an auxiliary integral In ≡ This integral can be also written as In = 2 ˜ d Vn e−R . . (7.7) and (7.2. we recover the equation of state. 83 ∏ dxi Θ −∞ i ∞ 2 R 2 − ∑ xi i . Since {C. ∂E (2π¯ )3N (3N/2 − 1)! h 5 V + log 2 N mE 3πN h2 ¯ 3/2 S(E) = N . H } = 0 any function f [H (q.7. p)] can be a proper stationary distribution function. (7. we recover old results. p). ˜ Vn = Vn (1) . ∏ dxi e−xi −∞ i=1 ∞ n 2 = πn/2 . Then. p) = Ci = const . 3N (3N/2)! (2π¯ ) h ∂Γ VN π3N/2 = (2mE)3N/2−1 . i (7.8) we get πn/2 = nVn ∞ 0 dR Rn−1 e−R 2 → Vn = πn/2 .2 Canonical ensembles Suppose that there are other than the energy E integrals of motion.8) ˜ ˜ Since Vn = Vn Rn we have d Vn = nVn Rn−1 . Γ(E) = D(E) = VN π3N/2 (2mE)3N/2 . 7. Different choice of the system gives rise to different ensembles.

∂β Grand canonical ensemble Now let as allow an exchange by particles with the surrounding medium. q) = (∆E)2 ≡ E2 − E 2 1 −βEl e . volume. 5.V (7. 5. U = 1 ∂ log Ξ ∑ ENseβ(µN−ENs) = − ∂β .84 CHAPTER 7. Ξ Ξ= N=0 ∑ ∞ eβµN ∑ e−βENl = l N=0 ∑ eβµN ZN . we have obtained (see Sec. Moving along the lines of Gibbs’ method we have obtained the canonical distribution (Sec.2. STATISTICAL ENSEMBLES Canonical ensemble In this ensemble. s = 1 1 ∑ Es2e−βEs − Z ∑ Ese−βEs Z s s 2 2 1 ∂Z 1 ∂2 Z − = 2 Z ∂β Z ∂β = ∂ ∂β 1 ∂Z Z ∂β =− ∂U = T 2CV . and number of particles are fixed. T. Z where Z is defined by the normalization condition. The fluctuations are given by the relation f (p.1) fNl = Again.9) .2) fl = In classical mechanics. the temperature. Ns 1 β(µN−ENl ) e .q) e . Along the Gibbs method. ∞ Here we have introduced the average particle number. 1 ∞ T ∂Ξ ¯ . N ≡ ∑ N fNs = ∑ NeβµN ZN = Ξ N=0 Ξ ∂µ Ns Introducing Ω ≡ −T log Ξ we get ∂Ω ¯ N=− ∂µ . Here the temperature is fixed. Ξ Ns ¯ S = − ∑ fNs log fNs = −β(µN −U) − log Ξ . Z Z = ∑ e−βEs . 1 −βH (p. while the energy fluctuate around the average energy U.

10) → N ¯ Ξ = eN .9) we have ¯ ∂ log Ξ = (V /Λ3 )eβµ N= ∂βµ Finally.2. . fN = that is the famous Poisson distribution. using the expression for N. T ∂ρ ∂P T ∂P ∂µ = T NρKT . T Poisson distribution. Z1 eβµN = N N .V 85 The latter expression can be converted into a more convenient form introducing the gas density. CANONICAL ENSEMBLES Then is becomes obvious that Ω coincides with the Landau free energy since ¯ Ω = U − T S − µN = −PV .7. It is easy to derive the fluctuation in the particle number (check!). Find ∆N ≡ ¯ sion for Eq. (7. ∞ V Λ3 N . Derive an approximate expres¯ ¯ 1. ¯ ρ = N/V .10) at N ∆N 2 from the Poisson distribution. ¯ ZN βµN e . N! (7. ρ≡ Then we obtain. Since fN ∝ eβµN ∑l e−βENl ∝ eβµN ZN we get fN = For an ideal gas. (7. ¯ N N −N ¯ e . |N − N| N. (∆N)2 = TV ∂ρ ∂µ = TV T ¯ N = V ∂P ∂µ . Ξ= N=0 ∑ eβµN ZN = exp ∞ V βµ e Λ3 . ¯ (∆N)2 = N 2 − N 2 = −T ∂2 Ω ∂µ2 . from Eq. T. An instructive exercise is to find the distribution of the particle number in a given volume V of an ideal gas. ZN 1 ZN = 1 = N! N! Now. Ξ Ξ= N=0 ∑ eβµN ZN . Exercise.

7. Here the integral of motion is the enthalpy rather that the energy.86 CHAPTER 7. A Other ensembles Consider the canonical ensemble with a fixed pressure rather that volume. VM where ZN (V. We observe that the pressure ensemble can be considered as Laplace-transformed canonical ensemble for fixed volume.1. the so-called pressure ensemble. T ) . Since for an ideal gas ZN ∝ V N the distribution function for the volume occupied by the gas under a given pressure is ZP = fV = CV N e−βPV . ∂β . T ) is the partition function calculated for fixed volume and fixed number of particles. q) + k2 + PAh . 2π¯ h VM VM = AΛM = A 2π¯ 2 h . ZP HP = H (p. dk −βk2 /2M dV e = . Now we can integrate out the kinetic energy of a piston an replace dh → dV /A to get dh As a result. Similarly. The system is shown in Fig. dV −βPV e ZN (V.1: On canonical ensemble with a fixed pressure. T ) = dE e−βE D(E) . 2M Here k is the piston momentum while M is its mass. Then we come to the distribution function f (q. we can define ∂ ¯ H = U + PV = − log ZP . STATISTICAL ENSEMBLES h Figure 7. ZP is again the normalization factor. ∂β One can easily check that G = −T log ZP . p) = 1 −βHP e . MT ∂ ¯ V = − log ZP . ¯ where C is the normalization constant (find it and express through V !) We can apply the above considerations to microcanonical and canonical ensembles to obtain the relation ZN (V.

Sz | ↓ = − | ↓ . Still the state can be expanded in eigenfunctions as θ θ |θ = cos | ↑ + sin | ↓ . Quantizing spin along z axis. |↓ = 0 1 . we find 2 basic states |↑ = Hence. ↓ | = (0 1) . 7. 1 1 ˆ ˆ Sz | ↑ = | ↑ . 2 2 Then we easily obtain. 1 θ θ ˆ Sz = cos2 − sin2 2 2 2 = 1 cos θ . the operator Sz in this representation can be expressed as 1 ˆ Sz = 2 1 0 0 −1 . 2 2 Consequently. 2 .3. Consider now the the ensemble where the particles are quantized along different direction. 1 0 .11) Exercise. Calculate the normalization constant. To find the energy distribution in the canonical ensemble let us expand the function u(E) ≡ − log[D(E)e−βE ] = βE − S(E) up to the second order in the difference E −U and using the relation ∂2 S ∂β = = −β2 /2CV 2 ∂E ∂E we obtain the energy probability distribution as fE ∝ exp − β2 (E −U)2 . 87 (7.7.3 Ensembles in quantum statistics Pure and mixed ensembles Let us consider a particle with spin S = 1/2. ENSEMBLES IN QUANTUM STATISTICS where D(E) is the partition function for the microcanonical ensemble. 2CV (∆E)2 = T 2CV . while the conjugated wave functions are just the strings ↑ | = (1 0) .

Now we can describe the general picture for a mixed ensemble. The operator Sx in the representation used has the form 1 ˆ Sx = 2 We can easily show that 1 ˆ Sx = sin θ . For different probabilities wi this average is finite. but ≤ 1/2. STATISTICAL ENSEMBLES This state corresponds to the system polarized along the axis n = cos θ ex + sin θ ey . ˆ ˆ Sα = ∑ wi i|Sα |i = 0 . βi i αβ αβ ˆ ˆ Here Aβα ≡ β|A|α is the matrix element of the operator A. For a pure state. The ensemble average is now ˆ ˆ A = ∑ wi i|A|i .88 CHAPTER 7. ˆ Here e is the unit vector. . Then. |i = ∑ ci |α α we rewrite the average as ˆ ˆ A = ∑ wi ∑ c∗ cαi β|A|α ≡ ∑ Aβα ραβ . The maximum is reached only for a pure state where one of the probabilities is zero.12) thus the density operator can have the eigenvalues 0 or 1. S · n = 0. It can be expressed as a matrix element of the density operator. ˆ ρ = |ψ ψ| . i=↑. while ραβ = ∑ wi c∗ cαi = ∑ wi α|i i|β βi i i is called the density matrix. i Now the average can be expressed as ˆ ˆˆ A = Tr A ρ . the ensemble is called mixed. ˆ ˆ ρ2 = ρ . (7. To see the difference with a pure ensemble let us consider a system there the probabilities for the states | ↑ and | ↓ are equal. ˆ for any direction. 2 The above example where all the spins are polarized along a given axis is called a pure state. If the spins occupy different states with a given probability. ˆ ρ = ∑ wi |i i| .↓ 0 1 1 0 . i Expanding the states |i is terms of any orthonormal basis.

3. ˆ σy = 0 −i i 0 . We can ˆ ˆ ˆ easily show that Tr ρ = 1 (since Tr 1 = 2 and Tr σi = 0). P is called the degree of polarization. ˆ ρ= ∑ i i i i ∑ αi i i ˆ ˆ ˆ Since Πi Π j = δi j Πi the projection operator Πi commutes with the Hamiltonian. while for a mixed ensemble P = |P| < 1. H = 0 . Z Z = Tr e−βH = ∑ e−βEα . B] = AB − BA is the commutator which in classical physics corresponds to the Poisson brackets. ˆ σz = 1 0 0 −1 . Wi = Tr Πi . ˆ α The grand canonical ensemble is defined in a similar way. Wi As usual. Now we can construct statistics similarly to the classical one replacing distribution functions by density operators. In this way. σy . Π = |α α | . 89 1 ˆ ˆ 1+σ·P . dt ∂t h ¯ ˆ ˆ ˆˆ ˆˆ Here [A. for the system having the energy Ei 1 ˆ ˆ ˆ ρ = Πi . 1 and the Pauli matrices ˆ σx = In general. 0 1 1 0 .7. . while P is called the polarization vector. For a pure state one has to require ˆ ρ= P = cos θ ex + sin θ ey . ˆ ˆ S = −Tr ρ log ρ . σ = (σx . we can write. σz ). ENSEMBLES IN QUANTUM STATISTICS Density operator for spin 1/2 ˆ Any 2 × 2 matrix can be generally expanded in the unit matrix. in the representation of eigenstates |αi belonging to ith level the Hamiltonian can be written as ˆ ˆ ˆ H = E Π . Density matrix for thermodynamic ensembles Since i¯ h we can show that ∂ ˆ |i = H |i ∂t ˆ ˆ d ρ ∂ρ i ˆ ˆ = + ρ. Ξ For a microcanonical ensemble. for canonical ensemble we obtain ˆ ρ= ˆ 1 −βH e . Thus. 2 ˆ ˆ ˆ ˆ ˆ where 1 is the unit matrix. ˆ 1 −β(H −µN) ˆ e .

A free particle in a box with volume V : ˆ ˆ H = p2 /2m Since → ˆ r|ρ|r = 1 ∑ r|p Z p. Since e−βH = ∑ ˆ n ˆ (−βH )n (βµBσz )n =∑ = cosh(βµB) + σz sinh(βµB) n! n! n (here we employ the fact that σ2k = 1. Z 2 ˆˆ σz = Tr ρσz = tanh βµB . STATISTICAL ENSEMBLES ˆ H = −µ · B = −µBσz . V ˆ p |e−βp |p = δpp e−βp 2 /2m we get ˆ r|ρ|r = 1 Z d 3 p −βp2 /2m+(i/¯ )p·(r−r ) 1 h e = (2π¯ )3 h Z m 2πβ¯ 2 h 3/2 h e−(m/2β¯ 2 )|r−r |2 . σ2k+1 = σz ) we obtain z z ˆ ρ= ˆ 1 −βH 1 ˆ ˆ e = 1 + σz tanh(βµB) . 1 h r|p = √ e(i/¯ )p·r .90 Examples Spin 1/2 in a magnetic field: CHAPTER 7.p 2 /2m ˆ p |e−βp 2 /2m |p p|r . Λ3 ˆ r|ρ|r = 1 −π|r−r |2 /Λ2 e . From the normalization conditions we obtain Z = Tr e−βH = ˆ d 3 r r|e−βH |r = ˆ V . V . where µ is the magnetic moment.

Chapter 8 Fluctuations In this section we discuss some additional aspect regarding fluctuations. Assume that x is some quantity of the subsystem.2) 1 2π x2 exp − x2 2 x2 dx .1 The Gaussian distribution Let us consider a closed system described by a microcanonical ensemble. x=0 ∂2 S < 0. 2 γ −γx2 /2 e dx . ¯ ∂S = 0. x = 0. (8. for any quantity φ (∆φ)2 = d2φ dx2 91 x2 . ∂x2 x2 . 8. 2π (8. ∂x Thus one can expand S(x) = S(x) − γ ¯ the normalized distribution being w(x) dx = As x2 = γ−1 . At this point entropy has a maximum.1) (8. we can write the result (the so-called Gaussian distribution) in the final form w(x) dx = As x is small.3) .

∂xi ∂xk Assume for a moment that the average xi = xi0 = 0. Now let us define thermal conjugates Xi = − ∂S = γik xk . n One can prove that it is equal to δik . 1 xi Xk = δik . Thus (∆φ)2 = For statistically-independent fluctuations ˆ (γ−1 )ik = 0 . ∂φ ∂φ −1 ˆ (γ )ik . γ ≡ det{γik } . ∂xi ∑ k The averages xi Xk can be easily calculated: xi Xk = γ ∑ (2π)n/2 l √ xi γkl xl exp − 1 γ pq x p xq 2∑ pq s=1 ∏ dxs . 1 Proof: Xi Xk = γik . i The last equality determines the normalization constant A. xk0 = 0 we obtain the above result. Then ¯ xi0 = γ (2π)n/2 √ ∏ dxi xi e−(1/2) ∑st γst (x−xs0 )(x−xt0 ) . i After differentiation of this equation with respect to xx0 and putting xi0 = 0. FLUCTUATIONS w ∏ dxi = 1 . Thus the distributions are uncoupled. From that we get ˆ xi xk = (γ−1 )ik . In this way we get w(x) = γ 1 exp − ∑ γik xi xk 2 ik (2π)n/2 √ .92 For several (n) variables we have w = Ae− ∑ik γik xi xk /2 . CHAPTER 8. .

1 ∂2U ∂2U ∂2U 2 ∆U = T ∆S − P∆V + (∆S) + 2 ∆S ∆V + 2 (∆V )2 2 2 ∂S ∂S∂V ∂V we get Wmin = Thus w ∝ exp 1 ∂U ∆S ∆ 2 ∂S + ∆V ∆ V ∂U ∂V S 1 = (∆S ∆T − ∆P ∆V ) 2 ∆P ∆V − ∆S ∆T 2T .8. (8.29). V ∆V .2 Fluctuations of fundamental thermodynamic quantities As the change in the entropy is proportional to minimal work to change the thermodynamic properties reversible the probability for a given configuration can be written as.2.4) This is the general formula to find fluctuations of various thermodynamic quantities.5) CV 1 (∆T )2 + 2 2T 2T ∂P ∂V (∆V )2 T . (8. (4. FLUCTUATIONS OF THERMODYNAMIC QUANTITIES 93 8. Regarding that all the quantities are the functions of S and V and expanding U in small deviations ∆S and ∆V from the equilibrium values. w ∝ e−Wmin /T ∝ exp − ∆U − T ∆S + P ∆V T .4). . (∆V )2 = −T (∂V /∂P)T . we get w ∝ exp − Thus we reproduce previous results: ∆T ∆V = 0 . T Substituting these expressions in Eq. see Eq. (∆T )2 = T 2 /CV . Examples Independent V and T : ∆S = ∆P = ∂S ∂T ∂P ∂T ∂S ∂V V ∂P ∆T + ∂V V ∆T + ∆V = T ∂P CV ∆T + T ∂T ∆V . and below we give several examples in which different parameters are considered as independent. (8.

(8.94 Independent P and S: ∂V ∂P ∂T ∂S ∂V ∂S S ∂T ∆S + ∂P P ∆P + CHAPTER 8. N ∂P = V ∂µ T. N2 Putting V = const we obtain (∆N) This formula can be re-written as N2 − 2 V Then N ∂ N ∂P V = T. S dH = T dS +V dP we have ∂V ∂S ∆V = Thus w ∝ exp As a result. S (∆P)2 − S 1 (∆S)2 2CP .V . T. T. S From Eq. (∆P)2 = −T 1 2T = P ∂2 H = ∂P∂S ∆P + S ∂T ∂P ∂T ∂P S and ∂V ∂P ∂V ∂P ∆S . ∆S ∆P = 0 .V Here we have employed the fact that the ratio N/V depends only on P and T (additivity principle). P ∆P = S T ∂T ∆S + CP ∂P ∆P . The second equality. ∂P ∂V .5) we get: V ∆ N 2 =− T (∂V /∂P)T .V ∂P ∂µ = T. FLUCTUATIONS ∆V = ∆T = As ∆S .V ∂N ∂µ .N 2 T N2 = − 2 (∂V /∂P)T V ∂ N ∂P V ∂V ∂P ∂N ∂P =N T.V ∂N ∂P T. Fluctuations per particle: (∆S)2 = CP .N N V = T.N .

The ˙ simplest function valid for small enough x is the linear one.t |x. It is clear from obvious symmetry considerations that φ(t) = φ(−t) .t |x. t > 0. Such ¯ correlation is characterized by the average x(t)x(t ) .t) .V 95 8. Finally we arrive at a very useful formula (∆N)2 = T ∂N ∂µ . Langevin method Consider correlation between the values of the quantity x (x = 0) at different instants.6) It is natural to assume that this equation is the same as the “equation of motion” for a macroscopic quantity x. If we assume that the state under consideration is determined only by the quantity x we can suggest that the rate x for restoring the equilibrium is some function of one variable x.8. λ > 0. dx x P(x . The quantity ξx (t) is defined through the conditional probability P(x . dx/dt = −λx . Now we are prepared to discuss the state far from the equilibrium.3 Correlation of fluctuations in time. (8.7) . ξx (t) = Consequently. T.3.t) to find the the value x of the variable x under condition that at time t it was x. To express this concept in a more mathematical form let us define the quantity ξx (t) as ¯ the mean value of x at time t > 0 under condition that at t = 0 its value was x. (8. ˆ ˆ ˆ ˆ 2 Now let us consider a state of partial equilibrium characterized by a given value x. CORRELATION OF FLUCTUATIONS IN TIME follows from the formula dΩ = −V dP = −S dT − N dµ . namely assume that x x2 . A proper definition for quantum mechanical case is φ(t − t ) = 1 x(t)x(t ) + x(t )x(t) . For simplicity let us also assume that x is a classical quantity. We have. φ(t) = xξx (t) = dx w(x) x ξx (t) . Since under stationary external conditions the average depends only on the difference t − t one can define φ(t) ≡ x(0)x(t) .

The above result can be easily generalized for the case of several fluctuating quantities. φik (t) = φik (−t) = φki (t) .2). we define φik (t − t) = xi (t )xk (t) . φik (t) = φki (−t) . Consequently. Substituting this expression into Eq. To do that we find a formal solution of the above equation with some initial condition. t0 dt1 t0 dt2 eλ(t1 +t2 ) y(t1 )y(t2 ) . (8.8) It the random quantity x is comparable with the mean square fluctuation one has to add the random force y to the “equation of motion” (8.2 Because of the time-reversal symmetry. see Eq.8) for x(0)x(t) at large x.8) for x(t)x(t ) for large t we get y(t1 )y(t2 ) = (2λ/γ) δ(t1 − t2 ) . x = −λx + y(t) . The only way to obtain x(t)x(t ) dependent only on |t − t | is to suggest that y(t1 )y(t2 ) = Cδ(t1 − t2 ) . ξx (t) = xe−λt → CHAPTER 8. xk . However. Here γ is the constant which enters the Gaussian distribution. (8. 2 The system is assumed to be not in a magnetic field and not rotating as a whole. at φ(t) = γ−1 e−λ|t| . . FLUCTUATIONS t > 0.6) for the fluctuation x. there is one more symmetry property which is a consequence of the invariance with respect to time reversal. Thus we can put t0 = −∞.96 For the values ξx x2 we obtain dξx (t)/dt = −λξx (t) . ˙ The correlation function y(0)y(t) must be defined in a way to provide correct expression (8. φ(t) = x2 e−λt = γ−1 e−λt . (8. This result must be independent of t0 provided λ(t −t0 ) 1 and λ(t −t0 ) 1. x(t) = e−λt The average x(t)x(t ) is then x(t)x(t ) = e−λ(t+t ) t t0 t t dτ eλτ y(τ) . Similarly. In general. t > 0.

Now we employ the Langevin method to find the velocity correlation function vi (t1 )vi (t2 ) = v2 e−λ|t1 −t2 | .9) with random forces yi defined by the correlation properties yi (t1 )yk (t2 ) = 2ηik δ(t1 − t2 ) . However. then there is an additional “-” sign. like the quantities proportional to velocities.11) dt1 0 dt2 vi (t1 )vk (t2 ) = δik t 0 t dt1 0 2 dt2 vi (t1 )vi (t2 ) = xi (t) δik . yi = fi /m . Here we employ the fact the different components of the velocity are uncorrelated. The fluctuations xk are to be determined from the set of differential equations xi = − ∑ λik xk + yi (t) ˙ k (8. We will discuss its physical meaning 2 later. i . of only one of two changes its sign.3. It is subjected to collisions with liquid molecules. xi (t) = v2 t 2 that corresponds to ballistic motion.10) ˆ Matrix elements of η are called the kinetic coefficients. 1 ˆ ˆ λ = (ν/m) δik . i As a result. Consequently. We have xi (t)xk (t) = t 0 t (8. (8. Again. i 3 Let us now consider a displacement x(t) = r(t) − r(0) and calculate xi (t)xk (t) for t > 0. The following generalization is straightforward. D ≡ v2 λ−1 . Due to such collisions the particle’s velocity v changes in time according to the equation m vi = −νvi + fi (t) ˙ where ν is the viscous friction coefficient while fi are random forces. ˆ ˆˆ η = λγ−1 . CORRELATION OF FLUCTUATIONS IN TIME 97 There is an important remark here. We have tacitly assumed that the quantities xk are not affected directly by time reversal. 2 xi (t) t 0 t1 (8. i (8.12) t = v2 i dt1 0 dt2 e−λ(t1 −t2 ) + t2 dt2 e−λ(t2 −t1 ) = 2 v2 λ−1 t − λ−1 + e−λt . i At large times. γ−1 = v2 (0) δik = v2 δik . At small times. Simple example: Brownian motion Consider a particle moving in a liquid. ˆ φik (0) = (γ−1 )ik . these quantities can change sign under time reversal. If both xi and xk possess such a property the above equation is valid. Their physical meaning will be discussed in the next section.13) The kinetic coefficient D is called the diffusion constant. t λ−1 we obtain (in general) 2 xi (t) = 2D|t| . The scheme discussed above is called the Langevin method. We introduce the matrix λik instead of the scalar λ.8.

h. If xi are comparatively small. A convenient way to do that is to study entropy generation. An example is given ˙ as S = T ˙ i below. ∂S ˙ ˙ (8. they are just responses to generalized forces. ηik (Ω) = ηki (−Ω) . ηik = ηki . We have xi Xk = δik . one can express the entropy generation through the heat release at a given temperature. xi = − ∑ ηik Xk . FLUCTUATIONS Onsager relations The “equations of motion” xi = − ∑ λik xk ˙ k have a deep internal symmetry. l The quantities ηik are called the kinetic coefficients. The latest equation is a consequence of the fact that the coefficients are second derivatives of the entropy. Then we get.s. ∂S Xi = − . ∂xi which were introduced earlier. Q. If the system is embedded into an external magnetic field H. The proof of the principle is based on accurate application of time-reversal symmetry and the property Xi xk = δik . Xi Xk = γik . through the thermodynamically conjugated quantities. ˙ k ˆˆ ˆ ηik = ∑ λil (γ−1 )lk = (λγ−1 )ik . i i ∂xi ˙ Thus. −1 Q and then express it through the quantities x in the form (8. According to the Onsager principle.14). Then the Onsager principle reads ηik (H) = ηki (−H) . ˆ xi Xk = (γ−1 )ik .14) ˙ S = ∑ xi = − ∑ Xi xi . k γik = γki . or it rotates as a whole with the angular velocity Ω then the time-reversal operation must include reversal of H or Ω. To reveal it let is express the r. Indeed. For proper application of the Onsager principle it is important to find proper thermodynamic forces Xi conjugated to the variables xi . Xi = ∑ γik xk . .98 CHAPTER 8.

Thus. the generalized forced conjugated to the currents j and w are: Xj = −E/T . ˙ ˙ ∂S Xi = − .16) Compare equations (8. we have derived an important relation 2. S.15) we arrive at the following matrix for kinetic coefficients: ˆ η= . As a result. α. An external electric field E = −∇ϕ (where ϕ is the electrochemical potential) with combination of the temperature gradient create electric current with density j.15) Our aim is to determine the corresponding thermodynamic forces.8. one has to find the entropy production rate. −σT −ηT 2 −γT −βT 2 γ = ηT . It can be shown that Onsager relations are also valid for the inverse matrix of kinetic coefficients. Π. w − ϕj = Πj − κ∇T . γ. κ through the quantities σ.3. Xw = −∇(1/T ) = T −2 ∇T . CORRELATION OF FLUCTUATIONS IN TIME 99 Determination of proper generalized forces ˙ According to the general principle. ˜ w ≡ w − ϕj = γ E − β ∇T . div ϕj = ϕ div j + j · ∇ϕ ≈ ϕ div j = 0. and express is as − ∑i xi Xi .17) (8. . Analyze Onsager relations between the kinetic coefficients given by the equations E = ρj − α∇T . Exercise: 1. (8. Thus. 3 Indeed. η. Express the coefficients ρ. ˆ ˆˆ ˆˆ ζ ≡ (λγ−1 )−1 = γλ−1 . Since the heat release per unit volume consists of the Joule heating j · E and of absorption −div w and in the ˜ linear response approximation div w = div (w − ϕj) = div w. and we have ˙ S= ˜ dV w · ∇ 1 T + j·E ≡− T ˜ dV w · Xw + j · Xj . Xj and Xw . The currents are conventionally expressed as j = σ E − η ∇T .16) and (8. 3 we have ˜ div w j·E ˙ S=− dV + dV T T Let us assume that the total heat flux through the surface is zero. Then the first integral can be calculated by integration by parts. (8. ∂xi ˙ Let us consider an example. β and check the Onsager relations explicitly. as well as energy current w.

are quadratic functions of their arguments. FLUCTUATIONS Dissipative function The problem of dissipation cannot be solved generally within statistical physics.s. Pi = − =− ∂Qi ∂Pk ∂Qi k k Now we can construct a quadratic form. ∂Pi ∂V (Qi ) ˙ Pi = − . = T ∂Pi T ∂Pi T As a result. Now we can formally write the equations of motion which will allow for dissipative processes.100 CHAPTER 8. and the potential one. One can derive the momenta in a usual way. ηQi Pi = −ηPi Qi = −T . In general. In this way we can write ∂K(Pk ) ∂V (Qi ) ∂V (Qi ) ˙ ˙ − ∑ ηik − ∑ ηik Qk . and that the motions consist of small oscillations around some equilibrium positions. the generalized forces are XQi = XPi 1 ∂Wmin 1 ∂V Pi = =− . On the other hand. ∂K(Pi ) ˙ Qi = . However. ˙ ∂ Qi ˙ Then we can express Qi in terms of Pi . the equation for Qi should be left unchanged because it is equivalent to the definition of momenta. and by the velocities. ηik = ηki . ∂Qi Now let us consider the quantities Qi and Pi as fluctuating quantities xk . a formulation becomes possible if one assumes that the state of the system is fully determined by ˙ macroscopic coordinates. D= 1 ˙ ˙ ηik Qi Qk 2∑ ik . of the Hamilton equations any linear combination ˙ of XPi . V (Qi ). otherwise the symmetry of kinetic coefficients will be violated. T . we can add to the r. ˙ Under that assumptions both kinetic energy.h. The coefficients satisfy the Onsager relations because one of the quantities (Pi ) changes its sign under the time reversal. In the absence of dissipation we come to the usual Hamilton equations of motion. However. Qi . K(Qi ). One has also to assume that the velocities are relatively small and high powers ˙ of Qi can be neglected. divided by the temperature of the thermal bath. T ∂Qi T ∂Qi T 1 ∂K Qi 1 ∂Wmin = = . XQi which possess the proper symmetry. Since the change in the entropy is equal to the minimal work. Wmin = K(Pi ) +V (Qi ). the additional ˙ terms to the equation for Pi can contain only XPi . Pi = ˙ ∂K(Qi ) . Qi . That means that high-order derivatives can be neglected.

The proportionality coefficient is usually denoted as 2π(x2 )ω .19) Since φ(t − t) = dω dω −i(ωt+ω t ) e xω xω . Obviously. 2 −∞ (2π) ∞ and it must be a function of t − t. we have xω xω ∝ δ(ω + ω ). 2π dω 2 = φ(0) = (x )ω . The dissipative forces lead to the equation of motion ˙ ∂V (Qi ) ∂D (Qi ) ˙ Pi = − − . 2π (8. φ(t) = x2 dω 2 (x )ω e−iωt .21) . 2π (8. xω ≡ If x(t) is real. then ∗ x−ω = xω . FLUCTUATION-DISSIPATION THEOREM 101 which is called the dissipative function. ∞ −∞ ∞ −∞ x(t) eiωt dt → x(t) = xω e−iωt dω . ˙ ∂Qi ∂Qi The energy conservation law has to modified as d (K +V ) = dt ∑ i i ∂K ˙ ∂V ˙ ˙ ˙ ∂V Pi + Qi = ∑ Qi Pi + ∂Pi ∂Qi ∂Qi i ∂D = −2D .1 Classical systems Let us start with a classical system and discuss properties of Fourier transformed fluctuation. ˙ ∂Qi ˙ = − ∑ Qi Note that we applied the symmetry principle which has to be modified in an external magnetic field. For that case out approach is strictly speaking not valid.4.20) (8. so xω xω = 2π(x2 )ω δ(ω + ω ) . 8.8.18) (8.4 Spectral properties of fluctuations and fluctuation-dissipation theorem 8.4.

XQ = T −1 ∂Wmin /∂Q = T −1 ∂U/∂Q = mω2 Q/T . Example: Fluctuations in one-dimensional oscillator.8) for φ(t) we obtain 4 2λ (x2 )ω = .24) (8. Since the minimal work against the “spring” is equal to the potential energy.23) If friction is present. ⇒ Q2 = T /mω2 . 0 (8. They are determined by the entropy generation. . For the correlation function of the Fourier components we obtain: yiω ykω = 2π(yi yk )ω δ(ω + ω ) . Consider one-dimensional oscillator which is at rest is in in the equilibrium position Q = 0. the equations of motion acquire the form ˙ Q = P/m . γ(ω2 + λ2 ) We can introduce also transform of random forces. Again. FLUCTUATIONS The quantity (x2 )ω is called the spectral density of fluctuations.102 CHAPTER 8. (8.22) . We write the potential energy as U = (1/2)mω2 Q2 where m is the “mass” while ω0 is the oscillator eigenfrequency. Wmin = (1/2)mω2 Q2 .25) ˙ where −γP/m = −γQ is the friction force. 0 Consequently.e. but able to perform small oscillations in some macroscopic coordinate Q. (8. i. 0 ˙ The generalized moment is then P = mQ. (8.26) (8. From the known expression (8.27) 0 −T T γT . Now let us take Q and P as random quantities and determine the conjugated thermodynamic forces. 0 XP = T −1 ∂Wmin /∂P = T −1 ∂K/∂Q = P/m . ˙ P = −mω2 Q − γP/m .27) we get the η-matrix has the form ˆ η= 4 Note (8. We have Wmin = P2 /2m + mω2 Q2 /2 . (y2 )ω = (ω2 + λ2 )(x2 )ω = 2λ/γ . 0 f (Q) ∝ e−Wmin /T = e−mω0 Q 2 2 /2T (yi yk )ω = ηik + ηki . ˆ Here γ is the matrix of kinetic coefficients. 0 (8.25) and (8. yω . XQ and XP . by the minimum work to bring the system from the equilibrium to the state with given Q and P.28) that according to general principles of quasistationary fluctuations this expression is valid for ω less that the inverse time for partial equilibrium to be established. the results can be expressed in the form of Langevin approach. ˆ Combining Eqs.

8. (8.32) (8. According to Eq. FLUCTUATION-DISSIPATION THEOREM In order to apply these equations for fluctuations we rewrite Eq. (8.24) is just a definition of the momentum must remain unchanged. ψ∗ (q) ≡ n| . 0 Multiplying by e−ωt and integrating over time.33) where H is the Hamiltonian.25) as ˙ P = −mω2 Q − γP/m + y .31) 8.34) i.35) (8. (8. Since eigenfunctions of the Schr¨ dinder equation form a complete set. First. (8. o i¯ (∂Ψ/∂t) = H Ψ h (8. n For a stationary state. Revisiting quantum mechanics According to quantum mechanics. we find (−mω2 − iωγ + mω2 )Qω = yω . (y2 )ω = 2η22 = 2γT . (8. 0 103 (8. (8.22). eigenstates of the Hamiltonian are called the stationary states.4.22). The states ψn which satisfy the stationary Schr¨ dinder equation. obtaining ¨ ˙ mQ + γQ + mω2 Q = y .36) .29) Equation (8. 0 and finally.2 Quantum systems Now let us turn to a quantum system.30) ˙ To calculate the required fluctuation spectrum (Q2 )ω we substitute P = mQ in (8.e.4. o dq ψ∗ (q)ψn (q) = 1 . h Ψn (t) = e−(i/¯ )Ent ψn (q) . the state of the system can be described by a time-dependent wave function Ψ(t) satisfying the Schr¨ dinder equation. we have to remind some basic concepts of quantum mechanics to be able consider time-dependent quantities. ψn (q) ≡ |n . (Q2 )ω = 2γT m2 (ω2 − ω2 )2 + γ2 ω2 0 . We will also use Dirac notations. o H ψn (q) = En ψn (q) .

45) → α (ω) = α (−ω) .3 The generalized susceptibility Let us now assume that the body is subjected to some perturbation. n n (8.4.42) (8.39) Now. Since x is real ¯ α(ω) = α∗ (−ω) . Then using Eq. . let us assume that the operator fˆ does not contain time explicitly. n (8.43) (8. α (ω) = −α (−ω) .46) 1 α(ω) f e−iωt + α(−ω) f ∗ eiωt .104 In general.44) Consequently.37) The mean value of any physical quantity. ˆ V (t) = − f (t) x . h ¯ (8. ¯ dt ∂t dt (8. fnm (t) = fnm eiωnmt . n ωnm ≡ En − Em . where x(ω) = α(ω) f (ω) . ˆ 2 and x(ω) = ¯ Then in general the dissipation is ˆ df dE ∂H = = −x . FLUCTUATIONS h Ψ(t) = ∑ an Ψn (t) = ∑ an e−(i/¯ )Ent ψn (q) . (8. 2 (8.41) 8. 1 ˆ V = − f e−iωt + f ∗ eiωt x . α(ω) ≡ α (ω) + iα (ω) is called the linear response function.36) we have. represented by the operator fˆ is then f¯(t) = ∑ a∗ am fnm (t) n nm (8.40) where fnm ≡ n| fˆ|m = dq ψ∗ fˆψm (q) . For a periodic perturbation.38) where fnm (t) = Ψ∗ (t) fˆΨm (t) . (8. ˆ Then the average value x(t) = ¯ 0 ∞ α(τ) f (t − τ) dτ . CHAPTER 8. ¯ α(ω) = ∞ 0 dt α(t) eiωt . (8. f¯.

8.4. FLUCTUATION-DISSIPATION THEOREM
Substituting x from Eq. (8.45), ¯ f (t) = and averaging over time we obtain Q= dE dt 1 1 = (α∗ − α)| f |2 = | f |2 ωα (ω) . 4 2 1 f e−iωt + f ∗ eiωt 2

105

(8.47)

(8.48)

t

Analytical properties of generalized susceptibility. Kramers-Kronig relations Let us regard ω as a complex variable, ω = ω + iω and consider the properties of α(ω) in the upper half of the ω-plane. From the definition (8.42) it follows that α(t) is finite at any positive t. Consequently, α(ω) is a one-valued regular function in the upper half of the ω-plane. Indeed, for ω > 0 the integrand of Eq. (8.44) contains the factor e−ω t and since α(t) is finite the integral converges. The function α(ω) has no singularities at the real axis (ω = 0), except possible at the origin. Note that the definition (8.44) is not valid for the lower half of the ω-plane since the integral diverges at ω < 0. The conclusion that α(ω) is regular in the upper half-plane is the consequence of the causality principle. It is because of this principle the integration over time in Eq. (8.42) is taken only over times previous to t. It is also evident from the definition (8.44) that α(−ω∗ ) = α∗ (ω) . (8.49) Hence for a pure imaginary frequency, ω = iω , we have α(iω ) = α∗ (iω ), i.e. α(ω) is real. The following theorem can be proved: The function α(ω) does not take real numbers at any finite point of the upper half-plane except on the imaginary axis, where it decreases monotonously from a positive value α0 > 0 at ω = i0 to zero at ω = i∞. In particular, it follows that α(ω) has no zeros in the upper half-plane. The proof is given in the book [1], §123. Let us now derive a formula relating the real and imaginary part of α(ω). To do so, we choose some real positive value ω0 of ω, integrate the expression α(ω)/(ω − ω0 ) round the contour C shown in Fig. 8.1, and then tend the outer contour radius R to infinity and the radii ρ of the inner circles to zero.
C R ω ρ 0 ωo

Figure 8.1: On the derivation of analytical properties of the susceptibility.

Since at infinity α → 0, and the ratio α(ω)/(ω − ω0 ) tends to zero more rapidly than ω−1 . Since the function α(ω) is regular and possible divergence points ω = 0 and ω = ω0 are excluded, the result must vanish, α(ω) dω = 0. ω − ω0 C

106

CHAPTER 8. FLUCTUATIONS

The integral over the infinite semicircle is also zero, so we are left with the integral over the real axis and the parts around the ω = 0 and ω = ω0 . The first part in not important since α(0) is finite. The integration around ω0 yields −iπα(ω0 ) since the direction is clockwise. Consequently,
ω0 −ρ ρ→0

lim

−∞

α(ω) + ω − ω0

∞ ω0 +ρ

α(ω) ω − ω0

− iπα(ω0 ) = 0 .

The first item is just the principle value of the integral, so we obtain iπα(ω0 ) = P
∞ −∞

α(ω) . ω − ω0

(8.50)

Here the variable of integration, ω has only real values. For convenience we replace the notations as ω → ξ , ω0 → ω , to obtain the famous Kronig-Kramers relations (1927): 1 P π 1 α (ω) = − P π α (ω) = α (ξ) , ξ−ω −∞ ∞ α (ξ) dξ . ξ−ω −∞ dξ

(8.51) (8.52)

Since α (ω) is the odd function, Eq. (8.51) can be rewritten as α (ω) = 2 P π


0

ξ α (ξ) . ξ2 − ω2

(8.53)

If the function α(ω) has a pole at ω = 0, near which α(ω) = iA/ω, then we get instead of Eq. (8.52): ∞ 1 α (ξ) A α (ω) = − P dξ + . (8.54) π ξ−ω ω −∞ The Kramers-Kronig relations are very important since they allow to reconstruct the whole response function from its real or imaginary part. This is a consequence of the causality principle. Quantum expression for susceptibility On the other hand,the probability for the transition n → m according to the Fermi golden rule is wmn = 2π | f |2 |xmn |2 [δ(ω + ωnm ) + δ(ω + ωmn )] . h 4¯ ¯ h (8.55)

The total absorbed power due to all transitions from the state n is then ¯ Qn = ∑ hωmn wmn =
m

πω| f |2 |xnm |2 [δ(ω + ωnm ) − δ(ω + ωmn )] . 2¯ ∑ h m

(8.56)

8.4. FLUCTUATION-DISSIPATION THEOREM
Comparing Eqs. (8.48) and (8.56) we write down the response function5 α for n-th state as αn (ω) = π |xnm |2 [δ(ω + ωnm ) − δ(ω + ωmn )] . h∑ ¯ m

107

(8.57)

This quantity should be averaged over the Gibbs distribution of the initial states occupations, fn = Z −1 e−βEn , as ˆˆ α (ω) = Tr ρα = ∑ fn αn (ω) (8.58)
n

ˆ where fn are diagonal elements of the density matrix ρ = Z −1 eβH , see Eq. (7.12). We have α (ω) = Immediately, we get α (ω) = or, in the complex form, α(ω) = 1 h ¯ 1 h ¯ π ∑( fn − fm) |xnm|2 δ(ω + ωnm) . h nm ¯ (8.59)

∑ |xnm|2 ωmn − ω ,
nm

fn − fm

(8.60)

∑ |xnm|2 ωmn − ω − i0 ,
nm

fn − fm

(8.61)

8.4.4 Fluctuation-dissipation theorem: Proof
According to quantum mechanics, one has to define a real quantity which is an average of a Hermitian operator, 1 1 {xω , xω } ≡ {xω xω + xω xω } . ˆ ˆ ˆ ˆ ˆ ˆ 2 2 Obviously, the correlation function must depend only on the time difference, and one can write down the spectral correlator as 1 xω xω + xω xω ≡ 2π(x2 )ω δ(ω + ω ) . ˆ ˆ ˆ ˆ 2 (8.62)

This relation can be regarded as a definition for (x2 )ω . Using the definition, as well as the fact that (x2 )ω is real and is an even function of ω we get φ(t) = In particular, φ(0) = x2 =

dω 2 −iωt x )ω e −∞ 2π

2 (xω =

∞ −∞

dt φ(t) eiωt .

dω 2 (x )ω . −∞ 2π

is the imaginary part of the susceptibility.

say n. m (8. h h As a result.65).108 CHAPTER 8. nm (8.64) Now. 2∑ m ∞ −∞ Here (xω )nm is the Fourier transform of the nm matrix element of the operator x(t). fn = Z −1 e−βEn . Averaging. we obtain (x2 )ω = π ∑ fn |xnm |2 [δ(ω + ωnm ) + δ(ω + ωmn )] nm h = π ∑( fn + fm )|xnm |2 δ(ω + ωnm ) = π ∑ fn 1 + eβ¯ ωnm |xnm |2 δ(ω + ωnm ) nm nm h = π 1 + e−β¯ ω ∑ fn |xnm|2 δ(ω + ωnm) . instead of (8. Let us consider a quantum body in a given state. nm (8.65) Comparing Eqs. ω (8. the expression for the fluctuations should be averaged over the Gibbs distribution of the initial state occupations.63) Again.67) . ˆ (xω )nm = ∞ −∞ dt xnm (t) eiωt = dt xnm ei(ωnm +ω)t = 2π xnm δ(ωnm − ω) . the average value to the correlator is its diagonal matrix element. when hω T . FLUCTUATIONS The aim of the present section is to relate the spectral correlation function (x2 )ω to the dissipation in the system. 1 xω xω + xω xω ˆ ˆ ˆ ˆ 2 nn = 1 [(xω )nm (xω )mn + (xω )nm (xω )mn ] . For small frequencies. ∗ Since x is real xnm = xmn and δ(ωnm + ω)δ(ωmn + ω ) = δ(ωnm + ω)δ(ω + ω ) we can extract 2π δ(ω + ω ) at get: (x2 )ω = π ∑ |xnm |2 [δ(ω + ωnm ) + δ(ω + ωmn )] . (8. (8. Then. we obtain (x2 )ω = hα (ω) coth(β¯ ω/2) ¯ h → x2 = h ¯ π α (ω) coth(β¯ ω/2) dω . let us note that Eq.64) and (8.59) can be rewritten in the form α (ω) = π h 1 − e−β¯ ω h ¯ ∑ fn |xnm|2 δ(ω + ωnm) . 1951).66) This is the famous fluctuation-dissipation theorem (Callen & Welton.66) we get (x2 )ω = T α (ω) . h (8. one can approximate coth(β¯ ω/2) as ¯ h coth(β¯ ω/2) ≈ 2T /¯ ω .

. 102) given by Eq. (I 2 )ω . Exercise: 1) Find spectrum of fluctuations of the current I through the circuit shown in Fig. In this case xi = ji . (I 2 )ω (dω/2π) and compare it with the square of the average current. Note that all the concept is valid only for small fluctuations near the equilibrium state.2. An important example is fluctuations of current density in a resistor.4. (8. The circuit is kept under constant voltage V12 . where σ(ω) is the complex conductivity.32) obey the classical limit (8. FLUCTUATION-DISSIPATION THEOREM 109 Exercise: Check that fluctuations in the one-dimensional classical oscillator (example in p. L is the inductance. for a fixed voltage V12 between the points 1 and 2) Calculate the integrated spectrum. while R is the resistance. Thus ( j2 )ω = hω ℜ σ(ω) coth ¯ hω ¯ 2T . I 2 .2: On the current fluctuations.8.66) of the fluctuation-dissipation theorem. α = ω ℜ σ(ω). 1 R 2 L C Figure 8. C is the capacitance. 8.

110 CHAPTER 8. FLUCTUATIONS .

Certainly R=0 ∑ PN (R) = (p + q)N = 1 . R!(N − R)! (9. In general. It can vary between +N and −N. 9.1 Random walks The simplest stochastic process is random walk in one dimension. 111 N . We are interested to find the displacement S after N 1 steps.1: On the 1D random walk. If there occurs 1 step to the left the displacement is S = N − 2.1). A particle can make a step to the right with the probability p or to the left with probability q = 1 − p.1) This distribution is called binomial or Bernoulli. walk with R steps to the right from total N steps is PN (R) = N! pR qN−R . the probability to find a Figure 9. The probability of the walk with S = +N is PN (N) = pN .Chapter 9 Stochastic Processes Now we discuss other aspects of random processes which take place in course of evolution of nonequilibrium states to equilibrium. 9. the probability being PN (N − 1) = N pN−1 q because the step to the left can occur at any of the N different times (see Fig.

2πN pq D≡ a2 .t) = 2 2 2 1 τ e−a τ/2a t = √ e−x /4Dt . Thus (∆S)2 = S2 − S 2 p ∂ ∂p 2 (p + q)N = (N p)2 + N pq = 4N pq → =N (for p = q = 1/2) . If the number of steps is very large then we can think that each step is very small and come to continuous description. STOCHASTIC PROCESSES As a result. we can easily calculate the average displacement S = R − (N − R ) = 2 R − N N! = −N + 2 ∑ RpR qN−R R!(N − R)! R = −N + 2p = −N + 2 ∂ N! ∑ R!(N − R)! pRqN−R ∂p R ∂ (p + q)N = N(p − q) . To obtain the statistical average one has to make M 1 runs. Using the Stirling’s formula for n! we obtain PN (R) = √ Since R = (N + S)/2 we have PN (S) = √ Here ¯ S = N(p − q) = N(2p − 1) . This set forms a statistical ensemble. one needs a random number generator. Then we obtain for the symmetric case the probability distribution P(x. 2τ . σ= (∆S)2 = 4N p(1 − p) . M m=1 In general case the way is more difficult. One makes N attempts.112 CHAPTER 9. One can also obtain R2 = to obtain S2 = 4 R2 − 4N R + N 2 = N 2 (p − q)2 + 4N pq . Assuming x = Sa and regarding the time step t = Nτ. 2t 2πa 4πDt 2 2 1 1 ¯2 e−[S−N(2p−1)] /8N p(1−p) = √ e−(S−S) /2σ . For very large N the Bernoulli distribution PN (R) = (N!/R!(N − R)!)pR (1 − p)N−R can be approximated by the Gaussian distribution. 8πN pq 2πσ 2 1 e−(R−N p) /2N pq . the average being 1 M S = ∑ Sm . ∂p The symmetric random walk can be generated by tossing a coin.

9. It is driven by gradients. i 2 1 e−r /4Dt .t) = We have xi (t)x j (t) = 2δi j D|t| . 0) = N δ(r) . 6τ 9. x = 0. (9. . J = −D gradC. Now we can map the above results on the results on Brownian motion obtained in Sec. The above expressions can be easily generalized for multidimensional space keeping in mind that the probability is multiplicative.2 0. The concentration of dye obeys the continuity equation. For 3D case we obtain.5 1 0. D = v2 λ ∼ a2 /λ . ∂C + div J = 0 ∂t where J is the current of colored particles. Thus.t) as a function of x at Dt = 1 and 3. P(r.2) ∂t Let us consider the initial condition C(r.9. Indeed.1 Relation to diffusion Consider a drop of colored dye in the water.4 0.1 3 –3 –2 –1 1 2 3 x This function is shown in Fig. In this way we get the diffusion equation ∂C − D∇2C = 0 . RANDOM WALKS 113 0.2: P(x.3 Figure 9.1.t) = 1. 3/2 (4πDt) D= a2 . 8.1. x2 = 2Dt .3 by the Langevin method. we can assume that during the time λ−1 between collisions the displacement is a ∼ v/λ. 0.2 We immediately obtain ∞ −∞ P(x. r2 = 3 x2 = 6DT .

while the r.h. kr Then. Consider a thin layer between x and x + ∆x. .t). Let the number of particles in this layer at time t will be denoted as N(x.t) = Ne−Dk 2t To derive this formula it is convenient to use spherical coordinates and to direct the polar axis along r. (9.t)eik·r . Einstein in 1905. The integral over the angles is π 0 sin θ dθ 2π 0 dφe−kr cos θ = 2π · 2 0 1 dµ cos krµ = 4π sin kr . (2π)3 and c(k. If we introduce the probability P(a.h. The solution with the above initial condition can be easily obtained by the Fourier method.s.t) = we obtain C(r.t + τ) = P(a.s.t + τ) = ∞ −∞ da P(a. Using the equalities da P(a. τ) = 1 . Since C(r.t) = N 2 d 3 k −Dk2t+ik·r N e−r /4Dt . Since C = ∆N/∆x we get an integral equation: C(x. we can write ∆N(x. to move a distance falling within a small interval between a and a + ∆a during a short time interval τ. and the definition 1 2τ da a2 P(a. 4π 8π3 0 ∞ k2 dk 2 sin kr −Dk2t 1 e−r /4Dt .t) .3) Since a and τ are small one can expand the l. τ) = x2 (τ) /2τ ≡ D . STOCHASTIC PROCESSES Here N is the total number of particles in the drop. in powers of a. da aP(a. e = 3/2 kr (4πDt) The relation between diffusion and random walk has been realized by A. τ) ∆a. e = 3 3/2 (2π) (4πDt) d3k c(k. in powers of τ. τ) = 0 (we assume symmetric distribution) we arrive at the same 1D diffusion equation as for the Brownian motion (with correct coefficients).114 CHAPTER 9. τ) ∆a ∆N(x − a) . τ)C(x − a. His reasoning is like that.

Fig. and the only random quantity is the instant at which a pulse starts. In the simplest case the pulses have the same shape.2. For our case φxk (u) = da wa (a) 1 tm tm /2 −tm /2 dt exp[iuF(t − tk . a)] . or pulsed random processes are encountered very often. dt i(t) = e . If n pulses took place during the time tm we obtain φx (u|n) = [φxk (u)]n . RANDOM PULSES AND SHOT NOISE F(t) 115 Figure 9. 9. Then the random quantity can be expressed as x(t) = ∑ F(t − tk .characteristic function of the random quantity. ak ) . e.2 Random pulses and shot noise Random pulses.9. g. tk being its starting time. The random quantities tk and ak are statistically independent for different k. while tm is the measuring time interval.3: On the random pulse process. The probability of the time tk being in the range between t and t + dt is constant and equal to ν dt. There are many examples of this type of process. Let us denote the number of pulse as k. t 9. 2. Below we shall use the above example to introduce the important concept . Here wa (a) is the distribution of the parameters a. k It is usually assume that 1.3. One well known example is the current in the anode circuit of the thermionic tube. impacts of the gas molecules on the wall of the container. Each electron escapes the cathode and moves in the space charge region toward the anode producing anode current i(t). Here p(x) is the probability distribution for x while the average is performed over realizations of the random process. For a quantity x it is defined as φx (u) ≡ eiux = ∞ −∞ dx p(x) eiux . and their distribution functions are k-independent.

t2 . . . . We can characterize the random process by the probability wn (x1 .3 Markov processes Many stochastic processes belong to the class of Markov.tn−1 ) ×Pn (xn . . The noise is conventionally measured as doubled Fourier component of the correlation function (9. . x(t) which is known at n instants t1 < t2 < . STOCHASTIC PROCESSES which then must be averaged over the Poisson distribution. To introduce a definition let us consider a random function of time. . . x2 . . xn .tn ) = wn−1 (x1 . . xn−1 .t1 . Let the time intervals be much greater than the microscopic time scale. nn −n ¯ Pn = e ¯ .5) (9.116 CHAPTER 9. Since F(0. x2 . xn .t2 . . n! Since e −n ¯ n = νtm .4) dt F 2 (t. . Having the characteristic function we can calculate any moment of x: x = ∂φx (u) ∂iu =ν u=0 2 da wa (a) ∂2 ln φx (u) ∂(iu)2 ∞ −∞ ∞ −∞ dt F(t. .tn |x1 . x2 .tn ) dx1 · · · dxn that the random quantity at ti lies within the range (xi . . < tn . x2 . xn−1 . or Markovian processes.tn−1 ) . x2 . we get a very universal formula Sx (0) = 2q2 ν → 2eI where I is the electric current in the case of thermionic tube (Schottky. . a) .t1 . a) . . a) ≡ q where q is the charge of the particle. (9. a) .t1 . . . . ¯ [nφxk (u)]n ¯ ¯ ∑ n! = exp {n[φxk (u) − 1]} = exp {νtm[φxk (u) − 1]} n ∞ −∞ we obtain for the well separated pulses φx (u) = exp ν da wa (a) dt [exp(iuF(t.tn |x1 . a)|2 . . a)) − 1] . .6). 1918). xn−1 .t2 .t1 . We can now relate wn to wn−1 by the conditional probability Pn (xn . We obtain (check!): Sx (ω) = 2ν da wa (a) |F(ω. 9. .t1 .t2 . .t2 . .tn−1 ) as wn (x1 . =ν da wa (a) ∞ −∞ (9. a)F(t + t1 .6) (δx)2 = x2 − x δx(t1 )δx(0) = ν = u=0 da wa (a) dt F(t. xi + dxi ).

9) 9. Both random walks and pulse processes are the Markov ones. (9. (9.tn |xn−1 .t|x j . p12 = p. θ)P(x j . Master equation There is an important class of the processes where the random quantity acquires only discrete values xi . xi ) . . θ|xi . dt The stationary solution of this equation is w0 = q . j t0 < θ < t . x2 .tn−1 ) . Let us consider a Markov process where the random variable ξ(t) acquires discrete value {xk }.8) This is the Markov equation.t0 ) . 9.t2 . MASTER EQUATION 117 The process is called the Markov one if the conditional probability depends only on the value xn−1 at the last instant tn−1 which precedes tn .t0 ) = ∑ P(xk . .tn−1 ) = P2 (xn . . (i = 1. j l < m < n. p+q (9. p21 = q. n − m)P(x j |m − l. Since w2 = 1 − w1 = 1 − w we can write the master equation for the probabilities in the form dw = −p w + q (1 − w) . w1 = w.4.tn−1 ) ≡ P(xn . xn−1 . x1 x2 x2 Time In the discrete case one can introduce the probability wi to find the system in the i-th state. Pn (xn .4: Random telegraph noise in the defect position.tn |x1 . . see Fig. . 2. Then the conditional probability can be uncoupled as P(xk .t1 . DISCRETE MARKOV PROCESSES. x x1 Figure 9. as well as transition probability pik .tn |xn−1 . .9. p+q w0 = 1 − w0 = 2 q .7) It means that the system forgets all but the last previous values of the random process.4 Discrete Markov processes. . . N). 0) = ∑ P(xk |x j .t|xi . For two-state system we can denote. An example is the random telegraph noise produced by a defect hopping between two states in the lattice. (9. For a homogeneous random quantity ξ(t) depending in discrete times tn the above equation acquires the form P(xk |n − l.4.10) .

10) we get λ = p + q.t1 ) is the probability It is natural to look for a non-stationary solution in the form g(t.t1 ) satisfies Eq. (9. (p + q)2 Let us for simplicity assume that p = q = λ/2. STOCHASTIC PROCESSES (δx)2 = pq(x2 − x1 )2 . P(x.t1 ). (9.t1 ) = e−(p+q)|t−t1 | .t|x1 .t1 ) = w(x) + [δ(x − x1 ) − w(x)]g(t. and for realistic values of ω the inequalities λ−1 max ω λ−1 min (9. Usually logarithm of λ is distributed almost uniformly in a wide region between λmin and λmax .11) The physical meaning of g(t.t1 )|t=t1 +0 = 1 . and from the initial condition (9. Since at t − t1 (p + q)−1 we have x(t)x(t1 ) = x 2 .t1 ) is the probability that the state remains unchanged during the time between t1 and t ≥ t1 . 2 + ω2 2π λ In many random systems the fluctuation are induced by structural defects with different transition rates λ. Substituting that into the master equation (9. Obviously. we obtain δx(t)δx(0) = (δx)2 e−(p+q)|t| = Consequently the noise (doubled Fourier transform) is S(ω) = (x1 − x2 )2 2pq p+q 1 · .118 Consequently. (p + q)2 Thus noise spectrum is a Lorenzian. The quantity g(t. x = qx1 + px2 . p+q CHAPTER 9. while g(t.11) g0 = 1.10) with the initial condition g(t. (p + q)2 We can also obtain the correlation function δx(t)δx(0) .t|x1 . 2 π (p + q)2 + ω2 (p + q) pq(x1 − x2 )2 −(p+q)|t| e . or (δx)2 = 0. at t ≥ t1 it has the form P(x. To calculate the correlation function one has to find conditional probability.12) .t1 ) where w(x) is the stationary probability to find the system in the state x. The integral noise is ∞ −∞ dω S(ω) = (x1 − x2 )2 2pq = 2 (δx)2 . g(t.t1 ) = g0 e−λ|t−t1 | . than S(ω) = (x1 − x2 )2 λ .

Fokker-Planck equation Let us return to the problem of random walks and note that Eq.t|y. CONTINUOUS MARKOV PROCESSES.t − τ) (x − y)2 = ··· .t|y.t0 ) = = ∞ −∞ ∞ −∞ dy P(y.t0 ) ∞ −∞ ∞ −∞ dx q(x)P(x. We get ∞ −∞ x → ±∞ dx q(x)P(x. Crossovers from random telegraph noise and flicker noise were observed in point contacts between metals. Then the distribution of λ is w(λ) dλ = nd dλ . 9.12).t0 ) dx q(x)P(x. We get P(x. We get P(x.14) is usually observed at low-frequencies.t − τ) = 1 .t|x0 .t − τ) dy P(y. Namely.13). Since logarithm the tunneling probability is proportional to the tunneling distance and the latter is uniformly distributed we arrive at the distribution (9. let us put θ = t − τ and then tend τ to zero.t|x0 .13) is so-called tunneling defects formed by an atom which can tunnel between two adjacent states in the material.t|y. Averaging the noise over the distribution (9.t − τ)P(y. (9. L λ 119 L ≡ ln(λmax /λmin ) . An example leading to the distribution (9.t|y. (9.t − τ|x0 .t − τ|x0 .3) can be reformulated for probabilities.5.t0 ) .t − τ|x0 . (9.t0 ) . θ|x0 .5 Continuous Markov processes. Noise with the spectrum (9. (9.13) we get S(ω) = nd (x1 − x2 )2 1 2L ω (9.t0 ) = dy P(x. t0 < θ < t .t0 ) = dy P(x.14) provided ω satisfies inequalities (9. 2 × q(y) + q (y)(x − y) + q (y) Since ∞ −∞ dx P(x.9.t|y. θ)P(y.15) This is the Smoluchowski equation. q → 0 at and then integrate over x. so let us make it close to t. it is called the flicker noise. The instant θ can be arbitrary.13) Here nd is the density of the defects.t|x0 . FOKKER-PLANCK EQUATION hold.16) Now let us multiply this equation by some arbitrary function q(x) which has the properties q(x) → 0.

t0 ) − P(x. τ=0 τ=0 τ −∞ τ A(y.22) We see that Eq.t0 ) q (y)A(y. we have ∞ −∞ dx q(x) ∂P(x. (9. Assume that they move independently.t0 ).t|x0 .s. ∂t ∂x 2 ∂x2 (9.t|y. (9.t|x0 .t0 )] ≈ . so it has different names: Einstein-Fokker equation.17) Here x−y 1 ∞ = lim dx (x − y)P(x.20) is the following.s. The diffusion equation (9. of Eq. 2 (9. τ ∂τ As a result.t0 ) ∂ 1 ∂2 = − A(x. (9. Let a stream (ensemble) of particles emerge form the point x0 at time t0 .t|y.t0 ) . The solution of Eq. Then we notice that 1 ∂P(x.t) = lim (9. . S = AP− 1 ∂BP .h.h.t|x0 . where A is the convection velocity. Then their concentration at the point x at time t is just P(x.t|x0 .t0 ) .20) This equation was derived in several contexts. (9. P(x. and of the diffusion part −(1/2)(∂BP/∂x) where B/2 has a meaning of the diffusion coefficient.22) is conventionally written as S = A− ∂B B ∂P(x. STOCHASTIC PROCESSES we can replace in this integral y → x move it into l. τ=0 τ=0 τ −∞ τ (x − y)2 1 ∞ B(y.t|x0 .t) + q (y) B(y. Fokker-Planck equation. ∂t ∂x Equation (9.t|x0 .18) (9. 2 ∂x (9. The the flow (mass current) S consists of the “convection” part AP. ∂P ∂S + = 0. Since q(±∞) = q (±∞) = 0 and q(x) is arbitrary we finally obtain ∂P(x.t − τ|x0 .t0 |x0 .t|x0 .t)P(x.t|x0 .20) must be non-negative and satisfy the initial condition P(x.23) The combination A − ∂B/∂x is called the drift velocity while B/2 is the diffusion constant.2) is a special case with constant A and B.19) Now we replace y → x in the r.t0 ) = δ(x − x0 ) .t)P(x.t − τ) .20) is equivalent to the particle conservation.t) +··· .t|x0 .t|x0 .21) The physical interpretation of Eq.120 CHAPTER 9.t0 ) + B(x.t0 ) − ∂x 2 ∂x (9. 2nd Kolmogorov equation.t0 ) = ∂τ ∞ −∞ dy P(y.17) and make integration by parts. (9.t) = lim = lim dx (x − y)2 P(x.t0 ) [P(x.t − τ) .

6. and the nonstationary part of the current is given by the expression Vg2 Ia = SVg 1 − 2 (9.27) can be rewritten as 4 x + x = µx p − x 2 ¨ ˙ ˙ x . τ ≡ ω0t.27) Now. Then Eq.26) (9. (9. see Fig.28) . I0 ≡ 2V0 /ω0 M.25) where M is the mutual inductance while prime means here the time derivative. the notations are clear from the picture. such that Vg = MI (9. FLUCTUATIONS IN A VACUUM-TUBE GENERATOR 121 9.9. (9. ω0 (MS − RC) ≡ p .it is convenient to introduce dimensionless variables.5. we have the following equation for the current through inductor L. ω0 MS ≡ µ. x ≡ I/I0 . 9.6 Fluctuations in a vacuum-tube generator Consider a simplest generator of electric oscillations based on 3-electrode vacuum tube. Since the charge at the capacitor C is just t ˜ Q= dt (Ia − I). LI + RI = C−1 R 1 SM I + I + I = I L LC LC t dt [Ia (t ) − I(t )] 1− M 2 (I )2 3V02 . where ω0 = (LC)−1 is the eigenfrequency of the circuit.5: On the vacuum tube generator. I a I Figure 9. The transformer facilitates feedback. → (9. L V g M R C The tube acts as an amplifier.24) 2Va where V0 is the DC anode voltage.

of this equation is proportional to µ we have to substitute the derivatives up to the lowest order. of Eq. τ0 ). (9. It can be shown that at √ p > 0 the only stable point r = p. ˙ −2˙u sin u − 2rϕ cos u = −µr(p − r2 ) sin u − µ r˙ r2 sin 3u .32) 2 So up to our approximation.s.6 the function r(p − r2 ) is plotted for p = −1. the solution of Eq.28) is ˙ u2 ≈ 1 + 2ϕ. In a straightforward way we get ˆ vector A and tensor B as . when we have several variables. 3 (9. r = 0. ϕ. In general. Then x = r cos u − ru sin u . u(τ) ≡ τ + ϕ(τ) . What does it mean will be clear a bit later. ˙ (9. In is clear that at p ≤ 0 there is only one stable point. the conditional probability can be expressed as a function of vectors x = {xα } and x0 = {x0α }. ˙ u = 1 + ϕ.122 CHAPTER 9.29) assuming the both r(τ) and ϕ(τ) are “slow” functions. we multiply these equation by sin u(τ) or by cos u(τ) and then integrate over τ from 0 to 2π assuming ϕ(τ) = const. 1. Now.s. Namely. Now we are prepared to write down the Fokker-Planck equation for conditional probability P(r. Namely. ˙ ˙ ˙ d x = ¨ (˙ cos u − ru sin u) ≈ −ru sin u − 2˙u sin u − ru2 cos u . STOCHASTIC PROCESSES Let us consider the situation when the feedback parameter µ is small and use the so-called method of slow perturbations. ϕ = 0 . τ|r0 . and we will search the general solution in the form x = r(τ) cos u(τ) . (9. ϕ0 .h.30) (9.31) Now we can average this equation over “rapid motion”. In Fig. r Since the r. From such a procedure we get µ ˙ r = r(p − r2 ) . ˙ u ≈ 0.28) at µ = 0 is x = r cos(τ + ϕ) .h. ¨ ˙ −2˙ sin u − 2rϕ cos u . ϕ is constant. 9. r ˙ ¨ r˙ ˙ dτ Here we neglected r since µ ¨ 1. as x = −r sin u to get ˙ 4 4 r2 x 1 − x2 = −pr sin u + r3 sin3 u = −r(p − r2 ) sin u − sin 3u . ˙ Then the l. ˙ ˙ 3 3 3 As a result we get. (9. 0.

t)P(x.2 1.6 0.2 0 –0. ∂t 2 ∂r 2 ∂r ∂r r r ∂ϕ P(r. τ=0 τ −∞ τ=0 τ (xi − yi )(xk − yk ) ˆ B(y. r is plotted.4 0.2 0.t0 ) . The function r(p − r2 ) vs.t0 ) = δ(r − r0 )δ(ϕ − ϕ0 ) .t − τ) .t|x0 .t − τ) . FLUCTUATIONS IN A VACUUM-TUBE GENERATOR 123 0.34) The Fokker-Planck equation acquires the form ∂P(x. ϕ0 .t0 ) + ∑ Bik (x.t|y. –0. and we get µ−1 µ Ar = r = r(p − r2 ) .t|x0 . (9.t0 ) ∂2 1 = −div A(x.t)P(x.t|x0 .2 0 0.35) µ ∂[r(p − r2 )P] B ∂ ∂P ∂ P 1 ∂2 P =− + r + 2 2 .36) It should be solved with the initial conditions . ˙ 2 Brϕ = 0 .8 1 1.6.9. ϕ.t) = lim 1 ∞ x−y = lim (dx) (x − y)P(x.6: On the stationary points.t0 |r0 .4 –0. ∂t 2 ik ∂xi ∂xk Below we shall use a simple model of isotropic random forces with Brr = r2 Bϕϕ = µB In our case. τ=0 τ −∞ (9.6 A(y.t|y.4 –1 Figure 9.4 1 0. (9.6 0.t) = lim τ=0 τ ∞ 1 = lim (dx) (xi − yi )(xk − yk )P(x.33) (9. Aϕ = 0 .

If the generator is strongly excited. (9. ϕ) dr dϕ = i. 2 (9. 2 erf x = √ π e−t dt . Since w is ϕ-independent C−1 = 2π Here rdr exp − (p − r2 )2 4B √ = π3/2 B 1 + erf x 0 p √ 2 B .39) The profile w(r) has a maximum at rm = 1 + 2 p2 +B ≈ 4 √ p(1 + B/2p2 ) for p B/|p| for p √ 2√B . and the solution is ϕ-independent.e.7. At negative x.t|r0 .124 Let us start from the stationary distribution CHAPTER 9. Then the equation (9. . w(r. (9. . ϕ) = lim P(r. the Gaussian distribution. πB 2π ρ2 = B/2p . (9. ϕ. STOCHASTIC PROCESSES w(r.37) 2 ∂r dr r dr r The only solution which is regular at r = 0 implies that the constant is equal to 0. . B 2π r2 = 2B|p| (9.40) 2 which is called the Rayleigh distribution. √ √ 2 B. (9. Then we obtain w(r. (9.38) The proportionality coefficient is given by the normalization condition. we can put r = p + ρ and neglect ρ2 in p −ρ2 p/B dϕ e dρ .h. of previous equation is zero. If ϕ is chosen in the interval (0. ϕ0 . e−x erf (−x) = −1 + √ − . Then we obtain.t0 ) t−t0 →∞ for which the l. 9.41) .36) acquires the form 1 ∂ d w d w r(p − r2 )w − Br = 0 or r(p − r2 )w − Br = const . and w(r) ∝ r exp 1 B r 0 ρdρ (p − ρ2 ) = Cr exp − (p − r2 )2 4B . −2 B . ϕ) dr dϕ = |p| −r2 |p|/2B dϕ e rdr . |x| π One can also neglect r4 in the exponent of Eq. p Eq.38). 2π) all the values of ϕ are equal.s. The function w(r) is plotted in Fig.38.

FLUCTUATIONS IN A VACUUM-TUBE GENERATOR 125 0.44) (9.9. φ.32) with respect to ρ = r − ˙ ρ = −pρ . φ0 . (9.t0 ) ≡ P(ρ. Denoting P(ρ. Pρ (ρ|ρ0 .43) This equation we can analyze for arbitrary initial conditions. rm = √ p ρ2 = B/2p . −1 ∂P ˙ ϕ = 0. shown near the curves. φ|ρ0 .6. The solution for Pϕ (ϕ|ϕ0 .t − t0 ) = Pρ (ρ|ρ0 .t −t0 ) we immediately get that fluctuations of ρ and ϕ are statistically-independent.46) 2p ¯ The ρ(t) is just the dependence due to mechanical equations of motion (9.4 –2 2 0.5 2 To find the generation spectrum let us consider the case of large amplitudes. φ0 .7: Plots of the distribution w(r) √ for different values of p/ B.t − t0 ) Pϕ (ϕ|ϕ0 . The fluctuations in the amplitude are Gaussian. σ2 = 2 1 ¯ 2 e−(ρ−ρ) /2σ 2πσ (9. 2π n=1 (9. ∂2 P 1 ∂2 P + ∂ρ2 p ∂ϕ2 The simplified Fokker-Planck equation acquires the form µ ∂ρP B =p + ∂t ∂ρ 2 .t − t0 ) = √ with ¯ ρ = ρ0 e−µp(t−t0 ) . √ p to get (9. P(ρ.42) In this case we can linearize the dynamic equations (9. φ0 . (9.2 0 Figure 9.45) B 1 − e−2µp(t−t0 ) .t − t0 ) has the form Pϕ (ϕ|ϕ0 .t|ρ0 .47) . 0 0.32) linearized near √ r = p. φ|ρ0 .t − t0 ) .t − t0 ) = ∞ 2 1 1 + 2 ∑ e−n (B/2p)µ(t−t0 ) cos n(ϕ − ϕ0 ) .5 1 1.

50) Since the joint probability w(ϕτ . Indeed. τ) = Here ¯ ρ2 (ρτ − ρτ )2 1 exp − 2 − 2 2πσ∞ στ 2σ∞ 2στ . σ2 = τ ψτ ≡ x(0)x(τ) . As a result. Hence cos ϕ = sin ϕ = 0.49) ∞ 2 1 1 + 2 ∑ e−n Dτ cos n(ϕτ − ϕ) . √ √ ψτ = ( p + ρ)( p + ρτ ) cos(t + ϕ) cos(t + τ + ϕτ ) 1 = (p + ρρτ ) [ cos(τ + ϕτ − ϕ) + cos(2t + τ + ϕτ + ϕ) ] . (9. τ. n=1 ∞ Having the conditional probabilities (9.126 CHAPTER 9.48) we get. 2p using these equations let us calculate the correlation function ¯ ρτ = ρ e−pµτ . Lets us for brevity denote x(0) = x. (9. Eq. ρ.46) and (9. ρρτ = dρτ dρ ρτ ρ w(ρτ . D = µB/2p . τ). 2 4π n−1 B 1 − e−2pµτ . τ) = w(ρ)P(ρτ |ρ.49) is an even function of ϕτ − ϕ one can easily show that cos(ϕτ + ϕ) = sin(ϕτ ± ϕ) = 0 . As a result. Then.52) (9. x(τ) = xτ . ρ. .47) one can find all necessary correlation functions. the joint probability. ρ. to find the values ρ and ρτ after time delay τ is w(ρτ . One can easily check that the t − t0 the system tends to stationary distributions. from Eq. (9.48) (9. cos(τ + ϕτ − ϕ) = cos(ϕτ − ϕ) cos τ. Since √ x(t) = [ p + ρ(t)] cos[t + ϕ(t)] and x(t) = 0 due to the facts that ρ =0. At t = t0 + 0 it gives proper initial conditions since 1 + ∑ cos n(ϕ − ϕ0 ) = 2π δ(ϕ − ϕ0 ) . τ) = w(ϕ)P(ϕτ |ϕ.51) 2 Then. (9. and the phase distribution is uniform.53) = e−Dµ|τ| . (9. ϕ. τ) = D e−pµ|τ| . the average deviation from the stationary solution. STOCHASTIC PROCESSES ¯ As t −t0 increases. decays while dispersion σ grows. cos(ϕτ − ϕ) 1 = 4π2 2π 0 2π ∞ dϕτ 0 dϕ cos(ϕτ − ϕ) 1 + 2 ∑ e−n n=1 2 Dµ|τ| cos n(ϕτ − ϕ) (9. τ) = w(ϕτ . ϕ. w(ρτ . ρ. ρ and ϕ are statistically independent at any time.

54) The first factor in given by the fluctuations in the amplitude while the second is due to the phase fluctuations. The oscillation spectrum is given by Fourier transform of the above function. h = µpω0 are dimensional phase diffusion constant and increment. FLUCTUATIONS IN A VACUUM-TUBE GENERATOR Collecting all the factors we finally obtain: ψτ = 1 p + De−pµ|τ| e−Dµ|τ| cos τ .6. In the dimensional variables. 2 + D2 2π α h α + (D + h)2 (9. That means a superposition of a narrow line with the half-width D with a broad line with the half-width D + h. τ → ω0 τ.55) D = µDω0 . . we get ψω = Φ(ω − ω0 ) + Φ(ω + ω0 ) ≈ Φ(ω − ω0 ) with Φ(α) = Here p D D D +h + · 2 .9. 2 127 (9.

128 CHAPTER 9. STOCHASTIC PROCESSES .

Such a pair can be extracted in N(N − 1)/2 ways. dVN ≈ N2 2V 2 e−βV12 − 1 dV1 dV2 . 1 and the gas density N/V Here V12 is the pair interaction potential. It is convenient to rewrite the above expression in the form F = Fid − T log 1 + 1 VN ··· e−βV (q) − 1 dV1 . Fi = −T log 1 VN ··· e−βV (q) ∏ dVa . The factor V −N in the argument of the logarithm is just because for the ideal gas the integral over the volume is V N . a Here Fid is the free energy of the ideal gas. dVN . Thus the integral can be expresses as N(N − 1) 2V N ··· e−βV12 − 1 dV1 . Fi = − 2 2V 129 .Chapter 10 Non-Ideal Gases 10. . Since log(1 + x) ≈ x at x is assumed to be small. Now let us make use of the assumption that the gas is rarefied and only two atoms can collide at the same time. The simplest case is the monoatomic gas where the interaction potential depends only on the distance between the atoms. . . Then the integrand is not small only if some two atoms are close together. Since p2 H (p. q) = ∑ a +V a 2m we can write the free energy as F = Fid + Fi . T N2 e−βV12 − 1 dV1 dV2 . .1 Deviation of gases from the ideal state We start from the case of a weakly non-ideal gas where the only pair interaction is important.

NON-IDEAL GASES Since the interaction depends only on the distance between the atoms we can integrate out the center of mass and get V . V B(T ) = 1 2 1 − e−βV12 dV . V123 = V12 +V13 +V23 . eβΩ = 1 βµN e N=0 N! ∑ ∞ dΓN e−βHN (p. For example. 2m H2 = p2 a +V12 . 10. Since P = −∂F/∂V we obtain the following correction to the equation of state. In this way we are left with the expression Fi = T N2 B(T ) . For the N-particle Hamiltonians we have. . H1 = where in general p2 .2 ∑ H3 = a=1 a ∑ 2m +V123 .1: A typical dependence of the normalized potential energy V /V0 versus the normalized interatomic distance r/r0 . To determine them it is convenient to use the Landau fee energy. A typical dependence of the interaction potential energy is shown in Fig. Gi = NBP.q) . B1 = 1 where coefficients Bn (T ) are called the virial coefficients. 0..130 CHAPTER 10.4 1. where dV stands for relative coordinates.8 2 0 -1 2r0 Expansion in powers of density In principle.1 3 2 V 1 Figure 10.8 1 1. 3 p2 . the equation of state can be expressed in the form of expansion in the density. P=T ∞ n=1 ∑ N V n Bn (T ) . V V The corrections to other thermodynamic potentials can be found using the principle of small increments. NT N P= 1 + B(T ) .6 1. 2m a=1.2 1.. Ω = −PV .

we can integrate out the center-of-mass motion to get Ω = −PV = −T log 1 + ζV + ζ2V 2! e−βV12 dV2 + ζ3V 3! e−βV123 dV2 dV3 + . . . . For a uniform system it can depend only on the difference r = |r1 − r2 |. The structure of the integrals Jn is clear. n dV being the particle number in the element dV .V ∞ N Jn =∑ ζn .10. . . they are not small only if n particles are close together. The correlation in the positions of the particles can be characterized by the average ¯ ∆n(r1 ) ∆n(r2 ) = n(r1 )n(r2 ) − n2 .V ∂P ∂µ T. Spatial correlation of density fluctuations Let us denote n(r) the fluctuating number density.. . Again. To obtain equation of state one has to relate the parameter ζ to the density. Since N=− we have ∂Ω ∂µ =V T. We arrive at the expansion in powers of ζ P=T with J1 = 1 . . J3 = J2 = e−βV12 − 1 dV2 . V n=1 (n − 1)! The set of equations P(ζ) and N(ζ) defines the equation of state. it must decay at r → ∞. n=1 ∑ n! ζn ∞ Jn e−βV123 − e−βV12 − e−βV23 − e−βV13 + 2 dV2 dV3 . DEVIATION OF GASES FROM THE IDEAL STATE If one denotes ζ≡ eβµ (2π¯ )3 h 131 d 3 p e−βp 2 /2m = mT 2π¯ 2 h 3/2 eβµ the expression for Ω can be written as Ω = −PV = −T log 1 + ζV + ζ2 2! e−βV12 dV1 dV2 + ζ3 3! e−βV123 dV1 dV2 dV3 + .1..

the particle belongs both to dV1 and dV2 . ∆n(r1 ) ∆n(r2 ) = nν(r) + n δ(r1 − r2 ) . ¯ ¯ ν(r) ≡ n[w12 (r) − 1] . Since for the pair interaction ν(r) = n e−βV12 (r) − 1 ¯ we get V ν(r) d 3 r . ¯ The quantities ∆n(r1 ) ∆n(r2 ) and ν(r) are called the correlation functions.2). Then the scattered wave at the detection point r is Fsc (r) = f i=1 ∑ F(ri) N N eikρi eikρi = f F0 ∑ ei(kri −ωt) . ¯ ¯ In a similar way. 10. Integrating the previous equation with respect to dV1 dV2 over a volume V we have to recover (∆N)2 . T This is zero for an ideal gas. From the formula above we find the normalization condition.2: On the scattering problem. Since the total probability of such event is n w12 (r) dV2 we might write ¯ n(r1 ) n(r2 ) = n2 w12 . N We see that the correlation function is an important property. Thus the correct formula is G(r1 − r2 ) = n(r1 ) n(r2 ) = n2 w12 + n δ(r1 − r2 ) . ρi ρi i=1 . only interaction leading to a finite compressibility leads to a correlation. It can be directly determined from the experiments on the light scattering. ¯ That would be a mistake because it ignores the fact that if the points 1 and 2 coincide. Let the scattering amplik’ k r i ρ i R Detector Figure 10. Consider a wave J2 = e−βV12 (r) − 1 dV = F(R) = F0 eikR−iωt which is scattered by the particles residing at points ri (see Fig. tude f be the same for all the particles. NON-IDEAL GASES The product n(r1 ) n(r2 ) can be expressed through the probability density w12 to find the particle 2 at the distance r from the particle one.132 CHAPTER 10. ν dV = (∆N)2 TN −1 = − 2 N V ∂V ∂P −1.

The quantity 2r0 has a meaning of the atomic diameter. To derive such a formula let us assume that the interaction is small in a sense that V0 T . Replacing the double sum by its ensemble average. DEVIATION OF GASES FROM THE IDEAL STATE 133 Here ρi = |R − ri | is the distance between the scatterer and the detector. In this way we get. We have ρi = R − ri cos φ where φ is the angle between ri and R Now let us take into account the fact that the detector is usually place far from the scattering medium. Van der Waals’ formula There is an interpolation formula which allows to follow the phase transition between gaseous and liquid state. i q ≡ k−k . Under this approximation we can replace ρi by R in the denominator of the previous formula. R ri . B(T ) = 2π 0 2r0 1 − e−βV12 r2 dr + 2π ∞ 2r0 1 − e−βV12 r2 dr . kri + kρi = kri + kR − kri cos φ = kR − ri (k − k ) . Thus one can expect singularities of the scattering cross section at the phase transition point. Fsc = f F0 eikR−iωt R ∑ e−iriq . Gi = NP(b − a/T ) . we can write it as S(q) = 1 N dV1 dV2 n(r1 )n(r2 ) eiq(r1 −r2 ) = 1 + d 3 r ν(r) eiqr . while the phase needs more exact treatment. Let us split the integral for B(T ) into 2 parts. 3 In the first integral we neglect the exponential and obtain for that b = 16πr0 /3. N S(q) is called the structure factor. In the second integral we expand the exponential to get −2πβ In this way. Fi = N 2 (bT − a)/V . Note that for small scattering angles q → 0 the structure factor is proportional to the compressibility of the system. B(T ) = b − a/T → ∞ 2r0 |V12 | r2 dr ≡ −2βa . Thus the ratio between the average intensity of the scattered wave and the intensity of the incident wave is | f |2 ∑ N i=1 j=1 ∑ eiq(ri−r j ) ≡ N| f |2 S(q) . To perform the integration for the virial coefficient let us assume that at r 2r0 we have V12 > T .1.10. This is the way to extract the correlation function from the experiment on light scattering. .

Let us assume that this expression holds also for large density! As a result. → N 2a P + 2 (V − Nb) = NT . V Nb → logV − Nb/V ≈ log(V − Nb) .134 The total free energy is CHAPTER 10. V . Now we get ∂F NT N 2a P=− = − ∂V V − Nb V 2 This the van der Waals’ equation. Fi = −NT log(1 − Nb/V ) − N 2 a/V . Since we have assumed that the gas is rarefied. NON-IDEAL GASES F = N f (T ) − NT log(e/N) − NT (logV − Nb/V ) − N 2 a/V . Now we do something approximate. This formula take account of the finite compressibility at large density.

Since the temperature and pressure remain constant. This state is non necessary homogeneous. This is illustrated in Fig. ∂T ∂P dT ∂T ∂P dT 135 . P1 = P2 .1 Conditions for phase equilibrium Consider an isolated body with the energy E and volume V .1 P II I T T I II Figure 11.V )-plane. T )-plane along which the phases coyexist. In the (T. T ) = µ2 (P.1: The hatched area corresponds to a mixture of 2 phases.Chapter 11 Phase Equilibrium 11. For the case of two phases. it can consist of two (or more) phase. the equilibrium conditions are T1 = T2 . V The Clapeyron-Clausius formula Differentiating the equilibrium condition µ1 = µ2 with respect to temperature we get ∂µ1 ∂µ1 dP ∂µ2 ∂µ2 dP + = + . µ1 = µ2 . there are 2 lines because at a given pressure and temperature can have different volumes. 11. T ) correspond to a line in the (P. the equality µ1 (P.

136 Since ∂µ ∂T ∂µ ∂P

CHAPTER 11. PHASE EQUILIBRIUM

= −s ,
P

=v
T

(s ≡ S/N, v ≡ V /N)

we obtain

dP s1 − s2 q = = dT v1 − v2 T (v1 − v2 )

where q = T (s1 − s2 ) is the heat of transition. This the Clapeyron-Clausius formula. This equation gives the temperature dependence of pressure along the phase equilibrium curve. Written as dT T (v1 − v2 ) = dP q it gives the change in the equilibrium temperature (freezing point, or boiling point) when the pressure changes.

The law of corresponding states
Let us discuss as an example the van der Waals equation of state, P= NT N 2a − 2 . V − Nb V

We can notice that there is a special temperature, Tc , at which the isotherm has an inflection point. We have: ∂P ∂V = 0,
Tc

∂2 P ∂V 2

=0
Tc

Tc =

8 a 1 a , Vc = 3Nb, Pc = . 27 b 27 b2

Measuring the P, V , and T in the units Pc , Vc and Tc , respectively, we can rewrite the equation of state in the form 3 P + 2 (3V − 1) = 8T . V The van der Waals’ isotherms determined by this equation are shown in Fig. 11.2. So we are able to express the equation of state in a form which is independent of the particular properties of the gas. We observe that at T < Tc there are regions with positive ∂P/∂V . In that regions the system cannot be stable. On the other hand, we know that the equilibrium can be achieved only at P = const. As a result we arrive at the situation shown in Fig. 11.3 To the right from the point e the system is in a gas state, while to the left from the pint b it is in a liquid state. The phase equilibrium takes place along the straight line b − e which corresponds to P = const. This line must be constructed in a way to keep µ1 = µ2 . This difference can be expressed as an integral along the transition path, µe − µb =
e b

dµ = 0 .

11.1. CONDITIONS FOR PHASE EQUILIBRIUM

137

1.4 1.05 1.2 1.0 1 0.95 0.8 p 0.6 0.4 0.2 0 0.4 0.6 0.8 1 1.2 1.4v 1.6 1.8 2 2.2 2.4 0.9

Figure 11.2: Van der Waals’ isotherms. The numbers near the curves are the ratios T /Tc .

P Liquid

a K d e c A B

b

Gas

Figure 11.3: On the gas-liquid transition.

f

V
Since dµ = (V /N) dP we come to the conclusion that the integral along the isotherm,
e b

V dP = 0 .

Geometrically it means that the areas between the isotherm and the line b − e below and above the line must be equal. This rule is called the Maxwell construction.

Van der Waals theory of the critical point
The calculations can be easily performed near the critical point, Tc . Introducing notations P = Pc (1 + p), T = Tc (1 + τ) N N = (1 + n) V Vc

and expanding in powers of τ and n up to non-trivial order we arrive at the equation of state p = 4τ + 6τn + (3/2)n3 , (∂p/∂n)τ = 6τ + (9/2)n2 .

138 The Maxwell condition can be written as
e b

CHAPTER 11. PHASE EQUILIBRIUM

V dP = 0

ne

n
nb

∂p ∂n

dn = 0 .
τ

Since the integrand is an odd function of n we get ne = −nb . From the condition p1 = p2 we obtain √ ne = −nb = 2 −τ , p = 4τ . The boundary of the metastable region are given by the equation nd = −nc = 2 −τ/3 . √ −τ.

According to the Clapeyron-Clausius formula, the heat of transition is proportional to

the work ˜ done by the material can be expressed through the total magnetic moment M ≡ V M as ∆W = −H ∆M . Then the magnetic field inside the solenoid is H = NI/L. Here A is the solenoid area. where I is the electric current. V is its volume.Chapter 12 Continuous Phase Transitions Introduction In the following set of lectures we shall discuss so-called continuous phase transitions. The second term is the work done by the field on the material. When the magnetic flux Φ through the solenoid is changing an electro-motive force. The first item is the energy of external magnetic field which is present also in the absence of material. while B ≡ Φ/A is the magnetic induction. Thermodynamics of a magnet Consider a long and thin solenoid with the length L and number of windings N. As a result. S 139 . or transitions of the 2nd order. Introduc˜ ˜ ing magnetization M from the relationship B = µ0 (H + M) we can express the work per volume as ˜ ∆Wb /V = ∆(µ0 H 2 /2) + µ0 H ∆M . A generic system to discuss the transitions is a magnet. E = −∂Φ/∂t. Consequently. M H= ∂U ∂M . the work made by the battery being ∆Wb = E I ∆t = I ∆Φ = NAI ∆B = V H∆B . Thus we start in recollection of thermodynamics of a magnet. the 1st law of thermodynamics reads as dU = T dS − H dM and T= ∂U ∂S .

1: The spontaneous magnetization as a function of temperature. while for d = 1 the magnetization never vanishes. or it is abrupt. At zero external magnetic field magnetization vanishes at certain temperature. It is the fight between the order and disorder that makes the critical state specifically interesting. According to that model. e. Below Tc a spontaneous magnetization M0 (T ) is observed. a vector. At low temperatures the energy U is more important and the system is a ferromagnet. T General considerations Uniaxial ferromagnet is a simplest example of phase transition. we can define the spontaneous magnetization M0 (T ) as the order parameter. Thus the character of transition cannot be specified without specifying the transition path. Tc . the above conclusion is true for the dimension d ≥ 2. ∂T M ∂M T ∂G ∂G G(T.H= . When an order parameter is specified. the next question is if the transition is of 1st order. Than the H = 0 the magnetic moment changes from −M0 (T ) to M0 (T ). In some cases it is difficult to specify the order parameter. Nearest neighbors interact in a way that is is favorable to point in the same direction. The dependence M0 (T ) is schematically shown in Fig. M=− ∂T H ∂H . U and −T S(T ). This figure can be explained in the framework the Ising M (T) o Figure 12.1. CONTINUOUS PHASE TRANSITIONS Generalizing of other thermodynamic potentials is more or less obvious: F(T. or it is continuous. At high temperature the free energy is minimized by disordered state in which the entropy is large. As we know. a tensor. The higher the dimension the greater the number of neighbors. it is the question if the spontaneous magnetization goes to zero at T → Tc− is a continuous way.140 CHAPTER 12. the free energy G = U − T S − HM has to be minimized. etc. i. H) = F − HM → dF = −S dT − M dH . In the example of the ferromagnet. M) = U − T S → dF = −S dT + H dM . Assume that we ramp the external magnetic field from a negative to a positive value. At H = 0 it consists of two terms. This is a delicate issue. its direction can be “up” and “down” with equal likelihood. In real life. S = − ∂F ∂F . 12. spins are located at lattice points in a regular lattice. That indicates that phase transition is a collective effect depending on interaction of many spins. Coming back to the case of a uniaxial ferromagnet. the transition is of the 1st order. S = − . the stronger the collective effects. T c T model. . In general it can be a scalar. They can point up and down.

Usually a number of quantities show power law behavior close to Tc . and on the average there are equally many spins pointing in opposite directions. An important concept of the theory of continuous phase transitions is the universality class. 1). In this case the corners of a (hyper)cube in spin space are energetically favored (or suppressed). Examples of phase transitions We shall classify the phase transitions by the spatial dimensionality and the number of independent components of the order parameter. 3). the universality class being (d. M0 (T ) ∼ (Tc − T )β . Systems of the same dimension and with a phase transition into an ordered state with the same symmetry belong to the same universality class. ms = ∑ (−1)k+l+m mklm . Heisenberg model: the magnetization is isotropic in 3 dimensions. cubic anisotropy. Note that the universality classes for antiferromagnets are the same as for ferromagnets. n). It became clear during 1960’s that a number of details in a given system are irrelevant as far as critical exponents are concerned. Magnetic systems Uniaxial ferromagnets: described by the Ising model. 2). T |M(H. So far we have assumed isotropy in spin space. They have essentially the same critical properties. l. This model is called the isotropic xy-model. the universality class being (d.141 At continuous phase transitions critical phenomena are observed. the universality class being (d. That gives rise to new universality classes. The order parameter on a 3d cubic lattice can be chosen as staggered magnetization. T = Tc )| ∼ H 1/δ . m). e. This assumption can be relaxed. The justification of that concept will be discussed later. Other examples of critical exponents for thermodynamic quantities are C ∼ |Tc − T |−α . and one can also expect. Hard axis ferromagnets: m0 is a vector in the xy-plane perpendicular to the hard axis. klm where mklm is the magnetic moment of spin at the lattice point (k. (d. χ= ∂M ∂H ∼ |Tc − T |−γ . g. (d. . say. Antiferromagnets: Neighboring spins prefer to point to different directions. 2).

liquid He4 undergoes a continuous phase transition from the normal phase (HeI) to a (partially) superfluid phase (HeII). M where H is the external magnetic field. Having in mind the magnetization as an example. percolation. we can specify magnetization M as an order parameter. 1) because the order parameter is a scalar. the critical region in the ”old” superconductors is so narrow that it is almost irrelevant form the experimental point of view.1 The critical phenomena were. T.1 Landau Theory Landau theory was designed to describe systems close to a phase transition where the order parameter ψ is small. Those 1 The situation is a bit different at low dimensions. The situation is more complex for liquid mixtures. . ψ) = G0 + αψ + Aψ2 +Cψ3 + Bψ4 . However. The simplest phase is nematic where the rods are oriented. then the free energy depends on T. e. observed in high-Tc superconductors. dynamical phase transitions. g. Otherwise the nematic is a disordered liquid. The period is about ˚ 3000 A. etc. Smectic liquid crystals combine orientation order with a spatial one along one direction. Liquid crystals Liquid crystals consist of long molecules. The relative fraction of superfluid component. CONTINUOUS PHASE TRANSITIONS The universality class is the same as for the Ising model. 2). Near the transition point we can expand the free energy as ˜ G(P. Structural phase transitions These transitions are changes in the spatial symmetry of the crystalline lattice. change in the symmetry of adsorbed monolayers. 12. There exist more complicated structure like cholesteric phase where a director rotates at a constant pitch in a plane perpendicular to the direction in which one moves. however. It appears that the proper order parameter is a two-component rotation invariant quantity. . A similar situation is relevant to superconductors. Superfluids and Superconductors At the temperature T = Tλ ≈ 2 K.142 Fluid systems CHAPTER 12. . Other phase transitions One can name many other phenomena which can be regarded as critical ones. so the universality class in (3. ψ can be a scalar. (d. H. . etc. ρs /ρ decreases continuously from 1 at T = 0 to 0 at T = Tλ . At the 1st approximation they can be considered as inflexible rods. vector.

. G(T. • The free energy must be properly defined and then expanded in a Taylor series up to the minimal relevant order (see below). M = ±M0 = ± −6aτ/u . see Fig.1.2: The free energy above (left) and below (right) the critical temperature Tc . ˜ • The spontaneous value(s) of the order parameter is found by minimization of G with respect to ψ keeping the thermodynamic parameters constant. and in the minimum it coincides with the Gibbs free energy. −M). it is a hypothetical free energy which would be the case if the system with the order parameter ψ could exist as an equilibrium one. At r(T ) = 0 the high symmetry of the disordered phase in broken and one of the low temperature ordered phases is chosen. 12. The total scheme of the Landau theory is the following. u(T ) = u(Tc ) ≡ u . ∂M T 3! This equation has 3 solutions.12. Uniaxial ferromagnets ˜ ˜ In zero magnetic field there is up-down symmetry.2. Thus ˜ ˜ ˜ ∆G(T. Obviously. This phenomenon is called the spontaneous symmetry breaking. Differentiation of G with respect to M yields ˜ ∂G u 0= = aτ + M 3 . 2! 4! Existence of stable minima requires u(T ) to be positive. r(Tc ) = 0. 0) = r(T ) 2 u(T ) 4 M + M +··· . M) = G(T. The ∆G t<0 ∆G t>0 M Figure 12. M) ≡ G(T. ˜ Here τ = (T − Tc )/Tc is the reduced temperature variable. At r(T ) > 0 the free energy has one stable minimum. The equilibrium is reached at the minimum of G. M) − G(T. M = 0. LANDAU THEORY 143 are the intensive parameters which characterize the state. while at r(T ) < 0 there are 2 equivalent minima ±MO . and near Tc we can assume r(T ) = at . -M 0 M0 M higher symmetry phase at r > 0 is replaced by a lower symmetry phase. Note that G is not a true Gibbs free energy.

while the other represent degenerate minima below Tc . 0. 0) + aτ 6a(−τ) u + 2 u 4! 6a(−τ) u 2 . M) . M) = −HM + r(T ) 2 u(T ) 4 M + M +··· . C+ + (T /Tc ) Thus we observe a negative jump in the specific heat if one passes the critical point from below. 0) = G(T. 0. 0) . 12. τ < 0 ˜ G(T. 0. τ>0 C= 2 · (3a2 /u) . H) = min G(T. C=T Here G is the usual free energy. M0 (T )) . 1 3 H = rMe + Me + · · · . 3! . 0)/∂T 2 we obtain C+ . The critical exponents following from the Landau theory are called “classical”. τ < 0 . τ>0 . M ∂S ∂T = −T H=0 ∂2 G ∂T 2 . H) = As a result. Magnetic Susceptibility Now let us study a response to a small external magnetic field. Denoting C+ = −T ∂2 G(T. Me . If we present the solution as M0 ∝ (−τ)β → ψ0 ∝ (−τ)β we obtain β = 1/2 for the critical exponent. 0.2 it is clear that the 1st solution represents a stable minimum above Tc . ˜ Now we easily obtain the specific heat by differentiation. Jump in Specific Heat Now let us consider the specific heat. ˜ G(T. G(T. H. Then we have ˜ ∆G(T. 2! 4! The minimization with respect to M leads to the equation for the equilibrium value of the magnetization.144 CHAPTER 12. H=0 At H = 0 we get G(T. CONTINUOUS PHASE TRANSITIONS From Fig. at τ < 0 we obtain G(T.

τ < 0 If we define the critical exponent of the quantity A(τ) as s ≡ − lim ln A(τ) τ→0+ ln τ we find the classical critical exponent γ = 1. 2a(−τ) . At the critical isotherm. 3! If we in general assume Me ∝ H 1/δ .12. LANDAU THEORY The inverse susceptibility is defined as χ−1 = ∂H ∂Me u 2 = r + Me . u 3 H = Me → Me ∝ H 1/3 .1. . 12. The plot of the free energy vs. then the classical value of δ is 3. M is shown in Fig. 2 145 T Substituting the equilibrium magnetization at H = 0 we obtain χ−1 = aτ . τ>0 . Later on we shall compare the values of critical exponents with more general theory and with experiment.3 for τ > 0 (solid line) and τ < 0 (dashed line). Conclusion: If we are interested in continuous phase transition the fields which linearly coupled with the order parameter must be removed. M minimum in both cases – magnetic field removed the system from its critical point. τ = 0. Role of External field Let us now discuss the situation in a finite external field. We observe that there is a stable G Figure 12.3: The free energy at a finite external field. We get χ(τ) ∝ |τ|−γ with the slope ratio A+ /A− = 2.

At low temperatures an additional minimum develops. M 1 G M 2 at τ = τ2 . CONTINUOUS PHASE TRANSITIONS Now let us consider the case when symmetry of the system allows ψ3 terms. At sufficiently large positive τ the profile has a shape shown in the left panel.146 Role of Cubic Invariants CHAPTER 12. .01 G Figure 12. τ = 0 does not have the significance of the transition temperature in this case. The analysis of that free energy becomes clear from Fig. 2! 3! 4! Let us put c > 0. 12.4: The free energy with a cubic invariant included. It has one stable minimum at ψ = 0. Thus there is a (weakly) first order transition from ψ = 0 to some ψ = ψ0 ˜ ∆G = G M 0.4. Conclusion: Cubic terms results in a first-order transition rather in a continuous one. and at some τ2 this minimum is degenerate with ψ = 0 (middle panel). At lower temperatures the second minimum is the only stable one (right panel). Then r(T ) 2 c 3 u 4 ψ − ψ + ψ +··· .

but u is negative. At this point the character of phase transition will change from 1st order to continuous. and it changes sign at p = pc . LANDAU THEORY Tricriticality 147 Now let us return to the simplest case where the free energy is even in the order parameter.5: The free energy with a cubic invariant included. p. At T = T0 the transition takes place to one of a symmetric set of states with lower symmetry at which the order parameter is ±ψ0 . 12. The temperature dependence of the order parameter is G M G M Figure 12. Above T0 the equilibrium order parameter is zero (see right panel of Fig. 12.12.6. u(pc ) = 0 is called the tricritical point. It is easy to find classical critical exponents at the tricritical point.1.5). Let us assume that r > 0. G M shown in Fig. Putting u = 0 and minimizing the free energy we obtain the . 2! 4! 6! At w > 0 the system undergoes a 1st order phase transition at some temperature T0 corresponding to r = r0 > 0. Suppose now that u is a function of some parameter. The point r = aτ = 0. To keep the stability one has to add the higher order term to get r u w ˜ ∆G = ψ2 + ψ4 + ψ6 · · · .

where all the terms compatible with the symmetries of the system are included. • Characterize the order of the low symmetry phase and define the order parameter (scalar. • Substitute ψ0 into the minimum.148 CHAPTER 12. or .). T w 5 ψ =0 5! 0 → ψ0 = 5! a (−τ) w 1/4 . . ψ) for a given space-independent order parameter. Summary of the Landau recipe • Given that the system has a continuous or weakly 1st order phase transition form a disordered phase to the phase of higher symmetry. The given value of this parameter does not have to be the equilibrium value. tensor. vector. T o expression for the order parameter.2 ˜ • Determine the equilibrium value(s). ˜ • Write down the Taylor expansion of G in powers of ψ. aτψ0 + Thus βt = 1/4. .6: The order parameter as a function of the temperature at u < 0. free energy and calculate equilibrium properties like critical exponents. by minimizing G with respect to ψ. CONTINUOUS PHASE TRANSITIONS ψo Figure 12. ˜ • Introduce the proper free energy G(T. . Include only the minimum number of terms. ψ0 . 2 Sometimes a parameter different from the dimensionless temperature can play a role of τ. • Write the coefficients of the ψ-expansion as a Taylor expansion in τ = (T − Tc )/Tc and keep only the leading ones.

h). After such an assumption the specific heat is C∼ ∂2 Gs (τ. λb h) = λGs (τ.2. However. The postulate allows one to express all the critical exponents through the constants a and b. SCALING LAWS 149 Critique • The Landau theory assumes the existence of the phase transition. .12. b/a |τ| |τ| .2 Scaling laws Consider a system close to continuous phase transition. Then. ∂h At τ > 0 spontaneous magnetization m0 vanishes. The physical reason for the postulate will be discussed later in connection with renormalization group arguments. h) = |τ|1/a Gs Since τ/|τ| = ±1 we can express Gs as Gs (τ. 1965) that Gs (λa τ. Gs (τ. 0) ∼ |τ|−α ∂τ2 α = 2 − 1/a. the transition point corresponding to (τ. The generalized magnetization is m∼ ∂Gs = |τ|2−α−∆ Φ± (h/|τ|∆ ) . τ h . Let us introduce reduced temperature. • Fluctuations of the order parameter are neglected. when τ < 0 Φ− (0) = B− = 0. while Φ± (0) = A± are constants. 12. As result. τ. • A general recipe for constructing the order parameter does not exist. This is the so-called homogeneity postulate. and external field h. β = 2−α−∆. h) = 0. • Spatial dependence of the order parameter is ignored. thus Φ+ (0) = 0. Since we are interested in singular behavior near the critical point let us separate the singular part of the free energy. The crucial point is to assume (following B. Widom. h) . To derive scaling laws we choose λa = |τ|−1 . These relations are called the scaling laws. and to find relationship between the exponents. ∆ = b/a . Gs (τ. h) = |τ|2−α Φ± (h/|τ|∆ ) .

β |τ| |τ|βδ 12. etc. the LG functional is constructed as GLG = dr r u g (∇m)2 + m2 (r) + m4 (r) + · · · . (∇m(r))2 .3 Landau-Ginzburg theory The Landau-Ginzburg (LG) theory is an attempt to take into account fluctuations of the order parameter in its simplest way. 2 2 4! . Then the free energy becomes a functional of m(r) rather the function of a total magnetic moment. Since the free energy must be an even function of magnetization one can imagine the combinations like m(r) ∇m(r). 2−α−∆ that is the second scaling law. ∂h2 γ = −2 + α + 2∆ . β = 1/2. h) ∼ |τ|2−α−∆−κ∆ hκ . As a result. Note that classical exponents are (α = 0. What is left is to determine the exponent δ for the critical isotherm. the isothermal susceptibility in zero field is given as χ= With Φ± (0) = C± = 0 we obtain ∂m ∂h ∼ T ∂2 Gs = |τ|2−α−2∆ Φ± (0) . Taking the uniaxial ferromagnet as an example we consider now the local magnetization m(r) as the order parameter. The problem is the argument of the function Φ± tends to infinity at |τ| → 0 for finite h. CONTINUOUS PHASE TRANSITIONS In a similar way. as has been derived by L. m ∇2 m. Both sets meet the scaling law.150 CHAPTER 12. → δ= ∆ . Onsager. Using the scaling laws we can express the behavior of the order parameter as a law of corresponding states as m h = Φ± . γ = 7/4). β = 1/8. and the 2nd and the 3d ones can be combined using partial integration. → α + 2β + γ = 2 . The 1st combination can be transferred into a surface integral. M. γ = 1) while the exact Ising one for a 2D system are (α = 0. The recipe is to add only simplest non-trivial terms. Excluding the ”gap exponent” ∆ we have ∆ = β+γ This is the first example of the scaling law. Assuming that Φ± (x) ∼ xκ for large x we obtain m(|τ|) → 0. For m to have a finite value at |τ| → 0 we have to put 2 − α − ∆ − κ∆ After a simple algebra we obtain ∆ = βδ = β + γ → α + η(1 + δ) = 2 .

g > 0. Consequently. (12. 2 2 (12.2) According to the Gaussian approximation. and the item proportional to δm vanishes.4) ∑ means the we restrict summation up to some k0 we have to do that because we neglect higher gradients. 3! 4! dr (12.3) ∑ A−kBk → k dk A−k Bk (2π)d ∑(gk2 + r)m−kmk . the spontaneous magnetization m(r) = 0. LANDAU-GINZBURG THEORY 151 As before. As a result. Then the fluctuations are Gaussian and various fluctuation modes are statistically independent. ∏k dmk exp(−GLG ) .1) Below Tc the fluctuation is δm(r) = m(r) − m0 (T ) where m0 (T ) = m(r) is the spontaneous magnetization. The latter inequality implies that the homogeneous state is the equilibrium one. Classical fluctuations In the classical liming discussed we keep only the terms quadratic in fluctuations.12. m(r) is itself a fluctuation. k (12. The LG theory allows for long-range fluctuations because it is an expansion in powers of the magnetization gradient. u > 0. At T > Tc . the items proportional to (δm)3 and (δm)4 are neglected.3. To order of m2 we have GLG = dr g r (∇m)2 + m2 (r) + · · · . m0 = −6r/u. and with h = 0. r = aτ. GLG = Since dr A(r)B(r) = we obtain above Tc GLG = 1 2V 1 V dr (−2r) 2 g (∇ δm)2 + m0 + · · · 2 2 . Now we are able to calculate correlation functions like Γm ≡ mq mq = qq ∏k dmk mq mq exp(−GLG ) . The expansion up to (δm)4 gives GLG = g r u (∇ δm)2 + m2 + m4 0 2 2 4! 0 u r u + r + m3 (δm) + + m2 (δm)2 0 3! 2 4 0 1 1 + um0 (δm)3 + (δm)4 + · · · . Furthermore.

Then we can introduce a new variable x = m2 .4) that Γm = V δq . After Fourier transform to real coordinates we obtain for d > 1 and r ξ Γm (r) ∼ r(1−d)/2 e−r/ξ for r r2−d for r ξ. d > 1 . ν = 1/2. As a result. The Ginzburg criterion Let us estimate the validity of the approximation made. d > 2 ξ= g/aτ In this way we predict critical exponents for the correlation length. dx = 2m dm and use the result for Gauss integral. Then ∏ dmk k = ∏ dm−kdmk k k ∑(gk2 + r) m−kmk k = 2 ∑(gk2 + r) m−k mk . φ) = 2im . 0 Γm = Γm = (aτ)−1 .152 CHAPTER 12. it −k is clear from(12. 0 q=0 . we are left with the integral Γm = q 1 V dm−q dmq m−q mq exp[−(gq2 + r)m−q mq /V ] . 1 Γm = 2 . CONTINUOUS PHASE TRANSITIONS Here ∏k means that one has to take into account only |k| ≤ k0 . mq ) e−iφ eiφ = −ime−iφ imeiφ ∂(m. q gq + aτ This formula has been obtained previously by Ornstein and Zernike in 1914 using other considerations. Since m(r) is real. mk = m∗ . ∏k dmk exp(−GLG ) Let us count only those k for which kx > 0.−q Γm q qq where Γm ≡ m−q mq = q ∏k dmk m−q mq exp(−GLG ) . ξ. As a result. dm−q dmq exp[−(gq2 + r)m−q mq /V ] Since m−q is a complex variable we can parameterize it as mq = meiφ and calculate the Jacobian ∂(m−q . The measure of the total strength of the long-wave fluctuations are given by the integral dr Γm (r) ≡ Γm .

A typical value of m when the 4th and 2nd order terms are comparable is (δm)2 −d −1 r 2 u m = m4 ¯ ¯ 2 4! The LG theory is valid at (δm)2 m2 .12. We see that at d > 4 this criterion is always satisfied because the r. is a constant depending on the concrete problem under consideration. 12. Since the typical diameter of correlated region is ξ we have Γm ≈ (δm)2 ξd . At d < dc = 4 the Ginzburg criterion is violated close to the critical point. n). This is the famous Ginzburg criterion. Because we are interested in the vicinity of Tc we can expand the coefficients in power of τ= T − Tc . so the LG theory is reasonable outside that region. Accordingly. one should define the order parameter. Thus dc = 4 is called the upper critical dimensionality. As the order parameter it is natural to chose the wave function of the Bose condensate. Near the critical temperature. According to the principles of SC. 1 a d/2 (d/2)−1 ∼ ξ (aτ) ∼ τ .h. The upper critical dimensionality dc = 4 is valid for all the universality classes (d. We postpone a discussion what is going on in a close vicinity of the critical point. the wave function being the same for all the Bose particles. GINZBURG-LANDAU THEORY OF SUPERCONDUCTIVITY (SC) 153 Let us denote a typical value of the fluctuation squared by (δm)2 . 0 Consequently. and one can expand the thermodynamic potential as Gs = Gn + a|Ψ|2 + |Ψ|4 + · · · . or at ¯ τ(d−4)/2 → m2 = ¯ 12aτ . the modulus of the order parameter is small. it is natural to assume that the ground state corresponds to the total momentum equal to 0. a g Now we can estimate when we can neglect 4th order terms in the expansion in powers of fluctuations.4 Ginzburg-Landau theory of superconductivity (SC) Let us formulate the theory of SC transition as the one for a general continuous transition.4. u a2 u g a d/2 . Ψ.s. all the critical exponents take their classical values. For d = 3 the criterion has the form τ u2 /sg3 . First. Tc b 2 .

. we get Gs − Gn = 2 (ατ)2 Hc = . Consequently. 2 = −(α/b)τ = |Ψ |2 at T < T . we describe the equilibrium values of the order parameter as Ψ=0 at T > Tc . To get the minimum we calculate variation with respect to Ψ∗ : dV 3 ατΨδΨ∗ + b|Ψ|2 ΨδΨ∗ + 1 2e 2e h −i¯ ∇ + A Ψ i¯ ∇ + A δΨ∗ h 4m c c α2 /2b = c1 g(εF ) (kB Tc )2 . c δG = dV b 1 ατ|Ψ|2 + |Ψ|4 + 2 4m −i¯ ∇ + h 2 H2 2e A Ψ + c 8π . Thus the coefficient a should change the sign at the transition point: a = ατ . Thus we should add both the energy of magnetic field H 2 /8π and the energy connected with the inhomogeneity. 3 The most interesting case is the case of external magnetic field. we get for δG = Gs − Gn −i¯ ∇ + h 2 2e A Ψ . The ratio α2 /2b can be related to the critical temperature Tc from the BCS theory of superconductivity. because of gauge invariance. CONTINUOUS PHASE TRANSITIONS According the definition of Tc . In this case Ψ is coordinate dependent. Here we come to the most important point: Cooper pairs are charged particles. Indeed Hc /8π is just the energy of magnetic field repelled from the sample in the Meißner state. where g(εF ) is the density of states at the Fermi level while c1 is a numerical constant. To make a proper dimension we write the corresponding term as 1 4m 0 Finally. Near the critical point it is enough to add |∇Ψ|2 . 0 Here Gn is the free energy of the normal state without magnetic field. above Tc the stable state corresponds to Ψ = 0.154 CHAPTER 12. α > 0. only combination 2e −i¯ ∇ + A h c is possible. 2b 8π 2 The last equality is a consequence of the definition of the critical field. |Ψ| c 0 Inserting the equilibrium value if |Ψ0 |2 into the expansion and comparing the result with the known relation for the critical field. As a result. while below Tc Ψ = 0.

c The variation with respect to δΨ leads to the complex conjugate expressions. To obtain it one should calculate the variation with respect to the vector potential A. Now it is important to get the equation for electric current. Using the Gauss theorem we get i¯ h 4m δΨ∗ −i¯ ∇ + h 2e 1 A Ψ dS + c 4m d V δΨ∗ −i¯ ∇ + h 2e A c 2 155 S Ψ. The variation of H 2 leads to δ(curl A)2 = 2 curl A curl δA. Then we put δΨ∗ = 0 at the surface and arbitrary inside. On the other hand. The second integral can be transformed into the surface one. and equating the total variation to zero we obtain j= ie¯ ∗ h 2e2 2 (Ψ ∇Ψ − Ψ∇Ψ∗ ) − |Ψ| A 2m mc This is just the quantum mechanical expression for the current for a particle with the charge (−2e) and the mass 2m. The set (1/4m) [−i¯ ∇ + (2e/c)A]2 Ψ + ατΨ + b|Ψ|2 Ψ = 0.4. As a result. Then we can use the relation div [a × b] = b curl a − a curl b to get δ(curl A)2 = 2δA curl curl A+2 div [δA × curl A] .12. The rest of the free energy can be calculated in a straightforward way. . If now we put δΨ∗ at the surface to be arbitrary. GINZBURG-LANDAU THEORY OF SUPERCONDUCTIVITY (SC) The item with the ∇δΨ∗ can be integrated by parts. thus it is not important. we obtain the boundary condition n −i¯ ∇ + h 2e A Ψ|S = 0. h j = (ie¯ /2m) (Ψ∗ ∇Ψ − Ψ∇Ψ∗ ) − (2e2 /mc)|Ψ|2 A h forms the Ginzburg-Landau equations. from the Maxwell equations curl curl A =curl H = 4π j c → δ(curl A)2 = 2jδA. we get the following equation for the order parameter 1 2e −i¯ ∇ + A h 4m c 2 Ψ + ατΨ + b|Ψ|2 Ψ = 0.

7: On the magnetic field penetration into a superconductor. (see Fig. 4π(2e)2 Ψ2 0 √ ατ Hc = 2 π √ . Ψ0 δ= H = H √ .156 CHAPTER 12. dx It is natural to consider Ψ as a function only of x. κ √ eHc δ2 κ=2 2 hc ¯ which is called the Ginzburg-Landau(GL) parameter. As a result. Solid line is the field profile while dashed one shows spatial dependence of the order parameter. To understand the meaning of the parameters let us consider the simplest problem. CONTINUOUS PHASE TRANSITIONS Dimensionless parameters and their meaning To make the equations simples new dimensionless parameters are usually introduced: Ψ = Ψ .7). We get H(x) = 1 d2Ψ − 2 2 − Ψ + A2 |Ψ| = 0 . b with 2mc2 . κ dx dΨ = 0. dx surface . We N z 2 Ψ − Ψ + |Ψ|2 Ψ = 0 . namely penetration of a weak field into the superconductor x > 0. In the following we will omit primes and use these variables. δ A = A √ Hc δ 2 α|τ| . i (Ψ∗ ∇Ψ − Ψ∇Ψ∗ ) − |Ψ|2 A = 0 2κ This set contains only one dimensionless parameter S Ψ H H Figure 12. b Ψ2 = 0 The quantity Ψ0 is a spatially-homogeneous solution of the GL equations in the absence of external magnetic field. Let H z. the set of equations can be expressed as i ∇−A κ curl curlA − with the boundary condition n i ∇ − A Ψsurface = 0 . A y. x ξ δ have dAy . Hc 2 r = r . 12.

2 If one keeps the external field H0 fixed the corresponding thermodynamic potential can be ob(0) 2 tained by subtracting H0 B/4π. GINZBURG-LANDAU THEORY OF SUPERCONDUCTIVITY (SC) 157 At κ 1 one can assume Ψ =const. Application of the GL theory Surface energy of N-S interface First. 2 dx2 dx κ dx dx dx dx dx 1 κ2 dΨ dx 2 We obtain dA Ψ4 + + Ψ (1 − A ) − 2 dx 2 2 2 = const = 1 2 . κ2 dx2 d2A − Ψ2 A = 0. while A ⊥ x. let us derive an auxiliary expression for the free energy which will be used below. As a result.12. H2 |Ψ|4 (ΩsH − ΩnH ) d V = c d V (H − H0 )2 − . The set of equations has the form (we’ll see that Ψ is real) 1 d2Ψ + Ψ(1 − A2 ) − Ψ3 = 0. As a result. the last one – by dA/dx and them add the equation and integrate over x: x dx 0 dΨ dΨ 3 d 2 A dA dA 1 d 2 Ψ dΨ +Ψ (1 − A2 ) − Ψ + 2 − Ψ2 A = 0. in usual units H = H0 e−x/δ . Immediately we get from the second GL equation d2A −A = 0 dx2 → H = H0 e−x . say x. (12. Using dimensionless notations we get (0) Ωs − Ωn H2 dV = c 4π dV |Ψ|4 −|Ψ| + + 2 2 2 i∇ − A Ψ + H2 . the surface integral being zero. we chose Ψ real. Let us multiply the first equation by dΨ/dx. In the rest integral we take into account that Ψ obeys the GL equation. in the bulk of material it is natural to assume |Ψ|2 = 1. dx surface It is easy to find the first integral of this set. In a normal phase. Thus δ is just the London penetration depth near Tc .5) 4π 2 Consider the problem where all the quantities depend only upon one co-ordinate.4. correspondingly ΩnH = Ωn − H0 /8π. dx2 dΨ | = 0. κ Then we integrate the item with ∇Ψ∗ by parts. Ωs − Ωn (0) dV = 2 Hc 4π dV H 2 − |Ψ|4 .

x → −∞ (N): Ψ = 0. Thus. We obtain H2 σns = c 8π In dimensional units that means √ 2 κx Hc 4 2 1 − tanh2 √ dx = . The result is that σns = 0 at κ = 1/ 2 √ and at κ > 1/ 2 it is negative. ξ is called the coherence length. (12. To find the surface energy we formulate the boundary conditions x → ∞ (S): Ψ = 1. H = 0. 2 it is wrong in the region where the field penetrates. dΨ/dx = 0. 8π 3κ 2 √ 2 4 2 Hc δ σns = . 3 8π κ √ The opposite limiting case can be treated only numerically. √ Now we can employ Eq.158 CHAPTER 12. we get 1 κ2 The solution is dΨ dx 2 + Ψ2 − Ψ4 1 = 2 2 → dΨ κ = √ (1 − Ψ2 ). CONTINUOUS PHASE TRANSITIONS (the constant being determined from the boundary conditions). √ dΨ/dx = 0. This property has very important implication on the magnetic field structure in superconductors. At κ 1 the most important region where A and H are small. but this region is small. H = H0 = 1/ 2. Quantization of magnetic flux Consider a hollow cylinder placed into a longitudinal magnetic field. Assume that the cylinder is Figure 12.8: On the flux quantization in a hollow cylinder in the Meißner state. H = A = 0. If the order parameter has the form Ψ = |Ψ|eiχ .5) and put H0 = 1/ 2. We observe that the typical length at which the order parameter is changed is (in dimensional units) ξ = κδ. dx 2 κx Ψ = tanh √ .

hi ({s}) = ∑ Ji j s j + hi . A given spin i “feels” the external field + the fields created by the neighbors. let us restrict ourselves with the generic Ising model with the partition function ZN = ∑ exp β ∑ Ji j si s j + β ∑ hi si s i< j i . For simplicity. h It is interesting that the effect has been predicted by F. m hc ¯ We can divide the current density by |Ψ| and integrate over a closed circuit in the bulk of superconductor: j=− j · dl e¯ h =− 2 |Ψ| m 0 Thus Φ = kΦ0 . Quantization of magnetic flux through a hollow cylinder has been experimentally observed and has many implications.07 · 10−7 G · cm2 . h In any case the quantized quantity is in fact fluxoid ∇χ dl which is equal to the total flux through the hole and the surface layer. ij where prime means that i = j. MEAN FIELD THEORY the current is 159 e¯ h 2 2e |Ψ| ∇χ + A .5 Mean field theory The mean field approximation This is a basic approximation which allows one to find main results regarding phase transitions leaving alone some subtle problems about correct critical exponents. dl · ∇χ + − 2πk 2e hc ¯ A dl .12. the flux quantization through the hole is not exact. As a result. i ij ij . The main idea is to look at the world from a point of view of a given spin. 12. He assumed the quantum to be π¯ c/e = 2Φ0 . flux Φ0 = π¯ c/e = 2. London (1950) before the concept of Cooper pairs.5. hi ({s}) → he = ∑ Ji j s j + hi = ∑ Ji j mi + hi . The first step is to replace the actual field by the average one.

h. • The mean field theory works better if the interaction range is large. At the critical point all 3 solutions merge.9 where the l. and that surface effects are not important.s. 12. We obtain ¯ m = tanh [β (Jm + h)] . to calculate mi = si .160 Then CHAPTER 12. Analysis of the free energy shows that the solution m = 0 is unstable.s. J¯ = ∑ Ji j . hi = h. and further simplifications are required. e. j Let us discuss the case h = 0. 12. the mean field theory can be used even when the order parameter is not small. In principle.6 Spatially homogeneous ferromagnets Assume the spatial homogeneity. The situation is illustrated in Fig. the other ones are degenerate at h = 0. So the critical exponents are classical. are plotted At small β (high temperatures) there is a unique solution m = 0. Then Ji j = J(ri . it is a set of many equations. m=0 → Tc = J¯ = ∑ Ji j . 4 4 At d = 1 the predictions of the mean field theory are wrong even qualitatively. while at low T (large β) 3 solutions exist. • The mean field approximation neglects the fluctuations of the order parameter. Before doing that we shall list some comments. Since mi = we have the equation mi = tanh β e 1 ∂ ln ZN = tanh(βhe ) i β ∂he i ∑ Ji j m j + hi j which should be solved for mi . • Contrary to the Landau theory. and r. j With nearest neighbor interaction and for a (hyper)cubic lattice J¯ = 2dJ where J is the nearest neighbor interaction. i e N s i=1 Then we have to close the scheme. i. This is why it can be used to discuss global properties of the phase diagram. CONTINUOUS PHASE TRANSITIONS ZN → Z e = ∑ eβ ∑i hi si = ∏ 2 cosh(βhe ) . That takes place at ¯ ∂ tanh(βJm) ∂m = βJ¯ = 1 . .h. r j ) = J(|ri − r j |).

K. van Leeuven. 1983). or the renormalization group transformation which transforms the Hamiltonian from one scale into one describing the system on a larger length scale. the critical state should have properties of a fractal. The N spins take the values si = ±1.M. Wilson who has developed a quantitative approach (Nobel Prize. Thus. 1 − βJ +β . Kadanoff.7 Renormalization framework The main idea (L. 12. The method is a mapping. Thus the fluctuations must have the similar appearance at all the scales. -1 Let us re-derive the same expression form another point of view which seems useful for following discussions.12. From the mean field equation we we can calculate the susceptibility χ= ∂m ∂h ∂m ¯ = cosh−1 [β (Jm + h)] βJ¯ ∂h β T ¯ = T − Tc . 1966) is that since ξ → ∞ at the critical point there is no intrinsic scale for the long range fluctuations. The proper way to perform such a scaling is to keep long-range (critical) fluctuations unchanged.6) . Geometrically speaking. Niemeijer and J. The Kadanoff’s idea was a starting point for K.9: Graphical solution of the mean field equation. and the critical exponent γ = 1. The mapping Let us consider a general class of Ising models. We shall briefly discuss only one way to perform such a scaling – the real space renormalization constructed by Th.7. T T Putting h = 0 and having in mind that at T > Tc m = 0 we obtain χ= In this way we obtain that susceptibility id divergent at the critical point. the Hamiltonian can be written as H (s) = ∑ Ki si + ∑ Kik si sk + ∑ Kikl si sk sl + · · · ≡ ∑ Kα Πα (s) . critical Hamiltonian should exist for the variables on any scale. i ik ikl α (12. RENORMALIZATION FRAMEWORK tanh βJm 1 T < Tc T=Tc T > Tc -m o mo m 161 Figure 12.

s Now we can determine a new Hamiltonian which is valid on the new scale as ∑ P(s . N→∞ N Here G is the free energy per spin.10: Cells in a triangular lattice. as shown in Fig. This mapping can be analytically expressed as a projection operator. 5 . Let us consider a triangle as a Figure 12. 0 otherwise Here i labels the cell while si are original spins in the cell i. s) = 1 . and there are 2N such terms. Then we can write. ∑k sk = 0. k ZN = ∑ ∑ P(s . s The equation (12.10. P(s . For every pair of sets α = β the spin sk summed over all the states gives zero. When β = α each term gives unity.6) into (12. Our goal is to map the effective Hamiltonian (12. s)e−βH (s) . new spin. s To prove this equation one can substitute (12. s G = −T lim 1 ln ZN .6) to another one relevant to a larger scale.162 CHAPTER 12. CONTINUOUS PHASE TRANSITIONS Let us fix the origin on the energy scale by the stipulation ∑ H (s) = 0 .6) can be inverted to express the coupling constants Kβ as5 Kβ = 1 Π (s)H (s) 2N ∑ β s (12. The “coarse-grained” spins are defined according to the algorithm that it is the same as the majority of spins in the covered triangle. i Pi = 1 if si = sgn (si + si + si ) 1 2 3 . s s ∑ P(s . s)e−βH (s) ≡ eβΩ1−βH (s ) . Consider spins on a 2D triangular lattice.7) The thermodynamics of the model is contained in the partition function.7). 12. s) = ∏ Pi . ZN = ∑ e−βH (s) .

s Here ZN is the partition function for the transformed Hamiltonian with N cells. s . Thus changing the temperature we can move radially in the e e plane of effective coupling constants Ke . G (K) = Ω1 + d G (K ) . Ω1 = Ω1 /N .7. In this √ way we map a triangular lattice on another triangular lattice increasing the scale by 3. at least in principle. so in shrinks by l after each iteration. Thus the Hamiltonian moves more and e e more away from criticality. K2 )-plane. Since N /N = l −d where l is the scaling factor. Eventually the point (K1 . So we can expect that singularities are the same for both Hamiltonians. To demonstrate the procedure. s) e−βH (s) = eβΩ1 ZN . In other words. Keeping only singular items we require 1 G (K) = d G (K ) .11). Since for every such direction there is a critical temperature. An example is a magnetic system in a zero magnetic field having nearest and nextnearest neighbor interaction. Suppose that we start from a supercritical Hamiltonian represented by point 1. so ξ(K ) = 1 ξ(K) . K2 ) = (0. So we know. 0) is reached. This is a unique definition. However while rescaling we use different yard sticks.7) we can determine a new set kα of the coupling constants. The spin-independent item is calculated from the initial configuration when the fluctuations are frozen. the 1 l . Since at the critical l point ξ(Kc ) = ∞ it follows that the critical Hamiltonian is mapped also on the another critical Hamiltonian. To discuss dynamics let us recall that the quantities Ke = −βK rather than K enter the the theory.12. RENORMALIZATION FRAMEWORK We have to introduce a spin-independent exponent βΩ1 to keep the property 163 ∑H s (s ) = 0 . Global properties of the mapping As a result of RG transform we get ZN = ∑ P(s . The correlation length of the initial Hamiltonian was finite. K = (K1 . and using the inversion scheme (12. the mapping K → K (K). The repeated RG transforms generate Hamiltonians 2. · · · . The origin of the (K1 . the e e subspace of the critical Hamiltonians is represented by a line of criticality in the (K1 . 3. while T = 0 corresponds to the infinity in the direction given by the ratio K2 /K1 (dashed line in Fig. The mapping H (s) → H (s ) is called the renormalization group transformation (RG). K2 ). 12. K2 )-plane corresponds to infinite temperature. let us choose the 2D set of the coupling constants. 4. l Before using the RG scheme let us recall that the physical correlation length remains the same. Thus it is not responsible for the singular behavior.

by subsequent transformations mapped on the Hamiltonians 2. H ∗ .12 The consideration above can be generalized for the case of multi- e K2 Figure 12.12: RG flow lines. This picture is usually represented as a RG flow diagram. 1.. An initial sub-critical Hamiltonian. is. is. That gives a new understanding of the concept universality class.. However. mapped on the Hamiltonians 2’. The full line defines the critical subspace. Near the fixed point one can linearize the mapping as δKα = ∑ Tαβ δKβ . Fig. Wilson. with fixed physical coupling constants. Following K. etc. 3’. An initial supercritical Hamiltonian. one can imagine several fixed points with its own basin of attraction. The critical Hamiltonian stays at the critical line. Initial Hamiltonians in this subspace converge to the fixed point. moving off to infinity. The dashed line represents the locus of the laboratory system. As we shall see. These considerations can be repeated for supercritical Hamiltonians which end at some of the fixed points at the infinity. the critical exponents are determined by the properties of RG transformation near fixed points. . with all the supercritical Hamiltonians belong to the basin of attraction associated with that fixed point.164 CHAPTER 12. so this line is an invariant space for RG transformation. e K1 dimensional parameter space which we shall not do here. e K2 1’ 1 2 3 4 2’ 3’ 4’ e K1 origin is a fixed point for the RG transformation. converging to the origin.11: Results of RG transformation. Fixed point properties The definition of fixed point is K∗ = K (K∗ ) . 4’. let us take for granted that under repeated mapping the critical Hamiltonian will converge to a fixed point Hamiltonian. β δK = K − K∗ . CONTINUOUS PHASE TRANSITIONS Figure 12. as the temperature changed. etc. 12. 1’. 4. 3.

Connection to Physics Scaling fields are convenient variable to analyze the equation 1 G (u ) . d l l This is nothing else then the Widom’s homogeneity postulate with replacement l d → λ. • Irrelevant. e. uτ → τ and uh → h. The left eigenvector belonging to the eigenvalue λi is defined as ˆ Φi · T = λi Φi . e. i. Thus we have G (τ. Usually 3 classes of the eigenvalues are discussed. yτ /d → a. l yh h) . RENORMALIZATION FRAMEWORK 165 ˆ ˆ Let us assume that eigenvalues of the matrix T are real and positive. [λi (l)]n = λi (l n ) . • Relevant. • Marginal. . yi > 0. with λi = 1. The solution of this equation has the form λi (l) = l yi where yi is the critical exponent. yi < 0. e. i.7. They grow exponentially and remove the system from the critical point. i. One can introduce the scaling field as ui = Φi · δK . with λi < 1.12. The corresponding scaling fields decay exponentially under RG transformations. yh /d → b . with λi > 1. h ) = d G (l yτ τ. yi = 0. Then one can assume that n repeated RG transforms with rescaling factor l are equivalent to a single transform with a scaling factor l n . Since T is asymmetric one must discriminate between left and right eigenvectors. h) = 1 1 G (τ . Then the rescaled field will be ˆ ui = Φi · δK = Φi · T · δK = λi Φi · δK = λi ui . ld G (u) = For magnetic system there are only 2 relevant fields.

Wilson. The precise conditions must be met to make the transform acceptable are less than clear. a seemingly fruitful way to think about critical phenomena. According to K. The main requirement is to leave long range fluctuations essentially untouched. .166 CHAPTER 12. the RG theory is only an appealing picture. “There is no RG cookbook!”. CONTINUOUS PHASE TRANSITIONS Summary At present time.

one can specify the coordinate r and the momentum p of the particle. Icoll ( fα ) = ∑ [Wαα fα (1 − fα ) −Wα α fα (1 − fα )] . fp (r. α Here α denotes the electronic state (in our case α ≡ {p.t).Chapter 13 Transport phenomena 13. We shall concentrate on the transport under influence of electric and magnetic field. 167 . while Icoll is the collision operator. s}). v ≡ ∂εp /∂p is the (group) electron velocity. According to classical physics. Here we denote (dp) ≡ d d p/(2π¯ )d . the current density being j(r.t) ∂ f ∂ f dr ∂ f dp ≡ + + + Icoll dt ∂t ∂r dt ∂p dt ∂f ∂f ∂f = + v + F + Icoll = 0 ∂t ∂r ∂p Here 1 F = e E + [v × H] c is the force acting upon the electrons.t) = e (dp) v fp (r. The distribution function is defined by the Boltzmann h equation d fp (r.t) .1 Classical transport Boltzmann equation Let us consider electrons in a solid state as an example. It expresses the changes in the state due to quantum collisions and can be expressed through the the transition probability Wi f between the initial state (i) and the final state ( f ). Thus one can calculate the non-equilibrium distribution function.

Since we assume E to be small. it is reasonable to search solution as fp (r) = f0 (εp ) + f (1) . Drude formula The simplest problem is to calculate the linear response to a small stationary electric field. E. τtr (1) 1 = Wpp τtr ∑ p pν − pν pν .168 CHAPTER 13. Substituting it into the collision operator we obtain a very important relation between the probabilities of the direct and reverse processes. pν = p cos φ . Using the expression for the relaxation time we can write fp = τtr e(E · v) − (1) d f0 (εp ) dεp → j = e2 (dp) v(E · v) τtr (εp ) − d f0 (εp ) dεp . ν ≡ E/E . | f (1) | f0 . p Since we are interested in the function odd in p (to create a current) and it must be a scalar it is natural to assume that (1) fp ∝ (p · ν) .1 we observe that pν = p cos θ cos φ . From the simple geometrical construction shown in Fig. Then. The quantity τtr is called the transport relaxation time. Wαα e−εα /kT = Wα α e−εα /kT . Under such an assumption Icoll ( f (1) fp )= . we can express the relaxation time in the form 1 1 = g(εp ) τtr 2 π 0 dθ sin θ (1 − cos θ)W (θ) . If the material is isotropic then W is dependent only on the scattering angle θ between p and p . 13. . Consequently. Since Icoll ( f0 ) = 0 we come to the integral equation Icoll ( f (1) ) = −eE ∂ f0 (εp ) d f0 (εp ) = eEv − ∂p dεp . (1) (1) Icoll ( f (1) ) = ∑ Wpp fp − fp . The linearized collision operator has the simplest form for elastic processes when Wpp = Wp p . TRANSPORT PHENOMENA In a stationary situation and in the absence of external fields the solution of the Boltzmann equation is the Fermi function.

the electro-chemical potential ζ + ϕ/e. Eq. Indeed. (dp)δ(ε − εp ) 1 1 = v2 τtr = v d d has an explicit physical meaning.2) is known as the Einstein relation. we arrive at the Ohm’s law. the phenomenological expression for the current density in the presence of the density gradient reads as j = σE − eD∇n . (13. j = σE with σ= ∞ 0 dε g(ε) v2 τtr E ε − d f0 (ε) dε . where D is th diffusion constant. At the same time. must be constant and ∇ϕ = −e∇ζ. while . for low temperatures we get σ = e2 g(εF )D(εF ) .1. .13. CLASSICAL TRANSPORT p’ 169 θ p Figure 13.2) On the other hand. . Thus we identify the quantity with the diffusion constant.1: On the calculation of the transport relaxation time. (13. φ ν As a result. In the equilibrium the current is zero and eD∇n = eDg(εF )∇ζ = −σ∇ϕ .1) ε Here vE means the projection of the velocity on the direction of the electric field. . This is just the diffusion constant for the electrons with a given energy ε. A(p) ε ≡ The quantity D = v2 τtr E ε (dp)A(p)δ(ε − εp ) . means the average over the surface of a constant energy ε. (13.

Another important assumption is that the system is connected to reservoirs by ideal quantum wires which behave as waveguides for the electron waves. then the conductance is 2 given by the expression G = 2e N. while its total entropy. |∂εn (kz )/∂kz | h If we take into account electron spin and N transverse modes are open. As we have seen. First of all let us specify what heat release is. S . δµ = δn/g(εF ). which is made only for the sake of . It will be convenient to consider an isolated system. Thus one can reduce the non-equilibrium transport problem to a quantum mechanical one. There is a capacitor which is discharged through the conductor of interest. The main principle of this approach is the assumption that the system in question is coupled to large reservoirs where all inelastic processes take place. The product RC of the whole system. This means that the result is independent of the assumption that the considered system is isolated. R and C being the resistance and capacitance respectively.2 Ballistic transport Landauer formula We start this chapter by a description of a very powerful method in physics of small systems so-called Landauer approach. i. U . even an ideal quantum wire has a finite resistance. Consequently. Ideal quantum wire Consider 2 large reservoirs of electron gas reservoirs having the difference δn in the electron density and separated by a pure narrow channel. So the difference between the chemical potential in α-th reservoir will be denoted as µα . the transport through the systems can be formulated as a quantum mechanical scattering problem. If the channel is long and uniform. is much bigger than any relaxation time characterizing the electron or phonon system of the conductor.170 CHAPTER 13. TRANSPORT PHENOMENA 13. through the applied generation is expressed through T ∂S voltage and characteristics of the nanostructure itself. In the following we shall use the Fermi level of non-biased system as the origin for the chemical potentials. Below we will discuss the physical picture of heat release by a current-carrying nanostructure. then the total current carried by the state characterized by a transverse mode n and a given direction of spin which propagates without scattering is Jn = e dkz ∂εn (kz ) 2 = 2π¯ ∂kz h 2π¯ h εF +µβ εF +µα dε ∂εn (kz )/∂kz 2 = δµ . This means that for all the practical purposes the conduction process can be looked upon as a stationary one. h We come to a very important conclusion: an ideal quantum wire has finite resistance h/2e2 N which is independent of the length of the wire. is conserved. That means a finite heat generation even in the absence of any inelastic processes inside the wire. The rate of heat ˆ /∂t. where T is the temperature. The total ˆ energy of the system.e. We start our analysis from the discussion of the properties of an ideal quantum wire. is growing. For small δn one can assume that there is a difference in a chemical potential. Therefore we will have in mind the following physical situation.

2 consisting of a barrier connected to reservoirs by ideal quantum wires. Tn . If there is some reflection only a part of the current is transmitted. also establish full equilibrium. That is the same mean free path that enters the Drude formula. which determines the conductivity of the reservoirs themselves. Tn = ∑ h n=1 h (13. and a magnetic rod can be drawn into the coil. This thermodynamically defined heat is generated in the classical reservoirs over the length having a physical meaning of the electron mean free path. The latter can in principle be measured by optical methods. BALLISTIC TRANSPORT 171 derivation. one can let the current flow through a coil. In this Figure 13. On the other hand. As a result of scattering. Only over those distances from the channel one can treat the results in terms of the true local temperature.13. to obtain (including spin degeneracy) N 2 J = δµ ∑ Tn . the equilibrium with respect to the lattice is established at the scales of electron-phonon and phononphonon mean free paths. G= 2e2 N 2e2 Tr tt† . and this means dissipation of mechanical energy into the heat. For instance. It is interesting to indicate that even purely elastic collisions can result in a heat generation although they of course cannot establish full equilibrium.3) . the amount of heat generated per second in both reservoirs being the same. All the above considerations do not mean that the collisions that give the main contribution to the heat release. It has been shown that the heat release is symmetric in both reservoirs even if the scatterers in the system are asymmetric. Such a distribution can be characterized by a local electro-chemical potential and sometimes an electron temperature. a local equilibrium electron distribution is established over the length scale determined by electron-electron interaction.2: On the resistance of a quantum resistor. case one can introduce the transmission probability of the mode n. 13. This has a clear physical meaning. The amount of order in the electron distribution resulting in electric current can bring about mechanical work. h n=1 As a result.2. What equilibrium needs is inelastic collisions which transfer the energy of electrons taking part in charge transfer to other degrees of freedom. In such a way the electrons transferring the current can execute a work on the rod. Resistance of a quantum resistor Consider a system shown in Fig. the amount of order in the electrons’ distribution diminishes. In particular. such as to other electrons and phonons.

The sources are assumed to be incoherent. The outgoing channels are fed up to thermal equilibrium population. . . we denote the proper amplitudes by primes. Ei + h2 ki2 ¯ = EF . . (13. the differences µ1 − µ2 are also assume small to yield linear transport.3). TRANSPORT PHENOMENA Here t is the matrix of scattering amplitudes while the expression is called two-terminal Landauer formula. this is the conductance which is measured between two reservoirs. j then from unitarity condition we get ∑ Ti = ∑(1 − Ri) .4) G= ∑ h n=1 1 − Tn However. 2 T 2e 2e T where R is the reflection coefficient. If we replace the incoming and outgoing channels. i i . Having in mind that the resistance of the connecting ideal wires (per one conducting mode) is h/2e2 we can ascribe to the scattering region the resistance h 1 h R −1 = 2 . Consequently. We introduce the scattering amplitudes ti j for the transmission from jth incoming to ith outgoing channel. Let each lead has the cross section A and have N⊥ transverse channels characterized by wave vectors ki so that. Now we derive the Landauer formula for finite-temperature and so-called multichannel case when the leads have several transverse modes. the quantity which is usually measured is given by Eq. Indeed. In a magnetic field the Onsager relation requires S(H) = S(−H). This very important and looking simple formula was confusing during a long period. 2m The incoming channels are fed from the electron baths with the same temperature and chemical potentials µ1 .172 CHAPTER 13. From the current conservation we must require unitarity while from time reversal symmetry ˜ S = S.. It one defines the total transmission and reflection into ith channel as Ti = ∑ |ti j |2 . We shall assume that the particles are absorbed in the outgoing baths. In this way it is convenient to introduce 2N⊥ × 2N⊥ scattering matrix as S= r t t r . (13. j Ri = ∑ |ri j |2 . in the original formulation the quantum resistance was described as 2e2 N Tn . Thus we have also SS = I where star stays for complex conjugation while tilde for ˜ transposition. µ2 . Consider ideal wires which lead to a general elastic scattering system. Reflection amplitudes ri j are introduces in a similar way for reflection into the ith incoming channel.

2.3 The conductance of QPC Figure 13.13. The quantization is not that accurate as in quantum Hall effect (about 1%) because of non-unit transparencies Tn and finite temperature. This is the two-terminal conductance measured between the outside reservoirs which includes contact resistances.3: A sketch of a QPC formed by split gates. A sketch of QPC is shown in Fig.short and narrow constrictions in 2d electron gas. . It is interesting to compare quantum and classical behavior of QPC. h π 1 dα cos α = W vF (δn) . π −π/2 2π π/2 Note that the integer part of the quantity kF W /π is just the number of occupied modes according to quantum mechanics. i Thus the conductance becomes G= 2e2 h dE − ∂f ∂E Tr tt† . BALLISTIC TRANSPORT 173 Since the densities of states in each channel are 1D like. In a classical picture one can write J = W (δn)vF Thus the “diffusion constant” is De f f = J 1 = W vF δn π → G = e2 g(εF )De f f = 2e2 kF W . gi (E) = (π¯ vi )−1 we write the current h through outgoing channels as I = = e π¯ ∑ h i dE f1 (E)Ti (E) + f2 (E)R (E) − f2 (E) dE − ∂f ∂E (µ1 − µ2 )e π¯ h ∑ Ti(E) . is quantized in the units of 2e2 /h. Point ballistic contact The most clean system is the so-called quantum point contact (QPC) . 13.

.5: On the classical conductance of a point contact. EF + eV α W EF Figure 13.4: Quantization of conductance of a point contact: Schematic picture (left) and experimental result (right).174 CHAPTER 13. TRANSPORT PHENOMENA 2 Conductance (2e /h) Gate Gate voltage Figure 13.

. . .pressure. p) .total entropy for the body and surrounding medium. S . U .state density.) .inverse temperature.mass of a molecule. Γ(E.(Helmholtz) free energy. q2 .volume of the box which contains the system. 175 . .generalized coordinates.classical Hamiltonian. P . E . β = 1/T . .number of states with the energy ≤ E .absolute temperature (we use also energy units). F .Appendix A Notations Microscopic variables q → (q1 .degeneracy of i-th state. T . q f ) . enthalpy. . H (q. .velocity of light. S . .) = ∂Γ/∂E = Tr δ E − H = V dΓ δ [E − H (p.entropy.conjugated momenta.Gibbs free energy. gi . . . c . Ω . . .quantum Hamiltonian.spin. Macroscopic variables V . 1 ˆ D(E. H . p → (p1 . f dΓ ≡ ∏ j=1 dq j d p j .Landau free energy.internal energy. p f ) . m . G . p2 .thermal function.element of phase space. ˆ H . . thermodynamic potential. q)] . St . .total energy.

6 at n 1. e .Bohr magneton. ¯ APPENDIX A.magnetic field.176 Electric and Magnetic Variables B . P.Planck constant. P.1: Thermodynamic potentials (summary) Basic distributions The Gibbs distribution wn = 1 −βEn e . s . N T. V. h . Thermodynamic relations Thermodynamic potentials Thermodynamic potential Internal energy Heat function (enthalpy) Helmholtz free energy Gibbs free energy Landau free energy Notation U H F G Ω Independent variables S.38 × 10−16 erg/K . µ = e/2mc . V. N T. µ Differential T dS − P dV + µ dN T dS +V dP + µ dN −S dT − P dV + µ dN −S dT +V dP + µ dN −S dT − P dV − N dµ Table A. h ≡ h/2π . Z Z = ∑ e−βEs . N S.electronic charge.Boltzmann constant. Constants k = 1. V. N T. NOTATIONS Useful mathematical formulas The Stirling’s formula: log n! ≈ n log n − n + 1 or more accurate n! ≈ 2π n + 1 nn e−n .

Relations between thermodynamic and statistical quantities Z = ∑ e−βEn . n F = −T log Z . .177 The Maxwell-Boltzmann distribution f (p. Ω = −T log ∑ eβµN ∑ eβEsN N s . . f p (p) = (2βπm)−3/2 e−β(px +py +pz )/2m . fq (q) = The Fermi-Dirac distribution fk = eβ(εk −µ) + 1 The Bose-Einstein distribution fk = eβ(εk −µ) − 1 dq e−βV (q) −1 2 2 2 β ≡ 1/T . q) = f p (p) · fq (q) . −1 The chemical potential µ is determined by the normalization V∑ k gk β(εk −µ) ± 1 e = N. e−βV (q) . −1 .

NOTATIONS .178 APPENDIX A.

1) If Y. We obtain ∂X ∂Z ∂Z ∂Y dZ = 0 .Y.2 by dX + − X ∂Z ∂Y X and subtract the results. while Z = Z(X. Consider 3 quantities. Now let us consider X. B. X Y ∂Z ∂Y ∂X ∂Y − Z Y Since dX and dY are independent. Y −1 (B. then −dX + Now we multiply Eq. X. X (B.Y as independent variables.3) Z X ∂X ∂Y = Z . Z (B. Z. Z) = const.Y. related by the equation of state K (X. Z are taken as independent variables. We have ∂Z ∂X dX + Y ∂Z ∂Y dY − dZ = 0 .Appendix B On the calculation of derivatives from the equation of state Important relations arise from the properties of partial differentiations. X = −1 .2) Z and Eq.1 by ∂Z ∂X ∂X ∂Y + Z ∂X ∂Y ∂X ∂Y dY + Z ∂X ∂Z dZ = 0 Y (B. B.Y ).4) 179 . this equation is compatible if ∂Z ∂X ∂X ∂Y or ∂X ∂Y ∂X + Y ∂Y Z ∂X + ∂Z Y Z ∂Y ∂Z ∂Z ∂Y ∂Z ∂Y ∂Z ∂X ∂Y ∂X = 0 X = 0.

(B.180APPENDIX B. .4) and the Maxwell relations are usually used for transformations and computation of derivatives from the equation of state.Y ). (dY = 0). We get ∂F ∂Z = Y ∂F ∂X . (B. ∂F ∂X = Z ∂F ∂X + Y ∂F ∂Y X ∂Y ∂X .6) Equations (B.3). ON THE CALCULATION OF DERIVATIVES FROM THE EQUATION OF STATE General scheme for transformation: can be expressed as dF = Consider any quantity F (X. the differential of which ∂F ∂X dX + Y ∂F ∂Y ∂X ∂Z dX . Y (B.5) Y Another important relation arises if one divides the expression for d F by xX. X Then we divide it by dZ and assume Y = const.5). (B.6) together with Eqs. Z (B.

Lifshitz.Tokyo). D.Bibliography [1] L. Landau and E.New York . 1980. Part 1 (Pergamon Press. Statistical Physics. M.Seoul . 3rd edition. Oxford . 181 .

Sign up to vote on this title
UsefulNot useful