This action might not be possible to undo. Are you sure you want to continue?

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

Leonard Susskind

Felix Bloch Professor of Physics

Stanford University

Warren C. Gibson, editor

Santa Clara University

September, 2009

Preface

A number of years ago I became aware of the large number of physics enthusiasts

out there who have no venue to learn modern physics and cosmology. Fat advanced

textbooks are not suitable to people who have no teacher to ask questions of, and the

popular literature does not go deeply enough to satisfy these curious people. So I started

a series of courses on modern physics at Stanford University where I am a professor of

physics. The courses are speciﬁcally aimed at people who know, or once knew, a bit of

algebra and calculus, but are more or less beginners.

This volume is based on notes prepared for a Continuing Education course I gave at

Stanford in the spring of 2009, and for a graduate course in statistical mechanics that

I gave at that same time.

– Leonard Susskind

I am one of those “physics enthusiasts,” an engineer who grew tired of mass-market

physics books shorn of all equations. I picked up the standard “Gravitation” text to try

to teach myself general relativity, but even though I have a fairly good math background,

it was too much. So I was delighted to ﬁnd Prof. Susskind’s course sequence at Stanford

and to have the opportunity to help with the preparation of this volume.

– Warren Gibson

Contents

1 Conservation of Information, E, S, and T 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Probability Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.3 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.4 Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.5 Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.6 The Zeroth and Second Laws of Thermodynamics . . . . . . . . . . . . . 8

2 The Boltzmann Distribution 9

2.1 Solving for the Lagrange Multipliers . . . . . . . . . . . . . . . . . . . . 12

2.2 Helmholtz Free Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.3 Why is T = 1/β? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.4 Ensembles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3 Fluctuations 15

4 Control Parameters 17

4.1 The Adiabatic Theorem and the First Law . . . . . . . . . . . . . . . . 17

4.2 Processes at Fixed Temperature . . . . . . . . . . . . . . . . . . . . . . . 19

5 Dilute Gas 23

5.1 Ideal Gas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

5.2 Almost Ideal Gas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

5.3 Ideal Gas in a Potential Field . . . . . . . . . . . . . . . . . . . . . . . . 26

5.4 Diatomic Molecular Gas . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

5.5 A Note about Volume and the van der Waals Gas . . . . . . . . . . . . 29

6 Simple Magnet and Magnetization 31

6.1 Ising Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

7 The Maxwell Relations 34

8 The Second Law 37

9 Quantum Considerations 41

10 Thermal Radiation 44

11 Chemical Potential and Grand Canonical Ensemble 46

12 Back to Magnets 49

12.1 Connection with Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

12.2 Low Temperature Expansion . . . . . . . . . . . . . . . . . . . . . . . . 55

12.3 Ising-like Systems in Higher Dimensions . . . . . . . . . . . . . . . . . . 57

12.4 High Temperature Expansion . . . . . . . . . . . . . . . . . . . . . . . . 62

12.5 One- and Two-Point Functions in the HTE . . . . . . . . . . . . . . . . 66

13 High-T/Low-T Duality 67

14 What Happens at T = T

C

? 72

14.1 High Temperature Expansion of the Correlation Function . . . . . . . . 72

14.2 Approach to T

c

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

14.3 Renormalization Group . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

14.4 Mean Field Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

15 Bose Einstein Condensation 83

A Appendix A: Some basic mathematics 88

B Appendix B: Phase space 92

C Appendix C: Ludwig Boltzmann 93

D Appendix D: for Engineers 95

List of Figures

1 Two simple “laws of motion.” . . . . . . . . . . . . . . . . . . . . . . . . 2

2 A law of motion with two trajectories. . . . . . . . . . . . . . . . . . . . 3

3 Sharply peaked occupation numbers. . . . . . . . . . . . . . . . . . . . . 10

4 Experiment Illustrating First Maxwell Relation . . . . . . . . . . . . . . 35

5 The Second Law. The phase space volume and topology are exactly

conserved but when coarse grained it grows. . . . . . . . . . . . . . . . . 39

6 Magnetization vs external Magnetic ﬁeld. . . . . . . . . . . . . . . . . . 51

7 Phase diagram of a magnet. The thick line represents the discontinuity

in the magnetization. In the 1-D Ising model the critical temperature

T

∗

is zero. In the analogy with molecular ﬂuids, negative magnetization

represents the gas phase, and positive magnetization represents the liquid

phase. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

8 2-dimensional lattice of spins. The o’s indicate up-spins, the x’s indicate

down spins and the •’s indicate dummy up spins. . . . . . . . . . . . . 57

9 The contributions to the HTE coming from a single link are represented

by a heavy line on that link. . . . . . . . . . . . . . . . . . . . . . . . . 63

10 Closed “curves” on the lattice . . . . . . . . . . . . . . . . . . . . . . . 65

11 The dual lattice is composed of points at the centers of the Plaquettes. 67

12 Surround each island of ﬂipped spins with a curve on the dual lattice. . 68

13 Two canceling graphs. . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

14 Another example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

15 Points a and b are separated along the x-axis by distance D and along

the y-axis by distance n. . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

16 The mean ﬁeld consistency relation at high T (left) and low T (right).

For low T the tanh-curve and the linear curve intersect twice: once for

positive s and once for negative s. . . . . . . . . . . . . . . . . . . . . . 79

17 A sawtooth wave approximated by one-term and four-term Fourier series. 90

18 Ludwig Boltzmann’s tomb, inscribed with S = log P . . . . . . . . . . . 93

Part I

General Principles

1

1 Conservation of Information, Energy, Entropy, and

Temperature

1.1 Introduction

Statistical mechanics is often thought of as an approach to thermodynamics, or some-

what more generally, a theory of how atoms combine to form gases, liquids, solids, and

even plasmas and black body radiation. But it is both more and less than that. It is

more because it can be used in many areas of science where a large number of vari-

ables is involved – too many to address individually. In these cases statistical methods

can be used to study the aggregate behavior of a system without knowing (or indeed

caring about) the behavior of each element of the system. My son, who studies neural

networks, uses it. “Financial engineers,” often refugee physicists who sought fame and

fortune on Wall St., perhaps with dubious results, also use statistical mechanics.

∗

So to say that Statistical Mechanics is the theory of gasses is rather like saying

calculus is the theory of planetary orbits. That’s how calculus may have started, but

it has of course found vastly more applications. Although our examples will focus on

thermodynamics and magnetics, other applications are possible.

1.2 Probability Theory

There are many good texts that explain the theory of probability, but for our purposes

a few simple concepts will suﬃce. We will review them now starting with coin ﬂipping.

The probabilities of a coin coming up heads (H) and tails (T) are both equal to 1/2.

Why do we say that? Taking an a priori view, we can say that the physical symmetry of

a coin allows us no reason to believe that either outcome is more likely than the other,

so we must assign 1/2 to each probability. Taking an a posteriori view, we might ﬂip

a coin several thousand times and observe that we got heads (or tails) very nearly half

the time and conclude that this will always be the most likely outcome.

Here is another example. Let’s take a die (as in dice) and color the six faces red,

yellow, blue, green, orange, and purple (R, Y, B, G, O, P). The cubic symmetry of the

die dictates that the probabilities of any one face appearing on top when the die is rolled

∗

J. Volt, “The Statistical Mechanics of Financial Markets,” Springer, 2005

2 1 CONSERVATION OF INFORMATION, E, S, AND T

are all 1/6. Again, this is obvious from physical symmetry, or could be inferred from

experiments. But what if we don’t have an obvious symmetry to rely on? How do we

assign a priori probabilities?

Suppose for example instead of the coloring scheme that I indicated above, I chose to

color the purple face red. Then there would be only ﬁve colors. Would the probability

of throwing a given color be 1/5? After all, if I just write (R, Y, B, G, O), the 5 names

are just as symmetric as the original 6 names. Nonsense, you say: the real symmetry is

among the 6 faces, and that is so. But what if there really is no obvious symmetry at

all, for example if the die is weighted in some unfair way?

If we still wanted to pursue a theoretical or a priori approach we would have to rely

on a bunch of details such as the precise way the die was thrown by the hand that threw

it, the wind, maybe even the surface that the die lands of (can it bounce?). As is often

the case, we have to think of the system in question as part of a bigger system. But

what about the bigger system? How do we assign its probabilities?

Here is another idea that involves some dynamics. Suppose there is a rule that

uniquely speciﬁes the conﬁguration of a system based solely upon its current conﬁgura-

tion. We may call this rule a “law of motion,” with time in this example proceeding in

discrete steps. For example R →B, B →G, G → P, P →O, O →Y , Y → R is such

a rule (Figure 1). What fraction of the time does that die spend in each conﬁguration?

Figure 1: Two simple “laws of motion.”

The answer is 1/6. In fact there are many possible laws for which the answer will be the

same, such as R → B, B → G, G → P, P → O, O → Y , Y → R. or R → Y , Y → P,

P →G, G →O, O →B, B →R.

But what about the rule R → B, B → G, G → R, P → O, O → Y , Y → P

1.2 Probability Theory 3

(Figure 2)? In this case there are two trajectories through the space of states. If we

Figure 2: A law of motion with two trajectories.

are on one of them we don’t ever jump to the other. So the probability will depend on

where we started. We may wish to assign probabilities for beginning on either of the

two cycles.

In this last case there is a conserved quantity. Suppose we assign the number one

to R, B, G and zero to O, Y, P. Let’s call this quantity the “Zilch number.” Obviously

Zilch is conserved. Whenever we have a conserved quantity like Zilch, we have to specify

its value. That’s not a problem, as there are many conserved quantities in nature, the

most important in statistical mechanics being energy. Once we specify all the conserved

Zilch numbers, we can proceed as usual, and say that all the states on a trajectory with

a particular Zilch numbers are equally probable.

That sounds good, but there are lots of counterexamples. Here is one: R → R

B →R, G →R, P →R, O →R, Y →R. No matter where you begin, you go to red in

the next instant and then stay there. There are no conserved quantities but obviously

after the ﬁrst step the probabilities are completely unequal; only red is possible.

This last law has something odd about it. It is an example of a law that does

not respect the “conservation of distinction,” or “conservation of information.” In the

previous examples distinct starting points always led to distinct outcomes. But in this

case we quickly lose track of where we started. Trajectories merge!

One of the central principles of classical mechanics (and it has an analog in quantum

mechanics) is that information (distinctions) is conserved. This principle is so important

to the validity of statistical mechanics and thermodynamics that I would call it the

“minus ﬁrst” law of thermodynamics (the “zeroth law” is already taken.)

Later we will use an abstract concept called “phase space” in which the coordinates

4 1 CONSERVATION OF INFORMATION, E, S, AND T

represent not physical space, but variables that may be unrelated to space. For now,

let’s just note that conservation of distinctions says that trajectories in phase space

never run into each other. Even stronger, if you start with a volume in phase space

and follow it using the Hamilton’s equations of motion, the volume is conserved. That

suggests that for a closed system, a priori probability is uniform over the phase space.

We will come back to this.

1.3 Energy

The First Law of Thermodynamics says is that energy is conserved. For a closed system,

†

dE

dt

= 0

If a system is composed of two very weakly interacting subsystems,

‡

the energy is

additive:

E = E

1

+E

2

Interactions between the subsystems can exchange energy between them. Thus the

energy of either subsystem is not conserved. In equilibrium the energy of a subsystem

ﬂuctuates although the total energy is ﬁxed.

1.4 Entropy

Entropy is an information-theoretic concept. It is a measure of how much we don’t know

about a system. Our ignorance may stem from the fact that the degrees of freedom are

too small to measure, too numerous, or too rapidly changing. The reason could just be

our own laziness. In any case, entropy measures our ignorance.

Suppose a system can be in any of N states. If we know nothing about the state of

the system (complete ignorance) the entropy S is deﬁned as

S = log N.

†

A closed system is one with a time-independent Hamiltonian and with insulating walls so that no

energy can enter or leave the system.

‡

Weakly interacting means that the energy of interaction is very small compared to the separate

energies of the subsystems by themselves. If the Hamiltonian is H1 + H2 + Hint, then the numerical

value of Hint should be negligible compared with either H1 or H2.

1.4 Entropy 5

Suppose we know only that the system is an any of M states with M < N. Then we

have some nontrivial knowledge. In this case the entropy is deﬁned to be

S = log M.

Evidently, the more we know, the smaller the entropy.

Here are some further examples. Consider a system of n distinguishable coins. Each

coin is in one of two states: H (heads) or T (tails). A state consists of an ordered

sequence of H’s and T’s such as

H H T H T T H

Suppose we know nothing at all. The total number of possible states is 2

n

and the

entropy is

S = log 2

n

= nlog 2.

Note that by taking the log we have made entropy an extensive property, that is,

proportional to the number of degrees of freedom n.

Entropy is measured in bits. If the entropy is S = log 2

n

= nlog 2, one says that

there are n bits of entropy.

As another example, suppose we know everything that there is to know about the

coins. In that case we know the exact state and M = 1. The entropy is log 1 which is

zero.

Next, suppose we know that n − 1 coins are in the state H and one is T, but we

don’t know which one. The system can be in any of n states and the entropy is

S = log n

In this case the entropy is not additive but that is because there are strong correlations

(to be deﬁned later).

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 1.1 Suppose that n is even and that we know that half the coins are H and

half are T, but that is all we know. What is the entropy?

Problem 1.2 For large n what is the entropy if we know that a fraction f of the coins

are H and fraction 1 −f are T?

6 1 CONSERVATION OF INFORMATION, E, S, AND T

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

In each case we are given a probability distribution on the space of states. If we

label the states with index i then the probability that the system is in state i is called

P(i). In the examples P(i) is zero for those states that are not possible and P(i) = 1/M

for the possible states. We can write a formula for S in terms of P(i),

S = −

¸

i

P(i) log P(i). (1.1)

Those states with P = 0 contribute nothing and those with P = 1/M contribute

(1/M) log M. Since there are M such states the sum gives log M.

For a general probability distribution we require only that P(i) ≥ 0 and

¸

i

P(i) = 1. (1.2)

The general deﬁnition of entropy is given by (1.1). Obviously each P(i) satisﬁes 0 ≤

P(i) ≤ 1 which implies that S is positive and that each contribution to (1.2) is positive

or zero.

The entropy deﬁned by (1.1) roughly measures the log of the number of states that

have non-negligible probability in the distribution P(i). In other words e

S

is the “width”

of the distribution. Note that the width is zero if and only if P = 1 for a single state

and zero for all others.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 1.3 Suppose you have a set of N coins. Each coin independently has a

probability 1/3 to be heads and 2/3 to be tails. What is the total entropy?

Problem 1.4 A variable q has a uniformly spaced spectrum of values with very small

spacing δ. The sum over states can be accurately approximated by

¸

i

→

1

δ

dq

Suppose the probability distribution for q is proportional to e

−q

2

. What is the entropy?

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

1.5 Temperature 7

1.5 Temperature

The average energy associated with a probability distribution (call it E) is given by

E =

¸

i

E

i

P(i) (1.3)

where E

i

is the energy of state i.

Now suppose we have a one-parameter family of probability distributions labeled

by the average energy P(i; E). For each value of E, P(i; E) satisﬁes the usual require-

ments of a probability distribution. Later we will think of it as the thermal equilibrium

distribution for given average E. But for now it is just a one-parameter family.

At each value of E we can compute the entropy so that S becomes a function of E,

S = S(E).

Consider the amount of energy that is needed to increase the entropy by one bit (by

log 2). It is given by

δE =

dE

dS

log 2 (1.4)

We call the quantity dE/dS the temperature, T.

T ≡

dE

dS

(1.5)

We should think of temperature as follows:

§

Apart from a factor of log 2, the temperature of a system is the amount of

energy needed to increase its entropy by one bit.

For example, if you erase a bit of information from your computer you are really trans-

ferring it from the computer out into the atmosphere, where it shows up as some heat.

How much heat? The answer is T = log 2.

Except in very unusual circumstances, the temperature is always positive, i,e,. en-

tropy is a monotonically increasing function of energy.

§

Our deﬁnitions of temperature and entropy diﬀer in essentially unimportant but potentially con-

fusing ways from engineering deﬁnitions of such quantities. See Appendix D.

8 1 CONSERVATION OF INFORMATION, E, S, AND T

1.6 The Zeroth and Second Laws of Thermodynamics

The Second Law of Thermodynamics says that entropy always increases. We can state

it more formally as

When a closed system which is out of thermal equilibrium comes to equilib-

rium the entropy of the system increases.

We will come back to the reasons for this, but for now let’s just accept it.

From the Second Law we can prove that heat always ﬂows from hot to cold. Consider

two isolated systems , A and B, at diﬀerent temperatures. Their energies, temperatures,

and entropies are E

A

, E

B

, T

A

, T

B

, S

A

, S

B

. Without loss of generality we can assume that

T

B

> T

A

.

Now bring them into contact so that energy (heat) ﬂows between them. Suppose

a small quantity of energy is exchanged. The total change in energy must be zero.

Therefore

T

A

dS

A

+T

B

dS

B

= 0 (1.6)

Since they must tend to equilibrium, if the entropy is not maximum it must increase.

Hence

dS

A

+dS

B

> 0 (1.7)

We can use (1.7) to eliminate S

B

from (1.6). We ﬁnd

(T

B

−T

A

)dS

A

> 0 (1.8)

Since the system B is initially the hotter of the two, (T

B

− T

A

) is positive. Therefore

dS

A

and also T

A

dS

A

are positive. Equation (1.6) then tells us that T

B

dS

B

is negative.

Equivalently, energy ﬂows from hot to cold as equilibrium is established. The ﬁnal

equilibrium conﬁguration in which energy has stopped ﬂowing must have T

A

= T

B

. In

other words, temperature must be uniform in a system in thermal equilibrium.

9

2 The Boltzmann Distribution

If a system A is in contact with (weakly interacting with) a much larger system called

a bath, then it can exchange energy with the bath. After a long time the combined

system will have jumped around over all states having the given total energy. Each

of these ways in which the energy can be shared will be equally probable. This is the

assumption of chaos and although it is very diﬃcult to prove for a given system, it is

almost certainly true for almost all systems. Whenever it is true, the small system A will

be in thermal equilibrium with a certain probability distribution for being in the state

i. (Note that i refers to states of the small system A and not the combined system.)

We will illustrate the consequences of this principle by choosing the bath to be

N −1 copies of A so that altogether we have a total system consisting of N copies of A.

The copies are labeled 1, 2, 3, , N and they are in states i

1

, i

2

, i

3

, i

N

with energies

E

i

1

, E

i

2

, E

i

3

, E

i

N

.

Let us deﬁne the “occupation numbers” n

i

to be the number of copies of A that

occupy the state i. The n

i

satisfy two conditions,

¸

i

n

i

= N (2.1)

and

¸

i

n

i

E

i

= E

total

. (2.2)

These equations state that the total number of copies adds up to N and that the total

energy adds up to some ﬁxed value, E

total

. Keep in mind that i sums over states, not

subsystems.

The assumption of chaos tells us that all of the conﬁgurations of the total system

with a given total energy are equally likely. Thus the probability for a given partitioning

of the energy (given set of n

i

) is equal to the number of conﬁgurations with that set

of n

i

. How many distinct conﬁgurations of the total system are there with a given set

(n

1

, n

2

, n

N

)? This is a combinatoric problem that we will leave to you. The answer

is

N!

n

1

!n

2

! n

N

!

=

N!

¸

i

n

i

!

(2.3)

The important point is that when N and n become large, subject to the constraints

(2.1) and (2.2), the quantity in (2.3) becomes very sharply peaked around some set of

10 2 THE BOLTZMANN DISTRIBUTION

occupation numbers as in Figure 3 . Before we compute the occupation numbers that

Figure 3: Sharply peaked occupation numbers.

maximize (2.3), let us introduce some changes of notation.

Deﬁne P(i) to be the fraction of copies in state i.

P(i) =

n

i

N

and let E be the average energy of a copy. Obviously

E =

E

total

N

.

Now (2.1) and (2.2) take a form identical to (1.2) and (1.3), namely

¸

i

P(i) = 1

and

¸

i

P(i)E

i

= E.

11

Now we will assume that N and n

i

are very large and use Stirling’s approximation

¶

in

(2.3). But ﬁrst let us take its logarithm. (Maximizing a positive quantity is the same

as maximizing its log.)

log

N!

¸

i

n

i

!

≈ N log N −

¸

i

n

i

log n

i

.

We want to maximize this subject to the constraints (2.1) and (2.2).

Substituting n/N = P and E

total

= NE we ﬁnd that this is equivalent to maximizing

−

¸

i

P(i) log P(i)

subject to

¸

i

P(i) = 1

and

¸

i

P(i)E

i

= E.

In other words the probability distribution for thermal equilibrium maximizes the en-

tropy subject to the constraint of a given average energy.

In order to ﬁnd an expression for P(i) that provides a constrained maximum entropy

value, we use the method of Lagrange multipliers (Appendix A). We introduce two

multipliers α and β, one for each constraint. We maximize

−

¸

i

P(i) [log P(i) −1] −α

¸

i

−E

total

P(i) −β

¸

i

P(i)E

i

(2.4)

At the end we ﬁnd α, β values that satisfy the constraints.

Diﬀerentiating with respect to P(i) and setting the result to zero gives

P(i) = e

−(α+1)

e

−βE

i

. (2.5)

This is the Boltzmann distribution.

Let us deﬁne

Z ≡ e

(α+1)

Then (2.5) has the familiar form

P(i) =

e

−βE

i

Z

. (2.6)

¶

Stirling’s approximation is n! ≈ n

n

e

−n

or nlog n! ≈ log n −n = n(log n −1).

12 2 THE BOLTZMANN DISTRIBUTION

2.1 Solving for the Lagrange Multipliers

Solving for α is equivalent to solving for Z which is done by setting the sum of the P

to 1. This gives the famous formula for the partition function,

Z =

¸

i

e

−βE

i

. (2.7)

The partition function is a generating function that is used to derive several useful

quantities, like a potential function. Numerical values of Z are generally not useful.

Next consider the equation

¸

i

P(i)E

i

= E

which becomes

1

Z

¸

i

e

−βE

i

E

i

= E.

Let us use e

−βE

i

E

i

= −∂

β

e

−βE

i

and we obtain

E = −

1

Z

∂

β

Z = −

∂ log Z

∂β

. (2.8)

This equation can be used to ﬁx β in terms of the average energyE.

**The Lagrange multiplier β is, of course, the inverse temperature. We will demon-
**

strate this, but ﬁrst let us derive another familiar thermodynamic formula.

2.2 Helmholtz Free Energy

Using S = −

¸

P log P we can write the entropy in the form

S =

¸

i

1

Z

e

−βE

i

(βE

i

+ log Z) = βE + log Z

The quantity A = −T log Z is called the Helmholtz free energy (reminder: T = 1/β.)

Thus we ﬁnd

S = −β(E −A)

or

A = E −TS (2.9)

**The notation ∂xy is shorthand for ∂y/∂x.
**

2.3 Why is T = 1/β? 13

In its diﬀerential form,

dA = dE −TdS −SdT

Using dE = TdS we ﬁnd

dA = −SdT (2.10)

A is called free energy because it represents the amount of useful work that can be

extracted from a system. The customary symbol A comes from the German arbeit,

work.

2.3 Why is T = 1/β?

We have proposed two deﬁnitions of temperature. The ﬁrst is equation (1.5) and the

second is the inverse of the Lagrange multiplier β. We would like to see that they are

really the same.

Consider a small change in the energy of a system,

dE = −d[∂

β

log Z]

Now use (2.9) in the form E = ST −T log Z to get

dE = TdS +SdT −T

d log Z

dβ

dβ −log ZdT

Using (2.9) again, and the deﬁnition of A, we ﬁnd that the last three terms cancel,

leaving

dE = TdS. (2.11)

This is of course equivalent to (1.5) and thus the two deﬁnitions of temperature are the

same.

2.4 Ensembles

For a closed system the energy does not ﬂuctuate. In that case the probability distri-

bution is a delta function of energy,

∗∗

P(E) = δ(E −E

0

)

∗∗

The Kronecker delta δ(x) is equal to one when x = 0 and zero otherwise.

14 2 THE BOLTZMANN DISTRIBUTION

This is called the Microcanonical Ensemble and governs the entire system including the

heat bath.

For a system embedded in a large heat bath the energy ﬂuctuates and the Boltzmann

probability distribution is called the Canonical Ensemble. Later we will also allow the

number of particles to ﬂuctuate. The ensemble in that case is called the Grand Canonical

Ensemble.

15

3 Fluctuations

Thus far we have been deriving classical thermodynamics from statistical mechanics.

We go beyond thermodynamics when we consider ﬂuctuations of quantities about their

averages. Such ﬂuctuations are observable – for example in Einstein’s theory of the

Brownian motion. In this section we will illustrate this by considering the ﬂuctuations

of the energy of a system in contact with a heat bath.

Given a probability distribution P(x) the ﬂuctuation in x (called ∆x) is deﬁned by

(∆x)

2

= '(x −'x`)`

2

(3.1)

which is also equal to

(∆x)

2

= 'x

2

` −'x`

2

(3.2)

where 'X` means the average value of X. For any function f(x) the average is deﬁned

by 'f(x)` =

¸

x

f(x)P(x). If x is continuous then the sum is replaced by an integral in

the obvious way.

Let us consider the ﬂuctuation of energy of a system in equilibrium. We use the

following:

'E` = −

1

Z

∂

β

Z

and

'E

2

` =

1

Z

∂

2

β

Z.

The ﬁrst identity is the usual identiﬁcation of average energy in terms of the derivative of

Z. The second identity is derived the same way as the ﬁrst, noting that each derivative

of e

(−βE)

brings down a factor of −E. Thus

(∆E)

2

= 'E

2

` −'E`

2

is given by

(∆E)

2

=

1

Z

∂

2

β

Z −

1

Z

∂

β

Z

2

.

But this expression is equivalent to

(∆E)

2

= ∂

2

β

log Z = −∂

β

E

Using T = 1/β we get

(∆E)

2

= T

2

dE

dT

.

16 3 FLUCTUATIONS

Now note that dE/dT is just the heat capacity of the system. Call it C. The ﬁnal

identity is

(∆E)

2

= T

2

C. (3.3)

Thus we ﬁnd that the ﬂuctuation of the energy is proportional to the speciﬁc heat.

It may seem odd that the ﬂuctuations should appear to be so large. But that is

because we have set the Boltzmann constant to 1. If we put k back into the equation it

becomes

(∆E)

2

= kT

2

C. (3.4)

k is a very small number in units of meters, kilograms, seconds, and degrees Kelvin:

k = 1.4 10

−23

. For a kilogram of water at room temperature,

∆E =

1.4 10

−23

J/K(293K)

2

4.18J/(kg K) 1kg = 2.2 10

−9

J

which is a very small amount of energy. Despite their small magnitude, however, ﬂuctua-

tions can be measured and in fact such measurements provide a method for determining

Boltzmann’s constant.

17

4 Control Parameters

So far we have considered closed systems characterized by constant values of the param-

eters in the Lagrangian such as the masses of particles, the values of external electric

and magnetic ﬁelds, and the shape and volume of the containers that enclose the system.

Some of these parameters such as the volume of the system and the external ﬁelds may

be controllable from the outside, for example by moving pistons to change the volume.

We will call such macroscopic control variables X

m

. For simplicity we will consider the

case of only one X although the principles are the same for several of them. If you want

to think of a speciﬁc example, X can represent the volume of the system.

4.1 The Adiabatic Theorem and the First Law

An adiabatic process

∗

means two things. First, the system is isolated so that no energy

in the form of heat can ﬂow into or out of the system. Secondly, it means that the

control parameters are varied very slowly. In that situation the system will remain

in equilibrium throughout the process, although typically the energy and temperature

will change. Those things which do not change during an adiabatic process are called

adiabatic invariants. It is a theorem that entropy is an adiabatic invariant. This theorem

is true in classical mechanics but it is most easily understood in quantum mechanics.

Consider the energy levels of the system E

i

. In general they will depend on X:

E

i

= E

i

(X). If a sudden change is made in X the system will not remain in an

eigenstate of energy.

†

But if the change is arbitrarily slow then the adiabatic theorem

says that the system remains in an eigenstate, simply tracking the slow time dependence

of the instantaneous energy levels. In fact the levels will not cross over or disappear.

The implication is that the probability function P(i) is constant for each level, even

as the energy along the way gradually varies. Obviously if P(i) is constant, so is the

entropy. That is why entropy is an adiabatic invariant.

Now consider the change in energy of the system during the adiabatic process. That

energy change, is by deﬁnition, the work done on the system by changing X. The most

familiar example is the work done in slowly compressing a gas in a insulated container.

∗

“Adiabatic” is from the Greek “impassable.”

†

A system is in an eigenstate if it is undergoing periodic motion free of any external disturbances.

18 4 CONTROL PARAMETERS

If the change in X is small (call it dX) we may assume the work done is small (call it

dW). We can express this idea in equations.

dW =

∂E

∂X

S

dX

In the general case of several control parameters this becomes

dW =

¸

n

∂E

∂X

n

S

dX

n

.

Let us deﬁne variables Y

n

that are conjugate to to the X

n

by the formula

Y

n

= −

∂E

∂X

n

S

(4.1)

and it follows that

dW = −

¸

n

Y

n

dX

n

(4.2)

The most familiar example of conjugate variables (X, Y ) are volume and pressure (V, P).

dW = −PdV

Let us suppose that an inﬁnitesimal adiabatic change is followed by a second process

in which energy is added to the system in the form of heat – in other words, a second

process in which the control parameters are constant but the entropy changes (just the

reverse of what happened in the ﬁrst process). For this second process dE = TdS so

that the combined eﬀect of the work (adiabatic process) and the added heat give a

change in energy,

dE = TdS −PdV

More generally

dE = TdS −

¸

Y

n

dX

n

(4.3)

This relation is called the First Law of Thermodynamics, but it is really an expression

of energy conservation. The term TdS (energy due to a change of entropy) is called heat

and is sometimes denoted dQ. But there is no function Q, and therefore, technically,

dQ is not an exact diﬀerential. To see this we write (4.3) in the form

dQ = dE +Y dX.

4.2 Processes at Fixed Temperature 19

If Q were a well deﬁned function of E and X then

∂Q

∂E

= 1

and

∂Q

∂X

= Y

would both hold true.

Consider ∂

2

Q/∂X∂E. Since the order of diﬀerentiation should not matter, one ﬁnds

(∂Y/∂E)[

X

= 0. For example one would ﬁnd that the pressure would not depend on

the energy of a gas with ﬁxed volume. This of course is false, so dQ cannot be an exact

diﬀerential.

The meaning of all of this is that it is possible to bring a system through a series

of changes that bring it back to its original equilibrium state in such a way that the

net input of heat is not zero. What must be zero for such a cycle is the change in the

energy. In other words

(dQ +dW)

must be zero.

4.2 Processes at Fixed Temperature

We have deﬁned the conjugate variables Y such as pressure in terms of adiabatic or

constant entropy processes. We can also deﬁne them in terms of constant temperature

processes (isothermal). Such processes usually mean that the system is in contact with

a heat bath which is so big that its temperature does not change during the process.

At this point we need a calculus theorem that we leave for you to prove. Suppose we

have a function S(T, X) of two variables T and X. Let there also be a second function

E(T, X). By solving for T in terms of S and X, we can think of E as a function of S

and X.

20 4 CONTROL PARAMETERS

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 4.1 Prove the following identity:

∂E

∂X

S

=

∂E

∂X

T

−

∂S

∂X

T

∂E

∂S

X

This identity is general but in the case where S, T, E have their usual thermodynamic

meaning, we can use

∂E

∂S

X

= T

to get an expression for Y ≡ −(∂E/∂X)

S

∂E

∂X

S

=

∂E

∂X

T

−T

∂S

∂X

T

=

∂(E −TS)

∂X

T

Finally, using E −TS = A this can be written (∂A/∂X)[

T

. Thus,

Y = −

∂A

∂X

T

(4.4)

Thus we can either deﬁne conjugate variables like pressure in terms of derivatives of

E with respect to X at ﬁxed entropy, or derivatives of A with respect to X at ﬁxed

temperature.

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

Part II

Some Simple Applications

23

5 Dilute Gas

In this section we study the properties of a dilute gas (as contrasted with very dense

forms of matter such as Bose-Einstein condensate, which we study in Section 15. We

begin with the well-known ideal gas, then add two-particle interactions. We then give

molecules rotational inertia and ﬁnally study the van der Waals gas.

5.1 Ideal Gas

An ideal gas is deﬁned as a system of non-interacting particles. Forces between particles

due to collisions, electromagnetic ﬁelds, etc. are ignored.

Consider a container of volume V containing N identical non-interacting point par-

ticles, each with mass m. In order to apply the reasoning of Part I we must know how to

deﬁne the state of the system and how to sum over states. Begin with a single particle.

The state of a classical system is described by a point in its phase space,

∗

i.e., a

value for each generalized coordinate and each generalized momentum. Thus in the

present case we have a value for each of 3N coordinates whose values are restricted to

the interior of the container, plus 3N momenta whose values are unrestricted.

To sum over states we imagine replacing the phase space by a discrete collection of

small cells. The phase-space “volume” of each cell (units of momentum times length to

the power 3N) would tend to zero at the end of a classical calculation. But quantum

mechanics suggests that we take it to be (¯h)

3N

. Thus we replace the sum over states by

¸

i

→

dx

3N

dp

3N

(¯ h)

3N

(where ¯h ≡ h/2π and h is Planck’s constant). The energy of a state at point x, p is

E(x, p) =

1

2m

3N

¸

n=1

p

n

2

(5.1)

and the partition function is

Z(β) =

1

N!

dx

3N

dp

3N

(¯ h)

3N

e

−βE(x,p)

. (5.2)

∗

See Appendix B for an introduction to the concept of phase space.

24 5 DILUTE GAS

The factor 1/N! is put in to avoid over-counting conﬁgurations of identical particles.

For example if there are two particles, there is no diﬀerence between the conﬁguration

particle-one at point x, p; particle-two at x

′

, p

′

and the conﬁguration

particle-two at point x, p; particle-one at x

′

, p

′

.

Thus we divide by the number of equivalent conﬁgurations which in this case is 2 and in

the general case is N!. We will use Stirling’s approximation, N! ≈ N

N

e

−N

, in evaluating

the partition function.

The x integrals in (5.2) are trivial since the integrand does not depend on x. The

integral gives V

3N

and when combined with the 1/N! we get

eV

N

N

Since N/V is the particle density which we call ρ, these factors combine to give

e

ρ

N

Notice how the N and V dependence nicely combine to give an expression which only

depends on the density which we will keep ﬁxed as the number of particles tends to

inﬁnity.

The momentum integral in (5.2) is a Gaussian integral over 3N variables. In fact it

is the 3N power of the one dimensional integral

dp e

−β

p

2

2m

=

2mπ

β

.

The ﬁnal result for the partition function is

Z(β) =

e

ρ

N

2mπ

β

3N

2

(5.3)

If we want to explicitly exhibit the dependence on volume we replace ρ by N/V .

Z(β) =

eV

N

N

2mπ

β

3N

2

(5.4)

5.2 Almost Ideal Gas 25

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 5.1 Given the partition function in (5.4), compute A, E, S, and P as functions

of the temperature. Derive the energy per particle and the ideal gas law P = ρT. What

is the average speed (magnitude of velocity) of a particle?

Problem 5.2 On a PV diagram (pressure on one axis, volume on the other) what are

the curves of constant temperature (isotherms) and constant entropy (adiabats)?

What is the relation between pressure and temperature at ﬁxed entropy?

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

5.2 Almost Ideal Gas

Now let us introduce interactions between particles. We consider only potential energy

due to interactions between pairs of particles. (More complicated 3-body, 4-body,

potentials are possible but we will ignore them).

U(x

1

, x

2

, , x

N

) =

¸

m>n

U(x

m

−x

n

) (5.5)

We will treat U as a small quantity and compute Z to a ﬁrst-order approximation:

Z(β) =

1

N!

dx

3N

dp

3N

(¯ h)

3N

e

−βE(x,p)

where

E(x, p) =

¸

n

p

2

n

2m

+

¸

m>n

U(x

m

−x

n

)

To linear order in U,

Z =

1

N!

d

3N

x d

3N

p e

−β

¸

p

2

2m

¸

1 −β

¸

m>n

U(x

m

−x

n

)

¸

which becomes

V

N

N!

2mπ

β

3N

2

−β

V

N−2

N!

2mπ

β

3N

2

N(N −1)

2

dxdx

′

U(x −x

′

)

26 5 DILUTE GAS

Using

dx dx

′

U(x − x

′

) = V

dx U(x) ≡ V U

0

, approximating N(N − 1) by N

2

, and

using Stirling’s approximation, one ﬁnds

Z =

e

ρ

N

2mπ

β

3N

2

¸

1 −

βN

2

ρU

0

(5.6)

To calculate log Z we use the ﬁrst-order Taylor series for log (1 −ǫ), namely log (1 −ǫ) ≈

−ǫ.

log Z = −N log ρ −

3N

2

log β −

Nβ

2

ρU

0

+ const (5.7)

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 5.3 Given the partition function in (5.7), compute A, E, S, and P as functions

of the temperature. Derive the energy per particle. Find the correction to the ideal

gas law P = ρT + . What is the average energy of a particle? On a PV diagram

(pressure on one axis, volume on the other) what are the curves of constant temperature

(isotherms) and constant entropy (adiabats)?

Problem 5.4 Calculate the log Z to the next order (second order) in U and ﬁnd the

next correction to the ideal gas law.

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

5.3 Ideal Gas in a Potential Field

Consider a box of gas subject to an external potential. For example the box could be

in a gravitational ﬁeld so that every particle has a potential energy U = mgy where y

is the vertical direction. More generally U = U(x). The value of the potential at every

point can be thought of as a control parameter. In the general case we have a con-

tinuous inﬁnity of control parameters. Ordinary derivatives are replaced by functional

derivatives.

What is the conjugate to the potential at point x? If we adiabatically increase the

potential a tiny bit in some region, every particle in that region will have its energy

increased. So the change in energy is

δE =

d

3

x ρ(x)δU(x).

5.4 Diatomic Molecular Gas 27

The variational derivative of E with respect to U(x) is just ρ(x). Thus the conjugate

to U(x) is −ρ(x).

Now consider the partition function. It is the same as for the ideal gas except that

the integral

d

3

x which gave a factor of volume for each particle in that case now

becomes

d

3

x e

−βU(x)

and we get

Z(β) =

d

3

x e

−βU(x)

N

2mπ

β

3N/2

(5.8)

The entire dependence of the free energy on U will be in the term

−TN log

d

3

x e

−βU(x)

**If we want the density at point y we functionally diﬀerentiate this expression with
**

respect to U(y).

ρ(y) =

TN

d

3

x e

−βU(x)

βe

−βU(y)

=

N

d

3

x e

−βU(x)

e

−βU(y)

(5.9)

The factor N/

d

3

x e

−βU(x)

is independent of position. It can be replaced by a normal-

ization constant K determined by the fact that the integral of the density must equal

the total number of particles in the box.

ρ(y) = Ke

−βU(y)

. (5.10)

Thus, as we might expect, the particles are most dense where the potential is lowest.

For a gravitational ﬁeld the density as a function of height is proportional to

e

−βmgy

.

5.4 Diatomic Molecular Gas

So far we have considered a gas composed of point particles. If the molecules of the gas

have structure, they are capable of internal motion such as rotation and vibration. To

illustrate this idea we will consider a gas composed of diatomic molecules modeled by

28 5 DILUTE GAS

two particles with mass m/2 connected by a rigid massless rod of ﬁxed length ℓ. The

mass moment of inertia about its centroid is

I = 2

m

2

ℓ

2

2

=

1

4

mℓ

2

We can use m and I to represent any rigid molecule having extension in space and thus

rotational energy.

The partition function will factorize into a translational part and an internal part.

The translational energy of the molecule is p

2

/2m where p represents its linear momen-

tum. In addition there is rotational energy which we can derive. Let’s suppose the rod

is oriented in space with angular coordinates u, v where 0 < u < π and 0 < v < 2π. You

can think of u as the polar angle, and v as the azimuthal angle within a hemisphere.

The rotational kinetic energy is

E

rot

=

I

2

( ˙ u

2

+ ˙ v

2

sin

2

u) (5.11)

The canonical momenta are

p

u

=

˙ u

I

,

p

v

=

˙ v

I sin

2

u

and the energy (Hamiltonian) is

E

rot

=

p

2

u

2I

+

p

2

v

2I sin

2

u

(5.12)

The integral

d

3

x d

3

p e

−βp

2

/2m

(there is one such factor for each molecule) is the same

as for the point particle. The internal factor for each molecule is

du dv dp

u

dp

v

e

−βErot

. (5.13)

First do the momentum integrals and obtain

2Iπ

β

du dv sin u =

4πI

β

The constant 4Iπ is of no importance and we can drop it. The important thing is that

there is a factor 1/β for each molecule. This means that the partition function in (5.3) is

changed by replacing β

−3N/2

by β

−5N/2

. The eﬀect is to change the energy per molecule

5.5 A Note about Volume and the van der Waals Gas 29

from

3

2

T to

5

2

T. In general we get an additional energy of

1

2

T for each internal degree

of freedom of the molecule (such as relative motion of two masses on an elastic link).

This leads to a famous paradox: as the distance between the parts of the molecule

tends to zero, one would expect the molecule’s behavior to approach that of the point

particle. But the internal energy we just derived does not depend on the particle’s size.

Without going into details, we simply note that the resolution of the paradox involves

quantum mechanics.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 5.5 Calculate the thermodynamic properties of the diatomic gas. In partic-

ular ﬁnd the adiabats on the PV diagram. Generalize the result to the case of n internal

degrees of freedom.

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

5.5 A Note about Volume and the van der Waals Gas

Consider the factor V

N

/N! in the ideal gas partition function. Using Stirling we can

write this as

V

N

N

One way to interpret this is to say that each particle moves in a box of size V/N.

Now let us suppose each particle is not really a point but rather an impenetrable

sphere of volume v (a “van der Waals gas”). In that case the volume over which the

center of mass can move is smaller and of order V/N−v. The partition function contains

the factor

V

N

−v

N

If we combine the idea of an impenetrable spherical particle with the corrections due to

a long range weak potential, equation (5.6) becomes

Z =

V

N

−v

N

2mπ

β

3N

2

¸

1 −

βN

2

ρU

0

(5.14)

and the free energy is

A = log Z = −N log

V

N

−v

−

3N

2

log β −

Nβ

2

N

V

U

0

+ const (5.15)

30 5 DILUTE GAS

Calculating the pressure and rearranging the equation gives the van der Waals equation,

(P −

u

0

2

ρ

2

)(V −Nv) = NT (5.16)

31

6 Simple Magnet and Magnetization

Consider a collection of N spins σ

1

, σ

2

σ

N

, each of which can be up or down along

the z axis (we will use the notation σ to indicate the z component of spin). There is

a magnetic ﬁeld H oriented along z and the energy of each spin is σµH. Note that in

this case the magnetic ﬁeld is a control parameter.

Let the number of up spins be n and the number of down spins be m where n+m =

N. The energy of such a conﬁguration is

E = (n −m)Hµ

and the number of such conﬁgurations is

N!

n!m!

=

N!

n!(N −n)!

.

The partition function is

Z(β, H) =

¸

n

N!

n!(N −n)!

e

−β(n−m)Hµ

.

We have written the partition function in terms of β and H in order to indicate its

dependence on temperature and the magnetic ﬁeld.

The sum is a binomial expansion and the result is

Z(β, H) = (e

βHµ

+e

−βHµ

)

N

= (2 cosh βHµ)

N

. (6.1)

and the free energy A is

A(β, H) = NT log (2 cosh βHµ). (6.2)

The conjugate variable to the control parameter H is the magnetization M:

M = −

∂A

∂H

T

. (6.3)

Using (6.2), we ﬁnd that (6.3) gives

M = −NµtanhβHµ.

32 6 SIMPLE MAGNET AND MAGNETIZATION

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 6.1 Compute the energy and entropy of the simple magnet. Find the func-

tional form of the energy in terms of entropy. Do you see anything strange?

Problem 6.2 Show that M is the average total magnetic moment of the assembly of

spins.

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

6.1 Ising Model

The so-called Ising magnet is equivalent to the simple magnet we just studied. Think

of a one dimensional line of N +1 spins, each interacting with its two nearest neighbors

except for the end-spins which have only a single neighbor. The system is deﬁned by

its energy

E = −j

N

¸

i=1

σ

i

σ

i+1

. (6.4)

The constant j is the strength of interaction between neighboring spins. At this point

we will not introduce an external ﬁeld although we could, but that makes the problem

harder. Note that we have chosen the sign so that for positive j the energy is minimum

when the spins are aligned. This is the ferromagnetic case. The anti-ferromagnetic case

is deﬁned by choosing j to be negative.

The trick is to write the partition function as the sum of two terms. The ﬁrst term

contains all conﬁgurations in which spin-one is up. To calculate this term let’s change

variables. Deﬁne “dual” spin variables µ

i

. (These µ

′

s are not magnetic moments. There

should be no confusion; µ without an index is a magnetic moment. With an index it is

a dual spin variable.)

µ

i

= σ

i

σ

i+1

.

There are N µ

i

-variables labeled µ

1

, µ

2

, µ

3

, µ

N

. They are all independent and deter-

mine the original spins by the transformation

σ

j

=

j−1

¸

i=1

µ

i

.

6.1 Ising Model 33

For example, in the sector with the ﬁrst spin up,

σ

1

= 1

σ

2

= µ

1

σ

3

= µ

1

µ

2

and so on. Note also that the µ

i

take on the values ±1 just as do the original spins.

The partition function in this sector is given by

Z =

¸

exp

βj

¸

µ

i

.

This is exactly the same as the earlier model with the following substitutions:

Hµ →−j

σ

i

→µ

i

Thus the partition function in this sector is

Z = (2 cosh βj)

N

Finally we have to add the sector with σ

1

= −1. But the entire problem is sym-

metric with respect to changing the sign of all the spins simultaneously. Therefore the

contribution of the second sector is the same as the ﬁrst, and we ﬁnd

Z = 2(2 cosh βj)

N

(6.5)

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 6.3 Add a term to (6.4) corresponding to an external magnetic ﬁeld. In

other words replace (6.4) by

E = −j

N

¸

i=1

σ

i

σ

i+1

+

N+1

¸

i=1

µHσ

i

. (6.6)

Compute the magnetization to linear order in the magnetic ﬁeld. Note what happens

in the limits of small and large temperature. Can you explain the behavior?

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

34 7 THE MAXWELL RELATIONS

7 The Maxwell Relations

Unlike heat,

∗

the total energy of a system is a well deﬁned function of the thermody-

namic equilibrium state. This means it is a function of all the control variables and one

other variable such as the temperature or the entropy. For simplicity we will work with

a single control variable, the volume, and freeze all the others.

The equation

dE = TdS −PdV

suggests that we think of E as a function of S and V . We can then write

∂E

∂S

= T

∂E

∂V

= −P (7.1)

Now from the fact that ∂

2

E/∂S∂V = ∂

2

E/∂V ∂S we derive the ﬁrst Maxwell Relation.

.

∂T

∂V

S

= −

∂P

∂S

V

(7.2)

Figure 4 illustrates a thought experiment to conﬁrm the ﬁrst Maxwell Relation. On

the left we have an insulated cylinder with a movable piston that can be used to change

the volume. We slowly move the piston and change the volume by dV. On the right we

have a heat-conducting box of ﬁxed volume and at temperature T. We allow an amount

of heat dQ into the box and the pressure changes by dP.

Equation (7.2) is a remarkable bit of magic. With very little input we derived the

general fact that for all systems the change in temperature under an adiabatic volume

change is the negative of the change in pressure when we add a bit of entropy at ﬁxed

volume. It is very general and applies to solids, liquids, gases, black body radiation and

more.

We can derive additional relation by focusing on the Helmholz free energy A =

E −TS:

dA = dE −TdS −SdT = (TdS −PdV ) −TdS −SdT

or

dA = −SdT −PdV. (7.3)

∗

We saw on page 19 that heat Q does not have an exact diﬀerential.

35

Figure 4: Experiment Illustrating First Maxwell Relation

36 7 THE MAXWELL RELATIONS

We think of A as a function of T and V . Thus

∂A

∂T

= −S

∂A

∂V

= −P

and by using the fact that derivatives commute, we derive the second Maxwell relation.

∂S

∂V

T

=

∂P

∂T

S

(7.4)

The ﬁrst Maxwell Relation follows from considering the energy as a function of the

pair (S, V ), and the second Maxwell Relation from A(T, V ). Two more relations can be

obtained from functions H(S, P) (enthalpy) and G(P, T) (Gibbs free energy).

H = E +PV

dH = TdS +V dP (7.5)

Enthalpy is useful when considering a process at ﬁxed pressure. For example, consider

a gas in a cylinder with a piston such that the pressure is just the weight of the piston.

The change in enthalpy if due to some added heat is just the heat added (TdS). If we

call the heat added ∆Q, then ∆E +P∆V = ∆Q.

The corresponding Maxwell relation, the third, is

∂T

∂P

S

=

∂V

∂S

P

. (7.6)

Finally the Gibbs free energy is deﬁned by

G = H −TS

dG = V dP −SdT (7.7)

and the fourth Maxwell relation is

∂V

∂T

P

= −

∂S

∂P

T

. (7.8)

When using the Maxwell relations we can replace a diﬀerential change in entropy by

dQ/T where dQ is the added heat energy.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 7.1 Design experiments for conﬁrming the second, third, and fourth Maxwell

Relations similar to the one shown in Figure 4.

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

37

8 The Second Law

Let us simplify the story a bit by considering an initial probability distribution on phase

space which is constant over some small “blob” and zero outside the blob. In this case

the entropy can be taken to be the logarithm of the phase space volume of the blob. In

some sense it increases with time, but that seems to violate basic principles of mechanics.

The Second Law has an interesting history.

∗

Boltzmann originally tried to deﬁne a

quantity called H that always increased (dH/dt ≥ 0) along a trajectory in phase space.

But Loschmidt said that mechanics is reversible, so if H increases along a trajectory it

will decrease along the time-reversed trajectory. Since the time-reversed trajectory is

also a solution to Newton’s equations, it is not possible to have a function on (p, x) that

always increases.

The answer to this apparent paradox comes from chaos theory and coarse graining.

Chaos is the fact that for most systems the orbits in phase space are unstable.

An example is a frictionless billiard system (ignoring the pockets). Start with the 15

numbered balls all arranged in a tight stack and carefully aim the Q-ball. If the initial

conditions are perfectly reproducible then every time you do the experiment the result

will be the same no matter how long you follow the orbits (the paths of the balls on the

tabletop). But the tiniest error, either in the way the numbered balls are stacked, or in

the initial trajectory of the Q-ball, will cause exponential growth, so that after just a

couple of collisions the outcome will be extremely diﬀerent.

This is a generic phenomenon in mechanics whenever there is more than one coordi-

nate. It is characterized by a quantity called the Lyapunov exponent. Let the distance

in phase space between two initial points be ǫ. After a small time the distance between

them will be e

λt

where λ is the (largest) Lyapunov exponent. It is easy to prove that the

Lyapunov exponent must be greater than or equal to zero. Zero is a very exceptional

value, implying no growth in ǫ, and for almost all systems with more than one degree

of freedom, it is positive. Then the system is said to be chaotic.

To see what chaos implies, imagine following a patch of phase space – say at a ﬁxed

energy – as it evolves. The Liouville theorem

†

says that the volume and topology of the

∗

See Appendix C.

†

See, for example, the entry of that name at www.wikipedia.org.

38 8 THE SECOND LAW

region does not change, but in general the shape does change. If the system is chaotic,

then points that were initially close will soon diverge. The shape branches out and

forms a crazy fractal made of thinner and thinner tendrils spreading our over the phase

space as in Figure 5. Eventually the tendrils will get arbitrarily close to any point on

the energy surface of the initial points.

Suppose we take into account our inability to resolve points of phase space with

arbitrary precision. In other words, we introduce a cutoﬀ, replacing each phase space

point by a small but ﬁnite sphere. We can coarse-grain the phase space blob by drawing

such a sphere over each point in the original blob. Assuming the original blob is bigger

than the cutoﬀ, then at ﬁrst, coarse graining has little eﬀect. But as the tendrils spread

the coarse-grained version ﬁlls a larger and larger volume, even thought the ﬁne-grained

blob has a ﬁxed volume. Eventually the coarse-grained blob will ﬁll the entire energy

surface. Liouville’s theorem does not tell us that the coarse-grained volume cannot

increase. Of course the course-grained volume cannot be smaller than the ﬁne-grained,

so the coarse-grained volume cannot decrease.

Now come back to Loschmidt. After a long time imagine time reversing for every

particle (this means reversing its momentum). If we do this for every point in the ﬁne-

grained blob, it will trace back to the original small round blob. Here is an example:

take all the air in the room and start it out in a very small volume in the corner of the

room. In a short amount of time the molecules will spread out over the volume and ﬁll

the room. Can the opposite happen? No, because this would violate the second law;

the entropy of a conﬁned gas is less than that of a gas that ﬁlls the room. But if we

time-reverse every molecule of the ﬁnal state, the air will rush back to the corner.

The problem is that if we make a tiny error in the motion of just a single molecule,

that error will exponentially grow with the Lyapunov exponent and instead of lurking

in one corner, it will traverse the entire room. We see that in the bottom of Figure 5.

Some time-reversed trajectories (long lines) lead back to the original blob. Most don’t,

Even trajectories that start very close to one which does go back to the blob, quickly

depart and go somewhere else. In fact if we run the coarse-grained blob backward in

time (or forward) and then coarse grain the result, the phase space volume will be even

bigger.

Freak accidents do happen, if you wait long enough. Given enough time the air in

the room will by accident congregate in the corner. The correct statement is not that

39

Figure 5: The Second Law. The phase space volume and topology are exactly conserved

but when coarse grained it grows.

40 8 THE SECOND LAW

unlikely things never happen, but only that they rarely happen. The time that you

would have to wait for this unusual air-event to take place is exponential in the number

of molecules. So don’t hold your breath!

Thus there are two kinds of entropies, ﬁne-grained and coarse-grained. The ﬁne-

grained entropy never changes because of Liouville’s theorem. The coarse-grained en-

tropy does increase.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 8.1 Suppose the room has volume of 100 cubic meters. Consider the pos-

sibility that all the air accidentally accumulates in a one cubic meter volume in the

corner. What is the probability for this to happen? How long would you expect to have

to wait to see it happen?

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

41

9 Quantum Considerations

There are a number of paradoxes in statistical mechanics that were only resolved by the

introduction of quantum mechanics. We saw one on p. 29. The energy per molecule

of an ideal gas depends on the molecular structure: for point molecules it is

3

2

T and

for the diatomic molecule with ﬁxed separation it is

5

2

T. But as the diatomic molecule

gets smaller and smaller, it should tend to the point molecule. It does not, at least

according to classical (non-quantum) statistical mechanics. Another example is the

apparently inﬁnite energy of black body radiation.

Let’s begin with a simple example. Suppose there is a single harmonic oscillator

(HO) in equilibrium with a heat bath. We can think of the HO as a molecule if we

ignore its translational degrees of freedom. The Hamiltonian is

E =

p

2

2m

+

kx

2

2

(9.1)

If we think of the HO as a molecule then p and x are not the position and momentum of

the center of the center of gravity, but rather the relative variables. The mass m would

be the reduced mass of the system. The partition function is

Z =

dx dp e

−β(p

2

/2m+kx

2

/2)

or

Z =

dp e

−β(p

2

/2m)

dx e

−β(kx

2

/2)

The integrals are Gaussian, and using ω =

k/m, give

Z =

2mπ

β

2π

kβ

=

2π

ω

β

−1

. (9.2)

An easy calculation gives the average energy of the oscillator to be

E = T. (9.3)

Two things to notice about this energy: First, it is independent of m and especially of

k. Second, in the molecular interpretation it is the internal energy of the molecule and

would be added to the usual

3

2

T.

42 9 QUANTUM CONSIDERATIONS

What is strange is that even in the limit of a rigid molecule (k → ∞) the internal

energy does not go away. One might have expected that when k is very large the molecule

should be indistinguishable from a point molecule which has no internal energy.

Let us redo the calculation taking account of quantum mechanics. All we really need

from QM is the fact that the energy spectrum is

E

n

= ¯ hω

n

(9.4)

where n = 0, 1, 2, and ω =

**k/m, the oscillator frequency.
**

The partition function is

Z =

∞

¸

n=0

e

−β¯hωn

.

The sum is a geometric series:

Z =

1

1 −e

−β¯hω

=

e

β¯hω

e

β¯hω

−1

(9.5)

Finally, the average energy is

E =

ω¯h

e

β¯hω

−1

(9.6)

Note that at high temperature (small β) the answer agrees with the classical answer

E = T. But for low temperature the energy is exponentially small:

E ≈ ω¯he

−ω¯h/T

. (9.7)

Notice that in the limit k →∞(or equivalently ω →∞) the energy of the oscillator goes

to zero. In other words, the HO molecule does behave like the classical point particle

which has no internal energy. Only when the temperature becomes large enough that the

classical internal energy T becomes comparable to a single quantum does the molecule

reveal its structure.

43

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 9.1 Compute the free energy and entropy of the classical and quantum os-

cillators. How do they compare for high and low temperature? Where is the transition

from quantum to classical?

Problem 9.2 In section 5.4 the energy for the diatomic molecule could have been

written

E =

L

2

2I

In quantum mechanics L

2

takes on the values n(n + 1)¯ h and each energy level has

degeneracy 2n + 1. The partition function for a single molecule can be written as an

inﬁnite sum over n. Estimate the sum in the limit of large and small temperature. Show

that at high temperature the classical diatomic result is reproduced for the energy. Show

that at low temperature the energy per particle tends to the point molecule limit.

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

44 10 THERMAL RADIATION

10 Thermal Radiation

Consider a box (cube) with sides of length L. The walls of the box are reﬂecting.

Electromagnetic radiation can exist in the box. The radiation can be decomposed into

standing waves with the form

F(x, y, z) =

¸

nx,ny,nz

X(n

x

, n

y

, n

z

) sin

n

x

π

L

x sin

n

y

π

L

y sin

n

z

π

L

z (10.1)

The dynamical degrees of freedom are the amplitudes X(n

x

, n

y

, n

z

) which behave like

harmonic oscillators of frequency

ω

n

=

nπc

L

(10.2)

where n ≡

n

2

x

+n

2

y

+n

2

z

The partition function is an inﬁnite product of harmonic oscillator partition func-

tions for each mode of oscillation. Classically

Z =

¸

nx,ny,nz

e

β¯hωn

e

β¯hωn

−1

Let us consider the energy of the system. Since log Z is a sum over the modes, the

energy is just the sum of the energy of the oscillators. As we have seen, a classical

oscillator has an energy equal to T = β

−1

, independent of its frequency. Therefore each

of an inﬁnite number of modes has energy T and the total energy is inﬁnite. Most of

the energy is in very short wavelengths (large n) and the apparently inﬁnite result is

called the “Ultraviolet Catastrophe.”

The resolution of the catastrophe is provided by quantum mechanics. As we have

seen, the energy stored in a quantum oscillator is much smaller than the classical value

when the frequency of the oscillator is large. Using (9.6) we ﬁnd the total energy is the

sum

E =

¸

nx,ny,nz

ω

n

¯h

e

β¯hωn

−1

(10.3)

which converges.

When the volume of the box is large the neighboring values of the ω

n

are very close

and the sum can be approximated as an integral. Let us deﬁne the wave vector k by

k

x

= n

x

π/L and similarly for y and z. The diﬀerence between neighboring wave vectors

45

is

∆k

x

=

π

L

and the sum is replaced by

¸

nx,ny,nz

→

L

3

π

3

d

3

k

We also note that ω

n

= ck so that (10.3) becomes

E =

L

3

π

3

d

3

k

ck¯h

e

β¯hck

−1

multiplying by Tβ = 1 gives

E =

L

3

π

3

T

d

3

k

βck¯h

e

β¯hck

−1

Changing integration variables to u = β¯hck we get

E =

L

3

(πc¯h)

3

T

4

d

3

u

[u[

e

u

−1

or

E = 4π

L

3

(πc¯h)

3

T

4

∞

0

du

u

3

e

u

−1

The integral can be evaluated and gives a factor of π

4

/15 leaving

E =

4π

2

15

L

c¯h

3

We have done the calculation as if the ﬁeld were a scalar with only one component.

In fact the ﬁeld has two components corresponding to the two possible polarizations.

The net eﬀect is that there are twice as many oscillators and the total energy is twice

the above value.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 10.1 Compute A, S, and P for thermal radiation. Work out all integrals.

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

46 11 CHEMICAL POTENTIAL AND GRAND CANONICAL ENSEMBLE

11 Chemical Potential and Grand Canonical Ensemble

In the remainder of this text we are going to study three topics: magnets, the classical

liquid-gas phase transition, and Bose-Einstein condensation. For all three applications

we will need the concept of chemical potential.

The chemical potential is a useful tool for studying systems in which the number of

particles ﬂuctuates. Roughly speaking it is related to particle number in the same way

that temperature is related to energy: as a Lagrange Multiplier.

Let’s return to the derivation of the Boltzmann distribution (Section 2). Suppose

there are some conserved quantities in addition to energy. For example suppose the sys-

tem is a container of helium. The total number of helium atoms is conserved (assuming

that the temperature is not high enough to break up the atoms).

There are situations in which the number of particles in each subsystem may ﬂuc-

tuate even though the total for all subsystems is ﬁxed. For example, in considering the

ensemble of N subsystems, we may allow helium atoms to be transferred from one sub-

system to another. In that case only the total number of atoms is conserved. Suppose

the number of atoms in the n

th

copy is called N

n

. The constraint takes the form

¸

N

n

= N

total

.

A conﬁguration of the (sub)system is now speciﬁed by an energy and a number of

atoms, N, and the probability function depends on N as well as on the energy level. To

implement the constraint of ﬁxed N

total

we introduce a new Lagrange multiplier. The

Lagrange multiplier is βµ which includes a factor of β as a matter of custom. µ is called

the chemical potential.

Following the same logic as in Section 2, it is easy to see that the probability function

takes the so-called Grand Canonical form,

P(i, N) = Z

−1

e

−β(E

i

+µN)

.

In this formula, the index i refers to the energy levels of the N-particle system and in

general these levels will depend on N.

The partition function is deﬁned so that the total probability is one. Thus

Z(β, µ) =

¸

N

e

−βµN

Z

N

(11.1)

47

where Z

N

is the partition function with ﬁxed number of particles N.

Recall that in the case of energy, the Lagrange multiplier β is determined so as to

ﬁx the average energy. In the same way the chemical potential is determined to ﬁx the

average number of atoms. It is easy to see that

N = −

∂A

∂µ

(11.2)

where, as usual, A = −T log Z.

Since N = V ρ we can also think of the chemical potential as the parameter that

allows us to vary the particle density.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 11.1 Prove the following relations for processes at ﬁxed volume and other

control parameters:

E = TS −µN −T log Z

dE = TdS −µdN (11.3)

Problem 11.2 Find an expression for the ﬂuctuation of the number of particles anal-

ogous to the relation in Section 3.

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

Let’s calculate the Grand Canonical Partition function (GCPF) for an ideal gas. For

a ﬁxed number of particles the Canonical Partition function is

1

N!

¸

V

2mπ

β

3/2

¸

N

The GCPF is gotten by multiplying this expression by e

−βµN

and summing over N.

The result is

Z = e

(2mπ/β)

3/2

V e

−βµ

(11.4)

and the free energy is

A = −T log Z = −T

2mπ

β

V e

−βµ

(11.5)

48 11 CHEMICAL POTENTIAL AND GRAND CANONICAL ENSEMBLE

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 11.3 Using (11.5), calculate the mean number of particles and the total

energy as a function of the chemical potential. Then show that E = 3N/T.

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

49

12 Back to Magnets

Let’s return to the Ising model in an external magnetic ﬁeld and compute the magne-

tization to lowest order in the ﬁeld. To simplify the notation we will set µH = h. We

will mostly be interested in the ferromagnetic case with j being positive. In this case

the energy is lowest when the spins line up in the same direction.

E = −j

N

¸

i=1

σ

i

σ

i+1

+

N+1

¸

i=1

hσ

i

.

The magnetization M (eq. 6.3) is given in terms of the derivative of log Z with respect

to h. For simplicity we replace (6.3) by

M = −

∂A

∂h

(12.1)

To compute M to order h we must compute log Z to order h

2

. Thus we write

Z =

¸

σ

e

βj(

¸

N

i=1

σ

i

σ

i+1

−

¸

N+1

i=1

hσ

i

)

and expand to order h

2

.

Z =

¸

σ

e

βj

¸

N

i=1

σ

i

σ

i+1

¸

1 −βh

¸

i

σ

i

+

1

2

β

2

h

2

¸

i,j

σ

i

σ

j

¸

(12.2)

It is easy to see that the term proportional to h vanishes. The original Ising model is

symmetric with respect to changing the signs of all spins and that insures that only

even powers of h survive. To order h

2

the logarithm of Z is given by

log Z = log Z

I

+

β

2

h

2

2Z

I

¸

σ

e

βj

¸

N

i=1

σ

i

σ

i+1

¸

i,j

σ

i

σ

j

. (12.3)

where Z

I

is the Ising partition function given by (6.5). For i = j the term σ

i

σ

j

= 1.

Thus we may write

¸

i,j

σ

i

σ

j

= N + 2

¸

i>j

σ

i

σ

j

or

¸

i,j

σ

i

σ

j

= N + 2

¸

i,m

σ

i

σ

i+m

50 12 BACK TO MAGNETS

where the m-sum starts at m = 1.

Now let us change to the dual variables µ

i

= σ

i

σ

i+1

. The term in (11.5) that is

quadratic in h is given by

β

2

h

2

2Z

I

¸

µ

e

−βj

¸

µ

i

¸

N + 2

¸

i,m

σ

i

σ

i+m

¸

. (12.4)

Now consider σ

i

σ

i+m

. In terms of the µ

i

it is given by

σ

i

σ

i+m

= µ

i

µ

i+1

µ

i

µ

i+m−1

Thus the second term in 12.4 gives

β

2

h

2

(2 cosh βj)

N

¸

µ

e

−βj

¸

µ

i

¸

¸

i,m

µ

i

µ

i+1

µ

i

µ

i+m−1

¸

. (12.5)

where we have replaced Z

I

by the value we computed in Section 6. For those µ outside

the range (i, i + m − 1) the sum over the values of µ just cancel the corresponding

factors of (2 cosh βj) in the denominator. For the points inside the range, the sum

simply replaces (2 cosh βj) by (2 sinh βj). Thus the net result for large N is

Nβ

2

h

2

¸

m=1

(tanh βj)

m

= Nβ

2

h

2

tanh βj

1 −tanh βj

+

1

2

**or more simply
**

N

β

2

h

2

2

e

2βj

The magnetization (−∂

h

A) is

M = βhe

2βj

. (12.6)

First of all, notice that for all ﬁnite temperature the magnetization is linear and

continuous in h for small h. This means that there is no spontaneous magnetization

when h is shut oﬀ. But at zero temperature the coeﬃcient e

2βj

diverges. The magne-

tization does not diverge at zero temperature but its derivative becomes inﬁnite. At

ﬁnite temperature the graph of M versus h looks like the solid line in Figure 6 while

for zero temperature it looks like the dashed line. In other words, as you shut oﬀ the

12.1 Connection with Fluids 51

Magnetic field h

M

a

g

n

e

t

i

z

a

t

i

o

n

M

T>0

T=0

Figure 6: Magnetization vs external Magnetic ﬁeld.

external ﬁeld the magnetization persists, but this phenomenon only occurs at zero tem-

perature for the Ising model. We will see that in higher dimensions the spontaneous

magnetization persists up to a critical temperature.

The Ising model is characteristic of all one-dimensional statistical systems with short-

range interactions. There is no phase transition at ﬁnite temperature.

Ising correctly solved the one-dimensional system and correctly told his advisor Lenz

that it had no transition at ﬁnite T. His mistake was to conclude that the corresponding

model in higher dimensions also had no transition.

12.1 Connection with Fluids

One of the interesting connections that we will explore is the connection between mag-

netic systems and the liquid-gas transition in molecular systems like water. Let us begin

with a system of particles with 2-body potentials that consist of a hard core repulsion

and a slightly longer range (but still short range) attraction. The hard core will be

implicit in the conﬁguration space integrations over position. No two particles will ever

be allowed to be closer than some core size. The longer range potential, U(x − y) will

52 12 BACK TO MAGNETS

be explicit.

Let the density of particles be ρ. The potential energy can be written in the compact

form

PE =

d

3

x d

3

y ρ(x)ρ(y)U(x −y). (12.7)

Since the potential energy is independent of momentum and the kinetic energy is

independent of position, the partition function factors into a position factor and a

momentum factor Z = Z

x

Z

p

. The momentum factor give the usual (2m/β)

3N

2

. It plays

no role in determining the pressure or in the gas-liquid phase transition. We can ignore

it for our present purposes. From now on we will just consider Z

x

and drop the subscript

x.

The partition function is

Z =

¸

conﬁg-

urations

e

−β

d

3

x d

3

y ρ(x)ρ(y)U(x−y)−βµ

d

x

ρ(x)

(12.8)

Now imagine a lattice approximation to space (in any number of dimensions.) We

will choose units in which the hard-core-size is unity. Thus, at each point in the lattice

there may either be a particle or not, and no site may have more than one particle. If

all sites are vacant we call the state the vacuum.

Let us introduce a variable at each point that indicates whether or not a particle is

present. Call it σ

i

where i indicates the site. If σ = −1 we say the site is vacant. If

σ = +1 we say a particle occupies the site. In this way we map the problem of a system

of identical molecules onto a magnetic system of the type we have studied.

If we work in units in which the lattice has unit spacing then the density is 1 on

occupied sites and 0 on empty sites. Thus

ρ =

1 +σ

2

We also specify that the potential energy U is negative and that its range is one lattice

spacing. In other words the potential is non-zero only for particles separated by a single

link. Replacing integrals by sums, the energy is

PE = −4J

∗

¸

ρ(i)ρ(j) (12.9)

12.1 Connection with Fluids 53

where the symbol

∗

¸

means a sum over neighboring sites on the d-dimensional lattice. We take J > 0 so that

if two particles are far apart the interaction energy is zero but if they are on neighboring

sites the energy is negative. This corresponds to an attractive short range potential.

Using ρ =

1+σ

2

we easily ﬁnd

PE = −J

∗

¸

σ

i

σ

j

−4J

¸

i

ρ(i) + constant.

From now on we drop the constant.

Let’s add to the potential energy the chemical potential term −µ

¸

i

ρ(i), which

apart from a constant is −

1

2

µ

¸

i

σ(i). The result is that the exponent in the expression

for Z is

−J

∗

¸

σ

i

σ

j

−

4J +

µ

2

¸

i

σ(i) + constant.

The partition function is the d-dimensional analog of the Ising model.

Z =

¸

σ

e

β(J

¸

∗

σ

i

σ

j

+

8J−µ

2

¸

i

σ(i))

Obviously this is the d-dimensional analog of the Ising model in the presence of an

external magnetic ﬁeld h =

1

2

(µ −8J).

Let us consider the magnetization M. It is simply the total number of sites times

the average value of σ. Thus we may identify the magnetization per site (average of σ)

as 2ρ −1.

Suppose a magnetic transition takes place so that at h = 0 there is a jump in

magnetization. In the Ising model this only happens at T = 0 but in higher dimensions

it happens at all temperatures up to some critical temperature T

∗

that depends on

dimension. What does this jump mean from the point of view of the molecular system?

It is a sudden jump in density as the chemical potential is varied. In fact it is the

transition from a gaseous phase to a more dense liquid phase. The phase diagram is

shown in Figure 7.

Notice that there are two ways to pass from one phase to another: cross the discon-

tinuity (the thick line) or go around it. When you cross it the magnetization (density)

54 12 BACK TO MAGNETS

Figure 7: Phase diagram of a magnet. The thick line represents the discontinuity in

the magnetization. In the 1-D Ising model the critical temperature T

∗

is zero. In the

analogy with molecular ﬂuids, negative magnetization represents the gas phase, and

positive magnetization represents the liquid phase.

12.2 Low Temperature Expansion 55

jumps in what is called a ﬁrst-order transition. Going around the critical point involves

no jump and the behavior is continuous. This is characteristic of both magnetic systems

and gas-liquid systems such as water.

One interesting point is that in one dimension there is no gas-liquid phase transition.

12.2 Low Temperature Expansion

Something strange happens at zero temperature in the Ising model. To analyze it we

can do a low temperature expansion (LTE). Actually the LTE is not an expansion in T

but in exp (−2Jβ). Let’s deﬁne the system by introducing extra dummy spins at the

ends of the spin chain. We will take the number of “active” spins to be N − 1 and

add two more frozen “dummy” spins at the ends. These boundary dummy spins are

assumed to be up σ

dummy

= +1. The question is whether the boundary conditions are

remembered far from the boundaries when N →∞.

Let us introduce new terminology: if two neighboring spins are the same we will say

the bond between them is unbroken, which is the same thing as saying µ = 1 for that

bond. If neighboring spins are opposite (µ = −1) the bond is broken. Note that there

are N bonds including the bonds involving the dummy spins.

Each broken bond increases the energy by 2J and introduces a factor exp (−2Jβ) in

the Boltzmann distribution. The LTE is an expansion in the number of broken bonds.

The ﬁrst strange thing we see in the LTE is that free energy is not proportional to

N. Since the boundary spins are frozen the number of broken bonds must be even. The

LTE for Z has the form

Z = 1 +

N(N −1)

2

exp (−4Jβ) +

and log Z is

log Z =

N(N −1)

2

exp (−4Jβ) +

The free energy seems to have a term proportional to N

2

. This is odd because the free

energy per site (−T log Z/N) is not ﬁnite but rather is proportional to N.

In fact the free energy density does not have a term proportional to N. For every

value of β it is ﬁnite. The point is that the things we are calculating in each order are

the derivative of the free energy with respect to the expansion parameter. One way of

saying this is that the ﬁrst non-vanishing derivative of the free energy density is inﬁnite

56 12 BACK TO MAGNETS

at exp (−2Jβ) = 0. You can go back and check that from the exact solution. This is

indicative of a singular behavior at T = 0.

Next consider the average value of a spin far from the boundary. At zero temperature

the only state with non-vanishing probability has all the spins up. So every spin has

average value +1.

For simplicity let’s assume that N is even and that we focus on the spin at the center

of the chain (just call it σ). There are N/2 bonds on either side of it. The average of σ

is

1

Z

=

¸

e

βJ

¸

σ

i

σ

i+1

σ.

To leading order Z

−1

= 1 −

N(N−1)

2

exp (−4Jβ) + To leading order (two broken

bonds) the sum in the numerator has four types of terms. The ﬁrst is a single term with

no broken bonds. In the second, both broken bonds are to the left of σ. In the third

term they are both to the right. In the fourth, one broken bond is on the left and one

is on the right. The ﬁrst term is 1.

For the second type of terms, σ = +1. We get

¸

2

=

N

2

(

N

2

−1)

2

exp (−4Jβ).

The third term – all broken bonds on the right – gives the same thing.

¸

3

=

N

2

(

N

2

−1)

2

exp (−4Jβ).

For the fourth type of terms σ = −1 and we get

¸

4

= −

N

2

2

exp (−4Jβ)

Altogether we ﬁnd that the average of σ to order exp (−4Jβ) is given by

1 −

N

2

2

+

¸

2

+

¸

3

+

¸

4

exp (−4Jβ) = 1 −

N

2

2

exp (−4Jβ)

Again, the lesson is that the derivative of the magnetization with respect to the expan-

sion parameter diverges (with a negative sign) as N →∞. What is going on is that the

magnetization vanishes for all T = 0. The smallest temperature destabilizes the spon-

taneous magnetization, i.e., the eﬀects of the inﬁnitely distant boundary conditions.

12.3 Ising-like Systems in Higher Dimensions 57

The reason for this behavior is simple. In one dimension we can ﬂip an arbitrary

number of adjacent spins, breaking only two bonds, at a cost in energy which is only

4βj. In other words a large block of ﬂipped spins costs an amount of energy which does

not grow with the size of the block.

12.3 Ising-like Systems in Higher Dimensions

In higher dimensions (number of dimensions = d) the situation is entirely diﬀerent. To

ﬂip M spins you must break at least ∼ M

1/d

bonds. This, of course, is a fact about the

surface-to-volume ratio in d dimensions. Thus you pay a large energy price for ﬂipping

large blocks of spins. We will ﬁnd that this stabilizes the low temperature behavior and

implies that the spontaneous magnetization of the ground state does not disappear at

ﬁnite T.

For deﬁniteness let’s take a square grid of lattice points with a total of n active

spins. In addition there are dummy spins at the boundary that are chosen to be up as

in Figure 8.

° = up

x = down

• = dummy

Figure 8: 2-dimensional lattice of spins. The o’s indicate up-spins, the x’s indicate down

spins and the •’s indicate dummy up spins.

58 12 BACK TO MAGNETS

Some deﬁnitions for the 2-dimensional case:

• The spins are located on the sites of the lattice.

• The energy is a sum over pairs of nearest neighbor (NN) spins. NN pairs are

associated with the links of the lattice.

• The squares of the lattice are called plaquettes. If you put a dot at the center

of each plaquette they form the dual lattice. The links of the dual lattice are in

one-to-one correspondence with the links of the original lattice. They cross the

links of the original lattice at right angles.

The energy is given by

E = −J

∗

¸

σ

i

σ

j

where the sum is over nearest neighbors. We can also think of the energy as residing on

the links of the lattice. The symbol

∗

¸

can be understood as the sum over links. If there were no dummy spins the ground

state would be 2-fold degenerate, i.e., all up or all down. But the dummies break the

degeneracy and the ground state is all up spins.

For notational simplicity we will take the case where all the boundary spins are

dummies.

Now consider the low temperature expansion. At zero temperature all the spins are

up and the partition function is Z = 1. The ﬁrst correction comes from ﬂipping a single

spin. This breaks four bonds on the links connected to that spin. The contribution

from single ﬂipped spins is

Z = 1 +Ne

−8βJ

+

log Z = Ne

−8βJ

+

Note that the free energy is extensive to this order. log???

Let’s check what happens to the next order. We can ﬂip two spins. There are two

possibilities; the two spins can be nearest neighbors or not. If they are not nearest

neighbors, eight bonds are broken and the added energy is 16j. If they are nearest

neighbors only six bonds are broken and the added energy is 12j. The reason that only

12.3 Ising-like Systems in Higher Dimensions 59

six bonds are broken is that the bond on the link connecting the pair is unbroken if

both spins are ﬂipped.

So the next order in the expansion comes from nearest neighbors (NN) and is of

order exp (−12βJ). The number of NN pairs is 2N for large N.

Z = 1 +Ne

−8βJ

+ 2Ne

−12βJ

log Z = N

e

−8βJ

+ 2e

−12βJ

**Note that the free energy is still extensive, i.e., ∼ N. This seems obvious since the
**

number of NN pairs is ∼ N.

The non-trivial test happens when we study non-nearest-neighbor pairs. The number

of such pairs is N(N − 5)/2. To see that, begin by picking a site and ﬂipping its spin.

The number of sites which are nearest-neighbor is ﬁve: the site itself and the four NN.

But we have to divide by two for the usual reason of not over-counting. Thus to order

e

−16βJ

we ﬁnd,

Z = 1 +Ne

−8βJ

+

N

2

e

−12βJ

+

N(N −5)

2

e

−16βJ

.

This looks dangerous because of the quadratic term in N. However, even if log Z is

linear in N, we expect Z to contain things of higher order. To calculate log Z we use

the formula

log (1 +x) = x −

x

2

2

+ +

x

n

n

+

and ﬁnd that the quadratic term in N cancels. Thus we ﬁnd

log Z/N = e

−8βJ

+

1

2

e

−12βJ

−

5

2

e

−16βJ

+

There are a few more terms that contribute in order e

−16βJ

. These terms come from

adjacent ﬂipped spins which break eight bonds. The spins can be in a line or in an

L-shape. The contributions are 2Ne

−16βJ

and 4Ne

−16βJ

respectively.

Unlike the 1-dimensional case there is nothing divergent in the power series coef-

ﬁcients of the free energy density. This suggests that there is nothing discontinuous

about the behavior at T = 0.

We can check this by examining the behavior of the average value of a spin far from

the boundary.

'σ` = Z

−1

¸

σ

e

βJ

¸

∗

σ

i

σ

j

σ

60 12 BACK TO MAGNETS

Again we sum over the number of broken bonds. The lowest order comes from a

single ﬂipped spin. If the ﬂipped spin is not σ, the contribution is the same as to the

partition function. But if the ﬂipped spin is σ, the sign is changed. Thus

¸

σ

e

βJ

¸

∗

σ

i

σ

j

σ = 1 + (N −1)e

−8βJ

−e

−8βJ

= 1 +Ne

−8βJ

−2e

−8βJ

Now we have to multiply by Z

−1

= 1 +Ne

−8βJ

. Expanding to order e

−8βJ

we ﬁnd

'σ` = 1 −2e

−8βJ

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 12.1 Compute the average value of σ to order e

−16βJ

and show that the

power series coeﬃcients are ﬁnite in the limit N →∞.

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

You may ask where we used the dummy spins. The only thing we used them for was

to break the degeneracy of the ground state. We could do something else that is often

more relevant for real magnets. Introduce an extremely small external ﬁeld h so that

each spin has an additional energy hσ. This will align the ground state. For h < 0 the

spins will line up in the up position. The energy cost of ﬂipping all the spins is 2Nh.

We can redo the calculations and then let h → 0. The average of σ will have a limit,

but at low temperature the limit will depend on whether you approach from positive of

negative h.

This phenomenon of spontaneous magnetization is a special case of spontaneous

symmetry breaking. The symmetry is deﬁned by changing the sign of all spins, σ

i

→

−σ

i

. It is a kind of “persistence of memory” in which an external magnetic ﬁeld will be

remembered even when it is switched oﬀ or made arbitrarily small.

We can diagnose the system without explicitly breaking the symmetry by studying

the response to an external ﬁeld. If we don’t break the symmetry then for every conﬁgu-

ration we must also allow the conﬁguration in which each spin is reversed. Since the two

conﬁgurations have the same energy, it is evident that they have the same probability

and the average of any spin must be zero. That is even true at zero temperature.

But there is a consequence of the spontaneous symmetry breaking. It is this: if we

ﬁnd a given spin to be up (down) then it is more likely that we will ﬁnd any other spin

12.3 Ising-like Systems in Higher Dimensions 61

to be up (down). In itself that is not really surprising. But we might expect the bias to

disappear as the distance between the two spins becomes arbitrarily large. However, if

there is spontaneous magnetization or symmetry breaking, then the correlation between

diﬀerent spins does not fade with distance.

We can say it in a precise way in terms of correlation functions. Deﬁne the average

value of the product of two spins 'σ

a

σ

b

` where a, b are two lattice sites.

C(a, b) ≡ 'σ

a

σ

b

` = Z

−1

¸

σ

e

βJ

¸

∗

σ

i

σ

j

σ

a

σ

b

(12.10)

Ordinarily, the bias introduced by ﬁnding one spin up (down) fades with distance, which

means that C(a, b) →0 as [a −b[ →∞. We deﬁne spontaneous symmetry breaking to

mean that the limit of C(a, b) is non-zero as the separation goes to inﬁnity.

Let us go back to the 1-dimensional case. Notice that the second term in equation

12.4 is nothing but β

2

h

2

C(i, j). The calculation that followed showed that

C(a, b) = tanh βJ

m

(12.11)

where m is the separation between a and b. For zero temperature C(a, b) = 1 but

for T = 0 the correlation tends to zero exponentially with distance. But in higher

dimensions C does not vanish at inﬁnity, at least for a range of temperature deﬁned by

the radius of convergence of the LTE.

Let’s calculate C(a, b) to order e

−8βJ

without introducing dummy spins at the

boundary. In this case the partition function is twice the previous value because we

must sum over the two starting points, i.e., all down or all up, and then start ﬂipping

spins. Thus to the order that we are working with,

Z = 2 + 2Ne

−8βJ

+ (12.12)

Next we consider

'σ

a

σ

b

` = Z

−1

¸

σ

e

βJ

¸

∗

σ

i

σ

j

σ

a

σ

b

(12.13)

Begin with the ground state with all spins up. The contribution to the sum in

(12.13) is of course the same as for the partition function, namely 1. Next, keeping σ

a

and σ

b

up, let us ﬂip any other spin. The contribution is

(N −2)e

−8βJ

62 12 BACK TO MAGNETS

Next we ﬂip σ

a

. This time the contribution has the opposite sign to what it would be

in the partition function. Combining it with the term coming from ﬂipping σ

b

(but not

σ

a

) we get

1 + (N −4)e

−8βJ

. (12.14)

Note that the term with both σ

b

and σ

a

ﬂipped is higher order in the expansion pa-

rameter. But this is just the term that arises from conﬁgurations that started with all

spins up. Since we have dropped the bias coming from the dummy spins, we must add

the contribution that arises from the opposite starting point.

Altogether we ﬁnd

C(a, b) = Z

−1

(1 + (N −4)e

−8βJ

)

= 1 −4e

−8βJ

(12.15)

Note that the result is independent of how far away the spins σ

b

and σ

a

are from one

another as long as they are not at the same point. In fact we can consider the result

(12.12) to be the limit of C(a, b) as the separation goes to inﬁnity.

In higher orders one ﬁnds that the correlation function continues to be ﬁnite at

inﬁnite separation. In fact the asymptotic limit is just the square of the average of a

single spin, assuming the dummy spins are put back.

The phenomenon of correlations remaining ﬁnite at inﬁnite separation is called long

range order and it persists over a range of temperature from T = 0 to some critical

temperature. In the next subsection we will see that it does not persist to arbitrarily

high temperature.

12.4 High Temperature Expansion

Now we come to the high temperature limit. Begin with Z. High temperature means

small (βJ). Therefore the high temperature expansion (HTE) is gotten by expanding

exp (−βE) in powers of (βJ).

Z =

¸

σ

1 −(βJ)E +

1

2

(βJ)

2

E

2

+ (12.16)

The n

th

term in the expansion is

Z

n

=

1

n!

(βJ)

n

¸

σ

∗

¸

(σ

i

σ

j

)

n

. (12.17)

12.4 High Temperature Expansion 63

Figure 9: The contributions to the HTE coming from a single link are represented by a

heavy line on that link.

Begin with the ﬁrst term, n = 0. All states contribute equally and we get

Z

0

= 2

N

log Z = N log 2 + (12.18)

Next, the n = 1 case. This involves a sum over nearest neighbors. Let us pick a

link on the lattice and consider the pair connected by the link. We will represent the

contribution of that link (connecting sites a, b) by drawing a line between sites as in

Figure 9 It is given by

2

N−2

(βJ)

¸

σa,σ

b

σ

a

σ

b

= 2

N

(βJ)

1

4

¸

σa,σ

b

σ

a

σ

b

(12.19)

This is obviously zero. The product σ

a

σ

b

must average to zero because for each value

of σ

a

, the other spin, σ

b

, can be 1 or −1. when summing, these will cancel.

64 12 BACK TO MAGNETS

Let’s go to the next order involving two links. Suppose the two links are not the

same. They could involve completely disjoint pairs or they may share one spin. In

either case there will be spins to sum over which are occurring to the ﬁrst power. Such

sums vanish since the unmatched spin can be ±1. But suppose the two links are the

same. Then the term involves (σ

a

σ

b

)

2

which equals 1. The doubled links contribute.

Their contribution is

2N(βJ)

2

(12.20)

The factor of 2N represents the number of links on the lattice. To see why the number

of links is twice the number of sites draw a link to the right of every site and a link

just below every site. As you range over all the sites you cover all the links. There are

corrections but they are sub-leading in N.

The next non-vanishing contribution comes from 4-link terms. Odd number of links

always leave unmatched spins so that when you sum over them the result vanishes.

The 4-link terms are of several types. For example, the four links can form a square,

or two links can be covered twice. A single link can be covered four times. Among

the cases in which two links are covered twice there are diﬀerent possibilities depending

on whether the links share a site. It would all be a lot easier if there were a rule that

forbade a link from being covered more than once.

In fact we can rewrite the sum over conﬁgurations so that we never do cover a link

more than once. Begin with the expression

Z =

¸

σ

e

βJ

¸

∗

σ

i

σ

j

(12.21)

which is the same as

Z =

¸

σ

¸

links

e

βJσ

i

σ

j

. (12.22)

Now consider the expression e

βJσ

i

σ

j

. Since σ

i

σ

j

can only take on the values ±1 all even

powers of σ

i

σ

j

are equal to 1 and all odd powers are −1. By expanding the exponential

and using this fact we ﬁnd

e

βJσ

i

σ

j

= cosh βJ +σ

i

σ

j

sinhβJ

= cosh βJ(1 +σ

i

σ

j

tanh βJ). (12.23)

Now we can write the partition function in the form

Z = (cosh βJ)

2N

¸

σ

¸

links

(1 +σ

i

σ

j

tanh βJ). (12.24)

12.4 High Temperature Expansion 65

Figure 10: Closed “curves” on the lattice

When the product is expanded in powers of tanh βJ we never cover the same link

twice. For example at fourth order (tanh

4

βJ) we get squares and nothing else. In

fact at any given order we get one or more closed “curves” in which no link is multiply

covered. Some examples are shown in Figure 10.

The partition function is given by

Z = (cosh βJ)

2N

¸

graphs

C(graph) tanh

m

βJ (12.25)

where the sum is taken over graphs consisting of any number of closed curves; C(graph)

is a counting factor that enumerates the number of such graphs and m is the number

of links that make up the curves.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 12.2 Calculate the partition function, free energy, and average spin in both

low and high T expansions for the case of the 3-dimensional Ising model. In the LTE

66 12 BACK TO MAGNETS

go to order e

−24βJ

. Show that the free energy is extensive and that the average magne-

tization is non-zero (zero) for low (high) enough temperature.

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

12.5 One- and Two-Point Functions in the HTE

Now let’s consider the averages 'σ` (1-pt. function) and 'σ

a

σ

b

` (2-pt. function) in the

HTE.

It is obvious that the 1-pt. function vanishes to all orders in the HTE Boltzmann

distribution in the form given in (12.24)

¸

links

(1 +σ

i

σ

j

tanh βJ). (12.26)

when expanded to a given order in tanh βj involves an even number of spins. Multiplying

by any odd number of spins and summing over the conﬁgurations will always give zero.

The 2-pt. function C(a, b) is not zero. Computing it involves graphs with any num-

ber of closed curves and one open curve connecting the points a, b. The lowest order

contribution comes from the shortest path connection them and has the value

C(a, b) = (tanh βJ)

n

(12.27)

where n is the number of links making up the shortest path. Thus we see an important

generic feature of correlation functions at high temperature: they decay exponentially

with distance. In this case it follows from the fact that tanh βJ < 1.

Since the asymptotic value of C at large separation satisﬁes C(∞) = 0 for high

temperature, and C(∞) = 0 for low temperatures, there must be a value of temperature

which separates the two regions. This point is the critical point.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 12.3 Compute the 2-pt function to leading order in the HTE for 2 points

on the same horizontal line and two points on the diagonal of a square. Extra Credit:

Compute it to the next order?

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

67

Figure 11: The dual lattice is composed of points at the centers of the Plaquettes.

13 High-T/Low-T Duality

The general properties of the Ising model in more than one dimension are independent of

dimensionality. What is special about two dimensions is that the model can be exactly

solved. We are not going to solve it but we are going to work out one special property

that occurs in d = 2. It is called duality and it is rather easily seen from the LTE and

HTE. We will focus on the partition function.

The HTE for Z is simple. We begin with the factor (cosh βJ)

2N

. Then multiply

this by a sum over all graphs involving any number of closed paths through the lattice.

Assign each graph a weight given by (tanh βJ)

n

where n is the number of links in the

graph. That’s it.

Now consider the LTE for Z. Start with all spins up. Now ﬂip a certain number.

That breaks a certain number of bonds. For each broken bond include the factor e

−2βJ

.

Sum over all such conﬁgurations.

The conﬁgurations of ﬂipped spins can be enumerated in a clever way. Divide the

ﬂipped spins into disconnected islands and draw a closed curve around each island.

To do that, it is convenient to deﬁne a dual lattice composed of the centers of all the

plaquettes (Figure 11).

The closed curves surrounding the islands of ﬂipped spins are composed of dual-

68 13 HIGH-T/LOW-T DUALITY

Figure 12: Surround each island of ﬂipped spins with a curve on the dual lattice.

lattice links. In fact they have the same form on the dual lattice as the curves on the

original lattice that were used to implement the HTE.

Each link on the dual curves is equivalent to a broken bond and the weight of such

a conﬁguration is (exp −2βJ)

n

where n is again the number of links forming the curves.

You may have realized by now that the LTE and HTE have exactly the same form

except for two diﬀerences. First, in the HTE the partition function contains a factor

of (cosh βJ)

2N

that has no analog in the LTE. Second, the LT expansion parameter is

exp −2βJ while the HTE parameter is tanh βJ. What this means is quite remarkable.

Suppose we compute the partition function or the free energy as a function of e

2βJ

from

the LTE. Then we automatically know the behavior at high T. Just take Ze

2βJ

and

replace the argument of the function by tanh βJ. Then multiply by (cosh βJ)

2N

and

you have the high temperature behavior.

If we assume that the critical point is the unique singularity on the T axis, it must

lie at the symmetry point of the duality exp −2βj ↔tanh βj. In other words the critical

69

point must be at exp −(2βJ) = tanh (βJ) or equivalently the critical point is at

sinh(2Jβ

crit

) = 1. (13.1)

Equivalently the critical temperature is

T

crit

≈ 2.27J. (13.2)

In higher dimensions the Ising model is dual to a theory living on the dual lattice but

in general that dual theory is not the same as the original theory. For this reason

the duality relates the critical temperatures of diﬀerent theories rather than giving an

equation for the critical temperature of the original theory.

There are many magical properties of the two-dimensional Ising model. We will

show you one more. In summing over lattice curves (connected and disconnected) we

must respect the rule that no link is covered more than once. Call this the “exclusion”

rule. Suppose there were no such rule and that we could freely draw curves. Call the

sum of all connected curves X. The sum of terms involving two disconnected curves

would be X

2

/2; three disconnected components, X

3

/3. The sum would exponentiate to

give exp X, and the logarithm of Z would just be X. Since any connected component

can be placed in N places, the free energy would be trivially seen to be extensive (∼ N).

In fact we can rearrange the sum so that there is no exclusion rule. Consider, for

example, a diagram with two adjacent squares. This is forbidden by the exclusion

rule. But let us combine it with a connected “ﬁgure-8” graph. The two are shown in

Figure 13. Neither graph is allowed according to the exclusion rule, but there is another

rule which allows such graphs but causes them to cancel. The rule is that every graph is

to multiplied by a factor (−1)

X

where X is the number of times the graph crosses itself.

With that rule the ﬁgure-8 graph cancels the graph with two adjacent squares. In fact

the new rule is exactly equivalent to the exclusion rule and allows the sum of connected

graphs to exponentiate. The rule can be used for the HTE directly, or for the LTE on

the dual lattice. In Figure 11 another example is illustrated: the two connected graphs

cancel leaving only the disconnected pair of squares with no overlapping links.

70 13 HIGH-T/LOW-T DUALITY

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 13.1 Use the new rules to calculate the log Z to order (tanh βJ)

8

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

The new rule also applies to correlation functions such as C(a, b).

Evidently the 2-D Ising model has some very simple features that make it exactly

solvable but we will not get into the exact solution here.

Figure 13: Two canceling graphs.

71

Figure 14: Another example.

72 14 WHAT HAPPENS AT T = T

C

?

14 What Happens at the Critical Temperature T

c

?

The behavior of a system at its critical temperature depends on how that temperature

is approached, i.e., from which direction in phase space. Let us begin by approaching

the critical point from the high temperature side.

14.1 High Temperature Expansion of the Correlation Function

The high temperature behavior of the correlation function C(a, b) exponentially falls

to zero with distance. For simplicity we take the two points to be on the horizontal

axis separated by D links. Let us also rename the expansion parameter by the simpler

symbol X. Note X < 1.

To leading order C(a, b) is given by

C(a, b) = (tanh βJ)

D

+ = X

D

(14.1)

Observe that at large separation the correlation function falls oﬀ exponentially.

C(a, b) ∼ e

−mD

m = −log X (14.2)

In general, for T > T

c

the correlation function exponentially tends to zero as with

a decay parameter m that depends on temperature. In the high temperature limit

m = −log (tanh βJ). The parameter m is the inverse to the “decay length” which is

often called the correlation length.

To ﬁnd the corrections to m we try a HTE for the logarithm of C.

The ﬁrst correction involves graphs in which the line of links makes an extra little

excursion with two additional links. The excursion can either be up one unit or down.

There are D(D + 1) such graphs and they give

C(a, b) =

X

D

1 +D(D + 1)X

2

(14.3)

To compute the inverse correlation length m we take the log of C, divide by D, and

let D tend to inﬁnity. Taking the log of (14.3) to leading order in X gives

log C = Dlog X +D(D + 1)X

2

(14.4)

14.1 High Temperature Expansion of the Correlation Function 73

D

n

a

b

Figure 15: Points a and b are separated along the x-axis by distance D and along the

y-axis by distance n.

This is bad because when we divide by D the result grows like D and in the limit it

diverges. The problem is that the correlation function does not exactly behave like an

exponential of D. As we will see, there is a prefactor multiplying the exponential which

gives (1/D) log C some extra non-analytic dependence on the expansion parameter.

To do the problem correctly we have to do it in “momentum space.” Let us consider

the two points a and b to be separated by D units along the x-axis and n units along

the y axis, as in Figure 15.

Call the correlation function C(D, n). Its Fourier transform

∗

with respect to n is

denoted C(D, k). Suppressing the D dependence, standard Fourier analysis gives

C(n) =

1

2π

+π

−π

dk e

−ikn

C(k)

∗

Appendix A

74 14 WHAT HAPPENS AT T = T

C

?

C(k) =

+∞

¸

−∞

C(n)e

ikn

(14.5)

We will calculate C(k) to order in X

2

. All diagrams contain the factor X

D

so let us

ignore it for now. Later we will restore it. To order X

2

we can only get to n = 0, ±1, ±2.

To get to larger values of [n[ requires more extra factors of X. Here is what we get.

C(0) = 1 +D(D + 1)X

2

C(±1) = (D + 1)X

C(±2) = (D + 1)(D + 2)X

2

(14.6)

and the Fourier transform C(k) is given by

C(k) = 1 + 2(D + 1)X cos k + (D + 1)

2

X

2

(1 + cos 2k) −(D + 1)X

2

(1 −cos 2k)

or, restoring the factor X

D

and using the identity 1 + cos 2k = 2 cos

2

k,

C(k) = X

D

1 + 2(D + 1)X cos k + 2(D + 1)

2

X

2

cos

2

k −(D + 1)X

2

(1 −cos 2k)

¸

(14.7)

Now take the logarithm. Note that there are no terms in log C(k) which are of order

D

2

. Dropping terms that are independent of D,

log C(k) ≈ 2DX cos k −2DX

2

sin

2

k +Dlog X (14.8)

Now let us Fourier transform back to C(0). From (14.5) we get

C(0) = X

D

1

2π

π

−π

dk e

2DX cos k−2DX

2

sin

2

k

. (14.9)

For large D the integral is calculated by the method of stationary phase. We ﬁnd the

maximum value of the exponent with respect to the integration variable k and expand

about it.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 14.1 Using stationary phase, show that the correlation function C(D, 0) has

the large D behavior

C →

1

√

DX

Xe

2X

D

The prefactor 1/

√

DX is the problem in expanding C(D, 0) directly in a HTE. But the

Fourier transform is free of such non-analytic behavior.

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

14.2 Approach to T

c

75

14.2 Approach to T

c

We have just seen that the inverse correlation length behaves like

m = [ log X[ −2X (14.10)

In the high temperature limit m is inﬁnite but as X increases the correlation length 1/m

increases. The critical point T

c

is the point at which m(T) vanishes and the correlation

length tends to inﬁnity. Typically near the critical point m(T) behaves like

m(T) ∼ (T −T

c

)

ν

(14.11)

For temperatures lower than T

c

the correlation function does not go to zero at large

D. Right at T

C

, C neither exponentially decays, nor is it non-vanishing at inﬁnity.

What does happen is that C falls to zero as a power of D. Thus at T = T

c

we write

C(a, b) = D

−η

(14.12)

The constants ν and η are called critical exponents. The constant η is often replaced

by η = 2∆ where ∆ is called the dimension of the “ﬁeld” σ.

The general theory of the critical behavior is the renormalization group (RG). I will

just touch on it and give the main idea.

14.3 Renormalization Group

For the moment forget that the spin σ can only take on the values ±1. In place of σ

we introduce a continuous variable f

i

at each lattice site. Think of f as a ﬁeld which

ranges over the real numbers. Consider a function F of all the spins. It represents

the integrand of the partition function, i.e., F = exp −βE(f). The partition function is

given by

Z =

¸

i

df

i

F(f). (14.13)

Next, let us coarse-grain the lattice by grouping four sites together and call the

resulting 2 2 square a supersite. On each supersite deﬁne a variable f

′

(sometimes

called a block variable).

f

′

=

λ(f

1

+f

2

+f

3

+f

4

)

4

(14.14)

76 14 WHAT HAPPENS AT T = T

C

?

where λ is a number to be chosen later.

Now let us carry out the sum in equation (14.9) in two steps. First integrate over

the f

i

holding ﬁxed the values of f

′

on each supersite. That gives us a function

F

′

(f

′

) (14.15)

of all the block variables. The result is a new statistical mechanical problem on a lattice

with twice the spacing as the original lattice and one quarter as many variables. In the

inﬁnite lattice limit we can just think of it as another problem on the exact same lattice

but we will have to remember that physically, it has been re-scaled.

Consider the operation of going from the function F to the function F

′

. Let us call

that operation B.

BF = F

′

. (14.16)

If we think of F being determined by a set of parameters ¦g¦ and F

′

by a similar set

¦g

′

¦ then B just represents the functional dependence of g

′

on g. Schematically,

g

′

1

= B

1

(g

1

, g

2

, g

3

, )

g

′

2

= B

2

(g

1

, g

2

, g

3

, )

g

′

3

= B

3

(g

1

, g

2

, g

3

, )

.

.

. (14.17)

Once we have the functions B we can iterate the process and thin the lattice by

another factor of 2 in spacing. Thus we can consider a sequence of F functions – call

them F

n

– that all represent the same problem after n lattice doublings.

F

n+1

= BF

n

. (14.18)

In other words we have a recursion relation for F. If we have an initial starting point we

may iterate the procedure and deﬁne a sequence of progressively coarse-grained versions

of the problem. To complete the problem we need to specify in initial condition. For

the Ising model the initial condition is

F(f) =

¸

i

δ(f

2

i

−1)e

βJ

¸

∗

f

i

f

j

(14.19)

Critical points correspond to ﬁxed points of B, i.e., values of the parameters of F

which reproduce themselves:

14.3 Renormalization Group 77

F

fp

= BF

fp

. (14.20)

Under certain conditions a given starting point may ﬂow to the ﬁxed point after

many iterations of the RG transformation B. That is what a critical point is; a starting

point that ﬂows to a ﬁxed point.

Let us consider the behavior of the correlation function C(D) at the ﬁxed point.

Consider a doubling of the lattice spacing. Obviously, since the problems are identical

on the original and the doubled lattice we can expect

'f

′

(0)f

′

(D)` = 'f(0)f(D)` (14.21)

but we must remember that on the left side, D represents a physical distance twice as

big as it does on the right side.

Rewriting (14.17) using

f

′

=

λ(f

1

+f

2

+f

3

+f

4

)

4

and the fact that for large D all the f

i

in the supersite at the origin are separated from

those in the supersite at D by distance 2D on the original lattice, we ﬁnd

C(2D) = λ

−2

C(D). (14.22)

It follows that C varies as a power of D as in (14.8) and the critical exponent η satisﬁes

η = 2 log λ (14.23)

or in terms of the dimension ∆

∆ = log λ (14.24)

The implication of all of this is that at the critical point the physics is invariant

with respect to re-scaling the lattice, as long as we re-scale the average (over the 4 sites

in the supersite) by a factor of λ. Moreover it is determined by the recursion relation

B, not the exact starting point. After many iterations the exact initial state becomes

irrelevant and the behavior tracks to the ﬁxed point. For example, the critical behavior

does not depend on whether there are second nearest neighbor interactions, or whether

the original lattice is a square, triangular, hexagonal lattice. In fact the long-distance

behavior of the correlation function at T

c

becomes isotropic (rotationally symmetric).

78 14 WHAT HAPPENS AT T = T

C

?

14.4 Mean Field Theory

Mean ﬁeld theory is another type of approximation that is useful for approaching the

critical point from the low temperature side. We have seen that the low T behavior

involves spontaneous magnetization, or spontaneous symmetry breaking. The symmetry

of the Ising model is the discrete reﬂection σ →−σ. The magnetization is proportional

to the average of σ and is non-zero even as the external ﬁeld tends to zero. The quantity

s ≡ 'σ` is an example of an order parameter, i.e., a quantity which must be zero unless

the symmetry of the system is spontaneously broken.

Mean ﬁeld theory is an approximation whose validity depends on the dimension of

the lattice: the higher the dimension d, the better the approximation. It becomes exact

in the limit of high dimension, but it completely fails for d = 1. It is order-of-magnitude

accurate for d = 2 and reasonably accurate for d = 3.

Let us begin by assuming that the order parameter is non-zero at temperature β

1

.

Let us focus on a given spin σ. It is surrounded by 2d neighbors which it interacts with

through the usual interaction −Jσσ

j

. If d is very large we can assume that the σ sees

an average spin s for each of its neighbors. In other words the energy of the spin is

E = −2dJsσ. (14.25)

It is also true that if the number of neighbors is large, the ﬂuctuations about the average

are small. Thus we can treat each spin as being in a mean ﬁeld (mean in the sense of

average) and calculate its average spin.

The partition function for the single spin is

Z =

¸

σ

e

2dβJsσ

= 2 cosh (2dβJs) (14.26)

and the average of s obtained by diﬀerentiating log Z is

'σ` = tanh (2dβJs). (14.27)

Consistency requires 'σ` = s, so we ﬁnd an equation for the order parameter:

s = tanh (2dβJs). (14.28)

This equation is illustrated in Figure 16. The critical temperature is the value at which

14.4 Mean Field Theory 79

Figure 16: The mean ﬁeld consistency relation at high T (left) and low T (right). For

low T the tanh-curve and the linear curve intersect twice: once for positive s and once

for negative s.

the slope of tanh (2dβJs) at s = 0 is unity. In other words

T

c

= 2dJ (14.29)

How good is the mean ﬁeld method? For d = 1 it’s very bad. It predicts a critical

temperature T

c

= 2J where the true value is zero. For d = 2 it’s better. Mean ﬁeld

theory gives T

c

= 4J and the exact solution gives T

c

= 2.27J. MFT is even better in

d = 3 where it gives T

c

= 6J compared to simulations that give T

c

= 4.6J.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 14.2 Here is a proposal for an improved mean ﬁeld approximation. Instead

of focusing on a single spin and treating it in the mean ﬁeld background, focus on a

nearest-neighbor pair in the MF background. Call the pair a and b. The energy of the

pair in the MF is

E = −Jσ

a

σ

b

−J(2d −1)s(σ

a

+σ

b

)

Compute Z for the two-spin system and then compute the average value of σ

a

. (The

average of σ

b

will be the same.)

By equating this average to s you will obtain a MF consistency condition. Use it to

compute the critical temperature and compare it with the previous approximation.

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

Part III

Quantum Gases

83

15 Bose Einstein Condensation

We are going to study quantum gases, i.e., gases at low enough temperature that the

quantum statistics of particles cannot be ignored. There are two kinds of particles in

nature, bosons and fermions.

†

When the temperature of a gas is high a gas of fermions

behaves the same way as a gas of bosons and both behave like a gas of classical particles.

To see why, consider the average kinetic energy of a classical molecule at temperature

T.

p

2

/2M = (3T)/2

We can also say that the square of the uncertainty in the momentum of a particle is of

order

(∆p)

2

= MT

From the uncertainty principle the spread in the wave function of a particle is at

least

¯h

(MT)

1/2

If this is much less than the spacing between particles, the wave functions of the particles

don’t overlap and are orthogonal. In this case the statistics does not matter. The

distance between molecules is of order ρ

−

1

3

. Thus the behavior is classical if

ρ <

(MT)

3/2

¯h

3

.

(15.1)

But for ﬁxed density (15.1) will become violated at low enough temperature and the

wave packets will overlap. Under those circumstances the fermionic or bosonic nature

of the molecules becomes important.

In the case of bosons, the statistics favor particles being in the same state. The more

particles in a given quantum state, the more likely it is that an additional particle will

wind up in that state. From the Maxwell Boltzmann distribution, the most likely state

for a particle is the state of lowest energy. In the bose gas at low temperature there will

be a much larger number of molecules in the lowest energy state than in the classical

gas.

†

Bosons (photons, gluons, etc.) can occupy the same quantum states whereas fermions (electrons,

protons, neutrons, etc.) cannot.

84 15 BOSE EINSTEIN CONDENSATION

At this point it might be a good idea to go back and review the theory of black body

radiation in Section 10. Photons are bosons and the same methods apply to ordinary

bosonic molecules.

Non-relativistic bosons of mass M are described by a wave ﬁeld Ψ that is analogous

to the electromagnetic ﬁeld although it has only a single (complex) component. The

ﬁeld can also be decomposed into oscillators. If the molecules are conﬁned to a box of

linear dimensions L the wave numbers satisfy

k

x

=

πn

x

L

k

y

=

πn

y

L

k

z

=

πn

z

L

(15.2)

with n being integers.

The momentum of a particle is given by p = ¯ hk and the energy by

E =

p

2

2M

=

¯h

2

k

2

2M

=

¯h

2

π

2

2ML

2

n

2

. (15.3)

The frequency of the oscillators is

ω

n

=

¯hπ

2

2ML

2

n

2

(15.4)

Using the same method that we used for thermal radiation, the partition function is

Z =

¸

n

1 −e

−β

¯ h

2

π

2

2ML

2

n

2

**However, there is one feature of the non-relativistic gas not shared by a gas of
**

photons. Non-relativistic molecules are conserved whereas photons are easily emitted

and absorbed. In order to account for the conservation of molecules we must introduce

a chemical potential. This is equivalent to shifting the energy of each particle by µ. The

grand canonical partition function has the form

Z =

¸

n

1 −e

−β

¯ h

2

π

2

2ML

2

n

2

−βµ

−1

(15.5)

85

The Helmholtz free energy is

A = −T log Z = T

¸

n

log

1 −e

−β

¯ h

2

π

2

2ML

2

n

2

−βµ

(15.6)

The next step is to replace the sum by an integral. Let us ﬁrst change variable from

the integers n

i

to u

i

where

u

i

=

¯hπ

√

2ML

n

i

Note that the relation between sums and integrals is

¸

n

→

√

2ML

¯hπ

3

d

3

u

Thus the free energy becomes

A = T

√

2ML

¯hπ

3

d

3

ulog

1 −e

−β(u

2

+µ)

(15.7)

or

A = T

π

2

√

2ML

¯hπ

3

u

2

dulog

1 −e

−β(u

2

+µ)

(15.8)

Another form for A is obtained by deﬁning v = u

2

(v is just the energy of a particle).

A = T

π

4

√

2ML

¯hπ

3

√

vdv log

1 −e

−β(v+µ)

(15.9)

The average number of particles is given by N = −

∂A

∂µ

. Dividing N by the volume

we get the density of particles ρ.

ρ =

π

4

√

2M

¯hπ

3

√

vdv

1

e

β(v+µ)

−1

. (15.10)

We can see from (1.10) that µ must be positive or else the integral diverges. But as

µ →0 the density approaches a ﬁnite critical limit.

ρ

0

=

π

4

√

2M

¯hπ

3

√

vdv

1

e

βv

−1

. (15.11)

86 15 BOSE EINSTEIN CONDENSATION

or, letting z = βv

ρ

0

=

π

4

√

2MT

¯hπ

3

√

zdz

1

e

z

−1

. (15.12)

The integral is approximately equal to 2.3. One ﬁnds

ρ

0

=

0.165

¯h

3

(MT)

3/2

. (15.13)

We can also think of varying the temperature at ﬁxed density. In this case a critical

limit is reached when the temperature reaches

T

0

=

3.3¯ h

2

M

ρ

2/3

. (15.14)

As that temperature is approached from the high temperature side, the chemical po-

tential drops to zero. Alternately, at ﬁxed temperature, when the density approaches

ρ

0

the chemical potential becomes zero. Since µ must remain greater or equal to zero,

it would appear that there is an obstruction to increasing the density or decreasing the

temperature past the critical value. However, this is not correct. What does happen is

that the approximation of going from a sum to an integral breaks down. In particular

the integrand in (1.9) vanishes like

√

v when v → 0, but in the sum, the ﬁrst term in

(1.6) does not vanish. The ﬁrst term represents particles with zero energy in the large

volume limit. When the density grows above ρ

0

all the extra particles go into the lowest

energy state. In other words, beyond the phase transition, the density is increased by

increasing the number of zero energy particles. The number of such particles is

N

0

= V (ρ −ρ

0

) (15.15)

where V is the volume. The zero energy state is the only state that has a macroscopic

number of particles (a ﬁnite fraction of the total), but only when the temperature is

below the critical temperature. The macroscopic occupation of the ground state is called

a Bose Einstein Condensate.

Below the phase transition in the Bose Einstein condensed phase, the chemical

potential is zero and the free energy is

A = T

π

4

√

2ML

¯hπ

3

√

vdv log

1 −e

−βv

(15.16)

87

or

A = T

5

2

π

4

√

2ML

¯hπ

3

√

zdz log

1 −e

−z

(15.17)

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Problem 15.1 For T < T

0

compute the energy, heat capacity, entropy, and pressure

of the B.E. condensed gas.

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

The number of particles with energy greater than zero (those not in the zero-energy

condensate) is obtained from (15.13):

N

E>0

= ρ

0

V ∼ V T

3/2

(15.18)

On the other hand for T ≥ T

0

the total number of particles is essentially equal to

the number with E > 0. So we expect that N

E>0

→ N

total

as T → T

0

. We can write

this in the form

N

E>0

= N

total

T

3/2

T

0

3/2

(15.19)

The remaining particles are those in the zero energy condensate:

N

E=0

= N

total

1 −

T

3/2

T

0

3/2

(15.20)

88 A APPENDIX A: SOME BASIC MATHEMATICS

A Appendix A: Some basic mathematics

Transcendental functions

You should be familiar with the exponential and log functions (natural logs, not base-10

logs), trigonometric functions, and the hyperbolic functions

sinhx ≡

e

x

−e

−x

2

cosh x ≡

e

x

+e

−x

2

tanh x ≡

sinhx

cosh x

=

e

x

−e

−x

e

x

+e

−x

These are called the hyperbolic sine, cosine and tangent respectively, and are typically

pronounced “cinch,” “cosh,” and “tanj” or “tan-h.” Important formulas include

d

dx

sinhx = coshx

d

dx

cosh x = sinhx

cosh

2

x −sin

2

x = 1

Calculus

We assume you know the rules for diﬀerentiation (total and partial) and integration

(deﬁnite and indeﬁnite). The notation we use may be unfamiliar to you:

dx dy f(x, y)

means the same as

f(x, y) dx dy

Also ∂

h

A means ∂A/∂h.

89

Probability

The idea of a probability distribution should be familiar to you. The area underneath

a probability distribution between x

1

and x

2

is the probability of a result between

those two limits. The best-known probability distribution is the normal or Gaussian

distribution

P(z) =

1

√

2π

e

−z

2

Necessarily,

∞

0

P(z)dz = 1

Fourier transforms

Fourier analysis is a standard technique for converting periodic functions into trigono-

metric series. Given a function f(t) deﬁned on an interval from −L to L, the Fourier

series representation of f(t) is

f(x) =

a

0

2

+

∞

¸

n=1

a

n

cos

nπx

L

+b

n

sin

nπx

L

where

a

n

=

1

L

L

−L

f(x) cos

nπx

L

dx, n = 0, 1, 2,

b

n

=

1

L

L

−L

f(x) sin

nπx

L

dx, n = 1, 2, 3,

Using the Euler identity e

iθ

= cos θ +i sin θ we may write

f(x) =

1

2

∞

¸

n=−∞

c

n

e

iωnx

where ω

n

= n ∗ pi/L and

c

n

=

1

L

L

−L

f(x)e

−i∗omeganx

dx, n = 0, ±1, ±2,

A sawtooth function is shown in Figure 17 with approximations to n = 1 and n = 7.

90 A APPENDIX A: SOME BASIC MATHEMATICS

−2π −π π 2π

Figure 17: A sawtooth wave approximated by one-term and four-term Fourier series.

Lagrange multipliers

The method of Lagrange multipliers allows us to ﬁnd minimum or maximum values

of a function subject to equality constraints. It is explained in many basic texts;

wikipedia.org has a good short article on the subject.

We will illustrate the method with a function of two variables subject to a single con-

straint: minimize f(x, y) subject to a constraint g(x, y) = 0 The minimum or maximum

value of a function is found by setting its derivative(s) to zero, provided the function is

suitably smooth.

The constraint function is a line in x, y space. The gradient at any point on that

line is a vector perpendicular to the line g(x, y) − c. The gradient of the f(x, y) is a

vector perpendicular to its contour lines. A constrained maximum or minimum is found

where the two vectors are perpendicular, i.e.,

∆f = λ∆g

91

where ∆f is deﬁned as the vector with components

∂f

∂x

,

∂f

∂y

**λ is called a Lagrange multiplier. We solve the problem by minimizing (maximizing) an
**

augmented function

Λ(x, y, λ) = f(x, y) +λ(g(x, y) −c)

Setting the three partial derivatives of this function to zero gives us three equations that

can be solved for x

m

, y

m

(the coordinates where the function is minimized or maximized)

and λ, the Lagrange multiplier:

∂f

∂x

= 0

∂f

∂y

= 0

∂f

∂λ

= 0 = g(x, y) −c

This last equation is of course a restatement of the constraint equation.

As an example,

‡

maximize f(x, y) = x+y subject to the constraint x

2

+y

2

= 1 The

constraint is the unit circle and f(x, y) is a tilted plane. Thus

Λ(x, y, λ) = f(x, y) +λ(g(x, y) −c) = x +y +λ(x

2

+y

2

−1)

∂Λ

∂x

= 1 + 2λx = 0

∂Λ

∂y

= 1 + 2λy = 0

∂Λ

∂λ

= x

2

+y

2

−1 = 0

Solving these equations yields two solutions, x

m

= y

m

=

√

2/2, for which f(x, y) =

√

2

and x

m

= y

m

= −

√

2/2, for which f(x, y) = −

√

2

‡

Source: wikipedia.org

92 B APPENDIX B: PHASE SPACE

B Appendix B: Phase space

We are all familiar of two-dimensional plots with some physical variable shown on each

axis, such as PV diagrams (pressure versus volumes). Phase space is a generalization

of this idea to multiple dimensions. Concepts like line, area, and volume are likewise

generalizations of geometric concepts. Thus an “area” on a PV diagram has dimensions

of pressure times volume (force per unit area times volume = force times length =

energy).

You should have no diﬃculty as long as you remember that areas and volumes in

phase space are not geometric concepts.

93

C Appendix C: Ludwig Boltzmann

Ludwig Boltzmann (1844-1906) was Professor of Physics at the University of Vienna and

the last of the great classical physicists. He viewed the Second Law as a manifestation

of probability, but struggled to establish a clear relationship. Then it occurred to

him that probabilities are multiplicative while entropy and other extensive quantities

are additive. What function takes multiplicative properties into additive properties?

One day it occurred to him the logarithm function just does that. For a brief time,

Boltzmann found himself in that elevated state, one that all scientists dream of and a

few occasionally achieve: having discovered something important that no one else yet

knew. His famous formula is inscribed on his tombstone in Vienna (Figure 18).

Figure 18: Ludwig Boltzmann’s tomb, inscribed with S = log P

Ludwig Boltzmann was a colorful character to say the least.

∗

. He was a large man,

accustomed to consuming copious quantities of food and drink. He was a ﬁne pianist

∗

von Baeyer, “Maxwell’s Demon,” Random House, 1998, Chapter 13

94 C APPENDIX C: LUDWIG BOLTZMANN

and once performed Schubert sonatas for William Randolph Hearst. In 1905, though in

ill health, he undertook a grueling journey from Vienna to Berkeley to deliver a summer

series of lectures. He found to his dismay that the City of Berkeley was dry. But he

soon learned to smuggle cases of wine across the line from Oakland.

Sadly, Boltzmann suﬀered from depression and perhaps bipolar disorder. Not long

after his return to Vienna in 1906, he took his own life.

95

D Appendix D: for Engineers

Engineers and physicists labor in somewhat diﬀerent worlds and each group has certain

conventions that suit their purposes. Readers who have been trained in engineering

thermodynamics may need to be careful about adjusting to the conventions in this

book, especially dimensions and units.

Some common engineering units in thermodynamics are as follows:

SI units English units

Temperature C, K F, R

Energy joule (J) ft-lb

Power watt (w) hp, ft-lb/sec

Entropy J/K ft-lb/R

Boltzmann’s constant k 1.38 10

−23

J/K 1.04 10

−23

ft-lb/F

(S = k log P)

Stefan-Boltzmann constant σ 5.67 10

−8

W/m

2

K

4

0.1714 10

8

BTU/hr ft

2

R

4

( ˙ q = σT

4

)

In this text, T has units of energy and can be thought of as equal to kt where t is

the “textbook temperature” in degrees K (SI units), and k is Boltzmann’s constant in

J/K. Also, we measure entropy in bits, a dimensionless quantity, as contrasted with the

engineering deﬁnition that involves a factor of k, again in J/K.

There should be no diﬃculty with units as long as you remember to include the

proper constants when performing numerical calculations.

Preface

A number of years ago I became aware of the large number of physics enthusiasts out there who have no venue to learn modern physics and cosmology. Fat advanced textbooks are not suitable to people who have no teacher to ask questions of, and the popular literature does not go deeply enough to satisfy these curious people. So I started a series of courses on modern physics at Stanford University where I am a professor of physics. The courses are speciﬁcally aimed at people who know, or once knew, a bit of algebra and calculus, but are more or less beginners. This volume is based on notes prepared for a Continuing Education course I gave at Stanford in the spring of 2009, and for a graduate course in statistical mechanics that I gave at that same time. – Leonard Susskind

I am one of those “physics enthusiasts,” an engineer who grew tired of mass-market physics books shorn of all equations. I picked up the standard “Gravitation” text to try to teach myself general relativity, but even though I have a fairly good math background, it was too much. So I was delighted to ﬁnd Prof. Susskind’s course sequence at Stanford and to have the opportunity to help with the preparation of this volume. – Warren Gibson

Contents

1 Conservation of Information, E, S, and T 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . 1.2 Probability Theory . . . . . . . . . . . . . . . . . . 1.3 Energy . . . . . . . . . . . . . . . . . . . . . . . . . 1.4 Entropy . . . . . . . . . . . . . . . . . . . . . . . . 1.5 Temperature . . . . . . . . . . . . . . . . . . . . . 1.6 The Zeroth and Second Laws of Thermodynamics . 2 The 2.1 2.2 2.3 2.4 Boltzmann Distribution Solving for the Lagrange Multipliers Helmholtz Free Energy . . . . . . . . Why is T = 1/β? . . . . . . . . . . . Ensembles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 1 4 4 7 8 9 12 12 13 13 15 17 17 19 23 23 25 26 27 29 31 32 34 37

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

3 Fluctuations 4 Control Parameters 4.1 The Adiabatic Theorem and the First Law . . . . . . . . . . . . . . . . 4.2 Processes at Fixed Temperature . . . . . . . . . . . . . . . . . . . . . . . 5 Dilute Gas 5.1 Ideal Gas . . . . . . . . . . . . . . . . . . . . . . . 5.2 Almost Ideal Gas . . . . . . . . . . . . . . . . . . . 5.3 Ideal Gas in a Potential Field . . . . . . . . . . . . 5.4 Diatomic Molecular Gas . . . . . . . . . . . . . . . 5.5 A Note about Volume and the van der Waals Gas

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

6 Simple Magnet and Magnetization 6.1 Ising Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 The Maxwell Relations 8 The Second Law

. .1 High Temperature Expansion of the Correlation Function 14. . . . . . . . . .3 Renormalization Group . . . . . . . .2 Low Temperature Expansion . . . . . . . . . . . . . .and Two-Point Functions in the HTE 13 High-T/Low-T Duality 14 What Happens at T = TC ? 14. . . . .2 Approach to Tc . . . . . . . . . . . . . . . . . . . .4 Mean Field Theory . . . . . . . . . . . . . 14. . . .1 Connection with Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . 12. . . . . . . . . . . . . . . . . . . .5 One. . . . . . 15 Bose Einstein Condensation A Appendix A: Some basic mathematics B Appendix B: Phase space C Appendix C: Ludwig Boltzmann D Appendix D: for Engineers 41 44 46 49 51 55 57 62 66 67 72 72 75 75 78 83 88 92 93 95 . .9 Quantum Considerations 10 Thermal Radiation 11 Chemical Potential and Grand Canonical Ensemble 12 Back to Magnets 12. . . . . . . . . . . . . . . . . . . 12. . . . . . . . . . . . . . . .4 High Temperature Expansion . . . . . . . . . . . . . . . . . . . . . . . . . 12. . . .3 Ising-like Systems in Higher Dimensions . . . . . . . . . . . . . . . . . . 12. . 14. .

. . . . . . . . . . . . . . . . The phase space volume and topology are exactly conserved but when coarse grained it grows. Two canceling graphs. . . The Second Law. . . . Phase diagram of a magnet. . . . . . . . . . . . . . . . . . . . The contributions to the HTE coming from a single link are represented by a heavy line on that link. Magnetization vs external Magnetic ﬁeld. . . . . . . . . . . . . . .” . . . . . . Closed “curves” on the lattice . . . . . . . .List of Figures 1 2 3 4 5 6 7 Two simple “laws of motion. . . . . . . . The mean ﬁeld consistency relation at high T (left) and low T (right). . . . . . . . . . . . . . . . . . . . . . . . . . . . . Experiment Illustrating First Maxwell Relation . 2-dimensional lattice of spins. In the 1-D Ising model the critical temperature T ∗ is zero. . . . The o’s indicate up-spins. . . Another example. . . . . . . . . . . . . . . . . . . . . . . . Surround each island of ﬂipped spins with a curve on the dual lattice. . . . . . . . . . . . . . . . . . . Ludwig Boltzmann’s tomb. . . . A law of motion with two trajectories. . . . . . . . . . . . . . . . . . . . . . . Points a and b are separated along the x-axis by distance D and along the y-axis by distance n. . . . The dual lattice is composed of points at the centers of the Plaquettes. . . . inscribed with S = log P . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . For low T the tanh-curve and the linear curve intersect twice: once for positive s and once for negative s. . In the analogy with molecular ﬂuids. the x’s indicate down spins and the •’s indicate dummy up spins. . . . . . . . . . . A sawtooth wave approximated by one-term and four-term Fourier series. . . 2 3 10 35 39 51 54 57 63 65 67 68 70 71 73 8 9 10 11 12 13 14 15 16 17 18 79 90 93 . . . . . . . . . . . . . . . . . . . . . . . . and positive magnetization represents the liquid phase. . . . . . . . . . . . . . . . . negative magnetization represents the gas phase. The thick line represents the discontinuity in the magnetization. . . . . . . . Sharply peaked occupation numbers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.

Part I General Principles .

.

1. solids. but for our purposes a few simple concepts will suﬃce.1 1 Conservation of Information. green.∗ So to say that Statistical Mechanics is the theory of gasses is rather like saying calculus is the theory of planetary orbits. Here is another example. Y. yellow. Although our examples will focus on thermodynamics and magnetics. blue.2 Probability Theory There are many good texts that explain the theory of probability. P ). we might ﬂip a coin several thousand times and observe that we got heads (or tails) very nearly half the time and conclude that this will always be the most likely outcome. also use statistical mechanics. orange. G.” often refugee physicists who sought fame and fortune on Wall St. a theory of how atoms combine to form gases. and purple (R. In these cases statistical methods can be used to study the aggregate behavior of a system without knowing (or indeed caring about) the behavior of each element of the system. so we must assign 1/2 to each probability.1 Statistical mechanics is often thought of as an approach to thermodynamics. The cubic symmetry of the die dictates that the probabilities of any one face appearing on top when the die is rolled ∗ J.” Springer. we can say that the physical symmetry of a coin allows us no reason to believe that either outcome is more likely than the other.. Volt. perhaps with dubious results. Entropy. liquids. My son. “The Statistical Mechanics of Financial Markets. We will review them now starting with coin ﬂipping. Taking an a posteriori view. uses it. other applications are possible. “Financial engineers. That’s how calculus may have started. and even plasmas and black body radiation. O. It is more because it can be used in many areas of science where a large number of variables is involved – too many to address individually. Why do we say that? Taking an a priori view. and Temperature Introduction 1. but it has of course found vastly more applications. The probabilities of a coin coming up heads (H) and tails (T) are both equal to 1/2. Energy. B. Let’s take a die (as in dice) and color the six faces red. 2005 . But it is both more and less than that. who studies neural networks. or somewhat more generally.

2 1 CONSERVATION OF INFORMATION. and that is so. Y → R. Y → R is such a rule (Figure 1). you say: the real symmetry is among the 6 faces. What fraction of the time does that die spend in each conﬁguration? Figure 1: Two simple “laws of motion. Suppose there is a rule that uniquely speciﬁes the conﬁguration of a system based solely upon its current conﬁguration.” The answer is 1/6. O → Y . S. I chose to color the purple face red. B → G. For example R → B. B → G. AND T are all 1/6. Nonsense. E. G → O. Again. we have to think of the system in question as part of a bigger system. P → O. As is often the case. or could be inferred from experiments. O → Y . for example if the die is weighted in some unfair way? If we still wanted to pursue a theoretical or a priori approach we would have to rely on a bunch of details such as the precise way the die was thrown by the hand that threw it. B → R. But what if we don’t have an obvious symmetry to rely on? How do we assign a priori probabilities? Suppose for example instead of the coloring scheme that I indicated above. O → B. this is obvious from physical symmetry. B.” with time in this example proceeding in discrete steps. O → Y . the 5 names are just as symmetric as the original 6 names. Y. G → P . Would the probability of throwing a given color be 1/5? After all. Y → P . such as R → B. P → G. We may call this rule a “law of motion. O). In fact there are many possible laws for which the answer will be the same. if I just write (R. G → R. maybe even the surface that the die lands of (can it bounce?). Then there would be only ﬁve colors. P → O. B → G. But what about the bigger system? How do we assign its probabilities? Here is another idea that involves some dynamics. the wind. G. But what about the rule R → B. Y → P . or R → Y . But what if there really is no obvious symmetry at all. P → O. G → P .

1.2

Probability Theory

3

(Figure 2)? In this case there are two trajectories through the space of states. If we

Figure 2: A law of motion with two trajectories.

are on one of them we don’t ever jump to the other. So the probability will depend on where we started. We may wish to assign probabilities for beginning on either of the two cycles. In this last case there is a conserved quantity. Suppose we assign the number one to R, B, G and zero to O, Y, P . Let’s call this quantity the “Zilch number.” Obviously Zilch is conserved. Whenever we have a conserved quantity like Zilch, we have to specify its value. That’s not a problem, as there are many conserved quantities in nature, the most important in statistical mechanics being energy. Once we specify all the conserved Zilch numbers, we can proceed as usual, and say that all the states on a trajectory with a particular Zilch numbers are equally probable. That sounds good, but there are lots of counterexamples. Here is one: R → R B → R, G → R, P → R, O → R, Y → R. No matter where you begin, you go to red in the next instant and then stay there. There are no conserved quantities but obviously after the ﬁrst step the probabilities are completely unequal; only red is possible. This last law has something odd about it. It is an example of a law that does not respect the “conservation of distinction,” or “conservation of information.” In the previous examples distinct starting points always led to distinct outcomes. But in this case we quickly lose track of where we started. Trajectories merge! One of the central principles of classical mechanics (and it has an analog in quantum mechanics) is that information (distinctions) is conserved. This principle is so important to the validity of statistical mechanics and thermodynamics that I would call it the “minus ﬁrst” law of thermodynamics (the “zeroth law” is already taken.) Later we will use an abstract concept called “phase space” in which the coordinates

4

1 CONSERVATION OF INFORMATION, E, S, AND T

represent not physical space, but variables that may be unrelated to space. For now, let’s just note that conservation of distinctions says that trajectories in phase space never run into each other. Even stronger, if you start with a volume in phase space and follow it using the Hamilton’s equations of motion, the volume is conserved. That suggests that for a closed system, a priori probability is uniform over the phase space. We will come back to this.

1.3

Energy

The First Law of Thermodynamics says is that energy is conserved. For a closed system,† dE =0 dt If a system is composed of two very weakly interacting subsystems,‡ the energy is additive: E = E1 + E2 Interactions between the subsystems can exchange energy between them. Thus the energy of either subsystem is not conserved. In equilibrium the energy of a subsystem ﬂuctuates although the total energy is ﬁxed.

1.4

Entropy

Entropy is an information-theoretic concept. It is a measure of how much we don’t know about a system. Our ignorance may stem from the fact that the degrees of freedom are too small to measure, too numerous, or too rapidly changing. The reason could just be our own laziness. In any case, entropy measures our ignorance. Suppose a system can be in any of N states. If we know nothing about the state of the system (complete ignorance) the entropy S is deﬁned as S = log N.

A closed system is one with a time-independent Hamiltonian and with insulating walls so that no energy can enter or leave the system. Weakly interacting means that the energy of interaction is very small compared to the separate energies of the subsystems by themselves. If the Hamiltonian is H1 + H2 + Hint , then the numerical value of Hint should be negligible compared with either H1 or H2 .

‡ †

1.4

Entropy

5

Suppose we know only that the system is an any of M states with M < N . Then we have some nontrivial knowledge. In this case the entropy is deﬁned to be S = log M. Evidently, the more we know, the smaller the entropy. Here are some further examples. Consider a system of n distinguishable coins. Each coin is in one of two states: H (heads) or T (tails). A state consists of an ordered sequence of H’s and T’s such as H H T H T T H ··· Suppose we know nothing at all. The total number of possible states is 2n and the entropy is S = log 2n = n log 2. Note that by taking the log we have made entropy an extensive property, that is, proportional to the number of degrees of freedom n. Entropy is measured in bits. If the entropy is S = log 2n = n log 2, one says that there are n bits of entropy. As another example, suppose we know everything that there is to know about the coins. In that case we know the exact state and M = 1. The entropy is log 1 which is zero. Next, suppose we know that n − 1 coins are in the state H and one is T, but we don’t know which one. The system can be in any of n states and the entropy is S = log n In this case the entropy is not additive but that is because there are strong correlations (to be deﬁned later). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problem 1.1 Suppose that n is even and that we know that half the coins are H and half are T, but that is all we know. What is the entropy? Problem 1.2 For large n what is the entropy if we know that a fraction f of the coins are H and fraction 1 − f are T?

. ... AND T .. Obviously each P (i) satisﬁes 0 ≤ P (i) ≤ 1 which implies that S is positive and that each contribution to (1....... Since there are M such states the sum gives log M... ... . The entropy deﬁned by (1. . . We can write a formula for S in terms of P (i). i (1..6 1 CONSERVATION OF INFORMATION....... . Problem 1....... . In other words eS is the “width” of the distribution.......... . ... .. . ......... . .... In each case we are given a probability distribution on the space of states.. E....3 Suppose you have a set of N coins. In the examples P (i) is zero for those states that are not possible and P (i) = 1/M for the possible states.... . ... .. ..2) is positive or zero. .... Homework ..1) roughly measures the log of the number of states that have non-negligible probability in the distribution P (i). ......2) The general deﬁnition of entropy is given by (1. ... .. .. . For a general probability distribution we require only that P (i) ≥ 0 and P (i) = 1. .... . Note that the width is zero if and only if P = 1 for a single state and zero for all others. .. ....... ... .... S. . . . ..... ......... ...1) Those states with P = 0 contribute nothing and those with P = 1/M contribute (1/M ) log M ... . . . .... .. .... ...... ... ........ . .. . .. .. ...... .. i (1. S=− P (i) log P (i). . .... .4 A variable q has a uniformly spaced spectrum of values with very small spacing δ. . Each coin independently has a probability 1/3 to be heads and 2/3 to be tails. ....1)... ... .. .. .... . .. . ... The sum over states can be accurately approximated by → 1 δ dq 2 i Suppose the probability distribution for q is proportional to e−q . If we label the states with index i then the probability that the system is in state i is called P (i)... What is the total entropy? Problem 1.. What is the entropy? .......... .. .. . . ... . ... .

E) satisﬁes the usual requirements of a probability distribution. Except in very unusual circumstances. S = S(E). E). Later we will think of it as the thermal equilibrium distribution for given average E. the temperature is always positive. T . For example. Our deﬁnitions of temperature and entropy diﬀer in essentially unimportant but potentially confusing ways from engineering deﬁnitions of such quantities. For each value of E. How much heat? The answer is T = log 2. See Appendix D.5 Temperature The average energy associated with a probability distribution (call it E) is given by E= i Ei P (i) (1.4) dS We call the quantity dE/dS the temperature. the temperature of a system is the amount of energy needed to increase its entropy by one bit. P (i.3) where Ei is the energy of state i. i.e. entropy is a monotonically increasing function of energy. Consider the amount of energy that is needed to increase the entropy by one bit (by log 2). where it shows up as some heat. Now suppose we have a one-parameter family of probability distributions labeled by the average energy P (i.. At each value of E we can compute the entropy so that S becomes a function of E.5) . § dE dS (1. But for now it is just a one-parameter family.1.5 Temperature 7 1. if you erase a bit of information from your computer you are really transferring it from the computer out into the atmosphere. T ≡ We should think of temperature as follows:§ Apart from a factor of log 2. It is given by dE δE = log 2 (1.

6) Since they must tend to equilibrium. We can state it more formally as When a closed system which is out of thermal equilibrium comes to equilibrium the entropy of the system increases. S. at diﬀerent temperatures. Now bring them into contact so that energy (heat) ﬂows between them. E. Their energies. TA . . Without loss of generality we can assume that TB > TA . Equation (1. Hence dSA + dSB > 0 (1.8 1 CONSERVATION OF INFORMATION.7) We can use (1.6). (TB − TA ) is positive.6) then tells us that TB dSB is negative. EB . Suppose a small quantity of energy is exchanged. TB . energy ﬂows from hot to cold as equilibrium is established. Equivalently. We ﬁnd (TB − TA )dSA > 0 (1.8) Since the system B is initially the hotter of the two. A and B. Consider two isolated systems . temperature must be uniform in a system in thermal equilibrium. From the Second Law we can prove that heat always ﬂows from hot to cold. SA . SB .7) to eliminate SB from (1. The total change in energy must be zero. The ﬁnal equilibrium conﬁguration in which energy has stopped ﬂowing must have TA = TB . Therefore dSA and also TA dSA are positive. Therefore TA dSA + TB dSB = 0 (1. but for now let’s just accept it. AND T 1. In other words. and entropies are EA . if the entropy is not maximum it must increase. We will come back to the reasons for this.6 The Zeroth and Second Laws of Thermodynamics The Second Law of Thermodynamics says that entropy always increases. temperatures.

Whenever it is true.3) n1 !n2 ! · · · nN ! i ni ! . How many distinct conﬁgurations of the total system are there with a given set (n1 . The ni satisfy two conditions.3) becomes very sharply peaked around some set of These equations state that the total number of copies adds up to N and that the total energy adds up to some ﬁxed value.2) The important point is that when N and n become large. · · · iN with energies Ei1 . not subsystems. subject to the constraints (2.) We will illustrate the consequences of this principle by choosing the bath to be N − 1 copies of A so that altogether we have a total system consisting of N copies of A. (Note that i refers to states of the small system A and not the combined system. This is the assumption of chaos and although it is very diﬃcult to prove for a given system. Ei3 . 2. 3. ni = N i (2. Etotal . Thus the probability for a given partitioning of the energy (given set of ni ) is equal to the number of conﬁgurations with that set of ni . it is almost certainly true for almost all systems. Each of these ways in which the energy can be shared will be equally probable. i3 . i (2. · · · nN )? This is a combinatoric problem that we will leave to you. After a long time the combined system will have jumped around over all states having the given total energy. the quantity in (2. · · · EiN . The copies are labeled 1. i2 . Let us deﬁne the “occupation numbers” ni to be the number of copies of A that occupy the state i.9 2 The Boltzmann Distribution If a system A is in contact with (weakly interacting with) a much larger system called a bath. Ei2 . n2 . the small system A will be in thermal equilibrium with a certain probability distribution for being in the state i. Keep in mind that i sums over states. The answer is N! N! = (2. N and they are in states i1 .2).1) and (2. · · · . The assumption of chaos tells us that all of the conﬁgurations of the total system with a given total energy are equally likely.1) and ni Ei = Etotal . then it can exchange energy with the bath.

i .3).3). N Now (2. namely P (i) = 1 i and P (i)Ei = E. P (i) = ni N and let E be the average energy of a copy. Obviously E= Etotal . maximize (2. let us introduce some changes of notation.2) and (1.10 2 THE BOLTZMANN DISTRIBUTION occupation numbers as in Figure 3 .1) and (2. Before we compute the occupation numbers that Figure 3: Sharply peaked occupation numbers. Deﬁne P (i) to be the fraction of copies in state i.2) take a form identical to (1.

This is the Boltzmann distribution.11 Now we will assume that N and ni are very large and use Stirling’s approximation¶ in (2.2). i In other words the probability distribution for thermal equilibrium maximizes the entropy subject to the constraint of a given average energy. But ﬁrst let us take its logarithm. Substituting n/N = P and Etotal = N E we ﬁnd that this is equivalent to maximizing − subject to P (i) = 1 i P (i) log P (i) i and P (i)Ei = E.3).4) At the end we ﬁnd α.5) has the familiar form P (i) = ¶ (2. Z (2. We introduce two multipliers α and β. .1) and (2.6) Stirling’s approximation is n! ≈ nn e−n or n log n! ≈ log n − n = n(log n − 1). we use the method of Lagrange multipliers (Appendix A).) log N! ≈ N log N − i ni ! ni log ni . We maximize − i P (i) [log P (i) − 1] − α i −Etotal P (i) − β P (i)Ei i (2. one for each constraint. Let us deﬁne Z ≡ e(α+1) Then (2. β values that satisfy the constraints. Diﬀerentiating with respect to P (i) and setting the result to zero gives P (i) = e−(α+1) e−βEi . In order to ﬁnd an expression for P (i) that provides a constrained maximum entropy value. i We want to maximize this subject to the constraints (2. (Maximizing a positive quantity is the same as maximizing its log.5) e−βEi .

12 2 THE BOLTZMANN DISTRIBUTION 2. Next consider the equation P (i)Ei = E i which becomes 1 Z = −∂β e−βEi e−βEi Ei = E. Numerical values of Z are generally not useful. (2. Z= i e−βEi .7) The partition function is a generating function that is used to derive several useful quantities. The Lagrange multiplier β is. E = − ∂β Z = − Z ∂β This equation can be used to ﬁx β in terms of the average energyE. This gives the famous formula for the partition function. i Let us use e−βEi Ei and we obtain (2. (2.1 Solving for the Lagrange Multipliers Solving for α is equivalent to solving for Z which is done by setting the sum of the P to 1.8) ∂ log Z 1 . like a potential function. of course. but ﬁrst let us derive another familiar thermodynamic formula.) Thus we ﬁnd S = −β(E − A) or A = E − TS The notation ∂x y is shorthand for ∂y/∂x. 2.9) . the inverse temperature.2 Helmholtz Free Energy P log P we can write the entropy in the form S= i Using S = − 1 −βEi e (βEi + log Z) = βE + log Z Z The quantity A = −T log Z is called the Helmholtz free energy (reminder: T = 1/β. We will demonstrate this.

10) A is called free energy because it represents the amount of useful work that can be extracted from a system. The customary symbol A comes from the German arbeit.5) and the second is the inverse of the Lagrange multiplier β.3 Why is T = 1/β? We have proposed two deﬁnitions of temperature. dE = −d[∂β log Z] Now use (2. 2. dA = dE − T dS − SdT Using dE = T dS we ﬁnd dA = −SdT (2. (2. 2. In that case the probability distribution is a delta function of energy. Consider a small change in the energy of a system.5) and thus the two deﬁnitions of temperature are the same.∗∗ P (E) = δ(E − E0 ) ∗∗ The Kronecker delta δ(x) is equal to one when x = 0 and zero otherwise. We would like to see that they are really the same.4 Ensembles For a closed system the energy does not ﬂuctuate.2. The ﬁrst is equation (1. work.9) in the form E = ST − T log Z to get dE = T dS + SdT − T d log Z dβ − log ZdT dβ Using (2. we ﬁnd that the last three terms cancel. and the deﬁnition of A. leaving dE = T dS.11) This is of course equivalent to (1. .3 Why is T = 1/β? 13 In its diﬀerential form.9) again.

14 2 THE BOLTZMANN DISTRIBUTION This is called the Microcanonical Ensemble and governs the entire system including the heat bath. For a system embedded in a large heat bath the energy ﬂuctuates and the Boltzmann probability distribution is called the Canonical Ensemble. Later we will also allow the number of particles to ﬂuctuate. The ensemble in that case is called the Grand Canonical Ensemble. .

The second identity is derived the same way as the ﬁrst. Z The ﬁrst identity is the usual identiﬁcation of average energy in terms of the derivative of Z. If x is continuous then the sum is replaced by an integral in the obvious way. Thus (∆E)2 = E 2 − E is given by (∆E)2 = But this expression is equivalent to 2 (∆E)2 = ∂β log Z = −∂β E 2 1 2 ∂ Z− Z β 1 ∂β Z Z 2 . In this section we will illustrate this by considering the ﬂuctuations of the energy of a system in contact with a heat bath.1) (3. We go beyond thermodynamics when we consider ﬂuctuations of quantities about their averages. noting that each derivative of e(−βE) brings down a factor of −E. Let us consider the ﬂuctuation of energy of a system in equilibrium.15 3 Fluctuations Thus far we have been deriving classical thermodynamics from statistical mechanics.2) where X means the average value of X. Such ﬂuctuations are observable – for example in Einstein’s theory of the Brownian motion. For any function f (x) the average is deﬁned by f (x) = x f (x)P (x). Using T = 1/β we get (∆E)2 = T 2 dE . Given a probability distribution P (x) the ﬂuctuation in x (called ∆x) is deﬁned by (∆x)2 = (x − x ) which is also equal to (∆x)2 = x2 − x 2 2 (3. We use the following: 1 E = − ∂β Z Z and 1 2 E 2 = ∂β Z. dT .

3) Thus we ﬁnd that the ﬂuctuation of the energy is proportional to the speciﬁc heat. But that is because we have set the Boltzmann constant to 1. Despite their small magnitude. seconds. . and degrees Kelvin: k = 1. ∆E = 1. (3. It may seem odd that the ﬂuctuations should appear to be so large. however. If we put k back into the equation it becomes (∆E)2 = kT 2 C. (3.16 3 FLUCTUATIONS Now note that dE/dT is just the heat capacity of the system.4 × 10−23 .4) k is a very small number in units of meters.2 × 10−9 J which is a very small amount of energy. kilograms. ﬂuctuations can be measured and in fact such measurements provide a method for determining Boltzmann’s constant.4 × 10−23 J/K × (293K)2 × 4. For a kilogram of water at room temperature.18J/(kg × K) × 1kg = 2. Call it C. The ﬁnal identity is (∆E)2 = T 2 C.

If you want to think of a speciﬁc example.” A system is in an eigenstate if it is undergoing periodic motion free of any external disturbances. That is why entropy is an adiabatic invariant. In that situation the system will remain in equilibrium throughout the process. even as the energy along the way gradually varies. and the shape and volume of the containers that enclose the system. X can represent the volume of the system. it means that the control parameters are varied very slowly. Obviously if P (i) is constant. The implication is that the probability function P (i) is constant for each level.1 The Adiabatic Theorem and the First Law An adiabatic process∗ means two things. so is the entropy.17 4 Control Parameters So far we have considered closed systems characterized by constant values of the parameters in the Lagrangian such as the masses of particles. the system is isolated so that no energy in the form of heat can ﬂow into or out of the system. Those things which do not change during an adiabatic process are called adiabatic invariants. 4. Some of these parameters such as the volume of the system and the external ﬁelds may be controllable from the outside. although typically the energy and temperature will change.† But if the change is arbitrarily slow then the adiabatic theorem says that the system remains in an eigenstate. In general they will depend on X: Ei = Ei (X). simply tracking the slow time dependence of the instantaneous energy levels. If a sudden change is made in X the system will not remain in an eigenstate of energy. Now consider the change in energy of the system during the adiabatic process. The most familiar example is the work done in slowly compressing a gas in a insulated container. for example by moving pistons to change the volume. We will call such macroscopic control variables Xm . That energy change. It is a theorem that entropy is an adiabatic invariant. Secondly. . Consider the energy levels of the system Ei . In fact the levels will not cross over or disappear. is by deﬁnition. This theorem is true in classical mechanics but it is most easily understood in quantum mechanics. For simplicity we will consider the case of only one X although the principles are the same for several of them. First. the values of external electric and magnetic ﬁelds. ∗ † “Adiabatic” is from the Greek “impassable. the work done on the system by changing X.

2) The most familiar example of conjugate variables (X.1) S (4. dW = −P dV Let us suppose that an inﬁnitesimal adiabatic change is followed by a second process in which energy is added to the system in the form of heat – in other words.3) in the form dQ = dE + Y dX. We can express this idea in equations. But there is no function Q. dW = ∂E ∂X dX S In the general case of several control parameters this becomes dW = n ∂E ∂Xn dXn .18 4 CONTROL PARAMETERS If the change in X is small (call it dX) we may assume the work done is small (call it dW ).3) This relation is called the First Law of Thermodynamics. P ). For this second process dE = T dS so that the combined eﬀect of the work (adiabatic process) and the added heat give a change in energy. Y ) are volume and pressure (V. dQ is not an exact diﬀerential. To see this we write (4. The term T dS (energy due to a change of entropy) is called heat and is sometimes denoted dQ. but it is really an expression of energy conservation. and therefore. technically. S Let us deﬁne variables Yn that are conjugate to to the Xn by the formula Yn = − and it follows that dW = − Yn dXn n ∂E ∂Xn (4. a second process in which the control parameters are constant but the entropy changes (just the reverse of what happened in the ﬁrst process). dE = T dS − P dV More generally dE = T dS − Yn dXn (4. .

By solving for T in terms of S and X. At this point we need a calculus theorem that we leave for you to prove. Let there also be a second function E(T. X) of two variables T and X. we can think of E as a function of S and X. This of course is false. Such processes usually mean that the system is in contact with a heat bath which is so big that its temperature does not change during the process. 4. X).2 Processes at Fixed Temperature We have deﬁned the conjugate variables Y such as pressure in terms of adiabatic or constant entropy processes. Consider ∂ 2 Q/∂X∂E. What must be zero for such a cycle is the change in the energy. We can also deﬁne them in terms of constant temperature processes (isothermal).2 Processes at Fixed Temperature 19 If Q were a well deﬁned function of E and X then ∂Q =1 ∂E and ∂Q =Y ∂X would both hold true. . one ﬁnds (∂Y /∂E)|X = 0. Suppose we have a function S(T. Since the order of diﬀerentiation should not matter. For example one would ﬁnd that the pressure would not depend on the energy of a gas with ﬁxed volume. The meaning of all of this is that it is possible to bring a system through a series of changes that bring it back to its original equilibrium state in such a way that the net input of heat is not zero. so dQ cannot be an exact diﬀerential. In other words (dQ + dW ) must be zero.4.

..1 Prove the following identity: ∂E ∂X = S ∂E ∂X T − ∂S ∂X T ∂E ∂S X This identity is general but in the case where S.. . .. . . . .. .. . . E have their usual thermodynamic meaning. . ... . . . . ... .... . Homework . .. .... . ...... we can use ∂E =T ∂S X to get an expression for Y ≡ −(∂E/∂X)S ∂E ∂X = S ∂E ∂X T −T ∂S ∂X = T ∂(E − T S) ∂X T Finally. ... . . .... ... .. . ... . ...... . . . . or derivatives of A with respect to X at ﬁxed temperature... . . . .. ... .. . ....... .. . .. . ..... . ... .20 4 CONTROL PARAMETERS .. . . ... .. ... T.. ... .. . . . .. .. . .. using E − T S = A this can be written (∂A/∂X)|T . .. .... . . . ...4) T Thus we can either deﬁne conjugate variables like pressure in terms of derivatives of E with respect to X at ﬁxed entropy... Problem 4... ... . Thus. Y =− ∂A ∂X (4....

Part II Some Simple Applications .

.

The energy of a state at point x. Thus in the present case we have a value for each of 3N coordinates whose values are restricted to the interior of the container. Begin with a single particle. But quantum mechanics suggests that we take it to be (¯ )3N .e. a value for each generalized coordinate and each generalized momentum. each with mass m. We then give molecules rotational inertia and ﬁnally study the van der Waals gas. plus 3N momenta whose values are unrestricted. then add two-particle interactions. The state of a classical system is described by a point in its phase space.p) e . 5. etc. The phase-space “volume” of each cell (units of momentum times length to the power 3N ) would tend to zero at the end of a classical calculation.1 Ideal Gas An ideal gas is deﬁned as a system of non-interacting particles. p is h E(x. Forces between particles due to collisions.∗ i. Consider a container of volume V containing N identical non-interacting point particles. are ignored.2) See Appendix B for an introduction to the concept of phase space. p) = and the partition function is Z(β) = ∗ 1 3N 2 pn 2m n=1 (5. Thus we replace the sum over states by h → dx3N dp3N (¯ )3N h i (where ¯ ≡ h/2π and h is Planck’s constant). We begin with the well-known ideal gas.23 5 Dilute Gas In this section we study the properties of a dilute gas (as contrasted with very dense forms of matter such as Bose-Einstein condensate. . To sum over states we imagine replacing the phase space by a discrete collection of small cells. electromagnetic ﬁelds. In order to apply the reasoning of Part I we must know how to deﬁne the state of the system and how to sum over states..1) 1 N! dx3N dp3N −βE(x. (¯ )3N h (5. which we study in Section 15.

particle-one at x′ . The x integrals in (5. In fact it is the 3N power of the one dimensional integral dp e−β 2m = The ﬁnal result for the partition function is Z(β) = e ρ N p2 2mπ . p′ . β 2mπ β 3N 2 (5.24 5 DILUTE GAS The factor 1/N ! is put in to avoid over-counting conﬁgurations of identical particles. The integral gives V 3N and when combined with the 1/N ! we get eV N N Since N/V is the particle density which we call ρ. For example if there are two particles.3) If we want to explicitly exhibit the dependence on volume we replace ρ by N/V . Thus we divide by the number of equivalent conﬁgurations which in this case is 2 and in the general case is N !.2) is a Gaussian integral over 3N variables. p′ and the conﬁguration particle-two at point x. p. We will use Stirling’s approximation. particle-two at x′ .4) . The momentum integral in (5. N ! ≈ N N e−N .2) are trivial since the integrand does not depend on x. in evaluating the partition function. p. Z(β) = eV N N 2mπ β 3N 2 (5. these factors combine to give e ρ N Notice how the N and V dependence nicely combine to give an expression which only depends on the density which we will keep ﬁxed as the number of particles tends to inﬁnity. there is no diﬀerence between the conﬁguration particle-one at point x.

. .. . .. . .... volume on the other) what are the curves of constant temperature (isotherms) and constant entropy (adiabats)? What is the relation between pressure and temperature at ﬁxed entropy? . .. ... . xN ) = U (xm − xn ) (5.. · · · ... We consider only potential energy due to interactions between pairs of particles.. .. .... .. .. . . .. . .... .. . . . Z= which becomes VN N! 2mπ β 3N 2 1 N! d3N x d3N p e−β p2 2m 1−β m>n U (xm − xn ) V N −2 −β N! 2mπ β 3N 2 N (N − 1) 2 dxdx′ U (x − x′ ) ... .. . . . .. · · · potentials are possible but we will ignore them)... . . . .. compute A. 4-body.. . . . ... Derive the energy per particle and the ideal gas law P = ρT . .. Problem 5.. 5...2 Almost Ideal Gas 25 .2 Almost Ideal Gas Now let us introduce interactions between particles. S. ....1 Given the partition function in (5. ... ..2 On a PV diagram (pressure on one axis. .. p) = n 1 N! dx3N dp3N −βE(x. .5. (More complicated 3-body...4). . ... .... . What is the average speed (magnitude of velocity) of a particle? Problem 5...... . .. . . . . .. E.. U (x1 . .. . .... .. Homework . ... .p) e (¯ )3N h p2 n U (xm − xn ) + 2m m>n To linear order in U .... x2 . .. . . .... ..... . .. .. . . ... . .5) m>n We will treat U as a small quantity and compute Z to a ﬁrst-order approximation: Z(β) = where E(x. .. . and P as functions of the temperature...

... ... For example the box could be in a gravitational ﬁeld so that every particle has a potential energy U = mgy where y is the vertical direction. . ... ... and P as functions of the temperature. .. .. Homework .. .. ... What is the conjugate to the potential at point x? If we adiabatically increase the potential a tiny bit in some region.... .7). every particle in that region will have its energy increased.. . Problem 5.... ... . Derive the energy per particle.. Find the correction to the ideal gas law P = ρT + · · ·.. . ..... . What is the average energy of a particle? On a PV diagram (pressure on one axis.7) .. .... ... . S.... .. . ..3 Given the partition function in (5....6) To calculate log Z we use the ﬁrst-order Taylor series for log (1 − ǫ)... . . . . . . . .. .4 Calculate the log Z to the next order (second order) in U and ﬁnd the next correction to the ideal gas law... So the change in energy is δE = d3 x ρ(x)δU (x).... . . . ... ... The value of the potential at every point can be thought of as a control parameter. . . volume on the other) what are the curves of constant temperature (isotherms) and constant entropy (adiabats)? Problem 5. . .. . .. and using Stirling’s approximation. log Z = −N log ρ − Nβ 3N log β − ρU0 + const 2 2 (5. .. . . . approximating N (N − 1) by N 2 .. . . one ﬁnds Z= e ρ N 2mπ β 3N 2 1− βN ρU0 2 (5.... . ...... .. . . .. . . .. ... compute A. . . E.3 Ideal Gas in a Potential Field Consider a box of gas subject to an external potential... . . . Ordinary derivatives are replaced by functional derivatives.. . .26 5 DILUTE GAS Using dx dx′ U (x − x′ ) = V dx U (x) ≡ V U0 ... In the general case we have a continuous inﬁnity of control parameters. . . More generally U = U (x).. 5. .... .. . .. . . namely log (1 − ǫ) ≈ −ǫ.. ..

4 Diatomic Molecular Gas 27 The variational derivative of E with respect to U (x) is just ρ(x). It can be replaced by a normalization constant K determined by the fact that the integral of the density must equal the total number of particles in the box. Now consider the partition function.9) The factor N/ d3 x e−βU (x) is independent of position. (5.10) Thus. the particles are most dense where the potential is lowest. To illustrate this idea we will consider a gas composed of diatomic molecules modeled by . Thus the conjugate to U (x) is −ρ(x). ρ(y) = Ke−βU (y) . as we might expect. ρ(y) = = TN βe−βU (y) e−βU (x) N e−βU (y) 3 x e−βU (x) d d3 x (5. It is the same as for the ideal gas except that the integral d3 x which gave a factor of volume for each particle in that case now becomes d3 x e−βU (x) and we get Z(β) = d3 x e−βU (x) N 2mπ β 3N/2 (5.5. If the molecules of the gas have structure. 5. For a gravitational ﬁeld the density as a function of height is proportional to e−βmgy .4 Diatomic Molecular Gas So far we have considered a gas composed of point particles. they are capable of internal motion such as rotation and vibration.8) The entire dependence of the free energy on U will be in the term −T N log d3 x e−βU (x) If we want the density at point y we functionally diﬀerentiate this expression with respect to U (y).

Let’s suppose the rod is oriented in space with angular coordinates u. The mass moment of inertia about its centroid is I =2× m × 2 ℓ 2 2 1 = mℓ2 4 We can use m and I to represent any rigid molecule having extension in space and thus rotational energy. The rotational kinetic energy is Erot = The canonical momenta are I 2 (u + v 2 sin2 u) ˙ ˙ 2 pu = (5. I v ˙ pv = I sin2 u p2 p2 v u + 2I 2I sin2 u and the energy (Hamiltonian) is Erot = 2 (5. and v as the azimuthal angle within a hemisphere.11) u ˙ . First do the momentum integrals and obtain 2Iπ β du dv sin u = 4πI β (5. The internal factor for each molecule is du dv dpu dpv e−βErot . The translational energy of the molecule is p2 /2m where p represents its linear momentum. The partition function will factorize into a translational part and an internal part. This means that the partition function in (5.3) is changed by replacing β −3N/2 by β −5N/2 . v where 0 < u < π and 0 < v < 2π. The important thing is that there is a factor 1/β for each molecule.28 5 DILUTE GAS two particles with mass m/2 connected by a rigid massless rod of ﬁxed length ℓ. The eﬀect is to change the energy per molecule . You can think of u as the polar angle.13) The constant 4Iπ is of no importance and we can drop it. In addition there is rotational energy which we can derive.12) The integral d3 x d3 p e−βp /2m (there is one such factor for each molecule) is the same as for the point particle.

. ..... Problem 5... . ... . .. ... .. ... . In general we get an additional energy of 2 T for each internal degree 2 of freedom of the molecule (such as relative motion of two masses on an elastic link). . 5. . . .. Generalize the result to the case of n internal degrees of freedom.5. ... . . ..14) . . ..... ... In that case the volume over which the center of mass can move is smaller and of order V /N −v.... . .. This leads to a famous paradox: as the distance between the parts of the molecule tends to zero. .. . . .... .. . Homework . The partition function contains the factor N V −v N If we combine the idea of an impenetrable spherical particle with the corrections due to a long range weak potential. ... ... . ... Using Stirling we can write this as V N N One way to interpret this is to say that each particle moves in a box of size V /N . . .... .. .. . .. . .. Without going into details. ... equation (5.5 A Note about Volume and the van der Waals Gas Consider the factor V N /N ! in the ideal gas partition function..... .. . . ... .15) V −v N N 2mπ β 3N 2 1− βN ρU0 2 (5.5 Calculate the thermodynamic properties of the diatomic gas.... . . . ..... . . . ..... .. .. . .. But the internal energy we just derived does not depend on the particle’s size. .6) becomes Z= and the free energy is A = log Z = −N log 3N Nβ N V −v − log β − U0 + const N 2 2 V (5. .. .. one would expect the molecule’s behavior to approach that of the point particle. ... . In particular ﬁnd the adiabats on the PV diagram..... . . ..... .. . .5 A Note about Volume and the van der Waals Gas 29 1 3 from 2 T to 5 T . we simply note that the resolution of the paradox involves quantum mechanics. .. . . ... Now let us suppose each particle is not really a point but rather an impenetrable sphere of volume v (a “van der Waals gas”).

(P − u0 2 ρ )(V − N v) = N T 2 (5.30 5 DILUTE GAS Calculating the pressure and rearranging the equation gives the van der Waals equation.16) .

and the free energy A is A(β. The energy of such a conﬁguration is E = (n − m)Hµ and the number of such conﬁgurations is N! N! = . T (6. n!m! n!(N − n)! The partition function is Z(β. n!(N − n)! We have written the partition function in terms of β and H in order to indicate its dependence on temperature and the magnetic ﬁeld. each of which can be up or down along the z axis (we will use the notation σ to indicate the z component of spin).1) (6.2) (6.2). Note that in this case the magnetic ﬁeld is a control parameter. Let the number of up spins be n and the number of down spins be m where n + m = N .3) . The conjugate variable to the control parameter H is the magnetization M : M =− Using (6. There is a magnetic ﬁeld H oriented along z and the energy of each spin is σµH. σ2 · · · σN . The sum is a binomial expansion and the result is Z(β. ∂A ∂H . H) = n N! e−β(n−m)Hµ .3) gives M = −N µ tanh βHµ. H) = (eβHµ + e−βHµ )N = (2 cosh βHµ)N . H) = N T log (2 cosh βHµ).31 6 Simple Magnet and Magnetization Consider a collection of N spins σ1 . we ﬁnd that (6.

. . .. ..... This is the ferromagnetic case. . . . .2 Show that M is the average total magnetic moment of the assembly of spins. .. each interacting with its two nearest neighbors except for the end-spins which have only a single neighbor... ... ... ... . . µ without an index is a magnetic moment. There should be no confusion.. . . . Do you see anything strange? Problem 6... . .1 Ising Model The so-called Ising magnet is equivalent to the simple magnet we just studied. ... .. Note that we have chosen the sign so that for positive j the energy is minimum when the spins are aligned. .. Deﬁne “dual” spin variables µi .. . .... . .. . .... .. .. . . . . ... .. Problem 6. . .4) The constant j is the strength of interaction between neighboring spins... . To calculate this term let’s change variables.. .. . The system is deﬁned by its energy N E = −j σi σi+1 ..... . .... There are N µi -variables labeled µ1 . They are all independent and determine the original spins by the transformation j−1 σj = i=1 µi . .. . At this point we will not introduce an external ﬁeld although we could. .. .. . . ..... ... .1 Compute the energy and entropy of the simple magnet. . .. Think of a one dimensional line of N + 1 spins... .. µ2 .. (These µ′ s are not magnetic moments.. . . µ3 . The ﬁrst term contains all conﬁgurations in which spin-one is up.... . Find the functional form of the energy in terms of entropy. Homework .... ... 6. . .. .32 6 SIMPLE MAGNET AND MAGNETIZATION .. . The trick is to write the partition function as the sum of two terms.. . · · · µN . .. . . .. .. .. . .. .. i=1 (6..) µi = σi σi+1 . The anti-ferromagnetic case is deﬁned by choosing j to be negative. but that makes the problem harder...... With an index it is a dual spin variable... .

. in the sector with the ﬁrst spin up... .. . Note what happens in the limits of small and large temperature. .. . .. . ... .. . . . .. . .... . Homework .. . ..... In other words replace (6.. . .... .. ... .3 Add a term to (6.. . .. . Therefore the contribution of the second sector is the same as the ﬁrst.. .... ... . But the entire problem is symmetric with respect to changing the sign of all the spins simultaneously. ..4) by N N +1 E = −j σi σi+1 + i=1 i=1 µHσi ....... . .. ..1 Ising Model 33 For example. . . .. .... . .. .. . . . .. .. .... .... . Can you explain the behavior? .... . . .4) corresponding to an external magnetic ﬁeld. ...6.. .. and we ﬁnd Z = 2(2 cosh βj)N (6.. . The partition function in this sector is given by Z= exp βj µi . ... .. .. ....6) Compute the magnetization to linear order in the magnetic ﬁeld. .. .. .. .. . .. Problem 6. . σ1 = 1 σ2 = µ 1 σ3 = µ 1 µ 2 and so on. .5) .. ... . . .... Note also that the µi take on the values ±1 just as do the original spins. (6. ... .. This is exactly the same as the earlier model with the following substitutions: Hµ → −j σi → µ i Thus the partition function in this sector is Z = (2 cosh βj)N Finally we have to add the sector with σ1 = −1. . .

For simplicity we will work with a single control variable. The equation dE = T dS − P dV suggests that we think of E as a function of S and V . We can then write ∂E ∂S ∂E ∂V = T = −P (7. On the right we have a heat-conducting box of ﬁxed volume and at temperature T .∗ the total energy of a system is a well deﬁned function of the thermodynamic equilibrium state. . . We can derive additional relation by focusing on the Helmholz free energy A = E − T S: dA = dE − T dS − SdT = (T dS − P dV ) − T dS − SdT or dA = −SdT − P dV. It is very general and applies to solids. We allow an amount of heat dQ into the box and the pressure changes by dP . liquids. the volume.2) ∂V S ∂S V Figure 4 illustrates a thought experiment to conﬁrm the ﬁrst Maxwell Relation. Equation (7. gases. and freeze all the others. ∗ (7.2) is a remarkable bit of magic. This means it is a function of all the control variables and one other variable such as the temperature or the entropy.3) We saw on page 19 that heat Q does not have an exact diﬀerential. With very little input we derived the general fact that for all systems the change in temperature under an adiabatic volume change is the negative of the change in pressure when we add a bit of entropy at ﬁxed volume. On the left we have an insulated cylinder with a movable piston that can be used to change the volume.1) Now from the fact that ∂ 2 E/∂S∂V = ∂ 2 E/∂V ∂S we derive the ﬁrst Maxwell Relation. black body radiation and more.34 7 THE MAXWELL RELATIONS 7 The Maxwell Relations Unlike heat. We slowly move the piston and change the volume by dV. ∂T ∂P =− (7.

35 Figure 4: Experiment Illustrating First Maxwell Relation .

. . .. Problem 7. . .. ... Homework .... The change in enthalpy if due to some added heat is just the heat added (T dS). Two more relations can be obtained from functions H(S. .. .. .. . . .... .. P ) (enthalpy) and G(P... . (7. . ... T ) (Gibbs free energy). If we call the heat added ∆Q.. . ..6) ∂P S ∂S P Finally the Gibbs free energy is deﬁned by G = H − TS dG = V dP − SdT and the fourth Maxwell relation is ∂S ∂V =− . ..... .. .. .... Thus ∂A = −S ∂T ∂A = −P ∂V and by using the fact that derivatives commute... ...... .. . ... . .. For example. ..36 7 THE MAXWELL RELATIONS We think of A as a function of T and V .... . . . . (7.. . . . consider a gas in a cylinder with a piston such that the pressure is just the weight of the piston... .. .1 Design experiments for conﬁrming the second. . then ∆E + P ∆V = ∆Q.. . H = E + PV dH = T dS + V dP (7..5) Enthalpy is useful when considering a process at ﬁxed pressure. . ...8) ∂T P ∂P T When using the Maxwell relations we can replace a diﬀerential change in entropy by dQ/T where dQ is the added heat energy. . .. .. .. .. .. . third. we derive the second Maxwell relation.4) ∂V T ∂T S The ﬁrst Maxwell Relation follows from considering the energy as a function of the pair (S... .. ∂S ∂P = (7.... .. . . . The corresponding Maxwell relation.. . . .. . .7) . and fourth Maxwell Relations similar to the one shown in Figure 4... . .. is ∂T ∂V = . . .. V ). ..... . .. V )... . and the second Maxwell Relation from A(T. . . the third... .. . . (7. ... ....

An example is a frictionless billiard system (ignoring the pockets). The Liouville theorem† says that the volume and topology of the ∗ † See Appendix C. See. will cause exponential growth. The answer to this apparent paradox comes from chaos theory and coarse graining. But Loschmidt said that mechanics is reversible. To see what chaos implies. Let the distance in phase space between two initial points be ǫ. Chaos is the fact that for most systems the orbits in phase space are unstable. . it is not possible to have a function on (p. x) that always increases. But the tiniest error. The Second Law has an interesting history. imagine following a patch of phase space – say at a ﬁxed energy – as it evolves.∗ Boltzmann originally tried to deﬁne a quantity called H that always increased (dH/dt ≥ 0) along a trajectory in phase space. Since the time-reversed trajectory is also a solution to Newton’s equations. either in the way the numbered balls are stacked. it is positive. and for almost all systems with more than one degree of freedom. so that after just a couple of collisions the outcome will be extremely diﬀerent. Then the system is said to be chaotic. for example. implying no growth in ǫ. Zero is a very exceptional value. the entry of that name at www. If the initial conditions are perfectly reproducible then every time you do the experiment the result will be the same no matter how long you follow the orbits (the paths of the balls on the tabletop). In this case the entropy can be taken to be the logarithm of the phase space volume of the blob.37 8 The Second Law Let us simplify the story a bit by considering an initial probability distribution on phase space which is constant over some small “blob” and zero outside the blob. Start with the 15 numbered balls all arranged in a tight stack and carefully aim the Q-ball. or in the initial trajectory of the Q-ball. It is easy to prove that the Lyapunov exponent must be greater than or equal to zero. This is a generic phenomenon in mechanics whenever there is more than one coordinate. It is characterized by a quantity called the Lyapunov exponent. In some sense it increases with time.wikipedia.org. so if H increases along a trajectory it will decrease along the time-reversed trajectory. but that seems to violate basic principles of mechanics. After a small time the distance between them will be eλt where λ is the (largest) Lyapunov exponent.

even thought the ﬁne-grained blob has a ﬁxed volume. But as the tendrils spread the coarse-grained version ﬁlls a larger and larger volume. Most don’t. the phase space volume will be even bigger. The correct statement is not that . If the system is chaotic. In fact if we run the coarse-grained blob backward in time (or forward) and then coarse grain the result. Given enough time the air in the room will by accident congregate in the corner. The shape branches out and forms a crazy fractal made of thinner and thinner tendrils spreading our over the phase space as in Figure 5. then at ﬁrst. it will traverse the entire room. that error will exponentially grow with the Lyapunov exponent and instead of lurking in one corner. Here is an example: take all the air in the room and start it out in a very small volume in the corner of the room. Liouville’s theorem does not tell us that the coarse-grained volume cannot increase. but in general the shape does change. Suppose we take into account our inability to resolve points of phase space with arbitrary precision. Assuming the original blob is bigger than the cutoﬀ. In a short amount of time the molecules will spread out over the volume and ﬁll the room. Some time-reversed trajectories (long lines) lead back to the original blob. Now come back to Loschmidt.38 8 THE SECOND LAW region does not change. In other words. replacing each phase space point by a small but ﬁnite sphere. We see that in the bottom of Figure 5. Freak accidents do happen. because this would violate the second law. We can coarse-grain the phase space blob by drawing such a sphere over each point in the original blob. After a long time imagine time reversing for every particle (this means reversing its momentum). coarse graining has little eﬀect. if you wait long enough. Eventually the tendrils will get arbitrarily close to any point on the energy surface of the initial points. we introduce a cutoﬀ. But if we time-reverse every molecule of the ﬁnal state. then points that were initially close will soon diverge. Of course the course-grained volume cannot be smaller than the ﬁne-grained. the entropy of a conﬁned gas is less than that of a gas that ﬁlls the room. Eventually the coarse-grained blob will ﬁll the entire energy surface. quickly depart and go somewhere else. If we do this for every point in the ﬁnegrained blob. it will trace back to the original small round blob. Can the opposite happen? No. Even trajectories that start very close to one which does go back to the blob. the air will rush back to the corner. The problem is that if we make a tiny error in the motion of just a single molecule. so the coarse-grained volume cannot decrease.

The phase space volume and topology are exactly conserved but when coarse grained it grows.39 Figure 5: The Second Law. .

.... . . .. .. .. . . but only that they rarely happen. ... Consider the possibility that all the air accidentally accumulates in a one cubic meter volume in the corner. .. . .. . . .. . .. ...1 Suppose the room has volume of 100 cubic meters. ..40 8 THE SECOND LAW unlikely things never happen... The ﬁnegrained entropy never changes because of Liouville’s theorem.. ... ... . ... . . ... .... . . ..... .. . . Homework . .. ......... ... . ... .. . . . The coarse-grained entropy does increase. ..... . .. .... . .. .. . . ... .... So don’t hold your breath! Thus there are two kinds of entropies. .. . . . . .. ... .. . .. .. ...... The time that you would have to wait for this unusual air-event to take place is exponential in the number of molecules. . . ..... . Problem 8. ﬁne-grained and coarse-grained. .. ... What is the probability for this to happen? How long would you expect to have to wait to see it happen? ... . . ..

It does not. 2 . The mass m would be the reduced mass of the system. in the molecular interpretation it is the internal energy of the molecule and would be added to the usual 3 T . The energy per molecule of an ideal gas depends on the molecular structure: for point molecules it is 3 T and 2 for the diatomic molecule with ﬁxed separation it is 5 T . But as the diatomic molecule 2 gets smaller and smaller. but rather the relative variables.41 9 Quantum Considerations There are a number of paradoxes in statistical mechanics that were only resolved by the introduction of quantum mechanics. We can think of the HO as a molecule if we ignore its translational degrees of freedom. We saw one on p.1) If we think of the HO as a molecule then p and x are not the position and momentum of the center of the center of gravity. (9. The Hamiltonian is E= p2 kx2 + 2m 2 (9. 29. Another example is the apparently inﬁnite energy of black body radiation. Second. Let’s begin with a simple example. Suppose there is a single harmonic oscillator (HO) in equilibrium with a heat bath. kβ ω (9. it should tend to the point molecule. it is independent of m and especially of k. and using ω = Z= 2mπ β 2π −1 2π = β . The partition function is Z= or Z= dp e−β(p 2 /2m) dx dp e−β(p 2 /2m+kx2 /2) dx e−β(kx k/m.2) An easy calculation gives the average energy of the oscillator to be E = T.3) Two things to notice about this energy: First. at least according to classical (non-quantum) statistical mechanics. give 2 /2) The integrals are Gaussian.

In other words. the average energy is E= h eβ¯ ω h 1 eβ¯ ω = β¯ ω h 1 − e−β¯ ω e h −1 (9. But for low temperature the energy is exponentially small: h E ≈ ω¯ e−ω¯ /T . Let us redo the calculation taking account of quantum mechanics. h (9. Only when the temperature becomes large enough that the classical internal energy T becomes comparable to a single quantum does the molecule reveal its structure.5) ω¯ h −1 (9. One might have expected that when k is very large the molecule should be indistinguishable from a point molecule which has no internal energy.7) Notice that in the limit k → ∞ (or equivalently ω → ∞) the energy of the oscillator goes to zero. the HO molecule does behave like the classical point particle which has no internal energy.6) Note that at high temperature (small β) the answer agrees with the classical answer E = T .4) Z= n=0 h e−β¯ ωn . ∞ (9. 1.42 9 QUANTUM CONSIDERATIONS What is strange is that even in the limit of a rigid molecule (k → ∞) the internal energy does not go away. . the oscillator frequency. The sum is a geometric series: Z= Finally. All we really need from QM is the fact that the energy spectrum is En = hωn ¯ where n = 0. · · · and ω = The partition function is k/m. 2.

.. . .. .. .... .. . The partition function for a single molecule can be written as an inﬁnite sum over n. Show that at low temperature the energy per particle tends to the point molecule limit. .......... Homework ... .. ..1 Compute the free energy and entropy of the classical and quantum oscillators.. . . . . . . ..2 In section 5..... ... . .. .4 the energy for the diatomic molecule could have been written L2 E= 2I In quantum mechanics L2 takes on the values n(n + 1)¯ and each energy level has h degeneracy 2n + 1.... .. ... .. .. . . . . .. ..... . . . . . . .. .. ... .. . . ........ Estimate the sum in the limit of large and small temperature. Show that at high temperature the classical diatomic result is reproduced for the energy.. . .. . .. .. .. .. . . .. . . . .... .... .. .. .. . . . .43 . . . . Problem 9.. ...... ... . . How do they compare for high and low temperature? Where is the transition from quantum to classical? Problem 9... ... ...

ny . When the volume of the box is large the neighboring values of the ωn are very close and the sum can be approximated as an integral. As we have seen.2) ωn = L where n ≡ n2 + n2 + n2 x y z The partition function is an inﬁnite product of harmonic oscillator partition functions for each mode of oscillation. ny . Let us deﬁne the wave vector k by kx = nx π/L and similarly for y and z. nz ) sin ny π nz π nx π x sin y sin z L L L (10.ny . Since log Z is a sum over the modes. Classically Z= nx . nz ) which behave like harmonic oscillators of frequency nπc (10. the energy stored in a quantum oscillator is much smaller than the classical value when the frequency of the oscillator is large.nz X(nx . As we have seen. independent of its frequency. y. The radiation can be decomposed into standing waves with the form F (x. The diﬀerence between neighboring wave vectors .ny . the energy is just the sum of the energy of the oscillators.nz h eβ¯ ωn h eβ¯ ωn − 1 Let us consider the energy of the system. Using (9. a classical oscillator has an energy equal to T = β −1 .nz e which converges. z) = nx .44 10 THERMAL RADIATION 10 Thermal Radiation Consider a box (cube) with sides of length L.” The resolution of the catastrophe is provided by quantum mechanics.6) we ﬁnd the total energy is the sum h ωn ¯ (10. Electromagnetic radiation can exist in the box. Most of the energy is in very short wavelengths (large n) and the apparently inﬁnite result is called the “Ultraviolet Catastrophe. The walls of the box are reﬂecting. Therefore each of an inﬁnite number of modes has energy T and the total energy is inﬁnite.1) The dynamical degrees of freedom are the amplitudes X(nx .3) E= β¯ ωn − 1 h nx .ny .

.... ... ... The net eﬀect is that there are twice as many oscillators and the total energy is twice the above value. ... ..ny . . Homework ..1 Compute A... .. . . . .. .. .. .. ..... .. . ...... .... .. . and P for thermal radiation.. . . ... ... ..... . .3) becomes E= multiplying by T β = 1 gives E= L3 T π3 d3 k βck¯ h h eβ¯ ck − 1 |u| −1 L3 π3 d3 k h eβ¯ ck ck¯ h −1 Changing integration variables to u = β¯ ck we get h E= or E = 4π L3 T4 (πc¯ )3 h L3 T4 (πc¯ )3 h d3 u ∞ eu du 0 The integral can be evaluated and gives a factor of π 4 /15 leaving E= 4π 2 15 L c¯ h 3 u3 eu − 1 We have done the calculation as if the ﬁeld were a scalar with only one component.. . In fact the ﬁeld has two components corresponding to the two possible polarizations. . ... Problem 10. . . . . .... . . .. . . .. .. ... . .. .45 is ∆kx = and the sum is replaced by nx .... .. ... Work out all integrals.. . . .. . .. . . ... . ..nz π L d3 k → L3 π3 We also note that ωn = ck so that (10.... .. .. .. .. . .. .. .. .. ... .. .... .. . . . S...

In this formula. µ is called the chemical potential. Suppose the number of atoms in the nth copy is called Nn . Following the same logic as in Section 2. in considering the ensemble of N subsystems. The total number of helium atoms is conserved (assuming that the temperature is not high enough to break up the atoms). the index i refers to the energy levels of the N -particle system and in general these levels will depend on N . To implement the constraint of ﬁxed Ntotal we introduce a new Lagrange multiplier. Let’s return to the derivation of the Boltzmann distribution (Section 2). and the probability function depends on N as well as on the energy level. There are situations in which the number of particles in each subsystem may ﬂuctuate even though the total for all subsystems is ﬁxed. The Lagrange multiplier is βµ which includes a factor of β as a matter of custom. the classical liquid-gas phase transition. In that case only the total number of atoms is conserved. For example suppose the system is a container of helium. N . A conﬁguration of the (sub)system is now speciﬁed by an energy and a number of atoms. The constraint takes the form Nn = Ntotal . The partition function is deﬁned so that the total probability is one. For example. and Bose-Einstein condensation. Thus Z(β. N ) = Z −1 e−β(Ei +µN ) . we may allow helium atoms to be transferred from one subsystem to another.1) . µ) = N e−βµN ZN (11. P (i. For all three applications we will need the concept of chemical potential. Suppose there are some conserved quantities in addition to energy. it is easy to see that the probability function takes the so-called Grand Canonical form. Roughly speaking it is related to particle number in the same way that temperature is related to energy: as a Lagrange Multiplier.46 11 CHEMICAL POTENTIAL AND GRAND CANONICAL ENSEMBLE 11 Chemical Potential and Grand Canonical Ensemble In the remainder of this text we are going to study three topics: magnets. The chemical potential is a useful tool for studying systems in which the number of particles ﬂuctuates.

.. .5) .. .2) where.. ....... ....47 where ZN is the partition function with ﬁxed number of particles N .. . . . . ... . In the same way the chemical potential is determined to ﬁx the average number of atoms.. A = −T log Z.. ... . . The result is 3/2 −βµ Z = e(2mπ/β) V e (11. ... .2 Find an expression for the ﬂuctuation of the number of particles analogous to the relation in Section 3. For a ﬁxed number of particles the Canonical Partition function is 1 V N! 2mπ β 3/2 N The GCPF is gotten by multiplying this expression by e−βµN and summing over N .. .. . . . ..... .. . . .3) Problem 11.. .. . . .. .. . ... . . . .... . . .4) and the free energy is A = −T log Z = −T 2mπ −βµ Ve β (11.. . . . . Since N = V ρ we can also think of the chemical potential as the parameter that allows us to vary the particle density. ... . . . It is easy to see that N =− ∂A ∂µ (11. . Let’s calculate the Grand Canonical Partition function (GCPF) for an ideal gas... . . .. .. .... .. . Recall that in the case of energy. . ... as usual. . ....... . ... ...... .. the Lagrange multiplier β is determined so as to ﬁx the average energy.. ....1 Prove the following relations for processes at ﬁxed volume and other control parameters: E = T S − µN − T log Z dE = T dS − µdN (11... . Problem 11...... .. . . Homework ... .... . .. ... . ..

5).. .. . .. Problem 11. . . . . ... . . . .. .... ... .. .. . . .48 11 CHEMICAL POTENTIAL AND GRAND CANONICAL ENSEMBLE ...... . . .. . ... ..... . .. ... .... . . .... . . .. . ... ... . ... ..3 Using (11.... .. .. ... . . ... ..... . calculate the mean number of particles and the total energy as a function of the chemical potential.. Then show that E = 3N/T . .. Homework . .. . . . .. . .... .. .. . .... .. .... ... . ... . . . . . . .......... ..

j σi σj . To order h2 the logarithm of Z is given by log Z = log ZI + β 2 h2 2ZI eβj σ N i=1 1 − βh i 1 σi + β 2 h2 2 i. We will mostly be interested in the ferromagnetic case with j being positive. In this case the energy is lowest when the spins line up in the same direction.m σi σi+m .1) To compute M to order h we must compute log Z to order h2 .2) σi σi+1 i. 6. For i = j the term σi σj = 1. Z= σ eβj N i=1 σi σi+1 It is easy to see that the term proportional to h vanishes. (12.3) by M =− ∂A ∂h (12. N N +1 E = −j σi σi+1 + i=1 i=1 hσi .3) is given in terms of the derivative of log Z with respect to h. Thus we write Z= σ eβj( N i=1 σi σi+1 − N+1 i=1 hσi ) and expand to order h2 . The magnetization M (eq. The original Ising model is symmetric with respect to changing the signs of all spins and that insures that only even powers of h survive. Thus we may write σi σj σi σj = N + 2 i.j i>j or σi σj = N + 2 i.3) where ZI is the Ising partition function given by (6.49 12 Back to Magnets Let’s return to the Ising model in an external magnetic ﬁeld and compute the magnetization to lowest order in the ﬁeld.j i. To simplify the notation we will set µH = h.j σi σj (12. For simplicity we replace (6.5).

(12. the sum simply replaces (2 cosh βj) by (2 sinh βj). In terms of the µi it is given by N + 2 i. as you shut oﬀ the . In other words.m σi σi+m . The magnetization does not diverge at zero temperature but its derivative becomes inﬁnite.6) First of all.50 12 BACK TO MAGNETS where the m-sum starts at m = 1. This means that there is no spontaneous magnetization when h is shut oﬀ.m where we have replaced ZI by the value we computed in Section 6.4 gives β 2 h2 (2 cosh βj)N e−βj µ µi i. For those µ outside the range (i. The term in (11.5) (tanh βj)m = N β 2 h2 tanh βj 1 + 1 − tanh βj 2 or more simply N The magnetization (−∂h A) is β 2 h2 2βj e 2 M = βhe2βj . Thus the net result for large N is N β 2 h2 m=1 µi µi+1 · · · µi µi+m−1 . (12. At ﬁnite temperature the graph of M versus h looks like the solid line in Figure 6 while for zero temperature it looks like the dashed line. (12. notice that for all ﬁnite temperature the magnetization is linear and continuous in h for small h. Now let us change to the dual variables µi = σi σi+1 . i + m − 1) the sum over the values of µ just cancel the corresponding factors of (2 cosh βj) in the denominator. For the points inside the range.4) σi σi+m = µi µi+1 µi · · · µi+m−1 Thus the second term in 12. But at zero temperature the coeﬃcient e2βj diverges.5) that is quadratic in h is given by β 2 h2 2ZI e−βj µ µi Now consider σi σi+m .

external ﬁeld the magnetization persists. His mistake was to conclude that the corresponding model in higher dimensions also had no transition. The hard core will be implicit in the conﬁguration space integrations over position. The Ising model is characteristic of all one-dimensional statistical systems with shortrange interactions. but this phenomenon only occurs at zero temperature for the Ising model. U (x − y) will .12. We will see that in higher dimensions the spontaneous magnetization persists up to a critical temperature. The longer range potential. No two particles will ever be allowed to be closer than some core size. Let us begin with a system of particles with 2-body potentials that consist of a hard core repulsion and a slightly longer range (but still short range) attraction.1 Connection with Fluids One of the interesting connections that we will explore is the connection between magnetic systems and the liquid-gas transition in molecular systems like water. There is no phase transition at ﬁnite temperature. Ising correctly solved the one-dimensional system and correctly told his advisor Lenz that it had no transition at ﬁnite T .1 Connection with Fluids 51 T>0 T=0 Magnetization M Magnetic field h Figure 6: Magnetization vs external Magnetic ﬁeld. 12.

Call it σi where i indicates the site.) We will choose units in which the hard-core-size is unity. If all sites are vacant we call the state the vacuum.8) Now imagine a lattice approximation to space (in any number of dimensions. (12. at each point in the lattice there may either be a particle or not. The partition function is Z= conﬁgurations e−β d3 x d3 y ρ(x)ρ(y)U (x−y)−βµ dx ρ(x) (12. In other words the potential is non-zero only for particles separated by a single link.7) Since the potential energy is independent of momentum and the kinetic energy is independent of position. Let the density of particles be ρ. Let us introduce a variable at each point that indicates whether or not a particle is present. The momentum factor give the usual (2m/β) 2 . It plays no role in determining the pressure or in the gas-liquid phase transition. Replacing integrals by sums. If we work in units in which the lattice has unit spacing then the density is 1 on occupied sites and 0 on empty sites. The potential energy can be written in the compact form PE = d3 x d3 y ρ(x)ρ(y)U (x − y). We can ignore it for our present purposes. the energy is ∗ PE = −4J ρ(i)ρ(j) (12. If σ = −1 we say the site is vacant. the partition function factors into a position factor and a 3N momentum factor Z = Zx Zp . and no site may have more than one particle.9) . If σ = +1 we say a particle occupies the site. Thus.52 12 BACK TO MAGNETS be explicit. In this way we map the problem of a system of identical molecules onto a magnetic system of the type we have studied. Thus ρ= 1+σ 2 We also specify that the potential energy U is negative and that its range is one lattice spacing. From now on we will just consider Zx and drop the subscript x.

In fact it is the transition from a gaseous phase to a more dense liquid phase. σi σj − 4J + 2 i The partition function is the d-dimensional analog of the Ising model. It is simply the total number of sites times the average value of σ. When you cross it the magnetization (density) . This corresponds to an attractive short range potential. i From now on we drop the constant. Thus we may identify the magnetization per site (average of σ) as 2ρ − 1. Using ρ = 1+σ we easily ﬁnd 2 ∗ PE = −J σi σj − 4J ρ(i) + constant. Let’s add to the potential energy the chemical potential term −µ i ρ(i).1 Connection with Fluids 53 where the symbol ∗ means a sum over neighboring sites on the d-dimensional lattice.12. What does this jump mean from the point of view of the molecular system? It is a sudden jump in density as the chemical potential is varied. Let us consider the magnetization M . Suppose a magnetic transition takes place so that at h = 0 there is a jump in magnetization. The phase diagram is shown in Figure 7. We take J > 0 so that if two particles are far apart the interaction energy is zero but if they are on neighboring sites the energy is negative. Notice that there are two ways to pass from one phase to another: cross the discontinuity (the thick line) or go around it. In the Ising model this only happens at T = 0 but in higher dimensions it happens at all temperatures up to some critical temperature T ∗ that depends on dimension. The result is that the exponent in the expression 2 for Z is ∗ µ −J σ(i) + constant. which apart from a constant is − 1 µ i σ(i). Z= σ eβ (J ∗ −µ σi σj + 8J2 i σ(i)) Obviously this is the d-dimensional analog of the Ising model in the presence of an 1 external magnetic ﬁeld h = 2 (µ − 8J).

negative magnetization represents the gas phase. In the analogy with molecular ﬂuids. The thick line represents the discontinuity in the magnetization. In the 1-D Ising model the critical temperature T ∗ is zero. and positive magnetization represents the liquid phase. .54 12 BACK TO MAGNETS Figure 7: Phase diagram of a magnet.

Each broken bond increases the energy by 2J and introduces a factor exp (−2Jβ) in the Boltzmann distribution.12. Going around the critical point involves no jump and the behavior is continuous. We will take the number of “active” spins to be N − 1 and add two more frozen “dummy” spins at the ends. The LTE for Z has the form Z =1+ and log Z is log Z = N (N − 1) exp (−4Jβ) + · · · 2 N (N − 1) exp (−4Jβ) + · · · 2 The free energy seems to have a term proportional to N 2 . which is the same thing as saying µ = 1 for that bond. The LTE is an expansion in the number of broken bonds. Actually the LTE is not an expansion in T but in exp (−2Jβ).2 Low Temperature Expansion 55 jumps in what is called a ﬁrst-order transition. 12. The question is whether the boundary conditions are remembered far from the boundaries when N → ∞. The ﬁrst strange thing we see in the LTE is that free energy is not proportional to N . This is odd because the free energy per site (−T log Z/N ) is not ﬁnite but rather is proportional to N . Let us introduce new terminology: if two neighboring spins are the same we will say the bond between them is unbroken. These boundary dummy spins are assumed to be up σdummy = +1. If neighboring spins are opposite (µ = −1) the bond is broken.2 Low Temperature Expansion Something strange happens at zero temperature in the Ising model. This is characteristic of both magnetic systems and gas-liquid systems such as water. One interesting point is that in one dimension there is no gas-liquid phase transition. In fact the free energy density does not have a term proportional to N . To analyze it we can do a low temperature expansion (LTE). One way of saying this is that the ﬁrst non-vanishing derivative of the free energy density is inﬁnite . Let’s deﬁne the system by introducing extra dummy spins at the ends of the spin chain. For every value of β it is ﬁnite. The point is that the things we are calculating in each order are the derivative of the free energy with respect to the expansion parameter. Since the boundary spins are frozen the number of broken bonds must be even. Note that there are N bonds including the bonds involving the dummy spins.

In the third term they are both to the right. The smallest temperature destabilizes the spontaneous magnetization. We get = 2 N N 2(2 2 − 1) exp (−4Jβ). the eﬀects of the inﬁnitely distant boundary conditions. both broken bonds are to the left of σ. In the fourth. In the second. = 3 N N 2(2 2 − 1) exp (−4Jβ). one broken bond is on the left and one is on the right. For the second type of terms. Z To leading order Z −1 = 1 − N (N −1) exp (−4Jβ) + · · · To leading order (two broken 2 bonds) the sum in the numerator has four types of terms.e. So every spin has average value +1. You can go back and check that from the exact solution. the lesson is that the derivative of the magnetization with respect to the expansion parameter diverges (with a negative sign) as N → ∞. Next consider the average value of a spin far from the boundary. For simplicity let’s assume that N is even and that we focus on the spin at the center of the chain (just call it σ). The ﬁrst is a single term with no broken bonds. σ = +1. For the fourth type of terms σ = −1 and we get 4 =− N 2 2 exp (−4Jβ) Altogether we ﬁnd that the average of σ to order exp (−4Jβ) is given by 1− N2 + 2 + 2 3 + 4 exp (−4Jβ) = 1 − N2 exp (−4Jβ) 2 Again. This is indicative of a singular behavior at T = 0. i.56 12 BACK TO MAGNETS at exp (−2Jβ) = 0. . The ﬁrst term is 1.. The average of σ is 1 = eβJ σi σi+1 σ. There are N/2 bonds on either side of it. What is going on is that the magnetization vanishes for all T = 0. The third term – all broken bonds on the right – gives the same thing. At zero temperature the only state with non-vanishing probability has all the spins up.

In other words a large block of ﬂipped spins costs an amount of energy which does not grow with the size of the block. For deﬁniteness let’s take a square grid of lattice points with a total of n active spins. Thus you pay a large energy price for ﬂipping large blocks of spins. of course. at a cost in energy which is only 4βj. This. the x’s indicate down spins and the •’s indicate dummy up spins. is a fact about the surface-to-volume ratio in d dimensions. ° = up x = down • = dummy Figure 8: 2-dimensional lattice of spins.3 Ising-like Systems in Higher Dimensions In higher dimensions (number of dimensions = d) the situation is entirely diﬀerent. To ﬂip M spins you must break at least ∼ M 1/d bonds. In one dimension we can ﬂip an arbitrary number of adjacent spins.3 Ising-like Systems in Higher Dimensions 57 The reason for this behavior is simple. We will ﬁnd that this stabilizes the low temperature behavior and implies that the spontaneous magnetization of the ground state does not disappear at ﬁnite T.12. breaking only two bonds. . 12. The o’s indicate up-spins. In addition there are dummy spins at the boundary that are chosen to be up as in Figure 8.

We can also think of the energy as residing on the links of the lattice.e. If they are not nearest neighbors. the two spins can be nearest neighbors or not.. eight bonds are broken and the added energy is 16j. But the dummies break the degeneracy and the ground state is all up spins. The energy is given by E = −J ∗ σi σj where the sum is over nearest neighbors. If there were no dummy spins the ground state would be 2-fold degenerate. If they are nearest neighbors only six bonds are broken and the added energy is 12j. NN pairs are associated with the links of the lattice. We can ﬂip two spins. If you put a dot at the center of each plaquette they form the dual lattice. The symbol ∗ can be understood as the sum over links.58 Some deﬁnitions for the 2-dimensional case: • The spins are located on the sites of the lattice. 12 BACK TO MAGNETS • The energy is a sum over pairs of nearest neighbor (NN) spins. This breaks four bonds on the links connected to that spin. Let’s check what happens to the next order. At zero temperature all the spins are up and the partition function is Z = 1. The reason that only . There are two possibilities. i. For notational simplicity we will take the case where all the boundary spins are dummies. all up or all down. They cross the links of the original lattice at right angles. The links of the dual lattice are in one-to-one correspondence with the links of the original lattice. Now consider the low temperature expansion. • The squares of the lattice are called plaquettes. The ﬁrst correction comes from ﬂipping a single spin. The contribution from single ﬂipped spins is Z = 1 + N e−8βJ + · · · log Z = N e−8βJ + · · · Note that the free energy is extensive to this order.

∗ σi σj eβJ σ σ = Z −1 σ . The contributions are 2N e−16βJ and 4N e−16βJ respectively. Z = 1 + N e−8βJ + 2N e−12βJ log Z = N e−8βJ + 2e−12βJ Note that the free energy is still extensive. The non-trivial test happens when we study non-nearest-neighbor pairs. This suggests that there is nothing discontinuous about the behavior at T = 0. i. But we have to divide by two for the usual reason of not over-counting. These terms come from adjacent ﬂipped spins which break eight bonds. Z = 1 + N e−8βJ + N −12βJ N (N − 5) −16βJ e + e . Thus we ﬁnd 1 5 log Z/N = e−8βJ + e−12βJ − e−16βJ + · · · 2 2 There are a few more terms that contribute in order e−16βJ . To calculate log Z we use the formula x2 xn log (1 + x) = x − + ··· + + ··· 2 n and ﬁnd that the quadratic term in N cancels. even if log Z is linear in N . So the next order in the expansion comes from nearest neighbors (NN) and is of order exp (−12βJ). The spins can be in a line or in an L-shape.. Thus to order e−16βJ we ﬁnd. The number of such pairs is N (N − 5)/2. We can check this by examining the behavior of the average value of a spin far from the boundary. begin by picking a site and ﬂipping its spin. The number of NN pairs is 2N for large N . The number of sites which are nearest-neighbor is ﬁve: the site itself and the four NN. ∼ N.12. However. we expect Z to contain things of higher order. To see that. Unlike the 1-dimensional case there is nothing divergent in the power series coefﬁcients of the free energy density. This seems obvious since the number of NN pairs is ∼ N.3 Ising-like Systems in Higher Dimensions 59 six bonds are broken is that the bond on the link connecting the pair is unbroken if both spins are ﬂipped. 2 2 This looks dangerous because of the quadratic term in N .e.

. That is even true at zero temperature. Thus eβJ σ ∗ σi σj σ = 1 + (N − 1)e−8βJ − e−8βJ = 1 + N e−8βJ − 2e−8βJ Now we have to multiply by Z −1 = 1 + N e−8βJ ... We can diagnose the system without explicitly breaking the symmetry by studying the response to an external ﬁeld. the sign is changed.... . . ... the contribution is the same as to the partition function. ..1 Compute the average value of σ to order e−16βJ and show that the power series coeﬃcients are ﬁnite in the limit N → ∞. Expanding to order e−8βJ we ﬁnd σ = 1 − 2e−8βJ . You may ask where we used the dummy spins. . .. If we don’t break the symmetry then for every conﬁguration we must also allow the conﬁguration in which each spin is reversed. . . . .. This phenomenon of spontaneous magnetization is a special case of spontaneous symmetry breaking. . . . But if the ﬂipped spin is σ.. ... .... .. The only thing we used them for was to break the degeneracy of the ground state. . ... The symmetry is deﬁned by changing the sign of all spins. .. .... . . ... . But there is a consequence of the spontaneous symmetry breaking. . ..... . .. . .. . It is a kind of “persistence of memory” in which an external magnetic ﬁeld will be remembered even when it is switched oﬀ or made arbitrarily small. ... . . . .. If the ﬂipped spin is not σ... . Since the two conﬁgurations have the same energy. but at low temperature the limit will depend on whether you approach from positive of negative h. ... . . For h < 0 the spins will line up in the up position..60 12 BACK TO MAGNETS Again we sum over the number of broken bonds. ... .. .. ... . .. . ... ....... We could do something else that is often more relevant for real magnets. This will align the ground state.. .. .. . Problem 12. We can redo the calculations and then let h → 0. ..... . .... .. ... Introduce an extremely small external ﬁeld h so that each spin has an additional energy hσ. . . .. σi → −σi ... The lowest order comes from a single ﬂipped spin. .. it is evident that they have the same probability and the average of any spin must be zero. . . The energy cost of ﬂipping all the spins is 2N h. . The average of σ will have a limit... . . .. . . ...... It is this: if we ﬁnd a given spin to be up (down) then it is more likely that we will ﬁnd any other spin . . Homework . ..... ..

Thus to the order that we are working with. at least for a range of temperature deﬁned by the radius of convergence of the LTE. In this case the partition function is twice the previous value because we must sum over the two starting points. Let us go back to the 1-dimensional case. The calculation that followed showed that C(a. i. The contribution to the sum in (12.13) Begin with the ground state with all spins up. b) to order e−8βJ without introducing dummy spins at the boundary. b) = 1 but for T = 0 the correlation tends to zero exponentially with distance. b are two lattice sites. b) is non-zero as the separation goes to inﬁnity. b) → 0 as |a − b| → ∞. namely 1. the bias introduced by ﬁnding one spin up (down) fades with distance. But we might expect the bias to disappear as the distance between the two spins becomes arbitrarily large. then the correlation between diﬀerent spins does not fade with distance.3 Ising-like Systems in Higher Dimensions 61 to be up (down). keeping σa and σb up. Z = 2 + 2N e−8βJ + · · · Next we consider σa σb = Z −1 σ (12. The contribution is (N − 2)e−8βJ . We can say it in a precise way in terms of correlation functions.. Let’s calculate C(a. But in higher dimensions C does not vanish at inﬁnity. b) ≡ σa σb = Z −1 eβJ σ ∗ σi σj σa σb (12.e.12) eβJ ∗ σi σj σa σb (12.10) Ordinarily.12. if there is spontaneous magnetization or symmetry breaking. Next. which means that C(a. let us ﬂip any other spin. all down or all up. j). For zero temperature C(a.4 is nothing but β 2 h2 C(i. and then start ﬂipping spins. b) = tanh βJ m (12.11) where m is the separation between a and b. C(a. We deﬁne spontaneous symmetry breaking to mean that the limit of C(a. In itself that is not really surprising. Deﬁne the average value of the product of two spins σa σb where a. However.13) is of course the same as for the partition function. Notice that the second term in equation 12.

(12. (12.62 12 BACK TO MAGNETS Next we ﬂip σa . In fact the asymptotic limit is just the square of the average of a single spin.16) The nth term in the expansion is 1 Zn = (βJ)n n! ∗ n (σi σj ) σ . Z= σ 1 1 − (βJ)E + (βJ)2 E 2 + · · · 2 (12. In the next subsection we will see that it does not persist to arbitrarily high temperature. assuming the dummy spins are put back. The phenomenon of correlations remaining ﬁnite at inﬁnite separation is called long range order and it persists over a range of temperature from T = 0 to some critical temperature.15) Note that the result is independent of how far away the spins σb and σa are from one another as long as they are not at the same point. High temperature means small (βJ). we must add the contribution that arises from the opposite starting point. b) = Z −1 (1 + (N − 4)e−8βJ ) = 1 − 4e−8βJ (12. But this is just the term that arises from conﬁgurations that started with all spins up. b) as the separation goes to inﬁnity. In fact we can consider the result (12.17) . Altogether we ﬁnd C(a. Since we have dropped the bias coming from the dummy spins. Begin with Z. Combining it with the term coming from ﬂipping σb (but not σa ) we get 1 + (N − 4)e−8βJ .12) to be the limit of C(a. This time the contribution has the opposite sign to what it would be in the partition function. Therefore the high temperature expansion (HTE) is gotten by expanding exp (−βE) in powers of (βJ). 12. In higher orders one ﬁnds that the correlation function continues to be ﬁnite at inﬁnite separation.4 High Temperature Expansion Now we come to the high temperature limit.14) Note that the term with both σb and σa ﬂipped is higher order in the expansion parameter.

All states contribute equally and we get Z0 = 2N log Z = N log 2 + · · · (12. can be 1 or −1.σb σa σb = 2N (βJ) 1 σa σb 4 σa . σb . This involves a sum over nearest neighbors.18) Next. the n = 1 case. these will cancel.19) This is obviously zero.σb (12. n = 0. b) by drawing a line between sites as in Figure 9 It is given by 2N −2 (βJ) σa .4 High Temperature Expansion 63 Figure 9: The contributions to the HTE coming from a single link are represented by a heavy line on that link. .12. when summing. Begin with the ﬁrst term. The product σa σb must average to zero because for each value of σa . Let us pick a link on the lattice and consider the pair connected by the link. the other spin. We will represent the contribution of that link (connecting sites a.

To see why the number of links is twice the number of sites draw a link to the right of every site and a link just below every site. As you range over all the sites you cover all the links. (12. or two links can be covered twice. But suppose the two links are the same. (12. For example. Odd number of links always leave unmatched spins so that when you sum over them the result vanishes.20) The factor of 2N represents the number of links on the lattice. The next non-vanishing contribution comes from 4-link terms. Suppose the two links are not the same. Begin with the expression Z= σ eβJ ∗ σi σj (12. links (12.23) Now we can write the partition function in the form Z = (cosh βJ)2N σ (1 + σi σj tanh βJ).22) Now consider the expression eβJσi σj .21) which is the same as Z= σ links eβJσi σj . In fact we can rewrite the sum over conﬁgurations so that we never do cover a link more than once. They could involve completely disjoint pairs or they may share one spin. The 4-link terms are of several types. In either case there will be spins to sum over which are occurring to the ﬁrst power. Since σi σj can only take on the values ±1 all even powers of σi σj are equal to 1 and all odd powers are −1. Among the cases in which two links are covered twice there are diﬀerent possibilities depending on whether the links share a site. Such sums vanish since the unmatched spin can be ±1.64 12 BACK TO MAGNETS Let’s go to the next order involving two links. The doubled links contribute. Then the term involves (σa σb )2 which equals 1. By expanding the exponential and using this fact we ﬁnd eβJσi σj = cosh βJ + σi σj sinh βJ = cosh βJ(1 + σi σj tanh βJ). Their contribution is 2N (βJ)2 (12. It would all be a lot easier if there were a rule that forbade a link from being covered more than once. There are corrections but they are sub-leading in N . the four links can form a square. A single link can be covered four times.24) .

.2 Calculate the partition function. . . In fact at any given order we get one or more closed “curves” in which no link is multiply covered. . . . . . . . . . . . . The partition function is given by Z = (cosh βJ)2N graphs C(graph) tanhm βJ (12. . . . . . . . Some examples are shown in Figure 10. . . .4 High Temperature Expansion 65 Figure 10: Closed “curves” on the lattice When the product is expanded in powers of tanh βJ we never cover the same link twice. . . . Homework . . . . . . . . and average spin in both low and high T expansions for the case of the 3-dimensional Ising model. . . . . . . . . . . . . .25) where the sum is taken over graphs consisting of any number of closed curves. . .12. . . . . . Problem 12. . . . . . . For example at fourth order (tanh4 βJ) we get squares and nothing else. . . . . . C(graph) is a counting factor that enumerates the number of such graphs and m is the number of links that make up the curves. . . free energy. . In the LTE . . . . .

.... . . .. . . . . In this case it follows from the fact that tanh βJ < 1. ..26) when expanded to a given order in tanh βj involves an even number of spins... function) and σa σb (2-pt............. ...... . Extra Credit: Compute it to the next order? . . function C(a.. . ......... .. .. .24) (1 + σi σj tanh βJ)..... ..27) where n is the number of links making up the shortest path...and Two-Point Functions in the HTE Now let’s consider the averages σ (1-pt... .. Computing it involves graphs with any number of closed curves and one open curve connecting the points a... Thus we see an important generic feature of correlation functions at high temperature: they decay exponentially with distance...... Since the asymptotic value of C at large separation satisﬁes C(∞) = 0 for high temperature. . ..... .. b) = (tanh βJ)n (12.. ....... ...5 One.... . function vanishes to all orders in the HTE Boltzmann distribution in the form given in (12.... . .. Multiplying by any odd number of spins and summing over the conﬁgurations will always give zero. .......... . . . It is obvious that the 1-pt.... .. ... . .. . Homework . .. .. function) in the HTE. Problem 12. . The lowest order contribution comes from the shortest path connection them and has the value C(a. Show that the free energy is extensive and that the average magnetization is non-zero (zero) for low (high) enough temperature. .. ...... . links (12. The 2-pt... there must be a value of temperature which separates the two regions. ... . . ... .. .. . and C(∞) = 0 for low temperatures.. . . ...... ....... b.. ......... .. .. . . ..66 12 BACK TO MAGNETS go to order e−24βJ . .. This point is the critical point... . . .. . .. .... . b) is not zero.... ... .....3 Compute the 2-pt function to leading order in the HTE for 2 points on the same horizontal line and two points on the diagonal of a square.. . 12. ...... ........ ..... . ........

Start with all spins up. For each broken bond include the factor e−2βJ . Divide the ﬂipped spins into disconnected islands and draw a closed curve around each island. What is special about two dimensions is that the model can be exactly solved. We begin with the factor (cosh βJ )2N . The conﬁgurations of ﬂipped spins can be enumerated in a clever way.67 Figure 11: The dual lattice is composed of points at the centers of the Plaquettes. It is called duality and it is rather easily seen from the LTE and HTE. The HTE for Z is simple. To do that. We will focus on the partition function. 13 High-T/Low-T Duality The general properties of the Ising model in more than one dimension are independent of dimensionality. Then multiply this by a sum over all graphs involving any number of closed paths through the lattice. Now consider the LTE for Z. We are not going to solve it but we are going to work out one special property that occurs in d = 2. Assign each graph a weight given by (tanh βJ)n where n is the number of links in the graph. it is convenient to deﬁne a dual lattice composed of the centers of all the plaquettes (Figure 11). Now ﬂip a certain number. Sum over all such conﬁgurations. That breaks a certain number of bonds. That’s it. The closed curves surrounding the islands of ﬂipped spins are composed of dual- .

You may have realized by now that the LTE and HTE have exactly the same form except for two diﬀerences. it must lie at the symmetry point of the duality exp −2βj ↔ tanh βj. In fact they have the same form on the dual lattice as the curves on the original lattice that were used to implement the HTE. Suppose we compute the partition function or the free energy as a function of e2βJ from the LTE. the LT expansion parameter is exp −2βJ while the HTE parameter is tanh βJ. Then multiply by (cosh βJ )2N and you have the high temperature behavior. Just take Ze2βJ and replace the argument of the function by tanh βJ. In other words the critical . lattice links. What this means is quite remarkable. If we assume that the critical point is the unique singularity on the T axis. Each link on the dual curves is equivalent to a broken bond and the weight of such a conﬁguration is (exp −2βJ )n where n is again the number of links forming the curves. Then we automatically know the behavior at high T. First. in the HTE the partition function contains a factor of (cosh βJ)2N that has no analog in the LTE.68 13 HIGH-T/LOW-T DUALITY Figure 12: Surround each island of ﬂipped spins with a curve on the dual lattice. Second.

69 point must be at exp −(2βJ) = tanh (βJ) or equivalently the critical point is at sinh (2Jβcrit ) = 1. Equivalently the critical temperature is Tcrit ≈ 2.27J. (13.2) (13.1)

In higher dimensions the Ising model is dual to a theory living on the dual lattice but in general that dual theory is not the same as the original theory. For this reason the duality relates the critical temperatures of diﬀerent theories rather than giving an equation for the critical temperature of the original theory. There are many magical properties of the two-dimensional Ising model. We will show you one more. In summing over lattice curves (connected and disconnected) we must respect the rule that no link is covered more than once. Call this the “exclusion” rule. Suppose there were no such rule and that we could freely draw curves. Call the sum of all connected curves X. The sum of terms involving two disconnected curves would be X 2 /2; three disconnected components, X 3 /3. The sum would exponentiate to give exp X, and the logarithm of Z would just be X. Since any connected component can be placed in N places, the free energy would be trivially seen to be extensive (∼ N ). In fact we can rearrange the sum so that there is no exclusion rule. Consider, for example, a diagram with two adjacent squares. This is forbidden by the exclusion rule. But let us combine it with a connected “ﬁgure-8” graph. The two are shown in Figure 13. Neither graph is allowed according to the exclusion rule, but there is another rule which allows such graphs but causes them to cancel. The rule is that every graph is to multiplied by a factor (−1)X where X is the number of times the graph crosses itself. With that rule the ﬁgure-8 graph cancels the graph with two adjacent squares. In fact the new rule is exactly equivalent to the exclusion rule and allows the sum of connected graphs to exponentiate. The rule can be used for the HTE directly, or for the LTE on the dual lattice. In Figure 11 another example is illustrated: the two connected graphs cancel leaving only the disconnected pair of squares with no overlapping links.

70

13

HIGH-T/LOW-T DUALITY

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Homework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Problem 13.1 Use the new rules to calculate the log Z to order (tanh βJ )8 ...................................................................................... The new rule also applies to correlation functions such as C(a, b). Evidently the 2-D Ising model has some very simple features that make it exactly solvable but we will not get into the exact solution here.

Figure 13: Two canceling graphs.

71

Figure 14: Another example.

Note X < 1. b) is given by C(a. b) = (tanh βJ)D + · · · = X D Observe that at large separation the correlation function falls oﬀ exponentially.3) To compute the inverse correlation length m we take the log of C. 14. Let us begin by approaching the critical point from the high temperature side. There are D(D + 1) such graphs and they give C(a.1) (14.3) to leading order in X gives log C = D log X + D(D + 1)X 2 (14. To ﬁnd the corrections to m we try a HTE for the logarithm of C. i.72 14 WHAT HAPPENS AT T = TC ? 14 What Happens at the Critical Temperature Tc ? The behavior of a system at its critical temperature depends on how that temperature is approached. In the high temperature limit m = − log (tanh βJ ).4) . b) ∼ e−mD m = − log X (14. and let D tend to inﬁnity. b) = X D 1 + D(D + 1)X 2 (14. To leading order C(a. from which direction in phase space. divide by D. b) exponentially falls to zero with distance. For simplicity we take the two points to be on the horizontal axis separated by D links.. C(a.2) In general. Let us also rename the expansion parameter by the simpler symbol X.1 High Temperature Expansion of the Correlation Function The high temperature behavior of the correlation function C(a.e. The ﬁrst correction involves graphs in which the line of links makes an extra little excursion with two additional links. for T > Tc the correlation function exponentially tends to zero as with a decay parameter m that depends on temperature. The parameter m is the inverse to the “decay length” which is often called the correlation length. The excursion can either be up one unit or down. Taking the log of (14.

as in Figure 15.14.1 High Temperature Expansion of the Correlation Function 73 b n a D Figure 15: Points a and b are separated along the x-axis by distance D and along the y-axis by distance n. k). Suppressing the D dependence. there is a prefactor multiplying the exponential which gives (1/D) log C some extra non-analytic dependence on the expansion parameter. This is bad because when we divide by D the result grows like D and in the limit it diverges. To do the problem correctly we have to do it in “momentum space. n). Its Fourier transform∗ with respect to n is denoted C(D. Call the correlation function C(D. standard Fourier analysis gives C(n) = ∗ 1 2π +π −π dk e−ikn C(k) Appendix A . The problem is that the correlation function does not exactly behave like an exponential of D.” Let us consider the two points a and b to be separated by D units along the x-axis and n units along the y axis. As we will see.

.. . ...... .. To get to larger values of |n| requires more extra factors of X. C(k) = X D 1 + 2(D + 1)X cos k + 2(D + 1)2 X 2 cos2 k − (D + 1)X 2 (1 − cos 2k) (14. . .. From (14.... ...7) Now take the logarithm. .. .. . . Here is what we get..... . .. To order X 2 we can only get to n = 0. .. ..... . . .5) we get C(0) = X D 1 2π π −π (14. . .. C(0) = 1 + D(D + 1)X 2 C(±1) = (D + 1)X C(±2) = (D + 1)(D + 2)X 2 and the Fourier transform C(k) is given by C(k) = 1 + 2(D + 1)X cos k + (D + 1)2 X 2 (1 + cos 2k) − (D + 1)X 2 (1 − cos 2k) or.1 Using stationary phase. . . . . show that the correlation function C(D.74 +∞ 14 WHAT HAPPENS AT T = TC ? C(k) = −∞ C(n)eikn (14.9) For large D the integral is calculated by the method of stationary phase.. 0) directly in a HTE..... . .. ±2. .. ... ...6) (14... . . D log C(k) ≈ 2DX cos k − 2DX 2 sin2 k + D log X Now let us Fourier transform back to C(0). .. (14.... .. .. . . . ... . Problem 14. .. .. ±1.. . Later we will restore it. . .. ... .. .. . All diagrams contain the factor X D so let us ignore it for now. . ... ... 0) has the large D behavior D 1 C→√ Xe2X DX √ The prefactor 1/ DX is the problem in expanding C(D.. .. Dropping terms that are independent of D.. .. . ... .5) We will calculate C(k) to order in X 2 .. . .. . .... .. Note that there are no terms in log C(k) which are of order 2 . . . We ﬁnd the maximum value of the exponent with respect to the integration variable k and expand about it. .... . . ..8) dk e2DX cos k−2DX 2 sin2 k ... . . restoring the factor X D and using the identity 1 + cos 2k = 2 cos2 k.. .. . .. Homework ... .. .. . . But the Fourier transform is free of such non-analytic behavior.

Typically near the critical point m(T ) behaves like m(T ) ∼ (T − Tc )ν (14. The general theory of the critical behavior is the renormalization group (RG).3 Renormalization Group For the moment forget that the spin σ can only take on the values ±1. On each supersite deﬁne a variable f ′ (sometimes called a block variable). In place of σ we introduce a continuous variable fi at each lattice site. Thus at T = Tc we write C(a. The partition function is given by dfi F (f ). i. b) = D −η (14. It represents the integrand of the partition function. Consider a function F of all the spins. The constant η is often replaced by η = 2∆ where ∆ is called the dimension of the “ﬁeld” σ.. λ(f1 + f2 + f3 + f4 ) f′ = (14.14.14) 4 .12) The constants ν and η are called critical exponents.2 Approach to Tc 75 14.2 Approach to Tc We have just seen that the inverse correlation length behaves like m = | log X| − 2X (14. let us coarse-grain the lattice by grouping four sites together and call the resulting 2 × 2 square a supersite.10) In the high temperature limit m is inﬁnite but as X increases the correlation length 1/m increases. I will just touch on it and give the main idea. Think of f as a ﬁeld which ranges over the real numbers. Right at TC .11) For temperatures lower than Tc the correlation function does not go to zero at large D. 14. (14.13) Z= i Next. nor is it non-vanishing at inﬁnity. What does happen is that C falls to zero as a power of D. F = exp −βE(f ). The critical point Tc is the point at which m(T ) vanishes and the correlation length tends to inﬁnity.e. C neither exponentially decays.

To complete the problem we need to specify in initial condition. ′ g1 = B1 (g1 . g2 . g3 .16) If we think of F being determined by a set of parameters {g} and F ′ by a similar set {g′ } then B just represents the functional dependence of g′ on g. it has been re-scaled.17) Once we have the functions B we can iterate the process and thin the lattice by another factor of 2 in spacing.15) of all the block variables. (14. · · ·) . i. g3 .19) Critical points correspond to ﬁxed points of B. In the inﬁnite lattice limit we can just think of it as another problem on the exact same lattice but we will have to remember that physically. . g2 . · · ·) ′ g3 = B3 (g1 . · · ·) ′ g2 = B2 (g1 . First integrate over the fi holding ﬁxed the values of f ′ on each supersite. Fn+1 = BFn .76 14 WHAT HAPPENS AT T = TC ? where λ is a number to be chosen later. Now let us carry out the sum in equation (14. (14. Let us call that operation B. Schematically. g3 . Consider the operation of going from the function F to the function F ′ . For the Ising model the initial condition is F (f ) = i δ(fi2 − 1)eβJ ∗ fi fj (14. . The result is a new statistical mechanical problem on a lattice with twice the spacing as the original lattice and one quarter as many variables. values of the parameters of F which reproduce themselves: .9) in two steps. Thus we can consider a sequence of F functions – call them Fn – that all represent the same problem after n lattice doublings. (14. That gives us a function F ′ (f ′ ) (14.18) In other words we have a recursion relation for F . If we have an initial starting point we may iterate the procedure and deﬁne a sequence of progressively coarse-grained versions of the problem.. BF = F ′ .e. g2 .

24) The implication of all of this is that at the critical point the physics is invariant with respect to re-scaling the lattice. we ﬁnd C(2D) = λ−2 C(D). Rewriting (14. triangular. For example. Consider a doubling of the lattice spacing. Obviously.23) . a starting point that ﬂows to a ﬁxed point. not the exact starting point. (14.20) Under certain conditions a given starting point may ﬂow to the ﬁxed point after many iterations of the RG transformation B.21) but we must remember that on the left side. the critical behavior does not depend on whether there are second nearest neighbor interactions. as long as we re-scale the average (over the 4 sites in the supersite) by a factor of λ.8) and the critical exponent η satisﬁes η = 2 log λ or in terms of the dimension ∆ ∆ = log λ (14. since the problems are identical on the original and the doubled lattice we can expect f ′ (0)f ′ (D) = f (0)f (D) (14. After many iterations the exact initial state becomes irrelevant and the behavior tracks to the ﬁxed point. D represents a physical distance twice as big as it does on the right side.14. Moreover it is determined by the recursion relation B. In fact the long-distance behavior of the correlation function at Tc becomes isotropic (rotationally symmetric). or whether the original lattice is a square.3 Renormalization Group 77 Ff p = BFf p . (14. Let us consider the behavior of the correlation function C(D) at the ﬁxed point. (14. That is what a critical point is.17) using f′ = λ(f1 + f2 + f3 + f4 ) 4 and the fact that for large D all the fi in the supersite at the origin are separated from those in the supersite at D by distance 2D on the original lattice. hexagonal · · · lattice.22) It follows that C varies as a power of D as in (14.

(14. In other words the energy of the spin is E = −2dJsσ.78 14 WHAT HAPPENS AT T = TC ? 14. The critical temperature is the value at which .4 Mean Field Theory Mean ﬁeld theory is another type of approximation that is useful for approaching the critical point from the low temperature side. or spontaneous symmetry breaking. so we ﬁnd an equation for the order parameter: s = tanh (2dβJs). It is surrounded by 2d neighbors which it interacts with through the usual interaction −Jσσj . but it completely fails for d = 1.27) This equation is illustrated in Figure 16.e. Mean ﬁeld theory is an approximation whose validity depends on the dimension of the lattice: the higher the dimension d. If d is very large we can assume that the σ sees an average spin s for each of its neighbors. It becomes exact in the limit of high dimension.26) and the average of s obtained by diﬀerentiating log Z is σ = tanh (2dβJs). The symmetry of the Ising model is the discrete reﬂection σ → −σ. the ﬂuctuations about the average are small. i. The quantity s ≡ σ is an example of an order parameter.. Thus we can treat each spin as being in a mean ﬁeld (mean in the sense of average) and calculate its average spin.25) It is also true that if the number of neighbors is large. It is order-of-magnitude accurate for d = 2 and reasonably accurate for d = 3. a quantity which must be zero unless the symmetry of the system is spontaneously broken. the better the approximation. Consistency requires σ = s. Let us begin by assuming that the order parameter is non-zero at temperature β 1 .28) (14. (14. We have seen that the low T behavior involves spontaneous magnetization. The partition function for the single spin is Z= σ e2dβJsσ = 2 cosh (2dβJs) (14. Let us focus on a given spin σ. The magnetization is proportional to the average of σ and is non-zero even as the external ﬁeld tends to zero.

29) How good is the mean ﬁeld method? For d = 1 it’s very bad.6J. It predicts a critical temperature Tc = 2J where the true value is zero. In other words Tc = 2dJ (14. the slope of tanh (2dβJs) at s = 0 is unity.14. Mean ﬁeld theory gives Tc = 4J and the exact solution gives Tc = 2.4 Mean Field Theory 79 Figure 16: The mean ﬁeld consistency relation at high T (left) and low T (right).27J. MFT is even better in d = 3 where it gives Tc = 6J compared to simulations that give Tc = 4. . For low T the tanh-curve and the linear curve intersect twice: once for positive s and once for negative s. For d = 2 it’s better.

... . .. . . . .. . . .... . .... . . .. . . . .... .. . .. .. Homework .. .) By equating this average to s you will obtain a MF consistency condition..... .. . . ...... . .. ... .. .. . . .. .. . .... Instead of focusing on a single spin and treating it in the mean ﬁeld background.. . .. ... .. ... ......... ... . focus on a nearest-neighbor pair in the MF background. .. .. ... .. . .. .. .. .2 Here is a proposal for an improved mean ﬁeld approximation.. (The average of σb will be the same. ....... ... Use it to compute the critical temperature and compare it with the previous approximation. . . . . .. .. . The energy of the pair in the MF is E = −Jσa σb − J(2d − 1)s(σa + σb ) Compute Z for the two-spin system and then compute the average value of σa . ... . .... Call the pair a and b.. . .. ... . . Problem 14.... .

Part III Quantum Gases .

.

protons. the statistics favor particles being in the same state. i. To see why.) can occupy the same quantum states whereas fermions (electrons. etc. Under those circumstances the fermionic or bosonic nature of the molecules becomes important. In the case of bosons. Bosons (photons.e. The 1 distance between molecules is of order ρ− 3 . There are two kinds of particles in nature.83 15 Bose Einstein Condensation We are going to study quantum gases. (15. bosons and fermions. etc..1) will become violated at low enough temperature and the wave packets will overlap. Thus the behavior is classical if ρ< (M T )3/2 h ¯ 3. † . the most likely state for a particle is the state of lowest energy. p2 /2M = (3T )/2 We can also say that the square of the uncertainty in the momentum of a particle is of order (∆p)2 = M T From the uncertainty principle the spread in the wave function of a particle is at least h ¯ (M T )1/2 If this is much less than the spacing between particles. gluons. the more likely it is that an additional particle will wind up in that state. In the bose gas at low temperature there will be a much larger number of molecules in the lowest energy state than in the classical gas.) cannot. gases at low enough temperature that the quantum statistics of particles cannot be ignored. From the Maxwell Boltzmann distribution. The more particles in a given quantum state. In this case the statistics does not matter.1) But for ﬁxed density (15. consider the average kinetic energy of a classical molecule at temperature T. the wave functions of the particles don’t overlap and are orthogonal.† When the temperature of a gas is high a gas of fermions behaves the same way as a gas of bosons and both behave like a gas of classical particles. neutrons.

If the molecules are conﬁned to a box of linear dimensions L the wave numbers satisfy kx = ky = kz = πnx L πny L πnz L (15. the partition function is Z= n 1 − e−β 2M L2 n h2 π 2 ¯ 2 However.2) with n being integers. Photons are bosons and the same methods apply to ordinary bosonic molecules. This is equivalent to shifting the energy of each particle by µ.4) Using the same method that we used for thermal radiation.5) .84 15 BOSE EINSTEIN CONDENSATION At this point it might be a good idea to go back and review the theory of black body radiation in Section 10. Non-relativistic bosons of mass M are described by a wave ﬁeld Ψ that is analogous to the electromagnetic ﬁeld although it has only a single (complex) component. The momentum of a particle is given by p = hk and the energy by ¯ E= p2 h ¯ 2 k2 h ¯ 2π2 2 = = n . In order to account for the conservation of molecules we must introduce a chemical potential. The grand canonical partition function has the form Z= n 1 − e−β 2M L2 n h2 π 2 ¯ 2 −βµ −1 (15. 2M 2M 2M L2 (15. The ﬁeld can also be decomposed into oscillators. there is one feature of the non-relativistic gas not shared by a gas of photons.3) The frequency of the oscillators is ωn = h ¯ π2 2 n 2M L2 (15. Non-relativistic molecules are conserved whereas photons are easily emitted and absorbed.

6) n The next step is to replace the sum by an integral. −1 (15.9) The average number of particles is given by N = − ∂A .8) Another form for A is obtained by deﬁning v = u2 (v is just the energy of a particle). π ρ0 = 4 √ 2M h ¯π 3 √ vdv eβv 1 . eβ(v+µ) − 1 (15. Dividing N by the volume ∂µ we get the density of particles ρ.85 The Helmholtz free energy is A = −T log Z = T log 1 − e−β 2M L2 n h2 π 2 ¯ 2 −βµ (15. π A=T 4 √ 2M L h ¯π 3 √ vdv log 1 − e−β(v+µ) (15.7) √ 2M L h ¯π u2 du log 1 − e−β(u 2 +µ) (15.10) that µ must be positive or else the integral diverges. But as µ → 0 the density approaches a ﬁnite critical limit. π ρ= 4 √ 2M h ¯π 3 √ vdv 1 .10) We can see from (1. Let us ﬁrst change variable from the integers ni to ui where h ¯π ni ui = √ 2M L Note that the relation between sums and integrals is → √ 2M L h ¯π 3 d3 u n Thus the free energy becomes A=T or π A=T 2 √ 2M L h ¯π 3 d3 u log 1 − e−β(u 3 2 +µ) (15.11) .

letting z = βv π ρ0 = 4 15 BOSE EINSTEIN CONDENSATION √ 2M T h ¯π 3 √ zdz ez 1 .6) does not vanish. the ﬁrst term in (1.3. the chemical potential drops to zero. In particular √ the integrand in (1. When the density grows above ρ0 all the extra particles go into the lowest energy state. it would appear that there is an obstruction to increasing the density or decreasing the temperature past the critical value.16) . The number of such particles is N0 = V (ρ − ρ0 ) (15. this is not correct. the density is increased by increasing the number of zero energy particles. The zero energy state is the only state that has a macroscopic number of particles (a ﬁnite fraction of the total).13) We can also think of varying the temperature at ﬁxed density. In other words. The macroscopic occupation of the ground state is called a Bose Einstein Condensate.14) As that temperature is approached from the high temperature side. Below the phase transition in the Bose Einstein condensed phase.165 (M T )3/2 .9) vanishes like v when v → 0. when the density approaches ρ0 the chemical potential becomes zero.12) The integral is approximately equal to 2. Since µ must remain greater or equal to zero.3¯ 2 2/3 h ρ . M (15. h ¯3 (15. but in the sum. Alternately. One ﬁnds ρ0 = 0.15) where V is the volume. What does happen is that the approximation of going from a sum to an integral breaks down.86 or. beyond the phase transition. In this case a critical limit is reached when the temperature reaches T0 = 3. at ﬁxed temperature. −1 (15. the chemical potential is zero and the free energy is π A=T 4 √ 2M L h ¯π 3 √ vdv log 1 − e−βv (15. The ﬁrst term represents particles with zero energy in the large volume limit. However. but only when the temperature is below the critical temperature.

. ..1 For T < T0 compute the energy.. .. and pressure of the B.. . So we expect that NE>0 → Ntotal as T → T0 . . .. .. . .. ....17) .. condensed gas.19) T0 The remaining particles are those in the zero energy condensate: NE=0 = Ntotal 1 − T 3/2 T0 3/2 (15. .. . Homework .. . . . ... heat capacity...... .... .. ..... . .. entropy... . .. .. . . .. . .20) . .... . . . .E.. .. ..... . .. .. . .13): NE>0 = ρ0 V ∼ V T 3/2 (15..... .... ... . ... ... Problem 15.. .. .. .. . . .. . .... . We can write this in the form T 3/2 NE>0 = Ntotal 3/2 (15... ..... . . . The number of particles with energy greater than zero (those not in the zero-energy condensate) is obtained from (15. .. . ... . .18) On the other hand for T ≥ T0 the total number of particles is essentially equal to the number with E > 0. .. ..... . . .. .87 or π A=T 4 5 2 √ 2M L h ¯π 3 √ zdz log 1 − e−z (15. .. ... ..

and are typically pronounced “cinch. cosine and tangent respectively. and the hyperbolic functions sinh x ≡ cosh x ≡ ex − e−x 2 ex + e−x 2 ex − e−x sinh x = x tanh x ≡ cosh x e + e−x These are called the hyperbolic sine.” and “tanj” or “tan-h.” “cosh. . not base-10 logs). y) dx dy Also ∂h A means ∂A/∂h. trigonometric functions. The notation we use may be unfamiliar to you: dx dy f (x.88 A APPENDIX A: SOME BASIC MATHEMATICS A Appendix A: Some basic mathematics Transcendental functions You should be familiar with the exponential and log functions (natural logs. y) means the same as f (x.” Important formulas include d sinh x = coshx dx d cosh x = sinh x dx cosh2 x − sin2 x = 1 Calculus We assume you know the rules for diﬀerentiation (total and partial) and integration (deﬁnite and indeﬁnite).

89 Probability The idea of a probability distribution should be familiar to you. · · · L nπx dx. ±1. 2. n = 1. 2. n = 0. · · · L f (x) sin −L = cos θ + i sin θ we may write f (x) = 1 ∞ cn eiωn x 2 n=−∞ where ωn = n ∗ pi/L and cn = 1 L L −L f (x)e−i∗omegan x dx. n = 0. 1. ±2. Given a function f (t) deﬁned on an interval from −L to L. 0 ∞ P (z)dz = 1 Fourier transforms Fourier analysis is a standard technique for converting periodic functions into trigonometric series. 3. . the Fourier series representation of f (t) is f (x) = where an = bn = Using the Euler identity eiθ ∞ nπx nπx a0 + an cos + bn sin 2 L L n=1 L 1 L 1 L f (x) cos −L L nπx dx. · · · A sawtooth function is shown in Figure 17 with approximations to n = 1 and n = 7. The area underneath a probability distribution between x1 and x2 is the probability of a result between those two limits. The best-known probability distribution is the normal or Gaussian distribution 1 2 P (z) = √ e−z 2π Necessarily.

wikipedia. The gradient of the f (x. y) = 0 The minimum or maximum value of a function is found by setting its derivative(s) to zero. ∆f = λ∆g . y space. The gradient at any point on that line is a vector perpendicular to the line g(x. i. y) is a vector perpendicular to its contour lines.org has a good short article on the subject. Lagrange multipliers The method of Lagrange multipliers allows us to ﬁnd minimum or maximum values of a function subject to equality constraints.90 A APPENDIX A: SOME BASIC MATHEMATICS −2π −π π 2π Figure 17: A sawtooth wave approximated by one-term and four-term Fourier series.. A constrained maximum or minimum is found where the two vectors are perpendicular. The constraint function is a line in x.e. y) subject to a constraint g(x. provided the function is suitably smooth. It is explained in many basic texts. We will illustrate the method with a function of two variables subject to a single constraint: minimize f (x. y) − c.

91 where ∆f is deﬁned as the vector with components ∂f ∂f . y) − c) Setting the three partial derivatives of this function to zero gives us three equations that can be solved for xm . y.org . ∂x ∂y λ is called a Lagrange multiplier. y) − c ∂λ This last equation is of course a restatement of the constraint equation.‡ maximize f (x. for which f (x. y) + λ(g(x. y) + λ(g(x. Thus Λ(x. λ) = f (x. for which f (x. the Lagrange multiplier: ∂f =0 ∂x ∂f =0 ∂y ∂f = 0 = g(x. y) = √ 2 ‡ Source: wikipedia. y) = − 2 √ 2/2. ym (the coordinates where the function is minimized or maximized) and λ. xm = ym = √ √ and xm = ym = − 2/2. y. λ) = f (x. As an example. y) is a tilted plane. y) − c) = x + y + λ(x2 + y 2 − 1) ∂Λ = 1 + 2λx = 0 ∂x ∂Λ = 1 + 2λy = 0 ∂y ∂Λ = x2 + y 2 − 1 = 0 ∂λ Solving these equations yields two solutions. We solve the problem by minimizing (maximizing) an augmented function Λ(x. y) = x + y subject to the constraint x2 + y 2 = 1 The constraint is the unit circle and f (x.

. such as PV diagrams (pressure versus volumes). Phase space is a generalization of this idea to multiple dimensions. Concepts like line.92 B APPENDIX B: PHASE SPACE B Appendix B: Phase space We are all familiar of two-dimensional plots with some physical variable shown on each axis. and volume are likewise generalizations of geometric concepts. You should have no diﬃculty as long as you remember that areas and volumes in phase space are not geometric concepts. area. Thus an “area” on a PV diagram has dimensions of pressure times volume (force per unit area times volume = force times length = energy).

Then it occurred to him that probabilities are multiplicative while entropy and other extensive quantities are additive. Boltzmann found himself in that elevated state. Figure 18: Ludwig Boltzmann’s tomb. Chapter 13 . inscribed with S = log P Ludwig Boltzmann was a colorful character to say the least. For a brief time.93 C Appendix C: Ludwig Boltzmann Ludwig Boltzmann (1844-1906) was Professor of Physics at the University of Vienna and the last of the great classical physicists. one that all scientists dream of and a few occasionally achieve: having discovered something important that no one else yet knew. “Maxwell’s Demon. His famous formula is inscribed on his tombstone in Vienna (Figure 18). He was a large man.∗ . 1998. but struggled to establish a clear relationship. What function takes multiplicative properties into additive properties? One day it occurred to him the logarithm function just does that. He viewed the Second Law as a manifestation of probability. accustomed to consuming copious quantities of food and drink. He was a ﬁne pianist ∗ von Baeyer.” Random House.

though in ill health. Boltzmann suﬀered from depression and perhaps bipolar disorder. he took his own life. In 1905. . But he soon learned to smuggle cases of wine across the line from Oakland. He found to his dismay that the City of Berkeley was dry. he undertook a grueling journey from Vienna to Berkeley to deliver a summer series of lectures. Sadly. Not long after his return to Vienna in 1906.94 C APPENDIX C: LUDWIG BOLTZMANN and once performed Schubert sonatas for William Randolph Hearst.

T has units of energy and can be thought of as equal to kt where t is the “textbook temperature” in degrees K (SI units). ft-lb/sec ft-lb/R 1. we measure entropy in bits.95 D Appendix D: for Engineers Engineers and physicists labor in somewhat diﬀerent worlds and each group has certain conventions that suit their purposes. especially dimensions and units. a dimensionless quantity.38 × 10−23 J/K English units F. as contrasted with the engineering deﬁnition that involves a factor of k. Some common engineering units in thermodynamics are as follows: SI units Temperature Energy Power Entropy Boltzmann’s constant k (S = k log P ) Stefan-Boltzmann constant σ (q = σT 4 ) ˙ In this text. 5. and k is Boltzmann’s constant in J/K. There should be no diﬃculty with units as long as you remember to include the proper constants when performing numerical calculations.67 × 10−8 W/m2 K4 0. Also.04 × 10−23 ft-lb/F . K joule (J) watt (w) J/K 1. again in J/K. Readers who have been trained in engineering thermodynamics may need to be careful about adjusting to the conventions in this book. R ft-lb hp.1714 × 108 BTU/hr ft2 R4 C.

- Leonard Susskind - The Black Hole War
- Statistical Mechanics
- Bernoulii Equation for Incompressible Liquid
- Intermediate Fluid Mechanics
- Kinematics.docx123
- HW 02 - The Law of Gauss
- Complex Potential Theory
- Help
- art%3A10.1007%2Fs10483-007-0702-1 (1)
- PROJECT1[1]
- Nonlinear Quantum Optical Springs and Their Nonclassical Properties
- Chapter 3
- Von Mises 19860019802
- Thermodynamic Equations
- Electromagnetic Waves and Antennas
- Crystallography
- emfesoln_chp01.pdf
- Chap 1
- fourier
- Vens a Bitch
- State Space Model of a Mechanical System in Matlab Simulink.pdf
- QMMitschrift (1)
- emfesoln_chp11.pdf
- QM1-1
- In Physics
- Calculus-Based Physics I - Schnick.pdf
- 36781327-ch4
- Densityfunctionaltheoryoftime-dependentphenomena
- 305Notes_II_1a

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd