Einstein’s early work on Statistical Mechanics

A prelude to the Marvelous Year

Luca Peliti

M. A. and H. Chooljan Member

Simons Center for Systems Biology
Institute for Advanced Study
Princeton (USA)

November 6, 2015 / UFRGS Porto Alegre

Outline

Introduction

The Papers

Einstein vs. Gibbs

Summary
Einstein before Einstein?
The 1902–04 Papers on the “Molecular Theory of Heat”

I Einstein’s approach to Statistical Mechanics is

independent and bolder than Gibbs’
I Einstein focuses on fluctuations as a tool for discovery,
rather than a nuisance
I The search for observable fluctuations leads him to focus
on black-body radiation
The Papers

I. Kinetic theory of thermal equilibrium and of the second law of

thermodynamics

II. A theory of the foundations of thermodynamics

III. On the general molecular theory of heat

Some biographical facts

I In 1902 Einstein had left the ETH having obtained a

diploma in 1900, but not the doctorate
I In spring 1902 his application for Technical Assistant, 3rd
Class, to the Federal Office for Intellectual Property in Bern
was accepted, and he started working there in June
I He married Mileva Marić, whom she had met as a fellow
student at ETH, in January 1903. Their first son was born
in May 1904
I Before the three papers which interest us, he had
published two papers in Annalen der Physik , which he
much later judged “worthless”
Atomism in the XIX Century

I Chemists: Dalton, Avogadro, Cannizzaro

The atomic idea becomes a scientific tool
I Early kinetic theory: Herapath, Waterston
Forgotten for lack of observable consequences
I Kinetic theory: Clausius, Maxwell, Boltzmann, Loschmidt
Maxwell: gas viscosity does not depend on density
Connections with thermodynamics, the problem of entropy
I “Energetists” (e.g., Ostwald and Mach): Atoms are a
concept and a calculating tool, not a reality
(Cf. the position of the Church on Copernicanism)
The man who trusted atoms

I 1870: Ergodic hypothesis and physical interpretation of the

temperature
I 1872: Boltzmann’s equation and the H-theorem
I 1877: S = kB log W and the Boltzmann distribution for
“complex molecules”
I 1884: Microcanonical and Canonical ensembles
(respectively called monode and holode)
Einstein’s motivations

I Einstein aims to “derive the postulates of thermal

equilibrium and the second principle using exclusively the
mechanical equations and the probability calculus”
I He provides “a generalization of the second principle,
which is useful for the application of thermodynamics”
I He also gives the “mathematical expression of entropy
from a mechanical point of view”
I The 1902–03 papers have similar structure: I’ll deal with
them in one go
Mechanical description

I General description of a mechanical system:

dpi
= φi (p1 , . . . , pn )
dt
I Energy is the unique integral of motion:

E(p1 , . . . , pn ) = const.

I (Liouville’s theorem is only implicitly assumed):

X ∂φi
=0
∂pi
i
Probabilistic description

I Observable quantities are given by temporal averages of

functions of state variables:

1 T
Z
A = lim dt A(p1 (t), . . . , pn (t))
T →∞ T 0

I For a given value of E, all observable quantities take on a

constant value at equilibrium
I Ergodic hypothesis: for any region Γ of state space, let τ
be the time spent in Γ during time T . Then
Z
τ
lim = const. = (p1 , . . . , pn ) dp1 · · · dpn
T →∞ T Γ
Probabilistic description

I Ensemble: Given N systems of the same type, the number

dN of systems in the small region g at any given time is
Z
dN = (p1 , . . . , pn ) dp1 · · · dpn
g

I From stationarity (and Liouville’s theorem) one obtains

(p1 , . . . , pn ) = const.

I Einstein has thus derived the microcanonical ensemble

Canonical ensemble
I Consider a small system σ in contact with a much larger
one Σ with total energy

E = η + H, E ∗ ≤ E ≤ E ∗ + δE ∗

I Consider g : πi ≤ πi ≤ πi + δπi (= 1, . . . , `) and

G : Πi ≤ Πi ≤ Πi + δΠi (i = 1, . . . λ):
I dN1 : number of systems that are found in g × G:

dN1 = C · dπ1 · · · dπ` dΠ1 · · · dΠλ

= const. e−2h(H+η) dπ1 · · · dπ` dΠ1 · · · dΠλ

I Number of systems for which the state of σ lies in g:

Z
−2hη
dN ∝ e dπ1 · · · dπ` e−2hH dΠ1 · · · dΠλ
E ∗ −η≤H≤E ∗ +δE ∗ −η
Canonical ensemble

e−2hH dΠ1 · · · dΠλ ' e−2hE ω(E)

R
I χ(E) = E≤H≤E+δE
I Choosing h such that

ω 0 (E)
2h =
ω(E)

χ is independent of the state of σ and we have

dN = const. e−2hη dπ1 · · · dπ`

I The system σ acts like a thermometer, and if σ1 and σ2 are

each in equilibrium with Σ, they are in equilibrium with
each other (“0-th principle”)
I Choosing σ as a single molecule, its average energy is
equal to 3/4h and thus 2h = 1/kB T (in modern notation)
The Entropy

Infinitely slow processes

I Einstein considers two kinds of transformations:
Adiabatic transformations: the evolution equations hold at
every time, but the φi ’s can vary by external
action via parameters λ
“Isopycnic” (=equal-density) transformations: correspond
to the thermal contact with a body at a
different temperature: the evolution equations
do not hold during the transformation, but
before and after
I Any infinitely slow process can be approximated by a
succession of adiabatic and isopycnic transformations
The Entropy
I During an infinitely slow process one has
X ∂E X ∂E
dE = dλ + dpν
∂λ ν
∂pν
| {z }
dQ

I The canonical distribution holds before and after an

infinitesimal transformation, thus from

dW = ec−2hE dp1 · · · dpn

one obtains from the normalization of W

Z P
ec+dc−2(h+dh)(E+ ∂λ E dλ) dp1 · · · dpn = 0

(neglecting fluctuations in E) leading to

2h dQ = d (2hE − c)
The Entropy

Thus, since 1/4h = κT

 
dQ E
dS = =d − 2κc
T T

leading to
Z
E
S = + 2κ log e−2hE dp1 · · · dpn
T

But what about ∆S ≥ 0?

On the growth of entropy

Even Einstein has some difficulties with entropy growth. . .

I Consider an ensemble of N systems of energy between E
and E + δE, and divide the available phase space into
regions gk of equal volume
I Define a “state” by assigning the number Nk of systems
which lie in gk
I Define the “probability” W of a state as the number of ways
of distributing the systems compatible with the state. One
has
Z
N!
log W = log ' const. − ρt log ρt dp1 · · · dpn
N1 ! · · · Nk ! · · ·
I Then W is maximal when the distribution is uniform
On the growth of entropy

I “We have to assume” that W never decreases: thus

Z Z
− ρt 0 log ρt 0 dp1 · · · dpn ≥ − ρt log ρt dp1 · · · dpn for t 0 ≥ t

I From this Einstein deduces (!) that − log ρt 0 ≥ − log ρt

(again neglecting fluctuations. . . )
I Consider a collection of systems σ1 , σ2 , . . . initially isolated
and let them exchange heat among themselves, then get
isolated again and reach equilibrium
P
The initial state dw = dw1 · dw2 · · · = e ν cν −2hν Eν dp
Q
I
evolves into the final stateP 0 0 0 Q
dw 0 = dw10 · dw20 · · · = e ν cν −2hν Eν dp
I Thus from ρt 0 ≤ ρt P
one obtains
0 0 0
P P 0 P
ν cν − 2hν Eν ≤ ν cν − 2hν Eν , i.e., ν Sν ≥ ν Sν
The 1904 Paper
On the general molecular theory of heat

I New expression
R from the entropy: given
ω(E) δE = E<E(p)<E+δE dp one has

ω 0 (E)
Z Z
dE
S= = 2κ dE = 2κ log[ω(E)]
T ω(E)

ω(E) is a property of the system, not of the environment

I A new (more restricted) derivation of the second principle
I Interpretation of the constant κ: the average kinetic energy
of a monoatomic gas at the temperature T is given by 3κT ,
yielding κ = R/(2NA ) = 6.5 · 10−24 J/K
(kB = 2κ ' 1.3 · 10−23 J/K)
The 1904 Paper
On the general molecular theory of heat

Here the lion’s paw starts to be felt. . .

I Einstein now considers fluctuations in E
I The “general” meaning of κ: from
dE (E − E)e−E/2κT ω(E) = 0 one obtains
R

2 dE
E 2 − E = ∆E 2 = 2κT 2
dT
I Application to radiation: where are the largest fluctuations
expected? dE/dT is maximal when radiation intensity is
2
maximal: ∆E 2 ' E
√ p
I But E = cvT 4 , then 3 v = 2 3 κ/c/T ' 0.42/T cm from
Stefan-Boltzmann, while λmax ' 0.293/T cm
I “This coincidence cannot be ascribed to chance, given the
generality of our hypotheses”
Einstein vs. Gibbs

It is well known that while theory would assign to the

[diatomic] gas six degrees of freedom per particle, in our
experiments on specific heat we cannot account for more
than five. Certainly, one is building on an insecure
foundation, who rests his work on hypotheses concerning
the constitution of matter.
Difficulties of this kind have deterred the author from
attempting to explain the mysteries of nature, and have
forced him to be contented with the more modest aim of
deducing some of the more obvious propositions relating to
the statistical branch of mechanics.

Gibbs, 1902, Preface

Einstein vs. Gibbs

Of special importance are the anomalies [fluctuations]

of the energies, or their deviations from their average
values. [. . . ]
It follows that to human experience and observation
[. . . ], when the number of degrees of freedom is of such
order of magnitude as the number of molecules in the
bodies subject to our observation and experiment, ε − ε,
εp − εp , εq − εq , would be in general vanishing quantities,
since such experience would not be wide enough to
embrace the more considerable divergencies from their
mean values [. . . ] In other words, such ensembles would
appear to human observation as ensembles of uniform
energy [. . . ]

Gibbs, 1902
Einstein vs. Gibbs

These relations [the “Jeans law”], found to be the

conditions of dynamic equilibrium, not only fail to coincide
with experiment, but also state that in our model there can
be no talk of a definite energy distribution between ether
and matter [. . . ]
We therefore arrive at the conclusion: the greater the
energy density and the wavelength of a radiation, the more
useful do the theoretical principles we have employed turn
out to be; for small wavelengths and small radiation
intensities, however, these principles fail us completely.
In the following we shall consider the experimental facts
concerning blackbody radiation without invoking a model for
the emission and propagation of the radiation itself.

Einstein, 1905
Einstein vs. Gibbs

What brought Einstein to the blackbody problem in 1904

and to Planck in 1906 was the coherent development of a
research program begun in 1902, a program so nearly
independent of Planck’s that it would almost certainly have
led to the blackbody law even if Planck had never lived.

Kuhn, 1978
Summary

I Einstein considers thermodynamics as a perfect example

of a “theory of principle”, starting from “empirically
observed general properties of phenomena”
I He holds fast to the validity of the statistical principles,
even in the presence of “insuperable difficulties”, but is
ready to renounce aspects of Maxwell electrodynamics
rather than the statistical principles (black-body radiation)
I He introduces (implicitly in 1904, explicitly in 1905) the
“backward reading” of S = kB log W
I In contrast with Gibbs, he welcomes fluctuations as a tool
for investigating microscopic physics
Summary

I The first two papers of his “miraculous year” stem directly

from his interest in fluctuations
I Looking for a “theory of principle”, analogous to
thermodynamics, for electrodynamics led him to the
Special Theory:
. . . we are by no means dealing with a ‘system’
here . . . but rather only with a principle which
allows one to reduce certain laws to others,
analogously to the second law of thermodynamics

to Ehrenfest, 1909