Lectures On Thermodynamics and Statistical Physics: Email: Gleb - Arutyunov@desy - de
Lectures On Thermodynamics and Statistical Physics: Email: Gleb - Arutyunov@desy - de
Lectures On Thermodynamics and Statistical Physics: Email: Gleb - Arutyunov@desy - de
and
Statistical Physics
Gleb Arutyunova∗
Abstract
This course covers the basic concepts of equilibrium thermodynamics and statistical
physics. Staring from foundations, we extensively discuss the laws of thermodynamics in-
cluding their modern and historical formulations. We introduce the method of thermo-
dynamic potentials and use it to establish the conditions of thermodynamic equilibrium.
Concerning statistical physics, we start with introducing the main statistical distributions
(Bose-Einstein, Fermi-Dirac and Maxwell-Boltzmann) for an ideal gas by means of a sim-
plified method of boxes and cells and later describe Gibbs’ general method of statistical
ensembles.
∗
Email: gleb.arutyunov@desy.de
1
Contents
I Equilibrium thermodynamics 8
1 Foundations 9
1.1 Thermodynamic systems – what they are . . . . . . . . . . . . . . . . . . . . 9
1.2 Choice of state parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Equations of state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4 Quasi-stationary, reversible and irreversible processes . . . . . . . . . . . . . . 19
1.5 Examples of thermodynamic systems . . . . . . . . . . . . . . . . . . . . . . . 19
2 Laws of thermodynamics 23
2.1 The first law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2 The second law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 The second law for non-quasi-static processes . . . . . . . . . . . . . . . . . . 34
2.4 Alternative formulations of the second law . . . . . . . . . . . . . . . . . . . . 41
2.5 Absolute temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.6 The third law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3 Thermodynamic potentials 53
3.1 Main thermodynamic potentials . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2 Extremal properties of thermodynamic potentials . . . . . . . . . . . . . . . . 61
3.3 Phase transitions and Gibbs’ phase rule . . . . . . . . . . . . . . . . . . . . . 67
3.4 Thermodynamics of chemical reactions . . . . . . . . . . . . . . . . . . . . . . 70
II Statistical physics 74
1 Foundations 75
2
2.2 Bose-Einstein and Fermi-Dirac distributions . . . . . . . . . . . . . . . . . . . 81
2.3 Boltzmann principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
2.4 Maxwell-Boltzmann distribution . . . . . . . . . . . . . . . . . . . . . . . . . 91
2.5 Classical ideal gas and degeneration criterion . . . . . . . . . . . . . . . . . . 94
2.6 Grand canonical potential for bose and fermi gases . . . . . . . . . . . . . . . 97
2.7 Quantisation of energy and Nernst theorem . . . . . . . . . . . . . . . . . . . 99
3
General references
[1] Herbert Callen, Thermodynamics and Introduction to Thermostatistics, Jonh Wiley &
Sons, Inc., 1985.
[2] Walter Greiner, Ludwig Neise, Horst Stöcker, Thermodynamics and Statistical Me-
chanics, Springer, 1994.
[3] Lev Landau and Evgeny Lifshitz, Statistical Physics: Volume 5 (Course of Theoretical
Physics, Volume 5), Elsevier, 3d edition, 1980.
[4] Franz Schwabl, Statistische Mechanik, Springer-Verlag Berlin Heidelberg New York,
2000.
[5] Hermann Schulz, Statistische Physik, Wissenschaftlicher Verlag Harri Deutsch GmbH,
Frankfurt am Mein, 2005.
4
Organization of the course
The lectures are scheduled on Mondays 8.30-10.00 and Thursdays 8.30-10.00 at Hörsaal I
(Wolfgang Pauli-Hörsaal), Jungius street 9. The first lecture is on 14.10.2019.
The final exam is scheduled on 10.02.2020 (Monday), 10.00-12.00 in Hörsaal I and Hörsaal
II. Retake will take place on 23.03.2020 (Monday), 11.00-13.00 in Hörsaal I.
5
Physical constants
Speed of light in vacuum 299792458 m/s ≈ 3 × 108 m/s
Planck’s constant h 6.62676 × 10−34 kg · m2 /s
Planck’s constant ~ 1.05458 × 10−34 kg · m2 /s
Avogadro’s number N0 6.02 × 1023 mole−1
Boltzmann constant k 1.38065 × 10−23 J/K
Gas constant R = kN0 8.314 J/(K · mole)
Notation
Absolute temperature (in kelvins) T
Absolute temperature (in Joules) θ = kT
Inverse absolute temperature β = 1θ = kT
1
Empirical temperature ϑ
Number of particles N
Chemical potential µ
Internal energy E
E
Specific internal energy ε= N
S
Specific entropy s= N
Energy of a single particle
Specific volume v = V /N
Particle density n = N/V
Free energy F = E − θS
Gibbs potential (free enthalpy) G = E − θS + pV
Enthalpy H = E + pV
Grand canonical potential Ω = E − θS − µN
One-particle partition function Z
Partition function Z
Pressure and momentum (clash of notation) p
6
Decimal multiplets of units
Milli 10−3 Kilo 103
Micro 10−6 Mega 106
Nano 10−9 Giga 109
Pico 10−12 Terra 1012
Femto 10−15
He H2 N2 O2 CO2 H2 O
C◦ -269 -252.87 -195.79 -182.96 -78.45 +100
K 4.2 20.28 77.36 90.19 194.7 373.15
7
Part I
Equilibrium thermodynamics
8
Chapter 1
Foundations
1. These are systems of many particles interacting with each other and with
external fields. The word “particle" stands here for molecules, if the corresponding system
is a gas or a liquid, and groups of atoms making sides of a crystal lattice in a solid body.
“Many particles" means that the number of particles in a system is measured in units of
Avogardo’s number N0
N0 = 6.022 . . . · 1023 .
In other words, in thermodynamics both the amount of matter N , i.e. the number of
structural units, and its change dN are measured by a number of moles n = N/N0 . Thus,
thermodynamic systems are the systems of laboratory size. From the point of view of
classical mechanics, an analytic description of motion of such multi-body systems is hopeless.
In practice, since N ∼ N0 1, in performing calculations with thermodynamic systems one
can neglect the terms of order O(1/N ) in comparison to 1.
Since for thermodynamic systems N is so large that N/N0 is visibly finite, such systems
are macroscopic and measurements of their physical parameters are naturally performed
with the help of macroscopic devices (thermometers, manometers and so on). This means
that a device simultaneously interacts with a macroscopically large number of particles and
measures an average value of the corresponding physical quantity.
9
1 2
2
3
1 3
external conditions. This statement constitutes the so-called zeroth law of thermodynam-
ics.
A state of thermodynamic equilibrium is the state where all macroscopic parameters of
the system do not change with time and where macroscopic flows of any kind are absent.1
The property of reaching the state of thermodynamic equilibrium is not characteristic to
mechanical systems, especially in view of the the Poincaré recurrence theorem (H. Poincaré,
1890). This theorem states that certain systems will, after a sufficiently long but finite time,
return to a state very close to (for continuous state systems), or exactly the same as (for
discrete state systems), their initial state. Estimates show, however, that this finite time
interval includes for one mole of a substance a factor of order 10N0 , which is tremendously
huge in comparison to the age of the universe (1017 − 1018 s).
The state of thermodynamic equilibrium has two important properties:
10
being successively in a thermal contact with equilibrium systems 2 and 3 does not
change its state, then bringing in a thermal contact systems 2 and 3 does not change
their equilibrium states.
The property of thermodynamic transitivity is in the heart of the notion of temperature
as a new non-mechanical characteristic of an equilibrium state. The numerical value
of this characteristics can be determined by the value of some mechanical parameter
of system 1, which for some reasons is chosen to play a role of a thermometer.
• if under such a devision the value of this quantity remains unchanged and is the same
for each of its parts, f1+2 = f1 = f2 , then this quantity is called non-additive (or the
0-th additivity class).
Examples of additive quantities constitute the number N of particles in a system, its volume
V , total energy E , heat capacity C, etc. To non-additive quantities belong temperature
of a system θ, its pressure p, the chemical potential µ and specific quantities of additive
characteristics, like specific energy ε = E /N , specific heat capacity c = C/N , etc. In the
framework of quasi-static and equilibrium thermodynamics macroscopic characteristics of
other additivity classes than 0 and 1 simply do not exist.
11
by means of which we single out the system from a surrounding media by putting some phys-
ical or imaginary walls. The procedure is not unique and essentially depends on concrete
problems one wants to solve. Let us stress from the very beginning that the thermodynamic
substance which we single out from surroundings remains the one and the same and there-
fore it keeps its properties regardless of the procedure used. The most important choices of
singling out the system from surroundings are the following.
(α) Adiabatically isolated system. This
is a system which is singled out by means of adi-
abatic walls (α) which are impenetrable to heat,
so that there is no heat exchange with surround-
ings. Singling out a system in such a way fixes E , V, a, N
2
its volume, number of particles N , an external
field a penetrating through an adiabatic wall and
the energy of E of the system which is the energy
of all particles inside the adiabatic shell. All the
parameters fixed by adiabatic walls are mechani- Figure 1.2: Adiabatically isolated system.
cal, among them there is no any specific thermo- All parameters are mechanical.
dynamic parameter. An adiabatically isolated
system can exchange energy with surroundings but only in the form of work, not in the form
of heat or matter.3 Graphically, we single out an adiabatically isolated system by double
walls like the walls of a dewar vessel, see Fig. 1.2.
(γ) System singled out by imaginary walls. We consider a big thermodynamic sys-
2
If there are particles of different sorts, then the number of particles of each sort Ni are fixed.
3
All thermodynamic systems can be split into three general classes: open, closed and isolated. Open
systems can exchange with surroundings both matter and energy. Closed systems can exchange only energy
but not matter. Finally, isolates systems do not interact in any way with surroundings. For instance, the
case α is a closed system, as it can exchange energy with surroundings in the form of mechanical work.
12
tem in equilibrium and single out by a mental effort some part of it which, will be an object of
our investigation. Thus, we fix volume V (which is a geometric factor), temperature θ, which
coincides with the temperature of the rest of the system θT and the field(s) a. The exact
number of particles N is not fixed. Because if this, one might get a feeling that a number
of parameters needed for a description of a ther-
modynamic state in the present situation get re-
duced in comparison to (α) and (β) (β). This is,
however, not so. Instead of N one can choose System
another parameter µ (in the case of a multi- ✓T ✓, V, a, µ
component system, µ = {µi }). This parameter
is called chemical potential and its meaning will
be clarified in due course. Thermostat
13
In the version (α) the parameter E is an energy of a system of N bodies as it is un-
derstood in mechanics (i.e. the sum of kinetic energies of all particles plus the sum of
interaction energies with each other and with external fields including walls of the vessel).
In the version (β) this characteristic of the system is not anymore an independent ther-
modynamic parameter but features as a function of θ, V, a, N and it does not have such a
simple mechanical interpretation (there is continuous energy exchange through walls of the
thermostat) but the leading N asymptotics this quantity (which in some sense in the mean
value) must coincide with the value E from the version (α)
(α):
as
E (α) = E (θ, V, a, N ) = N ε(θ, v, a) .
External and internal parameters. In thermodynamics one often speaks about external
and internal parameters. External parameters are those which are determined by positions
of bodies which do not belong to a thermodynamic system under study. Examples include
volume of a system (because it is determined by positions of external bodies), an exter-
nal electric field, etc. External parameters are functions of coordinates of external bodies.
Quantities which are determined by motion and distribution of of particles entering a sys-
tem under study are called internal. Examples include density, pressure, internal energy,
etc. Values of integral parameters depend and are determined by the values of external pa-
rameters. Depending on conditions imposed on system, the one and the same parameter can
be external or internal. For instance in the version (β) volume V is an external parameter
and pressure p = p(θ, V, a, N ) is an internal one, while in the version (δ) the role of V and
p is interchanged – p is an external parameter and V = V (θ, p, a, N ) is an internal one.
14
Indeed, say, under conditions (β)
(β), a system in thermostat with temperature θ and confined
in volume V can contain N particles of a gas or a liquid, or, for instance, play-doh. How
to detect and describe the properties which distinguish one substance from another? In
thermodynamics this is done by studying back-reaction of a system on an external macro-
scopic disturbance. Exerting influence on a system in naturally realised through the walls
that confine it in an a given experiment (α) − (δ)
(δ). This influence must be infinitesimal (yet
macroscopic) and must lead to infinitesimal changes of equilibrium thermodynamic param-
eters. Since in thermodynamics in addition to mechanical quantities, one also has specific
thermodynamic variables, a back-reaction of a system comes correspondingly in two different
ways: a back-reaction on a change of mechanical parameters and the related notion of work
of a thermodynamic system, and a back-reaction with respect to a thermal influence related
to a possibility to act on a system by transmitting heat through thermally-conducting walls.
Below we consider these two possibilities separately.
This is work in the mechanical sense against external forces that support definite values of
thermodynamic parameters of the system. Assuming for definiteness, that we deal with a
system in a thermostat (version (β)(β)), i.e. a state of the system is fixed by a set of variables
(θ, x, N ), where x = (V, a) ≡ (x1 , . . . , xk ),4 then work is a energy transmitted by the system
to surroundings under the change of its macroscopic parameters xi .
The differential expression for work δW un-
der an infinitesimal change dx = (dV, da) ≡
(dx1 , . . . , dxk ) has the same form as in me-
V = `S dV = Sd`
chanics · · · ·
· · ·
·
· · · · · · · ·
k ·· · · ·
X · · · ·
· · · · · ·· · · · ·
δW = Xi dxi . · · · · F = pS
· · · ·· · · ·
i=1 · ·· · ·· · ·
·
· · · · ·
· · · · ·
· · · ·
If, in analogy with mechanics to call xi ther- ·
` d`
modynamic “coordinates", then Xi should
be naturally regarded as thermodynamic
“forces” conjugate to xi . It is traditionally
accepted that δW > 0 if the system per- Figure 1.6: Work done towards shifting a pis-
forms work, and δW < 0 if work is done on ton by amount d`: δW = F d` = pSd` = pdV .
the system: δWext = −δW .
An example that often comes in consideration is an expression for δW related to the
change of its volume, like shifting a piston or changing the form of a vessel in a more
complicated way:
δW = pdV .
Note that pressure p is always positive. The work caused by changing electric or magnetic
fields has a similar expression
δW = Ada .
4
We assume that there could be several external fields, so a = {ai }.
15
Thus, we come to the general conclusion that since the value of δW is measured by macro-
scopic devices, the specification of a system by its reaction on the change on the parameters
(x1 , . . . , xk ) relies on the knowledge of the quantities Xi = δW/δxi as functions of the
thermodynamic parameters
Xi = Xi (θ, N, x1 , . . . , xk ) , i = 1, . . . , k . (1.1)
These equations are known as thermal equations of state. The number of these equations
coincide with the number of mechanical parameters used to describe a thermodynamic state.
For instance, for a system like a gas there
is only one thermal equation of state
X
1
p = p(θ, V, N ) = p(θ, v) , (1.2)
XC = XC (x)
where it was taken into account that pres-
sure is a non-additive variable. C
see Fig. 1.7. Since the value of the integral depends on a shape of the curve, the quantity δW
is not a complete differential of the variables x, θ, N . This is the reason why the expression
δW is used for work rather than dW .
A quantitative measure of this influence is related to the notation of heat δQ. This notion
arose from numerous experiments on calorimetry that in many cases are well-described by
the equation of thermal balance.
Let us assume for simplicity that a system is singled put by walls impenetrable for
particles so that N = const. If these walls are of type α, then, according to the energy
conservation law, the system performs work δW by diminishing its energy exactly as
δW = −dE .
This is the usual conservation law in mechanics. On the other hand, if a system is singled
out by thermally-conducting walls (β)
(β), then the balance of δW and dE breaks due to energy
transfer through the walls and we get
dE + δW = δQ 6= 0 .
16
This energy δQ is precisely what is called heat which the system receives if δQ > 0 (heating)
or looses if δQ < 0 (cooling) via thermally-conducting walls connecting the system with a
thermostat.
A quantitative determination of δQ can be done if we prohibit a system to perform work
δW . Then the thermal influence on the system δQ|δW =0 is defined by the change of its
internal energy dE . For instance, for a gas under a piston, when δW = pdV , fixing rigidly
the piston we achieve V = const, so that
(δQ)V = dE .
δQ = Cdθ . (1.4)
It was introduced in physics by Joseph Black (1770), the Scottish physicist and chemist who
also introduced calorie as a unit of heat, the notion of latent heat of phase transition and
discovered carbon dioxide.
Giving just δQ/dθ = C(θ, V, a, N ) does not have much sense because δQ and, therefore,
C does depend not only on the parameters of a state but also on a process which starts from
this state. The simplest examples include isothermal, θ = const, and adiabatic, δQ = 0,
processes. For an isothermal process with (δQ)θ ≷ 0 one has Cθ = ±∞ because dθ = 0.
For an adiabatic process Ca = 0. It is clear from these examples that the range of C is
infinite, −∞ < C < +∞, and in order to characterise the reaction of a system on heating,
one has to specify a definite procedure of heating which is different from the above discussed
possibilities. The most natural way is to heat a system by keeping the parameters x = (V, a)
and N fixed. Then the reaction of a system on heating will be given by the heat capacity
δQ
CV aN = = CV aN (θ, V, a, N ) = N cV aN (θ, v, a) . (1.5)
dθ
V aN
The following point should emphasised. Although, heat and work have physical dimen-
sion of energy, they do not represent the forms of energy, rather they are the form of energy
transfer. Once energy is transferred to a system, either as heat or as work, it is indistin-
guishable from energy which might have being transferred differently. Both W and Q are
non-zero only under a process that the system undergoes and it has no sense to ask what
value of W or Q is stored insider a thermodynamic body.5 A practical unit of energy is the
so-called 15◦ C calorie, or 4,186 joules: 1 cal15 ≈ 4,186 jules (Joules). This is the amount of
heat which warms 1g water from 14.5◦ C to 15.5◦ C at atmospheric pressure.
A thermodynamic system is fully defined if its thermal equations of state (1.1) and its caloric
equation of state (1.5) are given. The total number of equations of state (thermal+caloric) is
5
In this respect, see Callen [1] on his analogy with water transfer in a pond.
17
Q
Once energy isV,transferre
C does depend not only on the parameters of a state but also on a process which starts from
this state. The simplest examples include an isothermic T =
C aN
processes. For an isothermic process with ( Q)T ? 0 one has
= =C
this state. The simplest examples include an isothermic T = const and adiabatic Q = 0
V processes.(✓,
C✓ = ±1For
a, N ) = N c
VdTaN
an isothermic process with ( Q)T ? 0 one has C
d✓ because = 0. For
anAadiabatic
work 1W
quantitative
<. CThen
< +1,
process
determination
and ininfluence
the thermal order to from energy which might
Ca = 0.of ItQiscan
clear fromifthese
be done
oncharacterise
the system Q|a W
examples
we prohibit
reaction
an
V
of aby
=0 is defined
adiabatic
thattothe
aN
a system
1 <
system C
perform
<
process
+1, and
on heating,
the change
specify a
of
definite
C = 0.
range of C isa infinite,It is clear from these examples
one has toto characterise a reaction of a
in
procedure
order
of heating di↵erent from the abo
itsspecify
internalaenergy
definite
dE . procedure
Fo instance, of
for heating di↵erent
a gas under a piston,from
when the
W = above discussed
pdV , fixing rigidlypossibilities. The
This is the so-called caloric
the piston
Then
we achieve
the reaction
A quantitative
V = const, so
of a systemofon
determination
only
that
( Q)
equation
heating
QV can
under
. will
= dE be
a process that
of state.
be ifgiven
done
most
by the heat
we prohibit
natural
most natural way is to heat a system by keeping the parameters x = (V, a) and N fixed.
Then the
way
reaction
capacity
a system
is
of
to
a
to perform
heat
system
a system by keeping the param
on heating will be given by th
work W . Then the thermal influence on the system Q| W =0 is defined by the change of ⇣ ⌘
its internal
important to The
emphasise
heatingdE
following
or .cooling
CV aN Fo=instance,
point
that thermodynamics
the piston we achieve V = const, cannot or
and willQ
d✓ soQthat
for a=gas
should isreasons
CVunder stored
equal to the total number of independent parameters that define a thermodynamic state. It is
Usually,energy
emphasised.
not
= Cd✓ .
V aN
give insider aheat
⇣ ofQa ⌘system is related to the notion of heat capacity C,
a piston, when W = pdV , fixing rigidly Q
Although,
aN (✓, V, a, N ) = N cV aN (✓, v, a) . CV aN =
why a certain (1.4)
(1.5)
d✓
V aN
= CV aN (✓, V, a, N ) = N cV
equation of state describes a system, it restricts itself to making assertions concerning the
energy,ItThis
state quantities they do
was introduced
(observables). not represent
in physics
is the so-called
Equations
by Joseph
caloric
15
Black
equation
of state cannot bethe
( Q)
C forms
of(1770),
calorie,
= dEthe
V state.
derived of energy,or
. Scottish physicist
This and
within thermodynamics, rather
4,186 they
jo
chemist
is the so-called caloric equation
who also introduced calorie as a unit of heat, the notion of latent heat of phase transition
of state.
and discovered
Usually, carbonordioxide.
they can be found either from experiment or be established by methods of statistical physics.
heating
The following cooling
point of a system
should is related
emphasised. to the notion
Although, heat ofand
heatwork
capacity
haveC,dimension of
Once energy
are given,isthetransferred to athermodynamics
system, can either
be used as heat or a
The following point should emphasised. Although, heat
Once these Giving they
energy,
equations just Q/d✓
do not = represent
C(✓, V, a, N the
C does depend not only on the parameters of
Once energy is transferred to a system,
from energy which might have
to deduce all other thermodynamic characteristics, the
energy which might by
have
which
) does not have
forms
approach of macroscopic
Joseph Black
warms
much sense
of energy,
Q a=state
Cd✓but
either
rather
1g water
because
theyQare
. also on a processenergy,
as heat orOnce
beingof transferred
most important them are energy
this state. The simplest examples include an isothermic T = const and adiabatic
It was
fromintroduced in physics (1770), the ScottishBoth
fro
and,thetherefore,
formdoofnot
they
which starts from
as work,
di↵erently energy
physicist
it is
Q = is
energy transfer.
represent
(1.4)
indistinguishable
the forms of energy, rather they a
0 transferred to a system, either as heat or a
andQchemist
processes.
E and entropy S 6 , an
For seeisothermic
Fig. process
1.8. withbeing
This ( is transferred
Q)Tthe one has Cdi↵erently.
? 0 so-called ✓ = direct from
±1 because dT
problem =W0. and
energy which
For
of
are non-zero
might have being
thermodynamics – transferred di↵erently.
who alsounder
anonly introduced
adiabatic Ccalorie
a process
process thatItas aclear
isthe unit
system of heat, the notion
undergoes thatof
and latent
itthehas noheat
sense ofto
phase transition
ask what value of W
a = 0. from these examples range of C is infinite,
only under a process that the system undergoes and it has n
given equations
andor < ofstored
1discovered
QCis< state,
+1, carbon into
insider
and derive
dioxide.
ordera to all other
thermodynamic
characterise thermodynamic
body.
a reaction of4 aA practical
system
only under
unit one
on heating,
or Q is
a
characteristics.
of energy
process
has to is the
stored insider
that
This
a
the
problem undergoes and it has no
system
so-called
thermodynamic body.4 A practical
will be completely
specify
15
Giving justsolved
C acalorie,
definite in
procedure
or 4,186
Q/d✓ the
= C(✓, forthcoming
ofjoules:
heating
V, a, N1di↵erent
)cal
does fromlectures.
the above
⇡ 4,186
15 not have julesdiscussed
much (Joules).
sense possibilities.
because ThisQisand,The
the therefore,
amount of heat
or
Cmost
does Q is stored insider a thermodynamic
natural way is to heat a system by keeping the parameters x = (V,
whichdepend
warms not1gonly on the
water from parameters
14.5 C toof15.5 a stateC atbut
15a)C
also on a process
atmospheric body.
calorie,
and
A practical
N fixed. or 4,186 joules: 1 cal
which starts from
pressure.
4 15 ⇡ 4,186 jules (Joules)
Then the reaction of a system on heating will be given by the heat capacity
which warms 1g water
Q = from 14.5 C to 15.5 C at atmospheric
0 one The main problems of
this state. The simplest examples include an isothermic T = const and adiabatic 0
15 Cmain calorie, dT =⇡
ofor 4,186
processes. For ⇣ Q⌘
an isothermic process with ( Q) ?joules: has C = ±11 because
cal 0. For4,186 jules (Joules
is clear from these examples that the range of15
T ✓
CV aN = = CV aN (✓, V, a, N ) = N cV aN (✓, v, a) . (1.5)
an The
adiabatic problems
process C = 0. Itthermodynamics
ad✓ C is infinite,
V aN The main problems of thermodynamics
which warms
Alternatively, 1g
can be found by water
This is the so-called caloric equation of state.
specify a definite procedure of heating from
di↵erent from 14.5
the above discussedC to 15.5
1 < C < +1, and in order to characterise a reaction of a system on heating, one has to
possibilities.
Laws of thermodynamics
The C at atmospheri
methods
most natural
The
ofway
statistical
following point
mechanics
is toshould
heat aemphasised.
system byAlthough,
keeping theheat parameters
and work have x= (V, a) and
dimension of N fixed.
energy,
Then the
Once
they
energy
caloric
do
Equations of state
not represent
reaction of a system
A thermodynamic
is transferred
equation
the forms
system isoffully
to a (1.5)
of state
from energy which might⇣have
energy,
on heating
system,are
rather
will
defined
either
they
be given
as heat
given.
are
if its
The
bythe form
the
thermal
or as
of
heat energy
equations
work,number
total
transfer.
capacity of state (1.1) and its
it is indistinguishable
Aofthermodynamic
Q ⌘being transferred di↵erently. Both W and Q are non-zero
equations of state system
(ther-is fully defined if its thermal e
mal+caloric) CV is
aN equal
= thetosystem
the total
= CV number
aN (✓, of
V,ita, Nindependent
) =sense parameters
caloric
N cVtoaNask
(✓, that
. equation
v, a)value ofdefine
statea (1.5)
(1.5) ther- are given. The total numbe
The main problems of (thermal
only
ormodynamic
thermodynamics
under a process
Q is stored insider and caloric)
that d✓
state.a thermodynamic
undergoes
It is important
V aN
and
to 4emphasise
body.
has no
A practical that
what
thermodynamics
unit of mal+caloric)
energy
of W
cannot andtowill
is equal
is the so-called thenot
total number of independent p
15 Cis
4 calorie, or 4,186caloric 1 cal ⇡ of
joules: equation 4,186 jules (Joules). This is the amount of heat
This Inthe
thisso-called
respect, see Callen [1] 15 state.
on his analogy with water transfer modynamic
in a pond. state. It is important to emphasise that thermo
which warms 1g water from 14.5 C to 15.5 C at atmospheric pressure.
4
The following point should emphasised. Although, heat and work In this respect,
have see Callen
dimension of [1] on his analogy with water transfer in
Equations of state
(
energy,
The maintheyproblems
do not represent
Once energy is transferred
)
of thermodynamics
p = p(✓, V,to
14
the forms of energy, rather
! E , S, . . .
they are the form of energy transfer.
N ) a system, either as heat or as work, it is indistinguishable 14
(thermal
from and caloric)
thermal and caloric:
energy which might
CV = CVhave being
(✓, V, N ) transferred di↵erently. Both W and Q are non-zero
only under a process that the system undergoes and it has no sense to ask what value of W
or QAisthermodynamic
stored insidersystem is fully defined body.
a thermodynamic if its thermal equationsunit
4 A practical of state
4
(1.1) and
of energy its so-called
is the
A thermodynamic system isInfully
15caloric thisdefined respect,
equation of state (1.5) are given. The total number of equations of state (ther-
C calorie, or 4,186 joules: 1 cal15 ⇡ 4,186 jules (Joules). This is the amount of heat
mal+caloric) is equal to the total number of independent parameters that define a ther-
which warms 1g water from 14.5 C to 15.5 C at atmospheric pressure.
seethermal
if its Callen
modynamic state. It is important to emphasise that thermodynamics cannot and will not
caloric equation of state (1.5) are given. The total numb
4
In this respect, see Callen [1] on his analogy with water transfer in a pond.
The main problems of thermodynamics
mal+caloric) is equal to the total number of independent
Can be measured in experiment with the help
14
18
1.4 Quasi-stationary, reversible and irreversible processes
The process of transition of a thermodynamic system from an non-equilibrium state into
an equilibrium one is called relaxation. For each thermodynamic parameter there exists
a characteristic amount of time, called relaxation time, that is needed for this parameter
to reach its equilibrium value from a non-equilibrium state. The total relaxation time is
a maximum of all relaxation times. Computation of relaxation time cannot be done by
methods of thermodynamics because the relaxation mechanism is related to the transfer
by molecules (atoms, electrons, etc.) of energy, mass, momentum and analogous physical
quantities. Therefore, estimation of relaxation times is a problem of physical kinetics.
Let us now imagine a process taking place in a thermodynamic system with velocity
much less than relation velocity. This means that on any state of this process the values of
all parameters will equilibrate and such a process will be represented as a chain of infinitely
close to each other equilibrium states. The very slow processes of this kind are called
equilibrium or quasi-static. Obviously, all real processes in nature are not equilibrium and
can approximate quasi-static processes only to some extent.
Further we note that in a quasi-static process the gradients of any parameters are zero.
It follows from a symmetry principle, that the process can go either in a direct or in a
reverse order, i.e. in the direction of increasing or decreasing of any of the parameters of
the system. Thus, in an equilibrium system a process can be reversed in time and for the
corresponding reversed process the system will pass through the same equilibrium states as
in the direct process, but in the opposite direction. In connection with this property, quasi-
static processes are also called reversible. Most of the processes in nature are irreversible.
In our previous discussion of work and heat, infinitesimal changes δW and δQ are as-
sumed to be obtained as a result of quasi-static processes.
In thermodynamics the graphical method of representing states and processes is widely
use. For instance, in the case of homogeneous systems (gas or liquid) states of a system
are depicted by points and process by lines on the X−x plane, see Fig. 1.7 as an example.
Clearly, such a graphical presentation is possible only for equilibrium states and quasi-
static (reversible) processes, because the system has definite values of the thermodynamic
parameters in equilibrium states only. Equilibrium thermodynamics deals with quasi-static
processes only.
Ideal gas
An ideal gas is a gas of non-interacting point-like particles. This concept provides the
simplest model of a real gas, particles of which do not interact and have atomic dimensions.
The approximation of a real gas by an ideal one is the better the more dilute a gas is.
19
Combining the laws of R. Boyle (1664) - E. Mariotte (1676) and Gay-Lussac (1802), E.
Clapeyron obtained in 1834 the equation of state of a dilute gas in the form pV = cT , where
the constant c depends on the mass and the nature of the gas. Based on this and Avogadro’s
law, in 1874 Mendeleev established the equation of state
m
pV = RT , (1.6)
M
where M is the molar mass and R is a universal constant, i.e. it is the one and the same
for all gases R = 8.314J/(K · mol). One has
R = k · NA , (1.7)
where
k = 1.381 . . . · 10−23 J/K
is Boltzmann’s constant introduced by M. Planck (1900).
Since m/M = ν is a number of moles
and νNA = N is the number of particles, we
rewrite the Clapeyron-Mendeleev equation
p
as
✓ increases
pV = N kT = N θ . (1.8)
pv = θ , (1.9) V
0
where v = V /N is the volume per particle.
As to the caloric equation for an ideal Figure 1.9: Family of isotherms on the p− V
gas, experiments revealed that CV N for an plane.
ideal gas is a constant independent on θ and
V , so that
∂E
= CV N , CV N = cV N N . (1.10)
∂θ V N
so that for such a gas the energy density
∂ε(θ, v)
= cV N . (1.11)
∂θ
By using Joule’s law on independence of its energy on volume under constant temperature
(∂ε/∂v)θ = 0, we obtain that ε = cV N θ + ε0 . The value of cV N can be found by methods of
statistical physics and for an ideal gas cV N = 23 , so that
3
ε(θ, v) = θ , (1.12)
2
where we put an additive constant ε0 to zero.
20
✓ > ✓c ✓ > ✓c
✓ = ✓c
pcr
✓ < ✓c ⇥
⇥
⇥ ⇥
⇥
vcr
Figure 1.10: Isotherms of Van der Waals’ gas. For θ < θcr on the p − V plane one has the
region of unphysical states (parts of isotherms between crosses where (∂p/∂v)θ > 0), which
separates all isotherms on two families.
Real gases obey the equation pv = θ only approximately; this approximation is the better,
the higher temperature and the less density are. At low temperatures deviation from the
ideal gas law becomes essential. There exist a large number (∼150) of semi-empirical pro-
posals for the equation of state proposed by different authors for describing real gases and
playing this role with different level of accuracy. The simplest and most used, although not
the most accurate, is the Van der Waals equation (1873)
a
p + 2 (v − b) = θ .
v
This equation describes quantitatively correct the behaviour of real gases even around the
phase transition to liquid. The equation depends on two empirical constants a and b. The
constant b takes into account the proper volume of molecules themselves. Further, in the
ideal gas one neglects the interaction between the molecules, which is mainly attractive.
The constant a for a > 0 takes into account the extra “internal" pressure that arises due to
these interactions. For constant θ and v the pressure p is smaller the bigger a is.
Typical isotherms of Wan der Waals’ equation have form depicted in Fig. 1.10. There
exists a critical temperature θcr such that for θ < θcr one has (∂p/∂v)θ > 0 which means
that points lying on these parts of isotherms cannot correspond to any real equilibrium
states. The same is also applied to the region where p < 0. At points marked by crosses
one has (∂p/∂v)θ = 0, while on the critical isotherm these points merge into a saddle point
where (∂ 2 p/∂v 2 )θ = 0.
Among further empirical equations, we point out
21
• First Dieterici’s equation
a
p(v − b) = θ exp − α , α ≈ 1.27 .
θv
• Bertlo’s equation
a
p+ (v − b) = θ .
v2θ
22
Chapter 2
Laws of thermodynamics
Robert Emden
Why do we have Winter Heating, Nature,
Vol. 141, 1938, p. 108
23
In the adiabatic situation (α) the energy E is understood in the standard sense of me-
chanics – it is an integral of motion which is the energy of all particles inside the adiabatic
shell). In the sense of (β) it is a function E = E (θ, V, a, N ). There cannot be two dif-
ferent values of the macroscopic energy (that is in the leading asymptotics in N ), namely
E 0 (θ, V, a, N ) > E 00 (θ, V, a, N ), otherwise we could detect the energy difference ∆E = E 0 −E 00
without any change of a thermodynamic system or its surrounding.
Consider a quasi-static process in which a system undergoes a transition from one equi-
librium state into another one such that the corresponding thermodynamics parameters
undergo only an infinitesimal change. Denote by dE , δW and δQ the corresponding changes
of the internal energy, the work done by the system and the amount of heat received. All
these quantities are of the same additive type, δW and δQ describe connections of the sys-
tem to its surroundings. The first law of thermodynamics can be then formulated in the
form of the energy balance: An infinitesimal change of the internal energy dE is due to heat
δQ received, work δW done by the system and, if the number of particles in the system is
not fixed, due to the change of this number by dN
dE = δQ − δW + µdN . (2.1)
Note that here δW is work performed by a system, so that −δW is work performed on a
system by external forces.2
This equation introduces a new quantity µ that has the name “chemical potential". By
its definition, this quantity is the change of the internal energy caused by adding one particle
under the conditions that the system does not perform work and receives no heat
∂E
µ= . (2.3)
∂N
δW =0,δQ=0
This change of the particle number can happen both through a wall permeable to particles
or through an imaginary wall like in the version (γ)(γ). Taking into account the additive
character of energy, one might think that µ coincides with the energy density ε = E /N .
However, in general this is not so, as we will see after the discussion of the second law of
thermodynamics.
Suppose that our system is characterised by a set (θ, V, N ) of thermodynamic parameters.
Let us look at the first law in this case. Since energy E = E (θ, V, N ) is the complete
differential and work is δW = pdV , we have
δQ = dE + δW − µdN = dE + pdV − µdN
∂E ∂E ∂E
= dθ + dV + dN + pdV − µdN ,
∂θ V N ∂V θN ∂N θV
so that
∂E
∂E ∂E
δQ = dθ + + p dV + − µ dN . (2.4)
∂θ VN ∂V θN ∂N θV
2
The first law can be alternatively written as
δQ = dE + δW − µdN , (2.2)
and in this form it states that heat transmitted to a system increases its internal energy, performs work and
expunges particles.
24
Given caloric and thermal equations of state, some information on the right hand side of
the last expression is in fact known. Indeed, the thermal equation is p = p(θ, v). Also, if
V = const and N = const, (δQ)V N = dE and, therefore,
∂E (δQ)V N
= = CV N (2.5)
∂θ VN dθ
is given as the caloric equation of state CV N = CV N (θ, V, N ). The rest of quantities entering
in (2.4) remains unknown.
δQ = θdS . (2.6)
Combining the second law for a quasi-static situation with the first law yields a relation (for
simplicity we put the rest of parameters x to zero)
• From a mathematical point of view, the incomplete differential δQ has the integrating
factor equal to 1/θ.
• It follows from (2.6) that the new thermodynamic variable S has an additive type and,
therefore,
S(θ, V, N ) = N s(θ, v) .
25
• With an appearance of S one can finally exclude the symbol δ from thermodynamic
formulae so that S will feature in these formulae on equal footing with such parameters
as θ, p, V, µ, N . According to Gibbs, S plays the role of a thermodynamic “coordinate",
while θ is then a thermodynamic “force" conjugate to S. In the same way as work is
related with the change of mechanical coordinates, absorption or liberation of heat is
related to the thermodynamic coordinate – entropy.
• The area under a curve on the θ−S-plane of conjugate variables (θ, S) gives the quantity
of heat received by a system during a quasi-static process specified by this curve
Z 2
∆Q = θdS ,
1
analogously to finite work ∆W given by the area enclosed by the p−V diagram of a
process. The value of the integral depends on the path (process) which reflects the
fact that δQ is not a complete differential, see Fig. 2.1.
• While the first law of thermodynamics is universal and applied to any process regard-
less if they are quasi-static (reversible) or not. As to the second law, the form (2.6) is
valid only for quasi-static (reversible) processes and for non-quasi-static (irreversible)
processes it must be replaced by the inequality dS > 0, as will be explained later in
the discussion of the second part of the second law.
The most important consequence of the second law is that, in conjunction with the first law,
it allows to determine from equations of state both energy and entropy, by exploiting the
fact that the latter are well-defined (potential) functions of state variables. Indeed, from the
main equation (2.8) and the fact that energy is a function of state variables, we have
1 ∂E 1 ∂E 1 ∂E
dS = dθ + + p dV + − µ dN
θ ∂θ V N θ ∂V θN θ ∂N θV
∂S ∂S ∂S (2.9)
= dθ + dV + dN ,
∂θ V N ∂V θN ∂N θV
where on the other hand, on the second line we used the fact that entropy is a function of
state variables. The latter also means that the second derivatives of S must satisfy
∂ ∂S ∂ ∂S
= ,
∂V ∂θ ∂θ ∂V
which gives
" # " #
∂ 1 ∂E ∂ 1 ∂E p
= + .
∂V θ ∂θ V N ∂θ θ ∂V θN θ
26
Differentiating and taking into account that
E is potential, we find ✓
1
∂E ∂p
+p=θ . (2.10)
∂V θN ∂θ V N
✓ = ✓(S)
Thus, the partial derivative of an energy
over volume is fully determined via the ther- 2
Q
mal equation of state. On the other hand,
we also have the caloric equation of state 0 S
∂E
= CV N . (2.11)
∂θ V N Figure 2.1: Area on the θ − S diagram.
The potential nature of energy
∂ ∂E ∂ ∂E
=
∂θ ∂V ∂V ∂θ
implies a relation
∂CV N ∂2p
=θ . (2.12)
∂V ∂θ2 V N
that should be regarded as the compatibility condition of the thermal and caloric equations
of state.
These consequences of the second law allows us to unambiguously compute and thermo-
dynamic characteristic of the system once equations of state are given (direct problem of
thermodynamics). We have
where the integral is taken along any path on the θ − v plane connecting the point (θ, v) is
an arbitrary reference point (θ0 , v0 ) and ε0 = ε(θ0 , v0 ). The value of the integral does not
depend on the path because dε is a complete differential. The total internal energy is then
E = N ε(θ, v) + E0 , E0 = N ε0 .
27
b) The system of equations for the entropy density
This system follow immediately from (2.9)
∂s cV N
= ,
∂θ v θ (2.15)
∂s ∂p
= .
∂v θ ∂θ v
where we recall that cV N = cV N (θ, v). This system defines a solution up to an additive
constant
Z (θ,v) ∂p
cV N
s(θ, v) = dθ + dv + s0 , (2.16)
(θ0 ,v0 ) θ ∂θ v
where the integral is taken along any path on the θ − v plane connecting the point (θ, v) is
an arbitrary reference point (θ0 , v0 ) and s0 = s(θ0 , v0 ). The value of the integral does not
depend on the path because ds is a complete differential. The total entropy is
S = N s(θ, v) + S0 , S0 = N s0 . (2.17)
A priory there is no way to fix S0 . This constant is fixed only by the third law of thermo-
dynamics.
c) Chemical potential
From (2.9) we obtain
∂E ∂S
µ= −θ . (2.18)
∂N θV ∂N θV
We have
∂E ∂ ∂ε
= (N ε(θ, v)) = ε − v ,
∂N θV ∂N ∂v
∂S ∂ ∂s
= (N s(θ, v)) = s − v .
∂N θV ∂N ∂v
Therefore, by using (2.13) and (2.15),
∂ε ∂s
µ = ε−v − θs + vθ
∂v ∂v
∂p ∂p
= ε−v θ − p − θs + vθ
∂θ v ∂θ v
= ε − θs + pv , (2.19)
which gives an answer µ = µ(θ, v), provided ε = ε(θ, v) and s = s(θ, v) are already found.
28
From (2.15) we obtain
∂p ∂v ∂p 2
∂θ v
cpN − cV N =θ =θ ∂p
, (2.21)
∂θ v ∂θ p − ∂v θ
As an illustrative example let us perform calculations for an ideal gas which has the thermal
equation of state pv = θ and the caloric equation cV N ≡ cV = const. For the energy we
have
∂ε
= cV ,
∂θ v (2.25)
∂ε θ
= − p = 0 (Joules law) ,
∂v θ v
so that
ε(θ, v) = cV (θ − θ0 ) + ε0 , (2.26)
29
For the entropy we have
∂s cV
= ,
∂θ v θ (2.27)
∂s 1
= ,
∂v θ v
and integrating we get
cV
v θ
s(θ, v) = ln + s0 , (2.28)
v0 θ0
where s0 = s(θ0 , v0 ). For the difference of two specific heat we obtain Mayer’s equation
(1/v)2
cp − cV = θ = 1. (2.29)
−θ/(−v 2 )
Now we are ready to compute the chemical potential of an ideal gas by using the formula
cV
v θ
µ = ε − θs + pv = cV (θ − θ0 ) + ε0 − θ ln − θs0 + θ (2.30)
v0 θ0
This gives the chemical potential as a function of θ and v. Since for an ideal gas pv = θ, we
have
v θ p0
= ,
v0 θ0 p
we can obtain the chemical potential as a function of θ and p
p0 θ cp
µ = (cp − s0 )(θ − θ0 ) − θ ln + θ 0 + ε0 − θ 0 s 0 , (2.31)
p θ0
from where we conclude that
µ0 = µ(θ0 , p0 ) = θ0 + ε0 − θ0 s0 (2.32)
30
Substituting this relation into (2.33), we will get
cV θ0 − ε0 µ0 p0 θ cp
µ= (θ − θ0 ) + θ − ln . (2.34)
θ0 θ0 p θ0
Since for an ideal gas ε0 = cV θ0 , the first term in the last formula vanishes and we obtain a
slightly simplified expression
µ0 p 0 θ cp
µ(θ, p) = θ − ln . (2.35)
θ0 p θ0
A few comments are in order. Equation (2.26) shows that in an isothermal process the
internal energy of an ideal gas remains constant. As a result, heat transferred to the system
from a thermostat during this process is fully converted into work of an expanding gas (heat
−→ work).
An equation for adiabats of an ideal gas can be derived in a number of different ways.
For instance, this equation in the form vθcV = const follows immediately from constancy of
entropy (2.28) in an adiabatic process and it is equivalent to
cp 1
θv γ−1 = const , γ= =1+ . (2.36)
cV cV
Another way to characterise adiabats is to note that since δQ = 0, one has dε = −pdv,
which means that an adiabatic expansion happens due to decrease of an internal energy
(internal energy −→ work). Since for an ideal gas ε = cV θ + ε0 = cV pv + ε0 , we find
pv γ = const , (2.39)
which is of course consistent with (2.36) through the equation of state. Finally, we mention
also an isochoric process which runs under a constant volume v = const. In this case the
gas does not perform work and the heat received goes into internal energy, according to
δQ = CV dθ = dE .
According to the main equation of thermodynamics (2.8), the internal energy appears to
be a function of extensive parameters only, E = E (S, V, N ). Alternatively, we can consider
S = S(E , V, N ), i.e. S is also a function on extensive parameters E , V and N . From
mathematical point of view, this means the following homogeneity condition
E (λS, λV, λN ) = λE (S, V, N ) ,
(2.40)
S(λE , λV, λN ) = λS(E , V, N ) ,
31
valid for any λ. There could be more extensive parameters involved.
The homogeneous first-order property implies that (2.8) can be integrated to a particu-
larly convenient form. Taking λ = 1 + , where is an infinitesimal parameter, we see that
at the first order in λ the following relation holds
∂E ∂E ∂E
E = S+ V + N.
∂S ∂V ∂N
On the other hand, from the main equation of thermodynamics
we see that
∂E ∂E ∂E
= θ, = −p , = µ.
∂S ∂V ∂N
Substituting these values into the previous relation, we will get
E = θS − pV + µN . (2.42)
This is the so-called Euler equation because it represents a particularisation for thermody-
namics of the Euler theorem on homogeneous first-order forms. In fact it is this equation
which allowed us to determine the chemical potential once internal energy and entropy are
known. Indeed, we have
E S V
µ= − θ + p = ε − θs + pv .
N N N
Now we note that the existence of the homogeneous scaling relation (2.40), implies one
relation between intensive parameters. The differential form of this relation can be directly
obtained from the Euler equation and is known as the Gibbs-Duham relation. Taking the
differential of (2.42), we get
Sdθ − V dp + N dµ = 0 . (2.43)
dµ = −sdθ + vdp .
The variation of chemical potential is, therefore, is not independent of the variations of
temperature and pressure. The Gibbs-Duham relation presents the relationship among the
intensive parameters in differential form. Integration of this equation yields the relation in
an explicit form, this requires, however, the knowledge of the equations of state.
In general, the number of intensive parameters capable of independent variations is called
the number of a thermodynamic degrees of freedom of a given system. A simple system with
r components has r + 1 thermodynamic degrees of freedom.
32
Fundamental relation
An equation
S = S(E , V, N1 , . . . , Nr ) , (2.44)
which gives the entropy as a function of extensive variables is known as the fundamental
relation. Inverting this equation with respect to energy, one obtains another form of the
fundamental relation
E = E (S, V, N1 , . . . , Nr ) . (2.45)
E = E (S, V, N ) ,
The relationship between the fundamental relation and equations of state become now
clear. If all three equations of state are known, they may be substituted into the Euler
relation, thereby recovering the fundamental equation. Thus the totality of three equations
of state is equivalent to the fundamental equation and contains all dynamical information
about a system. Any single equation of state contains less thermodynamic information than
the fundamental equation. If two equations of state are known, the Gibbs-Duhem relation
may be integrated to obtain the third. The equation of state so obtained will contain an
undetermined integration constant. Thus, two equations of state suffice to determine the
fundamental equation, except for an undetermined constant.
3
In a sense, this is inverse of the main problem of thermodynamics.
33
Conditions of thermal equilibrium
To proceed, imagine an adiabatically isolated system that is divided into two parts that can
exchange with each other heat, work and particles. Let Ei , Si , Vi be state variables associated
to the corresponding phase i = 1, 2. We have
E1 + E2 = E = const S1 + S2 = S = const
V1 + V2 = V = const N1 + N2 = N = const .
Adding up the second equation from the first and taking into account that dE1 = −dE2 ,
etc., we obtain
Since the change of the variables S1 , V1 , N1 is unrestricted, the last equation remain true
provided
θ1 = θ2 , p1 = p2 , µ 1 = µ2 . (2.46)
All real processes are non-static, which leads, of course, to unavoidable errors when we
approximate them by quasi-static equilibrium processes.
Consider one representative example: expansion of a gas in a cylinder with a movable
piston put in a thermostat. Let us draw a diagram p0 − V , where V is a volume of the gas
and p0 is pressure on the piston. Under a quasi-static expansion the gas has everywhere the
one and the same temperature θ = θT and the one and the same pressure at all its points,
including the pressure on the piston p0 = p. Work done by the gas under its quasi-static
expansion is then given by
Z V2 Z V2
0
∆W = p (V )dV = p(θ, V, N )dV ,
V1 V1
where the equation of state p = p(θ, V, N ) is assumed to be given. If, on the other hand,
the piston was moved non-quasi-statically, that is fast enough that equilibrating of the
34
parameters of the gas does not happen, pressure on the piston under such an intermediate
volume V appears smaller that p(θ, V, N ) because of creation of a zone of low pressure just
behind the moving piston:
p0expand (V ) ≤ p(θ, V, N ) .
δW 0 ≤ δW . ✓ ✓ + d✓
35
the system and, therefore, the heat δQ0 transmitted to the system will be smaller than the
heat δQ needed to heat the system from θ1 = θ to θ2 = θ + dθ, that is
δQ0 ≤ δQ .
This is principle of maximal heat absorption and, as the principle of maximal work, it is a
phenomenological statement generalising our everyday experience. The principle of maximal
energy and maximal heat absorption are precursors of the second part of the second law of
thermodynamics.
Let 1 and 2 be two equilibrium states of a thermodynamic system, such that the corre-
sponding parameters of these states differ from each other by amounts which we denote by
differential symbols d and δ. Under a quasi-static transition 1 → 2 we have, according to
the second law,
δQ = θdS = dE + δW ,
where dS = S2 − S1 . On the other hand, under a non-quasi-static process, according to the
first law,
δQ0 = dE + δW 0 .
The internal energy E is an unequivocal function of thermodynamic state, so that dE 0 =
dE = E2 − E1 , while, according to the principle of maximal work, δW ≥ δW 0 . From here it
follows immediately that δQ ≥ δQ0 , so that we obtain the inequality
δQ0
dS ≥ (2.47)
θ
that expresses the 2nd part of the second law of thermodynamics.
Two comments are in order.
• With an account of (2.47) all thermodynamic relations of equilibrium theory take the
form of estimates, for instance, for N = const,
θdS ≥ δQ0 = dE + δW 0 ,
while the outcome of equilibrium thermodynamics looks like a limiting case of real
processes.
• The second law (2.47) determines not only the direction of evolution of a real process,
but also allows one to study a number of properties of equilibrium states as extremals.
In particular, in an isolated system, when the system does not receive any heat and does
not perform any work, δQ = δW = 0, non-equilibrium processes develops, according
to (2.47), such that
dS > 0 .
This inequality determines the direction of a non-equilibrium process, while an equi-
librium state will correspond to a maximal value of entropy provided the values of
36
thermodynamic parameters (E , V, a, N ) are fixed by α-walls. From a mathematical
point of view, this condition (maximum with respect to virtual changes of parameters
of a system that leave the relations δQ = 0, δW = 0 unbroken) implies two conditions:
the first (condition of extremum, equality) determines the state of thermodynamic
equilibrium itself, the second (condition of maximum, inequality) yields a criterium of
its stability.
Some examples
All these process are assumed to take place in an isolated system (under an adiabatic shell).
We discuss them one by one.
1&2. If initial temperatures of portions of liquids4 θ1 and θ2 are not much different, then
we can neglect effects related to the change of volume of each of the subsystems. In this
irreversible process the total system does not perform any work δW = 0 and two subsys-
tems exchange heat with each other but not with surroundings since the total system is
adiabatically isolated. Therefore, the total internal energy remains unchanged
dE = dE1 + dE2 = 0 .
Since dE1 = δQ1 = C1 dθ, dE2 = δQ2 = C2 dθ, we can determine the final temperature θ of
an equilibrium state,
Z θ Z θ
C1 dθ = − C2 dθ
θ1 θ2
so that
C1 θ1 + C2 θ2
θ= .
C1 + C2
Obviously, this is an arithmetic mean of θ1 and θ2 weighted by C1 and C2 . In particular,
we we mix the one and the same liquid, i.e. C1 = C2 , then θ = 21 (θ1 + θ2 ).
If the process of heat transfer would be reversible (happening under infinitesimal tem-
perature difference) then the individual entropies would undergo the change dS1 = δQ1 /θ
and dS2 = δQ2 /θ. However, because entropy of each body is an unequivocal function of its
state, its change cannot depend on the way heat δQ was taken – reversibly or irreversibly,
but it depends on the value of δQ = Cdθ itself. Therefore, even for a real process of mixing
4
Or of a hot body and cold water.
37
the expressions dS1 = δQ1 /θ and dS2 = δQ2 /θ remain true5 and, therefore, for the total
entropy change we find
Z θ Z θ
dθ dθ θ θ
∆Stot = ∆S1 + ∆S2 = C1 + C2 = C1 ln + C2 ln . (2.48)
θ1 θ θ2 θ θ1 θ2
Studying this expression, one can show that ∆Stot ≥ 0 and the equality is achieved only for
θ1 = θ2 (see exercises). In the case of the same liquid, C1 = C2 = C, we get
θ2 (θ1 + θ2 )2
∆Stot = C ln = C ln
θ1 θ2 4θ1 θ2
We see, in particular, that for a reversible process one obtains for equilibrium temperature
the geometric mean6 of θ1 and θ2 weighted by C1 and C2 . In this case both systems receive
the following amount of heat from surroundings
Z θ Z θ
∆Q = C1 dθ + C2 dθ = C1 (θ − θ1 ) + C2 (θ − θ2 ) .
θ1 θ2
The work performed by the heat engine in the reversible case for one cycle (∆E = 0 for a
cycle) is ∆W = ∆Q.
5
Indeed, for non-quasi-static processes we have the 1-st law δQ0 = dE + δW 0 . If δW 0 = 0, then δQ0 = dE .
Thus, even if δQ0 is received non-quasi-statically, it is the same δQ = dE as would be received in a quasi-static
heating process relating the same initial and final equilibrium states 1 and 2.
6 Qn 1 √
For for a set of numbers x1 , . . . , xn , the geometric mean is defined as i=1 xi
n
= n x1 x2 · · · xn .
38
3. For an ideal gas we have
ε = cV θ + ε0 , s = cV ln θ + ln v + s0 . ✓ 1 , v 1 , N1 ✓ 2 , v 2 , N2
E − E0 = N1 cv θ1 + N2 cv θ2 = (N1 + N2 )cv θ ,
V = v1 N1 + v2 N2 = (N1 + N2 )v .
Namely,
N1 θ1 + N2 θ2 v1 N1 + v2 N2
θ= , v= .
N1 + N2 N1 + N2
For the entropy change we find
∆S = S − S1 − S2
= (N1 + N2 )(cV ln θ + ln v) − N1 (cV ln θ1 + ln v1 ) − N2 (cV ln θ2 + ln v2 )
θ v θ v
= N1 cV ln + ln + N2 cV ln + ln .
θ1 v1 θ2 v2
In particular, if N1 = N2 = N , we have
" #
(θ1 + θ2 )2 (v1 + v2 )2
∆S = N cV ln + ln ≥0
4θ1 θ2 4v1 v2
and the equality is achieved only if θ1 = θ2 = θ and v1 = v2 = v, that is when removing the
wall the state of thermodynamic equilibrium does not change at all.
4. In the case when gases on different sides of the wall are different (two-component system),
the final state of the system is characterised not by two but rather by 3 parameters – final
temperature θ and two specific volumes of each of the components
V N1 v1 + N2 v2 V N1 v1 + N2 v2
v10 = = , v20 = = .
N1 N1 N2 N2
The change of entropy after removing the wall and reaching the state of thermodynamic
equilibrium will be
∆S = N1 (cV )1 ln θ + ln v10 + N2 (cV )2 ln θ + ln v20
− N1 (cV )1 ln θ1 + ln v1 − N2 (cV )2 ln θ2 + ln v2
θ θ v0 v0
= N1 (cV )1 ln + N2 (cV )2 ln + N1 ln 1 + N2 ln 2 .
θ1 θ2 v1 v2
39
In particular, if the mixing gases have the same temperature, we have
v10 v0 V1 + V2 V1 + V2
∆S = N1 ln + N2 ln 2 = N1 ln + N2 ln > 0. (2.49)
v1 v2 V1 V2
The entropy change ∆S appears as a function of three variables N1 , N2 and the ratio V1 /V2
This quantity is called mixing entropy (∆S)mix . Its extremum (∆S)0mix = 0 corresponds to
ξ = 1/2, i.e. N1 = N2 meaning equal amount of gases. Also (∆S)00mix = −4N < 0 at the
extremum, i.e. we deal with maximum. Thus, the mixing entropy is always positive and
reaches its maximal value
((∆S)mix )max = N ln 2 .
The mixing entropy does not depend on the concrete characteristics of gases, but only on N1
and N2 .7 Gases can be any but necessarily different. If::::
we::::::
would::::::
forget::::::
about:::::::::
example :3:::::
that
describes mixing of the one and the same gas, we could make an erroneous conclusion that
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
the formula for (∆S)(ξ) would be also valid when two gases are the same. If this would be
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
7
The reader might notice that our search for an absolute extremum of the function (2.50) can be done in
one go. We write
∆S(y, ξ) = N ξ ln(1 + 1/y) + (1 − ξ) ln(1 + y) .
The extremality condition is given by two equations
∂∆S ∂∆S
= 0, = 0,
∂y ∂ξ
1
whose simultaneous solution is y = 1, ξ = 2
and leads to the same value at the extremum: (∆S)ext = N ln 2.
40
so, then we would run in the situation when removing the wall in the system with θ1 = θ2 ,
v1 = v2 , N1 = N2 , which does not, of course, disturbs the state of equilibrium, entropy
would suddenly jump by the value 2N1 log 2, which contradicts to the to the statement that
entropy is a single-valued function of a thermodynamic state. This contradiction is known
in the literature as the Gibbs paradox. This paradox arises under an erroneous account
of the additive properties of entropy encoded into S(θ, V, N ) = N s(θ, V /N ). In fact, one
should clearly realise that the case 4 in no circumstances can limit to the case 3: a system
consisting from a few sorts of chemically neutral with respect to each other molecules (for
instance O2 and N2 ) by any thermodynamic means cannot be transformed in a system
containing molecules of one sort.
Without running into a contradiction with what have been said above, note that examples
3 and 4 have the common limiting case when a gas expands into vacuum, i.e. we deal just
with one sort of gas. In example 3 this is the case when N2 = V2 /v2 → 0 for V2 = const and
v2 → ∞. This gives θ → θ1 , v → (V1 + V2 )/N1 and
V1 + V2
∆S → N1 log . (2.51)
V1
In example 4 this is the limit N2 → 0 so that from (2.49) we get the same limiting case
(2.51).
Carnot theorem
The theory of heat was born by French physicist Sadi Carnot in 1824 in his self-published
book “Reflections on the Motive Power of Fire Ð and on Machines Fitted to Develop that
Power"8 . In this work Carnot put forward the general theory of heat engines. The basic
questions Carnot asked were – how a heat engine generates work and what are the limitations
on the amount of work produced. It is clear that different heat engines work differently:
some deliver more effect, with their help one can get more work than with others. Is it
possible to improve an engine endlessly? In an attempt to answer these questions, Carnot,
in fact, discovered the second law of thermodynamics in the form of the Carnot theorem
(40 years before the formulation of Clausius). Since the second law of thermodynamics has
been already formulated, we first derive the Carnot theorem from it.
To formulate the Carnot theorem, we need to consider the so-called Carnot cycle consist-
ing of two isotherms and two adiabats. In the θ −S diagram Fig. 2.5 it is simply a rectangle.
8
In the original version “Réflexions sur la Puissance Motrice du feu et sur les Machines propres à Dévelop-
per cette Puissance".
41
The first thermostat with temperature θ1 is called heater and the second one with θ2 < θ1
is called cooler. Let us demand a thermodynamics system consisting of a working body to
perform a quasi-static process along this cycle and determine ::::
the ::::::::::
efficiently ::::::::::
coefficient:::
of
this engine as the ratio of the work done to the heat received for one cycle. As is obvious
::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
from the diagram, the heat ∆QAB > 0 is received from the heater in the isothermal process
A → B, and in the process C → D the heat ∆QCD < 0 is expunged from the system to
the cooler. The processes B → C and D → A are adiabatic, so for them ∆QBC = 0 and
∆QDA = 0. According to the second law,
where ∆S = SB − SA > 0 is the corresponding entropy change. According to the first law,
the change ∆E of the internal energy E for the cycle, which is zero because the system
returns to its original state and energy is a single-valued function of state, is given by
where I
∆W = δW
∆W = ∆QAB + ∆QCD = θ1 ∆S − θ2 ∆S
and for the efficiently coefficient of the Carnot cycle one finds
1) The efficiency of the Carnot cycle does not depend on the nature of the working body
and on limiting adiabats, it is determined only by temperatures of a heater and a cooler.
2) If an engine cyclicly works under the same external conditions and receives in an irre-
versible cycle the same amount Q1 of heat an as in a reversible one, then the efficiency
of this irreversible engine, ηirrev = Wirrev /Q1 , is less than the efficiency of a reversible one
ηrev = Wrev /Q1 :
ηirrev < ηrev .
42
✓1 Isotherms
p
✓
IV I
✓2 A QAB = ✓1 S B
✓1
II QDA = 0 QBC = 0
III ✓2 C
D QCD = ✓2 S
V S
S
Adiabats
Figure 2.5: Carnot cycle in the variables p − V and θ − S. Here a thermostat with the
temperature θ1 > θ2 is called a heater and the one with θ2 < θ1 is a cooler. The working
(third) body performs work by returning to its initial state through the cycle.
It is interesting to point out that changing temperatures θ1 and θ2 have different influence
on the efficiency of the Carnot cycle. We see that
∂ηC θ2 ∂ηC 1 θ1
= 2, =− =− 2.
∂θ1 θ1 ∂θ2 θ1 θ1
Since θ2 < θ1 ,
∂ηC ∂ηC
< ,
∂θ1 ∂θ2
meaning that changing the temperature of the heater produces less effect that changing the
temperature of the cooler.
Let us now show how from the Carnot
theorem the statement of Clausius on the ✓
existence of a single-valued function of state 2 (i)
S such that dS = δQ/θ. Denoting ∆QAB = Q1 > 0
∆Q1 and ∆QCD = ∆Q2 , from (2.52) that
for any Carnot cycle
1 3
∆Q1 ∆Q2
+ = 0.
θ1 θ2
(i)
Q2 < 0
Assume now that a system performs a quasi- 4
static process on an arbitrary cycle. This S
cycle can be drawn in any coordinates, but
it is easier to use the θ−S coordinate system Figure 2.6: Approximating an arbitrary cycle
in view of a simple rectangular form of the by Carnot cycles.
Carnot cycle in these coordinates (although
43
the use of S is a bit “illegal"). We cut this cycle by adiabats as shown in Fig. 2.6, starting
from an adiabat touching the cycle at point 1 up to an adiabat touching it at point 3, and
approximate the original smooth cycle by a fat ladder cycle depicted in Fig. 2.6.
The finer this arrangement, the better approximation we get. The original cycle can be
represented as a sum of many (infinite in the limit) number of Carnot cycles attached to each
other (contributions of internal adiabats that are passed in opposite directions compensate
each other), for each of which one has
(i) (i)
∆Q1 ∆Q2
(i)
+ (i)
= 0.
θ1 θ2
Summing up and passing to the limit of infinite number of cycles, one gets
n Z Z I
∆Q2
X (i) (i)
∆Q1 δQ δQ δQ
lim (i)
+ (i) = + = = 0.
n→∞
i=1 θ1 θ2 123 θ 341 θ θ
The quantity δQ/θ is called reduced heat and the last formula has the name of Clausius
equality. Vanishing of the integral of a function over an arbitrary closed contour means that
::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
the integrand is a complete differential of some single-valued function S and dS = δQ
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::: θ . :::::
This
is precisely Clausius’ formulation of the second
::::::::::::::::::::::::::::::::::::::::::::::::::
law.
It is also instructive to study Carnot’ cycle on the p − V diagram by using an ideal gas
as a working body. It will give a further idea how the individual properties of a working
body disappear from the expression for the efficiency of Carnot cycle. On this diagram the
Carnot process is also performed in four reversible steps.
Step I. An ideal gas undergoes an isothermal expansion from volume V1 to volume V2 at
constant temperature θ1 . The energy of the gas remains constant since it depends on
temperature only and work ∆W1 is performed due to heat ∆Q1 reserved from a thermostat.
We have
Z V2
∆Q1 = ∆W1 = pdV
V1
Z V2
dV V2
= N θ1 = N θ1 ln > 0,
V1 V V1
where the equation of state pV = N θ for an ideal gas was used.
Step II. The gas undergoes an adiabatic expansion. There is no heat exchange with sur-
roundings and expansion happens due to diminution of the internal energy. For the work
∆W2 done by the gas in this process we have9
∆W2 = −∆E = −CV (θ2 − θ1 ) > 0 .
Step III. The gas is compressed at constant temperature θ2 . The amount of heat ∆Q3 given
back to a cooler is due to the work ∆W3 done on the gas
Z V4 Z V4
dV V4
∆Q3 = ∆W3 = pdV = N θ2 = N θ2 ln < 0,
V3 V3 V V3
9
For an ideal gas internal energy is E = CV θ and, therefore, ∆E = CV (θ2 − θ1 ).
44
because V4 < V3 .
Step IV. The gas is compressed adiabatically. The work done on the gas leads to the increase
of the internal energy
The quantities ∆W2 and ∆W4 cancel each other and we obtain
V2 V4
∆W = ∆W1 + ∆W3 = N θ1 ln + N θ2 ln <0
V1 V3
and the efficiency is
∆W N θ1 ln VV12 + N θ2 ln VV34
ηC = = .
∆Q1 N θ1 ln VV2 1
To find the relation between the ratio of volumes, we recall the equation for adiabats
V θcV = const.
45
Taking into account that simultaneous increase by a positive number of both the numerator
and denominator of a ratio increases it, we obtain
In other words, the Carnot cycle has the highest efficiency among all other cycles in the
same temperature range.
Other formulations
2. Kelvin’s formulation (1851): it is impossible to fully convert heat taken from one
thermostat into work without any changes in the surrounding nature (i.e. without
compensation).
It might seen that this formulation contradicts, for instance, the process of isothermal
expansion of an ideal gas. Indeed, all heat received by the gas from a thermostat goes
into work. However, receiving heat and converting it into work is not the only outcome
of this process, what also happens is the change of volume.
One can see that Kelvin’s formulation easily follows from that of Clausius. Indeed,
as we know, work can be completely transformed into heat, for instance, by means of
friction. Therefore, converting heat taken from a single body into work by a process
forbidden by Kelvin’s formulation, we can transform all this work into heat transmitted
by means of friction to another body with higher temperature. This would mean
violation of Clausius’ formulation of the second law.
46
space (θ, V, a), then adiabatically accessible states form a family of such non-intersecting
surfaces that foliate this space and that are solutions of the differential equation δQ = 0
with different values of the integral S(θ, V, a) = const conserved on each surface. This
formulation is equivalent to the existence of an integrating factor for the differential form
defining δQ
δQ = ϑ(θ, V, a, N )dS ,
From a mathematical point of view, the differential form for δQ
∂E
∂E ∂E ∂E
δQ = dθ + + p dV + da + − µ dN
∂θ V N a ∂V θN a ∂a θV N ∂N θV a
where the coordinates xi ’s stand for the all the parameters of a state. If such a form depends
only on two coordinates x1 , x2 , i.e. it is of the form
then an integrating factor always exists and Caratheodory’s principle is not an axiom,
but a derivable mathematical statement. For a large number of independent arguments,
which is actually the case of thermodynamics, the existence of an integrating multiplier is
only possible under certain rigid conditions on the coefficients ωi . :::::::::::::::
Caratheodory’s :::::::::
principle
requires an existence of an integrating factor always, for any number of independent variables
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
characterising a state of thermodynamic equilibrium, which means that this principle is really
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
an axiomatic statement.
::::::::::::::::::::::::
Let us now show how the general formula δQ = ϑ(θ, V, a, N )dS that follows from
Caratheodory’s principle can be reduced to Clausius’ formulation δQ = θdS. To keep
clarity, let us omit external fields and consider
∂E
∂E ∂E
ϑdS = dθ + + p dV + − µ dN .
∂θ V N ∂V θN ∂N θV
For the left hand side to fix the structural pattern (with respect to the N -dependence), the
quantity S must be additive, while ϑ must be a quantity of non-additive type:
ϑ(θ, V, N ) = ϑ(θ, v) .
47
For an infinitesimal quasi-static process δQ = δQ1 + δQ2 ,
where ϑ1 and ϑ2 are some functions of their non-additive variables. Since dS1 and dS2 are
independent, one has
ϑ1 (θ, v1 )
= 1,
ϑ(θ, v1 , v2 )
ϑ2 (θ, v2 )
= 1,
ϑ(θ, v1 , v2 )
for any v1 and v2 , which means that
ϑ = ϑ1 = ϑ2 = ϑ(θ).
Thus, ϑ(θ) is a universal quantity for all subsystems in equilibrium with each other and it
depends on the temperature θ only. An empirical scale ϑ can be used to define an absolute
temperature θ = θ(ϑ).
“Absolute temperature" implies that there exists temperature scale that does not depend
on the empirical scale ϑ(θ) which is used to define it. This means that θ = θ(ϑ) has the
one and the same value independently which empirical thermometer ϑ is used to define it.
To understand this, we take into account that fixing θ we are also fixing ϑ and θ = θ(ϑ).
Consider (2.10)
∂p ∂E ∂E ∂p dϑ
θ = + p(θ, v) = + p(ϑ, v) = θ .
∂θ v ∂V θ ∂V ϑ ∂ϑ v dθ
This yields the following differential equation for the absolute temperature
∂p
dθ ∂ϑ v
= dϑ = f (ϑ)dϑ .
θ ∂E
+ p(ϑ, v)
∂V ϑ
Here the expression in the denominator of f (ϑ) can be measured in the empirical (experi-
mental) scale ϑ. Indeed,
" #
1 δQ 1 δQ ∂p ∂v ∂E ∂v
− = (cp − cV )exp = ϑ = + p(ϑ, v) ,
N δϑ p N δϑ V ∂ϑ v ∂ϑ p ∂V ϑ ∂ϑ p
we find 2
∂p
∂ϑ v
f (ϑ) = ,
∂p
(cp − cV )exp − ∂v ϑ
48
that is the function f (ϑ) is expressed via the quantities measured in experiment (specific heat
measured in the empirical temperature scale ϑ as well as resilience (pressure-temperature,
Elastizität) and compressibility (pressure-volume, Kompressibilität) of a gas. This shows, in
fact, that the change of absolute temperature does not depend on the choice of a thermometer
(thermometric working body):
Z θ1 Z ϑ1 Z ϑ01
dθ θ1
= ln = f (x)dx = f 0 (x0 )dx0 = . . . .
θ0 θ θ0 ϑ0 ϑ00
Thus, the change of the absolute temperature ln θθ10 is defined by the area under a curve
f (ϑ). Different thermometers yield different curves but the one and the same area.
Note that for all thermodynamic systems experiment gives cp > cV and (∂p/∂ϑ)v < 0,
that is the function f (ϑ) is always positive and the area for ϑ1 > ϑ0 is also positive. Thus,
θ1
ln > 0, (2.55)
θ0
for any θ1 and θ0 , which is only possible if θ1 and θ0 are always of the same sign. By an
agreement, one usually chooses θ > 0. In this temperature scale a hotter body (that is the
body with a higher empirical temperature) has also a higher absolute temperature (in the
scale ϑ < 0 this would be opposite). The point θ = 0 is an essential point to be discussed
later.
Let us now establish the relation-
ship of the absolute θ-scale with the cV N
T -scale of Kelvin (W. Thomson, 1848)
related with the gas thermometer. In
this thermometer the working body is
an ideal gas with the equation of state
pV = RT .
We have a=1
∂E ∂p a = 3/2
=T −p = T R/V −p = 0 ,
∂V T ∂T V
0 ✓0 ✓
which is Joules law. Thus, for this ther-
mometer f (T ) = 1/T and we have
Figure 2.8: Temperature dependence of specific
θ T heat for ideal gases.
ln = ln ,
θ0 T0
that is θ = const · T . This means that
θ-scale is different from Kelvin’s scale only by the choice of dimensionality and the value of
grad. This scale is chosen to be
θ = kT ,
where k is the Boltzmann constant k = 1.38 · 10−23 J/K. The reference point of Kelvin’s
scale is a triple point of water to which one associates temperature 273.16 K.
49
Historically, temperature was measured by thermometers – by measuring the height of mercury or
ethanol or by a volume of a gas in a sealed device. All these ways of measuring cannot be considered as
fully satisfactory. Mercury and ethanol thermometers are not good at all for precision measurements: their
work is based on the assumption that expansion of mercury or ethanol is proportional to temperature, an
assumption which is true only approximately. In fact, checking this assumption, by itself, requires to have
a different way of defining temperature.
In reality physicists were lucky. Measurements of temperature by gas thermometers turns out to be a
good method due to the nobel properties of gases to behave themselves at low densities in nearly the same
way (they have the one and the same equation of state). The secret of this luck is in the fact that almost all
gases which are found in Nature are liquified at very low temperatures . In the times of Carno people even
thought that such gases as oxygen and nitrogen always remain in the gas phase. Far from the liquefaction
point they behave themselves as ideal. Therefore, up to now the gas thermometer serves as the main device
for the most precision measurements of temperature. On the other hand, even gas thermometer is not good
when it comes to measuring very high or very low temperatures. It is clear that temperature should be
defined, at least theoretically, without any reference to properties of concrete substances, even ideal gases.
In 1848 Thomson (Lord Kelvin) realised that if the work of the Carnot cycle depends only on temperatures
of a heater and a cooler, one can establish a new temperature scale independent on properties of the working
body. The Carnot cycle can be considered as a device which allows to measure the ratio of two temperatures
θ1 and θ2 from
θ1 ∆Q1
= .
θ2 ∆Q2
Measuring the ratio of heat taken from the heater and returned back to the cooler (or, which is the same,
measuring the ratio of works in two isothermal steps of the Carnot cycle), we obtain the ratio of temperatures
of the heater and cooler.
Thus, the Carnot cycle, in the case it can be realised between two bodies (using one body as a heater
and another as cooler) allows one to determine the ratio of temperatures of these bodies. The temperature
scale defined in such a way is called thermodynamic (absolute) scale of temperatures or Kelvin scale. In order
to define the thermodynamic temperature itself, rather than the ratio of temperatures, one needs to choose
a definite value for one point of a new thermodynamic scale, i.e. one numerical value of temperature can be
assigned arbitrarily. After that all temperatures are defined with the help of the Carnot cycle. Unfortunately,
in spite of the beauty of this theoretical construction, a practical realisation of the Carnot cycle, which is
an infinitely slow quasi-static process, is out of reach.
After Kelvin the question about temperature became quite clear. But the practical problem of building
up a reference thermometer suitable for real measurements remained. For many years for the temperature
scale two points were taken – the melting temperature of ice and the boiling point of water at the standard
atmosphere, which is the mean barometric pressure at the mean sea level. The interval between these
two points was divided into 100 pieces, and one piece was called 1 degree Celsius. This temperature scale
was developed by Swedish astronomer Andres Celsius in 1742, it has two reference points and was widely
accepted. Having adopted the Celsius scale, the relation to the Kelvin scale could be found from
Z ϑ
T
ln = f (x)dx ,
T0 ϑ0
where this time ϑ is measured in degrees Celsius and f (ϑ) is the corresponding empirically determined
function of ϑ. If the unit on the scale T is suitable chosen then we have the correspondence
ϑ = 0◦ C ↔ T0
ϑ = 100◦ C ↔ T = T0 + 100 .
In this scheme the degree of Celsius coincide with the degree of Kelvin and relation between this scales is
T = T0 + t◦ C.
50
However, the point of ice melting is not so
convenient, it depends on pressure and is not
good reproducible. Therefore, now only one
reference point is used – the so-called triple
point of water, which is the temperature at
which all three phases: steam–water–ice coex-
ist in equilibrium. At each temperature there
exists over ice saturated water steam. If we
slowly raise temperature at the moment when
ice stars to melt all three phases will be in
equilibrium. This state corresponds to tem-
perature 0.01◦ C. The point 0.01◦ C, which
can be relatively easily reproduced in the lab,
is adopted as the reference point of the ther-
modynamic (Kelvin) scale and is identified as
273.16 K exactly.11 Such a number has been
exclusively chosen for the purpose that values
of temperatures in the new scale did not prac- Figure 2.9: The phase diagram of water.
tically deviate from the ones in old Celsius scale with two reference points. A transition to the new scale
with the triple point of water12 as the unique reference point was done in 1954.
Exactifying the temperature scale, there appears an ambiguity in the size of degree. Degree of Kelvin
is strictly speaking not the same as degree of Celsius. According to the international convention of 1968,
one calls kelvin the 1/273.16 part of the temperature of the triple point of water on the Kelvin temperature
scale (thermodynamic temperature scale).
51
with the condition s(θ, v)|θ=0 = 0, we obtain
Z θ
cV (θ0 , v) 0
s(θ, v) = dθ . (2.56)
0 θ0
would get
0
s(θ, v) = α ln θ0 |θθ0 =θ a
=0 + βθ + . . . ,
with necessary conditions that a > 0 (the absence of singularity at θ = 0) and β(v) > 0
(positivity of specific heat).
The cases of non-integer dependence of cV N on θ (for instance, cV N ∝ θ ln θ or any other
construction compatible with cV N → 0 as θ → 0) are considered analogously.
Note that the caloric equation for ideal gases cV N = const does not satisfy the above-
established consequence of the 3d law. The point is that the equations of state for ideal
gas pv = θ and cV N = const are only approximate and they are valid at temperatures
higher than the so-called degeneration temperature θ0 . For θ < θ0 specific heat tends to
zero according to the power law (a = 1 or a = 3/2 depending whether a gas is Fermi or
Bose), see Fig. 2.8.
We can now argue that absolute zero is inaccessible. We do not have a thermostat
with θ = 0, so that the only way to try to reach zero is to demand a system to perform a
positive work by diminishing its internal energy. If, in addition, we prevent transfer of heat
from surroundings, i.e. the process will be adiabatic δQ = 0, then we will be decreasing
temperature in the most efficient way. Let us assume that under these idealised conditions
(δQ = 0) we were able to descend from an initial state 1 lying on the adiabat S1 to a state
2 lying on the isotherm θ = 0. But, according to the 3d law, isotherm θ = 0 coincides with
the zero adiabat S0 and the statement of accessibility of absolute zero means that adiabat
S = S1 has a common point with adiabat S = S0 , the situation which is now prohibited
by the 2nd law, see Fig. 2.10. Thus, an assumption about reachability of absolute zero is
incompatible with the 3d and 2nd laws of thermodynamics.
52
Chapter 3
Thermodynamic potentials
where for simplicity we assumed that the system is simple. This is a single equation which
relates seven state variables: θ, S, E , p, V , µ and N . On the other hand, a state of a
simple system is determined by three parameters only. Thus, choosing 3 from 7 variables
mentioned above, we find that the main equation contains 4 undetermined functions. For
their determination we could add, for instance, the thermal and caloric equation of state
and the equation for the chemical potential
p = p(θ, V, N ) ,
E = E (θ, V, N ) ,
µ = µ(θ, V, N ) ,
if θ, V and N are chosen as independent variables. However, under some other choice of
independent variables the main equation allows one to find all unknown functions if one
adds not three but just one equation.
(α)
(α). If (S, V, N ) play the role of independent variables, then for determination of other
4 variables, one needs to know just a single equation for the energy as a function of this
variables
E = E (S, V, N ) . (3.1)
This is precisely the fundamental relation that we have already discussed. Note that in the
main equation of thermodynamics the variables S, V, N enter under differentials. We write
53
from where θ, p and µ immediately follow if know E = E (S, V, N )
∂E ∂E ∂E
θ= , p=− , µ= .
∂S V N ∂V SN ∂N SV
Now the general definition is in order. We call a set of thermodynamic variables natural for
a thermodynamic function X, if the differentials of these variables enter the expression for
dX. With respect to these variables the function X is called a thermodynamic potential or
characteristic function.
Thus, the natural variables for internal energy are S, V, N and E (S, V, N ) is the thermo-
dynamic (adiabatic) potential with respect to these variables. The terminology “thermody-
∂E
namic potential" for the energy comes from analogy with mechanics – pressure p = −( ∂V )SN
is expressed via E in the same way as force F is expressed via potential energy U . 1
Taking into account that dE is the total differential, one can equate the mixed derivatives
∂2E ∂2E
∂S∂V = ∂V ∂S and get an equation
∂θ ∂p
=− , (3.2)
∂V S ∂S V
which relates two different properties of the system – change of temperature under its adia-
batic expansion and change of pressure under isochoric heating. The method of thermody-
namic potentials is designed for establishing relations of this kind.
If the independent variables are not S, V, N , then E considered as a function of these
variables is not a thermodynamic potential. It turns out, however, that even for some other
choices of independent variables one can find a proper thermodynamic potential.
In fact, the thermodynamic potentials are in correspondence with the physical ways to
single out the system from surrounding media. In subsection 1.2 we have labeled these ways
as (α) − (δ) and below introducing the thermodynamic potentials we will apply the same
labelling, although our discussion will not necessarily keep the alphabetic order.
(β)
(β). Choose (θ, V, N ) as independent variables. This set of state variables appears to be
the most convenient one for a “system in thermostat", the variant (β)
(β). We rewrite the main
thermodynamic identity as
dE = d(θS) − Sdθ − pdV + µdN
and introduce the so-called Helmholtz free energy (1882)
F = E − θS . (3.3)
Then
dF = −Sdθ − pdV + µdN , (3.4)
that is F = F (θ, V, N ) is the thermodynamic potential with respect to variables θ, V, N .
The potential character of F shows up in the relations
∂F ∂F ∂F
S=− , p=− , µ= .
∂θ V N ∂V θN ∂N θV
1
That is Fx = − ∂U
∂x
, Fy = − ∂U
∂y
, Fz = − ∂U
∂z
.
54
The second of these equations is the thermal equation of state p = p(θ, V, N ). The second
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
derivatives of F allows one to find caloric characteristics, in particular, CV :
∂2F
CV = −θ ,
∂θ2 V
(δ)
(δ). Independent variables are (θ, p, N ), which are used to describe a system in a thermo-
stat under a piston. The corresponding characteristic function is the Gibbs thermodynamic
potential or free enthalpy,
G(θ, p, N ) = E − θS + pV = F + pV . (3.5)
Obviously,
∂G ∂G ∂G
S=− , V = , µ= . (3.7)
∂θ pN ∂p θN ∂N θp
The middle equation allows one to find the thermal equation of state. The second derivative
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
gives specific heat at constant pressure
∂2G
Cp = −θ .
∂θ2 p
The change of the Gibbs energy under isothermal processes is equal to work in adiabatic
processes performed by an extended system. For instance, in the case of a gas in a closed
vessel such an extended system consists of a gas and a piston with weight over it. The
energy of this system is equal to the internal energy E of a gas and the potential energy
psh = pV of a piston with weight: E = E + pV , from where
dE = dE + pdV + V dp = θdS + V dp ,
2
Note that in mechanics energy consists of kinetic and potential energy, in thermodynamics the internal
energy is divided into free and binding.
55
so that for adiabatic processes S = const one gets (dE)S = V dp, which is the same as
the work of an extended system. On the other hand, from (3.6) it follows that under the
isothermal process (dG)θ = V dp.
The Gibbs potential depends on one extensive variable, N , and, therefore, it has the
structure G(θ, p, N ) = N g(θ, p). Taking into account the third formula in (3.7), we conclude
that ∂G
µ= = g(θ, p) ,
∂N θp
from which we conclude that
G = µN .
This formula straightforwardly follows from the definition of G and the Euler relation (2.42).
The importance of the Gibbs energy for thermodynamics lies in the fact, that in an
equilibrium state of a complex system the state variables p and θ are the same in all parts
of the system and therefore, they are the most convenient ones. In this respect we make the
following comment concerning the often arising formula
X
G= µi N i .
i
However, as it follows from (3.6), the second sum here must vanish. Indeed, taking into
∂G
account the potential character of G(θ, p, Ni ) and µi = ∂N i
, we get
" #
X X ∂2G X ∂ ∂G ∂G
Ni δµi = Ni δNj = Ni − δij δNj
∂Ni ∂Nj ∂Nj ∂Ni ∂Ni
i ij ij
X ∂ X
= µi Ni − G δNj = 0 .
∂Nj
j i
H = E + PV . (3.8)
The name was invented by H. Kamerling-Onnes (1909). For the differential one has
56
The middle equation in (3.10) is the equation for adiabats. Thus, if enthalpy is known as the
thermodynamic potential, the equation for adiabats immediately follows by differentiating
H with respect to p.
The physical meaning of enthalpy lies in the fact that under isobaric processes change
of enthalpy is equal to the heat absorbed (dH)p = (θdS)p = (δQ)p = Cp dθ, that is
∂H
Cp = .
∂θ p
so that ∂2H ∂θ θ
= = .
∂S 2 p ∂S p Cp
(γ)
(γ). The potentials we have introduced above form the so-called 1st group. In this group
the number of particles N occurs as a parameter, rather than a thermodynamic variable. If
we involve both N and µ, we obtain the 2nd group of potentials of the form X 0 = X − µN ,
where X is one of the potentials from the 1st group. In the 2nd group of potentials the
chemical potential µ plays the role of independent argument. The special role in the 2nd
group of potentials is played by the potential Ω = Ω(θ, V, µ) introduced by Gibbs and known
as grand canonical potential
Ω = F − µN = E − θS − µN , (3.11)
Thus, the situation when Ω(θ, V, µ) is given corresponds to singling out the system by
imaginary walls and describing it with the set (θ, V, µ), as in variant (γ)
(γ). We will have
∂Ω ∂Ω ∂Ω
S=− , p=− , N =− . (3.13)
∂θ V µ ∂V θµ ∂µ θV
The additive properties of Ω are obvious,
Ω(θ, V, µ) = V ω(θ, µ) ,
57
so that ∂Ω
p=− = −ω(θ, µ) ,
∂V θµ
which means that
Ω = −pV .
This formula straightforwardly follows from the definition of Ω and the Euler relation (2.42).
The transformation of the results from (θ, V, µ) to more convenient (θ, V, N ) is done by
solving
∂Ω
N =− = N (θ, V, µ)
∂µ θV
with respect to µ in favour of N , and further
∂Ω
p=− = p(θ, V, µ) = p(θ, V, µ(θ, V, N )) .
∂V θµ
The free energy is then reconstructed as follows
Finally, we note that the first letters of the first four potentials, EF GH, form the second
four laters of the Latin alphabet. Abbreviation is also clear – E -energy, F -free energy, G-
Gibbs, H-heat.
Thermodynamic potentials are not unrelated. If one of them is known, the others can be
found from it. For instance, we express the potentials via the free energy F = F (θ, V, N )
∂F
E = F + θS = F − θ ,
∂θ V N
∂F
G = F + pV = F − V
∂V θN
∂F ∂F
H = E + pV = F − θ −V .
∂θ V N ∂V θN
The first of these formulae is the Gibbs-Helmholtz relation and the others are of similar
nature. All potentials in this example appear as functions of θ, V, N . Further, one needs to
express the answer in terms of variables for which the corresponding potential is a charac-
teristic function. This will complete the program of restoring all the potentials via the one
given.
As follows from the above discussion, knowing just one thermodynamic potential, one can
establish both thermal and caloric properties of the corresponding system, i.e. to obtain full
information about its thermodynamic properties. Each thermodynamic potential contains,
therefore, all characteristics of the system.
58
Potential Derivatives w.r.t Maxwell relation
∂θ
∂p
E (S, V, N ) S, V ∂V SN =− ∂S V N
∂θ
∂µ
dE = θdS − pdV + µdN S, N ∂N SV = ∂S V N
∂p ∂µ
V, N ∂N SV =− ∂V SN
∂S
∂p
F (θ, V, N ) = E − θS θ, V ∂V θN = ∂θ V N
∂S
∂µ
dF = −Sdθ − pdV + µdN θ, N ∂N θV =− ∂θ V N
∂p ∂µ
V, N ∂N θV =− ∂V θN
∂S
∂V
G(θ, p, N ) = E − θS + pV θ, p ∂p θN =− ∂θ pN
∂S
∂µ
dG = −Sdθ + V dp + µdN θ, N ∂N θp =− ∂θ pN
∂V
∂µ
p, N ∂N θp = ∂p θN
∂θ
∂V
H(S, p, N ) = E + pV S, p ∂p SN = ∂S pN
∂θ
∂µ
dH = θdS + V dp + µdN S, N ∂N Sp = ∂S pN
∂V
∂µ
p, N ∂N Sp = ∂p SN
∂S
∂p
Ω(θ, V, µ) = E − θS − µN θ, V ∂V θµ = ∂θ V µ
∂S
∂N
dΩ = −Sdθ − pdV − N dµ θ, µ ∂µ θV = ∂θ V µ
∂p ∂N
V, µ ∂µ θV = ∂V θµ
Table 3.1: Maxwell relations resulting from five potentials discussed in the text.
59
ther allows one to determine the equations of state by methods of thermodynamics. For
gases thermodynamic potentials are often computed by methods of statistical mechanics,
while for liquids and solid bodies they are obtained experimentally with the help of caloric
determination of specific heat.
As an example, find an expression for the internal energy of an ideal gas considered as a
thermodynamic potential. It is known that for an ideal gas
E = CV (θ − θ0 ) + E0 .
This is expression is not, however, a thermodynamic potential. One cannot find from this
equation the thermal equation of state and other thermal properties. The internal energy
becomes a thermodynamic potential if it is expressed via S and V . For an ideal gas, this
expression is easy to find by invoking the expression (2.28) for entropy
cV
v θ
s(θ, v) = ln + s0 ,
v 0 θ0
from which we find the temperature
v γ−1 s − s0 1
0
θ = θ0 exp , γ−1= .
v cv cv
Thus, the energy as the thermodynamic potential is
γ−1
V0 S − S0
E (S, V, N ) = CV θ0 exp − CV θ0 + E 0 .
V CV
This is also nothing else but the fundamental relation. This thermodynamic potential allows
to find the thermal equation of state. Indeed,
∂E γ−1
V0 S − S0
θ = = θ0 exp ,
∂S V N V CV
∂E γ−1
θ0 V0γ−1 S − S0 N V0 S − S0
p = − = (γ − 1)CV γ
exp = × θ0 exp .
∂V SN V CV V V CV
In follows from these expressions that P V = N θ.
We can also determine the free energy of an ideal gas. The corresponding expression
follows straightforwardly from the definition F = E − θS, so that
cV
V θ
F (θ, V, N ) = CV (θ − θ0 ) − N θ ln − θS0 + E0 ,
V 0 θ0
where S0 = N s0 . For the Gibbs potential we will have
cV
V θ
G(θ, p, N ) = F + pV = CV (θ − θ0 ) − N θ ln − θS0 + E0 + N θ
V0 θ0
p 0 θ cp
= cp N (θ − θ0 ) − N θ ln + N θ0 + N ε0 − N θs0 = µ(θ, p)N ,
p θ0
60
where we have taken into account that cp = cv + 1 and used the formula (2.31) for the
chemical potential. Enthalpy is obtained in a similar way. Finally, to get Ω(θ, V, µ) = −pV
we need to exclude p in favour of µ. From (2.31) we find
( )
θ c p µ − µ0 θ − θ0
p = p0 exp − (cp − s0 ) ,
θ0 θ θ
E → E + dE , S → S + dS , θ → θ + dθ , V → V + dV, N → N + dN ,
and so on. For a quasi-static transition from the initial into the final state, according to the
first and the second law of thermodynamics, we have
Smax = Sequilib (E , V, a, N ) .
61
From the mathematical point of view, this physical result based on the zeroth and second
laws means the fulfilment of the following conditions: the necessary condition of extremum
– the condition of thermal equilibrium
(δS)E V aN = 0
and the condition of having maximum at the extremum – the condition of stability of an
equilibrium state of a thermodynamic system
(δ 2 S)E V aN < 0 .
It is important to stress how to understand these variations: variations are performed over
those parameters of the system which under fixed conditions (these conditions fix the fi-
nal equilibrium state of the system) can take non-equilibrium values. These could be, for
instance, n, θ in different parts of the system, an amount of matter in different phases,
chemical compounds in a reacting mixture and so on. One can also admit various thought-
ful constructions, walls of type (β) and (γ) (but not (α)
(α)), pistons and so on. The choice of
these internal parameters over which one varies is relatively arbitrary and can be done in
different ways depending on a concrete problem.
Recalling the definition of the free energy F = E − θS, we can rewrite the above inequality
as
[d(E − θS)]θV aN = (dF )θV aN < 0 ,
which means that non-equilibrium processes in a system (β) flow towards lowering its free
energy, and at equilibrium the latter reaches its minimal value (under fixed θV aN )
F (θ, V, aN ) = Fmin ,
62
Thus, non-equilibrium processes in a system surrounded by fictitious walls or by walls per-
meable to matter develop in the direction of lowering its thermodynamic potential Ω. The
equilibrium state Ω(θ, V, a, µ) = Ωmin is then determined by the conditions
so that an equilibrium state is reached when Gibbs’ potential G achieves its minimal value
where δWextirrev is an irreversible work done on the system by external forces. If especially for
irrev = 0, i.e. if no utilisable work
an irreversible process in the isobaric adiabatic system δWext
is performed, we have dH ≤ 0. This means that in an adiabatic isobaric system, which is left
to its own, irreversible process happens which decreases the enthalpy until in equilibrium it
reaches its minimum H = Hmin , which is characterised by the conditions
The notion of enthalpy is especially important for chemistry, since many chemical re-
actions happen in open vessels at constant pressure. Many of these reactions happen so
fast that an exchange of heat with the surroundings is nearly impossible. With the help
of enthalpy one can decide in such a case (p = const, S = const) whether a certain chem-
ical reaction is possible and whether it happens spontaneously under given conditions (for
instance, at room temperature). To this end, one compares the sum of enthalpies of the
reaction products with that of the reactants. If
the reaction happens spontaneously and irreversibly. To simplify such a comparison, pure
chemical elements at room temperature and atmospheric pressure have by definition the
enthalpy H(θ0 , p0 ) = 0, which determines an arbitrary additive constant. Reactions which
happen under constant pressure and where enthalpy is released are called exothermal, ∆H <
0, while reactions where which enlarge the enthalpy, i.e. which only happen if work is
63
performed, are called endothermal, ∆H > 0.3 As an example, if carbon is combusted
together with oxygen, carbon dioxide is formed according to the equation
We restrict ourselves to these five (α) − () practical possibilities, although other con-
structions of this type are possible. A general pattern originating from the 2nd part of the
second law and the zeroth law is that when reaching a state of thermodynamic equilibrium,
precisely that thermodynamic potential exhibits extremal properties which is a characteristic
function with respect to variables that were fixed as parameters of the final equilibrium state.
Extremal properties of the thermo-
dynamic potentials appear very hand- Gibbs potential
ful from a practical point of view. First
of all, the condition (δX)k = 0 for
a thermodynamical potential X deter-
mines a state of thermodynamic equi- G(S + S, V + V )✓pN
librium itself (in the case of multi-
component and multi-phase systems ·
V V + V
this is an interesting problem). The Volume
conditions of minimum (δ 2 X)k > 0 or S
maximum (δ 2 X)k < 0 provide the sta- S+ S
G(S, V )✓pN
bility criterium with respect to spon-
taneous or artificial perturbations of a Entropy
system that do not break fixed values of
certain parameters collectively denoted
Figure 3.1: Deviations of the Gibbs potential away
by k.
from equilibrium under spontaneous fluctuations of
volume and entropy
Stability of a system under me-
chanical influence and heating
As an example, consider a spatially homogeneous system of a gas type put into a cylinder
under a piston, i.e. fix the variables θ, p, a, N and in this way realise the variant (δ)
(δ). The
condition of thermal equilibrium for fixed temperature, pressure and number of particles
(we put a = 0 for simplicity) is minimality of the Gibbs potential G. We consider the
question of stability of the system (δ) under two types of perturbation: with respect to
volume change and with respect to heating. If in a state of thermodynamic equilibrium the
parameters V and S receive small variations δV and δS, then the corresponding change δG
of the thermodynamic potential G must be non-negative for any δV and δS. A necessary
condition for this is vanishing of the coefficients of δV and δS, because terms containing
δV and δS linearly change their sign under the change of signs of δV and δS. A sufficient
condition of minimality of G for small deviations from equilibrium is the requirement of
positivity of the second order differential.
3
Since dH = δQ|p , the measurement of reaction enthalpies in chemistry is simply done by measuring the
amount of heat released in the reaction under constant pressure via a calorimeter.
64
For small deviation from equilibrium we have
and performing the Taylor expansion of the Gibbs potential around equilibrium values S
and V up to the second order in variations, one finds
∂E ∂E
(δG)θpN = δ(E − θS + pV ) = δS + δV − θδS + pδV
∂S ∂V S
" V 2 #
1 ∂2E ∂ 2E ∂ E
+ (δS)2 + 2 δSδV + (δV )2 ,
2 ∂S 2 V ∂S∂V ∂V 2
where all derivatives of the internal energy are taken in the equilibrium state, in particular,4
∂E ∂E
= θ, = −p ,
∂S V ∂V S
where the last formula is one of the Maxwell relations (the first one from Table 3.1). Thus,
the stability condition results in the following inequality
∂θ 2 ∂p ∂p
(δS) − 2 δSδV − (δV )2 > 0 .
∂S V ∂S V ∂V S
This expression is a quadratic form of δS and δV . It will be positive for any δS and δV
provided
∂θ θ ∂p
= > 0, <0 (3.15)
∂S V CV ∂V S
65
of this volume would decrease. This would cause the flow of heat towards the fluctuation
region which would further increase its temperature. In the same way a fluctuation drop of
temperature in some region for CV < 0 would not dissipate rather it would have a tendency
to further amplify. Thus, a state with CV < 0 would be absolutely unstable with respect to
small fluctuations of temperature – in some regions a catastrophic heating of matter would
take place, in other regions – cooling to absolute zero.
Now we turn to inequality (3.16). Here pressure and temperature are treated as functions
p = p(S, V ), θ = θ(S, V ). Equation θ = θ(S, V ) can be inverted to find S = S(θ, V ), then
substituting this in pressure we will have
p = p(S(θ, V ), V ) .
so that
∂p
< 0.
∂V θ
Another way to work out the left hand side of (3.16) is to use the method of determinants.
We note that
∂p ∂p
D(p, θ) ∂V S ∂S V
= .
D(V, S) ∂θ
∂θ
∂V S ∂S V
66
Then using the standard rule for the product of determinants, we arrive at the same result
D(p, θ) D(p, θ) D(V, θ) ∂p ∂θ
= = .
D(V, S) D(V, θ) D(V, S) ∂V θ ∂S V
Thus, for stable states of a homogeneous system both adiabatic and isothermal compres-
sion lead to increase of pressure
∂p ∂p
< 0, < 0. (3.17)
∂V S ∂V θ
The physical meaning of these relations is obvious – in a stable state a gas must exhibit
“resilience" – lowering of its volume must be accompanied by increasing of its internal pres-
sure and vice versa, wherein small fluctuations of density will dissipate. In opposite, for
(∂p/∂V )θ > 0 density fluctuations do not dissipate but rather amplify. For instance, let in
some region a small compression of a gas happens. Then, for (∂p/∂V )θ > 0 the pressure in
this region will reduce and the external pressure continues to compress this region until there
appears a drop of liquid. And vice versa, a region, where a fluctuation decrease of density
happened, will expand due to raising internal pressure until it transforms to a normal gas
state with (∂p/∂V )θ < 0. An ideal gas pv = θ satisfies the requirement (∂p/∂V )θ < 0
everywhere, while phenomenological equations, like the Van der Waals equation
a
p + 2 (v − b) = θ .
v
or Dieterici’s equation
a
p(v − b) = θ exp − α , α ≈ 1.27
θv
have on their isotherms under critical temperature θ < θc the regions where this criterium of
stability is violated so that the points in these regions are unphysical and cannot correspond
to any equilibrium states. States with (∂p/∂V )θ > 0 are absolutely unstable with respect
to small fluctuations of density or decay on two phases.
Our considerations of stability serve as an illustration of the so-called Braun-Le Chatelier
principle: If a system is in a stable equilibrium, then all spontaneous changes of parameters
must invoke processes which bring the system back to equilibrium, i.e. which work against
these spontaneous changes.
67
phases are in equilibrium, small changes of external conditions (for instance, heating or
cooling) leads to transforming of a certain amount of substance from one phase into another
(melting, boiling). Thus, studying the conditions of phase equilibrium, we simultaneously
study the processes of phase transitions.
The necessary conditions of the thermodynamic equilibrium (2.46) are, of course, applied
to equilibrium of two phases. We consider a process of phase transition under constant
pressure and temperature. The thermodynamic potential to consider is G which depends
on θ, p and also on the number of particles N1 and N2 in the corresponding phases. In the
equilibrating process the thermodynamic potential G diminishes, dG < 0, where
From here we see that dN1 ≤ 0 if µ1 ≥ µ2 and vice versa dN1 ≥ 0 if µ1 ≤ µ2 . Thus, the
flux of particles is directed from a phase with bigger chemical potential to a phase with the
smaller one. At equilibrium G reaches its minimum, dG = 0, and the phases equilibrate:
This equation allows one, in principle, to express one of the arguments of the chemical
potential via the other, i.e. to find a curve p = p(θ) or θ = θ(p) and, therefore, define on
the p − θ plane the curve of phase equilibrium. To do so, one needs, however, to know the
analytic expressions for chemical potentials of both phases.
It turns out that even without the knowledge of the corresponding chemical potentials
one can find a differential equation for the curve of phase equilibrium. Equation (3.18) shows
that through the line of the phase transition the chemical potential changes continuously,
without jump. From the Gibbs-Duhem relation dµ = −sdθ + vdp, we see, however, that its
derivatives jump
∂µ ∂µ
= −s , = v, (3.19)
∂θ p ∂p θ
that is the molar volume of the first phase is not equal to the molar volume of the second and
the same is true for specific entropies. Such phase transitions are called phase transitions
of the first kind. Since the phase transition occurs under a constant temperature, from
δQ = θdS, we find that the jump s2 − s1 is related to the molar heat of phase transition
(latent heat)
λ = θ(s2 − s1 ) .
The statement that under the phase transition of the first kind s1 6= s2 is equivalent to
the statement that latent heat λ 6= 0. Such transitions comprise all aggregate changes of
matter (melting, boiling) and many transformations of crystal modifications of the one and
the same substance. We then have
that gives
68
so that we obtain the Clapeyron-Clausius equation
dp s2 − s1 λ
= = . (3.20)
dθ v2 − v1 θ(v2 − v1 )
The equations shows that for λ > 0 (transition with absorption of heat) the derivative
dp/dθ > 0 for v2 > v1 and vice versa, that is under transition with volume increase the
transition temperature grows with pressure (for instance, boiling of water), while under a
transition with volume decrease pressure decreases under increase of the transition temper-
ature, dp/dθ < 0 (for instance, melting of ice).
Above we considered a thermodynamic equilibrium in systems which were physically
inhomogeneous consisting of two (or more) phases but homogeneous with respect to chemical
compound. Now we extend out consideration to complex systems which contain several
chemical components. We assume that these components are chemically neutral with respect
to each other and that the amount of each of these components does not depend on the
presence of other components.5
Generalising the above discussion, we can obtain equations describing equilibrium in a
system that comprises P different phases and consists of K different chemical components.
(k) (k)
To this end, we need to introduce PK chemical potentials µi = ∂G(k) , where Ni is
∂Ni
the number of particles (moles) of the ith component in the kth phase. The condition
of mechanical equilibrium –immobility of boundaries between phases – requires equality
of pressure in all phases, so that the system can be described by unique pressure p. The
condition of thermal equilibrium requires equality of temperature in all phases, so that the
system is characterised by one temperature θ. Generalising equations for chemical potentials,
we will have a system of K (P − 1) equations
(1) (2) (P)
µ1 = µ1 = . . . = µ1 ,
(1) (2) (P)
µ2 = µ2 = . . . = µ2 ,
.........
(1) (2) (P)
µK = µK = . . . = µK .
The state of the system under given masses of all phases is characterised by values of intensive
parameters θ and p, and concentrations
(k)
(k) Ni
Xi = (3.21)
P
K
(k)
Ni
i=1
(k)
of ith components in kth phase. The numbers Xi satisfy the relation
K
X (k)
Xi = 1,
i=1
5
As an example of a two-component (binary) system one can consider a salt solution in water. This
system is two-component but single-phase (homogeneous). If this solution is saturated and crystals of salt
are precipitated, then the system is not homogeneous, it consists of two phases and it remains binary.
69
so that the number of independent concentrations in each phase is K −1. The full number of
parameters including θ and p is, therefore, 2 + (K − 1)P. At equilibrium these parameters
subject to (P − 1)K equations, so the number of independent thermodynamic degrees of
freedom, that is the number of free parameters N left is
N = 2 + (K − 1)P − (P − 1)K = 2 + K − P .
Equation
N =2+K −P (3.22)
N =3−P.
For P = 1 we have N = 2 and the free changing parameters are θ and p. For P = 2
we are on the curve of phase equilibrium and N = 1, so that only one of two variables,
θ or p, can be arbitrarily chosen. Indeed, if one speaks about equilibrium gas-liquid, then
pressure of saturated steam is a function of temperature psat. steam = f1 (θ); if we speak about
equilibrium liquid-solid, then melting temperature is a function of pressure θmelt = f2 (p).
Finally, if P = 3, then N = 0 and we find ourselves at the triple point with definite values
θ0 , p0 of temperature and pressure, which are determined by the equations
p0 = f1 (θ0 ) , θ0 = f2 (p0 ) .
Obviously, the value P = 4 is allowed in principle (equilibrium of a gas, liquid and two
different crystal phases) but forbidden by Gibbs’ phase rule.
Each chemical reaction goes, generally speaking, in both directions. Before an equilibrium
is reached, the direct reaction is taken over the inverse one. At equilibrium two opposite
reactions go with the same velocity, so that the mass of each sort of substance does not
change with time. Let us find the condition of chemical equilibrium.
Any chemical reaction is written in the form
X
νi Ai = 0 , (3.23)
i
70
where Ai are chemical symbols of reacting elements and νi are numbers of the corresponding
molecules Ai participating in this reaction. This numbers are called stoichiometric coeffi-
cients. For instance, for the reaction
A1 = SO2 , ν1 = −2 ; A2 = O2 , ν2 = −1 ; A3 = SO3 , ν3 = 2 .
where δNi are variations in the numbers of particles of reacting chemical elements. Since
Ni = νi N ,6 we have δNi = νi δN . Thus, replacing in (3.24) the variations δNi for νi , we
obtain the condition of chemical equilibrium
X
νi µi = 0 . (3.25)
i
Comparing (3.23) with (3.25), we see that to get the condition of chemical equilibrium, one
needs to replace in the equations of a chemical reaction symbols Ai with the corresponding
chemical potentials µi . If in a system several chemical reactions take place, then an equilib-
rium is determined by a set of equations of the type (3.25). Application of these equations
to concrete chemical reactions requires the knowledge of chemical potentials.
For ideal gases the chemical potential is known up to additive entropy constant, see (2.30),
µ(θ, p) = θ ln p + µ0 (θ) ,
where
µi = θ ln pi + µ0i (θ) ,
P
where pi is a partial pressure of i-th component. The total pressure of a mixture p = i pi .
For ith ideal gas component pi V = Ni θ where V is volume taken by the mixture, so that
Ni N X
pi = θ = Xi p , Ni = N .
N V
i
6
Multiplying νi by N in the equation (3.23) for a chemical reaction gives a totally equivalent equation
which describes the same reaction.
71
The condition of chemical equilibrium of a gas mixture takes the form
X
νi θ ln(Xi p) + µ0i (θ) = 0 .
i
From here
X X X
θ νi ln Xi + θ νi ln p + νi µ0i (θ) = 0 ,
i i i
or
X X 1X
ln Xiνi = − νi ln p − νi µ0i (θ) .
θ
i i i
Considering νi positive for reaction products and negative for reactants, we rewrite the last
formula in the form
Q ν
Xj j
j∈products
Q |ν |
= K(θ, p) . (3.27)
Xk k
k∈reactants
This is the law of mass action. The quantity K(θ, p) is called the reaction constant. The
more the reaction constant the more equilibrium is shifted on the side of products (substances
arising during the reaction) and vice versa.
One can establish how K depends on pressure P and temperature. The dependence of
pressure is fully determined by the multiplier p− i νi . We thus have three different options:
P
1. i νi > 0. As a result of reaction the number of molecules and, therefore, the volume
of gas, increases. In this case increasing pressure leads to decreasing the reaction
constant and, as a consequence, to decreasing the amount of reaction products.
P
2. i νi < 0. The number of molecules (and the volume of a gas) decreases during
the reaction. In this case with pressure increase the reaction constant, as well as the
amount of reaction products increase.
P
3. i νi = 0. Number of molecules (and gas volume) does not change in the reaction. In
this case the reaction constant and the amount of products do not depend on pressure.
72
Taking into account that
we obtain
∂ ∆H
ln K(θ, p) = < 0,
∂θ N θ2
so that the reaction constant and the amount of products decreases with temperature in-
crease. For ∆H > 0 (endothermal reaction)
∂ ∆H
ln K(θ, p) = > 0.
∂θ N θ2
With temperature increase the reaction constant and the amount of products increase.
73
Part II
Statistical physics
74
Chapter 1
Foundations
The main problem of statistical physics can be formulated as follows: knowing the laws
governing the behaviour of individual particles in a system, derive the laws governing the
behaviour of a macroscopic substance. Thus, statistical physics strives for theoretic justi-
fication of thermodynamic laws from the atomistic-molecular point of view. And not only
that. Thermodynamic methods do not allow to establish equations of state, which is neces-
sary in order to fill up the thermodynamic laws with a specific physical content. It is the
statistical approach which allows one, at least in principle, to find equations of state for any
thermodynamic system. However, mainly because of mathematical difficulties, this problem
has not yet been fully solved for a number of realistic systems.
In order to understand the idea of statistical approach, consider, as the simplest example,
a gas consisting of a very large number N of molecules. If we neglect the internal structure
of molecules, classically, this gas represents a many-body system of N interacting particles
moving according to Newton’s law. The corresponding equations of motion for each of the
molecules are
dvi X
mi = Fik ,
dt
k6=i
where vi is a velocity of i-th molecule and Fik is an interaction force of i-th and k-th
molecules. To integrate this system, one first needs to know the interaction strength or the
interaction potential between molecules, the problem of atomic physics which remains far
from been solved, and, second, to have 6N initial data that are initial coordinates xi , yi , zi
and initial velocities vxi , vyi , vzi for each molecule. Even if we would have this information
at our disposal, which is totally non-realistic, integration of Newton’s equations and finding
trajectories for each molecule is impossible in view of a large number of equations (the
number of molecules in 1m3 of a gas under normal conditions is 2.7 · 1025 ). Note also that
integration of Newton’s equations would give us essentially nothing because the knowledge
of trajectories of individual molecules does not yield any information about the properties
of a gas as a whole. However, here comes a new feature on the scene: :: in::a::::::::
system :::::
with
large number of particles there arise new, purely statistical, or probabilistic
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
patterns that
are absent in systems with a small number of particles.
::::::::::::::::::::::::::::::::::::::::::::::::::::::
To understand these patterns, consider the following idealised situation: let us assume
that we can measure very fast and very precise energy of each molecule of the gas. We will
75
be presenting results of such a measurement with the help of a histogram, Fig. 1.1, that is
constructed as follows. The axis of energy is divided onto equal intervals of length 0 . The
choice of 0 is more or less arbitrary and dictated by convenience (it is certainly reasonable
to choose 0 sufficiently small so that all energies in the interval l0 and (l + 1)0 could
be considered as almost equal and given by l0 ). Next, we determine a relative number of
molecules n(l) that have energies in the interval [l0 , (l + 1)0 ], namely,
N (l, l + 1)
n(l) = ,
N
where N (l, l + 1) is the number of molecules in the above interval and N is the total number
of molecules. The function n(l) is called the distribution function of molecules over energies
or distributions over energies for short. The graph we obtain will represent a system of
rectangles, as on Fig. 1.1.
Constructing such a histogram it is
convenient, in addition to partitioning
the energy axis into equally-sized in- n(l)
tervals 0 called in the following cells,
to also introduce a partition on larger
unequally-sized intervals called boxes.
This is needed, because the number
of molecules with very small and very
large energies is small and strongly fluc-
tuates from one experiment to another.
That is why it is convenient to com-
bine in these regions several cells to-
gether (for instance, m) in a larger
✏ 2✏
0 0
✏
l✏0 (l + 1)✏
0
76
where U (xi , yi , zi ) is the potential energy of a molecule. Some distributions over energies
wherein will correspond to a larger number of microstates, others - to a less number. ::::::
There
exists such a distribution over energies that is realised by the maximal number of microstates
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
and for macroscopic masses of a gas this maximum turns out to be extremely sharp. This
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
means that such a distribution, which we call most probable, will be realised most often,
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
and in the state with such distribution the gas spends almost all time. In the idealised
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
experiment described above we precisely obtain the histograms which with overwhelming
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
probability stay close to the histogram of the most probable distribution.
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
All above-said concerns not only the distribution over energy but also any other physical
observables. We could study, for instance, how molecules of a gas get distributed over
coordinates, momenta, angular momenta, etc.
Let us now consider a thermodynamic description of a gas, where an equilibrium state is
specified with the help of such macroscopic parameters as pressure, temperature and so on.
We want to understand what does this mean from the point of view of statistical physics. For
instance, pressure of a gas is defined as an avaraged force caused by collisions of molecules
with 1sm2 of a wall of a container p = F /S. The averaged kinetic energy of a gas is related
to absolute temperature as
mv 2 3
= θ,
2 2
i.e. the absolute temperature is a measure of mean kinetic energy of molecules of a gas. It
is also natural to to identify the thermodynamic internal energy E with the averaged energy
of molecules, atoms, ions, etc constituting the system.
These examples show that the macroscopic parameters p, θ, E and so on represent aver-
aged values of quantities related to individual molecules. One of the important problems of
statistical physics consists in finding such averaged values and relations between them.
The question now is which information should we possess in order to find, for instance,
the mean energy of a molecule. Obviously, this information is precisely the distribution over
energies. Indeed, if we know the function n(l), then
X N (l, l + 1)l X
= = n(l)l .
N
l l
Analogously, one can find mean values of any functions of energy, and also of any func-
tions of coordinates and momenta, provided distributions of molecules over coordinates and
momenta are known.
Therefore, besides the description of a macroscopic system by methods of phenomenolog-
ical thermodynamics, another description, the statistical one, based on the atomic-molecular
model and the use of statistical approach is possible. :: A ::::
link:::::::::
between ::::::
these ::::
two:::::::::
different
descriptions is provided by the basic postulate of statistical physics, namely, a distribution
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
which is realised by the largest number of microstates – the most probable distribution – is
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
the one which corresponds in the thermodynamics terminology to an equilibrium state.
::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
Note the principle difference between the notion of equilibrium state in thermodynamics
and statistical physics. In phenomenological thermodynamics we assume that in the absence
of external disturbances a system remains in a state of equilibrium infinitely long time. From
the point of view of statistical physics, the system spends in this state most of its time. In
77
other words, statistical physics predicts existence of fluctuations, i.e. spontaneous and very
rare deviations from an equilibrium state in macroscopic bodies, which implies violation of
the laws of macroscopic thermodynamics.
Thus, the fundamental problem of statistical physics consists in finding most proba-
ble distributions. It is worth stressing that results of statistical thermodynamics are valid
only for systems with a macroscopically large number of particles. The less the number
of particles in a system, the more distributions occur that sufficiently deviate from the
most probable one. This means that a system with not a large number of particles finds
itself in the most probable (equilibrium) macroscopic state and in other (non-equilibrium)
states comparable times and the use of the most probable distributions becomes unjus-
tified. ::::::::::
Therefore, ::::::::::
statistical:::::::::::::::::
thermodynamics :: is::::
an :::::::::::
asymptotic::::::::
theory, ::::::
which::::::::::::
conclusions
are exactly valid in the limit N → ∞, V → ∞ with the ratio N/V kept fixed.
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
This is
the so-called thermodynamic limit. In the formulas statistical thermodynamics
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
one should
neglect higher powers of 1/N leaving only the leading lowest power.
::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
78
Chapter 2
The main problem of equilibrium statistical mechanics is to determine most probable statis-
tical distributions. In this chapter we solve the easiest occurrence of this problem, namely,
finding the most probable distributions for collectives of non-interacting particles, i.e. for
ideal gases. Statistical properties of ideal gases can be studied with the help of a general
and, at the same time, a simple method – the method of boxes and cells. In spite of the fact
that the ideal gas case is rather specific, it encompasses many concrete physical problems
which solution can be found exactly.
which splits into the product of the volume element in the configuration space dV = dxdydz
and the volume element in the momentum space dVp = dpx dpy dpz . Depending on the
symmetry of the problem it might be convenient to choose for volume elements both in con-
figuration and in momentum spaces not cartesian parallelepipeds but, for instance, spherical
shells: dV = 4πr2 dr and dVp = 4πp2 dp.
79
Consider as the simplest example a one-dimensional gas in the gravitational field of the
earth. The lines of constant energy are determined by the equation
p2
= + mgx
2m
and they are parabolas. The phase space volume element is the area element dpdx. As a
result of collision of two molecules, the points that represent them pass from one parabola
into another with the conservation of the total energy.
The notion of the µ-space can be extended for cases of more complicated systems than
a monoatomic gas. Let an ideal gas consists of molecules that have f degrees of freedom
each.1 Then the µ-space will have not 6 but 2f dimensions and its coordinates will be
f generalised coordinates qi and f generalised momenta pi , where i = 1, . . . , f . These
generalised coordinates are chosen in the same way as is done in mechanics, according to
the symmetry of the problem, while the generalised momenta are determined in the canonical
way as pi = ∂L/∂ q̇i , where L is the corresponding Lagrangian. The energy of the gas can
be written as
X
= aik (ql )pi pk + U (qi ) ,
i,k
where the homogeneous quadratic form of generalised momenta is the kinetic energy and
U (qi ) is the potential energy. The hypersurface = const is the 2f − 1-dimensional hyper-
surface of the constant energy. The phase space volume element is
f
Y
dΓ = dV dVp = dpi dqi .
i=1
At equilibrium the points representing states of molecules fill the µ-space with a certain
density ρ, which depends on pi , qi but not on time. This density is defined as
dN
ρ(pi , qi ) = ,
dΓ
where dN is a number of representing points that occur in the volume element dΓ. This
density describes the most probable distribution of points in the phase space compatible
with external conditions and our nearest goal will be to find it. An important axiomatic
statement we rely upon is that ρ depends on the energy only, rather than on individual
pi , qi . Credibility of this statement will be discussed later, when the Liouville theorem will
be proved.
Now we introduce a division of the µ-spaces into boxes and cells. Since the distribution
function depends on energy only, it is meaningful to divide the phase space into boxes
by constant energy hypersurfaces. This should be done with caution. On the one hand,
energetic shells (boxes) should be thin enough in order to be able to assign to all representing
points within one shell the one and the same energy . On the other hand, the size of a
shell i must be big enough in order to encompass a large number Ni of representing points,
1
For instance, in the two-atom gas f = 6, in a three-atom gas has f = 9 degrees of freedom.
80
Ni 1. It is clear that because of these restrictions the boxes can not be all of equal size,
as energetic shells corresponding to large or small energies must be chosen fatter than those
which correspond to moderate energies and, therefore, contain more representing points.
The second element needed for statistical description is equal-size cells. The necessity of
their introduction is dictated by a requirement that the occupation numbers must be defined
with respect to equal-size cells rather than with respect to unequal boxes. This must be the
case if we want to treat all states on equal footing. Thus, we introduce a division of the
phase space on equally sized cells having the phase volume a. The geometric form of cells
plays no role. As to the cell volume a, we have the following situation. The size of boxes
was defined from the requirements Ni 1 and i+1 − i i and, as such, was fixed only
up to the order of magnitude, rather than exactly. Amazingly, the size a of cells appear to
be uniquely fixed by the laws of nature and will be determined later. Note, that the volume
of a cell turns out to be very small and, therefore, each energetic shell will contain not only
a large number Ni of particles but also a large number gi 1 of cells.
81
are indistinguishable and states of a system that differ from each other by permutations of
particles are, in fact, represent the one and the same state. We will proceed by relying on this
hypothesis and later consider the consequences of the hypothesis about distinguishability of
identical particles.
Finally, we point out that in nature there exist two type of particles essentially differ
from each other by their behaviour in ensemble (i.e. collective behaviour) – fermions and
bosons.
Fermions obey the fundamental law
of nature – the Pauli exclusion princi-
Nmax
ple. For our purposes it is convenient to
slightly idealise the situation and for-
mulate the Pauli principle in the fol-
lowing way. In a system of N iden-
tical fermions in any phase space vol- 5
ume Γ only a finite number of repre- 4
senting points that does not exceed a 3
certain integer number Nmax (Γ) can be 2
found. With diminishing of Γ the num- 1
ber Nmax (Γ) diminishes uniformly, as a 2a 3a 4a 5a 6a
in the Fig. 2.1. Therefore, in a phase
volume confined in between na and Figure 2.1: The gross of Nmax with Γ.
(n + 1)a there could be no more than n
points representing fermions. We stress
that this result does not depend on a
geometric shape of a given phase space volume but only on its value.
Let us now take a minimal phase volume per one representing point as the volume of
an elementary cell in µ-space, both for bosons and fermions. As we will see, a is uniquely
determined by the laws of nature and can be found experimentally.3 The Pauli exclusion
principle can be now formulated in the following way: In a system of N identical fermions
in one cell of µ-space there can not be more than one representing point. In opposite, in a
system of N identical bosons in one cell of µ-space any number of representing points can
be found.
In quantum field theory one proves an important theorem – the theorem of Pauli which
relates spin of a particles and the type of statistics it obeys. According to this theorem, all
particles with half-integer spin are fermions (from elementary particles these are electrons,
positrons, protons, neutrons, µ-mesons, hyperons and so on), while all particles with integer
spins are bosons (from elementary particles there are photons, π-mesons, K-mesons and
so on). Concerning composite particles (nuclei, atoms, molecules, ions), because they are
composed of particles of spin 1/2, for them the following rule is valid: particles which
consist of even number of elementary particles are bosons, particles which are made from
odd number of elementary ones are fermions.
3
Due to the Pauli exclusion principle, the value of a is already defined. One can, however, imagine a
world in which only bosons exist and fermions are absent. In this world the value of a can not be a priory
defined, but as we will see, however, the laws of nature dictate the value of a uniquely and it can in any way
be found from experimental data.
82
ab ab ab
a b b a a b
b a a b b a
+ + + + + +
++ ++ ++
+ + + + + +
Figure 2.2: Examples of distribution of distinguishable particles (upper figure), bosons (mid-
dle figure) and fermions (bottom figure).
Note that formulating the Pauli principle, we ignore the existence of spin. As is known, a
state of a particle with spin s is 2s+1-degenerate (the projection of spin on a given direction
can have 2s + 1 values). Thus, one should, in fact, allow no more than 2s + 1 representing
points in one cell of the µ-space. We can, however, do not change the formulation of the
Pauli exclusion principle but change the definition of the µ-space by adding to coordinates
x, y, x, px , py , pz one more discrete coordinate that is a projection of spin on a given direction.
This is the same as to have 2s + 1 exemplars of a cell defined by coordinates x, y, x, px , py , pz
in the old sense.
Let us illustrate the difference in statistical properties of bosons and fermions on a simple
example. Suppose we need to distribute two particles over three cells 1, 2 and 3. If these
particles are distinguishable (we denote them in this case as a and b), then 9 different ways
to distribute these particles, see Fig. 2.2 (up). For bosons, due to their indistinguishability,
6 distributions are possible (indistinguishable bosons are market by crosses), see Fig. 2.2
(middle). For fermions, due to their indistinguishability, and because of the Pauli exclusion
principle only 3 distributions are possible, Fig. 2.2 (bottom).
After these preliminary remarks, we come to the derivation of the distributions and we
start with the case of identical bosons. Consider i-th box in which there are Ni bosons.
Taking gi − 1 walls and setting them in an arbitrary order, we obtain a division of the box
on gi cells. Thus, we have two types of objects: particles and walls. Denote by Wi the
number of ways by means of which we can put Ni particles in gi cells. Because particles
are indistinguishable, these ways can only be different by a number of particles in cells
provided the number of particles Ni in the box is fixed. These ways are obtained from one
another by transport particles from a cell into another cell (an interchange of positions of
83
1 2 gi
particles and walls). Fixing each such distribution, we make all unessential permutations, i.e.
permutations which do not lead to new distributions. These unessential permutations are
all permutations of particles between themselves (their number is Ni !) and permutations of
walls between themselves (their number is (gi − 1)!). In this way we obtain all permutations
of a collective of Ni + gi − 1 objects, which are particles and walls together. Thus,
from where
(Ni + gi − 1)!
Wi = .
Ni !(gi − 1)!
If we have a number of boxes, then the number of ways W to distribute particles over boxes
such that in the first box there would be N1 particles, in the second N2 and so on, is equal
to the product of Wi . In this way we obtain the main formula of boson statistics
Y (Ni + gi − 1)!
W = . (2.1)
Ni !(gi − 1)!
i
To find the most probable distribution, we need to find the maximum of W . It is, however,
more convenient to find the maximal value of ln W , which is, of course, equivalent to finding
the maximum of W . For Ni 1 and gi 1 we can use the Stirling approximation discussed
in appendix III
X
ln W = ln(Ni + gi − 1)! − ln Ni ! − ln(gi − 1)!
i
X
≈ (Ni + gi ) ln(Ni + gi ) − Ni ln Ni − gi ln gi . (2.2)
i
which mean that the total number of particles and the total energy of the gas are fixed.
Using the method of Lagrange multipliers, we introduce a function
X X X
L= (Ni + gi ) ln(Ni + gi ) − Ni ln Ni − gi ln gi + α Ni − N − β Ni i − E
i i i
84
and find its extremum with respect to all Ni . We have
∂L
= ln(Ni + gi ) − ln Ni + α − βi = 0 .
∂Ni
From here we find the most probable particle numbers Ni
gi
Ni = −α+β
. (2.4)
e i −1
This distribution has been derived in 1924 by Einstein and has the name Bose-Einstein
distribution (Bose derived it for a particular case of photons). The parameters α and β are
determined from equations (2.3), which upon substituting (2.4), take the form
X gi
=N,
e−α+βi − 1
i
X gi i (2.5)
= E .
e−α+βi − 1
i
Ni !(gi − Ni )!Wi = gi ! ,
85
1 2 gi
The extremity conditions of this function with respect to the numbers Ni read as
∂L
= − ln Ni + ln(gi − Ni ) + α − βi = 0 ,
∂Ni
from which we find
gi
Ni = −α+β
. (2.8)
e i +1
This is the Fermi-Dirac distribution and it is different from the Bose-Einstein distribution
(2.4) only by the sign in from of 1 in the denominator. The parameters α and β are found
from the conditions
X gi
−α+β
=N,
e i + 1
i
X gi i (2.9)
= E .
e−α+βi + 1
i
86
identify these quantities as
S = ln W . (2.10)
This formula expresses the Boltzmann principle. According to (2.10), entropy is a dimen-
sionless quantity. Because of the formula ∆Q = θdS, such a choice of identification implies
that the temperature is measured in energetic units.4
Entropy of a system at equilibrium is
Since W < Wmax , one has S < Smax . To safe the notation, in the following we will be
denoting the equilibrium entropy by the same letter S. Thus, according to Boltzmann,
increase of entropy in the process of striving towards an equilibrium state is a consequence
of the transition of a system from less to more probable states.
It should be noted that there is a principle difference between the statistical and thermo-
dynamic treatment of an irreversible process. From the point of view of a phenomenological
thermodynamics a process which is reverse of an irreversible process is impossible by def-
inition. From the point of view of statistical physics, a process which is reverse to an
irreversible one in the thermodynamic sense, is nothing else but a transition from a more
probable state to a less probable one. In a macroscopic system sensible fluctuations happen
very rare and their probability is vanishingly small. In systems with not a large number of
particles probability of large fluctuations becomes noticeable and they happen pretty often.
Let us now show how the Boltzmann principle can be used to identify the physical
meaning of the parameters α and β. We need the formulae,
X gi X gi i
−α+β
=N, −α+β
=E. (2.12)
e i ∓1 e i ∓ 1
i i
Here the upper sign is for the Bose-Einstein distribution and the lower sign for the Fermi-
Dirac one. We now relate the entropy with the number N of particles and the internal
energy E .
All cells belonging to the one and the same box are equivalent. This can be seen, in
particular, from the fact for both, the Bose-Einstein and Fermi-Dirac, distributions the
number Ni of particles in the i-th box is proportional to the number of cells gi . Therefore, it
seems reasonable to use the occupation numbers with respect to one cell: ni = Ni /gi . These
occupation numbers are
1
ni = . (2.13)
e−α+βi ∓ 1
P
Using these numbers ni any sums over boxes can be brought to the form i gi F (i), where
F (i) is a function of a box label. With this way of writing, the equivalence of cells becomes
obvious.
4
Measuring temperature T in absolute units (grads of Kelvin) implies through the formula δQ = T dS
that entropy has the physical dimension of J/K. Then the relation between statistical quantity ln W and the
thermodynamic entropy S is given by S = k ln W , where k is the Boltzmann constant.
87
Let us now find entropy of an equilibrium state first nor a system of bosons. Using the
occupation numbers ni and formula (2.2), we write
X
S = (Ni + gi ) ln(Ni + gi ) − Ni ln Ni − gi ln gi
i
X
= gi (1 + ni ) ln(1 + ni ) − ni ln ni . (2.14)
i
The maximal value of entropy is obtained by substituting here ni for the Bose-Einstein
distribution
X −α + βi
α−βi
S= gi −α+β − ln(1 − e ) .
e i − 1
i
Analogously, for a system of fermions with the help of (2.17) one finds
X
S = gi ln gi − Ni ln Ni − (gi − Ni ) ln(gi − Ni )
i
X
= − gi ni ln ni + (1 − ni ) ln(1 − ni ) . (2.16)
i
To get an equilibrium distribution we substitute here the occupation numbers ni from the
Fermi-Dirac distribution and get
X
S = −αN + βE + gi ln(1 + eα−βi ) . (2.17)
i
Combining (2.15) and (2.17), we can write one formula that comprises the answer for the
equilibrium entropy for both distributions
Equilibrium
X entropy
S = −αN + βE ∓ gi ln(1 ∓ eα−βi ) , (2.18)
i
where the upper and lower signs are for the Bose-Einstein and Fermi-Dirac distributions,
respectively.
It is worth to stress the difference between the formulae (2.14) and (2.16) from the one
hand, and (2.15) and (2.17) from the other. Formulae (2.14) and (2.16), combined as
Entropy of an
X arbitrary state
S[ni ] = − gi ni ln ni ∓ (1 ± ni ) ln(1 ± ni ) , (2.19)
i
give entropy of a gas in an arbitrary state, which can be either non-equilibrium or equilibrium,
while formulae (2.14) and (2.16) are valid only for an equilibrium state with the maximal
entropy and most probable occupation numbers ni . We see that entropy (2.19) of a non-
equilibrium state is a function of an infinite set of occupation numbers ni that are related
88
P P
by the conditions i gi ni = N and i gi ni i = E , and arbitrary otherwise. In opposite,
the entropy of an equilibrium state is fully determined by two parameters α and β, which,
therefore, become of fundamental importance in equilibrium thermodynamics.
To find the physical meaning of α and β, let us commute the differential dS by using
(2.18) and compare it with the expression for dS in thermodynamics. For this, we first
differentiate (2.18) over α keeping β = const and then over β keeping α = const, considering
every time that i = const and gi = const.
If external field are absent (electric, magnetic, gravitational), then energy levels i and
degeneracies gi depend only of the volume V of a container that contains gas. If external
fields are present, then energy levels do also depend on other parameters than V , more
precisely, on the field strengths of the corresponding fields. Let us consider the case when
all external fields except the walls of a contained are absent. Then the conditions i = const
and gi = const are equivalent to the condition V = const. We have
∂S ∂N ∂E X gi i
= −α +E +β − −α+β
.
∂β α,V ∂β α,V ∂β α,V e i ∓1
i
Analogously,
∂S ∂N ∂E X gi
= −N − α +β + −α+β
∂α β,V ∂α β,V ∂α β,V e i ∓1
i
∂N ∂E
= −α +β , (2.21)
∂α β,V ∂α β,V
where the first formula in (2.12) was used. Multiplying (2.20) by dβ and (2.21) by dα and
adding up, we find
∂E ∂E
∂N ∂N
dSV = −α +β dβ + −α +β dα
∂β α,V ∂β α,V ∂α β,V ∂α β,V
= −αdN + βdE .
89
Substituting these identifications into (2.18), we get
X
θS = E − µN ∓ θ gi ln(1 ∓ e(µ−i )/θ ) , (2.23)
i
E = θS − pV + µN ,
The right hand side of this expression is always positive because for the upper sign (Bose-
Einstein statistics), the expression under the logarithm is negative.
Thus, the Bose-Einstein and Fermi-Dirac distributions can be written in the form
gi
Ni = ( −µ)/θ
, (2.25)
e i ∓1
while the additional conditions become
X gi X gi i
( −µ)/θ
=N, ( −µ)/θ
=E. (2.26)
i
e i ∓1 i
e i ∓1
Note that after the physical meaning of the parameters α and β has been established,
the physical interpretation of the relations (2.26) must be changed. The first equation still
Change of inter-
defines µ as a function of the number N of particles and temperature θ. The parameter pretation of N
β = 1/θ does not need any additional definition. Therefore, the second equation defines and E
the internal energy as a function of µ and θ, while from the first relation we can find
µ = µ(θ, N ). Thus, the second equation with the help of the first defines the internal energy
at the function of θ and N . In particular, if a system is homogeneous and external fields
are absent, the chemical potential µ and the specific energy ε = E /N are, as any intensive
parameters, functions of the density N/V and temperature θ.
The following important comment is in order. The derivation of the Bose-Einstein and
Fermi-Dirac distributions by the method of boxes and cells assumes that in the process of
equilibrating (thermalisation) particles can change their energy passing from one box into
another, otherwise, any non-equilibrium distribution in the µ-space would stay unchanged
and would not relax to equilibrium, so that maximisation of ln W would not have any sense.
Obviously, the possibility of passing from one box into another is due to interactions of
Distributions
particles with surroundings, because particles do not interact with each other (they are free). derived are for
This surrounding must be a thermostat with temperature θ = const and walls impermeable an ideal gas in
a thermostat
to particles N = const. This follows from the fact that deriving statistical distributions,
we regarded the total number N of particles and the total energy E as fixed, the latter for
fixed N depends for an ideal gas only on temperature. Therefore, the Bose-Einstein and
Fermi-Dirac distributions5 represent the most probable distributions of ideal gas particles in
the µ-space under the condition (β) (a gas in a thermostat).
5
Also the Maxwell-Boltzmann distribution which will be obtained in the next section.
90
2.4 Maxwell-Boltzmann distribution
Let us consider now the case of a distinguishable particles. As before, we have a system of
boxes with energies i and a number of cells gi and want to distribute Ni particles in the i-th
box. First, let us find a number of ways by means of which we can distribute N particles
over boxes without fixing cells in which they descend. Since the particles are distinguishable,
different ways of such distributings differ from each other by permutations of particles that
belong to different boxes. Denote the number of these distributions by W 0 . Fixing one
such distribution, we perform all unessential permutations of particles in each of the boxes,
therefore, obtaining all possible permutations
Y N!
W0 Ni ! = N ! , W0 = Q .
i i Ni !
Now we find the number of ways by means of which one can distribute N1 distinguishable
particles over g1 cells, N2 particles over g2 cells, and so on. Since for each of the Ni particles
there are gi possibilities of distributing over cells, this number is gi · gi . . . · gi = giNi and for
Q
all boxes i giNi . Therefore, the total number of distributing of N particles is
Y Y g Ni
W = W0 giNi = N ! i
(2.27)
Ni !
i i
we obtain an equation
∂L
= ln gi − ln Ni − 1 + (α + 1) − β = 0 ,
∂Ni
which solution for Ni gives the Maxwell-Boltzmann distribution
Ni = gi eα−βi . (2.30)
91
One can see that the Maxwell-Boltzmann distribution can be obtained as the limiting
case of the Bose-Einstein and Fermi-Dirac distributions
gi
Ni = ( −µ)/θ
. (2.32)
e i ∓1
Indeed, if for any i one has
then 1 in the denominator of (2.32) can be dropped and one obtains the Maxwell-Boltzmann dis-
tribution
Ni
Ni = gi e(µ−i )/θ , ni = = e(µ−i )/θ . (2.34)
gi
This result can be anticipated from simple physical agreement, namely, inequality (2.33) is
equivalent to ni = Ni /gi 1, which is nothing else but the condition of a dilute gas. In a
Maxwell-
dilute gas, on average particles are far from each other and in this situation the difference Boltzmann
between distinguishable or indistinguishable particles becomes negligible – particles do not distribution
corresponds to
practically approach each other and one can not mix them up. Therefore, it is natural that the dilute gas
under such conditions the statistical distributions for indistinguishable particles coincide approximation
with the Maxwell-Boltzmann distribution for distinguishable particles.
An important warning is that the Bose-Einstein and Fermi-Dirac distributions are always
true for any kind of non-interacting particles, while the Maxwell-Boltzmann distribution is
valid only in the limiting case of small occupation numbers.
An expression for the entropy of the Maxwell-Boltzmann gas, both for arbitrary and for
equilibrium states can be obtained in two different ways.
92
Indeed, mixing r portions of the one and the same gas with one and the same tem-
perature and density, to compute the entropy of the admixture we need to replace Ni
for rNi , gi for rgi and N on rN . Then the entropy giving by (2.35) will increase by a
factor of r and the total entropy of r portions will stay the same. As to the entropy
S ∗ defined by (2.36), it will acquire except the multiplier r also and additional term
N r ln r and, therefore, the total entropy of the admixture will increase by N r ln r.
This phenomenon is called the Gibbs paradox. This paradox was already discussed in
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
the context of entropy change in the process of irreversible mixing of two gases.
::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::: Gibbs paradox
This drawback of classical theory by Boltzmann, was corrected by Gibbs, who sug-
gested to consider states of a gas in the µ-space that are differ from each other by
permutations of the points as the one and the same state, as molecules are physically
identical. One needs, therefore, to divide the number of possible states W by the
number of permutations of N molecules, i.e. by N !. As a consequence, one should
subtract from S ∗ the term ln N !, so that, according to the Stirling formula, the result-
ing entropy should coincide with S. In doing so, the Maxwell-Boltzmann distribution
will not change, because we changed ln W by a constant N ln N − N . Gibbs, therefore,
anticipated the concept of indistinguishable particles already at the end of the XIX-th
century. From the logical point of view, however, the trick proposed by Gibbs to re-
solve the Gibbs paradox cannot be considered as self-consistent. Indeed, first, in the
derivation of the Maxwell-Boltzmann distribution we assumed that particles are dis-
tinguishable and only at the final result one introduces the correction that takes care
of identical states that differ from each other by permutations of identical particles.
The logical derivation is based on the hypothesis on non-distinguishability of identical
particles and leads to the Bose-Einstein and Fermi-Dirac distributions. The Maxwell-
Boltzmann distribution is obtained from these ones in the limit of small occupation
numbers.
1. Bose-Einstein distribution
gi X gi X gi i
Ni = ( −µ)/θ
, N= ( −µ)/θ
, E = ( −µ)/θ
. (2.37)
e i −1 i
e i −1 i
e i −1
93
2. Fermi-Dirac distribution
gi X gi X gi i
Ni = ( −µ)/θ , N= , E = . (2.38)
e i +1 i
e(i −µ)/θ +1 i
e(i −µ)/θ +1
3. Maxwell-Boltzmann distribution
X X
Ni = gi e(µ−i )/θ , N= gi e(µ−i )/θ , E = gi i e(µ−i )/θ . (2.39)
i i
1. Bose-Einstein distribution
Z Z
g dΓ g dΓ g dΓ
dN = , N = , E = . (2.41)
a e(−µ)/θ − 1 a e(−µ)/θ − 1 a e(−µ)/θ − 1
94
2. Fermi-Dirac distribution
Z Z
g dΓ g dΓ g dΓ
dN = , N= , E = . (2.42)
a e(−µ)/θ + 1 a e(−µ)/θ +1 a e(−µ)/θ +1
3. Maxwell-Boltzmann distribution
Z Z
g (µ−)/θ g g
dN = e dΓ , N= e(µ−)/θ dΓ , E = e(µ−)/θ dΓ . (2.43)
a a a
E =N Z ,
e−/θ dΓ
95
where we have used that p = (2m)1/2 .
The condition of validity of the Maxwell-Boltzmann distribution is e(µ−)/θ 1 for any
, in particular, for = 0. In the latter case, we have
eµ/θ 1 .
Therefore, the Maxwell-Boltzmann distribution applies better if the density N/V is lower,
while mass m of molecules is bigger and temperature θ is higher. In statistical physics one
uses the notion of the thermal wave-length λ given by
1/2
h2
λ= . (2.45)
2πmθ
With this notion, the criterion of applicability of the Maxwell-Boltzmann distribution be-
comes
λ3 λ3
N = 1, (2.46)
V v
where v is a specific volume. In other words,
Degeneration
r criterion –
3 V dilute gas
λ , (2.47)
N
that is the thermal wave-length must be much smaller than the length of free path between
collisions.
96
An opposite criterion
3/2
N h2
≥1 (2.48)
V 2πmθ
is the condition which means that the Maxwell-Boltzmann distribution is not applicable and
the corresponding gas obeys the Bose-Einstein or Fermi-Dirac distribution. In this case a
gas is called degenerate and criterion (2.48) has a name of criterion of degeneration. The
temperature
2/3
h2 N
θdeg = ,
2πm V
3/2
h2
defined from the condition NV 2πmθdeg = 1, is called the temperature of statistical de-
generation of a mono-atomic gas.6 For θ < θdeg a gas is degenerate.
Let us first consider criterion (2.44) for a gas that consists of ordinary atoms. Substi-
tuting the numerical values7
N/V ' 1025 m−3 , m ' 10−26 − 10−27 kg , h ' 6.6 · 10−34 m2 kg/s ,
we obtain T 0.1K. Therefore, for usual gases under pressures close to the normal one, the
Maxwell-Boltzmann distribution provides a good approximation up to rather low tempera-
tures.
Electron gas
The situation becomes different for light particles, for instance for electrons in metal. is degenerate
(quantum)
For electrons m ' 10−30 kg and, as a result, up to temperatures ∼ (104 −105 )K the Maxwell- at room temperature
Boltzmann distribution is not applicable and, therefore, electron gas is degenerate at room
temperature!
97
and
∂Ω ∂Ω ∂Ω
S=− , p=− , N =− . (2.50)
∂θ V µ ∂V θµ ∂µ θV
For a specific case of ideal bose and fermi gases, we find from (2.24)
X
Ω = −pV = ±θ gi ln(1 ∓ e(µ−i )/θ ) ,
i
where the upper sign for bosons and lower for fermions. Passing to integration, we obtain
Z
dΓ
Ω = ±gθ ln(1 ∓ e(µ−)/θ ) , (2.51)
a
Let us now show that for a gas in the absence of external fields Ω-potential is related to
the internal energy in a simple way. Assume that the dispersion relation has the form8
= κps (2.52)
where κ and s are constants. Such a form of the dispersion relation encompasses both non-
relativistic particles ( = p2 /2m, s = 2) and relativistic ones ( = cp, s = 1). Integrating
over volume, we get
Z ∞
4πgθ
Ω=± V dp p2 ln(1 ∓ e(µ−(p))/θ ) ,
a 0
Integrating by parts and taking into account that boundary terms vanish, we obtain
Z ∞ p3 ∂(p) Z ∞
4πg ∂p 4πgs p3 κps−1
Ω = − V dp ((p)−µ)/θ =− V dp ((p)−µ)/θ
3a 0 e ∓1 3a 0 e ∓1
Z ∞ 2 s Z ∞ 2
4πgs p κp s g 4πp (p)
= − V dp ((p)−µ)/θ =− · V dp ((p)−µ)/θ .
3a 0 e ∓1 3 a 0 e ∓1
It remains to notice that
Z ∞
g 4πp2 (p)
E = V dp (2.53)
a 0 e((p)−µ)/θ ∓ 1
is an expression for the internal energy. Thus, we obtain that
s
Ω=− E. (2.54)
3
In particular, for non-relativistic particles
2
Ω=− E, (2.55)
3
while for ultra-relativistic particles (for instance, photons)
1
Ω=− E. (2.56)
3
Note that formulae (2.54)-(2.56) together with (2.53) define the Ω-potential as the function
of its natural variables θ, V, µ. therefore, entropy, pressure and the number of particles can
be found from equations (2.50).
8
Here p is particle momentum, do not confuse with pressure!
98
while for ultra-relativistic particles (for instance, photons)
while for ultra-relativistic particles (for instance, photons)
1
⌦= E. (2.52)
1
3 ⌦= E.
3
Note that formulae (2.50)-(2.52) together with (2.49) define the ⌦-potential as the function
Notepressure
of its natural variables ✓, V, µ. therefore, entropy, that formulae (2.50)-(2.52)
and the number oftogether
particleswith
can (2.49) define the
of its natural variables ✓, V, µ. therefore, entropy, pressure and
2.7 Quantisation of energy and Nernst theorem be found from equations (2.46).
be found from equations (2.46).
while for ultra-relativistic particles (for instance, photons)
QM
classical
classical theory has two aspects, illustrated in theory has two aspects, illustrated in Fig. (??).
Fig. (??).
but rather gi different quantum- Indistinguishable particles,
Distinguishable particles,
mechanical states, each described discrete spectrum of energies discrete spectrum of energies
by its own wave function.
Concerning these two quantum-
mechanical features, transition
from quantum to classical theory
has two aspects, as illustrated in Figure 2.6: Transition from quantum to classical theory.
Fig. 2.6.
72
1. If occupation numbers are small, we can approximately treat72identical particles as
distinguishable and pass from formulae of Bose-Einstein and Fermi-Dirac distributions to
formulae of the Maxwell-Boltzmann statistics. The possibility for this passage is controlled
by the criterion of degeneration discussed above.
2. We can neglect discreteness of the energy spectrum of real systems and consider en-
ergy as a continuous variable. It is then necessary to find out when such an approximation 72
72 number n, where
is possible. Let us assume that energy of a particle depends on a quantum
the dependence (n) and the meaning of n depends on a specific problem. For instance, for
translational, rotatory and oscillatory motion both the physical meaning of n and the de-
pendence (n) are different. Obviously, quantisation of energy can be neglected, if distances
between neighbouring energy levels are small in comparison with the energy itself.
For large quantum number, practically in all most interesting cases (n) is a homogeneous
power law function of the number n: (n) = Anα . Then
(n)
∆(n) = A[(n + 1)α − nα ] ≈ Aαnα−1 = α .
n
Therefore, the condition ∆/ 1 leads to the inequality n 1. Therefore, high levels
corresponding to large n are always dense and their energy can be considered as continuously
99
changing function. Since at high temperatures most of particles occupy high energy levels,
neglecting quantisation of energy is justifiable. However, at low temperatures a sufficient
portion of particles occupies lower levels n & 1, which are separated from each other by big
intervals comparable to the value of energy, quantisation of energy becomes essential. The
exact meaning of the terms “high" and “low" temperatures can be established as follows.
Most of the particles have energies ∼ θ. From here we find values of n essential for at this
temperature
Then the condition neff 1 gives a characteristic temperature θ0 , above which quantisation
of energy is not essential: θ0 ∼ A. Note that criterion θ ≥ θ0 ∼ A is in general different
from the condition of non-degeneracy
3/2
N h2
1,
V 2πmθ
although both conditions are obeyed at sufficiently high temperatures. Thus, a gas with
discrete energy levels and a gas with continuously changing energy of particles can be either
degenerate or non-degenerate.
Note that the derivation of the Bose-Einstein and Fermi-Dirac distributions and, as a
consequence, of the Maxwell-Boltzmann distribution as their limiting case, can be repeated
without any changes and lead to the same results, although the notions of “boxes" and “cells"
will acquire a different meaning.
A“box" should be now understood as an energy level, i.e. all the states of a particle with
a given energy, while a “cell" will be now a distinct state with a given energy. If a level is
non-degenerate (that is to a given value of energy corresponds just one state), then the cell
coincides with the corresponding box. If there is degeneration, then to a given energy level
there correspond a certain non-trivial number of cells. In quantum mechanics, the ground
state, i.e. a state with the lowest possible energy, is typically non-degenerate. Note that in a
theory that takes into account quantisation of energy, numbers gi should not necessarily sat-
isfy the condition gi 1, necessary for application of the Stirling approximation. Therefore,
the method of boxes and cells used to derive the Bose-Einstein and Fermi-Dirac distributions
becomes inappropriate. However, the distributions themselves still remain correct, as can
be derived by proper alternative means.
Let us now prove an important theorem, namely, the Nernst theorem that has been al-
ready mentioned in phenomenological thermodynamics. According to this theorem, entropy
of a gas at θ = 0 is a constant that does not depend on any variable parameters (volume,
pressure, etc.). As it will be clear from the proof, in many cases this constant is actually
zero and, therefore, the Nernst theorem is formulated as
S(θ = 0) = 0 ,
100
such non-existing in nature particles this theorem would still hold. Since, at temperatures
low enough a gas is always degenerate, this observation does not have any practical meaning.
Let us consider the proof. We start with the Boltzmann formula for the entropy of an
equilibrium state S = ln Wmax . We write the exact formulae9 for the number of distributions
of particles over energy levels i with degeneracies gi for fermions, bosons and classical
(distinguishable) particles:
Y gi ! Y (Ni + gi − 1)! Y g Ni
i
Wf = , Wb = , Wcl = N ! . (2.58)
Ni !(gi − Ni )! Ni !(gi − 1)! Ni !
i i i
At absolute zero energy of a gas must be minimal, therefore, in a system of bosons and also
in a system of classical particles all particles occupy the lowest energy level Ni = 0, i 6=
1, N1 = N . In a system of fermions particles occupy the n lowest energy levels
P according to
the Pauli exclusion principle Ni = gi for i < n, Ni = 0 for i > n, where ni=1 Ni = N ; the
level i = n can be either fully occupied or partially, 0 < Nn ≤ gi . Since the factors in (2.58)
related with the i-the level turn to 1 for Ni = 0, and Ni = gi for fermions, we obtain in the
limit θ → 0
gn ! (N + g1 − 1)!
Wf = , Wb = , Wcl = g1N .
Nn !(gn − Nn )! N !(g1 − 1)!
Suppose in the case of bosons and classical particles the ground level is not degenerate g1 = 1
and in the case of fermions the level i = n is non-degenerate, i.e. gn = 1. Then in all three
cases W = 1. The physical meaning of this result is obvious: under these conditions the
state of a gas is uniquely defined because none of permutations of particles within one cell
do not lead to a new state of the gas, even for distinguishable particles. Therefore, according
to the Boltzmann formula we get S = 0.
If the degeneracies g1 and gn do not equal to one (for instance, for particles with spin
s degeneracy of any state is 2s + 1), then although entropy is not equal to zero at θ = 0,
rather it is a constant. For instance, for bosons using the Stirling approximation we have
(N + g1 − 1)!
S(θ = 0) = ln = (g1 − 1) ln N
N !(g1 − 1)!
S(θ = 0) = N ln g1 .
All consequences of the Nernst theorem are also valid for this case, as they exclusively based
on the fact that the entropy at absolute zero is independent on pressure, volume and other
variable parameters.
Finally, we note that the Nernst theorem is also valid for interacting particles as can be
derived in the framework of Gibbs’ method.
9
No Stirling approximation.
101
Chapter 3
Maxwell-Boltzmann gas
Substituting this expression into formula (3.1) for the energy, we get
3
E = Nθ .
2
102
Finally, we find the thermal equation of state for an ideal Maxwell-Boltzmann gas
2
pV = −Ω = E = N θ .
3
Further, for the chemical potential formula (3.2) gives
Na
µ = θ ln , (3.3)
gV (2πmθ)3/2
This is the Sacur-Tetrode formula (1911-1913). It fully determines the entropy constant s0 ,
the knowledge of which is necessary for considering problems of chemical thermodynamics.
This formula was confirmed experimentally, but its theoretic justification in the pre-quantum
era caused significant difficulties. The Sacur-Tetrode formula can also be written in the form1
" #
5 g(2πm) 3/2 V
S = N + N ln(θ3/2 v) + N ln 3
, v= . (3.6)
2 h N
This formula agrees with the thermodynamic entropy (2.28) for CV = 3/2. The additive
constant in (2.28) is essentially depends on the volume a = h3 of elementary cell. Note that
the formula for the entropy, as well as other formulas for thermodynamic potentials obtained
in this section are not valid at low temperatures. It is, therefore, not surprising that S →
−∞ as θ → 0 is in contradiction with the Nernst theorem. In general, thermodynamic
functions of a gas at low temperatures should be computed by taking into account degeneracy
of the gas, its non-idealness and quantisation of energy.
103
dΓ of the µ-space
dN e−/θ dΓ
dW = =Z , (3.7)
N −/θ
e dΓ
We consider a gas consisting of molecules with an arbitrary number n of atoms. Since here
we are interested in translational motion only, we single out from all generalised coordinates
only 3 coordinates of the centre of mass x, y, z and the corresponding projections px , py , pz
of its momentum, and represent the energy of the molecule in the form
p2
= + U (x, y, z) + 0 ,
2m
where p2 = p2x +p2y +p2z and U (x, y, z) is a potential energy of a molecule in an external field,
and 0 is the energy of remaining degrees of freedom, which does not depend on coordinates
and momenta that we singled out. Integrating (3.7) over all degrees of freedom except
translational, we find the probability that the centre of mass of a molecule lies in the volume
dV , while its momentum is inside the interval (p, p + dp). This probability splits into two
factors, one solely depends on momenta and the other on coordinates
2 /2mθ
e−p dVp e−U (x,y,z)/θ dV
dW = Z Z .
−p2 /2mθ −U (x,y,z)/θ
e dVp e dV
104
Passing from momenta to velocities, the last formula yields the distribution of molecules
over velocities
m 3/2 mv2
dW (v) = 4π e− 2θ v 2 dv . (3.9)
2πθ
Formula (3.8) and (3.9) are known as Maxwell distribution.
Let us find the most probable velocity vp of molecules. For this velocity the expression
(3.9) must have the maximum
d − mv2 2 mv 2
mv 2
(e 2θ v ) = ve− 2θ 2 − = 0,
dv θ
from where
r
2θ
vp = .
m
We can also find the mean velocity and the mean velocity squired
Z m 3/2 Z ∞ mv2 r
8θ
v = vdW (v) = 4π e− 2θ v 3 dv = ,
2πθ 0 πm
Z Z
m 3/2 ∞ mv2 3θ
v2 = v 2 dW (v) = 4π e− 2θ v 4 dv = .
2πθ 0 m
mv 2 3
Ek = = θ.
2 2
The Maxwell distribution has a profile shown on Fig. 3.1. With temperature increasing
the maximum shifts in the direction of higher temperature and the graphs itself becomes
less steep with a bigger tail extending in the region of large velocities.
If we divide the momentum space not by spherical shells but by parallelepipeds dvx dvy dvz ,
then the Maxwell distribution can be written as the product of three factors of the form
m 1/2 2
dW (vi ) = e−mvi /(2θ) dvi , i = x, y, z .
2πθ
This shows that distributions of projections of the velocity are statistically independent.
One can see that the mean value of the projections vx , vy , vz equals zero.
105
compared to characteristic lengths in the system. Here m is mass and v is velocity of a
particle. In a classical Maxwell-Boltzmann gas particle velocities are distributed according
to the Maxwell distribution and we can compute the average value of λ
Z
h 1 h m 3/2 1 − mv2 2 h
λ= = 4π e 2θ v dv = 2 √ . (3.10)
m v m 2πθ v 2πmθ
Up to the unessential coefficient 2 this expression coincides with the thermal wave length
introduced in (2.45).
and energy
Z
e−β dΓ
∂ ∂
E =N Z = −N ln Z = N θ2 ln Z ,
∂β ∂θ
e−β dΓ
remain valid, but in the volume element dΓ = dV dVp in the µ-space, dV and dVp should be
now understood as products of differential of all generalised coordinates and all generalised
momenta, respectively.
Energy of a molecule in the absence of external fields is equal to the sum of its kinetic
energy which, as in known from mechanics, is a homogeneous quadratic function of mo-
menta aij pi pj (coefficients aik generally depend on generalised coordinates qr ) and potential
interaction energy of atoms constituting the molecule. After singling out translational and
rotatory motion of a molecule as the whole, internal motion of atoms in a molecule rep-
resents small oscillations around an equilibrium position which corresponds to the energy
minimum. Usually, Umin is identified as zero, and the equilibrium position corresponds to
qr = 0. For an n-atomic molecule the number of these internal degrees of freedom is 3n − 5
if a molecule is linear (equilibrium positions of atoms are on the one and the same line),
and 3n − 6 if a molecule is non-linear. Indeed, in the case of a linear molecule its position
is totally specified by three coordinates x̄, ȳ, z̄ of its center of inertia and by two angles. In
the case of a non-linear molecule its orientation in space is specified by three (Euler) angles.
Therefore, for the potential energy one has an expression brs qr qs , where brs are constant
coefficients. Thus,
where for small deviations from an equilibrium position the coefficients aij can be regarded
as constant. Note that in the above expression summation over i, j goes from 1 to 3n, and
106
over r, s from 1 to 3n − 5 for linear molecules and to 3n − 6 for non-linear ones. From here
we have
Z Y Z Y
−aij pi pj /θ
Z = BV e dpl e−brs qr qs /θ dqn .
l n
The coefficient V (volume) arises as a result of integrating over coordinates of the center
of inertia, while a constant dimensionless coefficient B as a result of integration over angles
specifying the orientation of a molecule in space.
The temperature dependence of this integral can be made explicit if we make a rescaling
pi → pi θ1/2 , qr → qr θ1/2 , so that the integral transforms into
Z Y Z Y
−aij pi pj
Z = BV θ 3n/2+3n/2−5(6)/2
e dpl e−brs qr qs dqn .
l n
Thus,
` = |{z} 2 +2 (3n − 5) = 6n − 5
3 + |{z} ← linear molecule ,
| {z }
transl rot oscil
` = |{z} 3 +2 (3n − 6) = 6n − 6
3 + |{z} ← non − linear molecule .
| {z }
transl rot oscil
∂ ` ` `
E = N θ2 (ln θ`/2 ) = N θ , CV = N , cV = .
∂θ 2 2 2
This results can be formulated as the law of equipartition of energy over degrees of freedom.
Namely, each translational and rotational degree of freedom contributes into internal en-
ergy the amount 1/2N θ, while each oscillatory degree of freedom contributes twice bigger
quantity, i.e. N θ.
Therefore, from the point of view of classical physics, all modes of motions in a molecule
are accounted on equal footing. Twice bigger contribution of oscillatory degrees of freedom is
explained by the fact that oscillations are related to the presence of potential energy which on
average equals to the kinetic energy of oscillations, while translational and rotational motions
are related to kinetic energy only. In particular, for mono-atomic gases classical theory
predicts CV = 3/2N , Cp = 5/2N , γ = 5/3. At first sight it seems that experiment confirms
these predictions: the measured heat capacities of these gases are indeed close to 3/2N .
This agreement is not exact, however, because atoms are not material points with just three
degrees of freedom but they consist of a nucleus made of nucleons and the electronic shell. So,
the actual number of degrees of freedom is 3(Z +A), where Z is the position of an atom in the
periodic table and A is the number of nucleons. Nevertheless, the measured heat capacity is
close to 3/2N θ. This shows that in contradiction with classical physics electron and nucleus
107
degrees of freedom do not contribute to heat capacity, they are “frozen". Analogous situation
takes place for multi-atomic gases. For instance, for diatomic gases, ignoring electron an
nucleus degrees of freedom, the law of equipartition of energy predicts CV for all gases the
same value of CV = 7/2N , from where Cp = 9/2 and γ = 9/7. Experiment shows, however,
that for all diatomic gases at moderate temperatures CV = 5/2N , from where Cp = 7/2 and
γ = 7/5. With temperature decreasing, CV also decreases and for H2 and D2 its reaches
3/2N . For other gases this value cannot be reached because they liquify before. In opposite,
with temperature increasing, heat capacity increases, however, the theoretical prediction
CV = 7/2N cannot be reached because molecules dissociate into atoms before.
The reason for this disagreement between experiment and predictions of classical theory
::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
lies in quantisation of energy and if the latter is taken into account, this disagreement
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
disappears.
::::::::::::
Introducing the statistical sum (Zustandssumme), which is also called single particle partition
function,
X
Z= gi e−i /θ ,
i
Ω = −θeµ/θ Z .
In the formulae above we assume that the summation goes over quantum energy levels, so
that gi means degeneracy of the corresponding energy level. The rest of consideration is to
find the energy levels which is purely quantum-mechanical problem.
108
Quantisation of translational motion
To quantise translational motion, we assume for simplicity that the vessel has a form of a
cube with edge L. The energy levels are then found by solving the Schödinger equation for
a free particle in a box and they are given by
h2
(n1 , n2 , n3 ) = (n2 + n22 + n23 )
8mL2 1
and the corresponding wave function is
n πx n πy n πz
1 2 3
ψ(n1 , n2 , n3 ) = C(n1 , n2 , n3 ) sin sin sin ,
L L L
where n1 , n2 , n3 are integers taking values 1, 2, 3, . . . The value ni = 0 is not allowed as in
this case the wave function is identically zero, also negative ni should not be considered as
independent because under the change of the sign of ni the wave function simply multiplies
by −1. Each set of admissible numbers n1 , n2 , n3 corresponds to one sate of a particle,
therefore, in the statistical sum, we should sum over all n1 , n2 , n3 , each of these numbers
runs from 1 to infinity, and take degeneracy g(n1 , n2 , n3 ) = 1. The energy levels (n1 , n2 , n3 )
are degenerate with degeneracy equal to the number of integer solutions of the equation
n2 = n21 + n22 + n23 , but this degeneracy is automatically accounted because we are summing
not over the energy levels but over n1 , n2 , n3 , i.e. over the states. For the partition function
we have
∞ X ∞ X ∞
( ) ∞
!3
X h2 X
2 2 2 −π(θt /θ)n2
Z= exp − (n + n2 + n3 ) = e ,
8L2 mθ 1
n1 =1 n2 =1 n3 =1 n=1
109
which allows one to represent the partition function in the form
3/2 " r #3
1 θ θ θt
Zt = θ − .
8 θt θt θ
Therefore, for any finite θ all higher powers of θt /θ must be neglected and the function θ
should be replaced by 1 according to its asymptotic expansion above. Thus,
3/2
1 θ (2πmθ)3/2
Zt = =V . (3.12)
8 θt h3
Since at high temperatures one can neglect quantisation of energy, this expression must
coincide with the value of statistical integral divided by the volume a of elementary cell,
since under passing to integration gi goes into dΓ/a. Thus, classically, we must have
Z r
Zclass 4πV ∞ − p2 2 4πV π V (2πmθ)3/2
= e 2mθ p dp = · (mθ)3/2 = .
a a 0 a 2 a
a = h3 .
Thus, here we have found the volume of elementary cell a for a 6-dimensional µ-space
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
corresponding to 3 translational degrees of freedom: a = h3 . This result is general – it
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
appears that each degree of freedom (translational, rotatory and oscillatory) contributes to
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
the phase volume one multiple of h. In this section this result was obtained in the frame-
::::::::::::::::::::::::::::::::::::
work of the Maxwell-Boltzmann statistics for a non-degenerate gas but with the account of
quantisation of energy. In the next chapter we will find out that the volume of elementary
cell can be established experimentally, without taking into account quantisation of energy,
but for substances which obey the Bose-Einstein and Fermi-Dirac statistical distributions,
i.e. for strongly degenerate gases.
In conclusion, we note that since the characteristic temperature θt of translational motion
should be regarded as zero, quantisation of translational motion does not change, in fact,
the formulae for internal energy, entropy, heat capacity, chemical potential that have been
found in section 3.1.
We continue to consider a non-degenerate gas obeying the Boltzmann statistics. All differ-
ent forms of molecule energy (translational, rotatory and oscillatory) undergo quantisation.
However, minimal portions (quanta) of various forms of energy drastically differ by their
110
value. Denoting quanta of translational energy as ∆t , rotatory – ∆rot , oscillatory – ∆vib ,
electron – ∆e and nucleus – ∆n , one has the following hierarchy of energy scales
Therefore, the characteristic temperatures θt for translational, θrot for rotatory, θvib for
oscillatory, θe for electron and θn for nucleus motions satisfy similar inequality
θt θrot θvib θe θn .
In the previous sections we have studied quantisation of translational motion and saw
that the corresponding characteristic temperatures θt are very small and, therefore, for any
reachable temperatures θt θ and ∆t θ. This means that translational motion delivers
fully classical contribution to the internal energy (3/2N θ) and to heat capacity (3/2N ). We
recall that the degeneration temperature of a gas is much higher than the characteristic
temperature θt and at θ < θt the gas does not obey the Maxwell-Boltzmann statistics
anyway, not even saying that at such small temperatures all gases liquify or harden.
Now we consider qualitative be-
haviour of other degrees of freedom of cV
a diatomic gas. Until the mean en-
ergy of molecules ∼ θ is less than θrot
(θ < ∆rot ), rotatory and, moreover, 7/2
oscillatory degrees of freedom are not
excited under collisions and the gas be- 5/2
~2 ~2
H=− ∆1 − ∆2 + U (r) ,
2m1 2m2
111
where ∆1 and ∆2 are Laplacians over the coordinates of particles and r is the distance
between particles. One introduces instead of radius-vectors r1 and r2 of particles new
variables
m1 r2 + m2 r2
r = r1 − r2 , R = ,
m1 + m2
where r is a vector of relative distance and R is a radius-vector of the center of mass of the
two-body system. One can show that
~2 ~2
H=− ∆R − ∆ + U (r) ,
2(m1 + m2 ) 2m
where ∆R and ∆ are Laplacians over components of R and r and m = mm11+m m2
2
is the reduced
mass. Thus, the Hamiltonian factorises into the sum of two independent parts. Accordingly,
the wave function ψ(r1 , r2 ) can be searched in the product form φ(R)ψ(r), where φ(R)
describes the motion of the inertia center and ψ(r) gives the relative motion of particles.
For ψ we will get the following Schrödinger equation
2m
∆ψ + [E − U (r)]ψ = 0 .
~2
The Laplace operator in spherical coordinates yields
~2 1 ∂ 2 ∂ψ L2
− 2 r + 2 ψ + U (r)ψ = Eψ .
2m r ∂r ∂r r
For motion in a centra field the angular momentum L is conserved. The stationary states
are defined by the values of the angular momentum l defined via
L2 ψ = l(l + 1)ψ
and by its projection m which takes 2l + 1 values. Assuming that the distance between
atoms d remain constant (we neglect the influence of oscillations on the rotatory motion),
we have the following expression for the energy of a rotating body
~2 h2
rot = 2
l(l + 1) = 2 l(l + 1) ,
2md 8π I
where I = md2 is the inertia moment of the system. The energy levels are degenerate and
degeneracy is equal to 2l + 1. Therefore, for the partition function we have
∞
( )
X h2 l(l + 1)
Zrot = (2l + 1) exp − .
8π 2 Iθ
l=0
h2
θrot = ,
8π 2 I
we write
∞
( )
X θrot
Zrot = (2l + 1) exp − l(l + 1) .
θ
l=0
112
This series can not be summed in a closed form and we consider two limiting cases.
In this case the difference between energetic levels is small in comparison with the mean
energy
∆t /θ ∼ θrot /θ 1 ,
and, therefore, quantisation of energy can be neglected, so that rotational energy can be
regarded as continuous function. Thus, in this approximation the partition function must
coincide with the statistical integral Zclass . From the point of view of classical mechanics,
a diatomic molecule represents the so-called rotator3 with two degrees of freedom with the
following energy
I 2
rot = θ̇ + sin2 θφ̇2 , (3.13)
2
where θ and φ are angles defining the direction of the symmetry axis of a molecule. The
generalised momenta are
∂rot ∂rot
pθ = = I θ̇ , pφ = = I sin2 θφ̇ ,
∂ θ̇ ∂ φ̇
The energy in the Hamiltonian form is then
p2θ + p2φ / sin2 θ
rot = .
2I
The statistical integral takes the form
Z π Z 2π Z +∞ Z ( )
+∞ p2θ + p2φ / sin2 θ
Zclass = dθ dφ dpθ dpφ exp − = 8π 2 Iθ .
0 0 −∞ −∞ 2Iθ
On the other hand, the partition function in this limiting case can also be approximated
θrot
by an integral over l, because for θrot /θ 1 the summand f (l) = (2l +1)e−l(l+1) θ changes
very little for changing l by one. We use the Euler-Mac-Laurin formula (1.32) in the form
X∞ Z ∞
1 1 1 000
f (l) = f (l)dl + [f (0) + f (∞)] + [f 0 (∞) − f 0 (0)] − [f (∞) − f 000 (0)] + . . . .
0 2 12 720
l=0
This gives4
Z ∞ θr
Zrot = (2l + 1)e−l(l+1) θ dl
0
2 3 !
1 1 θrot 1 θrot θrot 1 θrot
+ + −2 − − + + ...
2 12 θ 60 θ θ 12 θ
θ 1 1 θrot
= + + + ... .
θrot 3 15 θ
3
A rigid body with principal moments of inertia I1 = I2 = I, I3 = 0.
4
One can see that the function f and all its derivatives at infinite are equal to zero.
113
Thus, in this approximation
" 2 #
θ 1 θrot 1 θrot
Zrot = 1+ + + ... .
θrot 3 θ 15 θ
This formula shows that Crot for θ θrot approaches its classical value N from above
and, therefore, has a maximum at some temperature θmax . Numerical integration gives
θmax ' 0.81θrot and a relative value of the maximum of Crot in comparison to the classical
value N is 1.1. This explains the small pick on the curve on Fig. 3.2.
θ
Finally, equating θrot = Zclass 2
a , we find a = h , i.e. the elementary phase volume for two
2
rotational degrees of freedom constitutes h .
In this case in the partition function successive terms tend to zero very fast with l growing,
so it is enough to restrict the consideration to the first two terms
Zrot = 1 + 3e−2θrot /θ .
Then we have
∂ h i
Erot = N θ2 ln 1 + 3e−2θrot /θ ≈ 6N θrot e−2θrot /θ
∂θ
and
2
θrot
Crot = 12N e−2θrot /θ .
θ
According to the Nernst theorem, Crot and E tend to zero for θ → 0.
Thus, for θ θrot rotatory degrees of freedom are fully excited and fully contribute to
the internal energy and heat capacity, N θ/2 and N/2 per each degree of freedom, and they
get frozen for θ θrot .
We note that rotational characteristic temperatures are inversely proportional to the
inertia moment I and, therefore, maximal values of θrot have gases, molecules of which are
build from light atoms and have small size. For most gases θrot:::::
::::::::::::::::::
have:::::: (1 ÷ 10)K, ::::
order::::::::::: i.e.
the characteristic temperature is lower than the condensation temperature, so that for
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
such
gases observation of freezing out of rotational degrees of freedom is not possible.
::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
114
Quantisation of oscillations
Here we treat oscillatory degrees of freedom of a diatomic molecule and restrict ourselves to
considering small oscillations, so that the potential energy can be regarded as a quadratic
function of deviations from equilibrium and oscillations themselves as harmonic.
In quantum mechanics the energy of a one-dimensional oscillator is
1
n = hν n + ,
2
where r
1 κ
ν=
2π m
is the frequency of oscillations, m is the reduced mass, κ is the Hooke constant and n is the
quantum number taking values n = 0, 1, 2, . . .. Energetic levels are non-degenerate, so all
gn = 1.
The partition function is therefore
∞
X e−hν/(2θ) e−θvib /(2θ)
Zvib = e−hν(n+1/2)/θ = = ,
n=0
1 − e−hν/θ 1 − e−θvib /θ
In this case quantisation of energy is not essential; and statistical sum must turn into
statistical integral. In classical mechanics the energy is
p2 κq 2
= + ,
2m 2
where q is the difference between the instantaneous distance r between atoms and the
equilibrium distance a. For the statistical integral we then have
Z ∞ Z ∞ r
−p2 /(2mθ) −κq 2 /(2θ) m θ
Zclass = dpe dqe = 2πθ = .
−∞ −∞ κ ν
On the other hand, expanding Zv for large θ, we will get
" #
θ 1 θvib 2 7 θvib 4
Zvib = 1− + + ... .
θν 24 θ 5760 θ
Comparing the leading term of this expansion with Zclass /a, we deduce θvib = hν = aν, so
that a = h. Thus, once again, on the example of oscillatory degrees of freedom, we have
demonstrated the rule of determining the elementary phase volume:
a = hf ,
115
where f is a number of degrees of freedom.
For energy and heat capacity of oscillatory degrees of freedom we have in this limiting
case
" 2 4 #
∂ 1 θ vib 1 θ vib
Evib = N θ2 ln Zvib = N θ 1 + − + ... ,
∂θ 12 θ 720 θ
" #
1 θvib 2 1 θvib 4
Cvib = N 1 − + + ... .
12 θ 240 θ
Thus, the oscillatory heat capacity at constant volume is Cvib ≈ N , which is the result
compatible with the equipartition law.
so that
∂ θvib 1
Evib = N θ2 − + ln 1 + e−θvib /θ = N θvib + e−θvib /θ ,
∂θ 2θ 2
2
θvib
Cvib = N e−θvib /θ .
θ
Finally, let us find the non-degeneracy criterium of diatomic ideal gas and compare it with
conditions of neglecting quantisation of energy of rotatory and oscillatory motion. As to
116
quantisation of translational motion, as we have seen it should not be taken into account
at arbitrary small temperatures. Conditions under which quantisation of energy can be
neglected are
h2
θ θrot = for rotatory motion (3.14)
8πJ
θ θvib = hν for oscillatory motion . (3.15)
Compare this with the degeneracy criterium of the Maxwell-Boltzmann gas. For multi-
atomic gases it is the same as for monoatomic ones e−µ/θ 1 and if it is satisfied, then
e−µ/θ can be found from
Z
g g
e−µ/θ = f
e−/θ dΓ = Zt Zrot Zvib ,
Na N
where f is the total number of degrees of freedom (translational, rotational, oscillatory).
Thus, the non-degeneracy criterium takes the form
g
Zt Zrot Zvib 1 .
N
Substituting for Zt , Zrot and Zvib their classical values, we will get
V 2πmθ 3/2 8π 2 Iθ θ
2 2
1. (3.16)
N h h hν
Substituting here the numerical values of density of a molecular gas N/V ∼ 1025 m−3 , mass
of a molecule m ∼ (10−27 ÷ 10−27 )kg , its linear size l ∼ (10−9 ÷ 10−10 )m (I ∼ ml2 ), the
frequency of molecular oscillations ν ∼ (1014 ÷ 1015 )s−1 , we obtain an estimate5
θ θdeg ≈ (1 ÷ 10)K .
Condition (3.14) allowing one to neglect quantisation of rotational energy, could ap-
pear stronger than the condition of applicability of the Maxwell-Boltzmann distribution,
or nearly the same. For instance, molecular hydrogen or deuterium characteristic rotation
temperatures are high: 85.4K and 43K, respectively, and in the interval from ∼ 10K till θrot
the rotation energy should be considered as quantised (rotational degrees of freedom freeze
out), while the gas as non-degenerate. For molecular oxygen and nitrogen (θrot = 2.85K and
θrot = 2.07K) criteria (3.14) and (3.16) practically coincide, so that above the corresponding
temperature the gas should be considered as non-degenerate and the rotational energy as
non-quantised.
Finally, condition (3.15) allowing one to neglect quantisation of oscillatory energy, is
much stronger than than the non-degeneracy condition, and, as we have already pointed
out, is not satisfied for any gas. Therefore, starting from temperatures (1 ÷ 10)K and up to
dissociation temperatures, the oscillatory energy of a diatomic molecule should be accounted
as quantised (oscillatory degrees of freedom freeze out), while the gas itself should be treated
with the Maxwell-Boltzmann statistics.
5
With pressure increasing, degeneration temperature grows as (N/V )2/7 ∼ p2/7 , that is rather slow and
the presented estimate could be sufficiently changed only at pressures that are so high that the gas can not
be anymore considered as ideal.
117
Chapter 4
Degenerate gases
In June 1924 Bose’s paper was published. He derived Planck’s radiation law by
counting states of photons in a novel way. As soon as Einstein saw this paper, he
generalized it to the counting of states of atoms, thereby predicting the phenomena
of Bose-Einstein condensation, a most daring and insightful extrapolation which has
only now been brilliantly experimentally confirmed.
A year and a half after Bose’s and Einstein’s papers, upon reading Pauli’s arti-
cle on the exclusion principle, Fermi realized in 1926 that he had now the concepts
in hand to discuss the thermodynamics of a collection of electrons. The results were
such fundamental concepts like the Fermi sea, the Fermi energy, etc.
118
the amount equal to θ. Therefore, to obtain ρ(ν, θ), one has to find the number of different
standing waves in a unit volume with frequencies in between ν and ν + dν and multiply it
by θ. Assume for simplicity that the cavity is a cube with the edge of length `, oriented
along the coordinate axes (the result at the end cannot depend on the form of cavity). The
equation for a standing wave is
y = A sin(k · r + δ) , (4.1)
where under y one can understand either electric or magnetic field, k represents the wave
vector with modulus
2π 2πν
k ≡ |k| = = .
λ c
The wave vector points in the direction of propagation of the corresponding wave. Here also
δ is an initial phase and A is an amplitude. Now we impose on the waves y(r) the periodic
boundary conditions. To this end, we introduce three unit vectors nα , α = 1, 2, 3, in the
direction of the coordinate axes x, y, z. Then the periodicity conditions read
y(r + ` nα ) = y(r) , α = 1, 2, 3 ,
that is
A sin(k · r + ` k · nα + δ) = A sin(k · r + δ) α = 1, 2, 3 .
Here k·nα = kα , where kα is a components of the vector k along the corresponding coordinate
axis. The solution of the condition above is obviously given by
As we can now see, these equations are equivalent to the statement that each edge of the cube
should accommodate an integer number of wave lengths,2 that is mα λα = `, λα = 2π/kα .
The frequencies of these waves take , therefore, discrete (quantum) values
q
ck c c
ν= = m21 + m22 + m23 ≡ m .
2π ` `
Introduce a space of numbers m1 , m2 , m3 . Denote by dn(ν) the number of different waves
with frequencies in the interval from ν to ν + dν. It is equal to the doubled (because of
two independent polarisations) number of points with integer coordinates mi lying between
two spheres with radii m and m + dm. Since each point accommodates a cube of unit
volume, then this number equals with good accuracy to the volume confined between these
two spheres, so that
8π`3
dn(ν) = 2 · 4πm2 dm = 3 ν 2 dν .
c
Taking in this expression `3 = V = 1 and multiplying by θ, we find the spectral energy
density
8πθ
ρ(ν, θ) = 3 ν 2
c
2
Physically, this is related to the fact that depending on reflecting properties of the walls, on the walls
nodes or anti-nodes of waves must be situated.
119
This is the Rayleigh-Jeans law. It agrees well with experiment for small frequencies while
it strongly disagrees with experimental data for large frequencies, the latter indicate that
ρ(ν, θ) as a function of frequency has a maximum for some value of frequency and then
decreases with ν increasing. ::::
The:::::::::::::::
Rayleigh-Jeans::::
law,:::::::
which :::::::::
indicates :::::::::::
unbounded:::::::
growth:::
of
ρ(ν, θ) with frequency leads to an absurd conclusion that the energy density
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
Z ∞
u(θ) = ρ(ν, θ)dν
0
at equilibrium state is infinite – a paradox that got historically the name of “ultraviolet
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
catastrophe".
:::::::::::::
In order to avoid this paradox one has to go beyond the framework of classical
physics (in the sense of introducing a new fundamental constant of Nature – the Planck
constant, as we see below).
Let us treat now radiation as a gas of photons, which have the following properties.
Photons move in vacuum with the speed of light and have the dispersion relation = cp.
They have spin equal to one, so that they are bosons. This means that the number of
photons in each elementary cell is unrestricted and can take any values. Moreover, ::::::::
photons
are permanently emitted and absorbed
P
:::::::::::::::::::::::::::::::::::::
by walls of a cavity, i.e. the total number of photons
::::::::::::::::::::::::::::::::::::::::::::::::::
is not fixed and the condition N i = N is not valid for the photon gas. Formally this
::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
means that the chemical potential related to the Lagrange multiplier α by µ = αθ should
be put to zero and
2dΓ/a
dN = .
e/θ − 1
Thus, from two equations
Z Z
1 2dΓ 1 2dΓ
N= /θ
, E = (4.2)
a e −1 a e/θ−1
the first equation changes its meaning: it does not define anymore the chemical potential
(which is now identically zero), rather it gives the mean value of photons in volume V .
Substituting here
4π
dΓ = 4πp2 dp · dV = 3 2 d · dV ,
c
integrating over volume and putting V = 1, we find for the spectral energy density the
following expression
8π 3
ρ(, θ) = (4.3)
c3 a e/θ − 1
The volume a of the elementary phase cell will be fixed later.
Integrating (4.3) over , we find
Z ∞ Z
8π ∞ 3 d 8π π4 θ4
u(θ) = ρ(, θ)d = 3 = × .
0 c a 0 e/θ − 1 c3 a 15
We briefly explain how to compute this integral. Expanding integrand in the geometric
series, we have
Z ∞ Z ∞ ∞ Z ∞
X
3 d 3 d
I= = = 3 e−(n+1)/θ d .
0 e/θ − 1 0 e/θ (1 − e−/θ ) n=0 0
120
Integrating, we get
∞
X X 1 ∞
4 1 4
I = 6θ = 6θ = 6ζ(4)θ4 ,
(1 + n)4 n4
n=0 n=1
π4
In particular, ζ(4) = 90 . Thus, ⇢ ✓ increases
8π 5 4
u(θ) = θ . (4.4) 6
15c3 a
5
8π 5 k 4 4σ 4
u(θ) = T4 = T , (4.5) 1
15c3 a c
5 10 15 ⌫
where k is the Boltzmann con-
stant3 and
2π 5 k 4 Figure 4.1: Spectral energy density and Wien’s dis-
σ= (4.6) placement law.
15c2 a
is the so-called Stefan-Boltzmann constant. Experimentally, the Stefan-Boltzamnn constant
was found to be
kg
σ = 5.67 · 10−8 . (4.7)
s3 · K4
From here we find for the elementary volume of the phase cell the following expression
2π 5 k 4 2π 5 1.384 · 10−92 KJ 4
3
4 −34
a= = · 2 = 6.62 · 10 J · s ,
15c2 σ 15 9 · 1016 m2 · 5.67 · 10−8 3kg 4
s s ·K
where
121
Let us now take into account the Einstein formula that relates energy of a photon with
frequency of the light wave: = hν.4 Then the Bose-Einstein distribution takes the form
8π ν 2 dν
dN (ν) = dV ,
c3 ehν/θ − 1
while the spectral energy density becomes
8πh ν3
ρ(ν, θ) = . (4.9)
c3 ehν/θ − 1
This is the celebrated Planck formula. Note the physical dimension [ρ(ν, θ)] = [h/`3 ] =
[E /`3 × t].
Consider now two limiting cases. In the region of small frequencies, where hν θ, we
expand the exponent in the denominator and recover the Rayleigh-Jeans law
8πθ 2
ρ(ν, θ) = ν .
c3
Naturally, in this limiting case of low-energy (soft) photons the particle properties of light do
not show up and we get the formula which coincides with the classical formula of Rayleigh
and Jeans. In the opposite limit, when hν θ, the Planck formula gets into
8πh 3 −hν/θ
ρ(ν, θ) = ν e . (4.10)
c3
This formula was proposed by Wien several years before the Planck formula (Wien’s radi-
ation law) and it agrees well with experimental data in the region of large frequencies. On
Fig. (4.1) the dependence ρ(ν) is depicted for various fixed temperatures.
From the Planck formula Wien’s displacement law follows. We find the maximum of the
spectral energy density from the equation
" #
∂ρ ν2 hν hν
8πh θ e
θ
= 3 hν/θ 3 − hν/θ = 0.
∂ν θ c e −1 e −1
3(ey − 1) = yey ,
which has a single root approximately equal to y = 2.822. From here we find Wien’s
displacement law in the form
hνmax
= 2.822 .
θ
Thus, for higher temperatures the maximum of the spectral energy density moves towards
higher frequencies (linearly with θ).
4
We recall the Einstein and de Broglie formulae = hν = 2π~ν = ~ω and p = hν/c = ~ ωc = ~k. These
formulae relate particle and wave properties of light and h can be regarded as the multiplier that converts
wave formulae into particle ones.
122
Let us now compute various thermodynamic characteristics of the photon gas. Earlier
we have found, see (2.56) that
1 1
Ω(θ, V, µ = 0) = − E = − u(θ)V .
3 3
4σ 4
Since Ω = −pV and u(θ) = ck4
θ , we find
Ω 1 4 σ 4
p=− = u(θ) = θ ,
V 3 3 ck 4
so that
1
pV = E .
3
Introducing the reduced Planck constant ~ through h = 2π~, the last formula can be written
as
3
2ζ(3) θ 3 θ
N= 2
V = 0.244 V,
π ~c ~c
5
The heat capacity at constant pressure does not have sense, since p = p(θ) and fixed p means fixed θ.
123
since ζ(3) = 1.202.
Looking at the expression for entropy of a photon gas, we deduce that under adiabatic
expansion or compression of a photon gas the temperature and volute are related as V θ3 =
const. Since p ∼ θ4 the equation for adiabats can be also written as pV 4/3 = const.
Let us now give some comments on the volume of the elementary phase cell that we found
to be a = h3 . Einstein proposed (1924) to extend this result for a on ordinary particles with
m 6= 0. He made this proposal basing on de-Broglie ideas that the dualism “wave-particle"
can be attributed not only to light (photons) but also to usual massive particles.
Note that in the general case of a system with f degrees of freedom, the elementary phase
volume is hf . This conclusion is in agreement with the Heisenberg uncertainty principle in
quantum mechanics. As is known, a classical description of a moving particle with the help
of coordinates and momenta which, strictly speaking, is invalid in quantum mechanics, can
still be applied approximately, if a coordinate and the conjugate projection of momentum
are determined together with uncertainties ∆x, ∆p, related by the Heisenberg uncertainty
principle
∆x∆p ≥ h .
This means that in the best possible case, when we have equality in the above formula, a state
on the phase plane (x, p) corresponds to a rectangular with area ∆x∆p = h, rather to a point.
The Heisenberg uncertainty relation is valid not only for cartesian rectangular coordinates
and momenta, but for any conjugate pairs of generalised coordinates and momenta for which
the Poisson bracket is equal to one. Therefore, for any quantum-mechanical object with f
degrees of freedom a state is described in a quasi-classical approximation not by a point
in the 2f -dimensional phase space, but by a cell with volume hf . In other words, we can
consider a motion of particle over a classical trajectory in the phase space, but we should
draw these trajectories with a certain density, such that every cell with volume hf will be
crossed by only one trajectory.
The example of a degenerate photon gas teaches us the following. For finding the numer-
ical value of a we do not need to appeal to any quantum-mechanical formulae. Indeed, the
internal energy of a gas of indistinguishable particles does depend on a as on parameter and,
therefore, a can be found from comparison with experiment, for instance, from the measured
value of the Stefan-Boltzmann constant (as we just did) or from the measurements of heat
capacity of fermi-gases.
This statement has s very deep meaning. It shows that the formal trick of dividing the
phase µ-space into cells in order to obtain a statistical description together with hypothesis
on indistinguishability of identical particles necessarily leads to the appearance of a new
fundamental constant of Nature with the physical dimension of action. This constant h,
which Planck has found by studying the theory of radiation and Einstein from matching
with experiment his formula for the photo-effect, could be already found from measuring of
heat capacity of a gas of bosons or fermions. This means that the main ideas of quantum
mechanics (Heisenberg uncertainty principle, quantisation of energy) are deeply rooted in
statistical physics.
124
4.2 Degenerate Bose gas and BEC
Here we consider a monoatomic degenerate non-relativistic Bose gas of non-interacting par-
ticles (ideal gas) with mass m. The Bose-Einstein distribution is
gi
Ni = ( −µ)/θ
. (4.11)
e i −1
The total number of particles and internal energy are
X gi
N = ( −µ)/θ
, (4.12)
i
e i −1
X i gi
E = ( −µ)/θ
. (4.13)
i
e i −1
First we note that the chemical potential of the Bose gas cannot be positive, as for µ > 0
the number of particles in the ground level 0 = 0 as well as on all levels with i < µ would
be negative, since ei −µ < 1. With temperature decreasing the chemical potential increases
remaining negative (so that it absolute value decreases). Indeed, we have
so that
P gi (i −µ)e(i −µ)/θ
∂µ 1 (e(i −µ)/θ −1)2
= − iP < 0.
∂θ θ gi e(i −µ)/θ
(e(i −µ)/θ −1)2
i
The passage to integration in (4.12) should be done with great caution. Indeed, assuming
energy to be continuous and taking into account that
where we have used that p = (2m)1/2 , an integral arising by replacing the sum is
Z ∞ 2m 3/2
ρ()d 1/2
, ρ() = AV = 2πg V 1/2 . (4.14)
0 e(−µ)/θ − 1 h2
One can now see that this integral misses the contribution of the ground state with 0 =
0. Indeed, passing to integration the first term in the sum (4.12) must be replaced with
R 0
an integral AV 0 d1/2 (e(−µ)/θ − 1)−1 , where 0 lies in between 0 = 0 and the first
quantised level 1 ∼ h2 /mL2 . Below we will find that under low temperatures |µ|/θ is
macroscopically small of order 1/N . Therefore, the contribution of the level 0 = 0 has
R 0 /θ
order AV θ3/2 0 dxx1/2 (ex − 1) and goes to zero together with θ. On the other hand, at
low temperatures there should be accumulation of bosons at the ground level 0 =P 0. This
follows from the fact that all Ni except N0 are exponentially small and their sum i6=0 Ni
can be replaced with good approximation by the integral (4.14) which also goes to zero as
125
θ → 0. Because of the normalisation condition, this means that the number of particles on
the ground level with 0 = 0 should be or order N , so that we have
g0 1 g0 θ
−µ/θ
∼N, → µ = θ log g0 ≈ − .
e − 1 1+ N N
In the thermodynamic limit N → ∞ this means that µ → −0 and we arrive at the conclusion
that there should exists an interval of low temperatures 0 < θ < θ0 in which the chemical
potential should be considered as zero in the macroscopic sense.
We, therefore, single out the contribution of the ground level and pass to integration
over other levels
Z ∞
g0 1/2 d
N = −µ/θ + AV . (4.15)
e −1 0 e(−µ)/θ − 1
The internal energy is determined only by particles with > 0 and passing to integration,
we get
Z ∞
3/2 d
E = AV . (4.16)
0 e(−µ)/θ − 1
Note that at any temperatures θ > θ0 the first term in (4.15) is macroscopically small
with respect to the second term (there ratio is ∼ 1/N ) and it should be discarded in the
thermodynamic limit. An equation
Z ∞
1/2 d
N = AV (4.17)
0 e(−µ)/θ − 1
can be solved with respect to µ only by approximate methods. At θ = θ0 the chemical
potential becomes macroscopically small (∼ 1/N ) and should be regarded as zero in the
thermodynamic limit . Putting in (4.17) µ = 0 and θ = θ0 , we obtain an equation for
finding θ0
Z ∞
1/2 d
N = AV
0 e/θ0 − 1
Z ∞ 1/2 √
3/2 x dx 3/2 π
= AV θ0 x−1
= AV θ 0 × ζ(3/2) .
0 e 2
Thus,
2mθ 3/2 √ 2πmθ 3/2
0 π 0
N = 2πg V × ζ(3/2) = gV ζ(3/2) .
h2 2 h2
From here
h2 N 2/3 0.084 h2 N 2/3 3.31 ~2 N 2/3
θ0 = = 2/3 = 2/3 . (4.18)
2πmg 2/3 ζ(3/2)2/3 V g m V g m V
At θ < θ0 the chemical potential of the bose gas remains macroscopically small – equal
to zero in the thermodynamic limit, while the number of particles on the ground level is
126
macroscopically large, comparable to N . Wherein the expression (4.17) with µ = 0 defines
not the total number of particles, rather the number of particles N>0 with > 0
Z ∞ 1/2 2mθ 3/2 Z ∞ x1/2 d
d
N>0 = AV = 2πg V . (4.19)
0 e/θ − 1 h2 0 ex − 1
The number of particles at the ground level = 0 is then N0 = N − N>0 . Dividing (4.19)
by (4.17), we find
N>0 θ 3/2 N0 θ 3/2
= , =1− . (4.20)
N θ0 N θ0
At θ → θ0 we have N0 → 0 and N>0 → N , all particles have positive energies. With
temperature decreasing below θ0 a transition of a macroscopic part of particles on the
ground level 0 = 0 starts and for θ → 0 we have N0 → N and N>0 → 0, i.e. all particles
occur on the ground level. This phenomenon is known as the Bose-Einstein condensation
and the tempretaure θ0 is called the temperature of condensation (BEC). It is clear, however,
that in the absence of an external field the term “condensation" should not be understood
literally, because one speaks about condensation in energetic (on the ground level 0 = 0) or
in momentum (on the level p = 0) space, rather than in 3-dimensional space. The totality
of particles with 0 = 0 is called condensate.
Let us find the energy of the bose gas at θ ≤ θ0 . We have
Z ∞ 3/2 Z ∞ 3/2 √
d 5/2 x dx 5/2 3 π
E = AV = AV θ = AV θ × ζ(5/2) . (4.21)
0 e/θ − 1 0 ex − 1 4
This gives
2m 3/2 √
5/2 3 π m3/2 θ5/2 V m3/2 θ5/2 V
E = 2πg θ V × ζ(5/2) = 31.69g = 0.128g . (4.22)
h2 4 h3 ~3
Let us now compute the main thermodynamic characteristics of the bose gas for θ ≤ θ0 . From
the energy formula we immediately obtain heat capacity under constant volume (caloric
equation of state)
∂E 5 m3/2 θ3/2 V 5E m3/2 θ3/2 V
CV = = × 31.69g = = 79.23g . (4.23)
∂θ V 2 h3 2θ h3
We see, in particular, that
127
For entropy we have
∂Ω 2 ∂E 5E 2 m3/2 θ3/2 V
S=− = = = CV = 52.82g . (4.26)
∂θ V 3 ∂θ V 3θ 3 h3
Finally, pressure is (thermal equation of state)
Ω m3/2
p=− = 21.13g 3 θ5/2 .
V h
We see that heat capacity and entropy tend to zero as θ → 0, according to the Nernst law.
Pressure does not depend on the volume, in this respect the bose bas is similar to saturated
vapour (Sattdampf). This similarity is explained by the fact that “condensed" atoms at the
state with 0 = 0 do not have momentum and, therefore, do not contribute to pressure.
At the point θ = θ0 all thermodynamic
quantities above are continuous. Below we
will show, however, that the derivative of µ ✓0
CV over temperature has a jump at this ✓
point. The function CV (θ) itself has a kink
and the value of CV achieves maximum at
θ0 .
In order to prove these statements, we
need to investigate the temperature region
lying a little bit higher than the condensa-
tion temperature 0 < (θ − θ0 )/θ0 1. In
this region N0 = 0 and |µ|/θ 1 is small.
However, in the expression Figure 4.2: Chemical potential of a bose gas
Z ∞ 1/2 in the thermodynamic limit N, V → ∞.
x dx
N = AV θ3/2 |µ|
(4.27)
0 ex+ θ − 1
We cannot perform an expansion over small R ∞|µ|/θ 1/2
because already the the coefficient in from
of |µ|/θ is given by the divergent integral 0 dxx ex (ex − 1)2 (diverges for small x). This
indicates that the right hans side of (4.27) is a non-analytic function of |µ|/θ. We transform
the integral as follows
Z ∞ Z ∞ 1/2 −x− |µ| X∞ Z ∞
x1/2 dx x e θ dx |µ|
Λ = |µ|
= |µ|
= x1/2 e−(n+1)x−(n+1) θ dx
0 ex+ θ − 1 0 1 − e−x− θ n=0 0
∞ Z ∞ √ X ∞ −n |µ| √ ∞ |µ| ∞
!
X |µ| π e θ π X 1 − e−n θ X 1
−n θ 1/2 −nx
= e x e dx = 3/2
= − 3/2
+
n=1 0 2
n=1
n 2
n=1
n n=1
n3/2
√ √ X ∞ |µ|
π π 1 − e−n θ
= ζ(3/2) − . (4.28)
2 2
n=1
n3/2
Introducing y = |µ| θ , we transform the sum above into an integral by applying the Euler-
Mac-Laurin formula
∞ |µ| Z ∞ " # " #
X 1 − e−n θ 1 − e−yx 1 1 1 e−y ye−y
3/2
= 3/2
dx + 1 − e−y + − − + ...
n 1 x 2 4 2 2 3
n=1
128
One can see that all non-integral terms in this expansion are of order y. In opposite, as we
√
will now show, the integral term is of order y, i.e. it is the leading one. We have
Z ∞ Z ∞ "Z Z y #
∞
1 − e−yx 1 − e−z 1 − e−z 1 − e −z
dx = y 1/2 dz = y 1/2 dz − dz .
1 x3/2 y z 3/2 0 z 3/2 0 z 3/2
Since
√
N π
3/2
= ζ(3/2) , (4.29)
AV θ0 2
we find
" # " #
ζ(3/2) 2 θ 3/2 2 θ 3/2 2
0 0
|µ| = √ θ 1− = 0.54θ 1 − .
2 π θ θ
129
We find cV N
" #
θ 3/2 θ 3
0 0
µ0 (θ) = −0.54 1 + −2 ,
θ θ kink
" #
0.54 3 θ0 3/2 θ 3
0
µ00 (θ) = − − +6 . 3/2
θ 2 θ θ
µ(θ0 + 0) = µ(θ0 − 0) ,
µ0 (θ0 + 0) = µ0 (θ0 − 0) .
Since µ and µ0 are continuous functions through θ0 , heat capacity is continuous as well.
Further differentiation of heat capacity over θ yields
Z ∞ 3/2 (−µ)/θ
∂CV e d
= AV (−µ)/θ
∂θ 0 (e − 1)2
" #2 " #2 " #
2e (−µ)/θ − µ µ 0 − µ µ 0 2( − µ) 2µ 0 µ 00
× (−µ)/θ + − + + − − 2 + .
(e − 1) θ2 θ θ2 θ θ3 θ θ
At θ0 we have µ = µ0 = 0, therefore,
Z " #
∞
∂CV 3/2 e/θ0 d 2e/θ0 2 2 2 µ00
= AV − − + .
∂θ 0 (e/θ0 − 1)2 (e/θ0 − 1) θ04 θ04 θ03 θ0
θ0
130
To compute the last integral, we use the fact that
Z ∞ 3/2 αx Z ∞ 1/2 √ √
x e dx ∂ x dx ∂ πζ(3/2) 3 πζ(3/2)
=− =− = .
0 (eαx − 1)2 ∂α 0 eαx − 1 ∂α 2α3/2 4α5/2
Taking into account (4.29), this gives
∂C 3√ 3 N
V 3/2
∆ = AV ∆(µ00 ) θ0 × πζ(3/2) = N ∆(µ00 ) = −3.66 .
∂θ 4 2 θ0
θ0
3/2
Below the condensation point we have CV = 1.92N θθ0 and therefore, (∂CV /∂θ)V
for θ → θ0 − 0 is +2.89N/θ0 . Taking into account the value of the jump, we then have
(∂CV /∂θ)V = −0.77N/θ0 for θ → θ0 + 0. At θ0 the function CV has a kink where it
achieves the maximal value, see Fig. (4.3). Thus, we see that BEC is an example of the
second order phase transition when µ0 is continuous and µ00 has a jump.
Let us now try to estimate the BEC tem-
perature θ0
~2 N 2/3
θ0 = kT0 ∼ 3.31
m V
Consider, for instance, such a bose gas as
He42 near the boiling point. The mass m of
He42 atom is
m = 6.645 · 10−27 kg
131
Finally, let us stress an essential dif-
ference in behaviour of bose and classical
Boltzmann gases at low temperatures. For
both gases the Pauli principle is no applied
and, therefore, with temperature decreas-
ing their particles must accumulate on the
ground level with 0 = 0. However, in the
case of a classical gas the fraction of parti-
cles on the ground level is
g0 g0
P ≈ .
−i /θ R∞
i gi e AV θ3/2 x1/2 e−x dx
0
Figure 4.5: λ-transition in He4 . He II is su-
To obtain the statistical thermodynamic de- perfluid of the Bose-Einstein condensate.
scription, we have to apply the thermody-
namic limit N → ∞, V → ∞. Obviously,
in this limit the fraction of particles on the ground level for any fixed temperature becomes
macroscopically small. In opposite, in bose gas the fraction of particles on the ground level
is g0 N/(e−µ/θ − 1) and already for finite temperatures θ ≤ θ0 this fraction becomes macro-
scopically large, i.e. comparable to 1, as is seen from 1 − (θ/θ0 )3/2 . Therefore, the BEC is a
phenomenon specific for bosons and it is related with the principle of indistinguishability of
identical particles and with the pole behaviour of the Bose-Einstein factor g0 (e−µ/θ − 1)−1 .
Let us first consider a fermi gas at zero temperature θ = 0. The fermi multiplier f () =
(e(−µ)/θ + 1)−1 behaves at θ → 0 as
1 for < µ0
1
lim = Θ(µ0 − ) = . (4.30)
θ→0 e(−µ)/θ + 1
0 for > µ
0
132
Here µ0 ≡ µ(θ = 0) is a limiting value of the chemical potential at θ → 0, while Θ(x) is the
Heaviside step-function. Thus, in the limit of zero temperature the Fermi-Dirac distribution
gi
Ni = gi f (i ) = (−µ)/θ
e −1
acquires the form
1 for < µ0
Ni
= Θ(µ0 − ) = (4.31)
gi
0 for > µ
0
133
From the equation of state for an ideal gas pV = 2/3E we can now find pressure of the gas
at zero temperature
3 3 p2F mv 2
E0 /N = F = = ,
5 5 2m 2
from where we get for the mean quadratic velocity
p r r
2
3 pF 3 3N 1/3 h
v = = .
5m 5 8πV m
Substituting numerical values, we get an estimate for the mean quadratic velocity
p r " #1/3
3 3 6.63 · 10−34 J · s
v2 = (1028 ÷ 1029 )m−3 ∼ 105 ÷ 106 m/s .
5 8π 9.1 · 10−31 kg
Switching on temperature
All the formulae obtained above are applied not only at θ = 0 but also at sufficiently low
temperatures. The range of applicability requires the ration µ/θ to be small. Indeed, at
non-zero temperatures part of electrons pass from levels below the Fermi boundary = µ0
on levels that lie above this boundary. As a result, the step-function gets a bit spread and
turns into the dashed curve depicted on Fig. 4.6. From the Fermi-Dirac distribution one
can see that semi-width of distribution ∆ ∼ θ and, for applicability of the formula above,
one needs the fulfilment of the inequality
θ θ0 = µ0 = F ,
where θ0 can be naturally identified with the degeneration temperature of electron gas. Nu-
merical estimate of F and θ0 gives F ∼ 10−19 J and T0 = θ0 /k ∼ (104 ÷ 105 ) K. Therefore,
electron gas in metals remains degenerate at any temperatures up to the melting point.
Note that degeneration criterium of electron gas can be written as
3N 2/3 h2 8 V h2 3/2
θ , → √ = λ3 ,
8πV 2m 3 πN 2πmθ
134
The properties of the degenerate Fermi-gas established above, namely, its huge internal
pressure, large mean velocities, high degeneration temperature are explained by the fact that
the fermi step-function is very elongated in the horizontal direction if the average distance
between levels is taken as a unit of energy. Indeed, at low temperatures the number of
occupied levels is ∼ N/2 and even for a piece of metal of 1 cm3 constitute 1022 ÷ 1023 . It is
clear that the ration of the blurring zone 2θ to µ and it represents a portion of electrons that
participate in the thermal motion and this portion remains very small up to the melting
point. This also explains a difficulty of the classical heat capacity theory based on the
Maxwell-Boltzmann distribution to explain heat capacities of metals.
If conductivity electrons in metal are considered as a classical electron gas, then, ac-
cording to the equipartition law of energy over degrees of freedom, these electrons must
bring an extra contribution to heat capacity equal to 3N/2θ. In reality the contribution of
electrons into heat capacity under the room temperature appear two orders of magnitude
smaller. Moreover, it appears that electron heat capacity Ce depends on temperature, as
Ce ∼ θ. These facts cannot be explained by the Maxwell-Boltzmann statistics and can be
understood from the point of view of the Fermi-Dirac statistics only.
Small heat capacity of electron gas is qualitatively explained by the fact that at low
temperatures θ θ0 those electrons take part in thermal motion, whose energy is close to
the fermi boundary F . Since, their portion is small – of order of the ration of blurring zone
2θ to µ0 , their contribution to heat capacity is also small, Ce ∼ 23 N µ2θ0 and linearly depends
on θ.
Now we give a quantitative description. We start from the formulae
Z ∞
1/2 d
N = AV , (4.32)
0 e(−µ)/θ + 1
Z ∞
3/2 d
E = AV , (4.33)
0 e(−µ)/θ + 1
3/2
where in the case of electron gas A = 4π 2m h2 . We will show that for low temperatures
θ µ0 the chemical potential can be excluded from these formulae. More generally, we
need to consider the integrals
Z ∞ Z ∞
ν d ν+1 xν dx
= θ ,
0 e/θ−µ/θ + 1 0 ex−y + 1
where we introduced y = µ/θ. Let us define
Z ∞
xν dx
J(ν) = .
0 ex−y + 1
According to our analysis of the zero temperature limit it is reasonable to subtract and add
the step function
Z ∞ Z ∞ h i
ν 1
J(ν) = x Θ(y − x)dx + xν dx x−y − Θ(y − x)
e +1
Z0 y Z y h
0
i Z ∞
1 xν dx
= xν dx + xν dx x−y −1 + x−y
0 0 e +1 y e +1
Z y Z ∞
y ν+1 xν dx xν dx
= − y−x
+ x−y
ν+1 0 e +1 y e +1
135
Here in the first integral we change the variable as y −x = u and in the second one x−y = v,
so that
Z 0 Z ∞
y ν+1 (y − u)ν du (v + y)ν dv
J(ν) = + + (4.34)
ν+1 y eu + 1 0 ev + 1
At θ0 we have y → ∞ and we can extend the integration region in the middle integral to
infinity. Thus,
Z ∞
y ν+1 (y + u)ν − (y − u)ν
J(ν) ≈ + du (4.35)
ν+1 0 eu + 1
Taking into account that
∞
X ∞
X ∞
X
ν ν ν
(y + u) − (y − u) = y Cνk (u/y)k −y ν
Cνk (−1)k (u/y)k = 2y ν
Cν2j+1 (u/y)2j+1 ,
k=0 k=0 j=0
We have
X∞
(−1)m+1
2(1+j)
= 1 − 2−1−2j ζ(2 + 2j) .
m=1
m
Thus, we obtain
X ∞
y ν+1
J(ν) ≈ +2 Cν2j+1 Γ(2 + 2j)ζ(2 + 2j) 1 − 2−1−2j y ν−2j−1 . (4.36)
ν+1
j=0
This is an asymptotic expansion of the integral. For ν = 1/2 the ratio of the coefficients of
power series (4.36) grow as aj+1 /aj ∼ 4j 2 for large j.8
R∞ ν
8
Passing from (4.34 to (4.35), we have omitted the integral y (y−u)
eu +1
du
. On can see, however, that this
integral produces corrections that are exponentially suppressed for large y. Indeed,
Z ∞ ∞ Z ∞
(y − u)ν du X
= (−1) m+1
(y − u)ν e−mu du
y eu + 1 m=1 y
∞ Z ∞ ∞
X X (−1)ν+m+1 −my
= (−1)m+1 (−1)ν e−my xν e−mx dx = Γ(ν + 1) e ,
m=1 0 m=1
mν+1
136
Using expansion (4.36) we find
Z ∞
ν d
N = AV = AV θ3/2 J(1/2)
e/θ−µ/θ + 1
0
ν=1/2
" #
3/2 π 2 µ −1/2 7π 4 µ −5/2
3/2 2 µ
= AV θ + + + ...
3 θ 12 θ 960 θ
" #
2 π 2 θ 2 7π 4 θ 4
= AV µ3/2 1 + + + ... .
3 8 µ 640 µ
Since
3/2
N = 2/3AV µ0 , (4.37)
We can perturbatively invert it, by searching for the inverse series µ = µ(θ) in the form
h i
µ = µ0 1 + a1 (θ/µ0 )2 + a2 (θ/µ0 )4 + . . . .
This is the form of the chemical potential away from θ = 0. With temperature growth the
chemical potential decreases and at a certain temperature θ• it changes the sign from positive
to negative. To find this temperature, we put in (4.32) the chemical potential µ(θ• ) = 0 and
get
Z ∞ 1/2 Z ∞ 1/2 √
d 3/2 x dx 2 − 2√ 3/2
N = AV /θ
= AV θ• x
= πζ(3/2)AV θ• .
0 e • +1 0 e +1 4
137
which matches the formula for the chemical potential of an ideal Maxwell-Boltzmann gas.9
Analogously, we compute the internal energy at low temperatures
Z ∞
ν d
E = AV /θ−µ/θ + 1
= AV θ5/2 J(3/2)
0 e
ν=3/2
" #
π 2 µ 1/2 7π 4 µ −3/2
5/2 2 µ
5/2
= AV θ + − + ...
5 θ 4 θ 960 θ
" #
2 5/2 5π 2 θ 2 7π 4 θ 4
= AV µ 1+ − + ... . (4.40)
5 8 µ 384 µ
138
of elementary cell and, therefore, measurements of heat capacity of electron gas allows to
find this volume and show that it is equal to h3 .
In order to find entropy and pressure of electron gas, we find the Ω-potential. Using
(4.40), we get
" #
2 4 5/2 5π 2 θ 2 7π 4 θ 4
Ω(θ, V, µ) = − E = − AV µ 1+ − + ... . (4.42)
3 15 8 µ 384 µ
These equations coincide with those for the usual monoatomic gas, hoever, the exponents
are not related with the ratio of heat capacities (because cp /cV = 5/3 and cp − cV = 1 are
not anymore valid).
Criterium of ideality
In view of the results obtained, one can ask to which extent at so higher densities and
pressures one can still consider an electron gas as ideal, i.e. to neglect interactions between
139
particles. It appears that degenerate fermi gas is peculiar – it becomes the more ideal the
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
higher its density.
::::::::::::::::::
Let us consider a gas on the background of a compensating charge of its nuclei (metals or
neutral plasma).10 The energy of Coulomb interaction of electrons with nuclei is Ze2 /a per
electron, where Ze is the charge of nucleus and a ∼ (V /(N/Z))1/3 = (ZV /N )1/3 is the mean
distance between electrons and nuclei. For a non-degenerate gas the criterium of ideality
would be in the requirement that the interaction energy should be smaller than the mean
kinetic energy which is or order θ, from where
Ze2 Ze2 N θ3
= θ → 2 6,
a (ZV /N )1/3 V Z e
that is the criterium of ideality would be satisfied for a dilute gas. In opposite, for a
2
2/3
degenerate gas the kinetic energy is of order µ0 = F ∼ hm N V , so that
10
A gas consisting of electrons alone would be unstable. Before we did not consider an effect of nuclei
because, due to the assumed ideality, the presence of nuclei does not effect the thermodynamic quantities of
the electron gas.
140
Chapter 5
Gibbs’ method
Multi-dimensional Γ-space
141
In a closed system, in which the Hamiltonian does not depend explicitly on time, the
energy is conserved and a phase space trajectory lies in the 6N − 1-dimensional hypersurface
of the constant energy
X X
W = Ki + U (qi ) + W (q1 , q2 , . . . , qN ) ,
i i
where Ki is kinetic energy of i-th particle, U (qi ) is its potential energy in an external field
and W (q1 , q2 , . . . , qN ) is a potential energy of interactions between particles. Note that
conservation of energy immediately follows from skew-symmetry of the Poisson bracket
dH
= {H, H} = 0 ,
dt
showing also that H is an integral of motion, that is the quantity that remains constant
along particle trajectories.
The product of all differentials gives the volume element of the Γ-space
Y
dΓ = dpj dqj .
j
The essence of the Gibbs method can be described as follows. Let us consider a given
system as immersed in surroundings (for instance in a thermostat). Due to interactions
with surroundings, a microstate of the system will change with time according to a very
complicated law. To foresee these changes is impossible because of huge number of degrees
of freedom. It is clear, however, that a point in the phase space representing a microstate
will move over a very tangled trajectory passing many times through any small volume of the
phase space. In general, this trajectory does not lie on a constant energy surface, because
due to interactions with surroundings the internal energy undergoes a slow change. One can
introduce the probability dW to find the point representing the system in any volume of the
phase space that is proportional to dΓ
where ρ(p, q) is the corresponding probability density also called the distribution function (p
and q here stand to denote all generalised coordinates of Γ). The probability density must
satisfy the normalisation condition
Z
ρ(p, q)dΓ = 1 .
Γ
142
If we will be watching the point representing the microstate of the system and mark its
position on the phase space trajectory after small time increments, then the entity of all
these instantaneous positions emerging after a long enough time fill the Γ-space with the
density ρ(p, q). Gibbs suggested that instead of looking how the position of a single point
representing the system changes with time, one can imagine many representative points
distributed in Γ-space with the density ρ(p, q). This means that we should imagine many
copies of the one and the same physical system that differ from each other only by the values
of qi , pi in some initial moment of time, which can be taken to be t = 0.2
For instance, if we study the behaviour of a gas in a vessel, we should imagine many such
vessels, if we consider a crystal, we should imagine many pieces of the same crystal, etc.3
Such a fanciful set of copies of the one and the same physical system, each copy being in
a particular microstate, is called statistical ensemble. In the following we consider different
examples of such ensembles depending on which parameters are fixed in the ensemble (energy,
volume, the number of particles, etc.). Due to different initial conditions and interactions
with surroundings, the state of each copy in the ensemble is changing differently with time.
This means that a point that describes a state of one species in the ensemble moves over
its own phase trajectory. The entity of all the points forms in Γ-space a gas, or, better to
say, a liquid with the density ρ(p, q, t). The average hf i of an observable f computed with
the help of (5.3) is called ensemble average. In a state of the thermodynamic equilibrium
the density should not have an explicit time dependence, ρ(p, q, t) = ρ(p, q), because the
ensemble averages (5.3) must be time-independent. To derive the equation for the density
in full generality, we assume nevertherless that ρ has an explicit time-dependence.
Let us now obtain the evolution equation for ρ(p, q, t), which is a function of coordinates
and time. Since the representative points can not be created or destroyed – the number of
specimens in the ensemble remains constant, then diminishing of the number of points4 in a
volume V per unit time should be compensated by the flux of points through the boundary
of this volume. In other words,
Z Z I
∂ ∂ρ
− ρ(p, q, t)dΓ = − dΓ = ρ vdσ .
∂t ∂t
V V
Here v is the velocity along the phase space trajectory, v = (ṗj , q̇j ) and the vector dσ is
directed along the normal to the surface towards outside the volume. By the Gauss theorem
we rewrite the last equality in the form
Z
∂ρ
dΓ + div(ρv) = 0 .
∂t
V
Since this equation must hold for integration over an arbitrary volume, we get an equation
∂ρ
+ div(ρv) = 0 .
∂t
2
Every point of the phase space can be thought of as initial data, and, according to Gibbs, initial data
are assumed to have a distribution with the density ρ(p, q).
3
Copies can even have different volumes and different number of particles but they are immersed in the
the one and the same surroundings.
4
R
Number of points in a volume V is ρ(p, q)dΓ.
V
143
This is the usual continuity equation and it only reflects the fact of constancy of the number
of points (ensembles). The information about the system contains in the vector v. Further,
one has
X ∂ ∂
X
∂
∂H
∂
∂H
div(ρv) = ρq̇j + ρṗj = ρ − ρ
∂qj ∂pj ∂qj ∂pj ∂pj ∂qj
j j
X ∂ρ ∂H X
∂ρ ∂H ∂2H ∂2H
= − + ρ − = {H, ρ} .
∂qj ∂pj ∂pj ∂qj ∂qj ∂pj ∂pj ∂qj
j j
Thus, the left hand side of the evolution equation for the density takes the Hamiltonian
form
∂ρ ∂ρ dρ
+ div(ρv) = + {H, ρ} =
∂t ∂t dt
and, therefore, the equation itself reads as
dρ
= 0.
dt
This is the Liouville equation. It says that the total derivative of the phase space density
vanishes along a phase space trajectory. The Liouville equation plays an important role for
studying time evolution laws of macroscopic systems. In statistical physics one considers
only equilibrium and, therefore, stationary states, which corresponds to the desity that is
constant at eqach point of the phase space, in other words, ∂ρ/∂t = 0. Then the Liouville
equation takes the form
v · ∇ρ = 0
or
{H, ρ} = 0 .
This means that in a stationary state the gradient of the density is normal to the dynamical
trajectory, so that along the trajectory the density remains constant. Thus, in this case
ρ is an integral of motion and depends only on the conserved quantities. Moreover, the
function ρ(p, q) is multiplicative, so that its logarithm is an additive integral of motion.
Indeed, if we split the system into two macroscopic parts I and II with linear dimensions or
order `, we can regard these parts as statistically independent. Due to a short interaction
range of forces between molecules, molecules which interact are in a thin layer between
the systems, but the number of them and their energy is of the order `2 , while the total
number of particles and their energy is proportional to `3 . Thus, for these subsystems their
interaction energy is very small in comparison to the internal energy. Thus, the subsystems
are quasi-closed and an instantaneous state of one subsystem does not influence the state
of the other. Note, however, that for a long time the interaction of the subsystems with
each other and with surroundings shows up in the fact that the energy of each subsystem
and also of the total system changes slowly and a representing point passes from one energy
hypersurface into another one. Because of statistical independence and according to the
theorem of multiplication of probabilities, the probability of finding a system in the element
dΓ of the phase space is
144
Since dΓ = dΓ1 dΓ2 , then
ρ = ρ1 ρ2 , ln ρ = ln ρ1 + ln ρ2 .
For most of physical systems there are 7 additive integrals of motion: energy, 3 projections
of the momentum P and 3 projections of the angular momentum M. Therefore, ρ(p, q) is
a function of these 7 integrals. In the following we will consider a macroscopic body in the
frame where it does not move as the whole, i.e. P = 0 and does not rotate, i.e. M = 0. In
this case ρ(p, q) is a function of energy only
ρ(p, q) = ρ(E ) .
For a given macrostate there is large number Γ(E , V, N ) of different microstates which are
consistent with this macrostate. Strictly speaking, there is an infinite number of microstates,
since the the phase space points in the thermodynamic limit V, N → ∞ are arbitrarily dense.
As the measure of a number of microstates, we can use the area of the energy surface, which
is at out disposal, and assume that Γ is proportional to this area
σ(E , V, N )
Γ(E , V, N ) = ,
σ0
where 1/σ0 is a proportionality coefficient. The number of microstates Γ(E , V, N ) is called
statistical weight.
The direct calculation of Γ through σ(E ) is inconvenient since it requires integration over
a complicated high dimensional energy surface. Typically, calculation of volume confined
by such surfaces is easier. Let V(E , V, N ) be the total phase space volume, the boundary of
which is given by the energy hypersurface E = H(p, q) and by the walls of the container in
the coordinate space. We then have
Z Z
V(E , V, N ) = dΓ = d3N p d3N q .
H(p,q)≤E H(p,q)≤E
For small ∆E , the volume between two energy surfaces with energies E and E + ∆E is given
by
∂V
∆V = V(E + ∆E ) − V(E ) = ∆E .
∂E V,N
145
On the other hand, the volume between two neighbouring surfaces with area σ(E ) and
distance ∆E is
∆E = σ(E )∆E .
Comparing the last two formulae, one gets5
∂V
σ(E ) = .
∂E
Thus, for the statistical weight we can write
σ(E , V, N ) 1 ∂V
Γ(E , V, N ) = = .
σ0 σ0 ∂E
The delta-function ensures that all points which are not on the energy surface with the area
σ(E) contribute with zero weight, while the factor 1/σ(E ) is the normalisation. This density
of a closed system defines an ensemble which is known as microcanonical.
For practical calculations (5.4) is not convenient because of the delta-function, and one
usually allows for a small energy uncertainty ∆E . One defines
( 1
Γ(E ,V,N ) if E ≤ H(p, q) ≤ E + ∆E
ρ(E , V, N ) = (5.5)
0 otherwise
In the case when the allowed microscopic energy runs over a discrete set {En } the definitions
are similar. One introduces the ∆-function as
(
1 if |ξ| ≤ ∆E
∆(ξ) =
0 if |ξ| > ∆E
Then ∆(E −En (x, N )) is a projector that projects a microscopic state n onto a fixed adiabatic
macroscopic state (E , x, N ). The total number of such states gives the statistical weight of
a given thermodynamic state
X
Γ(E , x, N ) = ∆(E − En (x, N )) ,
n
5
This equation is particularly clear and well-known for the case of the 2-sphere of radius R in 3-dimensions.
The volume is V(R) = 4π 3
R3 that yields the surface area as σ(R) = ∂V/∂R = 4πR2 .
146
that is the number of microstates by means of which a given macrostate can be realised.
Note that in the last formula summation is performed not over energy levels En but rather
over all microstates of the system described by the wave functions ψn (q; V, a, N ). Then the
probability wn of a microstate with energy En is given by the microcanonical distribution
( 1
Γ(E ,x,N ) if En is inside the shell ∆E ∆(E − En )
wn (E , x, N ) = = . (5.6)
Γ(E , x, N )
0 if outside the shell ∆E
Distributions (5.5) and (5.6) realises an equipartition principle: all microstates within an
energetic shell ∆E are equally probable. This statement cannot be justified within the
framework of equilibrium theory and, therefore, should be regarded as a hypothesis. Intu-
itively, it seems to be natural as a consequence of the macroscopic approach to a microscopic
situation where all microscopic states which are the same from a macroscopic point of view
should be regarded as equivalent. In order to understand this hypothesis, one has to un-
derstand how the equilibrium state of an N -body system is formed, which goes beyond the
framework of equilibrium theory.
In thermodynamic equilibrium the most probable macroscopic state in the one which cor-
responds to the largest number of consistent microstates. The basic hypothesis is that all
microstates with the same total energy appear with the same probability.
Consider a closed system which consists of two subsystems with the state variables Ei , Vi
and Ni . We have
i.e. the subsystems can exchange energy, volume and the particles. However, in equilibrium
Ei , Vi and Ni will adapt to certain mean values. The total number of all microstates
Γ(E , V, N ) results into a product of these numbers for the subsystems as the later are
statistically independent:
The most probable state, i.e. equilibrium state, is the one with the largest number of
microstates, i.e. Γ = Γmax and dΓ = 0. We have
d ln Γ = d ln Γ1 + d ln Γ2 , (5.7)
d ln Γ = 0 , ln Γ = ln Γmax . (5.8)
147
Considering the same system from a purely thermodynamic point of view, where the internal
energy of the system is identified with the total energy E , the entropy, being an additive
function of a thermodynamic state is
dS = 0 , S = Smax . (5.10)
Comparing equation (5.7) with (5.9) and equation (5.8) with (5.10), we observe the complete
analogy of ln Γ and entropy. The Boltzmann principle consists in identifying these quantities6
148
5.3 Canonical distribution
In the previous section we set up to describe a system in the framework α, where the
set of parameters (E , V, a, N ) was assumed fixed. From a practical point of view this is
not a convenient choice because we do not have a possibility to measure with the help
of simple devices internal energy E . As a consequence, results obtained with the help of
the microcanonical distributions should be re-calculated and expressed in terms of other
variables. From a macroscopic point of view, the most convenient way is β which operates
with the set (θ, V, a, N ) of independent variables.
Considering the variant β, we have now two systems in a thermodynamic equilibrium
– the one, for which we would like to find the distribution over microstates, and the other
which serves as a thermometer. Having two systems, we need to be careful about the
principle of thermodynamic additivity. The statistical weight has a definite property with
respect to division of a system on macroscopic subsystems Γ(E ) = Γ(E1 )Γ(E2 ). Since the
microcanonical distribution
∆ E − En (V, a, N )
wn (E , V, a, N ) = (5.13)
Γ(E , V, a, N )
depends on the form of the function ∆ which controls the structure of this distribution, it
is clear that the procedure of dividing a system on two macroscopic parts imposes certain
requirements on this function as well. To find these requirements, it is enough to consider
the simplest possibility: assume that the system singled out the by the adiabatic walls
(α) is divided by an adiabatic wall into two macroscopic subsystems in such a way that
thermodynamic equilibrium remains undisturbed. The subsystems appear adiabatically
isolated and, therefore, statistically independent from each other. If we denote their energies
as E1 , E2 and states as n1 , n2 , then, taking into account that N1 and N2 are large, then the
division does not influence neither macroscopic nor microscopic values of the corresponding
energies
as
E = E1 + E2
as
En = En1 + En2 .
Writing explicitly only the energy variables, for probability to find an adiabatically separated
system in a state (n1 , n2 ), we will have
Since for the statistical weights Γ(E ) = Γ(E1 )Γ(E2 ), for ∆ we will have
149
so that
f 0 (x1 ) f 0 (x2 )
= = const = β ,
f (x1 ) f (x2 )
and we find
f (x) = eβx ,
where it was taken into account that, according to the original equation, f (0) = 1. Thus,
the function ∆ must have the form
where we have introduced the statistical sum (Zustandssumme) or partition function for the
system under consideration
X
Z= e−βEn . (5.16)
n
ln Γ = S = βE + ln Z
is entropy, we find
1 ∂S
= =β,
θ ∂E V aN
so that
1
β= .
θ
Then, taking onto account the expression for the free energy F = E − θS, we have
F (θ, V, a, N ) = E − θS = −θ ln Z(θ, V, a, N ) ,
where
( )
X En (V, a, N )
Z(θ, V, a, N ) = exp − .
n
θ
150
or
n o
exp − En (x,N
θ
)
wn (θ, x, N ) = P n o. (5.18)
En (x,N )
exp − θ
n
8
Equipartition of probability of the same energy is immediately seen from the fact that wn ∼ eβEn .
151
where g(En ) denotes the degeneration of the En ’s level and wEn is the distribution function
over energies.
Let us pick up now O = H and define the width of the corresponding distribution wEn .
For the mean value of energy we obtain
1 X 1 ∂Z ∂ ln Z
E =H= En e−βEn = − =− ,
Z n Z ∂β ∂β
which is, of course, the same as (5.19). For the mean squared energy
1 X 2 −βEn 1 ∂2Z
H2 = En e = .
Z n Z ∂β 2
2 1 ∂2Z 1 ∂Z 2
(∆E)2 = H 2 − H = − .
Z ∂β 2 Z 2 ∂β
1 ∂2Z 1 ∂Z 2 ∂F ∂2F
− = −2 − β .
Z ∂β 2 Z 2 ∂β ∂β ∂β
Since
∂F ∂F ∂2F 2
2 ∂ 2 ∂F 3 ∂F 4∂ F
= −θ2 , = −θ − θ = 2θ + θ ,
∂β ∂θ ∂β 2 ∂θ ∂θ ∂θ ∂θ2
we obtain
∂2F ∂S CV N
(∆E)2 = −θ3 2
= θ3 = θ3 = θ2 CV = N θ2 cV N .
∂θ ∂θ θ
The relative fluctuation is
q √
1 θ cV N
δE = (∆E)2 = √ ∼ N −1/2 .
E ε N
This result is very characteristic to the whole statistical theory in the sense that the dis-
persion of an additive quantity is proportional to N , rather than N 2 . If we depict the
distribution wn in the energy scale ε = E/N then we get the graph depicted in Fig. 5.1. 9
The width of this distribution is
√
δE ' (∆E)2 = θ cV N N 1/2 .
9
Such a scale is chosen in order for the specific energy ε = E /N to stay at finite value in the thermodynamic
limit N → ∞.
152
Inversion theorem
Consider the statistical sum and write it via the sum over energies and then approximate
by an integral
X X Z ∞
−βEn −βEn
Z(β) = e = g(En )e = e−βE g(E)dE ,
n En 0
where g(E) is a density of microstates over energy and we have chosen the ground state to
have E0 = 0. From the mathematical point of view Z(β) is the Laplace transform of g(E).
The inverse Laplace transform is given by the Mellin transform
Z γ+i∞
1
g(E) = Z(u)eEu du , E ≥ 0 ,
2πi γ−i∞
where the integration path lies on the right from all singularities of Z(u) that means that
γ >Real (all poles of Z(u)).
To investigate this integral in the limit N → ∞, we will apply the method of steepest
descend. This method allows to obtain asymptotics of an integral I(N ) in the limit N → ∞,
where s
Z
N w(u) w(u0 ) 2π
I(N ) = e dz = N e 1 + O(1/N ) ,
L N w00 (u0 )
where u0 is defined from the equation w0 (u0 ) = 0.
as
In order to bring the integral g(E) for the form of I(N ), we recall that Z(u) = z(u)N ,
E = N and E = E = ε. Then
Z γ+i∞ Z γ+i∞
1 N N u 1
g(E) = z(u) e du = eN (ln z(u)+u) du .
2πi γ−i∞ 2πi γ−i∞
We, therefore, identify w(u) = u + ln z(u) and w0 (u) = + z 0 (u)/z(u). The point u0 is
determined by
ln z(β)
w0 (u0 ) = + = 0, β = u0 .
∂β
ln z(β) ∂ 1 1 ∂ε ∂s 1 cV
= −θ2 − ε + s = −ε + θ2 − = −ε + θ2 cV − = −ε .
∂β ∂θ θ θ ∂θ ∂θ θ θ
Thus, we have proved that
ln z(β)
ε=− .
∂β
153
As a result, = ε and the point u0 coincides with that inverse temperature β for which
the value of = E/N coincides with the thermodynamic energy ε = E /N . For the second
derivative at this point we will have
∂ ln z(β) ∂ ∂ε
w00 (u0 ) = =− = θ2 = θ2 cV .
∂β ∂β ∂β ∂θ
For the leading asymptotic we, therefore, find
r
1 N (β+ln z(β)) −2π eN (β+ln z(β))
g(E) = e = p .
2πi θ2 N cV 2πθ2 N cV
Using (5.20), we find that β + ln z(β) = s() is the entropy that the system would have if
its energy coincides with and temperature is θ = 1/β. Taking into account that N θ2 cV =
(∆E)2 , we arrive at
eN s() 1 Γ(E)
g(E) = p =√ q .
2πθ2 N cV 2π (∆E)2
g(En )e−βEn N1
wEn =
Z(β) " ✏
to the continuous distribution density w(E),
Figure 5.2: Squeezing of the distribution w()
we get
to the delta-function when N → ∞.
1 1
w(E) = g(E)e−βE = q e−βE+S(E)+βF (E ) .
Z(β)
2π(∆E)2
Since βF (β) = βE − S(β), for the expression in the exponent one finds
−βE + S(E) + βF (E ) = S(E) − S(E ) + β(E − E)
" #
∂S(E ) 1 ∂ 2 S(E ) 2
= − (E − E) + (E − E) + . . . + β(E − E) ,
∂E 2 ∂E 2
where we expand S(E) into Taylor series around E = E . Since ∂S(E )/∂E = β, the linear
term in the expression above cancels out, while the quadratic one includes the formula for
the dispersion of energy (∆E)2 = −∂E /∂β, thus,
( )
1 1 (E − E)2
w(E) = q exp − ,
2 2 (∆E)2
2π(∆E)
154
which is the standard Gaussian distribution with dispersion
∂E
(∆E)2 = θ2 = N cV N θ2 .
∂θ
It is not difficult to see that this distribution turns into the delta-function at N → ∞.
Singling out the N -dependence explicitly
E = Nε , E = N , (∆E)2 = N θ2 cV N , dE = N d
This distribution squeezes to the delta-function δ( − ε) in the limit N → ∞, see Fig. 5.2.
Concluding remarks
There exists other ways to introduce the microcanonical distribution, all of them are based
on the one and the same physical principles – thermodynamic understanding of macroscopic
parameters such as θ, s and so on, account of all characteristic properties of a statistical
system, in particular, the principle of thermodynamic additivity as well as independency of
a system on peculiarities of other systems that are in equilibrium with it, restriction to the
leading asymptotics in N in taking thermodynamic limit, accepting the microcanonical dis-
tribution as a departing point for the whole consideration. The popular ways of introducing
the canonical distribution include
• considering an ensemble of systems identical to the one under study and counting the
number of those which find themselves in a given microstate;
All these methods lead to the one and the same result for wn and they cast light on different
aspects of the canonical distribution.
155
5.4 Macrocanonical distribution
Here we introduce yet another important ensemble known as a grand canonical ensemble.
Our goal is as before – to find probability of finding a system in a certain microstate com-
patible with a given macrostate. The microcanonical ensemble describes systems with given
energy E and volume V and number of particles N , the canonical ensemble describes sys-
tems in a thermostat at given temperature θ, V and N . Now we want to consider open
systems that are singled out from surroundings by means of procedure (γ) (γ). Thus, we look
at a system in a thermostat confined in a volume V permeable to particles.10 The indepen-
dent variables are (θ, V, µ). A certain chemical potential is established by a large particle
reservoir which is in contact with the system. :: If ::::
the :::::::
particle:::::::::
reservoir:::
is :::::::::::
sufficiently ::::::
large,
the value of the chemical potential depends exclusively on the properties of the
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
reservoir,
in other words, µ is dictated to the system by the thermostat, in the same way as the
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
thermostat imposes its temperature. Since walls surrounding the system are transparent to
:::::::::::::::::::::::::::::::::::
particles, the particle number is not exactly conserved but in equilibrium some mean value
of N will be established.
The probability distribution wN,n (θ, V, a, µ) we are looking for should not only give the
probability to find our system in a given microstate n(N ) with an a priory fixed particle
number N , but should also contain the probability to find in the system exactly N particles
where all the possibilities N = 0, 1, 2, . . . , ∞ are allowed.
We derive the probability distribution wN,n (θ, V, a, µ) for a system by assuming that
the system and a thermostat are in equilibrium and form together an adiabatically isolated
system subjected to the microcanonical distribution. The energy E and the number of
particles N of the total system are fixed. We have two macroscopic relations
E = E + ET , N = N + NT ,
where E and ET are the mean (thermodynamic) energies of the system and the thermostat,
respectively, while N and NT are the mean number of particles in the system and in the
thermostat. In addition, we have two microscopic relations
E = En (N ) + ET (N − N ) , N = N + NT ,
156
Recalling the relation between statistical weight and statistical entropy, we can write
( )
wN,n (θ, V, a, µ) = exp ST E − En (N ), N − N − S E, N .
The thermostat and the system are in equilibrium, therefore, the total entropy is
S E, N = ST (EI , NT ) + S(E , N )
Further,
ST E − En (N ), N − N = ST ET + E − En (N ), NT + N − N . (5.21)
We require that the thermostat is very large compared to the system, i.e.,
E En N N
1, 1, 1, 1.
ET ET NT NT
This means that E − En (N ) ET and N − N NT and we can expand (5.21) into Taylor
series around ET , NT
ST E − En (N ), N − N = ST ET + E − En (N ), NT + N − N
∂ST ∂ST
= ST ET , NT + E − En (N ) + N − N + ... .
∂ET ∂NT
Taking into account that
∂ST 1 1 ∂ST µT µ
= = , =− =− , (5.22)
∂ET θT θ ∂NT θT θ
we obtain
( )
1 µ
wN,n (θ, V, a, µ) = exp E − En (N ) − N − N − S(E , N )
θ θ
( ) ( )
E − θS − µN 1
= exp exp − En (N ) − µN .
θ θ
Recalling the definition of the grand canonical potential Ω(θ, V, µ) = E −θS −µN , we finally
obtain
( )
Ω − En (N ) + µN
wN,n (θ, V, a, µ) = exp .
θ
157
The expression11
( ) ∞
X X En (N ) − µN X µ
N
Z (θ, V, µ) = exp − = eθ Z(θ, V, N )
θ
N n∈{n(N )} N =0
is called the grand canonical partition function. It represents the weighted sum of all canon-
ical partition functions Z(θ, V, N ). The weighting factor z = exp(µ/θ) is called fugacity.
The relation of the grand canonical potential to the grand canonical partition function is
then given by
( )
X X E n (N ) − µN
Ω(θ, V, µ) = − ln Z (θ, V, µ) = − ln exp − . (5.23)
θ
N n∈{n(N )}
Let k will be all possible values of energy for one particle of a gas. Since particles do not
interact, the energy and the number of particles can be written as
X X
E= nk k , N = nk , (5.24)
k k
where sums are taken over all possible particle states. Every microstate of an ideal gas is
uniquely characterised by a set of occupation numbers {n1 , n2 , . . . , nk , . . .}. Therefore,
∞
( )
X X µN − E{n1 ,n2 ,...}
Z (θ, V, µ) = exp
θ
N =0 n1 +n2 +...=N
∞
( )
X X µ(n1 + n2 + . . .) − (n1 1 + n2 2 + . . .)
= exp
θ
N =0 n1 +n2 +...=N
Due to summation over N , in this expression all possible sets of numbers ni occur with all
possible values of n1 + n2 + . . .. As a consequence, the sum can be rewritten as
XX (µ−1 )n1 (µ−2 )n2 YX (µ−i )ni
Z (θ, V, µ) = ...e θ e θ ... = e θ .
n1 n2 i ni
From now on one has to treat separately the case of bosons and fermions.
11
For simplicity we assume that external fields a are absent.
158
For a system of bosons numbers ni can take any values ni = 0, 1, 2, . . . and summing the
geometric series we get
X∞ h i
(µ−i ) ni 1
e θ = (µ−i )/θ
,
n =0
1 − e
i
from where
Y −1
Z (θ, V, µ) = 1 − e(µ−i )/θ .
i
For a system of fermions, because of the Pauli exclusion principle, numbers ni can take only
two values ni = 0, 1 and we have
X h (µ−i ) ini
e θ = 1 + e(µ−i )/θ ,
ni =0,1
so that
Yh i
Z (θ, V, µ) = 1 + e(µ−i )/θ .
i
Formula (5.23) implemented for each of the two cases above yields
X h i
Ω(θ, V, µ) = ±θ ln 1 ∓ e(µ−i )/θ .
i
159
Part III
Mathematical appendix
160
1. The volume of N -dimensional sphere
Consider an N -dimensional sphere of radius R. Its volume VN (R) is given by an integral
Z Z
N
VN (R) = dx1 . . . dxN = R dy1 . . . dyN
PN PN
i=1 x2i ≤R2 i=1 yi2 ≤1
VN (R) = RN CN ,
and evaluate it in an alternative way by using integration over spherical shells. We have
so that Z ∞
2
IN = N CN dR RN −1 e−R = CN N2 Γ( N2 ) .
0
Thus, the volume of a unit N -dimensional sphere is
π N/2
CN = N N
(1.25)
2 Γ( 2 )
RN π N/2
VN (R) = N N
. (1.26)
2 Γ( 2 )
R3 π 3/2 4
V3 (R) = √
π
= πR3 .
3
· 3
2 2
161
2. Stirling formula
The Stirling formula gives an approximation for the value of N ! for N large. To derive this
formula, we start with the following representation for the factorial
Z ∞
1
N ! = (−∂α ) N
= Γ(N + 1) = tN e−t dt .
α 0
α=1
where we introduced
√ Z ∞
χ= N e−N f (z) dz ,
−1
where
f (z) = 12 z 2 − 13 z 3 + 41 z 4 +
is a function that achieves its minimum at z = 0. We have
2 /2
e−N f = e−N z 1+ N 3
3z − N 4
4z + N 5
5z + ... .
Here the odd powers of t are integrated to zero in the limit N → ∞ because the integrand
for these terms in an odd function. Thus, In the large N limit
ln N ! = N ln N − N + 21 ln(2πN ) + O(1/N )
For large N the last term is small in comparison to the first two (for instance, already
for N = 103 it is of the order 0.4%). Thus, in statistical physics one uses the Stirling
approximation in the form
N
N
N! = e (1.27)
and
ln N ! = N ln N − N . (1.28)
162
Another way to derive the last formula is based on the equality
d
(x ln x − x) = ln x ,
dx
so that differentiating both sides we get
Z
dx ln x = x ln x − x ,
so that Z N
dx ln x = N ln N − N + 1 ≈ N ln N − N
1
for N large. The integral on the left hand side can be approximated by a partial sum
1 · ln 2 + 1 · ln 3 + . . . + 1 · ln N = ln 2 + ln 3 + . . . + ln N = ln(1 · 2 · 3 · . . . · N ) = ln N !
yields r
a
A= .
π
The mean value x̄ = 0 because the distribution is an even function. Then
r Z ∞ r r
a 2 −ax 2 a ∂ π 1
x2 = x e dx = − = .
π −∞ π ∂a a 2a
Excluding a the Gauss distribution can be written in the form
2
1 −x
dW (x) = p e 2x2 dx . (1.29)
2πx2
where −∞x, y < +∞. The normalisation condition can be found by completing the square
b b b2 b2 2 b 2 b2 2
a x2 + 2 xy + bcy 2 = a x2 + 2 xy + 2 y 2 + c − y =a x+ y + c− y .
a a a c a a
163
We have
Z #Z" " #
∞
b2 2 ∞ b 2
1 = A dy exp − c − y dx exp − a x + y
−∞ a −∞ a
r Z ∞ " # r s
π b2 2 π π πA
= A dy exp − c − y =A =√ ,
a −∞ a a c − b2 ac − b2 a
so that
1p
ac − b2 .
A=
π
Thus, the normalised distribution has the form
1p 2 2
dW (x, y) = ac − b2 e−(ax +2bxy+cy ) .
π
For the mean values x2 , y 2 and the correlation xy, we will have
1p ∂ π c
x2 = − ac − b2 √ = ,
π ∂a ac − b 2 2(ac − b2 )
1p ∂ π a
y2 = − ac − b2 √ = , (1.30)
π ∂c ac − b 2 2(ac − b2 )
1 p ∂ π b
xy = − ac − b2 √ =− .
2π ∂b ac − b2 2(ac − b2 )
164
We expand integrand in the power series in ξ and integrate over ξ
f (xi + ∆x) + f (xi ) f 0 (xi + ∆x) − f 0 (xi ) (∆x)2 f 00 (xi + ∆x) + f 00 (xi ) (∆x)3
Ii = ∆x − +
2 2 2 2 6
f 000 (xi + ∆x) − f 000 (xi ) (∆x)4 f IV (xi + ∆x) + f IV (xi ) (∆x)5
− +
2 24 2 120
f V (xi + ∆x) − f V (xi ) (∆x)6 f VI (xi + ∆x) + f VI (xi ) (∆x)7
− + + ...
2 720 2 5040
From this sum all the terms which contain half-sums of even derivatives can be excluded.
To show this, consider first
f 0 (xi + ∆x) − f 0 (xi ) f 0 (xi + ∆x − ∆x) − f 0 (xi + ∆x)
f 0 (xi + ∆x) − f 0 (xi ) = − .
2 2
We expand the first term on the right hand side at xi and the second term at xi + ∆x. This
yields
f 00 (xi + ∆x) + f 00 (xi ) f 000 (xi + ∆x) − f 000 (xi ) (∆x)2
f 0 (xi + ∆x) − f 0 (xi ) = ∆x −
2 2 2
f IV (xi + ∆x) + f IV (xi ) (∆x)3 f V (xi + ∆x) − f V (xi ) (∆x)4
+ −
2 6 2 24
VI VI
f (xi + ∆x) + f (xi ) (∆x) 5
+ + ...
2 120
Analogously,
f 000 (xi + ∆x) − f 000 (xi ) f 000 (xi + ∆x − ∆x) − f 000 (xi + ∆x)
f 000 (xi + ∆x) − f 000 (xi ) = − ,
2 2
and we expand here the first term on the right hand side at xi and the second term at
xi + ∆x
f IV (xi + ∆x) + f IV (xi ) f V (xi + ∆x) − f V (xi ) (∆x)2
f 000 (xi + ∆x) − f 000 (xi ) = ∆x −
2 2 2
f VI (xi + ∆x) + f VI (xi ) (∆x)3
+ + ...
2 6
One more term of the same type
f V (xi + ∆x) − f V (xi ) f V (xi + ∆x − ∆x) − f V (xi + ∆x)
f V (xi + ∆x) − f V (xi ) = −
2 2
f VI (xi + ∆x) + f VI (xi )
= ∆x + . . .
2
Starting from the last equation, we successively find
f VI (xi + ∆x) + f VI (xi ) f V (xi + ∆x) − f V (xi ) 2
= + ...
2 2 ∆x
f (xi + ∆x) + f IV (xi )
IV f 000 (xi + ∆x) − f 000 (xi ) 2 f V (xi + ∆x) − f V (xi ) ∆x
= + + ...
2 2 ∆x 2 6
f 00 (xi + ∆x) + f 00 (xi ) f 0 (xi + ∆x) − f 0 (xi ) 2 f 000 (xi + ∆x) − f 000 (xi ) ∆x
= +
2 2 ∆x 2 6
f V (xi + ∆x) − f V (xi ) (∆x)3
− + ...
2 360
165
Plugging these expressions into the formula for Ii , we obtain
f (xi + ∆x) + f (xi ) f 0 (xi + ∆x) − f 0 (xi ) (∆x)2 f 000 (xi + ∆x) − f 000 (xi ) (∆x)4
Ii = ∆x − +
2 2 6 2 360
V V
f (xi + ∆x) − f (xi ) (∆x) 6
− + ...
2 15120
The integral is written as the following sum of Ii ’s
Z b n−1
X
I= f (x)dx = Ii .
a i=0
(∆x) 2 n−1
X 0
− f (xi + ∆x) − f 0 (xi )
12
i=0
(1.31)
n−1
(∆x)4 X 000 000
+ f (xi + ∆x) − f (xi )
720
i=0
6 n−1
X
(∆x)
− f V (xi + ∆x) − f V (xi ) + . . .
30240
i=0
Second, we observe that in each of the remaining sums in (1.31) only the first and the last
terms contribute, while all the other terms cancel out leaving behind
" n #
X f (a) + f (b)
I = ∆x f (xi ) −
2
i=0
2
(∆x) (∆x)4 000 (∆x)6 V
− f 0 (b) − f 0 (a) + f (b) − f 000 (a) − f (b) − f V (a) + . . .
12 720 30240
Thus, we obtain
Xn Z b
1 f (a) + f (b)
f (xi ) = f (x)dx +
∆x a 2
i=0
∆x 0 (∆x)3 000 (∆x)5 V
+ f (b) − f 0 (a) − f (b) − f 000 (a) + f (b) − f V (a) + . . .
12 720 30240
166
Quite remarkably, the power series in ∆x arising in the right hand side can be written in a
compact form as
Xn Z b ∞ h i
1 f (a) + f (b) X B2k
f (xi ) = f (x)dx + + (∆x)2k−1 f (2k−1) (b) − f (2k−1) (a) ,
∆x a 2 (2k)!
i=0 k=1
The remainder term has an exact expression in terms of the periodised Bernoulli functions
Pk (x), which we will not discussed here and refer the reader to the mathematical literature.
Taking xi = a + i, so that ∆x = 1, we get
Z
f (a) + f (b) X B2k h (2k−1) i
b
X b ∞
f (i) = f (x)dx + + f (b) − f (2k−1) (a) , (1.32)
a 2 (2k)!
i=a k=1
which is an asymptotic expansion of the corresponding sum. Often the expansion remains
valid even after taking the limits a → −∞ or b → +∞ or both.
A good illustration of the last formula is the derivation of the Stirling approximation.
We have
XN Z N
ln 1 + ln N
ln N ! = ln(1 · 2 · . . . N ) = ln i = ln(x)dx +
1 2
i=1
X∞
B2k (−1)2k (2k − 2)! 2k
+ − (−1) (2k − 2)! ,
(2k)! N 2k−1
k=1
Finally, since all Bernoulli numbers on odd argument are zero, the last formula takes the
form
X∞
1 (−1)k Bk 1
ln N ! = N ln N − N + ln N + 1 + −1 .
2 k(k − 1) N k−1
k=2
167
5. Functional relation for the function θ
Recall the definition of θ(x)
∞
X 2x
θ(x) = e−πn
n=−∞
This function is periodic with the period equal to 1 and, therefore, it can be expanded into
the Fourier series
∞
X
S(x, y) = am (x)e2πimy ,
m=−∞
where Z Z
1 1 ∞
X
−2πimt 2 x−2πimt
am (x) = S(x, t)e dt = e−π(n+t) dt .
0 0 n=−∞
Changing the order of summation and integration and taking into account that e−2πinm = 1,
we get
∞ Z
X 1 ∞ Z
X n+1
−π(n+t)2 x−2πim(n+t) 2 x−2πimz
am (x) = e dt = e−πz dz
n=−∞ 0 n=−∞ n
Z +∞ Z +∞ 2
−πz 2 x−2πimz −πm2 /x −πx(z+im/x)2 e−πm /x
= e dz = e e dz = √ .
−∞ −∞ x
Therefore,
∞
1 X −πm2 /x+2πimy
S(x, y) = √ e .
x m=−∞
∞
1 X −πm2 /x 1 1
θ(x) = √ e =√ θ .
x m=−∞ x x
168