THERMODYNAMICS AND STATISTICAL PHYSICS

LECTURES

Events important for students of Statisti al Physi s A (English)

• 2nd De ember 2019 at 9.00 in room 0.03a - 1st olloquium

• 11th January 2020 at 9.00 in room 1.40 - 2nd olloquium

• 3rd February 2020 at 13.00 in room 1.40 - 1st written exam

• 17th February 2020 at 13.00 in room 1.40 - 2nd written exam

Fizyka Statysty zna A wersja polskojzy zna

• 25 listopada 2019 o 9.00, sale 1.02 i 2.21 - pierwsze kolokwium

• 13 sty znia 2020 o 9.00, sale B2.38 i 2.21 - drugie kolokwium

• 31 sty znia 2020 o 9.00, sala 0.06 - pierwszy egzamin pisemny

• 20 lutego 2020 o 9.00, sala 0.06 - drugi egzamin pisemny

 LITERATURE
• A.B. Pippard, Elements of Classi al Thermodynami s

• Adkins, Thermodynami s

• H. Callen, Thermodynami s

• E. Fermi, Thermodynami s

• R. Kubo, Thermodynami s (Problems with solutions)

• P. Landsberg, Thermodynami s and Statisti al Me hani s

• R. Kubo, Statisti al Me hani s An advan ed Course (Problems with solutions)

• K. Huang, Statisti al Me hani s

and in Polish

• J. Werle, Termodynamika

• K. Gumi«ski, Termodynamika

• K. Rejmer, Ciepªo−→Zimno, z±¢ pierwsza: Zasady, z±¢ druga: Zastosowania

• K. Huang, Me hanika Statysty zna

2
 LECTURE I (TMD)

Thermal physi s or statisti al thermodynami s in the large meaning of these terms

en ompasses all ways and methods of understanding and interpreting properties of matter
as far as they are inuen ed by the hanges of temperature. In this sense it is one of the
major subdivision of physi al s ien e as su h and it employs a variety of mathemati al
and experimental te hniques, as well as all available information about mi ros opi on-
stitution of matter to a hieve its goals - the explanation of observed properties of matter
at all temperatures and all onditions. This is, of ourse, what makes the subje t so
interesting as it en ompasses essentially all physi s and is important to pra ti ally all its
areas, from ondensed matter to osmology. Briey, all physi s unites it it. Thus all gen-
eral ourses of lassi al me hani s, quantum me hani s, ele trodynami s you have taken
in this Department an be onsidered introdu tions to statisti al thermodynami s.

This general goal has resulted in the quite widespread tenden y (in fa t not a modern one
- look at the Landau Lifs hitz theoreti al physi s ourse in whi h the subje t is organized
in this way) to present this subje t mixing statisti al physi s with thermodynami s.

I, however, prefer not to pro eed in this way, and would like to learly separate what
is alled lassi al thermodynami s from the statisti al physi s part in this le ture.
Perhaps this will make unhappy some of you - those who have already taken thermo-
dynami s ourse run here by experimentalists and have hoped to have this awful thing
aus dem Kopf  on e for ever - but I'm onvin ed (I'm not alone in this) this is the right
way. First of all, you have in fa t only a very poor knowledge of what thermodynami s
really is, and se ondly, at this level one annot go too far with true statisti al physi s -
all we are going to dis uss will be three statisti al ensembles applied mainly to systems
of nointera ting parti les. Going further would require hanging the format of this ourse
to 2x3 at least... But more importantly, good working knowledge of thermodynami s is
important be ause when the statisti al approa h is applied to a physi al system, the rst
goal is to re over its thermodynami s and on e some basi fun tions hara terizing this
system are omputed using statisti al methods, one applies to them the thermodynami s
formalism. sS one has rst to learn and understand it.

Classi al Thermodynami s (CTMD) is a phenomenologi al theory - it does not

enter into the mi ros opi stru ture of matter. Its goal is more modest: it is establish-
ing relations between observed (measured) properties of bodies and substan es thereby
redu ing the very large number of su h properties to only a few; the others an be then
treated as onsequen es of these few. For instan e a gas exhibits dierent behaviour under
various onditions, but given its equation of state of a gas in the form f (T, V, p) = 0 and
one of its spe i heats, say cp (T, p0 ) over a range of temperatures (for a xed value p0
of the pressure) it is possible to predi t quantitatively how it will behave under various
ir umstan es, e.g. how the temperature of the gas will hange when it is passed through
a throttle (a valve), how its volume will hange with temperature, what is its spe i heat
cv for the same range of T and any volume, ompute the work that an be extra ted when
the gas is expanded, et .

3
 CTMD deals only with the equilibrium states of ma ros opi physi al systems and
although onsiders pro esses in whi h systems undergo hanges, following the time
evolution of the systems is beyond its s ope. For this reason some would prefer the
name thermostati . Nevertheless, using essentially idealizations like quasistati pro esses,
adiathermal isolations, whi h are limiting ases of real situations, thermodynami s is able
to formulate basi restri tions whi h apply to real pro esses and annot be over ame.

Classi al Thermodynami s essentially rests on only four (but in most ases three are
su ient) laws whi h I denote nTMDL, n = 0, 1, 2, 3. (The law means here something
whi h annot be derived from other rules.) These laws onstitute a generalization of
a great number of experimental observations. Continued appli ation of the methods of
lassi al thermodynami s based on these laws to all kinds of pra ti al problems showed
these laws give orre t predi tions in all ases. This is the empiri al justi ation of these
laws as having very large (pra ti ally unlimited) range of validity (they were never falsied
- this is what the law means).

Classi al Thermodynami s does not seek to explain the origin of these laws, that is how
they emerge as onsequen es of more fundamental laws whi h govern the behaviour of
mi ros opi onstituents of matter - this is the goal of statisti al physi s whi h we are
going to treat in the se ond part of this le ture. But we will see that in fa t they annot
be - at least to my taste - fully derived from the laws of me hani s (be it lassi al or
quantum); they are rather repla ed by other postulates (laws) whi h again lead to or-
re t predi tions in most ases, but here the problem be omes more ompli ated be ause
appli ation of statisti al physi s methods relies in most ases on approximations and/or
on using simplied models of matter. Nevertheless, analyzing behaviour of systems of
systems onsisting of huge numbers of mole ules (parti les) it an be shown (as Pippard
puts it: with a fair degree of rigor - enough to satisfy most physi ists but few pure math-
emati ians) that these their general properties whi h an be treated as ma ros opi ones
follow only from statisti al averaging being to a large extent independent of the details of
the mi ros opi dynami s do indeed obey the laws of TMD.
Of ourse statisti al physi s, apart from reprodu ing thermodynami s and providing
inputs whi h otherwise (within purely thermodynami al treatment) would have to taken
from experiments, allow to ask and investigate theoreti ally questions whi h are entirely
beyond the s ope of lassi al thermodynami s. Here belong for example u tuations, their
spatial orrelations, riti al exponents ( hara terizing ontinuous phase transitions), et .
One an also go to kineti theories and onsider theoreti ally the pro ess of attaining
equilibrium.

One an think therefore, that lassi al thermodynami s whi h was mainly developed when
the ma ros opi stru ture of matter was not yet investigated fully and was even questioned
is already passé, has be ome obsolete at least for those who want to inquire into the
deepest mysteries of the physi al world and to dis over its most fundamental laws (laws
of quantum gravity for example). But

• Frequently, and almost as a rule in applied s ien es, it is of primary interest to know

4
 relations between properties of substan es than to know from whi h fundamental
rules these properties follow

• Thermodynami s greatly redu es the number of properties whi h have to be deter-

mined experimentally or have to be omputed theoreti ally employing the statisti al
physi s methods (as has been said: statisti al physi s will give ertain fun tions -
thermodynami al potentials - from whi h all follows thermodynami ally)

• As said, employing statisti al physi s to real physi al systems one is for ed to make
approximations or use simplied models; knowledge of thermodynami s allows us
to tell whi h of the obtained properties are general and whi h are valid only in
parti ular models

• It is pra ti ally impossible to analyze with methods of statisti al physi s very om-
plex systems (e.g, biologi al ones), whereas the simple rules of thermodynami s still
allow to make predi tions on erning su h systems and understand (even if only in
very general terms) su h systems

• It sometimes turns out (unexpe tedly) that thermodynami s has something to do

with the deepest fundamental mysteries - vide entropy of bla k holes

• Last but not least, the less pra ti al reason: the development of thermodynami al
ideas has a formal elegan e whi h is ex eedingly satisfying aestheti ally (Pippard
again). It approa h the ideal of mathemati al rigor (hen e attempts to mathemati-
ize it - we will not go this way!) loser than any other bran h of natural s ien e;
furthermore, its histori al development (fas inating as almost all history of s ien e,
but we have to omit it) and notions it introdu es (entropy!) have be ome part
of ulture - for all these reasons it should be an important part of edu ation of a
s ientist (and any physi ist).

Und somit fangen wir an - as says Thomass Mann in his Zauberberg.

Two more quotations for the good start:

• Thermodynami s is a funny thing: when you learn it for the rst time, you don't
understand it at all. On the se ond approa h you have the impression of under-
standing it all, ex ept for a few small details. At the third approa h you already
know that you don't understand it, but when you get a ustomed to this, you begin
not to are about. Arnold Sommerfeld

• Every mathemati ian knows that understanding of an elementary ourse of thermo-

dynami s is impossible. Wladimir Igorievi h Arnold

Basi notions

• Thermodynami system - this portion of the Universe whi h is sele ted for in-
vestigation. A system an be

5
 • Simple - a single homogeneous body or a substan e; apart from the number(s)
(single n in the ase of hemi ally pure systems - e.g. pure water - and n1 , . . . , nr
in the ase of a homogeneous mixture of r omponents - e.g, a mixture of phenol
and water) of moles of its material onstituents (some systems, e.g. the elmg eld,
are not hara terized by this variable) only two parameters are needed to fully
hara terize its equilibrium state - the most widely used example (the working
horse of thermodynami s) is the gas (usually the perfe t one) and the variables are
then p and V (at this stage), but one an also onsider a paramagneti substan e
with the variables M (total magnetization) and H0 or diele tri materials with the
variables P (total polarization) and E (when the volume ee ts an be negle ted,
eg. at p ≈ 0), or

• Not simple - a single homogeneous body or a substan e; more than 2 parameters

are needed; it will be seen later that the number of variables (parameters) is equal
1 plus the number of ways a quasistati work an be performed reversibly on the
system.

• NonHomogeneous systems onsist of many homogeneous parts (e.g. two on-

tainers with same gas or dierent gases). Sometimes a single nonhomogeneous body
(e.g. a gas in the gravitational eld) an be mentally split into small parts whi h
an be treated as homogeneous thus allowing to apply to it thermodynami s.

• Under ertain onditions a given homogeneous system or may split into Phase(s).
A phase is a homogeneous part (this means that splitting in reases the number of
parts of the system) with denite boundaries. A phase may be hemi ally pure
(one n) or be a mixture - e.g. the phenol-water mixture an split into phases with
dierent on entrations in ea h.

What is not in luded in the system onstitutes its surrounding. Less generally, as
the surrounding one takes into a ount those parts of the Universe whi h may somehow
inuen e the system under study (here the physi al intuition is ne essary). In thermo-
dynami al onsiderations the surrounding is usually taken to onsists of sour es of work
and heat whi h may be ex hanged with the system

Boundaries separating the system from its surrounding an be natural as in the ase of a
water droplet or, more frequently, they are arti ial or just mental onstru tions, and are
generally alled walls. They play the role of onstraints a studied system is subje t to.
Walls play ru ial roles in various thermodynami al reasonings They an be dierent and
an allow or inhibit dierent kinds of intera tions of the system with its surrounding. For
instan e, a rigid wall prevent performan e of me hani al work on the gas en losed by it.
Walls an also allow or inhibit ex hange of matter (of all kind or of only parti ular kind)
between the system and its surrounding.
Of parti ular interest for many thermodynami al reasonings are adiathermal walls.
Normally we would say su h walls inhibit thermal intera tion of the system with its
surrounding but sin e we (o ially) don't know yet what heat is, to explain this notion

6
it is better to use the words of Pippard who says: the walls of dierent vessels dier
onsiderably in the ease with whi h inuen es from without may be transmitted to the
system within. Water within a thin-walled glass ask may have its properties readily
hanged by holding the ask over a ame or putting it into a refrigerator; or the hange
brought about by the ame may be simulated (though not so easily) by dire ting an
intense beam of radiation onto the ask. If, on the other hand, the water is ontained
within a double-walled va uum ask with silvered walls (Dewar vessel), the ee t of the
ame or refrigerator or radiation may be redu ed almost to nothing. (...) it is not a
very daring extrapolation to imagine the existen e of a vessel having perfe tly isolating
walls, so that the substan e ontained within it is totally unae ted by any external
agen y. (ex ept the gravitational eld). Walls whi h do not have this property are alled
diathermal Two systems onta ted with one another through su h a wall are said to be
in thermal onta t.
One should add that walls adiathermal in the above sense an allow for some kinds of
work to be done on the system, e.g. ele tri ally (by passing a urrent through a resistant
wire inserted in the system), or by stirring.

Thermodynami variables omprise dire t observables (volume V, pressure p magneti-

zation M magneti eld strength H0 , et . and new thermodynami al quantities like
temperature T , hemi al potential(s) µ, enthalpy H , entropy S ... All they fall into two
lasses intensive ones of essentially lo al hara ter (p, H0 , µ) and extensive ones har-
a terizing the system as a whole (mass m, volume V , internal energy U ). If the onsidered
system have the property of extensiveness (most onsidered systems do have it), it is
onvenient to operate with spe i quantities and molar quantities whi h are the extensive
quantities referred to a unit of mass or to one mole, respe tively. (sometimes, somewhat
in orre tly, molar heat apa ities, et . are also alled spe i ). I denote molar quanti-
ties by a lower ase hara ter, and true spe i ones by a tilde. A quantity whi h value
depends uniquely on the (equilibrium) state of the system is alled a state fun tion.

Equilibrium
All thermodynami s rests on the following fa t: any system shielded by adiathermal walls
and not subje t to any external inuen es tends towards and eventually rea hes a state
in whi h no further hange (of its ma ros opi ally dened hara teristi s) is per eptible,
no matter how long one waits. This state is alled the thermodynami al equilibrium
state. It is the state of equilibrium whi h an be hara terized by only a small number of
quantities (intensive or extensive). One also requires that there be no ma ros opi ows
in the system like a steady ux of heat, or ele tri urrent passing through it. Water
owing down a tube is not stri tly speaking in equilibrium in equilibrium - this is the so
alled steady state - but we will see that to some extent TMD an be applied to it.
If the system is omplex (has several parts or some heat or work sour es are in luded
in it), in the approa h to equilibria some hanges in individual parts o ur - they an
ex hange heat or ex hange work between them (e.g. a mer ury in glass thermometer
inserted in water hanges but then it stabilizes) but eventually attain equilibrium. If the
external onditions are hanged, or walls through whi h various parts of a omplex system

7
 ommuni ate, that is the onstraints to whi h the system is subje ted has hanged, a
new equilibrium state is attained and one of the main roles of thermodynami s is to
determine this state (Callen's point of view)

Similarly as in me hani s one an ontemplate dierent kinds of thermodynami al equi-

libria.

• stable me hani al - tmd analog: a pure gas at uniform temperature and pressure
in a ylinder - upon small disturban e (e.g. the pressure and density of a gas made
somewhat nonuniform by a short external perturbation) it rea hes ba k the same
equilibrium state.

• unstable me hani al - no analogy

• neutral me hani al - a mixture of water and its vapour (saturated) in a ylinder at

spe ial T and p - moving the piston the proportions an be hanged

• metastable me hani al - super ooled vapour or a mixture of hydrogen H2 and oxygen

O2 ; look as fully stable - an remain un hanged per eptibly for a very long time
∼ 10100 years!, and an even by subje t to small disturban es (e.g. of pressure or
temperature) - but the ee ts of a nu leation enter or a spark show they are in
fa t not in equilibrium states. For many pra ti al purposes an be, if not arti ially
stimulated, treated as being in full equilibrium.

In fa t no unstable equilibrium exists in me hani s too: it is a purely mathemati al

on ept, for it relies on many idealizations (pointlike material points, negle t of internal
stru ture of bodies, negle t of minute external inuen es). Real physi al systems an
have only metastable equilibria but the range of displa ements whi h do not make them
hanging the state may be so narrow that we lassify them as unstable.
It will be seen that similarly to me hani s, also in thermodynami s equilibrium states
of the system realized in spe i onditions (spe i onstraints) minimize or maximize a
suitable fun tions alled thermodynami s potentials (examples are entropy, Helmholtz
free energy, Gibbs potential, et ).

Equilibrium in thermodynami s is never stati : mi ros opi examination of parti les

or mole ules in the system always reveals they are in a state of ontinuous agitation
(Brownian motion); lo al density of a uid u tuates a little around the mean value
If one waits long enough - theoreti ally, be ause the times involved are longer than the
Universes lifetime! - one might observe sizeable departures from the mean state of the
system; e.g. a gas lling 1 m an spontaneously ontra t to half of this volume and
−4
then in less than 10 se . revert to its average density, but this an o ur - it an be
1019
estimated - on e in 10 years. Of ourse the numbers quoted here annot be obtained in
pure thermodynami s, one needs to delve into the mi ros opi dynami s, that is, go over
to the statisti al physi s (or even the kineti theory of gases). And from it, in addition
to being able to ompute su h numbers, one draws a lesson that su h u tuations should
be treated as immanent part of the equilibrium state. Anyway, for most purposes after

8
reasonably short time one an treat any system as having attained equilibrium state. If
the onsequen es of treating it as su h are not orroborated by experiment, we must revise
this assumption - evidently the system has not attained equilibrium or its u tuations
play a ru ial role in the phenomenon we want to explain.

0TMDL and Temperature

To introdu e and dis uss this on ept, fundamental to TMD, it is onvenient to on-
entrate rst on an espe ially simple system - a homogeneous uid (liquid or gas). Its
simpli ity derives from the fa t that the shape of the ontainer is irrelevant - deformation
of the shape without any hange in volume do not require work. The shape of a solid an
only be hanged by the appli ation of stress and thermal properties are then usually also
ae ted.
We a ept a fa t of experien e that the equilibrium state of a xed mass of a uid
is ompletely spe ied by its volume V and pressure p (we assume it has no ele tri or
magneti properties or that elemg elds are absent altogether).
We an x the volume V of the uid kept in a ylinder and adjust p to any desired value
by pla ing the uid in an oven or a refrigerator or any other means. The important point
is: whatever the pro ess by whi h the given values of V and p have been rea hed, the nal
state is always the same: same olour, smell, sensation of warmth, thermal ondu tivity,
vis osity et . Anything ma ros opi that an be measured as the hara teristi of this
system an be treated as a unique fun tion of V and p.
Take now any two simple systems (two uids for deniteness). If they are isolated and
rea h equilibrium states separately, and then are brought into thermal onta t through an
diathermal (rigid, so that me hani al work is not possible) wall, hanges in general will
be observed to o ur in both until they rea h, as a ompound system, a new equilibrium
state. So we say they are now in thermal equilibrium with one another.
On the basis of our sensory experien e we say that this is so be ause initially they
in general do not have same temperature. But this word has no ontent yet. What is
important is that two systems may be separately in equilibrium but not in equilibrium
with one another.
Consider now two masses of uids, adjusted to have V1 and p1 and V2 and p2 , respe -
tively. They in general will not be in equilibrium with one another unless V1 , p1 . V2 and
p2 are orrelated: x V 1 , p1 and V2 then p2 has to be adjusted in order the two systems
are in equilibrium with one another (no hanges in their states are observed when they
are brought into thermal onta t). So, their equilibrium requires that the relation

F (V1 , p2 , V2 , p2 ) = 0 , (1)

be satised. The form of the fun tion F an be determined experimentally and depends,
of ourse, on both systems. To introdu e the on ept of temperature one has to show
that this relation always takes the form

φ1 (V1 , p2 ) = φ2 (V2 , p2 ) , (2)

with the fun tion φ1 being the property of only the rst system and φ2 of only the se ond
one. This is shown relying on

9
0TMDL
If of the three bodies A, B and C , A and B and are separately in thermal equilibrium with
C, then A and B are also in thermal equilibrium with one another.
It is useful to have it stated somewhat dierently:
Converse 0TMDL
If the three bodies A, B and C are pairwise in thermal onta t through appropriate diather-
mal walls and the whole system in equilibrium then any two of them taken separately are
also in equilibrium with one another.
One an illustrate this with an example: let C be a mer ury in glass thermometer
in whi h Hg is roughly at zero p (the thermometer tube is eva uated) so that h - the
height of Hg in the tube fully spe ies its state. If the height is the same when we put
it in thermal onta t rst with the body A and then with the body B, then nothing will
happen if A is brought into thermal onta t with B.
So let's pass to the argument. We rst give a physi al argument and only then sket h the
mathemati al reasoning (whi h is ni e, so worth showing).
Take two masses of uids S - the standard one whi h is kept at xed VS and pS , and T -
the one whi h is being tested. Vary VT and pT in su h a way as to maintain equilibrium
between T and S. In the plane (VT , pT ) this
determines a urve whi h will be alled
isotherm. The isotherm does not depend on the standard body, for if S ′ is taken whi h
is in thermal equilibrium with S, by the 0TMDL the same isotherm of the system T will
be obtained.
Changing now VS and pS we an produ e many isotherms of the tested body T. One an
now introdu e a system (however arbitrary) of labeling these isotherms by numbers t, so
that we dene a fun tion (whi h need not be analyti at this stage)

φT (VT , pT ) = t ,

and we all t the empiri al temperature. On e this is done, we an produ e isotherms

of all other bodies taking the test one for the standard. But if a onsisten y is to be
a hieved, there is no more freedom in labeling isotherms of other bodies: with ea h su h
body must be asso iated a fun tion φbody depending on the parameters fully hara terizing
equilibrium states of that body and this fun tion must take values equal φT (VT , pT ) if the
body is in equilibrium with the test body at VT and pT . in this way one establishes the
existen e of another state fun tion - (empiri al) temperature. From this it also lear that
if more than two parameters are needed to ompletely hara terize equilibrium states of
a body, say x1 , . . . , xr , its isotherms are not urves, but hypersurfa es of dimension r−1
(or odimension 1) determined by the equation

fbody (x1 , . . . , xr ) = t . (3)

Now the mathemati al reasoning. Consider three uids A, B and C. If A and C are in
equilibrium, then

F1 (VA , pA , VC , pC ) = 0 , so pC = f1 (VC , VA , pA ) .

10
Similarly, if B and C are,

F2 (VB , pB , VC , pC ) = 0 , so pC = f2 (VC , VB , pB ) .

It then follows that

f1 (VC , VA , pA ) = f2 (VC , VB , pB ) .

But a ording to 0TMDL A and B also are in equilibrium so

F3 (VA , pA , VB , pB ) = 0 ,

and f1 (VC , VA , pA ) = f2 (VC , VB , pB ) must be equivalent to this. But VC does not enter, F3 ,
so it must drop out from f1 = f2 . This is so if f1 (VC , VA , pA ) = φ1 (VA , pA )ψ(VC ) + η(VC )
and f2 (VC , VB , pB ) = φ2 (VB , pB )ψ(VC ) + η(VC ) (with some universal fun tions ψ and η ).
That 0TMDL enfor es dropping out of VC an be shown more formally as follows. From
F3 (VA , pA , VB , pB ) = 0 one an get pB = f3 (VB , VA , pA ) and write the equality f1 = f2 in
the form

f1 (VC , VA , pA ) = f2 (VC , VB , f3 (VB , VA , pA )) ≡ pC ,

whi h learly shows that the right hand side must be independent of the variable VB . So
we an x it and forget it hanging the notation to

f3 (VB , VA , pA ) ≡ gA (VA , pA ) .

so now the relation pC = f2 (VC , VB , f3 (VB , VA , pA )), from whi h VB , as argued, drops out,
an be written as

pC = f2 (VC , gA (VA , pA )) ,

and disentangling from it gA write it in the form

gA (VA , pA ) = gC (VC , pC ) .

In the analogous manner one obtains two other relations

hA (VA , pA ) = hB (VB , pB ) , and tB (VB , pB ) = tC (VC , pC ) ,

but we don't know yet, e.g. how the fun tion tB (VB , pB ) relates to hB (VB , pB ). One has
to show that these three relations between 6 fun tions an be redu ed to three relations
involving only 3 fun tions ea h depending on only one pair of variables V , p.
To this end from gA (VA , pA ) = gC (VC , pC ) we get pC = ψ(VC , VA , pA ) and put it into
tB (VB , pB ) = tC (VC , pC ):

tB (VB , pB ) = tC (VC , ψ(VC , VA , pA )) .

11
We now take VA , VB , VC i pA for independent variables - in other words we treat pB as
determined by these four) and dierentiate both sides of the above equality w.r.t. VB .
Be ause the right hand side is independent of VB , this gives

∂tB ∂tB ∂pB

+ = 0.
∂VB ∂pB ∂VB
In turn, dierentiating w.r.t. VB the relation between the fun tions hA and hB ( ontinuing
to treat VA , VB , VC i pA as independent variables) we get

∂hB ∂hB ∂pB

+ = 0.
∂VB ∂pB ∂VB
Eliminating now the derivative ∂pB /∂VB we nd

∂tB ∂hB ∂tB ∂hB ∂(tB , hB )

− ≡ = 0.
∂VB ∂pB ∂pB ∂VB ∂(VB , pB )
Vanishing of this Ja obian identi ally means in ee t that the mapping (VB , pB ) −→
(tB , hB ) ∈ R2 is degenerate (is of rank 1, instead of 2), or - saying it more a essibly to the
2 2
audien e - the image in R of this mapping (dened on R ) is a one-dimensional urve,
and not a two-dimensional domain. This in turn means that the fun tions hB (VB , pB )
and tB (VB , pB ) are not independent: there must exist a relation r(tB , hB ) = 0 whi h
determines this urve and this relation an be again inverted to give hB = χ(tB ) that
is,hB (VB , pB ) = χ(tB (VB , pB )). This an now be exploited in the relation linking the
−1
fun tions hA i hB : if we dene a new fun tion tA ≡ χ (hA ), this relation takes the form

tA (VA , pA ) = tB (VB , pB ) .

And this, ombined with the equality tB (VB , pB ) = tC (VC , pC ) yields tA (VA , pA ) = tC (VC , pC ).
This means that the third equality, gA (VA , pA ) = gC (VC , pC ), must be equivalent to this
one, that is gA (VA , pA ) = r(tA (VA , pA )) = r(tC (VC , pC )) = gC (VC , pC ).
S ales of temperature
It is be ause of the freedom in labeling the isotherms of the test body (in the rst reasoning
presented) or be ause one an always use t̃A = f (tA ), t̃B = f (tB ) and t̃C = f (tC ) (in the
mathemati al reasoning) the quantity t (a state fun tion) is alled empiri al. We will see
that TMD itself provides a mean of dening the absolute temperature T with the s ale
being the only freedom in its denition. The question is therefore how to relate empiri al
temperatures dened by dierent thermometri bodies to the absolute temperature.
Usually one hooses a thermometri body with suitable properties and labels its isotherms
by hanging x - one of the parameters hara terizing its equilibrium states, while keeping
the other parameters xed. The empiri al temperature t an be then taken to be related
to x as x = f (t) with f (·) an arbitrary monotoni fun tion. A parti ularly simple fun tion
is f (t) = at. There are then two ways of xing the proportionality onstant a. Either one
hooses two points and xes the number of units between them or one as ribes a on rete
value t to one parti ular point.

12
For instan e, one takes the Mer ury in glass at pressure p = 0, sets hHg = aHg tHg (hHg
being the height - above some onveniently hosen referen e level - of the olumn of
Mer ury in eva uated glass tube at p = 0) and determines aHg by requiring that there
be 100 degrees between the melting point of i e and the boiling point of water, both at
5
normal pressure 1.013 × 10 Pa (the famous 1013 HPa ≡ 1 atm). However, if one takes
another thermometri substan e, e.g. the ethyl al ohol and onstru ts h̃ = ãt̃ in the same
way, both thermometers an be made (by the appropriate hoi es of the referen e levels of
the height hHg andh̃) to give tHg = t̃ at the two hosen referen e points but will in general
dier (even if not too mu h in pra ti e) at all other point (tHg 6= t̃). This is be ause these
are two dierent empiri al temperatures.

Another hoi e of the thermometri substan e is gases at very low pressures. This is
simple be ause the isotherms of gases under this ondition are simple: pV = onst. This
is the Boyle-Marriotte law whi h is satised to a very good a ura y by real gases at
su iently low pressures. So one an set pV = nRt and determine nR as above. It then
turns out that t and t̃ obtained with dierent gases (at su iently low pressures) are
nearly the same not only at the referen e points but in a wide range of temperatures.
(Moreover, it happens experimentally that if the empiri al temperature t̃ is xed by the
Mer ury in glass thermometer by the rule t̃ = ãh, then over a wide range of temperatures:
t̃ ≈ t.) This is of ourse be ause the temperature dened in this way using the perfe t
gas (a theoreti al onstru t) is exa tly proportional to the absolute temperature T (the
one determined by the TMD itself ) and all gases at su iently low pressures behave as
the perfe t gas.
Before 1954 one dened the temperature s ale using the gas thermometry (at p → 0) as
des ribed above using the i e melting point and the water boiling point. Be ause it is
easier to reprodu e in laboratories the water triple point the denition was hanged and
now the absolute temperature s ale is xed by as ribing to this point (whi h orresponds
to p = 611.73 Pa) T = 273.16 K (exa t value by denition). This has the ee t (the
value 273.16 has been hosen to get this!) that between the water boiling point and the
i e melting point at 1 atm the temperature dieren e is (very nearly) 100 K and that
these points orrespond to 273.15 and 373.15 K, respe tively (although only within some
a ura y: more pre ise measurements may reveal small departures from these numeri al
values).

It should be noted that be ause the perfe t gas s ale relates dire tly to the absolute
temperature, the determination of temperature almost always is based on gas thermom-
etry. In general alibrating thermometers with respe t to the absolute temperature s ale
is too long a story to be told here.

On e the absolute temperature is established, the ommonly used Celsius s ale is dened
0
as t = T −273.15. On this s ale the triple point orresponds to 0.01 C. This approximately
(to a quite good a ura y) oin ides with the old Celsius temperature, whi h is now alled
the entigrade s ale, dened by the Mer ury in glass thermometer by h = at + b and
0 o
as ribing 0 C to the i e melting point (and therefore 100 C to the water boiling point).
Gas thermometers are in onvenient and di ult to use when high a ura y is required,

13
so they are used only to measure absolute temperature T. Other kinds of thermometers
are used - the hoi e depends on onvenien e and sensitivity required. To alibrate these
other thermometers w.r.t. the absolute temperature a number of referen e points have
been measured very a urately. Among these are: the triple point of Hydrogen (13.81 K),
triple point of Oxygen (54.361 K), the melting point of Zin at 1 atm (692.73) melting
point of Gold (1337.58 K).
Thermometers based on expansion of liquids (Mer ury, Ethyl Al ohol, Pentane) over
the range of−hundredo C up to + a few hundreds o C.
Resistant thermometers are based on the variation of ele tri al resistan e of a metal
with temperature over even larger range. E.g. Platinum is easy to purify, has high
o
melting point (1770 C); it is very a urate between 70 K and 1500 K.
Thermo ouple thermometers use variation of the e.m.f. with T. One keeps one jun -
tion at xed temperature and the e.m.f depends on the temperature of the other one.
Small voltages have to be measured - this make them di ult to work with if high a u-
ra y is needed but su h thermometers an be miniaturized and respond qui kly to hanges
of T.
Condu tivity of semi ondu tors. Current arriers must be thermally ex ited and the
semi ondu tor ondu tivity is ∝ exp(−ε/T ). They are good thermometers from well
−5
below 1 K up to ∼ 600 K. At low T su h thermometers have sensitivity of order 10 K
−3
and of order 10 K at room temperatures.
Carbon resistors. Useful below 20 K. Below 10 K also have sensitivity of order 10−5
K.
4 3
Below 5 K to somewhat below 1K liquid He is used. Between 1 K and 0.3 K He is
used. In both ases T is found by measuring the vapour pressure.
For even lower T one relies on paramagneti salts. Their sus eptibility goes like a/T
(Curie law)
Above the Gold melting point only measurements of radiation emitted by hot bodies
are used (radiation pyrometers).
The International Pra ti al Temperature S ale is the set of a urately measured ref-
eren e points plus a set of thermometers whi h should be used to interpolate between the
referen e points together with the interpolation pro edures,

Finally it should be stressed that the mere introdu ing the notion of temperature - as an
indi ator of whether two bodies will be in thermal equilibrium, if they are brought to
a onta t through an diathermal wall - does not mean yet orrelation of its values with
the sensation (experien ed by our bodily senses for example) of warmth and oldness.
Nothing as yet guarantees that higher (lower) t orresponds to what we feel as hotter
( older). Of ourse, one an arrange the perfe t gas s ale to ree t the degree of hotness
but this annot be demonstrated rigorously before dening the meaning of the terms
hotter and  older operationally that is, in a way whi h is not based on our subje tive
physiologi al sensation. And this requires to investigate rst what heat is.

14
 LECTURE II (TMD)

Relying on the fundamental fa t that all systems, when isolated (adiathermally by rigid
walls against inuen es from without, on whi h no other kind of work is performed by
any means) sooner or later attains an equilibrium state in whi h no hange in its state is
per eptible ma ros opi ally and on the 0TMDL stated in two equivalent ways), we have
inferred (by using a physi al as well as a mathemati al reasoning) the existen e of a new
(in addition to the obvious me hani al ones like pressure p or volume V) state fun tion -
the empiri al temperature t.
If the empiri al temperatures tA and tB of two isolated systems A and B are equal,
tA = tB , then no hange in their states will be observed when they are onta ted with one
another through a rigid (preventing their me hani al onta t) diathermal wall imperme-
able for matter. The two systems are then said to be in thermal equilibrium with one
another. (Noti e, that tA = tB is not su ient for full thermodynami al equilibrium
whi h requires in addition that their pressures be the equal as well as hemi al potentials
- to be introdu ed in due ourse - so that when the two systems are onta ted through
a movable wall whi h also permits transfer of matter between them, no hange in their
states will observed.) The empiri al temperature plays therefore the role of an indi ator
of a possible thermal equilibrium between dierent systems. There is a huge arbitrariness
in the denition of the empiri al temperature t: if an be always repla ed by t̃ = f (t)
where f (·) is any monotoni fun tion. But on e this arbitrariness of t has been xed (by
adopting some its denition - however arbitrary - by hoosing a thermometri body), it
is a single valued fun tion of the parameters x1 , x2 , . . ., like p, V , and other ones, like the
amount of matter in the system represented by the number of moles n1 , . . . , nr of its r
omponents needed to fully hara terize the equilibrium state:

t = t(x1 , . . . , xs , n1 , . . . , nr ) . (4)

Su h a fun tional relation is alled the equation of state of the system (of the body, of
the substan e). Naturally, thermodynami s by itself does not determine its form and it
has to be determined experimentally or else derived using the statisti al physi s approa h.

We re all also that as yet no orrelation between higher (lower) value of the empiri al
temperature and the subje tive sensations of hotness ( oldness) has been established.
This requires dening pre isely the notion of heat. We shall do it now, ontinuing to
onsequently develop thermodynami s as a phenomenologi al theory (as opposed to the
Callenian thermodynami s whi h is - somewhat absurdly to my taste - onstru ted as a
dedu tive theoreti al system).

Internal energy
We begin by introdu ing the notion of internal energy. Let us onsider the histori
experiments performed by Joule whi h originally were intended to measure the me hani al
equivalent of heat (but for us the word heat has as yet no ontent, so we will interpret
this experiment somewhat dierently).

15
 One realization of this sort of experiment an be the original Joule's paddle-wheel
immersed in an (as ideally as possible) adiathermally isolated alorimeter ontaining a
mass of a liquid (e.g. water). A measurable me hani al work an be performed on this
system by rotating the wheel by an angle α applying to it a known ouple D: W =
αD . Alternatively one an let a known mass m fall down the height h in the Earth's
gravitational eld g propelling the wheel: measuring its nal speed w one gets the work
mgh − mw2 /2 done on the system (the liquid) by the devi e. The state of the liquid is
found to hange (its temperature hanges, its pressure hanges) as a result of performing
the work.
Alternatively, a resistive wire an be inserted in the alorimeter and a known urrent
I passed through it during a period ∆τ . If the potential dieren e a ross the wire is E,
the (ele tri al) work done on the liquid equals W = E I∆τ .
Similar experiments, employing dierent kind of dire tly measurable work an be
performed. The important fa t is that the same temperature hange is obtained by the
performan e of the same amount of work. Pippard (whom I follow here) stresses that
none of su h experiments should be interpreted as transferring heat to the system: So
long as we take a ount only of what is observed, the dedu tion to be drawn from the
experiment is (...):
If a state of an otherwise isolated system is hanged by the performan e of work, the
amount of work needed depends solely on the hange a omplished, and not on the means
by whi h the work is performed, nor on the intermediate stages through whi h the system
passed between its initial and nal states (I would add here equilibrium states - see the
footnote below).
This is the 1TMDL as applied to adiathermally isolated systems.

Example: the state of a gas adiathermally isolated (against any inuen es from without)
hanges from A to B. Path 1: an amount of ele tri work is rst performed iso hori ally
on it, until it rea hes the state Y ; then the gas is adiathermally expanded performing some
measured work and attains the state B ; path 2: the gas is rst adiathermally expanded
rea hing the state X and doing on the way some measured work and then an amount of
work is done on it by, say, a paddle-wheel me hanism and it attains the same nal state
B. Then the statement is that

WA→Y →B = WA→X→B .

The pro esses onsidered here need not be reversible (a word not dened yet) and at
the intermediate stages the system may not be in equilibrium (parameters like pressure,
temperature of the system may not be dened on these stages). It is only required that
the system be adiathermally isolated, the works done on/by it measured, and the initial
1
and nal states be equilibrium states.

1 We require the states A and B to be equilibrium states be ause only then they an be spe ied by
giving the values of a few parameters only; the statement remains true also if one or both these states
are not equilibrium states but it would then be harder to be sure that the same state B has been rea hed
on both paths.

16
 It should be stressed that experiments of this kind ( he king arefully that indeed
the same work is needed to produ e the same hange) have never been really performed,
owing to the rapid universal a eptan e of the just stated form of 1TMDL. But its manifold
onsequen es are so well veried in pra ti e that it should be onsidered to be established
beyond any reasonable doubt (Pippard again).

Relying on the 1TMDL applied to adiathermally isolated systems one an introdu e in-
ternal energy U whi h by onstru tion is a state fun tion. If an adiathermally isolated
system is brought from the state A to another state B by performan e on it an amount
W of work, its internal energy shall be said to have in reased by

∆U ≡ UB − UA = W on adiathermal paths . (5)

The 1TMDL ensures that ∆U is determined by the states A and B only and not on the
path onne ting these two states. So xing for every system some referen e state R and
assigning to it (arbitrarily) an internal energy U0 , internal energy of any other state A of
this system is uniquely determined:

UA = U0 + WR→A on adiathermal paths . (6)

In reality it may prove di ult to measure WR→A dire tly but, again owing to 1TMDL,
it an always be measured indire tly: a suitable roundabout path an in prin iple always
be devised to get from a state A to another state B or the other way around, that
is from B to A. In all textbooks it is remarked at this point that as a onsequen e of
2TMDL (whi h will be introdu ed later) a given path onne ting the states A and B
may not ne essarily be tra ed in both dire tions (in given adiathermal onditions) but
to determine ∆U it is su ient that it an be tra ed in one way only. Sin e I always
had trouble to understand what this statement means, I advi e you to think again on the
paddle-wheel devi e: all the state A the state of the uid when the mass m is in the
higher position and B the state when the mass m is in the lower one. The path A to B
an be realized (in fa t in numerous ways). The path B to A evidently not and not only
with the help of the paddle-wheel devi e but by any other means, so long as the liquid is
adiathermally isolated.

In any ase, the important thing is that dieren es of internal energies in various states
and therefore, this quantity up to an additive onstant, an be measured by measuring
only me hani al or ele tri al works (not heat) and that U is a fun tion of state, so
it an be expressed (in ase of equilibrium states) as a fun tion of the parameters needed
to spe ify this state:

U = U(p, V ) , or U = U(p, t) = U(p, V (t, p)) , or U = U(t, V ) = U(p(t, V ), V ) ,

in the ase of a simple uid, et . In this way, on the manifold of equilibrium states
of the system (parametrized by some onvenient state variables) we superimpose the
network of internal energies. Of ourse, mi ros opi ally this internal energy (by

17
denition onsidered in the body's rest frame) onsists of kineti and intera tion energies
of the matter onstituents. (tu mo»e o addytywno± i U)
It also follows, sin e works an be added, that internal energy is an additive quantity:
if U1 and U2 are the energies of two bodies taken separately, the internal energy U of these
bodies ombined equals U = U1 + U2 .
Heat as the work defe t and 1TMDL
On e a value of internal energy U is unambiguously (when the referen e state R is hosen)
assigned to any (equilibrium) state in the way sket hed above, one an onsider hanges of
the system during whi h it is not adiathermally isolated. It is then possible to a omplish
a hange of the system's state from A to B in more dierent ways and some of these ways
involve amounts of work whi h are dierent than ∆U = UB − UA . For instan e, in the
onsidered example of a simple uid taken from A to B the hange from A to Y an be
a hieved by lightning a Bunsen burner under the (diathermal) ask ontaining the uid
and this ertainly does not involve performing work. So in this ase ∆U = UB − UA 6=
WA→Y →B . We then dene the quantity Q by

Q = ∆U − WA→Y →B , (7)

and all it (somewhat misleadingly, but this should ause no harm - we are already far
from the dis ussions of XIX entury!) heat taken by the uid on its way from A to B.
More properly one should all Q energy transferred to the uid in the form of
heat. In other words - and this is my great ontribution to tea hing thermodynami s! -
Heat is the defe t of work
by obvious analogy to the binding energy dened as the defe t of mass in nu lear physi s
(nu lei weight less than protons and neutrons out of whi h they are made and this de it,
alled defe t, multiplied by c2 , is identied with their binding energy).
Heat dened in this way has all properties whi h are usually attributed to it (and whi h
in the past supported treating heat as a kind of indestru tible uid owing from one body
to another). 1) when absorbed by a body, it hanges the state of this body (obvious -
existen e of diathermal walls proves that hanges of states are not always due to work), 2)
may be onveyed from one body to another by ondu tion, onve tion and radiation (ob-
vious - even va uum does not entirely inhibit hanges ee ted by other means than work
- hanges aused by radiation an be inhibited by walls perfe tly ree ting ele tromag-
neti waves) 3) in alorimetri experiments (in whi h one measures heat by the method
of mixtures - this is what all problems in undergraduate physi s ourses are about: one
puts into a alorimeter bodies whi h then ex hange heat and one is asked to ompute the
nal temperature, or given some data, the heat apa ities et .)
Only the property 3 requires some brief justi ation: onsider a typi al alorimetri
experiment in whi h two bodies, 1 and 2, at dierent temperatures t1 6= t2 are brought
into a thermal onta t within a alorimeter (a Dewar vessel) whose walls are adiathermal
and rigid (no work an be performed in the bodies within the alorimeter by external
agents). Therefore the total hange ∆U of the system's internal energy must be zero

∆U = ∆U1 + ∆U2 = 0 . (8)

18
But

∆U1 = W1 + Q1 , ∆U2 = W2 + Q2 , (9)

and sin e the only work that ould have been performed on the bodies was the work they
performed on ea h other, so that W1 = −W2 , one learns that

Q1 + Q2 = 0 . (10)

In alorimetri experiments (but not in general!) heat is onserved.

2
Thus we nally arrive at the general 1TMDL in the form

∆U = W + Q . (11)

In words: energy is onserved if heat Q is in luded and re ognized as a form of energy

transferred from one system to another. Callen (whose elebrated textbook on thermody-
nami s will be exploited later) des ribes the dieren e between energy transferred through
me hani al work and in the form of heat as the results of ouplings of dierent types of
degrees of freedom: me hani al work is due to oupling to external agents of the few
globally dened ma ros opi degrees of freedom (like the position of its enter of mass, or
volume) of the system, while heat is due to the oupling of mi ros opi degrees of freedom
of the system and its environment (this pi torial explanation does not, however, seem to
apply to work performed in magnetizing or polarizing ele tromagneti ally a tive bodies).

The equivalen e of heat and work an be illustrated by the example of a gas en losed
in a ylinder with piston. The system is here the gas and the ylinder and the whole
system is adiathermally isolated. If the gas expands doing some work W̄ on the piston
(not ne essarily reversibly - we still do not know what this word means) and next the
whole work W̄ is onverted into heat somehow (or just heat equivalent to this work is
used) and added ba k to the system, its internal energy will return to the initial value
(its state will not in general be the initial one).

Notions hotter,  older and their orrelation with the s ale of t

We an now onsider the problem of orrelating the introdu ed empiri al temperature
with the properties of oldness and hotness. In alorimetri experiments like the one just
onsidered, one body gains heat whi h the other body is loosing. This is alled heat
transfer though this does not imply existen e of heat as a physi al entity ( alori  or
ogiston) whose movement an be followed from one body to another.
In general if any two bodies are brought into thermal onta t in the onditions that
no work is performed on either, a transfer of heat will o ur (a ompanied by the hanges
of states of both bodies) unless their temperatures are equal in whi h ase they are in
thermal equilibrium with one another. In the following reasoning important will be the

2 Our onvention is that Q and W will always stand for heat and work added to the system; heat
and work extra ted from the system will be denoted Q̄ and W̄ , respe tively. Of ourse, any of these
quantities an be either positive or negative.

19
fa t, following from experien e, that the rate of the heat transfer may usually be varied
over a wide range, depending on the nature of diathermal wall through whi h the bodies
ex hange heat (the rate is a measure of thermal ondu tan e of the wall).

We adopt the denition ( ould one expe t it to be dierent than this?!) that the body,
all it A, whi h loses heat (negative QA ) is hotter and the body, all it B, whi h gains
heat (positive QB ) is by denition the older one.

So we now will show that the hierar hy of hotness ( oldness) dened in this way an be
onsistently orrelated with the s ale of temperature (the labeling of isotherms of bodies),
that is, that the values of t = t(p, V, . . .) of the empiri al temperature an be assigned to
isotherms in su h a way, that all bodies at temperature t2 will be hotter (in the sense
made pre ise above) than all bodies at temperature t1 if t2 > t1 .
This is done by redu tio ad absurdum. Let us assume this is not possible. Therefore it
should be possible to nd three bodies, A, B and C having temperatures tA and tB = tC
(so B and C are in thermal equilibrium with one another) and yet su h that A is hotter
than B while C is hotter than A.
It is then possible the break somewhat thermal equilibrium between B and C , varying
slightly (almost innitesimally) the temperature of B making it somewhat hotter than C
but still older that A (this should be possible be ause tA 6= tB = tC ; note we do not say
tA is greater or lower than tB = tC ).
If the three bodies are then brought to thermal onta t, heat will ow (in agreement
with the meaning of the words hotter and  older) from A to B, from B to C and
(be ause we have assumed C an from the beginning be hotter that A) from C to A.
By adjusting the diathermal walls (their ondu tion rates) separating the bodies it would
then be possible to establish a (dynami al) equilibrium of these three bodies. But this
would ontradi t the onverse of 0TMDL, for if any two of these bodies were separated
from the third one, they would not be in thermal equilibrium.
The on lusion must, therefore, be that it is possible to dene the empiri al temper-
ature so that if t2 > t1 than any body at t2 is hotter than any body at t1 .
In the reasoning we have assumed (analyzing the heat onservation in a alorimeter ex-
periment) we have allowed for the possibility that the two bodies do work on one another
(depending on other properties of the wall through whi h they are in onta t, while they
ex hanged heat between them). Therefore, if those parameters whi h together with t are
used to spe ify the states of these bodies remain onstant when the bodies are brought
into thermal onta t (e.g. p instead of V when the rigid diathermal wall separating them
is repla ed by a rigid movable one) The orrelation of t and hotness will still obtain. The
onsequen e of this is that absorption of heat by a body whose independent parameters
other than t remain onstant will always ause an in rease of t. Therefore the so alled
prin ipal heat apa ities (like CV , Cp ) hara terizing the body are always positive
(this will be important in establishing stability onditions of thermodynami systems) by
virtue of the operational denition of the notions hotter and  older and the ( onven-
tional) assignement of higher values of the (empiri al) temperatures to the hotter bodies.

20
Reversible and irreversible hanges (pro esses)
We have divided the hanges whi h systems an undergo into adiathermal ones during
whi h the system is adiathermally isolated and ones in whi h the system is not isolated
in this way. In this other, more general, kind of hanges ∆U = W + Q. We now
inquire, under whi h onditions in an innitesimal hange of the system's state from one
equilibrium state to another one the 1TMDL

dU = q + w , (12)

in whi h dU is an innitesimal hange (whi h is an exa t dierential) of the system's

internal energy and q and w are innitesimal (elementary as one alls them in thermody-
3
nami s) heat and work, an be written with w and q being forms (dierential one-forms
in thermodynami s alled Pfaan forms) on the spa e of the parameters x1 , . . . , xs har-
a terizing equilibrium states the onsidered system. Certainly, neither q nor w whi h
represent innitesimal heat and work added to the system in a hange in whi h its in-
ternal energy hanges by dU an be written as dfQ (x1 , . . . , xs ) of dfW (x1 , . . . , xs ), that
is, as exa t dierentials (or losed forms) be ause in nite hanges the heat and work do
depend on the path from the initial to the nal state. We will rst argue that under some
4
well dened ir umstan es w an be written as an inexa t form

ω̂W ≡ d¯W , (13)

so that in these onditions also q an be written as

ω̂Q ≡ d¯Q . (14)

Thus, if the ne essary onditions are met, one will have the right to write

dU = d¯Q + d¯W . (15)

(But, as we are going to diss uss in the next Le ture, owing to 2TMDL the hara ter of
the form d¯Q is dierent than that of d¯W .)
As usually it is onvenient to onsider a simple system, a gas en losed in a ylinder
tted with a movable piston of ross se tion area A. In the equilibrium state the for es
a ting on the piston must be balan ed so that the net for e is zero. These for es are: a)

3 We do not onsider here the possibility that separately w and q are large but mutually an el out
leaving only an innitesimal sum w + q.
4 Thermodynami al tradition requires the dierential forms whi h are not exa t dierentials of state
fun tions be written with the slashed d, d¯ (Kubo uses d′ W ), instead of ω̂ (whi h is the mathemati al
notation). Be ause when I started to write notes for lasses and problems I did not know how to produ e
this symbol in LaTeX, in some pla es dW or dQ may still be found (instead of d¯W and d¯Q). While I
will do my best to nd these pla es and repla e the symbols, I hope that students who en ounter dW or
dQ will be smart enough to not take them for exa t dierentials. (In our Department prof. Ci ho ki was
famous for assigning zero points to student works whenever he noti ed the la k of the slash on d in the
heat or work form...)

21
the for e whi h the gas exerts on the piston, equal p A, where p is the gas pressure, b) an
external for e Fext applied to the piston from outside ( an be provided by the pressure
of the environment or by other means). When the piston starts to move, however slowly,
also fri tional for e an enter into play. Now suppose, the piston has moved by dx while
being a ted upon by Fext . The work done by the external for e on the system (the gas,
the ylinder and the piston) ertainly is

Fext
Fext ·dx = −Fext dx = − d(Ax) ≡ −pext dV . (16)
A
The work done on the system, assuming the absen e of fri tion, is this minus the kineti
energy a quired by the piston on the distan e dx.
In general pext is not simply related to the pressure in the gas. Worse yet, if the piston
moves qui kly (or a elerates) the gas will not remain in equilibrium and it will not be
possible to hara terize it by two parameters only: the gas pressure will not be the same
in the entire gas volume - it will be dierent from point to point. Only if the piston moves
very slowly, without a quiring any a eleration, so that at any moment the gas inside the
ylinder remains pra ti ally in equilibrium and one an as ribe to it a unique pressure p
(almost the same in the whole volume) and there is no fri tion, an one say that pext = p
and write the work done by the external for e on the distan e dx in the form

d¯W = −p dV , or d¯W = −p(t, V ) dV , or d¯W = −p(V, U) dV , (17)

depending on the hoi e of independent variables - (p, V ), or (t, V ) or (U, V ) - hara -

terizing the equilibrium states. If this is possible, the work q be omes a well dened
dierential form on the spa e of the equilibrium parameters of the gas.
The equality pext = p ensures also that the piston does not a elerate; if so, then it,
stri tly speaking annot start to move - one is making here an idealization: the pro ess
in whi h one is allowed to write d¯W = −p dV is the limiting ase of a pro ess whi h an
be pra ti ally realized.
Thus, there are two onditions allowing to identify w with d¯W (p, V ):

• the hange must o ur vanishingly slowly - the pro ess must be quasistati , mean-
ing that at ea h stage the system an be treated as if it were in full equilibrium
(within itself and with its surrounding)

• there should be no fri tion - the pro ess must be reversible whi h in pra ti al
terms means that it o urs under an innitesimal dieren e between p and pext ;
innitesimal hange of this dieren e (whi h reverses its sign) su es to hange the
dire tion of the pro ess

This an be further illustrated by the following onsiderations. If the piston is with-

drawn suddenly, a rarefa tion in the gas will o ur and the work done by the gas will be
smaller than if the pressure in the gas remained uniform. The limiting example is the
pro ess in whi h the piston is repla ed by a wall impermeable to the gas parti les whi h

22
prevent the gas from diusing into an additional volume of the ylinder. If this wall is
removed the gas will ll the whole ylinder. The removal of the wall an by done at no
work ost at all. The gas will then not perform any work at all. And the pressure will
not be dened for a while. But if the additional volume is quite small, just an innites-
imal dV , the departure from equilibrium may be negligible and the pressure will remain
well dened - the gas will pra ti ally stay in equilibrium - the expansion pro ess will be
quasistati . The produ t −p dV will ertainly be non-zero. Yet it will not represent the
work done on or by the gas - the work will be exa tly zero! One an also make in this way
a nite hange of the volume o upied by the gas, just by opening to it su essively and
always quasistati ally, additional volumes dV until a nite
P∆V is rea hed.RThe work done
by the gas will still be zero irrespe tively of the fa t that (−p dV ) = − dV p 6= 0. The
pro ess, while quasistati , will not be reversible: one annot hange something innitesi-
mally to reverse its dire tion. This shows that quasistati ity does not imply reversibility
5
although, as should be lear, any reversible pro ess must be quasistati .
It is also instru tive to onsider the ee ts of the fri tion assuming that the piston is
removed or inserted with vanishing velo ity, without any a eleration and the equilibrium
of gas is maintained at ea h stage. When the gas is being ompressed (the volume V
de reases), pext must be greater than the pressure p of the gas, be ause the external for e
must over ome in addition the fri tion. When the piston is removed, pext must be smaller
than p, be ause now it is the gas pressure whi h must over ome the fri tion. As a result
of the omplete y le onsisting of a ompression followed by the de ompression, the total
work done by the external for e will be positive (make a plot)
I
W =− dV pext > 0 . (18)

If the nal gas temperature at the end of this pro ess is the same as the initial state (we
assume that the volume after the y le returns to its initial value), this means that W
had to be somehow taken away from the system formed by the gas and the ylinder. If
the nal temperature is higher (assuming the ylinder does not absorb heat) W is just
the dieren e of the gas nal and initial energies. In both situations we have to do with
onversion of work into heat (through the ee t of the fri tional for e). But irrespe tively
of this, the whole pro ess is in this ase irreversible, for hanging the dire tion of the
piston requires a nite hange (and not an innitesimal one) of the applied external for e.
And ertainly the work done by this for e is not given by
I
− dV p , (19)

so w - the work done on the gas annot be written as −p dV .

5 As always, there is a lot of onfusion in terminology here. Many authors, in luding my respe ted
olleagues from the 5th oor, put the equality sign between reversible and quasistati pro esses, alling
a pro ess whi h is quasistati in my sense but not reversible, a pseudostati pro ess. Somehow I annot
digest this semanti hair-splitting and maintain that linguisti ally it is mu h more natural to simply
distinguish quasistati ity and reversibility.

23
 It should be also remarked at this pla e that heat an be transferred reversibly
from one body to another one only if their temperatures dier innitesimally, so that
innitesimal hange of the temperature of one of these bodies would reverse the dire tion
of the heat ow. This is the always the ase when isothermi pro esses are onsidered
(provided other elements of su h pro esses are arried out reversibly, there is no fri tion,
et .): one assumes that the temperature t of the system, while it undergoes su h a hange
during whi h heat is added to or extra ted from it, remains equal to the temperature
tres of a reservoir ( alled also heat bath) whi h, owing to its very large size (innite in
the limit), stays un hanged (its temperature does not hange and its equilibrium is not
perturbed), now matter how mu h heat it loses or absorbs.
Pro esses in whi h the system's temperature hanges while they are ex hanging heat
(e.g. isobari , that is o urring at onstant pressure, expansion of a gas) an be treated
as reversible provided one imagines them as split into small subpro esses during whi h
the system's temperature an be treated as approximately onstant and the system is
su essively onta ted with a sequen e of reservoirs having temperatures adjusted to the
a tual system's temperature.

Thus when the uid undergoes a reversible hange, one an write

dU = q − p dV . (20)

This means that in su h a hange also q an be written as d¯Q, that is, as a dierential
form on the spa e of the systems' parameters, although at the moment we do not know
how to do it in the other way as dU + p dV . This requires 2TMDL.

Other types of reversible works

As said in the introdu tory part of these le tures, thermodynami s an be applied to
very dierent physi al systems on whi h dierent kinds of works an be done. Here we
list some of them to prepare ground for problems whi h will be assigned as lass, home
and olloquial works.

• Work required to enlarge the soap lm spanned on a frame (pi ture).

d¯W = γ dA , (21)

where A is the area of the lm and γ its surfa e tension. The relation γ = γ(T, A)
plays here the role of the equation of state. Usually γ depends on temperature only.
Again, the hange of the area must be made slowly to maintain equilibrium; fri tion
is almost absent here, so quasistati hange of A almost ne essarily is also reversible
• Work needed to deform a rubber band (many ni e problems with this systems an
be formulated) is

d¯W = KdL . (22)

L is here the rubber length and K its strain. The relation K = K(T, L) plays the
role of the equation of state.

24
• Work needed to stret h a thin wire.

d¯W = KdL , (23)

where L is the wire length and L its tension. This an be reversible if the deformation
is elasti (and not plasti ; no hysteresis is allowed). The relation K = K(T, L) plays
the role of the equation of state of the wire. If the deformation is small it takes the
form of the Hooke's law K = k(T )(L − L0 ), where L0 is the length of unstret hed
wire

• The pre eding example an be generalized to elasti deformations of a solid. In this

′
ase one onsiders the lo al displa ement ve tor u(x) = x (x) − x. The hanges of
of distan es in the body after the deformation are en oded in the tensor
!
∂ui ∂uj X ∂ul ∂ul ∂ui ∂uj
 
1 1
uij = + + ≈ + , (24)
2 ∂xj ∂xi l
∂xi ∂xj 2 ∂xj ∂xi

in terms of whi h dℓ′2 = dℓ2 + 2uij dxi dxj . The eigenvalues λ(1) (x), λ(2) (x), λ(3) (x),
ij
of the the tensor u determine the lo al hanges of the volume:

dV ′ = dV (1 + tr(uij ) = dV 1 + λ(1) (x) + λ(2) (x) + λ(3) (x) .

(25)

The elementary work done on the solid when it is elasti ally (not plasti ally!) de-
formed reversibly is given by
Z
d¯W = d3 x σ ij duij , (26)
V

where σ ij is the stress tensor. The Hooke's law (the equation of state) takes in this
ase the form
 
ij 1 ij ll 1 1
u = δ σ + ij
σ − σ ij σ ll (27)
9K 2µ 3
or
 
ij ll ij 1 ll ij
ij
σ = Ku δ + 2µ u − u δ . (28)
3

where K is the ompressibility modulus and µ is the shear modulus. Knowing their
temperature dependen e is equivalent to knowing the equation of state. In the ase
ij ij
of a uid under a hydrostati pressure p the stress tensor takes the form σ = −pδ
and the work redu es to −p dV . (See L&L, vol. 8, Elastome hani s.)

• Work needed for magnetization. Magneti materials (paramagneti or diamagneti )

put in an external magneti eld of strength H0 (the subs ript 0 indi ates this is the

25
 magneti eld produ ed by the urrents (in a oil, for example) a quire lo al mag-
netization M(x) and the work (whi h a battery must provide) needed to magnetize
a given magneti body is (using the illegal Gauss system of units and assuming
homogeneous in spa e magneti eld strength)

d¯W = H0 ·dM , (29)

M = d3 x M is the total magnetization of the body. The relation M =

R
where
α(T, p, H0 )H0 plays here the role of the equation of state. In spe ial ases one an
repla e M by V M and write equation of state as M = χ(T )H0 with χ(T ) being
the magneti sus eptibility of the material. Typi ally (if the temperature is not too
low) χ = a/T , a = onst. (Curie law).

• Work needed for polarization. A diele tri material pla ed in a uniform external
ele tri eld of strength E a quires a lo al polarization P and the work needed to
polarize a pie e a diele tri is

d¯W = E ·dP , (30)

d3 x P .
R
where P = Again the relation between P and E plays the role of the
equation of state.

A digression
Before we move further it is useful to make a short summary. We started with hara -
terizing equilibrium states of any system by parameters dire tly related to works whi h
an be performed on the system reversibly. In the ase of a uid these are V and p, or
M and/or P if systems with nontrivial ele tromagneti properties are onsidered. These
parameters are said to be state fun tions. This means that (equilibrium) states of the
system an be viewed (and this is, a ording to my experien e, the most onvenient view
to adopt in thermodynami s) as forming an abstra t manifold (in the mathemati al sense
of this term) on whi h the fun tions like volume, pressure, et . are dened. Next we
have introdu ed some additional quantities: the empiri al temperature t and the internal
energy U (there will be also others). These too should be treated as fun tions dened on
the manifold of system's states. A manifold, to be explored, requires introdu ing on it a
s
system of oordinates whi h, mathemati ally speaking, map its points onto a spa e R .
A oordinate system is therefore a set of s independent fun tions (their number s equals
to, or rather denes, the dimension of the manifold) dened on the manifold. Thus p,
V, or p, t, et . should be treated as oordinates on the manifold of the system states.
On e the system has been hosen, other fun tions like U, t or V be ome fun tions of the
oordinates used to identify the points on the manifold. It should, however, be lear that
the division into oordinated and fun tions is not xed on e for ever: in some appli ations
what formerly was treated as a fun tion an now be treated as one of the oordinates.
It is known from the theory of manifolds that some systems of oordinates may be ill
dened in some regions of the manifold - the same an happen - si puo dare ( ompressed

26
to podarsi) as Gioelle Botta would say - in some ases in thermodynami s: apparent
paradoxes, ontradi ting e.g. 2TMDL, are mostly problems related to the wrong hoi e
of oordinates on the manifold of states and not real paradoxes. Furthermore, dierential
forms d¯W , d¯Q dis ussed here, whi h in fa t are, like ve tors or tensors, geometri obje ts,
should also be viewed as forms dened over the manifold of the system's states and as
su h an always be expressed in the hosen oordinate system.

Other state fun tions

Already here one an introdu e other state fun tions although the roles they play in
thermodynami s annot be appre iated at this stage (here these fun tions are introdu ed
H =U+
purely formally). In the ase of a simple uid one frequently uses the Enthalpy
pV , the Helmholtz free energy F = U −T S or the Gibbs fun tion G = U −T S +pV .
They are alled thermodynami potentials (U and entropy S - to be introdu ed in the
next Le ture - also are alled thermodynami potentials) for in spe i situations they
1 s
indeed play analogous role the potentials V (q , . . . , q ) play in Me hani s - they determine
the equilibrium state of the system. As state fun tions, all of them an be expressed in
6
terms of dierent state parameters.

Let us briey demonstrate the usefulness of enthalpy in hemistry. Most of hemi al

rea tions o ur at onstant (atmospheri ) pressure. During su h rea tion the system,
onsisting of rea ting substan es (dierent at the end and initially) whi h hange their
volume in the ourse of the rea tion, does work on the surrounding atmosphere (against
its pressure pext . If the initial and nal states of the system are in equilibrium within
themselves and with the surrounding atmosphere (p = pext , t = text ), 1TMDL an be
written as Ufin = Uin + W + Q, with W the work done by the atmosphere on the system
and Q the heat absorbed by the system. Sin e the work done by the surrounding on the
system is −pext (Vfin − Vin ), and in the initial and nal states p = pext , one an write

Ufin = Uin − p (Vfin − Vin ) + Q , (31)

or

(Uin + p Vin ) − (Ufin + p Vfin ) ≡ Hin − Hfin = Q̄ . (32)

Thus the heat Q̄ released in a rea tion o urring at onstant pressure is given by the
dieren e of enthalpies of the initial and nal substan es. Noti e that it is not required
that the thermodynami equilibrium be maintained during the rea tion; only the initial
and nal states must be equilibrium states. (Examples in lasses.)
Enthalpy, as will be dis ussed in lasses, is onserved in the so- alled Joule-Kelvin
pro ess in whi h a gas passes irreversibly from a state in whi h its temperature and
pressure equal t1 and p1 , respe tively, to another state in whi h its pressure p2 is lower
and temperature equals t2 .
Finally, enthalpy is onserved in various ow pro esses - its onservation generalizes
to ompressible uids the Bernoulli law (whi h applies only to in ompressible uids).

6 Later we will see, however, that ea h of these thermodynami potentials has a preferred (in the sense
that will be elu idated) set of its variables.

27
 LECTURE III (TMD)

We have already introdu ed and dis ussed two of the four laws of thermodynami s. In
the pre eding Le ture it was found that in innitesimal reversible hanges it is possible
to represent the elementary work w in the generally valid form dU = q + w of 1TMDL as
a dierential form d¯W (or d¯w , if hanges du of the molar internal energies are onsidered)
dened on the spa e of parameters hara terizing equilibrium states of the onsidered
system. In su h hanges, sin e dU and d¯W are well dened dierential forms, also q must
be a dierential form d¯Q. However, on the basis of what has been done, there is no way
to write this dierential form dierently than as d¯Q = dU − d¯W , for example, onsidering
a simple uid, as

d¯Q = dU(p, V ) + p dV , (33)

if the variables p and V are taken for independent ones. The heat Q taken by the system
in a nite pro ess (a nite hange) in whi h it reversibly passes from the equilibrium state
A to another equilibrium state B is then given by

Z B Z B
Q= d¯Q = (dU(p, V ) + p dV ) , (34)
A A

the integral being taken along the urve representing the pro ess on the (V, p) plane.
Adiabati hanges of the system, that is (in my terminology) its reversible adiather-
mal hanges, are those hanges in whi h the form d¯Q vanishes. More mathemati ally,
d¯Q proje ted onto paths representing su h pro esses in the spa e of parameters is zero -
it gives zero on all ve tors tangent to su h paths. Adiabats of a simple uid are therefore
urves in the (p, V ) or (p, t) or (t, V ) spa es determined by the solutions of the dierential
equation

dU + p dV = 0 , (35)

written in the set of variables, (p, V ) or (p, t) or (t, V ), whi h has been hosen (on the
basis of onvenien e) to work with, and the initial point (state). In the general ase of a
system whose equilibrium states are determined by the parameters x1 , · · · , xs and Xi are
the generalized for es whi h an a t on the system, adiabats are urves (paths) on whi h
7
proje tions of the one-form

s−1
X
d¯Q = dU(x1 , · · · , xs ) − Xi (x1 , · · · , xs ) dxi , (36)
i=1

vanish.
7 Re all, that the number s of independent parameters xi equals 1 plus the number of works whi h
an be reversibly done on the system.

28
 Unfortunately in pra ti e one usually does not know U as the fun tion of the system's
parameters like p and V or, x1 , . . . , xs in the general ase - the assignment of the internal
energies to dierent states of the system is an example of the typi al paper and pen il
theoreti al onstru tion! - and therefore one annot go too far in this way with solving
8
various thermodynami al problems.

At this point it is amusing to say that the perfe t gas (and perfe t magneti material)
has been invented partly in order to have a system on whi h to torment students. Indeed,
in luding as part of its denition the information that its internal energy U is independent
9
of its volume V, the main obsta le for playing with this system is removed, be ause
passing to t and V as the independent variables, the heat form of the perfe t gas an be
expli itly written down as
    
∂U ∂U (t)
d¯Q = dt + + p(t, V ) dV = CV dt + p(t, V ) dV . (37)
∂t V ∂V t

(t)
CV is here the heat apa ity absorbed by the system when its empiri al temperature t
hanges by one unit, but sin e usually one in ludes in this denition the onstan y of
(t)
CV and, moreover, sin e the empiri al temperature dened by the perfe t gas equation
of state happens to be (proportional) to the absolute temperature T (to be introdu ed
in this Le ture), the fa t that one ( ons iously or in ons iously) repla es t by T has no
onsequen es.
The adopted denition of the perfe t gas allows to solve problems of the sort what heat
it absorbs when it isothermally and reversibly expands from the pressure p1 to p2 < p1 :
sin e the internal energy of the perfe t gas depends on the temperature only, it stays
onstant in the isothermal expansion and, by 1TMDL, Q = −W = W̄ , where
p2
dp
Z Z Z
W̄ = − d¯W = p dV (T, p) = −nRT = nRT ln(p1 /p2 ) ,
Γ Γ p1 p
upon using the equation of state in the form V = nRT /p. The same integral gives also
the answer dire tly, without 1TMDL, be ause as an be seen from (37), on isotherms,
owing to the assumption that (∂U/∂V )t = 0, the heat form of the perfe t gas takes the
form d¯Q = p dV . Su h a simple reasoning would not be true if the gas satised e.g. the
Van der Waals (VdW in short) equation of state, be ause, as it will be possible to show
by appealing to 2TMDL, the internal energy U of su h a gas does depend on the volume
V.
As a matter of fa ts, as a onsequen e of 2TMDL, the dependen e of the system's
internal energy U on the volume V is dire tly determined by this system's equation of

8 However, some of the thermodynami al relations an be obtained with this limited knowledge whi h
is en oded in 0&1TMDLs - see the home and olloquial problems to this ourse.
9 This is usually ba ked by the physi al argument that mole ules of the perfe t gas are mutually
nonintera ting and therefore the internal energy of the gas is just the sum of kineti energies of individual
mole ules; the volume V o upied by the gas determines only the relative positions of the mole ules but
sin e there is no ontribution to U of the intera tion energies (whi h would depend on the relative
positions of mole ules), hanges of the volume do not alter U.

29
state and it ould happen that the assumption of independen e of its internal energy
U on its volume V is in onsistent with the perfe t gas equation of state pV ∝ t ∝ T
(fortunately, it is onsistent, as you will be able to he k).
Furthermore, in reality, independen e of temperature of the heat apa ity CV (2TMDL
does not onstrain its dependen e on the temperature) is true only if the perfe t gas is
3
a monoatomi one - CV = ncv in this ase equals nR, where n is the number of moles.
2
Gases, mole ules of whi h are omposed of more than one atom (the majority of real
gases) have heat apa ities only (to a good degree of a ura y) pie ewise onstant: at
almost all temperatures at whi h gases exist as gases (and not as liquids into whi h
they eventually hange when temperature is lowered) the three-dimensional motion of gas
mole ules is quasi lassi al ( an be represented as in lassi al me hani s) and ontributes
3
2
R to CV /n; however owing to the prin iples of quantum me hani s whi h must be
employed to properly treat the internal motion (rotations and vibrations) of mole ules,
CV /n of multiatomi gases rises to 3R ( 52 R if the gas mole ules are omposed of two
atoms only) and then to even higher values depending on the number of the vibrational
degrees of freedom of the mole ule (+1R per ea h vibrational degree of freedom; the
number of vibrational degrees of freedom of a mole ule equals −3 − 3 + 3×the number of
atoms, or −3 − 2 + 3 × 2 in ase of two atoms). All this, as we will see, an be predi ted
within the statisti al physi s approa h (in phenomenologi al thermodynami s molar heat
apa ities together with their dependen e on temperature must be taken dire tly from
measurements or else an be related to other quantities taken from experiment) - the
gas of mutually nonintera ting mole ules is one of a few ompletely solvable problems.
Within the statisti al physi s approa h it will also be ome lear that the temperatures T
at whi h the rise of the molar heat apa ity o urs an be estimated from the simple rule

T ∼ Eexc /kB , (38)

where Eexc
are typi al energies of the rotational and vibrational ex itations of the gas
10
mole ule and kB is the Boltzmann onstant, kB = 8.617 × 10−5 eV/K. Typi al atomi
energies are of order of tens of ele tronowolts (energies of the rotational ex itations are
lower than energies of the vibrational ones - simple physi al intuition says that it is easier
to rotate something than to make it vibrate), so typi al temperatures at whi h CV rises,
are in the region of thousands or tens of thousands Kelvins - quantum ee ts in the
spe i heats of gases manifests themselves only at very high ( ompared to the room
ones) temperatures (not only, as one naively ould think, at very low temperatures)!

2TMDL
Only some of hanges of thermodynami al systems permitted by 1TMDL, that is, by
the onservation of energy, are observed to o ur in the real world. All hanges have a
lear tenden y to o ur preferentially in one dire tion (not in both); to take the simplest
phenomena: if the me hani al energy is lost as a result of fri tion or vis osity it annot
be re overed (without other hanges), hemi al rea tions o ur evidently irreversibly,

10 These are the right units in whi h the Boltzmann onstant should be remembered; giving it in J/K
is as useless as measuring the Warsaw-New York distan e in mi rons or atomi sizes in parse s...

30
mixing of dierent gases is also irreversible, et . The preferred dire tion of the hange
is perhaps most learly manifested in the distin tion between a hot and a old body:
although 1TMDL does not forbid the opposite (so long as energy is onserved), heat ows
between two bodies whi h initially are not in thermal equilibrium in su h a dire tion as
to eventually bring them into equilibrium (equalize their temperatures). This allowed us
to dene operationally whi h of the two bodies is hotter and whi h one is older. In fa t,
the basi assumption underlying the whole phenomenologi al thermodynami s (but also
the equilibrium statisti al physi s) is that systems left to themselves eventually attain an
equilibrium state; reversion to the original state is never observed (if it happened, the
11
notion of equilibrium would loose any sense).

Although in the pre eding Le ture we found it useful to single out reversible hanges
whi h an o ur in both dire tions - only in su h pro esses the form d¯W an be used as
representing the work done on the system - they are only theoreti ally useful idealizations
- they require stringent onditions, impossible to fulll in pra ti e (like p = pext , t =
text , quasistati ity et .). Normal and prevailing type of behaviour of real systems are
irreversible hanges.

These obvious (so obvious from the everyday experien e that it took longer to a ept
1TMDL - a epting that the me hani al energy is not lost but gets onverted into the
internal energy required ompli ated quantitative measurements - than 2TMDL) obser-
vations underlie the 2MDL whi h generalizes them and promotes to the universal law of
Physi s.

There are various formulations of 2TMDL and in view of the fundamental role it plays
in physi s, we will dis uss all of them (although not all in equal depth).

2TMDL (R. Clausius, 1850, in Pippard's words):

It is impossible to devise an engine whi h, working in a y le, shall produ e no ee t other
than the transfer of heat from a older body to a hotter body.

Of ourse, we have dened the notions of  older and hotter on the basis of the
dire tion the heat ows, so this statement of 2TMDL may appear somewhat tautologi al,
but dening these notions we had in mind only a dire t onta t of two bodies through a
diathermall wall; the Clausius' prin iple says that reversing this dire tion is impossible at
all, by any means and by using any roundabout physi al pro ess.
Cru ial lause in this (and in the Kelvin's one given below) formulation is working in
a y le: It is possible, for example, to expand a gas isothermally in onta t with a old
body, so that it takes some heat Q, then to ompress it adiathermally making it hotter,

11 It is perhaps fair to say already in this pla e, that within the statisti al approa h it turns out that at
the mi ros opi level returns of (ma ros opi ) systems to states ma ros opi ally indistinguishable from
the initial ones are not forbidden by the fundamental laws (of lassi al or quantum me hani s) but are
only very improbable - in statisti al approa h the dis ussed tenden y of hanges to o ur in only one
dire tion is a statisti al ee t. But probabilities of su h returns are so fantasti ally tiny (one talks here
−10n−teen
of numbers as small as 10 ) that they never happen in pra ti e and are never observed. Thus,
phenomenologi al thermodynami s whi h is based on what is really (and not what hypotheti ally ould
be) observed an safely rest on the dis ussed assumption.

31
and then to bring it into onta t with a hotter body (reservoir) and, ompressing it further
isothermally, transfer heat to the hotter body (making the two ne essary ompressions at
the ost of the work gained at the rst stage). Su h a pro ess does not violate 2TMDL
as formulated above, for it is not a y le - at the end the gas will be in a dierent state
than initially. Only if it were possible to bring the gas ba k to its original state without
undoing the heat transfer, ould violation of the Clausius' 2TMDL be laimed. Only
y li ity an guarantee that the pro ess would be repeatable and ould serve to transfer
an arbitrary amount of heat from a older body to a hotter one.

2TMDL (Kelvin, 1851, again in Pippard's words):

It is impossible to devise an engine whi h, working in a y le, would produ e no ee t other
than the extra tion of heat from a reservoir and performan e of an equivalent amount of
me hani al work.
In ontrast to the Clausius' formulation, whi h as all German philosophy of that time
(Hegel, Kant & S helling!) on entrates on rather abstra t notions, the one of Kelvin
learly bears an imprint of an utterly pragmati approa h, hara teristi of the times
of the British industrial revolution: in simple words it ommuni ated to engineers and
inventors of ma hines what they should not hoped to a hieve. Its alternative formulation
an read:
It is impossible to absorb heat from a reservoir and to onvert it all into work without
introdu ing other hanges in the system and its environment.

Again, by expanding isothermally a gas remaining in thermal onta t with a reservoir

of heat (the reservoir's internal energy) it is possible to perform some work using up the
extra ted heat but su h a pro ess is not y li al and therefore, it does not ontradi t the
Kelvin's 2TMDL.

The formulations of 2TMDL by Clausius and Kelvin are easily proved to be equivalent.
This is a sort of a s holasti exer ise, whi h I re all qui kly here.
1. Clausius−→Kelvin. We argue that if the Kelvin's prin iple were violated, one ould
violate also the Clausius one (proof by redu tio ad absurdum - a.a.). Suppose there are
two reservoirs (bodies), a older one at t1 and a hotter one at t2 > t1 . Suppose Kelvin is
wrong, and a heat Q̄ an be taken from the older reservoir and all onverted into work
W̄ . Then this work an be used up to run a reversible Carnot engine12 whi h takes heat
Q̄1 from the reservoir at t1 and transfers a positive amount Q2 = Q̄1 + W̄ of heat to the
reservoir at t2 .
2. Kelvin−→Clausius. We argue that if the Clausius' prin iple were violated, one ould
violate also the Kelvin's one (again a.a.). If Clausius were wrong, one ould transfer a
heat Q̄1 from the reservoir at t1 to another one at t2 > t1 and then run a Carnot engine
between the two reservoirs; it ould be arranged so as to take the heat Q̄2 > Q̄1 from
the reservoir at t2 , give ba k the heat Q̄1 to the reservoir at t1 produ ing the net work

12 It is assumed here that everybody knows what the Carnot y le or engine is. (Carnot y les will
appears here three pages further). We annot say that by means of a me hani al devi e, like the Joule's
paddle-wheel one, the work W̄ an all be transferred as heat to the reservoir at t2 , for we insisted (in
Le ture II) that using paddle-wheel-like devi es should be lassied as doing work!

32
W̄ = Q̄2 − Q̄1 , in ee t entirely at the ost of heat taken from the reservoir at t2 .
There is one more formulation of 2TMDL, due to Caratheodory, whi h, ompared to
the physi ally (and operationally) lear formulations of Clausius and Kelvin, sounds rather
abstra t and is immediately re ognized as the produ t of a mathemati ally formed mind.
Indeed, Caratheodory was a mathemati ian well edu ated in the theory of dierential
forms and his formulation is learly rooted in properties of these mathemati al obje ts.
Literally it reads:
2TMDL (C. Caratheodory, around 1909, again in Pippard's words):
In the neighbourhood of any equilibrium state of any thermodynami al system there are
states ina essible by any adiathermal pro ess.
The Caratheodory's formulation is obvious in some simple ases like e.g. the adiather-
mally isolated liquid in whi h the Joule's paddle-wheel is immersed - adiathermally work
an only be added to the system but evidently not extra ted from it keeping the vol-
13
ume un hanged and, therefore, states of lower internal energies are ina essible, but the
Caratheodory's prin iple is assumed to apply to any thermodynami al system, no matter
how ompli ated.

It is fairly straightforward to show that the Caratheodory's prin iple follows from the
Kelvin's prin iple. The proof is again by redu tio ad absurdum. Take a thermodynami al
system whi h does not obey the Caratheodory's prin iple at least in some domain of the
manifold of its states. Consider a state A of the system in this domain and let the system
make an isothermal hange after whi h it rea hes a state B in this domain, absorbing on
14
the way a positive heat Q (from a reservoir). The assumption that the Caratheodory's
prin iple is violated in the onsidered domain means, that all points are a essible from
B on adiathermal paths. So A must be a essible too. If the system has returned to A
by an adiathermal path, the total hange of its internal energy is zero and, by 1TMDL,
the positive heat Q whi h it absorbed from the reservoir in the isothermal transition
from A to B must have been onverted into a positive work W̄ = Q in the adiathermal
transition. But this ontradi ts the Kelvin's prin iple. (For the sake of omplete larity
one should admit that it might happen that in some domain the heat form d¯Q of a
system is identi ally zero. Then any two states within this domain an be onne ted by
an adiathermal path, but vanishing of d¯Q means that the system is entirely me hani al
and not thermodynami al, so in this domain 1TMDL redu es ∆U = W . Su h systems
indeed do not satisfy the Caratheodory's prin iple but this does not invalidate, of ourse,
2TMDL in the Kelvin's and Clausius' formulations.)

For ompleteness one should mention here the Callen's approa h to thermodynami s
in whi h 2TMDL takes the form of an axiom in whi h the existen e of entropy as a state
fun tion possessing ertain properties is postulated and then onsequen es of this axiom

13 Adiathermal expansion of the liquid an de rease (a bit) its energy, but in this way the liquid does
annot return to the initial state whi h is, therefore, ina essible adiathermally from the nal state.
14 If in the isothermal transition from A to B the system loses a positive heat Q̄, we an repeat the
reasoning ex hanging the roles of the states A and B, be ause isothermal pro esses an always be made
reversible, as I tried to explain in the pre eding Le ture.

33
are derived and ompared with experien e. We will will dis uss this formulation a bit
later.

The mathemati al onsequen es of 2TMDL are the following

• The dierential one-form of heat, d¯Q, of any thermally homogeneous system, no

15
matter how ompli ated, whi h represents the heat absorbed by the system in a
reversible hange is integrable (i.e. has an integration fa tor).

• Among many (mathemati ally) possible integration fa tors of d¯Q there is one, 1/T ,
whi h is given by a universal (i.e. independent of the system) fun tion T (t) of the
empiri al temperature t.
• The exa t dierential dS = d¯Q(x1 , . . . , xr )/T (t(x1 , . . . , xr )) denes, up to a on-
stant, a new state fun tion of the system - its entropy S(x1 , . . . , xs ).

• In all adiathermal pro esses (reversible or not) the entropy does not de rease: Sfin ≥
Sin .
• The fun tion T (t) is, up to a s aling fa tor (setting in fa t the units in whi h it is
measured) the absolute temperature as dened by the Carnot y le.

These onsequen es an be derived from any of the three formulations of 2TMDL

given above. The only dieren e is the degree of ompli ation of the reasonings involved.
Derivation of all these onsequen es from the Clausius' or Kelvin's formulations of 2TMDL
requires, as we will see below, onsidering gedanked experiments invoking y li pro-
esses (Carnot y les). The Caratheodory's formulation of 2TMLD seems to be rather
abstra t but it allows to derive its mathemati al onsequen es without appealing to y li
pro esses whose use (outside the theory of ma hines produ ing work or refrigerators) is
rather arti ial and for this reason is said to be more e onomi al. However, sin e in the
XXI entury we are - or at least we should be! - familiar with rudiments of the theory of
forms (I deliberately planned lasses so that you get a quainted with them), we an now
say that it simply goes in the dire tion of putting the thermodynami s on a postulatory
basis, as a purely dedu tive theoreti al system (this tenden y, as we will dis uss, eventu-
ally ulminates in the Callen's formulation, elebrated by many, in luding my olleagues
th
from the 5 oor) for the Caratheodory's prin iple dire tly postulates (through the se -
ond Caratheodory's theorem) the existen e of adiabati surfa es, that is, it dire tly states
that the heat forms d¯Q of physi al systems do have integrating fa tor and are therefore
integrable, with values of the entropy S labeling families of the orresponding solutions of
the equations d¯Q = 0. But, at least to me, the whole essen e of thermodynami s is to see
how these properties of the heat form and the existen e of entropy ne essarily follow from
the formulations of Clausius and Kelvin - the formulations whi h generalize our dire t
experien e!

As we will be dis ussing in these le tures, on the pra ti al side 2TMDL

15 As explained below, this statement is trivial in the ase of simple systems whose equilibrium states
are hara terized by two parameters only.

34
 • Determines the dire tion of real pro esses Sfin ≥ Sin .

• Expresses in a pre ise way and quanties the degree of irreversibility of physi al
pro esses

• Determines the equilibrium states of thermodynami systems under various ondi-

tions (we will elaborate on this Callenian point of view in due ourse)

• Puts upper limits on e ien ies of thermodynami ma hines.

• Limits possible use of huge amount of internal energy of bodies around us (e.g. the
energy of o eans).

• Is important for thermodynami s of hemi al rea tions and in biology and many
other pla es.

To quote R. Emden: In the huge manufa tory of natural pro esses the prin iple of
entropy o upies the position of manager, for it di tates the manner and method of the
whole business, whilst the prin iple of energy merely does the bookkeeping, balan ing
redits and debits.

35
 LECTURE IV (TMD)

We now set ourselves to infer the integrability of d¯Q and, in onsequen e, the existen e
of the entropy as a state fun tion, from the Kelvin's formulation of 2TMDL. (Sin e we
have shown that this formulation is equivalent to the Clausius' one and vi e versa, we
may laim, if we wish, that we start from the Clausius' formulation, as well).
At the beginning we will onsider adiabati (adiathermal reversible) hanges in whi h
d¯Q = dU − d¯W = 0. In the ase of simple systems, e.g. uids, hara terized by the
p and V , the
variables equation
"  #  
∂U ∂U
d¯Q = + p dV + dp = 0 , (39)
∂V p ∂p V

or, hoosing t = t(p, V ) andV as independent variables, the equation

    
∂U ∂U
d¯Q = dt + + p dV = 0 , (40)
∂t V ∂V t
16
determines uniquely in the parameter spa e a family of noninterse ting urves labeled
by the initial point whi h therefore makes the heat form d¯Q trivially integrable in su h
ases. If the system is the perfe t gas, the equation redu es to

(t)
CV dt + p(t, V ) dV = 0 , (41)

(t)
and, if in addition the onstan y of CV is assumed, yields immediately, upon using
(t)
nR/CV
the equation of state p(t, V ) = nRt/V , the adiabati urves t V = onst., or
(t) (t) (t) (t) (t)
1+(nR/CV ) Cp /CV
pV = pV = onst. (The relation Cp = CV + nR satised by the
prin ipal heat apa ities of the perfe t gas follows from 1TMDL.) The heat form d ¯Q is
made integrable by dividing it just by t (or onst×t), that is, as advertised, its integrat-
ing fa tor is indeed a fun tion (in this ase a linear one) of the empiri al temperature t
(dened by the perfe t gas thermometer).
However in lasses you have seen (at least those who attended...) an example (taken
from Pippard's book) of a one-form in three dimensions whi h is not integrable. Thus not
every one-form in more than two dimensions is integrable.

To establish integrability of the form d¯Q in the general ase, we will (as in the ase
of proving the existen e of the empiri al temperature as a fun tion of state) rst use a
reasoning whi h is more physi al (and uses virtually no mathemati s) and then a more
mathemati al one.

To show physi ally, using the Carnot y les, that the Kelvin's 2TMDL implies in-
tegrability of d¯Q, we start from two isothermal hyper-surfa es t(x1 , . . . , xs ) = t1 and

16 It should be known from Math II ourse (at least to those trained in math lasses by me) that the
equations of the type P (x, y)dx + Q(x, y)dy = 0 sometimes do have singular points through whi h more
than one integral urve (solution) passes; su h ases are here ex luded.

36
t(x1 , . . . , xs ) = t2 with t1 6= t2 (if the system has three parameters these are ordinary two-
dimensional surfa es in the three-dimensional parameter spa e). From 0TMDL we know
su h hyper-surfa es always exist and are not interse ting (the empiri al temperature is a
unique state fun tion - it does not have bran hes). We onsider on the surfa e t = t1
a urve onne ting the points A and B belonging to this surfa e. Then we onsider two
urves whi h are two solutions of the equation d¯Q = 0 passing through the points A and
B and interse ting the surfa e t = t2 at the points D and C , respe tively (The equation
d¯Q = 0 an always be integrated step by step starting from the points A and B and
ontinuing until it rosses the hypersurfa e t = t2 , although the ontinuation may not be
unique). In this way one obtains a reversible y le - the Carnot y le, alled also the
Carnot engine - whose segments AB and CD orrespond to isothermal hanges of the
onsidered system while its segments BC and DA orrespond to adiathermal reversible
hanges. Beginning from the state A, the system takes in the isothermal hange A→B
a heat Q1 from a reservoir at the temperature t1 and in the isothermal hange C →D
- a heat Q2 from another reservoir at the temperature t2 . As the system returns to the
state A, its nal internal energy is the same as the initial one and, by 1TMDL, the work
W̄ done by the system must be equal
W̄ = Q1 + Q2 . (42)

Using the Kelvin's statement of 2TMDL it will be now argued that the ratio −Q1 /Q2
is universal, that is, it is the same irrespe tively of the nature of the system performing
su h a y le, so long as the two isotherms remain at the temperatures t1 and t2 . The
negative sign of the ratio Q1 /Q2 (that is, the positive sign of −Q1 /Q2 ) is also a dire t
onsequen e of Kelvin's 2TMDL: if it were positive (Q1 and Q2 of the same sign), it
would be possible to a omplish the y le (whi h has been onstru ted as reversible) in
the sense that both Q1 and Q2 were positive; of the positive work W̄ = Q1 + Q2 its
amount Q1 ould be put ba k irreversibly (by a Joule's paddle-wheel devi e, for instan e)
into the reservoir at t1 and the net result of the pro ess would be only the extra tion of
a positive heat Q2 from the reservoir at t2 and performan e of a positive work W̄ = Q2 .
This is not possible a ording to Kelvin. To show that the ratio −Q1 /Q2 is universal,
one onsiders two su h reversible Carnot engines whi h may be onstru ted with the
help of two dierent thermodynami al systems working between the same two empiri al
temperatures t1 and t2 , of whi h the rst engine absorbs heats Q1 and Q2 and the other
′ ′ ′
one Q1 and Q2 . It is then possible to hose two integers k and k so that to the desired
′ ′
a ura y k|Q1 | = k |Q1 |. This is always possible be ause any real number |Q1 |/|Q2 | an
′
be approximated by a rational number k /k (Cau hy sequen es!). One an then treat the
two Carnot engines as a single system and onsider its y le onsisting of k runs of the
k ′ of the se ond one a omplished in su h senses that the heats Q1 and Q′1
rst y le and
are of opposite signs. In the omplete y le of the ompound system the total heat taken
from the reservoir at t1 is zero, while the total heat taken from the reservoir at t2 equals
k Q2 + k ′ Q′2 and by 1TMDL must be equal to the work done by the ompound system.
By Kelvin's 2TMDL this annot be positive, so

k Q2 + k ′ Q′2 ≤ 0 . (43)

37
But be ause the individual y les, and therefore the y le of the ompound system, are
reversible, also

−(k Q2 + k ′ Q′2 ) ≤ 0 , (44)

whi h follows from the possibility of a omplishing the y le of the ompound system in
′ ′
the opposite sense. Thus, k Q2 + k Q2 = 0 and from the two equalities

k Q1 = −k ′ Q′1 ,
k Q2 = −k ′ Q′2 ,
it readily follows that Q′1 /Q′2 = Q1 /Q2 . Therefore, as proposed, the ratio of heats an
only depend on the temperatures t1 and t2 :

−Q1 /Q2 = f (t1 , t2 ) , (45)

and the fun tion f (t1 , t2 ) must be universal (independent of the system a omplishing
the Carnot y le between the temperatures t1 and t2 ).
In the next step one shows that the universal fun tion f (t1 , t2 ) ne essarily fa torizes:
f (t1 , t2 ) = φ(t1 )/φ(t2 ). To this end one onsiders a ompound Carnot y le onsisting of
one Carnot y le C12 working between t1 and t2 onsisting as previously of the hanges
A → B → C → D , and another one, C23 , working between t2 and t3 and onsisting of the
17
hanges D → C → F → G. Viewing the ompound y le omposed of the y le C12
followed by the y le C23 as working between the temperatures t1 and t3 , on one hand
(12) (23)
one has −Q1 /Q3 ≡ −Q1 /Q3 = f (t1 , t3 ) while on the other hand
! !
(12) (23)
Q1 Q2
−Q1 /Q3 = f (t1 , t3 ) = − (12) − (23) = f (t1 , t2 ) f (t2 , t3 ) , (46)
Q2 Q3
(12) (23)
be ause Q2 = −Q2 . This is possible only if the fun tion f (t1 , t2 ) fa torizes as pro-
posed: f (t1 , t2 ) = φ(t1 )/φ(t2 ). Thus, in any Carnot y le working between temperatures
t1 and t2 , independently of the nature and degree of ompli ation of the system used to
onstru t it,

−Q1 /Q2 = φ(t1 )/φ(t2 ) ≡ T1 /T2 . (47)

Obviously T ∝ φ(t) an always be taken for an empiri al temperature, but in view of the
fa t that the ratio of heats taken by any system in the reversible Carnot y le is universally
given by the ratio of the temperatures dened in this way of the two reservoirs, T is alled
18
the absolute temperature. Thus

T = const. × φ(t) , (48)

17
It is not ne essary that these two y les be performed by the same system; it is enough that the heats
(12) (23)
Q2 and Q2 taken at t2 by the systems performing y les C12 and C23 were of equal absolute values
and of the opposite signs and taken from the same reservoir.
18 As dis ussed, the proportionality onstant in (48) is, sin e 1954, xed by assigning to the triple point
of water the absolute temperature equal exa tly 273.16 K. Before 1954 the onstant fa tor was xed by
the requirement that at 1 atm there be 100 units between the i e melting point and water boiling point;
this led to the temperature 271.15 K of the i e point.

38
 qi Tiext qi
Γi Ci

σ qi (T0 /Tiext )

T0

Figure 1: A realization of the fragment Γi of the y le exe uted by the body σ.

and operationally the temperature T is determined by the e ien y

W̄ Q1 + Q2 T1
η≡ = =1− , (49)
Q2 Q2 T2
of any Carnot y le working between t1 and t2 . This means that any empiri al temperature
t dened using a thermometri body an in prin iple be alibrated with respe t to the
absolute temperature, that is the form of the fun tion φ(t) established, by performing a
Carnot y le using this body as the working substan e under onditions as nearly ideal, as
possible. In pra ti e, sin e the perfe t gas temperature s ale turns out to be proportional
to the absolute one, it is easier (if the physi al onditions allow for this) to relate a given
absolute temperature to the perfe t gas temperature.

Above we have found that if a system performs a Carnot y le between absolute

temperatures T1 and T2 (T1 < T2 ), then

Q1 Q2
+ = 0. (50)
T1 T2
We now generalize this result, still relying on the Kelvin's formulation of 2TMDL.
Consider a system σ exe uting a y li al pro ess of any degree of omplexity whi h
we mentally an split into a large number (innite in the limit) of small (innitesimal)
segments (subpro esses) Γi , i = 1, . . . , M (M → ∞). At the i-th subpro ess Γi of the
y le a work is done on or by the system σ and some heat is transferred to or abstra ted
from this system. In the gedanken experiment this transfer of the amount qi of heat to
the system an be a omplished with the help of an innitesimal Carnot y le Ci working
between a single reservoir R0 of temperature T0 and another reservoir at a temperature
Ti (see Figure 1). One an imagine that the Carnot y le Ci delivers the amount qi of
heat from the reservoir at T0 to Ri and then the same amount of heat qi is transferred
from the reservoir Ri at Ti to the system σ . The last step - the transfer of qi to σ may
be irreversible; we do not assume that the system σ is in equilibrium with Ri nor even in
equlibrium in itself, so its temperature (as a single parameter hara terizing it) may well
not be denable at some or all stages of the y le.

39
 Sin e ea h Carnot y le Ci is perfe tly reversible, it takes, as follows from (50), from
R0 the heat

T0
qi . (51)
Ti
Moreover, sin e all Carnot y les Ci return to their initial states, and the same is assumed
about the system σ , the total amount W̄ of work obtained from the whole ompound y le
equals the total heat taken by all y les Ci from the reservoir at T0 . By Kelvin's prin iple,
the work W̄ annot be positive (for this would simply mean taking an amount of heat
from R0 and onverting it all into work). Thus

X T0
W̄ = qi ≤ 0. (52)
i
Ti

Taking the limit M →∞ and omitting T0 whi h is positive, one obtains in this way the
inequality

q
I
≤ 0, (53)
Text
19
known as the Clausius inequality (despite the fa t we derived it by relying on the
20
Kelvin's prin iple). The temperature under the integral has been here appended the
subs ript ext to stress (strongly!) the fa t that this is not the temperature of the
system σ (this may not be denable) but the temperature of the ( hanged in the pro ess)
reservoirs from whi h the heat q is supplied to the system σ and whi h all are in equilibrium
within themselves (they an be assumed to be su iently large).
Only if the y le exe uted by the system σ is reversible, an one identify the temper-
atures Text with the a tual (on a given stage of the system's σ y le) temperature of the
system σ (re all - Le ture II - that the heat transfer between two bodies an be realized
reversibly only if their temperatures are nearly equal). Moreover, if the y le exe uted by
σ is reversible, it an be exe uted in the opposite sense, leading to the inequality

q q
I I
− =− ≤ 0. (54)
Text T
In this ase, be ause the hanges of the system σ omposing the y le it exe utes are
reversible, the elementary heat q an be interpreted as the dierential form d¯Q on the
spa e of the system's σ equilibrium parameters (and, as said, Text identied with the
a tual system's temperature T ). Hen e, if the y le is reversible, the Clausius inequality
be omes the equality

d¯Q
I
= 0, (55)
T
19 This inequality is frequently erroneously written with the sign ≥, be ause students usually remember
that hanges of entropy are nonnegative and the fa tor under the integral q/Text is onfused with dS =
d¯Q/T ...
20 In lasses you will prove it relying dire tly on the Clausius's own prin iple.

40
whi h generalizes the equality Q1 /T1 + Q2 /T2 = 0 holding for Carnot y les.
The last equality allows to dene entropy S as a state fun tion, be ause it shows that
the integral of d
¯Q/T taken between the equilibrium states A and B of any thermodynam-
1
i al system does not depend on the reversible path between these states: If ΓA→B and
2
ΓA→B are two su h paths, then

d¯Q d¯Q d¯Q

I Z Z
0= = + , (56)
T Γ1A→B T Γ2B→A T

hen e

d¯Q d¯Q
Z Z
= . (57)
Γ1A→B T Γ2A→B T

The dieren e of entropies of the two equilibrium states A and B of a system an be then
dened as

B
d¯Q
Z
SB − SA = ,
A T
with the integral being taken over any reversible path onne ting A with B, and if
for every system a referen e state R is hosen and as ribed (arbitrarily) the value S0 of
entropy, then the entropy of any other state A an be dened as

Z B
d¯Q
SA = S0 + ,
R T
mu h in the same way as the internal energy U of every state has been dened with
respe t to the energy U0 of a referen e state (it is reasonable to take the same referen e
state R for U and S) by linking the states with adiathermal paths.

The existen e of entropy as a fun tion of state means that the heat form d¯Q is inte-
grable - the surfa es whi h are solutions of the equation (re all the material of lasses!)
d¯Q = 0 are simply the surfa es S = onst. and its integration fa tor, whi h is singled out
by its dependen e on the systems parameters x1 , . . . , xs only through the empiri al tem-
perature t(x1 , . . . , xs ) is just the absolute temperature T = φ(t(x1 , . . . , xs )). In reversible
adiathermal pro esses entropy S stays onstant. This also means that in a reversible
pro esses the heat form d¯Q an be written as

d¯Q = T dS . (58)

Using this relation whi h must hold, it is possible to show that if the integration fa -
tor of the form d¯Q is restri ted to depend only on t, the only its nonuniqueness and
nonuniqueness of entropy redu es to

1
T̃ = a T , S̃ = S + b. (59)
a
41
Indeed if alternative temperature T̃ and entropy S̃ are introdu ed so that d¯Q an be
written in two ways as

d¯Q = T dS = T̃ (T ) dS̃(x1 , · · · , xs ) , (60)

(sin e both, T and T̃ must be fun tions of t only, T̃ = T̃ (T )), one an pass to the new
set of variables ( oordinates on the manifold of states) with S being one of them and
y1 , . . . , ys−1 the remaining ones. In these new variables the above equality an be written
as
 ! ! 
s−1
T̃ (T )  ∂ S̃ X ∂ S̃
dS = dS + dyi  .
T ∂S i=1
∂yi
y1 S,y1′

From this, by omparing the oe ients of the dierentials on both sides, it readily follows
that (∂ S̃/∂yi )S,y1′ = 0, whi h implies that S̃ = S̃(S) and, moreover, that

T̃ (T ) dS̃(S)
= 1,
T dS
or that

dS̃(S) T
= .
dS T̃ (T )

Sin e the two sides of the above equality depend on dierent variables (S and T, respe -
tively) the equality an hold only if T /T̃ (T ) = 1/a = dS̃/dS . This gives the result.

We an now present another way of arriving at the same onsequen es of 2TMDL (still
relying on the Kelvin's prin iple) without using the Carnot y les and appealing more to
mathemati s (but understood physi ally, without mathemati al hieroglyphs). We will il-
lustrate it by onsidering a three parameter system, so that d¯Q = Y1 dx1 + Y2 dx2 + Y3 dx3 .
We begin by drawing in the system's parameter spa e an adiabat Γ ( onstru ted by mak-
ing steps orrelated by d¯Q = 0) rossing su essive (they are distributed ontinuously)
isothermal surfa es. Sin e the system has three parameters, it is possible to draw lines
beginning at the adiabat Γ and lying entirely within the respe tive isothermal surfa es.
We an therefore onsider two points P and P ′ lying on two su h lines on two dierent
but innitesimally lose isothermal surfa es ( orresponding to innitesimally lose tem-
′
peratures t and t ) and removed arbitrarily far from the points in whi h these lines start
from the adiabat Γ. The integral

Z P′
d¯Q ,
P

taken along the onstru ted adiabati lines (along the two ones lying on the isothermal
surfa es and the adiabat Γ) is zero. Then from the Kelvin's prin iple it follows that if P

42
and P′ are innitesimally lose to one another (one an arrange them so, be ause the two
isotherms on whi h they are situated are innitesimaly lose to one another), then the
′
segment P P must also give d ¯Q = 0 (more pre isely, the one-form d¯Q must give zero on the
′
innitesimal ve tor P P joining these two points) for otherwise one would have onstru ted
in this way a reversible y le in whi h the system takes the heat d ¯Q on the segment P P ′
and zero on its remaining parts. Sin e the y le would be reversible, this would violate
the Kelvin's prin iple. In this way we an onstru t a two-dimensional adiabati surfa es.
′
Moreover, sin e now we know that on the ve tor P P tangent to the surfa e the form d ¯Q
′′ ′′
gives zero, it annot give zero on a ve tor P P joining P with a neighbouring point P not
′′ ′′
lying in the onstru ted surfa e, so P P is not an adiabati line and, therefore, P annot
′′
be onne ted to P by any adiabati path (if it ould, then this path losed with P P would
be a reversible y le violating the Kelvin's prin iple). So the surfa e is entirely surrounded
by points ina essible on reversible adiathermal paths. Other adiabati surfa es an be
onstru ted similarly starting from any other urve along whi h d¯Q = 0. The reasoning
an be generalized to more ompli ated systems requiring more than three parameters
(one onstru ts in su h ases adiabati hypersurfa es of dimension s − 1).
The existen e of adiathermal surfa es an be also demonstrated taking as the start-
ing point the Caratheodory's formulation of 2TMDL, instead of the Kelvin's one. One
possible (more mathemati al) onstru tion is given in the material prepared for lasses.
Here we will briey show the same in a more intuitive way. The Caratheodory's prin iple
states that in the neighbourhood of any point (state) of any thermodynami al system
there are points ina essible in adiathermal hanges. From this it follows that there must
be even more points ina essible in reversible adiathermal hanges whi h form a more
narrow lass than the general adiathermal hanges. It is straightforward to see that if
all states of the system are on equal footing, the nearest points ina essible in reversible
adiathermal hanges from a given point P must be already innitesimally lose to it (and
not at a nite distan e from it). For if Q, the nearest ina essible point on su h paths
were at a nite distan e from P,
on the line representing a reversible pro ess onne ting
′
P with Q, there would have to exist a point P , arbitrarily lose to Q, a essible on
adiabati paths from P but not from Q (re all we onsider here reversible hanges, so
any reversible adiathermal hange an o ur in both dire tions). Then the nearest to
Q point ina essible from it adiabati ally would be infnitesimally lose to it (point P ′)
but not inntesimally lose to P - some points would be then distinguished. Hen e, the
Caratheodory's prin iple requires in fa t that innitesimally lose to any state of any
thermodynami al system there are points ina essible on adiabati (reversible adiather-
mal) paths. The onstru tion of adiabati surfa es pro eed then almost as previously:
starting from any point P one draws a d¯Q = 0 line through it, then an adiabati surfa e,
ontaining this line, then the three-dimensional hypersurfa e ontaining this surfa e, and
so on. Any surfa e an be then ordoned from both sides by other not interse ting it su h
hypersurfa es onstru ted by starting from points innitesimally lose to P on both sides
of the onstru ted hypersurfa e and ina essible from it adiabati ally.

As the adiabati surfa es an be onstru ted (the approa hes based on Kelvin's and

43
Caratheodury's prin iples merge at this point), one an now pro eed to showing that
in a reversible hange the heat form d¯Q an be written as T dS with T being the fun -
tion of the empiri al temperature t only. One starts by labeling the onstru ted adia-
bati (hyper)surfa es of dimension s − 1 by a parameter σ , onstru ting thereby a fun -
tion σ(x1 , . . . , xs ), mu h in the same way as the fun tion t(x1 , . . . , xs ) has been intro-
du ed by labeling dierent isothermal surfa es (and in dire t analogy with t the fun tion
σ(x1 , . . . , xs ) an be alled the empiri al entropy).
21
As the adiabati surfa es exist, from the rst Caratheodory's theorem it follows
that the form d¯Q(x1 , . . . , xs ) has an integration fa tor. The same an be also shown
here dire tly by using the method of Lagrange multipliers. Sin e on surfa es whi h are
solutions to d¯Q = 0, the fun tion σ(x1 , . . . , xs ) is onstant, in any adiabati (reversible
adiathermal) hange of the onsidered system

s
X ∂σ
dσ ≡ dxi = 0 , (61)
i=1
∂xi

22
similarly as

s
X
d¯Q = Xi dxi = 0 . (62)
i=1

Therefore,

s  
X ∂σ
d¯Q − λ dσ ≡ Xi − λ dxi = 0 , (63)
i=1
∂xi

where λ(x1 , . . . , xs ) an be an arbitrary state fun tion (on the spa e of states of the
system). In this relation only s−1 dierentials dxi are independent ( an be hanged
at will), be ause they are (at every point) orrelated by (61). If one regards, dx1 as
determined by the remaining dierentials, one an adjust the arbitrary state fun tion
23
λ(x1 , . . . , xs ) so as to make the oe ient of dx1 in (63) vanish:

∂σ
X1 (x1 , . . . , xs ) = λ(x1 , . . . , xs ) . (64)
∂x1
Then, sin e (63) must hold for any adiathermal reversible hange in whi h dx2 , . . . , dxs
an be hosen arbitrarily, one on ludes that

∂σ
Xi (x1 , . . . , xs ) = λ(x1 , . . . , xs ) , (65)
∂xi
21 Hopefully dis ussed in lasses.
22 Here X are not ne essarily the generalized for es be ause we in luded in them also the terms following
i
from dU (x1 , . . . , xs ).
23 And this is the essen e of the tri k with the Lagrange multipliers.

44
now for all i = 1, . . . , s. Hen e

s s
X X ∂σ
d¯Q = Xi (x1 , . . . , xs ) dxi = λ(x1 , . . . , xs ) dxi = λ dσ . (66)
i=1 i=1
∂xi

1/λ is therefore an integrating fa tor of d¯Q. One has, however, to show (and this does
not follow dire tly from the Caratheodory's rst theorem) that λ annot be an arbitrary
x1 , . . . , xs but ne essarily takes a form of a produ t of two fun -
fun tion of the parameters
tions: one whi h depends on x1 , . . . , xs through the empiri al temperature, and another
one whi h depends on these parameter through the fun tion σ itself.
To this end we onsider two systems, 1 and 2, the rst one hara terized by the
variables x1 , . . . , xs and the se ond one by y1 , . . . , yr , ea h of whi h has its own heat
form (d
¯Q1 and d¯Q2 ), its own system of labeling adiathermal (hyper)surfa es (fun tions
σ1 (x1 , . . . , xs ) and σ2 (y1 , . . . , yr )) and an integration fa tor of its heat form (fun tions
λ1 (x1 , . . . , xs ) and λ2 (y1 , . . . , yr )). We now imagine these two system brought into thermal
onta t through a diathermal wall and in equilibrium, so that their temperatures are equal

t1 (x1 , . . . , xs ) = t = t2 (y1 , . . . , yr ) . (67)

The ompound system, being thermally homogeneous, is therefore hara terized by s+

r − 1 variables for whi h one an take the ommon temperature t and x2 , . . . , xs and
y2 , . . . , yr . As any thermally homogeneous thermodynami al system it too must have its
fun tion Σ(t, x2 , . . . , xs , y2 , . . . , yr ) and its heat form d ¯Q must have an integrating fa tor
Λ(t, x2 , . . . , xs , y2 , . . . , yr ). Sin e d¯Q = d¯Q1 + d¯Q2 , the following relation must, therefore,
hold

Λ(t, x2 , . . . , xs , y2 , . . . , yr ) dΣ(t, x2 , . . . , xs , y2 , . . . , yr ) = λ1 (t, x2 , . . . , xs ) dσ1 (t, x2 , . . . , xs )

+λ2 (t, dy2 , . . . , yr ) dσ2 (t, y2 , . . . , yr ) .

Dividing both sides by Λ and going over to the variables t, σ1 , σ2 , x3 , . . . , y3 , . . . one has

λ1 λ2
dΣ(t, σ1 , σ2 , x3 , . . . , y3 , . . .) = dσ1 + + dσ2 .
Λ Λ
It is now lear that Σ is a fun tion of σ1 and σ2 only (be ause only dierentials of these
two variables appear on the right hand side), so any dependen e on x3 , . . . , y3, . . . and t
on the right hand side must drop out. But sin e λ1 does not depend on y3 , . . . , yr , neither
an Λ, and in the same way, sin e λ2 does not depend on x3 , . . . , xs , neither an Λ. It
then follows, that

λ1 = λ1 (t, σ1 ) , λ2 = λ2 (t, σ2 ) , Λ = Λ(t, σ1 , σ2 ) ,

and, moreover, the dependen e on the empiri al temperature t must also drop out from
the ratios λ1 /Λ and λ2 /Λ:
   
∂ λ1 ∂ λ2
= = 0,
∂t Λ ∂t Λ

45
and from this it follows that

∂ ∂ ∂
ln λ1 (t, σ1 ) = ln λ2 (t, σ2 ) = ln Λ(t, σ1 , σ2 ) . (68)
∂t ∂t ∂t
All these three derivatives must be therefore equal to the same fun tion of the empiri al
temperature (the only variable whi h is ommon in these derivatives), say g(t), whi h
must be a universal fun tion (the same for all thermodynami al systems). Integrating the
rst two equalities

∂ ∂
ln λ1 (t, σ1 ) = g(t) , ln λ2 (t, σ2 ) = g(t)
∂t ∂t
one gets that
Z 
λi (t, σi ) = Wi (σi ) exp dt g(t) ,

(The fun tions W1 (σ1 ) andW2 (σ2 ) are the integration onstants.) In this way
 Z 
1
d¯Qi = λi dσi = a exp dt g(t) Wi (σi ) dσi , (69)
a
It remains to show that the integrating fa tor Λ of the ompound system also has this
stru ture, with the same universal fun tion of the empiri al temperature and the fun tion
W whi h depends on σ1 and σ2 only through Σ. From (68), in the same way as above, it
follows that
Z 
Λ(t, σ1 , σ2 ) = W (σ1 , σ2 ) exp dt g(t) .
R

Sin e the fa tor exp dt g(t) drops out from the equality d¯Q = d¯Q1 + d¯Q2 , one an write

W (σ1 , σ2 ) dΣ(σ1 , σ2 ) = W1 (σ1 ) dσ(σ1 ) + W2 (σ2 ) dσ(σ2 ) ,

from whi h it is lear that

∂Σ ∂Σ
W (σ1 , σ2 ) = W1 (σ1 ) , W (σ1 , σ2 ) = W2 (σ2 ) .
∂σ1 ∂σ2
It is now su ient to dierentiate the rst of these two equalities with respe t to σ2 and
the se ond one with respe t to σ1 to arrive at the relation

∂W ∂Σ ∂W ∂Σ ∂(W, Σ)
− ≡ = 0,
∂σ1 ∂σ2 ∂σ2 ∂σ1 ∂(σ1 , σ2 )
whi h means that the mapping of (σ1 , σ2 ) into (W, Σ) is of rank one, that is, its image is
a urve, so that W (σ1 , σ2 ) = W (Σ(σ1 , σ2 )). Hen e,

W (σ1 , σ2 ) dΣ = W (Σ(σ1 , σ2 )) dΣ ,

46
whi h is just what we wanted to prove.
Thus the heat form d¯Q of any thermally homogeneous thermodynami al system an
be written as
 Z 
1
d¯Q = λ dσ = a exp dt g(t) W (σ) dσ ≡ T (t) dS ,
a

with T being a universal fun tion of the empiri al temperature t and

S
1
Z
S= dσ W (σ) ,
a
a fun tion whi h is onstant in reversible adiathermal hanges. S therefore the true
entropy of the system.

Summarizing, in reversible hanges the heat form of any thermodynami system an

be written as d¯Q = T dS and in su h hanges

s−1
X
dU = T dS + Xi dxi . (70)
i=1

This dierential relation an be now extended to all dierential hanges, reversible ones
and also irreversible ones, be ause it is simply the relation between the state fun tions at
innitesimally lose points (representing equilibrium states of the system). However one
should remember, that only in reversible hanges the dierentials T dS and Xi dxi have
the meaning of the heat absorbed by the system and of the work(s) done on it. In other
words, it is only in reversible hanges that one an make the identi ations

s−1
X
d¯Q = T dS , d¯W = Xi dxi . (71)
i=1

To give an example: onsider a gas expanding into an additional volume of the ylinder
when the gas (the whole ylinder) is adiathermally isolated. As dis ussed, this hange
an be a omplished quasistati ally, su essively opening additional volumes dV , but is
always irreversible. (This an now be quantied by simply omparing entropies of the
initial and nal states - see Le ture V). In the expansion of the gas into opened additional
volume dV , ertainly −p dV 6= 0, and, as is lear from 1TMDL also T dS 6= 0 be ause
dU = 0 (no work is done on the gas and, as the ylinder is assumed to be adiathermally
isolated, no heat is absorbed by the system) and the two terms must ompensate ea h
other: p dV = T dS .
The extension we have made here is the basis of the frequently used way of omputing
the hange of a state fun tion (be is energy U or anything else) in a pro ess A → B
in whi h the system passes from an equilibrium state A to another equilibrium state B ,
but whi h pro ess by itself may not be reversible, by saying that one an integrate the
appropriate forms along any reversible path whi h onne ts the states A and B . One does

47
not need to really point this reversible path be ause all one is doing is just omparing the
dierential hanges of the state fun tions.
Having introdu ed entropy as a state fun tion, we an express the heat apa ities of
systems in reversible pro esses in a onvenient way. Re all that in any quasistati hange
(in whi h the system passes through a sequen e of equilibrium states) the system's heat
apa ity C is dened by proje ting the form d¯Q onto the urve in the spa e of states
representing this pro ess:

d¯Q|projected = C(x1 (T ), . . . , xs (T )) dT . (72)

If the hange is reversible (and only then) one an repla e d¯Q by T dS and write

d
C(x1 (T ), . . . , xs (T )) = T S(x1 (T ), . . . , xs (T )) . (73)
dT
In most ases the pro ess is spe ied by onstan y of ertain parameters, e.g. V or p in
the ase of simple uids, and this is written as

   
∂S ∂S
CV = T , Cp = T . (74)
∂T V ∂T p

One more important thing follows from the presented reasonings: in the ase of systems
in thermal equilibrium (like the two systems onsidered above), that is having the same
t and therefore T,

d¯Q = d¯Q1 + d¯Q2 = T dS1 + T dS2 = T d(S1 + S2 ) = T dS , (75)

so the entropy is additive. If the systems are not at the same temperatures but are not
in thermal onta t (are separated by an adiathermal wall) one denes the entropy of the
ompound system as S = S1 + S2 .

48
 LECTURE V (TMD)

A epting 2TMDL as a generally valid law of Physi s whi h applies to any ma ros opi
thermodynami al system, we have shown that the heat form d¯Q of any (thermally homoge-
neous) system an, in reversible pro esses, be written as T dS , where T = T (t(x1 , . . . , xs ))
is the absolute temperature (so, the indi ator telling us whether two system brought into
a thermal onta t will be in equilibrium) and S(x1 , . . . , xs ) the new state fun tion - the
entropy. In innitesimal reversible pro esses 1TMDL, dU = q + w , an therefore be
written as the sum

s−1
X
dU = T dS + Xi dxi , (76)
i=1

of two (inexa t) dierential forms on the spa e of the system's equilibrium states, of whi h
the rst one represents the heat absorbed by the system in su h a pro ess, and the se ond
one is the sum of works. We have in this way obtained a onvenient representation of heat
apa ities hara terizing the system in reversible pro esses as derivatives of its entropy
 
∂S
CX = T . (77)
∂T X

We have also stressed that the formula (76) remains valid in any innitesimal hange, as
relating hanges of the state parameters ex ept that if the hange is not reversible, the
interpretation in terms of heat and works of its individual terms is lost.
Restri ting now the attention to the paradigmati example of a uid (as a simple
system) one immediately noti es that the fundamental relation (76), whi h in this ase
takes the form

dU = T dS − p dV , (78)

implies various, a priori not obvious, relations. For instan e, sin e U is a state fun tion
whi h an be treated as a fun tion of the variables S and V , one obtains from (78) that
   
∂U ∂U
T = , p=− . (79)
∂S V ∂V S

Furthermore, its mixed se ond derivatives must be equal, whi h gives the relation
   
∂T ∂p
=− . (80)
∂V S ∂S V

This is a relation between two measurable oe ients: the one on the right hand side
an be obtained experimentally by keeping the volume of the uid xed and transferring
to it reversibly heat whi h is measured; in this way the hange ∆S of the uid's entropy
an be determined as ∆S = ∆Q/T ; if this is divided by the measured resulting hange

49
∆p of the uid's pressure, the right hand side is (although not so easily, as is always the
ase with alorimetri measurements) obtained. The oe ient on the left hand side is
mu h easier to measure: one simply measures the hange ∆T of the uid's temperature
resulting from a hange ∆V of the uid's volume in the adiathermal (stri tly speaking
adiabati ) onditions.
The relation (80) and numerous other relations of this sort whi h an be derived treat-
ing U or S (or other state fun tions mentioned at the end of Le ture II) as a fun tion
of dierent pairs of variables, onstitute one of the most important predi tions of phe-
nomenologi al thermodynami s. From the mathemati al point of view all of them follow
rather trivially from the existen e of entropy as a state fun tion, that is, from the inte-
grability of the heat form d¯Q; from the physi al point of view they are highly nontrivial
onsequen es of 2TMDL whose validity is not (as we will show on an illustrative example
below) a mathemati al ne essity: 2TMDL is the ma ros opi ree tion of the intrinsi
working of the Nature.
One more, less trivial, example of su h relations: onsider a nonsimple system, a
uid or a solid under the hydrostati pressure, whi h exhibits diele tri properties. To
hara terize its equilibrium states three parameters, e.g. p, V and P are needed and the
fundamental relation (76) reads (P is the system's total polarization ve tor and E is the
ele tri eld in whi h the system is pla ed)

dU = T dS − p dV + E ·dP . (81)

Instead of onsidering the internal energy U of this system, one an onsider another
thermodynami potential (a state fun tion), all it Φ, dened as24 Φ = U −ST +pV −E ·P.
It is straightforward to see that

dΦ = −SdT + V dp − P·dE . (82)

In this way Φ is treated as the state fun tion of the independent variables T, p and E,
whi h are all easy to ontrol experimentally. It also follows that
     
∂Φ ∂Φ ∂Φ
= −S , =V , = −P . (83)
∂T p,E ∂p T,E ∂E T,p

Equality of the se ond mixed derivatives of the potential Φ now implies (among others)
the relation
   
∂P ∂V
=− . (84)
∂p T,E ∂E T,p

This shows that there must be a nontrivial relation between two seemingly unrelated
phenomena: piezoele tri ity - generation of a polarization of a material as a result of
squeezing it - and ele trostri tion - whi h is the hange of the material's volume in the
ele tri eld. The two are somehow intimately related by 2TMDL, that is, by the inner
working of the Nature.

50
 g g

A B B

Figure 2: Two identi al balls in the Earth's gravitational eld. On the right: alternative
view on the ball B as pla ed in the gravitational eld pointing upwards.

To illustrate better the deeply physi al hara ter of 2TMDL, let us onsider the
25
following example. Two identi al homogeneous balls A and B of mass M made of,
say, iron or another material of nonnegligible thermal expansivity property, have the
same initial temperature. One of them, A rests on a horizontal plane (in the Earth's
gravitational eld g), say, on a table, and the other one, B, is suspended on a thread (see
the left Figure 2). The same quantities δQ of heat are supplied to both balls. Whi h
one will have higher temperature? This problem was on e assigned at the International
Physi s Olympiad and then appeared in many sour es. The expe ted answer was that
the heat supplied to the ball A auses two ee ts: one is the raising of the ball's enter
of mass in the gravitational eld as a result of its thermal expansion and the other ee t
is the in rease of the ball's internal energy resulting in raising its temperature; so in this
ase part of the heat δQ is used up for a me hani al work. In the ase of the ball B
instead, thermal expansion lowers the position of its enter mass, so in addition to the
heat supplied, also the hange of its potential energy ontributes to in reasing its internal
energy. The (expe ted) on lusion, therefore, was that it is the ball B whi h will have at
lin
the end higher temperature. Putting all this in equations: let α = (1/R)(dR/dT ) be
the linear expansion oe ient of the ball and C0 its heat apa ity (whi h we an take to
be independent on the temperature; for simpli ity we an assume that all the experiment
is arried out at zero pressure) in the absen e of the gravitational eld. Then

ball A : δQ = (C0 + MgR αlin) δTA ,

ball B : δQ = (C0 − MgR αlin) δTB ,
from whi h it readily follows that δTB > δTA . In the following it will be onvenient to
treat the ball B as glued to the same horizontal plane as the ball A but in the gravitational
eld pointing upwards (see the right Figure 2): in this way the two situations A and B
are distinguished by the sign of g and one an onsider g as varying ontinuously between
positive (dire ted downwards) and negative (upwards) values. Using this onvention one
lin
an introdu e C(g) ≡ C0 + MgRα as the ball's heat apa ity in the gravitational eld.
Then
 
∂C
= MR αlin . (85)
∂g T
24 Φ is the Legendre transform of U. We will tell more about this transformation later.
25 G. De Palma, M.C. Sormani, Am. J. Phys. 83, 723 (2015).

51
 T1 = T2 − δT
g

2 δR

T1 T2 = T1 + δT T1

Figure 3: A heat engine exploiting the thermal expansion of the ball.

Yet this reasonably looking solution has been found (by the authors of the ited paper)
to be in oni t with 2TMDL! To see this one an onsider the following y le shown
s hemati ally in Figure 3. Start with the ball on the table at a temperature T1 . Then
bring it (not hanging its position in the gravitational eld) into the thermal onta t with
a heat bath of temperature T2 = T1 + δT > T1 . A ording to the dis ussed solution,
lin
the ball will absorb heat δQabs = (C0 + MgR α )δT and its enter of mass will raise by
δR = R αlinδT . Now a thread xed to the eiling an be atta hed to it without hanging
the ball's new position in the gravitational eld. Then the ball an be onne ted to a heat
bath at the temperature T1 . A ording to the  solution presented above, the ball will
δQlost
loose the heat = (C0 − MgR αlin)δT and its enter of mass will raise by another
lin
δR = R α δT so that part of the heat absorbed from the reservoir at T2 > T1 goes
into in reasing the balls potential energy by 2δR. This potential energy an be onverted
into a me hani al work δ W̄ bringing at the same time the ball to its initial position
and ompleting thereby the y le (operating between the reservoirs at the temperatures
T2 = T1 + δT > T1 and T1 with the ball as the working body). One an now ask what is
the e ien y η of the y le? This is easily omputed as the ratio of the work done to the
heat absorbed:

δ W̄ 2MgR αlinδT 2MgR αlin

η= = = .
δQabs (C0 + MgR αlin)δT (C0 + MgR αlin)

It is learly independent of the temperature dieren e δT of the two reservoirs! But

a ording to 2TMDL the e ien y of any y le annot ex eed that of the reversible
Carnot y le

T1 T1 δT
ηCarnot = 1 − =1− = ,
T2 T1 + δT T1 + δT
whi h de reases to zero as δT goes to zero. Choosing δT su iently small, on ould, if
the presented solution were right, beat the e ien y of the Carnot y le!
What was wrong? We have assumed that the ball in the gravitational eld does not
get deformed - that its shape is spheri al independently of whether it rests on the table or
is suspended on a thread. This is a eptable as a mathemati al assumption but not as a
physi al one. In reality the ball whi h rests on the table will be somewhat squashed while
the one suspended will be stret hed. These deformations do not disappear as δT → 0
and ultimately will save 2TMDL. Putting things the other way around: it is 2TMDL

52
whi h tells us that the deformations annot be negle ted. Mi ros opi ally, a solid out of
whi h balls are made is omposed of mole ules whi h intera t ea h with the other ones
(or at least with the nearest ones). This an be modeled by small masses onne ted with
springs (the simplest model of a solid). The internal energy of the ball is the sum of
kineti energies of the mole ules forming it and of potential energies of the springs. In the
gravitational elds these springs get either ompressed or stret hed and this ne essarily
has some impa t on the ball's internal energy independently of the hange of the height of
its enter of mass and the hange of the related potential energy whi h is not in luded in
the ball's internal energy. While a detailed mi ros opi analysis of all the ee ts involved
would be very ompli ated, thermodynami s allows to take these ee ts into a ount
phenomenologi ally without the need of delving into the mi ros opi onstitution of the
ball.
Applying thermodynami s to the ball we must only assume that the ball's internal
energy depends on its temperature and on the gravitational eld g (here the pi ture of
the ball glued to the table and allowing for the variable sign of g allows to make the
analysis simple): U = U(T, g) and that the relevant for the problem parameters T, Y -
the ball's enter of mass verti al position and g are related by an equation of state

f (T, Y, g) = 0 .
Applied to the ball 1TMDL takes the form

δQ = δU + MgδY ,
or, if reversible hanges are onsidered, the familiar form

dU = T dS − MgdY .
(The sign of the se ond term is the same as in 1TMDL dU = T dS − p dV applied to a
uid be ause in in reasing Y the ball must do a positive work against gravity - if g>0 -
just in the same way as the uid must do a work against an external pressure.) The rest
is the matter of rudimentary thermodynami al omputations: to redu e the ne essary
steps to minimum, it is onvenient (as always when one ontrols - here mentally only -
a ertain parameter like g) to form the analog of enthalpy, dened as H = U + MgY ,
whose dierential is
   
 
∂S ∂S
dH = T dS + MY dg = T dT + T + MY dg
∂T g ∂g T
   
∂H ∂H
≡ dT + dg .
∂T g ∂g T
Its se ond form shows that H is treated here a fun tion of T and g (that is, of the
ontrolled parameters). Sin e H is a fun tion of state, its se ond mixed derivatives must
be equal:
  !     
∂ ∂S ∂ ∂S
T = T + MY .
∂g ∂T g ∂T ∂g T g
T

53
This leads to the relation (one of the Maxwell identities)

   
∂S ∂Y
= −M .
∂g T ∂T g

Furthermore, sin e T (∂S/∂T )g , as we already know, is just the system's (here ball's) heat
apa ity at xed g , the dierential dH an be written, using the derived relation, in the
form
"  #    
∂Y ∂H ∂H
dH = Cg dT + M Y − T dg ≡ dT + dg .
∂T g ∂T g ∂g T

Applying to this form on e again the equality of the mixed se ond derivatives of H we
nd that
" #
∂2Y
    
∂Cg 2 ∂αlin
= −MT ≡ −MT Y αlin + , (86)
∂T T ∂T 2 g ∂T g

This is markedly dierent than the naive formula (85)! In parti ular, the sign of the
derivative is opposite if, as usually happens with real materials, (∂αlin /∂T )g > 0 (or
2
at least if this derivative is not too large negative to outweight the positive αlin term).
Therefore, the heat apa ity C of the ball is larger if g is negative (dire ted upwards, or
the ball is suspended on a thread) than when g is positive. As a result the orre t answer
to the problem is δTA > δTB .
This example is ni e be ause it learly illustrates the status of 2TMDL as the physi al
law. Mathemati ally one ould imagine a world in whi h the ball is innitely rigid and its
shape does not get deformed when it is pla ed in the gravitational eld. Yet, 2TMDL tells
us that in the real physi al world this is impossible. 2TMDL, as said, generalizes results of
many experiments and by this impli itly takes into a ount how the real matter behaves
and what are the ma ros opi al onsequen es of its mi ros opi (mole ular) onstitution.
The example also illustrates the working of thermodynami s as a phenomenologi al the-
ory: the internal energy of the ball must somehow be modied by the presen e of the
gravitational eld - and we do not need to investigate this in detail; we only know that
this is somehow ree ted in the equation of state f (T, Y, g) = 0 and this di tates how
U and C depend on g; we need to nd the equation of state and part of the informa-
tion needed for this is ontained in the thermal expansion oe ient αlin ; measuring this
oe ient is enough to tell what will be the answer to the assigned question.

We now return to the Clausius inequality whi h in Pippard's words holds the key to
the dieren e between reversible and irreversible pro esses ( hanges). It allows to de ide
whether a given pro ess (in an isolated system) is possible.
Let the two equilibrium states A and B of a system be onne ted by two paths (pro-
1 2
esses), ΓA→B and ΓA→B , of whi h the rst path is irreversible, while the se ond one is
reversible. The Clausius inequality (53) applied to the y le A → (1) → B → (2) → A

54
 B

ad

A
is C

VA =VB VB

Figure 4: Example of a y le exe uted by the perfe t gas. The pro ess A −→ C −→ B is
reversible. The iso hori pro ess A −→ B an be irreversible or reversible.

an be written in the form

q q
Z Z
+ < 0.
Γ1A→B Text Γ2B→A Text

But in the reversible hange q an be written as the heat form d¯Q over the spa e of the
system's parameters and Text may be identied with the system's temperature T, so

q d¯Q
Z Z
=− = −(SB − SA ) ,
Γ2B→A Text Γ2A→B T

be ause 2TMDL tells us that d¯Q/T = dS . Thus in the irreversible hange A→B
q
Z
< ∆S ≡ SB − SA . (87)
Γ1A→B Text

that is, the in rease of the system's entropy in any transition (reversible or irreversible)
between the states A and B is never smaller than the amount of the quantity q/Text
absorbed by the system in this transition. In innitesimal hanges (when the points A
and B are innitesimally lose to one another in the spa e of states)

q ≤ Text dS . (88)

The Clausius inequality (87) an be simply illustrated by a pro ess in whi h the perfe t
gas (for simpli ity of onstant heat apa ity CV ) passes at onstant volume VA (not
ne essarily reversibly) from the pressure pA to the pressure pB > pA (see Figure 4). The
same nal state an be rea hed reversibly by rst expanding the gas isothermally to the
volume VC > VA and pressure pC and then ompressing it adiabati ally (adiathermally
and reversibly). Sin e (as will be obtained in lasses) SB − SA = CV ln(TB /TA ), the
inequality (87) applied to the transition A→B takes the form

B
q TB
Z
≤ CV ln .
A Text TA
One an now ontemplate dierent iso hori pro esses A → B. If the internal energy
of the gas is in reased by doing on it work using the Joule paddle-wheel devi e, q ≡0

55
and the above inequality is trivially satised. The same an be a hieved heating the gas
irreversibly by onne ting it thermally with a reservoir of heat of onstant temperature
Text ; the heat will ow from the reservoir to the gas provided Text ≥ max(TA , TB ) = TB ,
so

B A
q 1 QA→B CV (TB − TA ) CV (TB − TA )
Z Z
= = = ≤
A Text Text A Text Text TB

(In the last step we have taken into a ount that QA→B ≡ UB − UA is also equal to
WC→B = CV (TB − TC ) = CV (TB − TA ), as an easily be omputed using the perfe t
gas adiabat equation p V CV /Cp = onst., and the relation Cp − CV = nR, be ause in
the isothermal transition A → C the energy of the perfe t gas does not hange). The
inequality is then equivalent to the inequality 1 − x ≤ − ln x whi h is true be ause
0 < x ≡ TA /TB < 1. Finally one an onsider the reversible iso hori heating of the gas
by onne ting it with a sequen e of reservoirs of appropriately in reasing temperatures;
then (using the fa t that in the iso hori pro ess d¯Q = dU and dU = CV dT in the ase
of perfe t gas)

B B TB TB
q d¯Q dU(T ) dT
Z Z Z Z
= = = CV ,
A Text A T TA T TA T

in whi h ase the inequality (87) is satised as equality.

Returning to the general onsiderations, the very important ase arises when the
system is adiathermally isolated during the hange A →B (whi h means that q = 0).
The inequality (87) takes then the form

∆S ≥ 0 . (89)

This is the elebrated entropy in rease law (or just the entropy law):
The entropy of an isolated system an never diminish.
Clausius expressed it in the ategori al German (Prussian) way: Energie der Welt is
konstant. Entropie der Welt strebt einem maximum zu.

The entropy law gives a thermodynami al riterion allowing to de ide whi h pro esses
of those whi h ould o ur (are allowed by 1TMDL) an a tually o ur. Following Pippard
we are now going to briey dis uss its operation on three simple examples.
Consider rst two bodies at dierent temperatures T1 and T2 , say T2 > T1 , whi h
together form an isolated system. If the bodies are brought momentarily into thermal
onta t, a portion of heat q>0 will ow from the hotter one (that at T2 ) to the older
one. As a result the entropy of the hotter one will de rease by ∆S2 = −q/T2 and that of
the older one will in rease by ∆S1 = +q/T1 . The total hange of entropy of the entire
isolated system is positive:

 
1 1
∆S = ∆S2 + ∆S1 = q − + > 0.
T2 T1

56
 26
The reverse ow of the heat is forbidden by the entropy law.
As the se ond example, onsider a quantity of a gas in one part of an adiathermally iso-
lated ylinder separated from the se ond empty part of the ylinder. If the wall separating
the two parts is removed, or just a hole is pier ed in it, the gas will expand (adiathermally
and) irreversibly and its entropy will in rease (as you will al ulate dis ussing in lasses
this so- alled Joule pro ess in more detail).
In these two simple examples the hanges were the transitions from one equilibrium
state to another equilibrium state as a result of altering (weakening of ) the on-
straints to whi h the systems were subje t (in the rst ase the onstraint ould be
represented by an adiathermal wall whi h gets then repla ed by a diathermal one and in
the se ond ase the onstraint was the wall impermeable to the gas mole ules whi h got
removed). In these two ases the operation of the entropy law is lear.
The third example is a moving body whi h omes to rest due to the presen e of fri tion.
In this ase the de rease of the body's kineti energy is a ompanied by the in rease of
its and of the surrounding's temperatures. The entropy of the entire system (body and
the surrounding) has in reased. In this example the initial state is not, stri tly speaking,
an equilibrium state, but here entropy of the moving body whi h is in equilibrium within
itself in its own rest frame an without any in onsisten y be dened as the entropy of
< c, this ould
the body in its rest system (in relativisti treatment of the body, when v ∼
be problemati ). In general, however, with the ex eption of trivial situations like that in
the third example, the entropy law is appli able to transitions between equilibrium states
only.
For a given set of onstraints to whi h it is subje t, a thermodynami system has
only one true equilibrium state and the entropy law an be formulated as the statement
(Pippard again)
Entropy law: It is not possible to vary the onstraints of an isolated system in su h a
way as to de rease the entropy.
This formulation stresses the role of onstraints and will be the basis of the Callen's
formulation of thermodynami s whi h we will dis uss shortly. Before, we will use it
(applied to the se ond example above) to onsider briey the role of u tuations. When
the gas lls the whole ylinder and is in equilibrium, its lo al density ρ seems to be uniform.
It is however subje t to ontinuous minute u tuations most of whi h are pra ti ally
undete table by ma ros opi measuring instruments. In reality there is a ontinuous
spe trum of u tuations ranging with de reasing probability from very small to very large.
So very rarely a large u tuation an o ur, e.g. su h that - taking things to the extreme -
27
the entire gas spontaneously on entrates in the smaller but ma ros opi volume (whi h

26 2TMDL gives in fa t the proof of the possibility to onsistently orrelate the s ale of temperature
with the dire tion of the heat ow (whi h served to dene operationally the notions hotter and  older).
Logi ally, this requires formulating 2TMDL without referen e to hotter and older, that is not to base it
on the Clausius formulation.
27 Of ourse the probability of su h a u tuation in a gas onsisting of ∼ 1023 mole ules is so fantasti ally
small that pra ti ally there is no han e to observe su h a u tuation in pra ti e, even waiting as long
as the universe's lifetime.

57
it originally left as a result of removing the separating wall or pier ing a hole in it). The
question may be then asked: what happens to the entropy of the gas during su h a large-
s ale u tuation? The orre t answer is: nothing! The thermodynami al entropy whi h
28
we are onsidering here stays un hanged. The ontinuous spe trum of u tuations,
from the minute to the largest possible ones, are part of the thermodynami equilibrium
state and not a departure from equilibrium. The entropy S whi h is as ribed to an
equilibrium state of the system is not as ribed to one (most probable) of its mi ros opi
ongurations but to the omplete set of mi ros opi ongurations the system an be
in. This will be ome lear in the statisti al approa h whi h we will dis uss later. The
important lesson whi h should be drawn from these onsiderations is that entropy S
(and other state fun tions) must be regarded as a property of the system and of its
onstraints -in the onsidered example of the gas and the ylinder. Only in this way
one an understand the statement that S is a fun tion of the gas internal energy U and
its volume V (the volume of the ylinder in ee t, and not of the volume o upied by
the gas at a parti ular instant). Therefore if the gas in the smaller volume be ause of the
wall (is subje t to stronger onstraint) its entropy has one value and it has another value,
when it is in the larger volume (weaker onstraint) and it is the very a t of removing the
wall (or pier ing the hole in it) whi h in reases the thermodynami al entropy, that is the
a t of hanging the onstraints. It follows that in thermodynami s we never talk about
 oming to equilibrium during whi h entropy gradually in reases. On e the wall blo king
the gas from expanding into the larger volume has been removed, the entropy in reases
and the mi ros opi onguration of the gas (whi h for a short while is still mostly in the
initial smaller volume) is now treated as a huge (very improbable to o ur spontaneously,
as we have said, but here fabri ated by an external agent) u tuation whi h however is
part of the set of all mi ros opi ongurations the gas an assume being subje ted to the
weakened onstraints.
Similarly in the rst example onsidered, it is the a t of repla ing the adiathermall
wall separating the two bodies by a diathermal one whi h in reases the system's entropy,
and not the subsequent ow of heat. On e the diathermall wall is introdu ed, the system
is treated as nding itself in a huge (very improbable to o ur spontaneously) u tuation
of the distribution of the kineti energy between all its mole ules, but a u tuation whi h
is part of the new equilibrium state.
Thus any thermodynami hange (a pro ess) should be viewed as a hange of the
onstraints and the entral problem of thermodynami s is, as Callen says, to determine
the equilibrium state orresponding to the given set of new onstraints. It is therefore
the se ond way of formulating the entropy law whi h is the most adequate one, be ause

28 In the kineti theory of gases one deals with a quantity alled H - introdu ed by Boltzmann - whi h
is sometimes identied with entropy or, more pre isely, with −S/kB . The famous Boltzmann H -theorem
states that H always (although this always also requires some quali ations) de reases with time, so
−kB H exhibits a property whi h makes it similar to the thermodynami al entropy. It should be stressed,
however, that −kB H should be more properly alled the kineti entropy and it is only in the innite time
limit (and in the thermodynami al limit) that it an legitimately be identied with the thermodynami
entropy (whose time evolution annot even be dis ussed).

58
it stresses the essential role of onstraints to whi h the system is subje ted.
It is interesting to follow here further the dis ussion of these matters presented by
Pippard for it is interesting and sheds light on how thermodynami s as a phenomenologi al
theory lives out of reasonable idealizations. If, says Pippard, one follows this point of view
on entropy to its ultimate logi al onsequen es, one should ome to the on lusion that
the entropy of the Universe is xed on e for ever be ause no real walls are absolutely
impermeable to matter nor no walls are perfe tly adiathermal. The state of the Universe
we are ontemplating should be then taken to be only a huge u tuation of a more or less
uniform density and temperature, that is a huge u tuation of an equilibrium state. But,
says Pippard, leaving aside the question whether the expanding universe an be treated
as an isolated system, su h a point of view is not useful and does not allow to make any
predi tions. A more pragmati attitude is to make reasonable ompromises, that is, to
relay on reasonable idealizations: although no walls are absolutely impermeable to matter,
on time s ales relevant to observed pro esses one an treat some portions of the Universe
as isolated. This is similar to the state of metastable equilibrium as that of a mixture of
oxygen and hydrogen whi h an for most purposes be treated as a true equilibrium state
be ause the hemi al rea tion between the two gases, if not arti ially stimulated, o urs
at a negligible rate. One is then able to dene entropy of physi al systems of interest,
apply to them the entropy law and to make predi tions.
Further, the point of view that entropy and other state fun tions of the system are
determined by the onstraints, whi h is natural in that u tuations nd their pla e in
the s heme of thermodynami s, entails a some somewhat strange onsequen e that the
entropy law is not universally valid: suppose the two bodies at dierent temperatures are
separated by an adiathermal wall. The total entropy of the system is S1 + S2 . Let them
to onta t through an diathermal wall. Then, a ording to the view adopted above, the
entropy of the system instantaneously in reases and be omes larger than S1 + S2 , most
likely by a signi ant amount. But if the thermal onta t of the bodies is broken before
their temperatures equalize, their entropy de reases to nearly the initial one. However, one
should rstly remark that in the above reasoning one talks about time whi h is nonexistent
in thermodynami s. Furthermore, one should noti e that in the omplete experiment the
total entropy nevertheless does not de rease and the dieren e of the temperatures of the
bodies annot in rease - no useful de rease of entropy an be obtained in this way. Su h
paradoxes an therefore be tolerated.
The origin of su h paradoxes is some in onsisten y in viewing large s ale u tuations.
Developing the laws of thermodynami s we have adopted the view that no u tuation an
lead to any observable temperature dieren e of the two bodies in thermal equilibrium
- this enabled us to as ribe them the same temperature. But now we are saying that
entropy is determined by the onstraints so the state of two bodies in thermal onta t but
not yet at the same temperature are treated as a huge u tuation (whi h has no han e
to be observed if it were to o ur spontaneously) whi h is part of the new equilibrium
state. To be onsistent one would have to distinguish the temperature of the new equlib-
rium state (also a fun tion of the onstraints) from the imperfe tly dened instantaneous
temperatures of the individual bodies.

59
 This dis ussion, says Pippard, leads us into rather deep waters and it is useless to
ontinue it within the framework of lassi al thermodynami s - the proper framework for
it is the statisti al thermodynami s, or even the kineti theory. But the di ulties just
dis ussed should not be taken as disqualifying the view that entropy is determined by
the onstraints. The entropy law, although it seems to be violated in the useless way
in experiments of the sort mentioned before, is always valid in pra ti e and orre tly
determines what hanges are permitted by 2TMDL.
The entropy law has in fa t larger range of validity that ould be supposed from the
foregoing dis ussion whi h might be taken to suggest that u tuations are not a ounted
by it and that they an in some ir umstan es lead to its violation. In reality there are
strong indi ations that u tuations annot be used to violate the entropy law or 2TMDL.
It is Maxwell himself who hypothesized a demon ( alled the Maxwell demon ever sin e)
whi h ould ontrol a trapdoor onne ting two vessels of gas at the same temperatures.
The demon was supposed to allow to pass from, say, the left vessel to the right one only
those mole ules whi h have velo ity above the average and in the opposite dire tion only
the mole ules of velo ity lower than the average. In this way the demon was supposed to
be able to rise the temperature of the gas in the right vessel and to lower the temperature
of the gas in the left vessel de reasing thereby the total entropy. This is nothing else
but an attempt to systemati ally exploit u tuations (here u tuations of the energy of
the gas in the region near the trapdoor) to violate 2TMDL. But this way of presenting
the problem assumes that the entropy of the demon itself does not enter the problem
nor that it does generate any entropy by its a tion. Brillouin has analyzed this problem
and found this is unjustied. To distinguish the position and velo ity of a mole ule (to
de ide whether to let it pass the trapdoor or not) the demon must be provided with a
small ash-lamp (in a uniform T the radiation is uniform and does not allow to distinguish
mole ules) and the ash-lamp by the radiation it emits operates irreversibly and in reases
entropy. Brillouin has shown that the de rease of entropy whi h an be a hieved in this
way owing to the segregation of the mole ules is always over ompensated by the entropy
generated by the demon's operation. (This gedanken experiment should be treated at the
same footing as the famous Heisenberg gedanken experiment with the mi ros ope, whi h
showed that the un ertainty prin iple ∆p∆q ≥ ~ annot be ir umvented). It is therefore
not true, says Pippard, that 2TMDL is only statisti ally true, being repeatedly violated
mi ros opi ally (but never seriously, on a ma ros opi ally per eptible s ales. If entropy
is understood as proposed here, as a fun tion of onstraints (whi h as said, allows to
in orporate u tuations into the s heme of phenomenologi al thermodynami s), 2TMDL
is universally valid.

60
 LECTURE VI (TMD)

The view dis ussed in the pre eding Le ture that it is the onstraints, to whi h ther-
modynami systems are subje ted, whi h determine the system's entropy is the basis of
29
the Callenian formulation of thermodynami s whi h we will dis uss now. To some, giv-
ing a fourth (after the ones of Clausius, Kelvin and Caratheodory) formulation of 2TMDL
may seem superuous, but this law is the heart of thermodynami s and deserves to be
understood from dierent points of view. The Callenian approa h will also allow us to
introdu e into the play the dependen e of thermodynami al state fun tions on the amount
of matter involved, quantied by the number (or numbers in the ase of multi omponent
systems) of moles. Moreover, by redu ing the omplete thermodynami al information
about a given system to the knowledge of a single thermodynami al potential (as a fun -
tion of its natural variables) it puts the ne essary order into its hara terization. Last
but not least, the Callen's formulation of thermodynami s, being dire tly inspired by the
equilibrium statisti al physi s approa h to thermodynami al problems whi h provides the
methods of al ulating the mentioned thermodynami al potentials, onstitutes a dire t
link between the two parts of this Course.
The Callen's formulation of thermodynami s is based as all previous ones on the
postulate that there exist equilibrium states, on 0TMDL whi h allows to introdu e an
empiri al temperature, and on 1TMDL in whi h the work Z of hange of the amount of
matter in the system will (ultimately) be in luded: ∆U = Q + W + Z (or dU = q + w + z
if innitesimal hanges are onsidered). In reversible pro esses it will be possible to write
as in the onventional approa h w = d¯W and q =
Prd¯Q. If the matter transfer is done
reversibly it will be possible to write z = d
¯Z = i=1 µi dni , where µi is the hemi al
potential of the i-th omponent.
2TMDL in this approa h takes the form of the following postulates

• There exists entropy S whi h is dened on all equilibrium states of the system and is
a fun tion of the system's internal energy U and other globally dened parameters
like its volume V, magnetization M, and like (denoted below olle tively Xi ) and
of the amount of matter ontained in the system and represented by the number(s)
ni of moles of its onstituents. If the system is homogeneous and possesses the
extensiveness property, entropy is the homogeneous fun tion of order one of its
extensive arguments.

• S is a (dierentiable, at least twi e) monotoni fun tion of the internal energy

 
∂S
> 0,
∂U Xj ,ni

• Entropy of a system onsisting of several subsystems is additive

29 In fa t the approa h we all Callenian omes dire tly from Gibbs and was subsequently developed
by Tisza; it was written up by Callen in his textbook on thermodynami s.

61
 • Entropy of an isolated system takes on the maximal value with respe t to all pos-
sible equilibrium states whi h ould be realized with the help of stronger onstraints
than the ones the system is a tually subje ted to.

We will now explain these postulates.

In most ases we deal with systems whi h are homogeneous or onsist of several sub-
systems whi h are homogeneous and posses the property of extensiveness whi h means
that their global parameters Xi (those whi h hara terize the system as a whole, e.g. its
volume, internal energy, total polarization, in ontrast to those whi h although uniform
throughout the whole system when it is in equilibrium, ould in prin iple lo ally take dif-
ferent values) and the amount of matter ontained in them (quantied by the number(s)
of moles ni ) s ale proportionally to their internal energy U: if U → λU , then Xi → λXi
and ni → λni . This amounts to the assumption that if su h a system is mentally divided
into two or more parts, its internal energy U is the sum U = U1 + U2 of the internal
energies U1 and U2 of these two (or more) parts taken separately (separated by a wall)
whi h in turn means that the energy U12 of the intera tion of these two parts is negligi-
ble ompared to U1 and U2 and also that in the total internal energy surfa e ee ts are
negligibly small. This of ourse requires that the intermole ular for es be (ee tively)
30
short range. In many ases (e.g. uids) s aling of the dimensions of the system with
energy an be reasonably arbitrary, and an hange the system's shape. This is not so in
the ase e.g. of magneti (or diele tri ) systems pla ed in an external magneti (ele tri )
eld, be ause the pre ise form of the magneti (ele tri ) eld inside a magneti spe imen
depends on its shape and so does the total magnetization (polarization); however owing
to the s ale (in fa t even onformal) invarian e of (free) lassi al ele trodynami s, mag-
neti and diele tri systems an be treated as extensive with respe t to s alings of their
size whi h do not hange their shapes. Global parameters U , Xi , nj of su h systems are
alled extensive parameters. One should also remark that thermodynami s an be ap-
plied to systems whi h are not hara terized by the numbers of moles, yet are extensive.
The most prominent example of su h a system is the ele tromagneti eld in equilibrium
with the walls of a avity whi h is hara terized only by the internal energy U and the
volume V of the avity. Other systems of this kind (of whi h the ele tromagneti eld is
the simplest example) are relativisti quantum elds whi h, although in popular a ounts
identied with dierent sorts of elementary parti les (the misleading and infamous eld-
31
parti le duality whi h should be denitely banished forever from serious treatments of
the quantum eld theory!) are nevertheless systems of elds and the number of parti les
ontained in su h a system is from the point of view of thermodynami s and equilibrium
statisti al physi s an ill dened quantity. For this reason no hemi al potential an be

30 Although ele tromagneti for es are long-range, they always get s reened in ele tri ally neutral ma-
terial systems. The gravitational intera tions annot be s reened in this way and therefore systems in
whi h they play important role are not extensive (see the bla k hole example below).
31 If there is any meaning to be atta hed to this notion, it is rather the astonishing fa t that quantum
states of most systems of relativisti elds at all exhibit features normally asso iated with parti les;
however the simplisti identi ation of parti les with a on rete type of eld is fundamentally wrong.

62
 32
asso iated with it.
As far as systems possessing the extensiveness property are on erned, it is in many
situations onvenient to work not with extensive quantities like U , Xi and nj but with
molar quantities u, v et . and the molar fra tions xi dened as

r
X
u = U/n , v = V /n , xi = ni /n , n≡ ni .
i=1

Internal energy (and any of the state fun tions like the already introdu ed enthalpy H,
Helmholtz free energy F or Gibbs fun tion G) of an extensive system whi h all are ne es-
sarily extensive quantities, when written as fun tions of extensive and intensive parameters
like pressure p, an be written in the form

U(p, V, n1 , . . . , nr ) = n u(p, v, x1 , . . . , xr ) ,

et . This works also the other way around: if a state fun tion of an extensive system is
given for one mole of it, say the molar Helmholtz fun tion f = f (T, v, x1 , . . . , xr ), its full
dependen e on the number n of moles an always be restored by writing

F (T, V, n1 , . . . , nr ) = n f (T, V /n, n1 /n, . . . , nr /n) .

Internal energy of the ele tromagneti eld must in turn take the form U(T, V ) = V u(T ).
Entropy of a system whi h possesses the property of extensiveness is (and this is part of
the postulate) a homogeneous fun tion of rst degree of the extensive (global) parameters:

S(λU, λXj , λni ) = λ S(U, Xj , ni ) . (90)

33
whi h means that it too an be written as

S(U, V, . . . , ni ) = n S(U/n, V /n, . . . , ni /V ) = n s(u, v, . . . , xi ) . (91)

One should be aware, however, that not all systems to whi h thermodynami s applies are
extensive. One important example of a nonextensive system is the bla k hole to whi h
we will devote below a digression.
The maximum entropy postulate, entral to the whole Callenian approa h, an be best
elu idated on a simple system of a gas en losed in a ontainer. The total volume V of
the ontainer is xed, similarly as the total gas energy U and the number of its moles n.
By introdu ing dierent kinds of auxiliary walls: movable or nonmovable, adiathermal or
diathermal, permeable or impermeable to mole ules, it is possible to divide the ontainer
into an arbitrary number of ells of dierent (ma ros opi ) sizes Vi in whi h dierent

32 In the statisti al physi s approa h we will see that systems of elds an give rise to onserved harges
- like the ele tri harge in quantum ele trodynami s - and it is with these harges that appropriate
hemi al potentials must be asso iated.
33 The usage of various hara ters representing physi al quantities in these le tures appears now a bit
onfusing. Will have to work on this, when time allows...

63
numbers of moles ni and portions Ui of energy are blo ked in equilibrium states, the only
onstraints being

X X X
ni = n , Ui = U , Vi = V .
i i i

34
One an imagine innitely many ways of blo king in this way various equilibrium states
of the system (whi h be omes in this way omposed of several subsystems). In agreement
with the rst postulate of the Callenian approa h, ea h su h equilibrium state is as ribed
a ertain entropy S whi h depends on U, V and n and on the distributions Ui , Vi , ni
of these quantities between the subsystems into whi h the system has been divided by
the auxiliary walls. It is given by the sum of entropies of the subsystems into whi h the
system has been split. The equilibrium state of the gas in the absen e of auxiliary walls is
that state among all equilibrium states whi h an be fabri ated by introdu ing auxiliary
walls in dierent ways, that has the largest entropy. This Callenian prin iple of maximum
entropy pertaining to isolated systems will be later generalized to apply also to systems
in dierent spe i onta ts (thermal, me hani al, material) with their surroundings.
One should note the similarity of this prin iple to the known prin iple of least a tion
i i
in Me hani s - the traje tory q (t) whi h the system goes from its position q (t1 ) at t1
i
to its position q (t2 ) at t2 is that traje tory whi h gives the smallest value of the a tion
i
fun tional I[q(t)]. Operationally, one an onsider dierent traje tories q (t) with xed
ends, ompute the value of the a tion on ea h of them and ompare these values: the
true traje tory is the one giving the lowest I. Be ause of this similarity, the Callenian
formulation of 2TMDL is sometimes alled variational. A tually, in me hani s the true
traje tory may not be the one orresponding to the smallest value of I: it must only
35
be the stationary one. This is not so in thermodynami s: the true equilibrium state
of the system maximizes its entropy on the set of all virtual equilibrium states whi h
an be fabri ated by imposing on it arbitrary auxiliary onstraints stronger than the
ones the system is a tually subje ted to. One should also noti e that in ontrast to
me hani s, in whi h the a tion I has in ea h ase a form known from the beginning, in the
variational formulation of 2TMDL the formula for entropy is not given. Only its existen e
36
is postulated. Another remark is that it is now lear that with respe t to isolated
systems, the entropy S plays the role of the thermodynami al analog of the potential
(hen e, it is one of thermodynami al potentials) as its maximum (over all possible virtual
equilibrium states) determines the equilibrium state of any isolated system
It will be noted that the entropy law, to whi h one arrives only through a ombination
of arguments, if the Clausius or Kelvin's formulations of 2TMDL are adopted, in the

34 One should not onsider divisions of the system into subsystems so small, that surfa e ee ts ould
be ome important.
35 The reason for this, as well as the justi ation of the variational prin iple on whi h me hani s an
be based is quantum me hani al.
36 Within thermodynami s the form of S must be re onstru ted from the empiri al data. It is the
statisti al me hani s whi h gives the on rete, unambiguous pres ription how to obtain entropy of an
isolated system, if its mi ros opi dynami s ( lassi al or quantum) is known.

64
Callenian formulation is built in it from the outset: if the onstrains to whi h an isolated
system is subje ted are weakened, the entropy of the new equilibrium state an only be
greater than in the old equilibrium state (or at most equal to it) be ause a larger set of
37
virtual equilibrium states be omes possible. In an isolated system reversible an only
be pro esses ( hanges of onstraints) whi h happen not to in rease the entropy.

The expression

S = S(U, X1 , . . . , n1 , . . .) , (92)

(S = S(U, V, n) in the ase of a one- omponent uid) is alled the fundamental relation
in the entropy representation. As will be ome lear, if it is known as a fun tion of
the global (extensive) variables U , Xi , nj , our thermodynami al information about the
system is omplete (this is the meaning of the word fundamental). The derivatives of
the entropy (of a simple, possibly multi omponent uid for deniteness) are
     
∂S 1 ∂S p ∂S µi
≡ , ≡ , ≡− . (93)
∂U V,ni T ∂V U,ni T ∂ni U,V,n′i T
Although this is suggestive (and obvious for those who already know 2TMDL), in the
Callenian approa h one has yet to demonstrate that the quantities T, p and µi (and
possibly others, if the system is more ompli ated) dened in this way do indeed have
the meanings of the ordinary (absolute) temperature, pressure and hemi al potentials.
To show that the parameter T dened by the derivative of entropy with respe t to
the internal energy has indeed the meaning of the possible thermal equilibrium indi ator,
we onsider a system whose energy U and the number of moles are xed, say a uid in
a ylinder of xed volume V. Introdu ing an adiathermal nonmovable wall whi h divides
the ylinder into two parts of xed volumes V1 and V2 ontaining n1 and n2 moles (one
an assume that V1 /n1 = V2 /n2 so that the gas densities in the two parts are equal; of
ourse V1 + V2 = V and n1 + n2 = n) we an fabri ate dierent virtual equilibrium states
orresponding to dierent distributions U1 and U2 of the total energy U between the two
subsystems into whi h the original single system has been split. Alternatively, one an
onsider a system whi h is from the beginning omposed of two subsystems, two (possibly
dierent) gases in two ontainers, of volumes V1 and V2 and numbers n1 and n2 moles,
whi h are in thermal onta t through a diathermall wall; the virtual equilibrium states
blo ked by repla ing the diathermal wall by an adiathermal one orrespond to dierent
distributions of the total energy U between the two gases. A ording to the Callenian
entropy maximum prin iple, the equilibrium state in the absen e of the wall in the rst
ase or when the wall is diathermal in the se ond ase, is that one of the virtual equilibrium
states that has the biggest entropy. Now, entropies of the virtual equilibrium states in
whi h the system onsists of two subsystems are given (using the postulated additivity of
entropy) by the formula

S = S1 (U1eq + δU, V1 , n1 ) + S2 (U2eq − δU, V2 , n2 ) ,

37 It is lear that if the domain of the arguments over whi h one seeks the maximum (minimum) of a
fun tion (fun tional) is enlarged, the maximum (minimum) an only in rease (de rease).

65
in whi h the departures of the virtual state energies U1 and U2 from the (unknown)
eq eq eq eq
equilibrium distribution U1 , U2 (U1 + U2 = U ) have been parametrized with δU
(automati ally taking into a ount the ondition U1 + U2 = U ). A tually, if as here, the
onsidered system is extensive, the two fun tions S1 and S2 are simply the same fun tion
S(·, ·, ·). The extremum ondition δS = 0 whi h, in view of the simplied way of seeking
38
the maximum ould simply be redu ed to dS/d(δU) = 0, gives the equality

     
∂S1  ∂S2 
δS = δU + (−δU) = 0 .
 
∂U1 ∂U2 V2 ,n2 
 
V1 ,n1 eq
U2 =U2eq

U1 =U1

whi h implies that in equilibrium the parameters 1/Ti = (∂Si /∂Ui )Vi ,ni |Ui =Uieq must be
39
equal. Thus the derivative of S with respe t to U plays the role of the thermal equilib-
rium indi ator i.e. of an empiri al temperature. Dening it to be 1/T and not T leads
to the assignment of higher temperatures to hotter bodies. Indeed, onsider one of the
virtual equilibrium states with δU 6= 0 but innitesimal |δU| ≪ U as a real state of two
bodies and assume that, say,
     
∂S1  ∂S2 
< ,
 
∂U1 ∂U2
 
V1 ,n1 U1 =U1eq +δU V2 ,n2 U2 =U2eq −δU
 

that is, that T1 > T2 . Then when the adiathermal wall is hanged to a diathermal one,
the system will rea h the new equilibrium state and in rea hing it the two bodies will
ex hange energy (in the form of heat) in su h a way that the entropy will in rease:
 
δU δU 1 1
∆S = ∆S1 + ∆S2 = − + = − δU > 0 .
T1 T2 T1 T1
This means that δU must be positive, i.e. that the body whi h had higher temperature
looses energy and that of lower temperature gains it.
Similarly, to show that the parameter p dened by the se ond of the derivatives (93)
is the ordinary pressure, we onsider virtual equilibrium states of the system (again, for
on reteness let it be a gas in the ylinder) whi h an be realized with the help of the
rigid adiathermal wall but now without orrelating the mole numbers ni with the volumes

38 This reasoning in whi h one onsiders only a ertain rather narrow lass of all possible virtual equi-
librium states whi h ould be fabri ated should be onfronted with the pra ti al approa h to the deter-
mination of the traje tory q i (t) of a lassi al parti le: assuming we do not know the variational al ulus
(whi h redu es the problem to solving the Euler-Lagrange dierential equations) we invent a trial traje -
tory whi h onne ts the initial and nal system's positions in the time interval t2 − t1 and depends on
one (or a few) free parameter(s) λ (λi ). The a tion omputed on this trial traje tory depends therefore
on this (these) parameter(s) and an be minimized with respe t to it (them). Of ourse, nding the true
traje tory in this way is possible only if it is one of the trial traje tories (whi h form only a very narrow
lass of possible virtual traje tories).
39 The assumption - whi h is part of the Callenian postulates - that entropy is a monotoni fun tion
of the internal energy automati ally means that that at Ui = Uieq the rst derivative dS/d(δU ) hanges
sign, so this point is the extremum.

66
Vi (that is, allowing for dierent densities of the gas in the two parts) and allowing for
dierent distributions of the internal energies U1 and U2 and of the volumes V1 and V2 .
Again one an alternatively onsider two (possibly dierent) gases, n1 n2 of ea h, in
and
eq eq
two parts of a ylinder separated by a movable diathermal wall. The values Ui and Vi
whi h will be realized in the equilibrium state whi h the system assumes in the absen e of
the wall in the rst ase and if the wall is movable and diathermal in the se ond ase, are
the same as in that virtual state obtained in the presen e of the wall, whi h maximizes
the total entropy

S = S1 (U1eq + δU, V1eq + δV, n1 ) + S2 (U2eq − δU, V2eq − δV, n2 ) ,

that is su h that
     
∂S1  ∂S2 
δS = δU + (−δU)
 
∂U1 ∂U2 V2 ,n2 
 
V1 ,n1 U =U eq ,V =V eq eq eq

U =U ,V =V
   1 1 1 1    2 2 2 2
∂S1  ∂S2 
+ δV + (−δV ) = 0 . (94)
 
∂V1 U1 ,n1  ∂V2 U2 ,n2 
 
eq eq eq eq
U1 =U1 ,V1 =V1 U2 =U2 ,V2 =V2

Sin e the departures δU and δV an be varied independently (in fabri ating dierent
virtual equilibium states), this implies equality (in the equlibrium state realized in the
absen e of the wall) of the two temperatures 1/Ti = (∂Si /∂Ui )Vi ,ni |Ui =Uieq ,Vi =Vieq and of
the two pressures pi = Ti (∂Si /∂Vi ) |
Ui ,ni Ui =Uieq ,Vi =Vieq .
The reason, why the derivative of S
with respe t to V is denoted p/T and not p, an be seeked in the sho king relation
       
∂S ∂S ∂U 1 ∂U
=− =− .
∂V U,n ∂U V,n ∂V S,n T ∂V S,n
To orroborate the interpretation of the derivative of S with respe t to V as the ratio
of the system's pressure to temperature, one an also onsider n moles of a gas in an
adiathermally isolated ylinder tted with the piston (of ross se tion area A) on a spring
satisfying the Hooke's law (see Figure 5). The total energy E of the entire system (the gas,
the piston and the spring) is xed. Blo king the piston in dierent positions ( hara terized
by the variable x) one an fabri ate dierent virtual equilibrium states of the system. The
equilibrium state assumed by the system when the piston is not blo ked, is that one of
the virtual states whi h has the maximal entropy (entropy of the me hani al elements of
the system is assumed to be zero)

1
S = Sgas (E − kx2 , V0 + Ax, n) .
2
Equating to zero the derivative of S with respe t to x one gets the ondition (the symbol
xeq stands for U = E − 12 kx2eq and V = V0 + Axeq )
     
∂S  ∂S 
−kx +A = 0.
 
∂U V,n  ∂V U,n 
 
xeq xeq

67
 k
V0

x=0 x = l0 x

Figure 5: Left: A gas in the adiathermally isolated ylinder losed with the movable
piston on the spring. Right: Two parts of the ylinder (isolated as a whole) ontaining
two gases are separated by the adiathermal movable piston.

Sin e kx/A is the me hani al pressure pext exerted on the gas by the piston, one learns
that in equilibrium state Teq (∂S/∂V )U,n |xeq = pext .
It is appropriate to omment in this pla e (also be ause this sheds some light on
the pre eding example) on the so- alled problem of the adiabati piston by whi h one
means the problem of establishing the onditions whi h should determine the equilibrium
state of the system onsisting of two (possibly dierent) gases, n1 and n2 moles of ea h,
en losed in two adja ent parts of a ylinder (isolated as a whole) separated by a movable
40
but adiathermal piston (Figure 5, right). As in the examples onsidered above, one an
imagine that in equilibrium the piston stays in some xed position and the two gases
eq eq eq eq
o upy the volumes V1 and V2 (V1 + V2 = V ) having some well dened internal
eq eq eq eq
energies U1 and U2 (with U1 + U2 = U ). As in the pre eding example one onsiders
then virtual equilibrium states of the system whi h an be obtained by blo king the
(adiathermal) piston in dierent positions. One has also to admit dierent energies U1 and
U2 (of ourse, respe ting U1 +U2 = U ) of these virtual equilibrium states be ause, although
the adiathermal piston inhibits heat transfer between the two gases, they nevertheless
an do work on one another through the pressure they exert on one another through the
movable piston. Thus as previously one an write as previous the ondition (94)

peq peq
   
1 1 1
δS = eq − δU + − 2 δV . (95)
T1 T2eq T1eq T2eq
However in this ase here the variations δV and δU annot be treated as independent.
The point is that we now onsider only those virtual equilibrium states whi h an be
realized by blo king the piston in the real system in whi h any transfer of energy between
the two parts must be due to the me hani al work done by the piston in the adiathermal
onditions. In these onditions δU = −p2 δV (the work on the gas number 1 equals the
hange δV of its volume times the external pressure whi h is provided by the pressure of
the gas number 2). Using this relation onverts the ondition (95) into

peq
1 − p2
eq
δS = δV = 0 .
T2eq
This shows that the equilibrium position of the piston must be su h as to make the
pressures of both gases equal, but does not impose any ondition on the equilibrium

40 A strange feature of a similar system was demonstrated in lasses.

68
temperatures T1eq and T2eq . Therefore the equilibrium state of the system is not determined
by the maximum entropy prin iple.
If a real system of this sort is prepared in a state in whi h the pressures of the two
gases are equal it super ially seems that the temperatures T1 and T2 ould indeed be
arbitrary, but this negle ts the role played by u tuations. If the initial pressures of the
two gases are not equal, the piston would os illate eternally, if there were no fri tional
and vis ous for es. In real systems fri tional and vis ous for es are always present and
will ultimately damp the os illations of the piston. This rst pro ess, whi h lies entirely
outside the domain of thermodynami s, will lead to equalization of the pressures. At
the se ond stage, it will be the u tuations whi h will lead to the equalization of the
temperatures: if the temperatures of the gases are unequal, the u tuations in two parts of
the ylinder will be dierent and the piston itself will a t as a Brownian parti le re eiving
unequal ki ks from both sides; this will result in energy transfer leading eventually to
equal temperatures.

Finally we dis uss the hemi al equilibrium, that is, equilibrium with respe t to possi-
ble matter transfer. Let n moles of a gas ll two parts (of volumes V1 and V2 ) of a ylinder
separated by an unmovable wall permeable to the gas mole ules. Sin e mi ros opi ally
the heat transfer o urs through ollisions of parti les, su h a wall is ne essarily diather-
eq eq
mal. We now seek the equilibrium distribution of the numbers n1 and n2 of moles of the
eq eq
gas and of the energies U1 and U2 between the two parts of the ylinder applying the
maximum entropy prin iple. To this end we onsider virtual equilibrium states realized
by an adiathermal and nonpermeable wall (repla ing the original one) and departures δn
of the number of moles of the gas ontained in the volume V1 and δU of its energy from
the equilibrium values. Sin e the equilibrium state maximizes entropy, the rst order
variation of the total entropy of the system in equilibrium

     
∂S1  ∂S2 
δS = δU + (−δU)
 
∂U1 V1 ,n1  ∂U2 V2 ,n2 
 
eq eq eq eq
U =U ,n =n U =U ,n =n
   1 1 1 1    2 2 2 2
∂S1  ∂S2 
+ δV + (−δV ) , (96)
 
∂n1 U1 ,V1  ∂n2 U2 ,V2 
 
eq eq eq eq
U1 =U1 ,n1 =n1 U2 =U2 ,n2 =n2

must vanish. As the departures δU and δn are independent, this entails the equality of
the temperatures of the gases in the two subvolumes and the equality of their hemi al po-
tentials: µ1 = µ2 . Thus the hemi al potential of a given material omponent (mole ules
of a given sort) plays the role of the indi ator whether two systems will be in the hemi al
equilibrium with respe t to the ow of this material omponent, when they are separated
41
by a wall permeable to this omponent mu h in the same way as the temperature T
plays the role of the indi ator of the possible thermal equilibrium. It follows also that the

41 Of ourse if a solid, made, say, of iron is in onta t with the air, the hemi al potential µFe of the air
(treated as one subsystem) is zero, similarly as is zero the hemi al potential µAir of the solid. Yet no
transfer of matter o urs be ause the iron surfa e should be treated as a wall impermeable to both air
and iron.

69
matter transfer (of a parti ular matter omponent) an o ur reversibly only between the
systems having the same value of the hemi al potentials (of this omponent). Finally,
in the same way as in the ase of heat ow, it an be shown that in maximizing entropy
matter ows from the system having the higher value of the hemi al potential to the
system having the lower value.

Sin e by assumption entropy of a system is a monotoni fun tion of the internal energy
U, the fundamental relation in the entropy representation (92) an be inverted to give

U = U(S, X1 , . . . , n1 , . . .) . (97)

This is alled the fundamental relation in the internal energy representation. If known as
a fun tion of the system's extensive (global) parameters it too, similarly to the relation
(92), ontains the omplete information on the system's thermodynami properties.
Owing to the standard mathemati al relation between partial derivatives of a fun tion
and of its inverse, the oe ients of the total dierential (if the system is a multi omponent
uid) of U
r
X
dU = T dS − p dV + µi dni , (98)
i=1

are, as we now know, the temperature, pressure and the hemi al potentials. It then
follows that in reversible hanges of the system, in whi h −p dV an be treated as a dier-
ential form d¯W of the work (in the ase of (98) the me hani al volume work) performed
on the system and the terms µi dni an be interpreted as dierential forms of works re-
lated to the hanges of the amount of the i-th omponent in the system, the term T dS
represent the heat taken by the system. Therefore integrability of the heat form d¯Q is in
this approa h to 2TMDL obtained almost for free!

Let us now demonstrate the fundamental role of the fundamental relation (92) whi h
makes it really fundamental (the fundamental form of this senten e is a joke, of ourse).
Suppose the fundamental relation of a hypotheti al simple, one- omponent system is
God-given (or given by the statisti al me hani s) in the form

S(U, V, n) = 3a (U V n)1/3 . (99)

The system is learly extensive, for S(λU, λV, λn) = λ S(U, V, n). Computing expli itly
the derivatives (93) one nds the relations

1 (V n)1/3 p (U n)1/3 µ (U n)1/3

=a , =a , − =a .
T U 2/3 T V 2/3 T V 2/3
Upon elimination of U, the rst two of these relations yield the equation of state

p2 V = n a3 T 3 ,

70
and, moreover, the rst relation alone dire tly yields the internal energy as a fun tion of
temperature, volume and the number of moles:

 1/2
3/2 3/2 1/2 1/2 3/2 3/2 V
U(T, V, n) = a T n V = na T .
n
Entropy an be now easily obtained either as a fun tion of T, V and n, or as a fun tion
of T, p and n:
 1/2
3/2 1/2 1/2 1/2 3/2 1/2 V T2 3
S(T, V, n) = 3a T n V = 3n a T , S(T, p, n) = 3n a .
n p
From these two forms of the entropy the two prin ipal heat apa ities CV and Cp an
readily be obtained

   1/2
∂S 3 3/2 1/2 V
CV = T = na T ,
∂T V 2 n
  2
 1/2
∂S 3 T 3/2 1/2 V
Cp = T = 6na = 6na T .
∂T p p n

Of ourse, they satisfy the (hopefully) well-known relation (whi h follows from 2TMDL)

    "  #2   −1
∂p ∂V ∂V ∂V
Cp = CV + T = CV − T ,
∂T V ∂T p ∂T p ∂p T

be ause
"  #2   −1  2 6 4
p3 2

∂V ∂V naT 9 3T
−T = −T 9 − = n a .
∂T p ∂p T p2 2na3 T 3 2 p

Finally, the last of the derivatives (93) gives the hemi al potential whi h, upon eliminat-
ing U from it, an be written either as a fun tion of T and V (in fa t v = V /n) or T and
p:
 1/2
3/2 3/2 V a3 T 3
µ(T, v) = −a T =− = µ(T, p) .
n p
It is left for the students to he k that the same results an be derived from the funda-
mental relation (97) in the energy representation whi h in this ase reads

S3
U(S, V, n) = .
27a3 n V
It should be also lear that if entropy S were known as a fun tion of T, V and n (or,
analogously, internal energy U as a fun tion of these variables), one ould ompute the

71
heat apa ity CV , but there would be no way to nd the equation of state or the heat
apa ity Cp . This illustrates the statement that S = S(U, V, n) ontains the omplete
thermodynami information about the system while S = S(T, V, n) does not.
The system onsidered in the example above (not a realisti one) was learly extensive
(all quantities like U , CV , Cp when expressed in terms of v = V /n were proportional to
n). To stress that although the majority of realisti systems an be treated as extensive,
there are nevertheless important nonextensive systems, we on lude this le ture with a
brief dis ussion of the thermodynami s of bla k holes.
42
Digression. Bla k hole entropy (example of a nonextensive system).
A bla k hole solution of the Einstein's equations of General Relativity was obtained by
K. S hwarzs hild in 1916. Just before the World War II J.R. Oppenheimer and H. Snyder
showed that a ollapsing su iently massive star ends up as a bla k hole. At present many
astrophysi al obje ts have been identied by astronomers as bla k holes (and re ently
even gravitational waves emitted by merging two bla k holes have been registered by
the LIGO Collaboration). In addition to the S hwarzs hild solution whi h represents
the simplest possible su h obje t hara terized entirely by its total mass M, there also
solutions representing rotating and/or ele tri ally harged bla k holes (known as the Kerr,
Reisner-Nordström and Kerr-Newman solutions, respe tively) whi h in addition to the
mass are hara terized by their total harge Q and/or angular momentum L. It is known
that M, Q and L are the only possible quantities whi h an hara terize a bla k hole
(famous saying that bla k holes have no hair).
Initially a bla k hole was viewed as an entirely passive obje t whi h annot emit
anything - it an only absorb. If this were true, bla k holes would lie outside the range of
appli ability of thermodynami s be ause, as was noted by J.A. Wheeler, 2TMDL would
not apply to systems in luding bla k holes: if a portion of matter of nonzero entropy were
dropped into a bla k hole, the external observer ould not be sure that the total entropy of
the system had not de reased, be ause the only hara teristi s of the bla k hole available
to him, M, Q and L, annot tell how mu h entropy the bla k hole has swallowed. This
43
has led J.D. Bekenstein, a Wheeler's student, to formulate in 1972 the onje ture that
the bla k hole has in fa t a nonzero entropy whi h is proportional to the area A of its
horizon surfa e (a surfa e separating from the rest of the spa e the region from whi h light
or any other obje t annot es ape), whi h in turn is a fun tion of M, Q and L. This was
prompted by a theorem of S. Hawking whi h says that a ording to the laws of lassi al
physi s, the total area A of horizons of bla k holes annot de rease and it in reases in
dynami al pro esses like e.g. merging of two bla k holes. Another hint was the observation
made by D. Christodoulou (another student of Wheeler): investigating the e ien y of
the so alled Penrose pro ess by whi h rotational energy an be extra ted from a rotating
bla k hole he found that it is the greatest if the pro ess is realized reversibly and that in
this ase the area A of the bla k hole horizon does not in rease - this indi ated that some
42 Based on the arti le by J.D. Bekenstein Physi s Today, 33, 24 (1980); another standard introdu tory
referen e is: B.R. Parker and R. M Leod Am.J.Phys. 48 1066 (1980); this se ond arti le is however not
very useful - its authors evidently had problems with lear formulation of their thoughts.
43 Just a year ago, or so, I saw his obituary in the CERN Courier.

72
sort of thermodynami s might be at play...
The general formula proposed by Bekenstein (the oe ient 1/4 has been xed later
by Hawking) whi h gives the bla k hole entropy reads:

1 A
S= kB 2 . (100)
4 ℓPl
p
kB is the Boltzmann onstant and ℓPl ≡ ~G/c3 is the Plan k length (G = 6.67 × 10−11
3 −1 −2 −35
m kg se is the gravitational onstant; ℓPl = 1.6 × 10 m). It is to be noted that this
formula relates a purely thermodynami al quantity S to a purely gravitational one (the
horizon area) and involves the Plan k onstant signaling existen e of a deep onne tion
between thermodynami s, gravitation and quantum phenomena. The formula for A of
the S hwarzs hild bla k hole (Q = 0 , L = 0) is very simple:

G2 2 ~G M 2 2 M
2
ASch = 4πRg2 = 16π M = 16π 2
= 16πℓ pl 2
. (101)
c4 c3 MPl MPl
p
here MPl = ~c/G = 2.2×10−8 kg= 1.22×1019 GeV/c2 is the Plan k mass. Rg = 2GM/c2
is just the radius whi h an be obtained using the Newtonian dynami s: it is the radius of
a planet of mass M for whi h the se ond osmi speed vII (determined by the ondition
1 2
2
mvII − GMm/r = 0) needed to rea h innity starting from its surfa e is equal c - sin e
this is the highest possible speed, Rg is just the radius of the horizon. Thermodynami s
of the S hwarzs hild bla k hole is very simple:

M2
S = 4πkB 2
,
MPl
U = Mc2 . (102)

This is in fa t the simplest of all thermodynami al systems sin e its equilibrium state is
hara terized by only one parameter M: no extra tion of energy from it by a reversible
work is possible. The equation

U2
S(M) = 4πkB 2 4
, (103)
MPl c
plays the role of the fundamental relation (92) of the S hwars hild bla k hole thermody-
nami s. It is to be noted that in this ase S is not a homogeneous rst order fun tion
of its only global parameter U. The simple s aling is violated by the presen e of the
fundamental mass (or length s ale) MPl (or ℓPl ). Applying to (103) the standard formula
dS = dU/T one gets

1 dS 8πkB
= = 2 4U, (104)
T dU MPl c
or
2 4
MPl c
U= , (105)
8πkB T

73
The larger the mass (the internal energy U ) of the bla k hole, the lower is its temperature!
This of ourse results in the negative its heat apa ity

2
MPl c2

dS(U(T )) dU kB
C=T = =− . (106)
dT dT 8π kB T

If a bla k hole has nonzero angular momentum L and/or ele tri harge Q, extra tion
of energy from it by a reversible me hani al or ele tri al pro ess is possible (we have
already mentioned the me hani al Penrose pro ess) by a oupling to L or to Q appropriate
external agents. Su h reversible works an be ontinued until L = 0, Q = 0. The formula
for the area of the Kerr-Newman bla k hole horizon surfa e is best expressed in terms of
the parameters (whi h both have dimension of mass squared; we use the normal Gauss
2 2
system of units so that Q /~c is dimenssionless - e /~c = αEM as every high energy
physi ist knows)

Q2 Q2 ~c Q2 2
Q̃2 ≡ = = M ,
G ~c G ~c Pl
2 2 2
L2 ~c L2 MPl
4
 
c L 1
a2 = 2 2 = = ,
G M ~2 G M2 ~2 M 2
and reads
" 2 #
G2
q
A = 4π 4 M + M 2 − Q̃2 − a2 + a2 . (107)
c

This formula together with (100) and the identi ation U = Mc2 , gives the fundamental
Callenian relation of the most general bla k hole thermodynami s:

S = S(U, L, Q) . (108)

It is again lear that the bla k hole is not an extensive system. All thermodynami al
information an be obtained from it in the standard way. In parti ular

1 φ 1
dS = dU − dQ − Ω·dL , (109)
T T T
whi h leads to the form

dU = T dS + φ dQ + Ω·dL , (110)

of 1TMDL as applied to reversible hanges of the bla k hole. φ is here the ele trostati
potential and Ω the bla k hole's angular velo ity. The se ond and the third term on the
right hand side represent innitesimal reversible works that an be done on or by the bla k
hole. The rest is a matter of ordinary thermodynami al ku-miku, whi h although an be
somewhat ompli ated be ause of the ompli ated form of the relation (107), should not
be more ompli ated from the point of view of prin iples, than any other thermodynami al

74
problem. A urious student may try to ompute for instan e the heat apa ity Cφ of the
Reisner-Nordström (L = 0) bla k hole or CΩ of the Kerr (Q = 0) one.
The formula for the S hwarzs hild bla k hole temperature an be written in terms of
the so- alled surfa e gravity κ

~ Mpl2 c3 c4
 
~ ~
T = = ≡ κ, (111)
2πkB c 4~M 2πkB c 4GM 2πkB c

whi h is the  g  (that isg = GM/RZ2 known from ordinary me hani s at the Earth
surfa e) of a planet of mass M and radius Rg .
If the bla k hole parameter T is to be regarded as the true temperature (and not
merely as an analog of it), the bla k hole should, as every body at a nonzero temperature
44
T radiate. Classi ally a bla k hole annot radiate, be ause nothing, in luding light, an
es ape from it. But the presen e of the Plan k onstant ~ in the Bekenstein formula (100)
strongly suggests, as has already been stressed, that the onne tion between gravity and
thermodynami s is not based on lassi al physi s. It was Hawking who by onsidering
45
quantum pro esses o urring near the horizon of a forming S hwarzs hild bla k hole
dis overed (in 1974) that it radiates just as does a body at temperature T given by
(111). In this pro ess the bla k hole horizon de reases (whi h lassi ally is impossible),
the bla k hole evaporates and its entropy de reases. This de rease of entropy is however
over ompensated (if the evaporation o urs in the surrounding of lower temperature) by
the entropy of the emitted thermal radiation (whi h onsist of all kinds of elementary
parti les). In this way the entropy law and thermodynami s as su h are appli able also
to bla k holes.

44 This is a point usually left aside in dis ussing equilibrium states of thermodynami al systems: to
be truly isolated adiathermally, a system must be pla ed in a shield whi h not only isolates it from all
inuen es from without in luding any in oming radiation, but must also ideally ree t ba k the thermal
radiation emitted by the system itself, when its temperature is not zero.
45 It is these onsiderations whi h allowed Hawking to x the oe ient in the Bekenstein formula (100).

75
 LECTURE VII (TMD)

We now pro eed to exploiting the entropy law and other properties of this quantity. We
will rst onsider the problem of maximal useful work whi h an be extra ted from a given
system in given onditions. These onsiderations will lead us to the formulation of the
prin iple whi h determines the equilibrium states of systems open to their surroundings,
generalizing thereby the Callenian maximum entropy prin iple whi h applies to isolated
systems only. Also the general onditions of stability of thermodynami systems will be
onsidered.

It is easy to show that entropy of an extensive system is a on ave ( on ave upwards)

fun tion of any of its extensive arguments. Indeed, onsider two systems A and B both
isolated and in ea h equilibrium within itself. Let us write their entropies as

SA = SA (λA UA , ξA VA , . . .) , SB = SB (λB UB , ξB VB , . . .) , (112)

that is, we onsider dierent extensive parameters of the two system s aled up by the
respe tive arbitrary fa tors λ, ξ , et . We now imagine that the two systems have been
brought into a thermal onta t and a new equilibrium state has been rea hed. By the
entropy law, the entropy of this new equilibrium state, SA+B (λA UA + λB UB , ξA VA +
ξB VB , . . .) annot be smaller than the sum of the entropies SA and SB :

SA+B (λA UA + λB UB , ξA VA + ξB VB , . . .) ≥ SA (λA UA , ξA VA , . . .) + SB (λB UB , ξB VB , . . .) ,

irrespe tively of the values (positive) of the fa tors λ, ξ , et . and irrespe tively of the
values of their energies, volumes, et . If the two systems are of the same kind (e.g.
two pie es of the same solid), the fun tions SA and SB are given by the same fun tion
S(·, ·, . . .) but evaluated at dierent values of its arguments. If in addition one sets λA = λ,
λB = 1 − λ, ξA = λ, ξB = 1 − λ, et . and onsiders VA = VB = V , et ., one obtains the
inequality

S(λUA + (1 − λ)UB , V, . . .) ≥ λ S(UA , V, . . .) + (1 − λ) S(UB , V, . . .) ,

whi h just means on avity upwards of the entropy S(·, ·, . . .) of the system as a fun tion of
its rst (energy) argument (Figure 6). In the same way one an demonstrate its on avity
as a fun tion of its volume argument and all other extensive parameters.
Con avity upwards of the entropy as a fun tion of the internal energy is equivalent to
the statement that

∂2S
 
< 0. (113)
∂U 2 V,...

Be ause of the rst of the equalities (93) this amounts to

   −1
∂ 1 1 ∂U 1
=− 2 ≡− < 0, (114)
∂U T V,... T ∂T V,... T 2C V

76
 S

UA UB U

Figure 6: Entropy of extensive systems is a on ave upwards fun tion of internal energy.
The dashed line shows the fun tion λS(UA ) + (1 − λ)S(UB ). Everywhere between UA and
UB the value of S(λUA + (1 − λ)UB ) is above this line.

whi h, in view of the positivity of the temperature (whi h is a dire t onsequen e of the
third Callenian postulate), in turn means that CV > 0. We have seen that the bla k hole
analog of CV is negative be ause the bla k hole is not an extensive system.

S and U as thermodynami al potentials

We will now show that the postulate that the equilibrium state of an isolated system
maximizes entropy on the set of all virtual equilibrium states (that an be fabri ated
with the help of onstraints stronger than the ones the system is a tually subje ted to)
at the xed value of the internal energy U (and xed values of other global parameters) -
the maximum entropy prin iple - is equivalent to the statement that the equilibrium
state minimizes internal energy on the set of all virtual states at the xed value of the
system's entropy S - this will be alled the minimum energy prin iple. This se ond
statement sounds a little bit more abstra t - it is easier to imagine keeping internal energy
xed (and distributing it among subsystems) than to imagine xed entropy and varying
internl energy. Nevertheless, the two formulations are equivalent. We will as usually give
two proofs: one physi al and general (but requiring some imagination) and another one
mathemati al, more on rete, but ne essarily restri ted to a spe i ase.
To prove that the maximum entropy prin iple implies the minimum energy prin iple,
we make the redu tio ad absurdum and assume that the equilibrium state in whi h the
entropy of the system is S0 and its energy is U0 does not minimize energy; there should
therefore exist another equilibrium state (whi h an be blo ked with the help of stronger
onstraints) in whi h the syste's entropy is still equal S0 but of energy U lower than U0 .
Therefore it would be possible to take the system adiathermally and reversibly (adiabat-
i ally) to this state of lower energy (be ause the two states have the same entropy S0 ,
logi ally this would be possible!) extra ting from it the work W̄ = U0 − U . The obtained
in this way energy ould be then put ba k irreversibly into the system bringing it into an
equilibrium state of energy U0 but of entropy higher than S0 . In ee t, the equilibrium
state would not maximize entropy at onstant energy be ause there would be a state
(possibly with stronger onstraints) of the same energy and higher entropy.
The proof in the reverse dire tion (that the minimum energy prin iple implies the
maximum entropy prin iple) goes similarly: we assume that the equilibrium state ( orre-
sponding to given onstraints) of entropy S0 and energy U0 does not maximize entropy

77
(on the set of virtual states with stronger onstraints but the same energy U0 ), so there
should exist a state of higher entropy S > S0 and the same energy U0 (realized when the
the system is subje ted to stronger onstraints); the system in this higher entropy state
ould be then brought into a onta t with a reservoir (a heat bath) transferring to it some
heat; entropy of the system would be thereby lowered to S0 (things ould be arranged so)
and, by 1TMDL, also the energy of the system would be lowered. One would therefore
end up with the system in equilibrium with the entropy S0 and energy lower than U0 (and
stronger onstraints). The assumed equilibrium state of entropy S0 and energy U0 would
not be, therefore, the state of lowest possible energy (for xed S0 ).
The mathemati al reasoning requires spe ifying a parameter, say x (or a ouple of pa-
rameters) whi h represents (represent) hanges of the onstraints imposed on the system;
one an assume that x = 0 orresponds to the a tual onstraints to whi h the system is
subje ted and x 6= 0 orresponds to stronger onstraints. The entropy maximum prin iple
means then that
 
∂2S
   
∂S  
= 0, < 0. (115)
 
∂x ∂x2
 
U,... U,...
 
x=0 x=0

Using the sho king relation one an write

       
∂U  ∂U ∂S 
=− = 0. (116)
 
∂x ∂S ∂x
 
S,... x,... U,...
 
x=0 x=0

This shows that (ex ept at T = 0) (∂U/∂x)S vanishes in the same

the derivative state in
whi h (∂S/∂x)U vanishes. The se ond derivative of U onsists of two terms
 2     2    "   #
∂ U ∂S ∂ U ∂U ∂ ∂S
= − − .
∂x2 S,... ∂x U,... ∂S ∂x ∂S x,... ∂x ∂x U,...
S,...

The rst term vanishes at x = 0; in the se ond term the derivative D ≡ (∂S/∂x)U,... whi h
is a fun tion of the variablesx and U should be now treated as D(U(x, S), x) so that
       
∂D ∂D ∂D ∂U
= + .
∂x S,... ∂x U,... ∂U x,... ∂x S,...

Again, the se ond term vanishes at x=0 (be ause (∂U/∂x)S,... = 0 there) and one ends
up with

∂2U ∂2S
   
= −T ,
∂x2 S,... ∂x2 U,...

whi h shows that if S(U, x, . . .) has a maximum at x = 0, then U(S, x, . . .) has a mini-
mum. The proof an be easily extended to more parameters representing onstraints, but
its weakness is that one onsiders the same onstraints for fabri ating virtual states at
onstant energy and at onstant entropy.

78
 The on lusion is that with respe t to isolated systems entropy and internal energy
play the role of thermodynami al potentials whi h (in the sense dis ussed in this and in
the pre eding Le tures) determine the system's equilibrium states.

Maximal and Minimal work

One of the main appli ations of the entropy law is putting an upper limit on me hani al
work whi h an be extra ted from a given thermodynami al system by hanging appro-
priately the onstrains to whi h the system is subje ted, thereby allowing it to rea h
another equilibrium state. The reverse problem is putting a lower limit on the work ne -
essary to bring a given system from one equilibrium state to another one (with stronger
onstraints).
The simplest situations in whi h the problem of the maximal (minimal) work an be
posed and analyzed is when a given system onsisting of several subsystems is as a whole
adiathermally isolated. With the environment (the rest of the world) it an only ex hange
me hani al work. The internal onstraints of the system (whi h keep its dierent parts
in equilibrium within themselves) an be weakened and on the way to the equilibrium
state orresponding to the new onstraints the system an deliver some work; sin e the
46
system is not ompletely isolated (it an ex hange work with the environment) the nal
equilibrium state and the work extra ted from the system depend on the pro ess by whi h
the new equilibrium is rea hed. The questions is, whi h pro ess will result in the maximal
extra ted work. In analyzing this we will assume that the nal and initial total volume
47
of all subsystems omprising the system is the same. Of ourse, it an vary during the
pro ess. In addition, some subsystems of the onsidered system may be subsidiary in the
sense that their initial and nal states are the same.
Sin e the system is supposed not to ex hange heat with the environment, by 1TMDL,
the work W̄ extra ted from it is just

W̄ = Uin − Ufin . (117)

The energy Uin is xed but the nal state and, therefore, Ufin depends on the pro ess.
It is, however, a fun tion of the nal state entropy. Sin e from the Callenian postulates
U is a growing fun tion of entropy (re all the assumption (∂S/∂U)X > 0 whi h implies
(∂U/∂S)X > 0), the biggest maximal work is obtained if entropy of the system does not
in rease, that is if the pro ess by whi h the system rea hes the nal equilibrium state
orresponding to the new weakened onstraints is reversible.
The standard illustration of this sort of situations is the system onsisting of two bodies
of unequal temperatures, say T1 and T2 > T1 , separated by an adiathermal wall. The nal
onstraint is a diathermal wall through whi h the two bodies are in thermal onta t and

46 Re all that if the system were ompletely isolated, its nal state would be uniquely determined by
the new onstrains.
47 In this way we ex lude the trivial work the system ould do hanging its volume - we are interested
only in the work whi h an be obtained by hanging the internal onstraints of the system: allowing it to
expand would mean hanging its external onstraints. In fa t the ondition of equal initial and nal total
volumes an be weakened by assuming that the pressure p0 of the environment is negligible (or zero) so
that expanding the system does not do any work.

79
have, hen e, the same nal temperature. If the two bodies are brought into the thermal
onta t dire tly (one possible pro ess of attaining the nal equilibrium state), the entropy
of the whole system will in rease (we talked about this in the pre eding Le ture) and no
48
work at all will be extra ted from the system. Assuming for simpli ity that the heat
49
apa ities C at onstant volume of the two bodies are equal and (to make things simple)
independent of the temperature, the nal ommon temperature Tfin will be determined
by 1TMDL:

Uin = (U0 + CT1 ) + (U0 + CT2 ) = (U0 + CTfin ) + (U0 + CTfin ) = Ufin , (118)

that is, Tfin = 12 (T1 + T2 ). The entropy hange in su h a pro ess will be

Tfin T1 T2 T2 (T1 + T2 )2
∆S = 2C ln − C ln − C ln = C ln fin = C ln > 0. (119)
T0 T0 T0 T1 T2 4T1 T2
If however the pro ess by whi h the two bodies attain thermal equilibrium with one
2
another is reversible, entropy will not hange and it is the formula ∆S = C ln(Tfin /T1 T2 ) =
√
0 whi h will determine the nal temperature Tfin = T1 T2 . By 1TMDL the work extra ted
from the system in this ase will be
 p 
W̄ = Uin − Ufin = 2U0 + C(T1 + T2 ) − 2 U0 + C T1 T2
 p  p p 2
= C T1 + T2 − 2 T1 T2 = C T2 − T1 . (120)

One an of ourse ask how to realize pra ti ally su h a reversible pro ess? In this
ase (but not ne essarily always) the answer is simple: it su es to run a Carnot engine
treating the body at T2 as the upper (higher temperature) heat reservoir and the body
at T1 as the lower one. The heat apa ity of the working substan e, whi h is a subsidiary
element of the system, should be small ompared to C to ensure that the bodies stay
pra ti ally in equlibrium when the heat is transferred between them and the engine. In
the al ulation one has of ourse to take into a ount that the ex hange of heat between
the bodies and the engine lowers the temperature of the upper reservoir and raises the one
of the lower reservoir, so that nally they will equalize. If the a tual (during the working
of the engine) temperature of the higher reservoir is T+ (Tfin ≤ T+ ≤ T2 ) and that of the
lower one is T− (T1 ≤ T− ≤ Tfin ) η(T− , T+ )
then (upon using the a tual e ien y of the
Carnot y le working between the temperatures T− and T+ )
 
T−
d¯W̄ = η(T− , T+ )d¯Q2 = 1 − (−C dT+ ) ,
T+
and the ne essary orrelation of T− T+ is provided by the
with ondition
 
d¯Q2 d¯Q1 dT+ dT−
+ = −C + = 0,
T+ T− T+ T−
48 The most general ase is treated in one of the homework Problems.
49 As said, one ould also onsider the heat apa ities at onstant internal pressure if the external
pressure is zero.

80
 RVS
p0

RHS
system
W̄
RWS
T0

RMS
µ0

Figure 7: The system and the set of reservoirs representing its environment.

of onstan y of the entropy (d

¯Q2 = −CdT+ > 0 and d¯Q1 = −CdT− < 0 are the heats
taken by the engine from the two bodies) whi h upon integration with the obvious initial
onditions yields T− = T1 T2 /T+ . Integrating then d¯W from T2 to Tfin gives the same
result (120).
The onverse situation to the one exemplied above is when the system is supposed
to attain, as a result of the hange, another equilibrium state of higher energy. The
analogous reasoning then shows that the minimal work W whi h must be done on the
system orresponds to the isentropi hange.

A more general situations is when a system, whi h an be omposed of several sub-

systems, an ex hange with the environment heat and perhaps also matter; one an also
admit that the total volume of (all parts of ) the system an be dierent in the nal state
than in the initial state. To analyze su h situations in general terms one an model the
system's environment as onsisting of a reversible heat sour e (RHS) - a reservoir so large
that it always remains in equilibrium at temperature T0 , no matter how big nite amount
of heat was extra ted from or added to it, the reversible volume sour e (RVS) - a very
large system in equilibrium (whose entropy stays always onstant and need not be taken
50
into a ount) at invariable pressure p0 and, nally, a sour e of matter (RMS) in equi-
librium (whose entropy also stays always onstant and need not be taken into a ount -
re all that the hemi al potential is dened by (93) as the hange of energy of the body,
here the RMS, resulting from hanging the amount of matter in it at onstant entropy)
at invariable hemi al potential µ0 (this general setting is s hemati ally illustrated in Fig-
ure 7). Owing to the onstraints (internal and external, separating the system from its

50 If there is more than one matter omponent whi h an be ex hanged with the environment, we
introdu e one su h sour e per omponent.

81
surrounding), the system onsisting of several subsystems is initially in equilibrium (the
individual subsystems an have dierent temperatures, pressures, hemi al potentials). If
the onstraints are hanged (are weakened or strengthened), the nal equilibrium state
ompatible with the new onstraints an depend on the pro ess by whi h the system at-
51
tains it. In the pro ess of rea hing the new equilibrium state the system may ex hange
heat with the RHS. We also assume that any hange in the total volume of the system (the
sum of volumes of its subsystems) is ompenstated by the opposite hange of the volume
of the RVS whi h is due to a me hani al onta t between the parts of the system and the
RVS. Similarly, any hange in the total matter ontent of the system is ompensated by
the opposite hange of matter ontent of the RMS due to a dire t ow of matter between
parts of the system and the RMS.
By 1TMDL, the work extra ted from the system and its environment represented by
RHS, RVS and RMS is

W̄ = −(∆U + ∆URHS + ∆URVS + ∆URMS ) , (121)

where −∆U = U in − U fin et . But sin e the reservoirs stay in equilibrium, ∆URHS =
T0 ∆SRHS , ∆URVS = −p0 ∆VRHS and ∆URMS = µ0 ∆nRHS . Furthermore, by assumption
∆VRHS = −∆V and ∆nRMS = −∆n (the total volume and the total amount of matter in
52
the system and its environment remain onstant )

W̄ = − (∆U + T0 ∆SRHS + p0 ∆V − µ0 ∆n) .

As far as the ex hange of heat (whi h need not be reversible) between the system and the
RHS is on erned, it follows from the entropy law that ∆SRHS + ∆S ≥ 0, or that

−T0 ∆SRHS ≤ T0 ∆S .

Therefore the work W̄ whi h an be extra ted from the system a omplishing a hange
(spe ied by ∆U , ∆V and ∆n) in the given environment ( hara terized by T0 , p0 and µ0 )
is limited form above:

W̄ ≤ −∆ (U − T0 S + p0 V − µ0 n) = −∆A . (122)

The quantity A ≡ U − T0 S + p0 V − µ0 n is alled availability of the system. Contrary

to this name, it is the property of the system and its environment. Re all also that in

51 In the setting onsidered here also the same nal state an be rea hed through dierent pro esses.
52 If there were more matter onstituents in the system (and orrespondingly more matter sour es
representing the environment) and hemi al rea tions were allowed to o ur in the system (as a result of
weakening its internal onstraint) and to hange its matter omposition, one would have to distinguish
the hanges of the number of moles of ea h of the onstituents o uring due to hemi al rea tions and
o uring due to the ex hanges with the matter sour es in the environment (the hanges of numbers of
moles due to rea tions would then be part of spe i ation of the pro ess by whi h nal equilibrium
state is rea hed); no problem in a ounting for hemi al rea tions arises, however, if the system does not
ex hange matter with the environment and the term ∆URMS in (121) is absent (formally one an set µ0
in all further formulae).

82
 Stot

∆Stot

Wmin

Utot

Figure 8: Interpretation of the minimal work in terms of the entropy. The solid line
represents the total entropy of the system and its environments as a fun tion of the total
energy with the initial (weaker) onstraints while the long-dashed line represents the total
entropy as a fun tion of the total energy with the nal (stronger) onstraints.

general the quantities U, V and S pertaining in (122) to the system are the sums of
energies, volumes and entropies of its dierent parts (subsystems).
If the hange of the system is innitesimal, repla ing ∆U by dU = T dS − p dV + µ dn
53
one gets

w̄ ≤ −(T − T0 )dS + (p − p0 )dV − (µ − µ0 )dn . (123)

If the hange is reversible, −T dS is the heat d¯Q lost by the system; its onversion into work
in the environment at temperature T0 (e.g. by running a suitable Carnot engine) yields the
useful work equal η(T0 , T )d ¯Q = −(T − T0 )dS . Similarly, if the system expands reversibly
hanging its volume by dV , the additional pressure p − p0 must be applied externally
(that is, p − p0 is the la king pressure ne essary to make the expansion reversible in order
to fully prot from it - re all the dis ussion in Le ture II) by external for es (provided
by the RWS) and the work done by the system on RWS is just (p − p0 )dV . Finally, to
make the hange of matter ontent of the system by −∆n reversible requires additional
54
hemi al potential µ − µ0 to be provided by the RWS and the work done by the system
on RWS due to the matter ow is −(µ − µ0 )dn.
Analogously, the minimal work W needed to a omplish a given hange (strengthen-
ing its onstraints and hara terized by ∆U , ∆V and ∆n) of the system in the given

53 We restri t here ourselves to a system onsisting of a single homogeneous body or to a situation in

whi h all parts of the system have the same temperature T and pressure p; in the general ase of a system
onsisting of several homogeneous bodies ea h of whi h ould have a dierent temperature, pressure and
hemi al potential the formula (123) would take the form

X
w̄ ≤ − [(Ta − T0 )dSa + (pa − p0 )dVa − (µa − µ0 )dna ] .
a

54 Here we enlarge a bit the meaning of the work sour e, allowing it to hange its energy also in the
form of matter ow. If one keeps the original denition of RWS, a reversible hange of matter ontent
of the system requires its hemi al potential µ to be equal to µ0 and the last term in (123) vanishes; if
the matter ow between the system and RMS is not reversible, T dS does not represent the heat transfer
to/from the system and (p − p0 )dV does not represent the work done by it (re all the dis ussion at the
end of Le ture IV) and (123) is valid only as the inequality.

83
surrounding is bounded from below

W ≥ ∆ (U − T0 S + p0 V − µ0 n) = ∆A . (124)

The minimal work Wmin needed to a omplish the hange is given an interpretation in
Figure 8 whi h sket hes the dependen e of the total entropy of the system and of its
environment on the total energy with weaker (solid line) and stronger onstraints (dashed
line); the horizontal dotted line shows the hange of the total energy needed to strengthen
the onstraints at xed entropy (it is thus equal Wmin); the verti al dotted line shows
the in rease ∆Stot > 0 of the total entropy if the stronger onstraints are repla ed by
the weaker ones and the system and its environment attain equilibrium without any
intervention from without (i.e. from RWS). If the system is very small ompared to
the environment (treated here as real and not as an innitely large) then Wmin ≪ Utot
and the in rease of the total entropy (in the spontaneous attaining the equilibrium after
weakening the onstraints) an be written as (here Utot and Stot stand for the total energy
and entropy of the system and of the surrounding)

 
∂Stot 1
∆Stot ≈ Wmin = (∆U − T0 ∆S + p0 ∆V − µ0 ∆n) . (125)
∂Utot Vtot ,ntot T0

This formula quanties by how mu h the total entropy of the system and of its environ-
ment (treated together as an isolated supersystem) with given onstraints diers from
the maximal entropy obtained when the system is in equilibrium with its environment
(without walls blo king the dire t onta t between them).

The formula (122) is very general and applies to most of the situations in whi h the
system exhanges heat with the environment. A few spe ial ases deserve to be dis ussed:

• The formula (122) has been derived assuming that the system does not ex hange
heat with the environment (RHS). If the system as a whole is thermally isolated
(whi h of ourse does not mean that in the pro ess of rea hing the nal state heat
ex hange annot o ur between its dierent subsystems) the inequality (122) be-
omes equality: the work done on the RWS in the pro ess in whi h the system
rea hes a given nal state equals simply the total hange of energy of the system, of
the volume sour e (RVS) and of the matter sour e (RMS). If however, as in the pre-
viously dis ussed situation, one is interested in the system hange spe ied only by
the nal onstraints (and not by the nal state), the work W̄ = −∆(U + p0 V − µ0 n)
done on RWS is maximal when the entropy of the system does not in rease; this
generalizes the previous onsiderations by allowing for a hange of the system's to-
tal volume and matter ontent due to its ex hange with the environment (RMS).
If the matter ontent of the system hanges as a result of hemi al rea tions (and
there is no ex hange of matter with the environment) the work done on RWS is just
W̄ = −∆(U + p0 V ) (again it is maximal for that nal state whi h orrespond to
the least entropy in rease - hemi al rea tions are nonequilibrium pro esses and as
su h always in rease entropy). Finally, if the initial and nal pressures of all parts

84
 of the system are the same and equal to the pressure p0 of the environment (e.g.
the hemi al rea tions in the system o ur uder the pressure of the environment),
then the work W̄ is given by the hange of enthalpy of the system

W̄ = −∆(U + p V ) = −∆H . (126)

• If the system ex hanges with the environment only heat ( hemi al rea tions o ur-
ring within the system are therefore allowed) and the nal total volume of all its
55
parts is identi al with the initial one, then W̄ ≤ −∆(U − T0 S). If in addition the
initial and nal temperatures of all parts of the system are the same and equal to
the temperature of RHS (of the heat reservoir with whi h the system may ex hange
heat in the pro ess), then, identifying T0 with T,

W̄max = −∆(U − T S) = −∆F . (127)

In this ase the maximal work is given by the hange of the Helmholtz free energy
of the system.

• If the system ex hanges heat and its volume hanges but there is no matter ex-
hange with RMS ( hemi al rea tions within the system are therefore allowed),
W̄ ≤ −∆(U − T0 S + p0 V ). If in addition the initial and nal states of all parts of
the system are in thermal and me hani al equilibrium with the environment (T = T0
and p = p0 ) then

W̄max = −∆(U − T S + pV ) = −∆G . (128)

To illustrate these results let us onsider one simple example. Suppose n moles of a
perfe t gas at temperature T en losed in the volume V are given. What maximal work
an be obtained by ooling it down to the temperature T0 < T of the environment not
hanging the volume V? As the nal volume of the gas is to be the same as the initial
one (during the pro ess of ooling down it may vary), the answer to the question is given
by

W̄max = −∆(U − T0 S) .

Using the formulae (whi h by now should already be well-known!)

U(T, V ) = CV T + const , S(T, V ) = CV ln T + nR ln(V /n) + const ,

(for simpli ity onstant heat apa ity of the gas has been assumed) we readily get the
answer

T0
W̄max = CV (T − T0 ) + T0 CV ln .
T
55 This an be relaxed if the pressure p0 of the environment vanishes.

85
The pro ess allowing to extra t this work an of ourse be realized with the help of
an innitesimal Carnot engine. One an also ask the question what maximal work an
be obtained by ooling the gas to the temperature T0 of the environment keeping it at
onstant pressure p0 (or, more generally, if the initial and nal gas pressures are equal
p0 ), if it initially had the temperature T. In this ase one has to onsider also the volume
hanges and ompute the entropy hange with temperature at onstant pressure

nR T0
∆V = (T0 − T ) , ∆S = Cp ln ,
p0 T
and

T0 T0
W̄max = CV (T − T0 ) + T0 Cp ln − nR(T0 − T ) = Cp (T − T0 ) + T0 Cp ln .
T T

Equilibrium state of a system open to its surrounding and the thermodynami al stability
onditions
The results obtained above an be used to formulate the ondition determining equilib-
rium states of systems open to their surrounding (represented by RHS, RVS and RMS)
generalizing thereby the Callenian maximum entropy prin iple (whi h applies to isolated
systems only) and to dis uss stability onditions whi h thermodynami al systems should
satisfy in typi al situations. Some ex eptions will be also mentioned.
Consider a thermodynami al system whi h owing to the internal onstraints is in
equilibrium within itself (various parts of the system are kept in equilibrium by suitable
walls). The external onstraints separate it from the surrounding represented by a heat
reservoir at the temperature T0 , a volume sour e at p0 and a reservoir of matter at
µ0 . Without the external onstraints the system might not be in equilibrium with the
surrounding. If it is possible to hange the onstraints, to whi h the system is subje ted,
so that its availability de reases, a useful (i.e. positive) work an in prin iple be extra ted
from the system through an appropriate reversible pro ess (the formula (122) gives only
an upper bound on the work whi h an be obtained; the system may well attain the new
equilibrium state spontaneously without delivering any useful work). If this is not possible,
one may say that the system is stable: any hange of onstraints requires supplying
to it a positive work from outside (the minimal su h work needed is of ourse equal
Wmin = +∆A > 0). It follows that in a given surrounding represented by T0 , p0 and
µ0 , it is the minimum of A (over the set all possible virtual equilibrium states that an
be fabri ated by applying to the system onstraints stronger than the ones it is a tually
subje ted to, when it is in equilibrium with its surrounding) whi h determines the stable
equilibrium state of the system.
To see what onditions in most typi al situations must be satised if the system is to
be in equilibrium with its surrounding (at T0 and p0 ), we assume (although this needs
not always be true - see the example of the equilibrium of a liquid drop with its vapour
dis ussed by Pippard) that in equilibrium all parts of the onsidered system have the
same temperature T and the same pressure p and ompute (restri ting ourselves for a
moment to systems whi h annot ex hange matter with the surrounding, so that the

86
variables n n1 , . . . , nr remain xed) the hange of the availability orresponding to
or
arbitrary departures ∆S and ∆V of the system's entropy and volume from the (supposed)
equilibrium state (the derivatives are taken at equilibrium values of S and V ):
"  # "  #
∂U ∂U
∆A = Aeq + − T0 ∆S + + p0 ∆V
∂S V,... ∂V S,...
1 ∂2U 2 1 ∂2U 2 ∂2U
+ (∆S) + (∆V ) + ∆V ∆S + . . . (129)
2 ∂S 2 2 ∂V 2 ∂V ∂S
It follows that the minimum of A is realized by the state in whi h the temperature T
and pressure p (assumed here to be uniform throughout the system) are equal to the
temperature T0 and pressure p0 T = T0 and p = p0 . If the
of the surrounding, that is, if
onsidered system is homogeneous this is lear: if T = T0
p = p0 , no useful (positive)
and
work an be extra ted from it. However, assuming by the availability A the minimal value
requires also that the quadrati form of the se ond derivatives of A (whi h translated into
the quadrati form of the se ond derivatives of the internal energy) be (stri tly) positive
denite. Applying the method of minors (who attended my Math II lasses should know
what this is) the following onditions are obtained:

∂2U
   
∂T T
= = > 0,
∂S 2 V,... ∂S V,... CV
 2   
∂ U ∂p 1
2
=− = > 0,
∂V S,... ∂V S,... V kS

and (writing(∂ 2 U/∂S∂V ) in one orner of the matrix of the se ond derivatives as −(∂p/∂S)V
and as (∂T /∂V )S in the other)
       
∂p ∂T ∂p ∂T ∂(p, T ) ∂(p, T ) ∂(V, T ) T
− + ≡ = = > 0.
∂V S ∂S V ∂S V ∂V S ∂(S, V ) ∂(V, T ) ∂(S, V ) V CV kT
Thus the stability of the system requires stri t positivity of its heat apa ity CV at on-
56
stant volume and of its adiathermal, kS , as well as isothermal, kT , ompressibilities.
Positivity of kT is obviously required by me hani al stability of the system (were it nega-
tive, the system would spontaneously ompress itself at onstant temperature). Thus
under normal ir umstan es the availability takes its minimal value at T = T0 and

56 The onditions formulated here an be also given another interpretation: one an onsider a small part
of an isolated homogeneous system whi h is in thermal and me hani al equilibrium at the temperature
T0 and pressure p0 ; with respe t to the small part onsidered the rest of the system plays pre isely the
role of the surrounding at T0 and p0 . This interpretation (en ountered in many sour es) is, however, only
a spe ial ase of the mu h more general situation ansidered here: we do not assume that the system is
homogeneous - it may onsist of several parts (subsystems) - but only that in equilibrium all its parts
have the same temperature T and the same pressure p; furthermore, departures from the equilibrium may
also be due to departures of other variables (other that the system's total entropy S and total volume V )
from their equilibrium values - these departures depend on the nature of the system and must, therefore,
be onsidered separately in ea h parti ular ase (see the examples below).

87
p = p0 , as expe ted, and stability is ensured by positive (and not innite) values of
CV , kS and kT . Similar stability onditions an be derived for other simple thermody-
nami systems like wires or rubber bands subje ted to stret hings (in these ases they
amount to the inequalities - see Le ture II for denitions of the variables - CL > 0,
(∂K/∂L)S > 0 and (∂K/∂L)T > 0) or magneti materials in external magneti elds
(C M> 0, (∂H0 /∂M)S > 0 and (∂H0 /∂M)T > 0).
There are however situations in whi h some of the derived inequalities be ome equali-
57
ties. One su h situation is when kT is innite, while CV and kS are nite. There is then
a dire tion in the (S, V ) spa e in whi h the se ond order term in the expansion of A does
not grow. This o urs at the riti al point of the liquid-vapour system; one then shows
that (∂ 2 p/∂V 2 )T must vanish while (∂ 3 p/∂V 3 )T must be negative in order that A has a
minimum at T = T0 and p = p0 (equal to the riti al values). Another possibility would
be that all inequalities be ome equalities (all the three, CV , kS and kT are innite). As
usually with the onditions for a minimum, the third order terms in the expansion of A
58
would then have to vanish and the fourth order terms, the tetra-form of departures ∆S
and ∆V would have to be positive denite. Detailed analysis of this ase (Pippard refers
here to the Landau and Lifshitz statisti al physi s textbook) shows that there would be
then more onditions than ould simultaneously be satised, so the on lusion is that CV
and kS an never be ome innite if A an be expanded in powers of the departures ∆S
and ∆V around the equilibrium state.
It may also happen that A annot be expanded in the power series. The typi al
situation is a (nonhomogeneous) system onsisting of two phases α and β in equilibrium
with the surrounding at T0 ad p0 whi h are su h that both phases may oexist with an
arbitrary proportion nα /nβ (nα + nβ = n) of the matter in ea h of the two phases. In
su h a ase kT = ∞ be ause hanging the volume does not hange the pressure - only
the proportion nα /nβ is thereby ae ted. So A has a valley of equal minima along a
dire tion in the (S, V ) plane and begins to raise only at the opposite ends of this valley
orresponding to the volumes in whi h one of the two phases ompletely disappears. As
long as both phases are present, the system's equilibrium is neutral (see the dis ussion
in Le ture I). It is amusing to show that onsidering this valley allows to derive the
Clapeyron-Clausius equation for the temperature dependen e of the pressure along the
oexisten e urve. The valley must be in the null dire tion dire tion of the quadrati form
of se ond derivatives in (129):

    
(∂T /∂S)V (∂T /∂V )S ∆S 0
= . (130)
(∂T /∂V )S −(∂p/∂V )V ∆V 0

From the rst line of this equality (the se ond line give a linearly dependent equation -
this is ensured by vanishing of the determinant of the above matrix, that is, by the innite

57 Re all the home Problems in whi h one shows that Cp = CV + T V α2p /kT and kS = (CV /CP )kT .
From these relations it follows that Cp ≥ CV (the equality requires either αp = 0 or innite kT ) and,
onsequently, kT ≥ kS .
58 Tetra-form like the tetradra hm - a Greek oin, an artefa t from antiquity - or the tetrar hy - the
system of governing the Roman empire established by Diokletianus.

88
value of kT ) one gets that the valley dire tion is su h that

   
∆S (∂p/∂V )S ∂S ∂p
=− = = , (131)
∆V (∂T /∂S)V ∂V T ∂T V

(on the right hand side rst the sho king relation has been used and then the well-
known Maxwell identity). Along the valley however, the entropy and the volume hanges
are given by

∆S = sα ∆nα + sβ ∆nβ , ∆V = vα ∆nα + vβ ∆nβ ,

59
but sin e nα + nβ = n is xed, ∆nα = −∆nβ and therefore (131) takes the form

dp sα − sβ
= ,
dT vα − vβ

whi h is just the Clapeyron-Clausius equation (to whi h we will return in the last Le ture
devoted to thermodynami s). In most ases neither the numerator nor the denominator
2 2 2 2
on the right hand side vanishes, whi h means that neither (∂ U/∂S )V nor (∂ U/∂V )S
vanish, that is CV and kS are nite (though kS and Cp are innite). On the melting urve
3
of 2 He (the lighter isotope of Helium) there is a point at whi h dp/dT = 0 whi h means
(0, ∆V ) dire tion; this in turn implies (sin e the matrix in (130)
that the valley lies in the
2 2
must vanish on this ve tor) that (∂ U/∂V )S = 0, that is, that kS = ∞. Similarly a point
at whi h dT /dp = 0 would probably exist on the melting urve of i e were it not for the
transformation of ordinary i e into its another phase (H2 O has many dierent phases),
2 2
whi h in the analogous way would maen vanishing of (∂ U/∂S )V = 0, that is CV = ∞.
This shows that the nitness of CV and kS are not absolute thermodynami requirements
(but ex eptions are rare).

Spe ial ases of the general stability riterion (determining in the variational way the
equilibrium states of thermodynami systems in dierent onditions) formulated above
deserve onsideration.

• If the system is isolated,U +p0 V stays onstant60 and the ondition of minimal value
of the availability A = U + p0 V − T0 S is equivalent to the ondition of maximum
entropy dis ussed before.

• If the volume of the system is xed (there is no onta t with RVM) and p − p0 in
the expansion (129) is indeterminate. The minimum of A is at T = T0 and the
other parameters spe ifying the state of the system are determined by minimizing
the free Helmholtz energy F (T, V ) at onstant V and onstant T = T0 .
59 Sin e along the valley the pressure does not depend on the volume, the derivative (∂p/∂T )V a quires
the meaning of the derivative dp/dT along the oexisten e urve.
60 If it is isolated both thermally and me hani ally, U and V are separately onserved; if it is only
thermally isolated but in me hani al onta t with RVS, the ombination U + p0 V is onserved, be ause
U + URVS is.

89
 • If the system is in thermal and me hani al onta t with its equilibrium at T0 and p0 ,
and if in equilibrium all its parts have the same temperature T and the same pressure
p, then T = T0 and p = p0 and the values of the remaining parameters spe ifying it
states are determined by minimizing the Gibbs fun tion G = U − T S + pV at xed
T = T0 and p = p0 .

As an example of appli ations of the stability onditions dis ussed above, we an

onsider the equilibrium of the system onsisting of n moles of a liquid and its vapour,
under dierent onditions. For simpli ity we an assume that there is only one matter
omponent. Let rst the vessel ontaining the system be open to a onstant external
pressure p0 and held at a temperature T0 by a thermal onta t with the environment
(playing the role of the heat-bath). The availability of the system orresponding to virtual
states in whi h the transfer of matter between the liquid and the vapour is blo ked by a
suitable wall is

A = nl (ul − T0 sl + p0 vl ) + nv (uv − T0 sv + p0 vv ) , (132)

where nl (nv ) is the number of moles of the liquid (vapour) in the vessel. Obviously,
nl + nv = n and all quantities ul,v et ., are omputed at T0 and p0 (they are therefore
xed) be ause we expe t that in equilibrium both subsystems (the liquid and the vapour)
should have the same temperatures and pressures and from the previous onsiderations
eq
we know that these must be equal to T0 and p0 , respe tively. The equilibrium values nl
eq eq eq
and nv (also satisfying the relation nl + nv = n) minimize A (or equivalently, the Gibbs
eq
fun tion, G(T0 , p0 , nl , nv ) at onstant T0 and p0 ). That is δA = 0 at nl = nl . From this
we nd the equilibrium ondition

ul (T0 , p0 ) − T0 sl (T0 , p0 ) + p0 vl (T0 , p0 ) = uv (T0 , p0 ) − T0 sv (T0 , p0 ) + p0 vv (T0 , p0 ) ,

that is, gl (T0 , p0 ) = gv (T0 , p0 ) µl (T0 , p0 ) = µv (T0 , p0 ). One

or, as we will see shortly,
should also noti e that in this ase the se ond derivative of A with respe t to nl vanishes
identi ally - the equilibrium is of the neutral nature (the value of nl is not xed by the
equilibrium ondition); this has already been dis ussed in this Le ture.
As the se ond situation we onsider the same mixture of n moles of a liquid and its
vapour but now as an isolated system of xed total energy U = nl ul + nv uv and xed
volume V = nl vl + nv vv . Minimization of the system's availability (132) redu es now to
maximizing the total entropy S = nl sl + nv sv , be ause U and V are xed. The external
pressure p0 and the temperature T0 do not play any role here (the system is isolated from
the external pressure and from the bath). However to seek the extremum of S respe ting
61
the onstraints we use the method of the Lagrange multipliers and equate to zero the

61 Of ourse, were we given the expli it forms of s (u , v ) and s (u , v ), we ould write down the total
l l l v v v
entropy S as an expli it fun tion S(U, V, n, ul, nl ), say, and maximize it dire tly treating 0 ≤ ul ≤ U/n
and 0 ≤ vl ≤ V /n as independent variables; in equilibrium both subsystems should have the same T and
p and, therefore, all the molar quantities ul , vl , sl and uv , vv , sv are parametrized by the (unknown yet)
temperature T and pressure p; these two variables an be onveniently traded for ul and vl .

90
variation of the the auxiliary fun tion in whi h the parameters T and p are the Lagrange
62
multipliers

δ(U − T S + pV ) = 0 .

The virtual equilibrium states whi h should be onsidered now may orrespond to dierent
partitions of the total internal energy U, the total volume V and numbers of moles (and,
hen e, to dierent entropies) between the two phases. Therefore the above variation an
be written as

nl (δul − T δsl + p δvl ) + nv (δuv − T δsv + p δvv )

+ δnl (ul − T sl + p vl ) + δnv (uv − T sv + p vv ) = 0 .

We know however, that the two Lagrange multipliers T and p an be hoosen in su h a

way as to kill the oe ients of nl and nv . This is possible, be ause δul , δvl and δsl are
hanges of the parameters between two equilibrium states of one mole of isolated liquid
and, as has been stated at the end of Le ture V, these hanges are always orrelated
in this way with T and p being the temperature and pressure of the liquid; the same
pertains also to the hanges δuv , δvv and δsv . Thus in this way, the Lagrange multipliers
a quire the proper meaning of the system's temperature and pressure and the equilibrium
ondition is

δnl (ul − T sl + pvl )|T,p + δnv (uv − T sv + pvv )|T,p = 0 ,

that is, be ause δnl = −δnv ,

gl (T, p) = gv (T, p) , (133)

whi h of ourse must be solved together with the onditions

nl ul (T, p) + nv uv (T, p) = U , nl vl (T, p) + nv vv (T, p) = V ,

to yield the equlibrium parameters T , p, nl and nv = n − nl . The ondition (133) is, as we

will dis uss, the same as µl (T, p) = µv (T, p) (be ause there is only a single omponent).
Finally on an also onsider the equilibrium of the liquid with its vapour in the vessel
of xed volume but held at onstant temperature T0 by a thermal onta t with the
environment. In this ase minimizing A redu es, be ause V is xed, to minimizing U−
T0 S , or be ause the external temperatiure T0 is also the temperature of the system, to
63
minimizing

F (T0 , V, nl , nv ) = nl fl (T0 , vl ) + nv fv (T0 , vv )

62 Maximizing S it would be more appropriate to write δ(S − λ1 U − λ2 V ) = 0, as in similar Math II
problems, with the Lagrange multipliers λ1 and λ2 , but this is learly equivalent to δ(U − T S + pV ) = 0.
63 Again, sin e in equilibrium both phases must have the same pressure, v and vv are not independent
l
and if the expli it formulae for vl = vl (T0 , p) nd vv = vv (T0 , p) were give, as well as the expli it forms of
fl (T0 , vl ) and fv (T0 , vv ), one ould dire tly minimize F (T, V, ul , nl ).

91
with the ondition of xed V = nl vl + nv vv whi h is taken into a ount by introdu ing
the Lagrange multiplier p, whi h then a quires the meaning of the internal pressure of
the system. Similar reasoning as above then leads to the onditions

gl (T0 , p) = gv (T0 , p) , nl vl (T0 , p) + nv vv (T0 , p) = V ,

whi h together determine the equilibrium pressure p and the numbers nl and nv of moles
in the two phases.

92
 LECTURE VIII (TMD)

Fundamental relations in the representations of F , G and H fun tions and of the Massieu-
Plan k fun tions
The onsiderations arried out in the pre eding Le ture showed that in spe i onditions,
when the system is open to its surrounding (is held at onstant temperature and/or
pressure through the onta t with a suitable reservoirs representing the surrounding), its
equilibrium states are determined by minima (over the set of all possible virtual states)
of the Helmholz free energy F (if T and V are xed), or the enthalpy H (if S and p
are xed) or the Gibbs fun tion G (if T and p are xed). The prin iples of minimum
of F, H or G in these situations repla e the Callenian prin iple of maximal entropy. In
this sense all these fun tions also play roles of thermodynami potentials. We have
also argued that if entropy S of an isolated system is known as a fun tion of its global
(extensive, if the system has this property) parameters U, V , . . . and n (or n1 , . . . , nr ),
the thermodynami al information about the system if omplete. We now want to argue
that if the termodynami potentials are known as fun tions of their natural variables:
F as a fun tion of T , V and n (or n1 , . . . , nr ), H as a fun tion of S, p and n or G
as a fun tion of T , p and n, the thermodnami al information about the system is also
omplete. The distinguished role of the parti ular thermodynami al potential in given
external onditions stems pre isely from the fa t that its natural variables are just those
whi h in the given situation are dire tly ontroled (by the environment).
From the mathemati al point of view the operations whi h one does passing from
the internal energy U known as a fun tion of S, V and n to the other potentials in
their natural variables is alled Legendre transformation whi h is pre isely the way of
swit hing from a given fun tion f (x) to another fun tion g(p), where p is the derivative
of f, without loosing the information about the form of f (only oding it dierently).
Sin e the Legendre transformation is used in many pla es in physi s, it is appropriate to
dis uss it here in the general way.
64
Let f represent a physi al quantity whi h is theoreti ally given as a onvex fun tions
(upwards or downwards) fun tion f (x) of some variable x. Suppose, however, that dire tly
ontrolled experimentally is not the variable x itself but the derivative p of f with respe t
to x. One ould then try to invert the relation

p(x) = df /dx ,

to get x = x(p) and to swit h to the fun tion

f˜(p) ≡ f (x(p)) . (134)

whi h would represent the same physi al quantity (be ause of this a physi ist would
just write f (p)). This is pre isely what one does expressing U in terms of T instead of

64 The generalization of the Legendre transform to non onvex fun tions is alled the Fen hel transform
and is dened as g(p) = supx (f (x) − p x) and plays the important role in the theory of phase transistions.

93
 y

Figure 9: The family of fun tions fa (x) whi h all yield the same fun tion f˜(x(p)) where
p is the slope of the tangent line.

expressing it in terms of S. But swit hing to f˜(p) one looses some information about the
form of f (x): the fun tions f (x) and fa (x) ≡ f (x − a) with an arbitrary shift onstant a
lead to the same f˜(p) (see Figure 9).
However the set of lines tangent at every point to a given a onvex urve on the
plane (x, y) determines this urve uniquely: geometri ally the plot of y = f (x) is just the
envelope of the set these tangent lines. In turn, every tangent line is uniquely determined
by its slope p (the variable we want to play with) and the value g of the interse tion of the
tangent with the y -axis. It su es therefore to give g as a fu tion of the slope (dening
in this way the family of the tangent lines) to retain the omplete information about the
form of the original fun tion f (x). To this end we onsider a point x0 and write down the
equation of the tangent to f (x) at this point (see Figure 10):

y = p x + f (x0 ) − p x0 .
The value g of the interse tion of this tangent with the y -axis is therefore equal (renaming
now x0 to x)
g = f (x) − p x .
If now x is written as x = x(p) inverting the relation (134) we will get the fun tion

g(p) = f (x(p)) − p x(p) ,

whi h ontains the same information about the dependen e of the quantity f on the
variable x as the original fun tion f (x), but is expressed through the variable whi h is
easily ontrolled experimentally. That the ombination f (x) − px is indeed a fun tion of
the variable p an be also seen by onsidering its dierential

df
dg ≡ d(f (x) − p x) = df − d(p x) = dx − p dx − x dp = −x dp .
dx
Thus, the hange of g depends only on the hange dp of the argument p and not on dx;
the fun tion g(p) an be therefore re onstru ted from its dierential.
Of ourse a fun tion of several variables, f = f (x1 , . . . , xn ) an be Legendre trans-
formed in an arbitrary number of its variables to obtain, say,

k
X
g(p1 , . . . , pk , xk+1 , . . . , xn ) = f (x1 , . . . , xn ) − pj xj , (135)
j=1

94
 y

f (x0 )

g(p)
x0 x

Figure 10: Geometri al onstru tion of the Legendre transform g value g(p) orresponding
to the value f (x0 ) of a ovex fun tion f.

where the variables x1 , . . . , xk should be expressed in terms of the variables pi by inverting

with respe t to them the k relations
∂f
pi = . (136)
∂xi

Students asso iate the Legendre transform primarily with lassi al me hani s in whi h
it is used to pass from the Lagrange to Hamilton's formulations of the equations of mo-
l l l
tion, i.e. to trade the generalized variables q and q̇ , l = 1, . . . , n for the variables q
l
and pl = ∂L/∂q . (Be ause of this, students have the tenden y to think that the Leg-
65
endre transform has something to do with the fa t that in me hani s variables ome
in pairsq i and q̇ l . The derivation above learly shows this is not so.) The Hamilto-
1 n
nian H(p1 , . . . , pn , q , . . . , q ) is just the (minus, in order that it has - in most ases
the interpretation of me hani al energy) Legndre transform of the Lagrangian fun tion
L = L(q 1 , . . . , q n , q̇ 1 , . . . , q̇ n ) in n its last variables
n
X
1 n
H(p1 , . . . , pn , q , . . . , q ) = pj q̇ j − L(q 1 , . . . , q n , q̇ 1 , . . . , q̇ n ) .
j=1

It is also interesting to illustrate the working of the Legendre transform on another

physi al example taken from ele trostati s. Let us determine the for e by whi h the two
plates of a apa itor attra t ea h other. We remember that the voltage ϕ between the
plates is related to the harge Q and the apa ity C by Q = C ϕ. If the apa itor is
being haged by su essively bringing onto its plate innitesimal portions dQ of harge
(e.g. in the at apa itor by transporting su essively and reversibly - in the sense that
an external for e only ountera ts the ele tri for es a ting on the transported harge -
portions dQ from one plate to the other) the work d¯W done on it by external for es is
equal d¯W = ϕ dQ. Charging in this way the apa itor with the harge Q requires doing
65 Legendre, Lagrange, Lapla e, Lavoisier - who an distinguish all these Fren h L-masters?

95
on it the work

Q Q Q
Q Q2
Z Z Z
d¯W = ϕ dQ = dQ = ≡U,
0 0 0 C 2C

whi h is therefore equal to the energy of the harged apa itor (its internal energy if we
treat the apa itor in the thermodynami way).
If the apa ity C of the apa itor is altered as a result of the a tion of an external
for e F (when the appli ation of the external for e makes possible hanging the distan e
between the plates - in the language of thermodynami s - reversibly), the work done by
this external for e is just the hange of the internal energy of the apa itor, provided the
apa itor is an isolated system, that is, has xed harge Q (we do not onsider here the
heat apa ity of the plates). This allows to nd the for e by whi h the plates attra t ea h
other. (This in Feynman's Le tures on Physi s is alled the virtual works prin iple but in
fa t this is pre isely the same as applying 1TMDL to isolated systems - in thermodynmi s
systems adiathermally isolated; here, however, the apa itor ould be not isolated also
be ause of being onne ted to a battery, so we need to ex lude this possibility and this is
just the ondition of onstan y of Q.) Let us write this statement in the thermodynami al
way
   
∂U ∂U
dU(Q, C ) = dQ + dC ≡ ϕ dQ + F dC ,
∂Q C ∂C Q

By F we have denoted here the generalized for e related to the hange of the apa ity at
xed harge on the apa itor. Expressing the apa ity C through the spatial hara teris-
ti s of the apa itor (the area of its plates, the distan e between them) allows to give the
genearlized for e F the ordinary me hani al meaning. For example if C hanges due to
hanging the distan e between the plates

dC
dU(Q, C )|Q=const. = F dC = F dz ≡ Fz dz .
dz
In this way we nd that the external for e Fz whi h pre isely balan es the for e by whi h
the plates of the at apa itor of area A and separated by the distan e z between the
pales (the apa ity of su h a apa itor is in the SI system equal C = ε0 (A/z)) attra t
ea h other is given by

Q2
   
∂U dC ∂U
Fz = = = .
∂C Q dz ∂z Q 2ε0 A

The for e Fz found in this way annot depend, of ourse, on whether the apa itor
has xed harge Q, or whether it is onne ted to a battery whi h keeps it at the voltage
ϕ of su h a magnitude that the harge on the plates is equal Q. If we write, however, the
energy U as a fun tion U(ϕ, C ), expressing simply the harge through the voltage and the

96
 apa ity and try to dene the generalized for e F a ting between the apa itor plates at
onstant voltage ϕ by the relation
 
∂U
dU(ϕ, C )|ϕ=const. = dC ≡ F dC (incorrectly),
∂C ϕ

we will get the for e Fz with the wrong sign. To obtain the right sign one has to perform
the Legendre transform, that is, to pass to the fun tion

Ũ(ϕ, C ) = U(Q(ϕ, C ), C ) − ϕ Q(ϕ, C ) ,

whose natural dierential is

! !
∂ Ũ ∂ Ũ
dŨ(ϕ, C ) = dϕ + dC ≡ −Q dϕ + F˜ dC ,
∂ϕ ∂C
C ϕ

As we an now fully ontroll the voltage, we an ompute the for e Fz as the oe ient
in dŨ of the dz dierential at xed voltage ϕ
  
 ∂ 1 2
dŨ(ϕ, C ) = C ϕ − ϕ Q(ϕ, C ) dC

ϕ=const. ∂C 2
  
∂ 1 2 1 dC
= − Cϕ dC = − ϕ2 dz ≡ Fz dz .
∂C 2 2 dz

In this way one gets the right expression for Fz (with the right sign).
The physi al reason for the ne essity to pass to the fun tion Ũ is, of ourse, that the
apa itor onne ted to the battery is not an isolated system and hanging its apa ity
C by moving its plates entails doing on it some work also by the battery, whi h has to
supply to the apa itor an additional harge (to maintain the voltage un hanged). The
energy balan e in this ase therefore reads

1 2 dC
Fz dz + d¯Wbat = dU(ϕ, C ) = ϕ dz .
2 dz
The work done by the battery onsist of supplying the additional harge ϕdC to the
¯Wbat = ϕ2 dC . Putting this work
apa itor at the voltage ϕ; therefore this work equals d
on the other side of the above equality yields the right for e Fz . In the thermodynami 
approa h onsisting of applying the prin iple of virtual works to the fun tion Ũ instaed of
U , the work done by the battery is already automati ally taken into a ount by ontrolling
the voltage ϕ. One should noti e here the analogy to using the Helmholtz free energy F
instead of the internal energy U in omputing the volume work done by a system whi h
is kepth in equilibrium with a heat bath at xed temperature T

In thermodynami s two families of potentials an be onstru ted. The rst one (more
ommonly used) is obtained starting from the fundamental relation (97) in the internal
energy representation U = U(S, V, . . . , n1 , . . . , nr ). In the ase of a simple system the

97
three basi potentials are the already introdu ed enthalpy (to simplify the notation
Pr n
stands for n1 , . . . , nr and µ dn for j=1 µj dnj )

H(S, p, n) = U + p V , (137)

dH = T (S, p, n) dS + V (S, p, n) dp + µ(S, p, n) dn , (138)

the Helmholtz free energy

F (T, V, n) = U − T S , (139)

dF = −S(T, V, n) dT − p(T, V, n) dV + µ(T, V, n) dn , (140)

and the Gibbs fun tion (the free enthalphy)

G(T, p, n) = U − T S + p V , (141)

dG = −S(T, p, n) dT + V (T, p, n) dp + µ(T, p, n) dn . (142)

One also denes the Grand potential

Ω(T, V, µ) = U − T S − µ n , (143)

dΩ = −S(T, V, µ) dT − p(T, V, µ) dV − n(T, V, µ) dµ . (144)

In the similar way one denes the fun tionsH , F and G pertaining to other simple systems
like wires and rubber bands (−p → K , V → L) lms (−p → γ V → A) of magneti s
(−p → H0 in the normal Gauss system or −p → µ0 H0 in the SI system, but if µ is used to
denote the hemi al potential, the SI system be omes learly in onvenient, and V → M ).
If the system is not simple and has more variables, for example a paramagneti gas whose
internal energy depends, in addition to the number of moles, of S, V and M one an form
more Legendre transforms and their names are not odied. We have already used one of
su h potentials in dis ussing the onne tion between piezoele tri ity and ele trostri tion
in Le ture V.
As follows from the onstru tion of the Legendre transform, all these potentials, if
known as fun tions of their natural variables (those whi h are expli itly indi ated in the
formulae above), ontain full thermodynami al information about systems to whi h they
pertain. Therefore the relations H = H(S, p, n), F = F (T, V, n) or G = G(T, p, n) an be
alled fundamental relations in the representations of enthalpy, free energy and Gibbs, re-
spe tively. To show this it is su ient to realize that from ea h of them U = U(S, V, . . . , n)
an be obtained by the repeated Legendre transform (the Legendre transform applied
twi e to a onvex fun tion is the identity operation). For istan e, knowing F = F (T, V, n)
one an write
   
∂F 2 ∂ F
U(T, V, n) = F + T S = F − T ≡ −T , (145)
∂T V,n ∂T T V,n

and inverting the relation S = −(∂F/∂T )V,n to get T = T (S, V, n) obtain U = (S, V, n).
Moreover the potentials allow to obtain Maxwell identities more straightforwardly than

98
does U = U(S, V, n). For instan e the identity (∂S/∂V )T,n = (∂p/∂T )V,n is an immediate
onsequen e of the fa t that F is a state fun tion and therefore its mixed se ond derivatives
must be equal. As another illustration of the usefulness of F we an obtain the dependen e
of the heat apa ity CV on the volume:
∂3F ∂2
         2 
∂CV ∂ ∂S ∂F ∂ p
= T = −T 2
= −T 2
= −T .
∂V T ∂V ∂T V T ∂V ∂T ∂T ∂V T ∂T 2 V
The distinguished role of the potentials F and Ω stems from the fa t that equilibrium
statisti al physi s gives dire t pres riptions to determine them on the basis of the mi-
ros opi dynami s ( lassi al or quantum) of the onsidered system subje t to spe i
onstraints: the free F Helmholtz energy if the system is isolated from the external pres-
sure but remains in thermal onta t with a heat bath at temperature T and the potential
Ω in the ase of systems ex hanging matter with a reservoir at the hemi al potential µ.
The other family of thermodynami potentials (used pra ti ally only by a very narrow
family of thermodynami s spe ialists) is onstru ted taking as the starting point the
fundamental relation in the entropy representation (92) whose dierential

1 p µ
dS = dU + dV − dn ≡ θ dU + η dV − ν dn , (146)
T T T
denes the variables θ, η and ν and performing the Legendre transforms to one or more
of these variables. One obtains in this way the so- alled Massieu-Plan k fun tions (po-
tentials). However, sin e with perhaps slightly more labour the same results an always
be arrived at with the help of the potentials H, F and G, we will not dis uss them here
any more.

Returning to the usual potentials it is good to noti e that although the me hani al
quantities like p in the dierentials dU and dF are denoted by the same letter, they are
in prin iple fun tions of dierent variables (the natural ones for the respe tive potentials)
and their experimental determination requires spe ifying the thermodynami al onditions.
As an example onsider the wire stret hed by the for e K and satisfying the Hooke's law.
If we are given its internal energy U = U(S, L), the tension K is dened as the for e
needed to reversibly stret h the wire under adiathermal onditions, that is at onstant
entropy of the wire: dU|S=const = KdL. In this way K = K(S, L). The oe ient k in
the Hooke's law whi h in me hani s is translated into the formula for the wire potential
1 2
energy Epot = k(L − L0 ) is however measured usually by stret hing the wire at onstant
2
temperature T (of the environment). In su h onditions the (minimal) work whi h must
be done by RWS to stret h the wire from its equilibrium state with the length L0 is, as
dis ussed in the pre eding Le ture, given by ∆F at xed T. in this way measured is the
oe ient k(T ) in the dierential (written using the Hooke's law)

dF = −SdT + k(T )(L − L0 )dL .

Integrating this formula one gets

1
F (T, L) = F (T, L0 ) + k(T )(L − L0 )2 .
2
99
Sin e S = −(∂F/∂T )L , one obtains

dF (T, L0 ) 1 dK 1 dK
S(T, L) = − − (L − L0 )2 ≡ S(T, L0 ) − (L − L0 )2 .
dT 2 dT 2 dT
Applying now the formula (145) one obtains
 
1 dK(T )
U(T, L) = F (T, L0 ) + T S(T, L0 ) + k(T ) − T (L − L0 )2
2 dT
 
1 dK(T )
= U(T, L0 ) + k(T ) − T (L − L0 )2 .
2 dT
1
Therefore the fa tor k (L − L0 )2 in the internal energy of
dened as the oe ient of
2
the wire (whi h one would naturally split into the thermal energy of the wire and its
me hani al energy) is not what is measured in typi al me hani al experiments ondu ted
at onstant temperature.

Consequen es of extensiveness: the Gibbs-Duhem relation and hemi al potentials

Re onstru ting the fundamental relation of a given system, that is, obtaining the full
thermodynami al information about it, requires exploiting various experimental data. In
the ase of a simple body onsisting of a xed amount of matter one prin ipal heat apa ity
at one xed value of volume or pressure must be measured as a fun tion of temperature
and the equation of state must be known (its form an be determined by measuring various
dierential oe ients). Together they provide su ient information on the oe ients
of the dierentials in the forms dU = CV dT + (∂U/∂V )T dV (or dU = (∂U/∂T )p dT +
(∂U/∂p)T dp) and dS = (CV /T )dT + (∂S/∂V )T dV (or dS = (∂S/∂T )p dT + (∂S/∂p)T dp)
to allow to integrate them up and to obtain U = U(T, V ) and S = S(T, V ) (or U =
U(T, p) and S = S(T, p)) whi h is equivalent to knowing the relations S = S(U, V ) or
U = U(S, V ). Re onstru ting the dependen e of entropy and of internal energy on ea h
additional parameter in the ase of nonsimple systems, e.g. on the magnetization M ,
if the work of magnetization d ¯W = H0 dM an be (reversibly) done on the system in
addition to the usual volume work −p dV , would require determining from data one more
66
fun tion. Similarly, re onstru ting the dependen e on the number(s) of moles n (ni ,
i = 1, . . . , r ) would in prin iple require suplementary data to determine the oe ient(s)
of the dierential(s) dn (dni ) in dU and/or dS . The extensiveness property of U and
S (if the system an be treated as extensive) puts, however, very stringent onstraints
whi h uniquely determine the dependen e of any state fun tion on the total amount of
Pr
matter represented by the total number of moles n = i=1 ni . Thus, if the system is
made up of only one sort of mole ules no additional data is needed beyond those related
to the reversible performan e of works and the reversible heat transfers. We now derive
the relations whi h follow from extensiveness of the system.

66 All su h fun tions are alled the equations of state. Similarly as e.g. CV , whose dependen e on the
volume in the ase of simple systems is entirely determined by the equations of state, they must satisfy
denite onsisten y onditions (following from the fa t that U and S are state funtions) whi h, however,
do not determine them ompletely, leaving room for experimental input.

100
 One of the Callenian postulates (Le ture VI) is that entropy of an extensive system is
a homogeneous fun tion of order one of its extensive arguments:

S(λU, λV, . . . , λn1 , . . . , λnr ) = λ S(U, V, . . . , n1 , . . . , nr ) . (147)

The same relation applies of ourse to U = U(S, V, . . . , n1 , . . . , nr ). One of the onse-

quen es, already dis ussed in Le ture VI, is that

S = n s(u, v, . . . , x1 , . . . , xr ) , U = n u(s, v, . . . , x1 , . . . , xr ) ,
where u, s, et ., are molar
Pr quantities (pertaining to one mole of the substan e) and
xi = ni /n, where n = i=1 ni , are molar fra tions. It also follows that the intensive
parameters T , p, µi whi h are dened as derivatives of U with respe t to the extensive
parameters must be homogeneous fun tions of order zero:

T = T (λS, λV, . . . , λn1 , . . .) = T (S, V, . . . , n1 , . . .) = T (s, v . . . , x1 , . . .) ,

p = p(λS, λV, . . . , λn1 , . . .) = p(S, V, . . . , n1 , . . . , nr ) = p(s, v . . . , x1 , . . .) , (148)

µi = µi (λS, λV, . . . , λn1 , . . .) = µi (S, V, . . . , n1 , . . . , nr ) = µi (s, v . . . , x1 , . . .) .

Sin e x1 + . . . + xr = 1, the 1+o+r intensive parameters (here o is the number of works
whi h an be done on the system) depends on only 1 + o + r − 1 = o + r variables, whi h
implies that there must be one relation linking the intensive variables T , p, . . ., µ1 , . . .,
µr . If there is only one omponent (r = 1), this relation uniquely determines the single
hemi al potential µ as a fu tion of the 1 + o intensive parameters T , p, . . ..
Furthermore, dierentiating the relation (147) and the analogous relation written for
U with respe t to λ and setting then λ = 1, one obtains that

r
X
U(S, V, n1 , . . .) = ST (S, V, n1 , . . .) − V p(S, V, n1 , . . .) + nj µj (S, V, n1 , . . .) , (149)
j=1
Xr
S(U, V, n1 , . . .) = Uθ(U, V, n1 , . . .) + V η(S, V, n1 , . . .) − nj νj (S, V, n1 , . . .) ,
j=1

(re all that θ ≡ 1/T , η ≡ p/T and νj ≡ µj /T ). Writing now the dierential of U in its
natural variables rst using U = U(S, V, n1 , . . .) and then using the form (149) of U we
get

r
X
dU = T dS − p dV + µj dnj ,
j=1
r
X r
X
dU = T dS + SdT − p dV − V dp + µj dnj + nj dµj .
j=1 j=1

Subtra ting now these two expressions side by side leads to the formula

r
X
SdT − V dp + nj dµj = 0 , (150)
j=1

101
known as the Gibbs-Duhem relation. Analogous operations done on the entropy writen
in two dierent ways give

  r
1 p X µ 
j
Ud +Vd − nj d = 0. (151)
T T j=1
T

Dividing (150) by the total number n of moles gives the orrelation of the dierentials of
the intensive parameteters whi h depends only on molar quantities:

r
X
s dT − v dp + xj dµj = 0 , (152)
j=1

and expresses the already noted fa t that of 1+o+r intensive parameters hara terizing an
extensive system, only o+r are independent (therefore their dierentials must be linearly
dependent). If there is only one omponent, so that x1 = 1, and the molar entropy s and
molar volume v ara known as fun tions of the temperature T and pressure p, the relation
(152) in written in the form

dµ = −s(T, p) dT + v(T, p) dp , (153)

an be integrated to give the hemi al potentials µ(T, p) up to a onstant (the hemi al

potential at some referen e values T0 , p0 ).
From the form (149) of the internal energy it immediately follows that

r
X
H(S, p, n1 , . . .) = ST (S, p, n1 , . . .) + nj µj (S, p, n1 , . . .) ,
j=1
r
X
F (T, V, n1 , . . .) = −V p(T, V, n1 , . . .) + nj µj (T, V, n1 , . . .) ,
j=1
r
X
G(T, p, n1 , . . .) = nj µj (T, p, n1 , . . .) , (154)
j=1
Ω(T, V, µ1 , . . .) = −V p(T, V, µ1 , . . .) ,

The same relations follow, of ourse, from the s aling properties of these fun tions: e.g.
dierentiating with respe t to λ the relation

F (T, λV, λn1 , . . .) = λF (T, V, n1 , . . .) ,

and setting λ=1 the se ond relation (154) is obtained. If there is only one omponent
in the system, the third of these relations, after division by n gives

1
µ(T, p) = g(T, p) ≡ G(T, p, n) = u − T s + p v . (155)
n

102
As advertised in the pre eding Le ture, in this ase the hemi al potential is just the
molar Gibbs fun tion. The formula (155) gives another method of al ulating the hemi al
potential of a system omposed of one omponent only.
If the system is omposed of more than one omponent, the Gibbs fun tion G is given
by (154) as the sum of hemi al potentials weighted by the respe tive mole numbers, or

r
X
g(T, p, x1 , . . . , xr−1 ) = xj µj (T, p, x1 , . . . , xr ) , (156)
j=1

and determination of the individual hemi al potentials requires more experimental input.

Mixture of perfe t gases and entropy of mixing

The simplest multi omponent system whi h an be analyzed is the mixture of r dierent
perfe t gases. Its thermodynami al fun tions an be expli itly written down by appealing
to the so alled Gibbs postulate (whi h repla es the experimental input) whi h says that
the internal energy U and entropy of su h a mixture in equilibrium is, when expressed
through temperature and volume in whi h it is en losed, simply the sum of internal
energies and entropies of the individual gases treated as independent (on a ount of the
fa t that they are mutually nonintera ting) and en losed in the same volume (we remem-
ber, however, that the internal energy of a perfe t gas depends only on its temperature,
and not on the volume it o upies):

r
X r
X r
X
U(T, n1 , . . . , nr ) = Ui (T, ni ) = ni ui (T ) = n xi ui (T ) ,
i=1 i=1 i=1
r
X r
X r
X
S(T, V, n1 , . . . , nr ) = Si (T, V, ni ) = ni si (T, vi ) = n xi si (T, vi ) , (157)
i=1 i=1 i=1

where n = n1 + . . . + nr , xi = ni /n and vi = V /ni .

The Gibbs postulate is onsistent with the fa t that in general the formulae for
U(T, V, n1 , . . . , nr ) and S(T, V, n1 , . . . , nr ) must (the Callenian approa h!) follow from
the fundamental relation (92): sin e the Gibbs postulate applies to perfe t gases only, it
is in prin iple possible to invert the formula

X X Z T
U= ni ui (T0 ) + ni dT ′ cv(i) (T ′ ) ,
i i T0

with respe t to T and to obtain in this way S = S(U, V, n1 , . . . , nr ). Moreover, the general
formula dS = (dU + p dV )/T is also satised:
X ∂Si X ∂Si X 1 X pi
dS = dUi + dV = dUi + dV
i
∂Ui i
∂V i
Ti i
Ti
! !
1 X 1 X 1 p
= d Ui + pi dV = dU + dV .
T i
T i
T T

103
 p1

n1 , n2 , n3
p p1

T, V
p3

Figure 11: Comparison of the total and partial pressures of a mixture of three gases using
the van 't Ho box. Dotted lines represents membranes permeable to mole ules of only
one gas ea h.

The immediate onsequen es of the postulate are the formulae for the molar heat apa ity
of the mixture:

r  
X 1 ∂U
cmix
v = xi c(i)
v , bo cmix
v ≡ ,
i=1
n ∂T V

and its pressure (we remember the formula for S(T, V, n) of the perfe t gas!):

r r
pmix
 
mix
X ∂S X R
p = pi bo ≡ = ni ,
i=1
T ∂V U,ni i=1
V

be ause the derivative at onstant internal energy U is here (that is for perfe t gases!)
the same as the derivative at onstant temperature T . As a onsequen e of the Gibbs
postulate, the pressure of a mixture of perfe t gases is therefore, onsistently with the
known Dalton law, the sum of the so- alled partial pressures, that is the pressures whi h
ea h of the gases would individually exert on the walls of the ontainer in the absen e of
the other onsituent gases (we drop from now the subs ript mix):

r r
X X RT RT
p= pi = ni ≡n .
i=1 i=1
V V

Both these onseqen es of the Gibbs postulate an be veried: the rst one by trivial
measurement of the molar heat apa ities of individual gases and of its mixture and the
se ond one by the devi e alled the van 't Ho box shown in Figure 11 - whi h in prin iple
allows to measure and ompare the total and partial pressures. The Gibbs postulate nds
its justi ation in statisti al me hani s - the ase of a mixture of perfe t gases, made of
67
mole ules whose mutual itera tions are negligibly weak is solvable within the so alled

67 But some intera tions must be present be ause otherwise the mixture of gases ould never ome to
equilibrium and dierent gases in the mixture ould have dierent temperatures as it happens in the
present day Universe in whi h perfe t and mutually nonintera ting gases of photons and neutrinos have
indeed dierent temperatures.

104
 anoni al ensemble approa h whi h gives the Helmholtz fun tion F (T, V ) of the system:
in the ase of nointera ting gases it automati ally gives F of the form

r
X
F (T, V, n1 , . . . , nr ) = Fi (T, V, ni ) .
i=1

We know (we already should!), that the fun tion F (T, V, n1 , . . . , nr ) ontains information
about all termodynami al hara teristi s of the system.

One more onseqen e of the Gibbsa postulate is the entropy of mixing, whi h is
easily identied when the formula (157):

r T
dT ′ (i) ′
 
vi
X Z
S(T, V, n1 , . . . , nr ) = ni si (T0 , v0 ) + ′
cv (T ) + R ln ,
i=1 T0 T v0

for the entropy of the mixture (molar volumes v0 of all perfe t gases taken at the same
referen e temperature and pressure are equal) is expli itly expressed (using the relations
vi = V /ni ) through the total volume V o upied by the mixture and the total mole
number n:

T
dT ′ mix ′ V
Z
mix
S(T, V, n, x1 . . . , xr ) = n s (T0 , v0 ) + n ′
cv (T ) + nR ln
T0 T nv0
r
X
+n (−R xi ln xi ) . (158)
i=1

smix(T0 , v0 ) =
P
(We have introdu ed here i xi si (T0 , v0 ).) The last term is the mixing
entropy whi h we will also all the Cinderella's (Kop iuszek) entropy. It is positive be ause
xi ≤ 1. That this term is related to the mixing of dierent gases in one ontainer follows
learly from the omparison of (158) with the entropy of n moles of a one- omponent
perfe t gas whi h (a identally) would have the the same molar heat apa ity and the
same molar entropy s0 as the mixture:

T
dT ′ V
Z
′
S(T, V, n) = ns(T0 , v0 ) + n c v (T ) + nR ln .
T0 T′ nv0

The mixing entropy an be seen in many ways (hopefully they will be dis ussed in lasses).
A one whi h oers an isight into its origin is as follows. Prepare n1 and n2 moles of two
dierent gases in two initially isolated ontainers at the same pressure and the same tem-
perature. If the two ontainers are joined together, ea h of the gasses expands freely into
the additional volume (undergoing essentially the Joule pro ess). The resulting entropy
hange is just the mixing entropy. It should be however noti ed, that if the two gases
were identi al, one woud say that being at the same temperature and pressure they are in
equilibrium and when the two ontainers are joined, nothing happens - the total entropy
does not in rease. This shows that the mixing entropy results from dierent treatement

105
of same and dierent gases and this pro edure nds its justi ation only in statisti al
physi s and in fa t in the quantum me hani al indistinguishability of identi al parti les.

Sin e the Gibbs postulate allows to expli itly onstru t all thermodynami al fun tions
of the mixture of perfe t gases, also the hemi al potentials an be obtained. It is a matter
of a simple al ulation to nd that

µi (T, p, x1 , . . . , xr ) = µ(T, xi p) ≡ µ(T, pi ) = µ(T, p) + RT ln xi , (159)

where µ(T, p) is the hemi al potential of a one- omponent perfe t gas at temperature T
and pressure p.
Another situation in whi h some approximate formulae for hemi al potentials of the
two omponents an be obtained without additional experimental (or statisti al physi s)
input is the one in whi h the number of moles of one of the two omponents is mu h
smaller than that of the other one. Let x = n2 /(n1 + n2 ) ≪ n1 /(n1 + n2 ) = 1 − x. The
molar internal energy and the molar volume of su h a weak solution, whi h an be treated
as fun tions od T, p and x, an be then expanded in the Taylor series:

u(T, p, x) = u0 (T, p) + x ∆u(T, p) + . . . ,

v(T, p, x) = v0 (T, p) + x ∆v(T, p) + . . . ,

in whi h u0 and v0 are the molar internal energy and volume of the pure solvent (the other
omponent is alled the solute). The dierential of the molar entropy (taken at onstant
x) an be then organized as follows
   
1 p 1 p
ds(T, p, x) = du0(T, p) + dv0 (T, p) + x d∆u(T, p) + d∆v0 (T, p) .
T T T T

As a whole, the right hand side must be an exa t dierential (as the dierental of the
molar entropy of the omplete solution). Moreover, the rst bra ket must also be an
exa t dierential of the molar entropy of the pure solvent. It follows, that also the se ond
bra ket must have this property and the one-form ds an be integrated up yielding

s(T, p, x) = s0 (T, p) + x ∆s(T, p) + f (x) ,

where f (x) is an integration onstant (whi h must, therefore, be independent of T and p).
One then invokes the Plan k argument that at su iently high temperature all substan es
turn into gases, and ultimately (at low pressures) behave as perfe t gases. This means
that the integration onstant f (x) must be (as the only term whi h is independent of T
and p) the mixing entropy of the mixture of two perfe t gases:

f (x) = −R(1 − x) ln(1 − x) − Rx ln x .

This allows to onstru t the molar Gibbs fun tion of the solution:

g(T, p, x) = u − T s + p v = g0 (T, p) + x∆g(T, p) + RT {(1 − x) ln(1 − x) + x ln x} ,

106
in whi hg0 = u0 −T s0 +pv0 and ∆g0 = ∆u0 −T ∆s0 +p∆v0 . The extensive Gibbs fun tion
of n = n1 + n2 moles of the solution is now onstru ted using the standard pres ription
(nx = n2 , n(1 − x) = n1 )

G(T, p, n1 , n2 ) = (n1 + n2 ) g(T, p, x)

n1 n2
= (n1 + n2 ) g0 (T, p) + n2 ∆g(T, p) + RT n1 ln + RT n2 ln .
n1 + n2 n1 + n2
The hemi al potentials an be then obtained as the derivatives

 
∂G
µ1 = = g0 (T, p) + RT ln(1 − x) ≈ g0 (T, p) − RT x ,
∂n1 T,p,n2
 
∂G
µ2 = = g0 (T, p) + ∆g(T, p) + RT ln x ≡ ψ(T, p) + RT ln x . (160)
∂n2 T,p,n1

107
 LECTURE IX (TMD)

Phase transitions
By phases of a substan e one understands dierent (spatially distinguished) forms in
whi h it may exist. Dierent phases of the same substan e have dierent physi al prop-
erties. In some range of intensive parameters a given phase is stable in another it may
be ome metastable and in yet some other ranges it simply annot exist. If in ertain on-
ditions hara terized by the values of temperature and pressure (and of other intensive
parameters, if the substan e is not simple) there may exist as stable more than one phase,
one speaks of the oexisten e of phases. It is in these onditions that phase transitions
68
normally o ur. Phases of a given substan e should not be onfused with its physi al
states (solid, liquid and gaseous): a given solid an have dierent phases ( alled allotropi
modi ations) in whi h atoms are dierently ordered in the rystalline latti e ells: for
example, sili on Si has 3 dierent allotropi phases, tin Sn two (white tetraghonal and
grey whi h is amorphous), H2 O has as mu h as 12 dierent rystalline modi ations ea h
of whi h is a separate i e phase! There are also known dierent mixtures of liquids (dif-
ferent phases of the mixture): one phase ri h in one omponent and another phase ri h
in another omponent.
If in given onditions a substan e may exist in more than one phase, the system is
usually inhomogeneous - the phases exist in spatially separated forms. They are then
treated as dierent parts of a ompound system separated by ( titious) walls allowing
69
for transfer of matter, energy and volume between the parts.
The onditions in whi h dierent phases of a simple one- omponent substan e an
oexist were already dis ussed at the end of Le ture VII: independently of whether the
system was isolated or open (intera ting in this or another way with its surrounding),
the oexisten e of two phases α and β of su h a substan e always required the equality
of the temperatures (thermal equilibrium with respe t to ex hanging energy), pressures
(me hani al equilibrium with respe t to ex hnging volume) and the equality of the hem-
i al potentials of the two phases. This extends to multi omponent systems, as well, the
only hange being that in su h ases the hemi al potentials hara terizing the same
omponent of both phases must be pairwise equal:
(α) (α) (α) (β) (β) (β)
µi (T, p, x1 , . . . , xr−1 ) = µi (T, p, x1 , . . . , xr−1 ) , i = 1, . . . , r. (161)

The dieren e between open and isolated systems is that in the former ase the equilibrium
of phases an be neutral (the rst example onsidered in Le ture VIII), while in the latter
ase it an uniquely determine the partition of matter omponents between the phases.

68 A phase whi h in a given range of the parameters is metastable may exist for a very time but
eventually turs out into the stable (in this range) phase. This means that the hange of one phase into
another one an also o ur in onditions in whi h only a single phase is stable.
69 The boundary separating the phases, although negle ted in further dis ussion here, has in fa t some
thi kness; it an be treated as onsisting of some quantity of matter, and as ribed entropy, energy and
other thermodynami fun tions. In this way it an be in luded into the analysis as yet another part of
the system. Its properties (e.g. the surfa e tension) an also lead to modi ations of the equilibrium
ondition; these an be also modied in presen e of external elds like the gravitational one.

108
 The questions how many phases of a given substan e an simultaneously be in equi-
librium nds the answer in the Gibbs phase rule. Let r be the number of omponents
(labeled by the index i) of a simple substan e whi h an exist in f phases (labeled by the
supers ript α).
If ea h phase is an isolated system, it is hara terized by 2 + r extensive
(α) (α) (α) (α)
parameters: its entropy S , its volume V and the numbers of moles n1 , . . . , nr of
(α)
the omponents. The rst two variables an be traded for the temperature T and pres-
(α)
sure p of ea h phase. In all, f isolated phases are hara terized therefore by (2 + r)·f
independent parameters. If the phases are to be in equilibrium, their intensive parameters
must satisfy a number of onditions. Firstly, they must have all the same temperatures
(thermal equilibrium)

T (1) = T (2) = . . . = T (f ) ≡ T ,

whi h gives f −1 onditions. Another f −1 onditions follow from the equality of their
pressures (me hani al equilibrium):

p(1) = p(2) = . . . = p(f ) ≡ p .

Equalities of the hemi al potentials ( hemi al equilibrium with respe t to matter ex-
hange)

(1) (2) (f )
µ1 = µ1 = . . . = µ1 ≡ µ1 ,
.........
µr = µr = . . . = µr(f ) ≡ µr ,
(1) (2)

give together (f − 1)·r further onditions. Finally, there are f onditions

r r
(1) (f )
X X
xi = 1, ... xi = 1,
i=1 i=1

folowing from the fa t that the equilibrium of pahases depends only on values of the
intensive parameters whi h in turn do not depend on all extensive parameters but only
on their ratios (see Le ture VIII). This is taken into a ount by expressing them in terms
of the molar fra tions xi whi h in ea h phase sum up to unity. Therefore there remains
the freedom of varying

k = (2 + r)·f − [(2 + r)·(f − 1) + f ] = 2 + r − f , (162)

70
independent parameters. Of ourse, the oexisten e of phases is possible only if

k ≡ 2 + r − f ≥ 0.
70 Another way of arriving at the same result is as follows: the intensive parameters whi h are at play
(α)
are: T and p f ·(r − 1) independent molar fra tions xi in all
whi h must be ommon to all phases and
f phases; the equalities of the hemi al potentials onstitute r·(f − 1) onditions whi h must be satised
if f phases oexist in equilibrium. This again gives 2 + r − f free parameters.

109
Thus two phases (f = 2) of a single omponent (r = 1) substan e (like pure H2 O) an
oexist along a urve be ause k = 2 + 1 − 2 = 1 whi h means that one parameter an be
varied freely, while three phases (f = 3) of a single- omponent substan e an oexist only
at an isolated point (isolated points). If there are more omponents, more phases an
oexist at one point, more along a urve et . If hemi al rea tions are allowed to o ur
between the omponents of the system, ea h possible rea tion imposes one ondition on
the molar fra tions and the number of degrees of freedom de reases by one per ea h
rea tion.
Below we will dis uss only phase transitions of one- omponent simple substan es.
Suppose a quantity (n moles) of su h a substan e is in a state ( hara terized by its internal
temperature T and pressure p) in whi h only one of its phases is stable. If this state is
hanged (by ompressing/expanding heating/ ooling the system) so that its temperature
T and p assume values at whi h two (or three) phases an exist, the proportion of the
substan e in the phases will hange - the phase transition will o ur.

Classi ation of phase transitions

Two phases an oexist only if their hemi al potentials are equal. Thus the hemi al po-
tentials are ontinuos along the oexisten e lines. P. Ehrenfest lassied phase transitions
a ording to the ontinuity of derivatives of hemi al potentials: In rst order transitions
ontinuous are hemi al potentials but their rst derivatives (at least one of them) - the
molar entropies and/or molar volumes are dis ontinuous. In se ond order phase transi-
tions the hemi al potentials and their rst derivatives, s and v , are ontinuos but at least
one of the derivatives of s or v, that is, c p , αp or kT , is dis ontinuous, and so on.
Of the transitions whi h t this lassi ation only the rst order ones are ommon.
Se ond order transitions are very rare - the main example being the transition ondu tor-
super ondu tor in zero external magneti eld H. There are, however, many transitions
whi h do not t the Ehrenfest lassi ation be ause derivatives of the hemi al potentials
are divergent at the transition: for instan e many rst-order phase transitions end up at
the riti al point at whi h derivatives be ome divergent (so it is hard to de ide whether
they are ontinuous or not). For this reason Landau simplied the lassi ation: one now
distinguishes only the rst-order transitions, while all others are alled ontinuous (on
a ount of the ontinuity of the rst derivatives of the hemi al potentials).

First-order phase transitions.

71
We shall now dis uss in general terms a typi al phase diagram of a simple, single ompo-
nent substan e whi h when onditions are varied, undergoes rst-order phase transitions
(perhaps with the ex eption of an isolated point(s)). The diagram shows regions on the
(T, p) plane of stable phases. Phase diagrams of multi omponent substan es an be on-
siderably more ompli ated: apart from T and p also molar fra tions xi enter the game
and one an only display proje tions of stability regions onto various two-dimensional
planes.
We onsider a simple substan e whi h in the onsidered ranges of the variables T
and p an exist in three modi ations (phases): solid, liquid and vapour. Ea h of the

71 Atypi al phase diagrams will be dis usses later.

110
 p p
S a) S b)

solid solid

liquid liquid

B cr B cr
C tr C tr

A vapour A vapour
O O
T T

Figure 12: a) S hemati view of a typi al phase diagram of a simple one- omponent
substan e whi h in the shown ranges of the temperature and pressure an exist in three
phases. b) S hemati view of a slightly less typi al phase diagram (H2 O, Bi).

modi ations is hara terized by its molar Gibbs fun tion g(T, p), that is, its hemi al
potential µ(T, p) whi h an be imagined to exist over the whole (T, p) plane. Thus there
are three g -surfa es orresponding to the three phases whi h pairwise interse t along some
lines (along whi h the two phases an oexist in equilibrium) and there is usually (but
not always) one point, alled the triple point, at whi h these three interse tions meet - at
this point ( f. the Gibbs phase rule) all the three phases an oexist. The proje tions of
these urves onto the (T, p) plane give the phase diagram of the type shown in Figures 12
a and b.
As follows from the dis ussion arried out at the end of Le ture VII, in a given region of
the (T, p) plane stable is that phase whi h has the lowest value of the molar Gibbs fun tion
(be ause all the substan e put in that phase gives the lowest value of the availability A).
To determine the regions of stability of dierent phases one an onsider the variation of
these fun tions with T and p whi h readily follow from the formulae
   
∂g ∂g
dg ≡ dµ = −s dT + v dp , = −s , = v.
∂T p ∂p T

Thus in the dire tion of in reasing pressure the g -surfa es (µ-surfa es) always slope up-
wards (v > 0): the g -surfa es orresponding to the solid and liquid phases omparatively
slowly (small v ), at nearly onstant rate (small isothermal ompressibility) and that of
the vapour phase fast (mu h larger v ) with the slope be oming steeper ( onsiderably
larger ompressibility) in the dire tion of de reasing pressure. If the arbitrary onstant
72
fa tor in the molar entropy of the substan e is xed so that s(T, p) ≥ 0, the g -surfa es
slope downwards (negative slope) at in reasing rate (cp > 0) in the dire tion of in reasing
temperature.
From the analysis of the slopes of the g -surfa es in the pressure dire tion it follows
that in the region of low pressures (just above the T -axis) stable is the vapour phase
(whose Gibbs fun tion falls most steeply as the pressure goes down), but as far as the

72 Of ourse, for ea h substan e there is only one arbitrary onstant entropy s .

0

111
region near the p axis is on erned, one has to rely on experimental observations to
identify the stable phase orre tly (3 He is an example that the liquid phase an be -
in some range of pressure - loser to the p axis than the solid phase). In the phase
diagram shown in Figure 12a rossing the line tr-S verti ally upwards one moves from
the region in whi h liquid is the stable phase to the region in whi h stable is the solid
(solid)
phase; this means that v < v (liquid) : the substan e expands on melting. Figure 12b
(solid)
shows the opposite situation in whi h v > v (liquid) : here the substan e (water is here
the most obvious example) expands on solidifying. When a oexisten e line in Figure
12a or 12b is rossed horizontally (at onstant pressure) the phase whi h is stable on the
right hand side must have larger entropy, whi h means that in Figure 12 the transitions
solid→liquid, solid→vapour or liquid→vapour heat is absorbed. The nite dieren e of
entropies of the adja ent phases means that if the system is in the lower entropy phase
and heat is supplied to it at a onstant rate (at onstant pressure, say), its temperature
rises until the oexisten e line is rea hed; then there is a halting of the temperature rise:
the new phase appears and the amount of substan e in it grows. Only when the phase
transition is a omplished, there is only the new phase, does temperature begin to rise
again (the same is observed when the heat is supplied and the volume of the system is
simultaneously hanged in su h a way that the oexisten e line is rossed not exa tly in
the verti al dire tion). The halting of the temperature grow is the distinguishing feature
of rst order phase transitions. The heat absorbed by the system in the transition is
alled the latent heat.
Even if in ea h of the regions (with the ex eption of the oexisten e urves) only one
phase is stable, it makes sense to onsider the g -surfa es as extending beyond the regions
of stability of the orresponding phases be ause from the experien e it is known that
the phases (parti ularily if the substan es are highly puried) an exist as metastable
states in the regions in whi h they should have already turned into another phase. Super-
ooled vapours an stay un ondensed at pressures signi antly higher than the equilibrium
vapour pressure (by whi h term one understands the pressure at the vapour-liquid oex-
isten e urve at a given temperature) - the histori devi e of parti le physi s, the Wilson
loud hamber, owed its operation to this phenomenon), puried liquids an be super-
o
ooled without solidifying - droplets of water in a loud may be ooled down to −40 C.
And onversely, liquids an be superheated - another histori devi e of parti le physi s,
the buble hamber exploited this possibility. It should be however kept in mind that the
analogous ontinuation of the g -surfa es is not ne essarily justied in the ase of higher
order transitions. Finally it should be mentioned that also the oexisten e lines an be
prolonged into the regions in whi h none of the two oexisting along it phases is stable
as indi ated by the dashed lines in Figure 12 - oexisten e of super ooled liquid with its
vapour is well known to meteorology (though it seems a solid-liquid oexisten e in the
vapour region has never been observed).

The Clapeyron-Clausius equation

The oexisten e urve of two phases of a simple single- omponent substan e is determined
by the equality of the two molar Gibbs fun tions ( hemi al potentials) and if these were

112
known pre isely, the form of the oexisten e urve would also be known. But sin e the
hemi al potentials as fun tions of T and p are usually not given, the Clapeyron-Clausius
equation allows to re onstru t the oexisten e urve on the basis of the experimental
data. It is derived straightforwardly by taking the total derivative with respe t to T of
the equality

µ(α) (T, p(T )) = µ(β) (T, p(T )) .

Another way of deriving this equation (better adapted to higher order transitions) onsists
of expanding in the Taylor series both sides of the equality

µ(α) (T + ∆T, p + ∆p) = µ(β) (T + ∆T, p + ∆p) .

Either way, if at least one of the two derivatives of the hemi al potential is dis ontinuous
one obtains

dp s(β) − s(α) qα→β

= (β) = , (163)
dT v − v (α) T (v (β) − v (α) )

where T (s(β) − s(α) ) has been identied with the latent heat qα→β of the transition. If
all quantities on the right-hand side of this equation are given as fun tions of T and p,
the equation an be integrated. One usually makes some rude approximations: eg. in
onsidering the solid-vapour transition the solid molar volume is negle ted, the vapour
is treated as a perfe t gas, qα→β is assumed to be onstant et . (This will be done in
lasses.)
The Clapeyron-Clausius (or, more familarly, Clapau ius) equation learly shows the
dis ussed already orrelation of the sign of the slope of the melting urve with the dier-
en e of the molar volumes (or densities ρ = mmol /v ) of the solid and liquid phases. Water
molar volume is smaller than that of i e (whi h has also lower entropy - the latent heat
qα→β is dened to be positive, so α = i e, β = water) whi h results in the negative slope
(negative right hand side of the equation (163)) of the i e melting urve (whi h is of the
type shown in Figure 12b). That the slope of the i e melting urve annot be innite (the
urve annot be verti al, that is, the melting temperature of i e annot be independent
of pressure, given that i e has larger molar volume than water) was understood by the
se ond Thomson, the brother of the lord Kelvin, around the middle of the XIX entury
to be required by 2TMDL. If it were verti al, one ould devise an i e-engine, shown in
gure ??, whi h would violate the Kelvin's 2TMDL: starting with a quantity of water kept
o
at 0 C one ould pla e a weight m on the piston losing the vessel with the water from
above (the water ompressibility is small and an be negle ted here). Then a quantity
Q̄ ould be extra ted from the water to a heat reservoir at 0o C, as a result of whi h the
water would solidify in reasing its volume (maintaining its initial temperature, the heat
extra ted from it being the latent heat) and raising thereby the weight to some height
h > 0; the weight ould be then moved to the side and brought down to its initial height
delivering some me hani al work (at the ost of its gained potential energy), while an
o
appropriate quantity Q of heat ould be supplied (again from the reservoir at 0 C) to the

113
i e ausing it all to melt; in this way the working substan e - the water - would be in its
initial state and the work W̄ = mgh would be done at the ost of the heat Q − Q̄ taken
entirely from the single reservoir at 0o C. This is something the Kelvin's 2TMDL forbids.
The resolution is of ourse that in the proposed y le the pro esses of solidifying water and
of i e melting o ur at dierent pressures - solidifying at p = p0 + mg/A, while melting at
p = p0 , where p0 is the pressure of the surrounding and A is the se tion area of the piston
- and 2TMDL is saved if the melting/solidifying temperature varies appropriately with
the pressure. The omparison of the measured pressure variation of the i e melting point
with that predi ted by the Clapau ius equation was one of the rst su esful appli ations
of thermodynami s to physi al problems and lend a great onden e to this developing
bran h of theoreti al physi s.

Criti al point
The line of oexisten e of the liquid and vapour phases (the line tr- r on the diagrams
12a,b) does not ontinue idenitely: as the line is followed to the right, the quantitative
hara teristi s (densities, heat apa ities, ompressibilities, et .) of the vapour and liquid
liquid phases be ome more and more similar to eventually disappear at the riti al point
marked  r (and hara terized by Tcr and pcr ) where the liquid and vapour be ome
indistinguishable. To dis uss the nature of the riti al point it is onvenient to plot
several isotherms of the vapour-liquid system on the (v, p) indi ator diagram.
If by de reasing its volume the vapour at T < Tcr is ompressed isothermally (one
moves verti ally upwards on the phase diagrams 12a,b) the isothermal ompressibility
kT = −(∂ ln V /∂p)T rst (far below the oexisten e line) de reases and then rises to
be ome innite at the oexisten e line, when the liquid begins to form; the system be omes
there inhomogeneous breaking up into two separate phases; subsequent de reasing the
volume does not in rease pressure (kT stays innite), only in reases the quantity of liquid
as ompared to the quantity of the vapour (on the diagram 12 one is staying all this
time on the oexisten e urve); when the vapour has dissapeared ompletely, the pressure
begins to rise rather steeply (kT falls abruptly to a small value) and one moves on the
diagram 12 upwards already above the oexisten e line.
When the analogous ompression of the vapour is ee ted exa tly at the riti al tem-
perature Tcr , the ompressibility kT rst de reases as the volume is redu ed (and pressure
rises) then rises to be ome momentarily innite at pcr and then steadily goes down. At
no point is there a separation of the system into two phases is observed. Above Tcr as
the volume is redu ed, the ompressibility de rases monotoni ally: at low values of the
pressure (large volume) it is large so the substan e resembles more the vapour, while at
high pressures (smaller volumes) it is small, so the system behaves more as a liquid.
It follows that it is possible by traveling around the point  r on the diagram 12
to make the system to pass from the vapour phase to the liquid phase, without any
73
dis ontinuous hanges in its properties. This means that the two g -surfa es: the gl (T, p)
73 It is appropriate to make a omment on the terminology: in the older literature the system at
temperatures T < Tcr is alled either liquid (at higher pressures) or vapour (at lower pressures) while the
system above Tcr is alled gas. More adequate seems to all the system above Tcr super riti al uid and

114
one and the gv (T, p) one whi h we ompared in establishing whi h of the phases is stable
in the given domain of the (T, p) plane, are in fa t to parts of one and the same g -surfa e
whi h simply interse ts itself for T < Tcr along a line whose proje tion onto the (T, p)
plane is the urve tr- r, but not for T > Tcr .
One way of visualizing the form of su h a self-interse ting g -surfa es is to use the Van
der Waals equation of state
 a
p + 2 (v − b) = RT , (164)
v
whi h qualitatively (though not very well quantitatively) models the equation of state of
real vapour-liquid systems. High T isotherms given by this equation are perfe tly mono-
toni and the relation of the pressure p to the molar volume is one-to-one, just as happens
with the T > Tcr isotherms of a real vapour-liquid system. Below some temperature,
whi h should therefore be identied with Tcr , to a given pressure p there orrespond three
dierent values (of whi h the smallest one and the largest one we, introdu ing their in-
terpretation, denote vl andvv , respe tively) of the molar volume. It an be also seen that
the inverse ompressibility kT−1 , whi h an be al ulated, given the equation of state,
 
−1 ∂p RT v 2a
kT = −v = 2
− 2,
∂v T (v − b) v

is always positive, if T > Tcr , but for T < Tcr it be omes negative somewhere between
vl and vv (therefore the ompressibility kT is there negative too). This means that the
intrinsi stability ondition (dis ussed in Le ture VII) is violated there and one is for ed
to admit that those parts of sub riti al VdW isotherms do not represent system's states
(are unphysi al). In agreement with what happens in real systems some parts of the
sub riti al VdW isotherms should be therefore repla ed by at horizontal lines, be ause
one knows that as the oexisten e urve is rea hed on the diagram 12 from below (from
above), the system breaks up into two phases and its number of moles n and the molar
volume v are given by

n = xl nl + xv nv , v = xl vl + xv vv xl + xv = 1 ,

with xv beginning at 1 (at 0) and de reasing to 0 (in reasing to 1) as the total volume is
redu ed (in reased).
The possible interpretation of the sub riti al VdW isotherms is therefore as follows
(Figure to be done - the position of the horizontal line will be dis ussed below):

• the segment at large v going through the point A up to the point B orrespond to
the vapour (gas);

• point B is where the phase separation should begin (it orresponds to the rea hing
the line tr- r on the diagram 12 from below);

the one below Tcr vapour or, ex hengeably, gas.

115
 • the segment B-C an be interpreted as representing the super ooled vapour (a
metastable state) - on Figure 12 the orresponding points lie above the tr- r line;

• the segment C-D-E is unphysi al (kT < 0);

• the segment E-F an be interpreted as representing the superheated liquid (a metastable
state) - on Figure 12 the orresponding points lie below the tr- r line;

• the at segment B-D-F represents the mixture of phases in neutral equilibrium

• the segment from the point F through G and further represents the system in the
single liquid phase

Having the VdW equation one an onsider integrating up the one-form

dg = −s(T, p) dT + v(T, p) dp ,
74
along the shown isotherm T < Tcr from the point, say A, on:
Z p
g(T, p) = g(T, pA ) + dp′ v(T, p′ ) . (165)
pA

As we integrate, the area under the urve v = v(p) (at xed T ) grows until the point C is
rea hed, so the plot of the fun tion g(T, p) raises. Then from C to E, as p de reases, the
area must be ounted as negative; therefore the plot of g(T, p) goes ba kwards and falls
down. Finally, from the point E on p again in reases and the area under the urve v = v(p)
must be ounted as positive, so the plot of g(T, p) again goes forward and raises. The
resulting self rossing urve g(T, p) (a T = onst se tion of the self- rossing g -surfa e) is
shown in Figure. Its part A-C is the onstant T se tion of what formerly was supposed to
be the vapour g -surfa e, while its part E-G is the se tion of the liquid g -surfa e; up to the
point at whi h they ross the vapour g -surfa e is lower and it is vapour whi h is the stable
phase, while from this point on stable is the liquid phase. As the temperature in reases,
the unphysi al segment C-E be omes shorter and shorter and disappears ompletely at
Tcr ; making plots of several su h self rossing g - urves one obtains the pi ture of the self
rossing g -surfa e of the liquid vapour system.
There remains only the question where to pla e the points B and F on a given isotherm?
The pres ription given by Maxwell (and alled the Maxwell onstru tion ever sin e) states
that these points should be pla ed in su h a way that the areas B-C-D-B and D-E-F-D
should be equal. In this way the value of the molar Gibbs fun tion at the point E (that
is of the hemi al potential of the vapour in equilibrium with its liquid at this value of
T) will be the same as the value of the molar Gibbs fun tion at the point F. Indeed, the
Maxwell rule requires that (geometri ally)

(area under B − C) − (area under C − D) = (area under D − E) − (area under E − F)

74 Of ourse the parametrization of the isotherm by the pressure is only pie ewise well dened and has
to be hanged on the way - in the dis ussion below this is taken into a ount in the s hool way, by
ounting the areas under the urve as positive or negative.

116
But this pre isely ensures that (re all how the integration (165) has been done)
Z pF
g(T, pF ) − g(T, pB ) = dp v(T, p) (166)
pB
= (area under B − C) − (area under C − D)
−(area under D − E) + (area under E − F) = 0 .

The onstru tion is however of doubtful validity for it learly exploits the unphysi al part
of the isotherm (segment C-E). The orre t pro edure would be to start from the point A
and move on along the isotherm omputing at every point the hemi al potential (whi h
does not involve the unphysi al part of the isotherm) and at the same time to start form
the point G and move ba k along the isotherm again omputing at every point the hemi al
potential. In this way one an nd the two points, one on the left part of the isotherm
and the other on the its right part, at whi h the two potentials have the same value. It is
these points, therefore, whi h should be onne ted by the at segment. It may be argued
that this shows that the Maxwell onstru tion annot be orre t be ause if the isotherm
between the points B and F is distorted, an operation whi h would not alter the equality
of the hemi al potentials at B and F found in the way proposed here, the two areas
ompared in the Maxwell onstru tion would be modied and it would predi t in orre tly
the position of the at part of the isotherm. This argument an be, however, dismissed
by appealing to the analyti ity property of the thermodynami al fun tions (everywhere
ex ept for the bifur ation point): the stiness of analyti fun tions does not allow for
distorting the unphysi al parts of the isotherm in a way un orrelated with the hanges of
the hemi al potentials found by the move and ontrol method.

The ondu tor-super ondu tor phase transition

We will briey onsider here the ondu tor-super ondu tor phase transition with the aim
of showing that in the zero external magneti eld it is of the se ond order a ording to
the Ehrenfest lassi ation. Many metals (whi h do not exhbit ferromagneti properties)
like tin (Sn), irydium (Ir), niobium (Nb), be ome super ondu ting at low temperatures
(3.73 K, 0.14 K and 9 K, in the ase of tin, irydium and nobium in zero magneti eld,
respe tively - we will not onsider hight temperature super ondu tors here), that is exhibit
zero ele tri resistan e. This is a ompanied, and one an show that the two ee ts
must ne essarily ome together, by their perfe t diamagnetism: if they are pla ed in
an external magneti eld, the magneti eld indu tion is expelled from their interior
altogether (i.e. B=0 inside the spe imen; this is alled the Meissner-O hsenfeld ee t)
unless the strength of the external eld is too strong: if it ex eeds some riti al value
Hc (whi h depends on the temperature, and to a mu h smaller extent, on the pressure),
the material be omes normal: the resistivity reappears and the magneti eld indu tion
B in the material is related to the strenght H inside by some magneti suss eptibility
χ: B = H + 4πM where M = χH. Below we will assume that the normal phase
suss eptibility χ is negligibly small, so that in the normal state simply B = H inside the
material (just as outside it).
Suppose a tin spe imen of a long ylindri al shape is pla ed in a weak external magneti

117
 75
eld H at a temperature lower than the riti al one, Tc (so that the spe imen is in
its super ondu ting phase). The magneti eld indu tion inside the spe imen is then
zero. If H is slowly in reased so that it ex eeds some riti al value, the metal undergoes
a rst order (as will be seen) phase transition to the normal state in whi h B = H
inside it. If the external eld is redu ed ba k below the riti al value Hc , the material
reverts to the perfe tly diamagneti state. The transition is therefore reversible in the
thermodynami sense and an be analyzed with standard methods. The dependen e of the
riti al eld strength Hc on the temperature (at xed pressure) is onveniently modeled
by the empiri al formula
"  2 #
T
Hc (T ) = H0 1 − , (167)
Tc
whi h orre tly aptures the two essential features of the real dependen e: its atness at
T = 0 (required by 3TMDL - of whi h we had no time to talk) and the nite slope at
H = Hc (whi h is ru ial for further onsiderations). The dependen e on the pressure is
rather weak: Tc of tin hanges from 3.73 K at p = 1 atm to 3.63 K at p = 1700 atm.
It is onvenient to onsider the molar magneti Gibbs fun tion of the metal whi h
exhibits super ondu tivity

g ′ = u′ − T s + pv − HMv , (168)

(we assume that the the magneti eld strenght H and the magnetization M are onstant
′
in the material) whose dierential dg = −sdT + vdp − vMdH implies that

∂g ′ ∂g ′ ∂g ′
     
= −s , = v, = −vM ,
∂T p,H ∂p T,H ∂H T,p
and whi h, being a fun tion of intensive parameters only, plays the role of the hemi al
potential.
Sin e the system is hara terized by three intensive parameters, its two phases, the
normal one denoted n and the super ondu ting one denoted s, in the three dimensional
spa e an oexist along a two dimensional surfa e at whi h their molar gibbs fun tions are
equal gs′ = gn′ . The oexisten e surfa e proje ted onto the three planes: (T, H), (p, H) and
the (T, p) one, gives rise to three oexisten e urves whi h are determined by the equations
76
analogous to the Clapau ius one - the method of derivation is exa tly the same:

sn − ss
 
∂Hc
=− ,
∂T p vn Mn − vs Ms
vn − vs
 
∂Hc
= , (169)
∂p T vn Mn − vs Ms
sn − ss
 
∂p
= .
∂T H vn − vs
75 Sin e we onsider the magneti eld aligned always in the same dire tion, we an play with the s alar
quantity H.
76 E.g. if the pressure is xed, on the oexisten e urve the hanges dHc and dT are orrelated so that
gn′ (T + dT, Hc + dHc , p) = gs′ (T + dT, Hc + dHc , p).

118
As in the super ondu tling (perfe tly diamagneti ) phase B=0 Ms = −H,
means that
while the equality B=H holding in the normal phase implies that in this phase Mn = 0,
the rst two of the above three equations an be rewritten in the forms
   
∂Hc 4π ∂Hc 4π
=− (sn − ss ) , = (vn − vs ) . (170)
∂T p vs Hc ∂p T vs Hc

The values ss , sn , et . should be taken on the oexisten e urve, but their variation
with the magneti eld is not large: Sin e, as readily follows from (168), (∂s/∂H)T,p =
(∂vM/∂T )p,H , neither sn nor ss are appre iably eld-dependent: the former be ause
Mn = 0 and the latter be ause Ms = −H/4π , so only a rather small temperature
dependen e of vs an ontribute to the variation of ss with the temperature. Similarly,
sin e (∂v/∂H)T,p = −(∂vM/∂p)T,H , vn is eld independent and to the eld dependen e
of vs an ontribute only the small ompressibility of the spe imen n this phase. Thus,
with the ex eption of the dieren e vn − vs one an use in the above equations ss and vs
taken at the zero magneti eld and similarly negle t the eld dependen e of sn and vn .
Now, the rst of the equations (170) rewritten in the form

vs Hc vs Hc
   
∂Hc ∂Hc
sn − ss = − , vn − vs = , (171)
4π ∂T p 4π ∂p T

show, be ause as stressed and as the formula (167) makes it lear, the slope (∂Hc /∂T )p
is not innite at Hc = 0, that at Hc = 0 the dieren e sn − ss vanishes. This shows that
in zero magneti eld the transition ondu tor-super ondu tor o urs without any latent
heat. In the same way the se ond equation shows that in zero magneti eld vanishes also
the dieren e vn −vs . Thus is in the ordinary two-parameter spa e (T, p) orresponding to
zero magneti eld, the ondu tor-super ondu tor transition esses to be rst order. This
77
means that the third of the equations (169) is indeterminate at zero eld. To show that
the transision is of se ond order a ording to the Ehrenfest lassi ation, it is su ient to
show that the derivatives of entropies and/or of molar volumes are dis ontinuous a ross
the oexisten e urve. Dierentiating the rst equation (171) with respe t to T at onstant
pressure (again negle ting the temperature dependen e of vs in their right hand sides) one
nds
"  2 #
∂ 2 Hc2 ∂ 2 Hc
   
vs T vs T ∂Hc
c(n)
p − c(s)
p =− =− Hc + . (172)
8π ∂T 2 p 4π ∂T 2 p ∂T p

Appealing to the empiri al formula (167) one sees that the dis ontinuity of the molar heat
apa ities is nite at zero magneti eld and is equal

vs H02
c(n) (s)
p − cp = − .
πTc
77 The transition remains rst order in all two-parameter spa es (T, p) orresponding to non-zero mag-
neti eld and on these planes the third equation (170) is well dened and does determine the oexisten e
urve.

119
Sin e the dis ontinuity of molar heat apa ities of the two phases is nite, the transition is
of se ond order. It remains to see what repla es the third, indeterminate equation (169) in
determining the ondu tor-super ondu tor oexisten e urve on the (T, p) plane (at zero
magneti eld). To this end one takes the derivatives at H=0 of both equalities (171)
with respe t to T and with respe t to p (again negle ting any small dependen e of vs on
these parameters) obtaining three relations (the fourth one is identi al with one of these
three, be ause of the Maxwell identity (∂s/∂p)T = −(∂v/∂T )p ) whi h an be written in
78
the forms:
 2
∂Hc 4π (n)
cp − c(s)


=− p ,
∂T p vT
 2
∂Hc 
(n) (s)

= −4π kT − kT ,
∂p T
   
∂Hc ∂Hc
= 4π αp(n) − αp(s) .


∂T p ∂p T

Taking their ratios and exploiting the standard sho king relation (∂Hc /∂p)T (∂Hc /∂T )p =
−(∂Tc /∂p)H , one obtains the two (mutually onsistent, as follows from the onstru tion)
equations

  (n) (s) (n) (s)

∂Tc αp − αp kT − kT
= vs T (n) (s)
= (n) (s)
,
∂p H=0 cp − cp αp − αp

whi h repla e the last of the equations (169) in determining the se ond order phase tran-
sition line on the (T, p) plane. The equation obtained here, alled the Ehrenfest equation,
is the analog of the Clapeyron-Clausius equation in the ase of se ond order phase tran-
sitions.

78 Re all that kT = −(1/v)(∂v/∂p)T and αp = (1/v)(∂v/∂T )p .

120
 LECTURE X (STAT)

We have nished the thermodynami part of this Course. Thermodynami s is a phe-

nomenologi al theory whose main aim is establishing relations between measurable quan-
tities hara terizing ma ros opi properties of bodies (systems). It also formulates general
rules allowig to tell whi h pro esses are possible and imposes limits on possible works that
an be extra ted from ma ros opi systems. Results obtained within this theory are very
general and are always true. The entral role in the stru ture of thermodynami s is played
by entropy whi h, however, does not nd a natural interpretation within this theory itself.
Furthermore, making quantitative predi tions on the basis of thermodynami s requires
that ertain fun tions (thermodynami potentials) be known but within the formalism of
pure thermodynami s these fun tions annot be omputed - they have to be laboriously
re onstru ted from experimental data (whi h however in applied and te hni al s ien es is
a good and fruitful method!).
We now turn to statisti al physi s (or statisti al me hani s) whose rst (but not the
only one!) purpose is the derivation of thermodynami s or, put dierently, providing
methods for omputing equilibrium properties of various physi al systems on the basis of
79
more fundamental laws (theories) governing the behaviour of mi ros opi onstituents
of these system (mole ules, atoms or physi al elds). In this approa h entropy will a quire
a on rete interpretation of a measure of disorder or rather of a measure of the la k of
information on the a tual mi ros opi state of the system. Our primary goal will there-
fore be understanding how this la k of information neverheless does not prevent making
denite predi tions for ma ros opi quantities hara terizing a given system and to pro-
vide on rete re ipies for obtaining with the help of statisti al methods thermodynami al
potentials su h as entropy, Helmholtz free energy et . (depending on the onditions in
whi h the system is pla ed) in their natural variables from the mi ros opi theories.
Within the statisti al approa h it is possible to go further and relying on mi ros opi
dynami al theories investigate also systems not in equilibrium, in parti ular study how
the equilibrium is rea hed in various situations. This part of the statisti al method is
alled kineti theory. In fa t the whole statisti al me hani s (also the equilibrium one)
partly grew out from the Boltzmann and Maxwell studies of su h problems within the
kineti therory of gases based on lassi al me hani s. The kineti approa h, essentially
still based on the lassi al Boltzmann equation supplemented only with rea tion rates
omputed within a suitable quantum theory, nds at present numerous appli ations for
example in modern osmology.

Thermodynami s was based on the assumption, whi h in fa t is an idealization, but

a one whi h ma ros opi ally seems possible to be approa hed as lose as it is required,
that by suitable walls a system an be isolated from any inuen es from without. A

79 It should be lear from the beginning that the ultimate laws of Nature are not known yet (will
they be ever known?) but they are largely irrelevant for understanding the behaviour of matter from
the statisti al standpoint: all that is needed is an ee tive theory su iently a urately predi ting the
behaviour of atoms and mole ules.

121
system isolated in this way attains then, after some time (whi h barring some ex eptional
systems is usually not too long on the ma ros opi s ale), the state of equilibrium. Ther-
modynami s applies stri tly speaking only to su h equilibrium states. In ontrast to the
assumption on whi h thermodynami s is based, the most realisti assumption on whi h
to base the statisti al me hani s is (probably) the statement that at the mi ros opi level
no real system an be absolutely isolated with respe t to highly random (un ontrolable)
minute external perturbations. This is parti ularily true of that ma ros opi part of the
system on whi h measurements are being made and whi h for short periods (when the
mesurements are made) an be treated as ma ros opi ally isolated from the rest of the
system.
When we are interested (we measure) those few properties of a ma ros opi system
whi h hara terize its equilibrium state, then the same their values ould be obtained by
measuring them on opies of the system whi h mi ros opi ally are in dierent states than
our system. In other words, typi ally the same ma rostate of the system is realized by a
very large number of its dierent mi rostates and this is true of equilibrium ma rostates
(whi h we onsider in this Course) as well as of nonequilibrium ma rostates. The notion
of the mi rostate depends on whether at the mi ros opi level the system is treated
lassi ally or quantum-me hani ally and will a quire proper meanings in both these ases
in due ourse. The ru ial onsequen e of the impossibility of putting the system in
absolute (from the mi ros opi point of view) isolation is that its mi rostate an never
be known pre isely: even if the system started in a well dened mi rostate it would
immediately make innumerable transitions to dierent mi rostates as a result of being
perturbed by randomly a ting external agents. Thus the most reasonable assumption on
whi h to base the foundations of statisti al me hani s of ma ros opi ally isolated systems
is to assume that all possible mi rostates of su h systems whi h realize the same their
ma rostates are equally probable. This is alled the equal probabilities postulate whi h
an be also given a justi ation on the basis of the information theory.

Foundations of lassi al statisti al me hani s

The mi ros opi onstituents of matter - mole ules, atoms, elds - obey the laws of quan-
80
tum me hani s. For this reason one ould think that the foundations of equilibrium and
nonequilibrium statisti al physi s should be based only on this theory. While this is prin-
ipially true, it is nevertheless worthwhile to onsider statisti al physi s based on lassi al
me hani s. Although some ingredients neessary to fully a ount for the behaviour of real
physi al systems are lost in this way (for example spin is lost, indistinguishability of iden-
ti al parti les an only approximately be a ounted for, phenomena at very low, T → 0,
temperatures annot be aptured orre tly, et .), developing statisti al approa h based
on lassi al me hani s is not ompletely useless. Firstly, as it is loser to our intuition,
it allows to better understand the basi prin iples and rules. Se ondly, it is also useful
in pra ti e: for instan e, it turns out that if properties of gases at temperatures not too

80 Or, at a yet deeper level penetrated to date, by the laws of quantum eld theory whi h is in a sense
(it takes me rougly half of the semester of tea hing to explain this properly) the appli ation of the general
prin iples quantum me hani s to systems of relativisti elds or to systems of relativisti parti les.

122
low are onsidered, the ontributions to the thermodynami potentials of the motions of
enters of masses of the mole ules an always be treated lassi ally (from the quantum
me hani s point of view this motion is quasi- lassi al) and ombined with the ontribu-
tions of the internal motions of these mole ules (rotations and vibrations) whi h (at least
in some ranges of the temperature) must be treated quantum me hani ally.

Thus we begin by onsidering a system onsisting of a xed number N of parti les

(mole ules) whi h obey the laws of lassi al me hani s. A mi rostatate of su h a system
is in prin iple fully spe ied by giving instantaneous values of 3N generalized oordinates
q i and 3N velo ities q̇ i (together they onstitute the omplete set of initial data for the
Euler-Lagrange equations whi h would uniquely determine further evolution of the sys-
tem if it were ompletely isolated). However, as the lassi al statisti al me hani s should
eventually emerge as a limiting ase of quantum statisti al me hani s whi h ne essarily
must be formulated in terms of Hamiltonians and so- alled anoni al variables, it is mu h
more onvenient to spe ify the lassi al system's mi rostate by giving instantaneous val-
i
ues of 3N oordinates q and 3N onjugated momenta pi and to adopt the Hamilton's
equations as the ones whi h would determine the time evolution of an isolated system.
The a tual state of a system of N parti les will be therefore represented by a point in a
i
6N dimensional phase spa e Γ whose 3N axes represent the system's positions q and
the remaining 3N axes represent its momenta pi . As these hange with time, the point
representing the system moves through the phase spa e Γ (irrespe tively of whether the
system is mi ros opi ally isolated or not).
A ma ros opi quantity (an observable) whi h hara terizes the system depends, of
ourse, on its a tual mi ros opi state and an be, therefore, represented by a fun tion
O = O(q, p) of its generalized oordinates and momenta. Be ause of some inertia of
all ma ros opi devi es, experimental measurments made on the system do not really
measure instantaneous values of the observable O but rather its time average of the sort
t0 +τ
1
Z
Oobs = dt O(q, p) , (173)
τ t0

in whi h τ is su iently long on the mi ros opi s ale to smooth out u tuations. Obvi-
ously, the formula (173) stays valid whether the system is ompletely isolated or not. In
prin iple, if the real system were ompletely isolated, the time evolution of its oordinates
q i and momenta pi would be determined by the anoni al Hamilton's equations of motion
∂H (q, p) ∂H (q, p)
q̇ i = , ṗi = − , i = 1, . . . , 3N, (174)
∂pi ∂q i
and the initial onditions q i (t0 ), pi (t0 ) set at some instant t0 . The measured quantity
(173) ould be then in prin iple omputed as

t0 +τ
1
Z
O= dt O(q(t), p(t)) . (175)
τ t0

It is lear however that in the ase of a real system this is not feasible: rstly, although
owing to the growing power of omputers one is at present able to simulate numeri ally the

123
 lassi al time evolution of systems onsisting of, say, N ∼ 104 parti les (su h simulations
are done in modeling e.g the formation of stru tures - galaxies, lusters of galaxies, et .
- in the expanding Universe), simulating the time evolution of systems onsisting of N ∼
NA = 6.022 × 1023 parti les is ertainly beoynd rea h. Se ondly, and this ultimately turns
i
out to be ru ial, one does not know the initial values of the oordinates q and momenta
pi .
Even worse is the situation when the system is not isolated from the mi ros opi point
of view (or even not isolated ma ros opi ally, when e.g. it remains in thermal onta t with
i
a heat bath). In this ase the time evolution of the variables q and pi is not determined
by a Hamiltonian depending only on these variables - to determine the evolution of the
i
variables q and pi of the onsidered system one would have to follow also the evolution
of the variables spe ifying the state of the entire surrounding.
In view of this situation one has to resort to the statisti al method (suggested by
Gibbs) whi h employs statisti al ensembles. By this one means a set of a very large
number N (impli itly the limit N → ∞ is to be taken) of opies of mi ros opi ally
isolated systems whi h are all ma ros opi ally indistinguishable from the onsidered real
system (whi h may be not isolated mi ros opi ally or even ma ros opi ally, when it is in
onta t with e.g. a heat bath) at an initial instant of time - that is, all have their ma ro-
s opi hara teristi s (these are rather easy to spe ify if the onsidered system is in full
equlibrium as an isolated system, or in equilibrium with its environment; slightly more
di ult is the spe i ation of ma ros opi hara teristi s of a system not in equilibrium)
identi al with those of the onsidered real system at some t0 . For example the ensemble
representing at the moment t0 a real system in equlibrium (either ma ros opi ally isolated
or in equilibrium with its environment) whose ma rostate does not hange in time, ould
be pra ti ally onstru ted by taking all points of the phase spa e Γ whi h this system
visits in onsequtive instants t0 + n∆t with some ∆t and n → ∞ and assinging them
all to the iniatial instant t0 . The systems of the ensemble populate, therefore, dierent
possible mi rostates whi h give rise to the same ma rostate (as spe ied by its ma ro-
s opi ally measured hara teristi s). Thus, elements of the ensemble an be represented
as a olle tion of points in the phase spa e Γ. The main task of the statisti al approa h
is however to theoreti ally onstru t an ensemble in su h a way that the distribution
of its systems over the phase spa e Γ (at some t0 ) is from the ma ros opi point of view
representative for the onsidered real system (at the same instant t0 ).
In general ensembles obtained in the way spe ied above an be hara terized by their
phase spa e distribution fun tion ρ(q, p, t) normalized to unity
Z
dΓ(q,p) ρ(q, p, t) = 1 , (176)
Γ
(the pre ise form - its normalization fa tor - of the measure dΓ(q,p) ∝ d3N q d3N p will be
spe ied later). It gives the fra tion of the number of the ensemble systems whi h at the
instant t are ontained in the innitesimal phase spa e volume element dΓ(q,p) around the
point (q i , pi ). In other words,
N ρ(q, p, t) dΓ(q,p) ,

124
is the number of systems of the ensemble ontained (at the instant t) in this innitesimal
phase spa e volume element. Given the distribution fun tion ρ(q, p, t), with every observ-
able quantity O whi h is determined by the mi ros opi state the ensemble average O(t)
over the phase spa e an be asso iated:
Z
O(t) ≡ dΓ(q,p) ρ(q, p, t) O(q, p) . (177)
Γ

It is the ensemble average whi h in the statisti al approa h is going to repla e the time
averages (173) and the justi ation of this repla ement will be dis ussed (to the extent
to whi h it is possible) below.
Sin e the systems forming an ensemble are by denition onstru ted as (mi ros opi-
ally) isolated, the time evolution of ea h of them an be analyzed with the help of the
anoni al equations of motion (174) with the Hamiltonian H (q, p) whi h in ludes only
the internal intera tions of the system (the terms Henv and Hsyst−env are absent from it)
and denes therefore the dynami s of an autonomous system. As ea h of the systems
of the ensemble evolves in time, the distribution fun tion ρ globally hara terizing their
distribution over the phase spa e hanges. Certain global features of these hanges of ρ
an be understood on the basis of the Liouville theorem.

Liouville theorem and its onsequen es

The Liouville theorem is a te hni al result obtained within lassi al me hani s whi h pro-
vides a global hara terization of the dynami al behaviour of a large number of identi al
systems forming an ensemble. It is needed in the development of the argument justifying
the statisti al approa h.
Let us onsider rst the total dierential of the introdu ed distribution fun tion:

3N  
∂ρ X ∂ρ i ∂ρ
dρ = dt + dq + dpi .
∂t i=1
∂q i ∂pi

The sum ρ + dρ is obviously the value of the distribution fun tion at an arbitrarily hosen
i i
neighbouring point (q + dq , pi + dpi ) of the phase spa e at the moment t + dt written in
i
terms of its value ρ at the point (q , pi ) at the instant t. One an, however ask what will
i i
be the value of the distribution fun tion at the instant t+dt at the point (q +dq , pi +dpi )
i i
whi h is liked to (q , pi ) by the motion of the system. This requires orrelating dq and
i i
dpi with dt by dq = q̇ dt, dpi = ṗi dt. The distribution fun tion at this neighbouring point
i
linked to (q , pi ) by the dynami s is equal ρ + (dρ/dt)dt with (the se ond form follows by
using the anoni al equations (174) and the denition of the Poisson bra ket)

3N  
dρ ∂ρ X ∂ρ i ∂ρ ∂ρ
= + q̇ + ṗi ≡ + {ρ, H }PB . (178)
dt ∂t i=1
∂q i ∂pi ∂t

The rst part of the Liouville theorem states that dρ/dt is zero.

125
 To prove this statement we onsider an arbitrarily hosen nite (in the mathemati al
language, of nite measure) domain ∆Γ of the phase spa e Γ and ask how the instanta-
neous number
Z
N∆Γ (t) = dΓ(q,p) N ρ(q, p, t) ,
∆Γ

of phase points (representing the systems of the ensemble) ontained in this domain
hanges between t and t+ dt. On one hand, this hange is simply given by (dN∆Γ(t)/dt)dt
where

dN∆Γ (t) d ∂ρ(q, p, t)

Z Z
= dΓ(q,p) N ρ(q, p, t) = dΓ(q,p) N .
dt dt ∆Γ ∆Γ ∂t
On the other hand, the same hange dN∆Γ(t) must be equal to the ux of the phase
points, given by the 6N dimensional ve tor

q̇ i
 
Nρu ≡ Nρ ,
ṗi
(whi h is the produ t of N ρ itself and the 6N dimensional velo ity ve tor u), integrated
over the boundary of the domain ∆Γ and multiplied by dt (dΣ is the 6N − 1-dimensional
dierential area ve tor normal to the surfa e and dire ted outwards):
Z
dN∆Γ (t) = −dt dΣ·u N ρ .
∂∆Γ

By the Stokes theorem this an be written in the form

dN∆Γ (t)
Z
=− dΓ(q,p) ∇·(u N ρ) .
dt ∆Γ

Equating the two forms of the rate dN∆Γ(t)/dt and taking into a ount that the domain
∆Γ was arbitrary (whi h implies that integrated equality an be repla ed by its lo al
dierential version), we obtain the equality

3N  
∂ρ ∂ρ X ∂ i ∂
0= + ∇·(u ρ) = + (q̇ ρ) + (ṗi ρ)
∂t ∂t i=1
∂q i ∂pi
3N   i  
∂ρ X ∂ρ i ∂ρ ∂ q̇ ∂ ṗi
= + i
q̇ + ṗi + i
+ ρ .
∂t i=1
∂q ∂p i ∂q ∂p i

Sin e

∂ q̇ i ∂ ∂H ∂2H ∂ ṗi ∂ ∂H ∂2H

= = , =− = − ,
∂q i ∂q i ∂pi ∂q i ∂pi ∂pi ∂pi ∂q i ∂pi ∂q i
the last term in the square bra kets is zero and we indeed obtain dρ/dt = 0. The vanishing
of the total time derivative of the distribution fun tion ρ means that this fun tion is

126
 onstant along traje tories of the phase points or, that the phase uid is in ompressible.
Alrnatively, this an be written as the equality

ρ(q, p, t) = ρ(q0 , p0 , t0 ) , (179)

holding for (q i , pi ) related at t to the values (q0i , p0i ) of the phase spa e oordinates at t0
i i
by the motion: q = q (t, q0 , p0 , t0 ) and pi = pi (t, q0 , p0 , t0 ).
To obtain the se ond part of the Liouville theorem we onsider a small (innitesimal
in fa t) domain ∆Γ of the phase spa e and follow over the time s the motion of all the
phase points whi h were ontained in it at some instant t0 . At the moment t = t0 +s these
points will o upy another domain od the phase spa e. The shape of this new domain is
marked by the points whi h at t0 formed the boundary of ∆Γ. Moreover, no one point of
those whi h at t0 formed the interior of ∆Γ ould have evolved outside of the new domain,
be ause to do so it would have to ross on the way the moving boundary of the domain:
at some moment between t0 and t0 + s it would have to oin ide with some boundary
point and sin e the Hamilton's equations of motion are ausal it would have to oin ide
all the time with this point of the boundary. It follows that the number of phase points
inside the moving domain ould not hange and sin e we have shown that along the phase
traje tories the distribution fun tion (i.e. the density of the points in the phase spa e) is
onstant, we on lude that although the shape of the moving domain ould have hanged
between t0 and t0 + s, its volume is onstant. This in turn means that

∂(q, p)
J= = 1. (180)
∂(q0 , p0 )
where J is the ja obian of the hange (transformation) of the integration variables from
(q i , pi ) to (q0i , p0i ) whi h is given by the motion: q i = q i (s, q0 , p0 ), pi = pi (s, q0 , p0 ) (with
the time s playing the role of an arbitrary transformation parameter).
Put dierently, the above argument shows that
Z Z
dΓ(q,p) N ρ(q, p, t0 + s) = dΓ(q0 ,p0 ) N ρ(q0 , p0 , t0 ) ,
∆Γ(t0 +s) ∆Γ(t0 )

where q i = q i (s, q0 , p0 , t0 ), pi = pi (s, q0 , p0 , t0 ). After using these formulae as dening the

transformation of the integration variables and using the equality (179) the left integral
takes the form

∂(q, p)
Z
dΓ(q0 ,p0 ) N ρ(q0 , p0 , t0 ) ,
∆Γ(t0 ) ∂(q0 , p0 )

Its equality to the right integral means therefore that (180) holds.

We now make the ru ial step in developing the statisti al approa h. We are going to
postulate the relation between a quantity (173)

t0 +τ
1
Z
Oobs = dt O(q, p) ,
τ t0

127
measured on the real system (isolated or not) and the appropriate ensemble average. The
latter is dened as follows. We onstru t an ensemble whi h at t0 is representative for
the real system. With ea h (isolated) system of the ensemble whi h from its mi rostate
(q0i , p0i ) at t0 evolves as di tated by the autonomous Hamilton's equations (174):

q i = q i (s, q0 , p0 ) , pi = pi (s, q0 , p0 ) ,

the time average

τ
1
Z
Oτ (q0 , p0 ) = ds O(q(s, q0, p0 ), p(s, q0 , p0 )) ,
τ 0

an be asso iated. The period τ (here and in (173)) an be taken su iently long from the
mi ros opi point of view (for instan e, in the ase of a gas τ should be large ompared
to the mean time between ollisions), so that Oτ
is pra ti ally independent of τ (though,
i 0
as indi ated by the notation, it an still depend on the initial point (q0 , pi )). The main
postulate is then that Oobs should be identied with the ensemble average
Z
Oτ ≡ dΓ(q0 ,p0 ) ρ(q0 , p0 , t0 ) Oτ (q0 , p0 ) . (181)

The validity of the identi ation of Oobs with Oτ depends of the magnitude of u -
tuations. If the real system is (mi ros opi ally) isolated (whi h is the limiting, not very
realisti , ase), it an be taken as one of the systems of the ensemble. Smallness of the
u tuations means that the probability that Oτ (q0 , p0 ) omputed taking a randomly ho-
sen element of the ensemble - in parti ular taking the real system - markedly deviates
from Oτ is very low, so the same must be true of Oobs . If the real system is not isolated
(either mi ros opi ally or be ause it intera ts with the surrounding), then its phase spa e
traje tory an be imagined to be pie ewise omposed out of parts of traje tories tra ed
out by dierent systems (whi h are mi ros opi ally isolated) of the ensemble - due to the
random external perturbations the real system jumps from time to time from a phase
spa e traje tory of one isolated system onto the traje tory of another isolated system. It
again follows that the probability that Oobs whi h is a omposition (in varying propor-
tions) of parts of dierent Oτ 's deviates signi antly from the ensemble average Oτ is low
if the u tuations (whi h are determined by the ensemble) of Oτ 's are small provided the
81
ensemble does not represent faithfully the real system. In either ase, the validity of the
identi ation of Oobs with the ensemble average Oτ is redu ed to the problem whether
the u tuations of Oτ 's around O τ are small, whi h an be investigated entirely in the
framework of the ensemble itself, without any further referen e to the real system. It
is expe ted that the u tuations are indeed small owing to the smoothing ee t of time

81 This is why the problem of hosing the representative ensemble is the entral one of the statisti al
approa h. Fortunately simple hoi es whi h are made in typi al situations of ma ros opi ally isolated sys-
tems, or systems ex hanging energy with a heat bath prove (by the results they lead to) to be su iently
good.

128
averaging involved in omputing individual Oτ 's and owing to the enormous number of
degrees of freedom involved.
The arguments given above are by no means rigorous. It is lear that the value of Oτ
depends on the form of the distribution fun tion ρ(q, p, t) whi h, at t0 must reasonably
well represent the real system. It an be however expe ted that the mean values Oτ of
observable quantities O (but not their u tuations) are to some extent insensitive to the
form of ρ.
The onstru tion of the statisti al ensembles and the pres ription for omputing the
ensemble average Oτ whi h a ording to the adopted postulate is to be identied with the
really measured value Oobs given above are fairly general and apply to (real) systems in
equlibrium (either as ma ros opi ally isolated system or with their surrounding) as well
to systems out of equilibrium. If, as in this Course, we are interested only in systems in
equilibrium, some simpli ations an be made. Firstly, the measured quantities Oobs are
in this ase independent of time (ma ros opi hara teristi s of equilibrium systems by
denition do not vary in time). Therefore also the ensemble averages Oτ given by (181)
should be independent of t0 . It is lear that this requires that the orresponding ensemble
distribution fun tion ρ(q, p, t) does not depend expli itly on time:

∂
ρ(q, p, t) = 0 . (182)
∂t
Sin e Liuoville theorem states that dρ/dt = 0, from the formula (178) it follows that the
distribution fun tion ρ representing a systems in equilibrium, in addition of being not
expli itly dependent on time, must have zero Poisson bra kets with the Hamiltonian of
the system (treated as isolated):

{ρ(q, p), H (q, p)}PB = 0 . (183)

This means that the distribution fun tion an depend on the dynami al variables qi and
pi only through onserved quantities spe i for the onsidered system. Most of omplex
(isolated) systems onsisting of a very large numbers of parti les have only a few of
82
onserved quantities: energy, represented by the Hamiltonian itself and perhaps the total
momentum and total angular momentum of whi h the last two are usually eliminated by
en losing the system in suitable spatial walls. Thus typi ally the distribution fun tion of
an ensemble representing a real system in equilibrium has the form

ρ(q, p, t) = ρ(q, p) = ρ(H (q, p)) . (184)

Moreover, the re ipe for al ulating the ensemble average (181) pertaining to the
properties of a system in equilibrium an be onsiderably simplied: in the expli it formula
Z τ
1
Z
Oτ = dΓ(q0 ,p0 ) ρ(q0 , p0 ) ds O(q(s, q0, p0 ), p(s, q0 , p0 ))
τ 0
1 τ
Z Z
= ds dΓ(q0 ,p0 ) ρ(q0 , p0 ) O(q(s, q0 , p0 ), p(s, q0 , p0 )) ,
τ 0
82 The ex eption are so alled integrable systems whi h an have up to 6N ( ompletely integrable
systems) onserved quantities. We will not be on erned here with su h systems.

129
one an use the result (179) withq i = q i (s, q0 , p0 ), pi = pi (s, q0 , p0 ) spe ied to distribution
fu tions satisfying (182) and the result (180) repla e dΓ(q0 ,p0 ) by dΓ(q,p) . After these
operations the integral over dτ fa torizes ompletely, an els out against the expli it 1/τ
fa tor and one obtains an expli itly τ -independent formula
Z
Oτ ≡ O = dΓ(q,p) ρ(q, p) O(q, p) . (185)

In the ase of systems in equilibrium, when taking the limit τ → ∞ in the denition
(173) of measured quantity is allowed (be ause the measured ma ros opi properties do
not depend on time) the formula (185) an be justied a la Landau: one an in this ase
imagine that the phase spa e traje tory of the real system is followed over a period τ and
the time periods ∆t(q,p) (τ ) the traje tory spends in dierent small domains ∆Γ(q, p) of
the phase spa e are ounted. If the distribution fun tion is onstru ted by taking

∆t(q,p) (τ )
ρ(q, p) ∝ lim , (186)
τ →∞ τ
and normalizing the resulting funtion of qi and pi to unity with respe t to the measure
dΓ(q,p) , the equality

τ
1
Z Z
lim dt O(q(t), p(t)) = dΓ(q,p) ρ(q, p) O(q, p) ,
τ →∞ τ 0

follows automati ally be ause the right integral over dt an be (somewhat heuristi ally)
written as

τ X ∆t(q,p)
1
Z
lim dt O(q(t), p(t)) = lim O(q, p) ,
τ →∞ τ 0 τ →∞ τ
(q,p)

with the sum over ells of the phase spa e in whi h the system spends overall periods ∆t(q,p)
in its journey over the phase spa e. Of ourse also in this derivation the question how to
theoreti ally onstru t the distribution ρ similar to the one dened here operationally is
not solved by the above reasoning.

On e the distribution fun tion ρ(q, p) of the ensemble representing a real system in
equilibrium is given, obtaining theoreti al predi tions for measured ma ros opi quantities
hara terizing this system redu es to the appli ation of the probability theory: the phase
spa e parametrized by the variables (q i , pi ) plays the role of the spa e of elementary events,
ρ(q, p)dΓ(q,p) plays the role of the probability distribution and observables O(q, p) be ome
random variables dened on the spa e of elementary events. The probability distribution
ρO (O) of a random variable (observable) O an be then onstru ted a ording to the

130
 83
standard pres ription
Z
ρO (O) = dΓ(q,p) ρ(q, p) δ(O(q, p) − O) . (187)

Owing to the properties of the Dira delta fun tion, the probability distribution ρO (O)
onstru ted in this way is automati ally normalized to unity. The mean value of O an
be then omputed either as in (185) or, as
Z
O= dO O ρO (O) . (188)

The standard estimate of the u tuations of the random variable O around its mean value
O is obtained by omputing the quantity
Z
2
σO2 ≡ (O − O)2 = dO (O − O)2 ρO (O) = O 2 − O . (189)

alled the mean quadrati u tuation and the taking its square root. The (dimen-
sionless) measure of u tuations is the relative u tuation given by the ratio
p
σO2
, (190)
O
whi h tells how large u tuations are ompared to the mean value. The su ess of the
statisti al method applied to large (N ∼ NA ) systems relies mostly on the fa t that even
if the distribution ρ(q, p) is at (as in the ase of the mi ro anoni al ensemble - to be
dened below), the probability distribution ρO (O) of any ma ros opi observable has an
enormously sharp peak at a value O ∗ whi h is therefore almost the same as O and the
relative u tuations are very tiny indeed.

Statisti al independen e
If the system (the spa e of elementary events) an be split into two nonintera ting parts
a and b whi h an be treated as not inuen ing one another in any way, these systems
should be statisti ally independent: their joint distribution fun tion ρa,b (qa , pa , qb , pb , t)
should fa torize, that is, the following equality should hold

dΓ(qa ,pa ,qb,pb ) ρa,b (qa , pa , qb , pb , t) = dΓ(qa ,pa ) ρa (qa , pa , t) dΓ(qb ,pb ) ρb (qb , pb , t) . (191)

The observables Oa = O(qa , pa ) and Ob = O(qb , pb ) pertaining to these two subsystems

are then independent whi h for example means that

Oa Ob = Oa Ob ,
83 This pres ription generalizes the ordinary summation of probabilities of dis rete elementary events
leading to the same value of the random variable; for instan e if the value of random variable f is −1 for
an odd number number of dots on a (perfe tly symmetri ) di e and and f = +1 for an even number, the
1 1 1 1 1 1
probabilities of the values f = ∓1 are p(−1) =
6 + 6 + 6 and p(+1) = 6 + 6 + 6 . In the ase of ontinuos
probability distributions the integration over dΓ(q,p) ombined with the Dira delta fun tion essentially
i
sums up probabilities of elementary events (q , pi ) whi h lead to the same value of the observable O .

131
et .
Statisti al independen e an be invoked to obtain two important results on erning
ma ros opi systems (bodies). Firstly, if the system is large it an be mentally divided into
84
a large ma ros opi parts whi h are to a good approximation statisti ally independent.
It is natural to assume that the number of statisti ally independent parts (whi h an still
be treated as ma ros opi ) into whi h the system is divided remains in a proportion to
the number N of its mole ules. If one is interested in an additive (extensive) quantity O
hara terizing the system - for instan e the system is in equilibrium with the heat bath
and therefore one is interested in its internal energy, or when the system is in me hani al
onta t with the atmosphere and one asks about its total volume - then from the statisti al
independen e of its ma ros opi parts it immediately follows that the relative u tuation
(190) of this quantity de reases with the size of the body. Indeed, an extensive quantity
O an be split into the sum
X
O= Oa ,
a

of quantities hara terizing its ma ros opi parts (labeled by a) and by the argument
given above,
X
O= Oa ∝ N ,
a

be ause all the parts are essentially identi al. Moreover,

!2
X XX
σO2 = (O − O )2 = (Oa − Oa ) = (Oa − Oa )(Ob − Ob ) .
a a b

In the double sum one an now single out the terms in whi h a=b and, appealing to the
statisti al independen e, write
X X
σO2 = (Oa − Oa )2 + (Oa − Oa )(Ob − Ob )
a a6=b
X X
= (Oa − Oa )2 + (Oa − Oa ) (Ob − Ob ) . (192)
a a6=b

The se ond sum vanishes be ause Ob − Ob = Ob − Ob = 0 and the rst sum, sin e all
parts into whi h the system has been split are essentially identi al, is proportional to N.
It follows therefore, that the relative u tuation of the extensive quantity s ales as
p √
σO2 N 1
∝ =√ , (193)
O N N
84 In general the statisti al independen e holds over not too long periods; in the ase of systems in
equilibrium whose ma ros opi state does not hange in time, it should however be always (approximately)
true.

132
and is really minute, if the system is large (N ∼ NA ).
Another important onsequen e of the statisti al independen e whi h is implied by the
fa torization (191) of the distribution fun tion is as follows. Suppose the two subsystems
an be treated as statisti ally independent, either be ause they are sepearated one from
another or be ause, as in the onsiderations above, they are two ma ros opi parts of
the same ma ros opi body. The Hamiltonian of the entire system in equilibrium treated
as mi ros opi ally isolated (even if the real system is not ne essarily su h - it an be in
equilibrium with its environment) is then just the sum H = Ha + Hb (Ha depends on
the variables of the subsystem a, while Hb on the variables of the subsystem b). In view
of the statisti al independen e of the subsystems a and b, the distribution fun tion of the
ensemble representing the entire system must have the property (re all the result (184) !)

ρa,b (H ) = ρa,b (Ha + Hb ) = ρa (Ha ) ρb (Hb ) ,

or

ln ρa,b (Ha + Hb ) = ln ρa (Ha ) + ln ρb (Hb ) . (194)

85
The only possibility of satisfying this requirement is

ln ρa (Ha ) = αa − βHa , (195)

or, in the ase the number of mole ules be omes a dynami al variable,

ln ρa (Ha ) = αa − βHa + γNa . (196)

Thus, statisti al independen e of systems whi h an be treated as isolated from one an-
other imposes stringent onstraints on the possible forms of the distribution fun tions of
statisti al ensembles representingsystems in equilibrium.

Classi al Mi ro anoni al Ensemble

We now introdu e the distribution fun tion of the statisti al ensemble whi h should ad-
equately represent the system whi h from the ma ros opi point of view is isolated and
in equilibrium. Be ause a ording to the point of view adopted here su h a system is
not isolated at the mi ros opi level, but is onstantly perturbed by its environment, it is
reasonable to assume that its internal energy is xed only up to some un ertainty, whi h
we will represent by allowing the system's energy to be in the range (E, E + ∆E) with
∆E ≪ E , and that all system's mi rostates with the energy in this range are equally
probable. In agreement with this assumption the distribution fun tion of the statisti al
ensemble representing a ma ros opi ally isolated system is taken in the form

const if E ≤ H (q, p) ≤ E + ∆E
ρ(q, p) = (197)
0 otherwise
85 Indeed, dierentiating the equalityln f (x + y) = ln f (x) + ln f (y) with respe t to x one obtains that
f (x + y)/f (x + y) = f ′ (x)/f (x); in the same way one nds that f ′ (x + y)/f (x + y) = f ′ (y)/f (y); hen e
′

f ′ (y)/f (y) = f ′ (x)/f (x) = −β = onst., be ause both sides are fun tions of a dierent independent
variable. From this the on lusion follows.

133
The ensemble dened by this distribution fun tion is alled mi ro anoni al. Sin e the
distribution fun tion must be normalized to unity, the onstant is equal 1/Γ(E, V, N, ∆E)
where
Z
Γ(E, V, N, ∆E) = dΓ(q,p) , (198)
E≤H (q,p)≤E+∆E

is the volume of the orresponding part of the phase spa e. In various onsiderations
useful are also two other quantities:
Z Z
Σ(E, V, N) = dΓ(q,p) ≡ dΓ(q,p) θ(E − H (q, p)) , (199)
H (q,p)≤E

whi h gives the phase spa e volume orresponding to the system's total energy less than
E and:

∂Σ(E, V, N)
Z
ω(E, V, N) = = dΓ(q,p) δ(H (q, p) − E) , (200)
∂E
whi h gives the area of the shell orresponding to the system's total energy equal E; in
the quantum ase ω(E) will have the lear interpretation of the density of quantum states
of the system. Sin e the energy allowan e ∆E is small ompared to E , a useful working
approximation is

Γ(E, V, N, ∆E) = Σ(E, V, N) − Σ(E − ∆E, V, N)

∂Σ(E, V, N)
≈ ∆E = ω(E, V, N) ∆E . (201)
∂E
As the quantity Γ(E, V, N, ∆E) hara terizing the spread of the systems of the mi ro-
anoni aal ensemble over the phase spa e is nite, one an dene the statisti al entropy
86
by the simple formula

Sstat = kB ln Γ(E, V, N, ∆E) , (202)

in whi h kB = 8.617 × 10−5 eV·K

−1
is the Boltzmann onstant. In this way the statisti al
entropy gives the measure of the disorder in the system, that is hara terizes how mu h
a given ma rostate of the system is spread out over the phase spa e, or how many mi-
rostates are asso iated with the given ma rostate (spe ied by the total energy E, the
volume V and the number N of parti les). Of ourse it still remains to be shown how the
statisti al entropy Sstat dened by (202) is related to the thermodynami entropy S, but
the resonings needed for this follow losely those employed in showing that the Callenian
entropy agrees with that introdu ed via the Clausius inequality. This will be done after

86 This is the famous Boltzmann formula whi h in the form

S = k ln W ,

is engraved on his tomb in der Haupstad Wien Friedhof (at the Viener emetery).

134
we onsider statisti al me hani s based on quantum me hani s. The result will be that
the thermodynami entropy S should be idented with
 
Sstat (E, V, N, ∆E)
S = N lim , (203)
∞ N
where the symbol lim∞ denotes the so- alled thermodynami limit whi h means N →
∞, E → ∞, V → ∞ with the ratios E/N and V /N kept xed. In this limit the
87
dependen e on ∆E drops out. In most ases the statisti al entropies omputed repla ing
Γ(E, V, N, ∆E) under the logarithm in (202) by Σ(E, V, N) or by ω(E, V, N) lead in the
thermodynami limit to the same thermodynami entropy.
Sin e the statisti al entropy (202) is proportional to the logarithm of Γ(E, V, N, ∆E),
it hanges by an additive onstant, when the normalization fa tor of the integration
measure dΓ(q,p) is altered. In lassi al physi s this fa tor is arbitrary. There is no su h
an ambiguity (apart from the hoi e of the energy allowan e ∆E ) in quantum statisti al
physi s in whi h entropy (in the thermodynami limit) is assigned unambigously to ea h
ma rostate. It will be seen that if lassi al statisti al me hani s is to be the limit of its
quantum ounterpart, the measure dΓ(q,p) should have the form

d3N qd3N p
dΓ(q,p) = . (204)
N! (2π~)3N
The fa tor (2π~)3N in the denominator makes the measure dimensionless. It is re-
lated to the rule (following e.g. from the Bohr-Sommerfeld quantization ondition)
that in one-dimensional quasi- lassi al motion of a single parti le every single quantum
state an be asso iated with the area 2π~ of the two-dimensional phase spa e. Thus
d3N qd3N p/(2π~)3N simply gives the number of 6N dimensional ells in the phase spa e
available to the system ea h of whi h would orrespond to one quantum state if the N
parti les were distinquishable. If parti les are indistinguishable, the rules of quantum
me hani s require that their quantum states be either symmetri (if parti les are bosons -
parti les of integer spin) or antisymmetri (if they are fermions - parti les of half-integer
spin) with respe t to inter hanging parti les (this will be ome more lear when we dis uss
the se ond quantization formalism). As a result the states whi h would be ounted as
distin t if the parti les were distinguishable, are, when the parti les are in distinguishable,
one and the same quantum state. Therefore the number of quantum states available to
88
parti les is redu ed roughly by the fa tor of N!. This explains the origin of the extra
fa tor of N! in the denominator of the measure (204). This fa tor was rst introdu ed
by Gibbs (before its quantum origin be ame lear) to save extensiveness of entropy of the
perfe t gas whi h would not hold without it (this will be seen in lasses; the problem is

87 In lassi al statisti al physi s one ould from the beginning set ∆E = 0, but this is not so in
quantum statisti al me hani s (unless the system is pe uliar). The dependen e of S on ∆E drops out in
the thermodynami limit in both ases.
88 This does not allow to ount the number of states orre ly but is a su iently good estimate in situa-
tions in whi h genuine quantum ee ts are not important for thermodynami s, that is, if the temperature
is not too low.

135
sometimes referred to as the Gibbs paradox). It should be also remarked that if the sys-
tem onsists of N1 (indistinguishable) parti les of a one kind and of N2 (indistinguishable)
parti les of a se ond kind (the two kinds are dierent, so distinguishable) the appropriate
measure would be

d3N1 q(1) d3N p(1) d3N2 q(2) d3N p(2)

dΓ(q,p) = ,
N1 ! (2π~)3N1 N2 ! (2π~)3N2

and it is pre isely the fa tor (N1 + N2 )!/N1 !N2 ! by whi h it diers from the measure asso-
iated with the system of N1 + N2 identi al (indistinguishable) parti les that leads to the
entropy of mixing by whi h thermodynami entropies of the two systems dier (Le ture
VIII). With the normalization spe ied in (204) the Boltzmann formula (202) states that
entropy (in units of kB ) of a ma ros opi ally isolated system is just the logarithm of the
number of (mi ros opi ) quantum states onsistent with the ma ros opi hara teristi s
of ( onstraints imposed on) the system.
If the system is not ma ros opi ally isolated and the distribution fun tion of the
representing it statisti al ensemble is not lo alized in the phase spa e as is (197), the
entropy annot be given by the formula (202) be ause the volume of the phase spa e
avalilable to the system is innite. One needs to invent then another measure of disorder
in terms of whi h entropy an be dened. We will dis uss this in general terms after we
onsider foundations of quantum statisti al me hani s.

Ergodi ity
It seems appropriate to lose this Le ture by dis ussing briey the so- alled ergodi prob-
lem and the pla e it a upied in the past in the foundations of ( lassi al) statisti al me-
hani s. We have not made any referen e to it in our onsiderations owing of our adoption
of the point of view, whi h physi ally seems very natural and moreover allows to put the
foundations of lassi al and quantum statisti al me hani s (see the next Le ture) on the
same basis, that no real system an be onsidered ompletely isolated at the mi ros opi
level. It served us to justify (somewhat heuristi ally, let us agree!) the postulate of a pri-
ori equal probabilities assigned to all mi rostates a esible to a ma ros opi ally isolated
system. We will see that this postulate an be introdu ed also without the ne essity of
relying (only) on the mi ros opi non-isolation of ma ros opi systems by appealing to
the information theory, if a somewhat more realisti view is taken on the possible out-
omes of real measurements made on ma ros opi systems. The mi ro anoni al ensemble
whi h results from this postulate leads to orre t physi al predi tions and there an be
no doubts about its orre tness. However in the past many physi ists subs ribed to the
position that a system whi h is isolated ma ros opi ally is also mi ros opi ally stri tly
isolated against any inuen es from without, however minute (and the relevan e of argu-
ments based on the information theory were not yet ommonly known), and were trying
to justify the postulate of a priori equal probabilities dire tly on the basis of mi ros opi
89
time evolution of large autonomous isolated dynami al systems. It is in this ontext
that the ergodi problem appeared.

89 If one was not ontented with the results the postulate leads to as its ultimate justi ation.

136
 p
∆Γ(t2 )
∆Γ(t1 ) D
C
B
A

Figure 13: .

90
Briey, the problem in whi h the ergodi problem appeared was to show rigorously
that if the ma ros opi system is isolated, the time average (173) with τ → ∞ is equivalent
(that is, equal in this limit) to the ensemble average with the onstant distribution fun tion
on the 6N − 1-dimensional hypersurfa e of onstant energy. The proof of the equivalen e
has two parts. One of it is showing that the point representing the time evolution of an
isolated system in the phase spa e spends in a given domain ∆Γ the time ∆t whi h is
91
proportional to its volume. The se ond part is showing that this is true independently
of where in the phase spa e su h a volume is lo ated. If both parts are shown then the
equivalen e of the (innite) time averages and the orressponding phase spa e averages
follows just as in the Landau's reasoning be ause this means that ρ(q, p) dened as in
(186) would be uniform.
The proof of the rst part is simple. Consider a set of points in the phase spa e whi h
at t1 ll the domain ∆Γ(t1 ) as in Figure 13. At a later instant t2 the same points ll the
domain ∆Γ(t2 ) whi h is uniquely determined by the positions at t2 of those phase points
whi h at t1 formed the boundary of ∆Γ(t1 ). By the Liouville theorem the volumes of the
two domains are equal (though their shapes may be dierent). Let us follow the points
A and B whi h lie on the same traje tory. The time tAC it takes the point A to rea h C
must be therefore the same as the time tBD it takes the point B to rea h D be ause these
two points and their images dene the boundaries of ∆Γ(t1 ) and ∆Γ(t2 ). Moreover

tAC = tAB + tBC , tBD = tBC + tCD ,

from whi h it follows that tAB = tCD : the phase point indeed does spend the same times
in∆Γ(t1 ) and ∆Γ(t2 ) in equal volumes. Furthermore, if τ is sent to innity, any time the
phase point rosses ∆Γ(t1 ) it must also ross ∆Γ(t2 ).
The di ulty lies in showing that this result holds wherever the volumes are lo ated.
The original hypothesis of Boltzmann ( alled by him the ergodi hypothesis) was that as
τ → ∞ the traje tory passes through every point of the onstant energy hypersurfa e. As

90 And this is what attra ted pure mathemati ians to this eld so that it ee tively be ame a bran h
of mathemati s.
91 We will simplify slightly the problem and will onsider the 6N -dimensional domains assuming impli -
itly that the limit ∆E → 0 is taken; otherwise we would have to onsider the proje tions of the measure
dΓ(q,p) onto the hypersurfa e of onstant energy.

137
this was untenable on mathemati al grounds, it has been repla ed by the so- alled quasi-
ergodi hypothesis (frequently naw alled just ergodi ) that as τ → ∞ the traje tory
passes through any arbitrarily small neighbourhood of every point of the onstant energy
hypersurfa e. Although this may seem physi ally obvious if the system is large, it is not
easy to prove mathemati ally. The important step was done by Birkho. The ru ial
problem is in showing that the onstant energy hypersurfa e does not split into parts
mutually ina essible from one another (the so alled metri inde omposability of this
hypersurfa e). Some results have been rea hed by mathemati ians (Sinai) in this dire tion
but on erning idealized systems like the model of N hard spheres (elasti ally s attering
one on one another). No general proof exists.
At present the ommon view is that, while being very interestig as a problem in
pure mathemati s and also as important part of the investigations of haoti phenomena
exhibited by omplex lassi al systems, the ergodi problem is not relevant to the physi al
foundations of lassi al statisti al me hani s and we mentioned it here only in order make
the students (with more mathemati al in linations) aware of its existen e as a potentially
interesting eld of resear h and to at least partly removing an atmosphere of misti ism
surrounding it.

138
 LECTURE XI (STAT)

Quantum me hani s
As it should be known, the primitive notion of quantum me hani s is the one of the
92
quantum state of the onsidered system. In the mathemati al formulation quantum
states are represented in a Hilbert spa e H - a omplete (in the sense of onvergen e of all
Cau hy sequen es) ve tor spa e over the eld of omplex numbers endowed with a s alar
produ t and therefore also a norm k · k - by lasses ( alled rays) of equivalen e of
(·|·)
′ iδ
ve tors: two ve tors Ψ and Ψ = e Ψ belong to the same ray (and represent the same
physi al state of the system). The proper Hilbert spa e H onsists of normalizable ve tors,
2
i.e. su h ve tors that k Ψ k = (Ψ|Ψ) < ∞, but in many ases it is onvenient to enlarge it
∗
(essentially repla ing it with its dual H - the spa e of linear forms over H - and identifying
elements of the proper Hilbert spa e with the one-forms one forms orresponding to
them via the Fre het-Riesz isomorphism) in luding also the nonnormalizable ve tors (also
alled generalized ve tors). Representants of rays are alled state-ve tors. Observables -
quantities whi h an be measured on the system - are represented by linear self-adjoint
operators Ô (also alled - not entirely orre tly - Hermitian operators), that is su h that
Ô † = Ô, where the operator Ô † adjoint with respe t to Ô is dened by the equality

(Φ|Ô † Ψ) = (ÔΦ|Ψ) , (205)

whi h must hold for all ve tors Ψ and Φ belonging to H. The spe trum - the set of its
eigenvalues ok i.e. the set of su h numbers that there exists ve tors satisfying the equation
ÔΨk = ok Ψk - of a self adjoint operator is real whi h enables one to identify it with possible
outputs of individual measurements on the system of the physi al quantity represented
by that operator. The most important operator is the (self-adjoint) Hamiltonian Hˆ
representing energy of the system, whi h determines the system's state time evolution via
the S hrödinger equation

d
i~ Ψ(t) = Hˆ Ψ(t) . (206)
dt
If the Hamiltonian operator is not expli itly time-dependent, the solution of the S hrödinger
equation is given by
 
i ˆ
Ψ(t) = exp − H (t − t0 ) Ψ(t0 ) , (207)
~
92 Let us remark that to some this notion remains still too abstra t to be immediately a epted as a
basis of the whole quantum me hani s. In onne tion with this it is amusing to read the rst hapter
of the Dira 's Prin iples of Quantum Me hani s and then the prefa e written to the russian translation
of this renowned monograph by the a ademi ian V.I. Fo k: the prefa e learly shows that its author,
who himself made a signi ant ontribution to the quantum theory - we will use in this Course the Fo k
spa e, when we ome to dis uss systems ex hanging matter with the environment - evidently ould not
liberate himself from the onventional notion of the wave fun tion understood narrowly in the spirit of
the S hrödinger wave me hani s.

139
with the state-ve tor Ψ(t0 ) playing the role of the initial onditions. The spe trum of the
Hamiltonian whi h in general an onsist of a dis rete part and (if the quantum me hani s
is formulated in the innite spa e) of a ontinuos part - the orresponding eigenve tors
∗
are nonnormalizable generalized ve tors belonging to H ) onstitutes always the most
important hara teristi s of the quantum system.
Wherever more subtle mathemati al issues do not intervene, onvenient is the Dira
notation in whi h state-ve tors are written as kets |Ψi and elements of the dual spa e
as hΦ|. The s alar produ t (Φ|Ψ) takes in this notation the form (whi h in fa t impli itly
employs the Fre het-Riesz isomorphism) hΦ|Ψi and the a tion of an operator Ô on a state
93
ve tor Ψ is written as Ô|Ψi. The quantities hΦ|Ô|Ψi are alled matrix elements of the
operator Ô between the states Ψ and Φ. As is usually the ase with ve tor spa es, also
in the Hilbert spa e it is possible to hose a basis, a set of ve tors Ψl , or just |li, labeled
by some label l ; the basis of the proper Hilbert spa e H an always be hosen so that
the label l , whi h an also stand for a multi-label of the sort l1 l2 . . . ln , is dis rete. Only
al ulational onvenien e di tates employing ontinuos labels (the orresponding ve tors
∗
are then generalized nonnormalizable ve tors, i.e. elements of H ). If the basis is ountable
(l runs over a nite or a ountably innite set of values) the Hilberst spa e H is separable.
If it is un ountable - if l = l1 l2 . . . ln n = ∞ and ea h li runs over a ountably innte set
- then H is nonseparable and innitely many mutually orthogonal separable subspa es
an be hosen in it; in statisti al physi s nonseparable Hilbert spa es enter the game with
the Grand Canoni al Ensemble and lie at the basis of several phenomena (Bose-Einstein
ondenstation, phase transitions), but we will not enter into these ne mathemati al
details in this Course. Any state-ve tor an be written as a linear ombination of basis
ve tors
X X
Φ= Ψ l cl , or |Φi = |li cl . (208)
l l

Usually one works with bases formed by mutually orthonormal ve tors: (Ψl′ |Ψl ) = hl′ |li =
δl′ l , whi h are eigenve tors of an observable Ô (in most ases of the Hamiltonian Hˆ of
the system under onsideration). Matrix elements of any operator Ô between the state-
ve tors |li of the basis formed by its own eigenve tors are diagonal that is, take the
form

hl′ |Ô|li = ol δl′ l , (209)

and the operator itself an be written in the form

X X
Ô = |li ol hl| = ol P̂ol , (210)
l ol ∈(spectrum)

where P̂ol = P̂o2l is the proje tor onto the subspa e (wni h may well be multidimensional)

of H orresponding to the Ô eigenvalue ol . Also important is the fa t that if two operators

93 The Dira notation is adapted to self adjoint operators; if Ô is not self adjoint one has to adopt the
onvention that the symbol hΦ|Ô|Ψi means (Φ|ÔΨ). The s alar produ t (ÔΦ|Ψ) an in this notation be
∗
written only as (hΨ|Ô|Φ̂i) .

140
Ô1 and Ô2 ommute, that is [Ô1 , Ô2 ] ≡ Ô1 Ô2 − Ô2 Ô1 = 0, it is possible to nd the basis
in whi h both are simultaneously diagonal.

In the previous Le ture we have formulated the approa h to lassi al statisti al me-
hani s based on the use of ensembles in general terms so as to make it appli able in
prin iple also to systems not in equilibrium. Only later we have narrowed down the dis-
ussion to systems in equilibrium (either ma ros opi ally isolated or in equilibrium with
their environments). Our dis ussion of statisti al methods based on quantum me hani s
will be from the beginning restri ted to systems in equilibrium be ause I'm not very famil-
iar (although I wish I would be!) with the treatment of nonequilibrium quantum systems.
(But most probably it the fomulation given below an be straightforwardly extended to
systems not in equilbrium).
There are two ir umstan es whi h make the assumption that systems ma ros opi ally
looking as perfe tly isolated annot be treated as su h at the mi ros opi level even
more true in the quantum ase than in lassi al physi s. The rst one is the extreme
density of the spe tra of Hamiltonians of quantum systems whi h are ma ros opi (i.e.
onsist of N ∼ NA elements - parti les or mole ules). Statisti al systems are always
assumed to be of nite (although ma ros opi ally large) spatial dimensions and therefore
the problems asso iated with nonnormalizable (generalized) eigenve tors of Hamiltonians
94
and their related ontinuous (parts of ) energy spe tra an in prin iple always be avoided.
Yet the spe tra of Hamiltonians of ma ros opi systems are so dense that they are nearly
ontinuous. Roughly, the energy levels of su h systems are separated by gaps of order

En+1 − En ∼ exp(−N) .

This is not seen if the N elements of whi h the system is omposed are mutually noinin-
tera ting (examples used in introdu tory ourses of statisti al physi s are mostly of this
kind) - in su h ases it is the degenera y of individual energy levels whi h grows expo-
nentially with N (this will be illustrated in lasses by the system of N non-intera ting
parti les onned in a ma ros opi nite volume V ) - but mutual intera tions, even if they
an be negle ted in omputing gross ma ros opi features of the system, must always be
present (otherwise the system ould not rea h equilibrium) and ause smearing degenerate

94 There is a strong physi al onvi tion that lo al physi al results annot depend on whether the theory
is formulated in the ontinuum (innite spa e) or in a nite (but su iently large) volume. Even problems
like the one of the Hydrogen atom energy spe trum should not depend on this, at least from the pra ti al
point of view, be ause in the nite volume the entire spe trum is dis rete and the density of states near
E = 0− is also signi antly modied. (Also in many problems in my own favourite bran h of theoreti al
physi s - in quantum eld theory - in many ases it is good to remember that in fa t the theory should
be formulated in a nite spa e.) Yet from the al ulational point of view formulation of the theory in
the innite spa e has many advantages. The problem however is that quantum me hani s in a nite
volume and in the ontinuum are mathemati ally very dierent! To give the simplest example: the
s attering theory whi h an be used if the quantum me hani s of a single parti le moving in a potential
in one dimension is formulated in the ontinum does not exist if the same problem is formulated on a
nite segment ofR1 , however large, and all quantities like transmition and ree tion oe ients must be
repla ed by other hara teristi s simply be ause the asymptoti (s attering) states annot be in this ase
dened.

141
energy levels so that the above estimate of the interlevel gaps be omes true. The estimate
remains true also in the ase of stronly intera ting systems, when the energy levels of the
Hamiltonians annot be seen as perturbations of the free Hamiltonian spe trum whi h
is interpretable in terms of individual energy levels of separate parti les. In view of this
extreme narrowness of gaps between energy levels of a ma ros opi system, any external
perturbation, however weak, is asso iated with the energy transfer to or from the sys-
tem whi h is mu h mu h larger than its energy gaps and therefore the system, even if it
ma ros opi ally seems isolated, is ontinuously making innumerable transitions between
dierent energy eigenstates.
The se ond ir umstan e has its origin in the energy-time un ertainty prin iple whi h
tells that preparing a quantum system in a state in whi h its energy un ertainly is ∆E
takes at least time ∆t related to ∆E by

> ~.
∆E∆t ∼

In view of the density of energy levels of a ma ros opi body (dis ussed above) preparing
the system in a states of denite energy (i.e. ∆E smaller than typi al energy splittings)
woud require time longer than the lifetime od the Universe.
For both these reasons one should treat all ma ros opi quantum systems, even the
ma ros opi ally isolated ones, as intera ting with their surrounding and ertainly not in a
stationary state of denite energy. Therefore we now develop the formalism of the density
operator and of mixed states whi h allows to treat su h ma ros opi systems.

Mixed states and the density operator

Suppose that the world is divided into the onsidered system and its environment. Even-
though elements whi h onstitute the system and those whi h onstitute the environement
an be separately identied and therefore the Hilbert spa e H of the entire universe ( on-
sisting of the system and the environment) an be represented as the tensor produ t
H = Hsys ⊗ Henv whi h in parti ular means that as the basis H one an take ve tors

|l, Li ≡ |li ⊗ |Li , (211)

where |li ∈ Hsys and |Li ∈ Henv , the quantum states of both, of the system and the
environment, i.e. of the Universe, are usually highly entangled, that is, state-ve tors
representing in H the state of the universe must be written as a general superpositions

XX
|Ψi = |li ⊗ |Li cL,l , (212)
l L

and in general annot be fa torized, i.e. are not of the form |ψsys i ⊗ |φenv i (the oe ients
sys
cL,l are not in general of the form cL,l = cenv
L cl ).
Suppose now that in the Hamiltonian of the world Hˆ = Hˆsys + Hˆenv + Hˆint in whi h

Hˆsys ≡ Hˆsys ⊗ 1̂env and Hˆenv ≡ 1̂sys ⊗ Hˆenv (whi h means that they a t essentially
ea h in only one spa e of the tensor produ t) the intera tion term is small and we are
interested in an observable measured on the system and not on the environment and

142
represented therefore by an operator of the sort Ô ≡ Ôsys ⊗ 1̂env . We would like to express
a quantum me hani al expe tation value of Ô in a state of the form (212) entirely by
obje ts pertaining to Hsys . This an be done with the help of the density operator
whi h is introdu ed in the following way. We start by writing
! !
X X X X
hΨ|Ô|Ψi = c∗L′ ,l′ hL′ | ⊗ hl′ | Ôsys |li ⊗ |Li cL,l = hl′ |Ôsys |li cL,l c∗L,l′ . (213)
l′ ,L′ l,L l′ ,l L

We have used the way Ô a ts on the state-ve tors of the produ t basis of the omplete
Hilbert spa e H:

Ô(|li ⊗ |Li) = Ôsys ⊗ 1̂env (|li ⊗ |Li) = (Ôsys |li) ⊗ (1̂env |Li) = (Ôsys |li) ⊗ |Li ,
95
and the orthogonality hL′ |Li = δL′ L of the basis of Henv . The statisti al operator ρ̂
a ting in Hsys is now dened by giving its matrix elements between the basis ve tors |li
of Hsys
X
(Ψl |ρ̂Ψl′ ) ≡ hl|ρ̂|l′ i ≡ cL,l c∗L,l′ . (214)
L

Using this operator the expe tation value (213) an be written in the form

X
hΨ|Ô|Ψi = hl|ρ̂|l′ ihl′ |Ôsys |li = Tr(ρ̂ Ôsys ) , (215)
l′ l

in whi h the tra e on the right hand side is restri ted to Hsys . Dened in this way the
operator ρ̂ a ting in Hsys
is Hermitian (self-adjoint). To see this, it su es to apply the
†
denition (205) to the matrix elements of ρ̂ between the basis ve tors |li, or Ψl :

!∗
X X
(Ψl′ |ρ̂† Ψl ) = (ρ̂ Ψl′ |Ψl ) = (Ψl |ρ̂ Ψl′ )∗ = cL,l c∗L,l′ = c∗L,l cL,l′ ,
L L

and to ompare it with the matrix elements of ρ̂ given by (214):

X
(Ψl′ |ρ̂ Ψl ) = hl′ |ρ̂|li = c∗L,l cL,l′ .
L

The two right hand sides are identi al, whi h shows that ρ̂† = ρ̂.
95 The s alar produ t of ve tors ′
|ψsys i ⊗ |φenv i and |ψsys i ⊗ |φ′env i belonging to Hsys ⊗ Henv is naturally
dened as

hφ′env | ⊗ hψsys
′
| (|ψsys i ⊗ |φenv i) = hφ′env |φenv i · hψsys
′


|ψsys i .

143
 As every self-adjoint operator, ρ̂ an be diagonalized, that is there exists in Hsys a
omplete set of orthonormal basis ve tors Φk ≡ |ki whi h are eigenve tors of ρ̂ and the
orresponding eigenvalues wk are real. In this basis

(Φk |ρ̂ Φk′ ) = hk|ρ̂|k ′ i = wk δk′ k (216)

If one now takes the expe tation value (213) of the unit operator 1̂ = 1̂sys ⊗ 1̂env so that
obviously hΨ|1̂|Ψi = 1 and writes the right hand side of the formula (213) using the
ve tors |ki as the basis of Hsys (instead |li), one will get the equality
of

X X
1 = hΨ|1̂|Ψi = Tr(ρ̂) = hk|ρ̂|ki = wk . (217)
k k

Thus, the sum of the eigenvalues of the statisti al operator equals one. Another informa-
96
tion about the eigenvalues wk an be obtained by realizing that ρ̂ an be written as
X
ρ̂ = |kiwk hk| , (218)
k

and using this representation to write in two ways Tr(ρ̂ Ôsys ) with the parti ular operator
Ôsys = |kihk|, i.e. a self-adjoint proje tor onto the ρ̂ eigensubspa e orresponding to the
eigenvalue wk . On one hand then
!
X X
′′
Tr(ρ̂ Ôsys ) = hk | |k i wk′ hk | (|kihk|) |k ′′ i = wk ,
′ ′

k ′′ k′

and on the other, using the denition (214),

 2
X XX X X 
Tr(ρ̂ Ôsys ) = hl|ρ̂|l′ ihl′ |Ôsys |li = c∗L,l′ cL,l hl′ |kihk|li = cL,l hk|li ≥ 0 .


 
l ,l
′ ′ L l ,l L l

Thus, 0 ≤ wk ≤ 1. Finally, written in the basis of the density operator eigenve tors |ki
the formula (215) takes the form

X
hΨ|Ô|Ψi = Tr(ρ̂ Ôsys ) = wk hk|Ôsys |ki . (219)
k

It follows that if the system intera ting with the environment is onsidered separately,
it annot be said to be in a well-dened quantum state, alled pure state, whi h in
the Hilbert spa e an be represented by a state-ve tor whi h an always be written as

96 Indeed,

!
X X
′′ ′ ′′
hk |ρ̂|k i = hk | |ki wk hk| |k ′ i = hk ′′ |ki wk hk|k ′ i = wk′ δk′ k′′ .
k k

144
a superposition (however ompli ated) of state-ve tors of some basis. Instead, its state
must be represented by a density operator ρ̂; the system is then said to be in a mixed
state. The dieren e between the two situations is best seen by writing a pure state
|ψi (of the system) as a superposition of the basis state-ve tors |ki of Hsys whi h are
eigenve tors of ρ̂. The expe tation value of an operator Ôsys in the pure state |ψi reads

X X X
hψ|Ôsys |ψi = c∗k′ ck hk ′ |Ôsys |ki = |ck |2 hk|Ôsys |ki + c∗k′ ck hk ′ |Ôsys |ki . (220)
k′ k k k ′ 6=k

The rst term resembles the formula (219) if the fa tors |ck |2 are identied97
with wk . But
the expe tation value of Ôsys |ψi involves also the se ond term whi h has
in the pure state
no ounterpart in (219). While the pure state |ψi is always a oherent superposition of
basis states |ki with denite relative phases of the superposition oe ients ck , the mixed
state an be viewed as their in oherent superposition, with random relative phases of
its oe ients, be ause (220) goes over into (219) upon averaging (with at probability
distribution in the [0, 2π] range) over the phases ϕk of the oe ients ck = |ck | exp(iϕk )
∗ ′
(the produ ts ck ′ ck with k 6= k are removed by su h averaging). Mixed states are gener-
alization of pure states be ause every pure state an be represented in the formalism of
mixed states by the density operator ρ̂ satisfying the equality

ρ̂2 = ρ̂ , (221)

whi h means that in the basis formed by its eigenve tors the operator ρ̂ is su h that
wk2 = wk and sin e the fa tors wk are nonnegative and must sum up to unity, this implies
that only one wk = 1 and all others are zero. In other words, in the formalism of mixed
states, a pure state normally represented by the normalized to unity state-ve tor |ψi is
represented by the density operator ρ̂ = |ψihψ| whi h is simply the proje tion operator
onto the subspa e spanned in the Hilbert spa e by this ve tor (it is evident that it is in
fa t the ray, and not the state-ve tor itself, whi h determines ρ̂).
Ensembles in quantum statisti al me hani s
Thus any real system whi h intera ts with its environment, if onsidered separately, must
be at a given instant t0 treated as being in a mixed state represented by some density
operator ρ̂ whi h is diagonal in some basis |ki and takes the form (219) at some initial
moment t0 . Of ourse, as the state of the entire universe evolves with time, the density
operator ρ̂ representing in Hsys the real system hanges too but, as in the lassi al ase,
these hanges annot be determined by the Hamiltonian of the system alone; this would
require following the time evolution of the entire universe.
As in the lassi al ase, in view of this situation, one resorts to the method of statisti al
ensembles. The statisti al ensemble orresponding to a real system intera ting with he
rest of the universe is a olle tion of N absolutely isolated systems ea h in one of
the pure states |ki. To be representative for the real system, the relative numbers Nk
of systems in dierent pure states |ki in the ensemble at the initial moment t0 (whi h

97 As wk , 0 ≤ |ck |2 ≤ 1.

145
be ause the systems in the ensemble are isolated, an be taken to be t0 = 0) are su h that
(the limits Nk → ∞ is understood)
X
Nk /N = wk , where N = Nk .
k

The mean values over the ensemble of expe tation values of operators representing ob-
servables are then formally identi al to the formula (219):

1 X X
O= Nk hk|Ô|ki = wk hk|Ô|ki , (222)
N
k k

and an be written as

O = Tr(ρ̂ Ôsys ) , (223)

with the statisti al operator ρ̂ whi h at t0 is formally identi al with the density operator
98
ρ̂ of the real system. The entral problem, as in the lassi al ase, is to theoreti ally
determine the form of the statisti al operator whi h would represent the mixed state of
the real system of interest in dierent situations (when the real system is ma ros opi ally
isolated, or in thermal equlibrium with its surrounding, et .).
Be ause the systems forming the ensemble are isolated, their time evolution is om-
pletely determined by the Hamiltonian Hˆ (from now on we drop the subs ript sys)
whi h a ts only in the system's Hilbert spa e. Ea h of the basis state-ve tors |ki hanges
with time a ording to (207):

|ki → |k(t)i = exp(−(i/~)Hˆ t)|ki .

as a result
i ˆ i ˆ
X
ρ̂ → ρ̂(t) = wk e− ~ H t |kihk|e+ ~ H t .
k

Dierentiating both sides with respe t to t gives the dierential equation

dρ̂(t) i
= − [Hˆ , ρ̂(t)] , (224)
dt ~
whi h in quantum statisti al me hani s plays the same role as does the Liouville equation

∂ ρ̂(q, p, t)
= {H (q, p), ρ(q, p, t)}PB , (225)
∂t
99
in lassi al statisti al physi s.

98 Be ause of this formal identity one an take the position that the ensemble is just a single system in
an appropriate mixed state.
99 The equation (224) looks similar to the Heisenberg equation

dÔH (t) i
= [Hˆ , ÔH (t)] ,
dt ~

146
 If the real system is in equilibrium (either as a ma ros opi ally isolated system or
as a system in equilibrium with its surroundig), whi h means that ma ros opi observ-
ables measured on it do not depend on time, the statisti al operator of the representative
ensemble should be time independent, just as were the distribution fun tions ρ(q, p) or-
responding to lassi al systems in equilibrium (only then the ensemble mean values (223)
will be stationary), From the equation (224) it follows that dρ̂/dt = 0 implies

[Hˆ , ρ̂] = 0 . (226)

A ording to the usual rules of quantum me hani s this means that there exist in H a
basis |ni in whi h both operators, Hˆ and ρ̂, are simultaneously diagonal:

Hˆ =
X X
En |nihn| , ρ̂ = wn |nihn| . (227)
n n

Another information on the form of the statisti al operator orresponding to a ma ro-

s opi system in equilibrium is provided by the statisti al independen e of its ma ros opi
subsystems. If the system is omposed of two mutually nointera ting, or intera ting negli-
gibly weakly, parts its Hilbert spa e H an be taken to be the tensor produ t H = H1 ⊗H2
and the Hamiltonian is Hˆ = Hˆ1 ⊗ 1̂2 + 1̂1 ⊗ Hˆ2 (the term Hˆint being either absent or
negligible). The basis of H built out of the Hamiltonian eigenve tors |ni an be then
hosen to be of the form

|ni ≡ |n1 , n2 i ≡ |n1 i ⊗ |n2 i ,

The statisti al operator of the orresponding ensemble should then have the form ρ̂ =
ρ̂1 ⊗ ρ̂2 whi h ensures that the mean value O of an observable represented by the operator
of the form Ô1 ⊗ Ô2 , that is whi h is the produ t of observables pertaining to dierent
systems, equals O 1 O 2 :
 
O = O1 O2 = TrH ρ̂1 ⊗ ρ̂2 Ô1 ⊗ Ô2
  X
= TrH ρ̂1 Ô1 ⊗ ρ̂2 Ô2 = hn2 , n1 |ρ̂1 Ô1 ⊗ ρ̂2 Ô2 |n1 , n2 i
n n
1 2
X X    
= hn1 |ρ̂1 Ô1 |n1 i hn2 |ρ̂2 Ô2 |n2 i = TrH1 ρ̂1 Ô1 TrH2 ρ̂2 Ô2 = O1 O2 .
n1 n2

satised by time-dependent Heisenberg pi ture operators ÔH (t) whi h are outerparts of time-
independent S hrödinger pi ture operators (for more details, see Chapter 1 of my notes to quantum
eld theory), but the sign of the right hand sides are dierent; in fa t the Heisenberg pi ture ounterpart
ρ̂H of the statisti al operator ρ̂(t) is time independent. It is also worth noting that the transition from the
lassi al Liouville equation (225) to the equation (224) an be done with the usual quantization rule a -
ording to whi h in lassi al equations all c-number fun tions are repla ed by their operator ounterparts
and the Poisson bra kets are repla ed by ommutators a ording to the pres ription

i
{·, ·}PB → − [·, ·] .
~

147
The operators ρ̂1 and ρ̂2 must be both Hermitian (the rst one in H1 and the se ond one
in H2 ) and both should be diagonal in the bases formed by the eigenve tors |n1 i and |n2 i
of the respe tive Hamiltonians:

wn(12) δn′ n = hn′ |ρ̂|ni = hn′2 , n′1 |ρ̂1 ⊗ ρ̂2 |n1 , n2 i

= hn′1 |ρ̂1 |n1 i hn′2 |ρ̂2 |n2 i = wn(1)1 δn′1 n1 wn(2)2 δn′2 n2 .
(12) (12) (1) (2)
In other words, wn ≡ wn1 ,n2 = wn1 wn2 , or

ln wn(12) ≡ ln wn(12)
1 ,n2
= ln wn(1)1 + ln wn(2)2 , (228)

whi h, sin e energies of the two (nonintera ting or intera ting negligibly weakly with
one another) parts of the system are additive, En1 n2 = En1 + En2 and in general the
probabilities an be viewed as fun tions of energies: wni = w(Eni ), means that

ln w(En ) = α − βEn , and ln w(Ena ) = αa − βEna , (229)

as this is the only way of satisfying the requirement (228) if En ≡ En1 n2 ,... = En1 +En2 +. . ..
100
All this an be on isely written in the form

ln ρ̂(12) = (ln ρ̂(1) ) ⊗ 1̂2 + 1̂1 ⊗ (ln ρ̂(2) ) . (230)

This imposes a strong onstraint on statisti al operators of ma ros opi systems and will
be ru ial in deriving the general formula for entropy.

The quantum mi ro ani al ensemble

We now onsider a real system whi h ma ros opi ally an be treated as isolated from
its surrounding and in equilibrium. Again, appealing to the prin iple of a priori equal
probabilities, we postulate that the statisti al operator ρ̂ of the representative ensemble
(of stri tly, that is, also at the mi ros opi level, isolated systems) has the form

X
ρ̂ = const · |nihn| , (231)

n
E ≤ En ≤ E + ∆E

where |ni are the eigenve tors of the ensemble systems Hamiltonian (that is, of the Hamil-
tonian of the real system from whi h the terms orresponding to its intera tion with the
rest of the universe have been removed). In (231) it has been taken into a ount that as
a result of its mi ros opi residual intera tion with the surrounding as well as a onse-
quen e of the mentioned restri tion imposed by the general quantum me hani al un er-
tainty prin iple, the energy of a real ma ros opi ally isolated system an be spe ied only
up to some toleran e ∆E (of ourse ∆E ≪ E , in the ase of a large system). Of ourse,

100 Re all that a fun tion of an operator is dened by giving matrix elements of this operator fun tion
in the basis in whi h the operator is diagonal.

148
sin e Tr(ρ̂) = 1, the onstant in the denition (231) must be given by (the elipses stand
for other ma ros opi variables like magnetization, et . whi h hara terize the system)

 
−1 number of quantum states
const = ≡ Γ(E, V, . . . , N, ∆E) . (232)
in the interval [E, E + ∆E]

Sin e the energy levels of the system are dis rete, and the volume V is nite, the number of
101
states orresponding to the energy interval [E, E + ∆E] should be nite. Together the
formulae (231) and (232) dene the quantum mi ro anoni al ensemble whi h should
be representative for systems ma ros opi ally isolated and in equilibrium. The form (231)
of the statisti al operator ρ̂ is, of ourse onsistent with the general requirements (226)
and (229).
As in the ase of the lassi al mi ro anoni al ensemble, the statisti al entropy of a
ma ros opi ally isolated system in equilibrium is dened to be

Sstat = kB ln Γ(E, V, . . . , N, ∆E) . (233)

As there, it a measure of disorder of a given equlibrium ma rostate quantied in terms of

the (logarithm of the) number of quantum mi rostates realizing that ma rostate (i.e. in
the Callenian language, onsistent with te ma ros opi onstraints the system is subje ted
to). Noti e, however, that in ontrast to the lassi al ase, the quantity Γ(E, V, . . . , N, ∆E)
is within the quantum mi ro anoni al ensemble dened in absolute terms, without any
arbitrary multipli ative onstant whi h implies that entropy is also dened without any
arbitrary additive onstants. As has been explained, the arbitrarines in the lassi al ase
is removed by requiring that results of the lassi al and quantum aproa hes mat h in the
appropriate limit.
As in the lassi al ase, in addition to Γ(E, V, . . . , N, ∆E) one an introdu e the
quantity Σ(E, V, . . . , N) whi h here is just the number of quantum states of energies En ≤
E and, owing to the fa t that the spe trum of the Hamiltonian of a ma ros opi system,
while being formally dis rete, is almost ontinuous, to dene the density ω(E, V, . . . , N)
of states (around En = E ) by

∂Σ(E, V, . . . , N)
ω(E, V, . . . , N) = , (234)
∂E
so that the relation (201) between Γ(E, V, . . . , N, ∆E), ω(E, V, . . . , N) and ∆E holds true
also in the quantum ase. Then, again as in the lassi al ase, one an dene two other
statisti al entropies repla ing in (233) Γ(E, V, . . . , N, ∆E) either by Σ(E, V, . . . , N) or by
the density ω(E, V, . . . , N) of states. In the thermodynami al limit, if the energy spe trum

101 Here it is important that the theory is formulated in the nite volume: for instan e, to the range
[−ε, 0] (with arbitrarily small positive ε) there orrespond innitely many dis rete energy eigenstates of
2 2 2
the Hydrogen atom Hamiltonian (re all that Enlml ,sz = −mc αEM /2n and the degenera y of the energy
2
levels is 2n ), but this is true only if the quantum me hani s of the Hydrogen atom is formulated in the
innite spa e; if the volume of the spa e is taken nite, the spe trum gets modied and the number of
states in the range [−ε, 0] is nite (I.I. Oppenheim and D.R. Hafeman, J.Chem.Phys. 39 (1963), 101).

149
of the system is typi al, all the three denitions of Sstat lead to the same thermodynami al
entropy
 
Sstat
STMD = N lim , (235)
∞ N
or, in ases in whi h the number of parti les annot be dened (e.g. be ause the system
is a relativisti quantum eld)
 
Sstat
STMD = V lim . (236)
∞ V
Before we demonstrate that the statisti al entropies dened by (233) have (in the
thermodynami al limit) the properties of the entropy dened within thermodynami s, we
will onsider the problem of dening the statisti al entropy of systems in equilibrium (but
not ne essarily isolated) in general. Owing to the uniform notation introdu ed, this an
be done jointly for the lassi al and the quantum ase.

The golden formula for entropy

Dening the statisti al entropy - a measure of the mi ros opi disorder whi h is asso i-
ated with a given equilibrium ma rostate - by the formulae (202) and (233) in terms of
the phase spa e volume or the number of quantum states in the lassi al and quantum
ases respe tively, was possible be ause the mi ro anoni al ensemble distribution fun -
tion ρ(q, p) in the lassi al ase is lo alized (has a nite support in the mathemati al
language) and the probabilities wn in the quantum ase are nonvanishing only for a nite
number of states. If the real system is in equilibrium with its surrounding (e.g. with a
heat bath), the distribution fun tion (in the lassi al ase) or the probabilities wn (in the
quantum ase) of the statisti al ensemble representing the real system do not have this
simple property and the quantity Γ(E, V, N, ∆E) employed in the denitions (202) and
(233) is ill dened (unlike the fun tions Σ(E, V, N) and ω(E, V, N) whi h are always nite
but aanot be dire tly used to dene entropy). Consequently one has to invent another
measure of disorder in terms of whi h to dene entropy.
In the general ase to introdu e the analog of the quantity Γ(E, V, N, ∆E), one an
resort to the following reasoning (Landau & Lifs hitz). One onstru ts rst the distribu-
tion fun tion ρE (E) of the random variable E dened on the spa e of elementary events.
In the lassi al ase it is dened using the pres ription (187) whi h here, be ause when
the onsidered system is in equilibrium ρ(q, p) = ρ(H (q, p)), gives
Z
ρE (E) = dΓ(q,p) ρ(H (q, p)) δ(H (q, p) − E)
Z
= ρ(E) dΓ(q,p) δ(H (q, p) − E) = ρ(E) ω(E) . (237)

In the quantum ase one an dene the distribution ρE (E) by the equality
X
ρE (E) dE = wn . (238)

n
E ≤ En ≤ E + dE

150
 P R
Sin e n wn = 1, the distribution ρE (E) dened in this way is normalized: dE ρE (E) =
1. On the other hand, owing to the quasi- ontinuity of the spe trum, the number of
quantum states in the interval [E, E + dE] is equal to ω(E)dE , and the probabilities
wn = w(En ) an be treated as a ontinuous fun tion w(E) of the system's energy. This
allows in the quantum ase to write

ρE (E) = w(E) ω(E) . (239)

The two expressions: the lassi al one (237) and the quantum one (239) be ome therefore
identi al if w(E) is identied with ρ(E).
If the onsidered system is large, the distribution ρE (E) is nonzero only in the lose
vi inity of a value E ∗ whi h is pra ti ally the same as the mean E omputed using
ρE (E). In other words, ρ(E) has a Dira delta-like peak pra ti ally at E = E. It is
therefore possible to hara terize the mi ros opi spread of an equilibrium ma rostate of
the system over the phase spa e in the lassi al ase and over the Hamiltonian spe trum
in the quantum ase in terms of the width ∆E of its energy distribution dened by the
relation

ρE (E) ∆E = 1 . (240)

This allows to dene the ee tive number of mi rostates (that is, the number of those
mi rostates whi h are really relevant in the ensemble) Γ(E, V, N) by the equality (in the
lassi al ase repla e w(E) by ρ(E))
1 = w(E) ω(E) ∆E = w(E) Γ(E, V, N, ∆E) , (241)

and then the statisti al entropy by

Sstat = kB ln Γ(E, V, N, ∆E) = kB ln[ω(E) ∆E] . (242)

It is easy to see that in the ase of the miro anoni al statisti al operator (231) or the
distribution fun tion (197) the denitions (233) and (202) oin ide with (242). Indeed,
to onsider only the quantum ase, in this ase w(E) = w(E) = 1/Γ(E, V, N, ∆E), whi h
implies the equality of Γ(E, V, N, ∆E) and Γ(E, V, N, ∆E).
Exploiting now the denition (241) of the ee tive number of mi rostates, the formula
(242) for entropy an be ast into the form

Sstat = −kB ln w(E) , (243)

It has been argued, however, that statisti al independen e of subsystems implies that
the logarithm of w(E) must be at most a linear fun tion of energy, ln w(E) = α − βE
(ln ρ(H ) = α − βH , lassi ally). This means that
ln w(E) = ln w(E) ,
whi h allows to write the formula (242) in the form

Sstat = −kB Tr(ρ̂ ln ρ̂) , (244)

151
in the quantum ase and as
Z
Sstat = −kB dΓ(q,p) ρ(q, p) ln ρ(q, p) (245)

in the lassi al ase. It is lear that in the ase of the lassi al mi ro anoni al ensemble,
when ρ(q, p) = const. = 1/Γ(E, ∆E), and

1 1
Z Z
dΓ(q,p) ρ(q, p) ln ρ(q, p) = ln dΓ(q,p) ,
Γ(E, ∆E) Γ(E, ∆E) E≤H ≤E+∆E

(245) redu es to (202). An analogous reasoning readily shows that (244) redu es to (233)
.
Expressions (244) and (245) are the general, truly golden, formulae for entropy valid
independently of the ensemble. Their virtue is that they an be (after some modi ations)
extended also to nonequilibrium situations (when ρ̂(t) of ρ(q, p, t) are not independent of
time), although we will not osider this extension in this Course.

152
 LECTURE XII (STAT)

Classi al energy equipartition theorem

One general onsequen e of the lassi al statisti al approa h applied to ma ros opi sys-
tems in equilibrium, whi h is independent of whether the system is ma ros opi ally iso-
lated or in equilibrium with its environment, is the equipartition of its mean energy
between its mi ros opi degrees of freedom. Here we demonstrate it using the lassi al
mi ro anoni al ensemble distribution fun tion (197).
We begin by omputing the ensemble average

∂H 1 i ∂H
Z
qi = dΓ (q,p) q
∂q j Γ(E, ∆E) E≤H ≤E+∆E ∂q j
∆E ∂ ∂H
Z
≈ dΓ(q,p) q i .
Γ(E, ∆E) ∂E H ≤E ∂q j

We have used the standard tri k

Z Z Z
dΓ(q,p) f (q, p) = dΓ(q,p) f (q, p) − dΓ(q,p) f (q, p)
E≤H ≤E+∆E H ≤E+∆E H ≤E
∂
= F (E + ∆E) − F (E) ≈ ∆E F (E) .
∂E
This an be also written as
 
∂H ∆E ∂ ∂  i
Z
i

qi j
≈ dΓ(q,p) q (H − E) − δ j (H − E) .
∂q Γ(E, ∆E) ∂E H ≤E ∂q j
j
The rst term under the integral (the one with the derivative ∂/∂q ) will then give zero
j
be ause the integral over dq whi h is part of the measure dΓ(q,p) an be written as the
boundary term but at the boundary, whi h is determined by the ondition H = E,
102
vanishes the expression under the derivative. The derivative ∂/∂E an be then put
under the integral (again, dierentiating with respe t to the dependen e on E of the
integration boundary would give the integrand evaluated at these boundary where the
integrand vanishes) leading to the result

∂H ∆E ∆E Σ(E) i
Z
qi ≈ dΓ(q,p) δ i j = δ i j Σ(E) = δ .
∂q j Γ(E, ∆E) H ≤E ω(E)∆E ω(E) j
102 More pre isely, the domain of the integrations over the variables q i should be ee tively restri ted by
introdu ing a smooth boundary potential V (q 1 , . . . , q 3N ) whi h is almost zero if all q i 's are in the volume
V and tends to innity if any of these variables is outside V ; then the limit

0 if all q i ∈ V

V → ,
∞ otherwise

should be taken.

153
(The relation (201) has been used here.) Using the relation ω(E) = ∂Σ(E)/∂E , this
result an be also written as
 −1
∂H ∂ ln Σ(E)
qi = δi j .
∂q j ∂E

Sin e entropy S within the Mi ro anoni al Ensemble an (from the thermodynami al

limit point of view) be dened in terms of the logarithm of Σ(E), using the general
thermodynami al relation (∂S/∂E) = 1/T , we obtain

∂H
qi = −q i ṗj = kB T δ i j . (246)
∂q j
103
Exa tly analogous omputation gives the result

∂H
pi = pi q̇ j = kB T δi j . (247)
∂pj

If the Hamiltonian has a regular form, e.g.

K f
1X X
H = Ai p2i + Bj qj2 ,
2 i=1 j=1

with some onstant Ai and Bj , so that

K f
1 X ∂H 1 X j ∂H
H = pi + q ,
2 i=1 ∂pj 2 i=1 ∂q j

104
then results (246), (247) imply that the mean system energy is related to its temperature
by

1
U ≡ H = (K + f )kB T , (248)
2
independently of the details of the system's mi ros opi dynami s. For instan e this
immediately implies that the heat apa ity cv per mole ule of a lassi al monoatomi
3
perfe t gas is kB (f = 0, K = 3N , so three degrees of freedom per mole ule) and of a
2
7
two-atomi gas kB (K = 6N , f = N , so 7 degrees of freedom per mole ule). Similarly
2
the heat apa ity of the rystal latti e of a solid built out of N mole ules and behaving

103 A tually even less ompli ated be ause one does not need to appeal to an arti ial boundary potential.
104 In the ase of the mi ro anoni al ensemble representative for a ma ros opi ally isolated system with
xed energy E (within the toleran e ∆E whi h in the lassi al ase an be set equal zero) the mean
energy H ≡ U is the same as E ; the result is valid however also if the system is in equilibrium but does
not have xed energy.

154
 105
as 3N independent harmoni os illators should be CV = 3NkB (K = 3N , f = 3N ;
this is the so- alled Dulong-Petit law) irrespe tively of the distribution of frequen ies of
these os illators.These results are heuristi ally formulated as the rule that lassi ally
1
ea h degree of freedom ontributes the amount kB T of energy to the mean system's
2
total energy, but it is lear that the impli ation of the general results (246), (247) for
106
the total energy depend on the pre ise form of the intera tions. Needless to say, that
the predi tions of the equipartition theorem for the heat apa ities exemplied above
(whi h are learly at varian e with 3TMDL) are experimentally veried only at su iently
high temperatures, at whi h genuine quantum ee ts (ex ept for the ones related to the
indistinguishability of identi al parti les) are unimportant and systems an be treated as
lassi al.

Canoni al (Gibbs) Ensemble

The Mi ro anoni al Ensemble introdu ed in the pre eding Le tures and appli able to
ma ros opi ally isolated systems is on eptually simple but very in onvenient in pra ti al
omputations. Mu h more onvenient in this respe t is the Canoni al Ensemble whi h
formally is representative for a system remaining in thermal equilibrium with its envi-
ronment (i.e. in thermal onta t with it through a diathermal wall) modeled by a large
(in the limit innitely large) heat bath of xed temperature T. Sin e in most ases, as
far as mean values of system's hara teristi s are on erned, that is from the point of
view of thermodynami s, there should be no dieren e between an isolated system whose
temperature (dened as (∂S/∂U)V,N,... ) is T and the same system ex hanging energy (in
the form of heat) with a heat bath at temperature T, the Canoni al Ensemble is in most
ases (at least lassi ally) the preferred way to perform a tual omputations also in the
ase of ma ros opi ally isolated systems.
To derive the Canoni al Ensemble we onsider the system and the heat bath (repre-
senting its environment) as a ompound ma ros opi ally isolated system and apply to it
the Mi ro anoni al Ensemble. A ording to the rules of the probability theory, the prob-
ability that the total energy Etot of the ompound system (xed up to small un ertainty
∆E ) is distributed as U and Etot − U between the system and the heat bath is given by
the number of mi rostates (of the ompound system) orresponding to su h a distribution
to the total number of all mi rostates orresponding to the total energy Etot :
1
P (U, dU) = ∆E ωsys (U) ωh.b. (Etot − U) dU , (249)
Γ(Etot , Nh.b. , Nsys , . . . , ∆E)
where
Z Etot
Γ(Etot , Nh.b. , Nsys , . . . , ∆E) = ∆E dU ωsys (U) ωh.b. (Etot − U) . (250)
0
105 A tually as 3N − 6 os illators be ause 6 generalized variables orresponding to translations and
rotations of the solid as a whole are eliminated if it is kept at rest; however 3N − 6 is pra ti ally the same
as 3N .
106 One an also remark that the notion of a degree of freedom, fairly lear in me hani s, be omes, as
one enters deeper into theoreti al physi s, more and more elusive and ultimately is used without any
on rete ontent.

155
Using the Mi ro anoni al Ensemble denition S(E) = kB ln(ωh.b (E)∆E) of the entropy
of the heat bath, the probability P (U, dU) an be written as

1
P (U, dU) ∝ dU ωsys (U) exp Sh.b (Etot − U) .
kB
Expanding now
 
∂Sh.b. (Etot )
Sh.b. (Etot − U) = Sh.b. (Etot ) − U + ..., (251)
∂Etot Nh.b. ,...

and negle ting all the terms of higher order in U than the rst one on a ount of the
fa t that in the limit of the very large heat bath the probability P (U, dU) of distributions
of energy markedly dierent than ones in whi h U/Etot ∼ Nsys /Nh.b. is negligible, while
Sh.b. ∼ Nh.b. and Etot ∼ Nh.b. (the n-th derivative of Sh.b with respe t to energy is then
1−n
∼ Nh.b. and the terms of higher order in U than the rst one are suppressed with respe t
to it by powers of Nsys /Nh.b. ), one an (in the limit of innitely large heat bath) write the
Canoni al Ensemble distribution ρU (U) of the system energy in the form

ρU (U) = Const. ωsys (U) e−U/kB T ≡ Const. ωsys (U) e−βU . (252)

We have introdu ed here the traditional symbol

1
β≡ . (253)
kB T
107
The normalization onstant is of ourse given by
Z ∞
Const. −1
= dU ωsys (U) e−U/kB T .
0

The Canoni al Ensemble energy distribution (252) is independent of whether the sys-
tem is treated lassi ally or quantum me hani ally. The orresponding phase spa e distri-
bution fun tion ρ(q, p) of the lassi al Canoni al Ensemble representative for the system
in thermal equilibrium with a heat bath at temperature T an be obtained by marginal-
ization (in the language of statisti s) of the joint distribution fun tion ρ(q, p, qh.b. , ph.b. ) of
the mi ro anoni al ensemble representative for the system and the heat bath as a single
ompound, ma ros opi ally isolated system:

1
Z
ρ(q, p) = dΓ(qh.b. ,ph.b.) ρ(q, p, qh.b , ph.b. )
Γ(Etot , Nh.b. , Nsys , . . . , ∆E)
 
1
Z
∝ dΓ(qh.b. ,ph.b.) = exp Sh.b. (Etot − Hsys (q, p)) .
Etot −Hsys ≤Hh.b. ≤Etot +∆E−Hsys kB
107 To make it lear: the probability distribution dened by (249) and (250) is normalized by onstru -
tion. However, on e the expansion (251) has been trun ated, the tails of the distribution are modied
(insigni antly from the pra ti al point of view) and its normalization must be readjusted.

156
It has been assumed here that the Hamiltonian of the ompound system is the sum
Hsys + Hh.b. ; in other words its part Hsys−h.b. - the intera tion of the system with the
heat bath ensuring in fa t the ssumed thermal onta t between them - has bee assumed
108
to be negligible. Expanding Sh.b. (Etot − Hsys (q, p)) in the power series in Hsys (q, p),
reje ting terms of the expansion of order higher than the rst one (again, on a ount of
the fa t that they ae t the onstru ted distribution fun tion ρ(q, p) vanishingly little -
i.e. insigni ntly from the pra ti al point of view - in the limit of innitely large heat
bath) and in luding the rst term of the expansion into the normalization onstant Zstat ,
one obtains the distribution fun tion ρ(q, p) of the Canoni al Ensemble in the form

1
ρ(q, p) = exp{−Hsys (q, p)/kB T } . (254)
Zstat
Its normalization fa tor
Z
Zstat (T, V, . . . , N) = dΓ(q,p) exp{−Hsys (q, p)/kB T } . (255)

whi h is a fun tion of the temperature T, the system's volume V the number N (num-
bers N1 , . . . , Nr ) of parti les onstituting it and possibly other ma ros opi parameters
(denoted by elipses) hara terizing the system is alled the Canoni al Statisti al Sum
or the Canoni al Partition Fun tion. By introdu ing under the integral in (255) the
R
unity written as 1 = dE δ(E − H (q, p)) and ex hanging the order of integrations, the
anoni al partition fun tion an also be written in the alternative form as
Z ∞
Zstat (T, V, . . . , N) = dE ω(E) exp{−E/kB T } . (256)
0
As will be shown below, it ontains the omplete thermodynami information about the
system.
The derivation of the statisti al operator ρ̂ of the quantum Canoni al Ensemble rep-
resentative for the system in thermal equilibrium with the heat bath is as follows. The
basis of the Hilbert spa e of the entire isolated ompound system H = Hsys ⊗ Hh.b. is
formed by the state ve tors of the form |li ⊗ |lh.bi whi h are assumed to be eigenve tors
of the ˆ
Hamiltonian Hsys ⊗ 1̂h.b. + 1̂sys ⊗ Hh.b. ; the intera tion term Hh.b.−sys ensuring the
thermal onta t between the system and the heat bath is, as in the lassi al ase and
with the same justi ation, assumed to be negligible. Hen e, energies of the two parts
of the system are additive. The statisti al operator of the orresponding Mi ro anoni al
Ensemble therefore reads (to make the notation easier we rename Etot to E)
1 X
ρ̂Micro = (|li ⊗ |lh.b. i)(hlh.b. | ⊗ hl|)
Γ(E . . . , ∆E)
lh.b. , l
E ≤ Elh.b. + El ≤ E + ∆E
108 In fa t, sin e the role of the heat bath is auxiliary only, this an freely be assumed. In onne tion
with this it should be re alled that the systems forming the ensemble are isolated - only the real system,
of whi h the systems of the ensemble are supposed to be representative, is in onta t with the heat bath
but it is assumed to have rea hed the equilibrium with the heat bath before it starts to be represented
by the ensemble; so rea hing this equilibrium might have taken arbitrarily long time.

157
 1 X X
= |lihl| ⊗ |lh.b. ihlh.b. |
Γ(E . . . , ∆E)
l lh.b.
El ≤ E + ∆E E − El ≤ Elh.b. ≤ E + ∆Etot − El
The Canoni al Ensemble statisti al operator ρ̂ is now obtained by taking the tra e of ρ̂Micro
109
with respe t to the Hilbert spa e Hh.b. of the heat bath. The sum over lh.b. restri ted to
the heat bath states of energies between E − El and E + ∆E − El , where El is the energy
of the system yields then just the fa tor Γ(E − El , Nh.b. , . . . , ∆E) = exp(Sh.b. (E − El )/kB .
This, upon expanding up to rst order in El and in luding the zeroth order term in the
110
normalization, leads to the nal result

1 X 1  
ρ̂ = |lihl| exp(−El /kB T ) ≡ exp −Hˆ /kB T , (257)
Zstat l
Zstat

with the normalization fa tor (quantum Canoni al Ensemble partition fun tion) Zstat
given by
  X
ˆ ˆ
Zstat = Tr e−H /kB T = hl|e−H /kB T |li
l
X X
= exp(−El /kB T ) = dn exp(−En /kB T ) . (258)
l En

The rst sum in the se ond line is over the system's Hamiltonian eigenve tors |li and the
se ond one is over the energy levels in luding their degenera y fa tor dn . This se ond form
of the formula is the quantum ase ounterpart of the lassi al formula (256). It should be
noted that both results (254) in the lassi al ase and (257) have the forms (195) and (229),
111
respe tively, established by onsidering general properties of ma ros opi systems.

We will now show that if the system is simple in the thermodynami al sense, that is
only volume work an be done on it, the statisti al sum given by (255) or (256) in the
lassi al ase and by (258) in the quantum ase, is dire tly related to the free Helmholtz
energy F whi h is already known to be the proper thermodynami potential hara terizing
a system kept at onstant temperature by thermal onta t with a heat bath. This readily
follows from the golden formulae (245) or (244). Indeed, onsidering for illustration the
lassi al ase, applied to (254) it yields:

1
S = −kB ln ρ = −kB (−H /kB T − ln Zstat ) = H + kB ln Zstat .
T
109 More pre isely, its matrix elements hl′ |ρ̂|li are identied with
X
hlh.b. | ⊗ hl′ |ρ̂Micro |li ⊗ |lh.b. i .
lh.b.

110 Justi ation of the reje tion of all higher order terms of the expansion is the same as in the lassi al
ase.
111 Noti e also that the symbol β in the forms (195) and (229) has now a quired the interpretation (253).

158
After a small rearrangement and identifying H with the system's internal energy U, this
means that (re all, that F = U − T S)

F (T, V, N) = −kB ln Zstat (T, V, N) . (259)

Obviously this formula stays valid in the quantum ase (the steps are in that ase om-
pletely analogous).
The rule (259) is also valid if the system onsists of more than one material omponent
- simply N should be repla ed by the numbers N1 , . . . , Nr of dierent kinds of parti les.
Of ourse, if a multi omponent system onsisting of r groups of identi al (hen e, quantum
me hani ally indistinguishable) parti les is analysed lassi ally, the measure over phase
spa e is

r
Y d3Na qa d3Na pa
dΓ(q,p) = .
a=1
Na ! (2π~)3Na

As a result, the Canoni al Ensemble formalism applied to a mixture of dierent lassi al

perfe t gases readily leads to the Gibbs Ansatz (157) and, as dis ussed in Le tures VIII
and XI, it is the produ t of fa tors N1 ! . . . Nr ! whi h leads to the mixing entropy.
As already dis ussed, on e the distribution fun tion ρ(q, p) in the lassi al ase or
the statisti al operator ρ̂ in the quantum ase are given, mean value of an observable O
(represented by a phase spa e fun tion O(q, p) in te lassi al ase and by a Hermitian
operator Ô in the quatum ase) an be omputed as

1
Z Z
O= dΓ(q,p) O(q, p) ρ(q, p) = dΓ(q,p) O(q, p) e−H (q,p)/kB T ,
Zstat
1 
−Hˆ/kB T

O = Tr(Ô ρ̂) = Tr Ô e , (260)
Zstat
in the lassi al and quantum ases, respe tively. Similarly an be omputed also its mean
quadrati u tuation (189) to estimate its relative u tuation (190). Sin e energy of a
system remaining in thermal equilibrium with a heat bath is not xed, it is of interest
to examine the possible magnitude of its u tuations. As the Canoni al Ensemble dis-
tribution fun tion ρ(q, p) or the statisti al operator ρ̂ are expressed through the system's
Hamiltonian, the formulae for the mean system's energy U and its u tuation an be ob-
tained in general terms without, using the expli it forms of these Hamiltonians. Indeed,
from the formulae (260) it follows that (for denitess we onsider the quantum ase -
lassi al al ulation is ompletely analogous)

1   1 ∂  
−Hˆ /kB T −β Hˆ 
U =E= Tr Hˆ e =− Tr e
Zstat Zstat ∂β 
β=1/kB T

∂ 
=− ln Zstat  . (261)
∂β β=1/kB T

159
Using the formula (259) and noti ing that ∂/∂β = −kB T 2 ∂/∂T this an be also written
in the form
 
2 ∂ F
U = −T ,
∂T T V,N

whi h is just the standard thermodynami identity (145). The mean quadrati u tuation
of U an be expressed in the similarly general way. To this end one starts with (we suppress
the instru tion to set β = 1/kB T at the end)

2
1 
ˆ 2 −Hˆ /kB T
 1 ∂ 
ˆ −β Hˆ

E = Tr H e =− Tr H e
Zstat Zstat ∂β
1 ∂
∂E ∂ ∂E 2
=− Zstat E = − −E ln Zstat ≡ − +E .
Zstat ∂β ∂β ∂β ∂β
It follows that
 
2∂E ∂U
σU2 ≡ E2 − E = − = kB T 2 = kB T 2 CV . (262)
∂β ∂T V,N

Hen e, in omplete generality, the mean quadrati u tuation of the system's energy is
determined by its heat apa ity (at onstant volume) CV . Sin e if the system is ma ro-
s opi ,
√ CV = Ncv and U = Nu, the relative u tuation of its energy is suppressed by
1/ N
√
p
σU2 1 p
= NkB T 2 cV ∼ 1/ N ,
U Nu
in agreement with the general result (193).
It is also worth to ask what thermodynami al potential is related to the Canoni al
Ensemble statisti al sum Zstat if the system is not simple but possesses e.g. magneti
properties. Let the (quantum) Hamiltonian of the system have the form (the subs ript
int stands here for internal)

N
Hˆ = Hˆint −
X
µ̂i ·H ,
i=1

where the se ond term is the oupling of magneti moments µi of the system's individual
elements (mole ules, for instan e) to an external magneti eld H through the operators
112
µ̂i representing them. Assuming su h a system to be in equilibrium with a heat bath

112 In su h a setting there is no distin tion between the magneti eld H0 produ ed by the experimental
setup - by a urrent passing through a oil, for instan e - in the absen e of the magneti material and
the a tual magneti eld H0 when the magneti material is present, be ause the magneti moments µ
of the system's elements are impli itly assumed not to produ e any magneti eld by themselves. Their
mutual oupling, whi h in reality o urs mostly through the magneti eld they produ e, is in statisti al
physi s models usually taken into a ount in terms of their dire t  onta t intera tions of the form
Hˆint ).
P
− i6=j Jij µ̂j · µ̂j (whi h, if present, are here in luded in

160
at temperature T and applying the golden formula (244) to the Canoni al Ensemble
representing it we get
!!
1 X
S = −kB − Eint − µi ·H + kB ln Zstat
kB T i
1 1 X
= Eint − H· µi + kB ln Zstat ,
T T i
P
Identifying now Eint with the system's internal energy U and i µi withe the system's
total magnetization M, one obtains

−kB T ln Zstat (T, V, H, N) = U − T S − M·H ,

113
whi h is the magneti Gibbs fun tion G(T, V, H, N) whose dierential is

dG = −SdT − p dV − M·dH + µ dN .

The dierential of the internal energy is then as usually (in this Course)

dU = T dS − p dV + H·dM + µ dN .

The reader should be warned, however, that there is another s hool of authors whi h
in lude the intera tion of magneti moments into what they all the system's internal
energy (let denote it Ũ ); the statisti al sum is then related to the fun tion F

−kB T ln Zstat (T, V, H, N) = F = Ũ − T S ,

This is however not a mere hange of the notation (using F in pla e of G): sin e

Ũ = U − M·H ,

the dierential of the internal energy Ũ is

dŨ = T dS − p dV − M·dH + µ dN .

so that the elementary work of magnetization is now −M·dH (if Ũ is used the Pippard's
derivation of the elementary work must be modied). In any ase the golden formula
always allows to properly identify thermodynami al quantities related to the used ensem-
ble.

Appli ation of the quantum Canoni al Ensemble

Computing the partition fun tion and the mean values of observables

ˆ
 1 
ˆ

Zstat = Tr e−H /kB T , O= Tr Ô e−H /kB T , (263)
Zstat
113 The hemi al potential µ should not be onfused with the magneti moment µ. It is lear that the
SI system of units whi h brings in yet another µ0 (the magneti sus eptibility of va uum) is utterly
in onvenient here!

161
of the quantum Canoni al Ensemble requires spe ifying the Hilbert spa e of the system
of interest in whi h a t the operators Hˆ , ρ̂, Ô. If the system is, as usually is the
ase, omposed of N parti les of some kind (we assume properties of these parti les,
like masses, spins, magneti moments are known), quantum states of the world an be
114
spe ied in terms of quantum states of individual parti les. The Hilbert spa e of
the system an be then onstru ted as a tensor produ t of single-parti le Hilbert spa es
spanned by state-ve tors representing states of individual parti les. That is, one rst
spe ies a basis of state-ve tors representing quantum states of a single parti le taking for
instan e generalized eigenve tors |xi of the position operator,115 |pi of the
or eigenve tors
momentum operator, or if the pariti le has a nonzero spin the ve tors |p, σi where σ is the
spin proje tion onto a hosen axis (usually the z -axis). We will denote these basis ve tors
|li (understanding the a single label l may stand for a ouple of independent labels). The
ve tors |li i thus span the single-parti le Hilbert spa e Hi of the i-th parti le of the system.
The Hilbert spa e H(N ) of the entire system is then onstru ted as H(N ) = H1 ⊗ . . . ⊗ HN ,
that is, it spanned by the basis state ve tors of the form

|l1 i ⊗ . . . ⊗ |lN i . (264)

Most of physi al systems however are omposed of identi al indistinguishable parti les
(or several groups of identi al parti les but for deniteness we will onsider only one type
of identi al parti les - extension to several groups is more or less straightforward) and in
this ase the rules of quantum me hani s whi h in the ourse of its developement were
abstra ted from the experimental fa ts di tate that the physi al states of su h systems are
represented not by arbitrary supepositions of state-ve tors (264) but only superpositions
of totally antisymmetrized state-ve tors (if the identi al parti les have half-integer spin,
that is, are fermions) or superpositions of only totally symmetrized state-ve tors (if the
identi al parti les have integer spin, that is, are bosons). The roots of this rule are in
spe ial relativity - in four spa e-time dimensions states of half-integer spin parti les must
be antisymmetri while those of bosons must be symmetri if relativisti ally invariant
quantum theories of their intera tions, satisfying the so- alled lo al ausality require-
ments, are to be onstru ted; in two spa e dimensions other symmetry properties are also
possible and the orresponding parti les are alled anyons. Therefore the Hilbert spa es
of a system of N fermions or of N bosons are spanned by the following basis ve tors:

1 X
√ (−1)P |lP (1) i ⊗ . . . ⊗ |lP (N ) i ≡ |l1 , l2 , . . . , lN i fermions ,
N! P
114 One an sometimes meet the statements to the ee t that the onstru tion of the Hilbert spa e of
the system as a tensor produ t of individual Hilbert spa es of its elements has something to do with the
assumption that mutual intera tions of these elements are (negligibly) weak. This is wrong. The hoi e
of the Hilbert spa e in whi h quantum me hani s of a given system is realized is something whi h belongs
to physi s and goes beyond mathemati s; one has rst physi ally de ide what are the possible quantum
states (in the abstra t sense) of the system - what states of its individual elements an be physi ally
identied - and only then model them by the hoi e of the appropriate Hilbert spa e in whi h the a tion
of operators is realized.
115 These better be avoided as the position operator does not exist in relativisti physi s!

162
 1 X
√ |lP (1) i ⊗ . . . ⊗ |lP (N ) i ≡ |l1 , l2 , . . . , lN i bosons . (265)
N! P

The symbol P stands here for permutations o fN labels and (−1)P denotes the permuta-
tion sign. These ve tors will be denoted |l1 , . . . , lN i. In both ases the labels li run over
a ountably innite set of values; below we will assume that li = 1, 2, . . . The number
of labels li in ea h ket is, of ourse, equal N. It is onvenient to adopt the onvention
(exploiting the antisymmetry or symmery of the ve tors (265)) that the labels in kets are
always ordered so that l1 < l2 < . . . < lN in the ase of fermions (no two labels li and lj
an be in this ase equal - this is just the Pauli ex lusion prin iple) and l1 ≤ l 2 ≤ . . . ≤ lN
in the ase of bosons (many bosons an be simultaneously in the same single-parti le
|li are normalized (hl′ |li = δl′ l ), the
state). Assuming thatthe single-parti le state ve tors
fermioni ve tors (265) are automati ally normalized to unity, while in the ase of they
116
require additional normalization if some labels li are equal.
As the Hamiltonian Hˆ(N ) of the system of N parti les is supposed to be built (as
appropriate tensor produ ts) of operators a ting in single-parti le Hilbert spa es of indi-
vidual parti les, the tra es in the formulae (263) an be omputed using the bases (265):

∞ ∞

ˆ
 1 X X ˆ
Tr e−H /kB T = ... hlN , . . . , l1 |e−H /kB T |l1 , . . . , lN i ,
N! l =1 l =1
1 N

and

∞ ∞ ∞ ∞
1 X X 1 X X
Tr(ρ̂ Ô) = ... ... hlN , . . . , l1 |ρ̂|l1′ , . . . , lN
′ ′
i , hlN , . . . , l1′ |Ô|l1 , . . . , lN i .
N! l =1 l =1 N! ′ ′
1 N l1 =1 lN =1

Noti e that in these sums the orderings of labels li are not respe ted: The fa tors 1/N!
an el then multiple ountings in these sums of the same state ve tors written with dif-
ferent orderings of the labels li .
It should be stressed that in this form the formulae are ompletely general and in lude
all possible quantum ee ts. In parti ular of the the mythi al quantum statisti s.
Yet the adopted notation |l1 , . . . , lN i of the basis state-ve tors is rather in onvenient.
It is mu h more pra ti al to pass to the so alled o upation number representation
in whi h the numbers in kets tell how many parti les o upies su essive single-parti le
states. Thus we set

1
|n1 , n2 , . . .i = √ |1, . . . , 1, 2, . . . , 2, . . .i .
n1 !n2 ! . . .
The square root of fa torials makes these ve tors well normalized. Of ourse in the ase of
fermions ni = 0 or 1 only. Moreover, the sum of the o upation numbers ni must always

116 If nl bosons are in the same single-parti le state |li, one has to multiply the state ve tor |l1 , . . . , lN i
√
by the fa tor 1/ nl !.

163
 √
be N. For example, the (normalized to unity) ve tor |1, 1, 1, 2, 3, 3, 7, 7, 11, 13i/ 3!2!2! of
10 bosons is in the o upation number representation written as

|3, 1, 2, 0, 0, 0, 2, 0, 0, 0, 1, 0, 1, 0, 0, 0, . . .i .
It should be stressed that this is not a hange of the basis in the Hilbert spa e but only
a hange of notation. It is also to important to note that in the notation |l1 , . . . , lN i the
number of entries in the ket is nite, but ea h label li an assume innitely many (dis rete
values); in the notation |n1 , n2 . . .i, the number of entries in the ket is innite but the
values of the labels ni are restri ted by the ondition n1 + n2 + . . . = N (and, moreover, if
parti les are fermions ea h ni is either zero or one). The o upation number notation of
state-ve tors of N bosons should be also ontrasted (in order to avoid onfution) with the
basis state-ve tors |n1 , . . . , nN i N quantum harmoni os illators: in this
of a system of
ase it is the number of the labels ni whi h is N but ea h ni an run from 0 to innity. We
will see that if the restri tion n1 + n2 + . . . = N is removed - and we will remove it passin
to the Grand Canoni al Ensemble - the system of bosons will be ome mathemati ally
identi al with a system of (innitely many) harmoni os illators.
The usefulness of the introdu ed notation will be ome evident when we introdu e
the reation and annihilation operators (asso iated with single-parti le states) through
whi h all operators of interest a tin in the system's Hilbert spa e - the Hamiltonian Hˆ
(in luding its intera tion terms) of the system as well as all observables - an be expressed.
Here we will osider only a system of N mutually nonintera ting identi al mole ules whose
Hamiltonian has the general form

N
HˆN = 1̂ ⊗ . . . ⊗ Hˆ (i) ⊗ . . . ⊗ 1̂ ,
X

i=1

(Hˆ (i) at the i-th position) of a sum of N operators ea h of whi h a ts essentially only
in one single-parti le Hilbert spa e onstru ted as a (anti)symmetrized produ t of N
(isomorphi ) single-parti le Hilbert spa es. If the ve tors |li are hosen to be eigenve tors
of the single parti le Hamiltonian H with with the eigenvalues εl , then the a tion of HˆN
ˆ
on the basis ve tors |n1 , n2 , n3 , . . .i of the Hilbert spa e of N mole ules is parti ularly
simple
!
HˆN |n1 , n2 , n3 , . . .i =
X
nl εl |n1 , n2 , n3 , . . .i .
l

The statisti al sum Zstat of su h a system of N mutually nonintera ting mole ules is given
by the expression

n
X max n
X max

Zstat = . . . δN,Pl nl e−n1 ε1 /kB T e−n2 ε2 /kB T . . . , (266)

n1 =0 n2 =0

in whi h nmax = 1, if the mole ules are fermions and nmax = ∞, if they are bosons.
Unfortunately, even in this simple ase the statisti al sum annot be omputed easily

164
(ex ept for spe ial forms of the spe tra εl of the Hamiltonian of a single mole ule) be ause
of the presense of the Krone ker delta whi h expresses the ondition of onstan y of
mole ules in the Canoni al Ensemble. There are two ways of going around this diu ulty:
one is to use the standard tri k of statisti al physi s (and thermodynami s): in order
to ontrol o quantity whi h is onserved by the system's dynami s we imagine that the
system is in onta t with a reservoir of this quantity allowing for ex hanging it between
the system and the reservoir (the Canoni al Ensemble is itself an example of this tri k: to
ontrol the system's internal energy we imagine it being ex hanged it with the heat bath
at xed temperature and by ontroling the temperature of the heat bath we are able to
ontrol the mean energy of the system). This will lead to the Grand Canoni al Ensemble
whi h is representative for the system ex hanging matter (mole ules) with a reservoir at
xed hemi al potential µ. Another way is to use the so alled Boltzmann approximation
whi h we now dis uss.
In the Boltzmann approximation the expression (266) is repla ed by

!N
1 X
Zstat = e−εl /kB T . (267)
N! l

In this approximation the statisti al sum fa torizes into the produ t of ontributions
of individual (mutually nonintera ting) elements just as it does in analogous lassi al
situations. The approximation (267) orresponds to the simple essentially lassi al in
its hara ter ounting of the system's mi rostates: the number of ways of hoosing n1
parti les whi h will be in the single-parti le states l=1 is
 
N
,
n1
then the number of ways of hoosing n2 parti les whi h will be in the single-parti le state
l=2 out of the remaining N − n1 parti les is
 
N − n1
,
n2
and so on. This gives as the statisti al weight of the level of energy n1 ε1 + n2 ε2 + . . . the
fa tor

    "∞ #−1
N N − n1 N − n1 − n2 Y
. . . = N! (na !) .
n1 n2 n3
a=1

Noti e that this ounting treats parti les as distinquishable (and does not take into a -
ount the Pauli ex lusion prin iple); it orresponds to taking as the Hilbert spa e of the
system of N identi al elements the spa e spanned by all tensor produ ts of the form (264)
and not only by its totally symmetrized or totally antisymmetrized subspa es. Indistin-
guishability of parti les is here only taken into a ount by dividing all these fa tors by
N!.

165
 With this ounting the sum (266) is repla ed by

N N
1 X X N!
Zstat = . . . Q∞ δN,Pl nl e−n1 ε1 /kB T e−n2 ε2 /kB T . . . , (268)
N! n =0 n =0 a=1 (na !)
1 2

117
This pre isely gives the expression (267).
It is instru tive to see on a simple example, what is the dieren e between the sums
(266) in the ases when parti les are bosons or fermions and the sum in (267). Let's take
N =3 parti les and introdu e the notation xl = exp(−εl /kB T ). If N =3 the formula
(266) yields

Zstat = x31 + x21 x2 + x1 x22 + x1 x2 x3 + . . . bosons,

Zstat = x1 x2 x3 + x1 x2 x4 + x1 x3 x4 + x2 x3 x4 + . . . fermions,
while from the formula (267) we obtain (in both ases)

1 1 3
(x1 + x2 + x3 + x4 + . . .)3 = x1 + 3x21 x2 + 3x1 x22 + 6x1 x2 x3 + . . . .


Zstat =
3! 6
It is lear that ontributions to the statisti al sum Zstat of singly o upied single-parti le
states in (266) are properly a ounted in the approximation (267). but those of multiply
o upied single-parti le states are not: if parti les are fermions they are totally absent in
(266) whereas if parti les are bosons they have higher weights.
It follows that the Boltzmann approximation (267) should be reasonable in situations
in whi h probabilities of multiple o upan y of the same single-parti le states are low.
This an be made quantitative only by going over to the Grand Canoni al Ensemble in
whi h the number of parti les of the system is not xed (this ensemble is representative
for the system whi h an ex hange energy and matter with a large (innitely large in the
limit) reservoir at temperature T and hami al potential µ. It is then possible to onsider
the mean number nl of parti les o upying a single-parti le state |li. The formula whi h
will be derived reads

1 bosons
nl = .
∓1 + exp((εl − µ)/kB T ) fermions
This is smaller than 1 (that is, the probability of the o upation of a given energy state
|li is low) independently of the number l of the state, if the hemi al potential (of the
reservoir, and hen e, also of the system whi h is in equilibrium with it) is large negative,
i.e. when the a tivity z is small ompared to unity:

z ≡ eµ/kB T ≪ 1 . (269)

117 The sum obtained in this way just the extension of the formula

N   N N
X N X X N!
(x1 + x2 )N = xn1 xN
2
−n
= δN,n1 +n2 xn1 1 xn2 2 ,
n n1 !n2 !
n=0 n1 =0 n2 =0

to the expression (x1 + x2 + x3 + . . .)N .

166
This is typi al in rareed perfe t gases at temperatures not too low, so that the gas is not
liqueed, and not too high so that the mole ules do not disso iate yet and atoms are not
ionized. The hemi al potential of a perfe t gas an be estimated by omputing it within
the lassi al Canoni al Ensemble:
"  3/2 #
V mkB T
µ = −kB T ln .
N 2π~2

It be omes large negative at high temperatures (and moreover, µ/kB T → −∞ as T → ∞).

More pre isely, the ondition (269) translates into the ondition

 3/2
N mkB T
≫ 1,
V 2π~2
or

1/2 1/3
2π~2
 
V
λT ≡ ≪ ,
mkB T N

whi h means that the so- alled thermal wavelength ΛT of mole ules should be small
ompared to the mean intermole ular distan e.

167