2

Methods of classical thermodynamics

as they apply to

SIMPLE SYSTEMS

Introduction. The “methods of thermodynamics” are, from a mathematical

point of view, all quite elementary. But because thermodynamics provides
no natural independent variable (analog of t in dynamics) to lend it a logical
beginning/middle/end,35 a “conceptual spine”. . . the blizzard of “patterned
relationships” to which it gives rise tends to defy all attempts to achieve
reduction to an orderly linear arrangment. In this respect, thermodynamics
is a bit like a Chinese dictionary. Moreover, getting from here to there on any
web -like structure poses strategic problems not encountered on linear
structures: those problems require explicit attention in thermodynamics, and
serve in part to explain why thermodynamic manipulations often seem so
obscurely ad hoc to physicists who come to the subject from dynamics.
It is with the expository intent of keeping semi-simple things as semi-simple
as possible that I will be illustrating the characteristic methods of classical
thermodynamics as they are encountered in the theory of systems with the

35
Rather surprisingly, that role is not taken over by T . Something like
it, however, is played (not within solitary systems subject to manipulation,
but within closed composite systems, within the universe as a whole) by the
never-decreasing numerical value of S. It has, in fact, been argued (most
notably by Einstein) that it is the growth of S that accounts for “time’s arrow.”
40 Thermodynamics of simple systems

least feasible number of independent state variables. Painting on such a small

canvas will serve to reduce the notational and conceptual clutter, but will
extract a price: certain topics of interest will have to be set temporarily aside
• no multi-phase systems (no ice cubes floating in icewater);
• no multi-species systems (no equilibrated hydrogen/nitrogen/ammonia
mixtures, no hot γ
e+ + e− systems).
But the theory that emerges will by itself be rich enough to embrace many
systems of practical interest, and will supply us with wealth sufficient to redeem
many of the topics we have pawned.
Initially, however, I will work on a relatively broad canvas, the better to
“frame” the work undertaken in the main body of the text.
1. From 2-function formalism to two flavors of 1-function formalism. Classical
mechanics does supply a population of statements about Lagrangians-in-general,
but in its principal role the Lagrangian serves as a “system-descriptor.” You
have only to ascribe some specific structure to L(q̇q , q )—and to describe the
physical meanings of the variables q—for me to know all I need to know about
whatever mechanical system you may have in mind. Within the Hamiltonian
formalism the function H(pp, q ) plays a similar role.
The 1st and 2nd laws of thermodynamics assert that with every system S we
can associate a pair of functions, U (ξξ ) and S(ξξ ). Those functions do enter into
a population of general, system-non-specific statements, but in their principal
role those functions serve as “conjoint system-descriptors.” You have only to
ascribe particularized structure to them—and to describe the physical meanings
of the variables ξ—to indicate the essentials of whatever thermodynamic system
is of momentary interest to you. Evidently the laws of thermodynamics invite
us to contemplate what might be called a “2-function formalism.”
Suppose, however, that—in reprise of a trick with we enjoyed some success
already on page 24—we were to promote the value of S(ξξ ) to the status of a
state variable; i.e., that we were to proceed as follows:
S = S(ξ1 , ξ2 , . . . , ξn )
↓
ξ1 = ξ1 (S, ξ2 , . . . , ξn ) by functional inversion
↓
U (ξ1 , ξ2 , . . . , ξn ) ≡ U (S, ξ2 , . . . , ξn )
We arrive then at a “1-function formalism,” within which system-specification
can be accomplished by presentation of a single function. Which is a conceptual
economy, but the principal merit of the formalism lies elsewhere:
Simultaneous statement of the 1st and 2nd laws yields
clausius’ differential equation† : dU = T dS + d¯W (25)

†
Seems a reasonable enough name for an equation among differentials, but
is misleading nonetheless.
Single-function system description 41

Simple calculus, on the other hand, supplies

     
dU = ∂U dS + ∂U dξ2 + · · · + ∂U dξn
∂S ∂ξ2 ∂ξn

so we have
 
T = ∂U (26.1)
∂S ξ2 ,...,ξn
n  
d¯W = ∂U  dξi (26.2)
i=2
∂ξi S,ξ2 ,...,ξi ,...,ξn

which provide the point of departure for much of the work (applied function
theory) that will soon command our attention.
The 1-function formalism comes to us in two primary (and, as will emerge,
many secondary) flavors. For by obvious adjustment of the procedure described
above (alternatively: by functional inversion of U (S, ξ2 , . . . , ξn )) we are led to
the system-specific function S(U, ξ2 , . . . , ξn ) which we can use in conjunction
with this variant of Clausius’ equation

dU − d¯W
dS =
T
to obtain
 
1
= ∂S (27.1)
T ∂U ξ2 ,...,ξn
n  
d¯W = −T ∂S  dξi (27.2)
i=2
∂ξi U,ξ2 ,...,ξi ,...,ξn

We will say we are working “in the U -representation” when we work from (26),
and “in the S-representation” when we work from (27). The two representations
lead ultimately to identical conclusions, but arguments that are simple in one
may be complicated in the other: choice of representation—and the list of
options will be greatly expanded with the introduction of “thermodynamic
potentials”—becomes therefore an “analytical degree of freedom,” an issue to
be addressed at the outset whenever one sets out to establish some specified
thermodynamic relationship/formula/identity. We will—at least initially—work
mainly in the U -representation.

REMARK: Notational conventions standard to thermodynamics. Given

a function f (x) and change of variable x = x(y) one would, in polite
mathematical society, never give the name f to the function f (x(y)),
for its dependence upon y differs from the dependence of f (x) upon x.
One would instead write something like F (y) = f (x(y)). That formal
nicety is suspended in thermodynamic practice. . . for the simple reason
that we change variables at every turn, and we would both exhaust the
resources of the alphabet and become confused if we changed the name
42 Thermodynamics of simple systems

of the function every time we did so. It was with this circumstance in
mind that on the preceding page we allowed ourselves to write

U (ξ1 , ξ2 , . . . , ξn ) ≡ U (S, ξ2 , . . . , ξn )

even though the U (•, •, · · · , •) on the left is a different function of its

arguments than is the U (•, •, · · · , •) on the right. No confusion will
result if one bears in mind that, by thermodynamic convention,

is not to be read “the function F of x and y ”
F (x, y)
is to be read “F described as a function of x and y ”

That practice requires that we depart slightly from some of the

notational conventions standard to the calculus. The symbol ∂F/∂x
informs us that we assume x to be an argument of F , but tells us
nothing about the other variables upon which F has been presumed
to depend. We resolve that ambiguity by writing
   
∂F , ∂F , etc.
∂x y ∂y x

and more generally

 
∂F
∂x explicit list of the variables being held constant

As, in fact, we did already at (26) and (27).

General cautionary note. Central to the processes that lead from the
2-function formalism to the various alternative 1-function formalisms
is a step that calls for functional inversion. Functional inversions are,
in fact, central to many of the arguments and manipulations that
are basic to thermodynamics. A point to be borne in mind is that
functional inversion, though always easy to talk about, is often difficult
or impossible to carry out.

EXAMPLE:
  gases in the single -function representations. We found earlier
Ideal
that T, V, N are variables sufficient to describe the state of an ideal gas, and
that presentation of the functions

U (T, V, N ) = N c T 
(28)
S(T, V, N ) = N c log T + N k log V 
T0 V0

serves in effect to define what we mean by an “ideal gas,” to distinguish such

systems from all other thermodynamic systems. Here

c ≡ kν ≡ CV /N0 = “isovolumetric specific heat per molecule”

= 32 k for monomolecular gases
General properties of bulk systems 43

and its occurance in (28) informs us that the term “ideal gas” refers actually
to the members of a c-parameterized family of systems.
The functional inversion of S(T, V, N ) poses in this instance no difficulty:
we obtain
 k/c
V
T (S, V, N ) = T0 0 exp S (29)
V Nc
giving
 k/c
V
U (S, V, N ) = N cT = N c T0 0 exp S (30)
V Nc
This solitary function—notable for its implausible appearance! (and entirely
typical in that respect!!)—serves to provide a complete characterization of the
thermodynamic theory of ideal gases (in the U -representation).
According to (26.1) we can expect to write
 
T = ∂U
∂S V,N

and indeed: when we use (30) to work out the derivative we promptly recover
precisely (29). But we can go further: according to (26.2) we should have
   
d¯W = ∂U dV + ∂U dN (31.1)
∂V S,N ∂N S,V
 
= ∂U dV if N is held constant
∂V S,N
= −p dV (31.2)

which supplies  
kU
p = − ∂U =
∂V S,N cV
k N cT
=
c V
The c’s cancel, and we obtain finally—for all ideal gases—the familiar statement

pV = N k T

We will return again and again to the ideal gases to illustrate points of principle,
as they arise. Here the lesson has been that
• the single -function formalism seems to work, but
• U (S, other variables) is, even in this simplest of cases, structurally bizarre:
it seems highly unlikely that one would, on intuitive grounds, ever guess
the design of (30)!

2. General properties of bulk systems. Let S be a thermodynamic system

(think, for example, of an equilibrated gas sample), and let S be a mentally
delimited fragment of S , as indicated in the following figure. Evidently S
constitutes a scaled-up version/replica of S. Let the scale-up—so far as it relates
44 Thermodynamics of simple systems

Figure 15: A system S in thermal equilibrium with S, a mentally

delimited fragment of itself.

to volume—be described
V −→ V  = λV
Particle number (equivalently : mass and mole number) scales similarly

N −→ N  = λN

and so—or so we expect, by the argument that gave (17)—does entropy

S −→ S  = λS

Finally, to the extent that the energetic mechanisms operative within the system
hinge on short-range effects that are repeated throughout its volume—that is:
2
to the extent that surface effects (which go as V 3 ) can be neglected—we expect
to have
U −→ U  = λU
It is, on the other hand, intuitively evident that (for example) temperature and
pressure scale by invariance:

T −→ T  = T
p −→ p  = p
General properties of bulk systems 45

The preceding remarks relate in the simplest ways—which, as it happens,

are also the most important ways—to the question: How do various properties
of a thermodynamic system scale?36 And they inspire now the introduction of
some terminology:
Extensive state variables (denoted generically by upper-case letters) are
variables that—like volume/mass—scale by the rule

X −→ X  = λX

Intensive state variables (denoted generically by lower-case letters) are variables

that—like pressure/temperature—scale by the rule

x −→ x  = x

A bulk system B is a thermodynamic system with the property  that it can

—need not, but can—be described in terms of state variables X1 , X2 , . . . , Xn
all of which are extensive. To describe such a system we (working “canonically”
within the U -representation) conventionally identify X1 with S and assign
specific structure to U (S, X2 , . . . , Xn ). But the internal energy is (by prevailing
assumption) itself extensive, so we have

U (λS, λX2 , . . . , λXn ) = λ1 U (S, X2 , . . . , Xn ) (32)

according to which the function U (S, X2 , . . . , Xn ) is homogeneous of degree one.

This fact will exert a major controlling force upon the shape of the ensuing
theory.
We have  
T = ∂U (33.1)
∂S X2 ,...,Xn
as a specialized instance of (26.1), and (see again page 8) will take
 
fi = ∂U  (33.2)
∂Xi S,X2 ,...,Xi ,...Xn
as the definition of the thermodynamic force conjugate to the variable Xi . In
this language “temperature” becomes the “thermodynamic force conjugate to
entropy.” It is clear (by extensive
extensive =
intensive) that the fi are intensive : as they

36
Though the mechanics of few-particle systems only rarely provokes interest
in that question (see, however, Problem 30 at page166 in classical mechanics
()), it is fairly central to the mechanics of distributed systems, especially
hydrodynamics and aerodynamics. It is of importance to all engineers who
would learn things about real-world systems by studying models, and inspires
the widespread interest among engineers in “dimensional analysis” (see, for
example, H. L. Langhaar, Dimensional Analysis & Theory of Models ();
C. L. Dym & E. S. Ivey, Principles of Mathematical Modeling ()). Scaling
theory contributes vitally to the understanding of turbulence, to the statistical
mechanical theory of critical phenomena, to all physical applications of the
theory of fractals.
46 Thermodynamics of simple systems

come to us from (33.2) they are described by functions that are homogeneous
of degree zero:

fi (λS, λX2 , . . . , λXn ) = λ0 fi (S, X2 , . . . , Xn ) (34)

The fi are dimensionally as diverse as the X’s, but in all cases we have

[Xi ][fi ] = energy

NOTATIONAL REMARK: We will consider ourselves free

henceforth to
write X1 in place of S
write f1 in place of T
whenever those adjustments serve expository clarity and
X)
simplicity. We will, for example, feel free to write U (X
in place of U (S, X2 , . . . , Xn ).

x) will
Euler’s “homogeneous function theorem” asserts that a function f (x
be homogeneous of degree n
  
∂f
λn f (x
x) = f (λx
x) x) =
if and only if nf (x xi (35)
i
∂xi

From (32) it follows therefore that

   
X) =
U (X Xi ∂U  = X)
Xi fi (X (36)
i
∂Xi X1 ···Xi ···Xn i

Variation of the preceding equation gives

 
dU = fi dXi + Xi dfi (37)
i i

But it is the upshot of Clausius’ differential equation (25)—i.e., of the combined

1st and 2nd laws—that


n 
dU = T dS + fi dXi = fi dXi
i=2 i

so from (37) if follows that—not generally, but for all bulk systems—we have
the so-called Gibbs-Duhem relation

Xj dfj = 0 (38.1)
j
General properties of bulk systems 47

If we write dfj = i fij dXi with fij ≡ ∂fj /∂Xi then the Gibbs -Duhem relation
becomes 
dXi · fij Xj = 0
i, j

which by the independent variability of the dX’s implies


fij Xj = 0 : i = 1, 2, . . . , n
j 
FXX = 0 : F ≡ fij = ∂i fj = ∂ 2 U/∂Xi ∂Xj
⇓
det F = 0 (38.2)
Evidently (38.1) and (38.2) say, in their separate ways,
 the same thing . . . which
can be phrased this way: the intensive variables f1 , f2 , . . . , fn cannot be
varied independently because they are, according to (38.1), subject to a solitary
differential constraint, which might be written

n
dT = − 1 Xj dfj
S j=2

If (!) the differential form on the right were exact then we would assuredly be
able (in principle) to write
T = T (f2 , f3 , . . . , fn ) (39)
I will, however, not attempt to establish exactness by direct argument, for in
the present instance it is both simpler and more illuminating to proceed by
indirection:
 
Notice first that the intensive variables f1 , f2 , . . . , fn —since they are
not independent—cannot serve  to provide a coordinatization of state space;
i.e., that the transformation X1 , X2 , . . . , Xn −→ f1 , f2 , . . . , fn is singular
 
 ∂( f1 , f2 , . . . , fn ) 

Jacobian  =0
∂(X1 , X2 , . . . , Xn ) 
. . . which is precisely the purport of (38.2). It is therefore evident that by
functional inversion of
T = T (S, X2 , . . . , Xn ) (40)
f2 = f2 (S, X2 , . . . , Xn )
..
.
fn = fn (S, X2 , . . . , Xn )
the best one can hope to achieve is something like
X2 = X2 (S, f2 , . . . , fn )
..
.
Xn = Xn (S, f2 , . . . , fn )
48 Thermodynamics of simple systems

which when inserted back into (40) give an equation of the form

T = T (S, f2 , . . . , fn ) (41)

Scale -up induces

↓
T = T (λS, f2 , . . . , fn )
which by “Euler’s trick”37 becomes
 
0 = S ∂T
∂S f2 ,···,fn
It follows (except at S = 0) that the seeming S-dependence of the function
on the right side of (41) is illusory, and therefore that (41) has precisely the
structure anticipated at (39).
The surviving companions of (40)—rewritten below

f2 = f2 (S, X2 , . . . , Xn ) 


.. (42)
. 


fn = fn (S, X2 , . . . , Xn )
—are called the equations of state of the system to which they refer . . . somewhat
confusingly, because some other things are too. They are n − 1 in number.

EXAMPLE: Ideal gases—revisited. How does the preceding formal material play
in the “simple” case of an ideal gas? Notice first that the internal energy
function encountered at (30) is homogeneous of unit degree if and only if V0 is
included in the list S, V, N ; V0 of extensive variables: we therefore write
 
V0 k/c
U (S, V, N ; V0 ) = N cT0 exp S (43)
V Nc
We then have
  
T = ∂U = U : temperature 


∂S V,N,V0 Nc 

  

∂U = −(k/c) U : negative pressure, denoted −p 

f2 = 
∂V S,N,V0 V
  (44)
f3 = ∂U = U − S
U : “chemical potential,” denoted µ 



∂N S,V,V0 N N 2c 

  

f4 = ∂U = +(k/c) U : fiducial pressure, denoted p0 

∂V0 S,V,N V0
from which it follows that

T S + f2 V + f3 N + f4 V0 = U S − (k/c)U + U − U S + (k/c)U
Nc Nc
= U after simplifications
37
Differentiate with respect to λ, then set λ = 1. See Problem 8.
General properties of bulk systems 49

We have here an instance of (36), by a calculation that would have failed if

the f4 V0 -term had been omitted. From the descriptions of f2 and f4 it follows
readily that
pV = p0 V0
while from the descriptions of f1 ≡ T and f2 we (again) recover
pV = N kT
With the invaluable assistance of Mathematica we compute
 
USS USV USN USV0
 UV S UV V UV N UV V0 
F≡ 
UN S UN V UN N UN V0
UV0 S UV0 V UV0 N UV0 V0

 
1
c2 N 2 − cNaV − c2SN 3 a
cN V0
 a(a+1) −S) 2 
 − cNaV − a(cN − VaV0 
=U·  V2 cN 2 V  : a ≡ k/c
 − 2S 3 a(cN −S)
− cN 2 V S 2 a(cN −S) 
 cN c2 N 4 cN 2 V0 
2 a(cN −S) a(a−1)
a
cN V0 − VaV0 cN 2 V0 V02

and  
 ∂(f1 , f2 , f3 , f4 ) 
det F =  =0

∂(S, V, N, V0 )
which provides a concrete instance of the general result reported two pages ago.
Now strike the first (which is to say: the S th ) row and column, and compute
 
 ∂(f2 , f3 , f4 )  a2 S 2
 
 ∂(V, N, V0 )  = −U c2 N 4 V 2 V 2 = 0
3
0

The implication is that—as anticipated at the bottom of page 47—it should in

principle be possible to write equations of the form
V = V (S, p, µ, p0 )
N = N (S, p, µ, p0 )
V0 = V0 (S, p, µ, p0 )
but in point of analytical fact it appers to be unfeasible to do so: the functional
inversion problem—even for this simplest of systems—appears to be intractable.
That awkward circumstance prevents our obtaining an equation of the form
T = T (p, µ, p0 )
contemplated at the top of the preceding page, though by the simplest of
arguments one does have
T = (p/p0 )T0 for isovolumetric processes
To summarize: ideal gas systems do conform to the points of general principle
developed in the text, but point up this important MORAL: The functional
inversion problem is a beast .
50 Thermodynamics of simple systems

To the (limited) extent that the results obtained above depend upon the
homogeneity assumption they are special to bulk systems. Though most of
the systems in which we will have physical interest are bulk systems (or their
lower-dimensional analogs, in which area/length play the role of volume), one
does occasionally encounter a system in which the “bulk system assumption” is
not justified. Think, for example, of a bubble of gas: an accounting of the system
energetics will have to include both volume effects and non-negligible surface
2
effects, which can be expected to go as V 3 . On similar grounds, we expect the
thermodynamics of “foam” to require methods beyond those supplied by the
theory of bulk systems.
The preceding discussion serves to demonstrate that the 1st and 2nd laws
(enriched here by occasional by homogeneity assumptions) place one in position
to “do function theory,” but supply no compelling evidence bearing on the
question “Function theory to what useful purpose?” Before addressing that
issue I will install—mainly as a notational convenience—the simplifying
assumptions that cause “bulk systems in general” to become “simple bulk
systems.” And it is to place those in context that I present some general
remarks pertaining to . . .

3. Systems of ascending complexity. If the state of S can be described by a

single variable ξ then the laws of thermodynamics have nothing useful to say.
For then (see again page 19) d¯W and d¯Q are both automatically exact. If ξ
refers to a mechanical property of the system then the system must necessarily
be purely mechanical , stripped of all thermodynamic properties (since no
variables remain available to describe such properties). The first law reads

dU = d¯W
d¯W = W(ξ)dξ = (dU/dξ)dξ

If, on the other hand, ξ refers to thermodynamic property of the system then the
system must necessarily be purely thermodynamic, stripped of all mechanical
properties. To describe (within the U -representation) the design of such a
system we would identify ξ with S and present U (S). Clausius’ equation then
reads
dU = T dS with T = dU/dS
Such systems are almost (not quite) too simple to be interesting.
If the state of S can be described by a pair of variables, ξ1 and ξ2 , then
the 1st law—which asserts the existence of a U (ξ1 , ξ2 ) such that

d¯W + d¯Q = dU

—conveys non-trivial information, but the 2nd law tells us nothing we did not
already know, for

the integrability of d¯Q = Q1 (ξ1 , ξ2 )dξ1 + Q2 (ξ1 , ξ2 )dξ2 is automatic

“Simple bulk systems” defined 51

To describe (within the U -representation) the design of such a system we would

present U (S, ξ), and write
   
dU = ∂U dS + ∂U dξ
∂S ξ ∂ξ S
     
d¯Q d¯W
Under circumstances in which the mechanical variable ξ can be considered
“frozen” we recover the (almost) trivial theory considered previously.
The simple point of this discussion:
The state space of S must be at least 3-dimensional if the
1st and 2nd laws are both to contribute non-trivially to the
development of the theory.

4. “Simple” bulk systems. These, in a nutshell, are systems U (S, ξ2 , ξ3 ) with

U (λS, λ ξ2 , λ ξ3 ) = λU (S, ξ2 , ξ3 )
In practice, we assign to ξ2 and ξ3 the names most commonly natural to physical
applications, writing U (S, V, N ).

REMARK: We noticed at (43) that to bring the theory of ideal

gases into the embrace of the theory of bulk systems we had
to write U (S, V, N ; V0 ), and that the simplest of systems is,
by this account, “not simple.” The formal theory of simple
bulk systems will, however, serve to illuminate most of those
aspects of the theory of gases in which V0 can be considered
to be “fixed/frozen.”

In this and the next few sections we will be looking to the analytical properties of
bulk-systems-in-general. We look to simple bulk systems because they require
us to keep mental track of only a few, utterly non-exotic variables concerning
which we possess already some ripe physical intuitions, variables which—because
they are few—will cause certain arguments and constructions to resolve into
manageably few “cases.”
Clausius’ differential relation becomes
   
dU = d¯Q + d¯W = T dS + − p dV + µ dN (45)
where
  
T = T (S, V, N ) ≡ ∂U 
: absolute temperature 

∂S V,N 

  
p = p (S, V, N ) ≡ − ∂U : negative pressure (46)
∂V S,N 

  

µ = µ(S, V, N ) ≡ ∂U : “chemical potential” 

∂N S,V
where T , −p and µ are the thermodynamic forces conjugate to the entropy S,
volume V and particle number N , respectively.
52 Thermodynamics of simple systems

REMARK: p wears a minus sign because to do mechanical

work on a system like a gas (i.e., to render d¯W > 0) one
must compress it (dV < 0). Confusion can result when the
minus sign built into the defintion of p collides with minus
signs that enter into our equations for other reasons. To
minimize that confusion I will sometimes (rarely) adopt the
non-standard notation

q ≡ −p

and call q the “negpressure.”

We can, by Euler’s theorem, always write (compare (36))

U = T S − pV + µN (47)

which is sometimes handy, but since the preceding statement follows simply
and directly from the assumed homogeneity of the function U (S, V, N ) it tells
us nothing about the specific structure of the internal energy function.
Equation (47) acquires diverse interpretations, depending uponhow one
has elected to coordinatize the space of states. The variables S, V, N come to
us as what might (within the U -representation) be called the “canonical point
of departure.” Taking classical mechanics as our model, we might contemplate
introducing “generalized curvilinear coordinates”

qi ≡ qi (S, V, N ) : i = 1, 2, 3

into state space, but in thermodynamics this turns out to be not useful, is not
done. Instead, one confronts in thermodynamics a relatively “crystaline” finite
set of coordinate transformations that can be described as follows:

S −→ T : do it or don’t  
V −→ −p : do it or don’t (48)


N −→ µ : do it or don’t

There would appear to be 23 = 8 possibilities, but in fact there are only 7, since
“do it, do it, do it” is excluded by
 
 ∂(T, p, µ) 
 
 ∂(S, V,N )  = 0

It will serve my expository purposes to exclude the last -listed option;38 i.e., to
restrict my attention to coordinate transformations of these four basic types:
38
I do this mainly to keep the discussion short, but have in mind also the fact
that in physical—as opposed to chemical—problems one often wants to retain
a direct and easy way to express the fact that N is a “frozen” parameter.
Theory of Legendre transformations 53

−−−−−→(S, V, N ) : trivial
−−−−−→(T, V, N )
(S, V, N ) (49)
−−−−−→(S, p , N )
−−−−−→(T, p , N )
Since T and p came into being at (46) as partial derivatives of U (S, V, N ) the
theory of Legendre transformations (see below) provides precisely the tools of
which we have just acquired need. The coordinate transformations (49) give rise
respectively to functions—partial Legendre transforms of the internal energy
function (and of each other)—which are standardly written and designated
−−−−−→U (S, V, N ) : energy
−−−−−→F (T, V, N ) : free energy
U (S, V, N ) (50)
−−−−−→H(S, p , N ) : enthalpy
−−−−−→G(T, p , N ) : free enthalpy

and are known collectively as “thermodynamic potentials.” The “home-base”

U -representation, natural receptacle for the laws of thermodynamics, acquires
thus—at least potentially (meaning except when the Legendre transformation
in question either fails to exist or cannot be executed)—the companionship of
• the F -representation,
• the H-representation,
• the G-representation, etc.
We have now to examine the detailed meaning those anticipatory remarks,
then to address the question: What has this excursion into “function theory”
purchased for us? The short answer: Lots!!
5. Theory of Legendre transformations. The mathematical literature provides
many instances of rules f (•) → F (•) for transforming functions into companion
functions (and back again), the point being that properties of f (•) may be more
easily developed as properties of F (•). Many of those rules39 are of the form

f (x) −→ F (y) = f (x)K(x, y) dx

but we will be concerned here with a rule that falls outside of that class (though
it can be obtained from Fourier’s rule by a certain delicate limiting process).
Your assignment, let us suppose, is to lend natural/useful meaning to
df (x)
f (x) −→ g(p) with p ≡
dx

39
See Ahmed I. Zayed, Handbook of Function and Generalized Function
Transformations () for an exceptionally nice and up-to-date survey of the
“integral transforms” most commonly encountered in physical applications.
54 Thermodynamics of simple systems

Figure 16: Above: the curve C—the “graph”—that results in the

usual way from plotting y = f (x). Below: the same curve displayed
as the envelope of its tangents. In the first representation, C is
a property of a set of points; in the second, it is a property of a
set of lines. The theory of Legendre transformations formalizes the
“duality” of the two constructions.

You might attempt to “do the obvious”. . . which is, I take it, to
• write x = x(p) by functional inversion (if possible!) of p = f  (x), then
• construct g(p) ≡ f (x(p))
but you would be hard-pressed to list “useful properties” of the transformation
f (x) → g(p) thus described; you would confront also the fact that the inverse
transformation f (x) ← g(p) is accomplished by quite a different procedure.
Theory of Legendre transformations 55

slope p

f (x)

g(p)

Figure 17: Notations used to describe the relationship between the

point-wise and line-wise descriptions of C. The function g(p) is the
“Legendre transform” of f (x).

Only a little bit less “obvious”—but culminating in what the experience of

two centuries has shown to be, in many applications, the method of choice—is a
procedure that takes a plane curve C to be the object that mediates the relation
between f (x) and g(p). Figure 16 captures the geometrical essence of the idea,
and Figure 17 sets the notation we will use to develop the details. The tangent
to C at x has slope p = f
(x) and intercepts the y-axis at a point we will call
g(p): it can therefore be described y = g(p) + p x, and at the point of tangency
we have

g(p) = f (x) − p x (51.1)

p ≡ f
(x) (51.2)

The idea now is to write

x = x(p) by functional inversion (51.3)

and then to construct

g(p) = f (x(p)) − p · x(p) (51.4)

EXAMPLE: Look to the case

f (x) = 12 a(x − b)2 (52.1)

p + ab
Here p ≡ f
(x) = a(x − b) =⇒ x(p) = a , so
 2
p + ab p + ab
g(p) = 12 a −b −p· = − 21a p2 − bp (52.2)
a a
56 Thermodynamics of simple systems

Look now to the differential of (51.1): we find

 
g
(p)dp = f
(x) − p dx − x dp
 
0 by the definition of p

according to which g
(p) = −x. What this says is that if we were to construct a
graph of g(p) then the slope of the tangent at p is −x: the tangent itself could
be described z = (intercept) − xp, which at the point of tangency becomes

g(p) = (intercept) − xp

which upon comparison with (51.1) supplies the information that

(intercept) = f (x)

In short: if

g(p) = f (x) − p x
: eliminate x to obtain f (x) −→ g(p) (53.1)
p = +f
(x)

it taken to describe Legendre’s rule for “promoting derivatives to the status of

independent variables” then

f (p) = g(x) + x p
: eliminate p to obtain f (x) ←− g(p) (53.2)
x = −g
(p)

informs us that a rule of—except for a reversed sign40 —identical design

accomplishes the inverse transformation.

EXAMPLE REVISITED: Inversion of x = −g
(p) = a

1
p + b gives
p = a(x − b) whence

f (x) = − 21a [a(x − b)]2 − b[a(x − b)] + x[a(x − b)]

= 12 a(x − b)2

which is precisely the f (x) introduced at (52.1).

40
Such sign-reversals are the rule rather than the exception. Recall from the
theory of Fourier transformations that if

g(p) = √1
2π
f (x)e+ipx dx

then
f (x) = √1
2π
g(p)e−ixp dp
Theory of Legendre transformations 57

In multivariable situations one proceeds similarly, writing (for example)


g(p, q, z) = f (x, y, z) − p x − q y 


∂f (x, y, z) 

p= (54)
∂x 


q=
∂f (x, y, z) 

∂y

By functional inversion of the last pair of equations one obtains

x = x(p, q, z)
y = y(p, q, z)

which when substituted into the first equation produces

f (x, y, z) −−−−−−−−−−−−−−−−−−−−−−−−−→ g(p, q, z)

double Legendre transformation

to which the z has been a mere spectator. The functional inversion may,
however, be impossible—will be, if
 
 ∂(p, q) 
 
 ∂(x, y)  = 0

—and, even when possible-in-principle, may well be intractable.

It is, by the way, by just such a scheme that in classical mechanics the
Lagrangian L(ẋ, ẏ, x, y) = 12 m(ẋ2 + ẏ 2 ) − U (x, y) gives rise to the (negative of
the) Hamiltonian:

−H(p, q, x, y) = L(ẋ, ẏ, x, y) − p ẋ − q ẏ

p = ∂L = mẋ ⇒ ẋ = m1
p
∂ ẋ
q = ∂L = m ẏ ⇒ ẏ = m
1
q
∂ ẏ
 
= − 2m 1
(p2 + q 2 ) + U (x, y)

Here ẋ and ẏ are participants in the transformation, x and y are spectators,

and the impossibility/intractability problem does not arise. The merit of the
exercise is that it converts the system of second-order Lagrange equations
into an expanded set of first-order equations of motion (Hamilton’s canonical
equations).41
41
For more on the elementary theory of Legendre transformations see
H. B. Callen Thermodynamics (), §5.2. More advanced material can be
found in J. V. José & E. J. Saletan, Classical Mechanics (), §5.1.2 and
V. I. Arnold, Mathematical Methods of Classical Mechanics (2nd edition ),
pages 61, 366 & 487.
58 Thermodynamics of simple systems

6. Construction of the thermodynamic potentials. It is to establish the pattern of

subsequent argument that I begin with a review of some of the already-familiar
essentials of the
U-representation Proceeding differentially from U = U (S, V, N ) we have

dU = T dS − p dV + µdN (55 · U )

with   
T = ∂U 


∂S V,N 

  
−p = ∂U
(56 · U )
∂V S,N 

  

µ = ∂U 

∂N S,V
The cross-derivative condition ∂U 2 /∂V ∂S = ∂U 2 /∂S∂V gives42
   
∂T ∂p
=− (57 · U )
∂V S,N ∂S V,N

while the homogeneity condition λU (S, V, N ) = U (λS, λ V, λN ) entails

U = T S − pV + µN (58 · U )

The function F (T, V, N ) is understood to be the “partial

F-representation
Legendre transform of U (S, V, N ) with respect to S ,” got by eliminating S
between
∂U (S, V, N )
T =
∂S
and F (T, V, N ) = U (S, V, N ) − T S
Differentially
   
dF = T dS − p dV + µ dN − T dS + S dT (55 · F )

where the red terms cancel,43 leaving in their wake the statements
42
This in addition to
   
∂p ∂µ
− =
∂N V,S ∂V N,S
   
∂µ
= ∂T
∂S N,V ∂N S,V

which I omit because it is our agreed upon intention to hold N constant.

43
Such a cancellation stands in the middle of every Legendre transformation:
it comprises the very heart of Legendre’s idea.
Thermodynamic potentials 59

 
−p = ∂F 


∂V T,N 

  
S=− ∂F
(56 · F )
∂T V,N 

  

µ= ∂F 

∂N T,V
The cross-derivative condition ∂F 2 /∂V ∂T = ∂F 2 /∂T ∂V gives
   
∂S ∂p
=+ (57 · F )
∂V T,N ∂T V,N

while—whether one argues from scale-up λF (T, V, N ) = F (T, λV, λN ) or from

F ≡ U − TS
U = T S − pV + µN

—one has
F = −pV + µN (58 · F )

H-representation The function H(S, p, N ) is produced by eliminating V

between
∂U (S, V, N )
−p =
∂V
and H(T, p, N ) = U (S, V, N ) + pV
Differentially
   
dH = T dS − p dV + µ dN + p dV + V dp (55 · H)

where again the red terms cancel, leaving in their wake the statements
  
T = ∂H 


∂S p,N 

  
∂H
V = (56 · H)
∂p V,N 

  

µ= ∂H 

∂N S,p

The cross-derivative condition ∂H 2 /∂p∂S = ∂H 2 /∂S∂p gives

   
∂T = + ∂V (57 · H)
∂p S,N ∂S p,N

while—whether one argues from scale-up or otherwise—one has

H = T S + µN (58 · H)
60 Thermodynamics of simple systems

G-representation Elimination of S and V from amongst

∂U (S, V, N )
T =
∂S
∂U (S, V, N )
−p =
∂V
and G = U (S, V, N ) − T S + pV
gives rise to the function of state G(T, p, N ). Differentially
     
dG = T dS − p dV + µ dN − T dS + S dT + p dV + V dp (55 · G)

where the red terms cancel and so do the blue terms, leaving in their wake the
statements   
S = − ∂G 


∂T p,N 

  
∂G
V = (56 · G)
∂p T,N 

  

µ= ∂G 

∂N T,p
The cross-derivative condition ∂G2 /∂p∂T = ∂G2 /∂T ∂p gives
   
∂S = − ∂V (57 · G)
∂p T,N ∂T p,N

while—whether one argues from scale-up or otherwise—one has

G = µN (58 · G)

Several remarks are now in order:

1. By similar manipulations one could “complete the transformational net.”
Which is to say: though I have, in the text, produced F , H and G as Legendre
transforms of U (red arrows in the following diagram), one could by similar
maneuvers produce any potential from any other . . . provided only that the

U F

H G
Thermodynamic potentials 61

Legendre transformation in question is well-defined (which requires that the

relevant Jacobian not vanish). Notice, by the way, that while the existence of
U is assured by the laws of thermodynamics, we are not similarly protected
from the possibility that (say) U −→ H may not be well-defined : in such a case
(though I know of no example) we would be forced to the conclusion that the
H-potential does not exist.
2. The so -called Born diagram (see the following figure) translates the problem
of remembering which variables associate naturally with which potentials into

U F

S T

H G

−p
Figure 18: Classic “Born diagram.” The potentials appear at the
vertices of a square, and the variables upon which they “naturally”
depend appear on the adjacent edges. Conjugate variables stand
diametrically opposite to one another.

the problem of remembering how to draw the Born diagram: this is usually
accomplished with the aid of personally-devised mnemonics, of which

G ood physicists Have S tudied Under V ery Fine Teachers

provides an insipid example (taken from the pages of the American Journal of
Physics). If N and µ are brought into play then the square expands into a
cube (Figure 19), and mnemonics lose their utility. While the Born diagram
serves to identify the variables “natural” to a potential, nothing (unless it be
the vanishing of a Jacobian) prevents—and practical considerations sometimes
recommend—the use of “unnatural” variables. For example: when developing
properties of the isovolumetric and isobaric specific heats we will find it useful
to write U (T, V, N ) and U (T, p, N ), which are obtained from U (S, V, N ) not by
Legendre transformations but by ordinary “scalar point transformations.”
62 Thermodynamics of simple systems

g h

U
F
µ

S T

f u

H G

−p
Figure 19: When N is brought into play the Born diagram becomes
a “Born cube.” There are then 8 potentials (unless homogeneity—
via the Gibbs-Duhem relation—intervenes to render u impossible).
We identify those with the vertices of the cube, and the arguments
“natural” to each with the faces that meet at that vertex. Conjugate
variables are associated with opposite faces.

3. The potentials U , F , H and G all have the same physical dimension

[U ] = [F ] = [H] = [G] = energy = [(variable)·(conjugate)]

and they are intimately interrelated: it follows, for example, from

F = U − TS
H = U + pV
G = U − T S + pV

that
U −F −H +G=0 (59)

4. Looking back to equations (55 –58·U/F/H/G) we see that the homogeneity

assumption (the characteristic “bulk system” assumption) entered into the
proof only of (58). The other statements remain in force even if that assumption
is dropped .
Thermodynamic potentials 63

5. A (relatively little used) formally parallel but distinct series of statements/

transformations/relations would have emerged if we had elected to work in
the “S-representation: ” elected, that is to say, to take not U (S, V, N ) but
S(U, V, N ) as our point of departure (see again page 41).

EXAMPLE: Ideal gas potentials. At (43) we found that the internal energy
function of an ideal gas can be described
   
V0 a
U (S, V, N ; V0 ) = N c T0 exp S (60.1)
V Nc
with a ≡ k/c. Functional inversion of
   a  
V
T = ∂U = T0 0 exp S
∂S V,N,V0 V Nc

gives
   a   a 
exp S = T V whence S = N c log T V
Nc T 0 V0 T0 V0
so the free energy of an ideal gas becomes

F (T, V, N ; V0 ) ≡ U (S, V, N ; V0 ) − T S
   a 
= N c T 1 − log T V (60.2)
T0 V0

In concrete illustration of (56·F) we by quick calculation find

 
p = − ∂F = NkT
∂V T,N,V0 V
    a 
S=− ∂F = N c log T V
∂T V,N,V0 T0 V0

which—gratifyingly—are correct, if not exactly news. Turning our attention

now from the free energy F to the enthalpy H. . . we found that to present ideal
gases as “bulk systems” we had to include V0 among the variables: it seems
appropriate, therefore, that if our objective is to eliminate V in favor of p then
we should at the same time eliminate V0 in favor of p0 ; i.e., that we should
undertake a double Legendre transformation. From44
   
p = − ∂U = a U and p0 = + ∂U = aU
∂V S,V0 ,N V ∂V0 S,V,N V0

it follows that pV = p0V0 = aU and that V0 /V = p/p0 , so from

H = U + pV − p0V0

(in which the red terms cancel) we obtain

44
See again (44).
64 Thermodynamics of simple systems

   
p a
H(S, p, N ; p0 ) = N c T0 exp S (60.3)
p0 Nc
In concrete illustration of (56·H) we find
   a  
∂H p
T = = T0 exp S
∂S p,p0 ,N p0 Nc
 
V = ∂H = a H = a U = N kT
∂p S,p0 ,N p p p
 
V0 = − ∂H = a H = a U = N kT
∂p S,p0 ,N p0 p0 p0
which are once again “old wine in new bottles.” A subsequent Legendre
transformation (designed to achieve S → T ) would give
   a 
p0
G(T, p, N ; p0 ) = N c T 1 − log T (60.4)
T0 p
7. The stuff we measure in the lab. The laws of thermodynamics—as formulated
—direct our attention to a certain function U (S, other state variables) from
which all the thermodynamic properties of the system S can be considered to
radiate. But how, in specific physical cases, are we to discover U ? The issue
can be approached in one or another (or some adroit mix) of three principal
ways:
• We might, on the basis of our experience, simply guess the structure of U ,
and then see how implications of our guess compare with the observational
facts. We might, for example, proceed from Uideal gas to a tentatively
conjectured Ureal gas .
• We might attempt to guess the underlying microphysics , then to bring into
play the methods of statistical mechanics. Those (as it turns out) deliver
into our hands not U but F , which serves quite as well as U as a point
of departure for thermodynamnic analysis . . . but from which we could (in
principle, if we thought it necessary) get to U by Legendre transformation.
• We might go into the lab and observe how S responds when “tickled” in
various ways, then attempt to devise strategies

U(state) ←−−−−−−−−−−−− measured data

that enable us to deduce the structure of U .

My business in this section and the next will be to describe (within the
illustrative context provided by simple systems) the last of those lines of attack.
We look here to some shared features of the diverse procedures that result
in “measured data” and to the elaborate network of identities that serves
invariably/inescapably to interrelate such data.
Let S be a simple system (not necessarily a “bulk system”) of fixed particle
number N (which in point of laboratory fact means of fixed mass). We ask—
because calorimetric techniques would permit us to observe—how S responds
Measurable “coefficients of susceptibility” 65

to the injection of differential heat d¯Q. The 1st law supplies

 
d¯Q = dU − − p dV + µ dN
dN = 0 because N is experimentally constrained
Writing
U = U (T, V, N ) : Note the non-standard arguments (61)
we have
d¯Q = CV dT + LV dV (62)
where  
CV ≡ ∂U : isovolumetric heat capacity (63.1)
∂T V,N
  
LV ≡ ∂U +p : volumetric latent heat (63.2)
∂V T,N
are typical of what thermodynamicists actually measure.
Definitions (63) are particularly well-adapted to the observational realities
of gas physics, but if one turns to liquids/solids one finds that pressure is more
susceptible to experimental control than volume, and is motivated therefore to
promote p to the status of independent variable, writing
V = V (T, p, N )
Returning in this light to (62) we obtain
    
d¯Q = CV dT + LV ∂V dT + ∂V dp
∂T p,N ∂p T,N
= Cp dT + Lp dp
with  
Cp ≡ CV + LV ∂V : isobaric heat capacity (64.1)
∂T p,N
 
Lp ≡ LV ∂V : barometric latent heat (64.2)
∂p T,N
If, on the other hand, we take (61) as our point of departure and write

scalar point transform, used
U = U (T, V (T, p, N ), N ) ≡ U (T, p, N ) : to obtain a different set of
“unnatural arguments”
then by the argument that gave (63) we obtain
   
Cp = ∂U + p ∂V (65.1)
∂T p,N ∂T p,N
   
Lp = ∂U + p ∂V (65.2)
∂p T,N ∂p T,N
Equations (65) do not much resemble equations (65)! Here we encounter
first evidence of the important fact that quantities of practical thermodynamic
interest can always be described in a great many alternative—and non-obviously
equivalent—ways . . . of which, inevitably, some are more useful than others;
utility is always conditional on the specific application at hand. We will need
to become familiar with the transformation-theoretic techniques that enable
one to move about efficiently within the network of equivalent identities.
66 Thermodynamics of simple systems

Looking to equations (63) and (64), we see that the quantities CV , LV , Cp

and Lp —which are, in general, not constants but functions of thermodynamic
state—are by nature coefficients of differential susceptibility : they tell us how,
when we hold all-variables-but-one fixed and differentially “tickle the system,”
the solitary unconstrained variable responds. The coefficients listed above all
arise (as it happens) from “tickle processes” that are by nature “calorimetric”
(meaning “performed with the aid of a calorimeter”).45, 46
What follows is a short list of some of the most frequntly encountered
non-calorimetric susceptibility coefficients:
 
coefficient of (cubic) thermal expansion : α ≡ +
1 ∂V (66.1)
V ∂T p,N
 
isothermal compressibility : κ ≡ −
1 ∂V (66.2)
V ∂p T,N
 
adiabatic compressibility : κs ≡ −
1 ∂V (66.3)
V ∂p S,N
 
coefficient of tension : β ≡ +
1 ∂p (66.4)
p ∂T V,N
..
.
etc.

The constructions CV , Cp , LV , Lp , α, κ, κs , β, . . . are accorded names because

they are intuitively informative and (depending on the system) conveniently
measurable functions of state.
They are—as previously remarked—interrelated by an elaborate network
of system-independent identities. For example: it follows from (66.1) and (66.2)
that  
d log V (T, p, N ) = α dT − κ dp + 1 ∂V dN
V ∂N T,p
and therefore that    
∂α = − ∂κ (67.1)
∂p T,N ∂T p,N

45
It is, by the way, from an instance of d¯Q = Cp dT that calorimetry derives
its basic system of units: let Cp refer to the isobaric heat capacity of one gram
of H2 O at atmospheric pressure and T = 15◦ C and let T + dT = 16◦ C. Then,
by definition, d¯Q = one calorie.
46
While CV , Cp , etc. serve well enough the needs of engineers, what one
finds tabulated in the handbooks are the corresponding specific heats :

CV ≡ m · cV ≡ (sample mass) · (isovolumetric specific heat)

Cp ≡ m · cp ≡ (sample mass) · (isobaric specific heat)

Physicists/chemists also speak frequently about the “specific heat per mole”
(or “. . . per molecule”). Note the implicit use made here of the “bulk system”
concept.
Inferring potentials from measured data 67

This is typical of the population of partial differential identities, and—since

latent in the definitions of α and κ—is clearly system-independent. Typical of
a population of algebraic identities are the statements

Cp − CV = αβpV T (67.21)
α = κβp (67.22)

—the derivations of which will be postponed.47 Results like (67) would seem
immensely surprising if presented as “experimental discoveries,” extracted from
data produced by laboratory study of a system S, and the further discovery
that study of a second system S
yields the same relationships might seem
astounding. Their derivation from first principles—as implications of the laws
of thermodynamics and of the relevant definitions—serves
• to render transparent their system-independent generality, and (since such
relations serve typically to declare that one measurement is obviated by a
suitably-chosen set of other measurements, that information about difficult-
to-measure system-properties may be inferred from easier-to-measure other
properties)
• to simplify the work of experimentalists.

8. Inferring potentials from measured data. While the functions of state S, U , F

etc. are “conceptually most primitive” (in the sense that it is they which enter
into the postulated foundations of thermodynamics), it is a distinct population
of objects—the “coefficients of differential susceptibility”—that are emperically
most accessible. The problem before us: How—from experimental knowledge of
the latter—does one deduce the structure of the former ? An illustrative example
will serve to make clear the essence of the general procedure:
Suppose the system S before us has (like an enclosed sample of gas) the
property that
• N is fixed (and will therefore be notationally surpressed), while
• T and V are subject to easy control.
Our stated objective—quite natural in such a circumstance—is to describe U as
a function of the (unnatural) variables T and V , and also to describe S(T, V ).
Differentially
   
dU (T, V ) = ∂U dT + ∂U dV
∂T V ∂V T
 
But ∂U ≡ CV (T, V ), while by a yet-to-be -established identity48
∂T V
     
∂U ∂p p
=T − p = T2 ∂ (68)
∂V T ∂T V ∂T T V
Suppose CV (T, V ) and p(T, V ) to be known—because measured—along some
47
See below, page 80.
48
See (86) page 77.
68 Thermodynamics of simple systems

V0

T0 T
Figure 20: Specialized path (T0 , V0 ) −→ (T, V ) used in (69) to
construct U (T, V ), and in (71) to construct S(T, V ). The path has
been made “fat” to provide a schematic reminder that the integrands
ask us to differentiate the data p(T, V ).
 
curve C that on the T, V -plane links (T0 , V0 ) −→ (T, V ): then

U (T, V ) − U (T0 , V0 ) = dU
C 
 p(T , V ) 
= CV (T , V )d T + T 2 ∂ dV
C ∂T T V

If, in particular, we take C to have the form shown in the figure, then
T V 
U (T, V ) = U0 + CV (T , V0 )d T + T 2 ∂ p(T, V ) dV (69)
T0 V0 ∂T T V

A similar argument—one that that exploits the identities

 
∂S C
= V (70.1)
∂T V T
   
∂S ∂p
= (70.2)
∂V T ∂T V
—gives
V
T
CV (T , V0 ) ∂p(T, V ) 
S(T, V ) = S0 + dT + dV (71)
T0 T V0 ∂T V

EXAMPLE: Sitting on our lab bench is (let us say) a fixed sample S of ideal
gas. By experiment we discover
• that CV (T, V ) is actually a constant, and
• that p(T, V ) = KT /V , where K is a second constant.
Equations (69) and (71) then supply
Inferring potentials from measured data 69

U (T, V ) = U0 + CV · (T − T0 )
S(T, V ) = S0 + CV · log T + K · log V
T0 V0
Experiments with scaled replicas of S lead us to write
CV = N cV and K = N k
Exercising our options to set U0 = N cV T0 and S0 = 0, we have
U (T, V ) = N cV T
 
S(T, V ) = N cV log T + a · log V : a ≡ k/cV
T0 V0
Eliminating T between those two equations we obtain finally
 a  
V
U (S, V, N ) = U0 0 exp a S
V Nk
This equation—which describes
 U as a function of its “thermodynamically
natural” variables S, V, N —is familiar already from (43), but was extracted
here from (idealized) “observational data.”49
The strategy—described above—for accomplishing
U(state) ←−−−−−−−−−−−− measured data

admits of many variations; 50 the trick, of course, to select the variation

appropriate to the situation at hand, and for that the only guide is experience.
It was remarked on page 29 that “. . . any system can be made to serve as
an ‘absolute thermometer.’ ” How one might proceed to do so is an issue that
has much in common with the issue discussed above, and that we are in position
now to discuss in more general terms what was possible in §13 of Chapter 1.
Suppose that for purposes of thermodynamic experimentation/measurement we
have adopted some conveniently/arbitrarily-graduated emperical temperature
scale T. Our problem is to execute the transformation
T −→ T = T (T)
Though T (T) is unknown, we do know (by the clever argument concluded on
page 26) that
1 dT ≡ G(T) is system-independent, universal
T dT
and that if G(T) were known then T (T) could be readily computed. We will
49
In more realistic situations the lab data would be discrete, and one would be
obliged to resort to numerical differention/integration techniques. One might,
however, proceed analytically if the data could be fit to conjectured formulæ.
50
For discussion of some of those see statistical physics (), Chapter 3,
page 61 or—better—A. H. Wilson, Thermodynamics & Statistical Mechanics
(), pages 37–39.
70 Thermodynamics of simple systems

proceed—non-obviously!—from a yet-to-be -established identity51 of such

general importance that it is known to some authors as the
   
∂U ∂p
thermodynamic equation of state : =T −p (72)
∂V T ∂T V

We proceed more particularly from the observation that our tentative adoption
of the T-scale entails that in place (72) we should write
   
∂U ∂p dT
=T −p
∂V T ∂T V dT
from which it follows quite simply that
 
∂p
G(T) ≡ 1 dT =  ∂T V (73.1)
T dT ∂U +p
∂V T
So G(T) is known, in the sense susceptible to direct emperical determination.
Upon feeding this experimental information into (compare (15))
 T 
T (T) = C · exp G(T) d T (73.2)

we find that we have accomplished a program that might be symbolized

T(state) ←−−−−−−−−−−−− measured data

It is on the basis of (73) that frontier thermodynamicists—those not in position

to employ off-the-shelf thermometers—do in fact conduct their business.

EXAMPLE: Suppose our “working substance” were (unbeknownst, perhaps, to

us) were a fixed sample of ideal gas, and that we had come experimentally to
the realization that (compare (18))
• U (T, V ) is actually V -independent
• p (T, V ) = V –1 · F(T), with F(T) known from data.
It would follow then from (73.1) that

V –1 · F
(T) d log F(T)

G(T) = =
0 + V –1 · F(T) dT
NOTE that all variables except T have dropped
away, as by general theory they must.

and therefore that T (T) = [C/F0 ] · F(T). From this point the story continues as
it continued at the bottom of page 31: the point is that we have here managed
to achieve by swift and general means a result that was there obtained by a
relatively ad hoc line of argument.
51
See (86 page 77. We have encountered this same identity already at (68).
Thermodynamic identities, made to order 71

9. Identity factories. Generalized “coefficients of susceptibility” (things like

CV : see (63.1)) possess (or are assembled from objects of) the characteristic
form  
∂A (74)
∂B C
while “coefficients of relative susceptibility” (things like α: see (66.1)) possess
the form    
1 ∂A ∂ log A
=
A ∂B C ∂B C
 
where A, B, C are drawn from the list

S, T, V, p ; U, F, H, G

It will serve the illustrative purposes of the present discussion to consider only
coefficients of the former type, of which there are a total of 8 · 7 · 6 = 336 (most
of which are of little or no practical interest). Our remarks will pertain to
simple systems with N held fixed: if N (and  its conjugate
 µ) were included in
the list of variables then the four potentials U, F, H, G would be joined by four
others (see again Figure 19), and the number total number of such coefficients
would expand to 14 · 13 · 12 = 2184.
Independent state variables can be selected from the short list

S, T, V, p

in a total of 6 ways:

U : (S, V ) both variables extensive 



G : (T, p ) both variables intensive
 non-conjugate pairs
F : (T, V )  


 

H : (S, p ) 
one extensive, other intensive
(S, T ) 


 conjugate pairs
(V, p )
 
The selected pair will, for purposes of generic argument, be called x, y .
Coordinate transformations can, in this context, be described

x a a(x, y)
−→ ≡ (75)
y b b(x, y)

The notations

  
∂a ∂a
∂(a, b)      
=  ∂x
∂b
y  ∂y x = ∂a
∂b
∂b − ∂b ∂a
∂(x, y) ∂x y ∂y x ∂x y ∂y x
∂x y ∂y x

provide alternative descriptions of the “Jacobian” of the transformation (75),

72 Thermodynamics of simple systems

and Jacobian algebra provides a powerful means of extracting what are for
our purposes the most important implications of (75). The key facts are the
following:
∂(a, b) ∂(b, a)
=− (76.1)
∂(x, y) ∂(x, y)
∂(a, b)
=− (76.2)
∂(y, x)
∂(a, b) ∂(u, v)
= · (76.3)
∂(u, v) ∂(x, y)
 ∂(x, y) –1
= (76.4)
∂(a, b)
∂(a, b)  ∂(x, y) –1
= · (76.5)
∂(u, v) ∂(u, v)
   
∂a = ∂(a, y) = ∂(a, y) · ∂(x, y) –1 (76.6)
∂x y ∂(x, y) ∂(u, v) ∂(u, v)
∂(x, y)
=1 (76.7)
∂(x, y)

I interpose here some remarks which will serve simultaneously to illustrate

the utility of (76) and to establish a property of the Maxwell relations which will
prove to be of high importance. Drawing upon (76.6) we find that the generic
Maxwell relation    
∂T ∂p
=− (57 · U )
∂V S ∂S V
∂(T, S) ∂(p, V )
can be formulated =− or, which is a bit neater (use (76.1/2))
∂(V, S) ∂(S, V )

∂(S, T ) ∂(V, p)
=
∂(S, V ) ∂(S, V )

∂(S, V )
Multiplication by gives (by (76.3))
∂(x, y)

∂(S, T ) ∂(V, p)
= : x and y arbitrary (77)
∂(x, y) ∂(x, y)
Remarkably, we have only to

 (S, V ) to recover the Maxwell relation (57· U )

(T, V ) to recover the Maxwell relation (57· F )
set (x, y) =

 (S, p ) to recover the Maxwell relation (57· H)
(T, p ) to recover the Maxwell relation (57· G )
By this argument, all four Maxwell relations are implicit in any one of them,
and can be obtained without reference to cross-derivative properties of collateral
potentials. But the annotated list on the middle of the preceding page displays
Thermodynamic identities, made to order 73

two further possibilities: if we set (x, y) = (S, T ) we are led from (77) to what, in
the absence of any standard terminology, I will call the unimodularity condition

∂(V, p)        
∂p ∂p
= ∂V − ∂V =1 (78)
∂(S, T ) ∂S T ∂T S ∂S T ∂T S

while if we set (x, y) = (V, p) we obtain its reciprocal:

∂(S, T )
=1
∂(V, p)

My terminology derives from the observation that if R is the region bounded

by a loop inscribed on the (S, T )-plane, and if R
is its image on the (V, p)-plane,

T p

S V

Figure 21: At left: an isolated set R of states identified by their

(S, T )-coordinates. (The set R is, as it happens, bounded by a
pair of isotherms and a pair of adiabats.) At right: the same set
of states—identified by their (V, p)-coordinates—define a region R
.
The unimodularity condition (78) asserts that

area of R = area of R

Because (S, T ) are conjugate variables—and so also are (V, p)—one

has
[area of R] = [area of R
] = energy
Orientation of the regions leads to easy physical interpretation of
the unimodularity condition.

then (see the figure)

  
∂(S, T )
area of R = dSdT = dV dp = dV dp = area of R

R R
∂(V, p) R

74 Thermodynamics of simple systems

The situation is clarified by the observation that

 
area of R =
T dS =
d¯Q = heat injected into S, per cycle

while
 


area of R =
p dV = −
d¯W = −(work done on S, per cycle)

= work done by S, per cycle

We can, in this light, look upon the unimodularity condition as a ramification

of the elementary observation that
 
d¯Q + d¯W = dU =⇒ d¯Q + d¯W = 0

And we can look upon the Maxwell relations as consequences ultimately of that
same circumstance.52
We are in position now to address what might be called the fundamental
problem in identity-generation theory , which is to write
   
∂A ∂a
= f x, y, a(x, y), b(x, y), ( ∂x )y , ( ∂a ∂b ∂b
∂B ∂y )x , ( ∂x )y , ( ∂y )x (79)
C

where  
• x and y are selected from S, T, V, p  
• a and b refer to the unselected elements of S, T, V, p
• the partials ( ∂x
∂a
)y , ( ∂a ∂b ∂b
∂y )x , ( ∂x )y and ( ∂y )x are subject to a single Maxwell
relation, so only three need/should appear, which is to say: an arbitrarily
selected one can/should be discarded.
The expression on the right side of (79) exists in 6 · 4 = 24 variant formulations,
so we confront a population of 336·24 = 8064 potential identities. The inclusion
of one or several additional state variables would increase dramatically the size
of the identity population. For present purposes it matters little that one has
practical interest in very few members of that population, for to analyse one
identity efficiently one must possess technique adequate to the analysis of all.
How to proceed? All “identity factories” are abstractly identical (and
lead, of course, to identical results); they differ one from another only in their
computational layout. I present a method that is more straightforward than
some others, and is due in outline to A. Tobolsky.53 We begin by writing
52
For further discussion, see pages 45–47 in the class notes cited previously.50
The unimodularity condition, and its relationship to Maxwell’s relations, is
discussed in D. J. Ritchie, “A simple method for deriving Maxwell’s relations,”
AJP 36, 760 (1968).
53
“A systematic method for obtain the relations between thermodynamic
derivatives,” J. Chem. Phys. 10, 644 (1942).
Thermodynamic identities, made to order 75

  
 ∂A
 X1 ≡ is the coefficient of interest
∂B C
dA = X1 dB + X2 dC where   (80)

 X2 ≡ ∂A is its “mate”
∂C B
Additionally 
dA = a1 dx + a2 dy 

dB = b1 dx + b2 dy (81)


dC = c1 dx + c2 dy
 
where a1 , a2 , b1 , b2 , c1 , c2 can be either
• read directly, or
• obtained by quick calculation
from the equations
dU = +T dS − p dV (82· U )
dF = −SdT − p dV (82· F )
dH = +T dS + V dp (82·H)
dG = −SdT + V dp (82·G)
and/or the equations
    
∂a dx + ∂a dy 
da = 
∂x y ∂y x
    (83)
db = ∂b dx + ∂b dy 

∂x y ∂y x
Returning with (81) (wherein the a’s, b’s and c’s can now be considered to be
known) to (80) we obtain
(X1 b1 + X2 c1 )dx + (X1 b2 + X2 c2 )dy = a1 dx + a2 dy
which (since dx and dy are independent) entails
b1 c1 X1 a1
=
b2 c2 X2 a2
By matrix inversion
X1 1 c2 −c1 a1
=
X2 b1 c2 − b2 c1 −b2 b1 a2
of which  
∂A a1 c2 − a2 c1
= X1 = (84.1)
∂B C b1 c2 − b2 c1
is the result of particular interest, while
 
∂A a1 b2 − a2 b1
= X1 = (84.2)
∂C B c1 b2 − c2 b1
is its B/C-reversed tag-along mate.
76 Thermodynamics of simple systems

A. N. Shaw54 has arrived at the same results by deft use of the properties
(76) of Jacobian determinants. He writes

( ∂A
∂x )y ( ∂A
∂y )x
∂(A, C) a1 a2
  ( ∂C ( ∂C
∂A ∂(A, C) ∂(x, y ) ∂x )y ∂y )x c1 c2
= = = =
∂B ∂(B, C) ∂(B, C) ( ∂B ( ∂B
∂x )y ∂y )x
C b1 b2
∂(x, y ) c1 c2
( ∂C
∂x )y ( ∂C
∂y )x

which clearly reproduces (in point of historical fact anticipates) precisely (84.1).
In any event—whether one follows in Shaw’s footsteps or Tobolski’s—one must
appeal to (82/83) to evaluate the a’s, b’s and c’s appropriate to the specific
problem in hand.
 
EXAMPLE: Our assignment is to describe CV ≡ ∂U ∂T V as a function of T and V
—variables which are, as it happens, not “natural” to U . Into

dU = a1 dT + a2 dV = T dS − pdV
   
we insert dS = ∂S dT + ∂S dV
∂T V ∂V T
to obtain    
a1 = T ∂S and a2 = T ∂S −p
∂T V ∂V T
Trivially,
dT = b1 dT + b2 dV =⇒ b1 = 1, b2 = 0
dV = c1 dT + c2 dV =⇒ c1 = 0, c2 = 1
so by Tobolski’s (84.1) we have
   
CV ≡ ∂U = T ∂S (85.1)
∂T V ∂T V

Though
 (T,
 V ) are not natural to U , they are natural to F : drawing upon
S = − ∂F
∂T V (see again (56 · F )) we find that we can, if we wish, write55

 2 
= −T ∂ F2 (85.2)
∂T V

54
“The derivation of thermodynamic relations for a simple system,” Phil.
Trans. Roy. Soc. A234, 299 & 378 (1935). A more lucid account of what
has come to be called “Shaw’s method” has been given by F. H. Crawford,
“Jacobian methods in thermodynamics,” AJP 17, 1 (1949). See also page 450
of the same volume, and additional papers cited there.
55
The following equation evidently serves to relate CV /T to the curvature of
isovolumetric curves inscribed on the (F, T )-plane.
Thermodynamic identities, made to order 77

We have—without further labor—also the “mate” of (85.1)

   
∂U = T ∂S −p
∂V T ∂V T
which with the aid of the Maxwell relation (57·F ) assumes the form
 
∂p
=T −p (86)
∂T V 
p
= T2 ∂
∂T T V
of what at (72) we called the thermodynamic equation of state, and of which we
made use already at (68).

REMARK: Equations (56) can be summarized

    
∂U = +T = ∂H 


∂S V ∂S p 

    

∂F ∂U 

= −p = 
 ∂V T  ∂V S (87)
∂G = −S = ∂F 




∂T p
 
∂T V
 



∂H = +V = ∂G 

∂p S ∂p T

and, because they yield Maxwell’s relations (57) by cross-partial

identification, are called the integrated Maxwell relations . They
often (as just above) prove quite useful.

EXAMPLE: At (64.1) we had

     
Cp = ∂U + ∂U + p ∂V
∂T V ∂V T ∂T p

Our assignment is to bring this to—compare (79)—the form


 
= f T, p, S(T, p), V (T, p), ( ∂T )p , ( ∂p )T , ( ∂T )p , ( ∂V
∂S ∂S ∂V
∂p )T

   ∂S 
∂T p = − ∂p T that has rendered the stricken term
where it is Maxwell’s ∂V
redundant and supplied our de facto point of departure:
     
Cp = ∂U − ∂U + p ∂S
∂T V ∂V T ∂p T

We will again use Tobolski’s method,56 and though I write out the details it
is to draw attention to how tedious they are . . . from which very fact we will learn
56
For an account of Shaw’s approach to the same problem see statistical
physics (), Chapter 3, page 49.
78 Thermodynamics of simple systems

something. Turning now to those details, we have

dU = T dS − p dV
          
= T ∂S dT + ∂S dp − p ∂V dT + ∂V dp
∂T p ∂p T ∂T p ∂p T
    
 ∂S ∂S
 a1 = T +p
∂T p ∂p T
= a1 dT + a2 dp with    

 a2 = T ∂S −p ∂V
∂p T ∂p T

b1 = 1
dT = b1 dT + b2 dp with
b2 = 0
   
dV = ∂V dT + ∂V dp
∂T p ∂p T
  
 ∂S
 c1 = −
∂p T
= c1 dT + c2 dp with  

 c2 = ∂V
∂p T

which, by (84.1), is found after simplification to give

   
  a1 ∂V + a2 ∂S
∂U ∂p T ∂p T
=  
∂T V ∂V
∂p T
  a2
∂U = 
∂V T ∂V
∂p T

Returning with this information to our announced point of departure, we find

       
∂S ∂S
 
 ∂p T   ∂p T 
Cp = a1 + a2    − a2   + p ∂S 
∂V ∂V ∂p T
∂p T ∂p T
 
=T ∂S (88.1)
∂T p
 
And since (56 · G) supplies S = − ∂G
∂T p we can, if we wish, write

 2 
= −T ∂ G2 (88.2)
∂T p

Equations (88) describe Cp in terms that bear a remarkable resemblance to

the descriptions (85) of CV , yet issue from an argument that is markedly more
tedious. Why? Because, while the variables (T, V ) are semi-natural to U , the
variables (T, p) are totally unnatural. The potential that stands to (T, p) in the
Thermodynamic identities, made to order 79

same semi-natural relationship that U (S, V ) stands to (T, V ) is H(S, p). The
implication is that we might fare better if we attempted to construct an H-based
theory of Cp . And that, indeed, is easily accomplished: write57

d¯Q = d(H − pV ) + p dV = Cp dT + Lp dp

and (note again the non-standard variables) H ≡ H(T, p). Then

  
Cp = ∂H 

∂T p
   (89)
Lp = ∂H −V 

∂p T

mimic the design of (63). Cp is now simple, and it is the description of CV that
has become relatively complicated:
 
∂p
CV = Cp + Lp
∂T V

Proceeding in direct imitation of the argument on page 76 we have

dH = a1 dT + a2 dp = T dS + V dp
   
dS = ∂S dT + ∂S dp
∂T p ∂p T

giving    
a1 = T ∂S and a2 = T ∂S +V
∂T p ∂p T
while trivially
dT = b1 dT + b2 dp =⇒ b1 = 1, b2 = 0
dp = c1 dT + c2 dp =⇒ c1 = 0, c2 = 1
Tobolski’s (84.1) now leads immediately to
     2 
Cp ≡ ∂H = T ∂S = −T ∂ G2 (90)
∂T p ∂T p ∂T p

which is (88.1) without sweat!58

MORAL: Representation-selection enters as a potentially
important consideration into in the efficient generation
of thermodynamic identities.

57
Compare page 65.
58
Since (T, p) are the variables natural to G it might seem reasonable to
anticipate that a G-based theory of Cp would be even more efficient than the
H-based theory (and similarly: that a F -based theory of CV would be more
efficient than the U -based theory). But try it: you will discover soon enough
that—for informative reasons—this is a pretty idea that leads nowhere.
80 Thermodynamics of simple systems

On page 67 I postponed the derivations of a pair of illustrative “algebraic

identities” which, as it happens, are so important—and the derivation of which
involves methods so important—as to merit close attention. At (64.1) we had
   
Cp − CV = ∂U + p ∂V
∂V T ∂T p
while the “thermodynamic equation of state” (86) supplies
    
∂U ∂p
+p =T
∂V T ∂T V
So we have (recall the definitions (66) of α and β)
   
∂p
Cp − CV = 1 ∂V ·1 · pV T
V ∂T p p ∂T V
= αβ pV T

. . . which is (67.21). The derivation of (67.22) is more interesting (because our

point of departure is so obscurely related to our destination): From (76.3) it
follows in particular that
∂(S, T ) ∂(S, T ) ∂(V, T )
= ·
∂(p , T ) ∂(V, T ) ∂(p , T )
which by (76.6) becomes
     
∂S = ∂S · ∂V
∂p T ∂V T ∂p T
↓
     
∂p
− ∂V = · ∂V by Maxwell relations
∂T p ∂T V ∂p T
Therefore59
        
1 ∂V = − 1 ∂V 1 ∂p p
V ∂T p V ∂p T p ∂T V
↓
α = κβp

. . . which is (67.22), and places us in position to write

Cp − CV = (α2 /κ)V T (91.1)

Since κ, V and T are—for physical reasons—necessarily positive, it follows that

Cp  CV , with equality if and only if α = 0 (91.2)

(which in H2 O occurs at about 4◦ C). A quantity of major importance—

59
For an entirely different line of argument leading to this same equation,
see pages 120 & 133 in mathematical thermodynamics ().
Thermodynamic identities, made to order 81

especially to the theory of gases—is the ratio of heat capacities

Cp isobaric specific heat
γ≡ = 1 (92)
CV isovolumetric specific heat
which can be described in a variety of ways: from (85.1) and (88.1) we have
 
∂S
∂T p
γ= 
∂S
∂T V
∂(S,p) ∂(p,T ) ∂(S,V ) ∂(p,S)  ∂S   ∂p   ∂S   ∂p 
which by ∂(T,p) · ∂(V,T ) = ∂(T,V ) · ∂(V,S) =⇒ ∂T p ∂V T = ∂T V ∂V S
becomes  
∂p
=  ∂V S

∂p
∂V T
 ∂x   ∂y  –1
and by ∂y z = ∂x z —a corollary of (76.6) & (76.4)—assumes the form
 
∂V
= 
∂p T κ
 =κ
∂V s
∂p S
We will later have occasion to consider why it is that γ occurs commonly in
the description of adiabatic processes, and why in particular it makes a famous
appearance in the formula that describes the velocity of sound in gases.
The “integrated Maxwell relations”(87) describe the first partial derivatives
of U , F , H and G with respect to their “natural” variables. But also of
importance are the various “unnatural” first partials of the potentials, as became
apparent when we wrote
   
CV = ∂U and again later Cp = ∂H
∂T V ∂T p
Such constructs yield to analysis by (for example) Tobolski’s method, which
directs our attention also to their “mates”
   
∂U and ∂H
∂V T ∂p T
and it is these latter (and objects like them) that I want now to approach by
an alternative method: write
   ∂[F + T S] 
∂U = : T and V are natural to F
∂V T ∂V T
= −p + T ∂S
!
 ∂V T by a Maxwell relation: we have at
∂p
= −p + T : this point recovered precisely the
 ∂T
 V “thermodynamic equation of state”
p
= T2 ∂
∂T T V
82 Thermodynamics of simple systems

and notice that this equation can be rendered

   
∂ U = ∂ p
∂V T 2 T ∂T T V

But the preceding equation can be read as an exactness condition, an assertion

that there assuredly exists a function A(T, V ) such that
   
∂A p
= U2 and ∂A =
∂T V T ∂V T T

and a little tinkering leads to the function that does the job:

F (T, V )
A(T, V ) ≡ −
T
What we have, in short, discovered is that
 
p
d F = − U2 dT − dV
T T T

from which (72)—the “thermodynamic equation of state”

   
∂U p
= T2 ∂
∂V T ∂T T V

—follows as a statement of the equality of cross-derivatives; i.e., as a kind of

“eccentric Maxwell relation.” So much for the “mate” of CV : if we look similarly
to the mate of Cp we are led to
 
G
= − H2 dT + V dp
T T T
⇓
   
∂H = −T 2 ∂ V
∂p T ∂T T p
 
= V − T ∂V
∂T p

which we may look upon as an unnamed sibling of (72).60

60
F (T, V )/T is a simple “assembled function” of T and V . Of course, one
could use the material at hand to construct assembled functions of arbitrary
complexity, give names X and Y to their partials, and stand back amazed by
the resulting “eccentric Maxwell relations”
   
∂X = ∂Y
∂V T ∂T V

but one cannot, in general, expect such activity to serve any useful purpose.
The lesson of the preceding discussion is that, however, it once in awhile does.
Born diagram as an identity transformation device 83

9. New identities from old by appeal to the formal symmetry-structure of

thermodynamics.61 The “identity-generation engines” described in the preceding
section are general-purpose tools, capable in principle of assembling a vast
array of products (made vaster still if one looks to derivatives of higher order,
and/or increases the number of thermodynamic variables and potentials). We
notice, however, the presence of certain repeated patterns in the design of those
products (compare, for example, (85) with (90)), and that (see again page 75)
the engines accept as input only the following very simple raw material:

dU = +T dS − p dV 



dF = −SdT − p dV
(93.1)
dG = −SdT + V dp 



dH = +T dS + V dp

U = F + TS = G + T S − pV = H − pV 


U − TS =F = G − pV = H − T S − pV
(93.2)
U − T S + pV = F + pV =G = H − TS 



U + pV = F + T S + pV = G + T S =H
U −F +G−H =0 (93.3)
Equations (93.1) inform us what we consider to be the variables “natural”
to each of the potentials, while the highly redundant system (93.2) describes
stipulated relationships among the potentials.62 Equation (93.3)—encountered
already at (59)—is an immediate but particularly useful implication of (93.2).
The point to which I would lay stress, however, is that all identities are latent
implications of (93) or, for more general systems (systems with more degrees of
freedom), latent in some expanded variant of (93). Our plan is
• to identify the substitutional symmetries inherent in the system (93), then
• to exploit the observation that those must be symmetries displayed also by
the collective implications of (93).
And to “identify the substitutional symmetries” we will press Born’s diagram
(Figure 18) into a new kind of service.63
The following operations send → , and in that sense serve to describe
the symmetries of a square:
61
What follows is an elaboration of material presented on pages 36–40 of
J. S. Lomont’s Applications of Finite Groups (). Lomont cites no source,
and his pretty idea (or whose? See the note on page 90.) seems never to have
found its way into the standard thermodynamic literature.
62
The second/third/fourth lines of (93.2) are mere rearrangements of the
first line: they have been included to provide escape from the “U -chauvinism”
seemingly latent in the standard formulation of laws of thermodynamics; i.e.,
to underscore our determination to treat the potentials as formally co-equal
entities.
63
In this respect my line of argument departs from Lomont’s.
84 Thermodynamics of simple systems

I : identity transformation
R1 : central -rotation through 90◦
R2 : central -rotation through 180◦
R3 : central -rotation through 270◦
A1 : reflect in horizontal axis ←→
A2 : reflect in vertical axis
D1 : reflect in diagonal
D2 : reflect in 
 diagonal
These operators (see Figure 22) can be looked upon as the elements of a finite
group of order 8—the so-called square group—but that interesting aspect of
the situation is of no direct concern to us. The point of interest is that if we
decorate the square a la Born, then the operations described above serve to
permute those symbols. Looking for the moment just to the potentials, we have
 
I ←→ U, F, G, H
 
R1 ←→ F, G, H, U
 
R2 ←→ G, H, U, F
 
R3 ←→ H, U, F, G
 
A1 ←→ H, G, F, U
 
A2 ←→ F, U, H, G
 
D1 ←→ G, F, U, H
 
D2 ←→ U, H, G, F
Notice that in all cases either U and G share the 1st & 3rd positions or they
share the 2nd & 4th positions in the string,64 from which it follows that (93.3)
is permutationally invariant . Butto describe
 the permutational transforms of
(93.1) and (93.2) we must adjoin V, T, p, S variables to the preceding list: we
must, in other words, work from
 
I ←→ U, F, G, H; V, T, p, S
 
R1 ←→ F, G, H, U ; T, p, S, V
 
R2 ←→ G, H, U, F ; p, S, V, T
 
R3 ←→ H, U, F, G; S, V, T, p
 
A1 ←→ H, G, F, U ; p, T, V, S
 
A2 ←→ F, U, H, G; V, S, p, T
 
D1 ←→ G, F, U, H; T, V, S, p
 
D2 ←→ U, H, G, F ; S, p, T, V

64
The circumstance just described is, as it happens, realized in all possible
ways, and reflects the elementary fact that symmetry operations can not disturb
the “diametric oppositeness” of U and G.
Born diagram as an identity transformation device 85

U F

H G

F G G H H U

U H F U G F

H G F U

U F G H

G F U H

H U F G

Figure 22: Explicit action of the elements of the square group,

displayed in the sequence
I
R1 R2 R3
A1 A2
D1 D2

The arrows proceed S −→ T and p −→

→ V , and will acquire an
importance evident in Figure 23.
86 Thermodynamics of simple systems

When (for example) the permutation R1 is applied to (93.1) one obtains


dF = ⊕pdV − SdT 


dG = V dp − SdT 
R1 (93.1)
dH = V dp + T dS 



dU = ⊕pdV + T dS

in which I have circled signs that are in disagreement with (93). All could be
rectified by the simple adjustment p → −p, which (compare Figures 22 & 23)
we might associate diagramatically with a restoration of ←
←− to its “canonical”
orientation. Similarly, the permutation R2 gives

dG = ⊕SdT  V dp 



dH = T dS  V dp
R2 (93.1)
dU = T dS ⊕ p dV 



dF = ⊕SdT ⊕ p dV

in which all the signs are wrong, a circumstance that we associate with the fact
that now (compare Figures 22 & 23) both arrows are misdirected, and rectify
by double adjustment: p → −p, T → −T . Look finally to the A1 permutation,
which gives

dH = +T dS  V dp  

dG = −SdT  V dp 
A1 (93.1)
dF = −SdT ⊕ p dV  


dU = +T dS ⊕ p dV

We associate the incorrect signs here with (compare Figures 22 & 23) a single
misdirected arrow, which when rectified leaves p → −p in its wake. So it goes
. . . though in fact we need go no farther, for from the group table

I R1 R2 R3 A1 A2 D1 D2
R1 R2 R3 I D1 D2 A2 A1
R2 R3 I R1 A2 A1 D2 D1
R3 I R1 R2 D2 D1 A1 A2
A1 D2 A2 D1 I R2 R3 R1
A2 D1 A1 D2 R2 I R1 R3
D1 A1 D2 A2 R1 R3 I R2
D2 A2 D1 A1 R3 R1 R2 I

we learn that all group elements can be assembled from R1 and A1 :

R1 R1 = R2 , R1 R2 = R3

R1 A1 = D1 , R1 D1 = A2 , R1 A2 = D2
Born diagram as an identity transformation device 87

U F

H G

F G G H H U

U H F U G F

H G F U

U F G H

G F U H

H U F G

Figure 23: Shown in red are arrows that had to be flipped to be

restored to “canonical orientation.” Flipping is associated in the
text with installation of a minus sign, and with the installation of
those signs we achieve permutational invariance—collectively, not
individually—of the fundamental equations (93).

It is easy to show, though I will omit the detailed demonstration, that the
permutations that preserve (93.1) also preserve (93.2), and therefore preserve
all corollaries of (93).
88 Thermodynamics of simple systems

Much more could be said concerning the formalism sketched above.

Suppose, for example, we introduce matrices
     
0 1 0 0 0 0 1 0 0 0 0 1
0 0 1 0 0 0 0 1 1 0 0 0
R1 ≡   , R2 ≡   , R3 ≡  
0 0 0 1 1 0 0 0 0 1 0 0
1 0 0 0 0 1 0 0 0 0 1 0
   
0 0 0 1 0 1 0 0
0 0 1 0 1 0 0 0
A1 ≡   , A2 ≡  
0 1 0 0 0 0 0 1
1 0 0 0 0 0 1 0
   
0 0 1 0 1 0 0 0
0 1 0 0 0 0 0 1
D1 ≡   , D2 ≡  
1 0 0 0 0 0 1 0
0 0 0 1 0 1 0 0
to describe how the operations of Figure 23 act (permutationally) upon the
elements of the  
U
F 
“potential vector”  
G
H
and that we introduce matrices
   −1
  
0 1 0 0 0 0 0 0 0 0 1
0 0 −1 0  0 0 0 1 1 0 0 0
R1 ≡   , R2 ≡   , R3 ≡  
0 0 0 1 1 0 0 0 0 −1 0 0
1 0 0 0 0 −1 0 0 0 0 1 0
 −1
  
0 0 0 1 0 0 0
0 1 0 0 0 0 0 1
A1 ≡   , A2 ≡  
1 0 0 0 0 0 1 0
0 0 0 1 0 −1 0 0
   
0 1 0 0 0 0 0 1
1 0 0 0 0 0 −1 0 
D1 ≡   , D2 ≡  
0 0 0 1 0 −1 0 0
0 0 1 0 1 0 0 0
to describe the signed permutational action of those operations upon elements
of the  
V
T 
“state-variable vector”  
p
S

We are not surprised to discover that the matrices I, R1 , R2 , R3 , A1 , A2 , D1 , D2
compose in conformity with the group table (page 86); i.e., that they provide
Born diagram as an identity transformation device 89

a 4 × 4 “matrix representation” of the square  group. We may, however, be

surprised by the observation that the set I, R1 , R2 , R3 , A1 , A2 , D1 , D2 does
not supply a representation: it is, in fact, not even multiplicatively closed, as
the following examples serve to illustrate:
− 
1 0 0 0
 0 −1 0 0 
R41 =  
0 0 −1 0
0 0 0 −1
−  + 
1 0 0 0 1 0 0 0
 0 +1 0 0   0 −1 0 0 
A21 =   , A2 = 
2

0 0 −1 0 0 0 +1 0
0 0 0 +1 0 0 0 −1
Clarity is brought to the matter by the realization that
       
V −V −V V
T   −T   T  −T 
 →  , else   , else  
p −p −p p
S −S S −S
describe an additional, trivial invariance property of the system (93). But I
will resist the temptation to pursue this and related (group-theoretic) matters,
and cut directly to the bottom line . . . which is that
Equations (93)—and therefore all implications of (93)—are collectively
invariant under the following rules of substitution (or “signed permutations”):

U F G H V T p S : original sequence
 ↓
 F G H U T −p S V : rule r1
 G H U F −p S V −T : rule r2
 H U F G S V −T p : rule r3
 H G F U −p T V S : rule a1
 F U H G V S p −T : rule a2
 G F U H T V S p : rule d1
U H G F S −p −T V : rule d2

EXAMPLE: At (85) we had

CV ≡ ∂U = T ∂S = −T ∂2F
∂T V ∂T V ∂T 2 V
Intuition suggests that a rule that sends V → p while preserving T should
provide information relevant to the description of Cp . Inspection of the list
shows the only candidate to be a1 , which gives
↓
∂H = T ∂S = −T ∂ 2 G
∂T p ∂T p ∂T 2 p
But this is precisely (90)! Notice also that if we apply a1 to the equations that
led to the former identity we obtain a derivation of (90).
90 Thermodynamics of simple systems

EXAMPLE: If we apply the rules serially to the first of the “integrated Maxwell
relations” (87) we obtain

∂U = +T = ∂H a1 : ∂H = +T = ∂U
∂S V ∂S p ∂S p ∂S V

r1 : ∂F = −p = ∂U a2 : ∂F = −S = ∂G
∂V T ∂V S ∂T V ∂T p
∂G = −S = ∂F ∂G
r2 : d1 : = +V = ∂H
∂T p ∂T V ∂p T ∂p S
∂H = +V = ∂G ∂U ∂F
r3 : d2 : = −p =
∂p S ∂p T ∂V S ∂V T

which is to say: we obtain two copies of each of the relations (87). In general
we can expect (for basic group-theoretic reasons) to obtain either
• one copy of each of a total of 8 identities;
• two copies of each of a total of 4 identities;
• four copies of each of a total of 2 identities, or
• eight copies of a single identity ((93.3) provides an example).
The previously-noted tendency of thermodynamic identities to coagulate into
structurally similar families begins to seem not so mysterious.
The Born diagram came to us (page 61) as a rudimentary mnemonic device,
but has been revealed to have latently a much deeper significance, to empower
us to do much more than remember a short list. In higher dimension (i.e., for
systems with n > 2 thermodynamic degrees of freedom) it loses its mnemonic
utility, but its identity-generating power is correspondingly enhanced: we find
ourselves talking then65 about the “symmetries of a hypercube,” which give
rise to a group of order 2n n!. Though the theory was thermodynamically
motivated, it pertains to multivariable Legendre transform systems in whatever
context they may be encountered, and also (with only slight modifications) to
multivariable Fourier transform systems.66
65
This topic is developed in my “ Hyper-octagonal Born diagrams” (Notes
for the Physics Seminar of  October , reprinted in transformational
physics & physical geometry, –).
66
When I worked out the preceding material I was aware only of the Lomont’s
work,61 which it was my objective to clarify. I have since learned that the
basic ideas were described a long time ago by F. O. Koenig in “Families of
thermodynamic equations. I. The method of transformations of the
characteristic group,” J. Chem. Phys. 3, 29 (1935). See also H. A. C. McKay,
“A generalized thremodynamic notation,” J. Chem. Phys. 3, 715 (1935) and
J. A. Prins, “On the thermodynamic substitution group and its representation
by the rotations of a square,” J. Chem. Phys. 16, 65 (1948), where it is
suggested that the entire subject was known already to Born. C. E. Reid,
in his Principles of Chemical Thermodynamics (), speaks in passing of
Koenig- Prins transformations . . . though I think very few poeple would know
what you were talking about if you alluded to those. Of these authors, only
McKay gives any attention to the n-dimensional case.
We take temporary leave of classical thermodynamics 91

Every well-trained musician should know something about the crumhorn,

but of that subject enough becomes easily too much. The same, I think, can be
said of the “theory of thermodynamic identities,” which seductively invites—but
does not reward—extended close attention. Some such identities are absolutely
indispensable, but they are well-known, and hardly need to be rederived every
time they are used. Others can be discovered in various dusty handbooks.67

10. Concluding remarks. We have acquired some familiarity with the basic
principles of classical thermodynamics, and with some of its characteristic
methods, at least as they relate to simple systems. But we have yet to consider
any of the major applications of those ideas. Before we are done we will want
to look in particular to
• the thermodynamics of some specific systems (especially real gases,
blackbody radiation);
• the theory of multi-phase systems and of phase transitions;
• the theory of heat engines and refrigerators (and to what the latter has to
say about the unattainability of absolute zero);
• the theory of superfluids and superconductors;
• the theory of irreversible processes;
• the theory of reactions in multi-component systems . . . the list goes on and
on.
But it has come time to cross to the statistical mechanical side of the street, for
our knowledge of thermodynamics is already sufficient to permit us to recognize
when statistical lines of argument have begun to say “thermodynamic” things.
And many of the topics listed above are most usefully illuminated by
considerations marked by a balanced interplay of statistical and thermodynamic
concepts and methods.

67
The classic compilation is P. W. Bridgman, A Condensed Collection of
Thermodynamic Formulas ().