Download as pdf or txt
Download as pdf or txt
You are on page 1of 24

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/225557667

On the Clausius Inequality

Article  in  Archive for Rational Mechanics and Analysis · September 1983


DOI: 10.1007/BF00250801

CITATIONS READS

24 1,431

1 author:

Miroslav Silhavy
Institute of Mathematics of the Czech Academy of Sciences
171 PUBLICATIONS   2,103 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Isotropy View project

Nonlocal and fractional models in continuum mechanics and geometry View project

All content following this page was uploaded by Miroslav Silhavy on 05 March 2016.

The user has requested enhancement of the downloaded file.


On the Clausius Inequality
M. II nAVq
Communicated by J. SERR~N

Summary

A condition equivalent to the existence of the absolute temperature scale satis-


fying the Clausius inequality is exhibited. The condition is close to the classical
statements of the second law of thermodynamics due to CLAUmUS and others.
It is formulated as an assertion about cyclic processes of bodies whose totality
is called a universe of bodies. The universe can contain general (local or not)
bodies with memory but the validity of the result (i.e., the mentioned equiva-
lence) requires that it must contain enough elastic bodies. With each universe
of bodies one can associate a collection of Borel measures on the real line, and
the restrictions imposed on the universe of bodies induce certain properties of
this collection. Also the condition equivalent to the existence of the absolute
temperature scale as well as the Clausius inequality can be stated completely
in terms of the collection of measures. The p r o o f of the main result is given in
this more general setting without recourse to bodies.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
2. Global Characteristics of Thermodynamic Processes . . . . . . . . . . . . 225
3. Universes of Bodies. Statement of the Result . . . . . . . . . . . . . . . 228
4. Classes of Measures . . . . . . . . . . . . . . . . . . . . . . . . . . 232
5. Abstract Version of the Result and the Proof . . . . . . . . . . . . . . . 236
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241

1. Introduction

In the last two decades the thermodynamics of materials with memory based
on the Clausius-Duhem inequality* was developed into great detail. Following
* The general form of the inequality first appears in the treatise by TRUESDrLL &
TOUPIN [1].
222 M. SILHAVY
the lead of COLEMAN • NOLL [2] and COLEMAN [3] in the first period of the re-
search the emphasis has been laid upon finding the restrictions which the Clausius-
D u h e m inequality places upon the constitutive functionals.* More recent
work has dealt with the implications of the inequality to wave p r o p a g a t i o n * *
and to the stability theory. * * * In summary the quoted research has shown that
the Clausius-Duhem inequality is a fruitful assumption with reasonable and far-
reaching consequences, although the questions of the experimental and theoretical
foundations were left open. In particular, the following questions have not been
answered until very recently:
(a) W h a t a priori arguments can be given in favour to the Clausius-Duhem
inequality ?
(b) To what extent does the Clausius-Duhem inequality reflect the classical
statements of the second law ?
Once the last question has been laid down, another one must be considered:
(c) What are the appropriate "translations" of the classical statements of the
second law into a precise mathematical language?
When applied to cyclic processes, the integrated form of the Clausius-Duhem
inequality reduces to the integral form of the Clausius inequality which does not
contain the entropy. The results of DAY [4, 10, 11] and COLEMAN & OWEN
[5, 12, 13, 14] indicate that for a large class of materials the Clausius inequality
also conversely implies the existence of non-equilibrium entropy satisfying the
Clausius-Duhem inequality. This reduces the question (b) to:
(d) To what extent does the Claushls inequality reflect the classical statements
of the second law?
FOSDICK & SERRIN [15] showed that the Clausius-Duhem inequality yields
propositions which cart be interpreted as the classical statements of the second
law due to CARNOT, CLAUSIUS,KELVIN,and PLANCI( and this proves that the
Clausius-Duhem inequality is compatible with the classical statements of the
second law. However, to show that the Clausius-Duhem inequality is the only
assumption compatible with the classical statements requires to derive it from a
condition close the the traditional statements of the second law.
For elastic materials in homogeneous situations various conditions of this
type are described in the literature, * * * * but for materials with genuine m e m o r y
the progress came only very recently. In the lecture notes [18] SERRIN deals with
universes of frictional simple bodies in homogeneous deformations and formulates
the conditions from which the existence of the absolute temperature and entropy
satisfying the Clausius-Planck inequality follows. By using a different frame-
w o r k * * * * * and techniques I have demonstrated in [20] the existence of the non-

* I make no attempt to cite all the literature along those lines. Partial bibliogra-
phies can be found in [4] and [5]. The general theories have been developed by GURrlN
[6] and COLEMAN & OWEN [7].
** See COLEMAN & GURTIN [8].
*** The general theory is due to GURT1N [9]. See also the literature cited therein.
**** See, e.g., [16] and [17].
***** In contrast to [18] the paper [20] does not employ the concept of empirical
temperature and the theory deals with a single body.
The Clausius Inequality 223

equilibrium entropy and temperature satisfying the Clausius-Planck inequality


from another "classical" condition for a class of materials with internal variables
in homogeneous processes. The proofs given in [18] and [20] seem to rest heavily
on the particular constitutive assumptions considered and thus the results apply
only to limited classes of materials.
General answers to the questions (a)-(d) are given independently by SERRIN
[19, 26, 27] and SILHAVq [28, 29, 30]. In [26, 27] SERRIN presents a condition
generalizing the statements of the second law due to CARNOT,CLAUSIUS,KELVIN,
and PLANCK. This condition implies the existence of the absolute temperature
scale satisfying the accumulation inequality which in turn implies the Clausius
inequality. The validity of a result of this type was conjectured in [19] and then
proved by SERRINin collaboration with R. HUMMEL.* The paper [19] also contains
the important idea that Borel measures can be used to make the foundations of
thermodynamics rigorous (see in particular the footnote on p. 427 of [19]).
In [28] I have presented a similar approach based on the extensive use of the Borel
measures associated with homogeneous cyclic processes of general thermodynamic
systems. A condition close to the Clausius statement of the second law was shown
to imply the existence of the absolute temperature scale satisfying the Clausius
inequality. The present paper, which was written in 1979, is a direct generalization
of [28]; in particular, the same condition as that given in [28] is used here to
establish the existence of the absolute temperature scale. Another approach is
developed in [29, 30], where the existence of the absolute temperature is shown
to be a consequence of various mathematical conditions corresponding to the
second laws due to CARNOT,CLAUSIUS,KELVIN,and PLANCK. The proofs are based
on a simple idea which clarifies the geometrical nature of the absolute tempera-
ture and sheds also a light on the first law of thermodynamics. * *
The purpose of the present paper is to offer an alternative proof of some of
the results of [29, 30]. Like the proof in [26], also the present p r o o f is close to
the standard textbook approach (though, of course, both the considerations
given in the subsequent sections and in [26] are carried with attention to rigor).
In contrast to [29, 30] the proof given here does not use the topology on the
space of measures associated with cyclic processes. The present paper also con-
tains a detailed discussion of the basic concepts of the theory which [29, 30]
presuppose.
The basic objects of the present theory are precisely those of classical thermo-
dynamics: heat, work, empirical temperature, and a universe of bodies endowed
with classes of processes. The mechanical background (motions, forces, stresses,
etc.) is irrelevant at the present level of generality and thus the theory is applicable
also to other than thermomechanical phenomena. No specific assumptions about
the nature of the response of bodies is necessary; the bodies can have memory

* See also [31]. A recent work of COLEMAN,OWEN, and SERRIN [32] generalizes
SERRIN'S ideas to systems with approximate cycles.
** A survey of the basic ideas of [29, 30] is given in [33]. A generalization of [29,
30] is contained in the forthcoming papers [34, 35] of the present author. In the last
two mentioned papers, I discuss in detail the consequences of and the relations among
the classical statements of the second law for systems which need not satisfy the first law.
224 M. ~ILHAVq

and need not be local. As the theory deals with only one aspect of the thermo-
dynamic behaviour of bodies, namely with the properties of cyclic processes,
even the concepts of state and the constitutive assumptions can be avoided, as
will be seen. The universe as a whole, however, is subject to certain requirements
which will be discussed later.
For the present theory it suffices that each body 9~ from the universe be
endowed with a class ~ ( ~ ) of processes which are the triples p = (0, r, h) of
time-dependent scalar fields over ~ with the following meaning: 0 is the field
of the empirical temperature, r is the body heat supply, and h (which is defined
only on the boundary of 9~) gives the heat flux into the body across the boundary.
For the interpretation it is understood that the triples p from ~(9~) correspond
only to cyclic processes of 8 . (Note that within the present framework the concept
of a cyclic process cannot be defined formally since this would require the concept
of state. Nevertheless, all the concepts necessary for the statement of the result
of the paper are defined in terms of the triples p and in terms of the collections
9~(&); hence there is no logical inconsistency nor incompleteness.)
The content of the subsequent sections is as follows. In Section 2 we
show that with each process p----(0, r, h) one can associate in a natural way
a Borel measure on the real line whose value on each Borel set gives the net gain
of heat of the body at the empirical temperatures from that set in the process p.
From this measure various global thermodynamic characteristics of the pro-
cess p are derived by employing some elementary tools of the measure theory.
By using those characteristics, in Section 3 two conditions for the universe of
bodies are formulated: Condition 1 is a statement which, appart from the fact
that it uses the new concepts defined in Section 2, resembles some of the statements
commonly attributed to CLAUSIUS, while Condition 2 is the assertion that there
exists an absolute temperature scale (which is a positive and non-decreasing
function of the empirical temperature) such that the Clausius inequality holds
for each process.
The main result of the paper--Theorem 1 in Section 3--asserts that for
certain universes of bodies, called admissible universes of bodies, Conditions 1
and 2 are equivalent. The requirements entering the definition of an admissible
universe of bodies guarantee, roughly speaking, (1) that the union of disjoint
bodies from the universe is again a body of the universe, (2) that the "union"
of processes which take place simualtaneously on disjoint bodies is a process for
the union of the bodies, and, finally, (3) that there are enough reversible homo-
geneous processes of some elastic bodies from the universe. The last requirement
is in a close analogy with the assumptions about the existence of various "quasi-
static heat reservoirs", made more or less explicitly in the standard treatises on
thermodynamics.
The fact that Condition 1 implies Condition 2 shows that the classical CLAUSIUS'
statement of the second law can be transformed into such a form that it really
implies the clausius inequality. In a more complete theory which includes the
concept of state it is possible to proceed further: By applying the results of DAY
and COLEMAN & OWEN mentioned above one can get, in addition to the exis-
tence of the absolute temperature scale, also the existence of the entropy which
satisfies the Clausius-Duhem inequality.
The Clausius Inequality 225

Instead of proving Theorem 1 directly I here prefer to state and prove a more
abstract and general Theorem 2 which contains Theorem 1 as a particular case.
Theorem 2 deals with measures and collections of them rather than with processes
and universes of bodies. The possibility of such a generalization is suggested by
the fact that Conditions 1 and 2 as well as the requirements imposed oft the
admissible universes of bodies can be restated completely in terms of the collec-
tions of measures associated with the processes of bodies f r o m the universe.
I believe that such an indirect procedure best reflects the nature of the main
result of the paper.
Therefore, in Section 4 we introduce a terminology concerning the meas-
ures which is analogous to the one set up in a special context in Sections 2
and 3 for processes; in particular the concept of an admissible collection of meas-
ures corresponds to the concept of an admissible universe of bodies. Theorem 2
is then stated and proved in Section 5. The discussion in Sections 4 and 5 is self-
contained: the reader can start in the second paragraph of Section 4. Neglecting
the few sentences which refer to previous sections he can proceed to Theorem 2
and then apply it in the special situation described in Sections 2 and 3.

2. Global Characteristics of Thermodynamic Processes

Consider a continuous body which performs a cyclic process in which it


absorbs heat and does work. F o r convenience we identify the body with a regular
region* ~ of a three-dimensional Euclidean space. The boundary and the interior
of ~ is denoted by 8& and ~ o respectively. The body is assumed to be endowed
with the mass-measure m(-); we write da for the element of area of 8 ~ . A process
p of the body is a triple p = (0, r, h) of continuous functions* * 0 : & • Ip ~ R,
r : ~ • Ip --+ R, h : 9& • ~ R, where /p C R is a closed, non-degenerate, and
bounded interval, called the time-interval of p. The interpretation of the quan-
tities 0, r, and h is as follows: for each X E ~ and tEIp the values O(X, t) and
r(X, t) give the empirical temperature and the body heat supply at the particle X
and time t, while for each X E 9 ~ and t E Ip the value h(X, t) is the area density
of the heat flux into the body across the boundary at the particle X and time t.
For brevity we shall sometimes use the word "temperature" for the empirical
temperature while when speaking about the absolute temperature we shall never
drop out the adjective "absolute". Note that we need not assume that h is of the
form h ---- - - q 9 n with q the heat flux vector and n the unit outward normal to
~ ; the function h is not defined in the interior of ~ . The integral

q(p) : f ( f h da Jr f r dm)dt
lp ,~ ,~

is the net gain of heat of the body ~ in the process p -----(0, r, h).

* i.e., closed, bounded region with smooth boundary.


** Throughout I~ = (-- 0% oo), I~+ __- [0, co), and ~++ = (0, oo).
226 M. S1LHAVY

In what follows we shall need the standard definitions and results of the meas-
ure theory; they can be found in the book by HALMOS[21]. Let A Q 1~ be a Borel
set and p ---- (0, r, h) a process of the body ~ . We introduce the following subsets
of ~ •

,Y-r A) = ((X, t) E ~ • Ip : O(X, t) E A),

~'b(P, A) = ((X, t) E ~o X Ip : O(X, t) E A}.

The set g-c(P, A) (or ~-'b(P, A)) is the set of all "events" (X, t) from # ~ • Ip (or
from ~ ~ Ip) at which the body has temperature in the set A (during the process
p). The number q(p, A), given by

q(p,A) = f h da dt -~ f r dm dt
J-c (p,A ) J'b (p,A )

is the net gain of heat of the body ~ during the process p at temperatures from
the set A. We have, of course, q(p) = q(p, R).
For a fixed process p the mapping q(p, .), which assigns to each Borel set
A ( K the number q(p, A), is a finite, real valued, signed Borel measure. We write
q(p, .) = q+(p, .) -- q-(p, .) for the Jordan decomposition of q(p, .); by the very
definition, q+(p, .) and q-(p, .) are finite, non-negative valued Borel measures.
The non-negative numbers q+(p) and q-(p), defined by

q+(p) = q+(p, R), q-(p) -----q-(p, R),

are referred to as the heat absorbed and the heat emitted by the body ~ in the pro-
cess p. Clearly
q(p) ---- q+(p) -- q-(p).

The supports of the measures q(p, .), q+(p, .), and q-(p, .) are denoted, respectively,
by s ~ ) , s+(p), and s-(p). These are the smallest closed subsets of R on which the
corresponding measures are concentrated. Obviously s(p), s+(p), and s-(p) are
subsets of 0(~ • Ip) which in view of the assumed continuity of 0 is a compact
set. Hence also s(p), s+(p), and s-(p) are compact sets. If s+(p) ~ 0 then the num-
ber
O+(p)=maxs+(p)

is referred to as the maximum temperature at which heat is absorbed by the body


:~ in the process p. Likewise, if s-(p) =~ 0 then the number

O-(p)-----mins-(p)

is referred to as the minimum temperature at which heat is emitted by the body


in the process p. It may, of course, happen that O+(p) < O-(p).
The Clausius Inequality 227

It should be stressed that the heat absorbed q+(p) and the heat emitted q-(p) defined
above differ from the numbers q+(p) and q~-(p) given by*

q~(p) = f ( f [h]+ da + ff jr]+ dm)dt,


lp
(2.1)
q~-(p)= f C f [ h ] - d a + ! [ r ] - d m ) d,.
11,
Usually the latter quantities are called the heat absorbed and the heat emitted.** They
again satisfy q(p) = q+(p) -- q~(p) with q+(p) => 0 and q~-(p) => 0, but in general
only the inequalities q+(p) ~ q+(p) and q-(p) <=q~(p) are valid.*** To explain the
difference between q-~ and qoi , note first that q0i can be defined independently of the
empirical temperature while the q:L cannot. Secondly note that if some positive amount
Qo of heat was absorbed at temperature 0o during some period of a process p and then
during another period of p the same amount Qo was emitted at the same temperature
0o then these two periods do not contribute to the values of q--(p), but they do to the
values of qoi(p), i.e., q• have the same values as if during the two periods heat was not
exchanged while the values of qoi(p) are increased by the two periods and the increment
is precisely Qo. In other words, the relations q+(p) = q~-(p) = 0 imply that the process
p is adiabatic (i.e., r = 0, h = 0) while the relations q+(p) = q-(p) = 0 mean only
that the absorption of heat during some period of the process was compensated by an
emission of the same amounts of heat at the corresponding temperatures during another
period of the process. It is just the insensitivity of the quantities q• to the mutually com-
pensated exchanges of heat which makes them more apt for the present purposes than
the quantities qo~ and some of the crucial arguments given in the succeeding sections would
no longer be valid if q• were replaced by qoi .

I n accord with the interpretation that the triples p correspoud to cyclic pro-
cesses, we define the work w(p) done by the b o d y 9~ in the process p by

w(p) ----q(p) ;
the last equality then expresses the first law for cyclic processes. (Note that
within the thermomechanical framework w(p) is the w o r k done by the true forces
plus the w o r k done by the "acceleration forces",

w(p) = -- f ( f Sn . i~ da + f (b - ~) . i~din)dt

with S the Piola stress tensor, b the density o f the b o d y force,/r the velocity field,
the acceleration field, and n the unit outward n o r m a l to ~ . )
I n conclusion o f this section we introduce the concept o f the Clausius integral
which will be used in the subsequent sections. Let p ---- (0, r, h) be a process
o f the b o d y 9~ and let T : 0 ( ~ • Ip)~ R++ be a non-decreasing and positive
function (not necessarily continuous). The Clausius integral ~gr(P) o f the process

* Here [f]+ and if]- denote the positive and the negative part of the function f.
** See, e.g., [20], [22], and [23].
*** These inequalities will not be used in the sequel and their proof is omitted here.
228 M. SILHAVY

p with respect to T is defined by

c~r(p) = da + dm dt.

If the function T is interpreted as the absolute temperature scale, i.e., if the com-
posed function To 0 is interpreted as the absolute temperature then (gr(P) is
just the expression which occurs in the thermodynamic inequalities. By the defi-
nition of the measures q(p, .), q+(p, .), and q-(p, .) we have for c~r(p) another
expression:

(dq(p, .) dq~ .) ( d q - ( p , .) (2.2)


CKT(P) = d ~ ' ) f d T(')

3. Universes of Bodies. Statement of the Result

In this section we deal with non-empty collections q/ of continuous bodies


such that each body & E q/is endowed with a non-empty class ~ ( ~ ) of processes
p = (0, r, h) of :~. Each such a collection with prescribed classes ~ ( ~ ) , & E q/,
is called a universe of bodies and is denoted, for brevity, by q/.
It is convenient to introduce the following notation and terminology. If q/
is a universe of bodies then ~(q/) denotes the set of all processes of bodies from
q/, i.e.,
~(~) = W (~(~) : ~ E ~);
S(q/) denotes the set of all empirical temperatures acquired in the processes from
:~(q/), i.e.,
S(~ = L/{0(~ X Ip) : p E ~ ( ~ ) , ~ E o//),
and ,/g(q/) is the set of all measures associated with processes from ~(q/),
.//r = {q(p, .) : p E ~(q/)}. (3.1)
If p E ~(&) is a process and I Q R an interval (possibly degenerate) then q1+(p, -)
and qF(P, ") denote the restrictions of the measures q+(p, .) and q-(p, .) to the inter-
val /, i.e.,
q+(p, A) = q+(p, If~ A), q~-(p, A) = q-(p, If~ A)
for any Borel subset A o f R . A Borel measure v on the real line is said to be con-
centrated at 0 E R if the support of v is a subset of the one-point set {0}.
A universe q / o f bodies is said to be an admissible universe of bodies if S(q/) =
kJ {s(p) : p E t~(~)} and if the set :~(q/) contains a subset t~ o with the following
properties:
U1. I f p E ~o then there is a ~ E ~o such that q(fi, .) = --q(p, .).
U2. I f p E ~(ql) and Po E t~o then there is a p' E ~(ql) such that q(p', .) =
q(P, ") q- q(Po, ").
The Clausius Inequality 229

U3. I f Pl, P2 E ~ o then there is a p' E ~ o such that q(p', .) = q(pl, .) -k q(P2, ").

U4. I f 1 ( R is an interval, p E ~ ( ~ ) and 0 E S(ql) then


(1) there is p+ E ~o and a measure r+ concentrated at 0 such that q(po+, .) =
q+(p, .) -- r + ;
(2) there is Po E ~o and a measure r_ concentrated at 0 such that q(Po, ") =
q~-(p, .) -- ~_.

We shall now explain informally the meaning of these requirements. To fix


the ideas, assume that the universe consists entirely from the thermomechanical
bodies (i.e., assume that other than the thermomechanical phenomena are ab-
sent).
Recall first that the classes ~ ( & ) are defined as the sets of all triples (0, r, h)
which correspond to cyclic processes.
The requirements that S(q/) = kY (s(p) : p E ~(q/)} plays a role of an "non-
triviality" assumption; it means that for each possible value of the empirical
temperature there is a process in which the body really exchanges heat with its
environment at that value of temperature.
Now assume that the collection 0//contains a subcollection q/o of elastic bodies.
We take for the set ~0 E ~(q/) the set of all processes of bodies from q/0 which
are piecewise homogeneous, i.e., homogeneous on each connected component
of the body. It is not hard to check formally that piecewise homogeneous processes
of elastic bodies are reversible, i.e., with each piecewise homogeneous process p
of an elastic body there is associated in a natural way another piecewise homo-
geneous process p - - t h e time-reversal of p - - s u c h that q(~, .) = --q(p, .). Hence
U1 holds.
Consider U2. Let p E ~ ( q / ) and p o E ~ o ; hence there is a body & E q /
such that p E ~(:~) and an elastic body &o E q/o such thatpo is a piecewise homo-
geneous process of ~o- Assume first that ~ f~ &o = 13 and that the time-intervals
of p and Po coincide, /p ---- Ipo. We suppose that the union &' = ~ kJ &o is
another body in o//; as the processes p and Po take place simultaneously, they
constitute a process p" of the body &' and we further suppose that p' E ~ ( ~ ' ) .
It is then easily seen that q(p', .) = q(p, .) + q(Po, ") and hence in this special
situation U2 holds. In the general case when :~ A g o 9 O and Ip ~ Ipo we make
use of the possibility of natural retarding or accelerating and "translating in
time" the piecewise homogeneous processes of elastic bodies without affecting
the corresponding measure. By applying such a transformation upon the process
Po we can achieve the coincidence of the time-intervals o f p and Po- By performing
then a suitable translation in space with the body ~ o we can achieve that ~ A ~ o
---- 13 and all the indicated transformations do not affect the measure q(Po, ").
The general case is then reduced to the special one described above, and U2 holds
generally. Note that if the two processes p and Po from U2 coincide then there is
another natural way to construct the process p', namely to take for p' twice
repeated cyclic process p = Po. For this construction the fact that p----Po
belongs to ~ o is unessential and thus one is led to the following condition:

U2 +. I f p E ~(qi) and i f n is any positive integer then there is a p' E ~(ql) such
that q(p', .) = nq(p, .).
230 M. SILHAVY
This condition follows from U2 when p E ~ o but generally U2 + is independent
of U 1 - U 4 ; it is worth noting that for the p r o o f of the main result of the paper the
entirely realistic condition U2 + is not necessary.
The requirement U3 says in addition to U2 that if the two processes p and Po
of U2 are piecewise homogeneous processes of elastic bodies then also the process
p ' can be choosen to be of this type, and the above construction of p ' shows that
this is really the case.
To explain the meaning of U4 we shall formulate another two conditions
whose meaning is more easily visualized and which in conjunction yield U4.
Note first that the restriction q~-(p, .) of the measure q+(p, .) to the interval I ~ E
gives the information what amounts of heat were absorbed by the body during
the process p at temperatures from the interval L The first assumption which stands
behind U4 is that there is in 0//0 an elastic body d o and a piecewise homogeneous
process p+ of 9~o such that ql+(p, .) ~ q(p+, .). The process p+ thus exchanges
the same amounts of heat as the process p absorbs at temperatures from the inter-
val I while outside of I the process p+ does not exchange heat. * I f q+(p, I) ~> 0
then the process p+ cannot be a cyclic process since this would be in conflict
with the second law (see Condition 1 in Theorem 1 below). The next assumption
which must be accepted is that any two piecewise homogeneous states of the elastic
body 9~o* * can be connected by a piecewise homogeneous process p ~ with the
property that 9~o exchanges heat only at a given temperature 0 E S(~//) and other-
wise proceeds adiabatically. * * * In particular we can choose the process p ~ in such
a way that it connects the final state of d o in the process p+ with the initial state
of d o in that process; if we denote by p+ the process of d o which consists of
p + followed by p+ then p0+ is a piecewise homogeneous and cyclic process. Clearly
q(p+, .) --- q(p+, .) q- q(p+, .) and as p+ satisfies q(p+, .) -----q+(p, .) and as in
view of the definition of p+ the measure ~,+ = --q(p+, .) is concentrated at 0,
U4(1) follows. A similar discussion applies to U4(2).
In conclusion of the discussion of the concept of an admissible universe of
bodies note that the requirements U 1 - U 4 could be weakened: an approximate
validity of U 1 - U 4 is entirely sufficient for the p r o o f of Theorem 1 below, but we
shall not obscure this p r o o f with too much generality.
Now we are able to state the main result.

Theorem 1. l f ql is an admissible universe of bodies then the following two con-


ditions are equivalent:
Condition 1. For each p E ~(~ll) we have:
(1) i f q+(p) ~ O then q-(p) ~ O;
(2) /f q+(p) ~ 0 and O+(p) ~ O-(p) then w(p) ~ O.

* Observe the analogy of this assumption with the assumptions about the existence
of quasi-static heat reservoirs of classical thermodynamics. Professor OWEN has kindly
sent me a manuscript of his lecture which contains an explicit construction of the pro-
cess p + in the case when Mo consists of a perfect gas and ql+(p, .) has a finite support.
** i.e., states at which the temperature field and the deformation are homogeneous
on each connected component of the body ~'o.
*** Cf. the Axiom in [23].
The Clausius Inequality 231

Condition 2. There exists a non-decreasing function T: S(~) -+ R ++ such that


for each p E ~(ql) we have fir(P) ~ 0.

Moreover, if the two conditions are satisfied then the function T from Condition 2
is unique to within an arbitrary constant positive factor.

The proof of Theorem 1 will be given in a somewhat more general situation


in Section 5; we now make several remarks.

(a) Item (1) of Condition 1 says, roughly speaking, that if in a cyclic process
p E ~(q/) heat is absorbed then heat must also be emitted in that process. A state-
ment of this type DAY [22] and ~ILI-IAV'? [20] attribute to KELVIN while FOSDICK
& SERRIN [15] to CARNOT and CLAUSlUS. In the cited works, however, the usual
definitions (2.1) of the heat absorbed and the heat emitted are employed rather
than those given in the previous section. More or less equivalent versions of Con-
dition 1(1) in different frameworks have been given by SERRIN [19, 26, 27], OWEN
[24], and ~ILnAVq [28, 29]. A generalization of Condition 1(1) for systems with
approximate cycles is given by COLEMAN,OWEN & SERR~r~[31 ]. Within the present
framework Condition 1(1) alone is insufficient to guarantee the existence of a
non-decreasing absolute temperature scale.

(b) A slight modification of item (2) of Condition 1 is the following:

(2+) if q+(p) > 0 and O+(p)< O-(p) then w(p) < O.

A process p for which O+(p) < O-(p) may be characterized in terms more suggesti-
tive than precise as a process in which heat passes from lower temperatures to
upper temperatures. Item (2+) then says that cyclic processes of this type do
negative work, and statements close to this are commonly attributed to CLAU-
SIUS (cf. DrOHEM [25], SERRIN [18, 19, 26, 27], and ~ILHAV'/" [28, 29]).

(c) Condition 2 asserts the existence of a positive universal absolute temperature


scale Twhich satisfies the Clausius inequality cgr(p) =< 0 for each processp E ~(q/).
The absolute temperature scale T provided by Condition 2 need not be a strictly
increasing function, asserted is only that T is a non-decreasing function. It is
not hard to show that if, in addition to Condition 1 also item (2+) formulated in
the preceding paragraph holds, then the function T necessarily must be strictly
increasing. T also need not be continuous, this would require additional assump-
tions.

(d) The proof that Condition 2 implies Condition 1 is rather trivial; the non-
trivial and more important part of Theorem 1 says that Condition 1 implies
Condition 2.

(e) Theorem 1 also says that the absolute temperature scale is as unique as
should be expected.
232 M. SILHAV~r

4. Classes of Measures

Observe that the quantities appearing in Condition 1 of Theorem 1 are defined


completely in terms of the measures q(p, .) associated with the processes p. By
(2.2) also the Clausius integral r appearing in Condition 2 can be expressed
through the measure q(p, .). Finally, also the requirements U 1 - U 4 from the
definition of an admissible universe of bodies can be restated as assertions about
the class J/l(ql) of measures associated with the processes from ~(q/). (See the
conditions M 1 - M 4 formulated below.) For the p r o o f of Theorem 1 the fact
that the measures q(p, .) arise from the processes of continuous bodies is irrelevant
and the purpose of this section is to introduce a suitable terminology and to state
a preliminary result for an abstract version of Theorem 1 which is Theorem 2
formulated in the next section.
Let B denote the tr-algebra of all Borel subsets of the real line. Henceforth
the term measure will denote any finite, real valued, and signed Borel measure
/~:B--~ 1% with compact support. I f # is a measure then the relation /z ~ 0
means that / ~ ( A ) ~ 0 for all A E B ; the relation / ~ > 0 means that / z > = 0
and /~ 4= 0. Similarly, if/~1 and #a are two measures then the relation #1 ~/~2
means that #2 -- #1 ~ 0 while/~1 <; [Z2 means that/~1 ~ lu2 and/~1 ~ /t't2"
We write /~ = #+ - - / ~ - for the Jordan decomposition of the measure #;
the measures #+ and # - are referred to as the positive and negative variation of #,
respectively; we have /~+ ~ 0 and /~- ~ 0. The supports of/~, #+, a n d / ~ - are
denoted by s(/~), s+(kt), and s-(/~), respectively.

An interpretation for the measures/~ which is appropriate for the present treatment
is that for any A E B the value/ffA) is the net gain of heat of a "thermodynamic system"
at temperatures from the set A.

I f ,u is a measure with s+(/~)% O, we define 0+(#) by 0+(/~)= max s + ~ ) ;


similarly if s-(/z) 4= 0 we define 0--(/~) by/z-(/~) = rain s-(#). (Recall here that
s+(#) and s-(/z) are compact sets.) We further define w(/z), q+(/~), and q- (/~) by
w(#) = / z ( R ) , q+(/z) = #+(R), and q-(#) = # - ( R ) ; hence w(#) = q+(/~) -- q-(/~).
I f 0 E R and/z is a measure, we say that/~ is concentrated at 0 if s(/z) Q {0}.
I f 0 +, 0- E R, 0+ > 0-, and i f # is a measure, we say that/z is a Carnot measure with
operating temperatures (0 +, 0-) if/z+ is concentrated at 0+ and/z- is concentrated at
0--.
The term interval is used to denote any connected subset of R. In particular,
any one-point set (0} Q R is an interval. If/~ is a measure and I Q g an interval
then the restriction of/z to I is defined to be the measure/zt given by #I(A) =
/J(I A A), A E B. It is not hard to see that for any measure we have (#+)t = (/~1)+
and the measure standing on both sides of this equality is henceforth denoted
b y / ~ [ ; similarly we write /z7 = ( [ A - ) i = ( ~ l ) - -
Let K be a compact subset o f R . A finite system 6 ~ of (non-void) intervals is
said to be a partition for K if (1) 50 is disjoint (i.e., L J E 6O and I @ J ~ 1/5 J ~- 0)
and (2) K ( LJ 6O. The norm II6Ot1 of 6O is the maximum of the lengths of the
intervals from 6O. The possibility of II6oll = ~ is not excluded.
Let tt be a measure and T a non-decreasing and positive function whose domain
The Clausius Inequality 233

contains s(/~). The Clausius integral Cgr(/Z) of/z with respect to T is defined by

It is further convenient to introduce the numbers fl+(b0 and fir(#) is by setting

fl+(#) = { lo/T(O+(~u)) ifotherwise,0+(#)isdefined,

is de ned
/0 otherwise.
(If T is interpreted as the absolute temperature scale then in essence fl+(/0 is the
reciprocal value of the maximum absolute temperature at which heat is absorbed
while fli(#) is the reciprocal value of the minimum absolute temperature at which
heat is emitted.) Finally, if 6e is any partition for s(#) then the crude Clausius inte-
gral ~r(#; 6a) of # with respect to T and Sp is defined by

~r(/z; Se) = ~ (flr+(/zl) q+(/z:) -- ~T(~I) q--(~l))"


IE6a

Ifr is a Carnot measure with operating temperatures (0 +, 0-) and Se any parti-
tion for s(/z) then
q-(I~)
~r(#) = ~r(#; ~ ) -- T(O+)
q+(#) T(O-) '
but generally only the inequality
(er(#; Se) < ~r(~) (4.1)
holds. The Clausius integral, however, can be approximated with an arbitrary
degree of accuracy by the crude Clausius integrals. More precisely, we have the
following

Proposition. I f tz is a measure and T a non-decreasing and positive function whose


domain contains s(ct) then
sup c~r(tt; 5e) = c~r(tt) (4.2)
where the supremum is taken over all partitions 6e for s(t O.

Proof of (4.2) requires first to prove (4.1)and then to find a sequence (~N)N~__I
of partitions for s(/z) such that
lim c~r(tt; 5air) = ~r(tt). (4.3)

To prove (4.1), observe that the definition of flr~(ttl) and the fact that Tis a posi-
tive and non-decreasing function yield the inequalities
fl+T([ll) ~ 1/T(O) for each 0 E S+~l)
234 M. ~u.nAvq

and
13~(l~i) ~ 1~TO) for each 0 E s-(/~l).
These inequalities imply

fl+(#,) q+(#,) --/3~-(#i) q-(/z,) ~ f ~


1

and summing the last inequalities over all I E 6 ~ yields (4.1).


Next find a sequence of partitions satisfying (4.3). Denote by ~/" the set of
all points of discontinuity of the function T which are contained in s(/z). As T is
a non-decreasing function, the well known theorem says that JV" is at most count-
able set. Suppose for definitness that JV" is infinite. (If~f" is finite or empty, the
proof given below is easily modified.) We can enumerate the points of JV" to obtain
a sequence (ON)N~=a. Then we take for the N-th partition 5aN any partition for
s(#) which satisfies 115aNII< 1/N and which contains the degenerate intervals
{01} , {02} . . . . . {ON). (Such a partition obviously exists.)The sequence ( s a N ) , ~ = l thus
constructed has the following properties:
(1) lim llsauH = 0;
N--->c~

(2) if 0 E JV" then there exists a No = No(O) such that for all N => No we have
{0} E YN.
As for each N = 1, 2 . . . . the system 5aN is a partition for s(#), for each 0 E s(#)
there is a unique interval I(N, O) E 5aN such that 0 E I(N, 0). We now define
+ oo
two sequences (f~,)N=I and ( f N ) ~ - I of functions from s(#) into R++ as follows:
If N = 1,2 . . . . . and 0 E s ( # ) then

f;(0) =
{
1/T(O+(m(N,O

l/T(0)
) if O+(#~(N,O)) is defined,
otherwise,

-(y1(lv, o))) if O--([Zl(N,O) ) is defined,


f u (O) = {1/T(O
1/T(O) otherwise.
If 0 E s ( # ) then
lim f + ( O ) = 1/T(O), limoJN(O ) = 1/T(O). (4.4)
N ~- or)

To prove (4.4)1, assume first that 0 E./r In view of the property (2) of the
sequence (sau)~- 1 we have for sufficiently large values of N (namely for N >=No(O))
the relation I(N, 0) = {0). If in this case 0+(/it(N,0)) is defined, it is necesssarily
equal to 0 and hence the definition o f f + ( 0 ) yields f+(O) = 1/T(O). If 0+(/~t(u,0))
is not defined then by the very definition f+(O) = 1/T(O). To summarize, if 0 E~f"
then for sufficiently large values of N the relation f+(O) = 1/T(O) holds and this
implies that (4.4)1 is valid. Next suppose that 0 E s(#) \./V'. Then the function T
and hence also lIT is continuous at 0. We shall show that for each e > 0 there
is a Nx such that
If+O) - 1/T(O)i < e. (4.5)
The Clausius Inequality 235

whenever N > N1. Accordingly, let e > 0 be a given number. As 1/T is con-
tinuous at 0, there is a ~ > 0 such that for each 0' in the domain of T satisfying
[ 0 ' - - 01 < 6 there holds ]I/T(O') -- 1/T(O)I < e. By the property (1) there is
a N1 such that for each N > NI the length of I(N, O) is less than & Therefore,
if N > N , and O+(/t,~N.O~) is defined then [0+(/t,~N.0})-- 01 < ~. The last in-
equality implies I1/T(O+(1*1{N.o))) -- 1/T(O)I < e, which in view of the definition
of f + ( 0 ) can be rewritten as (4.5). If N > N, but if 0+OZZ(N,0~)is not defined then
by the very definition o f f + ( 0 ) we have f+(O) = I/T(O) and hence (4.5) holds
again. This completes the p r o o f of (4.4)1. The proof of (4.4)2 is similar and is
omitted here.
In view of the definition of the crude Clausius integral it is not hard to see
that

r#r(/t; SeN) = f f u+ d1*+ - - f f~7 d/t- (4.6)

for each N = 1, 2 . . . . . By using (4.4) and the dominanted convergence theorem


(the existence of dominants being clear) we conclude that the right hand side of
(4.6) tends to

fd +
R R

as N---> oo and this proves (4.3), q.e.d.


In conclusion of this section we introduce the concept of an admissible collec-
tion of measures. Let J// be any non-empty set of measures; we write
S(~r = W (s(~) : 1. E .///}
for the union of the supports of measures from Jg and
D ( . e ) = {(0+, 0-) E • S( g) : 0+ > 0-}

for the set of all pairs (0 +, 0-) from S ( J / ) x S(d/) satisfying 0+ > 0-.
A non-empty set J / o f measures is said to be an admissible collection o f measures
if S(.///) 4= 0 and if.A/contains a subset -/go C ~ with the following properties:

M1. / f 1. E J [ o then --1. E ,-go.


M2. / f / t E . / r and 1*oEd/o then 1 * - k # o E d r
M3. I f /tl and /t2 E .glo then t'1 q- #2 E Jgo.
M4. I f 1 C R is an interval, /t E ~11 and 0 E S(#) then
(1) there is a measure v+ concentrated at 0 such that 1.+ -- v+ E -/go;
(2) there is a measure ~,_ concentrated at 0 such that 1.i- -- v_ E Jlto.

In particular, if ag is an admissible universe of bodies then the collection .4'l(~


of measures associated with processes from ~(aR) is an admissible collection of
measures: In view of the properties U1-U4, the set -g/o = {q(P, ") : P E ~o} is easily
seen to have the properties M 1 - M 4 .
236 M. SILHAV~"

5. Abstract Version of the Result and the Proof

Theorem 2. I f ~ l is an admissible collection of measures then the following


two conditions are equivalent:

Condition 1. For each # E d[ we have:

(1) if q+(#) > 0 then q-(#) > 0;


(2) i f q+(fO > 0 and 0+(#) <: 0-(#) then wOO <=o.

Condition 2. There exists a non-decreasing function T: S(#) --~ R++ such that
for each t ~ E ,//4 we have ~r(#) <: O.

Moreover, if the two conditions are satisfied then the function T from Condition 2
is unique to within an arbitrary constant positive factor.

If we set in this theorem ./g = .~,(o-//), where q/is an admissible universe of


bodies, we get Theorem 1 of Section 3.

Proof of Theorem 2. We first prove that Condition 2 implies Condition 1.


Let /z E dr' and q+(#) > 0. We have, by hypothesis, c~r(/z) ~ 0. The inequality
(4.1) with a special choice of a partition 5a, namely with 5e consisting of a single
interval I = B (i.e., 5P = {R}), yields c~r(/t;{R})~ 0. This means

fl+(#) q+(#) -- fl~-(/z) q-(#) ~ 0. (5.1)


As q+(#) > 0, the number fl+(#) is equal to 1/T(O+(#)) and hence is strictly posi-
tive. The validity of (5.1) requires that q-(/0 > 0 and this proves item (1) of
Condition 1. The fact that q-(#) > 0 in turn implies that fir(#) = 1/T(O-(t~)),
and (5.1) can be rewritten, by using the equality w(/z)= q+(/~)- q-(#), as

W(~) <= [1
_ ~T(O-(~))
] 1 q+(p). (5.2)

Therefore, if 0+(#) < 0-(#) then also T(O+(~)) < T(O-(IX)) (recall that T is a
positive and non-decreasing function) and (5.2) implies w(/x) < 0; hence item (2)
of Condition 1 holds and the first part of the proof is complete.
The converse implication, that Condition 1 implies Condition 2, has a more
difficult proof which is divided into Lemmas 1-4. To avoid repeated hypotheses,
assume henceforth that Condition 1 holds and that J/go is the subset of J / w h o s e
existence is guaranteed by the definition of an admissible collection of measures.
The plan of the proof is as follows. We first realize in Lemma 1 that there are non-
trivial (i.e., different from zero) Carnot measures in ~'o with arbitrary operating
temperatures. The function T, whose existence is to be proved, first appears in
Lemma 2; its characteristic property is that it satisfies the classical relation for the
heat absorbed and the heat emitted in a reversible Carnot process. Lemma 3 then
establishes the validity of the Clausius inequality for measures from J//which have
a finite support. In Lemma 4 we finally prove the inequality asserting that the crude
The Clausius Inequality 237

Clausius integral corresponding to any partition for the s u p p o r t o f any measure


from -g can never be positive; as by P r o p o s i t o n of the preceding section the crude
Clausius integrals a p p r o x i m a t e with arbitrary degree of accuracy the Clausius
integral, the Clausius inequality will follow.

L e m m a 1. For any (0 +, 0-) E D ( - g ) there is a Carnot measure /t E -go with


operating temperatures (0+, 0-) such that q+(/t) > 0 and q-(/t) ~ O.

Proof. As S ( ` g ) = ~ 0, there is a non-zero measure /to in all. Necessarily


/to > 0 for in view o f Condition 1(I) the a s s u m p t i o n # o = 0 would lead to
/ t + - - - - 0 and hence to /to----0. We n o w set I - - - - R and apply M4(2) to the
m e a s u r e / t o and to the t e m p e r a t u r e 0 + to show that there is a measure v+ concen-
trated at 0 + such t h a t / t o - - v+ E ./go. Similarly we prove by M4(2) the existence
o f a measure v_ concentrated at 0- such that /to -- v_ E ~#o. The m e a s u r e / t ,
whose existence is to be proved, is defined by /t ----~+ -- v_ = --(#~- - - v+) q-
(/to - - v _ ) . By using M1 and M 3 we find that /t E `go. F r o m the relations
/to -- v+ E - g o , /to - - v_ E - g o and /to > 0 we further deduce, by using once
m o r e Condition 1(1), that v+ > 0, v_ > 0. As the measures v+ and v_ are concen-
trated at 0 + and 0-, respectively, we see t h a t / t is a C a r n o t measure which has also
the other required properties, q.e.d.

L e m m a 2. There is a non-decreasing function T: S(Jg) ~ R++ such that for


any Carnot measure /t E d[o with operating temperatures (0 +, 0-)E D(J/[) we
have
T(O-)
q-(/t) -- T - ~ q+(/t)" (5.3)

Proof. We first prove an auxiliary result: I f /tl,/t2 E ~'~r are two C a r n o t


measures with operating temperatures (0 +, 0-) E D ( ` g ) then

q+(/tl) q-(/t2) = q+(/ta) q - ( / t 0 - (5.4)

I f q-(/tj) ---- 0 for some (or both) j E {1, 2} then by Condition 1(1)also q+(/tj) -----0
a n d (5.4) holds. We n o w assume that q - ( / t l ) > 0 and q - ( / t ~ ) > 0. Let r be
a n y rational n u m b e r satisfying 0 < r < q-(/tz)/q-(/tO. Writting r = nl/n2
with n~, n2 positive integers, we cart restate the inequality r < q-(/tz)/q-(/tO as

n2q-(#2) - - nlq-(/tl) > O. (5.5)

Since #1 a n d / t 2 belong to Jgo, M1 and M 3 imply that the measure # given by


# = nl/tl -- n2/t2 also belongs to Jgo- N o w /tl and /tz are C a r n o t measures
with operating temperatures (0 +, 0-) and hence

/ t j = q+(/tj) ~0+ - - q-(Pj) $0-


f o r j ---- 1, 2, where b0+ and b0- are the Dirac measures at 0+ and 0% respectively.
Consequently

/t ---- [nlq+(/tt) -- n2q-(~2)] ~o+ + [nzq-(/t2) -- nlq-(/tx)] ~0-.


238 M. SlLHAVY

The above expression together with (5.5) imply that q+(/z) ~ n2q-(/t2) -- nlq-(#l)
> 0 which in view of Condition 1(1) is possible only when q-(#) > 0; this re-
quires that nlq+(tzO -n 2 q + ( / h 2 ) < 0. Recalling that r = rtl/rt 2 we obtain from
-

the last inequality that


rq+(/z,) < q+(#2). (5.6)
This must hold for arbitrary rational r satisfying 0 < r < q-(tZa)/q-(#l), and
on letting r tend to q-(iz2)/q-(,ul) in (5.6) we get, after some rearrangement,

q+(/zl) q-(#2) ~ q+(#2) q-(/zl).


By interchanging the roles of/zl and #2 we obtain from this also the opposite
inequality and the proof of (5.4) is complete.
The auxiliary result (5.4) and Lemma 1 imply the existence of a unique func-
tion L : D(J/r -+ R++ such that for any Carnot measure /~ C J/go with operating
temperatures (0§ 0-)E D(~') we have
q-(tz) = L(O+, 0-) q+(iz). (5.7)
The next step of the proof is to show that L admits the decomposition
L(O+, 0-) = T(O-)/T(O+), (0+, 0-) E D ( ~ ) (5.8)
with T a positive function, T: S(~/) - + R ++. To do this, choose 01, 02, 03 E S(J/r
such that 01 < 0z < 03. By Lemma 1 there is a Carnot measure /~ E ~'o with
operating temperatures (03, 01) and with q+(~) > 0, q-(#) > 0. By M4(1) there
is a measure v concentrated at 02 such that /zz =
def/t~+ -- r E clio. By M1 then
def
also --/~2 E J~'o so that by M 3 / z l = # --/z2 = ~, - - / z - ~ ~1/o. It is not hard to
see that/~1 and/~2 are Carnot measures with operating temperatures (02, 01)
and (0a, 02) respectively, which satisfy
q+(/*2) = q+(/z), q-(/z2) = w(~),
q+(/za) = w(~), q-(/q) = q-(#).
By applying (5.7) to the measures/z,/~1, and/Zz we get
q-(I ~) = L(03, 01) q+(I-O,
q-(t z) = L(Oz, 01) w(~),
w(r) = L(03, 02) q+(#).
This is a system of three linear equations for q+(/~), q-(#) and w0'), which has a
non-trivial solution (recall that q+(/~) > 0). Hence the determinant of the system
is zero and this leads to the equality
L(O3, 0,) = L(03, 02) L(02, 00,
which must be satisfied for any 01, 02, 03 E S(dr such that 01 < 02 < 03.
From this the validity of the decomposition (5.8) follows by a standard argument.
Moreover, with (5.8) the relation (5.7) reduces to (5.3).
The only thing that now remains to be proved is that T: S(,///)-+ R++ is
a non-decreasing function. Let (0+, 0-) be in D(J//). By Lemma 1 there is a Carnot
The Clausius Inequality 239

measure # E ~ ' o with operating temperatures (0 +, 0-) such that q+(#) > O,
q-(/z) > 0. This measure satisfies (5.3). We also have 0+(--#) = 0-, 0 - ( - - # ) = 0+
and thus 0+(--/0 < 0+(--~). M1 implies that --/rE dr and Condition 1(2)
applied to --/t yields that 0 ~> w(--#) = --w(#) ---- q-(/~) - - q+(/z), i.e., q+(/~)
q-(#), which in conjunction with (5.3) implies T(O-) ~ T(O+), q.e.d.

Lemma 3. I f / , E ~g has a finite support then cgr(/~) ~ 0.

Proof. I f s(/O is a finite set then also s+(/O and s-(/O are finite sets and we
have
# = y~ z t o - ~ Zto (5.9)
(9 |

where for each 0 E s(#) = s+(/0 k] s-(#) the measure z~0 > 0 is concentrated
at 0 and where the symbols ~ and ~ stand for sums over all elements f r o m
(9 |
s+(#) and s-(/z), respectively. We set 0o = max s(/0. I f 0 E s+(#) and I = (0}
then #+ = zto and M4(1) tells us that there is a measure ro concentrated at
0o such that Z~o - - v o E dl'o. Similarly by M4(2) for each 0 E s-(#) there is a
measure vo concentrated at 0o such that Z~o -- ~'oE ~ ' o . To summarize, for each
0 E s(tz) there is a measure vo concentrated at 0o such that zro - - q'o E clio. Since
2to > 0, Condition 1(1) implies that also ro > 0 for each 0 E s(#). As the meas-
ures zoo and vo are concentrated at 0 and 0o, respectively, we see that for each
0 E sQz) the measure ~to given by
/*o = ~o -- ~o (5.10)

is a Carnot measure with operating temperatures (0o, O) which satisfies q+(/~o)


= W(Vo), q-(/~o) = w(~o). The relation ~ o - ro E J'/o and M1 imply that
#o E -go. By L e m m a 2 then
T(0)
w( o) = r- o) w(,,o)
for each 0 E s(/O. It follows that

w(=o)
- - -
w(~o) 1 w@o)] ,
= 7"(o)

i.e.,
1
= w(O (5.11)

where
t r = Y ~ ' o - - S~ vo. (5.12)

As for each 0 E s(#) the measure #o belongs to J / o , we deduce from M1 and M3


hat also the measure
0 = N / z o -- N # o (5.13)
240 M. SILHAVY
belongs to d/o. But in view o f (5.9), (5.10), (5.12), and (5.13) we have a = # + 9
and as /z E J/[ and 9 E d / o , M2 tells us that tr E d'r The measure a is concentrat-
ed at 0o and hence either tr ~ 0 or tr > O. Because a E de', the assumption
(r > 0 would be in conflict with Condition 1(1), for then we would have q+(a) > 0
and q-(a) ---- O. Hence necessarily a ~ 0 and the inequality Cgr(#) =< 0 follows
f r o m (5.11), q.e.d.

Lemma 4. I f t, E ~//l then f o r each partition 5" f o r s(#) we have

~ r ( ~ ; ~ ) _-< 0. (5.14)

Proof. W e define for each I E 6e the measure vl as follows. I f # + = 0 we


set vl ---- 0; if /z+ > 0 then 0+(#1) is defined and by M4(1) there is a measure
~'i concentrated at 0+(/,1) such that # + -- vl E Jgo. We shall now show that for
each 1E 6 ~ there holds

fl+(/*l) q+(&) ~< Cgr(V~). (5.15)


I n view o f the definition o f vz the inequality is obvious when # + = 0. Next
suppose that /z+ > 0. I f the m e a s u r e / , + is concentrated at 0+(#i) then also
/~+ -- vz (which belongs to Jgo) is concentrated at 0+(/~x) and hence the measure
/ ~ + - v z has a finite support. L e m m a 3 then tells us that 0 ~ c g r ( / z + - Vl)
= cgr(#+) -- cgr(vl); the relation cgr(#+ ) = fl+(ttl) q+(/,~) which holds in this
case shows that (5.15) holds again. Hence there remains the case when tt + > 0
and tt + is not concentrated at 0+(#i). As tt + -- Vl E Jllo, Condition 1(1) implies
that Vl > 0 and the well-known minimality o f J o r d a n decomposition together
with the assumption that # + is not concentrated at 0+(tq) yield

0 < ( / V - ~1) + =< F,,+, 0 < (t,,+ - v l ) - __< ~,.

It follows that 0 =~ s+(iz + -- vx) C s(Iz+) ~- s+(#l), 0 ~= s-(tz + - vl) Q s(v~) =


{0+(btl)}. Hence 0+(# + -- vl) = max s+(# + -- vl) ~ max s+(/zl) = 0+(#1), and
0-(/z + -- v,) = min s - ( p + -- vl) >~ min {0+(/Zl)} = 0+(#,) and thus O+(p[ -- v,)
0-(tz + -- v/). Condition 1(2) then yields w(tz+ -- vt) ~ 0, which can be
rewritten as q+(/zz) ~ w(v1). It follows

q+(#1) < w(rl)


r(O+(/~,)) = T(O+(t,,))
and the definitions of flr+(/tl) and o f (gr(vl) yield (5.15).
In the preceding part o f the p r o o f only the positive variation #+ o f / z has
been employed. The next-part deals in a similar way with the negative variation
# - . We first define for each I E 5~' the measure OZl as follows. We set z~1 = 0 if
Yi- = O, while if /z/- > 0 then by M4(2) there is a measure ~i concentrated at
0-(/.tl) such that #~- -- az, E d//o. By using the similar arguments as above one
can prove that for each I E 6 a there holds

--fl~-(/zl) q-(vl) <= --(~r(z~l) (5.16)

but the p r o o f is omitted here.


The Clausius Inequality 24I

With the help of (5.15) and (5.16) the p r o o f is now easily completed. First,
by adding (5.15) to (5.16) we get

fl+ (,u,,) q+ (ttt) -- fir(#1) q-(tzt) ~ c~r(vl -- ~h)


for each I E Y , and by summing these inequalities over all I E 6e we obtain
C~r(/Z; Y ) ~< ~r(~), (5.17)
where

IESe

As /~+ -- viEv//o and /~/- - - 2ttEdr for each I E S e , by M1 and M3 also


--p~ + vl - - z~1 = ( # i -- z~z) - - (/t+ -- vl) E d'r by using once again M2 we
deduce that --/z + r = ~ (--/zt + vl - - z~l) E dr Finally, as /_t E JAg and
IE6a
--/z + Q E de'o, M2 tells us that ~ = / z + (--/z + ~) E de. As ~ has a finite sup-
port, L e m m a 3 indicates that (fr(9) =< 0 and (5.17) implies (5.14) and the p r o o f
of L e m m a 4 is complete.
As has been pointed out in the paragraph preceding L e m m a 1, Proposition
stated in Section 4 and L e m m a 4 yield the Clausius inequality for any measure
f r o m ~ ' and thus the Condition 2 holds.
The only thing that now remains to be shown is that if the two equivalent
Conditions 1 and 2 are valid then the function T of Condition 2 is unique to within
arbitrary positive constant factor. Obviously if T satisfies Condition 2 then also
cT, with c > 0, satisfies that condition. Conversely, if T is any positive function
satisfying Condition 2, then the Clausius inequality reduces to the equality
~ r ( # ) = 0 for measures # from ~ ' o . In particular, if # E dt'o is a Carnot meas-
ure with operating temperatures (0 +, 0-) E D(J//') then the equality c~r(/z) = 0,
yields the equality (5.3). As there are, by L e m m a 2 (which holds since Condition 1
holds), Carnot measures with arbitrary operating temperatures and with q+(/~) > 0
q+(#) > 0, the equality (5.3) obviously determines T to within a constant positive
factor. The p r o o f of Theorem 2 is now complete.

Acknowledgement. I wish to express my deep thanks to Dr. J. KRATOCHVIL,CSc, for


numerous discussions on various aspects of thermodynamics and for suggesting many
improvements in the manuscript. I also thank Professor D. R. OW~N for sending me
the manuscript [24] and for drawing my attention to the paper [19].
Finally, I would like to thank Professor J. B. SERmNfor many valuable suggestions
concerning a previous draft of the paper.

References

1. TRUESDELL,C., & R. A. TOUPIN, The Classical Field Theories. FL~GGE'S Handbuch


der Physik, III/1. Bedin-Gtittingen-Heidelberg: Springer 1960.
2. COLEMAN,B. D., & W. NOEL, The Thermodynamics of Elastic Materials with Heat
Conduction and Viscosity. Arch. Rational Mech. Anal. 13, 167 (1963).
3. COLEMAN, B. D . , Thermodynamics of Materials with Memory. Arch. Rational
Mech. Anal. 17, 1 (1964).
242 M. SILHAVY

4. DAY, W.A., The Thermodynamics of Simple Materials with Fading Memory.


Springer Tracts in Natural Philosophy, Vol. 22. Berlin-Heidelberg-New York:
Springer 1972.
5. COLEMAN,B. D., • D. R. OWEN, A Mathematical Foundation for Thermodynamics.
Arch. Rational Mech. Anal. 54, 1 (1974).
6. GURTIN,M. E., On the Thermodynamics of Materials with Memory. Arch. Rational
Mech. Anal. 28, 40 (1968).
7. COLEMAN,B. D., & D. R. OWEN, On the Thermodynamics of Materials with Me-
mory. Arch. Rational. Mech. Anal. 36, 245 (1970).
8. COLEMAN,B. D., & M. E. GURTIN, Waves in Materials with Memory. III. Thermo-
dynamics Influences on the Growth and Decay of Acceleration Waves. Arch. Ra-
tional Mech. Anal. 19, 266 (1965).
9. GURTIN,M. E., Thermodynamics and Stability. Arch. Rational Mech. Anal. 59,
63 (1975).
10. DAY, W. A., A Theory of Thermodynamics of Materials with Memory. Arch. Ra-
tional Mech. Anal. 34, 85 (1969).
11. DAY, W.A., Entropy and Hidden Variables in Continuum Thermodynamics.
Arch. Rational Mech. Anal. 62, 367 (1976).
12. COLEMAN,E. D., & D. R. OWEN, On Thermodynamics and Elastic-Plastic Materials.
Arch. Rational Mech. Anal. 59, 25 (1975).
13. COLEMAN,B. D., & D. R. OWEN, On Thermodynamics and Intrinsically Equili-
brated Materials. Annali Mat. Pura Applicata (IV), 108, 189 (1976).
14. COLEMAN,B. D., & D. R. OWEN, On the Thermodynamics of Semi-Systems with
Restrictions on the Accessibility of States. Arch. Rational Mech. Anal. 66, 173
(1977).
15. FOSDICK,R.L., & J. SERRIN, Global Properties of Continuum Thermodynamic
Processes. Arch. Rational Mech. Anal. 59, 97 (1975).
16. ARENS,R., An Axiomatic Basis for Classical Thermodynamics. J. Math. Anal.
Appl. 6, 207 (1963).
17. BOYLING,J. B., An Axiomatic Approach to Classical Thermodynamics. Proc. R.
Soc. Lond. A. 329, 35 (1972).
18. SERRIN,J., Foundations of Classical Thermodynamics. Lecture Notes, Depart-
ment of Mathematics, University of Chicago, 1975.
19. SERRIN,J., The Concepts of Thermodynamics. In: Contemporary Developments
in Continuum Mechanics and Partial Differential Equations, G. M. de La Penha
& L. A. J. Medeiros (eds.), p. 411. Amsterdam: North-Holland 1978.
20. SILHAVY,M., A Condition Equivalent to the Existence of Non-Equilibrium Entropy
and Temperature for Materials with Internal Variables. Arch. Rational Mech. Anal.
68, 299 (1978).
21. HALMOS,P. R., Measure Theory. New York: D. Van Nostrand Co. 1950.
22. DAY, W. A., A Condition Equivalent to the Existence of Entropy in Classical
Thermodynamics. Arch. Rational Mech. Anal. 49, 159 (1972).
23. DAY, W. A., & M. SILHAV~',Efficiency and the Existence of Entropy in Classical
Thermodynamics. Arch. Rational Mech. Anal. 66, 73 (1977).
24. OWEN,D. R., An Elementary Proof of Serrin's Accumulation Inequality in Classical
Thermodynamics. Manuscript of a lecture. Private communication (1979).
25. DUIJEM,P., Trait6 d'Energetique ou de Thermodynamique G6n6rale. Paris: Gauthier-
Villars 1911.
26. SERRIN,J., Lectures on Thermodynamics, University of Naples, 1979.
27. SERRIN,J., Conceptual Analysis of the Classical Second Laws of Thermodynamics.
Arch. Rational Mech. Anal. 70, 355 (1979).
The Clausius Inequality 243

28. ~ILHAVq,M., On the Clausius Inequality. Lecture presented at Euromech 111 Sym-
posium, September 26-28, 1978, Mari~nsk6 L~izn6, Czechoslovakia. Abstracts,
p. 68. Czechoslovak Academy of Sciences and Skoda, National Corporation,
Plzefi 1978.
29. ~ILHAV~',M., On Measures, Convex Cones, and Foundations of Thermodynamics.
I. Systems with VectorValued Actions. Czech. J. Phys. B 30, 841 (1980).
30. SILHAVq,M., On Measures, Convex Cones, and Foundations of Thermodynamics.
II. Thermodynamic Systems. Czech. J. Phys. B 30, 961 (1980).
31. SERRIN,J., The Foundations of Thermodynamics. Gaz. des Sci. Math. Quebec (to
appear).
32. COLEMAN,]3. D., D. R. OWEN • J. SERRIN, The Second Law of Thermodynamics
for Systems with Approximate Cycles. Arch. Rational Mech. Anal. 77, 103 (1981).
33. KRATOCHViL,J., & M. ~ILHAVY, O termodynamice re,~ln~ch fyzik~lnich d~j~ (in
Czech). Czech. J. Phys. A 31, 99 (1981).
34. SILHAVY,M., On the Second Law of Thermodynamics. I. General Framework.
Czech. J. Phys. B 32, 987 (1982).
35. SILHAV3(,M., On the Second Law of Thermodynamics. II. Inequalities for Cyclic
Processes. Czech. J. Phys. B 32 (1982).

Mathematical Institute
Czechoslovak Academy of Sciences
~itn~ 25
11567 Prague 1
Czechoslovakia

(Received September 15, 1979; revised May 6, 1981)

View publication stats

You might also like