May 29, 2002 9:12 WSPC/103-M3AS 00186

Mathematical Models and Methods in Applied Sciences

Vol. 12, No. 5 (2002) 721736
c World Scientic Publishing Company
ON THE MEASURE-THEORETIC FOUNDATIONS OF
THE SECOND LAW OF THERMODYNAMICS
ALFREDO MARZOCCHI

and ALESSANDRO MUSESTI

Dipartimento di Matematica, Universit`a degli Studi di Brescia,

Via Valotti 9, I-25133 Brescia, Italy

amarzocc@ing.unibs.it

a.musesti@dmf.unicatt.it
Received 18 May 2001
Revised 28 November 2001
Communicated by N. Bellomo
Balance laws of the type of entropy are treated in the framework of geometric measure
theory, and a weak version, although conceptually simple, of the Second Law of Ther-
modynamics is introduced, allowing extensions to measure-valued entropy productions
and to sets of nite perimeter as subbodies.
Keywords: Cauchy interactions; entropy; sets with nite perimeter.
AMS Subject Classication: 74A15, 80A05, 80A17
1. Introduction
In this paper we reconsider the standard formulation of the Second Law of Ther-
modynamics in the general framework of elds with divergence measure, using
Geometric Measure Theory and some recent results on Cauchy interactions. To our
knowledge, the rst attempt to put the Second Law of Thermodynamics within the
framework of Geometric Measure Theory is Gurtin, Williams, Ziemer.
4
Recently,
5
the present authors studied the balance of heat and generalized it to elds having
divergence measure and measure-valued sources. Here, with respect to Ref. 5, more
general balances are considered, involving inequalities and superadditive entropy
production functions, as pointed out in Ref. 4. Two main improvements have been
achieved: rst, we obtain the existence of temperatures and the ClausiusDuhem
inequality with ux vector elds having divergence measure, a degree of generality
which seems to be the highest compatible with the representation results of general
uxes (see, for example, the discussion in Degiovanni, Marzocchi, Musesti
1
). To this
end, we dropped the absolute continuity of the uxes with respect to Lebesgue mea-
sure: this is not an idle generalization, since it is conceivable to have some form of
concentrated heat or entropy uxes. Second, and perhaps more important, we nd
that the usual statement of the Second Law of Thermodynamics has to be weakened
721
May 29, 2002 9:12 WSPC/103-M3AS 00186
722 A. Marzocchi & A. Musesti
in order to have the ClausiusDuhem inequality valid for almost all subbodies: in
fact, a particularly simple class of multi-intervals is sucient to extend the result
on almost all subbodies of nite perimeter, i.e. on a very wide class. In the present
context it appears that the class of multi-intervals is preferable to that of simple
n-intervals, since the lack of additivity of involved quantities (namely, the entropy
production) does not allow natural extensions from n-intervals.
After deniting our basic denitions in Sec. 2, we state the entropy inequality
in Sec. 3 and obtain the existence of temperatures in Sec. 4. Section 5 is devoted
to the extension result, and Sec. 6 contains some technical proofs.
2. Basic Denitions
In this section we recall the denitions and tools of Geometric Measure Theory
which extend the usual notions of e.g. normal, boundary, smooth vector eld, to-
gether with the main denitions we use in the sequel. For a deeper and more
systematic introduction, we refer the reader to Federer
2
or Ziemer.
8
We denote by L
n
the Lebesgue outer measure on R
n
and by H
k
the
k-dimensional Hausdor outer measure.
Let M R
n
. By cl M and int M we denote the closure and the interior of M
in R
n
, respectively. When M is a Borel set, we also denote by B(M) the -algebra
of Borel subsets of M.
Denoting by B
r
(x) the open ball with center x and radius r, we introduce
M

=
_
x R
n
: lim
r0
+
L
n
(B
r
(x)\M)
L
n
(B
r
(x))
= 0
_
and

M = R
n
\[M

(R
n
\M)

] ,
(the so-called measure-theoretic interior and measure-theoretic boundary of M, re-
spectively), which are Borel subsets of R
n
. If M

= M, we say that M is normalized.

Now let x

M. We call the unit exterior normal vector to M at x the vector

u
x
R
n
such that |u
x
| = 1, and
lim
r0
+
L
n
({ B
r
(x) M : ( x) u
x
> 0})
L
n
(B
r
(x))
= 0 ,
lim
r0
+
L
n
({ B
r
(x)\M : ( x) u
x
< 0})
L
n
(B
r
(x))
= 0
(no more than one such vector can exist). Hence we can dene a bounded Borel
map n
M
:

M R
n
, called the unit exterior normal to M, setting n
M
(x) = u
x
where it exists and n
M
(x) = 0 otherwise.
We say that M has nite perimeter if H
n1
(

M) < + (this implies the

L
n
-measurability of M). In that case, n
M
(x) = 0 for H
n1
-a.e. x

M and the
May 29, 2002 9:12 WSPC/103-M3AS 00186
Foundations of Second Law 723
GaussGreen Theorem
_
M
v gradf dL
n
=
_

M
f v n
M
dH
n1

_
M
f div v dL
n
holds whenever f : R
n
R and v : R
n
R
n
are Lipschitz continuous with
compact support (see Federer,
2
Theorem 4.5.6 or Ziemer,
8
Theorem 5.8.2).
Now let be an open subset of R
n
. We denote by L
1
loc
(; R
n
) the set of Borel
maps v : R
n
with
_
K
|v| dL
n
< + for any compact subset K of , by
L
1
loc,+
() the set of Borel functions h : [0, +] with
_
K
hdL
n
< +for every
compact subset K and by M() the set of Borel measures : B() [0, +]
nite on compact subsets of . Moreover, for , M() we will say that is
absolutely continuous with respect to , and write , if
E B() : (E) = 0 =(E) = 0 .
Let q L
1
loc
(; R
n
); we say that q has divergence measure, if div q is a distribution
on of order 0, i.e. for every compact subset K of there exists a constant c
K
such that

q gradf dL
n

c
K
max
K
|f|
for every f C

0
() with supt f K.
Throughout the remainder of this work, B will denote a bounded normalized
subset of R
n
of nite perimeter, which we call, in the spirit of

Silhav y,
6
a continuous
body.
Denition 2.1. Let M be the collection of all normalized subsets of B of nite
perimeter. We set
M
loc
= {A M: cl A int B} ,
N
loc
= M
loc
{A (R
n
\B)

: A M
loc
} ,
D
loc
= {(A, C) M
loc
N
loc
: A C = } .
Let furthermore h L
1
loc,+
(int B) and M(int B). We set
M
loc
h
=
_
A M
loc
:
_

A
hdH
n1
< +, (

A) = 0
_
,
N
loc
h
= {C N
loc
: (C B) M
loc
h
},
D
loc
h
= D
loc
(M
loc
h
N
loc
h
) .
Denition 2.2. We say that D M
loc
contains almost all of M
loc
, if M
loc
h
D
for some h L
1
loc,+
(int B) and M(int B), and that a property holds almost
everywhere in M
loc
, if the set
{A M
loc
: (A) is dened and (A) holds}
May 29, 2002 9:12 WSPC/103-M3AS 00186
724 A. Marzocchi & A. Musesti
contains almost all of M
loc
.
We say that a subset D of D
loc
contains almost all of D
loc
, if D
loc
h
D for some
h L
1
loc,+
(int B) and M(int B), and that a property holds almost everywhere
in D
loc
, if the set
{(A, C) D
loc
: (A, C) is dened and (A, C) holds}
contains almost all of D
loc
.
For a full discussion on the concept of almost all etc. the reader is referred
to

Silhav y
6
and Degiovanni, Marzocchi, Musesti.
1
The rough idea is to leave out
subbodies which may encounter singular sets of the measures involved, but not too
many.
Denition 2.3. A grid G is an ordered triple
G = (x
0
, (e
1
, . . . , e
n
),

G) ,
where x
0
R
n
, (e
1
, . . . , e
n
) is an ordered orthonormal basis in R
n
and

G is a Borel
subset of R. If G
1
, G
2
are two grids, we write G
1
G
2
if the rst two components
coincide and

G
1


G
2
. A grid G is said to be full, if L
1
(R\

G) = 0.
Let G be a grid; a subset I of R
n
is said to be an open n-dimensional G-interval,
if
I = {x R
n
: a
(j)
< (x x
0
) e
j
< b
(j)
j = 1, . . . , n}
for some a
(1)
, b
(1)
, . . . , a
(n)
, b
(n)

G. We set
I
G
= {I : I is an open n-dimensional G-interval with cl I int B} .
The following proposition is a particular case of Proposition 4.5 in Degiovanni,
Marzocchi, Musesti.
1
Proposition 2.1. Let x
0
R
n
and (e
1
, . . . , e
n
) be an ordered orthonormal basis
(e
1
, . . . , e
n
) in R
n
. Then for every h L
1
loc,+
(int B) and M(int B) there exists
a full grid G of the form G = (x
0
, (e
1
, . . . , e
n
),

G) such that I
G
M
loc
h
.
Denition 2.4. Given a grid G, a subset P of R
n
is said to be a G-gure, if
P = (

IF
I)

, where F is a nite family of open n-dimensional G-intervals.

For any grid G, we set
P
G
= {P : P is a G-gure with cl P int B} ,
P
G
= {(A, C) D : A, C P
G
, A C = }
{(A, C (R
n
\B)

) : A, C P
G
, A C = } .
The concepts of grid, G-gure etc. will be used in the discrete foundation of
the Second Law later on. Now we give the denitions concerning interactions and
superadditive functions dened on subbodies.
May 29, 2002 9:12 WSPC/103-M3AS 00186
Foundations of Second Law 725
Denition 2.5. Let A N
loc
. A function F : A R is superadditive if
F((A
1
A
2
)

) F(A
1
) + F(A
2
)
whenever A
1
, A
2
, (A
1
A
2
)

A and A
1
A
2
= . If the above relation holds with
the equal sign, F is said to be additive.
Let D D
loc
. We say that a function F : D R is biadditive if the functions
F( , C) : {A

M: (A

, C) D} R,
F(A, ) : {C

N : (A, C

) D} R,
are additive for every (A, C) D.
Denition 2.6. Let D D
loc
be a set containing almost all of D
loc
and consider
a function I : D R. We say that I is a Cauchy interaction, if the following
properties hold:
(a) I is biadditive;
(b) there exist h L
1
loc,+
(int B), M(int Bint B) and
e
M(int B) such that
the inequality
|I(A, C)|
_

_
_

C
hdH
n1
+ (AC) if C B ,
_

C
hdH
n1
+ (A(C B)) +
e
(A) otherwise ,
holds almost everywhere in D
loc
.
The last denition, motivated by a choice already made in Gurtin, Williams,
Ziemer,
4
covers the possibility that the exterior of the body, considered as a whole,
has a bulk interaction with its parts. If it is not the case, it is enough to set
e
= 0.
Denition 2.7. A Cauchy interaction I is said to be balanced, if there exists a
measure M(int B) such that

C =|I(A, C)| (A)

on almost all of D
loc
.
This concept is a generalization of the idea for a quantity to obey a balance law
(see Refs. 4 and 5).
If we take D = P
G
for some grid G and require I to satisfy Denitions 2.6 and
2.7 only on P
G
for suitable h, , , we shall call the function I a G-interaction. This
means that we have an interaction dened only on special subbodies, the G-gures.
In Marzocchi, Musesti
5
(Corollary 8.1) it is proved that if G is a full grid, then a
G-interaction can be extended in an essentially unique way to a balanced Cauchy
interaction.
May 29, 2002 9:12 WSPC/103-M3AS 00186
726 A. Marzocchi & A. Musesti
Remark 2.1. When I is balanced, it is possible to enlarge the domain of I to the
set
D
loc
h
{(A, C) : A M
loc
h
, C is normalized, (R
n
\C)

M
loc
h
, A C = }
for suitable h and (cf. Theorem 7.4 in Ref. 5). In particular the interaction between
a subbody A and its exterior, I(A, (R
n
\A)

), is dened for almost every A M

loc
.
3. Balance of Entropy
During a thermodynamic process, amounts of entropy (as well as of other quantities,
such as heat) transfer from a subbody to another, so that we can model this transfer
by means of a Cauchy interaction. In view of a balance law, we call entropy transfer
a balanced Cauchy interaction M(A, C) which can be interpreted as the amount
of entropy that the subbody C transfers to A. It is therefore a biadditive function.
However, due to the peculiar nature of entropy, an extra amount can be produced
in the process. Hence, we introduce an entropy production term, i.e. a real-valued
function dened on almost all of M
loc
.
In the standard frame, a balance law of the form

S(A) = M(A, A
e
) + (A)
is introduced, where A
e
denotes the exterior of the subbody A and

S is the rate
of change of internal entropy, and a local form for regular elds is deduced. For
instance, if all expressions involved make sense, one can write

S(A) =
_
A
s dL
n
, M(A, A
e
) =
_

A
j n
A
dH
n1
, (A) =
_
A
g dL
n
and get the usual dierential equality
s = div j + g .
The usual interpretation of the principle of increase of entropy implies 0;
Moreover, using the biadditivity of M, it is not hard to see that

S(A C)

S(A)

S(C) = (M(A, C) + M(C, A)) + (A C) (A) (C) .
If the terms M(A, C) and M(C, A) are surface integrals, then their sum vanishes
and

S(A C)

S(A)

S(C) = (A C) ((A) + (C)) .
In Truesdell
7
(Appendix G4), Gurtin and Williams called the left side binding
entropy; since in some cases it may be strictly positive (see again Truesdell
7
and
Sec. 6 of Ref. 4), we will suppose in the sequel that is superadditive. Moreover, we
do not want to be necessarily absolutely continuous with respect to the volume
measure.
Denition 3.1. An entropy production is a real-valued function dened on
almost all of M
loc
such that
May 29, 2002 9:12 WSPC/103-M3AS 00186
Foundations of Second Law 727
(a) 0 on almost all of M
loc
;
(b) there exists M(int B) with
(A) (A)
for almost every A M
loc
;
(c) is superadditive.
Of course, this denition belongs to every situation in which a balance law
holds with an extra production term. Since is positive, from (c) it follows that it
is monotone, i.e.
A
1
A
2
(A
1
) (A
2
) .
The following theorem states the main property of entropy productions.
Theorem 3.1. If is an entropy production, then there exists a Borel function
g : int B [0, +) such that :
(i) for almost every A M
loc
,
_
A
g d (A) ;
(ii) g is maximal, in the sense that if G is a full grid such that is dened on P
G
and f : int B [0, +) is a Borel function satisfying
P P
G
:
_
P
f d (P) ,
then f(x) g(x) for -almost all x int B.
Moreover, such a g is uniquely determined -a.e.
This is a plain consequence of Theorem 6.2, which we postpone to the last
section. It is quite a delicate theorem in Measure Theory; therefore, it is somewhat
technical.
In particular, the maximal density g produces a maximal measure
(A) =
_
A
g d
with (A) (A) on almost all of M
loc
.
Denition 3.2. The maximal measure is called the optimal entropy production.
Now we recall the integral representation of the entropy transfer M: by
Theorem 7.4 of Ref. 5 we have
M(A, C) =
_

_
_
AC
k d+
_

C
j n
A
dH
n1
if C B,
_
A(Cint B)
k d+
_
A
k
e
d
e
+c
_

C
j n
A
dH
n1
otherwise ,
(3.1)
May 29, 2002 9:12 WSPC/103-M3AS 00186
728 A. Marzocchi & A. Musesti
on almost all of D
loc
, where M(int B int B),
e
M(int B), the Borel
functions k : int B int B R and k
e
: int B R are such that |k| = 1 -a.e. and
|k
e
| = 1
e
-a.e. respectively, j L
1
loc
(int B; R
n
) has divergence measure.
In (3.1) k represents the density of the bulk entropy transfer, while j is the
surface density of the entropy ux. Finally, k
e
takes into account possible bulk
entropy exchanges with the exterior of the body.
For almost every A M
loc
we dene
M(A, A) :=
_
AA
k d,
M(A, R
n
) := M(A, (R
n
\A)

) + M(A, A)
=
_
Aint B
k d +
_
A
k
e
d
e
+
_

A
j n

A
dH
n1
.
Notice that, in order to speak of (R
n
\A)

as a subbody, we rely on Remark 2.1.

Now, combine the entropy production and the entropy transfer: clearly the mea-
sure + M( , R
n
) is additive and bounded by a Radon measure on almost all of
M
loc
, so it can be represented by a measure M(int B) and a Borel function u
with |u(x)| = 1 for -a.e. x int B, i.e.
(A) + M(A, R
n
) =
_
A
u d
on almost all of M
loc
. Since is positive, it is obvious that
_
A
u d M(A, R
n
).
Hence,
_
A
u d
_
A
div j +
_
Aint B
k d +
_
A
k
e
d
e
(3.2)
on almost all of M
loc
. This is the measure-theoretic version of the entropy inequality.
Remark 3.1. Clearly we have
(A) + M(A, (R
n
\A)

) (A) + M(A, (R
n
\A)

) .
The quantity (A) +M(A, (R
n
\A)

) is usually interpreted as the rate of change of

the internal entropy of the subbody A. We can hence look at (A)+M(A, (R
n
\A)

)
as the optimal rate of change of internal entropy of A, accordingly to Denition 3.2.
4. Existence of Temperatures
In this section we suppose that, beyond the entropy transfer, there exists a balanced
Cauchy interaction H which we refer to as the heat transfer. By Theorem 7.4 of
Ref. 5, there exist b, b
e
, ,
e
, q (with roles analogous to the corresponding list for
entropy) such that
May 29, 2002 9:12 WSPC/103-M3AS 00186
Foundations of Second Law 729
H(A, C) =
_

_
_
AC
b d +
_

C
q n
A
dH
n1
if C B ,
_
A(Cint B)
b d +
_
A
b
e
d
e
+
_

C
q n
A
dH
n1
otherwise ,
on almost all of D
loc
.
Physical evidence shows that when there is no heat transfer, no entropy can be
transferred. Therefore, the following assumption is a crucial part of the Second Law
of Thermodynamics. Without it, one could not have a link between temperature
and entropy (or even the concept of temperature).
Assumption 4.1. Let h L
1
loc,+
(int B), M(int B) be such that the represen-
tation formulas for M and H hold on D
loc
h
. Given (A, C) D
loc
h
, if H(

A,

C) = 0
for every (

A,

C) D
loc
h
such that

A A,

C C, then M(A, C) = 0.
The goal of this section is to prove that the measures associated with M in the
representation (3.1) are absolutely continuous with respect to those associated with
H; in this way, we will nd three functions playing the r ole of temperature in a
generalization of the classical ClausiusDuhem inequality. This is not trivial, since
we need to associate in some way a measure of AC to the surface integrals.
From now on, we will use the notation E

= E

E; this is the measure-

theoretic closure of E. First of all, we state a slight variant of Theorem 7.4 of Ref. 5
for balanced Cauchy interactions in terms of the measure-theoretic closed sets E

.
Theorem 4.1. Let H be an arbitrary balanced Cauchy interaction and let q,
b, b
e
, ,
e
be as in Theorem 7.4 of Ref. 5. Then the formula
H(A, C)=
_

_
_
A

b d+
_

q n
A
dH
n1
if C B ,
_
A

(Cint B)

b d+
_
A

b
e
d
e
+
_

q n
A
dH
n1
otherwise ,
holds on almost all of D
loc
.
Proof. Clearly, for every A B(int B) we have that A

is the disjoint union of A

and

A. Since we can choose h L

1
loc,+
(int B) and M(int B) in a way such
that (

Aint B) = (int B

C) =
e
(

C) = 0 for every (A, C) D

loc
h
,
it is apparent that the volume integrals involving

A and

C vanish. Finally,
notice that

A =

.
Now let (A, C) D
loc
h
with C B and let f : int B int B int B be such
that f(x) = (x, x); we dene two real-valued measures m
1
, m
2
on B(AC) setting
m
1
(E) =

2
_
E
q(f
1
(x, y)) n
A
(f
1
(x, y))dH
n1
(x, y) ,
May 29, 2002 9:12 WSPC/103-M3AS 00186
730 A. Marzocchi & A. Musesti
m
2
(E) =

2
_
E
j(f
1
(x, y)) n
A
(f
1
(x, y))dH
n1
(x, y) .
The previous denitions allow to express the surface integrals associated to H and
M, denoted by H
c
and M
c
respectively, in terms of the measures m
i
: by a change
of variables it results that for every (

A,

C) D
loc
h
with

A A and

C C, we have
H
c
(

A,

C) = m
1
(

A

) , M
c
(

A,

C) = m
2
(

A

) .
Details are omitted for the sake of brevity.
Let p
1
, p
2
be the densities of m
1
, m
2
, respectively. Combining this fact with
Theorem 4.1, it follows that
H(

A,

C) =
_

b d +
_

p
1
d|m
1
| ,
M(

A,

C) =
_

k d +
_

p
2
d|m
2
| ,
for every (

A,

C) D
loc
h
with

A A,

C C. Moreover, the supports of |m
1
|
and |m
2
| lie in (

C) (

C), while the supports of and lie in

AC\((

C) (

C)).
By Assumption 4.1, we have that + |m
2
| + |m
1
|; since the supports are
disjoint, it follows that
, |m
2
| |m
1
|
and there exist two Borel functions
A,C
, d
A,C
: AC R such that
M(

A,

C) =
_

A,C
b d +
_

d
A,C
p
1
d|m
1
|
for every (

A,

C) D
loc
h
with

A A,

C C. Our choice of having xed A and C is
not restrictive: it is readily seen that for (A
1
, C
1
), (A
2
, C
2
) D
loc
h
one has

A
1
A
2
,C
1
C
2
=
A
1
,C
1
|
(A
1
A
2
)(C
1
C
2
)
=
A
2
,C
2
|
(A
1
A
2
)(C
1
C
2
)
-a.e.
d
A
1
A
2
,C
1
C
2
= d
A
1
,C
1
|
(A
1
A
2
)(C
1
C
2
)
= d
A
2
,C
2
|
(A
1
A
2
)(C
1
C
2
)
|m
1
|-a.e.
Since
A,C
and d
A,C
agree on intersections, they can be extended in a unique (almost
everywhere) way to , d : int B int B R.
We are now in position to introduce the notion of the reciprocal of the contact
temperature, setting
c
= d f; we have that
M
c
(A, C) =
_

c
q n
A
dH
n1
on almost all of D
loc
.
a
a
M
c
can be called the contact part of M (cf. Denition 3.10 of Ref. 5).
May 29, 2002 9:12 WSPC/103-M3AS 00186
Foundations of Second Law 731
On the other hand, if (A, (R
n
\B)

) D
loc
h
we can repeat in an easier way the
same procedure, getting a Borel function
e
: int B R such that
M(A, (R
n
\B)

) =
_
A

e
b
e
d
e
.
Finally we have
M(A, C) =
_

c
q n
A
dH
n1
+
_
A

b d +
_
A

e
b
e
d
e
(4.1)
on almost all of D
loc
. The meaning of the integrands in (4.1) is that of reciprocal
of absolute temperatures related to interactions: 1/
c
is the usual one, since it is
associated to the contact part, while the others, associated to bulk and external bulk
heat transfer, appear in some applications, such as the theory of radiative transfer.
Notice that we introduce
c
and not its reciprocal: indeed, when
c
vanishes the
usual temperature does not exist.
We can now restate the entropy inequality of the previous section, involving the
densities of the heat interaction:
_
A
u d
_

c
(x) q(x) n
A
(x) dH
n1
(x)
+
_
Aint B
(x, y) b(x, y) d(x, y) +
_
A

e
(x) b
e
(x) d
e
(x) (4.2)
on almost all of M
loc
, which is a generalization of the integral form of the Clausius
Duhem inequality.
5. A Weaker Form of the Second Law
The aim of this section is to introduce the Second Law of Thermodynamics in a
sort of discrete context and deduce all results proved up to here.
In Sec. 8 of Marzocchi, Musesti,
5
a notion of balanced Cauchy interaction dened
only on almost all n-intervals is introduced, and an extension theorem is proved.
In that context, the choice between multi-intervals or n-intervals fell on the second
class, since the biadditivity bears immediately an extension to the rst. Here, on
the contrary, the entropy production is only superadditive and there is no natural
way to extend it. Because of this fact, the class of almost all multi-intervals seems
to be the simplest natural class of subbodies which gives enough information to
extend the notions to almost all normalized sets of nite perimeter, at least in the
framework of Thermodynamics.
Let G be a full grid, and consider two G-interactions M, H : P
G
R. These
functions represent the entropy transfer and the heat transfer, respectively. As
remarked in Sec. 2, in Marzocchi, Musesti
5
it is proved that M and H can be
extended in an essentially unique way to balanced Cauchy interactions.
Then consider a function : P
G
R, which takes the place of the entropy pro-
duction of Sec. 3. We restate the Second Law of Thermodynamics in the following
way.
May 29, 2002 9:12 WSPC/103-M3AS 00186
732 A. Marzocchi & A. Musesti
Axiom 5.1. (Second Law of Thermodynamics) For a full grid G there exist two
G-interactions M, H and a function : P
G
R such that :
(i) 0;
(ii) there exists M(int B) with
(P) (P)
for every P P
G
;
(iii) is superadditive;
(iv) given (A, C) P
G
, we have M(A, C) = 0 whenever H(

A,

C) = 0 for every
(

A,

C) P
G
such that

A A,

C C.
Clearly, this is a weaker version of the Second Law of Thermodynamics, since
each requirement involves only G-gures. The following theorem shows that this
axiom is sucient to prove the validity of the ClausiusDuhem inequality in its
general form. This is our main result.
Theorem 5.1. If Axiom 5.1 holds, then (4.2) holds on almost all of M
loc
.
Proof. Theorem 6.2 below allows to nd a unique optimal entropy production
such that (3.2) holds on almost all of M
loc
.
Moreover, by means of (iv) of Axiom 5.1 we can again prove that and
|m
1
| |m
2
| in Sec. 4.
Once more we stress the fact that testing the validity of the Second Law only
on almost all multi-intervals, we obtain the same results as testing it on almost all
of M
loc
and D
loc
.
6. Superadditive Functions
This section is essentially devoted to proving Theorem 3.1. In order to do so, we
need some preliminary material.
Let C(x, r) denote the closed cube with center x and edge 2r. Given a full grid
G, we denote by C
G
the set of all closed cubes C(x, r) such that x int B, r > 0
and int C(x, r) I
G
.
Remark 6.1. (Geometric property) For any r > 0 there exists a maximal number
of closed cubes with edges greater than r that do not contain the center of each
other and such that all of them intersect a given cube having edge r.
Denition 6.1. A subfamily C of C
G
is ne with respect to a set E int B, if
every x E is the center of an element C C and
inf{r : C(x, r) C} = 0
for every x E.
May 29, 2002 9:12 WSPC/103-M3AS 00186
Foundations of Second Law 733
The following, usually known as the Besicovitch Theorem, is a standard tool
in Measure Theory and strongly relies on Remark 6.1. For a proof the reader is
referred to Federer
2
2.8.15.
Theorem 6.1. (Besicovitch Theorem) Let M(int B), E B(int B) with
(E) < + and C be a subfamily of C
G
which is ne with respect to E.
Then for every open set A int B with E A, there exists a countable disjoint
subfamily G of C such that
_
CG
C A,
_
E
_
_
CG
C
_
= 0 .
The previous result also applies to the case of open normalized n-intervals and
normalized unions, by choosing a suitable full grid G, as stated below.
Corollary 6.1. Let M(int B) and E B(int B) with (E) < +. Let G be
a full grid such that I
G
M
loc
0
,
b
and C be a subfamily of C
G
which is ne with
respect to E.
Then for every open normalized set A int B with E A, there exists a
countable disjoint subfamily (I
k
) I
G
such that
_
_
kN
I
k
_

A,
_
E
_
_
_
kN
I
k
_

_
= 0 .
Proof. Let G = {C
k
: k N} be a countable subfamily of C as in Theorem 6.1 and
set I
k
= int C
k
. Then the family (I
k
) is contained in I
G
and, keeping into account
that A is normalized, one has
_
_
kN
I
k
_

A.
Moreover, M
loc
0
implies that

_
E
_
_
_
kN
I
k
_

_
=
_
E
_
_
kN
I
k
_
= 0
and the proof is complete.
Let now A M
loc
and : A [0, +) be a superadditive function. We
suppose that there exist a full grid G and a measure M(int B) such that
P
G
A, P
G
M
loc
0
and on P
G
. Moreover, we suppose that A is closed
by nite intersection. Note that if A = M
loc
h
for suitable h and , as in Sec. 3, or
A = P
G
as in Sec. 5, then in any case it satises the previous requirement.
b
Clearly, 0 means h = 0, M(int B).
May 29, 2002 9:12 WSPC/103-M3AS 00186
734 A. Marzocchi & A. Musesti
Now we set
g(x) = inf
>0
sup
_
{0}
_
(I)
(I)
: I C
G
, x I, diamI < , (I) > 0
__
. (6.1)
Lemma 6.1. The function g : int B [0, +) is Borel and bounded.
Proof. As , it is clear that 0 g(x) 1, hence it is bounded. Let us dene,
for > 0,

(x) = sup
_
{0}
_
(I)
(I)
: I C
G
, x I, diamI < , (I) > 0
__
.
Since we have g(x) = inf
>0

(x), it is sucient to prove that each

is a Borel
function (see Federer
2
2.2.15). Let c R and x int B; if

(x) > c, then there

exists I C
G
such that x I, diamI < , (I) > 0 and (I) > c(I). This means
that

(y) > 0 for every y I, which is an open set, and so

(]c, +[) is an open

set; in particular, it is Borel.
For A R
n
and > 0, we denote with N

(A) the set

{x R
n
: d(x, A) < } .
Lemma 6.2. Let c > 0 and suppose that K is a compact subset of int B such that
K {x : g(x) > c}. Then for every G-gure P P
G
such that K P, we have
c(K) (P) .
Proof. Let
C = C
G

_
cl I : I I
G
, I K = , I P, (I) > 0,
(I)
(I)
> c
_
;
then C is ne with respect to K. Applying Corollary 6.1, we nd a countable disjoint
subfamily {I
k
: k N} I
G
such that
_
_
kN
I
k
_

P ,
_
K
_
_
_
kN
I
k
_

_
= 0.
For every l N we have
(P)
__
l
_
k=1
I
k
_

k=1
(I
k
) c
l

k=1
(I
k
) = c
_
l
_
k=1
I
k
_
.
As l , it follows that
(P) c
_
_
kN
I
k
_
= c
__
_
kN
I
k
_

_
c(K) ,
where we used the fact that (

I
k
) = 0 for every k N.
May 29, 2002 9:12 WSPC/103-M3AS 00186
Foundations of Second Law 735
Now we are ready to prove our main theorem.
Theorem 6.2. Let : A [0, +) be a superadditive function, G a full grid with
P
G
A and M(int B) with P
G
M
loc
0
and on P
G
. Then there exists a
Borel function g : int B [0, 1] such that :
(i)
_
E
g d (E)
for every E A;
(ii) if f : int B [0, +) is a Borel function satisfying
P P
G
:
_
P
f d (P) ,
then f(x) g(x) for -almost all x int B.
Moreover, such a g is uniquely determined -a.e.
Proof. Let g be dened as in (6.1). Let 0 < t < 1 and > 0. For h N consider
the sets
E
h
= {x E : t
h+1
< g(x) t
h
} ;
since is a regular measure, there exists a sequence (K
h
) of compact sets in int B
such that K
h
E
h
and (E
h
\K
h
) 2
h1
for every h N; in particular, (K
h
)
is a disjoint sequence. Then we have
_
E
g d =

hN
_
E
h
g d

hN
t
h
(E
h
)
l

h=0
t
h
(E
h
) +
l

h=0
t
h
(K
h
) + 2
for a suitable l N.
By Lemma 6.2, for every h N there exists a set P
h
P
G
such that K
h
P
h
and t
h+1
(K
h
) (P
h
). Setting d(A, B) = inf{|x y| : x A, y B}, let
=
1
2
min{d(K
h
, K
j
) : 0 h < j l} ;
since the sets K
h
are compact and pairwise disjoint, we have > 0. Without loss of
generality, we may assume that P
h
N

(K
h
), so that (P
h
) is a disjoint sequence.
Moreover, by substituting P
h
with P
h
E, which again belongs to A, we may also
assume that P
h
E. Hence we can continue the inequalities:
l

h=0
t
h
(K
h
) + 2 t
1
l

h=0
(P
h
) + 2 t
1
(E) + 2 .
Letting 0
+
and t 1

, (i) is proved.
Now let f be as in the statement; let x int B, > 0 and I I
G
be such that
x I, (I) > 0 and diamI < . Since
1
(I)
_
I
f d
(I)
(I)
,
taking the supremum limit as 0
+
the proof is complete.
May 29, 2002 9:12 WSPC/103-M3AS 00186
736 A. Marzocchi & A. Musesti
Acknowledgment
The authors wish to thank Marco Degiovanni for very helpful discussions and sug-
gestions, and the referees for useful remarks.
The research of the authors was partially supported by M.U.R.S.T. COFIN2000
project Modelli matematici per la scienza dei materiali and Gruppo Nazionale
per la Fisica Matematica.
References
1. M. Degiovanni, A. Marzocchi and A. Musesti, Cauchy uxes associated with tensor
elds having divergence measure, Arch. Rational Mech. Anal. 147 (1999) 197223.
2. H. Federer, Geometric Measure Theory, Die Grundlehren der Mathematischen Wissen-
schaften, Vol. 153 (Springer-Verlag, 1969).
3. M. E. Gurtin and W. O. Williams, An axiomatic foundation for continuum thermody-
namics, Arch. Rational Mech. Anal. 26 (1967) 83117.
4. M. E. Gurtin, W. O. Williams and W. P. Ziemer, Geometric measure theory and the
axioms of continuum thermodynamics, Arch. Rational Mech. Anal. 92 (1986) 122.
5. A. Marzocchi and A. Musesti, Decomposition and integral representation of Cauchy
interactions associated with measures, Cont. Mech. Thermodyn. 13 (2001) 149169.
6. M.

Silhav y, Cauchys stress theorem and tensor elds with divergences in L
p
, Arch.
Rational Mech. Anal. 116 (1991), 223255.
7. C. A. Truesdell, Rational Thermodynamics (Springer-Verlag, 1984), 2nd edition.
8. W. P. Ziemer, Weakly Dierentiable Functions, Graduate Texts in Mathematics,
Vol. 120 (Springer-Verlag, 1989).