Beruflich Dokumente
Kultur Dokumente
002
is obtained. Here
RN is an open, bounded set of class C 2 , T is the trace operator on @
and
HN 1 is the N 1 dimensional Hausdor measure. The main hypotheses on the functions h and are
that h(x; u; ) is quasiconvex and has linear growth, and that (x; ) is Lipschitz. The understanding
of nucleation phenomena for materials undergoing phase transitions leads to the study of constrained
minimization problems of the type
inf H(u;
) +
(x; u(x)) dx : u 2 BV (
; K ) ;
Key words. functions of bounded variation, nucleation, relaxation, bulk and contact energies,
generalized co-area formula
AMS subject classications. 49J45, 49Q20, 49N60, 73T05, 73V30
1. Introduction.
This paper is divided into two parts. In the rst part we obtain an integral
representation formula in BV (
; Rp ) for the relaxation H(u;
) with respect to the
L1 topology of functionals of the general form
(1.1)
Z
Z
H (u;
) := h(x; u(x); ru(x)) dx + (x; T u(x)) dHN 1 (x); u 2 W 1;1 (
; Rp );
@
where
and HN 1 is the N 1 dimensional Hausdor measure. The main hypotheses on the
functions h and are that h(x; u; ) is quasiconvex and has linear growth, and that
(x; ) is Lipschitz.
190
191
H(u;
) = h(x; u(x); ru(x)) dx + h1 (x; u(x); dC (u))
(1.2)
ZS(u)\
+
Kh (x; u (x); u (x); u (x)) dHN 1 (x)
where ru is the density of the absolutely continuous part of the distributional derivative Du with respect to the N {dimensional Lebesgue measure LN , (u+ u ) is the
jump across the interface S (u), and C (u) is the Cantor part of Du. For the canonical
model where h(x; u; ru) := jruj, > 0, the relaxed energy H(u;
) reduces to
(1.3)
H(u;
) = jDuj +
(x; T u) dHN 1 ;
u 2 BV (
; Rp ):
In the scalar case where p = 1 the lower semicontinuity of the functional (1.3) was
proved by Massari and Pepe [MP] when (x; u) := ^ juj, with j^ j , and by Modica
[Mo2] under the assumption that
(1.4)
j(x; u) (x; u1 )j ju u1 j
for all x 2 @
and all u, u1 2 R.
One of the motivations for the introduction of a relaxed energy is that nonconvex variational problems may not have a minimizer in the space of smooth functions,
therefore in order to apply the direct method of Calculus of Variations one has to extend the original functional. Although Sobolev spaces are considered to be the natural
extension to the space of smooth functions, in recent years the theory of phase transitions, and the need to determine eective energies for materials exhibiting instabilities
such as fractures and defects, have led us to further extend the domain of functionals
of the form (1.1) in order to include functions u which present discontinuities along
surfaces. Motivated somewhat by Lebesgue's denition of surface area, Serrin in [Se1,
Se2] proposed the following notion for the relaxed energy of H (u;
) (in the case where
0)
192
walls, where (x; u) is the contact energy per unit area when the density is u (see [C,
G]).
We present here two relaxation results. In Theorem 3.5 we show that, without
any a priori coercivity on the function h, the functional on the right hand side of (1.2)
actually gives the integral representation for the following relaxed energy
lim inf H (u ;
) : u 2 W 1;1(
; Rp ); u !> u in L1(
; Rp );
Hb (u;
) = finf
n
n
n
u g n!1
n
Z
H (u;
) = h(x; u(x); ru(x)) + '(x) ruT (x)ru (x; u(x)) dx
Z
Z
+
Z
jru(x)j + '(x) ruT (x)ru (x; u(x)) dx
H (u;
) =
Z
Z
+
and it is clear that the functional H is not bounded from below in general, unless one
assumes a condition of the type
jru (x; u)j for a. e. x 2
and for all u 2 Rp ;
193
which is essentially the condition found by Massari and Pepe [MP] and by Modica
[Mo2]. We note that once (1.1) is written in the form (1.6), well known integral
representation results for relaxation of bulk energy functionals apply (see [FM2, FR]).
This is illustrated rst on Section 2 in the scalar case p = 1 and for a class of energies
including the typical ones considered by Visintin (see [V1, V2]),
h(x; u; ) := j j; (x; u) := ^ u:
More general models, relevant to the study of anisotropic materials and nonlinear
contact forces, are studied in Section 3. In particular, degenerate bounds for the bulk
energy density h, as introduced in [FM2] in the vectorial case, may prove to be useful
in the analysis of phase transition problems.
In the second part of the paper, Sections 4 and 5, we are concerned with constrained minimization problems of the type
inf H(u;
) +
(x; u(x)) dx : u 2 BV (
; K ) ;
RN , is given by
(1.7)
Z
Z
Z
E" (u) := "2 jruj2 dx + W1 (u) dx + " W2 (T u) dHN 1 ; u 2 W 1;1 (
; R);
where the coarse{grain energy W1 (u) is a double well potential vanishing only at and
and corresponding to the stable two{phase conguration of the
uid, the gradient
term "2 jruj2 models the interfacial energy across a smooth transition layer, with "
a small parameter, and W2 represents the contact energy between the
uid and the
container walls. The stable congurations of the
uid correspond to solutions of the
problem (see [C])
u dx = m :
Conrming a conjecture of Gurtin [G], Modica in [Mo2] was able to show that if a
sequence of minimizers fu" g converges in L1 to a function u0 , then u0 solves the
liquid{drop problem
inf fH(u;
) : u 2 BV (
; f; g)g ;
where H(u;
) is the relaxed energy of
H (u;
) =
jruj dx + ^
T u dHN 1 ;
u 2 W 1;1 (
; R):
194
"!0
and W2 is a double well potential which vanishes only at 1 and 1 , then the limit
problem is given by a dierent model for capillarity with line tension. It is worth
noting that in this case the eective energy takes the form
H(u;
) =
jD(G(u))j
Z
p
for u 2 BV (
; f; g) and H(u;
) = 1 otherwise. Here G is a primitive of 2 W1 . It
+ inf
can be seen immediately that in this capillarity model the contact energy is nonlocal
and strongly nonlinear, and again this leads us to considering functions other than
(x; u) = ^ u (see [V1, V2] and Section 3).
In the last section of the paper we prove some minimization results which are
related to solid nucleation. For a complete description of this phenomenon we refer to
the recent monograph of Visintin [V1] and to the bibliography contained therein. By
solid nucleation we mean the formation of a new solid phase, that is of a connected
component of solid in a liquid. If the new solid phase is formed in the interior of the
liquid, the nucleation is called homogeneous, while if it is also in contact with other
substances, such as the container, impurities dispersed in the liquid or nucleants, then
we name it heterogeneous nucleation (cf. [V1, Ch. VII.2]). By thinking of these
impurities or particles as holes in the domain
, we can represent the contact energy
by an integral term over the boundary of
. Furthermore, since the new solid phase
is formed through crystallization,
and crystals are
R anisotropic, the classical isotropic
R
interfacial energy
jDuj is now replaced by
h(x; Du). In the applications one
sees often h(x; Du) = jA(x)Duj, where A(x) is a nonnegative denite N N tensor
(cf. [V1, p. 157]).
The main results of this part are Theorems 5.1 and 5.4, where we show that
minimizers u of the total energy are given by pure phases, that is, there exists
u
195
H (u;
) :=
h(x; ru(x)) dx +
as a bulk energy and then apply known relaxation results from W 1;1 to BV (see, e.g.,
[DM, FM2]).
We start with some notation. For any 2 S N 1 := fx 2 RN : jxj = 1g let
f1 ; ; N 1 ; g be an orthonormal basis of RN varying continuously with , and
Q := fx 2 RN : jx i j < 1=2; jx j < 1=2; i = 1; ; N 1g be a unit cube centered
at the origin with two of its faces orthogonal to the direction .
If g is a positively homogeneous function of degree one and if is a Rm {valued
measure then we dene
g(d) :=
g((x)) dj(x)j;
@' (x) dx =
ui (x) @x
j
'(x) dij
and
"!0+ "N
196
1 Z
lim
r!0 rN
and
ju(y) u (x)jN=(N 1) dy = 0:
fy2B(x0 ;r):(y x)u (x)<0g
Note that in general (ui )+ 6= (u+ )i and (ui ) =
6 (u )i . Moreover, the Sobolev inequality
Z
(N
ju(x)jN=(N 1) dx
1)=N
C (N )jjujjBV
holds in BV (
; Rp ) when N > 1. Finally, Du may be represented as
Du = ru LN + (u+ u )
u HN 1 b S (u) + C (u);
where ru is the density of the absolutely continuous part of Du with respect to the
N {dimensional Lebesgue measure LN . These three measures are mutually singular.
Lemma 2.1. There exists ' 2 C01 (RN ; RN ) with j'(x)j < 1 in
such that for
any v 2 BV (
; Rp )
v(x)div '(x) dx +
'(x) d(D(v(x))):
Proof. Since @
is compact and of class C 2 , we can nd a nite open covering
fUj gj of @
, where Uj are balls centered at points of @
, j = 1; ; P , and for each
Uj there is a C 2 dieomorphism j : Uj ! j (Uj ) such that j (Uj ) B (0; Rj ) RN
for some Rj > 0,
(2.1)
Let be a partition of the unity for [Pj=1 Uj subordinate to fUj gj . For any 2
there exists j 2 f1; ; P g such that 2 C01 (Uj ), and we dene
rT (x) e
(2.2)
' (x) := jrTj (x) eN j 1 + (jR(x))N (x);
j
N
j
then ' (x) 2 C01 (Uj ; RN ) and j' (x)j < 1 for x 2
\ Uj . If we set ' to be zero
outside Uj we obtain that ' (x) 2 C01 (RN ; RN ), and thus we can apply the Trace
Theorem (cf. [EG,Th. 5.3.1]) to the BV function v, and we have
' (x)n(x;
) T v(x) dHN 1 (x)
=
197
Hence
XZ
2 @
0
1
1
Z 0X
X
= v(x) div @ ' (x)A dx + @ ' (x)A d(D(v(x))):
2
2
P
The proof of Lemma 2.1 is complete if we show that '(x) := 2 ' (x) satises
j'(x)j < 1 in
. Fix x 2
. If x 2= [Pj=1 Uj then '(x) = 0. If x 2 [Pj=1 Uj then
P
(x) = 1, and so there exists at least one 2 such that (x) > 0. Let
Z
Using this lemma, we are now in position to obtain an integral representation for
the relaxed energy
n
o
1;1 (
; Rp ); un ! u in L1 (
; Rp ) :
lim
inf
H
(
u
;
)
:
u
2
W
H(u;
) := finf
n
n
u g n!1
n
Theorem 2.2.
H(u;
) = h(x; ru(x)) dx + h1 (x; dC (u))
ZS(u)\
H (u;
) = [h(x; ru(x)) + ^ '(x) ru(x)] dx +
=
f (x; ru(x)) dx +
^ u(x)div'(x) dx
^ u(x)div'(x) dx
198
there is an upper bound j'(x)j < < 1 and thus f is coercive. Therefore, and in view
of the strong continuity in L1 of the mapping
u 7!
^ u(x)div'(x) dx;
H(u;
) =
+
=
+
ZS(u)\
f (x; ru(x)) dx +
f 1 (x; dC (u))
h(x; ru(x)) dx +
S (u)\
h1 (x; dC (u))
^ u(x)div'(x) dx
Z
@
where we have used once again Lemma 2.1 in the last equality.
H (u;
) :=
199
200
Kh (x; a; b; ) := inf
Z
Q
For a detailed study of the properties of the function Kh (x; a; b; ) we refer to [FR].
For u 2 BV (
; Rp ) we dene the functional
L(u;
) :=
Z
+
Kh(x; u (x); u (x); u (x)) dHN 1 (x) + (x; T u(x)) dHN 1 (x):
+
@
S (u)\
H(u;
) = L(u;
):
H(u;
) =
h(x; ru(x)) dx +
+
+
ZS(u)\
h1 (x; dC (u))
201
L(u;
) = jDuj + ^
Consider the sequence
T u dH1 ; u 2 BV (
; R):
0
if x1 + x2 1=n
n if x1 + x2 < 1=n:
p
Then un (x) ! 0 in L1(
; R) but L(un ;
) = 2 + 2^ < L(0;
) = 0, and this shows
that H(u;
) 6= L(u;
) since H(u;
) is lower semicontinuous in L1 .
It is worth noting that in the special case where (x; u) = ^ ju (x)j in (1.3),
with j^ j and 2 L1(@
; R), one can still prove lower semicontinuity of L for
Lipschitz subdomains of RN . The rst result in this direction is due to Massari
and Pepe [MP] who treated the case where 0. Modica [Mo2] then extended it to
include 2 L1 (@
; R). The idea in [MP, Mo2] is to nd a function ^ 2 BV (RN n
; R)
whose trace Ris and then use an extension theorem (see [EG, Th. 5.4.1]) to rewrite
the integral @
jT u j dHN 1 as
un (x1 ; x2 ) :=
where
jT u T ^j dHN 1 =
u^(x) :=
RN
u(x)
^(x)
jDu^j
jDuj
RN n
jD ^j;
if x 2
if x 2 RN n
:
(ii) Without condition (H7 ) Theorem 3.2 may fail. As an example, let
:=
(0; 1) R and take h and as in (3.6). In this case condition (H7 ) is equivalent to
the inequality j^ j . Assume that < ^ and consider the sequence
n3(x 1) n if 1 1=n2 x 1
un (x) :=
0
otherwise:
Then un (x) ! 0 in L1(
; R) but L(un ;
) = ( ^ )n < L(0;
) = 0.
Theorem 3.5. Let (H1 ) (H6 ) hold, with (3:1) and (H7 ) replaced by the weaker
hypothesis
(3.7)
jru (x; u)j j j h(x; u; ) C g(x; u)(1 + j j)
for all (x; u; ) 2
Rp M pN , and some C > 0. Then the relaxation of H (u;
)
inf H (u ;
) : u 2 W 1;1(
; Rp );
Hb (u;
) = inf lim
n
n
n!1
fun g
un ! u in L1(
; Rp ); sup jjun jjW 1;1 < 1
n
in BV (
; Rp ) with respect to the L1 topology has the integral representation
Hb (u;
) = L(u;
):
Remark 3.6. Under the assumptions of Theorem 3.5, the functional L(u;
) provides the correct integral representation for Hb (u;
) but not necessarily for H(u;
).
Indeed, in the scalar case where p = 1 and when 0, Dal Maso has shown in
202
H (u;
) =
Z
h(x; u(x); ru(x)) + '(x) ruT (x)ru (x; u(x)) dx
Z
Z
+
and let
n
o
1
;
1
p
1
p
F (u;
) := finf
lim
inf
F
(
u
;
)
:
u
2
W
(
;
R
)
;
u
!
u
in
L
(
;
R
)
:
n
n
n
un g n!1
Lemma 3.8. If u 2 BV (
; Rp ) then
H(u;
) = F (u;
) + (x; u(x)) div '(x) dx + '(x) rx (x; u(x)) dx:
203
lim
inf H (un ;
) = lim
inf F (un ;
)
n!1
n!1
by (3.4)
By (H6 ), by virtue Rof the Sobolev inequality, and due to the fact that ' is bounded,
the functional u 7!
' rx (x; u(x)) dx is continuous in L1 (
; Rp ) (see [K, Th. 2.1])
and thus
Z
Z
lim
'
r
(
x;
u
)
dx
=
'(x) rx (x; u) dx:
x
n
n!1
204
F (u;
)
f (x; u; ru) dx +
+
f 1 (x; u; dC (u))
S (u)\
The proof of this inequality relies on the blow{up argument introduced in [FM1]
which is a local argument, in the sense that in order to prove the three main pointwise
inequalities (2.10){(2.11) in [FM2] at points x0 2
, one is only interested in what
happens in a ball B (x0 ; "). Since in our case j'(x0 )j < 0 < 1 for some 0 > 0, if we
take " suciently small we can assume that j'(x)j 0 for all x 2 B (x0 ; ") and thus
(3.10) reduces to
g(x; u)j j(1 0 ) f (x; u; ) 2C g(x; u)(1 + j j)
for all (x; u; ) 2 B (x0 ; ") Rp M pN , which is the local version of (H3 ) in [FM2].
In conclusion, we may apply Theorem 2.16 of [FM2] (see Remark 3.10 below) to
obtain that for u 2 BV (
; Rp )
F (u;
) =
Z
Z
h(x; u; ru) + ' ruT ru (x; u) dx + h1 (x; u; dC (u))
Z
Z
+
where
('
ru (x; u)) dC T (u) +
Kf (x; a; b; ) = inf
S (u)\
Kf (x; u ; u+ ; u ) dHN 1 ;
Z
h1 (x; (y); r (y)) + '(x) r T (y)ru (x; (y)) dy :
Q
2 A(a; b; ) :
Given any 2 A(a; b; ) we have
Q
'(x) r
T (y )ru (x; (y )) dy =
205
and so
lim inf F (u ;
) : un 2 W 1;1 (
; Rp ); un ! u in L1(
; Rp ) ;
F" (u;
) := finf
u g n!1 " n
n
where
F" (u;
) :=
We claim that
u 2 W 1;1 (
; Rp ):
lim F (u;
) = Fb (u;
):
"!0 "
Fix u 2 L1 (
; Rp ). For any given > 0 there exists a sequence fung W 1;1 (
; Rp ),
with supn jjun jjW 1;1 = M < 1, such that un ! u in L1 (
; Rp ) and
Fb (u;
) + nlim
!1 f (x; un (x); run (x)) dx:
Fb (u;
) + lim
inf f" (x; un (x); run (x)) dx " M;
n!1
206
Conversely, x u 2 L1 (
; Rp ) and " > 0. Then there exists a sequence fu"ng
1
;
1
W (
; Rp ) such that u"n ! u in L1(
; Rp ) as n ! 1 and
(3.11)
Z
Z
" ; ru" ) + "jru" j] dx lim inf f (x; u" ; ru" ) dx:
[
f
(
x;
u
F" (u;
) + " nlim
n
n
n
n
n
n!1
!1
F" (u;
) =
S (u)\
jdC (u)j
lim F (u;
) = f (x; u; ru) dx + f 1 (x; u; dC (u))
"!0 "
Z
Kf (x; u ; u+ ; u ) dHN 1 ;
+
S (u)\
S (u)\
Z
S (u)\
Kf (x; u ; u+ ; u ) dHN 1 :
This convergence follows easily from Lebesgue Dominated Convergence Theorem and
because we may nd a constant C1 independent of " such that (see the proof of Lemma
2.15 in [FM2])
0 Kf" (x; a; b; ) C1 ja bj:
This concludes the proof of the theorem.
Remark 3.10. In the proof of Theorem 2.16 of [FM2] the inequality
(3.14)
F (u; S (u) \
)
S (u)\
was derived by using a result of [FR] which requires the function f (x; u; ) to be
coercive, that is to satisfy the inequality
(3.15)
f (x; u; ) c1 j j c2
207
F (u; S (u) \
) F" (u; S (u) \
)
S (u)\
If we now let " ! 0 and use (3.13), we conclude that (3.14) holds also for f .
4. Mesoscopic scale.
inf H(u;
) +
(x; u(x)) dx : u 2 BV (
; Rp ); u(x) 2 K for LN a.e.
x2
;
IK (u;
) :=
+1
otherwise:
Lemma 4.1. If (H1 ) (H7 ) hold then the functional I (u;
) is lower semicontinuous in L1(
; Rp ).
Proof. Consider un, u 2 L1 (
; Rp ) such that un ! u in L1 (
; Rp ).
If lim
inf I (un ;
) = 1 there is nothing to prove. Assume that lim
inf I (un ;
) < 1
n!1
n!1
and take a subsequence funk g which converges pointwise to u for LN a.e x 2
, and
such that
lim I (u ;
) = lim
inf I (un ;
) < 1:
n!1
k!1 nk
I (unk ;
) = H(unk ;
) + (x; unk (x)) dx
(x; u(x)) dx
208
and
(4.3)
Let
Then
D(I ) := fu 2 L1 (
; Rp ) : I (u;
) < 1g:
D1 := u 2 BV (
; Rp ) : u(x) 2 K for LN a.e. x 2
D(I )
but in general the two sets do not coincide, unless one assumes that h(x; u; ) is coer-
cive.
I (u;
) inf fI (w;
) : w 2 D1 g :
Proof. Let fung D1 be a minimizing sequence, that is
lim I (un ;
) = inf fI (w;
) : w 2 D1 g < M < 1:
n!1
I (un ;
) = H(un ;
) + (x; un (x)) dx M:
we get a contradiction. Therefore the claim holds, and by (4.5), Theorem 3.2, and
209
(3.5) we have
h(x; un ; run ) dx +
(4.6)
Z
Z
h1 (x; un ; dC (un ))
S (un )\
h1 (x; un ; run ) dx
Z
C 0 jjgjjL1 + C 0 jjgjjmL1 h1 m (x; un ; run) dx + M1:
Using Holder's inequality and (4.6) again, we conclude that there exists M2 2 (0; 1)
such that for all n
(4.7)
h1 (x; un ; run ) dx +
+
h1 (x; un ; dC (un ))
S (un )\
(4.9)
jDvn j =
+
Z
Z
vn (x) 2 BV (
; Rp )
on
nS (un)
on S (un ):
S (vn )\
j(vn+
vn )
vn jdHN 1 :
By Remark 2.17 in [FM2] and the fact that S (vn ) = S (un) and vn = un , we can
rewrite the last integral as
S (un )\
where
(Z
Qun
jr (y)j dy :
Given 2 A(un (x); u+n (x); un ), it is clear that the function (y) := (x; (y)) belongs
to A(vn (x); vn+ (x); un ) and that r (y) = ru (x; (y))r(y). By (4.3) this implies
210
that
(Z
inf
jru (x; (y))j jr(y)j dy : 2 A(un (x); u+n (x); un )
Qun
Kh (x; un ; u+n ; un ):
(4.10)
jDvn j
+
S (un )\
211
juj
I : v 2 L1 (
; K ) 7! H(v;
) + (x; v(x)) dx
u(
) K nS:
In particular, if K = [a; b], if 2 is concave in [a; b], and if u is a minimizer of I , then u
must have a 2-phase structure, i.e. there exists a set
0
such that u(x) = a for LN
a.e. x 2
0 and u(x) = b for LN a.e. x 2
n
0. This result has important applications
u(x) =
(5.3)
k
X
i=1
ci !i (x);
)
k
X
1
S1 := u 2 L (
; R) : u = ci !i ; !i 2 E ; k 2 N D(V )
i=1
(
where
and
D(V ) := fu 2 L1 (
; R) : V (u) < 1g;
(ii) For any u 2 S1 of the form (5:3) (5:4), with k 2, there holds
(5.5)
V (u) =
kX1
i=1
(ci+1 ci )V ([kr=i+1 !r ):
212
V (u) = V (fx2
: u(x)tg ) dt:
R
It is easy to see that (5.6) reduces to (5.5) for functions u of the form (5.3)-(5.4).
Therefore (5.5) is weaker than (5.6). On the other hand, conditions (i) and (5.6) do
not seem to be related. Indeed, consider the functional
R maxfu(x); 0gdx if HN 1(S (u) \
) = 0
Z
V (u) := jDuj + 0
otherwise:
From the proof of Theorem 5.4 below it follows that V satises hypotheses (i)|(iii) of
Theorem 5.1. Take u(x) := 1 in (5.6); then V (1) = LN (
), while the right hand side
of (5.6) is innite. Therefore (5.6) fails. We note that V is not lower semicontinuous
in L1 .
We remark that Theorem 5.1 may be applied to a large class of functionals of the
form (1.1), for which the co-area formula might not hold.
(iii) If, in addition to hypotheses (i){(iii) in Theorem 5.1, we assume that V is
lower semicontinuous in L1(
; R), that K = R, and that there exist a set ! 2 E with
0 < LN (!) < LN (
), then V satises the following properties:
1) V (c) = 0 for all c 2 R;
2) V ( u) = V (u) for all > 0 and u 2 D(V );
3) V (u + c) = V (u) for all c 2 R and u 2 D(V );
R
4) V (u) R V (fx2
: u(x)tg ) dt for all u 2 D(V ).
In order to prove the rst property, dene
c + "n if x 2 !
un (x) :=
c
if x 2
n!,
where "n := n1 minf1; 1=V (! )g if V (! ) > 0, and "n := n1 otherwise. Clearly un ! c
in L1 (
; R), therefore by the lower semicontinuity of V and (5.5)
0 V (c) lim
inf V (un) = nlim
n!1
!1 "nV (! ) = 0;
where we have used the fact that V (! ) < 1 because S1 D(V ).
We omit the proofs of properties 2) and 3) since they follow quite easily >from
hypotheses (i) and (ii) of Theorem 5.1 and from the lower semicontinuity of V .
213
V (u) nlim
inf V (fx2
: un (x)tg) dt
!1 V (un ) = nlim
!1 V (fx2
: un (x)tg ) dt lim
n!1
R
R
N
by (5.5) and Fatou's Lemma. Since L (fx 2
: u(x) = tg) = 0 for all t 2 RnM ,
where L1 (M ) = 0, we x t 2 RnM and take a subsequence funk g of fung such that
lim
inf V (fx2
: un (x)tg) = klim
V (fx2
: unk (x)tg ):
n!1
!1
Then ffx2
: unk (x)tg g converges pointwise to fx2
: u(x)tg for LN a.e. x 2
and,
by Lebesgue Dominated Convergence Theorem, also strongly in L1 (
; R). Therefore
by the lower semicontinuity of V
lim
inf V (fx2
: un (x)tg ) V (fx2
: u(x)tg )
n!1
for L1 a.e. t 2 R, and we conclude that
R n!1
Z
R
V (fx2
: u(x)tg ) dt:
We do not know if the reversed inequality of 4) holds, i.e. if the co-area formula
(5.6) is satised.
Let
:= inf V (u) + IK (u;
) : u 2 D(V ); u(x) 2 K nS for LN a.e. x 2
:
Lemma 5.3. If u 2 S1 then
V (u) + IK (u;
) :
Proof. As IK (u;
) = 1 for u 2= L1 (
; K ) it suces to prove the result for
u 2 S1 \ L1(
; K ). By (5.1) we can decompose the open set S as a disjoint union of
bounded intervals
S = [r2R (ar ; br ):
Following Visintin [V2] we replace the function 2 by
8 (u)
if u 2 RnS
< 2
~2 (u) := : 2 (br ) 2 (ar )
(u ar ) + 2 (ar ) if u 2 (ar ; br ),
b a
r
i := V ([kr=i+1 !r );
Then by (5.3), (5.4), and (5.5)
k 1
X
i=1
i (c) :=
!i
(ci+1 ci )i +
~(x; c) dx:
k
X
i=1
i (ci ):
Let r 2 R be such that ci 2 (ar ; br ) for some i 2 f1; ; kg. There can only be nitely
many such r. Assume that k 2, and suppose that cl 2 (ar ; br ), l 2 f2; ; k 1g,
214
kX1
i=1; i6=l 1; l
k
X
i=1; i6=l
i (ci ) + l (t)
for t 2 [ar ; d], where d := cl+1 if cl+1 br and d := br if cl+1 > br . Since 2 (u) ~2 (u)
by construction, then clearly V (u) + IK (u;
) V (u) + I~K (u;
) = (cl ). Observe
that since ~(x; ) = 1 (x; ) + (x) ~2 () is concave in [ar ; d], then the function l ()
is also concave in [ar ; d], and (t), being the sum of a linear function and a concave
function, attains its minimum at one of the endpoints Q of [ar ; d]. It follows that
V (u) + IK (u;
) V (u) + I~K (u;
) = (cl ) (Q) = V (u) + I~K (u;
)
where
8 Pk c (x) + a (x)
>
r !l
< Pki=1; i6=l i !i
u(x) := > i=1; i6=l ci !i (x) + br !l (x)
: Pk
ci !i (x) + cl+1 !l [!l+1 (x)
i=1; i6=l;l+1
V (u) + IK (u;
) V (u) + I~K (u;
) = V (c) +
if Q = ar
if Q = br
if Q = cl+1 :
~(x; c) dx:
Assume that c 2 (ar ; br ) for some r 2 R. Since by (iii) and by the construction of ~2
the right hand side of (5.7) is a concave function of c 2 [ar ; br ], its inmum is attained
at one of the endpoints, say at br , and thus we can replace u(x) by u(x) := br 2 K n S .
We conclude that it is energetically possible to reduce at least by one the number
of values ci between ar and br . Repeating this procedure for the nite number of
intervals (ar ; br ) which contain at least one of the ci , by means of a nite induction
argument we can construct a simple function u^ of the form
u^(x) =
k^
X
i=1
Moreover, by hypothesis
nlim
!1
(x; un (x)) dx =
215
S (u)\
f (x; (u+
u )u ) dHN 1 + ^
where, we recall,
f (x; ) = h(x; ) + ^ '(x) for all (x; ) 2
RN :
Furthermore, we assume that the potential also satises the growth condition
for LN a.e. x 2
and all u 2 K ,
(5.8)
(x; u) b1 (x) + M1 (1 + jujqc )
where b1 2 L1(
; R), M1 > 0 and, as before, qc is the Sobolev exponent qc :=
N=(N 1) if N > 1 and qc < 1 if N = 1. Dene
R f (x; Du) if u 2 BV (
; R)
V (u) := 1
otherwise;
where
Z
Z
Z
Z
f (x; (u+ u )u ) dHN 1 ;
f (x; Du) := f (x; ru) dx + f (x; dC (u)) +
S (u)\
and take E to be the algebra generated by the class of open polyhedral subsets of
.
If u 2 S1 \ L1 (
; K ) has the form (5.3), then either !i = Ei or !i =
nEi , where Ei
is an open polyhedral set of
. Therefore, in both cases @!i \
= @Ei \
, which is
given by the intersection of a nite union of hyperplanes. Consequently, if i < j for
HN 1 a.e. x 2 @!j \ @!i the outward unit normal j (x) to the set !j at the point x
coincides with i (x). Moreover, for j < i < l we have
(5.9)
HN 1 (@!j \ @!i \ @!l \
) = 0;
and
(5.10)
= [ki=1 !i \
;
@!i \
= [j6=i @!i \ @!j \
;
@ [kl=i !l \
= @ [kl=i+1 !l \
n [kl=i+1 @!l \ @!i \
[ [ij=11 @!j \ @!i \
:
These properties will be useful in the sequel.
216
inf fI (u;
) : u 2 BV (
; K )g = inf fI (u;
) : u 2 BV (
; K nS )g :
Remark 5.5. Theorem 5.4 no longer holds in general if (x; ) is non linear.
Indeed, consider the simple case where
:= (c; d), K := [ 1; 1],
H(u;
) := jDu(x)j
and (x; u) := a(1 u2 ). Here K n S = f 1; 1g, and if > then all conditions
(H1 ) (H7 ) are satised.
Let u 2 BV (
; R), with u(x) 2 f 1; 1g for LN a.e. x 2
.
R
Then I (u;
) =
jDu(x)j 0. On the other hand, if we take u(x) 21 then
I (u;
) = 2 + 43 a(d c) < 0 provided a(d c) < 83 .
For the proof of the lemma below we refer to [F], [LM], and [Re].
Lemma 5.6. Let f :
RN ! R be a continuous function such that 2 RN 7!
f (x; ) is positively homogeneous of degree one for all x 2
, and
0 f (x; ) C (1 + j j) for some C > 0, all x 2
and 2 RN .
Let fn g be a sequence of Radon measures converging weakly-? to a Radon measure
and
nlim
!1 jn j(
) = jj(
):
Then
lim
n!1
f (x; dn ) =
f (x; d):
(5.4)
jDun j =
(i;j)2Jn
217
P n d (x), where
Let un(x) := ki=1
n;i !n;i
8 sup K
>
<
dn;i = > cn;i
: inf K
Since 0 (dn;j
quently
if cn;i sup K
if inf K < cn;i < sup K
if cn;i inf K:
jDun j
jDuj:
f (x; Dun ) =
f (x; Du):
V (u) =
(i;j)2J
(cj ci )
@!i \@!j \
f (x; j ) dHN 1
V (u) = ck
kX1 Z
k Z
X
f (x; k ) dHN 1 c1
f (x; l ) dHN 1
j =1 @!k \@!j \
1
0i 1 Z
k Z
kX1
X
X
f (x; l ) dHN 1 A :
f (x; i ) dHN 1
+ ci @
j =1 @!i \@!j \
i=2
l=i+1 @!l \@!i \
By (5.9) and (5.10) we can rewrite the rst two terms as, respectively,
ck
@!k \
c1
@ ([kl=2 !l )\
f (x; 1 ) dHN 1 ;
218
j =1 @!i \@!j \
f (x; i ) dHN 1
=
@ ([kl=i !l )\
k Z
X
l=i+1 @!l \@!i \
f (x; l ) dHN 1
@ ([kl=i+1 !l )\
where ^i and ^i+1 are, respectively, the outward unit normals to the sets [kl=i !l and
[kl=i+1 !l . It now follows from (5.11){(5.12) that
kX1
i=2
ci V ([kl=i!l )
kX1
i=2
ci V ([kl=i+1!l );
which is (ii).
R
Finally, by (5.2), (5.8) and the Sobolev inequality, the functional u 7!
(x; u(x)) dx
is continuous in BV (
; K ) (see [K, Th. 2.1]). Therefore we can now apply Theorem
5.1 (with (x; u) replaced by (x; u) + ^ u div '(x)) to obtain that
inf fI (u;
) : u 2 BV (
; K )g = inf fI (u;
) : u 2 BV (
; K nS )g :
Corollary 5.7. Assume that K = [a; b] in Theorem 5.4. Then there exists a
function u 2 BV (
; [a; b]nS ) such that
I (u;
) = inf fI (u;
) : u 2 BV (
; [a; b])g :
Proof. By Theorem 5.4
inf fI (u;
) : u 2 BV (
; [a; b])g = inf fI (u;
) : u 2 BV (
; [a; b]nS )g = :
To complete the proof it suces to apply Corollary 4.3, with K := [a; b]nS , to nd
u 2 BV (
; [a; b]nS ) such that I (u;
) = .
Remark 5.8. If we assume that 2 is concave in [a; b], then S = (a; b) and
consequently the minimizer u in Corollary 5.7 has the property that u(x) 2 fa; bg for
LN a.e. x 2
.
Acknowledgments. Part of this research was undertaken during G. Leoni's visit to
the Center for Nonlinear Analysis under the CNA/CMU{CNR project. The research
of I. Fonseca was partially supported by the National Science Foundation through the
Center for Nonlinear Analysis, and by the National Science Foundation under grant
No. DMS{9500531. The authors are deeply indebted to Luc Tartar for many fruitful
and stimulating discussions in the subject of this work.
REFERENCES
[ABS] Alberti G., G. Bouchitte and P. Seppecher, Phase transition with line tension eect,
submitted to Arch. Rat. Mech. Anal.
[ADM1] Ambrosio L. and G. Dal Maso, A general chain rule for distributional derivates, Proc.
Amer. Math. Soc., 108 (1988), pp. 691-702.
[ADM2] Ambrosio L. and G. Dal Maso, On the relaxation in BV (
; Rm ) of quasi{convex integrals, J. Funct. Anal., 109 (1992), pp. 76-97.
219
[AMT] Ambrosio L., S. Mortola and V.M. Tortorelli, Functional with linear growth dened
on vector{valued BV functions, J. Math. Pures et Appl., 70 (1991), pp. 269-322.
[BFM] Bouchitte, G. I. Fonseca and L. Mascarenhas, A global method for relaxation, submitted to Arch. Rat. Mech. Anal.
[C] Cahn J. W., Critical point wetting, J. Chem. Phys., 66 (1977), pp. 3667-3672.
[CH1] Cahn J. W. and J. E. Hilliard, Free energy of a non{uniform system I: interfacial
energy, J. Chem. Phys., 28 (1958), pp. 258-267.
[CH2] Cahn J. W. and J. E. Hilliard, Free energy of a non{uniform system, J. Chem. Phys.,
28 (1959), pp. 688-699.
[DM] Dal Maso G., Integral representation on BV (
) of {limits of variational integrals,
Manuscripta Math., 30 (1980), pp. 387-416.
[EG] Evans L. C. and R. F. Gariepy, Measure Theory and Fine Properties of Functions, CRC
Press, 1992.
[F] Fonseca I., Lower semicontinuity of surface energies, Proc. R. Soc. Edin., 120A (1992),
pp. 99-115.
[FM1] Fonseca I. and S. Muller, Quasi{convex integrands and lower semicontinuity in L1 ,
SIAM J. Math. Anal., 23 (1992), pp. 1081-1098.
[FM2] Fonseca I. and S. Muller, Relaxation of quasiconvex functionals in BV (
; Rp ) for integrands f (x; u; ru), Arch. Rat. Mech. Anal., 123 (1993), pp. 1-49.
[FR] Fonseca I. and P. Rybka, Relaxation of multiple integrals in the space BV (
; Rp ), Proc.
Roy. Soc. Edinb. Sect.A, 121 (1992), pp. 321-348.
[GSe] Goffman C. and J. Serrin, Sublinear functions of measures and variational integrals,
Duke Math. J., 31 (1964), pp. 159-178.
[GMS] Giaquinta M., G. Modica and J. Soucek, Functional with linear growth in the Calculus
of Variations, Comment. Math. Univ. Carolinae, 20 (1974), pp. 143-172.
[G] Gurtin M.E., Some results and conjectures in the gradient theory of phase transitions,
IMA, Preprint, 156 (1985).
[K ] Krasnosel'skii M.A., Topological Methods in the Theory of Nonlinear Integral Equations,
Pergamon Press, 1964.
[LM] Luckhaus S. and L. Modica, The Gibbs-Thompson relation within the gradient theory of
phase transitions, Arch. Rat. Mech. Anal., 107 (1989), pp. 71-83.
[MP] Massari U. and L. Pepe, Su di una impostazione parametrica del problema dei capillari,
Ann. Univ. Ferrara, XX (1975), pp. 21-31.
[Mo1] Modica L., Gradient theory of phase transitions and minimal interface criterion, Arch.
Rat. Mech. Anal., 98 (1987), pp. 123-142.
[Mo2] Modica L., Gradient theory of phase transitions with boundary contact energy, Ann. Inst.
Henri Poincare, 4 (1987), pp. 48 5-512.
[M] Muller S., On quasiconvex functions which are homogeneous of degree 1, Indiana Univ.
Math. J., 41 (1992), pp. 295-301.
[Re] Reshetnyak Y. G., Weak convergence of completely additive vector functions on a set,
Siberian Math. J., 9 (1968), pp. 1386-1394; (translation of Sibirsk. Mat. Z^ ., 9 (1968),
pp. 1386-1394).
[Se1] Serrin J., A new denition of the integral for non{parametric problems in the Calculus of
Variations, Acta Math., 102 (1959), pp. 23-32.
[Se2] Serrin J., On the denition and properties of certain variational integrals, Trans. Amer.
Math. Soc., 161 (1961), pp. 139-167.
[V1] Visintin A., Models of Phase Transitions, Birkhauser, 1996.
[V2] Visintin A., Nonconvex functionals related to multiphase systems, SIAM J. Math. Anal.,
21 (1990), pp. 1281-1304.
[vdW] van der Waals J. D., The thermodynamic theory of capillarity under the hypothesis
of a continuous variation of density, in Dutch, Verhandel. Konink. Akad. Weten.
Amsterdam (Section 1), 1 (1893); (translated by Rowlinson J.S., J. Stat. Phys., 20
(1979), pp. 197-244).
[Z] Ziemer W., Weakly Dierential Functions, Springer, 1989.