Sei sulla pagina 1di 19

AppI Math Optim 27:i05-123 (1993)

Applied Mathematics
and Optimization
9 1993 Springer-Verlag New York Inc.
Asymptotic Development by F-Convergence
Gabri el e Anzellotti and Sisto Baldo
Dipartimento di Matematica, Universitfi Degli Studi di Trento,
38050 Povo (Trento), Italy
Communicated by D. Kinderlehner
Abstract. A descri pt i on of the asympt ot i c devel opment of a family of mini-
mum probl ems is pr oposed by a suitable i t erat i on of F-limit procedures. An
exampl e of asympt ot i c devel opment for a family of funct i onal s rel at ed to phase
t r ansf or mat i ons is also given.
Key Words. F-convergence, Asympt ot i c devel opment s, BV functions.
AMS Classification. Pr i mar y 49J45, Secondary 41A60.
O. Introduction
It is very common, bot h in pur e and appl i ed mat hemat i cs, to have to deal with
a family of probl ems dependi ng on a par amet er e > 0, being i nt erest ed in t he
asympt ot i c behavi or of t he probl ems as e - , 0. Typically, in t he calculus of
vari at i ons we are given a family of mi ni mum probl ems
mi n{~( u) : u ~ X} (0.1)
and we are i nt erest ed in finding a "l i mi t pr obl em"
min{~-(u): u e X}. (0.2)
Of course, from the poi nt of view of t he calculus of variations, the limit pr obl em
must be such t hat its minimizers are closely rel at ed to the possible limit points of
minimizers {u~}~ of pr obl em (0.1).
A not i on of convergence for functionals, which is very well suited to the
vari at i onal setting, is the wel l -known F-convergence, i nt r oduced by De Gi orgi
[ DF] . In fact, if a funct i onal ~ is the F-limit of the ~ and if u s are minimizers
of ~ and u s ~ u, t hen u is a mi ni mi zer of ~ (see Section 1 for a precise statement).
106 G. Anzei!otti and S. Ba!do
Roughl y speaking, we have
{limits of minimizers} ~ {minimizers of the F-limit}, (0.3)
where the inclusion may well be proper, as we can see by very simple and nat ural
examples. Hence the F-limit, t hough very useful, fails in general to characterize
c o m p l e t e l y the asympt ot i c behavior of the family ~ .
Our remark is t hat in fact the F-limit is onl y the first step t oward the
description of the asympt ot i c behavior o f ~ , and t hat we may try to pursue further
the description l ooki ng for an asympt ot i c development
= o~(o) + e , ~ 1) + e 2 y ( a ) + . . . + eko~(k) + o(sk), (0.4)
where the first-order t erm ff(o) is j ust the F-limit Y of the family Y~ and each
one of the higher-order terms y ( 0 is a functional defined by a nat ural recursive
procedure on the space ~,~-1) of the minimizers of f f ( ~- l ) ( s ee the following
section). When a development as in (0.4) holds, then we have the following
situation:
{limits of minimizers} c {minimizers of ~(k)}
c {minimizers of c~ (k- 1 ) } C ' ' " C {minimizers of ~,~t0)}. (0.5)
This may provide a considerable i mprovement of (0.3), and in some cases may
give a complete characterization of the asympt ot i c behavior of ~ .
In Section 1 we give the general definitions and theorems about the not i on
of asympt ot i c development of a family of functionals. In Section 2 we illustrate
the general t heory by a simple but not completely obvious example, related to the
well-known example by Modi ca and Mor t ol a [ MM] . It was in fact in the context
of this example t hat we first got the idea of characterizing the asympt ot i c behavior
t hrough a sequence of functionals j~(k) defined on nested spaces. The idea t hat the
functionals o~(k) could be t hought formally as an asympt ot i c development (written
as in (0.4)) was suggested by an incidental remark by De Giorgi. On the other
hand, the fact t hat some not i on of asympt ot i c development by F-convergence
could be useful, must have been known more or less explicitly by many people
working in the field. For instance, somet hi ng which is close to an asympt ot i c
development can be found in a work by Buttazzo and Pereivale [BP], and a first
at t empt at a definition can be found at the end of a paper by Modi ca [M]. After
all, even the scaling of the functionals in the first paper by Modi ca and Mor t ol a
[ MM] may be t hought of as an unconscious order-one development.
1. The Asymptotic Development of a Family of Functionals by F-Convergence
Let X be a topological space for which the first axiom of count abi l i t y holds, and let
~ : X~ R
be a family of functionals, with 8 a positive parameter.
For such a family the following definition of the F-limit is well known (see
[ DF] and [DM]).
As ympt ot i c Devel opment by F- Conver gence 107
D e f i n i t i o n 1.1. A funct i onal ~(o): X ~ I] is said to be the F( X- ) - l i mi t of the family
. ~ at a poi nt f i e X i fft he following st at ement s are fulfilled for each sequence es ] O:
(i) For each sequence {@ = X with u s ~ ~ we have
lim i nf ~, (us) > ~(o)(~).
]--, + ce
(ii) Ther e exists a sequence {us} ~ X with u s --* fi and
lim sup ~( us ) <_ ~m)(fi).
j--, + ~
In this si t uat i on we write
F( X- ) lim ~( ~) = ym)(fi).
e--~O
D e f i n i t i o n 1.2. We write
F( X- ) lim ~ = g( o) in E ~ X
~+ 0
if ~( o) is the F( X- ) - l i mi t of ~ at each poi nt f i e E.
When no confusi on may arise we often omi t the specification of the space X.
The i nt r oduct i on of F-convergence in t he calculus of vari at i ons is justified by t he
following wel l -known result, whose easy pr oof can be f ound in the papers quot ed
above.
Theor em 1.1. Le t e s ~ 0 be a f i x e d sequence, let {us} c X be such t hat ~, ( u j ) =
mi n{~( u) : u ~ X} . I f ~ ~~ is t he F- l i mi t o f ~ on t he whol e space X and uj ~ f~ in
X, t hen fi is a mi ni mi z er o f ~<o) and we have
lim ~ , ( u s ) = f f ( o ) ( f i ) .
j ~+c e
Unfort unat el y, we may well have many minimizers of the F-limit, which are
not limit poi nt s of minimizers of the functionats ~ . A trivial exampl e of such a
phenomenon is t he following one.
Exampl e 1.1, Consi der the case X = R and
~ ( u ) = ~ l u l .
We can easily check t hat the functions . ~(u) F-converge to the const ant function
y~o) _ 0. Clearly, every poi nt in R is a mi ni mum poi nt of y{0~, white the onl y
limit poi nt of the minimizers of ~ is the poi nt u = 0.
Now, t he idea is t hat of i nt roduci ng a not i on of "asympt ot i c expansi on"
= ~ o ) + ~ y , ) + . . . + ~k~(k~ + O(ek)
108 G. An z e l l o t t i a n d S. Ba l d o
of a fami l y J~(u) in such a way t hat the knowl edge of t he funct i onal s ~(k) gives
addi t i onal i nf or mat i on on the limit poi nt s of minimizers. Precisely: any limit poi nt
of a sequence of mi ni mi zers us will also be a mi ni mi zer of each one of t he
funct i onal s ~(k) appear i ng in the devel opment above. Just by const ruct i on, the
sequence of sets
~//k = {minimizers of ~(k)}
will be noni ncreasi ng, and we ma y hope t hat in some cases t he mi ni mi zers of
some t er m ~-(k) of t he devel opment are exact l y al l the possi bl e limit poi nt s of
mi ni mi zers u~.
Now, let us discuss how t o define a sui t abl e not i on of asympt ot i c devel opment .
For instance, suppose we want t o give meani ng to the expressi on
= y( o) + ~ ( 1 ) + o(8). (1.1)
Nai vel y, we shoul d like t o say t hat (0,I) is equi val ent to
lim _ ~( i ) ,
s ~O 8
where the limit shoul d be t aken in t he sense of F-convergence. Unf or t unat el y such
a definition makes little sense, as we can convi nce oursel ves by t ryi ng to appl y it
to si mpl e situations. For exampl e, it may happen t hat ~ and j~(o) are finite on
disjoint domai ns (see the case in [ MM] ) . However , it t urns out t hat we can give
a simple and very good subst i t ut e for the nai ve definition:
Set
~-(o) = F - lim ~ in X
e ~O
(we assume the the F-l i mi t exists) and also set
m o = inf ~(o), ~o = {u ~ X: ~(~ = too}.
X
Def i ni t i on 1.3.
We say t hat the fi rst -order asympt ot i c devel opment
= ~( o) + s~(1) + o(e)
holds, if we have
F - lira ~ - m~ - ~( 1) in ago.
~ 0 8
Wi t h the not at i on above, assume t hat m o < + oo and 0g o # ~ .
(1.2)
Fr om now on we use the not at i on
~ , ) = ~% - m 0
8
The definition (1.2) is mot i vat ed mai nl y by the following very si mpl e results.
As y mp t o t i c D e v e l o p me n t b y F - C o n v e r g e n c e 109
T h e o r e m 1.2. Suppose the first-order asymptotic development (1.2) holds, and let
ej .[ 0 be a sequence f or which there exists a sequence {u j} = X, uj ~ fi in X, and
~ ( u j ) = min{Y~,(v): v ~ X}. Then ~t ~ ago and ft minimizes ~(1) in ago. Moreover, i f
m~ denotes the infimum of ~ on X and m 1 denotes the infimum of ~ (1) on ago we have
m~ =mo + ejml + o(ej). (1.3)
Proof. T h e f a c t t h a t ~ e ago is a c o n s e q u e n c e o f T h e o r e m 1.1. Le t v e ago a n d l et
{ @ c X b e a s e q u e n c e c o n v e r g i n g t o v i n X s u c h t h a t ~(1)(v.~--, ff(1)(v): f r o m
t he F - c o n v e r g e n c e o f @~1) i t f ol l ows t h a t s u c h a s e q u e n c e exi st s. F o r e a c h f i xed
i n d e x j we h a v e t h a t ~j~(1)(Va)>_ ff{~)(u.~j, a,, a n d s o t h e F - c o n v e r g e n c e yi e l ds
f i r e ( v ) l i m ~ ( : ) " '
= ~ j (~;~ >_ l i mi n f ~ ( 1 ) .
~ (~;I >- ~-(1)(~).
j ~ + ~ j ~ + m
I n p a r t i c u l a r we h a v e
j ~ + c o
Th i s r e a d s l i mj ~ + o~[(m~j - mo)/ej - m: J = 0, wh i c h i mpl i e s (1.3). [ ]
R e ma r k . N o t e t h a t (1.3) i n g e n e r a l is t r u e o n l y f o r s e q u e n c e s ea $ 0 f o r wh i c h t h e r e
is c o mp a c t n e s s f o r mi n i mi z e r s : i n p a r t i c u l a r , i t is false t h a t
m~ = m o + em 1 q- o(e), (1.4)
as t he f o l l o wi n g e x a mp l e s hows .
Le t X = R, a n d c o n s i d e r t h e f o l l o wi n g f a mi l y o f f u n c t i o n a l s
f
0 i f u = 0 ,
.7~(u) = - e 1/2 i f u = 1/e a n d ~ is r a t i o n a l ,
{e 2 o t h e r wi s e i n R.
Of c o u r s e e a c h f u n c t i o n a l ~ ha s a u n i q u e mi n i mi z e r , wh i c h is 0 i f e is i r r a t i o n a l
a n d is 1/e i f e is r a t i o n a l . T h e o n l y l i mi t p o i n t o f mi n i mi z e r s is O, a n d we h a v e
c o mp a c t n e s s o n l y f o r s e q u e n c e s ej $ 0 s u c h t h a t ej is de f i ni t i ve l y i r r a t i o n a l . I n
~21/2
p a r t i c u l a r , (1.4) is f al se f o r a r a t i o n a l s e q u e n c e ej J. 0, f o r i n t hi s c a s e m,, = - v j ,
whi l e y ( o ) _ 0 a n d ~.~-(1) _= 0.
T h e p r o c e s s a b o v e c a n be i t e r a t e d i n t h e f o l l o wi n g way. S u p p o s e t h e a s y mp t o -
t i c d e v e l o p me n t o f t h e f i r s t o r d e r (1.1) hol ds , r e c a l l t h a t mi = inf,4/o ~( ~) , s et
d~ 1 = {U E 0~r f f ( 1 ) ( U ) = m l } ,
a n d s u p p o s e m a < + oo a n d ag~ r ~ . T h e n c o n s i d e r t h e f a mi l y o f f u n c t i o n a l s
. ~ - ( 1 )
g
I f we h a v e
y ( 2 ) = F - l i m f f ~ ) i n agl ,
e-+ 0
110 G. Anzellotti and S. Baldo
t hen we say t hat the asympt ot i c devel opment of the second or der
~ ( 0 ) ~_ g ~ ( 1 ) _[_ g 2 ~ ( 2 ) _}_ O(g2)
holds.
Obvi ousl y, in this si t uat i on the anal ogs of Theor ems 1.2 and 1.3 hold, so t hat
the limit poi nt s of t he mi ni mi zers of t he funct i onal s ~ mus t mi ni mi ze the
funct i onal ~ 2 on the set d//1.
The definition of t he asympt ot i c devel opment of order k is obt ai ned si mpl y
by recursi on in the above process: if t he asympt ot i c devel opment of order k - 1
holds, we set m k _ ~ =i nf {~( k- *) }, ~//k-1 = {us~#k- 2: 5~(k-1)(u) =i nk_l } and,
assumi ng m k _ 1 < + cc and ~ k - ~ V: ~ ,
I f
y(k) = F - lim ~-(f) in q/k-~,
~--~0
we write
= ~( o) + ~ ( 1 ) + . . . + ~k~(k) + o(~k),
the devel opment of or der k. Of course, a limit poi nt of mi ni mi zers of the fami l y
. ~ has t o be a mi ni mi zer of ~(k) in ~(k-1) and
gg/ej = H'/O -~- ~ j m l -t . . . . . "~ ~km k -'~ O(/3~).
We concl ude this section by giving a few very si mpl e exampl es of asympt ot i c
devel opment s of families of functionals, showi ng how the above t heoret i cal set t i ng
can give useful i nf or mat i on in some situations. A mor e compl ex exampl e is t reat ed
in Section 2.
We also give an exampl e which show t hat our met hod may need i mpr ove-
ment s for mor e general classes of functionals.
Exampl e 1.2. Consi der the fami l y of funct i onal s in Exampl e 1.1. We have shown
t hat ~~ = 0 is the F-l i mi t of or der 0, and so m o = 0 and ~?l ~ = R. On the ot her
hand, Y(1)(u) = lul, and so m I = 0 and ~ = {0}.
Thi s way we al ready have the compl et e i nf or mat i on about t he limit poi nt s of
mi ni mi zers by arrest i ng our devel opment t o t he fi rst -order term.
I f we consi der a sequence of the f or m
~ ( u ) = d I u I,
we have t hat all t he t er ms of the asympt ot i c devel opment s of or der < k are zero
functional, while t he asympt ot i c devel opment of or der k is [ u I + o ( ~ k ) , and so we
need t o comput e k F-l i mi t s in or der t o have the compl et e i nf or mat i on on the limit
poi nt s of minimizers.
Of course, in this par t i cul ar case this phenomenon is due to a bad choi ce of
Asymptotic Development by Y-Convergence 111
the par amet er e: if we consi der e ~ as a new par amet er we reach this i nf or mat i on
in one step only. The following exampl e shows t hat it is not always so.
Exampl e 1.3. Let X = [ - 1, 1-] c R, and consi der the following family of func-
tionals:
< ( u ) =
0 if u = 0 i 1
1 1
e k if j U[ ~ ' 2 k- 1 "
For t he devel opment of or der zero we have the same si t uat i on as before:
F-converges t o y( 0) _ 0, and so q/o = [ _ 1, 1], mo = 0 and we have no i nforma-
t i on at all. We have
{0 i f u = ~ l ~ l ) ( u ) = 1 1
e k-1 if l ul ~ , 2 ~ .
On the set ~//o t he family y~l ) F-converges to the following funct i onal :
.T/1)(u) = {01 if u ~ [ - 89189
otherwise in ~o.
Hence we have oR* = I-_21_, 89 and m 1 = 0. By i nduct i on on the above process we
get ~k = [ _ 1/2k, 1/2a], m k = 0, and
y{k)(u)={01 if I x l < _ l / 2 k,
otherwise in ogk- 1.
As ~ff=~ q/k = {0}, onl y the compl et e asympt ot i c devel opment of Y, gives us the
desired i nf or mat i on about t he limit poi nt s of minimizers.
Exampl e 1.4. In some cases our process does not work: this, we believe, is
essentially due to the necessity of the right choice of s c al i ng in our family. For
instance, if we modi fy t he family of Exampl es 1.1 and 1.2 as follows:
~ ( u ) = e - l l ~ l u l ,
we have t hat all the t erms of the asympt ot i c devel opment are the functional
const ant l y equal to zero, and so with this ill-omened scaling we cannot hope to
get any i nf or mat i on at all.
A similar si t uat i on is the following:
1
g ~ ( u ) = l u l + - .
g
In this case the F-limit is const ant l y equal to + az, and so we cannot devel op
our family further.
112 G. Anzellotti and S. Baldo
Of course a simple rescaling obt ai ned by mul t i pl yi ng the functionais by o;
eliminates the probl em, but in general det er mi nat i on of the right scaling is far
from easy.
Remark. A si t uat i on as t hat of Exampl e 1.5 may arise if we have a count abl e
sequence of functionals {YJ} instead of a cont i nuous family.
In this case, once we have defined the funct i onal ~-(o) as the F-limit of the
family, we may define ~(~) as the F-limit of the sequence
o)j
where coj is a suitable vanishing sequence. In this case we do not have a natural
scaling given by t he par amet er e, and we must carefully choose a sort of "or der
of zer o" for the sequence ~ .
2. A Concrete Example of Asymptotic Development in a
Problem Related to Phase Transitions
In this section we st udy a family ~ of funct i onal s very closely related to those
al ready consi dered by Modi ca and Mor t ol a [ MM] , and afterwards studied in
many papers, also in view of their physical meani ng in the t heor y of phase
transitions, see, for example, [ MI ] , [ M2] , IS], [ FT] , [B], [ABV], and [BB].
In the papers quot ed above, the i nformat i on about the asympt ot i c behavi or
of the minimizers of the family of functionals was obt ai ned simply by comput i ng
the F-limit of a suitable rescaling of the family itself: in the setting we established
in the previ ous section this corresponds to the fi rst -order asympt ot i c devel opment
of the family. In this section we show that, in di mensi on one, the second- or der
devel opment of ~7 can also be comput ed and t hat this gives furt her i nf or mat i on
about the asympt ot i c behavi or of the minimizers. We show, furt hermore, with an
example, t hat in hi gher dimensions t he seemingly most nat ural conj ect ure for the
second F-limit 2 (27 is indeed false. For the moment we are not able to conj ect ure
the correct form of g( 2) in di mensi on great er t han or equal to 2.
Thr oughout this section we have X = L~(f~), with .q c R n a bounded open
set with C2-regular boundar y. In the second par t of the section we restrict ourselves
t o n = l .
Suppose we have a funct i on (o: R -~ [0, + oo) with the following propert i es:
( i ) ~o e C ~
(ii) Z = { u : ( p ( u ) = 0 } = [ a , b ] u [ c , d ] wi t h a < b < c < d .
(iii) Ther e is a K > 0 such t hat (p(u) decreases for u < - K, but increases for
u>K.
a b c d
Asymptotic Development by F-Convergence 1t3
Let g: ~?fl -* [a, d] be a l i pschi t z-cont i nuous function, and define our family
of functionals as follows:
~ ( u ) = I e 2 f n , D u l 2 d x + f n c p ( u ) d x if u ~ H ~ ( ~ ) , u , o n = g ,
+ Go otherwise in Ll(f~).
Up t o rescaling our family of funct i onal s is the same as in the paper by Modi ca
and Mor t ol a, but in our case t he funct i on (p vanishes in t wo intervals instead of
two points. We shall see t hat this difference makes it possible and interesting to
comput e a step mor e in t he asympt ot i c devel opment . Her e we also have a
boundar y condi t i on which is not present in [ MM] , but this makes onl y a technical
difference.
We r emar k t hat each funct i onal ~ has at least one mi ni mi zer in L~(f~) and
that, under suitable hypot hesi s on t he growt h of q) at infinity (a superl i near gr owt h
is enough), each family of minimizers {u~}~ is relatively compact in Ll(f~) (see [ M1] ,
[FT-1, and [B]). Hence we are lawfully al l owed to speak of "l i mi t points of
minimizers. "
Th e o r e m 2.1. We have F - l i m~o ~ = if(o) in Ll (f ~), where
g~~ = ~ ~ (u (x )) dx .
J ~
P r o o f We fix a sequence ej i 0 and a funct i on u ~ L~(~).
Fi r s t st ep. We must show first that, for each sequence {uj } c LI(f~) with uj ~ u
in Ll(f~), t he following i nequal i t y holds:
lim inf ~ j ( u j ) >_ J ( ~
j~+o~
Thi s follows easily f r om Fat ou' s lemma.
S e c o n d st ep. We must const ruct a sequence {uj} c L~( ~) with uj ~ u in L~(fl)
and such t hat
lim ~ j ( u j ) - - ~( ~
j ~+~
To do this, we first choose a sequence {vj} ~ W 1' ~(~) such t hat vj converge to u
in L~(f~) and al most everywhere in f~. We may suppose t hat vjl~, = g (this can be
obt ai ned by modi fyi ng the original sequence vj on small nei ghbor hoods of ~?fl).
If u is bounded, {v j } can also be t aken uni forml y bounded in L ~~ and we
obvi ousl y have
l i m f a ~o(vj) d x = f n ~o(u) dx.
j ~ + ~
If u is not bounded, we can obt ai n the same result by appr oxi mat i ng the
t runcat i ons
W~(X} = max{- - m, min{u(x), m}}
t 14 G. Anzellotti and S. Baldo
of u, and by t aki ng a suitable di agonal sequence. This is possible by propert y (iii)
of q~.
If the following inequality holds
f Dv j [ 2 ~ - (2.1)
~j
our claim is proved.
This is obt ai ned simply by "sl owi ng down" the convergence of the sequence
{@: the sequence {@ is obt ai ned from {@ simply by repeating each term Vk
many times in order to have the estimate (2.1). []
The F-limit (or the asympt ot i c development of order zero, which is the same)
does not tell us very much about the asympt ot i c behavior of the minimizers of
the family ~.~ as z ~ 0. In fact the only thing we can say at this point is t hat a
limit point of minimizers has to be a minimizer of y(o), t hat is an element of the
very large space
~/o = {u c Ll(f2): u ( x ) ~ Z for almost all x c f2}.
Hence we l ook for higher-order developments, according to the definitions
established in Section 1.
By using the not at i on of the previous section we have
t oo= inf ~ ~
ucLl(~)
~ ( u ) - mo
Y ~ ) ( u ) -
[ D u ] 2 d x + - q ) ( u ) d x if u e H t ( f ] ) , u = g o n ~ ? ~ ,
, + oo otherwise.
We remark t hat the functionals ~ ) , apart from the boundar y condition, are the
same ones considered in [ MM] .
We use the following not at i on:
. ~ ( t ) = f ] q ; / 2 ( s ) d s , c o = f f q ; / 2 ( s ) d s = ~ ( c ) .
I f u s N ~ we set
A , , = { x c ~ : u ( x ) e [ c , d ] } .
Notice t hat in this case
r = COZA, ,
(.
j alD(q~(u))[ = Co Pn ( Au ) ,
where Pa(A,) is the perimeter of the set A, (see [G]).
Asymptotic Development by F-Convergence !15
We consider the functional ~o) defined on 4 `0 as follows:
( 2c oPa( Au) + 2 f IcI)(u) - (I)(g)[ d~Vf"-i if Pn(Au) < + ~ ,
~( 1) ( U)
L + oe otherwise in 4`(0),
where (l)(u) denotes the trace of (I)(u) on Of~, which is well defined whenever the
perimeter of A, is finite.
Theorem 2.2. We have
F - lira -~(~) = ~ o ) in 4`0.
~;-*0
Pr o o f Fix ej ~ 0 and u ~ 4`0.
Fi rst step. We show that, for each sequence uj -* u in Ll(f~), we have
lira inf ~( t ) - ~ j ( - ; > 5 ( " @ .
j ~+o c
Wi t hout loss of generality, we can restrict ourselves to the case {@ c Hl(f~) with
ust~ = g, and we can also assume t hat the limit
lim ~-o). ,
j-~ + oe
exists and is finite. In particular we may assume t hat ~ (. jj _< C.
Fr om the inequality
f n f a 1 f n f ~
~slDu~l ~ + - ~0(uj) ~ 2 I Du j l ~' ~( u s ) = 2 IV(q~(uJx)))l
gj
and the lower semi cont i nui t y of the t ot al variation we get t hat 9 o u e BV( D) . In
particular, the trace of (q~ o u) on 8~ is defined. Applying again the same inequality
we obt ai n
infY~))(uj) >_ 2 lim inf t ]D(~ o u i)[ lim dx
S~+oo j ~+oe d f~
_> 2 ]D(qbou)[ + [O(u) - q)(g)l d~/ f "-~ = ~ ( 1 ) ( u ) .
The last inequality is a consequence of the lower semicontinuity in L~(f2) of the
functional
f n l D v l + f o ] v - h l d ~ ~
defined for v ~ BV(f ~) and with h s Ll(Of~) fixed (see [G]).
Se c ond step. We must exhibit a sequence {uj} converging to u in L~(f~) such
t hat
lim ~ o ) .
5' qq (uj ) = /~(1)(U).
j ~+o~
The other case being obvious, we can assume t hat ~~ < + oc.
116 G. Anzellotti and S. Baldo
We can reduce the pr oof to a si t uat i on in which t he set ~?*A u is regul ar and
intersects Off t ransversal l y: to do t hat we can use appr oxi mat i on results as in [BB],
[B], and [ OS] .
As in the pr oof of Theor em 2.1 we can get a sequence {fj} c W 1, ~(f~) such
t hat fjlo~ = 9, f j --' u in Ll (fl ) and a.e., and such t hat t he fol l owi ng est i mat e holds:
IIDf~llL~ --< C ~ / 4 .
Consi der t he following par t i t i on of n:
A j = { x ~ A . : d( x, A c) > K e j and d( x, 3 n ) > K @ ,
V~ = { x ~ A . : d( x , A c) <_ K @ w { x ~ f i : d( x, On) <_ Ke j } ,
B j = n\(A Vj),
where K is a sui t abl e posi t i ve const ant . We now define
u~(x) = max{min{f~(x), b}, a} if x e B~,
uj (x) = max{mi n{fj (x), d}, c} if x E Aj.
Finally, we may ext end the funct i on uj wi t h funct i ons const r uct ed as in [B],
[BB], and [ OS] : these funct i ons are responsible, at t he limit, for the per i met er
and t he boundar y i nt egral which appear in ~-(~). On the ot her hand, the i nt egral s
on A: and Bj vanish.
A simplified discussion of the ideas used in t he paper s above can be found
in a r emar k on p. 186, of [ABV], where t he pr oof of case n = 1 is outlined. [ ]
Wi t h this last F-l i mi t we have gai ned furt her i nf or mat i on on the asympt ot i c
behavi or of the mi ni mi zers of our fami l y of functionals, in part i cul ar, we can now
say that, for a limit poi nt u of mi ni mi zers, the set A u has mi ni mal boundar y. In
ot her words, by empl oyi ng t he usual not at i on we have
q/1 = {u ~q/ o: A~ mi ni mi z e s H(A)},
where
H( A) : = 2 c o P n ( A ) + 2 ( ICo;~a - qb(g)[ d ~ , , - ~
.J o
9Z
is defined on the subsets of ~ wi t h finite per i met er and
m~ -- inf ~(*~ = mi n H( A) .
a
As we have said before, we are abl e t o comput e a furt her st ep in t he as ympt ot i c
devel opment of o~ when ~ is an open i nt erval of R, for exampl e, n = (0, 1).
Put
)Cu o if x = 0,
g(x)
u 1 if x = l ,
As y mp t o t i c De v e l o p me n t by F - Co n v e r g e n c e 117
wi t h u0 e (a, b) a n d Ul e (c, d). I n t hi s c a s e we h a v e
m 1 = i n f ~ ( i ) = 2Co,
q/o
ag* = {u e a//~ A, i s o f t h e t y p e (2, 1) wi t h 2 ~ [0, 1]}.
We d e f i n e
c ~ 2 ) ( U) - - ~ 1 ) ( u ) - - ' ~1
8
a n d
I f n (u' ) 2 d x i f u e ~/~,
~(2>(u) = \{~)
V + co o t h e r wi s e i n q/~,
wh e r e
"J~ ~- (U e d~/1 wi t h 2 # 0 a n d 2 # I, u ~ HI ( Au ) ~ Hl ( f ~\ Au) ,
u(0) = Uo, u(1) = ul , u - ( 2 ) = b, u+( 2) = c}.
He r e u - ( 2 ) a n d u+( 2) d e n o t e t h e t r a c e s o f u f r o m t h e l ef t a n d f r o m t h e r i g h t a t
t h e p o i n t 2.
Theorem 2. 3. I f f l = (0, 1) c R, we have
F - l i r a ~ 2 ) = y ( 2 ) in ql l .
8--+0
Pr oof . As u s u a l , f i x ej $ 0 a n d u E q/1.
Fi r s t st ep. We s h o w t h a t f o r e a c h s e q u e n c e uj --+ u i n Ll ( f ~) we h a v e t h e
i n e q u a l i t y
l i r a i n f ~-(~2)(uj) _> ~,~(2)(u).
j ~ + o o
Wi t h n o l os s o f g e n e r a l i t y , we m a y s u p p o s e t h a t t h e l i mi t
l i m ~ ' ( 2 ) . X
~j ( Uf l
j - ~ + oo
e xi s t s a n d i s f i ni t e. I n p a r t i c u l a r we a s s u me .~~2)(uj) < C < + o0.
T h e s e t A u is o f t h e f o r m (2, 1], wi t h 2 ~ [-0, 1]. S u p p o s e f o r t h e m o m e n t t h a t
2 e (0, 1), l a t e r we s h a l l s ee t h a t i f t hi s i s n o t t h e c a s e we mu s t h a v e
l i r a i n f o~(2). ~/ (uj~ = + oo.
j -+ + oo
118 G. Anzellotfi and S. Baldo
Next put
X 1 = ess sup{x ~ [-0, 1]: u ( x ) < b } ,
x 2 = ess inf{x e [0, 1]: u ( x ) > c } .
Of course, x I < ~ < x z .
As a consequence of our position, for each 6 > 0 t here exists a a > 0
and a subset of positive measure of (x2, x2 + 8) such t hat u ( x ) > c + or. As uj
can be supposed t o converge to u al most everywhere in ~, we can find a poi nt
x ~ ~ ( X z , x 2 + 6 ) such t hat u j { x * ) > c for j large enough.
For the same reason, for every 6 > 0 t here is a poi nt x* e (xl - (5, Xx) such
t hat u j ( x * ) < b for j large enough.
By denot i ng I* the interval Cx* x *~
1 ~ 2 1 w e h a v e
= - - , ~ j Ai- - - ~O( Uj ) - - m I + (U))2 + 25 r
- - " e j ' , , I ; ~ ' J ' ~ J , ) ~ \ I , ~ C , j \ i ~
(2.2)
The expression bet ween the square bracket s is positive: in fact, the sum of the two
integrals is great er t han or equal to
2 f ~ I D(~o uj) l,
and this last expression is great er t han rnl because u j ( x T ) < b and u j ( x * ) > c.
Because t he last integral of (2.2) is also positive, we get the estimate
i
(u)) 2 d x < C ,
,J f l \ l ~ '
hence as c5 is ar bi t r ar y and the const ant C does not depend on 6 we get t hat
u ~ Ha( f l \ ( {xl } w {2} w {x2})) and t hat the L2- nor m of the derivative on this open
set is domi nat ed by the const ant C (recall t hat the function u is i ndeed const ant
in the intervals (Xl, 2) and (2, x2) ).
To concl ude the pr oof in the case 2E( 0, 1) we onl y need to show t hat
u+(x2) = c and u - ( x 0 = b. In fact, this implies t hat u s ~##, while (2.2) gives the
desired estimate on t he mi ni mum limit of the sequence g~(2)~ (,jj..
Suppose by cont r adi ct i on t hat u+( x2) = c + h with h > 0. By using the
regul ari t y of u pr oved above we have t hat t here exists a const ant fl > 0 such t hat
u ( x ) > c + ~ h in (x2, x 2 + fl) and u ( x ) < c in ( x 2 - f l , x2). As the uj converge to u
al most everywhere, for each fixed k ~ N we can find an index Jk and t wo poi nt s
Z'k, Z;, ~ ( X 2 - - l / k , x 2 + 1 / k ) with z~, < z{ such t hat Jk - ~ + 0 % c < Uj~(Z'k) < C + 89
and uj k(z' i ) > c + 3 h . In this si t uat i on we have the estimate
~j, j ~ > - - 2 [D(~ouj ~)l dx + (u~) 2dx.
Asympt ot i c Devel opment by F-Convergence 119
T h e p a r t b e t we e n t he s q u a r e b r a c k e t s is p o s i t i v e b e c a u s e uj~(z'k) > c, wh i l e t h e l a s t
i n t e g r a l i s g r e a t e r t h a n o r e q u a l t o
, [ h/3 ,~2 hZk
1~"
As t h e l a s t e x p r e s s i o n g o e s t o + oe as k ~ + oe we ge t a c o n t r a d i c t i o n wi t h t h e
a s s u mp t i o n t h a t
l i r a i n f ~- ~) ( uj ) < + ~ .
j ~+~
Wi t h a s i mi l a r p r o o f we c a n s h o w t h a t i f 2 = 0 o r 2 = 1 we n e c e s s a r i l y h a v e
l i r a i n f ~( 2)
j---~ cO
a n d t h e f i r s t s t e p o f t h e p r o o f is c o mp l e t e .
Second step. We e x h i b i t a s e q u e n c e {us} s u c h t h a t uj ~ u i n L I ( ~ ) a n d
lim ~ E ( 2 ) . '~
j-~ + oo
We c a n a s s u me ff(2)(u) < + oc, b e c a u s e o t h e r wi s e t h e c o n s t r u c t i o n i s t r i vi a l , a n d
s o we h a v e u s H l ( f ~ \ { 2 } ) a n d u(O) = u o, u(1) = ul , u - ( 2 ) = b, u+( 2) = c.
We de f i ne
f
u(x) i f 0 ~ x < 2 ,
u j ( x ) = t l j ( X - 2) i f 2 _ < x _ < 2 + ~ j ,
l ( u 1 - 2 - ~ j 1 - 2 ( x - 2 - r i f 2 + ~ y < x < _ l .
T h e f u n c t i o n s r/j a n d t h e s e q u e n c e Ca a r e d e f i n e d i n t h e f o l l o wi n g l e mma .
L e m m a 2. 4. I t is possible to f i n d increasing solutions o f the f ol l owi ng sequence o f
differential problems:
~ / 1 ) = q r + @ j ,
(P~) lt/~(0) = b, r/j(~j) = c,
with ~j --, O. The sequence 6 i abov e is defined by
fi~ = me a s ( { s z [ b, c] " (p~12(s) <_ z)/2}).
Obviously, g)) -~ O.
P r o o f C o n s i d e r a s o l u t i o n o f t he C a u c h y p r o b l e m c o r r e s p o n d i n g t o ( P j) wi t h o u t
t h e c o n d i t i o n t/j(~j) = c. T h e s o l u t i o n o f t hi s p r o b l e m i s g l o b a l l y d e f i n e d a n d i s
s t r i c t l y i n c r e a s i n g . We cal l ~i t h e t i me t hi s s o l u t i o n t a k e s t o r e a c h t he v a l u e c.
Mo r e p r e c i s e l y we p u t ~j = ( i nf {t ~ R: t/j(t) = c}).
120 G. Anzellotti and S. Ba]do
We have to show t hat the sequence {j goes
Ej = {se [b, c]: (pl/2(s) > e)/2}. We have
f / ~j ds
~J = e l / 2 ( S ) - ~ g j 6 j
- rE ] g j d s f [ g j d s
- - q ) l / 2 ( S ) + g~jtSj AV b , c ] k E j ~ 0 1 / 2 ( S ) - } - E j 6 j
meas([b - c ] \ E j )
<_ e)/2(c - b) +
aj
~ - F .1 /2 (C - - b) + @
As the last expression vanishes as j --+ + so the l emma is proved.
to zero as j - ~ +oo, Put
D
We are now ready to conclude the second step of the proof of Theorem 2.3,
by estimating the maxi mum limit of o~2)(uj) (of course the sequence uj converges
to u in Ll)f~)).
On (0, 1)\[2, 2 + ~j] we get
lira sup ( @2 + ~ p( uf l d x
j ~ + o o . ) ( 0 , I ) \ [ s + { A L ~;j _1
_< lim sup ~ (u)) 2 d x + lira sup @2 d x = u') z dx.
j ~ + a o , j 0 j ~ + o o d 2 + { i d ( O , 1)\{s
Finally, on (2, s + {fl we have the estimate
g~ ~+r 1 1 2 f [ t
lim s u p | | /(u)) 2 + 25 ~o(uj) d x - - @/2(s) d s
j + + m k J ~ L g j g j
= l i m s u p ffJ[(tl'j)z + l~ q~(tlj) - 2 q'jq~l/2(~lj)] d x
j--+ + ae g j
= l i ms u p (' e~[, qr 2 - - d x
s
= lim sup a~ d x = O.
j - + + oo
By addi ng these estimates we get the claim, and the proof is complete. []
In the more general situation fl c R" with n > 1, the simplest generalization
of the limit functional would be
f f n ,V u ] 2 d x i f u E r
k + oo otherwise in ~1,
Asymptotic Development by F-Convergence 121
wh e r e
~U = {u ~ ~gl c~ BV(f~): ( D u - DUle,A, ) ~ L2 ( ~) ; (I)(u) = (I)(g) ~ " - 1-a.e.
wh e n e v e r 9 e [ a, b [ u ] c , d ] ; u - ( x ) = b a n d
u + ( x ) = c ~ " - 1-a.e. o n 8A, }.
He r e u - a n d u d e n o t e t h e t r a c e s o f u o n t h e t wo s i de s o f 0*u. We r e m a r k t h a t
i n t h e c a s e wh e r e 0 A, i s c l o s e d we s i mp l y h a v e
= {u ~ J#~ c~ H~( f ~\ 0A=) : O(u) = q~(g) ~ " - ~-a.e. wh e n e v e r g e Ea, b [ w ] c , d ] ;
u - ( x ) = b a n d u + ( x ) = c ~ " - 1-a.e. o n ~A, }.
T h e f o l l o wi n g e x a mp l e s h o ws t h a t t hi s f u n c t i o n a l Z (=) c a n n o t b e t h e F - l i mi t
o f t h e s e q u e n c e Z ~ ~).
E x a m p l e 2. 1. L e t ~ = { x ~ R 2 : l x ] < 1} a n d g ( y ) = c t o n 0f~, ~ ( b , c ) . O n t h e
i n t e g r a n d cp we m a k e t h e f u r t h e r a s s u mp t i o n t h a t ~ q) - 1/ 2( s ) ds < + oo ( t hi s is
e s s e n t i a l l y a n a s s u m p t i o n o f t h e o r d e r o f z e r o o f (o a t c).
I f we c h o o s e e ~ (b, e) i n s u c h a wa y t h a t O(c) - O(~) < ~( c 0, we h a v e
~,(1) = {u e Ll ( f ~) : u ( x ) e Ec, d ] a. e. i n f~}.
m 1 = 4rc(O(c) - (I)(~)).
I n p a r t i c u l a r , t he f u n c t i o n u ~ - c i s i n ~/(1). Wh a t we d o n o w is b u i l d a s e q u e n c e
{uj} c HI(F~) wi t h u s -= g o n 8f~ a n d u s ~ uo~ i n L I ( ~ ) s u c h t h a t Y(2)(u.~ = - f l ,
wi t h fi > 0. Th i s s h o ws t h a t t h e p o s i t i v e f u n c t i o n a l ~(2) c a n n o t b e t he F - l i mi t o f
t h e s e q u e n c e Z ~ 2).
Le t us c o n s i d e r t h e s o l u t i o n o f t h e f o l l o wi n g C a u c h y p r o b l e m:
{ ~
' ( t ) = e ( 1 / 2 ) ( ~ ( t ) ) ,
( 2 . 3 )
v ( O ) = c ~ .
v ( t ) is a s t r i c t l y i n c r e a s i n g f u n c t i o n o n t he i n t e r v a l [ 0, K ] wi t h K = S~ q) - t / z ( s ) d s ,
a n d v ( K ) = c . I f h ( x ) d e n o t e s t h e d i s t a n c e o f x ~ f~ f r o m 0s we de f i ne
o t h e r wi s e i n f~.
By u s i n g t u b u l a r n e i g h b o r h o o d c o o r d i n a t e s , o b s e r v i n g t h a t
O( c ) - , ( ~ ) =
J o , , , , ~ ; / / ~ ~ c / r ,
122 G. Anzellotti and S. Baldo
and recalling (2.3) we get
j ~ C.TJ.t gsj
=2~z f f ~J {[ d v / t \ ' 2 L ~ ) + ~ 5 1 q ~
ej \ \ e j / / J
e j \ \ e j / / & k e j / )
- - 4 ~ ~[ K'jo t - e j l q o t / 2 ( v ( t - ~ ) d v ( ~ ) \ \ e j / / d t d t
= - - 4 7 C zq) i/2(V(Z')) ~ V('C) dz --= --4re zq0(v(z)) dr = - / 3.
It is heuristically convi nci ng t hat sequences of this ki nd can be const ruct ed
whenever Of] has a curved port i on, or whenever u j umps on a curved surface, and
the negative t erm we get seems to depend on some manner on an integral of
the curvat ures of the surface itself. In part i cul ar, a similar phenomenon may occur
ar ound the intersections between i nt eri or j umps and the boundar y of .Q, where
we may t hi nk t here is some concent r at ed curvat ure of c3A,. Anyhow, we have no
generally reliable conj ect ure about t he correct form of t he second F-limit in a
di mensi on hi gher t han one.
Re f e r e nc e s
[A]
[AMT]
[ABV]
[B]
[BB]
[BP]
[DM]
[DF]
IF]
[FT]
[G]
L. Ambrosio: Metric space valued functions of bounded variation. Preprint, Scuola Normate
Superiore, Pisa, 1989.
L. Ambrosio, S. Mortola, V. Tortorelli: Functionals with iinear growth defined on vector
valued BV functions. J. Math. Pures Appl., to appear.
G. Anzellotti, S. Baldo, A. Visintin: Asymptotic behaviour of the Landau-Lifshitz model of
ferromagnetism. Appl. Math. Optim., 23 (1991), 171-192.
S. Baldo: Minimal interface criterion for phase transitions in mixture of a Cahn-Hilliard
fluids. Ann. Inst. H. Poincar~ Anal. Non Lin6aire, to appear.
S. Baldo, G. Bellettini: F-convergence and numerical analysis: an application to the minimal
partition problem. Ricerche Mat., to appear.
G. Buttazzo, D. Percivale: Preprint.
G. Dal Maso, L. Modica: A general theory of variational integrals. "Topics in Functional
Analysis 1980-81" Qnaderno, Scuola Normale Superiore, Pisa, 1982, pp. 149-22t.
E. De Giorgi, T. Franzoni: Su un tipo di convergenza variazionale. Rend. Sere. Mat. Brescia,
3 (1979), 63-10l.
H. Federer: Geometric Measure Theory. Springer-Verlag, Berlin, 1968.
I. Fonseca, L. Tartar: The gradient theory of phase transitions for system with two potential
wells. Proc. Roy: Soc. Edinburgh, l l l A (1989), 8%102.
E. Giusti: Minimal surfaces and functions of bounded variation. Birkhauser, Boston, 1984.
Asymptotic Development by F-Convergence 123
[ Mi ]
[M23
[ MM]
[os]
is]
L. Modica: Gradi ent theory for phase transitions and the minimal interface criterion. Arch.
Rat. Mech. Anal., 98 (t987), 123 142.
L. Modica: Gradi ent theory of phase transitions with boundary contact energy. Ann. Inst.
H. Poincar6 Anal. Non Lin6aire, 4 (1987), 487-512.
L. Modica, S. Mort ol a: Un esempio di F-convergenza, Boll. Un. Mat. Ital. B (5), 14 (1977),
285-299.
N. Owen, P. Sternberg: Nonconvex variational problems with anisotropic perturbations, to
appear.
P. Sternberg: The effect of a singular pert urbat i on on nonconvex variational problems. Ph.D.
Thesis, New York University, 1986.
Accepted 25 September 1991

Potrebbero piacerti anche