HAL Id: hal-02083808
https://hal.archives-ouvertes.fr/hal-02083808
Submitted on 4 Aug 2020
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of
sci-entific research documents, whether they are
pub-lished or not. The documents may come from
teaching and research institutions in France or
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
Homogenization and localization for a 1-D eigenvalue
problem in a periodic medium with an interface
Grégoire Allaire, Yves Capdeboscq
To cite this version:
Grégoire Allaire, Yves Capdeboscq. Homogenization and localization for a 1-D eigenvalue problem
in a periodic medium with an interface. Annali di Matematica Pura ed Applicata, Springer Verlag,
2002, 181 (3), pp.247–282. �10.1007/s102310100040�. �hal-02083808�
a periodi medium with an interfa e GregoireAllaire Yves Capdebos q y Mar h 6, 2002 Abstra t
Inonespa e dimensionwe addressthehomogenizationof thespe tralproblemfor asingularly
perturbeddiusion equationinaperiodi medium. Denotingbythe period,thediusion
oeÆ- ient is s aledas
2
. Thedomainismade oftwopurelyperiodi mediaseparatedbyaninterfa e.
Dependingonthe onne tionbetweenthetwo ellspe tralequations,threedierentsituationsarise
whengoestozero. First,thereisaglobalhomogenizedproblemasinthe asewithoutinterfa e.
Se ond,thelimitismadeoftwohomogenizedproblemswithaDiri hletboundary onditiononthe
interfa e. Third,thereisanexponentiallo alizationneartheinterfa eofthersteigenfun tion.
1 Introdu tion
Thispaperisdevotedtothehomogenizationoftheeigenvalueproblemforasingularlyperturbeddiusion
equation in a periodi medium. Although this problem is of interest in higher spa e dimensions, we
restri t ourselves to the one-dimensional ase be ause of the diÆ ulty of the analysis. In parti ular,
oneof ourkey tool is the theory of Hill's ordinary dierential equation [Eas73℄ for whi h there is no
equivalentin higherdimensions. Denotingby theperiod, thediusion oeÆ ientisassumedto beof
theorder of
2
. Thus,we onsiderthefollowingmodel
8 > > < > > : 2 d dx a x; x d dx + x; x = x; x in; =0on; (1.1) where ;
isaneigenvalueandeigenfun tion(throughoutthispaper,theeigenfun tionsarenormalized
byk k L 2 ()
=1). In(1.1)the oeÆ ientsareperiodi ofperiod1withrespe ttothefastvariablex=.
Thegeneralstudy ofthehomogenizationof(1.1)isfar frombeing omplete. Whenthe oeÆ ientsare
notrapidlyos illating(i.e.,theydependontheslowvariablexbutnotonx=),itisaproblemofsingular
Centre de Mathematiques Apliquees, E ole Polyte hnique, 91128 Palaiseau, Fran e |
allaire mapx.polyte h niqu e.f r
y
e.g.,[Pia98℄). When the oeÆ ientsarepurelyperiodi fun tions (i.e., theydepend solelyon x=),the
homogenizationof(1.1)(andsimilarmodelsin higherdimension) hasbeena hievedin [AB99℄, [AC00℄,
[AM97℄. Inthe aseofsmooth oeÆ ientswitha on entrationhypothesis,partialresultshavere ently
beenobtainedin[AP02℄(againinanyspa edimension). Herewefo usonthedierent ase(ofpra ti al
aswellastheoreti al importan e)where the oeÆ ientsaredis ontinuous. Morepre isely,wefo uson
the simplest possible model in this ontext, assuming that the domain is omposed of two periodi al
mediaseparatedbyaninterfa e.
Thedomainisoftheform( l;L),wherelandLarestri tlypositive onstants,andweintrodu ethe
twosub-domains
1
=( l;0)and
2
=(0;L)separatedbyaninterfa elo atedatthepoint0. Denoting
by
i
(x) the hara teristi fun tion of
i (satisfying 1 + 2 =1and 1 2 =0in ), the oeÆ ients
areassumedtobegivenas
8 < : a(x;y)= 1 (x)a 1 (y)+ 2 (x)a 2 (y); (x;y)= 1 (x) 1 (y)+ 2 (x) 2 (y); (x;y)= 1 (x) 1 (y)+ 2 (x) 2 (y): (1.2) All fun tions a 1 ;a 2 ; 1 ; 2 ; 1 and 2
are assumed to be measurable, 1-periodi , bounded from above
andbelowbypositive onstants. Underthese assumptions,itis wellknownthatequation (1.1)admits
a ountable innitenumberof non-trivialsolutions(
m ; m ) m1
. Bystandardregularityresults,ea h
eigenfun tion m belongs to H 1 0 ()\C 0;s
(), with s > 0, and by the Krein-Rutman theorem the
rsteigenvalue is simpleand the orrespondingeigenfun tion anbe hosenpositive. Be ause of this
property, the rst eigenpair has a spe ial physi al signi ation, and we are mostly interested in its
behavior,althoughthe aseof higherleveleigenpairsisalsotreatedinsomeo asions.
Themotivationforstudyingthismodel omesfromseveralappli ations. First,it anbeseenasa
semi- lassi allimitproblemforaS hrodinger-typeequationwithperiodi potential,aswellasperiodi metri
(thisis the so- alledground-state asymptoti problem, see, e.g., [KP93℄, [Pia98℄). Se ond, it plays an
importantrole in the uniform ontrollability of thewaveequation (see, e.g., [CZ℄). Third, and this is
ourmainmotivation,itisasimplemodelfor omputingthepowerdistributioninanu learrea tor ore.
Thisis theso- alled riti alityproblem for theone-groupneutron diusionequation (formoredetails,
wereferto[AC00℄,[Cap99℄andreferen estherein). Inalltheseappli ations,theassumptionofapurely
periodi medium(i.e.,nodependen eonxofthe oeÆ ients)ismu htoostrong. Ontheotherhandthe
oeÆ ientsarenotsmoothlyvarying butexhibit jumps at materialinterfa es. This makesmodel(1.1)
withassumptions(1.2)physi allyrelevant.
Thelimitbehaviorof(1.1)ismainlygovernedbythersteigenpair(
i ;
i
)intheunit ellof
i ,i=1;2, solutionof ( d dy a i (y) d dy i + i (y) i = i i (y) i in[0;1℄; y! i
(y) 1 periodi andpositive:
(1.3)
Beforeweexplainourmainresults,letusre allwhatwasalreadyprovedin[AM97℄inthepurelyperiodi
ase,namelywhena
1 =a 2 , 1 = 2 ,and 1 = 2
. Asymptoti ally,thema ros opi trendof
isgiven
by anhomogeneouseigenvalueproblem, whereasits os illatorybehaviorisgovernedby
1 (
x
)(we all
Theorem1.1. Assuming that a 2 = a 1 , 2 = 1 , and 2 = 1 , the m eigenpair m ; m of (1.1) satises m (x)=u m (x) 1 ( x )and m = 1 + 2 m +o 2 ;
where, uptoasubsequen e, thesequen eu
m onvergesweakly inH 1 0 () tou m ,and( m ;u m )isthem th
eigenvalueandeigenve torfor the homogenizedproblem
D d 2 dx 2 u m = m u m in; u m =0 on: (1.4)
Thehomogenized oeÆ ientsaregiven by
D= Z 1 0 a 1 (y) 2 1 (y) 1+ d dy (y) dy and= Z 1 0 1 (y) 2 1 (y)dy; (1.5)
where the fun tion isthe solutionof
( d dy a 1 (y) 2 1 (y) d dy +1 =0 in[0;1℄;
pny!(y) 1 periodi :
(1.6)
Letus summarizeourresultsinthe aseofequalrsteigenvaluein the ells,
1
=
2
. Inthesequelwe
hoosetonormalizetherstperiodi eigenfun tionsasfollows
1
(0)=
2
(0)=1: (1.7)
Weintrodu easo- alleddis ontinuity onstantdenedby
=a 1 (0) d 1 dy (0) a 2 (0) d 2 dy (0): (1.8) Notethata i d i dy belongstoH 1 ( i )whi hisembeddedin C( i
)(in 1-D)andthereforeiswelldened
asthe tra eofa ontinuousfun tion at theorigin. Threedierentsituations arepossiblea ordingto
thesignof.
If=0, then thetwoperiodi media aresaid to bewell- onne ted. Inparti ular, thefun tion equal
to a i (d i )=(dx) in i
is ontinuous through the interfa e (as well as
i
be ause of the normalization
ondition(1.7)). Therefore,Theorem1.1extendseasilytothis ase,andthedis ontinuityattheinterfa e
isnot seenin thelimit. Introdu ing afun tion (x=)=
1 (x) 1 (x=)+ 2 (x) 2 (x=),the eigenpairs ( m ; m ) m1 satisfy m = 1 + 2 m +o( 2 )and m (x)=u m (x) x ; (1.9) whereu m onvergesweaklytou m ,and( m ;u m ) m1
aretheeigenpairsofthehomogenizedproblem(see
Theorem3.1andFigure3.1)
8 > > < > > : d dx 1 (x)D 1 + 2 (x)D 2 du dx =( 1 (x) 1 + 2 (x) 2 )u in; u=0 on:
vergen eresult(1.9)stillholdstrue,butthehomogenizedproblemhasanadditionalDiri hletboundary
onditionatx=0. Morepre isely,thelimithomogenizedproblemis(seeTheorem3.1 andFigure3.2)
8 > > > > > > > < > > > > > > > : D 1 d 2 dx 2 u= 1 u in 1 ; D 2 d 2 dx 2 u= 2 u in 2 ; u=0 on 1 [ 2 :
If < 0, the situation is ompletely dierent sin e the rst eigenfun tion on entrates exponentially
fastattheinterfa e. Inthislatter ase,thereisnofa torizationprin ipleasinTheorem1.1, butrather
alo alization prin ipleat the dis ontinuity(see Theorem 3.5 and Figure 3.3). The rsteigenvalue
1 onvergesto alimit 0< 1 < 1 = 2 , and0< 1 1
<Cexp( =),whereasthe rstnormalized
eigenve torsatises d dx 1 (x) 1 p d dx x L 2 () + 1 (x) 1 p ( x ) L 2 () Cexp :
Thelimitfun tion 2H
1
(R ) de reasesexponentiallyawayfromtheinterfa e,sin eitisgivenby
(x)= 1;1 (x) forx<0; 2;2 (x) forx>0; with 1 = 1 ( 1 )= 2 ( 2
), and ea h of the eigenpairs (
i (
i );
i;i
) beingthe rst eigen oupleof the
followingspe tral ellproblem
8 > > < > > : d dx a i (x) d i; i dx + i (x) i;i = i ( i ) i (x) i;i in[0;1℄; x! i; i (x)e ix 1-periodi . (1.10)
Therequiredproperties ofthe-parameterizedfamilyof spe tral ellproblems(1.10) aregiven in
se -tion2.
Wenowturn to the ase
1
6=
2
,and withno lossof generalityweassume
1 >
2
. Inthis ase too,
thespe tral ellproblems (1.10)governthelimit behaviorof(1.1). Weintrodu eapositiveparameter
0 >0,su h that 1 ( 0 )= 2
,andanotherdis ontinuity onstant(seeLemma3.9)
( 0 )=a 1 (0) d 1;0 dy (0) a 2 (0) d 2 dy (0):
Thesignofthisnewdis ontinuity onstantdeterminestheasymptoti behaviorof(1.1).
If(
0
)>0,theeigenfun tions
m
on entrateinthesub-domain
2
wheretherstperiodi eigenvalue
isthe smallest(seeTheorem 3.8 in thesimpler asewhen 0,and Theorem 3.11when(
0 )>0).
Morepre isely, thelimitof m
vanishesin thesub-domain
1
. Introdu ing thefa torization
m (x)= u m (x) 2 (x=)in 2
,thehomogenizedproblem forthelimitofu
m
issimply(seeFigure3.4)
8 > < > : D 2 d 2 dx 2 u= 2 u in 2 ; u=0 on 2 : The ase( 0
)=0 orrespondstothelimitbetweenlo alizationat theinterfa eand on entrationin
2
. Thelimitoftheeigenfun tion
m
stillvanishesin
1
,butinthehomogenizedproblemtheDiri hlet
boundary onditionatx=0isrepla edbyaNeumannboundary ondition(seeTheorem3.12)
8 > > > < > > > : D 2 d 2 dx 2 u= 2 u in 2 ; u(L)=0and du dx (0)=0: Finally, when ( 0
)< 0, alo alization phenomenon appears, and therst eigenfun tion on entrates
exponentially fast at the interfa e. The result is then similar to the oneobtained when
1
=
2 and
<0(seeTheorem3.10).
Ourmainresultsarestatedinse tion3when
1
isequalornotto
2
. Previously,inse tion2wegivea
fewte hni alresultsonthespe tral ellproblemsthatare ru ialnotonlyfortheproof,butalsoforthe
statementofourmainresults. Se tion4 ontainstheproofswhenthedis ontinuity onstantispositive,
0, while se tion 5 fo us on the lo alization phenomena, namely < 0 or(
0
) < 0. Se tion 6
ontainstheproofsinthespe ialsituationwhen<0but nolo alization o urs((
0 )0),asit an happen when 1 isnotequalto 2
. Se tion 7 ontainsthe proofofa ru ialte hni alresultaboutthe
Hillequationin onedimension.
2 Cell problems
Inordertostatepre iselyour onvergen eresults,theknowledgeofthespe tral ellproblem(1.3)isnot
enough. Asin[Cap98℄, weneedtointrodu eaparameterizedfamilyofspe tral ellproblems. Theyare
reminis entoftheso- alledBlo hwavede omposition(seee.g. [CPV95℄, [RS78℄),but theyinvolvereal
exponentialsinstead of omplexones. All the resultsin this se tionare proved under the assumption
thatthe periodi oeÆ ientsa
i
;
i ;
i
are positive, bounded, measurablefun tions, ex eptProposition
2.2whi h asksformoresmoothnessorpie ewise onstant oeÆ ients.
Lemma2.1. For ea h2R thereexistsauniquerst eigen ouple (
i; ; i ()), of theproblem 8 < : d dx a i (x) d i; dx + i (x) i; = i () i (x) i; in [0;1℄; x! i; (x)e x
1 periodi andpositive ;
(2.1)
whi h isnormalizedby
i;
i 2 i (0) i ()C 2 ;
where C and arepositive onstants,independent of .
A further property of the rst eigenfun tion
i;
; is given in the next Proposition. Its proof is quite
deli ateandrelies onpurely 1-Darguments(wepostponeittose tion7). Wegivetwodierentproofs:
rstinthe aseofC
2
oeÆ ients,whi hallowstoperformaLiouvilletransformationandtouse lassi al
resultsonthe 1-DHillequation,se ond inthe aseof pie ewise onstant oeÆ ients, whi h permitsto
doexpli it omputations.
Proposition2.2. Assumingthatthe oeÆ ients areC
2
or pie ewise onstant, forea h2R the rst
eigenve tor i;
ofproblem(2.1) withthe normalization
i; (0)=1satises lim ! 1 d i; dx (0)= 1and lim !+1 d i; dx (0)=+1:
Proof ofLemma2.1. Byintrodu ingthe hangeofvariable
i; (x)= i; (x)e x ; equation(2.1)isequivalentto 8 < : d dx a i d i; dx d dx (a i i; )+a i d i; dx + i a i 2 i; = i () i i; in[0;1℄; x! i;
(x)1 periodi andpositive;
(2.3)
withthesamenormalization ondition
i;
(0)=1:
Theexisten eofauniquerstpositiveeigen oupleforproblem(2.3)isknown,seee.g. [GT83,Theorem
8.38℄,andwehave i; 2H 1 # ([0;1℄)\C 0;s
([0;1℄),withs>0. Inparti ular,thisimplythatC>
i; (x)>
>0in [0;1℄. Itis provedin [Cap98℄ thatthefun tion !
i
()issmooth,stri tly on aveonallR,
andrea hesitsmaximumat=0.
Toobtainthegrowth onditionon
i
(),weperformthefollowing hangeofunknown
u (x)= i; (x) i;0 (x)
whi hisli itbyvirtueofProposition4.1. Then,u
is solutionofthefollowingproblem
8 < : d dx b(x) du dx =()s(x)u in [0;1℄; x!u (x)e x 1 periodi ; (2.4) withb(x)=a i (x) 2 i;0 (x), s(x)= i (x) 2 i;0 (x),and()= i () i
(0). These oeÆ ientsarebounded,
onstantM>m>0. Forea h 2R the rsteigenvalue ()of problem (2.4)satises m M 2 () M m 2 :
Proof. Wealreadyknowthat()<0forall6=0. We anassumethat>0sin e hangingthesignof
in (2.4)is equivalentto onsideritsadjointequationwhi hhasthesamersteigenvalue. Be ausewe
areworkingin onespa edimension,(2.4) anbewritten asasystemofordinarydierentialequations.
Namely,denotingby0 thex-derivation,
Y 0 (x)=A(x)Y(x)andA= 2 4 0 b 1 ()s 0 3 5 and Y = 0 Y 1 =u Y 2 =bu 0 1 A : (2.5)
By enfor ing the normalization u
(0) = Y
1
(0) = 1, the Krein-Rutman Theorem implies that Y
1 is
positive,andthus Y
2 isin reasing. Sin eY 2 (n)=e n Y 2
(0),and>0,thisimpliesthat Y
2
(0)>0,and
thusY
2
(x)>0forx0. Thisin turngives,bytherstequation, thatY
1 isin reasingthus Y 1 1for x0. Be auseY 1 andY 2
arepositivefun tionsonR
+ ; we anwrite A Y Y 0 A + Y withA + = 2 4 0 m 1 ()M 0 3 5 ; andA = 2 4 0 M 1 ()m 0 3 5 :
Sin ethematri esA
+
andA have onstant oeÆ ients,itisstraightforwardtoobtainthesolutionsof
theinitial valueproblems
Z 0 = (A ) T Z ; Z(0)=Z 0 ; andX 0 = (A + ) T X; X(0)=X 0 :
Inparti ular,the hoi e Z
0 =X 0 = 1;( ()mM) 1=2
leadstothepositivesolutions
Z(x)=Z 0 exp x r ()m M ! andX(x)=X 0 exp x r ()M m ! :
We an ompute that (Y Z)
0 = Y 0 Z +Y Z 0 = ( Y 0
A Y)Z 0 sin e Z is positive. Thus
Y Z Y(0)Z(0)forallx0,and hoosingx=n2N leadsto
Y(n)Z(n)=exp n r ( ())m M !! Y(0)Z(0)Y(0)Z(0); and therefore q ( ())m M . Similarly, we have (Y X) 0 =(Y 0 A + Y)X 0sin e X is positive,
whi hgivesinturn foralln,
Y(n)X(n)=exp n r ( ())M m !! Y(0)X(0)Y(0)X(0); andtherefore q ( ())M m .
Itisprovedin[Cap02℄thatin general!()isastri tly on avefun tion,i.e.,thatonanybounded
subset K R
N
(with N the spa e dimension) the Hessian matrix H =
2 ij 1i;jN is negative
denite andHxx C(K)xx with C(K)>0. The fun tion () a hievesits maximumin 0 and
lim jj!1
()= 1.
3 Main results
Inthespirit of the method of proofof Theorem 1.1 (see [AM97℄), weintrodu ein (1.1)the hange of
unknown u (x)= (x) x; x ;
withafun tion (x;y)denedby
(x;y)= 1 (x) 1 (y)+ 2 (x) 2 (y); (3.1) where( 1 ; 1 )and( 2 ; 2
)arethersteigen ouplesinea hperiodi ellof(1.3). Byournormalization
ondition(1.7),thefun tion (x;x=)is ontinuousattheinterfa ex=0. Onthe ontrary,thefun tion
a(x;x=)(d (x;x=))=(dx)isnotne essarily ontinuousanditsjump attheinterfa eismeasuredbythe
dis ontinuity onstantintrodu edin (1.8).
Therstresult on ernsthespe ial asewhentherst elleigenvaluesof(1.3)areequal,
1
=
2
, and
thedis ontinuity onstantisnon-negative,0. Under these assumptions,weobtainageneralization
ofTheorem1.1. Theorem3.1. Let m and m
be the m-th eigenvalue andnormalized eigenve tors of (1.1). Assume
thatthe dis ontinuity onstantdenedin(1.8)isnon negative0, andthat
1 = 2 . Then m (x)=u m (x) x; x and m = 1 + 2 m +o 2 ; uptoasub-sequen e, u m onvergesweakly in H 1 0 () towards u m ,and( m ;u m )isthe m-th eigen ouple
ofthe homogenizedproblem, whi h,if =0, is
8 > > < > > : d dx 1 (x)D 1 + 2 (x)D 2 du dx =( 1 (x) 1 + 2 (x) 2 )u in ; u=0 on ; (3.2) and,if >0, is 8 > > > > > > > < > > > > > > > : D 1 d 2 dx 2 u= 1 u in 1 ; D 2 d 2 dx 2 u= 2 u in 2 ; u=0 on 1 [ 2 : (3.3)
−20
−10
0
10
20
0
0.2
0.4
0.6
0.8
1
Figure3.1: Firsteigenfun tionforproblem(1.1)inthe aseoftwowell- onne tedmedia,i.e.,=0.
−50
−40
−30
−20
−10
0
10
20
30
40
50
0
0.2
0.4
0.6
0.8
1
Figure3.2: Firsteigenfun tionforproblem(1.1)inthe aseofnonwell- onne tedmediawithapositive
dis ontinuity onstant>0.
AsanillustrationofTheorem3.1, wepresentsomedire t omputationsof thersteigenfun tion
1 of
problem (1.1). The ase
1
=
2
and =0is shownon Figure 3.1. (The domain is omposed of an
homogeneousmediumontheleftandanheterogeneousoneontheright). The ase
1
=
2
and>0is
shownonFigure3.2(thedomainis omposedoftwoheterogeneousmediawiththesame ell oeÆ ients
but witha onstantphaseshiftbetween therightand theleft). Thedata usedfor the omputation is
presentedinRemark3.7.
Remark3.2. Of ourse, sin e the homogenized oeÆ ients are onstant in ea h sub-domain we an
omputeexpli itlytheeigenvaluesofthehomogenizedproblemsinTheorem3.1.
Remark3.3. There is asimple suÆ ient onditionfor havingwell- onne tedmedia, i.e., =0. If all
oeÆ ientssatisfy a entral symmetry ondition, i.e., are symmetri with respe t to the enter of the
unit ell [0;1℄, then it is easy to he kthat
i
satises a Neumannboundary onditionat x =0 and
x =1, and therefore = 0. A tually, Theorem 3.1 wasalready proved by Malige [Mal96℄ under this
assumption. Thesymmetrywasusedforthe onstru tionoftheexampleshowninFigure3.1: in
2
,the
periodi oeÆ ientsarepie ewise onstanton(0:3;0:7)and(0:1;0:3)[(0:7;1:0); in
1 , a 1 1, 1 1 and 1 = 2 .
1 2
Inother words,therearetwode oupledhomogenizedproblems. Therefore,therealwaysexisttwo
non-negativeeigenfun tionswithdisjointsupports,u
1 (x)=sin( l x) 1
(x) orrespondingtotheeigenvalue
1 = 2 D 1 1l 2 and u 2 (x) = sin( L x) 2
(x) orresponding to the eigenvalue
2 = 2 D 2 2L 2 . If the rst
eigenvaluesin ea h sub-domain are distin t, e.g., L
2 2 D 1 > l 2 1 D 2
, the rstfa torized eigenfun tion
u
1
will tend to u
2
, i.e., will on entrate in the sub-domain that has the smallest rst eigenvalue and
onvergetozeroin theotherone. Intheother asewhere thersteigenvaluesin
1
and
2
areequal,
thersteigen-subspa eisofdimension2,spanbyu
1
andu
2
andtheuniquenessofthelimitof
1 islost
(onFigure3.2thelimitseemstobealinear ombinationofthersteigenfun tionsonea hsub-domain).
Ourse ondresult ompletesthe ase
1
=
2
whenthedis ontinuity onstantisnegative,<0. Under
theseassumptions,weobtainalo alizationphenomena.
Theorem3.5. Let( 1 ; 1
)betherstnormalizedeigen oupleof(1.1). Assumethat
1
=
2
and<0.
Then,thereexistsaunique
1
>0and auniquepositive (x)2H
1 (R ) su hthat 0 1 1 Cexp and d dx 1 (x) 1 p d dx x L 2 () + 1 (x) 1 p ( x ) L 2 () Cexp ;
where C and are positive onstant,independentof . The limiteigenvaluesatises
1 < 1 = 2 ,and
thelimit eigenfun tionisdenedby
(x)= 1; 1 (x) for x<0; 2; 2 (x) for x>0; with 1 >0and 2 <0and( 1 ; i; i
)isthe rsteigen ouple ofthe ellproblem (2.1), i.e.,
8 > > < > > : d dx a i (x) d i;i dx + i (x) i;i = 1 i (x) i;i in[0;1℄; x! i; i (x)e ix 1 periodi :
Remark3.6. Theorem 3.5 is illustrated by Figure 3.3: the rst eigenve torof system (1.1) onverges
exponentially fasttowardsa lo alized eigenfun tionnear the interfa ebetweenthe twodomains.
Fur-thermore, the orresponding eigenvalueis smallerthan
1
=
2
, whi h is the limitobtainedin all the
other ases. In ontrast with Theorem 3.1, no fa torization, or limit homogenized problem appear in
thewordingofTheorem3.5. Thelimiteigenfun tion ontainsboththeperiodi alos illationsandthe
ma ros opi trend.
Remark3.7. The omputationsshownonFigure 3.2andFigure3.3 wereperformedwiththesametwo
media,but theirpositionsareswit hed withrespe ttotheinterfa ewhenpassingfromone asetothe
other. Wetake l=L=1with100periodi ity ells,whi hyields=0:02. Allthemoretheperiodi ell
oeÆ ientsforthetwomediaarethesameuptoaphaseshiftintheunit ell. Morepre isely,inFigure3.2
the oeÆ ients are a
1 (y) = a(y), a 2 (y) = a(y+ ), 1 (y) = (y), 2 (y) = (y+ ), 1 (y) = (y), 2
(y) = (y+ ), while in Figure 3.3 they are a
1 (y) = a(y+ ), a 2 (y) = a(y), 1 (y) = (y+ ), 2 (y)=(y), 1 (y)=(y+ ), 2
(y)=(y), where =0:6 isa onstantphaseshift, anda;and
−50
−40
−30
−20
−10
0
10
20
30
40
50
0
0.2
0.4
0.6
0.8
1
Figure3.3: Firsteigenfun tionforproblem(1.1)inthe aseofnonwell- onne tedmediawithanegative
dis ontinuity onstant<0.
spe iedorderasfollows
(a;;)= 8 > > < > > : (a I ; I ; I ) if0<y<0:1 (a II ; II ; II ) if0:1<y<0:5 (a III ; III ; III ) if0:5<y<0:8 (a I ; I ; I ) if0:8<y<1 with ConstituentI a I =0:9666 I =2:1080 I =2:8283 ConstituentII a II =2:0086 II =2:3878 II =2:9451 ConstituentIII a III =2:0444 III =2:9945 III =1:1493
Notethat,by onstru tion,
1
=
2
1:3863. Theshapeofthersteigenve tor
1 onFigure3.2(with eigenvalue 1
1:3899), orrespondstowhatisannoun edbyTheorem3.1: asymptoti ally,bothmedia
tendtoseparatewhen>0. Therefore,bysymmetryFigure3.3 orrespondstoasituationwhere<0:
thersteigenve tor on entratesexponentiallyat theinterfa ebetweenthetwomedia. Thenumeri al
al ulation onrms that the orresponding eigenvalue (
1
1:3720) is below that of the periodi ity
ell. This phenomenonis explainedby Lemma 5.4whi h givesane essaryandsuÆ ient onditionfor
theexisten eofalo alizedeigensolution.
Wenowturntothegeneral ase
1
6=
2
. Inthesequel,weshallassume,withoutlossofgenerality,that
1
> 2
:
Ifthedis ontinuity onstantis non-negative,i.e., 0,the eigenfun tions on entrateasymptoti ally
inthesub-domain
2
wheretherstperiodi eigenvalueisthesmallest.
Theorem3.8. Let m and m
bethe m-theigenvalue andnormalizedeigenfun tion of (1.1). Assume
that0and 1 > 2 . Then, m (x)=u m (x) x; x and m = 2 + 2 m +o 2 ;
−20
−10
0
10
20
0
0.2
0.4
0.6
0.8
1
Figure3.4: Firsteigenfun tionof(1.1)inthe aseoftwomediawith
1 > 2 and=0. where, upto asubsequen e, u m onverges weakly inH 1 0 () tou m , with u m =0 in 1 and( m ;u m ) is
them-th eigenpair of the following homogenizedproblem
D 2 d 2 dx 2 u m = m 2 u m in 2 ; u m =0 on 2 ; (3.4)
andthehomogenized oeÆ ientsarestill given by(1.5).
Figure3.4 illustratesTheorem 3.8. It displaysthersteigenfun tion
1
in the aseoftwomedia with
symmetri periodi stru tures (so that =0), with
1
'1:58 and
2
'0:43and 20 periodi ellson
ea hsideoftheinterfa e.
When
1 >
2
, a lo alization phenomena an also o ur. Let us rst remark that, as an obvious
onsequen eofLemma2.1, wehavethefollowingresult.
Lemma3.9. For all
1 >
2
thereexistsaunique
0 >0su hthat 1 ( 0 )= 2 .
Indeed, Lemma 3.9 is obvious by remarking that
1
(), dened in Lemma 2.1, is a on ave fun tion
with quadrati growth at innity and rea hing its maximum at = 0,
1 (0) = 1 > 2 . The rst eigenve tors orrespondingto 2 and 1 ( 0 )aredenoted by 2 and 1; 0
. Theyare ontinuousat the
interfa e,i.e.,
2
(0)=
1; 0
(0)=1,andweintrodu eanewdis ontinuity onstantthatwill hara terize
thelo alization phenomena
( 0 )=a 1 (0) d 1;0 dy (0) a 2 (0) d 2 dy (0): Theorem3.10. Let( 1 ; 1
)betherstnormalizedeigen oupleof(1.1). Assumethat(
0
)<0. Then,
thereexistsaunique
1
>0andauniquepositive 2H
1 (R ) su hthat 0 1 1 Cexp and d dx 1 (x) 1 p d dx x L 2 () + 1 (x) 1 p ( x ) L 2 () Cexp ; (3.5)
1 2 1
thelimit eigenfun tionisdenedby
(x)= 1;1 (x) for x<0; 2;2 (x) for x>0; with 1 >0and 2 <0and( 1 ; i;i
)isthe rsteigen ouple ofthe ellproblem (2.1), i.e.,
8 > > < > > : d dx a i (x) d i;i dx + i (x) i; i = 1 i (x) i; i in[0;1℄; x! i;i (x)e i x 1 periodi :
Finally, in the remaining ase < 0 and (
0
) 0, there is no lo alization, and the eigenfun tions
still on entrateasymptoti allyinthesub-domain
2
wheretherstperiodi eigenvalueisthesmallest.
When (
0
) > 0, the limit problem has Diri hlet boundary onditions. When (
0
) = 0, the limit
problemhasaNeumannboundary onditionat theinterfa e.
Theorem3.11. Let m and m
bethem-th eigenvalueandnormalizedeigenfun tionof(1.1). Assume
that 1 > 2 ,<0and( 0 )>0. Then, m = 2 + 2 m +o 2 ; m (x)!0in L 2 ( 1 ) and m (x)=u m (x) 2 x ; where, uptoa subsequen e, u m onverges weakly in H 1 ( 2 )tou m ,and ( m ;u m )is the m-th eigenpair
ofthe followinghomogenized problem
D 2 d 2 dx 2 u m = m 2 u m in 2 ; u m =0 on 2 ; (3.6)
andthehomogenized oeÆ ientsarestill given by(1.5).
Theorem3.12. Let m and m
bethem-th eigenvalueandnormalizedeigenfun tion of(1.1). Assume
that 1 > 2 ,<0and( 0 )=0. Then, m = 2 + 2 m +o 2 ; m (x)!0in L 2 ( 1 ) and m (x)=u m (x) 2 x ; where, uptoa subsequen e, u m onverges weakly in H 1 ( 2 )tou m ,and ( m ;u m )is the m-th eigenpair
ofthe followinghomogenized problem
D 2 d 2 dx 2 u m = m 2 u m in 2 ; u m (L)=0; and du m dx (0)=0; (3.7)
andthehomogenized oeÆ ientsarestill given by(1.5).
Remark3.13. NotethatthehomogenizedproblemsofTheorems3.11and3.12( orrespondingto(
0 )>
0 and (
0
) = 0, respe tively) are similar ex ept the boundary ondition at x = 0. Then a simple
omputationshowsthat thersthomogenizedeigenvalue
1
is fourtimessmallerwhen(
0
)=0than
when(
0 )>0.
fa torizedeigensolutionsu
m
havebounded gradientsin all of , but simplywithin
2
. It is therefore
diÆ ult(at least for us)to study the possible o urren eof boundary layersin
1
. In thelimit ase
of Theorem 3.12, be ause of the homogenized Neumann boundary ondition at x = 0, we expe t a
nontrivialboundarylayerin
1 .
Remark3.15. Thegeneralizationoftheresultsofthisse tiontohigherspa edimensionsisnotobvious
foratleasttworeasons. First,Theorems3.5 and3.10reliesonProposition2.2whi hisprovedonlyin
onedimension (by usingo.d.e. te hniques). Se ond, evenTheorems 3.1and 3.8 (whi h donotdepend
onProposition 2.2) arenotstraightforwardin higher dimensionsbe ause we annot assumeaperfe t
transmission ondition(1.7) at theinterfa e. Of ourse, ifit happens by han e that, for adimension
N>1,wehave 1;0 (0;y 0 )= 2 (0;y 0 ) (3.8)
foralmost everyy
0 2[0;1℄ N 1 ,and 0(y 0 )=a 1 (0;y 0 ) d 1;0 dy (0;y 0 ) a 2 (0;y 0 ) d 2 dy (0;y 0 )M <+1; (3.9)
thenTheorems3.1and3.8extendeasilysin eattheinterfa etheproblemisessentiallyone-dimensional
(see [Cap99℄). Of ourse, these onditions are very stri t and almost never satised in pra ti e. In
general,webelievethatboundarylayersat theinterfa emustbetakenintoa ount.
Remark3.16. Throughoutthispaperweassumethat, afterres alingby,theperiodi ityonbothsides
of the interfa e is exa tly one. The fa t that the period is the samein
1
and
2
is not important,
andthisispurelyby onvenien ethatwemadethis hoi e. All ourresultsapplyifthetwoperiods are
dierent,providedthat thedis ontinuity onstantsand(
0
)areproperlydened.
4 Proofs in the ase 0
Inorderto proveTheorem 3.1and 3.8,werstneedto justifythefa torization
(x)=u (x) x; x .
ThisisthegoalofthenextPropositionwhi hisageneralizationofapreviousresultof[AM97℄(seealso
[AC00℄).
Proposition4.1. Let (x;y)bethe fun tion denedby (3.1). Then, the linear operator T denedby
T :H 1 0 () ! H 1 0 () (x) ! (x) x; x
isbounded,invertible andbi ontinuous.
Proof. Thankstothenormalization ondition(1.7)thefun tion (x;x=)is ontinuousonR. Byvirtue
ofLemma2.1weknowthatthereexisttwopositive onstantsC> >0su hthatC
1 (y);
2
(y)
forally2[0;1℄,andtheseboundsalsoholdsfor . Therefore,forall2H
1 0 (),ifwedeneu=T() , wehave C 1 kk L 2 () kuk L 2 () 1 kk L 2 () ; (4.1)
andT isanhomeomorphism onL (). Ontheotherhand, Z a d dx d dx = Z 1 a 1 2 1 du dx du dx + Z 2 a 2 2 2 du dx du dx + Z 1 a 1 d 1 dx d(u 2 1 ) dx + Z 2 a 2 d 2 dx d(u 2 2 ) dx : (4.2) Equation(1.3)dening 1 (x=)tested againstu 2 (x) 1 (x=)writes Z 1 a 1 d 1 dx d(u 2 1 ) dx 1 a 1 (0) d 1 dy (0)u 2 (0)= 1 2 1 Z 1 1 2 1 u 2 Z 1 1 2 1 u 2 ; (4.3)
andsimilarlywehave
Z 2 a 2 d 2 dx d(u 2 2 ) dx + 1 a 2 (0) d 2 dy (0)u 2 (0)= 1 2 2 Z 2 2 2 2 u 2 Z 2 2 2 2 u 2 : (4.4)
Whenwerepla e(4.3)and(4.4)into (4.2)weobtain
Z a x; x d dx d dx dx+ 1 2 Z x; x 2 dx = 2 X i=1 Z i a i x 2 i x du dx du dx dx + 1 2 2 X i=1 Z i i i x 2 i x u 2 dx + 1 u 2 (0): (4.5)
Where is the dis ontinuity onstant given by (1.8). If 0, all the left hand side terms are
non-negativein(4.5). Sin e a 1 ;a 2 ; 1 and 2
arebounded belowbypositive onstants,we andedu ethat
du dx 2 L 2 () +kuk 2 L 2 () C() d dx 2 L 2 () +kk 2 L 2 () ! : (4.6) Conversely,wehave 0u 2 (0)C Z du dx 2 dx; (4.7)
thereforewealsoobtainfrom(4.5)that
d dx 2 L 2 () +kk 2 L 2 () C() du dx 2 L 2 () +kuk 2 L 2 () ! (4.8)
andthis on ludestheproofofthepropositionfor0. If0,notethatthankstothenormalization
ondition(1.7),u
2
(0)=
2
(0). Consequently,identity(4.5)isalso
Z a x; x d dx 2 dx+ 1 2 Z x; x 2 dx 1 2 (0) = 2 X i=1 Z i a i x 2 i x du dx 2 dx + 1 2 2 X i=1 Z i i i x 2 i x u 2 dx:
Ifwepro eedto the hangeofunknownu
=T(
),problem(1.1)istransformedintoaneweigenvalue
problem,wherethesingularperturbationinfrontofthedivergen etermhasdisappeared. Proposition4.2
givestheformofthisnewproblemafter somesimplealgebra.
Proposition4.2. Introdu ing u (x) = (x)= x; x
, (1.1) is equivalent to the following eigenvalue
problem 8 > > > > > > > < > > > > > > > : d dx D x; x du dx + 1 2 2 1 (x)B x; x u + 1 u (0)Æ(x)= B x; x u u =0on in (4.9)
where Æ(x) istheDira fun tion,the (positive)diusion oeÆ ient isdenedby
D x; x = 1 (x) 2 1 x a 1 x + 2 (x) 2 2 x a 2 x ;
the(positive) oeÆ ient B by
B x; x = 1 (x) 2 1 x 1 x + 2 (x) 2 2 x 2 x ;
andthenew eigenvalueby
= 2 2 :
Remark4.3. WeprovedProposition4.1regardlessofthesignof,thereforeProposition4.2isalsovalid
when<0. Weshallusethisequivalentformof(1.1)in Se tion6.
Following astrategy already used in [AC00℄, [AC98℄, the asymptoti study of the eigenvalue problem
(4.9)relies onthedetailedhomogenization,astendtozero,ofthefollowingproblem
8 > > > > > > < > > > > > > : d dx D x; x du dx + 1 2 2 1 (x)B x; x u + 1 u (0)Æ(x)=f u =0on; in (4.10)
witharighthandsidef
whi hisaboundedsequen eofL
2
(),weakly onvergingtoalimitf 2L
2 ().
Werstobtainaprioriestimates
Proposition4.4. If0,the solution u
of equation(4.10) satises ku k H 1 0 () + 1 2 ku k L 2 (1) + r ku (0)kCkf k L 2 () (4.11)
where C is a onstant independent of . Therefore, up toasubsequen e, u
onverges weakly toa limit
uinH
1
0
if 1
> 2
,the limituvanishes in
1
andthusbelongstoH
0 ( 2 ), if 1 = 2
and > 0, the limit satises u(0) = 0 and thus an be written u = u
1 +u 2 with u 1 2H 1 0 ( 1 )andu 2 2H 1 0 ( 2 ).
Proof. Ifwetest variationallyequation(4.10)dening u
againstu ,weobtain Z D x; x du dx 2 dx+ 1 2 2 Z 1 B x; x (u ) 2 dx+ 1 (u ) 2 (0)= Z f u dx:
Sin eD andBareboundedbelowbyapositive onstant,andsin eweassumethat
1 2 and0, weobtain du dx 2 L 2 () + 1 2 2 ku k 2 L 2 ( 1 ) + ku (0)k 2 Ckf k L 2 () ku k L 2 () ;
whi hyieldsthedesiredresultthankstoPoin areinequality.
Lemma4.5. Let S
bethe operator denedby
S :L 2 ()! L 2 () f ! u unique solutioninH 1 0 () ofequation (4.10) withr.h.s. f: (4.12) Forallxed>0, S
isalinear ompa t operator inL
2 ().
This resultis a onsequen e of the apriori estimate (4.11) and of the ompa t in lusion of H
1
0
() in
L 2
(). Weshallshowthefollowingresult
Proposition4.6. Letf
beaweakly onvergingsequen etoalimitfinL
2 (). Thesequen eu =S (f ) weakly onvergesinH 1 0 () towardsu 0 dened byu 0 =S(f). 1. If =0and 1 = 2
thenS isthe following ompa t operator
S: L 2 () !L 2 () f !uunique solutionof d dx 1 (x)D 1 + 2 (x)D 2 d dx u(x) =f in ; u=0 on : whereD 1 andD 2 aregiven by(1.5). 2. If 0and 1 > 2
thenS isthe following ompa t operator
S: L 2 () !L 2 () f !uuniquesolutionof D 2 d 2 dx 2 u(x)=f in 2 ; u=0 on n 2 :
1 2 S: L 2 () !L 2 () f !uuniquesolutionof 8 < : D 1 d 2 dx 2 u(x)=f in 1 ; D 2 d 2 dx 2 u(x)=f in 2 ; u=0 on 2 [ 2 :
Proof. Theproofisquitestandardinhomogenizationtheory. Forexample,usingthenotionoftwo-s ale
onvergen e (see [All92℄, [Ngu89℄) it is an easyexer ise that we safely leaveto thereader (the details
an be found in [Cap99℄ if ne essary). Let us simply remark that, if
1
=
2
and = 0, then the
homogenization of (4.10) is ompletely obvious. If
1
=
2
and >0, then the apriori estimates of
Proposition4.4 showsthat u
(0) goesto zero,while, if
1 >
2
and0,theyimply thatu
goesto
zeroin
1 .
Wearenowableto on ludetheproofofTheorem 3.1andTheorem3.8.
Proof ofTheorem3.1 andTheorem 3.8. LetusrstremarkthatProposition4.6implythatthesequen e
ofoperatorsS
,denedby(4.12),uniformly onvergestothelimitoperatorS. Theasymptoti analysis
oftheeigenvalueproblem(4.9)istrulygivenbythat ofT
givenby T :L 2 () !L 2 () f !S (B x; x f): Theeigenvaluesof T
being theinverseof that of (4.9). Introdu ing (x) =
R 1
0
B(x;y)dy whi h is the
weaklimitofB(x;
x
);wedenethelimitoperatorT by
T :L 2 () !L 2 () f !S(f): The sequen e T
does not uniformly onverge to T, but the sequen e T
is nevertheless sequentially
ompa t,in thesensethat
8f 2L 2 () lim !0 kT (f) T(f)k L 2 () =0; thesetfT (f); kfk L 2 () 1;0gissequentially ompa t:
Theorems3.1and3.8arethen onsequen esofTheorem4.7(seealso hapter11in[JKO95℄).
Theorem4.7. (seee.g. [Ans71℄,[Cha83℄)LetT
n
beasequen eof ompa toperatorsthat onvergesto
T. Assumethat(T
n )
n1
is olle tively ompa t andT is ompa t. Let 2C be aneigenvalue ofT,of
multipli itym. Let beasmooth urveen losing inthe omplexplane andleavingoutsidethe restof
thespe trumof T. Then, for suÆ ientlylargevalues ofn , en losesalso exa tly meigenvaluesof T
n
andleavesoutside therestof the spe trumof T
n .
0
The goal of this se tion is to proveTheorems 3.5 and 3.10. To understand the asymptoti behavior
of problem (1.1) when the dis ontinuity onstant (
0
) is negative, we rst res ale the equations by
introdu ingthe hange ofvariables y=
x
. Then,problem(1.1)isequivalentto
8 > > < > > : d dy a(y) d' dy +(y)' = (y)' in ; ' ( 1 l)=' ( 1 L)=0; (5.1) with =℄ 1 l; 1 L[,' (y)= ( x ),and
a(y)(resp: (y);(y))=
a 1 (y)(resp: 1 (y); 1 (y)) ify<0; a 2 (y)(resp: 2 (y); 2 (y)) ify>0:
Asgoesto0,thedomain
onvergestoR, andformallythelimitproblemof(5.1)is
8 > > < > > : d dx a(x) d dx +(x) =(x) inR; 2H 1 (R) : (5.2)
Werstre all someproperties of thespe trum of (5.2). Weintrodu ethe Greenoperator S a tingin
L 2 (R) denedby S:L 2 (R) ! L 2 (R) f ! uuniquesolutioninH 1 (R) of d dx a(x) du dx +(x)u=(x)f inR: (5.3)
Theeigenvaluesof S arepre iselythe inverseof thoseof(5.2). Nevertheless,to simplify thedis ussion
weshallsaythat is aneigenvalueofS,or(5.2),ifitsinversebelongstothespe trumofS.
Proposition5.1. The operator S isself-adjoint and non- ompa t. Its spe trum an be de omposed in
itsdis reteandessential part, (S)=
dis
(S)[
ess
(S). The lower bound of the essential spe trumis
equaltothe smallest ell rsteigenvaluein(1.3), namely
min ess (S)=min( 1 ; 2 ):
If(; ) isan eigen ouple inthe dis rete spe trum,thenthere exist
1 >0and 2 <0su hthat (x)= 1 (x) ifx<0; 2 (x) ifx>0; and(; i )isaneigen oupleof 8 > > < > > : d dx a i (x) d i dx + i (x) i = i (x) i in[0;1℄; x! i (x)e i x 1 periodi : (5.4)
tipli ity,whileitsessentialspe trumis hara terizedbyWeyl riterion, i.e.,82 ess (S)thereexistsa sequen efu n g2L 2 (R ) su hthat ku n k L 2 (R) =1; u n !0in L 2 (R ) weakly; (S Id)u n !0inL 2 (R ) strongly:
Proposition5.1tellsus inparti ularthat
ess
(S)isnotemptyandthatanydis reteeigenve torde ays
exponentiallyatinnity. Remarkthat equation(5.4)issimilar to(2.1).
Proof. Thestudyofthespe trumofS is lassi al. Theexponentialde ayofthedis reteeigenfun tions
isobtainedthroughFloquetTheory(see, e.g.,[MR73℄, [RS78℄). The sametoolyields thelowerbound
ofthe essentialspe trum(see [AC98℄, [CPV95℄). Note that these results areobtained under themere
assumptionthat the oeÆ ients of equation (5.3)are positivemeasurable fun tions (nosmoothness is
required).
Inordertopasstothelimit!0in (5.1),wealsointrodu eanoperatorS
a tingin L 2 (R) denedby S :L 2 (R) ! L 2 (R ) f ! u uniquesolutionin H 1 0 ( )of 8 > > < > > : d dx a(x) du dx +(x)u =(x)f; in u (x)=0on : (5.5) The operator S
is ompa t and its eigenvalues are the inverses of that of (5.1). Unfortunately, the
onvergen e of thesequen e S
toS is notuniform, so that the limitof the spe trum ofS
is notthe
spe trumofS. Nevertheless,thislimit anbe hara terizedexpli itlyandwere allthefollowingresult
thatmaybefoundin[AC98℄.
Proposition5.3. For allf 2L 2 (R) , S (f) onvergesstronglytoS(f)inL 2 (R ), andwehave lim !0 (S )=(S)[ BL :
Furthermore, therst eigenvalue
1
onvergestoa limit
1
whi h does belong tothe spe trumof S and
isthusthe smallestelement of(S). Wealso have
min BL =min ess (S)=min( 1 ; 2 ): (5.6)
That part of the limit spe trum, denoted by
BL
, is alled the boundary layer spe trum. It an be
hara terized ompletely in terms of an equation similar to (5.2) but in the half-line (for details, see
[AC98℄). We do notdwell on this boundary layerspe trum sin e we only need to know (5.6) in the
sequel.
Lemma5.4. Let
0
be dened asin Lemma3.9, i.e.,
1 ( 0 )= 2 ,and 1;0
the orresponding
eigen-ve tordenedby (2.1). Let(
0 )bedenedby ( 0 )=a 1 (0) d 1; 0 dy (0) a 2 (0) d 2 dy (0):
( 0 )<0; (5.7) thelimit 1 ofthe rsteigenvalue 1 of problem (1.1)satises 1 <min( 1 ; 2 ):
Remark5.5. Inparti ular,thisLemmaapplieswhen
1
=
2
,and(
0
)<0. Itimpliesthat, when
thedis ontinuity onstantisnegative,thelimitrsteigenvalue annotbepredi tedbythehomogenized
modelsobtainedunderstri tperiodi ityassumptiononea hsideoftheinterfa e.TheproofofLemma5.4
reliesonProposition2.2,thatwehavenotbeenabletoproveinthegeneral ase,butundertheadditional
assumptionthatthe oeÆ ientsareC
2
,orpie ewise onstant.
Proof ofLemma5.4. Forall 2 [
0
;+1[, where
0
isdened in Lemma 3.9, be auseof the on avity
of
1
()and
2
(), we anasso iatetoea haunique
0 0su hthat 1 ()= 2 ( 0 )and 2; 0 isthe rsteigenve tordened by 8 > > > > > > < > > > > > > : d dx a 2 (x) d 2; 0 dx + 2 (x) 2; 0 = 2 ( 0 ) 2 (x) 2; 0 in[0;1℄ x! 2; (x)e 0 x 1 periodi ; 2; 0(0) =1: (5.8)
Notethatfor=
0
,wehave
0
=0. Thepair(; ) denedby
= 1 ()= 2 ( 0 ) = 1; forx>0; = 2;0 forx<0;
isaneigen oupleforproblem(5.2)ifand onlyif
()=a 1 (0) d 1; dx (0) a 2 (0) d 2; 0 dx (0)=0: (5.9)
Thanksto Proposition2.2,wehave
lim !+1 ()= lim !+1 a 1 (0) d 1; dx (0) lim 0 ! 1 a 2 (0) d 2; 0 dx (0)=+1: Therefore,ifweassume( 0
)<0then Equation(5.9)admitsasolution,for some <
0
, and
0 >0.
We have thus obtained a value of su h that < min(
1 ; 2 ). Finally, sin e 1 , by virtue of Proposition5.1,wehave 1 2 dis (S). Conversely, if 1 < min ( 1 ; 2
) we know from Proposition 5.6 that on both sides of the origin the
orrespondingeigenfun tion hasanexponentialde ay. ThenProposition5.1showthatitmustofthe
form =
1;
and =
2;0
for someand
0
on ea h half line. Sin e identity (5.9)isa ne essary
then 1
isin thedis retespe trumofproblem(5.2). Theorems3.5and 3.10arethena onsequen eof
Proposition 5.6. Indeedtheeigenfun tion
1
(x) in Theorems 3.5 and3.10 isequal to
1 p ' 1 x where ' 1
isthersteigenfun tionin Proposition5.6. Inequality(5.10)thenbe omes
2 d dx 1 (x) 1 p d dx x L 2 () + 1 (x) 1 p x L 2 () Cexp
whi hin turnimply
d dx 1 (x) 1 p d dx x L 2 () + 1 (x) 1 p x L 2 () C 0 exp 0 forany 0 <.
Proposition5.6. Assume that problem (5.2), or equivalently operator S, admits a rst positive
nor-malized eigen ouple ( 1 ; ) su hthat 1 <min( 1 ; 2
). Then the rst positive normalized eigen ouple
( 1 ;' 1 )of (5.1), orofS ,satises 0 1 1 Cexp and d dx ' 1 d dx L 2 ( ) +k' 1 k L 2 () Cexp (5.10)
where C and arestri tlypositive onstant independentof .
Proof. Sin eweassumed
1 <min ess (S),wehave 1 = min '2H 1 (R) 6=0 Z R a(x)j d dx 'j 2 dx+ Z R (x) 2 dx Z R (x)' 2 dx ;
andthisminimumisattainedfor'= whi h belongstothedis retespe trumof S. Wealsohave
1 = min '2H 1 0 ( ) 6=0 Z a(x)j d dx 'j 2 dx+ Z (x)' 2 dx Z ( x)' 2 dx ;
and this implies, by in lusion of spa es that
1
1
. Let be asmooth ut-o fun tion, vanishing
outside = l ; L ,equalto1on l +1; L 1
,su hthat 01;and
d
dx
doesnotdependon
(seeFigure5.1). Wethenhave 2H
1 0 ( ),and 1 Z a(x)j d dx ( )j 2 dx+ Z (x)( ) 2 dx Z ( x)( ) 2 dx : (5.11)
N N +1 N 1 N
Figure5.1: Cut-ofun tion .
By onstru tion
d
dx
hasitssupportin[
l ; l +1℄[[ L 1; L
℄andinequality(5.11)be omes
1 Z R a(x)j d dx j 2 dx+ Z R (x) 2 dx+R 1 (1 R 2 ) Z R (x) 2 dx ; (5.12) with R 1 =2 Z [ l ; l +1℄[[ L 1; L ℄ a(x)j (x)j d dx d dx + d dx dx and R 2 = Z l +1 1 (x) (x) 2 + Z +1 L 1 (x) (x) 2 Z R (x) (x) 2 :
Thanksto Proposition5.1,weknowthat
sup x2( 1; l ) j (x)jCexp 1 l ;and sup x2( L ;+1) j (x)jCexp 2 L with 1 > 0 and 2
< 0. We an dedu e that R
1 Cexp and R 2 Cexp with = min (lj 1 j;Lj 2
j),andinsertingtheseinequalitiesin (5.12)weobtain
1 1 1+Cexp : (5.13)
Letusnowshowthat'
1
onvergesto . Inordertoobtainanapproximationof thatvanishesonthe
boundariesofthedomain
,weaddto anaÆnefun tionwhi h ompensatesitsvaluesatbothends
ofthedomain. Wedene
(x)= (x)+`
(x)where`
istheaÆnefun tionsu hthat
l +` l =0and L +` L =0: By onstru tion, 2H 1 0 ( ),and
issolutionofthesameproblemthan'
1 uptoaperturbationr . 8 > > > < > > > : d dx a(x) d dx +(x) = 1 ( x) +r in℄ l ; L [ ( l )= ( L )=0: (5.14)
Theperturbation is r =( 1 1 ) +` 1 ` d dx a d` dx 2H 1 (
). The oeÆ ientsbeing
bounded, weobtainthat forall2H
1 0 ( ), Z r C j 1 j+sup j` j kk L 2 () +C d` dx L 2 () d dx L 2 () ;
whereCis a onstantwhi hdoesnotdependon. Fromtheexponentialde ayof wededu ethat
sup j` jCexp and d` dx L 2 () C p exp ; (5.15)
andwiththehelp ofestimate(5.13)weobtain
Z r Cexp kk L 2 ( ) + d dx L 2 () ! : (5.16) Thersteigenvalue 1
beingsimple,byaFredholmalternativewe ande ompose
intoa omponent
proportionalto'
1
anda omponentorthogonalto'
1 . Wewrite = 1 +g ,where isa onstant, and kg k L 2 ( ) + dg dx L 2 () C kr k H 1 ( ) where C is the norm of S 1 1 Id 1
, a bounded operator dened on the orthogonal of the line
generated by ' 1 . We haveC C j 1 2 j
, where C is a onstantindependent of , and
2 is the next eigenvalueofS . Ifweobtainthatj 1 2
j> >0,with independentof,wethendedu e,withthe
helpofinequality(5.16) kg k L 2 ( ) + dg dx L 2 () C exp : (5.17)
Fromthede omposition
= ' 1 +g ,weget j jk' 1 k L 2 ( ) kg k L 2 ( ) k k L 2 ( ) j jk' 1 k L 2 ( ) +kg k L 2 ( ) : Wehavek' 1 k L 2 ( ) =1andk k L 2 (R) =1thus k k L 2 () 1 = k` k L 2 () k k L 2 (Rn) k` k L 2 () +k k L 2 (Rn) C 1 exp
thankstoestimate (5.15)and theexponentialde ayof . Asa onsequen e,jj
j 1jC 1 exp and and' 1
beingpositives,wealsohave
j 1jC 1 exp : (5.18) Finally,ifwewrite' 1 (x) (x)=(1 )' 1 (x) g (x)+` (x)on
andusingestimates(5.15),(5.17)
and(5.18)weobtain d' 1 dx d dx L 2 () +k' 1 k L 2 ( ) C 1 exp Cexp 0
Letusnowshowthatthespe tralgapisuniformly bounded, i.e., 0< < 2 1 <C. Weknowthat 2 onvergestoalimit 2
whi h either belongs to
BL
[
ess
(S)orto
dis
(S). In thelatter ase,the
eigenvaluesofthedis rete spe trum areisolated so that 0< <
2
1
<C. In theformer ase, we
know from (5.6) that
2
min(
1 ;
2
) whi h is stri tly larger than
1
by assumption, so that again
0< < 2
1
<C. ThisyieldsthedesiredresultforsuÆ ientlysmall.
6 Proofs in the ase < 0 and (
0
) 0.
Inthis se tion we proveTheorems 3.11 and 3.12 following the strategy used in se tion 4for the ase
0. A ording to Proposition 4.2 and Remark 4.3, the original problem (1.1) is equivalent to the
fa torizedproblem (4.9) forany valueof thedis ontinuity onstant. Introdu ing, asin Lemma 4.5,
anoperatorS
,the onvergen eof(4.9)isgovernedbythehomogenizationofproblem(4.10)withgiven
right hand side. The key element for the proof of Proposition 4.6, and in turn Theorem 3.8, is thea
prioriestimategivenbyProposition4.4. Itdoesnotholdfor<0. Nevertheless,theargumentsofthe
proofofProposition4.4yieldsasimilarresultthat westatein Proposition6.1below.
Proposition6.1. Thesolution u
of equation (4.10) satises ku k 2 H 1 0 () + 1 2 2 ku k 2 L 2 (1) + ju (0)j 2 Ckf k L 2 () ku k L 2 () (6.1)
where C isa onstant independentof .
Sin eweassumed<0,(6.1)alonedoesnotfurnish suÆ ientaprioriestimatesfor on luding. Thus,
fortheproofofTheorems3.11and3.12weneedanadditionallemma.
Lemma6.2. Assume that
1 >
2
, <0 and (
0
) >0. Then, the solution u
of equation (4.10) satises du dx L 2 () Ckf k L 2 () ; ku k L 2 ( 1 ) Ckf k L 2 () ; and j u (0)jC p kf k L 2 () : Assumethat 1 > 2 ,<0and( 0
)=0. Then, the solutionu
ofequation(4.10) satises du dx L 2 ( 2 ) Ckf k L 2 () ; ku k L 2 (1) C p kf k L 2 () ; and e 0 x dv dx L 2 ( 1 ) Ckf k L 2 () ; where v =u 1 (x;x=)= 1;0 (x;x=)in 1 .
Proof ofTheorem3.11. Thanks to the a priori estimate of Lemma 6.2, the ase
1 > 2 , < 0 and ( 0
)>0is ompletelysimilarto the ase
1 >
2
and0,whi hisalreadysolvedinse tion4.
Proof of Theorem 3.12. Let u
be the solution of (4.10) with right hand side f
whi h is a bounded
sequen ein L
2
(). Weintrodu ethefun tion
0 (x;y)= 1 (x) 1;0 (y)+ 2 (x) 2 (y); (6.2)
v (x)=u (x) x; x 0 x; x : Remarkthat v =u in 2 , and v (0)=u
(0) (be ause ofthenormalization ondition(2.2)). Testing
variationallyequation(4.10)against
0 (x; x ) (x; x ) (x),where isatest fun tionin H 1 0 (), weobtain Z D x; x du dx d dx 0 x; x x; x ! dx+ 1 2 2 Z 1 B x; x u 0 x; x x; x ! dx (6.3) + 1 u (0) (0)= Z f 0 x; x x; x ! dx: Repla ingu byv
in itslefthandside,identity(6.3)be omes
Z 1 a 1 x 2 1 x d dx v 1; 0 x 1 x ! d dx 1; 0 x 1 x ! dx+ Z 2 D x; x dv dx d dx dx (6.4) + 1 2 2 Z 1 1 x 2 1;0 x v dx+ 1 v (0)u (0)= Z 1; 0 x 1 x f dx: Notethat Z 1 a 1 x 2 1 x d dx v 1; 0 x 1 x ! d dx 1; 0 x 1 x ! dx= Z 1 a 1 x 2 1;0 x dv dx d dx dx + Z 1 a 1 x d dx 1;0 x d dx v 1;0 x Z 1 a 1 x d dx 1 x d dx 1;0 x 2 1 x v ! ;
and,byintegrationbyparts anddenition(2.1)of
1; ,wehave Z 1 a 1 x d dx 1;0 x d dx v 1;0 x Z 1 a 1 x d dx 1 x d dx 2 1;0 x 1 x v ! = 1 ( ( 0 ) )v (0) (0)+ 2 1 2 Z 1 1 x 2 1; 0 x v :
Asa onsequen e,identity(6.4)be omes
Z 1 a 1 x 2 1;0 x dv dx d dx dx+ Z 2 D x; x dv dx d dx dx (6.5) + 1 ( 0 )v (0) (0)= Z 1;0 x 1 x f dx
0 Z 1 a 1 x 2 1;0 x dv dx d dx dx+ Z 2 D x; x dv dx d dx dx= Z 1;0 x 1 x f dx
Notethatforanyboundedsequen e
inW 1;1 (), Z 1 a 1 x 2 1;0 x dv dx d dx dx C e 0 x dv dx L 2 ( 1 ) e 0 x L 2 ( 1 ) !0 sin eke 0 x dv dx k L 2 ( 1 )
isbounded, thanksto Lemma6.2. Of ourse
R 1 1; 0 ( x ) 1( x ) f goesto0
exponen-tiallyfast. Forsu hbounded
,(6.3)thereforewrites Z 2 D x; x du dx d dx dx= Z 2 f dx+o(1): (6.6)
Sin ethetestfun tions inthetwo-s ale onvergen emethodareofthetype
(x)= 0 (x)+ 1 (x;x=)
withsmoothfun tions
0 ;
1
,theyareuniformlyboundedinW
1;1
()andone anuse(6.6)topassto
thelimit. Classi alargumentsofhomogenizationallowto on lude.
Proof ofLemma6.2. With the hoi e
=v in(6.5)weobtain Z 1 a 1 x 2 1;0 x dv dx 2 dx+ Z 2 D x; x dv dx 2 dx+ 1 ( 0 )(v ) 2 (0)= Z f u dx: (6.7) If( 0
)>0,thisimpliesthat
jv (0)j 2 Ckf k L 2 () ku k L 2 () : (6.8) Be auseu (0)=v
(0),plugging(6.8)in (6.1)yieldsthedesiredresults.
If(
0
)=0,identity(6.7)onlyimpliesthat
e 0 x dv dx 2 L 2 (1) Ckf k L 2 () ku k L 2 () ; and du dx 2 L 2 (2) Ckf k L 2 () ku k L 2 () : (6.9)
Wewill nextshowthat
ku k 2 L 2 () C kf k 2 L 2 () +ku k 2 L 2 ( 2 ) (6.10)
and,togetherwith(6.9)andPoin areinequalityin
2
,thisyieldsthedesiredresults.
Notethat u (0) 2 j 2 j 2 Z 2 du dx 2 Ckf k L 2 () ku k L 2 () :
Usingthisinequalityin (6.1)gives
ku k 2 L 2 (1) Ckf k L 2 () ku k L 2 () C kf k 2 L 2 () +ku k 2 L 2 ()
7.1 The ase of C 2
oeÆ ients.
TherststepissimilartotheproofofLemma2.1,namelywetransform(2.1)into(2.4). Ifweassumethat
the oeÆ ientsa i ; i and i areC 2
periodi fun tionson[0;1℄,thenthersteigenfun tion
i;0
isa tually
twotimesdierentiable,andthusthe oeÆ ientsb andsof(2.4)arealsoof lassC
2
. Proposition2.2is
thena onsequen eof Lemma7.1.
Lemma7.1. Let b and s be periodi positive fun tions on [0;1℄ su h that their se ond derivative b
00
ands
0 0
existand are pie ewise ontinuous. Denote by M >m>0twopositive onstant whi h arethe
upper andlower bounds of b and s. For ea h 2 R the rst eigenve tor u
of problem (2.4) with the
normalizationu (0)=1satises C 1 C 2 p () +C 3 p () jj u 0 (0)C 3 p ()+C 1 + C 2 p () ; (7.1)
where the positive onstantsC
1 ;C
2
andC
3
dependonly onb ands.
Proof. Theassumedsmoothnessofb andsenablesustoperformaLiouvilletransformationofproblem
(2.4). Introdu ing t= 1 Z x 0 s(z) b(z) 1 2 dz = Z 1 0 s(z) b(z) 1 2 dz; and f (t)=(s(x)b(x)) 1 4 u (x); (7.2)
thetransformedequationis,see[Eas73℄,
8 < : d 2 f dt 2 (t)+( 2 ()+Q(t))f =0in[0;1℄; t!f (t)e t 1 periodi ; (7.3) with Q(t)= 2 b 1 4 (x)s 3 4 (x) d dx b(x) d dx (b(x)s(x)) 1 4 :
We anassumewithoutlossofgeneralitythat =1. Theboundary onditionsarepreservedsin ethis
hangeof variablepreservesperiodi ity. Weshalluse thefa t that Qisabounded 1-periodi fun tion.
ItissuÆ ientto prove(7.1)for>0,sin ein theother asethefun tiong
(t)=f
( t) issolutionof
(7.3),with >0,ifQis repla edbyQ( t),whi his alsoabounded1-periodi fun tion. Byaddinga
onstant to Q (andsubtra ting it from ()), we an alwaysassume that M <Q(t)< 1. On the
otherhand,thankstoLemma2.3,forsuÆ ientlylargewe analsoassumethat()translatedbythe
above onstantisnegative.
Next,weintrodu eg
1
andg
2
asthetwofundamentalsolutionsoftheCau hyproblem fortheordinary
dierentialequation d 2 g dt 2 +(()+Q(t))g=0,satisfying g 1 (0)=1; g 0 1 (0)=0;andg 2 (0)=0; g 0 2 (0)=1:
Itis a lassi alresultofFloquettheorythat X 1
=e andX
2
=e are therootsof the hara teristi
equation X 2 (g 1 (1)+g 0 2 (1))X+1=0:
By linearity, we an write f
(t) = f (0)g 1 (t)+f 0 (0)g 2 (t). Sin e > 0, e = + 2 1 1 2 where 2=(g 1 (1)+g 0 2 (1)) . Consequently,>1and g 1 (1)+g 2 0(1)=2>e ande >2 1 >2 2e =g 1 (1)+g 2 0(1) 2e : (7.4)
Fromtherelationf
(1)=e f (0)wededu ethat f 0 (0) f (0) g 2 (1)=e g 1
(1). Usingrelation(7.4)wehave
obtainedthat g 0 2 (1)> f 0 (0) f (0) g 2 (1)>g 0 2 (1) 2e : (7.5)
FollowingPi ard'siterationmethod(seee.g. [MW79℄),wedenere ursivelyasequen e(v
n (t)) n2N by v 0 (t)= 1 ! sinh(!t)and v n (t)= 1 ! Z t 0 sinh( !(t ))Q()v n 1 ()d foralln1: For!= p (),wend that g 2 (x)= P +1 n=0 v n
(x). Sin e sinh(!(t ))Q()>0forall0< <t,
andv
0
(t)>0for allt >0, byindu tion,we an on ludethat W
n (x)v n (x)w n (x), foralln 0 andx0,whereW n andw n
aretwoothersequen esdened byW
0 =v 0 =w 0 , and W n = M ! Z t 0 sinh(!(t ))W n 1 ()d; w n = 1 ! Z t 0 sinh(!(t ))w n 1 ()dforn1: NotethatW(t)= P +1 0 W n (t)(resp. w(t)= P +1 0 w n (t))isasolutionof d 2 W dt 2 +( M)W =0 resp. d 2 w dt 2 +( 1)w=0 ;
andthereforeisgivenbyW(t)=sinh t
p M (resp. w(t)=sinh t p 1 )and onsequently sinh p M =W(1)g 2 (1)w(1)=sinh p 1 : (7.6) SimilarlyW 0 n (t)v 0 n (t)w 0 n (t),and p M osh p M =W 0 (1)g 0 2 (1)w 0 (1)= p 1 osh p 1 : (7.7)
Usinginequalities(7.6)and(7.7)in(7.5)yields,
Ce M 1 2 p p M f 0 (0) f (0) p 1 e M 1 2 p 2e :
Usingthe hange of variables (7.2),and using the resultof Lemma 2.3 to bound e
in termsof ()
Asin theprevioussubse tion,itissuÆ ient to onsider system(2.4),whi h isequivalentto(2.1), and
tostudythe ase goingto+1. Asin Lemma2.3,werewrite(2.4)asarst-ordersystem
Y 0 (x)=A(x)Y(x) Y(1)=e Y(0) ;A= 0 b 1 ()s 0 ; andY = Y 1 =u Y 2 =bu 0 :
Here weassume that the oeÆ ients b;sare pie ewise onstant fun tions. Morepre isely,there exists
a number N, a family of points (x
i ) 0iN satisfying x 0 = 0 < x i 1 < x i < x i+1 < x N = 1 for
2iN 2,andpositivevalues(b
i ) 1iN and( s i ) 1iN su hthat b(x)=b i ands(x)=s i forx2(x i 1 ;x i ); 1iN:
Thegoalis to provethat Y
2
(0) growslinearlyas goes to +1,whi hin turn provesProposition2.2,
sin e d i; dx (0)=b 1 (0)Y 2 (0)+ d i;0 dx (0):
By Lemma 2.3 we already know that () < 0 for 6= 0 and has quadrati growth at innity. A
straightforward omputationyieldsforanyx2(x
i 1 ;x i ) Y(x)=M i (;x)Y(x i 1 ); M i (;x)= 2 6 4 osh' i (x) 1 p ()b i s i sinh' i (x) p ()b i s i sinh' i (x) osh' i (x) 3 7 5 ; (7.8) with' i (x)= q ()si b i (x x i 1 ). Thus Y(1)=M()Y(0)=e Y(0);with M()= N Y i=1 M i (;x i ): Ea hmatrixM i
(;x)hasitsdeterminantequalto1,aswellasM(). ThusthetwoeigenvaluesofM()
aree
ande
. Letus omparetheseexa teigenvalueswiththoseoftheleadingordertermofM()as
goesto+1. Introdu ing D()=diag
p () ;1 ,wehave M i (;x i )=e 'i(xi) D() 1 M 0 i D() 1+O e ;with M 0 i = 1 2 2 4 1 1 p bisi p b i s i 1 3 5 ; and=min 1iN 2(x i x i 1 ) q s i m b i M
>0. Therefore,noti ing that
P N i=1 ' i (x i )=C p ()where
C>0doesnotdependon,weobtain
M()=e C p () D() 1 M 0 D() 1+O e ;withM 0 = N Y i=1 M 0 i :
Uptoasmallremainder,theeigenvaluesofM()arethusequaltothoseofM timesthemultipli ative
fa torexp(C
p
() ). Sin eM
0
doesnotdepend on , this provesthat ()=
2
+o(1) forsome
positive onstant >0. Ontheotherhand,theeigenve torsofM()areequaltoD()
1
timesthoseof
M 0
(uptoasmallremainder). ChoosingthenormalizationY
1
(0)=1,thisyieldsthatY
2
(0)=
0
+o(1)
forsome onstant
0
,whi h ispositiveasalreadyremarkedintheproofofLemma2.3.
NoteAddedin Proof. After submission of this paper for publi ation, wefound an alternative proof of
Lemma5.4,whi hdonotrelyonProposition2.2. ThisenablesustoproveTheorem3.5andTheorem3.10
assumingonlythattheperiodi oeÆ ientsarepositive,bounded,measurablefun tions. Thisproofwill
bepresentedinafutureworkin ollaborationwithA. Piatnitski.
Referen es
[All92℄ G.AllaireHomogenizationandtwo-s ale onvergen e. SIAMJ.Math.Anal.23(6):1482{1518,
(1992).
[AB99℄ G. Allaire, G.Bal Homogenization of the riti ality spe tral equation in neutrontransport.
M2AN 33,pp.721-746,(1999).
[AC00℄ G. Allaire, Y. Capdebos q Homogenization of a spe tral problem in neutroni multigroup
diusion.Comput.MethodsAppl.Me h. Engrg.187:91-117,(2000).
[AC98℄ G. Allaire and C. Con a Blo h wavehomogenization and spe tral asymptoti analysis. J.
Math.PuresetAppli. 77:153-208,(1998).
[AM97℄ G.Allaire,andF.MaligeAnalyseasymptotiquespe traled'unproblemedediusion
neutron-ique.C.R. A ad.S i. ParisSerie I,t.324:939{944,(1997).
[AP02℄ G.Allaire,A.PiatnitskiUniformspe tralasymptoti sforsingularlyperturbedlo allyperiodi
operators,Commun.inPartial Dierential Equations,27,pp.705-725(2002).
[Ans71℄ P.AnseloneColle tivelyCompa tOperatorApproximationTheoryandAppli ationstoIntegral
Equations. Prenti e-Hall,EnglewoodClis,N.J.,(1971).
[Cap98℄ Y.Capdebos qHomogenizationofadiusionequationwithdrift.C.R.A ad.S i.ParisSerie
I,t.327:807-812,(1998).
[Cap99℄ Y.Capdebos q Homogeneisationdesmodelesdediusionen neutronique. ThesedeDo torat
del'UniversiteParisVI,(1999).
[Cap02℄ Y.Capdebos q Homogenizationofaneutroni riti aldiusionproblemwithdrift. Roy.So .
Edin.Pro .A, 132,pp.1{28(2002).
[CZ℄ C.Castro, E. Zuazua Lowfrequen y asymptoti analysis of astringwith rapidly os illating
density. SIAMJ.Appl. Math.60(4):1205-1233,(2000).
[Cha83℄ F. Chatelin Spe tral Approximation of Linear Operators. Computer S ien e and Applied
Sons,(1995).
[Eas73℄ M.S.P. Eastham The Spe tral Theory of Periodi Dierential Equations. S ottish A ademi
Press,(1973).
[GT83℄ D.Gilbarg andN. Trudinger Ellipti Partial Dierential Equations of Se ondOrder,2nd ed.
ComprehensiveStudies inMathemati s.Springer-Verlag,(1983).
[JKO95℄ V.Jikov,S.Kozlov,O.OleinikHomogenizationof dierentialoperatorsandintegral
fun tion-als,Springer,Berlin,(1995).
[KP93℄ S. Kozlov, A. Piatnitski Ee tive diusion for aparaboli operator with periodi potential,
SIAMJ.Appl.Math.53,401{418,(1993).
[Mal96℄ F. Malige Etude mathematique et numerique de l'homogeneisation des assemblages
om-bustiblesd'un urde rea teurnu leaire. ThesedeDo toratde
l'
e olePolyte hnique,(1996).
[MW79℄ W. Magnus et S. Winkler Hill's Equation. Corre ted reprint of the 1966 edition, Dover
Publi ations,(1979).
[MR73℄ J. Mawhin and N. Rou he Equations dierentielles ordinaires, Tome 1 : theorie generale.
Masson,Paris,(1973).
[MT78℄ F.Murat andL.Tartar H- onvergen e. Seminaire d'Analyse Fon tionnelle etNumerique de
l'Universited'Alger, mimeographednotes (1978). Englishtranslationin Topi s in the
mathe-mati al modellingof omposite materials, CherkaevA., KohnR.V.,Editors,Progressin
Non-linearDierentialEquationsandtheirAppli ations,31,Birkhauser,Boston,(1997).
[Ngu89℄ G.Nguetseng A general onvergen e resultfor afun tionalrelated to thetheoryof
homoge-nization. SIAMJ.Math. Anal.,20:608{623,(1989).
[Pia98℄ A.Piatnitski Asymptoti behaviourofthegroundstateofsingularlyperturbedellipti
equa-tions. Commun.Math. Phys.197:527-551,(1998).
[RS78℄ M.ReedandB.Simon Methodsofmodern mathemati alphysi s. A ademi Press,NewYork,