• Aucun résultat trouvé

Asymptotics of the eigenvalues of the Dirichlet-Laplace problem in a domain with thin tube excluded

N/A
N/A
Protected

Academic year: 2021

Partager "Asymptotics of the eigenvalues of the Dirichlet-Laplace problem in a domain with thin tube excluded"

Copied!
11
0
0

Texte intégral

(1)

HAL Id: hal-01120422

https://hal.archives-ouvertes.fr/hal-01120422v2

Submitted on 7 Apr 2016

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

abroad, or from public or private research centers.

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

publics ou privés.

Asymptotics of the eigenvalues of the Dirichlet-Laplace

problem in a domain with thin tube excluded

Xavier Claeys

To cite this version:

Xavier Claeys. Asymptotics of the eigenvalues of the Dirichlet-Laplace problem in a domain with

thin tube excluded. Quarterly of Applied Mathematics, American Mathematical Society, 2016, 74 (4),

pp.595-605. �hal-01120422v2�

(2)

Lapla e problem in a domain with thin tube ex luded

X.Claeys

∗†

Abstra t

We onsider aLapla eproblem withDiri hlet boundary ondition inathree dimen-sionaldomain ontaininganin lusiontakingtheformofathintubewithsmallthi kness

δ

. Weprove onvergen einoperatornormoftheresolventofthisproblemas

δ → 0

, estab-lishingthattheperturbationindu edbythein lusionontheresolventisnotgreaterthan

O(| ln δ|

−γ

)

for some

γ > 0

. We dedu e onvergen eof theeigenvaluesoftheperturbed operatortowardthelimitoperator.

1 Introdu tion

Analysis of singular perturbations of ellipti boundary value problems by in lusions of small volume has re eived a lot of attention in the past de ades due to its numerous appli ationstomodelingandnumeri alsimulationofmulti-s aleproblems,anditspossible usein e ientinverseproblem[1,2℄orshapeoptimization[20℄ strategies.

For Lapla e equation, many works of theexisting literature havebeendedi ated to asymptoti analysis involving thepresen e of an in lusion that has small size in all di-re tionsofspa e(typi allysmallballs),inthe aseofeither impenetrable(homogeneous Diri hlet,Neumann,orRobin)orpenetrable(transmissionproblem)boundary onditions. Su hasymptoti shavebeenprovidedforboth2-Dand3-Dproblems,inthe aseofeither a singlein lusion ormany, see [19, 18, 17, 16, 10℄. Thebooks [14, 9℄are landmarkson this typeofproblem.

As regards Lapla e equation perturbed by a small in lusion that takes the form of a thin tube, mu h less work is available in the literature. Some works an be found in the ase of a tube entered around a smooth losed ontour without self-interse ting point. The analysis of Lapla e equation in 3-Dwith Neumann and Diri hlet boundary ondition was provided in [5, 6℄ where the author onsiders either a tube entered at a non-self-interse ting smooth losed ontour, oratube entered at astraightsegment. Thisanalysisisrenedandgeneralizedtoarbitrarydimensionsin[15℄and[14,Chap.12℄. Similarresultswereestablishedfora ousti s andHelmholtzequationin [3, 7,22℄.

It shouldbepointedoutthat,in the aseofa3-D Lapla eequation perturbedby a thin elongatedin lusion,the asesofthehomogeneousDiri hletorRobinboundary on-dition leadtomu hmore hallenginganalysisthanthe aseofNeumannortransmission onditions: while Neumann/transmission onditionslead to ratherstandardasymptoti onstru tions, mat hingof asymptoti sin theDiri hlet/Robin aseleadstoan ill-posed 1-Dintegralequation.

LaboratoireJa ques-LouisLions,UPMCUniv. Paris6andCNRSUMR7598,75005,Paris,Fran e

(3)

ary onditionhasmanyimportantappli ationsinparti ularin ele tri alengineeringand antenna based ele tri /ele troni devi es, as it is ommonly taken as a rough model for ele tromagneti dira tion by perfe tly ondu ting thin wires, see for example [11, Chap.6℄. Inthis ontext,thestudy ofeigenvaluesis ofparti ularimportan easitis re-latedtoresonan ephenomena. Yet,inthe aseofthinelongatedin lusionswithDiri hlet boundary ondition, onlyPlanida [21℄ hasaddressedthis issueso far. Inthis referen e, the authorestablishes onvergen eof eigenvaluesin the asewhere thein lusion is en-teredaroundasmooth losed urvethatdoesnotself-interse t. Howevernoexpli itrate of onvergen eisprovided.

Inthepresentarti le, westudyLapla eequation withhomogeneousDiri hlet ondition, and theasso iated eigenvalueproblem, perturbedby the presen e of a small elongated in lusion that takesthe form of a thin tube. Weestablish onvergen e in the operator norm,oftheresolventoftheperturbedproblemtowardtheresolventofthelimitproblem. A rstnovelingredienthere omparedtopreviouslyestablishedresults on ern the geo-metri al ongurationsunder onsiderationthatonlyrequirethein lusionto on entrate around aparametrized urve. Contrary to pre-existingworks, thepresent analysisdoes not require the in lusion to be onne ted, or admit any symmetry of revolution. The limit urve onsideredheremayadmit self- rossingpointswhi h, to ourknowledge,has neverbeeninvestigatedin theexistingliterature(atleastin the aseofanhomogeneous Diri hletboundary ondition). Inaddition,weestablisha onvergen eresult(witherror estimates) for eigenvalueswith expli it error estimate showing that the presen e of an elongatedin lusionindu esashiftoforder

O(| ln δ|

−γ

)

oftheeigenvalues,forsome

γ > 0

. Theoutlineofthepresentarti le isasfollows. InSe tion 2wedes ribein detailthe geometry under onsideration. In Se tion 3 we introdu e weighted

δ

-dependent norms adapted to our analysis, and establish a stability property for the Lapla e operator in terms ofthese norms. The nextse tion isdedi ated to provingHardytype inequalities thatwillbene essaryforthesubsequenterrorestimates. Inthelastse tionwederivean upperboundforthedieren e,inoperatornorm,betweentheresolventsoftheperturbed problem andthelimitproblem.

2 Geometry under onsideration

Westartbydes ribingindetailthegeometri alsettingunder onsideration. Let

Ω ⊂ R

3

refer to aLips hitz domain. Consider a

C

1

-fun tion

γ : R 7→ Ω

that is

L

-periodi for some

L > 0

, andsu h that

0 < α

≤ |

dz

(z)| ≤ α

+

, ∀z ∈ R

for xed onstants

α

±

> 0

. Thenweset

Γ := γ(R)

,whi hisaLips hitz urve. Weshallneedtorefertothedistan e fun tion

d(x) := inf

y∈Γ

|x − y|.

(1) Then we onsider

δ

)

δ>0

asany family of Lips hitz domains su h that there exists a xed onstant

C > 0

independentof

δ

forwhi h

sup

x∈Ξ

δ

d(x, Γ) ≤ C δ.

(2)

This onditionimposesthat,as

δ → 0

,thesets

Ξ

δ

aremoreandmore on entratedaround

Γ

. Changing the index parameter

δ

throughmultipli ation by somefa torif ne essary, wemayassumethat

C < 1

in(2)withoutrestri tinggenerality. Thenwedenote

(4)

lems in avities ontainingin lusions whose geometry is similar to wires. Note that we do not require

γ

to be one-to-one overaperiod so that, under the assumptions above, the urve

Γ

mayself-interse t. Note alsothat

Ξ

δ

donotneedto be onne tedand may degenerate,as

δ → 0

,toonlyapartof

Γ

withtips. Asa onsequen e,thepresentsetting overssu hgeometri al ongurationsastheonespresentedin pi turesbelow.

Figure1Threeexamplesofgeometri al ongurations overedbytheanalysispresentedhere.

3 Weighted Sobolev spa es GivenanyboundedopenLips hitzdomain

ω ⊂ R

3

,theset

L

2

(ω)

shallrefertothespa eof square integrablefun tions over

ω

equipped withthepairing

u, v 7→ (u, v)

ω

:=

R

ω

u v dx

and the norm

kuk

L

2

(ω)

=

p

(u, u)

ω

. As usual,

H

1

0

(ω)

shall refer to the ompletion of

C

0

(ω) := {ϕ ∈ C

(R

3

), ϕ = 0

on

R

3

\ ω, supp(ϕ)

bounded

}

forthenorm

kϕk

H

1

0

(ω)

:=

kϕk

L

2

(ω)

+ k∇ϕk

L

2

(ω)

. Finally

H

−1

(ω)

will refertothetopologi aldualto

H

1

0

(ω)

.

Foragivenparameter

δ > 0

smallenough,weareinterestedinstudying theeigenvalues oftheoperator

A

δ

: H

1

0

(Ω

δ

) → H

−1

(Ω

δ

)

denedby

hA

δ

(u), vi := (∇u, ∇v)

∀u, v ∈ H

1

0

(Ω

δ

).

(3)

Weshall ompareitseigenvalueswiththoseof thelimitoperator

A

0

: H

1

0

(Ω) → H

−1

(Ω)

that is dened similarly but in a domain without any perturbation i.e.

hA

0

(u), vi :=

(∇u, ∇v)

∀u, v ∈ H

1

0

(Ω)

.

Tostudythesetwooperators,weneedtointrodu eadapted

δ

-dependentweightedSobolev norms. Thesearenot lassi alweightednormssu hasthose hara terizingKondratiev's spa es,seee.g. [13℄. Thenormswe onsiderhereinvolve

δ

-dependentlogarithmi weights: givena

β ∈ R

,andaLips hitzopenset

ω ⊂ R

2

,wedene

kvk

2

V

1

β,δ

(ω)

:= |v|

2

V

1

β,δ

(ω)

+ |v|

2

V

0

β,δ

(ω)

where

|v|

2

V

0

β

(ω)

:=

Z

ω

| ln d

δ

(x)|

−2β

|v(x)|

2

|d

δ

(x) ln d

δ

(x)|

2

dx

|v|

2

V

1

β

(ω)

:=

Z

ω

| ln d

δ

(x)|

−2β

|∇v|

2

dx

with

d

δ

(x) := d

(x) + δ

where

d

(x) := min(d(x), e

−1

/2).

(4)

For

δ > 0, β ∈ R

xed, the norm above is obviously equivalent to

k k

H

1

(ω)

. However thesenorms"behave"dierentlyas

δ → 0

. Thenorms(4)arespe i allytailoredforour analysis.

(5)

The ondition

d

(x) ≤ e

−1

/2 ∀x ∈ Ω

guarantees that

|d

(x) ln d

(x)| ≤ |d

δ

(x) ln d

δ

(x)|

for all

x

∈ Ω

and for

δ < e

−1

/2

. Ontheother hand,re allthat

d

isadistan efun tion to a ompa t setsoit isLips hitzwith

k∇dk

L

(Ω)

= 1

, see[4, Prop.1D.4℄forexample. Hen e

d

is aLips hitz fun tion,and

d

δ

is non-vanishing with

d

δ

(x) ≥ δ, ∀x ∈ Ω

. We dedu e that

Υ

δ

β

(v) := v(x) ln

β

d

δ

(x)

isomorphi ally maps

H

1

0

(Ω)

onto itself, and there exist onstants

c

±

, δ

0

> 0

independentof

δ

su hthat

0 < c

δ

β

(v)k

V

1

β,δ

(Ω)

kvk

V

1

0

(Ω)

≤ c

+

∀v ∈ H

1

0

(Ω),

∀δ ∈ (0, δ

0

).

(5)

Weshallalso onsiderthe orrespondingdualnorms,ea hofwhi hprovidinganormfor

H

−1

(Ω)

,and arryingatthesametimesomedependen ywithrespe tto

δ

,

kf k

V

−1

β,δ

(ω)

:=

v∈H

sup

1

0

(ω)\{0}

|hf, vi|

kvk

V

1

β,δ

(ω)

.

Let us examine the ontinuity properties of the operators

A

0

, A

δ

with respe t to these norms. Firstofall, both

A

0

and

A

δ

easily appear asuniformly ontinuouswith respe t to thesenormsi.e. adire t al ulusshowsthat

lim sup

δ→0



sup

v∈H

1

0

(Ω)\{0}

kA

0

(v)k

V

−1

β,δ

(Ω)

kvk

V

1

β,δ

(Ω)

+

sup

v∈H

1

0

(Ω

δ

)\{0}

kA

δ

(v)k

V

−1

β,δ

(Ω

δ

)

kvk

V

1

β,δ

(Ω

δ

)



< +∞.

Uniforminvertibilityof theseoperatorsalsoholdunder ertainhypothesisontheweight exponent

β

.

Proposition3.1.

There exists

β

> 0

su h that, for

|β| < β

, the inverses of both

A

0

and

A

δ

remain uniformly boundedwith respe t to

δ

intermsof the weighted norms

lim sup

δ→0



sup

f∈H

−1

(Ω)\{0}

kA

−1

0

(f )k

V

1

β,δ

(Ω)

kf k

V

−1

β,δ

(Ω)

+

sup

f∈H

−1

(Ω

δ

)\{0}

kA

−1

δ

(f )k

V

1

β,δ

(Ω

δ

)

kf k

V

−1

β,δ

(Ω

δ

)



< +∞ .

Proof:

Weneedtoprovideanupperboundforboth

A

−1

δ

and

A

−1

0

. Weprovesu habound only for

A

−1

δ

, asthe derivation for

A

−1

0

is very similar and slightly easier. First of all observethat,a ordingto theuniformbounds (5),thisisequivalentto provingthat,for

β ∈ R

xedsu hthat

|β| < β

,thereexistsa onstant

C

β

> 0

independentof

δ

su hthat

C

β

sup

v∈H

1

0

(Ω

δ

)\{0}

|hA

δ

· Υ

δ

β

(u), Υ

δ

−β

(v)i|

kuk

V

1

0,δ

(Ω)

kvk

V

1

0,δ

(Ω)

.

(6) Wehave

hA

δ

· Υ

δ

β

(u), Υ

δ

−β

(v)i = (∇Υ

δ

β

(u), ∇Υ

δ

−β

(v))

δ

,a ordingtothedenitionof

A

δ

. Expanding thisexpressionyields

(∇Υ

(u), ∇Υ

−β

(v))

δ

= (∇u, ∇v)

δ

Z

∇d

(x) ·

u∇v − v∇u

d

δ

(x) ln d

δ

(x)

dx − β

2

Z

u v

|d

δ

(x) ln d

δ

(x)|

2

dx.

(7)

Observe that, taking

v = u

, the se ond term in the right hand side above is purely imaginary. In addition, sin e

|d

(x) ln d

(x)| ≤ |d

δ

(x) ln d

δ

(x)|

for all

x

∈ Ω

, using Proposition4.1belowandthefa tthat

v ∈ H

1

0

(Ω

δ

) ⊂ H

1

0

(Ω)

,weobtain

ℜe{ (∇Υ

(u), ∇Υ

−β

(v))

δ

} ≥ (1 − |β/β

|

2

) k∇uk

2

L

2

(Ω

δ

)

1 − |β/β

|

2

1 + 1/β

2

kuk

2

V

1

0

(Ω

δ

)

.

(8)

(6)

witha onstantindependentof

δ

.



4 Hardy type inequalities

This se tionisdedi ated toprovingtwoinequalitiesthat takeaform similartoHardy's inequality,butadapted toourgeometri alsetting. Therstresultbelowmaybe under-stoodasa ylindri alversionofHardy'sinequality. Theproof,though,ismadetri kyby the rather generalgeometry under onsiderationhereand, in parti ular, the possibility for

Γ

toadmitself- rossingpoints.

Proposition4.1.

Thereexistsa onstant

β

> 0

su hthat

1

β

2

:=

sup

v∈H

1

0

(Ω)\{0}

n

1

k∇vk

2

L

2

(Ω)

Z

|v(x)|

2

dx

|d

(x) ln d

(x)|

2

o

< +∞.

(9) Proof:

Duringthisproof,weshallreferto

e

3

(t) := ∂

t

γ(t)/|∂

t

γ(t)|

, onsidera

C

0

-ve toreld

t 7→ e

1

(t) ∈ R

3

su hthat

e

1

(t) · e

3

(t) = 0, ∀t ∈ R

,andset

e

2

(t) := e

3

(t) × e

1

(t)

. Denoting

I

ǫ

:= (−ǫ, +ǫ)

and

D

ǫ

⊂ R

2

thediskof enter

0

andradius

ǫ

, forea h

t ∈ R

,there exists asmall ylinder

Q

ˆ

t

= D

ǫ

t

× I

ǫ

t

with

ǫ

t

> 0

su h thatthemap

φ

t

: ˆ

Q

t

→ R

3

denedby

φ

t

(x, y, z) := γ(t + z) + xe

1

(t) + ye

2

(t)

isanimmersion. Pi knitelymany

t

1

, t

2

, . . . t

n

su hthat

[0, L] ⊂ ∪

n

j=1

(t

j

− ǫ

t

j

, t

j

+ ǫ

t

j

)

. Denote

Q

ˆ

j

:= ˆ

Q

t

j

,

Q

j

:= φ

t

j

( ˆ

Q

j

)

and

Γ

j

:= Γ∩Q

j

sothat,inparti ular,

Γ ⊂ Q

1

∪· · ·∪Q

n

. Let

ω

η

= {x ∈ Ω | d(x) < η}

. Letusprovethat,if

η

is hosensmallenough,thenforany

x

∈ ω

η

thereisone

j

su hthat

x

∈ Q

j

and

d(x) = inf

y∈Γ

j

|x − y|

. Toshowthis,pro eed by ontradi tion,assumingforamomentthat su hisnotthe ase.

This means that there exists a sequen e

x

p

∈ Ω

with

lim

p→∞

d(x

p

) = 0

and su h that, for any

x

p

∈ Γ

satisfying

|x

p

− x

p

| = d(x

p

)

, none of the

Q

j

's both ontain

x

p

and

x

p

. Sin e

Γ

is ompa t,extra tingasub-sequen eifne essary,wemayassumethat

lim

p→∞

x

p

= lim

p→∞

x

p

= x

∈ Γ

. Takea

Q

m

ontaining

x

. Then, sin e

Q

m

isan openneighborhood of

x

, forsu ientlylarge

p

wehaveboth

x

p

∈ Q

m

and

x

p

∈ Q

m

whi h ontradi tsourinitialassumption.

Fromnow on, weassume that

η > 0

is hosensmall enough to guaranteethe property dis ussedinthepreviousparagraph. Denote

d

j

(x) = inf

y∈Γ

j

|x − y|

,andlet

1

Q

j

referto the hara teristi fun tion of

Q

j

. Whatpre edesshowsthat,forany

x

∈ ω

η

,wehave

1

|d

(x) ln d

(x)|

n

X

j=1

1

Q

j

(x)

|d

j

(x) ln d

j

(x)|

.

(10)

Choose a smooth open set

Q

0

⊂ R

3

su h that

Q

0

∪ Q

1

· · · ∪ Q

n

a hieves a overingof

. Using a partition of unity subordinated to this overing, one may de ompose any

v ∈ H

1

0

(Ω)

in theform

v = v

0

+ · · · + v

n

where

v

j

∈ H

1

0

(Q

j

)

. Thisremark,togetherwith Inequality(10),showsthatitsu estoprovetheexisten eofa onstant

C > 0

su hthat

Z

Q

j

|v(x)|

2

dx

|d

j

(x) ln d

j

(x)|

2

≤ C

Z

Q

j

|∇v|

2

dx

∀v ∈ H

1

0

(Q

j

) ∀j = 1 . . . n.

(11)

(7)

Fromnowon,anduntiltheendoftheproof,wetakea

j

xed. Letusrelabelforamoment

ǫ = ǫ

t

j

,

φ = φ

t

j

andset

Q = ˆ

ˆ

Q

j

= D

ǫ

× I

ǫ

,

Σ := {0} × I

ǫ

and

v = v ◦ φ

ˆ

. Forany

x

ˆ

∈ ˆ

Q

wealso set

d(ˆ

ˆ

x) = inf

y∈Σ

ˆ

x

− ˆ

y|

. Thedieomorphism

φ

and its inverse

φ

−1

are both Lips hitz,sothereare onstants

c

±

> 0

su hthat

c

x

− ˆ

y| ≤ |φ(ˆ

x) − φ(ˆ

y)| ≤ c

+

x

− ˆ

y|

. Routineveri ationsshowthatthisimpliesexisten eoftwo onstants

c

, c

′′

> 0

su hthat

c

d(ˆ

ˆ

x) ≤ d φ(ˆ

x)



≤ c

′′

d(ˆ

ˆ

x)

∀ˆ

x

∈ ˆ

Q.

(12)

Let

r, θ

refertothepolar oordinatesin

R

2

sothatthe ylinder

Q

ˆ

isparametrizedbythe ylindri al oordinates

(r, θ, z) ∈ [0, +ǫ) × [0, 2π] × I

ǫ

. The lassi al Hardy's inequality applied in

Q

ˆ

,see[8℄,showsthat

Z

ˆ

Q

v|

2

rdrdθdz

|r ln(r)|

2

≤ C

Z

ˆ

Q

|∇ˆ

v| + |ˆ

v|

2

x

∀ˆ

v ∈ H

1

( ˆ

Q).

(13)

There only remains to observe that

d(ˆ

ˆ

x) = r

, to use (12), and to apply the hange of variables

x

= φ(ˆ

x)

in the integrals (13). Denoting

the dierential of

φ

, sin e

kDφk

L

( ˆ

Q)

and

kDφ

−1

k

L

(Q

j

)

arebothbounded,thisnally yieldsinequality(11)with

k∇vk

2

L

2

(Q

j

)

+ kvk

2

L

2

(Q

j

)

insteadof just

k∇vk

2

L

2

(Ω)

. We on ludeby using Poin are's

in-equalityin

Q

j

.



Wewillalsoneedanother

δ

-dependentweightedinequality. Thisoneinvolvesaweighted

L

2

-normevaluatedonlyovera oronal ylinderofradius

δ

,andisnotprimarilybasedon the lassi alHardy'sinequality. InsteaditisderivedbymeansofKondratiev'sanalysis. Proposition4.2. Denoting

Q

δ

:= {x ∈ Ω | δ < d(x) < 2δ}

wehave

lim sup

δ→0

sup

v∈H

2

(Ω)∩H

1

0

(Ω)\{0}

n

1

k∆vk

2

L

2

(Ω)

Z

Q

δ

v(x)

d

(x)

2

dx

o

< +∞.

Proof:

Toestablishthisresult,wefollowapathsimilartothatoftheproofofProposition4.1, andusethesamenotations. Takeany

x

∈ ω

η

. A ordingtotherstpartofthepre eding proof,there exists

j

su h that

x

∈ Q

j

and

d

j

(x) = d

(x)

, hen e

δ < d(x) < 2δ ⇒ δ <

d

j

(x) < 2δ

. As a onsequen e

Q

δ

⊂ ∪

n

j=1

Q

j

δ

where

Q

j

δ

:= {x ∈ Q

j

, δ < d

j

(x) < 2δ}

whi h implies

1

Q

δ

P

n

j=1

1

Q

j

δ

. Hen e it su es to prove, for ea h

j

, the existen e of onstants

C, δ

0

> 0

independentof

δ

su hthat

Z

Q

j

δ

v(x)

d

j

(x)

2

dx ≤ C

Z

|∆v|

2

dx

∀v ∈ H

2

(Ω) ∩ H

1

0

(Ω),

∀δ ∈ (0, δ

0

).

Fixing

j

, andusinga hangeofvariableslikein thepreviousproof,theinequalityabove boilsdowntoestablishingtheexisten eof onstants

C, δ

0

> 0

independentof

δ

su hthat, forany

v ∈ H

2

( ˆ

Q) ∩ H

1

0

( ˆ

Q)

andany

δ ∈ (0, δ

0

)

wehave

Z

ˆ

Q

δ

|v|

2

r

2

rdrdθdz ≤ C

Z

ˆ

Q

|∆v|

2

+ |∇v|

2

+ |v|

2

x.

(14)

where were allthat

Q = D

ˆ

ǫ

× I

ǫ

,thevariables

r, θ, z

refertothe ylindri al oordinates in

Q

ˆ

,and

Q

ˆ

δ

:= {ˆ

x

∈ ˆ

Q | δ < r < 2δ}

. Toestablish(14),set

v = r

−2

((r∂

r

)

2

+ ∂

θ

2

)v

. De omposeea h point

x

ˆ

∈ ˆ

Q

as

x

ˆ

= (ˆ

x

, z)

,sothat

x

ˆ

= (r cos θ, r sin θ)

and

r = |ˆ

x

|

. Introdu etheFourierde ompositionof

v

inthe

z

variable,setting

v

p

x

) :=

1

Z

−ǫ

(8)

Then wehave

v

p

∈ H

2

(D

ǫ

) ∩ H

1

0

(D

ǫ

)

and

−∆

v

p

∈ L

2

(D

ǫ

)

for all

p ∈ Z

, if

v ∈ H

2

( ˆ

Q) ∩

H

1

0

( ˆ

Q)

. Applying Kondratiev's analysis in

D

ǫ

\ {0}

, see Chapter 6 of [13℄, we nd the existen eof oe ients

α

p

∈ C

anda onstant

C > 0

independentof

p

su hthat

p

|

2

+

Z

D

ǫ

|v

p

x

) − α

p

|

2

x

|

3

x

≤ C

Z

D

ǫ

|∆

v

p

|

2

x

∀p ∈ Z.

Parseval identity asso iated to De omposition (15), together with the estimate above, indi ates that

P

+∞

p=−∞

p

|

2

< +∞

sothat thereexistsafun tion

α = α(z) ∈ L

2

(I

ǫ

)

and onstants

C, C

> 0

independentof

v

satisfying

kαk

2

L

2

(I

ǫ

)

+

Z

ˆ

Q

|v(ˆ

x

, z) − α(z)|

2

x

|

3

x

dz

≤ Ckf k

2

L

2

( ˆ

Q)

= Ck∆

vk

2

L

2

( ˆ

Q)

≤ C

k∆vk

2

L

2

( ˆ

Q)

∀v ∈ H

2

( ˆ

Q) ∩ H

1

0

( ˆ

Q).

(16)

The inequality aboveis justied by standardellipti apriori estimates for the Lapla e operator,see[13,Chap.3℄forexample.Finally,letuspi kanarbitrary

v ∈ H

2

( ˆ

Q)∩H

1

0

( ˆ

Q)

. Plugging(16)into theleft handsideof(14)yields

Z

ˆ

Q

δ

|v|

2

r

2

rdrdθdz =

Z

ˆ

Q

δ

|v(ˆ

x)|

2

x

|

2

x

Z

ˆ

Q

δ

|α(z)|

2

x

|

2

x

+

Z

ˆ

Q

|v(ˆ

x) − α(z)|

2

x

|

3

x

≤ kαk

2

L

2

(I

ǫ

)

 Z

δ

dr

r



|

{z

}

=ln(2)

+Ck∆vk

2

L

2

( ˆ

Q)

≤ C

k∆vk

2

L

2

( ˆ

Q)

.

Sin e

v

was hosenarbitrarilyin

H

2

( ˆ

Q) ∩ H

1

0

( ˆ

Q)

,andthe onstant

C

> 0

isindependent of

δ

,this on ludestheproof.



5 Norm onvergen e of the resolvent Wewillnowusethepreviousanalysistoshowthat

A

−1

δ

strongly onvergestoward

A

−1

0

in someappropriateoperatornorm. Beforestatingthisresultletusjustpointoutthat,using extension by

0

, we have

H

1

0

(Ω

δ

) ⊂ H

1

0

(Ω)

sothat

H

−1

(Ω) ⊂ H

−1

(Ω

δ

)

. The expression

A

−1

δ

(f )

with

f ∈ H

−1

(Ω)

should beunderstooda ordingto thesein lusions. Proposition5.1.

For

β

> 0

as in Proposition 3.1, and for ea h

β ∈ R

satisfying

|β| < β

, there exist onstants

c

β

, δ

0

> 0

independentof

δ

,su hthat

sup

f∈L

2

(Ω)\{0}

kA

−1

0

(f ) − A

−1

δ

(f )k

V

1

β,δ

(Ω)

kf k

L

2

(Ω)

c

β

| ln δ|

β

∀δ ∈ (0, δ

0

).

Proof:

Firstofall onsidera

C

ut-ofun tion

χ : R → R

su hthat

χ(t) = 0

for

t ≤ 1

and

χ(t) = 1

for

t ≥ 2

,andset

χ

δ

(x) := χ( d(x)/δ )

and

ψ

δ

:= 1 − χ

δ

. Intheremainingofthis proof,weshalldenote

Q

δ

:= supp(∇χ

δ

)

sothat, forany

x

∈ Q

δ

,wehave

δ ≤ d(x) ≤ 2δ

. Weintrodu etheoperator

R

δ

: H

−1

(Ω) → H

1

0

(Ω

δ

)

dened by

R

δ

(f ) := χ

δ

A

−1

0

(f )

∀f ∈ H

−1

(Ω).

Nowtake any

f ∈ L

2

(Ω)

, and set

u

δ

:= A

−1

δ

(f ) ∈ H

1

0

(Ω

δ

)

,

u

ˆ

δ

:= R

δ

(f ) ∈ H

1

0

(Ω

δ

)

and

(9)

u

δ

− ˆ

u

δ

. There exist onstants

c, c

> 0

independentof

δ

su hthat, forany

v ∈ H

1

0

(Ω

δ

)

, wehave

|hA

δ

(u

δ

− ˆ

u

δ

), vi| = |(∇(u

δ

− ˆ

u

δ

), ∇v)

|

= |(f, ψ

δ

v)

+ (∇χ

δ

, u

0

∇v − v∇u

0

)

Q

δ

|

≤ kf k

L

2

(Ω)

δ

vk

L

2

(Ω)

+ |(∇χ

δ

, u

0

∇v − v∇u

0

)

Q

δ

|.

(17)

To derive an upper bound for the se ond term in the right hand side above, observe that

δ ≤ d(x) ≤ 2δ

for

x

∈ supp(∇χ

δ

)

. Sin e

|∇d| ≤ 1

, we have

sup

x∈Ω

|∇χ

δ

| ≤

2d(x)

−1

sup

x∈Ω

|∂

t

χ|

. So Cau hy-S hwarzinequality, together with Proposition 4.1 ap-pliedto

∇u

0

,andProposition 4.2applied to

u

0

,yield theexisten eof a onstant

C > 0

independentof

δ

su hthat

|(∇χ

δ

, u

0

∇v − v∇u

0

)

Q

δ

| ≤ C( k∇vk

L

2

(Q

δ

)

+ kvk

L

2

(Q

δ

)

)k∆u

0

k

L

2

(Ω)

≤ C

| ln δ|

−β

kvk

V

1

−β,δ

(Ω)

kf k

L

2

(Ω)

.

(18)

Thersttermintherighthandsideof(17) anbeboundedbynotingthat

δ ≤ d

δ

(x) ≤ 2δ

on

supp(ψ

δ

)

. SoapplyingHardy'sinequality(9)yieldsa onstant

C > 0

independentof

δ

su hthat

δ

vk

L

2

(Ω)

≤ Cδ| ln δ|

1−β

kvk

V

1

β,δ

(Ω)

forall

v ∈ H

1

0

(Ω

δ

)

. Pluggingthistogether withEstimate (18)into Inequality(17)providesa onstant

C > 0

independentof

δ

su h that

|hA

δ

(u

δ

− ˆ

u

δ

), vi| ≤ C| ln δ|

−β

kvk

V

1

β,δ

(Ω)

kf k

L

2

(Ω)

.

Observethat

A

δ

(u

δ

− ˆ

u

δ

) = (Id − A

δ

R

δ

)f ∈ H

−1

(Ω)

. Sin e

v ∈ H

1

0

(Ω

δ

)

wasarbitraryin the al ulusabove,whatpre edesshowsthat thereexists a onstant

C > 0

independent of

δ

su hthat

k(Id − A

δ

R

δ

)f k

V

−1

β,δ

(Ω

δ

)

≤ Ckf k

L

2

(Ω)

| ln δ|

−β

.

(19) Followingthesame al ulus,but hoosing atestfun tion

v ∈ H

1

0

(Ω)

, thesameresultas (19) holdswith

A

δ

, Ω

δ

repla ed by

A

0

, Ω

. To on ludetheproof, there onlyremains to usethestabilityestimatesofProposition3.1thatyield onstants

C > 0

independentof

δ

su hthat,forany

f ∈ L

2

(Ω)

, wehave

kA

−1

0

(f ) − A

−1

δ

(f )k

V

1

β,δ

(Ω)

≤ kA

−1

0

(f ) − R

δ

(f )k

V

1

β,δ

(Ω)

+ kA

−1

δ

(f ) − R

δ

(f )k

V

1

β,δ

(Ω

δ

)

≤ C( kf − A

0

R

δ

(f )k

V

−1

β,δ

(Ω)

+ kf − A

δ

R

δ

(f )k

V

−1

β,δ

(Ω

δ

)

)

≤ C| ln δ|

−β

kf k

L

2

(Ω)

.



The proposition above yields onsisten y estimates for the asymptoti sour e problem. Indeed, let

f ∈ L

2

(Ω)

refer to a xed fun tion not depending on

δ

, let

u

0

∈ H

1

0

(Ω

δ

)

satisfy

−∆u

0

= f

in

, andlet

u

δ

∈ H

1

0

(Ω

δ

)

satisfy

−∆u

δ

= f

in

δ

. Then Proposition 5.1 impliesthat

ku

δ

− u

0

k

H

1

(Ω\U)

= O(| ln δ|

−β

)

for any neighborhood

U

of

Γ

, and for

|β| < β

.

Sharper results an be obtained in terms of the

L

2

-norm. Observethat there exists a onstant

C > 0

independent of

δ

su h that

kvk

L

2

(Ω)

≤ Ckvk

V

1

β,δ

(Ω)

for any

β ∈ R

. PluggingthisintotheestimateofProposition5.1yieldsthefollowingresult.

Corollary 5.1.

(10)

independentof

δ

su hthat

sup

f∈L

2

(Ω)\{0}

kA

−1

0

(f ) − A

−1

δ

(f )k

L

2

(Ω)

kf k

L

2

(Ω)

c

ǫ

| ln δ|

β

−ǫ

∀δ ∈ (0, δ

0

).

Re allthat

L

2

(Ω) ⊂ H

−1

(Ω

δ

)

,and

H

1

0

(Ω

δ

) ⊂ L

2

(Ω)

,sothat

A

−1

δ

isa ontinuousoperator mapping

L

2

(Ω)

to

L

2

(Ω)

. Assu h,itisself-adjointand ompa tand,as anbe he kedby routineveri ations,itseigenvalueswiththoseof

A

−1

δ

onsideredasanoperatormapping

L

2

(Ω

δ

)

to

L

2

(Ω

δ

)

. Then, sin e

A

−1

0

: L

2

(Ω) → L

2

(Ω)

is also self-adjoint and ompa t, straightforwardappli ationofTheorem 4.10ofChapter Vof[12℄ yieldsthatthespe tra of

A

−1

δ

and

A

−1

0

are losedtoea hother. Proposition5.2.

For

β

> 0

asinProposition 3.1,andforany

ǫ ∈ (0, 2β

)

,thereexist onstants

δ

0

, c

ǫ

> 0

independentof

δ

su hthat

sup

µ∈S(A

0

)

inf

λ∈S(A

δ

)

1

µ

1

λ

+

sup

µ∈S(A

δ

)

inf

λ∈S(A

0

)

1

µ

1

λ

c

ǫ

| ln δ|

β

−ǫ

∀δ ∈ (0, δ

0

).

Referen es

[1℄ H. Ammari. An introdu tion to mathemati s of emerging biomedi al imaging, vol-ume 62 of Mathématiques & Appli ations (Berlin) [Mathemati s & Appli ations℄. Springer,Berlin,2008.

[2℄ H. Ammari and H. Kang. Re onstru tion of small inhomogeneities from bound-ary measurements,volume 1846of Le ture Notesin Mathemati s. Springer-Verlag, Berlin,2004.

[3℄ I.I.ArgatovandF.J.Sabina.A ousti dira tionbyathinsofttorus.WaveMotion, 45(6):846856,2008.

[4℄ A.L. Dont hev and R.T. Ro kafellar. Impli it fun tions and solution mappings. Springer Series in Operations Resear h and Finan ial Engineering. Springer, New York,se ondedition,2014. Aviewfromvariationalanalysis.

[5℄ M.V.Fedoryuk.Asymptoti softhesolutionoftheDiri hletproblemfortheLapla e andhelmholtzequationsintheexteriorofaslender ylinder. Izv.Akad.NaukSSSR Ser.Mat.,45(1):167186,1981.

[6℄ M.V.Fedoryuk. Theory ofCubatureFormulas and the Appli ations of Fun tionnal Analysisto Problems of Mathemati al Physi s,volume126ofAMSTrans., hapter The Diri hlet problem for the Lapla e operator in the exterior of a thin body of revolution. Ameri anMathemati alSo iety,1985.

[7℄ J.Geer. The s atteringof as alarwavebyaslenderbody ofrevolution. SIAM J. Appl.Math.,34(2):348370,1978.

[8℄ G.H.Hardy,J.E.Littlewood,andG.Pólya. Inequalities. CambridgeMathemati al Library.CambridgeUniversityPress,Cambridge,1988. Reprintofthe1952edition. [9℄ A.M.Il

in. Mat hing of asymptoti expansionsof solutionsof boundary value prob-lems, volume 102of Translations of Mathemati al Monographs. Ameri an Mathe-mati alSo iety,Providen e,RI,1992. Translatedfrom theRussianbyV.Mina hin [V.V. Minakhin℄.

[10℄ A.M.Il'in. Studyoftheasymptoti behaviorofthesolutionofanellipti boundary valueprobleminadomainwithasmallhole.TrudySem.Petrovsk.,(6):5782,1981.

(11)

OxfordUniversityPress,NewYork,1979. OxfordEngineeringS ien eSeries. [12℄ T.Kato.Perturbationtheoryforlinearoperators.Classi sinMathemati s.

Springer-Verlag,Berlin,1995. Reprintofthe1980edition.

[13℄ V. A. Kozlov, V. G.Maz'ya,and J. Rossmann. Ellipti Boundary Value Problems inDomainswithPointSingularities,volume52ofMathemati alSurveysand Mono-graphs. AMS,Providen e,1997.

[14℄ V.Maz

ya,S.Nazarov,andB.Plamenevskij. Asymptoti theoryofellipti boundary value problems in singularly perturbed domains. Vol. I & II, volume 111 &112 of OperatorTheory: Advan esandAppli ations. BirkhäuserVerlag,Basel,2000. [15℄ V.G. Maz

ya,S. A.Nazarov,and B.A.Plamenevski. Theasymptoti behaviorof solutionsof the Diri hlet problem in a domain witha utout thin tube. Mat. Sb. (N.S.),116(158)(2):187217,1981.

[16℄ V.G.Maz'ya,S.A.Nazarov,andB.A.Plamenevski.Asymptoti expansionsof eigen-valuesof boundary valueproblems forthe Lapla eoperatorin domains with small openings. Izv.Akad. NaukSSSRSer. Mat.,48(2):347371,1984.

[17℄ S.A.Nazarov. Asymptoti onditionsatapoint,selfadjointextensionsofoperators, andthemethodofmat hedasymptoti expansions.InPro eedingsofthe St. Peters-burgMathemati al So iety, Vol.V,volume193ofAmer.Math. So .,Transl. Ser.2, pages77125,Providen e,RI,1999.

[18℄ S.A.NazarovandJ.Sokoªowski.Asymptoti analysisofshapefun tionals. J.Math. PuresAppl.,82(2):125196,2003.

[19℄ S.A.NazarovandJ.Sokoªowski. Self-adjointextensionsfortheNeumannLapla ian andappli ations. A ta Math.Sin. (Engl.Ser.),22(3):879906,2006.

[20℄ A.A.NovotnyandJ.Sokoªowski. Topologi al derivativesin shape optimization. In-tera tionofMe hani sandMathemati s.Springer,Heidelberg, 2013.

[21℄ M.Yu. Planida. On the onvergen eof solutionsofsingularlyperturbed boundary valueproblemsfortheLapla ian. Mat. Zametki,71(6):867877,2002.

[22℄ G.V. Zhdanova. TheDiri hlet problemfor theHelmholtz operatorin the exterior ofathin body ofrevolution. Dierentsial

Références

Documents relatifs

The main purpose of this work is to study and give optimal estimates for the Dirichlet problem for the biharmonic operator A 2 on an arbitrary bounded Lipschitz domain D in R\ with

The proof of Theorem 3 about the asymptotics of the first eigenvalues of the cone is over, when, using a Feshbach-Grushin projection, we justify that the Born-Oppenheimer

In [17], Exner, ˇ Seba and ˇSˇtov´ıˇcek prove that for the L-shaped waveguide there is a unique bound state below the threshold of the essential spectrum. Two of the authors,

We consider the solution of an interface problem posed in a bounded domain coated with a layer of thickness ε and with external boundary conditions of Dirichlet or Neumann type..

Tandis que le deuxième chapitre a eu pour objectif principal l’élaboration de la relation hauteur- débit ainsi que celle du coefficient de débit par une approche théorique en

In the Eastern Mediterranean Region, in response to the growing public concern related to EMF in the mid 1990s, particularly among countries of the Gulf Cooperation Council (GCC) and

This quest initiated the mathematical interest for estimating the sum of Dirichlet eigenvalues of the Laplacian while in physics the question is related to count the number of

This proves that V (t) given by (3.4), can be used in Chernoff’s Theorem in order to compute an approximating expression for the system