† ‡ ‡

‡

†

‡

@

1

ological

atten

deriv

in

ativ

er,

es

lev

in

spaces

shap

e

o

optimization

ely

Piotr

FULMANSKI

assured

top

,

allo

An

ev

toine

er

LA

set

URAIN

pur

with

,

is

Jean-F

in

ran?ois

for

SCHEID

Mosco

d

,

of

Jan

ds

SOKO?O

of

WSKI

set

metho

y

set

el

Univ

ersiti

the

of

and

??d?,

solution

F

set

acult

b

y

ords.

of

optimization

Mathematics

t

with

ha

and

22,

The

90-232

problems

??d?,

e.g.

P

of

oland,

the

fulmanp@imul.math.uni.lodz.pl

ev

el

y

Institut

tly

Elie

to

Cartan

the

UMR

top

7502,

of

Nancy-Univ

is

ersit?,

mo

CNRS,

b

INRIA,

equations.

B.P

nature,

.

the

239,

onen

54506

optimization

V

the

ando

on

euvre-l?s-Nancy

is

Cedex,

additional

F

b

rance,

to

lev

d

sokolows

on

Lev

v

set

Key

ological

e

optimal

tro

d,

ativ

1.

A

Shap

elliptic

of

imp

shap

h

e

v

optimization

umerous

problems

is

solution

solv

erse

ed

of

n

e

umerically

b

b

mild

y

the

the

v

lev

Sob

el

set

metho

ho

d

optimalit

more

bined

b

with

domains.

the

a

top

is

ological

umerical

deriv

h

ativ

pro

es

for

for

top

domains.

ology

optimization.

framew

A

lev

whic

tually

the

,

and

the

determined

top

appropriate

ology

w

v

y

ariations

obtained

are

function

in

um

tro

of

on

during

the

basis

v

of

asymptotic

d

analysis,

lev

b

asymptotic

y

y

an

ev

in

aluation

the

of

etter

extremal

p

the

oin

e

ts

el

(lo

metho

maxima

based

for

the

the

oundary

sp

ariation

hnique.

problem)

W

of

Shap

the

optimization,

so-

el

metho

top

top

ological

deriv

deriv

e,

ativ

design.

es

In

in

tro

e

for

b

equations

y

an

Sok

ortan

olo

wski

of

and

of

Zo

ariations,

n

ho

applications

wski

[24

℄

to

for

of

elliptic

v

b

problems.

oundary

existence

v

solutions

alue

shap

problems.

optimization

T

op

e

ological

under

deriv

ativ

under

es

so-called

are

giv

ergence

en

the

for

olev

energy

asso

functionals

with

of

minimizing

linear

of

b

domains,

oundary

w

v

er

alue

y

problems.

require

W

regularit

e

of

presen

oundaries

t

admissible

results,

,

n

particular

umerical

tion

examples,

paid

whic

n

h

metho

whic

that

during

the

optimization

application

of

w

top

the

ological

ology

deriv

hanges

ativ

es

One

in

the

the

for

framew

h

ork

ork

of

the

the

el

lev

function

el

h

set

dels

metho

domain

d

olution

really

impro

e

v

b

es

solving

the

Hamilton-Jacobi

Ho

of

ev

the

b

metho

its

d.

the

Examples

lev

sho

set

w,

that

n

the

b

lev

of

el

set

ts

metho

resulting

d

domains

the

bined

pro

with

T

the

impro

asymptotic

e

analysis,

p

is

of

robust

metho

for

based

the

the

shap

el

e

function,

optimization

analysis

problems,

emplo

and

ed,

it

lo

allo

of

ws

holes

to

iden

tifylo

domains

p

is

emplo

determined

resp

b

metho

y

the

analysis

top

of

ery

the

the

so-called

turns

top

used

ological

osed

deriv

[17],

ativ

for

es

b

of

n

shap

a

e

ological

functionals

for

under

b

study

us,

.

domain

In

ev

the

references

pap

y

er

Steklo

w

t

e

describ

or

e

whole

in

it

details

hanges

es

h

a

determined

one

bination,

analysis

used

b

already

of

in

ativ

literature,

er.

and

also

pro

deriv

vide

the

some

giv

examples

for

whic

,

h

the

sho

of

w

℄

that

The

the

in

metho

9.

d

scalar

is

functional.

robust,

dieren

in

the

particular

e

impro

en

v

h

es

ho

the

our

optimal

the

v

y

alue

top

of

functional.

the

holes

shap

actual

e

the

functional

mization

obtained

W

as

h

a

ev

result

whic

of

a

if

F

e

or

b

the

order

v

the

enience

is

of

w

the

results

reader

determine

w

o

e

hnique

presen

the

t

oincar?

as

an

w

ological

ell

e

the

℄

elemen

t

ts

ation

of

shap

mathematical

in

analysis

timization

asymptotic

[15

analysis

[24

of

asymptotic

sp

erator

for

b

presen

oundary

problem

v

metho

alue

v

problems

yp

with

h

resp

analysis

ect

expansions

to

parameter

small

of

parameter

of

whic

mo

h

of

mo

dened

dels

domains,

singular

unfortunately

p

e

erturbations

ev

of

satised

During

domains.

pro

ology

h

dened

analysis

of

is

maxima

required

deriv

for

the

determination

this

of

,

top

b

ological

to

deriv

domain

ativ

ters

es.

oin

The

y

n

the

umerical

ativ

metho

d

at

for

of

shap

ho

e

the

and

the

top

allo

ology

optimization

n

of

of

an

and

energy

to

functional

The

is

holes

pre-

determi-

sen

the

ted

ological

in

and

the

is

pap

ed

er.

t

The

pap

metho

tained,

d

app

pro

bines

the

h

shap

order

e

top

gradien

es.

t

end

osition

hnique

applied

and

of

the

expansion

asymptotic

Steklo

analysis

erator

for

maximization

h

of

of

an

ativ

energy

t

functional

is

for

in

elliptic

[27].

equations.

e

The

list

lev

the

el

application

set

analysis

metho

optimization.

d

of

is

framew

used

e

for

studied

the

y

ev

℄

olution

℄

of

[16

[19

domains

[25],

with

of

the

for

mo

oincar?

ving

giv

b

oundaries

e

determined

optimization

from

in

the

The

shap

e

e

n

sensitivit

for

y

b

analysis

problems,

of

energy

the

shap

energy

or

shap

w

e

vide

functional,

shap

the

and

top

solutions

ology

to

h

hanges

p

are

domain.

p

sak

erfor-

y

med

2

in

ving

addition

oundaries

b

giv

y

functions

analysis

in

of

the

whic

top

is

ological

v

deriv

ativ

assumption,

es.

w

It

er

is

is

still

in

an

op

the

en

umerical

problem

to

top

devise

ho

are

w

b

the

analysis

the

bination

of

of

b

ological

oundary

ativ

v

for

ariations

energy

and

In

singular

w

p

y

erturbations

small

of

e

domains

in

en

the

ters

in

with

a

general

at

framew

p

ork

ts

of

b

shap

maxi-

e

of

optimization.

top

One

deriv

p

es.

ossibilit

e

y

only

is

hole

giv

en

etap

b

the

y

d,

the

w

so-called

er

self-adjoin

asymptotic

t

furnish

extensions

information

of

h

dieren

ws

tial

the

op

of

erators,

nite

where

um

the

er

small

holes,

hole

is

it

appro

out

ximated

b

b

useful.

y

size

singular

the

p

erturbation

e

of

ned

the

y

top

ts

deriv

of

es

dieren

this

tial

hnique

op

not

erator.

y

W

in

e

presen

refer

pap

the

The

reader

er

to

[18],

th

[11],

in

[14],

endix

[16

e

℄

vide

[17

some

℄

hnical

for

whic

some

are

results

in

in

to

this

the

ological

W

ativ

e

T

this

only

the

steady

state

b

is

oundary

and

v

pro

alue

of

problems,

asymptotic

ho

for

w

related

ev

v-P

er

op

the

is

same

en.

results

h

b

for

e

aluation

obtained

top

for

deriv

the

es

ev

energy

olution

yp

problems

functionals

describ

prop

ed

e.g.,

b

[12

y

and

parab

Finally

olic

w

and

presen

h

a

yp

of

erb

on

olic

deriv

p

and

de's.

of

The

asymptotic

Hamilton-Jacobi

in

nonlinear

e

h

Singular

yp

erturbations

erb

domains

olic

the

equation

ork

mo

shap

dels

op-

the

are

ev

b

olution

man

of

authors,

the

℄

lev

[10

el

[13

set

[14],

function.

℄

The

℄

normal

[18],

sp

℄

eed

℄

of

[26].

the

mo

the

ving

expansion

b

the

oundaries

v-P

are

op

determined

is

from

en

the

[28].

shap

results

e

shap

gradien

and

ts

ology

obtained

are

for

ted

the

energy

2.

functional

linear

[29].

W

Since

presen

the

the

shap

umerical

e

d

gradien

a

ts

elliptic

serv

oundary

e

alue

as

and

the

the

t

e

ts

e

for

F

the

h

problem

yp

e

erb

pro

olic

equation,

on

the

e

standard

tiation

assumption

asymptotic

is

of

to

with

assume

ect

that

small

the

whic

shap

measures

e

singular

gradien

erturbations

ts

are

F

giv

the

en

e

b

y

w

to

theU V

2 2V ⊂⊂ U ω ⊂ #ω

ω

O ={Ω =U \ω; ω , ω⊂V,#ω≤k}.k

Ω∈O k≥ 1 Ω ∂Ω = Γ ∪∂U Γ =∂ωk N N

∂U Γ ΓD N

Γ ωD

ω

ΓN

ΓN

ω

Ω

∂U = ΓD

Ω

∞f ∈C (U)

−Δu+u = f Ω,

u = 0 Γ ,D

∂ u = 0 Γ ,n N

n ∂Ω ∂n

∂Ω u

Z Z Z Z

1 1 12 2 2 2E(Ω,u) := |∇u| +u dx− fudx =− |∇u| +u dx =− fudx,

2 2 2Ω Ω Ω Ω

2J(Ω) :=E(Ω,u)+λA(Ω)−μP (Ω) ,c

A(Ω) P (Ω)c

A(Ω) := |Ω|,

P (Ω) := max(0,ℓ(∂Ω)−c).c

2|Ω| Ω ℓ(∂Ω)

∂Ω λ μ

in

oundary

dimensions,

erformed

spatial

and

is

ation

also

shap

denoted

is

b

b

y

alue

o

w

w

op

.

The

the

b

tal

oundary

shap

t

tial

to

o

for

es

for

Neumann

the

b

in

oundary

spatial

admissible

while

holes

w

b

e

denote

ha

F

v

e

tro

subsets

hlet

e

mo

on

egin

es

and

and

y

.

ological

Let

h

us

reader

p

w

oin

e

t

esgue

out

b

that

dimensional

the

and

op

of

en

set

set

w

ourselv

ts

b

of

Constan

n

The

y

.

w

based

op

set

p

an

e

the

allo

that

us

R

tak

us

in

the

functional

3

op

PSfrag

w

replacemen

dieren

ts

problem.

to

the

w

holes

(4)

in

dimensions.

with

in

the

dened

form

b

of

ativ

a

of

the

In

general

to

Fig.

refer

1

y

(6)

An

o

h

the

domain

the

to

of

F

e

or

and

in

the

splitted

of

e

dimensions

b

arbitrary

e

a

domains

of

of

,

the

let

e

us

and

of

the

onen

follo

wing

er

problem

um

oundary

the

b

analysis

the

b

,

e

,

b

R

y

on

,

en

F

y

ourier

or

in

.

analysis

fundamen

w

solutions

y

h

an

for

or

(3)

on

let

F

in

(1)

e

regularized

e

obtain

of

the

en

explicit

ounded

solutions

o

on

t

for

b

auxiliary

and

problems

Let

(2)

del

where

the

in

tro

is

e

the

with,

unit

b

out

with

w

T

ards

spatial

normal

arbitrary

v

problems

ector

v

to

b

rings.

oundary

set

elliptic

,

es

and

deriv

en

top

op

(5)

stands

deriv

for

to

the

normal

a

deriv

[18]

ativ

e.g.,

e

the

on

e

Otherwise,

,

for

a

.

same

In

In

addition,

ab

let

v

us

denitions,

in

the

denotes

energy

Leb

functional

measure

dep

p

ending

R

on

and

the

solution

erators

an

is

to

1-

(2)

measure

:

y

admissible

.

is

ts

not

and

are

an

ositiv

y

and

w

set

to

as

e

illustrated

to

in

t

Figure

1.c

c

k≥ 1

max{J(Ω) : Ω∈O }.k

Ω ∂U Ω

ΓN

J

J

J

J

Ω

ΓD

Ω = U \ ωi i

ω ωi i

ω ui

−Δu +u = f Ω ,i i i

u = 0 ∂U.i

∂ u = 0 ∂ω .n i i

u Ω |Ω| → |Ω|i

u = 0 ω u = 0 ωi i

2L (U)

k≥ 0 Ω∈Ok

{Ω}, i = 0,1,2,··· ,i

Ω Ω → Ω i→ +∞i i

H

Ω −→ Ωi

mo

p

With

osed

ast

in

b

the

to

sense

metric.

that

y

there

e

exists

to

an

).

optimal

domain,

furthermore,

the

the

obtain

optimal

shap

the

e

and

e

b

F

e

domain.

shifted

that

b

Similar

y

giv

appropriate

of

our

optimalit

e

y

et

en

F

erform

or

of

a

oblem

giv

(7)

en

oundary

top

on

ology

l

of

teger

an

plac

optimal

pro

domain

p

there

Pro

are

p

t

still

w

erimeter

o

dieren

t

o

sort

Sv

of

hanges

optimalit

w

y

℄

p

On

since

the

Th

mo

theorem

ving

of

b

oundaries,

the

the

shap

b

e

ges

gradien

to

ts

of

.

b

determined

e

,

ev

xed

aluated

(8)

and

of

the

only

optimalit

the

y

e

setting

sa

y

y

onver

that

activ

the

is

optimal

existence

v

or

alue

(7)

of

,

the

i.e.

b

b

There

e

impro

resp

v

the

ed

ws,

b

denoted

ell

y

for

admissible

b

b

t

oundary

dimensions

v

b

ariations.

[30

On

few

the

the

other

this

hand,

the

from

sign

℄

of

It

top

without

ological

dicult

deriv

b

ativ

is

es

w

is

follo

xed

1

inside

(7)

of

mathematical

the

a

of

domain,

in

er

h

onents

a

is

w

d.

a

y

the

that

Hausdor

an

,

y

three

Neumann

of

the

a

b

small

y

hole

actually

domains

impro

to

v

for

e

of

the

exterior

v

onver

alue

solution

of

oblem

the

if

problem

The

wing

.

functions

is

extende

v

o

ery

let

strong,

xed

it

or

seems,

the

therefore

e

it

in

is

b

a

w

v

ossible

ery

e

dicult

an

task

2

to

nd

Pr

an

at

optimal

solution

domain

ey

whic

Let

h

erimeter

satises

to

e

h

is

a

a

Problem

A

a

b

,

w

ely

e

ts

study

,

the

of

existence

what

of

h

an

ergence

optimal

y

domain

the

in

result

the

3,

hlet

and

oundary

then

in

the

w

shap

spatial

e

is

sensitivit

en

y

y

in

erak

℄

4,

a

the

top

in

ological

pro

deriv

of

ativ

result,

es

e

of

apply

the

results

functional

[3

is

and

are

to

obtained

problem.

in

is

ossible

5.

an

The

additional

y

of

the

a

oundary

lev

W

el

xed.

set

us,

represen

e

tation

the

is

wing

based

Theorem

on

L

the

.

shap

problem

e

analysis

deriv

the

ativ

e

es

se

of

e

(7)

op

,

sets

as

that

it

numb

is

of

describ

onne

ed

d

in

omp

of

6.

p

This

uniformly

form

ounde

ulation

If

pro

w

vides

onver

a

in

sense

w

the

a

metric

y

to

then

solution

the

wing

v

the

alues

pr

of

follo

the

In

shap

oundaries

e

ving

functional

their

problem

b

.

in

The

are

app

satisfying

earance

the

of

on

new

equals

holes

all

is

not

is

p

on

ossible

b

with

the

the

only

ges

use

the

of

Since

the

pr

lev

(8)

el

(7)

set

and

metho

if

d

optimization

based

shap

on

follo

the

shap

(A

e

l

deriv

ar

ativ

implicitely

es.

d

Therefore,

zer

an

by

application

us

that

,

w

in

in

F

and

the

an

of

in

4

and

allo

ws

genc

us

takes

to

e

e.

new

ecomes

holes

t

in

No

order

it

to

p

to

the

v

v

the

alues

of

of

optimal

w

Theorem

.

F

3.

any

Existence

erimeter

of

,

an

oblem

optimal

admits

domain

le

It

one

the

b

ond

e

.

sho

of.

w

b

ed

functional,

that

p

there

a

exists

onds

a

and

domain

ositiv

sho

also

,

t

solution

e

of

maximizing

(7).

for

W

(7).

e

exists

mak

subsequence

e

denoted

use

y

of

The

a

h

result

.

from

ectiv

for

and

V

p

in

hon

sense

℄

Hausdor

and

In

℄

follo

for

the

a

Neumann

v

b

is

oundary

b

and

on

area

the

holes.

of

.

top

or

ological

denition

deriv

Hausdor

ativ

ecΩ

ℓ(∂Ω)≤ liminfℓ(∂Ω ).i

i

ℓ(∂Ω)<∞ J(Ω) P (Ω)c

Ω Ωi

1L (U)H

Ω −→ Ω =⇒χ −→ χ .i Ω Ωi

i→ +∞

|Ω|→|Ω|.i

R

1E(Ω ,u ) =− fui i i2 Ωi

Z Z Z Z