HAL Id: hal-01646003
https://hal.inria.fr/hal-01646003v2
Submitted on 3 Dec 2018
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.
Domain with a Thin Coating
Hanen Boujlida, Houssem Haddar, Moez Khenissi
To cite this version:
Hanen Boujlida, Houssem Haddar, Moez Khenissi. The Asymptotic of Transmission Eigenvalues for
a Domain with a Thin Coating. SIAM Journal on Applied Mathematics, Society for Industrial and
Applied Mathematics, 2018, 78 (5), pp.2348-2369. �hal-01646003v2�
WITH A THIN COATING
2
H. BOUJLIDA ∗, H. HADDAR†, AND M. KHENISSI ‡ 3
Abstract. We consider the transmission eigenvalue problem for a medium surrounded by a thin layer of 4
inhomogeneous material with different refractive index. We derive explicit asymptotic expansion for the transmission 5
eigenvalues with respect to the thickness of the thin layer. We prove error estimate for the asymptotic expansion 6
up to order 1. This expansion can be used to obtain explicit expressions for constant index of refraction.
7
Key words. transmission eigenvalues, asymptotic expansions, thin layers, inverse scattering problems 8
AMS subject classifications. 35P30, 35P25 9
1. Introduction. This work is a contribution to the study of transmission eigenvalues [11, 4,
10
6] and their relation to the shape and material properties of scatterers. These special frequencies
11
are associated with the existence of an incident field that does not scatter. They can be equiva-
12
lently defined as the eigenvalues of a system of two coupled partial differential equations posed on
13
the inclusion domain. One of these equations refers to the equation satisfied by the total field and
14
the other one is satisfied by the incident field. The two equations are coupled on the boundary
15
by imposing that the Cauchy data coincide. This eigenvalue problem can then be formulated as
16
an eigenvalue problem for a non-selfadjoint compact operator. Although non intuitive, it can be
17
shown that this problem admits an infinite discrete set of real eigenvalues without finite accumu-
18
lation points [7, 26]. These special frequencies can be identified from far field data as proved in
19
[5, 19, 4]. Since they carry information on the material properties of the scatterer, transmission
20
eigenvalues would then be of interest for the inverse problem of retrieving qualitative information
21
on the material properties from measured multistatic data [14, 15]. In this perspective, it appears
22
important to study the dependence of these eigenfrequencies with respect to the material prop-
23
erties and the geometry. Several works in the literature have addressed this issue by considering
24
asymptotic regimes and quantifying the dependence of the first leading terms in the asymptotic
25
expansion of the transmission eigenvalue with respect to the small parameter [10, 8, 21, 16]. We
26
here consider the case of a scatterer made of a thin coating which corresponds to frequently en-
27
countered configurations in the stealth technology for instance. The goal is to characterize the
28
dependence of the first order term on the material properties and the thickness of the coating. A
29
first work on this topic was done in [10] where the case of coated perfect scatterer is considered.
30
One proves in particular for the latter case that the first order term depends only on the thickness.
31
We here address the more complicated configuration of a coated penetrable media. The analysis
32
indicates that the first order asymptotic resembles to the shape derivative for the buckling plate
33
equation [17] and contain non trivial dependence on the material properties. More importantly,
34
this expansion allows us to obtain explicit (approximate) expressions for the thin layer index of
35
refraction in terms of the thickness of the layer, the transmission eigenvalue for the coated medium
36
that can be extracted from the measurements and the transmission eigenvalues and eigenvectors
37
for the coated free medium that can be evaluated numerically. This indeed can be useful for the
38
solution of the inverse problem.
39
Although the formal derivation follows the systematic procedure using the classical scaled
40
expansion method (as in [3, 2, 13] for instance), the rigorous justification is much more involved.
41
For instance the arguments in [10] are hard to extend to the present case since special uniform
42
estimates have to be obtained for the transmission problem. We restrict ourselves here to the jus-
43
tification of the first two terms in the asymptotic expansion using the abstract theory developed
44
in [23, 21]. We follow the procedure developed in [8] for the case of small obstacles asymptotic.
45
The main technical point in the proof is to obtain the corrector for the main operator, which is
46
∗LAMMDA-ESST Hammam Sousse, University of Sousse,Tunisia (boujlida.hanen12@gmail.com).
†INRIA Saclay Ile de France/CMAP Ecole Polytechnique,France(houssem.haddar@inria.fr).
‡LAMMDA-ESST Hammam Sousse, University of Sousse, Tunisia (moez.khenissi@gmail.com).
1
here the biharmonic operator. Our main result provides explicit expansion for simple transmis-
47
sion eigenvalues and for multiple transmission eigenvalues that are associated with a generalized
48
eigenspace spanned only by eigenvectors.
49
We analyze the problem where the contrast in material properties affect only the lower order
50
term in the Helmholtz equation. We finally indicate that although the problem is considered
51
only in dimension 2, the results of the main theorem (including the expression of the first order
52
asymptotic term) remain true for three dimensions (up to more complicated technicalities in the
53
proof related to differential geometry).
54
The paper is organized as follows. We first introduce the transmission eigenvalues and write
55
them as the eigenvalues of a non selfadjoint operator. We then present the main result of our paper
56
and discuss applications to the inverse problem. We present next the outline of a classical formal
57
procedure to obtain the expression of the asymptotic expansion. We give the expression till the
58
second order term. We explain in particular why the expression of the second order term would
59
have less interest in practice. We then proceed with the main part of the paper that provides
60
explicit expressions and an error estimate for the first two terms in the asymptotic expansion.
61
2. Problem statement and main results. Let Ω ⊂ R
2be a bounded domain with a
62
smooth boundary Γ. We denote by
63
Ω
0= {x ∈ Ω, d(x, Γ) > }
64
and its boundary
65
Γ
= {x ∈ Ω, d(x, Γ) = } = ∂Ω
0,
66
for > 0 a small enough parameter, where d(x, Γ) denotes the distance of a point x to the boundary
67
Γ. Let Ω = Ω\Ω
0be the layer of thickness around Ω
0(see Figure 1).
Fig. 1.Stretch of the geometry 68
We consider the following transmission eigenvalue problem:
69
(1)
∆w
+ k
2n
(x)w
= 0 in Ω,
∆v
+ k
2v
= 0 in Ω,
w = v on Γ,
∂w
∂ν = ∂v
∂ν on Γ,
70
where k denotes the unknown eigenfrequency and ν the unitary normal to Γ directed to the
71
interior of Ω. The index of refraction n is defined as follows:
72
n (x) =
( n
0(x) in Ω
0, n
1(x) in Ω ,
73
where n
0and n
1are non negative real valued functions ∈ L
∞( R
2) that are independent from .
74
For the sake of simplicity, we assume that the restriction of n
0and n
1to Ω are constant functions
75
along the normal coordinate to Γ for sufficiently small. We finally assume that the function
76
1/(1 − n ) is either positive definite or negative definite on Ω. We remark that this assumption
77
also implies that 1/(1 − n
0) is either positive definite or negative definite on Ω and that
78
(2) 1/|1 − n (x)| ≥ γ > 0 for a.e. x ∈ Ω
79
with γ being independent from (sufficiently small) .
The main goal of this paper is to find the asymptotic expansion of eigenfrequencies k with respect to . Assuming that 1
1 − n
∈ L
∞(Ω), the transmission eigenvalue problem (1) can be reformulated as the nonlinear eigenvalue problem for λ
:= k
2∈ R and u
:= w
− v
∈ H
02(Ω) such that
(∆ + λ
n
) 1 1 − n
(∆ + λ
)u
= 0 in Ω,
which in variational form, after integration by parts, is formulated as finding λ ∈ R and non-trivial
80
function u ∈ H
02(Ω) such that
81
(3)
Z
Ω
1 1 − n
(∆u
+ λ
u
)(∆φ + λ
n
φ)dx = 0, ∀φ ∈ H
02(Ω).
82
The space H
02(Ω) denotes the closure in H
2(Ω) of the set of regular compactly supported functions
83
in Ω. We shall work with the reformulation of (3) as a linear eigenvalue problem for a non
84
selfadjoint compact operator [4]. First observe that (3) can be written as
85
(4) A u + λ B u + λ
2C u = 0 in H
02(Ω)
86
where
87
A : H
02(Ω) → H
02(Ω), B : H
02(Ω) → H
02(Ω), C : H
02(Ω) → H
02(Ω)
88
are defined by the Riesz representation theorem as
89
(5) (A u , φ)
H20(Ω)
:=
Z
Ω
1 1 − n
∆u ∆φdx,
90 91
(6) (B u , φ)
H20(Ω)
:=
Z
Ω
1
1 − n
(u ∆φ + n ∆u φ)dx,
92
and
93
(7) (C u , φ)
H20(Ω)
:=
Z
Ω
n
1 − n
u φdx.
94
Note that A : H
02(Ω) → H
02(Ω) is a bounded, self-adjoint and invertible operator (thanks to (2)),
95
B : H
02(Ω) → H
02(Ω) is a bounded, compact and self-adjoint operator and C : H
02(Ω) → H
02(Ω)
96
is a (non negative or non positive) bounded, compact and self-adjoint operator. Observe that
97
since A is invertible, λ 6= 0. In order to avoid distinguishing the cases of 1 − n being positive or
98
negative we shall abusively set C
1
2
≡ −(−C
12) in the case where 1 − n non positive.
99
Setting U
= (u
, λ
C
12u
), the transmission eigenvalue problem (4) can be transformed into
100
the linear eigenvalue problem, τ
∈ R , U
∈ H
02(Ω) × H
02(Ω) such that
101
(8) (T − τ
I)U
= 0, with τ
= 1
λ
,
102
for the compact non-selfadjoint operator T : H
02(Ω) × H
02(Ω) → H
02(Ω) × H
02(Ω) defined by
103
(9) T = −A
−1B
−A
−1C
12C
120
! .
104
We set
105
(10) T
0= −A
−10B
0−A
−10C
1 2
0
C
1 2
0
0
!
106
where A
0, B
0and C
0are defined by (5), (6) and (7) respectively for n = n
0in Ω. We state here
107
the main result of this paper which will be proven in Section 4. In the following a transmission
108
eigenvalue λ
0is called simple if the corresponding τ
0= 1/λ
0has an algebraic multiplicity equal
109
to 1. We refer to Theorem 4.11 for the case where λ
0has an associated eigenspace formed
110
only by eigenvectors (and therefore an algebraic multiplicity that coincides with the geometrical
111
multiplicity).
112
Theorem 2.1. Assume that n
0, n
1∈ C
4(Ω). Let λ
0∈ R be a simple transmission eigenvalue of (3) with n = n
0in Ω and let u
0∈ H
02(Ω) be an associated eigenfunction. This implies in particular that
β
0:=
Z
Ω
1 1 − n
0λ
20n
0|u
0|
2− |∆u
0|
2dx 6= 0.
If we suppose in addition that u
0and A
−10u
0are in C
6(Ω), then, for sufficiently small > 0, there
113
exists a transmission eigenvalue λ of (3) such that
114
λ = λ
0+ λ
1+ O(
32)
115
where λ
1is given by the following expression
116
λ
1:= λ
0β
0Z
Γ
n
0− n
1(1 − n
0)
2|∆u
0|
2ds(x).
117
This theorem is an immediate consequence of Theorem 4.8 that is stated and proven in the
118
last section of this paper.
119
The formal calculations in Section 3 show that the formula for λ
1is generically valid whenever
120
β
06= 0. However, we remark that in the case of transmission eigenvalues with multiplicity greater
121
than 1, this is not automatically ensured (See Theorem 4.11 for a rigorous expression of λ
1that
122
involves all eigenvectors associated with λ
0).
123
From the practical point of view, this theorem implies in particular that λ
1gives a measure for
124
the contrast n
0− n
1. For instance, if n
1is constant and n
0is constant on Γ, one can approximate
125
the value of n
1using the identity
126
(11) n
1|
Γ= n
0|
Γ− λ − λ
0α
0Z
Ω
1 1 − n
0λ
0n
0|u
0|
2− 1 λ
0|∆u
0|
2dx + O(
12)
127
with
α
0:=
Z
Γ
|∆u
0|
2(1 − n
0)
2ds(x).
For the inverse problem where one would like to determine n
1from multistatic measurements of
128
scattered waves, the value of λ
can be approximated using sampling methods as in [5, 4] (see
129
also [19] for an alternative approach). The values of λ
0and u
0can be computed numerically if
130
one has a priori knowledge of n
0and Ω (see for instance [12, 18, 20] for numerical methods to
131
approximate λ
0and u
0). We finally indicate that, although not carefully checked, we conjecture
132
that the expression for λ
1remains true in three dimensions (corrections due to the curvature of Γ
133
only affect higher order terms).
134
3. Formal asymptotic expansion. In this section, we derive the formal asymptotic ex-
135
pansion for transmission eigenvalues and give explicit formulas for the terms up to order 2. The
136
idea here is to provide a systematic formal way to quickly obtain the explicit expression of λ
1in
137
Theorem 2.1 and also higher order terms. The latter turn out to have complicated expressions
138
that would be of marginal interest for the solution of the inverse problem mentioned above. This
139
formal stage will also be helpful in establishing the rigorous proof based of Osborn’s theorem [23].
140
It allows one to have an intuition for the expression of the corrector in the asymptotic of the main
141
operator A
.
142
We assume the following expansions for the transmission eigenvalues :
143
(12) λ =
∞
X
j=0
jλ
j,
144
and then follow a classical technique for thin layers asymptotics based on rescaling and asymptotic
145
expansion with respect to the thickness . We shall mainly follow the approach in [10].
146
3.1. Scaling. We assume that the boundary Γ is C
∞-smooth, although much less regularity
147
is needed if we restrict ourselves to only few terms in the expansion. The issue of optimal regularity
148
assumptions for Γ is not discussed here. However, one can check that at least a C
2regularity is
149
needed to get the expression of λ
1. We parametrize Γ as
150
Γ = {x
Γ(s), s ∈ [0, L[},
151
with L being the length of Γ and s is the curvilinear abscissa. At the point x
Γ(s), the unit tangent
152
vector is τ(s) := dx
Γ(s)
ds , the curvature κ(s) is defined by:
153
dτ (s)
ds = −κ(s)ν (s) or dν (s)
ds = κ(s)τ(s).
154
Within these notations, the boundary of Ω
0is parametrized as
155
Γ = {x
Γ(s) + ν(s), s ∈ [0, L[}.
156
This parametrization of the surface Γ is equivalent to the definition of Γ , for > 0 a small enough
157
parameter.
158
For a function u defined in Ω , we consider ˜ u defined on [0, L[×]0, [ by
159
(13) u(s, η) = ˜ u(ϕ(s, η)) where ϕ(s, η) := x
Γ(s) + ην(s).
160
Then, the gradient and Laplace operators are expressed in the local coordinates as:
161
∇u = 1 (1 + ηκ(s))
∂
∂s τ(s) + ∂
∂η ν (s)
˜ u,
162
(14) ∆u = 1
(1 + ηκ)
∂
∂s 1 (1 + ηκ)
∂
∂s + κ (1 + ηκ)
∂
∂η + ∂
2∂η
2˜ u.
163
To make the formal calculations, we need to separate the thin layer and scaled it with respect
164
to the thickness so that the equation are posed on a domain independent from . We therefore
165
rewrite the transmission eigenvalue problem (1) in the following equivalent form
166
(15)
∆w
++ k
2n
1w
+= 0 in Ω ,
∆w
−+ k
2n
0w
−= 0 in Ω
0,
∆v
+ k
2v
= 0 in Ω,
w
+= w
−, ∂w
+∂ν = ∂w
−∂ν on Γ ,
w
+= v on Γ,
∂w
+∂ν = ∂v
∂ν on Γ.
167
We denote by ξ = η
the stretched normal variable inside Ω
and define
168
ϕ : G = [0, L[×]0, 1[ → Ω
(s, ξ) 7→ ϕ (s, ξ) = x
Γ(s) + ξν(s).
169
Then the expression of the Laplace operator in the scaled layer is:
170
(16) ∆u = 1
(1 + ξκ)
∂
∂s 1 (1 + ξκ)
∂
∂s + κ (1 + ξκ)
∂
∂ξ + 1
2∂
2∂ξ
2ˆ
u =: ∆
s,ξu ˆ
171
for ˆ u(s, ξ) := u(ϕ (s, ξ)).
172
The next step is to write the equation for w
+in the scaled domain and solve for the asymptotic
173
expansion of w
+in terms of the boundary values on Γ. These boundary values are given by the
174
asymptotic expansion of v
. More specifically, setting ˆ w
(s, ξ) := w
+(ϕ
(s, ξ)), we have that
175
(17) ∆
s,ξw ˆ + λ n
1w ˆ = 0 in G
176
together with the boundary conditions
177
(18)
ˆ
w
(s, 0) = v
(x
Γ(s)) s ∈ [0, L[, 1
∂ w ˆ
∂ξ (s, 0) = ∂v
∂ν (x
Γ(s)) s ∈ [0, L[.
178
We assume that
179
(19) w ˆ (s, ξ) =
∞
X
j=0
jw ˆ
j(s, ξ), (s, ξ) ∈ G and v (x) =
∞
X
j=0
jv
j(x), x ∈ Ω
180
for some functions ˆ w
jdefined on G and v
jdefined on Ω that are independent from . Multiplying
181
(17) by
2(1 + ξκ)
3and using (12), we obtain
182
5
X
k=0
kA
kw ˆ
= 0,
183
where (A
k)
k=0...5are differential operators of order 2 at maximum with the following expressions
184
for the first fourth terms:
185
A
0= ∂
2∂ξ
2,
186
A
1=3ξκ ∂
2∂ξ
2+ κ ∂
∂ξ ,
187
A
2= ∂
2∂s
2+ 3ξ
2κ
2∂
2∂ξ
2+ 2ξκ
2∂
∂ξ + λ
0n
1,
188
A
3=ξ
3κ
3∂
2∂ξ
2+ ξ
2κ
3∂
∂ξ − ξ ∂κ
∂s
∂
∂s + ξκ ∂
2∂s
2+ 3λ
0n
1ξκ + λ
1n
1.
189 190
Inserting the ansatz (19) in (17) and (18) we obtain after equating the terms of same order in
191
and using the convention ˆ w
j= v
j= 0 for j < 0,
192
(20)
∂
2∂ξ
2w ˆ
j= −
5
X
k=1
A
kw ˆ
j−kin G, ˆ
w
j(s, 0) = v
j(x
Γ(s)) s ∈ [0, L[,
∂ w ˆ
j∂ξ (s, 0) = ∂v
j−1∂ν (x
Γ(s)) s ∈ [0, L[.
193
These equations can be solved inductively to get the expressions of ˆ w
jin terms of the boundary
194
values of v
l, l ≤ j. One gets for j = 0, 1, 2 and 3
195
ˆ
w
0(s, ξ) = v
0(x
Γ(s)),
196
ˆ
w
1(s, ξ) = ∂v
0∂ν (x
Γ(s))ξ + v
1(x
Γ(s)),
197
ˆ
w
2(s, ξ) = − ξ
22
κ ∂w
−0∂ν (x
Γ(s)) + ∂
2w
−0∂s
2(x
Γ(s)) + λ
0n
1w
0−(x
Γ(s)) + ∂v
1∂ν (x
Γ(s)ξ + v
2(x
Γ(s)), (21)
198199 200
and
ˆ
w
3(s, ξ) = ξ
36
− 2κ
2∂w
0−∂ν (x
Γ(s)) − 3κ ∂
2w
−0∂s
2(x
Γ(s)) − κλ
0n
1w
−0(x
Γ(s)) + λ
0n
1∂v
0∂ν (x
Γ(s))
201
+ ξ
36
∂
3v
0∂s
2∂ν (x
Γ(s)) − κ ˙ ∂w
−0∂s (x
Γ(s))
202
+ ξ
22
κ ∂v
1∂ν (x
Γ(s)) + λ
0n
1v
1(x
Γ(s)) + λ
1n
1w
−0(x
Γ(s)) + ∂v
2∂ν (x
Γ(s))ξ + v
3(x
Γ(s)).
(22)
203 204
Now, we also postulate the following expansion for w
−:
205
(23) w
−(x) =
∞
X
j=0
jw
j−(x)
206
with w
−j: Ω → R are functions independent of . Then (w
−j, v
j) solves
207
(24)
∆w
−j+ λ
0n
0w
−j= −
j
X
l=1
λ
ln
0w
−j−lin Ω,
∆v
j+ λ
0v
j= −
j
X
l=1
λ
lv
j−lin Ω.
208
Note that the functions w
−jare defined in all Ω and not only Ω
0and therefore (23) gives a extension
209
of w
−to all Ω. The continuity conditions at Γ can be written as
210
˜
w
−(s, ) = ˆ w (s, 1) and ∂ w ˜
−∂η (s, ) = 1
∂ w ˆ
∂ξ (s, 1)
211
where ˜ w
−is defined from w
−using the local change of variables (13) in a neighborhood of Γ.
212
Using Taylor’s expansion (up to the second order, which is sufficient to compute the first three
213
terms in the asymptotic expansion) we get
214
(25) w ˜
−(s, ) = ˜ w
−(s, 0) + ∂ w ˜
−∂η (s, 0) +
22
∂
2w ˜
−∂η
2(s, 0) + o(
2) = ˆ w (s, 1)
215 216
and
(26) ∂ w ˜
−∂η (s, ) = ∂ w ˜
−∂η (s, 0) + ∂
2w ˜
−∂η
2(s, 0) +
22
∂
3w ˜
−∂η
3(s, 0) + o(
2) = 1
∂ w ˆ
∂ξ (s, 1).
217
Injecting (19) and (23) into (25) and (26), we respectively obtain the following continuity conditions
218
on Γ,
219
w
0−(x
Γ(s)) = ˆ w
0(s, 1),
220
w
1−(x
Γ(s)) + ∂w
0−∂ν (x
Γ(s)) = ˆ w
1(s, 1), (27)
221
w
2−(x
Γ(s)) + ∂w
1−∂ν (x
Γ(s)) + 1 2
∂
2w
−0∂ν
2(x
Γ(s)) = ˆ w
2(s, 1),
222 223
and
224
0 = ∂ w ˆ
0∂ξ (s, 1),
225
∂w
0−∂ν (x
Γ(s)) = ∂ w ˆ
1∂ξ (s, 1), (28)
226
∂w
1−∂ν (x
Γ(s)) + ∂
2w
−0∂ν
2(x
Γ(s)) = ∂ w ˆ
2∂ξ (s, 1).
227 228
System (24) coupled with the boundary conditions (28) and (27) provide an inductive way to
229
determine (w
j−, v
j). We obtain the set of equations satisfied by these terms after substituting the
230
expressions of ˆ w
j(s, 1) given by (21),(22). We hereafter summarize the set of equations obtained
231
for (w
j−, v
j) and how to use them to get the expressions of λ
j, j = 0, 1, 2.
232
We first obtain that the couple (w
0−, v
0) solves
233
(29)
∆w
−0+ λ
0n
0w
0−= 0 in Ω,
∆v
0+ λ
0v
0= 0 in Ω, w
0−− v
0= 0 on Γ,
∂w
−0∂ν − ∂v
0∂ν = 0 on Γ.
234
This means in particular that λ
0is a transmission eigenvalue for the limiting problem where the
235
thin layer is removed. We then obtain that the couple (w
1−, v
1) satisfies
236
(30)
∆w
−1+ λ
0n
0w
1−= −λ
1n
0w
−0in Ω,
∆v
1+ λ
0v
1= −λ
1v
0in Ω,
w
1−− v
1= 0 on Γ,
∂w
−1∂ν − ∂v
1∂ν = λ
0(n
0− n
1)w
−0on Γ.
237
Since λ
0is an eigenvalue of the associated homogeneous system, this problem is solvable only if a compatibility condition is satisfied by the right hand sides. This compatibility condition can be obtained by multiplying the first equation with w
0−and the second equation with v
0, taking the difference then integrating by parts and using (29). One ends up with
λ
1= Z
Γ
λ
0(n
0− n
1)|w
−0|
2ds(x) Z
Ω
n
0|w
−0|
2− |v
0|
2dx
which coincides with the expression of given in Theorem 2.1 expressed in terms of u
0= w
0−− v
0.
238
Although not covered by the analysis of convergence, we also provide the expression of the third
239
term in the asymptotic expression. One get that the couple (w
2−, v
2) solves
240
(31)
∆w
2−+ λ
0n
0w
−2= −λ
1n
0w
−1− λ
2n
0w
0−in Ω,
∆v
2+ λ
0v
2= −λ
1v
1− λ
2v
0in Ω,
w
−2− v
2= h
1on Γ,
∂w
2−∂ν − ∂v
2∂ν = h
2on Γ,
241
where
242
h
1= − 1 2
∂
2w
0−∂ν
2− 1
2 λ
0(n
0− n
1)w
0−243
and
244
h
2= κ ∂
2w
0−∂ν
2− 7κ 2
∂
2w
−0∂s
2+
2κ
2+ λ
0(n
0+ n
12 ) ∂w
−0∂ν − 3κ 2
∂w
−0∂s + 3 2
∂
3w
−0∂ν∂s
2245
+
λ
1(2n
1− n
0) + λ
0(κ( n
12 − n
0))
w
−0− ∂
2w
1−∂ν
2+ κ ∂w
−1∂ν .
246247
Writing the compatibility condition for (31), we obtain the following formula for λ
2 248λ
2Z
Ω
1 1 − n
01
λ
0|∆u
0|
2− λ
0n
0|u
0|
2dx = −λ
21Z
Ω
1
λ
0∆u
0u ¯
0+ 1
1 − n
0|u
0|
2dx
249
− λ
1Z
Ω
1 1 − n
0u
1∆¯ u
0+ n
0∆u
1u ¯
0+ 2n
0λ
0u
1¯ u
0dx
250
+ Z
Γ
h
1∂
∂ν 1
1 − n
0(∆ + λ
0)¯ u
0ds(x) −
Z
Γ
h
21
1 − n
0(∆ + λ
0)¯ u
0ds(x).
(32)
251 252
This complicated expression shows in particular a nonlinear dependence of λ
2in terms of n
1. It
253
suggests that the use of λ
2for solutions to the inverse problem of determining n
1may not be
254
appropriate.
255
4. Convergence analysis. The main goal of this section is to prove Theorem 2.1 that
256
provides a rigorous mathematical justification of the formal asymptotic expansion for simple real
257
transmission eigenvalues up to the first order. The proof is split into several steps. The first one is
258
to establish the convergence in norm of the operator T to T
0. This ensures the convergence of λ
259to λ
0. In order to get to the term of order 1 in , we shall apply the Osborn theorem which requires
260
for instance a characterization of the pointwise asymptotic expansion of T (U ) up to order 1 in
261
(for some given function U ∈ H
02(Ω) × H
02(Ω)). The latter can be obtained from the asymptotic
262
expansions of A
−1u, B u and C u for some u ∈ H
02(Ω). The difficult part to get the expansion of
263
A
−1u since for the two others, the first order terms are vanishing. This critical result is provided
264
by Lemma 4.5.
265
In all the following we use the notation
266
(f, g) := (f, g)
H20(Ω)
= Z
Ω
∆f ∆gdx and kgk := (g, g)
1 2
H02(Ω)
.
267
For an operator A : V → V , kAk denotes the operator norm. To simplify the writing, C will
268
denote a generic constant whose value may change but remains independent from as → 0.
269
4.1. Pointwise convergence of the spectrum of T . In this first step, we prove pointwise
270
convergence of the spectrum of the operator T to the spectrum of T
0. This is a direct consequence
271
of the following convergence in norm [23, 8].
272
Theorem 4.1. Assume that n
0∈ C
2(Ω). Let T and T
0be defined by (9) and (10) respectively.
273
Then T converges to T
0in the operator norm.
274
Proof. The proof follows from Lemma 4.2 and Lemma 4.4 below, using the definition of T
275and T
0.
276
In the first lemma we prove norm convergence for B and C .
277
Lemma 4.2. Let B
, C
, B
0and C
0be the operators defined by (6) and (7). Then, for suffi-
278
ciently small ,
279
(33) kB
− B
0k ≤ C
12and kC
12− C
012k ≤ C.
280
Proof. From the definitions of B and B
0, we have that for u, φ ∈ H
02(Ω)
281
((B
− B
0)u, φ) = Z
Ω
1 1 − n
u∆φ + n
∆uφ dx −
Z
Ω
1 1 − n
0u∆φ + n
0∆uφ dx
282
= Z
Ω
1
1 − n
1− 1 1 − n
0u∆φ + ∆uφ dx.
283 284
Therefore,
285
|((B − B
0)u, φ)| ≤C
kuk
L∞(Ω)k∆φk
L1(Ω)+ kφk
L∞(Ωk∆uk
L1(Ω).
286287
Using the Sobolev embedding theorem and the Cauchy Schwartz inequality, we get
288
|((B − B
0)u, φ)| ≤ C
12(kuk
H20(Ω)
kφk
H2 0(Ω)).
289
By choosing φ = (B − B
0)u, we get
290
k(B − B
0)uk
H20(Ω)
≤ C
12kuk
H2 0(Ω).
291
The proof is similar for the second inequality. For u, φ ∈ H
02(Ω), we have
292
((C − C
0)u, φ) = Z
Ω
1 1 − n
1− 1 1 − n
0uφdx ≤ C
|Ω |kuk
L∞(Ω)kφk
L∞(Ω)293 294
From the Sobolev embedding theorem, we obtain
295
((C − C
0)u, φ) ≤ C kuk
H20(Ω)
kφk
H2 0(Ω).
296
By choosing φ = (C
− C
0)u, we have
297
(34) k(C
− C
0)uk
H20(Ω)
≤ Ckuk
H2 0(Ω).
298
Using the square root Lemma in [24] and the fact that C
nconverges to C
0nat the same order
299
O(), we conclude that C
1
2
converges to C
1 2
0
at the same order O(). Hence we have
300
(35) k(C
12− C
1 2
0
)uk
H20(Ω)
≤ Ckuk
H20(Ω)
.
301
Now we show convergence in the H
02(Ω) norm for A
−1f assuming smoothness of f . This will be
302
useful in the proof of Lemma 4.4 since the operators B and C are regularizing.
303
Lemma 4.3. Let A and A
0be defined by (5) for > 0 and = 0, respectively and f ∈ H
02(Ω).
304
If A
−10f ∈ C
2(Ω), then for sufficiently small ,
305
(36) kA
−1f − A
−10f k ≤ C
12.
306
Proof. For a fixed f ∈ H
02(Ω), define z and z
0in H
02(Ω) as z = A
−1f and z
0= A
−10f . Since
307
A z = A
0z
0= f , we have that for φ ∈ H
02(Ω)
308
(37) (A
(z
− z
0), φ) = (A
0z
0− A
z
0, φ) = Z
Ω
1 1 − n
0− 1 1 − n
1∆z
0∆φdx.
309
If z
0∈ C
2(Ω), we get
310
Z
Ω
1
1 − n
0− 1 1 − n
∆z
0∆φdx ≤ Ck∆z
0k
∞Z
Ω
∆φdx ≤C
12kφk
H2 0(Ω).
311 312
Thus, we have shown that
313
(A (z − z
0), φ) ≤ C
12kφk
H2 0(Ω).
314
By plugging in φ = z − z
0, we obtain the desired convergence using the coercivity of A .
315
Lemma 4.4. Assume that n
0∈ C
2(Ω). Let A , B , C , A
0, B
0and C
0be defined by (5), (6)
316
and (7) for > 0 and = 0, respectively. Then for sufficiently small ,
317
kA
−1B − A
−10B
0k −→
→0
0 and kA
−1C
1
2
− A
−10C
1 2
0
k −→
→0
0.
318
Proof. From (37), we have that for f, φ ∈ H
02(Ω) and with z = A
−1f and z
0= A
−10f
319
(A (z − z
0), φ) ≤ Ck∆A
−10f k
L2(Ω)kφk
H2 0(Ω).
320321
Furthermore,
322
kA
−1B
f − A
−10B
0f k
H20(Ω)
≤k(A
−1− A
−10)B
0f k
H20(Ω)
+ kA
−1(B
− B
0)f k
H2 0(Ω) 323≤Ck∆A
−10B
0fk
L2(Ω)+ kA
−1kk(B − B
0)kkf k
H2 0(Ω). (38)
324325
For estimating k∆A
−10B
0f k
L2(Ω), observe that B
0u ∈ H
02(Ω) is the weak solution
326
∆∆B
0u = ∆ n
01 − n
0u
+ 1
1 − n
0∆u in Ω.
327
Classical regularity results [22, 25] and the fact that n
0∈ C
2(Ω) imply that B
0u ∈ H
4(Ω) ∩ H
02(Ω)
328
and therefore
329
k∆A
−10B
0f k
H1(Ω)≤ Ckf k
H2(Ω).
330
By the Sobolev embedding theorem, this implies that
331
k∆A
−10B
0f k
Lp(Ω)≤ Ckf k
H2(Ω),
332
for p > 2 . Let ˜ p =
p2> 1 and q such that 1
˜ p + 1
q = 1.
333
k∆A
−10B
0f k
2L2(Ω)≤ k∆A
−10B
0f k
2Lp(Ω)|Ω |
1q≤ C
2q1kf k
H2(Ω). (39)
334335
From (33) we obtain
336
(40) kA
−1kk(B − B
0)kkf k
H20(Ω)
≤ C
12kf k
H20(Ω)
.
337
Using (38), (39) and (40) we have that
338
kA
−1B
− A
−10B
0k −→
→0
0.
339
The second convergence result follows from similar arguments.
340
Now we would like to obtain explicit formula for the correction term in the asymptotic expansion
341
for the operator T . More precisely, we define explicit formula for the corrector term associated
342
with A
−1− A
−10.
343
4.2. Corrector term for A
−1− A
−10. In this subsection, we construct a corrector function
344
and use it to estimate the convergence rate of z = A
−1u for u ∈ H
02(Ω). Let z
0= A
−10u ∈ H
02(Ω),
345
i.e z
0∈ H
02(Ω) solution of
346
(41) ∆ 1
1 − n
0∆z
0= ∆∆u in Ω.
347
Inspired by the formal calculations on the previous section, namely problem (30), we define z
1348
solution of
349
(42)
∆ 1
1 − n
0∆z
1= 0 in Ω,
z
1= 0 on Γ,
∂z
1∂ν = 1 − n
11 − n
0− 1
∆z
0on Γ.
350
We expect that z = z
0+ z
1+ O(
2) in Ω
0. We extend z
1in Ω as ˜ z
1defined by
351
(43) z ˜
1=
( z
1in Ω
0, z
1− ψ in Ω
352
where ψ is a polynomial of order ≤ 3 and satisfying the boundary conditions:
353
(44)
ψ = 0 , ∂ψ
∂ν = 1 − n
11 − n
0− 1
∆z
0on Γ, ψ = ∂ψ
∂ν = 0 on Γ
.
354
This gives the following expression of ψ (that plays the role ˆ w
2in the formal calculations)
355
ψ(x) = ψ(ϕ(s, ξ)) = ˆ ψ(s, ξ) = 1 − n
11 − n
0− 1
∆z
0(ϕ(s, 0))ξ(1 − ξ)
2.
356
The choice of ψ ensures in particular that ˜ z
1∈ H
02(Ω). To simplify the notation we set m := 1
1 − n
0− 1 1 − n
1. Now we have the following Lemma.
357
Lemma 4.5. Assume that n
0and n
1are in C
4(Ω). Let u ∈ H
02(Ω) then set z = A
−1u and
358
z
0= A
−10u. We define z ˜
1as in (43) and assume that z
0∈ C
6(Ω). Then we have, for sufficiently
359
small ,
360
kz − z
0− ˜ z
1k
H20(Ω)
≤ C
32.
361
Proof. For any φ ∈ H
02(Ω) we have that
362
(45) (A (z − z
0− ˜ z
1), φ) = (A (z − z
0), φ) − (A z ˜
1, φ).
363
We recall that
364
(A (z − z
0), φ) = Z
Ω
1 1 − n
0− 1 1 − n
1∆z
0∆φdx.
365 366
Furthermore, we have that
367
(46) (A z ˜
1, φ) = Z
Ω0
1 1 − n
0∆z
1∆φdx + Z
Ω
1 1 − n
1∆(z
1− ψ)∆φdx.
368
Using the fact that ∆ 1 1 − n
0∆z
1= 0 and the Green formula yield,
369
(A z ˜
1, φ) = Z
Γ
m∆z
1∂φ
∂ν ds(x) + Z
Γ
∂
∂ν 1
1 − n
1∆z
1− ∂
∂ν 1
1 − n
0∆z
1φds(x)
370
+ Z
Γ
1 1 − n
1∆ψ ∂φ
∂ν ds(x) − Z
Γ
∂
∂ν 1
1 − n
1∆ψ
φds(x) − Z
Ω
∆ 1
1 − n
1∆ψφdx.
371 372
Using the expression of ψ we have
373
1
1 − n
1∆ψ|
Γ= 1
1 − n
1∆
s,ξψ(s, ˜ 1) = 2
m∆z
0(ϕ(s, 0)),
374
∂
∂ν ( 1 1 − n
1∆ψ)|
Γ= ∂
∂η ( 1 1 − n
1∆ψ)|
Γ= 1 1 − n
11
∂
∂ξ (∆
s,ξψ)(s, ˜ 1) = 6
2m∆z
0(ϕ(s, 0)).
375
We then get after substitution of these expressions
376
(A z ˜
1, φ) = Z
Γ
m
∆z
1(ϕ(s, )) + 2
∆z
0(ϕ(s, 0)) ∂φ
∂ν ds(x)
377
− Z
Γ
m ∂
∂ν (∆z
1(ϕ(s, ) + 6
2∆z
0(ϕ(s, 0)))φds(x) − Z
Ω
∆ 1
1 − n
1∆ψφdx
378
= Z
Γ
φ
1∂φ
∂ν ds(x) − Z
Γ
φ
2φds(x) − Z
Ω
∆ 1
1 − n
1∆ψφdx
379 380
where we have set
381
φ
1(s) := m∆z
1(ϕ(s, )) + 2
m∆z
0(ϕ(s, 0)),
382
φ
2(s) := m ∂
∂ν (∆z
1(ϕ(s, ))) + 6
2m∆z
0(ϕ(s, 0))
383
using the parametrization of the curve Γ , s 7−→ ϕ(s, ) with ϕ defined by (13). Using this
384
parametrization and setting ˜ φ(s, η) := φ(ϕ(s, η)) in Ω we have
385
Z
Γ
φ
1∂φ
∂ν ds(x) = Z
L0
φ
1∂ φ ˜
∂η (s, )(1 + κ)ds = Z
L0
Z
0φ
1∂
2φ ˜
∂η
2(s, η)(1 + κ)dsdη.
386 387
From the definition of φ
1we then get for φ ∈ H
02(Ω),
388
Z
Γ
φ
1∂φ
∂ν ds(x) = 2
Z
L 0Z
0m∆z
0(ϕ(s, 0)) ∂
2φ ˜
∂η
2(s, η)dsdη + O(
12)kφk
H2(Ω).
389 390
Here and in all the following O(
r) denotes a function such that O(
r) ≤ C
rfor a constant C
391
independent from the test function φ but that may depend on kz
0k
C6(Ω). Using Taylor’s expansion
392
we also get for φ ∈ H
02(Ω),
393
Z
Γ
φ
2φds(x) = 2
Z
L 0Z
0φ
2∂
2φ ˜
∂η
2(s, η)(1 + κ)dsdη + O(
32)kφk
H2(Ω) 394= 3
Z
L 0Z
0m∆z
0(ϕ(s, 0)) ∂
2φ ˜
∂η
2(s, η)dsdη + O(
12)kφk
H2(Ω) 395396
where the last equality is obtained after substituting the expression of φ
2. One ends up with
Z
Γ
(φ
1∂φ
∂ν − φ
2φ)ds(x) = − Z
L0
Z
0m∆z
0(ϕ(s, 0)) ∂
2φ ˜
∂η
2(s, η)dsdη + O(
32)kφk
H2(Ω). Equation (45) then gives
397
(A (z − z
0− ˜ z
1), φ) = Z
Ω
m∆z
0∆φdx − (A z ˜
1, φ)
398
= Z
L0
Z
0m∆z
0(ϕ(s, η))∆φ(ϕ(s, η))(1 + ηκ)dsdη − Z
L0
Z
0m∆z
0(ϕ(s, 0) ∂
2φ
∂η
2(ϕ(s, η))dsdη
399
− Z
Ω
∆ 1
1 − n
1∆ψφdx + O(
32)kφk
H2(Ω).
400 401
We use the expression of the Laplacien in local coordinates
402
(1 + ηκ)∆φ(ϕ(s, η)) = ∂
∂s 1
1 + ηκ
∂ φ ˜
∂s (s, η) + κ ∂ φ ˜
∂η (s, η) + (1 + ηκ) ∂
2φ ˜
∂η
2(s, η)
403
to make the decomposition
404
Z
L 0Z
0m∆z
0(ϕ(s, η))∆φ(ϕ(s, η))(1 + ηκ)dsdη = Z
L0
Z
0m∆z
0(ϕ(s, η)) ∂
∂s 1
1 + ηκ
∂ φ ˜
∂s (s, η)
405
+ Z
L0
Z
0m∆z
0(ϕ(s, η)) κ ∂ φ ˜
∂η (s, η) + ηκ ∂
2φ ˜
∂η
2(s, η) +
Z
L 0Z
0m∆z
0(ϕ(s, η)) ∂
2φ ˜
∂η
2(s, η).
406 407
To estimate the first term, we integrate by parts on [0, L[, we obtain
408
Z
L 0Z
0m∆z
0(ϕ(s, η)) ∂
∂s 1
1 + ηκ
∂ φ ˜
∂s (s, η) dηds
409
= − Z
L0
Z
01 1 + ηκ
∂
∂s (m∆z
0(ϕ(s, η))) ∂ φ ˜
∂s (s, η)dsdη
410
= − Z
L0
Z
1 0m 1
1 + ξκ
∂
∂s ∆z
0(ϕ(s, ξ)) ∂ φ ˜
∂s (s, ξ)dsdξ
411
= − Z
L0
Z
1 0m ∂
∂s ∆z
0(ϕ(s, 0)) Z
ξ 0∂
2φ ˜
∂η∂s (s, η)dη
dsdξ + O(
2)kφk
H2(Ω) 412= O(
32)kφk
H2(Ω).
413414
For the last term we proceed similarly to obtain
415
Z
L 0Z
0m∆z
0(ϕ(s, η)) ∂ φ ˜
∂η (s, η)dsdη = Z
L0
Z
1 0m∆z
0(ϕ(s, ξ)) ∂ φ ˜
∂η (s, ξ)dsdξ
416
= Z
L0
Z
1 0m∆z
0(ϕ(s, 0)) Z
ξ 0∂
2φ ˜
∂η
2(s, η)dη
dsdξ + O(
2)kφk
H2(Ω)= O(
32)kφk
H2(Ω) 417418419
Observing in addition that
420
Z
L 0Z
0ηκm(∆z
0(ϕ(s, η) ∂
2φ ˜
∂η
2(s, η))dsdη = O(
32)kφk
H2(Ω),
421
one ends up with
422
(A
(z
− z
0− ˜ z
1), φ) = Z
L0
Z
0m
∆z
0(ϕ(s, η)) − ∆z
0(ϕ(s, 0)) ∂
2φ ˜
∂η
2(s, η)dsdη
423
− Z
Ω
∆ 1
1 − n
1∆f ψ φdx + O(
32)kφk
H2(Ω).
424 425
To conclude we just observe that the two remaining terms are also of the form O(
32)kφk
H2(Ω). For the first term, we simply use a Taylor expansion for ∆z
0while for the second one we just use that, due to the regularity of n
0and n
1,
∆ 1
1 − n
1∆ψ ∈ L
∞(Ω).
In conclusion,
426
(A (z − z
0− ˜ z
1), φ) ≤ C
32kφk
H2(Ω).
427
Choosing φ = z − z
0− ˜ z
1, since the coercivity constant associated with A is independent from
428
, we get
429
kz − z
0− ˜ z
1k
H20(Ω)
≤ C
32430
which ends the proof.
431
Lemma 4.6. Assume that n
0and n
1are in C
4(Ω). If u ∈ C
6(Ω) ∩ H
02(Ω), then for sufficiently
432
small ,
433
(47) kB
0(A
−1− A
−10)uk
H20(Ω)
≤ C and kC
012(A
−1− A
−10)uk
H20(Ω)
≤ C
434
where C independent of .
435