Uniqueness for an hyperbolic inverse problem
with time-dependent coefficient
Mourad Bellassoued * — Ibtissem Ben Aïcha**
* ENIT-LAMSIN
University of Tunis El Manar, National Engineering School of Tunis
B.P.37, 1002 Tunis
TUNISIA
mourad.bellassoued@enit.utm.tn
** ENIT-LAMSIN
University of Aix-Marseille
85 Boulevard Charles Livon, 13284 Marseille
FRANCE
&
University of Carthage, Faculty of sciences of Bizerte
7021, Jarzouna Bizerte
TUNISIA
ibtissem.benaicha.enit@utm.tn
ABSTRACT. This paper deals with an hyperbolic inverse problem of determining a time-dependent
coefficient a appearing in a dissipative wave equation, from boundary observations. We prove in
dimension n greater than two, that a can be uniquely determined in a precise subset of the domain,
from the knowledge of the Dirichlet-to-Neumann map.
RÉSUMÉ. Dans ce travail, on étudie le probl ème inverse de la détermination d’un coefficient dépendant de la variable d’espace et du temps apparaissant dans une équation d’onde dissipative, à partir
des mesures faites sur tout le bord du domaine. On démontre que ce coefficient peut être déterminé
d’une manière unique dans une partie précise du domaine à partir des mesures de type Neumann.
KEYWORDS : Inverse problems, Dissipative wave equation, Time-dependent coefficient, Uniqueness .
MOTS-CLÉS : Problèmes inverses, équation d’onde dissipative, coefficient qui dépendent du temps,
Unicité
Special issue, LEM2I
Mourad Bellassoued, Nabil Gmati, Mohamed Jaoua, Gilles Lebeau, Editors.
ARIMA Journal, vol. 23, pp. 65-78 (2016)
66 ARIMA - volume 23 - 2016
1. Introduction
1.1. Statement of the problem
The present paper is devoted to the study of the following hyperbolic inverse problem:
Given T > 0 and a bounded domain Ω ⊂ Rn , n ≥ 2, with C ∞ boundary Γ = ❇ Ω, determine
the absorbing coefficient a present in the following initial boundary value problem for the
wave equation from boundary observations
⎧
❇t2 u − ∆u + a(x, t)❇t u = 0 in Q = Ω × (0, T ),
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨ u(x, 0) = 0, ❇t u(x, 0) = 0 in Ω,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
on Σ = Γ × (0, T ),
⎩ u(x, t) = f (x, t)
(1)
where f ∈ H 1 (Σ), and the coefficient a ∈ C 2 (Q) is assumed to be real valued. It is well
known (see [10] and [9] ) that if the compatibility condition is satisfied, that is f (⋅, 0) = 0,
then, there exists a unique solution u to the equation (1) that belongs to the following
space
u ∈ C([0, T ], H 1 (Ω)) ∩ C 1 ([0, T ], L2 (Ω)).
Moreover, there exists a constant C > 0 such that we have
∥❇ν u∥L2 (Σ) ≤ ∥f ∥H 1 (Σ) ,
(2)
where ν denotes the unit outward normal to Γ at x and ❇ν u stands for ∇u⋅ν. In the present
paper, we focus on the uniqueness issue in the study of the inverse problem of determining the time-dependent absorbing coefficient a from the knowledge of the Dirichlet-toNeumann map.
From a physical view point, the inverse problem under consideration consists in recovering the absorbing coefficient a in an homogeneous medium by probing it with disturbances generated on the boundary. The data are the responses of the medium to these
disturbances measured on all the boundary. Here the coefficient a can be seen as one of
the medium properties and we aim to recover it in a specific subset of the domain from
boundary measurements, after probing the medium by a Dirichlet data f . The medium is
assumed to be quiet initially.
The problem of recovering coefficients that depend only on the spatial variable is
considered by many authors. In [16] Rakesh and Symes proved a uniqueness result in
recovering a time-independent potential appearing in a wave equation from measurements
made on the whole boundary. The main tools in the derivation of this result are first, the
construction of geometric optics solutions and second, the relation linking the hyperbolic
Dirichlet-to-Neumann map to the X-ray transform. As for the uniqueness from local
Neumann measurements, we refer to Eskin [7]. One can also see the paper of Isakov [11],
in which a uniqueness result was proved in the determination of two time-independent
coefficients appearing in a dissipative wave equation.
The stability in the case where the Neumann data are observed on a subdomain of the
boundary was considered by Bellassoued, Choulli and Yamamoto [1], where a stability
estimate of log-type was proved in recovering a time-independent coefficient appearing
in a wave equation. In [13], Isakov and Sun established a stability result of Hölder type in
determining a coefficient in a subdomain from local Neumann data. As for the stability in
Uniqueness for an hyperbolic inverse problem with time-dependent coefficient 67
the case where Neumann data are observed on the whole boundary, we refer to Sun [21],
Cipolatti and Lopez [6].
When the coefficients depend also on the time variable, there is a uniqueness result
proved by Ramm and Rakesh [17], in which they proved that a time-dependent potential
appearing in a wave equation can be uniquely determined in a precise subset made of lines
making an angle of 45○ with the t-axis and meeting the planes t = 0 and t = T outside Q,
from global Neumann-data. It’s clear from [10] that this coefficient can not be recovered
over the whole domain Q and this is actually due to the homogeneous initial conditions
imposed in the system. However, Isakov proved in [12], that the time-dependent coefficient may be uniquely determined over the whole domain Q, but he needed to know
much more information about the solution of the wave equation. We can also refer to
[8, 18, 19, 20].
Inspired by the work of Bellassoued and Dos Santos [2], Waters [22] proved recently
that one can stably recover the X-ray transform of a time-dependent lower order term
present in a wave equation from the knowledge of the Dirichlet-to-Neumann map, in
the Riemmanian case. In the euclidian case, Ben Aïcha [5] and Kian [14, 15], showed
by taking inspiration from the work of Bellassoued-Jellali-Yamammoto [3, 4], stability
results in the recovery of a zeroth order time-dependent coefficient appearing in a wave
equation.
In this paper, we prove that the time-dependent absorbing coefficient a can be uniquely
determined with respect to the Dirichlet-to-Neumann map in a specific subset of the domain Q, provided that a is known outside this subset.
1.2. Main results
In order to state our main result we first introduce the following notations.
Let r > 0 be such that T > 2r and Ω ⊆ B(0, r/2) = {x ∈ Rn , ∣x∣ < r/2}. We set
Qr = B(0, r/2) × (0, T ). We consider the annular region around the domain Ω,
Ar = {x ∈ Rn ,
r
r
< ∣x∣ < T − } ,
2
2
and the forward and backward light cones:
r
r
Cr+ = {(x, t) ∈ Qr , ∣x∣ < t − , t > } ,
2
2
r
r
− t, T − > t} ,
2
2
r
r
Cr = {(x, t) ∈ Qr , ∣x∣ ≤ − t, 0 ≤ t ≤ }.
2
2
Cr− = {(x, t) ∈ Qr , ∣x∣ < T −
Finally, we denote
Q∗r = Cr+ ∩ Cr− and Qr,∗ = Q ∩ Q∗r .
We remark that the open subset Qr,∗ is made of lines making an angle of 45○ with the
t-axis and meeting the planes t = 0 and t = T outside Qr . We notice that Qr,∗ ⊂ Q. Note,
that in the particular case where Ω = B(0, r/2), we have Qr,∗ = Q∗r (see Figure 1 in [5]).
Our set of data will be given by the the Dirichlet-to-Neumann map Λa defined as
follows
Λa ∶ H 1 (Σ) Ð→ L2 (Σ)
f z→
❇ν u,
68 ARIMA - volume 23 - 2016
By (2) we have that Λa is continuous from H 1 (Σ) to L2 (Σ). We denote by ∥Λa ∥ its
norm in L(H 1 (Σ), L2 (Σ)). Let us now introduce the admissible set of the absorbing
coefficients a. Given a0 ∈ C 2 (Qr ) and M > 0 we set
A(a0 , M ) = {a ∈ C 2 (Qr ), a = a0 in Qr ∖ Qr,∗ , ∥a∥C 2 (Q) ≤ M }.
Having said that we may state the main results of this paper
Theorem 1.1 (Non uniqueness) For any a ∈ A(a0 , M ) such that Supp(a) ⊂ Cr , we have
Λa = Λ 0 .
Theorem 1.2 (Uniqueness) Let T > 2Diam(Ω) and ai ∈ A(a0 , M ), i = 1, 2. Then, we
have
Λa2 = Λa1 implies a2 = a1 on Qr,∗ .
The outline of this paper is as follows. In Section 2, we develop the proof of Theorem
1.1. Section 3 is devoted to the construction of geometric optics solutions to the equation
(1). Using these particular solutions, we prove in Section 3 Theorem 1.2.
2. Non uniqueness in determining the time-dependent
coefficient
In this section we aim to show that it is hopeless to recover the time-dependent coefficient a over the whole domain in the case where the initial conditions are zero.
2.1. Preliminary
This section is devoted to the proof of a fundamental result which is borrowed from
[10]. Let us first introduce the following notations. We define
V = ⋃ D(τ ) = ⋃ (Cr ∩ {t = τ }),
0≤τ ≤t′
0≤τ ≤t′
where 0 < t′ < r/2. Moreover, we denote by
S = ❇ Cr ∩ (Ω×]0, t′ [), and ❇ V = S ∪ D(t′ ) ∪ D(0).
Lemma 2.1 Let us denote by u the solution of the dissipative wave equation
⎧
(❇t2 − ∆ + a(x, t)❇t ) u(x, t) = 0 in Q,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨ u(x, 0) = 0 = ❇t u(x, 0)
in Ω,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
on Σ.
⎩ u(x, t) = f (x, t)
Then, u(x, t) = 0 on the set Cr .
We denote by P = ❇t2 − ∆ + a(x, t)❇t . A simple calculation gives us
∫ 2P u(x, t) ❇t u(x, t) dx dt
V
= ∫ 2❇t2 u(x, t)❇t u(x, t) dx dt − ∫ 2∆u(x, t) ❇t u(x, t) dx dt
V
V
+ ∫ 2a(x, t)∣❇t u(x, t)∣2 dx dt
V
n
= ∫ ❇t (∣❇t ∣2 + ∣∇u∣2 ) dx dt + ∫ ∑ ❇j (❇t u ❇j u) dx dt
V j=1
V
+ ∫ 2a(x, t)∣❇t u(x, t)∣ dx dt.
2
V
Uniqueness for an hyperbolic inverse problem with time-dependent coefficient 69
Then, using the above identity, we see that
n
2
∫ 2P u(x, t)❇t u(x, t) dx dt = ∫ ❇t e(x, t) dx dt+ ∫ ∑ ❇j Xj (x, t) dx dt+ ∫ 2a(x, t)∣❇t u(x, t)∣ dx dt,
V j=1
V
V
V
where e(x, t) = ∣❇t u(x, t)∣2 + ∣∇u(x, t)∣2 and Xj (x, t) = −2❇t u(x, t)❇j u(x, t). Next, by
applying the divergence theorem, one gets
n
∫ 2P u(x, t)❇t u(x, t) dx dt = ∫ (e(x, t)η + ∑ Xj (x, t)µj ) dσ + ∫
S
V
D(t′ )
j=1
+ ∫ 2a(x, t)∣❇t u(x, t)∣2 dx dt,
e(x, t′ ) dx − ∫
e(x, 0) dx
D(0)
(3)
V
where dσ denotes the surface element of S and the vector (η, µ1 , µ2 , ..., µn ) ∈ Rn+1 is
the outward unit normal vector at (x, t) ∈ S such that
⎛n
⎞
η = ∑ µ2j
⎝j=1 ⎠
1/2
(4)
.
On the other hand, from Cauchy-Schwarz inequality and (4), we can see that
n
∫ (e(x, t)η + ∑ Xj (x, t)µj ) dσ
S
j=1
≥ ∫ (∣❇t u(x, t)∣2 + ∣∇u(x, t)∣2 ) η − 2∣❇t u(x, t)∣∣∇u(x, t)∣ η dσ
S
≥ 0.
(5)
Then, since e(x, 0) = 0 we get from (3) and (5) this estimation
∫
D(t′ )
e(x, t′ ) dx
≤ ∫ 2P u ❇t u(x, t) dx dt − ∫ 2a(x, t)∣❇t u(x, t)∣2 dx dt
V
V
Now, using the fact that P u(x, t) = 0 for any (x, t) ∈ V , we get
∫
D(t′ )
t′
e(x, t′ ) dx ≤ C ∫
0
∫
D(t)
(e(x, t) + ∣u(x, t)∣2 ) dx dt,
(6)
where, the positive constant C is depending on M . Now bearing in mind that
∣u(x, t′ )∣2 = ∣u(x, 0)∣2 + ∫
0
t′
❇t (∣u(x, t)∣2 ) dt ≤ ∫
t′
e(x, t) dx.
(7)
0
Thus, from (6) and (7) we deduce that
∫
D(t′ )
(e(x, t′ ) + ∣u(x, t′ )∣2 ) dx ≤ ∫
0
t′
∫
D(t)
(e(x, t) + ∣u(x, t)∣2 ) dx dt.
In view of Gronwall’s Lemma we end up deducing that u(x, t) = 0 for any x ∈ D(t′ ) and
t′ ∈ (0, r/2). This completes the proof of the lemma.
70 ARIMA - volume 23 - 2016
2.2. Proof of Theorem 1.1
Let a ∈ A(a0 , M ) such that Supp(a) ⊂ Cr . Let f ∈ H 1 (Σ) and u satisfy
⎧
❇t2 u − ∆u + a(x, t)❇t u = 0 in Q,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨ u(x, 0) = 0, ❇t u(x, 0) = 0 in Ω,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
on Σ,
⎩ u=f
Since from Lemma 2.1, we have u = 0 in the conic set Cr and using the fact that Supp(a) ⊂
Cr , we deduce that u solves also the following hyperbolic boundary-value problem
⎧
❇t2 v − ∆v = 0
in Q,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨ v(x, 0) = 0, ❇t v(x, 0) = 0 in Ω,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
on Σ.
⎩ v=f
Then, we conclude that Λa (f ) = Λ0 (f ) for all f ∈ H 1 (Σ).
3. Construction of geometric optics solutions
In this section, we construct suitable geometrical optics solutions for the dissipative
wave equation (1), which are key ingredients to the proof of our main result. We first state
the following lemma that will be used in order to prove the main statement of this section.
Lemma 3.1 (see [10]) Let T, M1 , M2 > 0, a ∈ L∞ (Q) and b ∈ L∞ (Q), such that
∥a∥L∞ (Q) ≤ M1 and ∥b∥L∞ (Q) ≤ M2 . Assume that F ∈ L1 (0, T ; L2 (Ω)). Then, there
exists a unique solution u to the following equation
such that
⎧
❇t2 u − ∆u + a(x, t)❇t u + b(x, t)u(x, t) = F (x, t) in Q,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
in Ω,
⎨ u(x, 0) = 0 = ❇t u(x, 0)
⎪
⎪
⎪
⎪
⎪
⎪
⎪
on Σ,
⎩ u(x, t) = 0
(8)
u ∈ C([0, T ]; H01 (Ω)) ∩ C 1 ([0, T ]; L2 (Ω)).
Moreover, there exists a constant C > 0 such that for t ∈ (0, T ) we have
∥❇t u(⋅, t)∥L2 (Ω) + ∥∇u(⋅, t)∥L2 (Ω) ≤ C∥F ∥L1 (0,T ;L2 (Ω)) .
(9)
φ(x, t) = ϕ(x + tω),
(10)
By the use of the above lemma, we may construct suitable geometrical optics solutions
to the equation (1) and to the retrograde problem. We shall first consider a function
ϕ ∈ C0∞ (Rn ). Notice that for all ω ∈ S n−1 the function φ given by
solves the following transport equation
(❇t − ω ⋅ ∇)φ(x, t) = 0.
Let us now prove the following Lemma.
(11)
Uniqueness for an hyperbolic inverse problem with time-dependent coefficient 71
Lemma 3.2 Let M1 , M2 > 0, a ∈ A(a0 , M1 ) and b ∈ W 1,∞ (Q) such that ∥b∥W 1,∞ (Q) ≤
M2 . Given ω ∈ S n−1 and ϕ ∈ C0∞ (Rn ), we consider the function φ defined by (10). Then,
for any λ > 0, the following equation
❇t2 u − ∆u + a(x, t)❇t u + b(x, t)u = 0
in Q,
(12)
admits a unique solution
u+ ∈ C([0, T ]; H 1 (Ω)) ∩ C 1 ([0, T ]; L2 (Ω)),
of the following form
u+ (x, t) = φ(x, t)̃
a+ (x, t)eiλ(x⋅ω+t) + r+ (x, t),
(13)
1 t
∫ a(x + (t − s)ω, s) ds),
2 0
(14)
where ̃
a+ (x, t) is given by
̃
a+ (x, t) = exp ( −
and r+ (x, t) satisfies
r+ (x, 0) = ❇t r+ (x, 0) = 0, in Ω,
r+ (x, t) = 0 on Σ.
(15)
Moreover, there exists a positive constant C > 0 such that
λ∥r+ ∥L2 (Q) + ∥❇t r+ ∥L2 (Q) ≤ C∥ϕ∥H 3 (Rn ) .
(16)
In order to prove this lemma, it will be enough to prove the existence of r+ satisfying
⎧
⎪
(❇t2 − ∆ + a(x, t)❇t + b(x, t))r+ = g(x, t),
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨ r+ (x, 0) = ❇t r+ (x, 0) = 0,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
+
⎪
⎪
⎩ r (x, t) = 0,
(17)
and obeying the estimate (16), where g(x, t) is given by
g(x, t) = −(❇t2 − ∆ + a(x, t)❇t + b(x, t))(φ(x, t)̃
a+ (x, t)eiλ(x⋅ω+t) ).
Bearing in mind that ̃
a+ (x, t) solves the following equation
2(❇t − ω ⋅ ∇)̃
a+ (x, t) = −a(x, t)̃
a+ (x, t),
one can see from (11) that
g(x, t) = −eiλ(x⋅ω+t) (❇t2 −∆+a(x, t)❇t +b(x, t))(φ(x, t)̃
a+ (x, t)) = −eiλ(x⋅ω+t) g0 (x, t),
where g0 ∈ L1 (0, T, L2 (Ω)). Hence, in light of Lemma 3.1, we deduce the existence of a
unique solution r+ such that
r+ ∈ C([0, T ]; H01 (Ω)) ∩ C 1 ([0, T ]; L2 (Ω)),
72 ARIMA - volume 23 - 2016
and satisfying (17). We set
t
w(x, t) = ∫
0
r+ (x, s) ds.
(18)
Then, in light of (17) and (18), we get
(❇t2 − ∆ + a(x, t)❇t + b(x, t))w(x, t) = ∫
0
t
+∫
g(x, s) ds + ∫
t
0
t
0
(b(x, t) − b(x, s))r+ (x, s) ds
❇s a(x, s)r+ (x, s) ds,
Then, the function w is a solution to the following equation
⎧
⎪
(❇t2 − ∆ + a(x, t)❇t + b(x, t))w(x, t) = F1 (x, t) + F2 (x, t) in Q,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨ w(x, 0) = 0 = ❇t w(x, 0)
in Ω,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
on Σ.
⎩ w(x, t) = 0
Here F1 and F2 are given by
F1 (x, t) = ∫
t
(19)
g(x, s) ds,
0
and
F2 (x, t) = ∫
t
0
(b(x, t) − b(x, s))r+ (x, s) ds + ∫
t
❇s a(x, s)r+ (x, s) ds.
0
Considering τ ∈ [0, T ] and applying Lemma 3.1 on the interval [0, τ ], we obtain
∥❇t w(., τ )∥2L2 (Ω)
≤ C(∥F1 ∥2L2 (Q) + T (M12 + 4M22 ) ∫
τ
0
∫ ∫
Ω
0
t
∣r+ (x, s)∣2 ds dx dt).
Therefore, in view of (18), we deduce that
∥❇t w(., τ )∥2L2 (Ω)
≤ C(∥F1 ∥2L2 (Q) + ∫
0
τ
≤ C(∥F1 ∥2L2 (Q) + T ∫
∫
0
t
τ0
∥❇s w(., s)∥2L2 (Ω) ds dt)
∥❇s w(., s)∥2L2 (Ω) ds).
As a consequence, we find out from Gronwall’s Lemma that
∥❇t w(., τ )∥2L2 (Ω) ≤ C∥F1 ∥2L2 (Q) .
Hence, from (18), one deduce that ∥r+ ∥L2 (Q) ≤ C∥F1 ∥L2 (Q) . In view of (19), one can
easily see that F1 can be written as follows
t
t
1
iλ(x⋅ω+s)
) ds.
∫ g0 (x, s)❇s (e
iλ 0
0
Therefore, by integrating by parts with respect to s, we get
F1 (x, t) = ∫
g(x, s) ds =
∥r+ ∥L2 (Q) ≤
C
∥ϕ∥H 3 (Rn ) ,
λ
for some C > 0. Since ∥g∥L2 (Q) ≤ C∥ϕ∥H 3 (Rn ) and using the energy estimate (9) associated to the problem (17) we obtain the following estimation
∥❇t r+ ∥L2 (Q) + ∥∇r+ ∥L2 (Q) ≤ C∥ϕ∥H 3 (Rn ) .
This completes the proof of the lemma.
Uniqueness for an hyperbolic inverse problem with time-dependent coefficient 73
Lemma 3.3 Let M1 , M2 > 0, a ∈ A(a0 , M1 ) , and b ∈ W 1,∞ (Q) such that ∥b∥W 1,∞ (Q) ≤
M2 . Given ω ∈ S n−1 and ϕ ∈ C0∞ (Rn ), we consider the function φ defined by (10). Then,
the following equation
❇t2 u − ∆u − a(x, t)❇t u + b(x, t)u = 0 in Q,
(20)
admits a unique solution
of the following form
u− ∈ C([0, T ]; H 1 (Ω)) ∩ C 1 ([0, T ]; L2 (Ω)),
u− (x, t) = ϕ(x + tω)̃
a− (x, t)e−iλ(x⋅ω+t) + r− (x, t),
(21)
1 t
̃
a− (x, t) = exp ( ∫ a(x + (t − s)ω, s) ds),
2 0
(22)
where ̃
a (x, t) is given by
−
and r− (x, t) satisfies
r− (x, T ) = ❇t r− (x, T ) = 0, in Ω,
Moreover, there exists a constant C > 0 such that
r− (x, t) = 0 on Σ.
λ∥r− ∥L2 (Q) + ∥❇t r− ∥L2 (Q) ≤ C∥ϕ∥H 3 (Rn ) .
(23)
(24)
We prove this result by proceeding as in the proof of Lemma 3.2. Putting
̃
g (x, t) = −(❇t2 − ∆ − a(x, t)❇t + b(x, t))(φ(x, t)̃
a− (x, t)e−iλ(x⋅ω+t) ).
Then, it is easy to see that if r− (x, t) is solution to the following system
⎧
⎪
g (x, t) in Q,
(❇t2 − ∆ − a(x, t)❇t + b(x, t))r− (x, t) = ̃
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨ r− (x, T ) = 0 = ❇t r− (x, T )
in Ω,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
−
⎪
⎪
on Σ,
⎩ r (x, t) = 0
+
−
then, r (x, t) = r (x, T − t) is a solution to (17) with g(x, t) = ̃
g (x, T − t) and a(x, t),
b(x, t) are replaced by a(x, T − t) and = b(x, T − t).
4. Proof of Theorem 1.2
This section is devoted to the proof of Theorem 1.2. The proof is based on the geometric optics solutions constructed in Section 3 and the following preliminary identity.
We need first to introduce the following notations. Let ω ∈ S n−1 , a1 , a2 ∈ A(a0 , M ). We
set
1 t
̃
a(x, t) = (̃
a− ̃
a+ )(x, t) = exp ( − ∫ a(x + (t − s)ω, s) ds),
2 0
where ̃
a−1 and ̃
a+2 are given by
1 t
1 t
̃
a+2 (x, t) = exp (− ∫ a2 (x+(t−s)ω, s) ds).
a−1 (x, t) = exp ( ∫ a1 (x+(t−s)ω, s) ds), ̃
2 0
2 0
Moreover, we define a in Rn+1 by a = a2 − a1 in Qr and a = 0 on Rn+1 ∖ Qr .
74 ARIMA - volume 23 - 2016
4.1. An identity for the absorbing coefficient
The main purpose of this section is to give a preliminary identity for the absorbing
coefficient a .
Lemma 4.1 Let ϕ ∈ C0∞ (Ar ) and a1 , a2 ∈ A(a0 , M ). Assume that Λa2 = Λa1 , then, the
following identity holds
2
a(x, t) dx dt = 0.
∫ a(x, t)ϕ (x + tω)̃
(25)
Q
In light of Lemma 3.2, there exists a geometrical optics solution u+ to the equation
in the following form
⎧
❇ 2 u+ − ∆u+ + a2 (x, t)❇t u+ = 0 in Q,
⎪
⎪
⎪ t
⎨
⎪
+
+
⎪
⎪
in Ω,
⎩ u (x, 0) = ❇t u (x, 0) = 0
u+ (x, t) = ϕ(x + tω)̃
a+2 (x, t)eiλ(x⋅ω+t) + r+ (x, t),
(26)
corresponding to the coefficients a2 , where r+ (x, t) satisfies (15), (16). We denote by fλ
the function
fλ (x, t) = u+ (x, t)∣Σ = ϕ(x + tω)̃
a+2 (x, t)eiλ(x⋅ω+t) .
We denote by u1 the solution of
⎧
❇t2 u1 − ∆u1 + a1 (x, t)❇t u1 = 0 in Q,
⎪
⎪
⎪
⎪
⎪
⎪
⎪
in Ω,
⎨ u1 (x, 0) = ❇t u1 (x, 0) = 0
⎪
⎪
⎪
⎪
⎪
⎪
⎪
on Σ.
⎩ u1 (x, t) = fλ (x, t)
Putting u = u1 − u+ . Then, u is a solution to the following system
⎧
❇t2 u − ∆u + a1 (x, t)❇t u = a(x, t)❇t u+
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨ u(x, 0) = ❇t u(x, 0) = 0
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎩ u(x, t) = 0,
in Q,
in Ω,
(27)
on Σ.
where a = a2 −a1 . On the other hand Lemma 3.3 guarantees the existence of a geometrical
optic solution u− to the adjoint problem of (1)
⎧
❇ 2 u− − ∆u− − a1 (x, t)❇t u− − ❇t a1 (x, t)u− = 0 in Q,
⎪
⎪
⎪ t
⎨
⎪
−
−
⎪
⎪
in Ω,
⎩ u (x, T ) = ❇t u (x, T ) = 0
corresponding to the coefficients a1 and −❇t a1 , in the form
u− (x, t) = ϕ(x + tω)e−iλ(x⋅ω+t)̃
a1 − (x, t) + r− (x, t),
(28)
where r− (x, t) satisfies (23), (24). Multiplying the first equation of (27) by u− , integrating
by parts and using Green’s formula, we obtain
−
+ −
∫ a(x, t)❇t u u dx dt = ∫ (Λa2 − Λa1 )(fλ ) u dσ dt.
Q
Σ
(29)
Uniqueness for an hyperbolic inverse problem with time-dependent coefficient 75
On the other hand, by replacing u+ and u− by their expressions, we get
iλ(x⋅ω+t) + −
+ −
̃
a2 r dx dt
∫ a(x, t)❇t u u dx dt = ∫ a(x, t)❇t ϕ(x + tω)e
Q
Q
a−1 )dx dt
a+2 r− dx dt + ∫ a(x, t)❇t ϕ(x + tω)ϕ(x + tω)(̃
a+2 ̃
+ ∫ a(x, t)ϕ(x + tω)eiλ(x⋅ω+t) ❇t̃
Q
+ ∫ a(x, t)ϕ
Q
2
a−1 dx dt +
a+2 ̃
(x + tω)❇t̃
iλ ∫
Q
T
0
iλ(x⋅ω+t) + −
̃
a2 r dx dt
∫ a(x, t)ϕ(x + tω)e
Ω
+ ∫ a(x, t)ϕ(x + tω)e−iλ(x⋅ω+t)̃
a−1 ) dx dt
a−1 ❇t r+ dx dt + iλ ∫ a(x, t)ϕ2 (x + tω)(̃
a+2 ̃
Q
Q
a dx dt + I(λ),
+ ∫ a(x, t)❇t r+ r− dx dt = iλ ∫ a(x, t)ϕ2 (x + tω)̃
Q
Q
a−1 . Then, in light of (29), we have
where ̃
a=̃
a+2 ̃
a(x, t) dx dt = ∫ (Λa2 − Λa1 )(fλ ) u− dσ dt − I(λ).
iλ ∫ a(x, t)ϕ2 (x + tω)̃
(30)
Σ
Q
Note that for λ sufficiently large, we have
∣Iλ ∣ ≤ C∥ϕ∥2H 3 (Rn ) .
(31)
Hence, using the fact that Λa2 = Λa1 , we deduce from (30), (31) and by taking λ → +∞
the desired result.
4.2. End of the proof
In this section we complete the proof of Theorem 1.2 by the use of the results we have
already obtained in the previous sections. Let us first consider the following set
E = {(ξ, τ ) ∈ Rn ∖ {ORn } × R, ∣τ ∣ < ∣ξ∣},
and denote by F̂ the Fourier transform of F ∈ L1 (Rn+1 ) as follows:
F̂(ξ, τ ) = ∫ ∫
R
Rn
F (x, t)e−ix⋅ξ e−itτ dx dt.
In light of (25), we have as λ goes to +∞, the following identity
2
∫ a(x, t)ϕ (x + tω) exp ( −
Q
1 t
∫ a(x + (t − s)ω, s) ds) dx dt = 0.
2 0
(32)
Then, using the fact a(x, t) = 0 outside Qr,∗ and making this change of variables y =
x + tω, one gets
∫
T
0
∫
Rn
a(y − tω, t)ϕ2 (y) exp ( −
1 t
∫ a(y − sω, s) ds) dy dt = 0.
2 0
Bearing in mind that
∫
0
T
∫
Rn
1 t
∫ a(y − sω, s) ds) dy dt
2 0
T
1 t
d
= −2 ∫ ∫ ϕ2 (y) [ exp ( − ∫ a(y − sω, s) ds)] dy dt
dt
2 0
0
Rn
1 T
2
= −2 ∫ ϕ (y)[ exp ( − ∫ a(y − sω, s) ds) − 1] dy.
2 0
Rn
a(y − tω, t) ϕ2 (y) exp ( −
76 ARIMA - volume 23 - 2016
we conclude that
∫
Rn
ϕ2 (y)[ exp ( −
1 T
∫ a(y − sω, s) ds) − 1] dy = 0.
2 0
(33)
Now, we consider a nonnegative function ψ ∈ C0∞ (Rn ) supported in the unit ball B(0, 1)
such that ∥ψ∥L2 (Rn ) = 1. We define
ϕh (x) = h−n/2 ψ(
x−y
),
h
(34)
where y ∈ Ar . Then, for h > 0 sufficiently small one can see that Supp ϕh ⊂ C0∞ (Ar ) and
satisfies
Supp ϕh ∩ Ω = ∅, and Supp ϕh ± T ω ∩ Ω = ∅.
Then, as h goes to 0 we deduce from (33) with ϕ = ϕh that
exp ( −
1 T
∫ a(y − sω, s) ds) − 1 = 0.
2 0
Since a = a2 − a1 = 0 outside Qr,∗ , we conclude that
∫ a(y − tω, t) dt = 0,
R
a.e y ∈ Ar , ω ∈ S n−1 .
(35)
r
On the other hand, if ∣y∣ ≤ , we notice that
2
a(y − tω, t) = 0, ∀ t ∈ R.
Indeed, we have
(36)
r
(37)
∣y − tω∣ ≥ ∣t∣ − ∣y∣ ≥ t − ,
2
hence, (y − tω, t) ∉ Cr+ if t > r/2, from (37). As (y − tω, t) ∉ Cr+ if t ≤ r/2, then we have
(y − tω, t) ∉ Cr+ ⊃ Qr,∗ for t ∈ R. This and the fact that a = a2 − a1 = 0 outside Qr,∗ , yield
(36), and consequently,
∫ a(y − tω, t) dt = 0,
R
r
∣y∣ ≤ .
2
By a similar way, we prove for ∣y∣ ≥ T − r/2, that (y − tω, t) ∉ Cr− ⊃ Qr,∗ , for t ∈ R. Then
we obtain
n−1
(38)
∫ a(y − tω, t) dt = 0, a.e. y ∉ Ar , ω ∈ S .
R
Thus, by (35) and (38) we find
n
n−1
∫ a(y − tω, t) dt = 0, a.e y ∈ R , ω ∈ S .
R
We now turn our attention to the fourier transform of a. Let ξ ∈ Rn . In light of (38) and
by the use of Fubini’s Theorem, we get
∫ ∫
R
Rn
a(x − tω, t)e−ix⋅ξ dx dt = 0.
Making the change of variables y = x − tω, one gets
∫ ∫
R
Rn
a(y, t)e−iy⋅ξ e−it(ω⋅ξ) dy dt = 0.
Uniqueness for an hyperbolic inverse problem with time-dependent coefficient 77
Let us now consider ξ ′ ∈ S n−1 such that ξ ⋅ ξ ′ = 0. Setting
¿
2
Á
τ
À1 − τ ⋅ ξ ′ ∈ S n−1 ,
ω = 2 ⋅ξ+Á
∣ξ∣
∣ξ∣2
then (ξ, τ ) = (ξ, ω ⋅ ξ) ∈ E. We then deduce that ̂
a(ξ, τ ) = 0 in the set E. By an argument
of analyticity, we extend this result to Rn+1 . Hence, by the injectivity of the Fourier
transform we get the desired result. This completes the proof of Theorem 1.2.
5. References
[1] M. B ELLASSOUED, , M. C HOULLI, , M. YAMAMOTO, Stability estimate for an inverse
wave equation and a multidimensional Borg-Levinson theorem, J. Diff. Equat., vol. 247,
2,num. 465-494, 2009.
[2] M. B ELLASSOUED, , D. D OS S ANTOS F ERREIRA, Stability estimates for the anisotripic wave
equation from the Dirichlet-to-Neumann map , Inverse Probl. Imaging, 5, 4, 745-73, 2011.
[3] M. B ELLASSOUED, , D. J ELLALI „ M. YAMAMOTO, Lipschitz stability for a hyperbolic
inverse problem by finite local boundary data , Applicable Analysis, 85, 1219-1243, 2006.
[4] M. B ELLASSOUED, , D. J ELLALI, , M. YAMAMOTO, Stability estimate for the hyperbolic
inverse boundary value problem by local Dirichlet-to-Neumann map , J. Math. Anal. Appl.,
343, 2, 1036-1046, 2008.
[5] I. B EN A ÏCHA I BTISSEM , Stability estimate for a hyperbolic inverse problem with timedependent coefficient , Inverse Problems, 31, 125010, 21pp, 2015.
[6] R. C IPOLATTI, , I VO F. L OPEZ , Determination of coefficients for a dissipative wave equation
via boundary measurements , J. Math. Anal. Appl., 306, 317-329, 2005.
[7] G. E SKIN , A new approach to hyperbolic inverse problems ,
815-831, 2006.
Inverse problems, 22 no. 3,
[8] G. E SKIN , Inverse hyperbolic problems with time-dependent coefficients , Commun. Partial
Diff. Eqns., 32, 11, 1737-1758, 2007.
[9] M. I KAWA , A mixed problem for hyperbolic equations of second order with a first order
derivative boundary condition , Publ.RIMS, Kyoto Univ,1969.
[10] M. I KAWA , Hyperbolic Partial Differential Equations and Wave Phenomena , Providence, RI
American Mathematical Soc., 2000.
[11] V. I SAKOV, An inverse hyperblic problem with many boundary measurements , Commun.
Partial Diff. Eqns., 16, 1183-1195, 1991.
[12] V. I SAKOV, Completeness of products of solutions and some inverse problems for PDE ,
J.Diff. Equat., 92, 305-316, 1991.
[13] V. I SAKOV, , Z. S UN , Stability estimates for hyperbolic inverse problems with local boundary data , Inverse problems, 8, 193-206, 1992.
[14] Y. K IAN , “ Unique determination of a time-dependent potential for wave equations from
partial data, preprint, arXiv:1505.06498.
[15] Y. K IAN , “ Stability in the determination of a time-dependent coefficient for wave equations
from partial data , arXiv:1406.5734.
[16] R AKESH, , W. S YMES ,“ Uniqueness for an inverse problem for the wave equation , Comm.
in PDE, 13, 1, 87-96, 1988.
[17] A. G. R AMM, , R AKESH , “ Property C and an Inverse Problem for a Hyperbolic Equation ,
J.Math. Anal. Appl., 156, 209-219, 1991.
78 ARIMA - volume 23 - 2016
[18] A.G. R AMM, , S JÖSTRAND , An inverse problem of the wave equation , Math. Z., 206,
119-130, 1991.
[19] R. S ALAZAR , Determination of time-dependent coefficients for a hyperbolic inverse problem
, Inverse Problems, 29, 9, 095015, 2013.
[20] P. S TEFANOV, Uniqueness of the multi-dimentionnal inverse scattering problem for timedependent potentials, Math. Z., 201, 4, 541-559, 1994.
[21] Z. S UN , On continuous dependence for an inverse initial boundary value problem for the
wave equation , J. Math. Anal. Appl., 150, 188-204, 1990.
[22] A. Waters, Stable determination of X-ray transforms of time-dependent potentials from partial
boundary data , Commun. Partial Diff. Eqns., 39, 2169-2197, 2014.