Simultaneously determine elastic impedance and

shape by a Newton-type iterative method

Yao Sun†,∗ , Pan Wang†
∗ Corresponding author: sunyao10@[Link]
arXiv:2401.00236v3 [[Link]] 18 Mar 2024

† College of Science, Civil Aviation University of China, Tianjin, China

Abstract. This paper focuses on an indirect boundary integral equation method

for the inverse elastic impedance and the geometry problem by a Cauchy data pair
on the access part of the boundary in a two-dimensional case. A uniqueness result
is given for the corresponding problem, and a non-iterative algorithm is proposed
to solve the data completion problem using a Cauchy data pair on an accessible
part of the solution domain’s boundary. Next, we introduce a Newton-type iterative
method for reconstructing the missing boundary and the impedance function using
the completion data on the unknown boundary, which is governed by a specific type
of boundary conditions. This method should not deal with the singularities of the
kernels of hypersingular integral from Fréchet derivative. Finally, we provide several
examples to demonstrate the effectiveness and accuracy of the proposed method.
Key words: Indirect boundary integral equation, Cauchy data, Collocation
method.

1 Introduction
The study of elastic waves has become an important research area in mathemat-
ical physics. This type of problem is widely used in nondestructive testing, medical
imaging and seismic exploration [3, 4]. A typical inverse problem involves determin-
ing the boundary shape of an elastic obstacle from the Cauchy data on the acces-
sible part boundary. In practical applications such as physics and engineering, due
to multiple factors including measurement technology limitations, it is not possible
to obtain complete information about the surface of obstacles. Therefore, it is nec-
essary to consider supplementing the information about the non-accessible/missing
part boundary. In the case of electrostatics, the model yields an inverse boundary
value problem related to the Laplace equation. In the case of elastic obstacles, an
inverse boundary value problem arises in relation to the Navier equation. Litera-
ture references can be consulted for both cases [1, 13, 23, 35]. The inverse acoustic
scattering can be found in [20, 21, 29]. In many inverse problems, it is not realistic

1
to assume that the boundary conditions are known. This is the physical motivation
for simultaneously determine the elastic impedance and shape stems from the prac-
tical problems. Carrying this idea further we can ask if it is possible, by modifying
the impedance values of the boundary of an obstacle, to make it appear from a
experiment standpoint as some object of a different shape.
Various methods have been investigated in recent years to solve the inverse
elastic obstacle problem. In [28], Li et al. investigate the inverse elastic scattering
problem for biperiodic surfaces in three dimensions. A local uniqueness is proved
for the inverse problem when the scattering surface has been assumed to be a small
and smooth perturbation of a rigid planar surface. Louër [32,33] derives the domain
derivatives of elastic scattering, which can be employed to reconstruct unknown
elastic obstacles through the design of gradient descent methods. Additionally, Alves
and Kress [2] examine a linear sampling method, which is based on far-field operator
decomposition, for inverse elastic scattering problem. Factorization method has
been extensively researched by Hu et al. in [17, 18]. The inverse elastic impedance
problem, as discussed in [30, 42], has found widespread application in seismology,
oilfield exploration, and other related fields. There are some other research papers
for the inverse elastic problem, such as [14, 31, 40] and etc.
In [25, 27], Kress and Rundell consider the inverse problems of time-harmonic
acoustic or electromagnetic waves. A regularized Newton iteration is given to solve
the inverse problem of simultaneously determining the impedance and shape of a
two-dimensional scatterer, which is based on a knowledge of the far-field pattern.
The main idea is to establish the ill-posed nonlinear equation for the operator that
maps the boundary and the boundary impedance of the scatterer onto the far-field
pattern. Later, Kress and Rundell [26] determine the shape of a perfectly conducting
inclusion via a pair of nonlinear and ill-posed integral equations for the unknown
boundary. The integral equations can be solved by the linearization corresponding
to the regularized Newton iterations. This idea is to extend by Cakoni and Kress [7]
to a simply connected planar domain from a pair of Cauchy data on the known
boundary curve. This method is also used to complete the Cauchy data on the
unknown boundary curve, and then an impedance profile is given by a point to point
method. Thus the boundary curve is to recover by a known impedance profile.
Motivated by the idea in [25] proposed by Kress and Rundell, we aim to demon-
strate a Newton-type iterative method for reconstructing the elastic impedance func-
tion and the shape of the elastic body based on the integral equations. Different
from the previous work [25] by a direct integral approach for the mixed problem, we
consider an indirect integral equation by the approach of having an integral operator
defined on a virtual boundary enclosing the scatterer. Though the direct integral
approach would provide some more information about the curve. It should deal
with the singularities of the kernels on the boundary especially the strongly singular
integral resulting the hypersingular integral from Fréchet derivative. The indirect
integral approach will easily give the Fréchet derivative of the boundary integral
operator. The contributions of this paper are as follows:

2
 (i) A uniqueness result is given for the corresponding problem.
(ii) A Newton-type iterative method for reconstructing the boundary and the
impedance function.
The paper is organized as follows. In Section 2, we introduce the inverse
impedance problem and the inverse shape problem. In Section 3, some theoretical
results including a unique result are given by introducing the Helmholtz decomposi-
tion of the displacement. Then in Section 4, we specifically demonstrate linearization
of the integral equations and the Newton-type iterative method for solving inverse
shape problem. Then some numerical examples are provided to demonstrate the
effectiveness of the proposed linearized iterative scheme.

2 Problem formulation
Let D ⊂ R2 be an open bounded domain with Lipschitz boundary ∂D, which
is occupied by homogeneous and isotropic elastic medium with the density ρ. As-
suming that the boundary ∂D be analytical and can be given as ∂D = Γ0 ∪ Γm ,
Γ0 ∩ Γm = ∅, where Γ0 and Γm are the known part and the unknown part of the
boundary ∂D. λ and µ, usually called Lemé constants, satisfy µ > 0 and λ + µ > 0.
ω designates the frequency of vibration. The displacement field u corresponding to
the stress tensor σ(u) inside D satisfies the following Navier equation

∇ · σ(u) + ρω 2 u = 0, in D. (2.1)

For a linear isotropic elastic medium, the components of the stress tensor σ(u)
be
σı = λδı εℓℓ + 2µεı , ı,  = 1, 2,
with the strain tensor εı given by
 
1 ∂uı ∂u
εı = + ,
2 ∂x ∂xı

where δı is the Kronecker delta function.

On the boundary ∂D, the displacement u satisfies the following boundary con-
ditions:

• Dirichlet boundary condition on Γ0

u = f (x, ω), on Γ0 . (2.2)

• Impedance boundary condition on Γm

Tn u + iωχu = g(x, ω), on Γm . (2.3)

3
The impedance function χ ≥ 0 is continuous.
For a fixed point x ∈ ∂D, T the traction operator, defined on the boundary
∂D, is given by
 
∂u ∂uℓ ∂u
(Tn u)ℓ := λnℓ + µn + , ℓ = 1, 2,
∂x ∂x ∂xℓ
where n is the unit outward normal vector.
For a scalar function w, we have
 ⊤  ⊤
∂w ∂w ⊥ ∂w ∂w
∇w = , , ∇ w= ,− .
∂x1 ∂x2 ∂x2 ∂x1
It is well-known that the displacement field u can be splited into the compressional
and shear parts, i.e. the Helmholtz decomposition as follows

u = ∇up + ∇⊥ us . (2.4)

up and us , respectively, are the solutions of the following Helmholtz equations

∆up + κ2p up = 0, ∆us + κ2s us = 0.

κp and κs are the

qwave numbers of qthe compressional part and the shear part with
ρω 2 2
the values κp = λ+2µ and κs = ρωµ .
Together with the equations (2.1) − (2.4), we get the problem considered as
follows


 ∆up + κ2p up = 0, in D,

∆u + κ2 u = 0,
s s s in D,
(2.5)
u = f (x, ω),
 on Γ0 ,


Tn u + iωχu = g(x, ω), on Γm .

Inverse Problem: The inverse problem we considered is to determine the

shape of the unknown portion Γm and impedance function χ from the displacement
field f (x, ω) and the (measured) traction t(x, ω) = Tn u on Γ0 .

3 Theoretical results
We denote Φ(κ|x − y|) be the fundamental solution of the Helmholtz equation
associated with the wavenumber κ, namely
i (1)
Φ(κ|x − y|) = H0 (κ|x − y|), (3.1)
4
(1)
with H0 be the Hankel function of the first kind of order zero.

4
 The displacement is u = ∇up + ∇⊥ us . The vector components of the displace-
ment u will be
Z
up (x) = Φ(κp |x − y|)ϕ1 (y)ds(y), x ∈ D, (3.2)
∂B

and Z
us (x) = Φ(κs |x − y|)ϕ2 (y)ds(y), x ∈ D, (3.3)
∂B
where ϕ1 , ϕ2 are the density functions, D ⊂⊂ B, ∂B is the so called virtual bound-
ary.
Combining (3.2), (3.3) and the definition of ∇⊥ for a scalar function, we have
Z
u(x) = E(x, y)ϕ(y)ds(y), x ∈ D, (3.4)
∂B

with ϕ = (ϕ1 , ϕ2 )⊤ and

!
∂Φ(κp |x−y|) ∂Φ(κs |x−y|)
∂x1 ∂x2
E(x, y) = ∂Φ(κp |x−y|) ∂Φ(κs |x−y|) .
∂x2
− ∂x1

Denote the unit outward normal vector by n at x ∈ ∂D, and the corresponding
traction Tnx u is given
Z
(Tnx u)(x) = T(x, y)ϕ(y)ds(y), x ∈ ∂D, (3.5)
∂B

where the elements of the matrix T(x, y) are given by

   
∂E1 (x, y) ∂E2 (x, y) ∂E1 (x, y) ∂E2 (x, y)
T1 (x, y) = (λ + 2µ) +λ n1 + µ + n2 ,
∂x1 ∂x2 ∂x2 ∂x1
   
∂E1 (x, y) ∂E2 (x, y) ∂E1 (x, y) ∂E2 (x, y)
T2 (x, y) = µ + n1 + λ + (λ + 2µ) n2 .
∂x2 ∂x1 ∂x1 ∂x2
Denote by
Z
Sϕ(x) := E(x, y)ϕ(y)ds(y), x ∈ R2 \ ∂B, (3.6)
∂B

Z
Kϕ(x) := Tnx E(x, y)ϕ(y)ds(y), x ∈ ∂D (3.7)
∂B

and
Z
Sϕ(x) := E(x, y)ϕ(y)ds(y), x ∈ ∂D. (3.8)
∂B

We have the following result, and the proof is similar to the proof of three dimen-
sional case [41].

5
Theorem 1. The operator S is compact, injective and has dense range if ρω 2 is not
a Dirichlet eigenvalue of the negative lamé operator in the interior D.

On Γ0 , define the operator N by

!
Sϕ(x)
N ϕ(x) : = , x ∈ Γ0 .
Tnx Sϕ(x)

Theorem 2. The operator N : L2 (∂B) → H1/2 (Γ0 ) × H−1/2 (Γ0 ) is compact, injec-
tive and has dense range.

Proof. The operator N : L2 (∂B) → H 1/2 (Γ0 ) × H −1/2 (Γ0 ) is compact by the kernel
function of N being analytic [6, Theorem 2.7].
If we let N ϕ = 0, it should be injective by ϕ = 0. The uniqueness of the
Cauchy problem yields that Sϕ = 0 in D. A unique continuation argument yields
that Sϕ = 0 in B.
In fact
Z Z
⊥
Sϕ(x) = ∇x Φ(κp |x − y|)ϕ1 (y)ds(y) + ∇x Φ(κs |x − y|)ϕ2(y)ds(y). (3.9)
∂B ∂B

Denote Z
φ(x) = Φ(κp |x − y|)ϕ1(y)ds(y), x ∈ R2 \ ∂B,
∂B
and Z
ψ(x) = Φ(κs |x − y|)ϕ2(y)ds(y), x ∈ R2 \ ∂B,
∂B
we have

∇φ = −∇⊥ ψ, in B. (3.10)

Taking the dot product with ∇ and ∇⊥ , respectively, thus we have

△φ = 0, △ψ = 0, in B. (3.11)

Noting that φ and ψ satisfy the Helmholtz equations, we get

φ = ψ = 0, in B, (3.12)

and thus
∂φ− ∂ψ−
= = 0, on ∂B, (3.13)
∂ν ∂ν
where ν denotes the unit outward normal vector defined on ∂B and − denotes the
limits as x → ∂B from inside of ∂B.

6
 From the jump relation of the single-layer potential functions (see e.g. [6]), we
can get
∂φ− ∂φ+
φ+ = φ− and − = ϕ1 , on ∂B, (3.14)
∂ν ∂ν
where + denotes the limits as x → ∂B from outside of ∂B.
From equation (3.14), we know φ+ = 0 and − ∂φ +
∂ν
= ϕ1 on ∂B. This means
that φ satisfies the homogeneous exterior Dirichlet boundary value problem with the
Sommerfeld radiation condition, the uniqueness for the exterior Dirichlet problem
yields φ = 0 outside of ∂B, and thus ∂φ ∂ν
+
= 0. Combining with equation (3.14),
we have ϕ1 = 0. The same augment will give ϕ2 = 0. Therefore the operator N is
injective.
Next we prove that N has dense range. We consider the adjoint operator
N ∗ : H1/2 (Γ0 ) × H−1/2 (Γ0 ) → L2 (∂B) defined by

hN ϕ, [ξ1, ξ2 ]i = hϕ, N ∗ [ξ1 , ξ2 ]i. (3.15)

We want to show that N ∗ is injective which implies that N has dense range.
Let N ∗ (ξ1 , ξ2) = 0. Noted that
Z Z
∗ ⊤
N (ξ1 , ξ2 ) = E (x, y)ξ1(x)ds(x) + T⊤ (x, y)ξ2(x)ds(x), y ∈ ∂B. (3.16)
Γ0 Γ0

Denote
Z
′
w (y) = E⊤ (x, y)ξ1(x)ds(x), y ∈ R2 \ Γ0 , (3.17)
Γ0

and
Z
′′
w (y) = T⊤ (x, y)ξ2(x)ds(x), y ∈ R2 \ Γ0 . (3.18)
Γ0

We set W (y) = w ′ (y) + w ′′ (y). The components of W (y) are W1 (y) and
W2 (y). W1 (x) satisfies
(
△W1 + κ2p W1 = 0, in R2 \ B,
(3.19)
W1 (y) = 0, on ∂B.

and the Sommerfeld radiation condition. The uniqueness for the exterior Dirichlet
problem yields W1 (y) = 0 outside of ∂B. A unique continuation argument yields
that W1 (y) = 0 in R2 \ Γ0 , thus also W2 (y) = 0 in R2 \ Γ0 .
Denote
f (y) = ∇W1 (y) + ∇⊥ W2 (y), y ∈ R2 \ Γ0 ,
W (3.20)

7
we have
f (y)+ = W
W f (y)− = 0 and Tn W
f (y)+ = Tn W
f (y)− = 0, on Γ0 . (3.21)

On the other hand, we have

Z Z
f (y) = ρω
W 2
Γ(x, y)ξ1(x)ds(x) + ρω 2
Tnx Γ(x, y)ξ2(x)ds(x),
Γ0 Γ0

where Γ(x, y) is the fundamental matrix (see e.g. [17, 34]).

Let α1 and α2 be the extensions of ξ1 and ξ2 by zero to the whole boundary
∂D. We have
Z Z
f (y) = ρω
W 2
Γ(x, y)ξ1(x)ds(x) + ρω 2
Tnx Γ(x, y)ξ2(x)ds(x) (3.22)
Γ0 Γ0
Z Z
2 2
= ρω Γ(x, y)α1(x)ds(x) + ρω Tnx Γ(x, y)α2(x)ds(x).
∂D ∂D

From the jump relations of single-layer and double-layer potential functions (see
e.g. [14, 24, 34]), we can get

f (y)+ − W
W f (y)− = ρω 2 α2 , on ∂D, (3.23)

and
f (y)− − Tn W
Tn W f (y)+ = ρω 2 α1 , on ∂D. (3.24)

Thus, by the equations (3.21), (3.23) and (3.24), we have

α1 = α2 = 0.

Thus we have
ξ1 = ξ2 = 0.
It means N ∗ is injective, and N has dense range.

Remark 3.1. In practice, given the Cauchy data pair f and t, we solve the corre-
sponding ill-posed equations by using regularization methods such as Tikhonov regu-
larization. Since the L2 -norm is the appropriate norm to measure the data error, it
is natural to apply the regularization scheme in the space of square integrable func-
tions. For this reason, we consider the operator N : L2 (∂B) → L2 (Γ0 ) × L2 (Γ0 ),
and theorem 2 is also correct in L2 -norm

After this, we write the problem considered in this paper. First, we approximate
the solution u of equation (2.1). In this step, we should find the density function
ϕ ∈ L2 (∂B) by
N ϕ = h, (3.25)

8
here h = (f |Γ0 , t|Γ0 )⊤ , and then Sϕ(x) will approximate the displacement u in D.
Second, a Newton type method is used to recover the missing boundary Γm and the
impedance function χ.
Before giving the uniqueness result, we should first give a property of the solu-
tion u(x, ω).
Lemma 3.1. The displacement u(x, ω) depends analytically on ω ∈ C, if h(x, ω)
depends analytically on ω ∈ C.
Proof. We seek the displacement field in the form as
Z
u(x) = E(x, y)ϕ(y)ds(y), x ∈ D, (3.26)
∂B

with ϕ = (ϕ1 , ϕ2 )⊤ and

!
∂Φ(κp |x−y|) ∂Φ(κs |x−y|)
∂x1 ∂x2
E(x, y) = ∂Φ(κp |x−y|) ∂Φ(κs |x−y|) .
∂x2
− ∂x1

Since the fundamental solution to the Helmholtzq

equation depends analytically
q on
ρ ρ
wavenumber κp or κs [6], together with κp = ω λ+2µ and κs = ω µ , and thus
it depends analytically on ω. Since h(x, ω) depends analytically on ω ∈ C and N
has dense range, we know that ϕ depends analytically on ω ∈ C. From the above
representation (3.26), it can be seen that the displacement field u(x, ω) depends
analytically on ω.
In the end of this section, we state a uniqueness theorem for the inverse problem
under the assumption that we have the knowledge of the Cauchy data pair.
Theorem 3. Assume that D1 and D2 are elastic bodies with impedance χ1 and χ2 ,
respectively, ∂D1 ∩ ∂D2 = Γ0 . The corresponding displacements u1 (x, ω), u2 (x, ω)
satisfy
u1 (x, ω)|Γ0 = u2 (x, ω)|Γ0 , Tn u1 (x, ω)|Γ0 = Tn u2 (x, ω)|Γ0
for the frequency ω ∈ (a, b). Then D1 = D2 and χ1 = χ2 .
Proof. Assume that there are two different domains D1 , D2 with the boundary co-
efficients χ1 and χ2 generating the same Cauchy data pair denoted by (f , t) on Γ0 ,
(j)
Γm = ∂Dj \Γ0 . Let uj (x, ω), j = 1, 2 be the solutions

 2
∇ · σ(uj ) + ρω uj = 0, in Dj ,
uj = f (x, ω), on Γ0 , (3.27)

 (j)
Tnj uj + iωχj uj = 0, on Γm .
(1) (2)
First, we prove Γm = Γm . It should be noted that f (x, ω) 6= 0. Assume on the
(1) (2)
contradiction that Γm 6= Γm . Without loss of generality we suppose that there is

9
a point in D1 \ D2 . Otherwise we may switch the notations D1 and D2 if necessary.
(1)
Then we can always find a domain D ∗ ⊂ D1 such that ∂D ∗ = Λ1 ∪Λ2 , Λ1 ⊂ Γm , Λ2 ⊂
(2)
Γm , Λ2 ∈ (D1 ∩ D2 ).
On Λ1 , we have
Tn1 u1 (x, ω) = −iωχ1 u1 (x, ω). (3.28)
On Λ2 , we have
Tn2 u2 (x, ω) = −iωχ2 u2 (x, ω). (3.29)
Since u1 (x, ω) = u2 (x, ω) on Γ0 , the uniqueness of the Cauchy problem for elliptic
equations will give u1 (x, ω) = u2 (x, ω) in D ∗ . By Λ2 ∈ (D1 ∩D2 ), we get u1 (x, ω) =
u2 (x, ω), and then
Tn2 u1 (x, ω) = −iωχ2 u1 (x, ω). (3.30)
Let ν = −n2 , χ = −χ2 on Λ2 and ν = n1 , χ = χ1 on Λ1 , combining equations
(3.28) and (3.30), we have

Tν u1 (x, ω) = −iωχu1 (x, ω), on ∂D ∗ . (3.31)

Since D ∗ ⊂ D1 , we know that u1 (x, ω) satisfy the Navier equation in D ∗ with the
boundary condition (3.31). The first Betti formula yields
Z Z
2 2


E(u1 , u1 ) − ρω |u1 (x, ω)| dx = Tν u1 (x, ω) · u1 (x, ω)ds,
D∗ ∂D ∗

where

E(u1 , u1 ) = ∇u1 (x, ω) : C : ∇u1 (x, ω) (3.32)

2 
X ∂up ∂up ∂uq 
= λdivu1 divu1 + µ +
p,q=1
∂xq ∂xq ∂xp
! !
2 2 2
∂u 1 ∂u 2 ∂u 1 ∂u 2
= λ|divu1 |2 + 2µ + +µ + .
∂x1 ∂x2 ∂x2 ∂x1

Combining the boundary condition (3.31), we have

Z Z
2 2


E(u1 , u1 ) − ρω |u1 (x, ω)| dx + iωχ |u1 (x, ω)|2ds = 0.
D∗ ∂D ∗

Taking the real part in the above equation, we have

Z


E(u1 , u1 ) − ρω 2 |u1 (x, ω)|2 dx = 0, for all ω ∈ (a, b). (3.33)
D∗

The analytical dependance of u1 (x, ω) on ω implies that equation (3.33) is valid for
ω ∈ C.

10
 Now choosing ω = 1 + iδ we get ω 2 = 1 − δ 2 + 2iδ, then the equation (3.33) will
be
Z Z
2 2


E(u1 , u1 ) − ρ(1 − δ )|u1 (x, ω)| dx + 2iρδ |u1 (x, ω)|2dx = 0, (3.34)
D∗ D∗

which implies
Z
ρδ |u1(x, ω)|2dx = 0, for all ω = 1 + iδ, δ ∈ R (3.35)
D∗

by taking the imaginary part of the equation (3.34). Again using the analytical
dependance, we get

u1 (x, ω) = 0 in D ∗ (3.36)

for all ω ∈ C. By the uniqueness of the analytic continuation, thus u1 (x, ω) = 0 in

D1 for all ω ∈ C. This yields

f (x, ω) = u1 (x, ω)|Γ0 = 0,

(1) (2)
which contradicts the fact that f (x, ω) 6= 0. Hence Γm = Γm = Γm , i.e. D1 =
D2 = D.
Next we will check χ1 = χ2 . On Γ0 , we have

u1 (x, ω) = u2 (x, ω)

and
Tn u1 (x, ω) = Tn u2 (x, ω),
the uniqueness of the Cauchy problem for elliptic equations will give

u1 (x, ω) = u2 (x, ω) in D. (3.37)

Together with Tn u1 (x, ω) = −iωχ1 u1 (x, ω) and Tn u2 (x, ω) = −iωχ2 u2 (x, ω) on

Γm , we have iωχ1 u1 (x, ω) = iωχ2 u1 (x, ω), i.e. iω(χ1 − χ2 )u1 (x, ω) = 0 on Γm . If
u1 (x, ω) = 0 on any open subset Γ ⊂ Γm , we have Tn u1 (x, ω) = −iωχ1 u1 (x, ω) =
0 on Γ. The uniqueness of the Cauchy problem for elliptic equations will give
u1 (x, ω) ≡ 0 in D. This is a contradiction to the behavior of u1 (x, ω) on Γ0 with
f (x, ω) 6= 0. Thus we have χ1 − χ2 = 0 on Γm . The proof is completed.

4 Recovery method
As discussing in the previous section, we should solve a Cauchy problem first.

11
4.1 The Cauchy problem
In order to solve the Cauchy problem, we should consider the perturbed equa-
tions
N ϕδ = hδ . (4.1)
Here, hδ = (f δ , tδ )⊤ ∈ L2 (Γ0 ) × L2 (Γ0 ) is measured noisy data satisfying

khδ − hk ≤ δ × khk := ǫ,

where δ is the noise level.

The minimum norm solution (see e.g. [22]) of (3.25) is to solve the following
equation
αϕδα + N ∗ N ϕδα = N ∗ hδ , α > 0.
and choosing the regularization parameter by Morozov discrepancy principle [36,
38, 39], i.e. via finding the zero of G(α) := kN ϕ − hδ k2 − ǫ2 . The regularization
parameter will be obtained numerically by Newton’s method as following:

1. Set n = 0, and give an initial regularization parameter α0 > 0.

2. Get ϕδαn from (αn I + N ∗ N )ϕδαn = N ∗ hδ .

d d
3. Get ϕδ
dα αn
from (αn I + N ∗ N ) dα ϕδαn = −ϕδαn .
′
4. Get F (αn ) and F (αn ) by

F (αn ) =k N ϕδαn − hδ k2 −ǫ2

and
2 2
′ d δ d δ
F (αn ) = 2αn N ϕ + 2αn2 ϕ ,
dα αn dα αn
respectively.
F (αn )
5. Set αn+1 = αn − ′
F (αn )
. If |αn+1 − αn | < ε(ε ≪ 1), end. Else, set n = n + 1
return to 2.

If the regularization parameter is fixed by α∗ , we can achieve the regularized solution

ϕδα∗ = Rα hδ by introducing the regularization operator

Rα∗ := (α∗ I + N ∗ N )−1 N ∗ .

The displacement u can be given by uδ (x) = Sϕδα∗ (x), which is the treated data in
the iterative method.

12
4.2 Newton-type method
In the above subsection, we have given the approximation of the displacement
u in B by solving equation (4.1), namely uδ (x). Then we can get the gradient
explicitly. The gradient of the approximated displacement uδ (x) can be calculated
by
Z
δ
∇u (x) = (∇∇⊤ Φ(κp |x − y|)ϕ1 (y) + ∇∇⊥ Φ(κs |x − y|)ϕ2(y))ds(y). (4.2)
∂B

Based on the gradient of the approximate scattered field, we are now ready to
propose the Newton-type iteration scheme for recovering ∂D and impedance function
χ. For a given function g (g can be zero) and the approximated displacement uδ (x),
we can define the mapping operator F from the boundary contour γ and impedance
function χ to the given function g by

F (γ, χ) = Tn uδ (γ) + iωχuδ (γ).

In addition, it should to seek the impedance boundary where the boundary curve γ
and impedance function χ satisfy

F (γ, χ) = g.

Our reconstruction algorithm can be summarized as follows:

First, for a given Cauchy data pair (f , t) on Γ0 , we can get the density function
ϕ by

N ϕ(x) = h, x ∈ Γ0 . (4.3)

Then we can get the treated approximated displacement uδ (x) = Sϕδα∗ (x) and
traction tδ (x) = Tn Sϕδα∗ (x).
Second, for the current approximation to the missing boundary curve γn , n =
0, 1, 2..., and the impedance χn we compute MD (γn ) = Sϕδα∗ (x)|γn and MT (γn ) =
Tn Sϕδα∗ (x)|γn . Denote F = MT + iωχn MD be the combination. We update the
boundary curve γn via the following procedure
(
F (γn , χn ) + F ′ (γn , χn )△γn = g,
(4.4)
γn+1 = γn + △γn .

Third, for the boundary parameterization γn+1 , get the update χn+1 for the
impedance function by
(
F (γn+1, χn ) + iωMD (γn+1 )△χn = g,
(4.5)
χn+1 = χn + △χn .

Once we have gotten △γn and △χn , we have updated γn and χn to γn+1 and χn+1 .

13
 The Fréchet derivatives of uδ (x) and tδ (x) with respect to x can be calculated
by  
δ ∂x1 uδ1 ∂x2 uδ1
∇u (x) = ,
∂x1 uδ2 ∂x2 uδ2
 
δ ∂x1 tδ1 ∂x2 tδ1
∇t (x) = ,
∂x1 tδ2 ∂x2 tδ2
where ϕδα∗ ,1
Z
∂x1 uδ1 = (∂x1 E11 (x, y)ϕδα∗ ,1 (y) + ∂x1 E12 (x, y)ϕδα∗ ,2 (y))ds(y),
Z∂B
∂x2 uδ1 = (∂x2 E11 (x, y)ϕδα∗ ,1 (y) + ∂x2 E12 (x, y)ϕδα∗ ,2 (y))ds(y),
Z∂B
∂x1 uδ2 = (∂x1 E21 (x, y)ϕδα∗ ,1 (y) + ∂x1 E22 (x, y)ϕδα∗ ,2 (y))ds(y),
Z∂B
∂x2 uδ2 = (∂x2 E21 (x, y)ϕδα∗ ,1 (y) + ∂x2 E22 (x, y)ϕδα∗ ,2 (y))ds(y),
∂B

Z
∂x1 tδ1 = ∂x1 T11 (x, y)ϕδα∗ ,1 (y) + ∂x1 T12 (x, y)ϕδα∗,2 (y)ds(y),
Z∂B
∂x2 tδ1 = ∂x2 T11 (x, y)ϕδα∗ ,1 (y) + ∂x2 T12 (x, y)ϕδα∗,2 (y)ds(y),
Z∂B
∂x1 tδ2 = ∂x1 T21 (x, y)ϕδα∗ ,1 (y) + ∂x1 T22 (x, y)ϕδα∗,2 (y)ds(y),
Z∂B
∂x2 tδ2 = ∂x2 T21 (x, y)ϕδα∗ ,1 (y) + ∂x2 T22 (x, y)ϕδα∗,2 (y)ds(y).
∂B

In fact, the forms of the elements ∂xı T1 (x, y) and ∂xı T2 (x, y) are given by

∂x T1 (x, y)
ı   2 
∂ 2 E1 (x, y) ∂ 2 E2 (x, y) ∂ E1 (x, y) ∂ 2 E2 (x, y)
= (λ + 2µ) +λ n1 (x) + µ + n2 (x)
∂x1 ∂xı ∂x2 ∂xı ∂x2 ∂xı ∂x1 ∂xı
   
∂E1 (x, y) ∂E2 (x, y) ∂n1 (x) ∂E1 (x, y) ∂E2 (x, y) ∂n2 (x)
+ (λ + 2µ) +λ +µ + .
∂x1 ∂x2 ∂xı ∂x2 ∂x1 ∂xı

∂xı T2 (x, y)

 2   2 
∂ E1 (x, y) ∂ 2 E2 (x, y) ∂ E1 (x, y) ∂ 2 E2 (x, y)
= µ + n1 (x) + λ + (λ + 2µ) n2 (x)
∂x2 ∂xı ∂x1 ∂xı ∂x1 ∂xı ∂x2 ∂xı
   
∂E1 (x, y) ∂E2 (x, y) ∂n1 (x) ∂E1 (x, y) ∂E2 (x, y) ∂n2 (x)
+ µ + + λ + (λ + 2µ) .
∂x2 ∂x1 ∂xı ∂x1 ∂x2 ∂xı

14
 For this iterative algorithm, we have to create a stopping criterion. For conve-
nience, the stopping of the iterative process we perform with the following relative
error
k∆γn kL2 k∆χn kL2
En = + . (4.6)
kγn−1kL2 kχn−1 kL2
We choose a constant ε > 0. If En > ε, the iteration process is continued. Otherwise,
it is stopped.
For the sake of simplicity, we assume that the boundary is starlike with respect
to the origin, i.e. ∂D can be represented in the parametric form
∂D = {r(ϑ)(cos ϑ, sin ϑ), ϑ ∈ [0, 2π)}
where r : R → R is a positive, twice continuously differentiable and 2π-periodic
function. In this paper we assume that
Γ0 = {r(ϑ)(cos ϑ, sin ϑ), ϑ ∈ [0, π)}
as the known part coincides with the measured Cauchy data and
Γm = {r(ϑ)(cos ϑ, sin ϑ), ϑ ∈ [π, 2π)}
as the missing part of the boundary with the absorbing boundary condition.
To numerically approximate r(ϑ), we assume that r(ϑ) is represented as the
trigonometric polynomials of degree less than or equal to N, namely
N
X
r(ϑ) = a0 + (a cos ϑ + b sin ϑ). (4.7)
=1

In the n-th iterative step, the radial function rn will be updated to the rn+1
by ∆rn . This process will be achieved by updating the Fourier coefficients a =
(n) (n) (n) (n) (n)
(a0 , a1 , ..., aN , b1 , ..., bN )⊤ . If we remember a(n) = (a0 , a1 , ..., aN , b1 , ..., bN )⊤ be
the Fourier coefficients in the n-th step, then the Fourier coefficients in iterative
process step 2 will be updated by a(n+1) = a(n) + △a(n) by the following procedure
(


tδ (Aa(n) x̂) + iωχn uδ (Aa(n) x̂) + ∇ tδ (Aa(n) x̂) + iωχn uδ (Aa(n) x̂) A∆a(n) x̂ = g(Aa(n) x̂),
a(n+1) = a(n) + △a(n) ,
where x̂ = (cos ϑ, sin ϑ)⊤ and A = (1, cos ϑ, ..., cos Nϑ, sin ϑ, ..., sin Nϑ).
Whilst the update impedance function χn+1 in iterative process step 3 will be
as following
(
tδ (Aa(n+1) x̂) + iωχn uδ (Aa(n+1) x̂) + iωuδ (Aa(n+1) x̂)△χn = g(Aa(n+1) x̂),
(4.8)
χn+1 = χn + △χn .
For convenience, the stopping rule will be replaced by the following relative error
k∆a(n) kL2 k∆χn kL2
En = + . (4.9)
ka(n−1) kL2 kχn−1 kL2
We choose a constant ε > 0. If En > ε, the iteration process is continued. Otherwise,
it is stopped.

15
5 Numerical Examples
In this part, we will give some examples to show the effectiveness of the present
method. The Lemé constants λ and µ are given by λ = 1 and µ = 1. The inside of
the elastic obstacle is assumed to be filled with a homogeneous and isotropic elastic
medium with a unit mass, i.e. ρ = 1. The wavenumbers κp and κs are usually
determined
q by the relationship
q between the wavenumbers and the Lemé constants
2
ρω 2
κp = λ+2µ and κs = ρωµ . The initial guess of the boundary is a half circle, and
the initial guess of the impendence function is a constant function. The number of
the coefficients is 17, i.e. N = 8 in equation (4.7).
In the practice, we known all the information of the known part including the
Cauchy data pair (f , t) about Γ0 . And we can easily get the equation Tn u + iωχu =
g on Γ0 . Thus we can deal with iterative procedure on the whole boundary ∂D.
The initial guess of the boundary is a circle. The results about the known part can
give a reference result for the present method.

Example 5.1. In this example, D is a apple-shaped domain, and the boundary ∂D

be
1 + 0.8 cos ϑ + 0.2 sin 2ϑ
z(ϑ) = (cos ϑ, sin ϑ)
1 + 0.7 cos ϑ
with the parameterized surface impedance χ given by χ(ϑ) = sin4 ϑ, ϑ ∈ [π, 2π). The
components up and us of the displacement u inside the elastic obstacle are generated
by
(1) (1)
up (x) = J0 (κp |x − y0 |), us (x) = J0 (κs |x − y0 |), x ∈ D,
where y0 = (1, 0), i.e. u(x) = ∇up (x) + ∇⊥ us (x). We fix the frequency by ω = 3.
The so called virtual boundary ∂B is a circle, and its radius is 4. The initial guess
of the boundary is a circle, and its radius is 0.3. Figure 1 shows the numerical
reconstructions of the missing boundary and the impendence function with different
noise levels δ ∈ {0%, 1%, 5%} for Example 5.1. We can see that the numerical
reconstructions are satisfactory. The reconstruction improves as the noise decreases.

Example 5.2. The components up and us of the displacement u inside the elastic
obstacle are generated by
i (1) i (1)
up (x) = H0 (κp |x − y0 |), us (x) = H0 (κs |x − y0 |), x ∈ D,
4 4
i.e. u(x) = ∇up (x) + ∇⊥ us (x). The so called virtual boundary ∂B is a circle, and
its radius is 4. The initial guess of the boundary is a circle as in Example 5.1.

First, D is a peanut-shaped domain, and the boundary ∂D be

1
z(ϑ) = R0 4 cos2 ϑ + sin2 ϑ 2 (cos ϑ, sin ϑ)

16
 Table 1: The iterative steps and the regularization parameter.

Noise level Iterative steps Regularization parameter

0% 127 1.2019e−13
1% 128 9.7500e−02
5% 147 2.5117e−00

with R0 = 0.5. We fix the frequency by ω = 5, y0 = (4, −9). The parameterized

surface impedance χ is given by
χ(ϑ) = sin4 ϑ, ϑ ∈ [π, 2π).
Second, D is a starfish-shaped domain, and the boundary ∂D be
z(ϑ) = (1 + 0.2 cos 5ϑ) (cos ϑ, sin ϑ).
We fix the frequency by ω = 3, y0 = (4, 9). The parameterized surface impedance
χ is given by a constant function
χ(ϑ) = 1, ϑ ∈ [π, 2π).

Table 2: The iterative steps and the regularization parameter.

Noise level Iterative steps Regularization parameter

0% 88 8.2341e−13
1% 98 2.5200e−02
5% 76 0.3063e−00

Figures 2 and 3 show the numerical reconstructions of the missing boundary and
the impendence function with different noise levels δ ∈ {0%, 1%, 5%} for the first
and second case. We can see that the impedance function χ is sensitive depending
on the noise level either for a function or for a constant function. The numerical
reconstructions of the missing boundary are satisfactory. From table 1 and 2, we
can also see that the number of the iterative steps using the noise data are not much
more than using the exact data.
Example 5.3. In this example, we consider D be a circle, and its radius is 1.2.
Consider the following boundary value problem

 2
∆up + κp up = 0, in D,
∆us + κ2s us = 0, in D, (5.1)


Tn u + iωχu = g, on ∂D.

17
The impedance function χ is given by

χ(ϑ) = sin2 ϑ, ϑ ∈ [0, 2π), (5.2)

and the input boundary function g(x) = (g1 , g2 )⊤ is

g1 (x(ϑ)) = g2 (x(ϑ)) = sin2 ϑ, ϑ ∈ [0, 2π). (5.3)

In order to solve the inverse problem, we first get the Cauchy data f = u|Γ0
and t = Tn u|Γ0 by solving the direct problem (5.1). In this step, we use ∂B =
{x : |x| = 3}. After this we use ∂B = {x : |x| = 7} for solving the inverse
problem. The initial guess of the boundary is a circle, and its radius is 0.6. The
initial guess of the impendence function is a constant function χ0 = 0.5.
We fix the frequency by ω = 2. The iterative steps and the regularization
parameter chosen by Morozov discrepancy principle are given in Table 3. Figure 4
shows the numerical reconstructions of the missing boundary and the impendence
function with different noise levels δ ∈ {0%, 1%, 5%} for Example 5.3. We can see
that the numerical reconstructions are satisfactory. From Table 3, we can also see
that the number of the iterative steps using the noise data are not much more than
using the exact data.

Table 3: The iterative steps and the regularization parameter.

Noise level Iterative steps Regularization parameter

0% 44 9.8186e−17
1% 47 1.0717e−05
5% 50 8.4172e−05

Example 5.4. As in example 5.3, we consider the same impedance boundary value
problem. Different from the example 5.3, D is a peanut-shaped domain, and the
boundary ∂D be

1
z(ϑ) = 0.5 4 cos2 ϑ + sin2 ϑ 2 (cos ϑ, sin ϑ).

The initial guess of the boundary is a circle, and its radius is 0.3. The initial guess
of the impendence function is a constant function χ0 = 1. We fix the frequency
by ω = 1. The iterative steps and the regularization parameter chosen by Morozov
discrepancy principle are given in Table 4. Figure 5 shows the numerical reconstruc-
tions of the missing boundary and the impendence function with different noise levels
δ ∈ {0%, 1%, 5%} for Example 5.4. We can see that the numerical reconstructions
are satisfactory. From Table 4, we can also see that the number of the iterative steps
using the noise data are not much more than using the exact data.

18
 Table 4: The iterative steps and the regularization parameter.

Noise level Iterative steps Regularization parameter

0% 11 1.5751e−15
1% 12 1.1075e−11
5% 13 2.2101e−10

6 Concluding remarks
In this paper, we have studied the inverse elastic problem by a Newton-type
iterative algorithm. This present problem is divided into two parts. The first part
is to solve a Cauchy problem through using an indirect integral equation method
combining a regularization technique. The second part is simultaneously to recover
the elastic impedance and the shape by a Newton-type iterative method. The effec-
tiveness of the method has been shown by solving some examples. In the numerical
examples, we construct the exact solution of the scattered field of the displacement
to check the feasibility and accuracy of the presented method.

Acknowledgements
The research was supported by the Natural Science Foundation of China (No:
11501566), the Fundamental Research Funds for the Central Universities (No. 3122019159)
and Tianjin Education Commission Research Project (No: 2022KJ072).

ORCID ID
Yao Sun: [Link]

Declarations
Conflict of interest: The authors declare that they have no conflict of interest.
Data availability: The datasets generated during the current study are avail-
able on reasonable request.

References
[1] G. Alessandrini, A. Morassi and E. Rosset, Detecting cavities by electrostatic
boundary measurements. Inverse Problems, 18(5):1333, 2002.

19
 [2] C. Alves and R. Kress, On the far-field operator in elastic obstacle scattering.
IMA Journal of Applied Mathematics, 67:1–21, 2002.

[3] H. Ammari, E. Bretin, J. Garnier, H. Kang, H. Lee and A. Wahab, In Mathe-

matical Methods in Elasticity Imaging. Princeton University Press, 04 2015.

[4] H. Ammari, E. Bretin, J. Garnier, H. Kang, H. Lee and A. Wahab, In Theory

of Elasticity. 1986.

[5] S. Andrieux, T. N. Baranger and A. B. Abda, Solving cauchy problems by

minimizing an energy-like functional. Inverse Problems, 22:115 – 133, 2006.

[6] D. Colton and R. Kress, Inverse Acoustic and Electromagnetic Scattering The-
ory, 4th edn (New York: Springer), 2019.

[7] F. Cakoni and R. Kress, Integral equations for inverse problems in corrosion
detection from partial cauchy data. Inverse Problems and Imaging, 1:229–245,
2007.

[8] F. Cakoni, R. Kress and C. Schuft, Simultaneous reconstruction of shape and

impedance in corrosion detection. Methods and Applications of Analysis, 17,
2010.

[9] Y. Chang, Y. Guo, H. Liu, D. Zhang, Recovering source location, polarization,

and shape of obstacle from elastic scattering data, Journal of Computational
Physics, 489:112289, 2023.

[10] P. Deuflhard, H. W. Engl and O. Scherzer, A convergence analysis of iterative

methods for the solution of nonlinear ill-posed problems under affinely invariant
conditions. Inverse Problems, 14(5):1081, 1998.

[11] H. W. Engl, M. Hanke and A. Neubauer. Regularization of inverse problems.

1996.

[12] J. Elschner and M. Yamamoto, Uniqueness in inverse elastic scattering with

finitely many incident waves. Inverse Problems, 26(4):045005, 2010.

[13] D. Fasino and G. Inglese, Discrete methods in the study of an inverse prob-
lem for Laplace’s equation. IMA Journal of Numerical Analysis, 19(1):105–118,
1999.

[14] S.N. Fata, B.B. Guzina, A linear sampling method for near-field inverse prob-
lems in elastodynamics. Inverse Problems, 20:713–736, 2004.

[15] S.N. Fata, B.B. Guzina, Elastic scatterer reconstruction via the adjoint sam-
pling method. SIAM Journal on Applied Mathematics, 67(5):1330–1352,2007.

20
[16] T. Hohage, Iterative Methods in Inverse Obstacle Scattering: Regularization
Theory of Linear and Nonlinear Exponentially Ill-Posed Problems. PhD thesis,
1999

[17] G. Hu, A. Kirsch and M. Sini, Some inverse problems arising from elastic scat-
tering by rigid obstacles. Inverse Problems, 29(1):015009, 2013.

[18] G. Hu, A. Mantile, M. Sini and T. Yin, Direct and inverse time-harmonic
elastic scattering from point-like and extended obstacles. Inverse Problems and
Imaging, 14(6):1025–1056, 2020.

[19] P. Jakubik and R. Potthast, Testing the integrity of some cavity –the cauchy
problem and the range test. Applied Numerical Mathematics, 58(6):899–914,
2008.

[20] A. Karageorghis, D. Lesnic, L. Marin, The method of fundamental solutions for

the identification of a scatterer with impedance boundary condition in interior
inverse acoustic scattering. Eng. Anal. Bound. Elem. 92: 218–224, 2018.

[21] A. Karageorghis, B.T. Johansson, D. Lesnic, The method of fundamental solu-

tions for the identification of a sound-soft obstacle in inverse acoustic scattering.
Appl. Numer. Math. 62(12):1767–1780, 2012.

[22] R. Kress, Linear Integral Equations, Springer-Verlag, Berlin, 1989.

[23] R. Kress, Uniqueness and numerical methods in inverse obstacle scattering.

Journal of Physics: Conference Series, 73(1):012003, 2007.

[24] V.D. Kupradze, Three-dimensional problems of the mathematical theory of elas-

ticity and thermoelasticity. Amsterdam: North-Holland; 1979.

[25] R. Kress, W. Rundell, Inverse scattering for shape and impedance. Inverse
Problems, 17(4): 1075–1085, 2001.

[26] R. Kress and W. Rundell, Nonlinear integral equations and the iterative solution
for an inverse boundary value problem. Inverse Problems, 21(4):1207, 2005.

[27] R. Kress, W. Rundell, Inverse scattering for shape and impedance revisited.
Journal of Integral Equations and Applications, 30(2):293–311, 2018.

[28] P. Li, Y. Wang, Y. Zhao, Near-field imaging of biperiodic surfaces for elastic
waves. Journal of Computational Physics, 324: 1-23, 2016.

[29] X. Liu, J. Sun, Data recovery in inverse scattering: From limited-aperture to

full-aperture. Journal of Computational Physics, 386:350-364, 2019.

21
[30] Y. Liu, W. Jiang J. Yang, G. Li, The application of elastic impedance inversion
in reservoir prediction at the jinan area of tarim oilfield. Applied Geophysics,
pages 201–206, 2007.

[31] J. Liu, X. Liu, J. Sun, Extended sampling method for inverse elastic scat-
tering problems using one incident wave. SIAM Journal on Imaging Sciences,
12(2):874–892, 2019.

[32] F.L. Louër, On the fréchet derivative in elastic obstacle scattering. SIAM Jour-
nal on Applied Mathematics, 72(5):1493–1507, 2012.

[33] F. L. Louër, A domain derivative-based method for solving elastodynamic in-

verse obstacle scattering problems. Inverse Problems, 31(11):115006, 2015.

[34] W. McLean, Strongly Elliptic Systems and Boundary Integral Equations, Cam-
bridge: Cambridge University Press, 2000.

[35] A. Morassi and E. Rosset, Stable determination of cavities in elastic bodies.

Inverse Problems, 20(2):453, 2004.

[36] Y. Sun, F. Ma, D. Zhang, An integral equations method combined minimum

norm solution for 3D elastostatics Cauchy problem. Computer Methods in Ap-
plied Mechanics and Engineering, 271: 231–252, 2014.

[37] Y. Sun. Indirect boundary integral equation method for the cauchy problem of
the laplace equation. Journal of Scientific Computing, 71(2):469–498, 2017.

[38] Y. Sun, L. Marin, An invariant method of fundamental solutions for two-

dimensional isotropic linear elasticity. International Journal of Solids and Struc-
tures, 117: 191–207, 2017.

[39] Y. Sun, F. Ma, X. Zhou, An Invariant Method of Fundamental Solutions for

the Cauchy Problem in Two-Dimensional Isotropic Linear Elasticity. Journal
of Scientific Computing, 64: 197–215, 2015.

[40] Y. Sun, L. He, B. Chen, Application of neural networks to inverse elastic

scattering problems with near-field measurements. Electronic Research Archive,
31(11): 7000–7020, 2023.

[41] Y. Sun, Y. Wang, A highly accurate indirect boundary integral equation solu-
tion for three dimensional elastic scattering problem. Eng. Anal. Bound. Elem.
159: 402–417, 2024.

[42] J. Zhang, Y. Wang, B. Xu, Q. Zhen and Y. Wu, Elastic impedance inversion
based on improved particle swarm optimization. Journal of Physics: Conference
Series, 1069(1):012042, 2018.

22
 1 1 1

0.8 0.8 0.8

0.6 0.6 0.6

0.4 0.4 0.4

0.2 0.2 0.2

0 0 0

-0.2 -0.2 -0.2

-0.4 -0.4 -0.4

-0.6 -0.6 -0.6

-0.8 -0.8 -0.8

-1 -1 -1
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1

(a) Noise free (b) Noise 1% (c) Noise 5%

Impedance function Impedance function Impedance function

2.2 2 2
exact exact exact
reconstructed reconstructed reconstructed
2
1.8 1.8

1.8
1.6 1.6

1.6
1.4 1.4
1.4

1.2 1.2
1.2

1 1
1

0.8 0.8 0.8

3 3.5 4 4.5 5 5.5 6 6.5 3 3.5 4 4.5 5 5.5 6 6.5 3 3.5 4 4.5 5 5.5 6 6.5

(d) Noise free (e) Noise 1% (f) Noise 5%

Figure 1: The numerical reconstructions of the shape and the impendence function
with different noise levels δ ∈ {0%, 1%, 5%} in Example 5.1.

23
 0.5 0.5 0.5

0 0 0

-0.5 -0.5 -0.5

-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1

(a) Noise free (b) Noise 1% (c) Noise 5%

Impedance function Impedance function Impedance function

1.2 1.2 1
exact exact exact
reconstructed reconstructed reconstructed
1 1
0.8

0.8 0.8
0.6

0.6 0.6
0.4
0.4 0.4

0.2
0.2 0.2

0
0 0

-0.2 -0.2 -0.2

3 3.5 4 4.5 5 5.5 6 6.5 3 3.5 4 4.5 5 5.5 6 6.5 3 3.5 4 4.5 5 5.5 6 6.5

(d) Noise free (e) Noise 1% (f) Noise 5%

Figure 2: The numerical reconstructions of the shape and the impendence function
with different noise levels δ ∈ {0%, 1%, 5%} for peanut-shaped domain in Example
5.2.

24
 1 1 1

0.8 0.8 0.8

0.6 0.6 0.6

0.4 0.4 0.4

0.2 0.2 0.2

0 0 0

-0.2 -0.2 -0.2

-0.4 -0.4 -0.4

-0.6 -0.6 -0.6

-0.8 -0.8 -0.8

-1 -1 -1

-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1

(a) Noise free (b) Noise 1% (c) Noise 5%

Impedance function Impedance function Impedance function

1.001 1.06 1.15
exact
1.04 reconstructed
1.0008
1.1
1.02

1.0006 1
1.05
0.98
1.0004
0.96 1
1.0002
0.94
0.95
1 0.92

0.9
0.9
0.9998
0.88 exact
exact
reconstructed
reconstructed
0.9996 0.86 0.85
3 3.5 4 4.5 5 5.5 6 6.5 3 3.5 4 4.5 5 5.5 6 6.5 3 3.5 4 4.5 5 5.5 6 6.5

(d) Noise free (e) Noise 1% (f) Noise 5%

Figure 3: The numerical reconstructions of the shape and the impendence function
with different noise levels δ ∈ {0%, 1%, 5%} for the starfish-shaped domain in
Example 5.2.

25
 1 1 1

0.5 0.5 0.5

0 0 0

-0.5 -0.5 -0.5

-1 -1 -1

-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1

(a) Noise free (b) Noise 1% (c) Noise 5%

Impedance function Impedance function Impedance function

1 1 1.2
exact exact exact
reconstructed 0.9 reconstructed reconstructed
1
0.8
0.8

0.7 0.8
0.6
0.6
0.6
0.4 0.5
0.4
0.4
0.2
0.3 0.2

0.2
0
0
0.1

-0.2 0 -0.2
3 3.5 4 4.5 5 5.5 6 6.5 3 3.5 4 4.5 5 5.5 6 6.5 3 3.5 4 4.5 5 5.5 6 6.5

(d) Noise free (e) Noise 1% (f) Noise 5%

Figure 4: The numerical reconstructions of the shape and the impendence function
with different noise levels δ ∈ {0%, 1%, 5%} in Example 5.3.

26
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
-0.1 -0.1 -0.1
-0.2 -0.2 -0.2
-0.3 -0.3 -0.3
-0.4 -0.4 -0.4
-0.5 -0.5 -0.5

-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

(a) Noise free (b) Noise 1% (c) Noise 5%

Impedance function Impedance function Impedance function

1 1 1
exact exact exact
0.9 reconstructed 0.9 reconstructed 0.9 reconstructed

0.8 0.8 0.8

0.7 0.7 0.7

0.6 0.6 0.6

0.5 0.5 0.5

0.4 0.4 0.4

0.3 0.3 0.3

0.2 0.2 0.2

0.1 0.1 0.1

0 0 0
3 3.5 4 4.5 5 5.5 6 6.5 3 3.5 4 4.5 5 5.5 6 6.5 3 3.5 4 4.5 5 5.5 6 6.5

(d) Noise free (e) Noise 1% (f) Noise 5%

Figure 5: The numerical reconstructions of the shape and the impendence function
with different noise levels δ ∈ {0%, 1%, 5%} in Example 5.4.

27
 Impedance function
1
exact
reconstructed
0.8

0.6

0.4

0.2

-0.2
3 3.5 4 4.5 5 5.5 6 6.5
 1

0.5

-0.5

-1

-1 -0.5 0 0.5 1 1.5

 Impedance function
1

0.8

0.6

0.4

0.2

0
exact
reconstructed

-0.2
3 3.5 4 4.5 5 5.5 6 6.5
0.4

0.2

-0.2

-0.4

-0.6
-1 -0.5 0 0.5 1
 1

0.5

-0.5

-1

-1 -0.5 0 0.5 1
 Impedance function
1.2
exact
reconstructed
1

0.8

0.6

0.4

0.2

-0.2
3 3.5 4 4.5 5 5.5 6 6.5
 1

0.5

-0.5

-1

-1 -0.5 0 0.5 1 1.5

 Impedance function
1
exact
reconstructed
0.8

0.6

0.4

0.2

-0.2
3 3.5 4 4.5 5 5.5 6 6.5
0.4

0.2

-0.2

-0.4

-0.6
-1 -0.5 0 0.5 1
 Impedance function
1.2
exact
reconstructed
1

0.8

0.6

0.4

0.2

0
3 3.5 4 4.5 5 5.5 6 6.5
 1

0.5

-0.5

-1

-1 -0.5 0 0.5 1
 Impedance function
1.2
exact
reconstructed
1

0.8

0.6

0.4

0.2

-0.2
3 3.5 4 4.5 5 5.5 6 6.5
 1

0.5

-0.5

-1

-1 -0.5 0 0.5 1 1.5

 Impedance function
1
exact
reconstructed
0.8

0.6

0.4

0.2

-0.2
3 3.5 4 4.5 5 5.5 6 6.5
0.4

0.2

-0.2

-0.4

-0.6
-1 -0.5 0 0.5 1
 Impedance function
1.2
exact
reconstructed
1

0.8

0.6

0.4

0.2

-0.2
3 3.5 4 4.5 5 5.5 6 6.5
 1

0.5

-0.5

-1

-1 -0.5 0 0.5 1
 Impedance function
1.3
exact
1.25 reconstructed

1.2

1.15

1.1

1.05

0.95

0.9

0.85
3 3.5 4 4.5 5 5.5 6 6.5
 1

0.8

0.6

0.4

0.2

-0.2

-0.4

-0.6

-0.8

-1

-1 -0.5 0 0.5 1
 Impedance function
1

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1 exact
reconstructed
0
3 3.5 4 4.5 5 5.5 6 6.5
0.4

0.2

-0.2

-0.4

-0.6
-1 -0.5 0 0.5 1
 Impedance function
1.2
exact
reconstructed
1

0.8

0.6

0.4

0.2

0
3 3.5 4 4.5 5 5.5 6 6.5