Estimate of an inclusion in a body with discontinuous conductivity ∗
Tu Nguyen
†Jenn-Nan Wang
‡Abstract
We study the problem of estimating the size of an inclusion embed- ded inside a two dimensional body with discontinuous conductivity by one voltage-current measurement. This problem is practically impor- tant because the conductivity of a human body is discontinuous. The proofs rely on quantitative uniqueness estimates for the conductivity equation with discontinuous coefficients.
1 Introduction
An important clinical problem is to estimate the size of a cancerous tumor inside an organ by noninvasive methods. In this paper, we study this problem by the method of electrical impedance tomography (EIT) with one measure- ment. Previous works on this problem assumed that the conductivity of the studied body is Lipschitz continuous (see, for example, [5, 6]). However, this is not guaranteed in reality, for example, the conductivities of heart, liver, intestines are 0.70 (S/m), 0.10 (S/m), 0.03 (S/m), respectively. In this paper, we show that in the two dimensional case, the assumption on the regularity of the conductivity can be weaken.
We briefly outline the framework, following [6]. Let Ω ⊂ R2 be an open bounded domain with Lipschitz boundary. Assume that the background
∗This work is dedicated to Professor Neil Trudinger for his 70th birthday.
†Department of Mathematics, National Taiwan University, Taipei 106, Taiwan. Email:
anhtu@uw.edu
‡Institute of Applied Mathematical Sciences, National Taiwan University, NCTS (Taipei), Taipei 106, Taiwan. Email: jnwang@ntu.edu.tw
conductivity σ(x) is elliptic, i.e. for some λ > 0,
λ−1|y|2 ≤ hσ(x)y, yi ≤ λ |y|2, ∀y ∈ R2, a.e. x ∈ Ω. (1.1) Let D be a subdomain of Ω and ˜σ be a matrix-valued function on D with bounded measurable coefficients, representing the conductivity of the inclu- sion. Let v be the electric potential with boundary value φ, i.e.
(div((σ(x)χΩ\ ¯D+ ˜σ(x)χD)∇v) = 0 in Ω,
v = φ on ∂Ω. (1.2)
The energy required to maintain voltage potential φ on ∂Ω is W :=
ˆ
∂Ω
φ hσ∇v, νi ds.
Let u be the electric potential with the same boundary value when there is no inclusion, i.e.
(div(σ(x)∇u) = 0 in Ω,
u = φ on ∂Ω. (1.3)
Similarly, we define the energy W0 :=
ˆ
∂Ω
φ hσ∇u, νi ds.
In [6], it is shown that if σ is Lipschitz continuous and for some ζ, η > 0 either
(1 + η)σ ≤ ˜σ ≤ ζσ a.e. in Ω, (1.4) or
ζσ ≤ ˜σ ≤ (1 − η)σ a.e. in Ω, (1.5) then the size of D can be estimated using the normalized power gap
W −W0
W0
. More precisely, the following estimate holds
K1
W − W0 W0
≤ |D| ≤ K2
W − W0 W0
1 p
, (1.6)
where p > 1, K1and K2are constants depending on a priori data. If moreover D satisfies the fatness condition (4.3), then a better estimate holds
K1
W − W0 W0
≤ |D| ≤ K2
W − W0 W0
. (1.7)
We will show that in two dimension, the method of [6] works even when σ is only piecewise H¨older continuous. Essentially, this is because in two dimension, the three-ball and doubling inequalities for solutions of (1.3) hold for bounded σ; and a gradient estimate needed in proving the propagation of smallness for ∇u was proved in [15] for piecewise H¨older σ (in any dimension).
We would like to mention that size estimates have also been derived for other systems, for example, [2] for the isotropic elasticity, [16, 17, 18] for the isotropic/anisotropic thin plate, [11, 10] for the shallow shell.
The paper is organized as follows. In next section, we define several notations and list several assumptions used in the paper. In Section 3, we prove some quantitative estimates for solutions of (1.3). In Section 4, we prove (1.6) and (1.7).
2 Notations and assumptions
Definition 2.1. Let Ω be an open bounded domain of R2. Given 0 < α < 1, we say that ∂Ω is of class C1,α with parameters r0, M0, if for any P ∈ ∂Ω, there exists a rigid coordinates transform under which P = 0 and
Ω ∩ Br0(0) = {z = (z1, z2) ∈ Br0(0) : z2 > ψ(z1)}, where ψ(z1) ∈ C1,α(−r0, r0) satisfying ψ(0) = 0 and ∇ψ(0) = 0 and
kψkC1,α(−r0,r0) ≤ M0. Recall that
kψkC1,α(−r0,r0) = kψkL∞(−r0,r0)+k∇ψkL∞(−r0,r0)+ sup
x,y∈(−r0,r0)
|∇ψ(x) − ∇ψ(y)|
|x − y|α . We now state the assumptions used in the paper.
Assumptions
• Ω ⊂ R2 is an open bounded C1,α domain with parameters r0 and M0.
• There exist disjoint C1,α domains Ωj ⊂ Ω, 1 ≤ j ≤ m such that Ω =
∪mj=1Ωj and for some µ > 0, we have σj(x) := σ(x)χΩj ∈ C0,µ(Ωj), 1 ≤ j ≤ m. For α0 = min{µ,3(α+1)α }, let M1 = supjkσjkC0,α0(Ωj).
• For any x ∈ Ω, there exist r > 0 and an appropriate rotation of co- ordinates such that the set (∪mj=1∂Ωj) ∩ Br(x) consists of the graphs of `(x, r) functions of class C1,α, whose C1,α norms are bounded by L(x, r). We assume that
L := sup
x∈Ω r>0inf
L(x, r) + `(x, r) + 1 r
< ∞.
• d = dist(D, ∂Ω) > 0.
• For some Γ ⊂ ∂Ω of positive measure, φ|Γ = 0.
Remark 2.2. The boundaries of subdomains may touch each other. The inclusion D is only required to stay away from the boundary ∂Ω, it may intersect ∂Ωj’s (see Figure 2.1).
Figure 2.1: Ωj’s may touch each other and D is allowed to intersect the interfaces.
We also define for h > 0,
Ωh = {x ∈ Ω | dist(x, ∂Ω) > h}.
3 Quantitative uniqueness estimates
In this section, we prove quantitative uniqueness estimates for solutions of (1.3) that will be used in the next section. We first recall the three ball inequality of [4].
Lemma 3.1. [4, Theorem 3.11] For all 0 < r1 < r2 < r3, there exist con- stants C > 0 and 0 < τ < 1 depending only on λ, rr1
3, and rr2
3 such that for any solution of (1.3) in Br3(x), we have
kukL2(Br2(x))≤ CkukτL2(Br1(x))kuk1−τL2(Br3(x)). (3.1) Using this three-ball inequality, we can prove
Lemma 3.2. (propagation of smallness) Assume that the assumptions in Section 2 holds. Let u ∈ H1(Ω) be the solution of (1.3). For any ρ > 0 and every x ∈ Ω4ρ, we have
ˆ
Bρ(x)
|∇u|2 ≥ C ˆ
Ω
|∇u|2, (3.2)
where C depends on Ω, Γ, λ, α, µ, r0, M0, M1, L, ρ, and kφkH2(∂Ω)/kφkH1/2(∂Ω). Proof. We follow the arguments presented in [6, Lemma 2.2]. We first ob- serve that it suffices to consider the case ρ is small, so we can assume that Ωρis connected. Using Caccioppoli and Poincar´e inequalities, we can deduce from Lemma 3.1 that
k∇ukL2(B3r(x)) ≤ Ck∇ukτL2(Br(x))k∇uk1−τL2(B4r(x)). (3.3) Given x, y ∈ Ω4ρ, let γ be a curve in Ω4ρjoining x and y. We define a sequence xk’s as follows: Let x1 = x. For k > 1, let xk = γ(tk) where tk = max{t :
|γ(t) − xk−1| = 2ρ} if |xk− y| > 2ρ; otherwise let xk = y, N = k and stop the process. Note that since the balls Bρ(xk) are disjoint, N ≤ N0 = πρ|Ω|2. Using (3.3), noting that Bρ(xk+1) ⊂ B3ρ(xk) because |xk+1− xk| ≤ 2ρ, we can deduce that
k∇ukL2(Bρ(xk+1)) k∇ukL2(Ω) ≤ C
k∇ukL2(Bρ(xk)) k∇ukL2(Ω)
τ
. By induction, we obtain
k∇ukL2(Bρ(y))
k∇ukL2(Ω) ≤ C1/(1−τ )
k∇ukL2(Bρ(x)) k∇ukL2(Ω)
τN
. (3.4)
Since we can cover Ω5ρ by no more than 2ρ|Ω|2 balls of radius ρ, we obtain k∇ukL2(Ω5ρ)
k∇ukL2(Ω) ≤ C
k∇ukL2(Bρ(x)) k∇ukL2(Ω)
τN0
, (3.5)
where C depends on λ, |Ω|, and ρ.
By Corollary 1.3 in [15], k∇uk2L∞(Ω) ≤ Ckφk2C1,1/2(Ω), hence by the em- bedding H2(∂Ω) ,→ C1,1/2(∂Ω), we get
ˆ
Ω\Ω5ρ
|∇u|2 ≤ C|Ω \ Ω5ρ|kφk2C1,α0(∂Ω) ≤ Cρkφk2H2(∂Ω). (3.6)
Here we have used |Ω\Ω5ρ| . ρ since ∂Ω is Lipschitz.
Using the Poincar´e inequality of [9, Theorem 6.1-8 (b)], recalling that ϕ|Γ= 0, we have
kφk2H1/2(∂Ω) ≤ Ckuk2H1(Ω) ≤ Ck∇uk2L2(Ω). (3.7) Combining this and (3.6), we see that if ρ is small enough depending on Ω, Γ, λ, r0, M0, M1, α, µ, L, and kφkH2(∂Ω)/kφkH1/2(∂Ω),
k∇uk2L2(Ω5ρ)
k∇uk2L2(Ω)
≥ 1 2. The lemma follows from this and (3.5).
Next, we derive a local doubling inequality for solutions of (1.3).
Lemma 3.3. For any ρ > 0, there exist δ = δ(ρ, λ) ∈ (0, ρ) and a constant C = C(ρ, λ) such that for all x ∈ Ωρ and r ∈ (0, δ) and any non-trivial solution u of (1.3), we have
kukL2(B4r(x))
kukL2(Br(x))
≤ CkukL∞(Bρ(x))
kukL∞(Bδ(x))
. (3.8)
Proof. The proof, using the theory of quasiconformal maps, follows the ideas of the proof of Proposition 2 in [1]. We first note that it suffices to consider the case u is real-valued. Let v ∈ Hloc1 (Ω) be a σ-harmonic conjugate of u, i.e.
∇v = Jσ∇u where J = 0 −1
1 0
. Then f = u + iv satisfies
∂zf = ν1∂zf + ν2∂zf , where
ν1 = bc − ad + 1 + i(b − c)
(a + 1)(d + 1) − bc , ν2 = d − a + i(b + c) (a + 1)(d + 1) − bc.
It is easy to check that |ν1| + |ν2| ≤ κ < 1. Here κ is a constant depending only on λ.
By Bers-Nirenberg representation theorem (see [8], p. 259), there exists a quasiconformal map χ : Ω → χ(Ω) and an analytic function h : χ(Ω) → C
such that f = h ◦ χ. Furthermore, there exist K, α > 1 depending on κ such that
K−1|x − y|α ≤ |χ(x) − χ(y)| ≤ K |x − y|α1 , ∀x, y ∈ Ω.
Let δ = (10K2)−αρα2 and R = (10K)−1ρα, then we have χ(Bδ(x)) ⊂ BR(χ(x)) and B10R(χ(x)) ⊂ Ω.
By Theorem 3.6.2 in [7], there exists an increasing function γ depending only on κ with γ(0) = 0 such that if x1, x2, x3 ∈ B(x, δ) then
|χ(x1) − χ(x2)|
|χ(x1) − χ(x3)| ≤ γ |x1− x2|
|x1− x3|
.
Let c = γ(8) > 1 then for any x ∈ Ωρ and r ∈ (0, δ), there exists s ∈ (0, R/c) such that if y = χ(x) then
Bs(y) ⊂ χ(Br/2(x)) and χ(B4r(x)) ⊂ Bcs(y). (3.9) Since ˜u = <h is harmonic on χ(Ω), by Hadamard’s three-circle theorem, there exists an absolute constant C such that
k˜ukL∞(Bcs(y))
k˜ukL∞(Bs(y))
≤ Ck˜ukL∞(B4R(y))
k˜ukL∞(B3R(y))
.
By Theorem 3.1.2 in [7], |E| = 0 iff |χ(E)| = 0, hence (3.9) implies kukL∞(B4r(x))
kukL∞(Br/2(x))
≤ CkukL∞(Bρ2(x))
kukL∞(Bρ1(x))
.
Here
ρ1 = (3R/K)α = 3αδ > δ, ρ2 = (4KR)α1 = (2/5)1/αρ < ρ.
Using well-known estimates for elliptic equations with measurable coeffi- cients, we have
kukL2(B4r(x))
kukL2(Br(x))
≤ C kukL∞(B4r(x))
kukL∞(Br/2(x))
≤ CkukL∞(Bρ(x))
kukL∞(Bδ(x))
.
4 Size estimates
To begin, we recall the following energy inequalities proved in [6].
Lemma 4.1. [6, Lemma 2.1] Assume that σ satisfies the ellipticity condition (1.1). If either (1.4) or (1.5) holds, then
C1 ˆ
D
|∇u|2dx ≤ |W0− W | ≤ C2 ˆ
D
|∇u|2dx, (4.1) where C1, C2 are constants depending only on λ, η, and ζ.
We now state and prove the main theorem.
Theorem 4.2. i/ Suppose that the assumptions in Section 2 hold. Then there exist constants K1, K2 > 0 and p > 1 depending only on Ω, Γ, λ, α, µ, r0, M0, M1, L, d, η, ζ, ρ, and kφkH2(∂Ω)/kφkH1/2(∂Ω) such that
K1
W0 − W W0
≤ |D| ≤ K2
W0− W W0
1 p
. (4.2)
ii/ If moreover, there exists h > 0 such that
|Dh| ≥ 1
2|D| (fatness condition). (4.3) then
K1
W0− W W0
≤ |D| ≤ K2
W0− W W0
, (4.4)
where K1 and K2 depend on the various constants as in i/ and also on h.
Proof. The proof closely follows the arguments of [6].
We first establish the lower bound. Let c = 1
|Ωd/4|
´
Ωd/4u. By the gradient estimate of [15, Theorem 1.1], the interior estimate of [14, Theorem 8.17] and the Poincar´e inequality for the domain Ωd/4, we have
k∇ukL∞(Ωd/2) ≤ Cku − ckL∞(Ωd/3) ≤ Cku − ckL2(Ωd/4)≤ Ck∇ukL2(Ω). From this, the trivial estimate k∇uk2L2(D) ≤ C|D|k∇uk2L∞(Ωd/2) and Lemma 4.1, the lower bound follows.
Next, we establish the upper bounds.
i/ We will first establish that |∇u|2 is an Ap-weight, following the proof of Theorem 1.1 in [13]. Let ρ = d/5 and δ be the constant appears in Lemma 3.3. By Cacciopolli inequality and (3.2), for any x ∈ Ω5ρ we have
ku − ckL∞(Bδ(x))≥ C ku − ckL2(Bδ(x)) ≥ C k∇ukL2(Bδ/2(x)) ≥ C k∇ukL2(Ω). (Note that C depends also on δ). By interior estimate, we have
ku − ckL∞(Bρ(x)) ≤ 2 kukL∞(Bρ(x))≤ C kϕkH1/2(∂Ω).
For r ∈ (0, δ), applying the doubling inequality of 3.3 to u − c where c =
1
|Br|
´
Br(x)u, we get ku − ckL2(B2r(x))
ku − ckL2(Br(x))
≤ Cku − ckL∞(Bρ(x))
ku − ckL∞(Bδ(x))
≤ C kϕkH1/2(∂Ω)
k∇ukL2(Ω)
≤ C.
At the last inequality we have used (3.8). We note that the constant C depends on various constants, including kϕkH2(∂Ω)/ kϕkH1/2(∂Ω) but is inde- pendent of r.
This and the Caccioppoli inequality give
r−1k∇ukL2(Br(x))≤ C ku − ckL2(B2r(x))≤ C ku − ckL2(Br(x)). Combining this with the Poincar´e inequality
1
|Br(x)|
ˆ
Br(x)
|u − c|2
12
≤ Cr−1
1
|Br(x)|
ˆ
Br(x)
|∇u|32
23 , we get
1
|Br(x)|
ˆ
Br(x)
|∇u|2
12
≤ C
1
|Br(x)|
ˆ
Br(x)
|∇u|32
23 .
This reverse H¨older inequality shows that |∇u|2 is an Ap-weight for some p > 1 (see [12, Chapter 7]).
We cover D with internally nonoverlapping closed squares Qk, 1 ≤ k ≤ I, with side length 2ρ. Since |∇u|2 is and Ap-weight, by [12, (7.2)], we have
|D ∩ Qk|
|Qk| ≤ C ´
D∩Qk|∇u|2
´
Qk|∇u|2
!1/p
.
Summing over k and using (3.2), we get
|D| ≤ C
´
D|∇u|2 mink´
Qk|∇u|2
!1/p
≤ C
´
D|∇u|2
´
Ω|∇u|2
!1/p
.
The upper bound of |D| now follows from (4.1).
ii/. Let ρ = 14min{d, h} and cover Dh with internally nonoverlapping closed squares {Qk}Jk=1 of side length 2ρ. It is clear that Qk ⊂ D, hence
ˆ
D
|∇u|2dx ≥ ˆ
∪Jk=1Qk
|∇u|2dx ≥ |Dh| ρ2 min
k
ˆ
Qk
|∇u|2dx.
≥ C|D|
ρ2 ˆ
Ω
|∇u|2dx.
Here we have used Lemma 3.2 and the fatness condition at the last inequality.
The upper bound of |D| follows from this and Lemma 4.1.
Acknowledgements
The authors were supported in part by the National Science Council of Tai- wan grant 99-2115-M-002-006-MY3. We are grateful to Sergio Vessella for carefully reading this paper and for many insightful comments.
References
[1] G. Alessandrini, L. Escauriaza, Null-Controllability of One-Dimensional Parabolic Equations, ESAIM Contr. Op. Ca. Va. 14 (2008) 284-293.
[2] G Alessandrini, A Morassi, and E Rosset, Detecting an inclusion in an elastic body by boundary measurements, SIAM J. Math. Anal., 33 (2002), 1247-1268.
[3] G Alessandrini, A Morassi, E Rosset, and S Vessella, On doubling in- equalities for elliptic systems, J. Math. Anal. Appl., 357 (2009), 349-355.
[4] G Alessandrini, L Rondi, E Rosset, and S Vessella, The stability for the Cauchy problem for elliptic equations, Inverse Problems 25 (2009) 123004 (47pp).
[5] G Alessandirni and E Rosset, The inverse conductivity problem with one measurement: bounds on the size of the unknown object, SIAM J. Appl.
Math., 58 (1998), 1060-1071.
[6] G Alessandrini, E Rosset, and J K Seo, Optimal size estimate for the inverse conductivity problem with one measurement, Proc. AMS, 128 (1999), 53-64.
[7] K. Astala, T. Iwaniec and G. Martin, Elliptic Partial Differential Equa- tions and Quasiconformal Mappings in the Plane, Princeton University Press, 2008.
[8] L. Bers, F. John and M. Schechter, Partial Differential Equations, In- terscience, New York, 1964.
[9] P G Ciarlet, Mathematical Elasticity. Volume I: Three-Dimensional Elasticity, Elsevier Science Publishers, B.V., 1988.
[10] M Di Cristo, C L Lin, S Vessella, and J N Wang, Size estimates of the inverse inclusion problem for the shallow shell equation, SIAM J Math Anal, in press.
[11] M Di Cristo, C L Lin, and J N Wang, Quantitative uniqueness for the shallow shell system and their application to an inverse problems, to appear in Ann. Sc. Norm. Super. Pisa Cl. Sci.
[12] J Duoandikoetxea, Fourier analysis, GMT 29, Springer 2000
[13] N Garofalo and F H Lin, Monotonicity properties of variational integrals, Ap weights and unique continuation, Indiana Univ. Math. J., 35 (1986), 245-268.
[14] D Gilbarg and N Trudinger, Elliptic Partial Differential Equations of Second Order, 2nd Ed., Springer 1998.
[15] Y Y Li and M Vogelius, Gradient estimates for solutions to divergence form elliptic equations with discontinuous coefficients, Arch. Rational Mech. Anal., 153 (2000), 91-151.
[16] A Morassi, E Rosset, and S Vessella, Size estimates for inclusions in an elastic plate by boundary measurements, Indiana Univ. Math. J. 56 (2007), 2325-2384.
[17] A Morassi, E Rosset, and S Vessella, Detecting general inclusions in elastic plates, Inverse Problems, 25 (2009).
[18] A Morassi, E Rosset, and S Vessella, Estimating area of inclusions in anisotropic plates from boundary data, Dis. Cont. Dyn. Sys., Series S, 6 (2013), 501-515.