PLEASE SCROLL DOWN FOR ARTICLE
On: 23 June 2009Access details: Access Details: [subscription number 908165586] Publisher Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Communications in Partial Differential Equations
Publication details, including instructions for authors and subscription information:
http://www.informaworld.com/smpp/title~content=t713597240
Uniqueness and Stability of Determining the Residual Stress by One
Measurement
Victor Isakov a; Jenn-Nan Wang b; Masahiro Yamamoto c
a Department of Mathematics and Statistics, Wichita State University, Wichita, Kansas, USA b Department of
Mathematics, National Taiwan University, Taipei, Taiwan c Department of Mathematical Sciences, University
of Tokyo, Meguro, Tokyo, Japan Online Publication Date: 01 May 2007
To cite this Article Isakov, Victor, Wang, Jenn-Nan and Yamamoto, Masahiro(2007)'Uniqueness and Stability of Determining the Residual Stress by One Measurement',Communications in Partial Differential Equations,32:5,833 — 848
To link to this Article: DOI: 10.1080/03605300600718453
URL: http://dx.doi.org/10.1080/03605300600718453
Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf This article may be used for research, teaching and private study purposes. Any substantial or systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material.
ISSN 0360-5302 print/1532-4133 online DOI: 10.1080/03605300600718453
Uniqueness and Stability of Determining
the Residual Stress by One Measurement
VICTOR ISAKOV
1, JENN-NAN WANG
2,
AND MASAHIRO YAMAMOTO
31Department of Mathematics and Statistics, Wichita State University,
Wichita, Kansas, USA
2Department of Mathematics, National Taiwan University, Taipei, Taiwan
3Department of Mathematical Sciences, University of Tokyo,
Meguro, Tokyo, Japan
In this paper we prove a Hölder and Lipschitz stability estimates of determining the residual stress by a single pair of observations from a part of the lateral boundary or from the whole boundary. These estimates imply first uniqueness results for determination of residual stress from few boundary measurements.
Keywords Continuation of solutions; Elasticity theory; Inverse problems. Mathematics Subject Classification 35R30; 74B10; 35B60.
1. Introduction
We consider an elasticity system with residual stress. Let be an open bounded
domain in 3 with smooth boundary . The residual stress is modelled by a
symmetric second-rank tensor Rx= rjkx3jk=1 ∈ C7which is divergence free
· R = 0 in (1.1)
and satisfies the boundary condition
R= 0 on (1.2)
Received November 16, 2005; Accepted March 7, 2006
Address correspondence to Victor Isakov, Department of Mathematics and Statistics, Wichita State University, Wichita, Kansas 67206-0033, USA; E-mail: victor.isakov@ wichita.edu
833
where · R is a vector-valued function with components given by · Rj= 3 k=1 krjk 1≤ j ≤ 3
In this paper x∈ R3 and =
1 2 3 is the unit outer normal vector to .
Here and below, differential operators and without subscript are with respect to x variables. Let ux t= u1 u2 u3 Q→ 3 be the displacement vector in
Q = × −T T. We assume that ux t solves the initial boundary value problem:
ARu = 2tu− + · u − u − · uR = 0 in Q (1.3)
u= u0 tu= u1 on × 0 (1.4)
u= g0 on × −T T (1.5)
where is density and and are Lamé constants satisfying
0 < 0 < 0 < + (1.6)
The system (1.3) can be written as
2tu− · u = 0
where u= tr I + 2 + R + uR is the stress tensor and = u + u/2
is the strain tensor. Note that the term · R does not appear in (1.3) due to (1.1).
Also, by the same condition, we can see that · uRi=
3
jk=1
rjkjkui 1≤ i ≤ 3
Since we are only concerned with the residual stress and we are motivated by applications to the material science we suppose that density and Lamé coefficients
and are constants. To make sure that the problem (1.3) with (1.4), (1.5) is
well-posed, we assume that
RC1< 0 (1.7)
for some small constant 0>0. The assumption (1.7) is also physically motivated
Man (1998). It is not hard to see that if 0 is sufficiently small, then the boundary
value problem (1.3), (1.4), (1.5) is hyperbolic, and hence for any initial data u0u1∈
H1× L2 and lateral Dirichlet data g0∈ C1−T T H1, u0= g0 on
× 0, there exists a unique solution u· R u0u1g0∈ C−T T H1 to
(1.3)–(1.5).
In this paper we are interested in the following inverse problem:
Determine the residual stress R by a single pair of Cauchy data u u on × −T T, where u = u· R u0u1g0and ⊂ .
We will address uniqueness and stability issues. The focus is on the stability since the uniqueness follows immediately from it. Our method is based on Carleman estimates techniques initiated by Bukhgeim and Klibanov (1981). For works on
Carleman estimates and related inverse problems for scalar equations, we refer to books Bukhgeim (2000) and Klibanov and Timonov (2004) for further details and references. Here we only want to mention some related results for the dynamical Lamé system and the residual stress system (1.3). For the Lamé system, the first step has been made by Isakov (1986) where he proved the Carleman estimate and established the uniqueness for the inverse source problem. It should be noted that Isakov (1986) transformed the principal part of the system into a composition of two scalar wave operators. It is well-known that the Lamé system is principally
diagonalized as a system of equations for u and div u. Based on this fact, L2
-Carleman estimates were derived in Eller et al. (2002) and Ikehata et al. (1998) for the Lamé system and applications of to the Cauchy problem and the inverse problem were given. Recently, Imanuvilov et al. (2003) obtained a Carleman estimate for the Lamé system by considering a new principally diagonalized system for u div u curl u. In Imanuvilov et al. (2003), they used this estimate to study the problem of identifying the density and Lamé coefficients by two sets of data measured in a boundary layer and a Hölder-type stability estimate. The continuation of this work is in Imanuvilov and Yamamoto (2005).
For the dynamical residual stress model (1.3), an L2-Carleman estimate has been
proved when the residual stress is small (Isakov et al., 2003). The system with the residual stress is no longer isotropic. In other words, this system is strongly coupled, and it is not possible to decouple the leading part without increasing the order of the system. There are almost no results on Carleman estimates and inverse problems for anisotropic systems which are very important in applications. In Isakov et al. (2003), we used the standard substitution u div u curl u and reduced (1.3) to a new system where the leading part is a special lower triangular matrix differential operator with the wave operators in the diagonal. The key point is that the coupled terms in the leading part contain only the second derivatives of u with respect to x variables and they can be absorbed by div u curl u when the residual stress is small. Using similar Carleman estimates, Lin and Wang (2003) studied the problem of uniquely determining the density function by a single set of boundary data. The unique continuation property for the stationary case of (1.3) was proved in Nakamura and Wang (2003).
In this work we study the problem of recovering the (small) residual stress in(1.3)–(1.5) by single set of Cauchy data. We will derive a Hölder stability estimate in convex hull of the observation surface and a Lipschitz stability estimate for R in
when = and observation time T is large. There are other results concerning
the determination of the residual stress by infinitely many boundary measurements, i.e. by the Dirichlet-to-Neumann map, we refer to Hansen and Uhlmann (2003), Rachele (2003), and Robertson (1977, 1988).
We are now ready to state the main results of the paper. Denote d= inf x and
D= sup x over x ∈ . We assume that
0 < d (1.8)
Let0 Ebe the class of residual stresses defined by
0 E= RC6< E Ris symmetric and satisfies (1.1), (1.2), and (1.7)
To study the inverse problem, we need not only the well-posedness of (1.3)–(1.5) but also some extra regularity of the solution u. To achieve the latter property,
the initial data u0u1 and the Dirichlet data g0 are required to satisfy some
smoothness and compatibility conditions). More precisely we will assume that u0∈ H9u
1∈ H8 and g0∈ C8−T T H1∩ C5−T T H4 and
they satisfy standard compatibility assumptions of order 8 at × 0. By using
energy estimates (Duvaut and Lions, 1976) and Sobolev embedding theorems as in Imanuvilov et al. (2003) one can show that
x
tuC0≤ C (1.9)
for ≤ 2 and 0 ≤ ≤ 5.
By examining the equation (1.3), we can see that the residual stress tensor appears in the equation without first derivatives because of (1.1). It turns out that a single set of Cauchy data is sufficient to recover the residual stress. To guarantee the
uniqueness, we impose some non-degeneracy condition on the initial data u0u1.
More precisely, we assume that
det M= det 21u0 212u0 213u0 22u0 223u0 23u0 2 1u1 212u1 213u1 22u1 223u1 23u1 > E−1 on (1.10)
Note that Mx is a 6× 6 matrix-valued function. For example, one can check that
u0x= x2
1 x22 x32and u1x= x2x3 x1x3 x1x2satisfy (1.10).
We will use the following notation: C are generic constants depending only on T 0 Eu0u1g0, any other dependence is indicated, · kQis the
norm in the Sobolev space HkQ. Q= Q ∩ < x2− 2t2− d2
1 and =
∩ < x2− d21. Here d1 is some positive constant. u 1 and u 2 denote
solutions of the initial boundary value problems (1.3), (1.4) associated with R 1 and R 2. Finally, we introduce the norm of the differences of the data
F = 4 =2 tu 2− u 15 2× −T T + tu 2− u 132× −T T
Due to (1.6) we can choose positive so that 2< 4< d2 T2 (1.11)
Theorem 1.1. Assume that the domain satisfies (1.8), satisfies (1.11), and for
some d1
x2− d2
1<0 when x∈ \ and D
2− 2T2− d2
1<0 (1.12)
Let the initial data u0u1 satisfy (1.10).
Then there exist 0 and constants C <1, depending on , such that for R 1 R 2∈ 0 E one has
R 2 − R 10≤ CF (1.13)
The domain is discussed in Isakov (2006, Sec. 3.4).
If is the whole lateral boundary and T is sufficiently large, then a much stronger (and in a certain sense best possible) Lipschitz stability estimate holds.
Theorem 1.2. Let d1= d. Assume that the domain satisfies (1.8),
D2<2d2 (1.14)
and
D2− d2
2 < T
2 (1.15)
Let the initial data u0u1 satisfy (1.10). Let = .
Then there exist an 0 and C such that for R 1 R 2∈ 0 E satisfying the
condition
R 1= R 2 on × −T T (1.16)
one has
R 2 − R 10≤ CF (1.17)
Let us show compatibility of conditions (1.15) and (1.11). From conditions (1.11) and (1.14) we have D2− d2 2 < d2 4
and hence we can find T2 between these two numbers.
As mentioned previously, the proofs of these theorems rely on Carleman estimates. Using the results of Isakov et al. (2003) we will derive needed Carleman estimates in Section 2. Using this estimate we will prove in Section 3 the Hölder stability estimate (1.13). In Section 4, we demonstrate the Lipschitz stability of the Cauchy problem for the residual stress model. This estimate is one of key ingredients to derive the Lipschitz stability estimate for our inverse problem in Section 5.
2. Carleman Estimate
In this section we will describe Carleman estimates needed to solve our inverse
problem. Their proofs can be found in Isakov et al. (2003). Let x t= x2−
2t2− d2
1 and x t= exp
2x t, where is choosen in (1.11) and < C is a
large constant to be fixed later.
Theorem 2.1. There are constants 0and C such that for R satisfying (1.7)
Q xtu2+ xtv2+ xtw2+ 3u2+ 3v2+ 3w2e2 ≤ C Q ARu2+ ARu2e2 (2.1)
for allu∈ H3
0Q and
Q
2u2+ div u2+ curl u2+ −1ue2≤ C
QAR
u2e2 (2.2)
for allu∈ H2 0Q.
Carleman estimates of Theorem 2.1 is our basic tool for treating the inverse problem. Lemma 2.2. For u∈ H2 0 satisfying u= f0+ 3 j=1jfj with f0 fj 1≤ j ≤ 3 belonging to L2, we have 1 u 2e2dx≤ C f2 0 2 + fj2 e2dx
Proof. For the weight function with large , we can use Theorem 3.1 of Fabre and Lebeaux (1996) to get
u2e2dx≤ C f2 0 2 + fj2 e2dx (2.3)
Now we will bound u. Observe that ue= ue− ueand hence
1 u 2e2dx≤ 2 ue 22dx+ C u 2e2dx (2.4) We have
ue= ue+ 2u · e+ 22+ ue = f0+
jfje+ 2u · e+ 22+ ue
Multiplying this equality by −1
ue
, integrating by parts, and using the
Cauchy-Schwarz inequality, we obtain 1 ue22dx = − 1 f0+ jfjue2dx− 2 u· ue2dx− 2+ u2e2dx ≤ 1 2 f02 2e 2dx+1 2 u 2e2dx+1 fjju+ 2uje2dx + 2 u· ue2dx+ C u2e2dx ≤ 1 2 f2 0 2e 2dx+1 2 u 2e2dx+ 1 fj2e2dx + u 2e2dxC u 2e2dx
where > 0 is arbitrary and we used thatab ≤ a2+ 1 4b
2. Choosing sufficiently
small > 1
C to absorb the term with u in the right side by the left side and using
(2.3), (2.4) we yield the bound of Lemma 2.2.
Now we are ready to prove Theorem 2.1.
Proof of Theorem2.1. We consider (1.3) with a source term,
ARu= f in Q (2.5)
Using that are constants we have from Isakov et al. (2003), Section 2, the new system of equations
P1u=+ v+ f P2v= 3 jk=1 rjk · jku+ div f P1w= 3 jk=1 rjk × jku+ curl f (2.6) where P1= 2t− 3 jk=1−1 jk+ rjkjk, P2= 2t − 3 jk=1−1+ 2 jk+
rjkjk, where jk is the Kronecker delta. Due to (1.6) and (1.7) with small 0,
P1 and P2 are hyperbolic operators. Using (1.6), (1.11) by standard calculations
one can show that is strongly pseudo-convex in Q provided < C is sufficiently large and 0 is sufficiently small (see Isakov et al., 2003, or Isakov, 2006). Observe
that according to the first condition (1.11) the function x t= x2− 2t2− d2 is
pseudo-convex on Q and according to the second condition (1.11) the gradient of
is non-characteristic on Q with respect to operators 2
t − ,
2
t − + 2 ,
and hence with respect to P1 P2 provided 0 is sufficiently small. We fix such
observing that it depends only on Q and . It follows from Theorem 3.1 in Isakov et al. (2003) that there exists a constant C > 0 such that for all > C we have Q xtu2+ xtv2+ xtw2+ 3u2+ 3v2+ 3w2e2 ≤ C Q f2+ f2e2+ C0 Q 3 jk=1 jku 2e2 (2.7) for all u∈ H3
0Q. As well known, u= v − curl w. Therefore, by Theorem 3.2 in
Isakov et al. (2003) and by (2.7), Q 3 jk=1 jku 2e2 ≤ C Q u2e2 = C Q v − curl w2e2 ≤ C Q f2+ f2e2+ C0 Q 3 jk=1 jku 2e2
Thus for small 0, we yield Q 3 jk=1 jku 2e2 ≤ C Q f2+ f2e2
and the estimate (2.7) leads to the first Carleman estimate (2.1).
To prove the second estimate we will use Carleman estimates for elliptic and hyperbolic operators in Sobolev norms of negative order. Applying Theorem 3.2 in Imanuvilov et al. (2003) to each of scalar hyperbolic operators in (2.6) we obtain
Qu 2e2 ≤ C Q v2+ −2f2e2 Q v2e2 ≤ C Q 0u2+ f2e2 Qw 2e2 ≤ C Q 0u2+ f2e2 Adding these inequalities we arrive at
Q u2+ v2+ w2e2≤ C Q 0u2+ f2e2 (2.8)
To eliminate the first term in the right side we use again the known identity u=
v− curl w, apply Lemma 2.2, and integrate with respect to t over −T T to get
Qu
2e2≤ C Q
v2+ w2e2
Using this estimate and choosing 0small and > C we complete the proof of (2.2).
In order to use (2.1), it is required that the Cauchy data of the solution and the source term vanish on the lateral boundary. To handle non-vanishing Cauchy data, the following lemma is useful.
Lemma 2.3. For any pair of g0g1∈ H52× −T T × H
3
2× −T T, we can find a vector-valued functionu∗∈ H3Q such that
u∗= g0 u∗= g1 ARu∗= 0 on × −T T
and
u∗
3Q≤ Cg05
2× −T T + g132× −T T (2.9) for some C >0 provided 0in (1.7) is sufficiently small.
Proof. By standard extensions theorems for any g2∈ H12× −T T we can find
u∗∗∈ H3Qso that
u∗∗= g0 u∗∗= g1 2u∗∗= g2 on × −T T
and u∗∗
3Q≤ Cg212× −T T + g132× −T T + g052× −T T
Since × −T T is non-characteristic with respect to ARprovided 0 is small, the
condition ARu∗∗= 0 on × −T T is equivalent to the fact that g2 can be written
as a linear combination (with C1 coefficients) of 2
tg0 and tangential derivatives of
g0 (of second order) and of g1 (of first order) along . In particular,
g212× −T T ≤ Cg132× −T T + g052× −T T
Choosing g2 as this linear combination we obtain (2.9).
3. Hölder Stability for the Residual Stress
In this section we prove the first main result of the paper, Theorem 1.1. Let u 1 and u 2 satisfy (1.3), (1.4), (1.5) corresponding to R 1 and R 2, respectively. Denote u= u 2 − u 1 and F = R 2 − R 1 = fjk, j k= 1 3. By
subtracting equations (1.3) for u 1 from the equations for u 2 we yield
AR 2u= u 1F on Q where u 1F = 3 jk=1 fjkjku 1 (3.1) and u= tu= 0 on × 0 (3.2)
Differentiating (3.1) in t and using time-independence of the coefficients of the system, we get AR 2U= U 1F on Q (3.3) where U= 2 tu 3 tu 4 tu and U 1 = 2 tu 1 3 tu 1 4 tu 1
By extension theorems for Sobolev spaces there exists U∗∈ H2Qsuch that
U∗= U U∗= U on × −T T (3.4)
and U∗
2Q≤ CU32× −T T + U12× −T T ≤ CF (3.5)
due to the definitions of u U, and F . We now introduce V= U − U∗. Then
AR 2V= F − AR 2U∗ on Q (3.6)
and
V= V= 0 on × −T T (3.7)
To use the Carleman estimate (2.1), we introduce a cut-off function ∈ C2R4
such that 0≤ ≤ 1, = 1 on Q
2and = 0 on Q\Q0. By Leibniz’ formula
AR 2 V= AR 2V+ A1V= F − AR 2U∗+ A1V
due to (3.6). Here (and below) A1denotes a first order matrix differential operator
with coefficients uniformly bounded by C. By the choice of , A1V= 0 on Q2.
Because of (3.7) the function V∈ H2
0Q, so we can apply to it the Carleman
estimate (2.2) to get Q V2e2≤ C Q F2+ AR 2U∗2e2+ C Q\Q 2 A1V 2e2 ≤ C QF 2e2+ F2e2!+ Ce21 (3.8)
where != sup over Q and 1= e
4. To get the last inequality we used the bounds
(3.5) and (1.9).
On the other hand, from (1.3), (3.1), (3.2) we have 2tu=fjkjku 3tu=fjktjku
on × 0. So using the definitons of M F we obtain 2
tu 3
tu= MF on × 0,
and from the condition (1.10) we have
F2≤ C =23 tu 0 2 (3.9) Since T= 0, tux 0 2 e2x0dx= − T 0 t tux t 2e2xt dx dt ≤ Q 2 2+1 t u tu + t tu 2e2 + 2 Q\Q 2 tu 2 t e 2
where = 2 3. The right side does not exceed
C Q U2e2+ C Q\Q 2 U2e2 ≤ C Q V2e2+ C Q\Q 2 U2e2+ QU ∗2e2
because U= V + U∗. Using that = 1 on
2, < 1 on Q\Q2and < ! on
Qfrom these inequalities, from (3.8), from (3.5), and from (1.9) we yield
2 tu 2 0e20≤ C QF 2e2+ Ce21+ e2!F2 (3.10) Using that = 1 on
2, from (3.9) and (3.10) we obtain
2 F2e20≤ C Q 2 F2e2+ e2!F2+ Ce21 (3.11) where we also split Q in the right side of (2.7) into Q
2 and its complement, and
used thatF ≤ C and < 1 on the complement. To eliminate the integral in the
right side of (3.11) we observe that Q 2 F2xe2xtdx dt≤ 2 F2xe2x0 T −Te 2xt−x0dt dx
Due to our choice of function we have x t− x 0 < 0 when t = 0. Hence
by the Lebesgue Theorem the inner integral (with respect to t) converges to 0 as
goes to infinity. By reasons of continuity of , this convergence is uniform with
respect to x∈ . Choosing > C we therefore can absorb the integral over Q 2in
the right side of (3.11) by the left side arriving at the inequality
F
2e20≤ Ce2!F2+ Ce21
Letting 2= e2 ≤ on and dividing the both parts by e22 we yield
F
2≤ Ce2!−2F2+ e−22−1≤ Ce2!F2+ e−22−1 (3.12)
since e−22< C. To prove (1.13) it suffices to assume that F < 1
C. Then
= !+−log F
2−1 > C and we can use this in (3.12). Due to the choice of ,
e−22−1= e2!F2= F2!2−1+2−1
and from (3.12) we obtain (1.13) with = 2−1
!+2−1. The proof of Theorem 1.1 is now
complete.
4. Lipschitz Stability in the Cauchy Problem
Now we will prove a Lipschitz stability estimate for the Cauchy problem for the system (2.5). This estimate is a key to prove the estimate (1.17) in the inverse problem. Before going to the main result of this section, we state a lemma concerning the boundary condition for auxiliary functions v and w. We refer to Lin and Wang (2003) for the proof.
Lemma 4.1. Let a solution u∈ H3Q to systemA
Ru= f in Q satisfy
f= u = u= 0 on × −T T
and let R satisfy (1.7) with 0sufficiently small.
Then
ku= jku= 0 on × −T T for 1 ≤ i j k ≤ 3 Now we can prove the following result.
Theorem 4.2. Suppose that and T satisfy the assumptions of Theorem 1.2. Let u∈
H3Q3solve the Cauchy problem
ARu= f in Q
u= u= 0 on × −T T (4.1)
with f∈ L2−T T H1 and f= 0 on × −T T. Furthermore, assume that
(1.7) holds for sufficiently small 0.
Then there exists a constant C >0 such that u2
H1Q+ v2H1Q+ w2H1Q≤ Cf2L2−TTH1 (4.2)
By virtue of (4.2) and equivalence of the normsu1and of
div u0+ curl u0
in H1
0 (e.g., Duvaut and Lions, 1976, pp. 358–369), it is not hard to derive the
following
Corollary 4.1. Under the conditions of Theorem 4.2
u0Q+ xtu0Q+ tu0Q≤ Cf2L2−TTH1 (4.3) Proof of Theorem4.2. By standard energy estimates for the system (2.6) we get
C−1 E0− C ×0tf 2+ f2 ≤ Et ≤ C E0+ ×0tf 2+ f2 (4.4) for some C where
Et=
tu2+ tv2+ tw2+ u2+ v2+ w2+ u2+ v2+ w2 t
To use the Carleman estimate (2.1) we need to cut off u near t= T and t = −T.
We first observe that from the definition
1≤ x 0 x ∈
and from the condition (1.15)
x T= x −T < 1 when x ∈
So there exists a > 1
C such that
1− < on × 0 < 1 − 2 on × T − 2 T (4.5)
We now choose a smooth cut-off function 0≤ 0t≤ 1 such that 0t= 1 for
−T + 2 < t < T − 2 and t = 0 for t > T − . It is clear that AR 0u= 0f+ 2t 0tu+ 2t 0u
By Lemma 4.1 and basic facts about Sobolev spaces 0u∈ H03Q, we can use the
Carleman estimate (2.1) to get Q 3 0u2+ 0v2+ 0w2+ xt 0u2+ xt 0v2+ xt 0w2e2 ≤ C Q f2+ f2e2+ ×T−2<t<Ttu 2+ tu 2+ u2+ u2e2
Shrinking the integration domain on the left side to × 0 where = 1 and
1− < and using that < 1 − 2 on × T − T we derive that e21− 0 Etdt ≤ C Q f2+ f2e2+ Ce21−2 T T−2 tu2+ tu2+ u2+ u2 (4.6) To eliminate the last integral in (4.6) we remind that
curl tu= tw div tu= tv tu= tv− curltw and use the standard elliptic L2-estimate
t u2 t≤ C tv2+ tw2 t Now using the energy bound (4.4) we derive from (4.6)
e21− CE0− Ce 2! Q f2+ f2≤ Ce2! Q f2+ f2+ Ce21−2E0
Choosing so large that e−2<
C2 and fixing this we eliminate the term with E0
on the right side. Using again (4.4) we complete the proof.
5. Lipschitz Stability for the Residual Stress
In this section we prove the second main result of the paper, Theorem 1.2. We will use the notation of Section 4.
In view of Lemma 2.3, there exists U∗∈ H3Qsuch that
U∗= U U∗= U AR 2U∗= 0 on × −T T (5.1)
and U∗
3Q≤ CU52× −T T + U32× −T T ≤ CF (5.2)
due to the definition of F . We introduce V= U − U∗. Due to (5.1),
V= V= 0 AR 2V= 0 on × −T T (5.3)
Applying Corollary 4.1 to (3.6), (3.7) and using (5.2) gives V2 0Q+ xtV 2 0Q+ tV 2 0Q≤ C F2 1+ F 2 (5.4) On the other hand, as in the proof of Theorem 1.1 we will bound the right side by V.
We will use the cut off function 0 of Section 4. According to Lemma 4.1 and
(5.3), 0V∈ H03Q. By Leibniz’ formula
AR 2 0V= 0 U 1F − 0AR 2U∗+ 2t 0tV+ 2t 0V and by the Carleman estimate (2.1)
Q 2 0 3V2+ V2e2 ≤ C Q F2+ F2+ AR 2U∗2+ AR 2U∗2e2 + ×T−2<t<TV 2+ xtV 2+ tV 2e2 ≤ Q F2+ F2e2+ e2!F2+ e21−2 F2+ F2
where we let != supQand used (4.5) and (5.4). Since U= V + U∗ from (5.2) we
obtain Q 2 0U 2+ U2e2 ≤ Ce2!F2+ T −Te 2xtdt+ e21−2 F2+ F2xdx (5.5)
Utilizing (3.2) and (1.10), similarly to deriving (3.9), we get from (3.1) that 2
tu 3
tu= MF on × 0. Therefore, using (1.10) we will have
F2+ F2e2s0≤ C =23k=01 t ku 02e20 = −C T 0 t 2 0t ku2e2dx dt
≤ C Q 2 0 t ku+1 t ku + tt ku2e2 + C ×T−2T 0t 0 t ku2e2
Now as in the proofs of Section 3 the right side is less than C Q 20U2+ U2e2+ ×T−2TU 2+ U2e2 ≤ C Q 20U2+ U2e2+ e21−2F21+ F2
where we used equality U= U∗+ V and (5.2), (5.4). From two previous bounds we
conclude that F2+ F2e20≤ C e2!F2+ T −Te 2tdt+ e21−2 F2+ F2 (5.6)
Due to our choice of , 1≤ 0 t − 0 < 0 when t = 0. Thus by the
Lebesgue theorem as in the proofs of Section 3, we have 2C T −Te 2tdt+ e21− ≤ e20
uniformly on when > C. Hence choosing and fixing such large we eliminate the second term on the right side of (5.6). The proof of Theorem 1.2 is now complete.
By using Carleman estimates on functions satisfying the homogeneous zero
boundary conditions (g0= 0 or zero stress boundary condition) one can replace
in Theorem 1.2 by its “large” part (Isakov, 2006, Section 4.5, for scalar equations).
6. Conclusion
Using similar methods one can expect to demonstrate uniqueness and stability for both variable and residual stress most likely from two sets of suitable boundary data. Motivation is coming from geophysical problems. Our assumptions exclude zero initial data. So far it looks like a very difficult question to show uniqueness from few sets of boundary data when the initial data are zero. The stability guaranteed by Theorems 1.1 and especially by Theorem 1.2 indicates a possibility of a very efficient algorithms with high resolution for numerical idenitification of residual stress from single lateral measurements. It would a very good idea to run some numerical experiments to understand possibilities of practical applications of these stability properties.
Acknowledgments
The work of Victor Isakov was in part supported by the NSF grant DMS 04-05976. The work of Jenn-Nan Wang was partially supported by the grant
of National Science Council of Taiwan NSC 94-2115-M-002-003.The work of Masahiro Yamamoto was partly supported by Grant 15340027 from the Japan Society for the Promotion of Science and Grant 15654015 from the Ministry of Education, Cultures, Sports and Technology.
References
Bukhgeim, A. L. (2000). Introduction to the Theory of Inverse Problems. Utrecht: VSP. Bukhgeim A. L., Klibanov, M. V. (1981). Global uniqueness of a class of multidimensional
inverse problems. Soviet Math. Dokl. 24:244–247.
Duvaut, G., Lions, J. L. (1976). Inequalities in Mechanics and Physics. Springer-Verlag. Eller, M., Isakov, V., Nakamura, G., Tataru, D. (2002). Uniqueness and stability in the
Cauchy problem for Maxwell’s and elasticity systems. In: Cioranescu, D., Lions, J.-L., eds. Nonlinear Partial Differential Equations and Applications. College de France Seminar, 14, Pitman Res. Notes in Math, pp. 329–350.
Fabre, C., Lebeaux, G. (1996). Prolongement unique des solutions de l’equation de Stokes. Comm. Part. Diff. Equat.21:573–596.
Hansen, S., Uhlmann, G. (2003). Propagation of polarization in elastodynamics with residual stress and travel times. Math. Annalen 326:536–587.
Ikehata, M., Nakamura, G., Yamamoto, M. (1998). Uniqueness in inverse problems for the isotropic Lame system. J. Math. Sci. Univ. Tokyo 5:627–692.
Imanuvilov, O., Isakov, V., Yamamoto, M. (2003). An inverse problem for the dynamical Lamé system with two sets of boundary data. Comm. Pure Appl. Math. 56:1–17. Imanuvilov, O., Yamamoto, M. (2005). Carleman estimates for the non-stationary Lame
system and the application to an inverse problem. ESAIM Control Optim. Calc. Var. 11:1–56 (online).
Isakov, V. (1986). A nonhyperbolic Cauchy problem for bc and its applications to elasticity theory. Comm. Pure Appl. Math. 39:747–767.
Isakov, V. (2006). Inverse Problems for Partial Differential Equations. New York: Springer-Verlag.
Isakov, V., Nakamura, G., Wang, J. N. (2003). Uniqueness and stability in the Cauchy problem for the elasticity system with residual stress. Contemp. Math. AMS 333:99–113. Klibanov, M. V., Timonov, A. (2004). Carleman Estimates for Coefficient Inverse Problems
and Numerical Applications. Utrecht: VSP.
Lin, C.-L., Wang, J.-N. (2003). Uniqueness in inverse problems for an elasticity system with residual stress. Inverse Problems 19:807–820.
Man, C.-S. (1998). Hartig’s law and linear elasticity with initial stress. Inverse Problems 14:313–320.
Nakamura, G., Wang, J.-N. (2003). Unique continuation for an elasticity system with residual stress and its applications. SIAM J. Math. Anal. 35:304–317.
Rachele, L. (2003). Uniqueness in inverse problems for elastic media with residual stress. Comm. Partial Diff. Eq.28:1787–1806.
Robertson, R. (1997). Boundary identifiability of residual stress via the Dirichlet to Neumann map. Inverse Problems 13:1107–1119.
Robertson, R. (1998). Determining residual stress from boundary measurements: a linearized approach. J. Elasticity 52:63–73.