PLEASE SCROLL DOWN FOR ARTICLE

This article was downloaded by:

On: 22 January 2010
Access details: Access Details: Free Access
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
Numerical Functional Analysis and Optimization
Publication details, including instructions for authors and subscription information:
http://www.informaworld.com/smpp/title~content=t713597287
NUMERICAL RECONSTRUCTION OF INTERNAL STATES BY
BOUNDARY MEASUREMENTS FOR THE LAPLACE EQUATION
Qingchuan Wang
a
; Kazuei Onishi
a
a
Department of Mathematical Sciences, Ibaraki University, Japan
Online publication date: 30 June 2001
To cite this Article Wang, Qingchuan and Onishi, Kazuei(2001) 'NUMERICAL RECONSTRUCTION OF INTERNAL
STATES BY BOUNDARY MEASUREMENTS FOR THE LAPLACE EQUATION', Numerical Functional Analysis and
Optimization, 22: 3, 441 453
To link to this Article: DOI: 10.1081/NFA-100105112
URL: http://dx.doi.org/10.1081/NFA-100105112
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.
NUMERICAL RECONSTRUCTION OF INTERNAL
STATES BY BOUNDARY MEASUREMENTS FOR
THE LAPLACE EQUATION
Qingchuan Wang
1
and Kazuei Onishi
2
1
Graduate School, Department of Mathematical Sciences,
Ibaraki University, Japan
E-mail: nd9412@mcs.ibaraki.ac.jp
2
Department of Mathematical Sciences, Ibaraki University, Japan
E-mail: onishi@mito.ibaraki.ac.jp
ABSTRACT
The purpose of the paper is to present a unied numerical method
for problems consisting of the conventional boundary value problem.
Cauchy problem, under-determined problem, and over-determined pro-
blem. The method is based on the direct variational approach, which
paraphrases the problems into the primary and adjoint boundary value
problems that can be tackled by commonly used computer programs for
the numerical solution of the Laplace equation.
Key Words: Inverse problem; Direct variational approach; Laplace
equation.
AMS Subject Classication: 65L08.
1. BOUNDARY VALUE PROBLEMS OF THE
LAPLACE EQUATION
Let be a domain boundary by a smooth curve in the two-dimen-
sional spaces R
2
with the rectangular coordinates x =(x
1
, x
2
). Consider an
441
Copyright
#
2001 by Marcel Dekker, Inc. www.dekker.com
NUMER. FUNCT. ANAL. AND OPTIMIZ., 22(3&4), 441453 (2001)
D
o
w
n
l
o
a
d
e
d

A
t
:

1
0
:
3
0

2
2

J
a
n
u
a
r
y

2
0
1
0
ORDER REPRINTS
unknown function u(x) which satises the Laplace equation:
u x 0, x 2 . 1.1
The Dirichlet data u " uu are given on a non-empty arc
u
of the boundary ,
and the Neumann data ou,on " qq are given on a non-empty arc
q
of the
boundary . Here n denotes the outward unit normal to the boundary . The
problem statement is shown in Figure 1.
Problem 1. For given " uu 2 H
1,2

u
and " qq 2 H
1,2

q
, nd u 2 H
1
which
satises the Laplace equation (1.1).
According to the dierent arrangement of
u
and
q
along the boundary
, Problem 1 can be classied into the following:
I. If
u
=
q
, Problem 1 is called a Cauchy problem or an initial
value problem;
II. If
u
\
q
and
u
[
q
, Problem 1 is called a (mixed)
boundary value problem. Especially, if
u
= (i.e.
q
=) it is
called a Dirichlet problem (the rst kind boundary value prob-
lem), and if
q
= (i.e.
u
=) it is called a Neumann problem
(the second the boundary value problem):
III. If
u
\
q
,
u
[
q
6 as shown in Figure 1, Problem 1 is
called an under-determined problem;
IV. If
u
\
q
,
u
6
q
, Problem 1 is called an over-determined
problem.
We know some results [1] on the existence of solution of the Cauchy
problem. For Laplace equation (1.1), if the Dirichlet data uj

u
0, uj

c
u
o
are given, then the set fou,on j o 2 H
1,2

c
u
g is dense in H
1,2

u
.
Regarding uniqueness of the solution we know the following result [2]: Let
the boundary
u
coincide with horizontal axis x
1
. If u ou,ox
2
0, and
u, ou,ox
2
are continuous in [
u
, then ux 0 in .
442 WANG AND ONISHI
Figure 1. Problem statement.
D
o
w
n
l
o
a
d
e
d

A
t
:

1
0
:
3
0

2
2

J
a
n
u
a
r
y

2
0
1
0
ORDER REPRINTS
The rst kind boundary value problem is equivalent to the following
variational problem [3]: Find u x 2 H
1
which minimizes the Dirichlet
integral
_

ru x

2
d:
inf
u
u j j
_

ru x

2
d. 1.2
The Laplace equation is a typical elliptic equation. Besides the bound-
ary value problem and the Cauchy problem, we need to solve the under-
determined problem and the over-determined problem as well in many prac-
tical cases. To solve those various kinds of problems, we present in this paper
a direct variational method from the point of view of an inverse problem [4].
We also give an example in application of our methods to an elliptic system
of equations a magnetostatic problem.
2. DIRECT VARIATIONAL METHOD
Let
c
u
n
u
,
c
q
n
q
. To solve the Problem 1:
u x 0, x 2 ,
u

u

" uu,
ou
on
j

q
qj
q
. " qq,
we identify o 2 H
1,2

c
u
which minimizes the following functional:
J o
_

q
q x; o " qq x

2
d j
_

ru x; o

2
d. 2.1
Here u(x; o) is the solution of the problem called a primary problem:
u x; o 0, x 2 , 2.2
u

u

" uu, u

c
u

o, 2.3
where ou,on x; o q x; o on
q
. Here j ! 0 is a regularization
parameter.
We use the direct method to minimize the functional J(o). That is,
taking a suitable initial guess o
0
2 H
1,2

c
u
for k 0, 1, 2, . . . , we proceed:
o
k1
o
k
o
k
J
0
o
k
. 2.4
LAPLACE EQUATION 443
D
o
w
n
l
o
a
d
e
d

A
t
:

1
0
:
3
0

2
2

J
a
n
u
a
r
y

2
0
1
0
ORDER REPRINTS
Here the Fre chet derivative J
0
o 2 H
1,2

c
u
is dened by:
J o oo J o J
0
o , oo
_ _
o oo k k ,
oo k k
2

c
u
oo x

2
d. 2.5
We use the Armijo criterion [5] to control the step size o
k
.
Theorem. Assume J
0
o 2 H
1,2

c
u
, and let v x be the solution of the adjoint
problem:
v x 0, x 2 ,
v

q

2
ou
on
" qq
_ _
, vj

c
q
. 0.
Then J
0
o has the following expression [6]:
J
0
o

c
u

ou
on
2j
ou
on
2.6
Proof. We note:
Jo oo Jo

_
q
jqx; o oo " qqxj
2
jqx; o " qqxj
2
_ _
d
j
_

jrux; o ooj
2
jrux; oj
2
_ _
d

_
q
qx; o oo qx; o 2" qqxgfqx; o oo qx; o
_
gd
j
_

ou
ox
1
x; o oo
ou
ox
1
x; o
_ _
ou
ox
1
x; o oo
ou
ox
1
x; o
_ _ _

ou
ox
2
x; o oo
ou
ox
2
x; o
_ _
ou
ox
2
x; o oo
ou
ox
2
x; o
_ __
d

_
q
fqx; o oo qx; o 2qx; o " qqxgoqx; o d
j
_

oou
ox
1
x; o 2
ou
ox
1
x; o
_ _
oou
ox
1
x; o
_

oou
ox
2
x; o 2
ou
ox
2
x; o
_ _
oou
ox
2
x; o
_
d
444 WANG AND ONISHI
D
o
w
n
l
o
a
d
e
d

A
t
:

1
0
:
3
0

2
2

J
a
n
u
a
r
y

2
0
1
0
ORDER REPRINTS

q
foqx; o 2qx; o " qqxg oqx; o d
j
_

oou
on
x; o 2
ou
on
x; o
_ _
oux; o d
_

foux; o 2ux; og oux; o d

_

q
2qx; o " qqx oqx; o d
_

q
joqx; oj
2
d
j
_

c
u
2
ou
on
x; o oox; o d j
_

c
u
oqx; ooox; o d.
In the above, we put oux; o ux; o oo ux; o, and accordingly
oqx; o qx; o oo qx; o. We notice that ou 0 in , ou 0
on
u
, and o
u
oo on
c
u
. From Greens integral theorem:
_

v ou d
_

ov
on
ou d
_

v
oou
on
d
_

v ou d
we have
0
_

c
u
ov
on
ood
_
q
2 q " qq oq d.
Consequently we know that
o oo Jo
_

c
u
ov
on
ood j
_

c
u
2
ou
on
ood okook

ov
on
2j
ou
on
. oo
_ _
L
2

c
u

o oo k k .
The way of choosing functional is not restricted to the form of (2.1). for
the example, if we take the following functional:
J o
_

u
u x; o " uu x

2
d, 2.7
then the primary problem becomes:
u x; o 0, x 2 ,
ou
on

q

" qq, uj

c
q
o,
LAPLACE EQUATION 445
D
o
w
n
l
o
a
d
e
d

A
t
:

1
0
:
3
0

2
2

J
a
n
u
a
r
y

2
0
1
0
ORDER REPRINTS
and the adjoint problem becomes:
v x; o 0, x 2 ,
ou
on

u
2 u " uu , v

c
u

0.
The corresponding Fre chet derivative has the following form:
J
0
o j

c
q
.
ov
on
.
If the values of function u(x) are available at n internal points and
the Dirichlet data are given on a non-zero measures part
u
of the boundary
[7], the functional to be minmized is as follows:
J o

n
j1
u x
j
; o
_ _
" u u
j

2
. 2.8
The relevant primary problem is
u x; o 0, x 2 ,
u

u

" uu, u

c
u

o.
The adjoint problem is:
vx; o

n
j1
2fux; o " uuxgox x
j
, x 2 ,vj

0.
Here ox x
j
is the Dirac delta function with unit measure at the point x
(j)
.
Then the corresponding Fre nchet derivatives has the following form:
J
0
o

c
u

ou
on
. 2.9
3. NUMERICAL EXAMPLES
In the boundary value problem
u
\
q
and
u
[
q
, Let
be a unit disk,
u
be the upper semicircle, and
q
be the lower semicircle. The
Dirichlet data " uu cos 2 are given on the boundary
u
and the Neumann
data " qq 2 cos 2 are given on the boundary
q
. Take o
0
1 - - 2,
is the central angle). By using boundary element collocation method we use
24 nodes on the boundary (see Figure 2). Our numerical result is shown in
446 WANG AND ONISHI
D
o
w
n
l
o
a
d
e
d

A
t
:

1
0
:
3
0

2
2

J
a
n
u
a
r
y

2
0
1
0
ORDER REPRINTS
Figure 3. u* denotes the exact solution on the boundary. Black points denote
the nodes on the boundary
u
, and triangles denote nodes on the boundary

q
. At points =0 and = on the boundary , we use double nodes.
Although we did not use many boundary elements, our numerical solution
u
10
is in fairy good agreement with the exact solution on the boundary.
Next, we consider an under-determined problem in the unit disk . The
Dirichlet data " uu cos 2 are given on the boundary
u
0 - - ,2 , and
Neumann data " qq 2 cos 2 are given on the boundary
q
- - 3,2 . In
the values of function u(x) are available [8] at ve points
x
j
j 1, 2, . . . , 5 as shown in Figure 4. The black points on the boundary
denote nodes on . The ve small circles in denote the internal points. The
numerical process is stable. On the boundary the convergent solution u
10
is in
good agreement with the exact solution u

cos 2 in the vicinity of the

boundary
q
(see Figure 5). In this case, the functional to be minimized in
(2.1) should be revised as follows:
J o
_

q
q x; o " qq x

2
d

5
j1
u x
j
; o
_ _
" uu
j

2
j
_

ru x; o

2
d subject to u

u

" uu. 3.1

LAPLACE EQUATION 447
Figure 2. Problem statement and boundary elements of the boundary value problem.
Figure 3. Numerical solution in the case of
u
\
q
and
u
[
q
.
D
o
w
n
l
o
a
d
e
d

A
t
:

1
0
:
3
0

2
2

J
a
n
u
a
r
y

2
0
1
0
ORDER REPRINTS
The corresponding Fre chet derivatives has the following form:
J
0
o

c
u

ou
on
x; o 2j
ou
on
x; o . 3.2
Here v(x) is the solution of the Poisson equation:
v x

n
j1
2 u x; o " uu x ox x
j
, 3.3
subject to the boundary conditions:
vj

q
2 q x; o " qq x
_ _
. v

c
u

0. 3.4
448 WANG AND ONISHI
Figure 4. Problem statement and boundary elements of the under-determined problem with
internal information.
Figure 5. Numerical solution of the under-determined problem with internal information.
D
o
w
n
l
o
a
d
e
d

A
t
:

1
0
:
3
0

2
2

J
a
n
u
a
r
y

2
0
1
0
ORDER REPRINTS
As the third example, let be a circlet,
u
be the external circle.
c
u
be
the internal circle, and let
u
=
q
. We take the radius of the external circle as
2, 3, 5 respectively. We take boundary data of the harmonic function:
u x
1
, x
2

1
2
x
1
2x
2

1
4
x
2
1
x
2
2
_ _
on the external circle as Cauchy data. Suppose that u(x
1
, x
2
) is unknown in
. The relevant numerical solutions are shown in Figure 6. The narrower the
circlet, the better our numerical result is.
LAPLACE EQUATION 449
Figure 6. Numerical solutions of the Cauchy problem.
D
o
w
n
l
o
a
d
e
d

A
t
:

1
0
:
3
0

2
2

J
a
n
u
a
r
y

2
0
1
0
ORDER REPRINTS
4. APPLICATION TO A MAGNETOSTATIC
INVERSE PROBLEM
Let H(A/m) be magnetic eld, B(Wb/m
2
) be magnetic ux density,
j(H/m) be magnetic permeability, and J(A/m
2
) be electric current along
vertical direction in two dimensions. Let be a domain bounded by a
smooth simple closed curve . Let n denote outward unit normal to the
boundary , and let t denote the unit tangent along . We consider the
following problem:
Problem 2. On a non-zero measure part of the boundary
d
, the tangential
component Ht of the magnetic eld and the normal component B n of the
magnetic ux density are given as
"
HH
t
and
"
BB
n
respectively. Find H, which
satises the following equations:
rot H J, div B 0. x 2 . 4.1
Here B=jH, and we assume that J and j are given
We can regard this problem as the Cauchy problem. The problem can be
solved by nding oj

c
d
n
d
which minimizes the following functional [9]:
J o
_

d
H x; o t
"
HH
t

2
d j
_

B x; o H x; o d. 4.2
The relevant primary problem is:
rot H x; o J, div B x; o 0, x 2 , 4.3
B n

d

"
BB
n
, H tj

c
d
. o. 4.4
The corresponding adjoint problem is:

^
AA x 2jjJ, x 2 , 4.5
^
AA

d

2 H t
"
HH
t
_ _
,
o
^
AA
on

c
d

2jjo. 4.6
Here
^
AA is the solution of the adjoint problem (4.5) and (4.6). the Fre chet
derivative is given as J
0
o j

c
d
.
^
AA.
For example consider a square domain with each side length equal to 1.

d
is the union of side x
1
=1 and side x
2
=1 of the square.
"
BB
n
2,3x
2
1,2
and
"
HH
t
1,2x
2
5,3 on the side x
1
=1.
"
BB
n
5,3x
2
1,2 and
"
HH
t

1,2x
1
2,3 on the side x
2
=1. Uniformly random errors of the magnitude
50% are added to
"
BB
n
and
"
HH
t
. For simplicity, take J =1 and j=1. We take
an initial guess o
0
10, and use triangular nite elements and edge elements.
450 WANG AND ONISHI
D
o
w
n
l
o
a
d
e
d

A
t
:

1
0
:
3
0

2
2

J
a
n
u
a
r
y

2
0
1
0
ORDER REPRINTS
Figure 7 shows the numerical solutions under dierent values of the par-
ameter j. We can conrm that the regularization term with suitable value
on j yields good approximate solutions even if errors are contained in the
boundary data.
5. CONCLUSIONS
We present a numerical reconstruction of internal states by boundary
measurements for the Laplace equation not only for the traditional Cauchy
LAPLACE EQUATION 451
Figure 7. Mesh and nite element solutions of magnetostatic inverse problem.
D
o
w
n
l
o
a
d
e
d

A
t
:

1
0
:
3
0

2
2

J
a
n
u
a
r
y

2
0
1
0
ORDER REPRINTS
problem and the boundary value problem, but also for the under-determined
and the over-determined problems. We use the direct variational method to
transform the problems into the corresponding primary and adjoint prob-
lems. We show the concrete expressions of the Fre chet derivative of the
functional to be minimized. Through numerical examples we show that
stable numerical solutions can be obtained for various inverse problems
even though they are ill-posed problems. In the example with internal
points, the adjoint problem is expressed as a problem of solving the
Poisson equation.
Our method has two advantages. On is that we can use a unied method
to solve boundary value problems, no matter whether they are traditional or
not. The other one is that without any changes of computer codes we can
employ the computer programms of nite element and boundary element
methods which are widely used for solving boundary value problems. We
can mention two disadvantages of the method. One is that the convergent
rate is slow in our interative process. The other one is that we still choose
suitable values of regularization parameter j experimentally.
ACKNOWLEDGMENTS
The authors would like to thank Professor Y. Ohura for providing the
numerical results in the Cauchy problem. We also thank Dr. T. Shigeta for
his work in Section 4.
REFERENCES
1. I.L. Lions, Optimal Control of Systems Governed by Partial Dierential
Equations, Springer-Verlag, Berlia, 1971.
2. F. John, Partial Dierential Euqations (Second Edition), Springer-Verlag New
York, 1975.
3. J. Jost, Postmodern Analysis (Translated by Azad, H.), Springer-Verlag Berlin
Heidelberg, 1998.
4. N. Tosaka, K. Onishi, and M. Yamamoto, Mathematics and Solution of Inverse
Problems Inverse Analysis of Partial Dierential Equations (in Japanese),
Tokyo University Press, 2000.
5. L. Armijo, Minimization of functions having Lipachitz-continuous rst
partical derivatives, Pacic Journal of Mathematics, 16 (1966) 13.
6. Q. Wang, Y. Ohura, K. Shirota, and K. Onishi, Boundary element solutions of
under-determined Laplace equation in two dimensions, Proceedings of the Ninth
BEM Technology Conference, JASCOME (Tokyo, Japan, 1999), 5761.
7. K. Kobayashi, K. Onishi, and Y Ohura, On identifying Dirichlet condition for
2D Laplace equation by BEM, Engineering Analysis with Boundary Elements, 17
(1996) 223230.
452 WANG AND ONISHI
D
o
w
n
l
o
a
d
e
d

A
t
:

1
0
:
3
0

2
2

J
a
n
u
a
r
y

2
0
1
0
ORDER REPRINTS
8. Q. Wang, Y. Ohura, K. Shirota, and Onishi, Boundary element solutions of
under-determined Laplace equation in two dimensions (II), Proceedings of the
Sixteenth Japan National Symposium on Boundary element Methods, JASCOME
(Tokyo, Japan, 1999), 8186.
9. T. Shigeta, Mixed nite element solution for a magnetostatic inverse problem,
Computational Methods for Solution of Inverse Problems in Mechanics, ASME,
AMD 228 (1998), 6571.
Received March 31, 2000
Revised September 14, 2000
LAPLACE EQUATION 453
D
o
w
n
l
o
a
d
e
d

A
t
:

1
0
:
3
0

2
2

J
a
n
u
a
r
y

2
0
1
0




















Order now!

Reprints of this article can also be ordered at
http://www.dekker.com/servlet/product/DOI/101081NFA100105112
Request Permission or Order Reprints Instantly!
Interested in copying and sharing this article? In most cases, U.S. Copyright
Law requires that you get permission from the articles rightsholder before
using copyrighted content.
All information and materials found in this article, including but not limited
to text, trademarks, patents, logos, graphics and images (the "Materials"), are
the copyrighted works and other forms of intellectual property of Marcel
Dekker, Inc., or its licensors. All rights not expressly granted are reserved.
Get permission to lawfully reproduce and distribute the Materials or order
reprints quickly and painlessly. Simply click on the "Request
Permission/Reprints Here" link below and follow the instructions. Visit the
U.S. Copyright Office for information on Fair Use limitations of U.S.
copyright law. Please refer to The Association of American Publishers
(AAP) website for guidelines on Fair Use in the Classroom.
The Materials are for your personal use only and cannot be reformatted,
reposted, resold or distributed by electronic means or otherwise without
permission from Marcel Dekker, Inc. Marcel Dekker, Inc. grants you the
limited right to display the Materials only on your personal computer or
personal wireless device, and to copy and download single copies of such
Materials provided that any copyright, trademark or other notice appearing
on such Materials is also retained by, displayed, copied or downloaded as
part of the Materials and is not removed or obscured, and provided you do
not edit, modify, alter or enhance the Materials. Please refer to our Website
User Agreement for more details.


D
o
w
n
l
o
a
d
e
d

A
t
:

1
0
:
3
0

2
2

J
a
n
u
a
r
y

2
0
1
0