Integral Equations 0378-620X/82/040562-i1501.50+0.

20/0
and Operator Theory © 1982 Birkh~user Verlag, Basel
Vol.5 (1982)

INTEGRAL EQUATIONS WITH NON INTEGRABLE KERNELS

J.C. NEDELEC

We study here some integral equation s linked to the

Laplace or the Helmholtz equation, or to the system of elasticity
equations. These equations lead to non integrable kernels only
defined as finite parts, so that they are quite difficult to appro-
ximate. In each case, we introduce a variational formulation which
avoid~this difficulty and allow us to use stable finite element
approximations for these p r o b l e m s

I. Neumann's problem using a double layer representation

Let ~ be a bounded regular domain in ~3 which boun-
dary is the closed surface F We denote by ~' the domain
exterior to ~ . We are interested in the following problem :

Find the function u such that

(1.1) Au = 0 in ~ and in ~' ;

~U i DU e
- = g on F ;
--~lr ~n Ir
where u.1 is the interior limit of u in ~ , and ue the exte-
rior one, ~ being the u n i t a r y exterior normal to the surface £
Nedelec 563
/

We suppose that

(1.2) j[r g dy = 0 .

For any f u n c t i o n v , regular in ~ and in ~' , we w r i t e

Iv] = v.
l
- v
e

We can then prove that the p r o b l e m (1.1) has a u n i q u e solution in

the f o l l o w i n g Hilbert space X :
3

(Ixl 2 = xf)
i=l
WI(~ ') ={v' T ~v 6 L 2 (~') ; Vv 6 (L 2 ( ~ ' ) ) 3 }

W = v £ HI(£)/R* W1(£ ' ) ; Av=0 in ~ and ~' ;[~-~] = 0

The solution of this p r o b l e m admits a double-layer representation

of d e n s i t y ~(x) = [u(x)] ; x £ F . So that we have

(1.3) u(y) - 47 <o(x) ~ ( ) dy(x) ; y 6 193- F

The expression of the normal derivative of such a poten-

tial is not an integral• It is a finite part expression :

( 1 •4 ) 3u
~-n ( y ) = - 4--ff
1~ ~(x)
~2
~nx~ n (
~2~ ) dy(y) ; y £ F .
Y
1
Its kernel has a singularity in T ~ when x and y are
close .

Variational formulation
Let ~ be a f u n c t i o n on the surface F , in the H i l b e r t
!
space H2(F)/~ (see L I O N S - M A G E N E S [4] for the d e f i n i t i o n of this
space). We can solve

I
Au = 0 in ~ and in
(i .s)
[u] = ~ on r
Nedelec 564

This problem has the following variational formulation

J~[ 7u.Tv dx + ,7u'Vv dx = %0 ~-~ dy , V v E X .

In fF Bv .

There exists a unique solution in the Hilbert space X .

!
Now, let %0 and ~ be in H 2 ( F ) / R . We denote by
b(%0,~) the following bilinear form

F F
(1.6) b(%0,~) = ,|~7u(%0)'Vu(~) dx + ~|,Vu(%0)'Vu(*) d x

Then, a variational formulation of the integral equation (1.4) is

(1.7) b(%0,¢) = rJ
| g ~ dv ; v ~U E H½CF)/R
l"
From the properties of the space X , it also results that

(1.8) i
b(%0,~) = b(~,%0) ; v %0,, E H ' C r ) / ~

b(%0,%0) > B 11%0II H~CF)/~

z ; V %0 E H~(F)/I~
I ; B>O

The formulation (I..7) will be useful if we can obtain an

easy expression for the bilinear form b(%0,~). This expression
exists and, in order to get it, we need some differential operators
on the surface F .
To any function %0 on the surface F , we associate a
fqn~tion ~ defined in a neighbourhood of F in ~3 by

~(x) = %0(~(x)),
where ~(x) is the local projection of x onto F
We then define

gYad r %0(x) = gYad ~ ( x )

(1.9)
l
c~rlv%0(x)

To any tangential
= ~(x) A g~adF%0(x).

vector J of the surface r , we

associate a vector ~ defined in a neighbourhood of F in ~3

by
Nedelec 565

~(x) = 3(~(x))
and
curl F ~(x) : n ( x ) . c u + r l ~'(x)
(1.1o)
A F~o= curl F c~rl F ~

We have the following result :

THEOREM 1 : The bilinear form b(~,t) given by (1,6) has the

expression
I ~ I cu~IF~(x)'cu~iF*(Y) Z

(1.11) b(~,¢) =T~ F F IX - Yl --dy(x)dy(y);¥~,~£ EHffF)/~.

We also have

(1.1z) I ar ~ (x) I ^

Proof : Let us consider the vector h = Vu in ~ and in ~' We

!
have div h = 0 in ~3 , so that h = curl A We can choose A
unique such that div A = 0. We then have

AA = curl curl A - grad div A = curl h

But since curl h = O in ~ and in ~', curl h is a d i s t r i b u t i o n

on the surface F Using Green's formula, it can be proved that

curl h = curl F ~ ~F

where 6F represents the Dirac d i s t r i b u t i o n of the surface F

We can then give the following representation of A :

I curl F ~(x)
A(y) dy(x)
~" r ix - y i - -
Using again Green's formula, we obtain

f
b(~,¢) = j A(~) curlr ~ dy ,
F

and this proves the result. []

Nedelec 566

Numerical approximation
In order to approximate equation (1.7), we use finite
element techniques to construct an approximate surface Fh and a
finite dimensional subspace of H½(Fh)/R t

In the simplest case, the surface £h is built using

the finite element PI with three degrees of freedom

£h is a union of plate triangles, adjacent by their edges and

with a vertex on the surface F . In this case, the subspace Vh
is exactly the space of continuous functions on the surface £h
which are polynomials of degree 1oneach triangle. The approximate
problem is then

(I .13) 4-~ J~ ~ Ix yI dy(x)dy(y)

= £hgh~hdY ; V ~ h £ V h vh/ .
The corresponding matrix has coefficients bij which are a simple
combination of the following coefficients :

1 f f d~(x)d~(y!
akZ - 4~ JTkJT£ Ix- y]

In this case, it is possible to prove an error estimate

of the following type

(1.14) I£0- £0hlL2(F ) <~ (ci h + c2 h 2) II£OIIH2(F) ,

where h is the maximum diameter of the triangles. The first term

comes from the geometrical error and the second one, from the
interpolation error.

2. Neumann's problem for the Helmholtz' equation

Let us use the same notations as in Section I. Instead
of (I.I), we consider the following problem
Nedelec 567

iAu + k 2 u = 0 in a and ~' ;

(2.1) i~ u

The f u n c t i o n u i s now c o m p l e x and we add to (2.1) the so-called

radiation condition

(2.2) ~u - i k u = O( ) , when r ~

The double layer representation can be written under as

1 [ ~o(x) ; ~eiklx-Yl
(2.3) u(y) = - ~ J r ~ ~ Ix-y[ ) dy(x); ~ y e N 3 - r.

Again, the integral equation corresponding to this representation

involves a finite part, analogous to the one of equation (1.4).
We have the following result (due to HAMDI [2]) :

THEOREM 2 : The equation (2.1) can be replaced by an integral

equation using the representation (2.3). This equation has the
following variational expression

(2.4) b(~o,~) = F~I g 0 dy ; V ~ E H ½ ( F ) / N

F
where the bilinear form b is given by

I [ I e ik]x-yl
(2.5) b(w,*) = ~ Jr r Ix - y] c u r l r ~o(x) . c u r l F ¢'(Y) dy(x)dy(y')
_ k2 I I e i k l x - y l ~o(x) ,(y)(~x.~y) dy(x)d~F(y).
r r Ix y [

The approximation of the equation (2.4) using finite

element techniques is then similar to (1.13). We also obtain
similar error estimates when k is not an eigenvalue of the
Dirichlet interior problem.
Nedelec 568

3. Elasticity in ~3
In this section, we consider the integral equations asso-
ciated to the system of elasticity in R 3 , in the case correspon-
ding to the double layer representation.
We are looking for a vector u = (u~,u2,u3), defined in
and in ~' . We define the tensors e and ~ by

1 /~u ~uj'~ .
(3.1) sij(~) = ~,k~-~j.
I +3~xi } , I ~ i,j ~< 3 ;

(3.z) oij = 2p Eij k~.~ ~-.

The equilibrium equations are then

3

(3.3) ~ ~ oij (~1 = 0 in and ~' ; i = 1,2,3

Here, we are interested in the case where the boundary

conditions are the forces, that is
3

(3.4) )_2 o..1]

j=1
n. = gi ; i= 1,2,3
]
(or o.n = ).

The double-layer representation is the solution of the following

problem
3

j~l
•=
~-~j ~ij
(~) = 0 in ~ and a' • i= 1 ' 2,3 •'
'

(3.s) [~.~] = o ;

= $ ;

Let ~ be the rigid body displacements in ~3 :

(3.6) .~= u = a + ^ ~ ; ~, g = constant vectors .

Let X be the following Hilbert space

S

X = ~ 6 (HI(a}3/~ x (WI(a'3 a ;.= ~xj ~ij (~) =0 in a and a';~.n]=0

Then, using Korn's inequality, it can be proved that the problem

(3.5) has a unique solution in the space X , when ~ is given in
Nedelec 569

!
the space (H 2 ( F ) ) 3 ~ •

A fundamental solution of the elasticity system is a

symmetrical tensor U such that
3

(3.7) j~l
.= ~ ~ij(Uk. ) = ~ki ~(0)"

It is given by

(3.8) Uk£(X) = 4 ~ gk£ - 2(2~+X) ~Xk§X£ ( [ x l )

Using Green's formula, we t h e n o b t a i n a representation of the solu-

tion of (3.5) written as

(3.9) u k (y) = i , 2.~

j = l J Ir ij(Uk.(Y-X)) nj(x) ~i(x) dT(x) ; y 6 N 3- r .

The i n t e g r a l equation linking ~ and g is then

3

(3.10) ~ ~ Xijkz(Y-X) nk(x) ng(y) ~i(x) dy(x) = gj(y) ; y£ r,

i,k,g=1 r

where Z is the 3 x ,3 symmetrical tensor of order 4

(3.11) Z(x-y) = ox ey(U(x-y)).

T h i s t e n s o r has s i n g u l a r c o m p o n e n t s w h i c h a r e i n ~ 1 when
and y a r e c l o s e , so t h a t t h e i n t e g r a l in ( 3 . 1 0 ) has in f a c t
finite parts.

Variational formulation

We can o b t a i n a v a r i a t i o n a l formulation of the equation

( 5 . 1 0 ) , u s i n g t h e p r o b l e m ( 3 . 5 ) . L e t u be t h e s o l u t i o n o f ( 3 . 5 )
+
associated to ~ and l e t ~ be t h e s o l u t i o n a s s o c i a t e d to
L e t us s e t
3

-~f o1
(3.12) b(;,~) -i,j=lJaU 2 .j(~) E i j ( { ) dx

An e q u i v a l e n t formulation of equation (3.10) is then

Nedelec 570

(3,13) : ' •

THEOREM 3 : The b i l i n e a r form b given by (3.12) is symmetric,

coercive on the space (H ~ (F))3~. It is given by

(3.14) b($,~) =i,~IIFfFGij .(x-y)c~rl F ~i(x).c~rlF~j(y)dy(x)dy(y

where Gij.. is a tensor of order 4 given by

(3.15) Gij.. = c~rl x c~rly Ai.j.(x-y),

A being a tensor of order 4 given by

Aijk~(X) "~ " xi j3Xk3 Z 2880

+ 21V { 3 2 f ~ ~ + 32 fx_~J~
\ ~ ~ 96~ ~ ~k~ ~-i~Xk ~ 96~ ~ ~ij)

96~ ~ ~£j 3xj3x k 96~ ~ ~ i

+ 32 + 32 lx-~-~ ~kj)
3xj~x---
7 (9x-~16J)
~ki 3xi3x~ ~ 96~
+~ 8~ (~ki ~Zj + ~ i ~jk ) + 2 W ~ ( 8~ ) ~ij ~kZ "

We then have

(3.17)

4. Plane elasticitx
We consider here the problem of elasticity in ~3 . We
are looking for the displacement of the elastic body ~ = (ul,u2).
Exactly as in the previous section, the equations are
3
3
(4. I ) E ~ oij (~) = 0 in ~ and ~' ,• i = 1,2 ,•
j~l j
(4.2) 2_~ °ij n.] = g i on F '• i = 1,2 "
y.=~
Nedelec 571

A fundamental solution is the following tensor of order 2

(4.3) Ukz(X) = F 1 ( _1 L o g ( i x l ) 6 k Z + 2~+X aXkaXg( Ixl! ~ '(Ix I))).

Let £ be the t e n s o r of o r d e r 4 such t h a t
(4.4) £(x-y) = a x ~y ( U ( x - y ) ) .
Then a r e p r e s e n t a t i o n of the s o l u t i o n as a " d o u b l e - l a y e r " p o t e n t i a l
is g i v e n by
2

(4.5) uk(Y) = - L [ eij(Uk.(Y-x))nj(x) ~i (x) dy(x) ; yE N3-F.

i,j=1J
The integral equation linking W and g has the expression
2

(4.6) - E ~ liD ~(y-x) nk(x ) ng(y) ~i(x) dy(x)= gj(y);yEr.

i,k,g=1
We have the following result [I] :

THEOREM 4 : The equation (4.6) admits a variational formulation

which is

(4.7) b(~o,~) = ~ ~ d~ ; V ~ E (H½(r))e/~;?

÷÷ f
-X2
~¢~= ," = a + B ; ~, B = c o n s t a n t s } ;
+X1

and the symmetric bilinear form b is given by

2

(4.8) b(~,~) =- E J I d~0ic d , ~ 4

i,j=1 r F Gij(x-Y) -~,x~ ~ s (y) as(x) ds(y)

(4.9) Gij (x) = 2~ (X+~) (Log

(I+2~)~ I xl ~ ij _ -~x--~-]
xi x 3 "

We also have

(4.10) b(<o,•) t> BIIR°ll2H½(V))2k/SP-- ; ¥ ~0 £ (H½(F)) 2 ; B > 0

Nedelec 572

REFERENCES

[1] BONNEMAY, P., Equations int~grales pour l'~lasticit~ plane,

Th~se de 3~me cycle, Universit~ de Paris VI, 1979.

[2] HAMDI, M.A., Une formulation variationnelle par ~quations

pour la r@solution de l'~quation de Helmholtz avec des
conditions aux limites mixtes, Note au C.R.A.S., Paris,
S~rie II, T. 292 (1981).

[3] HA DUONG, T., A finite element method for the double-layer

potential solutions of the Neumann's problem, Math.
Meth. in the Appl. Sc~., 2 (1980), 191-208.

[4] LIONS, J.L., MAGENES, E., Probl@mes aux limites non h o m o ~ n e s

et Applications, T.I, Dunod, Paris, 1968.

[5] NEDELEC, J.C., R@solution par potentiel de double couche

du probl~me de Neumann ext~rieur, Note au C.R.A.S., Paris,
S~rie A, T. 286 (1978), 103-106.

Centre de Math~matiques Appliqu~es (ERA/CNRS 747)

91128 PALAISEAU CEDEX FRANCE
-