Académique Documents
Professionnel Documents
Culture Documents
generalized Laplace
equation by coupling the
boundary element method
and the perturbation
method
R. Rangogni
ENEL-DSR-CRIS. Via Ornato 90/14, Milan. Italy
(Received January 1986)
The boundary element method (BEM) has, in general, some advantages with respect to domain methods, in so far as no internal discretization of the domain is required. Such an advantage, however,
is lost if the BEM is used directly for the numerical solution of the
generalized Laplace equation (GLE). This paper demonstrates that
the GLE can also be dealt with advantageously by coupling the BEM
with the perturbation method (PM). The technique involves transforming the starting equation (GLE) into an equation without partial
first derivatives and then solving a few problems by BEM. The
method requires an internal network but the unknowns are only
on the boundary. The procedure is applied to numerically solve two
test problems with known analytical solutions.
Key words: mathematical models, generalized Laplace equation,
boundary element method
The boundary element method (BEM) has become a
powerful and alternative method to the well known
methods of finite differences (FDM) and finite elements
(FEM). This is due mainly to the simplicity in the
arrangement of the input data, to a lesser number of
unknowns to be dealt with and to its applicability to
problems defined on infinite domains.
The application range of BEM, however, is confined
to simple differential equations owing to the difficulty
of finding a fundamental solution (FS) for a general
differential operator. The applications of BEM mainly
concern constant coefficient differential equations such
as the Laplace, Poisson, Helmholtz, wave, heat equations etc. It is well known that the Laplace equation,
the first one to which BEM was applied, is a particular
case of a more general equation called the generalized
Laplace equation (GLE).I
Unfortunately, it is not advantageous to apply BEM
266
u = [a(x,y)p-gl = G,
Ou
1 Oa
On
2a On
on Si
u + (a)-~g:
on $2
Ou (
l Oa ]u
- - + g3-0,,
~Tn/ = 0
(6)
onS3
on D
(7)
on
(1)
where V. and V denote the divergence and gradient operators. Associated with equation (1) are the general
boundary conditions:
V(x,y) = gt(x,y)
on S,
avian = gz(x,y)
on $2 S~ u $2 u S3= S
aV/an =g3(x,y)V
on S 3
(2)
(3)
(4)
V2a
2a
(5)
(8)
Perturbation method
Instead of equation (5), the following family of equations is considered:
0~<e~<l
V2llq-Ef(lg)=O
(9)
or, in other words, equation (5) is thought of as a perturbation of the Laplace equation; for e = 0, equation (9)
becomes the Laplace equation while for e = 1, equation
(5) is obtained.
The solution of equation (9) is sought in the form:
lg = bl0 "4- ELl I q- g21.12 4- . . . = ~ U j g
j=0
(10)
0
(11)
Ilo ~
G I
Ouo
On
0u0
-
on D
on Sl
10a
Uo + (a)~gz
2a On
(
1 0a
-
)uo = O
on S 2
P0
02)
on $3
267
on D
UI=0
on S t
OU1
On
Oul
on S 2 P l
-0
a,, +
1 Oa~
=0
(13)
on $3
G(P,Q)=log(r)
r=[P-Q[
PeD
Numerical
Q~D
(14)
au,(P) =
( / Ouj u OG'~ds
JsIG
foGf(ui-l)dD(15)
J~n)
k=l
k=l
o%
son
6k =
e2(%)
max I Vi - 9"i]
i
100
maxV
D
--
F~i(Vi - 15"i2}X
maxV
(18)
100
D
Subscript i extends over points Pi --- (xi,yi) c D U S
where a numerical value is computed.
In the two problems, parameter R is very high so
the technique of reference 2 is not applicable. To compare the two procedures, however, the errors given by
equations (18) obtained using the reference 2 technique
are shown between brackets ( ) for both problems. For
all problems, only solutions u0 and ut are computed.
The computer times refer to a Vax 750.
Example 1
V2V+Ivv.Va=O
on
D - - {(x,y):
0<x<5;
0<y<5}
a(x,y) = (x + 1)(y + 2)
V(x -= 0,y) = 0
(17)
GfuodD
(19)
V(x = 5,y) = 5
OV
On
--=0
y=0,5;0~<x~<5
D k
268
e=(%)-
(16)
s[~n dS
In this section, two problems, both with known analytical solutions and solved numerically in reference 2, are
solved numerically using the procedure presented in the
previous paragraphs. To measure the error between the
analytical solution V and the numerical solution V, two
percentage errors are defined as:
01k
examples
--
&Ok=fGdS
s~
V(x,y) = ~
log (1 + x)
f(x,y) = ~
+ (2 +
(20)
'
'
'
'
op,
IF;--"
o
o
o
21- \',
Figure 1 N e t w o r k s u s e d in e x a m p l e 1
The boundary S is discretized by 40 linear equal elements. To check the influence of the internal discretization two internal meshes are used: the first very coarse
and the second finer as shown in Figure 1. The results
are checked on the 40 boundary points as well as on
the internal points (five for the coarse mesh and 56 for
the finer one). The errors given by equations (18) are:
e~, = 2.34%
e2
Mesh a
1.76%
e~ = 2.27%
(e= = 7.4%)
Mesh b
(22)
(e2 = 17.9%)
e2 = 2.46%
The maximum error e= has about the same value for
the two computational networks considered. This
explains the fact that e_, for network b (96 points) is
larger than e, for network a (45 points). The computing
time is 37.8s using mesh b against 30.5 s using the reference 2 technique. The improvement in the results should
be noted.
Example 2
The electrical potential is required of a cylindrical capacitor with non-homogeneous dielectric. Mathematically, the problem is defined by:
1
V2V + - V V . V a = O
a
on
a(x,y) = 1 + r
V= 1
r= 1
V=5
r=5
OV
r 2=x2 + y2
(23)
fory=O,l<~x<~5;x=O,
an
0 <~y <~5
The analytical solution of problem (23) is:
4
[ 2r \
V(x,y) = 1 -I log ~/3ilog~]--~r)___
(24)
1{ 1
vn
Figure 2 Errors e= a n d e 2 a g a i n s t n u m b e r o f c o m p u t e d s o l u t i o n
The errors obtained with this problem, using 36 boundary points and 43 internal points are:
e= = 0.35%
(2.68%)
e2 = 0.12%
(12.2%)
(26)
e2 = 1.8%
(27)
Conclusions
-0
f(x,y)=~ 2(l~r) 2
r(l+r
R=19.3
(25)
269
References
1 Morse, P. M. and Feshbach, H. "Methods of theoretical physics',
vols I-II, McGraw-Hill, New York, 1953
2
3
4
5