Two-dimensional boundary element kernel

integration using series expansions

M. H. Aiiabadi

Computational Mechanics Institute, Ashurst Lodge, Ashurst, Southampton, S04 2AA, UK

W. S. Hall

Department of Mathematics and Statistics, Teesside Polytechnic, Middlesbrough, Cleveland, UK

Series expansions are applied to integrations arising in the two dimensional Boundary Element
Method. It is shown that, for the derivative kernel, exact integrals can be obtained for quadratic
and higher order shape functions. It is proved incidentally that the derivative kernel contains no
singularity.
For the logarithmic kernel the series expansions are truncated and used to integrate the first
singular term and some early, badly behaved terms, with the well behaved remainder being
integrated numerically. A test example shows how term by term subtraction leads quickly to an
accurate solution.

Key Words: boundary element method, series expansion integration, two dimensional potential
problems, logarithmic kernel, derivative kernel

INTRODUCTION
f a M(4)J(~) n. r d4 (7)
The two dimensional boundary element method1 requires - 1 r2

the evaluation of the following integrals where J(~) is the Jacobian of transformation. Suffix i on
the shape function has been dropped for convenience.
fAs q ~ ( x ) l o g ( ! ) a s ( x ) (1)
SERIES EXPANSION TO EVALUATE THE
SINGULAR INTEGRALS
The expansion method of Aliabadi, Hall and Phemister 3
where q~(x) is some unknown function which will be for three dimensional problems may be used here in
represented over an element by shape functions, x is a simplified form to express the vector r from a variable
point on the boundary, x o is a general point which may point x to a fixed point x o as a series expansion in the
coincide with x, AS is an element in which x and x o variable 4 arising from the shape functions of the
coincide and r is the distance from x and x o given by boundary representation (5). The expansion is about ~o,
which corresponds to x o. Thus
r = Ix - Xo[ (3)
In the isoparametric formulation of boundary element r=x(~)-X(4o)
method 2, the function and boundary may be represented = x'6~ + ½x"642 (8)
using quadratic shape functions as
where 64 = ¢ - ~ o and the derivative coefficients x' and
3
x" are evaluated at 4o. The expansion is limited because
q~= ~ M'(4)c]J' (4) the derivatives are obtained by differentiating expression
i=1
(5) so that, for example,
and
3
3 ~"'= y~ M"'(4)x'
x= ~ g'(4)x' (5) i=1
i=1
where M~= ~ ( 4 - 1)/2, M 2 = 1 - ~2, M a = 4( 1 + 4)/2. =0 since M"i = 0 i = I, 2, 3
Thus integrals (1) and (2) will be made up from integrals
Thus, from expression (8), the real plane distance
of the form
r=l_x-xol may be found as
~ 1 M(4)J(4)log(r)d~
-1
(6) r 2 -----d o ~ 2 + d l ~

where
3 + d26~ 4 (9)

Paper accepted May 1989. Discussion ends February 1990. do =x'.x_', d I =x'.x", d2=x".x"/4 (10)

140 Emtineering Analysis with Boundary Elements, 1989, Vol. 6, No. 3 ,(,i 1989 Computational Mechanics Publications
 Two-dimensional boundary element kernel integration using series expansions: M. H. Aliabadi and W. S. Hall

Similarly M(~) may be expanded in a limited form as where

M (C)= rho + m~b~ + m26~ 2 (11) f o = moh 1. x' + ½moh o . x"
where mo, m 1, m2 are evaluated at ~o. f l = ½moh1. x" + mlh 1 . x' + ½mlho . x"
Since
f2 = m2hl. x' + ½mzho . x" (21)
J = Ix' x k I (12) Although a term of order 1/6C would be expected in this
and expansion for the integrand it has disappeared. Its coefficient
is mob o . x' and from equation (16), h o is in the direction of
x'xk the normal to the boundary at Co and x' is the direction of
n= - - (13)
Ix' x k I the tangent to the boundary at Co so that moho . x ' = 0 .
the combination J(~)n appearing in integral (7) is given by Equation (20) also contains a simplification which has
removed the term in 6C3. Its coefficient ½mzh1 . x" is zero
J(¢)n = x ' × k (14) since h a = x " x k is a vector perpendicular to x" making
where the out of plane unit vector k has been introduced. h 1 . x " = 0.
Now expanding x'(¢) as a Taylor expansion gives The integrals coming from the expression (20) are all of
the form
x'(¢) = x' + x " 6 C (15)
with x', x" evaluated at 40, which is again limited since
x"' =0. Thus
L. =

where
f5 d~ n = 0, 1, 2, 3 (22)

J(~)n = x' x k + x" x k6~

A = (do + d~6~ + d2~C2)
=ho+h16 ~ (16)
which can be found from the recurrence formula
where
ho=x'xk, hl=X"xk (17) t~n-1 ~22Ln_1 do
L. - (n - 1)d 2 -d22 L._ 2 (23)
The Jacobian term J(C) when taken by itself, as it
appears in integral (6), cannot be expanded in limited 1 dl L
form because it is defined from equation (12) by L1 =~d~2 log A - ~ 2 o (24)

j2 = (x' x k). (x' x k) 2 - 2d26~+d1 (25)

-- (x'. x')(k, k) - (x'. k)(k. x') L o = [4dod2 _ d2]~ tan 1 [4dod2 _ d2] ½

by Lagrange's identity. Thus since k and x' are Thus the integral (7), involving the derivative kernel
perpendicular and k. k = 1 can be evaluated in closed form and no numerical
integration is needed. Although this analysis has been
J2(C)=x'.x' carried out for quadratic shape functions it will clearly
Substituting the limited expansion (15) for x'(C) gives extend to higher order shape functions and then will
involve larger values of n in expressions (22) and (23).
j2(~) = [x'+ x"6~]. [x' + x"6C] It may be noted from equation (9) that
=x'.x' + 2x'.x"b~ + x".x"6C z r 2 = 6~2(d0 + dlb~ + d26C 2)
= d o + 2dlb ~ + 4d26~ 2 is a positive quantity and thus the expression
and A - - d o + d16~-1- dzt~ 2
J(C) = (do + 2dt6~ +4d26~2) 1/2 (18) can have no roots. Hence
which thus uses the same constants do, dl, d2 as r 2 in 4dod 2 - d 2 >~0
equation (9). A series expansion of expression (18) for J(¢)
is given by and the square roots of expression (25) give real values.
If shape functions of higher order (n>~3) are used, A
becomes more complicated and equation (23) does not
J(~)=d~/2 ( l +2d13c +4ff~2o apply directly. In these higher order cases, A is a
polynomial of degree 2 ( n - 1 ) which may be factorized
__dol/2[1+~3~+(2~ 1d2'~ o~2 +...] and A -1 expressed as partial fractions. Equation (23)
would apply to each of the partial fractions.

INTEGRATION OF THE LOGARITHMIC KERNEL

EXACT INTEGRATION OF THE DERIVATIVE
KERNEL Substituting expressions (9), (11) and (18) into integral (6)
gives
Substituting expressions (8), (9), (11) and (16) into the
integrand of integral (7) gives MJ log r = (mo + ml6 C+ m23~ 2)
x (do + 2dlb C+ 4d2t~C2) 1/2
M(~)J(~) -~f = (fo +flaC +f26~ 2) (20)
r- (d o + dl3 ~ + d26C 2) x (log 6C+½1og(do+dlaC+d2bC2)) (26)

Engineering Analysis with Boundary Elements, 1989, Vol. 6, No. 3 141

Two-dimensional boundary element kernel integration using series expansions: M. H. Aliabadi and W. S. Hall

Using the expansion (19) of the square root term and

the expansion of the logarithmic term gives an expansion + f ~ t T#° d~a~al--f ~, Tq° d~num
M J log r = ago log 6~ + aol~ ~ log ~ + ag2t~ 2 log ~
+ 7"ol d~anal- To1 dCnum
+ao 1 -1
+aa6~+... ..~-...
= Too + To1 + Tog + T o + T 1 + . . . (27) =Do+D a+D 2+.-. (29)
where The term D o = S l l (MJlogr)d{.um is the straight-
aoo=d~/Zm o forward numerical integration of the singular integrand
with no help from the subtraction process. This is
improved stage by stage by difference terms such as
aox = ~/2 mo + d~/2ml
ao
1"2 dx ( d 2 1 d21~ D 1 = f 1 Too d~anal- f 1 Too d~num
-1 -1
ao2=do / m 2 + d - - ~ m l W . 2 d o 2d2oJ m°
Such difference terms should decrease in magnitude as the
a o = ½domo log d o correct evaluation of the integral is approached.
a 1 = (dim o log d o + dom 1 log d o + dlmo)/2d~/2
PERFECT SQUARE INTEGRATION TEST
In expansion (27) it can be seen that only the first term
contains the singularity log 6~ which tends to - ~ as It is possible to test integral (6) for a special case in which
fig ~ 0. The second and third terms may not be very well it can be integrated exactly. Such a special case is obtained
behaved, but all further terms will be straightforward to taking a particular combination ofd o, dl, d2 which makes
integrate. the expression (18) for J({) a perfect square.
The truncation of expansion (27) is denoted by If
( M J l o g r ) , and is used in the subtraction method to d~-4dod2 = 0 (30)
produce accurate evaluations of integral (6). Thus
consider then
J(~) = dl/2 q_2d1/Z6~
J'~lMJlogrd~=f ~ ( M J l o g r ) t d ~ and also, incidentally, the distance r given by equation (9)

+ fl
-1
[(MJ log r ) - (MJ log r)J d~
reduces to
r(~) = a~(d~/a + d~/26~)
(28) Taking first a simple case in which no shape functions
in which the second term can be integrated exactly. The M appear, i.e. too= 1, m 1 =0, mz=O the integrand (26)
third term can be integrated accurately using standard becomes
numerical integration procedures such as Gauss M J log r = (d~/2 + 2d~/26{){log 6~ + log(do1/2 + d~/26¢)}
integration, since it becomes non-singular even if only one
term of expansion (27) is used and increasingly (31)
well-behaved as more terms are included. which is straightforward to integrate exactly. For values
It is important to note that expansion (27) is not a series d o = 1, d I = 2, d 2 = 1 which satisfy equation (30), Table 1
approximation for M J l n r over the whole of the gives the appropriate terms in series (29).
integration interval [ - 1, 1] since it is only in the neigh- After D 2 has been added the exact and approximate
bourhood of one point, Xo, that 6~ is small and the integrations agree to 6 decimal places.
integrand could reach a maximum value of 2. The usefulness
of expansion (27) is that it allows the singularities and bad
behaviour of the integrand to be subtracted out so that CONCLUSIONS
accurate integration can be achieved for the remainder The series expansion method has been applied to integrals
integral. Nevertheless it is informative to see how arising in the two-dimensional Boundary Element
additional terms of the expansion provide corrections to Method. For the derivative kernel n. r/r 2, exact closed
the integration. This is done by rewriting the subtraction integrations can be carried out for quadratic and higher
process in equation (28) term by term as

fl (MJlogr)d~l Table 1. Magnitude of correction terms

Term Comment Value
=f1_1 (Tg° + Tgl +" " ") d~a"a'
Exact integral 4.7505568
DO Unaided numerical integration 4.7676876
D1 First correction term -0.0175219
+ [(MJ log r) - (Tgo + TgI + . . . ) ] d~.. m D2 Second correction term 0.0003911
-1
D3 N o t tested since ag2=0 0
D4 Constant term 0
= (MJ log r) d~..m D5 Linear term 6 x 10-14
-1

142 Engineering Analysis with Boundary Elements, 1989, Vol. 6, No. 3

 Two-dimensional boundary element kernel integration using series expansions: M. H. Aliabadi and W. S. Hall

order shape functions. Incidentally it is shown that the Khiabany for providing the perfect square integration test
derivative kernel has no singularity. and C. Flood and E. Sinclair for carrying out associated
When applied to the logarithmic kernel the first, calculations.
singular term of the integrand expansion and other,
possibly badly behaved, early terms are integrated
exactly, although the integrand expansion must be
truncated and the remainder integrated by standard REFERENCES
numerical methods. A test example shows how successive 1 Brebbia, C. A. The Boundary Element Method for Engineers,
terms in a term by term subtraction process decrease Pentech Press, 1978
quickly in magnitude. 2 Lachat,J. C. and Watson, J. O. EffectiveNumerical Treatment of
Boundary Integral Equations: a Formulation for Elastatics, Int. J.
Numer. Methods Eng., 1976, 10, 991-1005
ACKNOWLEDGEMENTS 3 Aliabadi,M. H., Hall, W. S. and Phemister,T. G. Taylor expansions
for singular kernelsin the BoundaryElementMethod, Int. J. Numer.
The authors would like to thank M. Nourbahksh- Methods Eng., 1985, 21, 2221-2236

Engineering Analysis with Boundary Elements, 1989, Vol. 6, No. 3 143