You are on page 1of 8

Cambridge Books Online

http://ebooks.cambridge.org/

Fast Multipole Boundary Element Method

Theory and Applications in Engineering

Yijun Liu

Book DOI: http://dx.doi.org/10.1017/CBO9780511605345

Online ISBN: 9780511605345

Hardback ISBN: 9780521116596

Paperback ISBN: 9781107655669

Chapter

APPENDIX A - Analytical Integration of the Kernels pp. 177-183

Chapter DOI: http://dx.doi.org/10.1017/CBO9780511605345.008

Cambridge University Press


APPENDIX A

Analytical Integration of the Kernels

A.1 2D Potential Boundary Integral Equations

For 2D potential problems, we have the following four kernels for the CBIE
and HBIE:
 
1 1
G(x, y) = log , (A.1)
2π r
∂G(x, y) 1
F(x, y) = =− r,k nk (y), (A.2)
∂n(y) 2πr
∂G(x, y) 1
K(x, y) = = r,k nk (x), (A.3)
∂n(x) 2πr
∂ 2 G(x, y) 1
H(x, y) = = [nk (x)nk (y) − 2r,k nk (x)r,l nl (y)] . (A.4)
∂n(x)∂n(y) 2πr 2
The integrations of the four kernels on a line segment S shown in Figure A.1
(from point 1 to point 2) can be evaluated analytically as follows (note that on
S, r = d/ cos θ, dS = r dθ/ cos θ ):

1
G(x, y)dS = [−(θ2 − θ1 )d + 2R − T2 log r2 + T1 log r1 ] , (A.5)
S 2π

1
F(x, y)dS = − (θ2 − θ1 ), (A.6)
S 2π
    
1 r2
K(x, y)dS = (θ2 − θ1 )nk (y) + log tk (y) nk (x), (A.7)
S 2π r1
      
1 T2 T1 1 1
H(x, y)dS = − − n k (y) + d − t k (y) nk (x),
S 2π r2 2 r1 2 r2 2 r1 2
(A.8)

in which 2R(= T2 − T1 ) is the total length of the line element and tk is the
component of the tangential vector t (Figure A.1). These results can be used
to evaluate directly the coefficients of the CBIE and HBIE for 2D potential

177

Cambridge Books Online © Cambridge University Press, 2010


Downloaded from Cambridge Books Online by IP 192.17.220.90 on Wed Jul 22 01:02:31 BST 2015.
http://dx.doi.org/10.1017/CBO9780511605345.008
Cambridge Books Online © Cambridge University Press, 2015
178 Appendix A: Analytical Integration of the Kernels

r2 t T2
n(y)
y
r ΔS

r1 1 T1

θ
θ2
θ1 d
2
x
n(x)

1
Figure A.1. Analytical integration on an arbitrary line segment.

problems using constant elements. If the source point x is on the element of


integration (at the midpoint), we have:

θ2 − θ1 = π, d = 0, r1 = r2 = R, T1 = −T2 = −R,

and the four integrals have the following values:



R
G(x, y)dS = (1 − log R) , (A.9)
S π

1
F(x, y)dS = − , (A.10)
S 2

1
K(x, y)dS = , (A.11)
S 2

1
H(x, y)dS = − . (A.12)
S π R

Note that in the preceding results, the second (F) and third (K) integrals are
equal to the CPV integrals plus the jump terms, and the last (H) integral is a
HFP integral.

A.2 2D Elastostatic Boundary Integral Equations

For 2D elasticity, we have the following four kernels for the CBIE and
HBIE:
   
1 1 1
Ui j (x, y) = (3 − 4ν)δi j log + r,i r, j − δi j , (A.13)
8π μ(1 − ν) r 2

Cambridge Books Online © Cambridge University Press, 2010


Downloaded from Cambridge Books Online by IP 192.17.220.90 on Wed Jul 22 01:02:31 BST 2015.
http://dx.doi.org/10.1017/CBO9780511605345.008
Cambridge Books Online © Cambridge University Press, 2015
Appendix A: Analytical Integration of the Kernels 179

 
1 ∂r
Ti j (x, y) = − [(1 − 2ν)δi j + 2r,i r, j ] − (1 − 2ν) (r,i n j − r, j ni ) ,
4π (1 − ν)r ∂n
(A.14)
1
Ki j (x, y) = [(1 − 2ν)(δi j r,k + δ jk r,i −δik r, j ) + 2r,i r, j r,k ] nk (x),
4π (1 − ν)r
(A.15)

μ ∂r
Hi j (x, y) = 2 [(1 − 2ν)δik r, j + ν(δi j r,k + δ jk r,i ) − 4r,i r, j r,k ]
2π (1 − ν)r 2 ∂n

+ 2ν(ni r, j r,k +nk r,i r, j ) − (1 − 4ν)δik n j



+ (1 − 2ν) (2n j r,i r,k +δi j nk + δ jk ni ) nk (x). (A.16)

To evaluate the integrals of these kernels over the straight-line segment S (a


constant element) shown in Figure A.1, we use the local coordinate system n–t
at y on S. In this local coordinate system, we have:


(n−t) 1 (n−t)
Ui j (x, y)dS = (3 − 4ν)I0 δi j + Ii j − Rδi j , (A.17)
S 8π μ(1 − ν)

(n−t) 1 (n−t) (n−t)
Ti j (x, y)dS = − (1 − 2ν) 0 δi j − i δ1 j +  j δ1i
S 4π(1 − ν)

(n−t)
+ 2i j , (A.18)

(n−t)
Ki j (x, y)dS = Ci jk nk (x), (A.19)
S

(n−t)
Hi j (x, y)dS = Di jk nk (x), (A.20)
S

where:
(n−t) (n−t)
4π(1 − ν)C111 = (1 − 2ν)1 + 211 ,
(n−t) (n−t)
4π(1 − ν)C112 = (1 − 2ν)2 + 212 ,
(n−t) (n−t)
4π(1 − ν)C121 = −(1 − 2ν)2 + 212 ,
(n−t) (n−t)
4π(1 − ν)C122 = (1 − 2ν)1 + 222 , (A.21)
(n−t) (n−t)
4π (1 − ν)C212 = −(1 − 2ν)1 + 222 ,
(n−t) (n−t)
4π(1 − ν)C222 = (1 − 2ν)2 + 2222 ,

C211 = C112 , C221 = C122 ;

Cambridge Books Online © Cambridge University Press, 2010


Downloaded from Cambridge Books Online by IP 192.17.220.90 on Wed Jul 22 01:02:31 BST 2015.
http://dx.doi.org/10.1017/CBO9780511605345.008
Cambridge Books Online © Cambridge University Press, 2015
180 Appendix A: Analytical Integration of the Kernels

2π(1 − ν)
D111 = − 12 − 2d2 14 ,
μ

2π(1 − ν)
D112 = −d 02 + 2d3 04 ,
μ

2π(1 − ν)
D122 = − 12 + 2d2 14 , (A.22)
μ

2π(1 − ν)
D222 = 3d 02 − 2d3 04 ,
μ
D211 = D121 = D112 , D212 = D221 = D122 .

In the preceding expressions:

I0 = −d(θ2 − θ1 ) + 2R − T2 log r2 + T1 log r1 ,


(n−t)
I11 = d(θ2 − θ1 ),
(A.23)
(n−t) (n−t)
I12 = I21 = d log(r2 /r1 ),
(n−t)
I22 = T2 − T1 − d(θ2 − θ1 );

0 = θ2 − θ1 ,

(n−t)
1 = θ2 − θ1 ,

(n−t)
2 = log(r2 /r1 ),

(n−t)
11 = (θ2 − θ1 )/2 + d 12 /2, (A.24)

(n−t) (n−t)
12 = 21 = 22 /2,

(n−t)
22 = (θ2 − θ1 )/2 − d 12 /2,

(n−t)
222 = log(r2 /r1 ) + d2 02 /2;

02 = 1/r22 − 1/r12 ,

04 = 1/r24 − 1/r14 ,

12 = T2 /r22 − T1 /r12 , (A.25)

14 = T2 /r24 − T1 /r14 ,

22 = T22 /r22 − T12 /r12 ;

Cambridge Books Online © Cambridge University Press, 2010


Downloaded from Cambridge Books Online by IP 192.17.220.90 on Wed Jul 22 01:02:31 BST 2015.
http://dx.doi.org/10.1017/CBO9780511605345.008
Cambridge Books Online © Cambridge University Press, 2015
Appendix A: Analytical Integration of the Kernels 181

in which all the parameters are as defined in Figure A.1. Once the integrals of
the kernels are determined in the local n–t system, they need to be transformed
to the global x–y system.
When the source point x is on the element of integration, we have:

θ2 − θ1 = π, d = 0, r1 = r2 = R, T1 = −T2 = −R,

and the four integrals have the following results:



(n−t) R
U11 (x, y)dS = [2(3 − 4ν)(1 − log R) − 1] ,
S 8π μ(1 − ν)
 
(n−t) (n−t)
U12 (x, y)dS = U21 (x, y)dS = 0, (A.26)
S S

(n−t) R
U22 (x, y)dS = [2(3 − 4ν)(1 − log R) + 1] ,
S 8π μ(1 − ν)

(n−t) 1
Ti j (x, y)dS = − δi j , (A.27)
S 2

(n−t) 1
Ki j (x, y)dS = δi j , (A.28)
S 2

(n−t) μ
Hi j (x, y)dS = − δi j . (A.29)
S π(1 − ν)R
Similar to the potential case, the second (T) and third (K) integrals are equal
to the CPV integrals plus the jump terms, and the last (H) integral is a HFP
integral.

A.3 2D Stokes Flow Boundary Integral Equations

For 2D Stokes flow problems, we have the following four kernels for the CBIE
and the HBIE:
   
1 1 1
Ui j (x, y) = δi j log + r,i r, j − δi j , (A.30)
4π μ r 2
1
Ti j (x, y) = − r,i r, j r,k nk (y), (A.31)
πr
1
Ki j (x, y) = r,i r, j r,k nk (x), (A.32)
πr
μ
Hi j (x, y) = [(δi j r,k + δ jk r,i − 8r,i r, j r,k ) r,l nl (y)
πr 2
+ ni r, j r,k + nk r,i r, j + δik n j ] nk (x). (A.33)

The integrals of these kernels over the straight-line segment S shown in


Figure A.1 can be obtained from the results for 2D elasticity problems by

Cambridge Books Online © Cambridge University Press, 2010


Downloaded from Cambridge Books Online by IP 192.17.220.90 on Wed Jul 22 01:02:31 BST 2015.
http://dx.doi.org/10.1017/CBO9780511605345.008
Cambridge Books Online © Cambridge University Press, 2015
182 Appendix A: Analytical Integration of the Kernels

setting Poisson’s ratio ν = 12 . In the local coordinate system, we obtain (see


results in the previous section):

(n−t) 1 (n−t)

Ui j (x, y)dS = I0 δi j + Ii j − Rδi j , (A.34)


S 4π μ

(n−t) 1 (n−t)
Ti j (x, y)dS = − i j , (A.35)
S π

(n−t)
Ki j (x, y)dS = Ci jk nk (x), (A.36)
S

(n−t)
Hi j (x, y)dS = Di jk nk (x), (A.37)
S

where:

(n−t)
2πC111 = 211 ,
(n−t)
2πC112 = 212 ,
(n−t)
2πC122 = 222 , (A.38)
(n−t)
2πC222 = 2222 ,

C121 = C211 = C112 , C212 = C221 = C122 ,

π
D111 = − 12 − 2d2 14 ,
μ
π
D112 = −d 02 + 2d3 04 ,
μ
π
D122 = − 12 + 2d2 14 , (A.39)
μ
π
D222 = 3d 02 − 2d3 04 ,
μ

D211 = D121 = D112 , D212 = D221 = D122 ,

and all the parameters I, , and are as defined earlier for elasticity ker-
nels in Eqs. (A.23)–(A.25). Once the integrals of the kernels are deter-
mined in the local n–t system, they need to be transformed to the global x–y
system.
When the source point x is on the element of integration, we have the
following results for the four integrals:

Cambridge Books Online © Cambridge University Press, 2010


Downloaded from Cambridge Books Online by IP 192.17.220.90 on Wed Jul 22 01:02:31 BST 2015.
http://dx.doi.org/10.1017/CBO9780511605345.008
Cambridge Books Online © Cambridge University Press, 2015
Appendix A: Analytical Integration of the Kernels 183


(n−t) R
U11 (x, y)dS =
[2(1 − log R) − 1] ,
S 4πμ
 
(n−t) (n−t)
U12 (x, y)dS = U21 (x, y)dS = 0, (A.40)
S S

(n−t) R
U22 (x, y)dS = [2(1 − log R) + 1] ,
S 4πμ

(n−t) 1
Ti j (x, y)dS = − δi j , (A.41)
S 2

(n−t) 1
Ki j (x, y)dS = δi j , (A.42)
S 2

(n−t) 2μ
Hi j (x, y)dS = − δi j . (A.43)
S πR
Similar to the potential and elasticity cases, the second (T) and third (K) inte-
grals are equal to the CPV integrals plus the jump terms, whereas the last (H)
integral is a HFP integral.

Cambridge Books Online © Cambridge University Press, 2010


Downloaded from Cambridge Books Online by IP 192.17.220.90 on Wed Jul 22 01:02:31 BST 2015.
http://dx.doi.org/10.1017/CBO9780511605345.008
Cambridge Books Online © Cambridge University Press, 2015

You might also like