Theoratical Basis for Boundary Element Method

Dhananjay Ghangale
LEAM-UPC Terrassa
November 1, 2018

Abstract
The report present the theoretical basis for Boundary Element Methods
for 3D elastodynamics.

1 The Integral Representation

The Poisson’s equation given by

4u + b = 0 (1)

(4 is the Laplace operator, u unknown field var able also called potential and
b the given source)
defined for all y ∈ Ω; Ω ∈ R3 ( bounded domain). The unknown u must obey at
any point of boundary dΩ a given boundary condition. The boundary condition
could be one where flux q(y) on whole boundary is given (Neuman boundary
condition)

q(y) = 5u(y) ∗ n(y) = q D (y) (2)

or could be one where potential is known through the whole boundary (Dirichlet
boundary condition)

u(y) = uD and (y ∈ dΩ) (3)

Also mixed boundary conditions are considered sometimes i.e. flux know over
some subset Sq of the boundary dΩ and potential over complimentary subset Su =
dΩ
Sq
.
Consider a bonded domain Ω ⊂ R3 in which the boundary dΩ is assumed to
be piecewise continuously differentiable and may contain finite number of edges or
corners. The unit normal n to dΩ is oriented outward from the domain Ω.

1
 A boundary value problem defined on a geometrical domain Ω ⊂ Rn (n = 1, 2, 3)
having an unknown variable u and associated with second order partial differential
operation L having a structure:

Lu + f = 0 in Ω (4)
u = g1 on Su (5)
n
T (u) = g2 on ST (6)
where g1, g2 and source f are known and T n represent first order partial differ-
ential operator linear with respect to n. Su and ST define partition of the boundary
such that Su ∪ ST = dΩ and Su ∩ ST = φ . The Operator L and T satisfy the
following property:
Z Z
(L(u) ∗ v − L(v) ∗ u)dV = (T n (u) ∗ v − T n (v) ∗ u)dS (7)
Ω dΩ
The Laplace operator 4 also satisfies the property given by eq 7
The Poisson equation with unknown u1
4u1 + b = 0 (8)
multiplying it with function u2

4u1 ∗ u2 + b ∗ u2 = 0 (9)
Integrating the result over Ω by using integration by parts and making use of
the identity eq 7

Z Z
1 2
(4u ∗ u )dV + (b ∗ u2 )dV = 0
Z Ω ZΩ
(4u1 ∗ u2 − 4u2 ∗ u1 )dV + (b ∗ u2 + 4u2 ∗ u1 )dV = 0
Z Ω Z Ω Z

(q 1 ∗ u2 )dS − (T n (u2 ∗)u1 )dS + (b ∗ u2 + 4u2 ∗ u1 )dV = 0 (10)

Z dΩ Z dΩ Z Ω Z
1 2 n 2 1 2
(q ∗ u )dS − (T (u ∗)u )dS + (b ∗ u )dV + (T n (u2 ∗)u1 )dS
dΩ dΩ Z Ω dΩ

− (5u2 ∗ 5u1 )dS = 0

Ω

We get the reciprocity identity

Z Z
q u dS + bu2 − 5u1 ∗ 5u2 dV = 0
1 2
(11)
dΩ ω

2
2 Fundamental Solution
Let E denote an open subset R3 . A fundamental solution is any solution in E
to Poisson equation with b = δ(y − x) (unit point source) applied at fixed point
x ∈ E denoted G(x, y) which is true for the equation

G,jj (x, y) + δ(y − x) = 0 (12)

The fundamental solution is a function of field point y and a source point also
called fixed support or collocation point x which is treated as a parameter.
Let the flux at y in direction n normal to boundary dΩ associated with G(x, y)
be denoted by H(x, y) then

H(x, y) = G,j (x, y) ∗ nj (y) (13)

For global equilibrium given a domain Ω ⊂ E and source point in x ⊂ E and
x∈/ dΩ integrating
RRR the fundamental
RR solution over the domain Ω using divergence
theorem (5 ∗ F )dV = (F ∗ n)dS

G,jj (x, y) − δ(y − x) = 0 and y∈E (14)

Z Z
G,jj (x, y)dV + δ(y − x)dV = 0 (15)
Ω Ω
Z
G,j (x, y) ∗ nj (y)dSy + k = 0 (16)
dΩ
we get the following important identity
Z
k+ H(x, y)(x, y)dSy = 0 (17)
dΩ

with k = 1 for all x ∈ Ω and k = 0 x ∈ / Ω. The equation implies that flux

across dΩ is in equilibrium with unit point source applied at x

3 The Boundary Integral equation for point in-

side or outside domain
Now the reciprocity identity is applied to an unknown state (u, q, b) and a known
fundamental solution G(x, y) the point x being interior to the domain Ω or in
R3 − Ω leads to

3
 Z Z Z
δ(y−x)u(y)dV y = q(y)G(x, y)−u(y)H(x, y)dSy+ b(y)G(x, y)dV y (18)
Ω dΩ Ω

Using the definition of Dirac-delta function

(
k=1 x∈Ω
δ(y − x)u(y)dV y = ku(x) and (19)
k=0 x∈
/Ω

which implies
Z Z
ku(x) = q(y)G(x, y) − u(y)H(x, y)dSy + b(y)G(x, y)dV y (20)
dΩ Ω

The equation gives the value of u(x) of a solution to the Poisson equation in
terms of source b and values of (u, q) on the boundary.

4 Boundary Integral equation for point on Bound-

ary
To find the value of u(x) in the domain values of (u, q) on the boundary must
be known however in many situation only either of them are known. So the task
is to define a boundary integral equation for values of (u, q) on boundary. The
BIE described in previous section cannot be used for points x ∈ dΩ as the kernel
functions G(x, y) and H(x, y) are singular so the integral over dΩ is non-convergent.

Z Z
0= b(y)G(x, y)dV y + q(y)G(x, y) − u(y)H(x, y)dSy (21)
Ω (Ω −e )−s

The static identity eq 17 takes the form

Z
u(x) H(x, y)dSy = 0 (22)
(Ω −e )−s

subtracting eq 21 from eq 22 we get

Z Z
0= b(y)G(x, y)dV y + q(y)G(x, y) − [u(y) − u(x)]H(x, y)dSy (23)
dΩ (Ω −e )−s

assuming u is C 0,α continuous at x

∃(α, c) < 0, |u(y) − u(x)|≤ C|y − x|α (24)

4
 then the normal derivative is such that

∃(α, c) < 0, |q(y)|≤ C|y − x|α−1 (25)

The singularity properties imply

|[u(y) − u(x)]H(x, y)|≤ C|y − x|α−2 (26)

|[q(y)G(x, y)|≤ C|y − x|α−2 (27)

The integral over s and dΩ − e are analysed separately

4.1 Integral over s

from the singularity considerations and |y − x|  and dSy 2 dω (ω is solid angle
seen from point x) we have:

[u(y) − u(x)]H(x, y) − q)(y)G(x, y)dSy = O(α )dω (28)

So, Z
lim [u(y) − u(x)]H(x, y) − q)(y)G(x, y)dSy = 0 (29)
→0 s

4.2 Integral over dΩ − e

From singularity consideration the integrand behaves near x like

[u(y) − u(x)]H(x, y) − q)(y)G(x, y)dSy |y − x|α−2 (30)

the limit when  → 0 of integral over dΩ − e is simply the integral over dΩ
which is convergent in ordinary sense

Z Z
lim [u(y) − u(x)]H(x, y) − q(y)G(x, y)dSy = [u(y) − u(x)]H(x, y) − q(y)G(x, y)dSy =
→0 (dΩ −e )−s dΩ
(31)
Gathering previous results leads to the required boundary integral equation
which relates the values of (u, q) on the boundary surface dΩ only and is given by
Z Z
[u(y) − u(x)]H(x, y) − q(y)G(x, y)dSy = b(y)G(x, y)dV y (32)
dΩ dΩ

5
5 Discretisation
The discretisation scheme consist of collocation method which enforces the integral
equation at finitely many collocation points xc . After the mesh is generated the
discritisation is performed by inserting element wise representations:

Ne Z
X
[u(y) − u(xc )]H(xc , y) − q(y)G(xc , y)dSy = 0 (1 ≤ c < Nc ) (33)
e=1 Ee

The above integral has both singular and non-singular integration. For x ∈ dΩ
define a subset I(x) such that an integral over element Ee is singular for e ∈ I(x)
and non singular for e ∈
/ I(x)
The decritised boundary integral has the form

X NX
I (e)
X
AR(e, p)um(e,p) − B(e, p)um(e,p) − u(xc ) AR0 (e)
e ∈I(x)
/ p =1 e ∈I(x)
/
(34)
NI (e)
X X
+ AS(e, p)um(e,p) − B(e, p)um(e,p)
e ∈I(x) p =1

where Z
AR(e, p) = Mp (ξ)H(xc , y)dSy (35)
Ee
Z
B(e, p) = Mp (ξ)G(xc , y)dSy (36)
Ee
Z
0
AR (e, p) = H(xc , y)dSy (37)
Ee
Z
AS(e, p) = [Mp (ξ) − Mp (η)]H(xc , y)dSy (38)
Ee

6 The Boundary Integral equation for elastody-

namics
In frequency domain, the elastodynamics state given by (u, σ, F ) takes the follow-
ing form

divσ + ρω 2 u + ρF = 0 (39)

6
 The displacement u(x) at a given point x ∈ E not located on dΩ is given by
the integral representation formula:
Z Z
k k
kuk (x) ti (y)Ui (x, y, ω) − ui (y)Ti (x, y, ω)dSy = ρFi (y)Uik (x, y, ω) (40)
dΩ Ω

The derivation of this integral representation follows the same procedure as that
given in section 3. These formula can be employed in conjunction with any fun-
damental solution provided that ω ∈ E
Because the singular behaviour of fundamental solution at y = x the repre-
sentation formula cannot be directly used for x ∈ dΩ so the BIE for x ∈ dΩ is
established by limiting process similar to one in statics.
Z
ui (y)Tik (x, y, ω) − ti (y)Uik (x, y)dSy = 0 (41)
dΩ−e −s

Using the static identity 17 and treating u(x) as a scalar constant:

Z
u(x) Tik (x, y)dy = 0 (42)
dΩ−e −s

The equation is reformulated as

Z Z
T (x, y, ω)ui (y) − T (x, y) ∗ ui (x)dSy− Uik (x, y, ω) ∗ ti (y)dSy = 0
dΩ−e −s dΩ−e −s
(43)
The singular term can be written as

T (x, y, ω)∗u(y)−T (x, y)∗u(x) = T (x, y, ω)∗u(y)−T (x, y)∗u(y)+[u(y)−u(x)]∗T (x, y)

(44)
The difference between corresponding dynamic and static function is non sin-
gular. Thus the limiting process leads to weakly singular equation given by
Z
ui (y)Tik (x, y, ω) − ui (x)Tik (x, y) − ti (y)Uik (x, y)dSy = 0 (45)
dΩ

for unbounded domain

Z
ui (x) = ui (y)Tik (x, y, ω) − ui (x)Tik (x, y) − ti (y)Uik (x, y)dSy (46)
dΩ

7 Discretisation of 3D BEM equation

The boundary integral equation is established by taking point (x0 , y0 , z0 ) on bound-
ary dS leads to

7
 Z
0 0 0
u(x , y , z ) = U (x0 , y 0 z 0 , x, y, z, ω)t(x, y, z, ω)dS
dS
(47)
Z
− [T (x0 , y 0 z 0 , x, y, z, ω)t(x, y, z, ω) − T (x0 , y 0 , z 0 , x, y, z, ω
dS
= 0)u(x0 , y 0 , z 0 , ω)]dS

In this section discretization of the above equation is proposed based on Let

uj = [u1x , u1y , u1z , u2x , u2y , u2z , u3x , u3y , u3z , u4x , u4y , u4z ]T (48)

tj = [t1x , t1y , t1z , t2x , t2y , t2z , t3x , t3y , t3z , t4x , t4y , t4z ]T (49)
be two 12 ∗ 1 vectors of nodal displacement and traction for 4 nodes in element j
and let

x, y j = [x1 , y1 , x2 , y2 , x3 , y3 , x4 , y4 ]T (50)
be a 8 ∗ 1 vector consisting of coordinates of element nodes. Then displacement,
traction and coordinates along the boundary element is approximated as
8
X
uj = Ni uij (51)
i=1

where j = (x, y, z) and Ni are the shape functions similarly the traction are
written as
X8
tj = Ni tij (52)
i=1

8
X
x, y j = Ni ∗ (x, y)ij (53)
i=1

inserting equation and letting point (x0 , y0 ) be on node i yields

NE
X NE
i T
 j X
[U (ξ, η)]Tij ∗ [N (ξ, η)]dS tj
  
u + [T (ξ, η)]ij ∗ [N (ξ, η)]dS u = (54)
j=1 j=1

where, ui is displacement vector of node i, and [T (ξ, η)]ij and [U (ξ, η)]ij are
traction and displacement Greens function matrices with source at node i and
observer located by eq 53
For a system with N nodes the equation 54 takes the form

8
    1
u1x tx
 u1   t1 
 y  y
 u1   t1 
 z  z
 ..   .. 
 .  .
 .  .
 .  .
 .  .
ui + [[T ]i1 [T ]i2 ......[T ]iN ]  .  = [[U ]i1 [U ]i2 ......[U ]iN ]  .  (55)
 ..   .. 
   
 ..   .. 
 .  .
 N  N
u x  tx 
 N  N
u y  ty 
N
uz tNz

where i = 1, 2, ..., N . [T ]ij and [U ]ij are element matrices which are 3 ∗ 3 are
assembled in finite element solution process

8 The algorithm
The objective is to compute the stiffness matrix of BEM to be used in the FEM-
BEM equation
[K + K̄s − ω 2 M]Ū = F̄ , (56)
where K and M are the stiffness matrix and mass matrix of FEM and K̄s is the
frequency dependent stiffness matrix of the soil obtained with BEM.
K̄s is obtained by
K̄s = Tm Ū−1
b [T̄b + I], (57)
where Tm is a transformation matrix that transforms the tractions on the
boundary
R T to forces and is only dependent on the shape function and Tm =
N N dS. Ub and Tb are matrices of traction and displacement obtained by
BEM on the boundary which can be obtained by following the discritisation scheme
mentioned in section 5,7