_
V
G
k
P; Qf
k
QdV: 2
Here, G
k
and H
k
represent rst and second fundamental solutions;
u
k
, t
k
and f
k
are displacement, traction and body force components,
respectively, in Fourier transform space. P and Q are xed (source)
and varied (integration) points, respectively. c
k
is a constant
depending on the location of P. The indicial notation is used for
writing Eqs. (1) and (2) and it is assumed that the indices appearing
in these equations have the range from 1 to 2 for two-dimensional
analysis. Repeated index implies summation over the range of that
index.
Boundary integral equation, Eq. (2), can be rewritten in matrix
form as
cuP
_
S
GP; QtQdS
_
S
HP; QuQdS
_
V
GP; Qf QdV: 3
In Eq. (3), c is a matrix which is
c
I if P is an interior point;
0 if P is outside of V;
1
2
I if P is on the boundary S;
_
_
4
where I is identity matrix. It may be noted that the last expression
above does not hold when P is a corner point [3]. The fundamental
solutions G and H of elastodynamics have been presented in the lit-
erature by taking the auxiliary system as innite [8].
3. Numerical solution of boundary element equation
Solving boundary element equation using boundary conditions,
the unknown displacement and traction components (i.e., bound-
ary quantities) on the boundary S can be determined.
The boundary element equation can be solved numerically by
discretizing the boundary S into small elements (boundary ele-
ments). In this analysis, the distribution of boundary quantities
over the elements is approximated using some shape functions.
In the present study, discontinuous boundary element (constant,
linear and quadratic) formulations are used. Constant element
contains one node, linear element contains two nodes and
quadratic element contains three nodes located on the element
(Fig. 2).
For the numerical solution, the boundary S is discretized into N
boundary elements (Fig. 3). When boundary integral equation, Eq.
(3), is written at the node P
m
(m = 1, . . . , N) of the mth boundary ele-
ment the following equation can be obtained (in the absence of
body force),
1
2
u
m
N
n1
G
mn
t
n
N
n1
H
mn
u
n
m 1; . . . ; N; 5
where
x
1
x
2
P
r
r
r
r
n
n
Q (x
i
)
Q (x
i
)
S
V
Fig. 1. Typical body.
+1 0 -1
=0.00
a) Constant
+1 0 -1
1
1
2
= 0.67
b) Linear
+1 0 -1
2
= 0.75
= 0.00
c) Quadratic
1
1
1
3
Fig. 2. Interpolation functions and position of nodes.
848 I.O. Deneme et al. / Advances in Engineering Software 40 (2009) 847855
G
mn
_
Sn
GP
k
m
; QdS; H
mn
_
Sn
HP
k
m
; QdS; 6
in which S
n
is the boundary of the nth element (Fig. 3). In Eq. (5), u
n
and t
n
are associated with the nth boundary element. It should be
noted that the node P
k
m
is not a corner point. Thus, c = I is used.
According to the boundary element formulation; coordinates,
displacement and traction components of a point can be dened as
x
i
q
k1
/
k
nx
k
i
; 7
u
i
q
k1
/
k
nu
k
i
; 8
t
i
q
k1
/
k
nt
k
i
; 9
where x
k
i
is vector position of the end nodes; u
k
i
and t
k
i
are nodal dis-
placements and stresses; n is parametric variable coordinate and /
k
is shape function (k = 1, . . . , q).
The shape function for constant element (q = 1) is given as
/
1
1: 10
The shape functions for linear element (q = 2) are given as
/
1
n
n
2
n
n
2
n
1
_ _
; /
2
n
n n
1
n
2
n
1
_ _
1 n
1
< 0 and 0 < n
2
1; 11
where n
1
and n
2
denote local coordinate of rst and second node of
elements, respectively.
The quadratic shape functions for discontinuous element (q = 3)
are given as
/
1
n
1
2
n
a
n
a
1
_ _
; /
2
n 1
n
a
_ _
2
;
/
3
n
1
2
n
a
n
a
1
_ _
12
where 1 6 n 6 1 and a is a parameter (0 < a < 1) which denes the
ratio of the distance between boundary nodes to the length of the
element. The nodes 1 and 3 are allowed to vary symmetrically by
a. If a = 1, then the shape functions will reduce to the standard qua-
dratic shape functions, thus the element becomes a continuous
element.
By using Eqs. (8) and (9), Eq. (5) can be rewritten as
c
k
m
uP
k
m
N
n1
q
s1
G
mn
ks
tQ
s
n
N
n1
q
s1
H
mn
ks
uQ
s
n
13
where tQ
s
n
and uQ
s
n
are the traction and displacement vectors at
sth node of nth element, respectively. In addition
G
mn
ks
_
1
1
JnGP
k
m
; Q/
s
dn; H
mn
ks
_
1
1
JnHP
k
m
; Q/
s
dn; 14
where J is Jacobian and /
s
is shape function matrix and dened as
/
s
/
s
0
0 /
s
_ _
s 1; . . . q: 15
By writing Eq. (13) for the xed points P
k
m
with k = 1, . . . , q
(which are the nodes of S
m
) and combining them, the following
equation can be obtained:
c
m
u
m
N
n1
G
mn
t
n
N
n1
H
mn
u
n
: 16
When Eq. (16) is written for all boundary elements (m = 1, . . . , N)
and combined, the system equations of BEM can be obtained in
matrix form as,
H~ u
G
~
t; 17
where
G G
mn
;
H H
mn
1
2
Id
mn
;
~ u u
n
;
~
t t
n
m; n 1; . . . ; N
18
with d
mn
is Kroneckers delta. The elements G
mn
and H
mn
may be
computed numerically by using Gaussian quadrature formula.
The solution of Eq. (17), together with the prescribed boundary
conditions, determines numerically the unknown boundary quan-
tities. Having determined unknown boundary quantities, if desired,
the interior displacements and stresses can be computed numeri-
cally using the boundary quantities.
4. Calculation of singular integrals
The singular integrals will occur when xed point P and varied
point Q are on the same element. Therefore, diagonal matrices in
Eq. (17) will be singular. These singular matrices G
mm
ks
and H
mm
ks
can be written as,
G
mm
ks
_
Sm
GP
k
m
; Q/
s
dS; H
mm
ks
_
Sm
HP
k
m
; Q/
s
dS; 19
where k is xed point number and s is shape function number. The
integrals in Eq. (19) are to be understood in the sense of Cauchy
principal value, i.e., they do not contain the contributions when var-
ied point Q coincides with the xed point P
k
m
[8]. But the fundamen-
tal solutions involve the distance r between the xed and varied
points, and become singular when r = 0. In Eq. (14) (written for
m = n) the varied point Q is on the element S
m
and the distance r
is smaller compared to case in which mn. So that, care should
be given for the evaluation of the integrals in Eq. (19) and they
should be computed by using some special numerical integration
techniques [3].
The diagonal matrices G
mm
ks
and H
mm
ks
(for the xed point P and
the varied point Q) can be rewritten as
G
PQ
_
G
k
r
P
n/
Q
nJndn; 20
H
PQ
_
H
k
r
P
n/
Q
nJndn: 21
x
2
x
1
S
m
Q(x
i
)
S
n
S
k
n
Q
r
r
k
m
P
Fig. 3. Boundary element discretization of the body.
I.O. Deneme et al. / Advances in Engineering Software 40 (2009) 847855 849
Here = 1, 2 and k = 1, 2 for two-dimensional problems, G
k
and H
k
are, the rst and the second fundamental solution for two-dimen-
sional elastostatics, given in Eqs. (22) and (23). The number of shape
functions should be the same with varied point number
G
k
1
8pl1 m
3 4m ln
1
r
_ _
d
k
r
r
r
k
r
_ _
; 22
H
k
1
4p1 mr
@r
@n
1 2md
k
2
r
r
r
k
r
_ _
_
1 2m
r
k
r
n
r
n
k
_ __
; 23
where d
k
, m, l, n
k
and n
s
k
; 24
where R denotes half of the length of the element. s
and s
k
are unit
tangential vector components.
The second fundamental solution can be integrated using
standard Gaussian quadrature for determining H
mn
matrices. When
the xed point P and the varied point Q at the same element the
terms H
mm
can be determined using rigid-body motion as follows
[3]:
H
mm
N
n1
H
mn
mn 25
4.2. Singular integrals for linear element in static case
The rst fundamental solution G (when P and Q on the same
element) includes ln(1/r) and (1/r) singularity cases. The integral,
that contains rst fundamental solution, will be considered in
two parts. The rst part includes ln(1/r) and the second part con-
tains (1/r) singularity case. The ln(1/r) term can be calculated ana-
lytically and converting the integral limits (1, +1) to (0, +1) [11]. In
addition the second part, that includes (1/r) singularity, can be
integrated using standard Gaussian quadrature. Therefore the rst
fundamental solution can be integrated [11].
The second fundamental solution, in Eq. (23), can be integrated
using standard Gaussian quadrature for determining H
mn
matrices.
When the xed point P and the varied point Q at the same element
the terms H
mm
can be determined using rigid-body motion, as
shown in Eq. (25).
4.3. Singular integrals for quadratic element in static case
For calculating the rst fundamental solution G (when P and Q
on the same element) there is need for special techniques. For this
purpose, the integral that contains rst fundamental solution in
two parts will be considered. The rst part includes ln(1/r) and
the second part contains (1/r) singularity case.
If the fundamental solution is written, which was given at Eq.
(22), into Eq. (20), G
PQ
will contain a weak singularity in the form
of natural logarithm. After all G
PQ
can be rewritten as
G
PQ
_
1
8pl1 m
3 4m ln
1
r
P
_ _
d
k
r
r
P
r
k
r
P
_ _
/
Q
nJndn:
26
The ln(1/r
p
) term can be manipulated by conversion into a form,
where the logarithmic quadrature can apply. This conversion can
accomplish by using change of variables [12]. The second part of
Eq. (26) can be integrated using standard Gaussian quadrature.
The second fundamental solution, in Eq. (23), can be integrated
using standard Gaussian quadrature for determining H
mn
matrices.
When the xed point P and the varied point Q at the same element
the terms H
mm
can be determined using rigid-body motion, as
shown in Eq. (25).
4.4. Singular integrals in dynamic case
The orders of singularity in the dynamic fundamental solutions
are the same as in the static fundamental solutions and can be
dealt with in the following manner:
_
Dynamic
dS
_
Static
dS
_
Dynamic
Static
dS; 27
where the second integral on the right-hand side of Eq. (27) is non-
singular and can be evaluated using standard Gaussian quadrature.
The rst integral on the right-hand side of Eq. (27) can be evaluated
using the method described in static case. Eq. (27) can be rewritten
for both of the fundamental solutions in concise form as:
G
D
G
S
G
D
G
S
; 28
H
D
H
S
H
D
H
S
; 29
where G
D
and H
D
are dynamic fundamental solutions, G
S
and H
S
are
static fundamental solutions. The fundamental solutions G
D
and H
D
for two-dimensional elastodynamics are given in Eqs. (30) and (31)
G
D
1
2pl
Wd
k
vr
r
k
; 30
H
D
1
2p
dW
dr
v
r
_ _
d
k
@r
@n
r
k
n
_ _
2
r
v n
k
r
2r
r
k
@r
@n
_ _ _
2
dv
dr
r
r
k
@r
@n
c
2
p
c
2
s
2
_ _
dW
dr
dv
dr
v
r
_ _
r
n
k
_
; 31
where
W K
0
a
s
1
a
s
K
1
a
s
c
s
c
p
K
1
a
p
_ _
; 32
v K
2
a
s
c
2
s
c
2
p
K
2
a
p
; 33
where
a
s
ixr
c
s
; a
p
ixr
c
p
:
Here i is imaginary number, c
p
and c
s
are dilatational and shear
wave velocities. K
i
s are modied Bessel functions of second kind.
The singularity in H
D
can be manipulated using Eq. (29) instead
of the rigid-body motion mentioned in static case.
5. Computer programs
Based on the formulations, presented in this study, three gen-
eral purpose computer programs are developed by using FOR-
850 I.O. Deneme et al. / Advances in Engineering Software 40 (2009) 847855
TRAN77 codes for analyzing elastic or visco-elastic 2D elastody-
namic soilstructure interaction problems. The structures of the
three programs are basically the same and the ow chart is shown
in Fig. 4. The programs perform the analysis in Fourier transform
space. If desired, the obtained results can be transformed to time
space with the use of FFT algorithm [9].
6. Numerical examples
In this section, some numerical examples are solved for the
comparison of the various discontinuous boundary element formu-
lations. To this end, based on the above mentioned formulations,
some general purpose computer programs have been developed
for each type of discontinuous boundary elements. All the pro-
grams perform the analysis in Fourier transform space. Although
the programs are developed for dynamic analysis, it can also be
used for static analysis by assigning a very small value close to zero
for the frequency.
Two examples are handled to demonstrate the suitability and
accuracy of the present formulation comparing with the results ob-
tained by independent numerical methods and/or exact solutions
available in the literature. In the example problems, mesh trunca-
tion is used for the analysis of unbounded domains and the funda-
mental solutions are integrated with 10 point quadrature rule for
both standard and logarithmic Gaussian quadrature formulas.
Boundary element model have been represented with the parame-
ters of n
1
= 0.67 and n
2
= 0.67 (for linear element) also n
1
= 0.75,
n
2
= 0, n
3
= 0.75 and a = 0.75 (for quadratic element).
In the formulation hysteretic damping ratio is used for damp-
ing. The hysteretic damping in the material can be included into
Compute terms of G
mm
and H
mm
matrices
Compute the resultants of boundary tractions
Determine the displacements
and stresses at interior points
START
Read input data and perform data generations
Form the system matrices for defined boundary in problems
Singular element (m = n)?
Compute G
mn
matrices
Interior points?
Compute the unknown boundary quantities by solving Eq.17
Write the results
STOP
No
No
Yes
Compute H
mn
matrices
Yes
Fig. 4. Flow chart of computer programs.
I.O. Deneme et al. / Advances in Engineering Software 40 (2009) 847855 851
a system by replacing, in Fourier transform space, the shear mod-
ulus l of a material by l(1 + 2iz). Here, i is imaginary number and z
is the hysteretic damping ratio of the material, and all the param-
eters involving the elastic constant l are determined by this com-
plex valued shear modulus [13].
6.1. Rigid strip foundation
In this example, a rigid strip foundation on half-space shown in
Fig. 5 is considered. Here, compliances of rigid strip foundation are
investigated.
This problem has been solved, as a plane strain problem. The
top surface of the problem is discretized into 80 equal elements
for all types of discontinuous boundary elements.
The analysis is carried out in nondimensional space. The nondi-
mensional variables are dened as: