You are on page 1of 10

Numerical implementation of the integral-transform solution

to Lamb's point-load problem


H. G. Georgiadis, D. Vamvatsikos, I. Vardoulakis
Abstract The present work describes a procedure for the
numerical evaluation of the classical integral-transform
solution of the transient elastodynamic point-load (axi-
symmetric) Lamb's problem. This solution involves inte-
grals of rapidly oscillatory functions over semi-innite
intervals and inversion of one-sided (time) Laplace
transforms. These features introduce difculties for a nu-
merical treatment and constitute a challenging problem in
trying to obtain results for quantities (e.g. displacements)
in the interior of the half-space. To deal with the oscilla-
tory integrands, which in addition may take very large
values (pseudo-pole behavior) at certain points, we follow
the concept of Longman's method but using as accelerator
in the summation procedure a modied Epsilon algorithm
instead of the standard Euler's transformation. Also, an
adaptive procedure using the Gauss 32-point rule is in-
troduced to integrate in the vicinity of the pseudo-pole.
The numerical Laplace-transform inversion is based on
the robust Fourier-series technique of Dubner/Abate-
Crump-Durbin. Extensive results are given for sub-surface
displacements, whereas the limit-case results for the sur-
face displacements compare very favorably with previous
exact results.
1
Introduction
As Achenbach (1973) notes in his well-known treatise on
elastodynamics, `One of the contributions of lasting sig-
nicance in the area of wave propagation in elastic solids
is the article entitled On the Propagation of Tremors Over
the Surface of an Elastic Solid by H. Lamb (1904). In this
work, Lamb investigated the wave motion generated at the
surface of an elastic half-space by the application of con-
centrated loads at the surface or inside the half-space.
Most fully discussed were the surface motions generated
by a line load and a point load applied normally to the
surface. Both loads of harmonic time dependence and
impulsive loads were considered'. We add that in the case
of slow application of the surface loading and an associ-
ated linear elastostatic material response, the above situ-
ation reduces to another classical problem, namely the
Flamant-Boussinesq problem (see e.g. Fung 1965; Barber
1992). Also, the anti-plane shear version of Lamb's prob-
lem was considered by Achenbach (1973) within the
context of classical linear elasticity, and by Georgiadis and
Vardoulakis (1998) within the context of gradient elastic-
ity. Obviously, the several variants of Lamb's problem are
of great interest in such areas as Geodynamics, Contact
Mechanics and Ultrasonic Nondestructive Evaluation. In
addition, these solutions may yield the necessary eld
quantities required to apply the Boundary Element
Method in more complicated problems (see e.g. Beskos
1987, 1997; Manolis and Beskos 1988).
Here, we concentrate on the transient point-load case and
present an approach for the numerical implementation of
the associated integral-transform solution of Lamb's prob-
lem. In this axisymmetric situation, the rapid (aperiodic)
application of a concentrated vertical load on the surface of
an elastic half-space is considered and one is usually in-
terested in the transient displacements at the surface and in
the interior of the half space. Although a good deal of studies
examining the generation of surface motions has been
published (see e.g. Lamb 1904; Cagniard 1939; Pekeris 1955;
Chao et al. 1961; Achenbach 1973; Mooney 1974; Eringen
and Suhubi 1975), there is a rather limited amount of work
on the more difcult case of sub-surface motions (Eason
1966; Gakenheimer 1970; Johnson 1974; Apsel 1979; Mi-
khailenko 1985; Laturelle 1991). Obtaining results for the
interior of the half space is, however, very useful in the
design of buried structures like pipes and tunnels.
From the general Laplace/Hankel transformed solution
of Lamb's point-load problem, results for stresses and
displacements typically follow through inversions involv-
ing contour integration and Cauchy's residue theorem.
Even in the case of surface motions the latter procedure is
complicated (Kupradze 1963; Achenbach 1973; Eringen
and Suhubi 1975) and the nal expressions are available
only in an integral form. The case of sub-surface motions
generated by a load varying in time as a Heaviside step
function was worked out by Gakenheimer and Miklowitz
(1969) and Gakenheimer (1970), who ended up with rather
involved integral expressions. Unfortunately, if a different
load variation is considered, then their laborious inversion
procedure of the Cagniard-deHoop type (Cagniard 1939;
deHoop 1960) should again be repeated and numerical
integration should also be performed. Notice that the
aforementioned difculties for an exact (i.e. analytical)
transform inversion will become unsurmountable for the
Computational Mechanics 24 (1999) 9099 Springer-Verlag 1999
90
Received 20 October 1998
H.G. Georgiadis (&), D. Vamvatsikos, I. Vardoulakis
Mechanics Division, National Technical University of Athens,
5, Heroes of Polytechnion Ave.,
Zographou, GR-15773, Greece
HGG gratefully acknowledges the support of NATO
(CRG 972116).
more general but common cases of a viscoelastic (Geor-
giadis 1993) and a layered (Cetinkaya and Vakakis 1996)
half-space. Thus, due to the complexities occurring in
analytical transform inversions for the homogeneous
purely-elastic problem and due to the need for general
procedures which will be applicable in more complex sit-
uations than the latter one, we propose here a numerical
approach for generating results for the surface and sub-
surface displacements in the case of the transient point-
load Lamb's problem.
The present approach uses an adaptive Gauss rule and
Longman's (1956) method, the latter being combined with
the Epsilon algorithm (MacDonald 1964; Davis and
Rabinowitz 1984), to treat the so-called wavenumber in-
tegrals (Xu and Mal 1985) in the one-sided Laplace-
transform plane, and the Fourier-series technique of
Dubner-Abate (1968) Crump (1976) Durbin (1976) to
invert the one-sided Laplace transform. Results are pre-
sented for loadings with both a Heaviside step-type and a
triangular-type variation in time. Our results compare very
favorably with results based on exact inversions for surface
(Eringen and Suhubi 1975) and sub-surface (Gakenheimer
1970) displacements. Also, the present approach is appli-
cable in more difcult problems (like the viscoelastic and
the layered half-space) involving wavenumber integrals in
the one-sided Laplace-transform plane.
2
Problem statement and analytical solution
We consider a homogeneous, elastic and isotropic half-
space z _ 0, whose free surface is subjected, at time t = 0,
to axisymmetric tractions. Then, with respect to a system
of cylindrical polar coordinates r; h; z ( ) attached to this
body (see Fig. 1), the generated axially symmetric motions
in the half-space can be described by the following dis-
placement and stress eld (Eringen and Suhubi 1975)
u
r
=
ou
or

o
2
v
or@z
(1a)
u
z
=
ou
oz

o
2
v
or
2

1
r
ov
or
(1b)
r
rr
= k
ou
r
or

u
r
r

ou
z
oz

2l
ou
r
or
(2a)
r
zz
= k 2l ( )
ou
z
oz

k
r
o ru
r
( )
or
(2b)
r
zr
= l
ou
r
oz

ou
z
or

(2c)
where u
r
; u
z
( ) and r
rr
; r
zz
; r
zr
( ) are the non-vanishing
components of, respectively, the displacement vector and
the stress tensor, k and l are the Lame constants
(k = Em=[ 1 m ( ) 1 2m ( )[ and l = E=[2 1 m ( )[ in terms of
the elastic modulus E and Poisson's ratio m), and the
functions u and v are the Helmholtz potentials satisfying
the wave equations
\
2
u =
1
c
2
1
o
2
u
ot
2
; (3a)
\
2
v =
1
c
2
2
o
2
v
ot
2
(3b)
with c
1
= k 2l ( )=q [ [
1=2
and c
2
= l=q ( )
1=2
denoting, re-
spectively, the dilatational (or longitudinal) and the shear
(or transverse) wave velocity, and with \
2
= o
2
=or
2

1=r ( ) o=or ( ) o
2
=oz
2

being the Laplace operator. All
eld quantities above are functions of r; z; t ( ), whereas (3)
are the eld equations of the problem. Finally, we notice
that the representation in (1) and (3) is different than the
usual displacement-potential representation found, for
instance, in other classical texts as those of Achenbach
(1973) and Miklowitz (1978).
Next, boundary conditions are assumed of the following
form
r
zz
r; z = 0; t ( ) = T
1
r; t ( ) (4a)
r
zr
r; z = 0; t ( ) = T
2
r; t ( ) (4b)
with T
1
and T
2
being known functions of bounded varia-
tion as r; t (the latter assumption secures that the
functions are amenable to integral transformations),
whereas the problem statement is completed by the fol-
lowing initial conditions (which are pertinent to the qui-
escent past that we assume here)
u r; z; t = 0 ( ) = ou r; z; t = 0 ( )=ot = v r; z; t = 0 ( )
= ov r; z; t = 0 ( )=ot = 0 : (5)
The problem (1)(5) will be attacked by means of the one-
sided Laplace transform and the Hankel transform. The
appropriate denitions are as follows

f r; z; p ( ) =
Z

0
f r; z; t ( ) e
pt
dt ; (6a)
f r; z; t ( ) = 1=2pi ( )
Z
Br

f r; z; p ( ) e
pt
dp (6b)

f
+
n; z; p ( ) =
Z

0

f r; z; p ( ) J
m
nr ( ) r dr ; (7a)

f r; z; p ( ) =
Z

0

f
+
n; z; p ( ) J
m
nr ( ) n dn (7b)
where (6a) and (7a) provide the direct transforms, (6b)
and (7b) provide the inverse transforms, Br denotes the
Bromwich inversion path in the complex p-plane, and J
m
()
is the rst-kind Bessel function of order m.
Now, successive application of (6a) and (7a) with m = 0
to (3), in view of (5), and subsequent inversion according
to (7b) yield the following general expressions for the
Laplace-transformed displacements (Eringen and Suhubi
1975)
Fig. 1. An elastic half-space under a normal time-dependent
point load
91
u
r
=
Z

0
nA exp k
2
a
1
z ( ) k
2
na
2
B [
exp k
2
a
2
z ( )[ J
1
nr ( )dn ; (8a)
u
z
=
Z

0
k
2
a
1
A exp k
2
a
1
z ( ) n
2
B

exp k
2
a
2
z ( )[ J
0
nr ( )dn ; (8b)
where A = A n; p ( ) and B = B n; p ( ) are yet unknown
functions (which will be determined by enforcing the
boundary conditions), k
j
= p=c
j
(with j = 1; 2), and
a
j
= (n
2
k
2
j
)
1=2
=k
2
. Also, u
r
and u
z
remain bounded as
z provided that branch cuts have been introduced
for the functions k
2
a
j
such that re k
2
a
j

> 0 in the complex
n-plane. Expressions analogous to (8) can be obtained for
the transformed stresses, and when these are combined
with the transformed tractions, the functions A and B are
found as
A = n l R n; p ( ) [ [
1
2n
2
k
2
2


T
+
1
n; p ( )

2k
2
a
2
n

T
+
2
n; p ( )

; (9a)
B = l R n; p ( ) [ [
1
2k
2
a
1
n

T
+
1
(n; p)

2n
2
k
2
2


T
+
2
n; p ( )

; (9b)
where

T
+
1
n; p ( ) =
Z

0

T
1
r; p ( ) J
0
nr ( ) r dr ; (10a)

T
+
2
n; p ( ) =
Z

0

T
2
r; p ( ) J
1
nr ( ) r dr (10b)
and R n; p ( ) is the well-known Rayleigh function given as
R n; p ( ) = 2n
2
k
2
2

2
4k
2
2
a
1
a
2
n
2
: (11)
Finally, combining (8) with (9) provides the Laplace-
transformed solution of the problem. In the case
T
1
r; t ( ) = F=2pr ( ) d r ( ) H t ( ) and T
2
r; t ( ) = 0, where
d( ) is the Dirac delta distribution and H( ) is the Heaviside
step function, (i.e. for a suddenly applied concentrated
normal force on the half-space surface), Eringen and Su-
hubi (1975) obtained an exact inversion for the surface
z = 0 ( ) displacements in terms of elliptic integrals,
whereas Gakenheimer and Miklowitz (1969) obtained an
exact general (z = 0 or z ,= 0) inversion, which results
however in complicated integrations handled only by nu-
merical means (Gakenheimer 1970). Both these exact in-
versions were very laborious despite the simple form of the
boundary conditions. In the case of more general loading
and in the presence of material damping (viscoelastic ef-
fects) and/or non-homogeneity (layered half-space), exact
transform inversions are impossible and the associated
wavenumber integrations and Laplace-transform inver-
sions (like the ones in (8) here) should be done numeri-
cally. Indeed, this is the object of the present study. The
procedure will be described for the simplest possible case,
i.e. for Lamb's point-load problem in a homogeneous and
purely elastic half-space.
Below, we deal with boundary conditions of the form
T
1
r; t ( ) =
F
2pr
d r ( ) g t ( ) ; (12a)
T
2
r; t ( ) = 0 (12b)
where F has dimensions of force, g t ( ) is dimensionless and
is assumed to be Laplace transformable. Explicit numerical
results will be obtained for the following two cases of g t ( )
g t ( ) = H t ( ) (13)
g t ( ) =
t=t
0
( ) for 0 < t=t
0
( ) < 1;
c 1 ( )
1
c t=t
0
( ) [ [ for 1 < t=t
0
( ) < c,
0 for t=t
0
( ) = 0; c [ [
8
>
<
>
:
(14)
where H( ) is the Heaviside step function, t
0
is a constant
with dimensions of time, and c is a dimensionless con-
stant. Notice that the loading case (14) implies a triangu-
lar-type variation in time and may model thus explosive
loadings. Further, the Laplace-transformed displacements
are obtained from (8)(12) as
u
r
r; z; p ( ) =
F g p ( )
2pl

Z

0
N
r
n; p; r; z ( )dn ; (15a)
u
z
r; z; p ( ) =
F g p ( )
2pl

Z

0
N
z
n; p; r; z ( )dn (15b)
where
N
r
= 2n
2
k
2
2

exp k
2
a
1
z ( )

2k
2
2
a
1
a
2
exp k
2
a
2
z ( )

n
2
J
1
nr ( )
R n; p ( )
; (16a)
N
z
= 2n
2
k
2
2

exp k
2
a
1
z ( )

2n
2
exp k
2
a
2
z ( )

nk
2
a
1
J
0
nr ( )
R n; p ( )
: (16b)
The numerical problem we face now is the evaluation of
the integrals in (15) (wavenumber integrals) and the La-
place-transform inversion of u
r
and u
z
. This is exactly the
subject of the next sections.
3
Numerical procedure
In this section we rst identify the difculties encountered
when one tries to implement numerically the solution (15)
and then we provide means to deal with them. More
specically, we notice the following features of the trans-
formed-solution behavior.
The functions N
r
and N
z
given in (16) involve:
1. Branch points at re n ( ) = b [ [=c
1
and re n ( ) = b [ [=c
2
due
to the presence of the terms k
2
a
1
and k
2
a
2
, respectively,
where p = a ib is the Laplace-transform variable, with
a and b having dimensions of inverse time. These branch
points occur both in the numerators of N
r
and N
z
and in
the denominator R n; p ( ). In the physical problem, these
branch points correspond to the arrivals (at the observa-
tion point) of longitudinal and shear waves marking the
associated non-smoothness.
2. A pole at re n ( ) = b [ [=c
R
, which is equal to the positive
real root of the Rayleigh function R n; p ( ). This occurs
when p is purely imaginary, i.e. p = 0 ib. The quantity c
R
is the velocity of the Rayleigh waves (see e.g. Rivlin and
Hayes 1962; Achenbach 1973; Eringen and Suhubi 1975)
92
and, therefore, the pole marks the arrival of the Rayleigh
wavefront. Note that the following inequalities hold be-
tween the characteristic wave speeds in the material,
c
R
< c
2
< c
1
, whereas Rahman and Barber (1995) and
Nkemzi (1997) provided exact expressions for evaluating
c
R
in terms of (c
2
=c
1
)
2
= (1 2m)=2(1 m). We also note
that, in a more general case, analogous integral expres-
sions providing the speed of the thermoelastic Rayleigh
waves were given in Brock (1995, 1997) and Georgiadis
et al. (1998). Finally, if a ,= 0 with p = a ib; a > 0 and
b ; ( ), a pole never occurs. However, large val-
ues of the real and imaginary parts of N
r
and N
z
arise when
b > 2a, and the situation often reminds a pole-like be-
havior. The respective stations along the x-axis are called
pseudo-poles. The largest absolute values of the functions
reN
r
; imN
r
; reN
z
and imN
z
occurs for the same b for
which a pole at re(n) = [b[=c
R
occurs when we set
p = 0 ib.
3. The Bessel functions J
1
nr ( ) and J
0
nr ( ) that produce
rapid oscillations of the integrands in (15). These oscilla-
tions are of aperiodic type with very slowly decaying am-
plitudes. The latter property follows from the asymptotic
expression of the rst-kind Bessel functions for large ar-
guments (see e.g. Abramowitz and Stegun 1982) which
shows that J
m
(n) = O(n
1=2
) as n . In addition, the
presence of the radial distance r in J
0
nr ( ) and J
1
nr ( ) makes
the number of oscillations to increase. For example, while
J
0
n ( ) and J
1
n ( ) have 3 roots in the interval (0,10], J
0
10n ( )
and J
1
10n ( ) exhibit nearly 30 zeros in the same interval.
Integrals with such oscillatory integrands are called
wavenumber integrals.
The difculties described above can be better realized if
one plots the real and imaginary parts of the functions
R n; p ( ); N
r
n; p; r; z ( ) and N
z
n; p; r; z ( ) vs the variable n for
several values of the complex parameter p (whose inverse
real and imaginary parts are expressed in seconds),
r=z = 10 and Poisson's ratio m = 0:25. Some representative
graphs are presented in Figs. 24. The branch points at
n = [b[=c
1
; n = [b[=c
2
and the pseudo-pole at n = [b[=c
R
are denoted by arrows at the stations bp
1
(branch point 1),
bp
2
(branch point 2) and rp (Rayleigh peak), respectively.
For instance, Fig. 2 for N
r
(with p = 1 i10) shows a
rather smooth behavior but Fig. 3 for N
z
(with
p = 1 i10
3
) and Fig. 4 for N
r
(with p = 1 i10
4
) indicate
non-smoothness, pseudo-pole behavior and oscillations.
Of course, the latter features introduce serious difculties
in the numerical evaluation of the integrals in (15).
3.1
Pseudo-poles in the integrands
First, we have to deal with the pseudo-pole behavior
(Rayleigh peak) of the integrands. It should be noted that
if 2a b, the pseudo-pole behavior is eliminated. Indeed,
taking 2a b in the Dubner/Abate-Crump-Durbin
method (used below to numerically invert the Laplace
transform in (15)) seems to be feasible but, in this way,
one avoids to consider the times of arrival of the Rayleigh
wave and thus fails to see the most interesting phase of the
dynamic process. Therefore, the numerical scheme has to
fully account for the pseudo-pole behavior. We may also
observe that, in the vicinity of the pseudo-pole say
n
p
; reN
r
and reN
z
behave like y n
p

1
, with y n
p
but
y ,= n
p
, whereas imN
r
and imN
z
behave like y n
p

2
.
The fact that, in the general case of a Bromwich path not
coincident a ,= 0 ( ) with the im p ( )-axis, reN
r
and reN
z
do
not contain a genuine Cauchy principal-value singularity
in the vicinity of the Rayleigh peak (since y ,= n
p
as stated
before) precludes the possibility of applying here certain
numerical techniques developed for Cauchy principal-
value integrals (see e.g. Davis and Rabinowitz 1984;
Okecha 1987). Instead, we follow an adaptive technique
Fig. 2. Variation of reN
r
(n; p; r; z) and imN
r
(n; p; r; z) with n for
the values p = 1 i10, r=z = 10 and m = 0:25
Fig. 3. Variation of reN
z
(n; p; r; z) and imN
z
(n; p; r; z) with n for
the values p = 1 i10
3
, r=z = 10 and m = 0:25
Fig. 4. Variation of reN
r
(n; p; r; z) and imN
r
(n; p; r; z) with n for
the values p = 1 i10
4
, r=z = 10 and m = 0:25
93
(based essentially on ideas of Brock, Georgiadis and
Charalambakis 1994) according to which one rst evalu-
ates the abscissa of the Rayleigh peak (this is accomplished
here through the Newton-Raphson method) and then
considers small sub-intervals to the left and right of that
abscissa. These sub-intervals are taken to be symmetrically
situated with respect to the Rayleigh-peak station and
become smaller and smaller as we approach that point.
Extensive numerical calculations based on the Gauss 32-
point rule showed that robust results can indeed be ob-
tained with 12 integration sub-intervals along an interval
[a
p
,b
p
[ around the Rayleigh peak (six sub-intervals to the
left and six to the right) in the following manner
a
p
; b
p

= a
p
; n
1

' n
1
; n
2
[ [ ' . . . ' n
5
; n
p
= [b[=c
R

' n
p
= [b[=c
R
; n
6

' . . . ' n
10
; b
p

:
(17)
In each case, [a
p
,b
p
[ is chosen as the interval between the
two nearest to the pseudo-pole roots of the respective
Bessel functions so as to avoid in the numerical treatment
the interference of the oscillations with the Rayleigh-peak
behavior. Also, in (17) each of the rst four intervals is
taken to be equal to the one quarter of the interval
[a
p
,(n
p
d)[, each of the last four intervals equal to the one
quarter of the interval [(n
p
d),b
p
[, while the four smaller
central intervals are taken to be equal to d=2, where
d =
min n
p
a
p

; b
p
n
p

10
: (18)
In this way, we fully exploit the fact that the Gauss rule
end-points do not coincide with the end-points of the
integration interval. As for the robustness and efciency of
the technique, the nal judgement is provided, of course,
by the comparison with available exact results. Generally,
the results presented in the next section demonstrate an
excellent agreement with limit-case exact results.
3.2
Branch points in the integrands
The difculty associated with the branch points can easily
be overcome by the same means as in the previous situ-
ation, i.e. by splitting up the integration interval at the
particular branch point and integrating to the left and
right of it with the Gauss rule. In particular, the small
intervals [a
1
,b
1
[ and [a
2
,b
2
[ are xed around the branch
points n
b1
= [b[=c
1
and n
b2
= [b[=c
2
, respectively. Then,
the Gauss 32-point rule is applied separately for the sub-
intervals [a
1
,n
b1
[, [n
b1
,b
1
[, [a
2
,n
b2
[ and [n
b2
,b
2
[. These in-
tervals are again determined by the location of the two
nearest to each branch point roots of the corresponding
Bessel function. In the case, however, where both branch
points or one branch point and the pseudo-pole or even
the two branch points and the pseudo-pole are located
between two consecutive roots (this happens only when
b a, with p = a ib being the Laplace-transform vari-
able), we consider an initial partition of the interval [a
p
,b
p
[
in three appropriate sub-intervals and then follow the
scheme described above for each sub-interval. For exam-
ple, if the two branch points and the pseudo-pole lie be-
tween two consecutive roots of the relevant Bessel function
n
b1
< n
b2
< n
p

, the initial partition is performed in the
way:
a
p
; n
b1
n
b2
( )=2

' n
b1
n
b2
( )=2; n
b2
n
p

=2

' n
b2
n
p

=2; b
p

:
Of course, an accurate location of the pertinent zeros of
the Bessel functions J
0
and J
1
, for real and positive argu-
ments, is a prerequisite in the above procedures. Use was
made for the rst 20 points of Table 9.5 in Abramowitz
and Stegun (1982) and, for larger arguments, of the
McMahon formula (see e.g. Sect. 9.5 in the same treatise).
Both sources provide an accuracy of 10 decimal digits.
Also, convenient polynomial approximations for the Bessel
functions with 78 decimal digits accuracy (see e.g. Section
9.4 in Abramowitz and Stegun 1982) were utilized
throughout the numerical integrations. Finally, all inte-
grations between the zeros were performed with the Gauss
32-point rule.
3.3
Wavenumber integrations
The third difculty in evaluating numerically the integrals
in (15), and the more serious one, is associated with the
oscillations of the Bessel functions J
0
nr ( ) and J
1
nr ( ). For
such integrals, the standard device is to compute the
positive and negative contributions individually and to
sum the resulting innite series (Davis and Rabinowitz
1984). However, if no special summation technique is
applied, this procedure exhibits a rather slow convergence
and introduces appreciable round-off errors. Therefore,
various techniques, such as the Euler transformation
(Longman 1956), V-transformation (Levin 1973),
W-transformation (Sidi 1982), Aitken's method (Green-
berg 1975), and e-transformation (Davis and Rabinowitz
1984), have been proposed to transform the slowly con-
vergent series into one more rapidly convergent. Here, we
tested in our particular problem the Euler transformation
(ET) and the e-algorithm (EPAL), which are considered
two of the most effective accelerating techniques. Never-
theless, both techniques need to be appropriately modied
in order to become applicable to our specic problem.
Regarding ET, a rst modication (with respect to the
standard form) to effectively accelerate summation of the
series
P

n=0
1 ( )
n
V
n
= V
0
V
1
V
2
V
3
V
4
,
where V
n
> 0 and V
n
< V
n1
(with n = 0; 1; 2; . . .), is
X

n=0
1 ( )
n
V
n
=
X
m1
n=0
1 ( )
n
V
n

X

n=m
1 ( )
n
2
nm1 ( )
D
nm
V
n
;
(19)
with D being the forward difference operator dened by
D
0
V
n
= V
n
; DV
n
= V
n1
V
n
and D
r1
V
n
= D D
r
V
n
( ) =
D
r
V
n1
D
r
V
n
. This modication makes the transforma-
tion more effective when small differences occur between
the absolute values of the terms of the series (which is the
case here). A disadvantage of ET is the recurrence manner
of calculating D
r
V
n
. Especially for large r, stack overow
may occur in running the computer program. We have
therefore to modify again the algorithm so that a
`marching' scheme (calculations `tree') to be followed
94
from a column (containing the rst terms of the series
which will be used in the acceleration algorithm) at the far
left to a column (containing the `nal' term D
nm
V
n
) at the
far right, in the same manner as with the e-algorithm (see
e.g. the relative sketch in p. 44 of Davis and Rabinowitz
1984). For instance, if an 8-term series is considered
and m = 2 is taken in (19), we proceed to calculate
DV
2
= V
3
V
2
, DV
3
= V
4
V
3
; . . . ,DV
6
= V
7
V
6
for the second column, D
2
V
2
= DV
3
DV
2
; D
2
V
3
=
DV
4
DV
3
; . . . ; D
2
V
5
= DV
6
DV
5
for the third column
and so on. In this way, there is an optimization of the
procedure of calculations with no undue repetition.
However, the algorithm should not be supplied with an
excessively large number of terms n because of the pres-
ence of (1)
n
=2
(nm1)
in (19). The latter factor may cause
overow or underow error and therefore may lead to a
collapse of the program. Finally, m = 6 was taken in the
relative calculations of the present problem in order to
avoid the inuence of the pseudo-pole on the amplitude of
oscillations of the Bessel functions.
Regarding EPAL, this algorithm should also be modied
before its application to our problem. In particular, the
non-linear EPAL proceeds as follows: If S
m
=
P
m
n=1
W
n
is
the mth order partial sum of the series
P

n=1
W
n
, the series
can be approximated by using the rst 2N 1 partial
sums, where m = 1; 2; . . . ; 2N 1. To this end, one de-
nes e
(m)
k1
= e
(m1)
k1
(e
(m1)
k
e
(m)
k
)
1
with e
(m)
0
= 0 and
e
(m)
1
= S
m
. Then, it turns out (Davis and Rabinowitz 1984)
that e
(1)
1
; e
(1)
3
; e
(1)
5
; . . . ; e
1 ( )
2N1
is a sequence of approxima-
tions of the series sum which constitutes a better ap-
proximation of the sequence of the partial sums S
m
. Two
difculties usually occur if one tries to implement the
EPAL, in its standard form presented above, in the present
problem. Both drawbacks are related to the use of large
numbers N. More specically, although increasing N
generally leads to an increased accuracy, after a certain
value the accuracy deteriorates (i.e. divergence occurs)
because of round-off errors due to the division operation
in the equation giving e
(m)
k1
above. Since the EPAL is ap-
plied in a different series in each case, the optimum N
should be determined automatically and not simply by
`trial and error'. Determining N will be done in relation to
the other difculty. The latter occurs in a non-predictable
way when computationally (e
(m1)
k
e
(m)
k
) becomes zero at
a certain step of the calculations `tree'. Then, the program
stops because of `division by zero'. Again, the fact that
EPAL is used most frequently in our program precludes a
`trial and error' procedure. Instead, we follow an adaptive
procedure which obtains the maximum value N
max
which
does not cause `singular error' in implementing the
algorithm. According to this, the term (e
(m1)
k
e
(m)
k
) is
checked at every step and when this becomes zero the
corresponding right triangular array is automatically cut
out. This results in a continuing adjustment of the right
`edge' of the array to the value of N
max
causing no `singular
error' in the program.
Finally, a comparison was performed between the
methods of modied ET, standard EPAL, and modied
EPAL used to evaluate numerically the integrals
I
0
=
R

0
J
0
(n)dn and I
1
=
R

0
J
1
(n)dn (for which the exact
values I
0
= 1 and I
1
= 1 are available), as test cases. This
comparison demonstrated the superiority of the modied
EPAL over all other techniques. Thus, the nal results
presented in Sect. 4 are based on the modied EPAL .
3.4
Numerical Laplace-transform inversion
Survey articles containing comparisons of various tech-
niques on numerical Laplace-transform inversion (Davies
and Martin 1979; Narayanan and Beskos 1982; Duffy 1993)
and previous experience (Georgiadis 1993; Georgiadis and
Rigatos 1996) suggest the Dubner/Abate-Crump-Durbin
technique (Dubner and Abate 1968; Crump 1976; Durbin
1976) as a powerful means for inverting (15) according to
the operation (6b). Of course, with such a numerical
technique, one may obtain reliable results only for a lim-
ited time-interval and not for the whole time domain. This
is because the associated integral equation for obtaining
the original function (i.e. f (r; z; t) in (6a)) is an equation of
the rst kind and, therefore, inevitably unstable to any
numerical treatment.
Figure 5 shows representative graphs of the real and
imaginary parts of the Laplace-transformed displacement
u
z
(p; r; z) for the step-type loading and in the case of
r = z = 1 m; p = 6000 ib and m = 0:25. The functions
exhibit an oscillatory behavior but are bounded every-
where.
The Dubner/Abate-Crump-Durbin technique is briey
outlined here. The starting point is the following alterna-
tive form of (6b)
f (t) =
e
wt
p
Z

0
[re(

f (w iu)) cos(ut)
im(

f (w iu)) sin(ut)[du (20)


where p = w iu: In this way, we now face a real integral
which then can be approximated by applying the trape-
zoidal rule (Davis and Rabinowitz 1984). This leads to an
approximation for the original function of the following
Fourier-series form
Fig. 5. Variation of re u
z
(p; r; z) and im u
z
(p; r; z) with im(p) for
the values re(p) = 6000, r=z = 1 and m = 0:25
95
f (t)
e
wt
T

1
2

f (w)
X

k=1
[re(

f (w ikp=T))
"
cos(kpt=T) im(

f (w ikp=T))
sin(kpt=T)[

: (21)
Crump (1976) has presented a systematic error analysis in
the above procedure from which f (t) can be computed to a
predetermined accuracy. First, the period T is chosen so
that 2T)t
max
(where t
max
is the time up to which results are
to be obtained) and then w is obtained by
w = q [ln(E)[=2T, where q is a number slightly larger
than max[re(p
0
)[ (where p
0
is a pole of

f (p)) and the rel-
ative error is to be no greater than E (E = 10
8
was used in
the present computations).
Finally, the necessary increase of the convergence rate of
(21) is again obtained by using the modied EPAL.
In our specic problem the following choice of pa-
rameters was made: (1) t
max
= s
max
(r
2
z
2
)
1=2
=c
2
, where
s
max
is the normalized time up to which we are interested
in obtaining results. The value s
max
= 2 is taken since for
s _ 1:145 the arrival of all types of waves (longitudinal,
shear, head and Rayleigh) at the observation point (r; z)
has already occurred for any choice of Poisson's ratio. (2)
T = 0:8t
max
which is a value also recommended by Crump
(1976). (3) q = 0 since u
r
(p; r; z) and u
z
(p; r; z) do not
exhibit poles. (4) N = 50 i.e. introducing 2N 1 = 101
terms in the EPAL as applied to (21). This produces suf-
ciently accurate results (see next section for comparisons
with limit-case exact results).
4
Results and comparisons
In all graphs which will be presented below, normalized
quantities appear. We employ: (1) Normalized time
dened by s = c
2
t= r
2
z
2
( )
1=2
, and (2) Normalized dis-
placements dened by u
norm
r
= pl r
2
z
2
( )
1=2
u
r
=F and
u
norm
z
= pl r
2
z
2
( )
1=2
u
z
=F. In these graphs also, arrows
indicate the arrival of the longitudinal or primary P ( ),
shear or secondary S ( ), head or von Schmidt SP ( ) and
Rayleigh R ( ) waves at the station r; z ( ).
First, in order to check the accuracy obtained by the
present approach we compare our numerical results with
the exact results given by Mooney (1974) and reproduced
by Eringen and Suhubi (1975). The latter results are for the
surface displacements u
r
and u
z
, i.e. for z = 0, and for
impact loading (see Eq. (13)) and Poisson's ratio m = 0:25.
These are given in terms of elliptic integrals, which are
here computed through Bulirsch's algorithm (Press et al.
1986). Figures 6 and 7 depict graphs of both approximate
u
norm
r
; u
norm
z

and exact u
exact
r
; u
exact
z

normalized dis-
placements at the half-space surface z = 0 ( ). One may
observe the excellent accuracy of the approximate results
given by the present approach. It is also worth noticing
that both radial and normal displacements take on innite
values at the arrival of the Rayleigh wave. In the normal-
ized time-range near this innite discontinuity, the ap-
proximate displacements exhibit some small oscillations
around the ``correct'' curve which inevitably occur because
of Gibb's phenomenon (see e.g. Greenberg 1975) in sum-
ming the Fourier series in (21).
Results for sub-surface displacements z ,= 0 ( ) are rare in
literature because no simple exact transform-inversion
arguments of the Cagniard-deHoop type apply, as was
explained in Sect. 2. Thus, our numerical approach is of
much importance in this case. Figures 8 and 9 depict
Fig. 6. Variation of the normalized approximate and exact radial
displacement with the normalized time for impact loading and
z = 0, m = 0:25
Fig. 7. Variation of the normalized approximate and exact ver-
tical displacement with the normalized time for impact loading
and z = 0, m = 0:25
Fig. 8. Variation of the normalized radial displacement with the
normalized time for impact loading and r=z = 1:732 (i.e. u =
60

), m = 0:25
96
graphs of u
norm
r
; u
norm
z

for g t ( ) = H t ( ) (impact loading),
m = 0:25 and r=z = 1:732 (i.e. u = tan
1
r=z ( ) = 60

),
whereas Figs. 10 and 11 depict similar graphs for
r=z = 0:577 (i.e. u = 30

).
We close the presentation of results by giving some
graphs of the displacement u
norm
z
generated by the loading
in (14), i.e. loading with a triangular-type variation in
time. This type of loading is convenient for modeling ex-
plosive surface loadings. Our results were obtained for
t
0
= r=2c
R
and c = 2 in the expression of g t ( ) in (14), and
for Poisson's ratio m = 0:25. Figure 12 corresponds to the
case z = 0 (i.e. u = 90

), whereas Figs. 13 and 14 corres-


pond to the cases r=z = 1:732 (i.e. u = 60

) and
r=z = 0:577 (i.e. u = 30

), respectively. Generally, one


may observe here that the displacements are not of so
steep variation as in the previous case of impact loading.
Finally, we should mention that the computation time
on a PC Pentium/120 MHz for obtaining each graph was
typically 5 min.
5
Conclusions
A procedure is described for numerically implementing
the integral-transform solution of Lamb's point-load
problem. This transform solution involves integrals of
rapidly oscillatory functions, which may also exhibit
spikes at certain locations, over semi-innite intervals and
inversion of one-sided Laplace transforms. The scheme
Fig. 9. Variation of the normalized vertical displacement with
the normalized time for impact loading and r=z = 1:732 (i.e.
u = 60

), m = 0:25
Fig. 10. Variation of the normalized radial displacement with the
normalized time for impact loading and r=z = 0:577 (i.e. u =
30

), m = 0:25
Fig. 11. Variation of the normalized vertical displacement with
the normalized time for impact loading and r=z = 0:577 (i.e.
u = 30

), m = 0:25
Fig. 12. Variation of the normalized vertical displacement with
the normalized time for triangular-type loading and z = 0,
m = 0:25
Fig. 13. Variation of the normalized vertical displacement with
the normalized time for triangular-type loading and r=z = 1:732
(i.e. u = 60

), m = 0:25
97
employed involves Longman's method, an adaptive Gauss
rule, a modied Epsilon algorithm, and a Fourier-series
technique for the Laplace-transform inversion. The dis-
placement components at different locations of the half
space are presented for two different load histories. The
results can be useful in the design of buried structures like
pipes and tunnels.
References
Abramowitz M, Stegun IA (1982) Handbook of Mathematical
Functions, New York: Dover
Achenbach JD (1973) Wave Propagation in Elastic Solids, New
York: North-Holland
Apsel RJ (1979) Dynamic Green's Function for Layered Media
and Application to BVPs. Ph.D. Thesis, Univ. California San
Diego (Univ. Microlms, Ann Arbor, Michigan)
Barber JR (1992) Elasticity, Dordrecht: Kluwer Academic Pub-
lishers
Beskos DE (1987) Boundary element methods in dynamic anal-
ysis. ASME Appl. Mech. Reviews 40:123
Beskos DE (1997) Boundary element methods in dynamic analysis:
Part II (19861996). ASME Appl. Mech. Reviews 50:149157
Brock LM (1995) Slip/diffusion zone formation at rapidly-loaded
cracks in thermoelastic solids. J. Elasticity 40:183206
Brock LM (1997) Transient three-dimensional Rayleigh and
Stoneley signal effects in thermoelastic solids. Int. J. Solids
Structures 34:14631478
Brock LM, Georgiadis HG, Charalambakis NC (1994) Static
frictional indentation of an elastic half-plane by a rigid un-
symmetrical punch. ZAMP 45:478492
Cagniard L (1939) Reexion et Refraction des Ondes Seismiques
Progressive, Paris: Gauthier-Villard; English translation by
E.A. Flinn and C.H. Dix (1962): Reection and Refraction of
Progressive Seismic Waves, New York: McGraw-Hill
Cetinkaya C, Vakakis AF (1996) Transient axisymmetric stress
wave propagation in weakly coupled layered structures.
J. Sound Vibration 194:389416
Chao CC, Bleich HH, Sackman J (1961) Surface waves in an
elastic half space. ASME J. Appl. Mech. 28:300302
Crump KS (1976) Numerical inversion of Laplace transforms
using a Fourier series approximation. JACM 23:8996
Davies B, Martin B (1979) Numerical inversion of the Laplace
transform: A survey and comparison of methods. J. Comp.
Physics 33:132
Davis PJ, Rabinowitz P (1984) Methods of Numerical Integration,
New York: Academic Press
deHoop AT (1960) The surface line source problem. Appl. Sci.
Res. B8:349356
Dubner H, Abate J (1968) Numerical inversion of Laplace
transforms by relating them to the nite Fourier cosine
transform. JACM 15:115123
Duffy DG (1993) On the numerical inversion of Laplace trans-
forms: Comparison of three new methods on characteristic
problems from applications. ACM Trans. Math. Software
19:333359
Durbin F (1976) Numerical inversion of Laplace transforms: An
efcient improvement to Dubner and Abate's method. The
Computer Journal 17:371376
Eason G (1966) The displacements produced in an elastic half-
space by a suddenly applied surface force. J. Inst. Maths Appl.
2:299326
Eringen AC, Suhubi ES (1975) Elastodynamics, vol. 2, New York:
Academic Press
Fung YC (1965) Foundations of Solid Mechanics, Englewood
Cliffs, NJ: Prentice-Hall
Gakenheimer DC (1970) Numerical results for Lamb's point load
problem. ASME J. Appl. Mech. 37:522524
Gakenheimer DC, Miklowitz J (1969) Transient excitation of an
elastic half-space by a point load traveling on the surface.
ASME J. Appl. Mech. 36:505515
Georgiadis HG (1993) Shear and torsional impact of cracked
viscoelastic bodies A numerical integral equation/transform
approach. Int. J. Solids Structures 30:18911906
Georgiadis HG, Rigatos AP (1996) Transient SIF results for a
cracked viscoelastic strip under a concentrated impact load-
ing An integral-transform/function-theoretic approach.
Wave Motion 24:4157
Georgiadis HG, Brock LM, Rigatos AP (1998) Transient con-
centrated thermal/mechanical loading of the faces of a crack
in a coupled-thermoelastic solid. Int. J. Solids Structures
35:10751097
Georgiadis HG, Vardoulakis I (1998) Anti-plane shear Lamb's
problem treated by gradient elasticity with surface energy.
Wave Motion 28:353366
Greenberg FD (1975) Foundations of Applied Mathematics, New
Jersey: Prentice Hall
Johnson LR (1974) Green's function for Lamb's problem. Geo-
phys. J. Royal Astron. Soc. 37:99131
Kupradze VD (1963) Dynamical Problems in Elasticity, Amster-
dam: North-Holland
Lamb H (1904) On the propagation of tremors over the surface of
an elastic solid. Phil. Trans. Royal Soc. A203:142
Laturelle FG (1991) The stresses produced in an elastic half-space
by a pressure pulse applied uniformly over a circular area:
role of the pulse duration. Wave Motion 14:19
Levin D (1973) Development of non-linear transformations for
improving convergence of sequences. Int. J. Comp. Math.
B3:371388
Longman IM (1956) Note on a method for computing innite
integrals of oscillatory functions. Proc. Camb. Phil. Soc.
52:764768
Mikhailenko BG (1985) Numerical experiment in seismic inves-
tigations. J. Geophysics 58:104124
MacDonald JR (1964) Acceletated convergence, divergence, iter-
ation, extrapolation and curve tting. J. Appl. Phys. 35:3034
3041
Manolis GD, Beskos DE (1988) Boundary Element Methods in
Elastodynamics, London: Unwin-Hyman (Chapman & Hall)
Miklowitz J (1978) The Theory of Elastic Waves and Waveguides,
Amsterdam: North-Holland
Mooney HM (1974) Some numerical solutions for Lamb's prob-
lem. Bull. Seism. Soc. Amer. 64:473491
Narayanan GV, Beskos DE (1982) Numerical operational meth-
ods for time-dependent linear problems. Int. J. Num. Meth.
Eng. 18:18291854
Fig. 14. Variation of the normalized vertical displacement with
the normalized time for triangular-type loading and r=z = 0:577
(i.e. u = 30

), m = 0:25
98
Nkemzi D (1997) A new formula for the velocity of Rayleigh
waves. Wave Motion 26:199205
Okecha GE (1987) Quadrature formulae for Cauchy principal
value integrals of oscillatory kind. Maths of Comp. 49:259
268
Pekeris CL (1955) The seismic surface pulse. Proc. Natl. Acad. Sci.
41:469475
Press WH, Flannery BP, Teukolsky SA, Vetterling WT (1986)
Numerical Recipes, Cambridge: Cambridge University Press
Rahman M, Barber JR (1995) Exact expressions for the roots of
the secular equation for Rayleigh waves. ASME J. Appl. Mech.
62:250252
Rivlin M, Hayes RS (1962) A note on the secular equation for
Rayleigh waves. ZAMP 23:8083
Sidi A (1982) The numerical evaluation of very oscillatory innite
integrals by extrapolation. Maths of Comp. 38:517529
Xu P, Mal AK (1985) An adaptive integration scheme for irreg-
ularly oscillatory functions. Wave Motion 7:235243
99

You might also like