Soil Dynamics & Earthquake Engineering 15 (1996) 387-397

Copyright 1996 Elsevier Science Limited

ai,W PII: S0267-7261(96)0001 3-9
Printed in Great Britain. All fights reserved
0267-7261/96/$15.00
ELSEVIER

Scattering of waves by subterranean structures via

the boundary element method
Ant6nio J. B. Tadeu
University of Coimbra, Department of Civil Engineering, 3000 Coimbra, Portugal

Eduardo Kausel
Massachusetts Institute of Technology, Department of Civil and Environmental Engineering, Cambridge, MA 02139, USA

&

Christos Vrettos
Technical University of Berlin, Section of Geotechnical Engineering, 10623 Berlin, Germany

(Accepted for publication 2 April 1996)

Tile boundary element method (BEM) is used to study the two-dimensional wave
field generated when buried structures of arbitrary shape (i.e. inclusions) in an
ehtstic medium are illuminated (or insonified) by dynamic line sources. Both
steady-state responses and time-domain transients are presented. The problem is
formulated in the frequency domain by means of appropriate Green's functions.
The evaluation of the singular integrals is achieved (and to the best of the
writers' knowledge, for the first time in the technical literature) in analytical
form, which results in improvements in computational efficiency and accuracy.
C},osed-form solutions for regular geometries are then used to validate the
method. The interaction of two cavities, the formation of shadow zones by
inclusions and the complexity of the scattered field from bodies with irregular
shapes are used as examples to demonstrate the versatility of the method. The
responses computed in the time domain were invariably found to be causal,
even for non-convex domains, which belies a recent assertion by some
researchers that the application of boundary element methods to concave
domains is associated with non-causal effects. Copyright 1996 Elsevier
Science Limited

Key words: wave propagation, soil, boundary-element method, embedded

structures, dynamic response.

INTRODUCTION subterranean structures and cavities as well as variations

M a n y studies in the past two decades have addressed the
phenomenon of wave propagation in the vicinity of
geological and topographical irregularities. While most
previous efforts have been devoted within the field of
engineering seismology, some recent researches have
also focused on the development of dynamic testing
procedures for the detection of subsurface anomalies in
the ground; such methods are generically referred to as
seismic tomography. The principal objective in seismic
t o m o g r a p h y is the location and identification of
387
in soil properties by means of acoustic (or even
electromagnetic) sources and receivers placed on the
ground's surface or inside holes. Such testing methods
have been aided by the enormous progress made in
recent years in computer-based data acquisition and
storage capabilities as well as in signal processing. This
has in turn stimulated research aimed at developing
appropriate inversion algorithms for the extraction of
accurate images of the subterranean structures from the
collected data. Clearly, such inverse problems pose great
theoretical difficulties and challenges, because one
388 A. J. B. Tadeu, E. Kausel, C. Vrettos
cannot establish a priori that their solutions are unique
or even exist, in contrast to direct problems, for which
such attributes are usually taken for granted.
Some of the first analytical studies on wave diffraction
and scattering were concerned with the problem of wave
motion and reverberations in alluvial basins of regular
shape 1'2 and with the issue of wave scattering induced by
cavities. 3-6 More recently, semi-analytical methods have
been used also for the analysis of wave diffraction
caused by geological irregularities of arbitrary shape
within globally homogeneous media. 7-9 By contrast, the
application of purely numerical methods (i.e. finite
elements or differences combined with transmitting
boundaries) has been restricted, for the most part, to
situations where the response is required only within
localized irregular domains, as is the case in soil-
structure interaction problems. 1-12 Discrete methods
have occasionally been used also for the modeling of
large alluvial basins, but only in plane-strain. 13 Finally,
hybrid methods involving combination of finite elements
to model the interior domain containing the inhomogene-
ities and semi-analytical representations for the exterior
domain have been used as well. 14 A detailed review on the
above methods was presented by Sanchez-Sesma. 15
Perhaps the best tool for the analysis of wave
propagation problems in unbounded, infinitely large
media is the boundary element method (BEM); the
reason can be found in its inherent ability in satisfying
the far-field radiation conditions and its intrinsic virtue
allowing description of the medium merely in terms of
nodes at material discontinuities. Depending on how the
algorithm is formulated, it is customary to distinguish
between direct and indirect boundary element methods.
In the direct method, the relationship between displace-
ments and forces on the physical boundary result from
consideration of virtual point sources and singular
integrals on the boundary, while in the latter they are
obtained by means of fictitious source densities with no
immediate physical significance. 16'17 The majority of
applications is based on the direct formulation.
Whatever the choice, boundary element methods
require the availability of appropriate fundamental
solutions, or Green's functions, relating the field
variables (stresses, displacements) in a homogeneous
medium to sources placed at some location in the
medium. The fundamental solution most often used is
that of an infinite homogeneous space, because it is
known in closed-form and has a relatively simple
structure. While the Green's functions for a half-space
satisfy automatically the zero stress conditions at the
free surface so that no boundary elements are required
here, their disadvantage is that they can only be
obtained by semi-analytical means. Full-space funda-
mental solutions, on the other hand, are much simpler
and easier to implement. Boundary element methods
based on the Green's functions for either an infinite
space or a half-space have been used in the past to
solve a variety of problems involving diffraction of
waves by surface irregularities of arbitrary shape. 18-2s
Boundary elements have also been used in studies on
plane harmonic or transient body and surface wave
diffraction by cavities and buried,structures. 29-36 For a
recent review of this subject, the reader is referred to
Beskos. 37
A major difficulty associated with the application of
the direct boundary element method lies in the evalua-
tion of the singular integrals, a topic which has received
a great deal of attention in the technical literature. 38 One
strategy used to overcome this difficulty consists in
decomposing the integrand into two parts, namely a
singular and a nonsingular component. The singular
part is determined analytically, while the non-singular
part is evaluated numerically by means of a standard
Gauss quadrature. An alternative approach 39'4 consists
in combining the direct boundary element method with
discrete wave-number representations of the Green's
functions;41 while this strategy allows evaluating the
element integrals in analytical form for each wave-
number, this advantage comes at the expense of having
to carry out cumbersome numerical integrations over
wave-number.
In the sections that follow, the direct boundary
element method is used together with the two-dimensional
full-space Green's functions in an evaluation of the
displacement field associated with waves illuminating a
cavity or inclusion within a homogeneous elastic
medium (i.e. a full-space, a half-space and a homo-
geneous slab). The problem is formulated in the
frequency domain, and the response in the time
domain is obtained by means of the fast Fourier
transform. Particular attention is given to the deriva-
tion of analytical expressions for the singular integrals.
After validation, the boundary element algorithm is
used to study four wave diffraction problems involving
inclusions within homogeneous media.
BOUNDARY ELEMENT FORMULATION
Considering the voluminous literature currently available
on the boundary element method, it does not appear
necessary to repeat yet again the details of the formulation
required for the type of scattering problems presented
herein (see for example, Ref. 42). It suffices to state that
the BEM requires evaluation of the two integrals
= fc, a,j( k, x,)dG (1)
H~' =/G Hij(Xl~,xl, n,)dCt (2)
in which Gij(xk, xt) and Hu(xk, xt, nl) are, respectively,
the components of the Green's tensor for displacement
and traction components at xk in direction i due to a
 Subterranean structures and the BEM
concentrated load at xt in direction j, and nl is the unit
outward normal for the/th boundary segment C1.
In the case of a medium under plane-strain conditions
subjected to in-plane and anti-plane line loads, the
required Green's functions are as follows: 43
(a) Anti-plane unit line load (i,j = 3):
G33(x , Xo) = + H (2) (k~r)
(b) In-lane unit line load (i,j = 1,2):
Gij(X,Xo)=+ {-'ijH(2)(kor)-l-~)r [Hl2)(kj3r)-'~Hl2)(kar)]}
i for Or [H(E2)(k~r)_,72H~2)(k~r)]}
4~ [Oxi Oxj
In these equations 7 = / 3 / a , where a = v/(A + 2#)/p
and/3 = ~ are the velocities of compressional and
shear waves, respectively; A and # are the Lam~
constants; p is the mass density; ks =w/a and
ka = w//3 are the wave-numbers; w is the circular
frequency; r = I x - x0[ is the source-receiver distance;
6o- is Kronecker's delta; the leading factor i = C'L-]- is
the imaginary unit; H~(2)( ) are Hankel functions of the
second kind and n-th order. These Hankel functions,
together with an implicit exponential factor eiwt repre-
sent waves in the far-field that travel outwardly toward
infinity. (Note: some researchers instead formulate the
Green's functions using a factor e -~'t and Hankel
functions of the first kind H~fl)( ) in eqns (3) and (4);
such choice requires a reversal of the usual positive sign
in the Fourier transformation from the frequency
domain into the time domain.)
The corresponding expressions for the tractions Hij,
which may be obtained from G;j by taking partial
derivatives to deduce the strains and then applying
Hooke's law to obtain the stresses, are omitted here for
the sake of brevity.
v i [
/
~
nl
= +L
t
,<
S =- -L
X2
Fig. 1. Local coordinate system for element integration.
389
Element integration
When the element to be integrated in eqns (1) and (2) is
not the loaded element, the integrands are non-singular
and the integrations are best carried out using standard
Gaussian quadrature. For the loaded element, however,
the integrands exhibit a singularity, but it is then
possible to carry out the integrations in closed form,
(3) as will be shown.
To demonstrate this assertion, consider the lth
singular element of length 2L shown in Fig. 1 for the
anti-plane loading case. Since for this case r . n / = 0,
the singular terms H ~t3 vanish. On the other hand, the
quantities G ~t3 can be evaluated from the expressions
(Ref. 44, p. 480)
(4)
f+LLH~2)(kflr)ds:2/L[jo(k~r)-iYo(k~r)]ds
with b = k~L, and
Ii(b) =Jo(b)-t-2[So(b)Jl(b)-Sl(b)Jo(b)] (6a)
h(b)= Yo(b)+2[So(b)Y,(b)-S,(b)Yo(b)] (6b)
where So( ) and $1( ) are Struve functions, and Jn( )
and Yn( ) are n-th order Bessel functions of the first and
second kind, respectively.
A similar procedure can be followed to evaluate the
singular integrals for the in-plane case. In this case, it
can first be observed that the t er m s Hi11 for i , j = 1,2
(which correspond to the loaded segment) vanish
because the tractions are anti-symmetric with respect
to the nodal point at the center of the element.
Consider next the integration of G it~ for i,j = 1,2,
which in view of eqn (4) may be written as:
G~l = /? Gij(x, xl)ds
Xl
Or Or ]
= ~ 6i](-B1 + B2) Oxi Oxj B3 (7)
with
f L 2
B, = Jo H~ )(k~r)dr (8a)
B2=~oL+[Hl2)(kr)-'THl2)(kj)]dr (8b)
The partial derivatives Or/Oxi, Or/Oxj in eqn (7)
represent the slopes of the line connecting the nodal
point with the integration point. Since for the loaded
390 A. J. B. Tadeu, E. Kausel, C. Vrettos
element they equal the slope of the segment, it follows
that they are constant and do not affect the integration.
This is the reason for their placement outside of the
integral B3 in eqn (7).
The first integral (B,) is obtained directly from eqn
(5), that is,
B, = Bi(b) = L[Ii(b) - iI:(b)]
To evaluate BE and B3, it is necessary to use the well
known recurrence relations for the Bessel functions, and
to dispose of an apparent singularity at the lower limit
of the integrals by considering the ascending series for
the Bessel functions and observing that the terms
contributing to the singularity vanish exactly on account
of the fact that 7 = 171a = k~lk n. For example,
B2 = L i m I L { [ H ~ 2 ) ( k s r ) - ~ r r1 d H(2)/k
i t ~r)]
-(k~2[H~2)(kar)-l dHIE)(kar)])dr
= SoL[H:2)(kjr)-72H:2)(k=r)]dr
1
,-Lim I L [~rHl2)(kjr)--~nk=~Hl2)(k=r)]d r
= B, (k~ L) - 7 2B, (k~L)
1Lim[Hl2)(knr) -~Hl2)(k~r)]~
kn ~ o I
Lr ~ r 1
BE = B'(b)-')'2Bl(a)--b H
| ~a)(bT)-'HEi)(a)Li -I
with a = k~L. By similar manipulations, one obtains
B3=BI(b)-72BI(a)-2--~-~ [HI2)(b)-THI2)(a)]
Equations (5) and (9)-(11) provide the exact integrals
for the singular terms associated with line sources in a
two-dimensional homogeneous space (plane-strain). To
the best of the writers' knowledge, this constitutes the
first such symbolic representation. In principle, similar
equations could also be written for a half-space, a
stratum, or a slab under anti-plane loads, because the
Green's functions for such problems can be obtained
from those of the homogeneous full-space by recourse to
the method of images.
Evaluation of response in the time-domain
In the following sections, the method outlined pre-
viously is applied to the solution of four scattering
problems. In each case, the computations are first
performed in the frequency-domain, and then Fourier-
transformed into the time-domain by means of the well
known FFT algorithm. The variation with time of the
dynamic sources is assumed to be given by a Ricker
wavelet, as defined below. This wavelet is chosen not
only because it decays very rapidly in both time and
frequency, but because it facilitates the interpretation of
(9) the computed time signatures, as they are then localized
in space-time. The Ricker wavelet function is given by
u(7-) = A(1 - 2T2)e-r2 (12a)
where A is the amplitude, -c = ( t - ts)/t o and t denotes
time; ts is the time at which the maximum occurs, while
zrt0 is the characteristic (dominant) period of the
wavelet. Its Fourier transform is
U(w) = A[2xl-~toe-i~ts]l) 2 e -: (12b)
in which fl = Wto/2.
As is well known, the frequency increment A f =
Aw/27r determines the total time duration for the
numerical analysis, namely T = 1/Af. If T is smaller
than the actual duration of motion, the response beyond
T wraps-around (i.e. is aliased) in the time window and
may be seen as a non-causal event occurring before the
first true wave arrivals, as detailed in textbooks on signal
analysis. One way to overcome this problem is to reduce
the frequency increment Af , but in most cases this
alternative may result in excessive computing time. A
more effective strategy is to use the complex frequency
method proposed by Phinney;45 in this method, a
fictitious attenuation is introduced by adding an
imaginary component to the frequency, i.e. & = ~ - iA,
(lO)
the effect of which is an exponential attenuation of the
response in the time domain. This fictitious damping is
then removed by multiplying the computed response by
the exponential window ea t. Clearly, A cannot be made
(11) too large, since doing so would lead to substantial
numerical errors that would show up toward the end of
the time window.46 The numerical calculations presented
herein were made with an imaginary component of
frequency = 0.7Aw, which implies a 1/80 attenuation
at the end of the time window t = T (i.e. the wrap-around
is attenuated by two orders of magnitude).
Validation of the algorithm
The method and expressions described previously were
implemented and validated by applying them to simple
problems with known analytical solution, namely a
cylindrical inclusion in an infinite, homogeneous space
subjected to SH waves (particle motion parallel to the
axis of the inclusion) and SV-P waves (particle motion
perpendicular to the axis). These problems were
modeled by means of boundary elements placed at the
 Subterranean structures and the B E M 391

Xl 0.30 . . . . . . . . . . . .

Ro,v : Receiver 3
(8,1) O
.-->
o
0.15

=e

E 0
t-
O
cJ

t1=200m/s o -0.15

[~l=100m/s
P l=lMg/m3 Imaginary part "~/ w x~
-0.30 , , , , = , , , i , ,
0 50 1 O0 a)

0.30

~ 0.15
Fig. 2. Cylindrical inclusion in an unbounded medium.

interface between the inclusion with the surrounding

medium, as shown in Fig. 2.
As a first test, the inclusion was assigned the same
material properties as the surrounding medium. Clearly,
this test should reproduce the free-field conditions, and
in fact it did. The program was subsequently tested by
considering an elastic inclusion with material properties
as listed in Fig. 2, and which is subjected to SV-P waves
caused by a blast (dilatational) source at point O. Such a
source elicits compressional waves with concentric
wave-fronts having a center at O, which in turn induce
particle motions in the radial directions. Upon impinge-
ment of this incident field on the inclusion, waves are
scattered in all directions. The motion components are
then computed and recorded at the indicated receivers,
both with the implemented boundary elements and with
the analytical solution reported by Pao & Mow. 47
The results are computed at 64 frequencies in the
range from 2 to 128 Hz, that is, for compressional (P)
waves whose wavelength are, respectively, between 33"33
and 0.5208 times the radius of the structure. Figure 3
E -0.15
x~
-0.30
0 50 100
b)
Frequency [Hz]
Fig. 3. Scattered field for a cylindrical inclusion in an
unbounded medium: (a) x2-component at receiver 1; (b) XE-
component at receiver 3. The solid lines are for the exact
closed-form solution, the dashed lines are for the BEM
solution.
displays the real and imaginary parts of the response for
the numerical and closed-form solutions and reveals
the very good accuracy of the BEM approach. The
comparison shows an excellent agreement for low
frequencies, and only slight differences at high
frequencies. This behavior was expected, because the
accuracy of the BEM solution depends on the ratio

Source 1 O(o,o) Source 2 0 (lo,o)

(-I0,5)(-8,5)(-6,5)(-45)(-2,5) (0,5) (2,5)(45) (6,5) (8,5)(IO,5)(12,5)(14,5)(16,5XI8.5)(20,5X22,SX24,5)(26,5)(28,5)(30,5)

+ + + + + + + + + + + + + + + + + + + + +
1 2 3 4 5 6 7 8 9 10 _ 1 1 12 13 14 15 16 17 18 19 20 21

KN~ Receivers

~l=lOOm/s

P] =lMg/m3

[rig. 4. Two cylindrical cavities in an unbounded medium.

392 A. J. B. Tadeu, E. Kausel, C. Vrettos

between the length of the incident waves and the length
of the boundary elements; in our example, this ratio
decreases from 265 to 4-1 in the range of frequencies
considered.
APPLICATIONS
Two cylindrical cavities in unbounded medium
Next, we consider the interaction of two cylindrical
cavities in a homogeneous medium when they are both
illuminated by SH (anti-plane) line sources placed first
at one position and then at another, as shown in Fig. 4.
The computations are performed in the range from 1 to
256 Hz with an increment of 1 Hz. Thus, the total time
duration of the numerical analysis is T = 1 s. A total of
220 boundary elements is used to model both cavities.
Each line source elicits an incident wave field with
cylindrical wavefronts and center at the source, which
consists of a Ricker wavelet with a characteristic
frequency of 100Hz. The motion is computed and
recorded at 21 receivers lying in a plane above the
cavities but below the sources, as indicated.
Figure 5 displays, for each source, the response
recorded at the receivers. The first wave arrivals
observed in this figure correspond to wave trains that
r
are directly reflected by each of the cavities. These are
followed by pulses having progressively lower ampli-
tudes, which are the result of reverberations between the
cavities. As these pulses reflect back and forth between
the cavities, they scatter and loose energy to the
surrounding medium, and eventually dissipate. By
measuring the arrival times of the pulses and echoes,
one can infer the relative location and size of the
cavities.
Another interesting aspect of this problem is the
creation of shadow zones behind the cavities. Figure 6
illustrates this phenomenon: it shows the amplitude of
the displacements in the surrounding medium that is
produced by a harmonically vibrating source in position
2. The total and scattered responses are shown as
contours in Fig. 6 for three values of the excitation
frequency, namely, 2, 50 and 100 Hz, which correspond
to incident waves with wavelengths of 50, 2 and 1 m,
respectively. The results for 2Hz do not show any
shadow zone, while those for 50 and 100Hz depict
distinct shadows behind the cavities. This behavior was,
of course, very much expected since it is very well known
that waves whose wavelength is large compared to the
obstacles they encounter suffer little scattering (i.e. they
do not see the inclusions). These drawings illustrate,
however, the virtues and advantages of the boundary
element formulation for the study of complicated wave
diffraction phenomena such as those being considered
here. They also clearly demonstrate the intense degree
of interaction between the cavities, which are the result
of the previously cited reverberations.

Irregularly shaped cavity in an unbounded medium

1 - , i i - ,

0 0.2 0.4 0.6 0.8 1.0 a)

One important aspect in the study of waves impinging
21 Z on obstacles is the effect of the shape of the inclusion on
the scattered wave-field. Basically, the question is
whether or not one can make inferences about the
oa) shape of the inclusion from measurements of the
rc m - ~, scattered field at selected recording stations.. As an
initial exploration of this subject, we consider a kidney-
1
0 0.2
, k ~ "
0.4 016 018 1.0 b)
shaped cavity in an unbounded elastic medium which is
illuminated, as before, by anti-plane (SH) line sources
placed at three different positions, as shown in Fig. 7.
The response is recorded at 25 receivers placed between
~v~-
the source and the cavity. Computations are once more
0 performed in the frequency range from 1 to 256 Hz, with
a frequency increment of 1 Hz, and the time variation of
the sources is a Ricker wavelet with a characteristic
0.2 0.4 0.6 0.8 1.0 c)
frequency of 100 Hz. A total of 400 boundary elements
are used to model the inclusion. The computed
T i m e [s] responses are displayed in Fig. 8.
Fig. 5. Time response for two cylindrical cavities in an Analysis of the results obtained reveals complicated
unbounded medium: (a) total field for excitation by source 1; wave patterns which originate in reverberations within
(b) scattered field for excitation by source 1; (c) scattered field the concave section of the cavity. The primary reflections,
for excitation by source 2. coming from the left and right sides of the inclusion (i.e.
 Subterranean structures and the BEM 393

F- 20 Hs 3D- .0 m CONOR,I"IP F= 2,0 Hs 30- .0 m CONDR.FIP

IO 1.0

40 40

"to 70

Ioo I0.0

130 13.0

~160 ~,,o
~190 ~18.0

~ 22.0

~ 2~.0

310

~,o F=2Hz \ ~ 3?0 F=2Hz/f

~ o - ~ -;.0 ~'0 ~b t*'* t~o ,8'o ~o%o'o ~oo ,oo oo ~o ~.'o e.'o 1o'.o ~4.o ,.'.o ~'.o ~ o ~o.
r- 60.0 Jla ~1~- .0 m COHDR.FIP W- 00,0 H i 81)- .0 m COfiDaR./~

_,oo O _: .o .o
- , . , , . - ..,o.o _,oo _.o _ ,.o ., o . .o. ,oo
. ,.o . , , . . . . .

I 0 r F- IOO,O H i [ 1 0 - .0 m CONDILnP 10 e F- I00.0 Ha 8D- 1.0 m COIMOI~IqP

'~I.0 " .~

-~o.o-.o -~.o ~o .o to.o t~'.o ~'.o -';.o ~e'.o ~o.o -ao.o-e.o -~o ~.o e.o ~o.o ~.o n.o ~.o ~.o :,,,.o
Total field. Range (m) ~eattered field. Range (m)

Fig. 6. Steady state response for two cylindrical cavities in an unbounded medium: total field (left) and scattered field (right).

the convex parts), are identified with the labels SL and
S D , respectively. The time-of-flight o f each o f these
pulses corresponds to the travel path o f a pulse f r o m the
source to the inclusion and back to the receivers. Hence,
the position o f the sources defines whether SL or S D
pulses are the first events to arrive at a given receiver
location, and ray theory can be used to verify the
accuracy o f the numerical prediction, at least as far as
arrival times o f pulses is concerned. Even then, a ray
analysis o f the c o n c a v e part can be c u m b e r s o m e indeed.
394 A. J. B. Tadeu, E. Kausel, C. Vrettos

Source 1 Source 2 1 Source 3

Q
(-16o0) (o,0) x1 (16,0)

(-12.5) Receivers
I
Tx2
- (12,5)
4. 4, 4- 4- 4- 4- 4- 4- 4- 4- 4- 4- 4- 4- 4- 4- 4- 4- 4- 4- ~ ~ ~ ~ ~
|
1

13 1 = lOOm/s
P 1 = 1Mg/m3

(0,15) ( 1 , 1 5 ) ~

Fig. 7. Irregularly shaped cavity in an unbounded medium.

Nonetheless, a careful analysis of the earliest wave

arrivals which are consistent with ray analysis, demon- 25
,_ = , -

strated that all computed responses were causal. In

other words, the signals did not arrive at the receivers
earlier than the shortest possible ray paths would have
allowed, even though the inclusion considered included
a concave region. This observation is in contradiction
with some recently expressed opinions that the use of t i

boundary element techniques to non-convex domains 012 0.4 0.6 0.8 1.0 a)

may lead to violations in causality.

25 I ' .. ,~" . ~ ,

I
Cylindrical cavity in a stratum
n-
The previous examples dealt with inclusions embedded
in an unbounded homogeneous medium. Applications
1 i
in the field of geomechanics, however, involve soils that 0 0.2 0.4 0.6 0.8 1.0

have a free surface and perhaps also a rock interface.

Thus, it is of interest to consider the problem of waves
scattered by inclusions in a homogeneous stratum. The
material discontinuities in this medium complicate the
problem in at least two ways. First, the numerical
solution becomes more involved because the Green's
functions must account for the presence of the bounding
interfaces,48 unless they are also modeled with boundary 1
0 0.2 0.4 0.6 0.8 1.0
elements (an option which increases the computational
burden). Second, these interfaces reflect the incident as Time Is]
well as the scattered waves, which leads to complex wave Fig. 8. Time response for an irregularly shaped cavity in an
patterns, as will be seen. unbounded medium: scattered field for three different source
Figure 9 shows a cylindrical cavity in a homogeneous positions: (a) source 1; (b) source 2; (c) s o u r c e 3.
 Subterranean structures and the B E M 395

I Source 10,o)
,,,
Fig. 9. Cylindrical cavity in a stratum.
soil layer underlain by rigid bedrock. The cavity is
modeled with 140 boundary elements, and is illuminated
from the free surface by two SH Ricker wavelet sources,
one placed directly above the cavity, the other at a
horizontal distance of 20m. Computations are per-
formed in the frequency range from 1 Hz to 128 Hz, with
a frequency increment of 0.5Hz. The surface and the
interface between the stratum and the rigid bedrock
were not discretized with boundary elements because the
Green's function used were based on the method of
virtual images (for details, see Ref. 48).
Total field
_ZLe'ss/ . . . .
Sovtc 2 (2o,o) Figure 10 depicts the total and scattered displacement
I l l~lll I l I I I I llil
,, i, , . . . .
~:iv~Ts
,1 , , w
fields for this case. In contrast to the problems
considered previously, the pulse trains at each receiver
in
l~*=lOOm/s
p 1 =IMghn3
result now from the reflection and diffraction of waves
at the boundary of the cavity, at the free surface and at
35m the rigid interface. The incident and the first two
reflected pulses from bedrock are tagged with labels I,
P, and PP, respectively, while those that reflect once or
twice at the surface are identified with labels S and SS. It
is interesting to observe that each time a pulse hits
bedrock the reflected pulse changes phase. The primary
reflection, which results as the incident pulse strikes the
boundary of the cavity and returns to the surface, is very
clear in the plots.
When the source is above the cavity (source 1), the
total response at the receivers located close to it do not
provide information about the rigid interface: in this
case the cavity acts as a barrier to the waves and creates
a clear shadow zone underneath. As the distance of the
receivers from the source increases, the reflections
returning from bedrock start to be visible. When the
source is located at position 2, however, the reflections
from the rigid interface are dearly evident, although
now some shadowing becomes evident at receivers 4 to 8
Scattered field on the side opposite to the source.
":
Cyfindrical cavity in a half-space

As a final application example, consider a cylindrical

.>_ ~-,-----,-4 cavity in a half-space illuminated by a dilatational line
,~
o=
0~
, . 4

t source (P-waves) placed between the surface and the

cavity, as detailed in Fig. 11. The source generates a
A m, ---
Ricker wavelet pressure pulse with a center frequency of
22Hz. Calculated frequencies range from 1 to 6 4 H z
~___,,_______~

, , i , h , , , i
3, ,s ,7, ,?,11 Receivers
0.5 1.0 1.5 0.5 1.0 1.5

Time [s] I.-.4

5m
- iOm
1. . . . s '/,s's . . . . iPp

4 ~ ",' 4.- lOOm "~ Source

~l--400m/s
.t.
al =800ra/s
, q- Pl =lMg/n~
~........~ A
,m

b)
, , , , i , , , , i

0.5 1.0 '1.5 0.5 1.0 1.5

Time [s]

Fig. 10. Time responses for a cavity in a stratum. Total field

(left) and scattered field (right) for two different source
positions: (a) source 1; (b) source 2. Fig. 11. Cylindrical cavity in a half-space.
396 A. J. B. Tadeu, E. Kausel, C. Vrettos

;:2 component be used to in the context of research on seismic

11 tomography problems, the results of which will be
reported in a companion paper. In the first phase
presented here, this program was used to compute the
.>_ ~v
(!) wave-fields in the vicinity of obstacles (inclusions)
0
11)
n-
embedded in homogeneous media when they are
illuminated by acoustic (SH and SV-P) line sources in
various relative positions. The cases considered
1 included cavities and elastic media embedded in a
i i I i full-space, in a half-space, and in a stratum. The
o 0.5 1.0 results obtained were consistent with predictions by
ray acoustics, and were used by the writers to elucidate
the most important aspects of wave acoustics, with an
i x 1 component
11 eye on the development of non-destructive testing and
imaging methods.
An important observation made in the course of this
.> research is that the response signals computed in the
(1)
O
(1)
time domain are invariably causal, even for non-convex
rr
domains. Hence, this result belies the controversial view
held by some researchers that the boundary element
method fails to satisfy causality for non-convex
i
domains. It is conceivable, however, that computations
i

o 0.5 1.0 made with coarse discretizations (i.e. few boundary

Time [s]
Fig. 12. Time responses for a cylindrical cavity in a half-space.
with an increment of 1 Hz. We compute the response at
receivers placed on the surface, using boundary elements
with Green's functions for the elastic half-space
obtained via discrete wave-number summation (see
Ref. 48 for details).
Figure 12 shows the vertical and horizontal displace-
ments at the receivers, which in the near field result
mainly from body waves on account of the depth of the
source. Because of symmetry, the horizontal response is
zero at receiver 1. As waves impinge on both the cavity
and the free surface, they scatter back into the medium
as P and SV waves (the latter as a result of P-SV
conversion). The first set of pulses recorded at the
receivers correspond to the incident field, second arrivals
are due to the initial P waves reflected from the cavity~
while third arrivals are SV mode converted waves
resulting from the initial P wave incidence on the
cavity. Higher-order reflections can also be seen and can
be identified by computing the travel time for different
ray paths and mode conversions.
CONCLUSIONS
A boundary element formulation for plane-strain
elastodynamic problems was developed and used to
efficiently evaluate problems of wave diffraction and
scattering. The aim was to implement a tool that could
elements) may lead to errors that could be misconstrued
as violations of causality.
Of central importance to the work presented in this
paper, and to the theory of boundary dements in
general, is the evaluation of the singular integrals, which
was here achieved in closed form. To the best of the
writers' knowledge, this is the first such evaluation,
which holds the promise of improved accuracy and
efficiency for elastodynamic problems.
REFERENCES
1. Trifunac, M. D. Surface motion of a semi-cylindrical
alluvial valley for incident plane SH waves, Bull. Seism.
Soc. Am., 1971, 61, 1755-70.
2. Wong, H. L. & Trifunac, M. D. Surface motion of a semi-
elliptical alluvial valley for incident plane SH-waves, Bull.
Seism. Soc. Am., 1974, 64, 1389-1403.
3. Lee, V. W. On deformations near circular underground
cavity subjected to incident plane SH waves, Syrup. Appl.
Computer Methods in Eng., Univ. of Southern California,
Los Angeles, 1977, pp. 951-61.
4. Datta, S. K. & Shah, A. H. Scattering of SH-waves by
embedded cavities, Wave Motion, 1982, 4, 265-83.
5. Lee, V. W. Three-dimensional diffraction of elastic waves
by a spherical cavity in an elastic half-space. I: Closed-
form solutions, Soil Dyn. Earthq. Eng. 1988, 7, 149-61.
6. Lee, V. W. & Karl, J. A. Diffraction of SV waves by
underground circular, cylindrical cavities, Soil Dyn.
Earthq. Eng., 1992, 11, 445-56.
7. Sanchez-Sesma, F. J. Diffraction of elastic waves by three
dimensional surface irregularities, Bull. Seism. Soc. Am.,
1983, 73, 1621-36.
8. Moeen-Vaziri, N. & Trifunac, M. D. Scattering and
diffraction of plane P and SV waves by two-dimensional
inhomogeneities: Part II, Soil Dyn. Earthq. Eng., 1988, 7,
189-200.
 Subterranean structures and the B E M
9. Lee, V. W. & Wu, X. Application of the weighted residual
method to diffraction by 2-D canyons of arbitrary shape:
II. Incident P, SV and Rayleigh waves, Soil Dyn. Earthq.
Eng., 1994, 13, 365-75.
10. Waas, G. Linear two-dimensional analysis of soil
dynamics problems in semi-infinite layered media, PhD
dissertation, University of California, Berkeley, California,
1972.
11. Lysmer, J., Udaka, T., Seed, H. B. & Hwang, R. L U S H - -
A computer program for complex response analysis of
soil-structure systems, Report No. EERC 74-4, Earthquake
Engineering Research Center, University of California,
Berkeley, California, 19'74.
12. Kausel, E. Forced vibrations of circular foundations in
layered media, MIT Research Report 74-11, Department of
Civil Engineering, Massachusetts Institute of Technology,
Cambridge, Massachusetts, 1974.
13. Ohtsuki, A. & Harumi, K. Effect of topography and
subsurface inhomogeneities on seismic SV waves, Earthq.
Eng. struct. Dyn., 1983,, 11, 441-62.
14. Shah, A. H., Wong, K. G. & Datta, S. K. Diffraction of
plane SH waves in a half-space, Earthq. Eng. struct. Dyn.,
1982, 10, 519-28.
15. Sanchez-Sesma, F. J. Site effects on strong ground motion,
Soil Dyn. Earthq. Eng., 1987, 6, 124-32.
16. Beskos, D. E. Boundary element methods in dynamic
analysis, Appl. Mech. Rev., 1987, 40, 1-23.
17. Kobayashi, S. Elastodynamics. In Boundary Element
Methods in Mechanics, ed. D. E. Beskos, North-Holland,
Amsterdam, 1987, pp. 192-255.
18. Wong, H. L. & Jelmings, P. C. Effect of canyon
topography on strong ground motion, Bull. Seism. Soc.
Am., 1975, 65, 1239-5"7.
19. Sanchez-Sesma, F. J. & Rosenblueth, E. Ground motion
on alluvial valleys under incident plane SH waves, Earthq.
Eng. struct. Dyn., 1979, 7, 441-50.
20. Dravinski, M. Scattering of plane harmonic SH wave by
dipping layers of arbitrary shape, Bull. Seism. Soc. Am.,
1983, 73, 1303-19.
21. Niwa, Y. & Hirose, S. Scattering problems of plan waves
in wedge-shaped elastic domains, Theor. appl. Mech., 1985,
33, 393-406.
22. Dravinksi, M. & Mossessian, T. K. Scattering of plane
harmonic P, SV and Rayleigh waves by dipping layers of
arbitrary shape, Bull. Seism. Soc. Am., 1987, 77, 212-35.
23. Mossesian, T. K. & Dravinski, M. Scattering of elastic
waves by three-dimensional surface topographies, Wave
Motion, 1989, 11, 579--92.
24. Vogt, R. F., Wolf, J. P. & Bachmann, H. Wave scattering
by a canyon of arbitrary shape in a layered half-space,
Earthq. Eng. struct. Dyn., 1988, 16, 803-12.
25. Luco, J. E., Wong, HI. L. & de Barros, F. C. P. Three
dimensional response of a cylindrical canyon in a layered
half-space, Earthq. Eng. struct. Dyn., 1990, 19, 799-817.
26. Sanchez-Sesma, F. J. & Campillo, M. Diffraction of P, SV
and Rayleigh waves by topographical features: a boundary
integral formulation, Bull. Seism. Soc. Am., 1991, 81,
2234-53.
27. de Barros, F. C. P & Luco, J. E. Diffraction of obliquely
incident waves by a cylindrical cavity embedded in a
layered viscoelastic half-space, Soil Dyn. Earthq. Eng.,
1993, 12, 159-71.
28. Pedersen, H. A., Sanchez-Sesma, F. J. & Campillo, M.
Three-dimensional scattering by two-dimensional topo-
graphies, Bull. Seism. $oc. Am., 1994, 84, 1169-83.
29. Niwa, Y., Kobayashi, S. & Fukul, T. Applications of
integral equation method to some geomechanical problems.
397
In Numerical Methods in Geomechanics, ed. C. S. Desai,
ASCE, New York, 1976, pp. 120-131.
30. Kobayashi, S. & Nishimura, N. Transient stress analysis of
tunnels and caverns of arbitrary shape due to travelling
waves. In Developments in Boundary Element Methods--II,
ed. P. K. Banerjee and R. P. Shaw, Applied Science,
London, 1982, pp. 177-210.
31. Manolis, G. D. A comparative study on three boundary
element method approaches to problems in elasto-
dynamics, Int. J. num. Meth. Eng., 1983, 19, 73-91.
32. Rice, J. M. & Sadd, M. H. Propagation and scattering of
SH-waves in semi-infinite domains using a time-
dependent boundary element method, J. appl. Mech.,
1984, 51, 641-45.
33. Rizzo, F. J., Shippy, D. J. & Rezayat, M. A boundary
integral equation method for radiation and scattering of
elastic waves in three dimensions, Int. J. num. Meth. Eng.,
1985, 21, 115-29.
34. Gonzalves, I. R., Shippy, D. J. & Rizzo, F. J. Direct
boundary integral equations for elastodynamics in 3-D
half-spaces, Comput. Mech., 1990, 6, 279-92.
35. von Estorff, O., Stamos, A. A., Beskos, D. E. & Antes, H.
Dynamic interaction effects in underground traffic
stystems, Eng. Analysis Boundary Elements, 1991, 8,
167-75.
36. Bencheikh, L. Modified fundamental solutions for the
scattering of elastic waves by a cavity, Int. J. num. Meth.
Eng., 1993, 36, 3283-302.
37. Beskos, D. E. Wave propagation through ground. In
Boundary Element Techniques in Geomechanics, ed. G. D.
Manolis and T. G. Davies, Computational Mechanical
Publications, Southampton and Elsevier Applied Science,
London, 1993, pp. 359-406.
38. Hall, W. S. Integration methods for singular boundary
element integrands. In Boundary Elements X, ed. C. A.
Brebbia, Computational Mechanical Publications,
Southampton, 1989, Vol. 1, pp. 219-36.
39. Kawase, H. Time-domain response of semi-circular
canyon for incident SV, P and Rayleigh waves calculated
by the discrete wavenumber boundary element method,
Bull. Seism. Soc. Am., 1988, 78, 1415-37.
40. Kim, J. & Papageorgiou, A. Discrete wave-number
boundary-element method for 3-D scattering problems,
J. eng. Mech. ASCE, 1993, 119, 603-24.
41. Bouchon, M. & Aki, K. Discrete wave-number representa-
tion of seismic-source wave field, Bull. Seism. Soc. Am.,
1977, 67, 259-77.
42. Manolis, G. D. & Beskos, D. E. Boundary Element
Methods in Elastodynamics, Unwin Hyman (now Chapman
& Hall), London, 1988.
43. Dominguez, J. & Abascal, R. On fundamental solutions
for the boundary integral equations method in static and
dynamic elasticity, Eng. Analysis, 1984, 1, 128-34.
44. Abramowitz, M. & Stegun, I. A. Handbook of Mathematical
Functions, Dover, New York, 1964.
45. Phinney, R. A. Theoretical calculation of the spectrum of
first arrivals in layered elastic mediums, J. Geophys. Res.,
1965, 70, 5107-23.
46. Kausel, E. & Roesset, J. M. Frequency domain analysis
of undamped systems, J. eng. Mech. ASCE, 1994, 118,
721-34.
47. Pao, Y. H. & Mow, C. C. Diffraction of Elastic Waves and
Dynamic Stress Concentrations, Crane and Russak, 1973.
48. Tadeu, A. J. B. Modelling and seismic imaging of buried
structures, PhD Thesis, M.I.T., Department of Civil
Engineering, Cambridge, Massachusetts, 1992.