ArticleGJI 2005 2

Geophys. J. Int. (2005) 163, 192224 doi: 10.1111/j.1365-246X.2005.02713.
Two-dimensional ground motion at a soft viscoelastic layer/hard

substratum site in response to SH cylindrical seismic waves radiated
by deep and shallow line sourcesII. Numerical results
Jean-Philippe Groby1 and Armand Wirgin2

1 Laboratoire de Mecanique et dAcoustique, UPR 7051 du CNRS, 31 chemin Joseph Aiguier, 13402 Marseille cedex 20, France
2 LMA/CNRS, 31 chemin Joseph Aiguier, 13402 Marseille cedex 20, France. E-mail: wirgin@lma.cnrs-mrs.fr
Accepted 2005 June 14. Received 2005 February 9; in original form 2004 January 23
SUMMARY
We consider, using theory and associated synthetic seismograms, the seismic response of a
site comprising a horizontal, homogeneous, soft viscoelastic layer of infinite lateral extent
overlying, and in welded contact with, a homogeneous, hard elastic substratum of half-infinite
radial extent. We show that for shear-horizontal motion: (1) coupling to Love modes is all
the weaker the deeper the source (modelled as a line, assumed to lie in the substratum) is
from the lower boundary of the soft layer, (2) for a line source close to the lower boundary of
the soft layer, the ground response is characterized by possible beating phenomena, and is of
significantly longer duration than for excitation by cylindrical waves radiated by deep sources.
Numerical applications of the theory show, for instance, that a line source, located 40 m below
the lower boundary of a 60-m-thick soft layer in a hypothetical Mexico-City-like site, radiating
a SH pulse of 4 s duration, produces substantial ground motion during 200 s, with marked
beating, at an epicentral distance of 3 km. Results from this modelling study are supported by
GJI Seismology
field examples taken from published literature.

Key words: duration, interference maxima, Love modes, regional path effects, site response,
source position.
1 I N T RO D U C T I O N
This investigation is relevant to two topics of broad interest in seismic wave propagation.
1.1 Regional path effects
The first topic is that of regional path effects in connection with seismic response in urban environments (Hisada et al. 1988; Novikova &
Trifunac 1993, 1995; Faeh et al. 1994; Furumura & Kennett 1998; Panza et al. 2001; Cardenas & Chavez-Garcia 2003; Balendra & Kong
2004; Celebi 2004; Savage 2004; Shoji et al. 2004).
Research on this topic was rekindled by efforts to explain some puzzling features of the devastating Michoacan earthquake which struck
Mexico City in 1985 (Furumura & Kennett 1998). Other than the fact that the response in downtown Mexico varied considerably in a spatial
sense, was quite intense and of very long duration (as much as 3 min) at certain locations, and often took the form of a quasi-monochromatic
signal with beating, a remarkable feature of this earthquake was that such strong (in the sense just mentioned) response could be caused by
a seismic disturbance so far (its epicentre was in the subduction zone off the Pacific coast approximately 350 km) from the city (Furumura
& Kennett 1998). A part of the cause of the large intensity and long duration (see Fig. 1) was attributed to multipathing between the source
and the site (Singh & Ordaz 1993), while being associated (Chavez-Garcia & Bard 1994) with surface wave propagation of the Rayleigh and
Love types, presumably between the source and the entry to the Mexico City basin, via the intervening crust.
Numerical results obtained by Faeh et al. (1994) and Faeh & Panza (1994) with a rather complete, 2-D hybrid model of the propagation
path between the source and the Mexico City basin, and of the action of the basin on the incident wave, stressed the important role of regional
path effects on anomalous response. The 3-D equivalent of these numerical simulations, carried out by Furumura & Kennett (1998), backed
up these findings and stressed the role of higher-order surface waves, which propagate in the relatively high Q layer of the Trans Mexican
Volcanic Belt (TMVB, composed of low-velocity volcanic lava and tuff overlying higher-velocity limestone) underlying/surrounding the soft
clay basin of Mexico City in producing large response (particularly with respect to duration) in the city. More recently, an analysis (Cardenas
192
C 2005 RAS
Seismic site response: a canonical problem II 193
Figure 1. Regional path effects as described in Furumura & Kennett (1998) for an off-the-southwestern Mexican coast to Mexico City regional path. The
various curves represent time histories of the transverse component of ground velocity at increasing epicentral distances. One notes the generally increasing
peak ground velocity and duration with increasing epicentral distance. This behaviour agrees with that observed by Shoji et al. (2004) for regional paths in
Japan, and Novikova & Trifunac (1993, 1995) for regional paths in California. One also notes the long duration of the wave just before hitting Mexico City
and the comparably long duration of the motion within the city (at station RMCS). This gives evidence for the fact that a part, or most, of the long duration
of motion in the city is due to the long duration of the incoming wave, the latter having travelled horizontally over a considerable (regional) distance from the
location of the source.
Figure 2. Regional path effects as described in Furumura & Kennett (1998) for: (1) an off the southwestern Mexican coast source-to-Mexico City recording
location-regional path (dots) and (2) an off the southwestern Mexican coast source-to-coastal recording location-regional path (circles). One notes (beyond
150 km up to an including observer locations within Mexico City) the increasing peak ground velocity with increasing hypocentral distance.
& Chavez-Garcia 2003) of seismograms recorded at various sites in central Mexico, for earthquake sources located in the subduction zone
off the Pacific coast, has shown that the crustal structure (including that of the TMVB) between the source and observation points acts as
a waveguide for surface waves coming from distances greater than 200 km, leading, by an unexplained mechanism, to amplification and
increase of duration of motion at various sites, notably in Mexico City.
Anomalous response in other cities such as Beijing, Bucharest, Rome, etc. has been studied in great detail, principally in a numerical
manner, within the framework of the UNESCO-IGCP project 414 (Panza et al. 2001). The features of this response were attributed to the
specifics of the source parameters, regional path effects, and the soil distribution and geometry in the urban basins (see next paragraph).
1.2 Effects of ground and underground irregularities
The second topic is that of the effects of the underlying soil heterogeneities, lateral variations of the underlying soil layer, and built environment
on seismic ground response at various (particularly urban) sites (Hisada et al. 1988; Chen et al. 1998; Furumura & Kennett 1998; Panza et al.
2001; Tsogka & Wirgin 2003; Celebi 2004).
This topic is associated with a class of alternative or complementary (so-called local) paradigms for explaining seismic motion in urban
sites built on soft soil. Even though the anomalous response in 1985 in Mexico City originated from a subduction zone source whose epicentral
distance was some 350 km from the city, it has been common to seek explanations of this response and others such as in Nice (Semblat et al.
2000) by employing models involving vertically propagating or nearly vertically propagating plane waves. This requires that the focal depth be
large and that the epicentral distance from the source to the city be rather small. Although both of these conditions are often not met in practice

C 2005 RAS, GJI, 163, 192224
194 J.-P. Groby and A. Wirgin
Figure 3. Regional path effects as measured by Celebi 2004. NS (top), EW (middle) and vertical (bottom) components of ground acceleration time histories
(left panel) and spectra (right panel) at a free-field station (approximately 1.5 km from the built area of the city of Anchorage, Alaska) due to an earthquake
arising from a source with focal depth of 5 km and epicentral distance (to the station) of 286 km.
Figure 4. Velocity time history of the 11 and 19 May 1962 earthquakes at the Alameda park location (surrounded by high rise buildings) in central Mexico
City arising from distant (250 km) sources (Jennings 1962). The question is whether the characteristic double-peaked nature of maximum response is due
more to the nature of the underground, to the nature of the incoming wave, and/or to coupling effects of the latter with the vibrations of the buildings, and/or to
coupling effects of the vibration modes of individual buildings, and/or to coupling between the vibrational modes of pairs of buildings. At present, there exists
no clear-cut answer to this question.
300 15
250 10
200 5
2 |S()| (mm)
ui(x,t) (mm)
150 0
100 5
50 10
0 15
0 0.5 1 1.5 2 2.5 3 3.5 0 5 10 15 20 25
(Hz) t (s)
Figure 5. 2 times the modulus of the spectrum (left panel) and time history (right panel) of the incident pseudo-Ricker pulse displacement field in the
substratum (considered to fill all space and wherein 0 = 2000 kg m3 , 0 = 600 m s1 , Q 0 = ) for 0 = 0.5 Hz.
(and, in particular, as concerns the 1985 Michoacan earthquake), the vertically propagating plane wave solicitation has often prevailed in the
theoretical/numerical studies (Balendra & Kong 2004; Semblat et al. 2000), apparently because it simplifies the analysis (another reason is
that it facilitates comparison with the so-called 1-D model of normally incident plane waves on a vertically layered half-space). This has the
effect of putting the focus on what occurs in the structure vertically below the city, namely, on the soft basin on which the earthquake-prone
cities are built. Thus, a considerable number of studies (see Chavez-Garcia & Bard 1994 for comprehensive reviews) examine the (local) effect
of the soft basin on the incident wave. For urban sites in the epicentral region of earthquakes, there exists considerable empirical evidence

C 2005 RAS, GJI, 163, 192224
1.6
0.1
1.4
0.08
1.2
0.06
1
0.04
2 |S()| (mm)
u (xg,t) (mm)
0.8
0.02
i
0.6
0
0.4
0.02
0.2 0.04
0 0.06
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 5 10 15 20 25 30 35 40 45 50
(Hz) t (s)
Figure 6. 2 times the modulus of the spectrum (left panel) and time history (right panel) of the incident first-derivative-of-Gaussian pulse displacement field
in the substratum (considered to fill all space and wherein 0 = 2000 kg m3 , 0 = 600 m s1 , Q 0 = ) for d0 = 0.5 Hz.
Table 1. Material and geometrical parameters of the site, source parame-

ters, and observation point parameters of the different configurations studied
herein.
Site/Parameter Mexico City Nice Regional
(MC) (N) (R)
0 2000 2000 2600
(kg m3 )
c0 600 1000 3000
(m s1 )
Q0
1 1300 1800 1690
(kg m3 )
c1 60 200 2000
(m s1 )
Q1 30 30 600
h 20, 40, 50, 50 10 103
(m) 60, 90
focal depth 100, 3000 100, 3000 12 103
(FD) (m)
epicentral distance (100, 6000, 100, 3 105
(ED) (m) 200 3000
0 Hz 0.1, 0.25, 0.3, 0.5, 0.9 0.05
0.5, 0.9, 1. 0.1, 0.2
d0 (Hz) 0.1, 0.25, 0.5
for h = 90 m
Methods: FE, NQ, CT NQ NQ
finite element (FE)
numerical
quadrature (NQ)
Cauchy theorem (CT)
(Lee et al. 1980) that the ground response at so-called free-field locations (and/or in the basements of buildings) can actually be accounted
for by the 1-D model (with a suitable number of layers in the underground). This 1-D model (and its variants) does not, however, seem to be
able to account for the typically double-peaked nature of maximum spectral response which is often observed in the ground response in urban
areas (such as Mexico City) subjected to earthquakes arising from distant sources (see e.g. Figs 14. At present, it is thought that local effects
account for only part of the anomalous response (see the RMCS time history in Fig. 1, characterized by strong motion of long duration, with
beatings) (Chavez-Garcia & Bard 1994; Panza et al. 2001; Cardenas & Chavez-Garcia 2003). On the other hand, Hisada et al. (1988) show
that the position of the source relative to the basin is a critical factor for determining the ground response in the basin. Another idea that has
been explored in the past few years is that the buildings of the city, in interaction with the soft soil and with each other, may also amplify
and lengthen the duration of the ground motion (see Tsogka & Wirgin 2003) for reviews of this subject). All of these studies (including or
excluding the buildings) point to the central role of surface waves, qualified either as locally launched surface (e.g. Love) waves (at the basin
edges or at heterogeneities of the soft soil) (Faeh et al. 1994) or as quasi-Love waves (excited at the base of the buildings and re-amplified
by interaction with neighbouring buildings Tsogka & Wirgin 2003; Groby et al. 2004) as a possible causal agent of anomalous response, but
little, if any, theoretical evidence has been given to back up these assertions.

C 2005 RAS, GJI, 163, 192224
250
200
150
2 |u (xg,)| (cm)
1
100
50
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
(Hz]
Figure 7. Comparison of 2 times the frequency domain ground displacement response (i.e. 2 u 1 (x g , = /2 ) versus (Hz)) at large epicentral distance
x = (3000 m, 0 m), for a shallow pseudo-Ricker 0 = 0.5 Hz source at xs = (0 m, 100 m), in the MC site with h = 50 m. The full curve was obtained by the
NQ method described in Appendix A herein, whereas the dashed curve was obtained by the finite element (FE) time domain technique described in Groby &
Tsogka (2003), Groby et al. (2004).
25
20
15
10
2 |TF(x,)|
10
15
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
(Hz)
Figure 8. 2 times the displacement transfer function relative to the MC site with h = 50 m, as computed by the Cauchy Theorem (CT) and (NQ) methods, for
xs = (0 m, 100 m), x = (1200 m, 0 m). The thin continuous curve applies to the modulus of the transfer function as computed by the CT method. A thin dashed
curve, which is invisible due to its identity with the thin continuous curve, applies to the modulus of the transfer function as computed by the NQ method.
The thin continuous curve with circles applies to the CT and NQ computations of the real part of the transfer function, whereas the thin continuous curve with
crosses applies to the CT and NQ computations of the imaginary part of the transfer function.
1.3 The questions addressed herein
We shall be concerned with a (deceivingly ) simple canonical scattering problem: that of a cylindrical SH pulse wave impinging on a soft
homogeneous layer, the latter being horizontal, of infinite lateral extent, bounded above by the free surface and below by an interface with a
half-space filled with hard homogeneous rock. The questions we address, and that we think can be answered with the help of such a simple
model, are:
(i) is it possible to obtain anomalous (in the sense mentioned above in connection with the Michoacan earthquake) response without any
lateral heterogeneity (arising from volumetric inclusions or unevennes of interfaces) in the underground medium?
(ii) what is the relation of 1-D to 2-D response and how adequate is it to model the general response of the configuration by its response
to a (nearly) vertically incident plane wave?
(iii) how does the focal depth of the source affect the response?
(iv) how does the epicentral distance affect the response?
(v) how does the contrast of mechanical properties between the layer and the half-space affect the ground response?
(vi) how does the thickness of the layer affect the response?
(vii) how do the spectral characteristics of the incident pulse affect the response?

C 2005 RAS, GJI, 163, 192224
3.5
2.5
2 |TF|
2
1.5
0.5
0
0 1 2 3 4 5 6
(Hz)
2.5
2
2 |TF|
1.5
0.5
0
0 1 2 3 4 5 6
(Hz)
5
2 |TF|
0
0 1 2 3 4 5 6
(Hz]
Figure 9. 2 times the moduli of displacement transfer functions of ground response in the Nice (N) site for various source locations and observation locations.
Top: deep source and small epicentral distance xs = (0 m, 3000 m), x = (100 m, 0 m). Middle: deep source and large epicentral distance xs = (0 m, 3000 m),
x = (3000 m, 0 m). Bottom: shallow source and large epicentral distance xs = (0 m, 100 m), x = (3000 m, 0 m). In this figure and in those which follow, the
graph of the modulus of 2 I 11 (x g , )/S() versus frequency is designated by dots, the graph of the modulus of 2 I 12 (x g , )/S() versus is designated by
dashes, the graph of the modulus of 2 (I 2 (x g , )1 + I 13 (x g , ))/S() versus frequency is designated by dot-dashes, and the graph of the modulus of the
ground displacement 2 u(x g , )/S() versus frequency by a continuous line.
Questions (i)(iii) were already addressed in the companion paper (Groby & Wirgin 2005); the emphasis herein is again on question (iii) and
more so on questions (iv)(vii). It will be substantiated, essentially by numerical means, that a source radiating cylindrical waves in a soft
layer/hard half-space medium produces a ground response which is the sum of three terms corresponding to various combinations of two
types of waves in the soft layer (SL) and hard half-space (HHS):
(1) standing body waves (SBW) in the SL and body waves (BW) in the HHS,
(2) standing body waves in the SL and surface waves (SW) in the HHS,
(3) standing surface waves (SSW) in the SL and surface waves in the HHS.

C 2005 RAS, GJI, 163, 192224
7
100
6 80
60
5
40
2 |TF| 4 20
u (x,t) (mm)
0
3
1
20
2 40
60
1
80
0 100
0 1 2 3 4 5 6 0 5 10 15 20 25 30 35 40 45 50
(Hz) t (s)
Figure 10. Spectra and time histories of ground displacement response in Nice (N) for constant source-to-observation point distances; deep source and small
epicentral distance xs = (0 m, 3000 m), x = (100 m, 0 m) (solid line curves) and shallow source and large epicentral distance xs = (0 m, 100 m), x = (3000 m,
0 m) (dotted line curves). Left: 2 times the transfer functions. Right: time histories (i.e. u 1 (x, t) versus t(s)) to a 0 = 0.5 Hz pulse.
500 250
400 200
300 150
200 100
100 50
u (x,t) (mm)
u1(x,t) (mm)
0 0
1
100 50
200 100
300 150
400 200
500 250
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
t (s) t (s)
Figure 11. Time histories of the ground displacement response in Nice (N) for constant source-to-observation point distances and a pseudo-Ricker input pulse
with 0 near the lowest-frequency maximum of transfer function. Left: deep source and small epicentral distance xs = (0 m, 3000 m), x = (100 m, 0 m), 0 =
1.0 Hz. Right: shallow source and large epicentral distance xs = (0 m, 100 m), x = (3000 m, 0 m), 0 = 0.9 Hz.
It was shown in the companion paper (Groby & Wirgin 2005) that only type (2) waves correspond to Love modes (at the resonance
frequencies of these modes); here we establish the conditions for optimal excitation and maximal contribution of these modes. It will be
shown that large-duration (i.e. anomalous) response generally requires a preponderant contribution of at least one (usually the lowest-order)
of the Love modes to the overall response. The type (1) waves dominate in the situation in which the focal depth is large and do not usually
produce long-duration response, although they can produce strong (although normal) response when the contrast of mechanical properties
between the SL and HHS is large. Beating phenomena will be shown to be a consequence of interference between type (1) and type (2) waves,
which both lead to maxima in response at nearly the same (low) frequency. Type (3) waves turn out to have negligible contribution to overall
response. Most of these features carry over to the case in which the layer is lossy. The practical consequences of these results, in relation to
the two topics of interest herein, will be discussed.
2 P R E L I M I NA R I E S
2.1 Description of the generic bare urban site
Fig. 5 in the companion paper (Groby & Wirgin 2005) represents a cross-section (sagittal plane) view of a generic bare (all constructions are
eliminated) urban site wherein g is the ground. All other features of the problem are the same as in the theoretical study of Groby & Wirgin
(2005). The present investigation addresses the numerical computation of the elastodynamic wavefield on the ground (i.e. on g ) resulting
from the cylindrical seismic wave solicitation of the site.
2.2 Spacetime and spacefrequency domain representations of the impulsive force
The general governing equations and notations of our problem are the same as in the companion paper (Groby & Wirgin 2005). The assumption
that the support of the applied force is a line, means that
f (x, t) = (x xs )S(t), (1)

C 2005 RAS, GJI, 163, 192224
9
6
5
7
4 6
2 |TF|
2 |TF|
3
4
2 3
1
1
0 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
(Hz) (Hz)
14 45
40
12
35
10
30
8
25
2 |TF|
2 |TF|
20
6
15
4
10
2
5
0 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
(Hz) (Hz)
Figure 12. 2 times the transfer functions of ground displacement response in the MC site with h = 50 m for various source locations and observation
points. Upper-left: deep source, large epicentral distance xs = (0 m, 3000 m), x = (3000 m, 0 m). Upper-right: deep source, small epicentral distance xs = (0 m,
3000 m), x = (100 m, 0 m). Lower left: shallow source, large epicentral distance xs = (0 m, 100 m), x = (3000 m, 0 m). Lower-right: shallow source, small
epicentral distance xs = (0 m, 100 m), x = (100 m, 0 m).
wherein ( ) is the Dirac delta distribution and f is the applied force density associated with the seismic source. The Fourier spectrum of f (x,
t) is

1
f (x, ) = f (x, t)eit dt
2

1
= (x xs ) S(t)eit dt = (x xs )S(). (2)
2
The pulse S(t) is chosen either to be a pseudo-Ricker function which is 6 times the first derivative of a Gaussian
d3 2 2
S(t) = 6 3 e (tt0 ) = 24 4 3 (t t0 ) + 2 2 (t t0 )3 e (tt0 ) ,
2 2
(3)
dt
to which corresponds
1 it
2
S() = 6i3 e 0 42 , (4)
2
wherein = 0 and t 0 = 1/ 0 , or the first-derivative-of-Gaussian function
d d 2 d 2
S(t) = e( ) (tt0 ) = 2( d )2 t t0d e( ) (tt0 ) ,
d 2 d 2
(5)
dt
to which corresponds
2
1 it0d d 2
S() = i e 4( ) , (6)
2 d
wherein d = d0 and t d 0 = 1/ d0 .
2.3 Spectra and time histories of various input pulses

We exhibit: (i) in Fig 5 the spectrum and time history of a pseudo-Ricker pulses having frequency 0.5 Hz and ii) in Fig. 6 the spectrum and
time history of a single first-derivative-of-Gaussian pulse having frequency d0 = 0.5 Hz.

C 2005 RAS, GJI, 163, 192224
8
14
7
12
6
10
2 |TF|
2 |TF|
4
6
3
4
2
2
1
0
0 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
(Hz) (Hz)
18 25
16
20
14
12
15
10
2 |TF|
2 |TF|
8
10
4
5
0 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
(Hz) (Hz)
Figure 13. 2 times the transfer functions of ground displacement response in the MC site for a shallow source xs = (0 m, 100 m), large epicentral distance
x = (3000 m, 0 m), and various layer thicknesses h. Upper left: h = 20 m. Upper right: h = 40 m. Lower left: h = 60 m. Lower right: h = 90 m.
15
60
40
10
20
u (x,t) (mm)
2 |TF|
0
1
5 20
40
0 60
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 20 40 60 80 100 120 140 160 180 200
(Hz) t (s)
Figure 14. 2 times the displacement transfer functions (left panel) and time histories for 0 = 0.5 Hz pseudo-Ricker input pulse (right panel) of ground
displacement response in the MC site with h = 50 m for two combinations of source locations and observation points: a deep source and small epicentral
distance xs = (0 m, 3000 m), x = (100 m, 0 m) (solid line curves), a shallow source and large epicentral distance xs = (0 m, 100 m), x = (3000 m, 0 m) (dotted
line curves).
The pulses of a given type and frequency have the same shape for all source and observation point locations since the source is assumed
to be located in the substratum and the latter is assumed to be an elastic (i.e. non-dispersive) medium; however the instants at which are
produced the maxima of the pulses change as a function of these locations.
Of particular interest is the fact that the input pulse duration (of e.g. the pseudo-Ricker function) is approximately 2/ 0 , which corresponds
to 8 s for a 0.25 Hz pulse, 4 s for a 0.5 Hz pulse, and 2 s for a 1.0 Hz pulse. As will be seen hereafter, the response to these pulses in the
layered configuration is generally of much longer duration.
2.4 Site parameters

The computations of seismic ground response were carried out for three bare urban sites. The first site (MC) is thought to be representative
of downtown Mexico city. The second site (N) is thought to be representative of downtown Nice (France). The third site (R) is a regional
one (300 km epicentral distance) which is a crude representation of the underground between, for instance, the southwestern coast of Mexico
(location of the source) and Mexico city.
The parameters of the sites, sources, and observation points employed in the computations are given in Table 1. Note that the computations
for all but the MC site were carried out via the numerical quadrature (NQ) method. The computations for the MC site were carried out by
means of the finite element (FE), Cauchy theorem (CT) and NQ methods. A full description of the FE method can be found in (Groby et al.
2004). The numerical aspects of the NQ method are described in Appendix A, and those of the CT method in Appendix B.

C 2005 RAS, GJI, 163, 192224
50 300
40
200
30
20
100
10
u (x,t) (mm)
0 0
u1(x,t) (mm)
1
10
100
20
30
200
40
50 300
0 10 20 30 40 50 60 0 10 20 30 40 50 60
t (s) t (s)
Figure 15. Time histories of ground displacement response at the MC site (with h = 50 m) to a 0 = 0.5 Hz pseudo-Ricker input pulse for large (left panel:
xs = (0 m, 3000 m), x = (3000 m, 0 m)) and small (right panel: xs = (0 m, 100 m), x = (100 m, 0 m)) source-to-observation point distances.
2.5 Method of computation of time histories of ground response from Fourier spectra of this response
The time history of ground displacement response is related to the Fourier spectrum of this response via the Fourier transform

u 1 (xg , t) = u 1 (xg , )eit d. (7)

The NQ and CT methods provide the spectrum of ground displacement response u 1 (x g , ) so that it is necessary to obtain u 1 (x g , t) by means
of (2.7) (note that the FE method yields u 1 (x g , t) directly). This is done numerically by standard fast Fourier transform (FFT) software.
2.6 Method of computation of transfer functions

Let u 1 (x g , ) be the spectrum of total ground response and I 1j (x g , ) a part of this response. Then the transfer function relative to the total
ground response is defined as u 1 (x g , )/S() and the transfer function relative to the partial ground response is defined as I 1j (x g , )/S().
In all the graphs relating to transfer functions which follow, designates the frequency (related to the angular frequency by = 2 ),
the graph of the modulus of I 11 (x g , )/S() versus frequency is designated by dots, the graph of the modulus of I 12 (x g , )/S() versus
is designated by dashes, the graph of the modulus of (I 12 (x g , )1 + I 13 (x g , ))/S() versus frequency is designated by dot-dashes, and the
graph of the modulus of the ground displacement u(x g , )/S() versus frequency by a continuous curve.
2.7 Comparison of the results of the NQ and FE methods for determining the frequency-domain response on the ground
In order to be reasonably sure that the NQ technique employed herein gives valid results for a viscoelastic layer, we compared these results
to those obtained by a finite element (FE) time-domain viscoelastic code developed by the present authors with C. Tsogka (Groby & Tsogka
2003; Groby et al. 2004). An example of these results is given in Fig. 7 wherein it is seen that the NQ results are very close to the FE results.
2.8 Comparison of the results of the NQ and CT methods for determining the frequency-domain response on the ground
The computations made via the Cauchy theorem (CT) method are compared in Fig. 8 to those made with the help of the numerical quadrature
(NQ) method for the Mexico City (MC) site. It is seen in this figure that the two methods give the same results. The same good agreement
between the CT and NQ method results was obtained (results not shown here) at the same site for x = (3000 m, 0 m).
3 G RO U N D R E S P O N S E I N T H E B A R E N I C E S I T E
The results apply here to a configuration thought to be representative of that in the central (bare) portion of the city of Nice (N).
3.1 Comparison of the transfer functions for various source and observer locations
We first place the source at a relatively large depth of 3 km on the x 2 axis, that is, xs = (0 m, 3000 m) and evaluate the moduli of the ground
transfer functions relatively near the epicentred, that is, x = (100 m, 0 m) (top subfigure in Fig. 9) as well as relatively far from the epicentred,
that is, x = (3000 m, 0 m) (middle subfigure in Fig. 9), and then place the source at a relatively small depth of 100 m on the x 2 axis, that is,
xs = (0 m, 100 m) and evaluate the ground transfer functions relatively far from the epicentred, that is, x = (3000 m, 0 m) (bottom subfigure
in Fig. 9).
It will be noticed that, in this and practically all subsequent results, the curve relative to ||(I 12 (x g , ) + I 13 (x g , ))/S()|| is coincident
with that relative to ||I 12 (x g , )/S()|| which means, as predicted by the analysis in the companion paper (Groby & Wirgin 2005), that the

C 2005 RAS, GJI, 163, 192224
0.8
0.6
0.4
0.2
u (x,t) (mm)
0
1
0.2
0.4
0.6
0.8
0 5 10 15 20 25 30 35 40 45 50
t (s)
2
u1(x,t) (mm)
8
0 5 10 15 20 25 30 35 40 45 50
t (s)
80
60
40
20
u1(x,t) (mm)
20
40
60
80
0 10 20 30 40 50 60 70 80 90 100
t (s)
Figure 16. Time histories of ground displacement response for 0 = 0.1 Hz (top panel), 0 = 0.25 Hz (middle panel), 0 = 0.5 Hz (bottom panel) pseudo-Ricker
input pulses at the MC site with layer thickness h = 20 m for a shallow source and large epicentral distance xs = (0 m, 100 m), x = (3000 m, 0 m).
contribution to overall ground response of the standing surface waves in the layer is negligible. Thus, we restrict the following discussion to
the sole contribution of the standing bulk waves of the first (SBW1) and second kinds (SBW2). The top and middle panels in Fig. 9 show
that when the focal depth is large the ground response is dominated by the contribution of the SBW1 (i.e. by ||I 11 (x g , )/S()||), and, in fact,
the SBW2 have no influence on the response beyond 1 Hz. However, the right panel in Fig. 9 gives just the opposite result when the focal
depth is small, since the ground response is largely dominated by the SBW2 (i.e. by ||I 12 /S()||) and the SBW1 have little influence beyond
1 Hz. Another interesting feature of these results is that the total response curves have noticeably different appearances when the source is
deep or shallow (notice that these appearances are qualitatively the same for small and large epicentral distances, assuming the same, large
focal depths in the two cases).
3.2 Comparison of frequency- and time-domain responses for constant source-to-observation point distances
Two constant source-to-observation point distance situations are considered: (1) xs = (0 m, 3000 m), x = (100 m, 0 m) (solid line curves in
Fig. 10), and (2) xs = (0 m, 100 m), x = (3000 m, 0 m) (dotted line curves in Fig. 10). We notice in the left panel of Fig. 10 that the first peak

C 2005 RAS, GJI, 163, 192224
0.8
0.6
0.4
0.2
u (x,t) (mm)
0
1
0.2
0.4
0.6
0.8
0 5 10 15 20 25 30 35 40 45 50
t(s)
15
10
5
u1(x,t) (mm)
10
15
0 20 40 60 80 100 120 140 160 180 200
t (s)
50
40
30
20
10
u1(x,t) (mm)
10
20
30
40
50
0 20 40 60 80 100 120 140 160 180 200
t (s)
Figure 17. Time histories of ground displacement response for 0 = 0.1 Hz (top panel), 0 = 0.25 Hz (middle panel), 0 = 0.5 Hz (bottom panel) pseudo-
Ricker input pulses at the Mexico City site with layer thickness h = 40 m for a shallow source and large epicentral distance xs = (0 m, 100 m), x = (3000 m,
0 m).
(apart from the one at 0 Hz) of the transfer function occurs at a lower frequency when the source is near to the layer than when it is far from the
layer, which fact suggests (on account of the discussion in the companion paper Groby & Wirgin 2005) that the lower-frequency peak is due to
the (resonant) excitation of the fundamental Love mode (SBW2) whereas the higher-frequency peak is associated with the first (non-resonant)
interference (SBW1) maximum. The same remarks apply to the higher-order peaks. Moreover, the value of the transfer function at the first pair
of peaks is much larger due to Love mode excitation than to constructive interference effects, and since the width at half-height of the Love
mode peak is smaller than that of the interference peak the sharpness of the Love mode peak is larger than that of the interference peak. The
translation of this into the time domain is that the signal associated mainly with the fundamental Love mode resonance is more intense and
of longer duration than the signal associated mainly with the fundamental interference peak.
This last remark should be tempered by consideration of the spectrum of the input pulse, since the transfer functions do not take this
spectrum into account whereas the temporal signals do. Thus, when the location of the maximum of the spectrum of the input pulse is closer
to the location of the maximum of the transfer function, the time-domain response is larger, as seen in Fig. 11, this being true for signals that

C 2005 RAS, GJI, 163, 192224
0.8
0.6
0.4
0.2
u1(x,t) (mm)
0
0.2
0.4
0.6
0.8
1
0 20 40 60 80 100 120 140 160 180 200
t (s)
10
2
u1(x,t) (mm)
10
0 20 40 60 80 100 120 140 160 180 200
t (s)
50
40
30
20
10
u1(x,t) (mm)
10
20
30
40
50
0 20 40 60 80 100 120 140 160 180 200
t (s)
Figure 18. Time histories of ground displacement response for 0 = 0.1 Hz (top panel), 0 = 0.25 Hz (middle panel), 0 = 0.5 Hz (bottom panel) pseudo-Ricker
input pulses at the MC site with layer thickness h = 60 m for a shallow source and large epicentral distance xs = (0 m, 100 m), x = (3000 m, 0 m).
are essentially due both to Love resonances or to constructive interference effects (note that the location of the pulse maxima were adjusted so
as to be close to the locations of the transfer function fundamental peaks). Actually, this figure reveals the existence of a beating phenomenon
in the ground response temporal signal for a source near the layer, which is probably due to the combined (amplitude modulation) effects of
the fundamental Love mode peak and the fundamental interference peak.
4 G RO U N D R E S P O N S E I N T H E B A R E M E X I C O C I T Y S I T E
We next consider the Mexico City (MC) site; this urban site is bare, contrary to the case treated in (Tsogka & Wirgin 2003) in which the
buildings were taken into account.

C 2005 RAS, GJI, 163, 192224
1.5
0.5
u (x,t) (mm)
0
1
0.5
1.5
2
0 50 100 150 200
t (s)
10
2
u1(x,t) (mm)
10
0 50 100 150 200
t (s)
50
40
30
20
10
u1(x,t)(mm)
10
20
30
40
50
0 20 40 60 80 100 120 140 160 180 200
t (s)
Figure 19. Time histories of ground displacement response for 0 = 0.1 Hz (top panel), 0 = 0.25 Hz (middle panel), 0 = 0.5 Hz (bottom panel) pseudo-
Ricker input pulses at the MC site with layer thickness h = 90 m for a shallow pseudo-Ricker source and large epicentral distance xs = (0 m, 100 m), x =
(3000 m, 0 m).
4.1 The cumulative contributions of the SBW1, SBW2 and SSW to the transfer functions of the total ground response
The discussion here centers on the transfer functions of partial and total ground response.
In all except the lower right hand panel of Fig. 12 we again observe that the ground response is dominated by the SBW1 when the source
is deep and by the SBW2 when the source is shallow. The exceptional case is that of a shallow source and small epicentral distance, for which
the contributions of the SBW2 and SBW1 to the overall response are of comparable magnitude, especially near the first low-frequency peak.
A plausible cause of this behaviour is the rather large contrast of body-wave velocities between the layer and substratum, thus giving rise to
a large contribution of the individual SBW1 at the fundamental Haskell frequency (recall that this contribution is all the larger the greater is
the body wave velocity contrast).
Fig. 13 gives a measure of the effect of changes in the layer thickness h for a shallow source and large epicentral distance. We observe in
this figure that the response is dominated by the cumulative contribution of the SBW2 for all the layer thicknesses. Furthermore, the number

C 2005 RAS, GJI, 163, 192224
1
0.8
0.6
0.4
0.2
u (xg,t) (mm)
0
0.2
1
0.4
0.6
0.8
1
0 50 100 150 200
t (s)
1.5
0.5
u (xg,t) (mm)
0
1
0.5
1.5
2
0 50 100 150 200
t (s)
2.5
1.5
0.5
u (xg,t) (mm)
0
1
0.5
1.5
2.5
0 50 100 150 200
t (s)
Figure 20. Time histories of ground displacement response at the MC site with layer thickness h = 90 m for 0 = 0.1 Hz (top panel), 0 = 0.25 Hz (middle
panel), 0 = 0.5 Hz (bottom panel) first-derivative-of-Gaussian input pulses and xs = (0 m, 100 m), x = (3000 m, 0 m).
and height-to-width ratio of the resonance peaks in the interval [0, 2 Hz] increases with h, the dominant peak always being the one associated
with the resonant excitation of the first (lowest-frequency) Love mode and being located at a frequency that is all the lower the larger is h.
4.2 Comparison of the spectra and time histories of ground response for constant source-to-observation point distances
Again, the two source/observation point pairs are: xs = (0 m, 3000 m), x = (3000 m, 0 m) (solid line curve in Fig. 14) and xs = (0 m, 100 m),
x = (3000 m, 0 m) (dotted line curve in Fig. 14). All that was said in the two previous examples concerning the transfer functions holds in the
present case. Likewise, the repercussions on the temporal signals are the same as in the previous two cases (short duration signal for a remote
source and relatively long signal for a near source.

C 2005 RAS, GJI, 163, 192224
45
40
35
30
25
2|TF|
20
15
10
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
(Hz)
40
35
30
25
2|TF|
20
15
10
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
(Hz)
Figure 21. 2 times the the modulus of the displacement transfer function ( the frequency) relative to the MC site with with h = 50 m, as computed by the
NQ method, for xs = (0 m, 100 m), x = (0 m, 0 m) (top panel) and xs = (0 m, 100 m), x = (200 m, 0 m) (bottom panel), and excitation by a pseudo-Ricker pulse
with 0 = 0.5 Hz. The contribution to this function due to the integral I 11 is designated by dots, the contribution of I 12 is designated by dashes, the contribution
of the modulus of I 12 + I 13 is designated by dot-dashes, and the total transfer function is the continuous curve.
4.3 Time histories of ground response for large and small source-to-observation point distances
We concentrate our attention on Fig. 15.
Although in Fig. 12 (upper left and lower right panels) we observed that the transfer functions for large and small source-to-observation
point distances are qualitatively very similar, we are now somewhat surprised to find that the two corresponding signals in Fig. 15 are so
qualitatively similar, due to the fact that the transfer function corresponding to the left panel in Fig. 12 is dominated by the SBW1 contribution,
whereas the transfer function corresponding to the right hand panel in Fig. 12 has strong contributions from both the SBW1 and SBW2. The
clue to this unexpected result resides in the spectrum of the 0 = 0.5 Hz input pulse (see left panel of Fig. 5), since the maximum of the latter
is around 0 = 0.6 Hz and this frequency is both far-removed from the peaks of the transfer functions and characterized by a predominant
contribution of the SBW1 (the latter fact providing an explanation of the relatively short duration of the response signals in Fig. 15 (note that
the intensity of the very close source-to-observation point signal is much larger than that of the other signal, as it should be).
4.4 Time histories of ground response for various layer thicknesses and input pulses
A more thorough study of the influence of the input spectrum must take into account variations of the layer thickness h. This is done for a
pseudo-Ricker source that is 100 m below the ground) in Fig. 16 for a 20 m thick layer, in Fig. 17 for a 40 m thick layer, in Fig. 18 for a
60m thick layer, and in Fig. 19 for a 90 m thick layer. We observe quite different responses, varying from a short pulse quite similar to the
input pulse (for the thinnest layer and the lowest frequency input pulse) to a very long duration pulse (as much as 200 s as compared to the 4
s duration of the input pulse) with pronounced beating (for the thickest layer and a medium frequency input pulse). Note that the 90 m layer

C 2005 RAS, GJI, 163, 192224
30
25
20
2|TF|
15
10
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
(Hz)
20
15
2|TF|
10
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
(Hz)
Figure 22. 2 times the the modulus of the displacement transfer function ( the frequency) relative to the MC site with h = 50 m, as computed by the NQ
method, for xs = (0 m, 100 m), x = (400 m, 0 m) (top panel) and xs = (0 m, 100 m), x = (1000 m, 0 m) (bottom panel), and excitation by a pseudo-Ricker pulse
with 0 = 0.5 Hz.
also corresponds to the case in which the source is closest to the layer (10 m from the bottom face of the layer), which may also be a factor
contributing to the pronounced anomalous character of the response in this configuration.
4.5 The influence of the incident wave pulse shape on ground response
It is of interest to ascertain to what extent the nature of the source spectrum affects the ground response, notably when the source is near
the layer. This can be appreciated by comparing Fig. 19 (for a pseudo-Ricker incident pulse) with Fig. 20 (for a first-derivative-of-Gaussian
incident pulse).
4.6 The influence of the epicentral distance on ground response
We now keep the source at a fixed location (xs = (0 m, 100 m)) and vary the epicentral distance.
4.6.1 Transfer functions

The transfer functions are computed on the ground at x = (0 m, 0 m) and xs = (200 m, 0 m) in Fig. 21, xs = (400 m, 0 m) and xs = (1000 m,
0 m) in Fig. 22, xs = (2000 m, 0 m) and xs = (3000 m, 0 m) in Fig. 23, xs = (6000 m, 0 m) and xs = (7000 m, 0 m) in Fig. 24. We observe that
at very short epicentral distances, both I 11 and I 12 contribute substantially to the principal peak of u1 . At less short epicentral distances, the
contribution of I 11 to the principal peak becomes much smaller than that of I 12 , and oscillations, presumably due to interference between the
I 11 and I 12 terms, appear on the high-frequency end of the principal peak. At moderate epicentral distances, the sharpness of the first peaks of

C 2005 RAS, GJI, 163, 192224
18
16
14
12
10
2|TF|
8
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
(Hz)
14
12
10
8
2|TF|
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
(Hz)
method, for xs = (0 m, 100 m), x = (2000 m, 0 m) (top panel) and xs = (0 m, 100 m), x = (3000 m, 0 m) (bottom panel), and excitation by a pseudo-Ricker
pulse with 0 = 0.5 Hz.
these oscillations is considerable and their amplitudes are fairly large. The oscillations are more and more rapid and of decreasing amplitude
as the epicentral distance increases. The sharpness of the principal resonance peak increases with epicentral distance, but a saturation effect
sets in for large epicentral distances.
4.6.2 Spectra and time histories of ground response for various epicentral distances
The spectra and corresponding time histories are computed on the ground at x = (0 m, 0 m) and xs = (200 m, 0 m) in Fig. 25, xs = (400 m,
0 m) and xs = (1000 m, 0 m) in Fig. 26, xs = (2000 m, 0 m) and xs = (3000 m, 0 m) in Fig. 27, xs = (6000 m, 0 m) and xs = (7000 m, 0 m) in
Fig. 28.
It can be observed, that although some oscillations (satellite peaks) appear at the high-frequency end of the principal peak of the response
spectrum at very short epicentral distances (<400 m), the sharpness of these satellite peaks is not large enough to produce a noticeable
beating effect in the time history of response. Also, at these short epicentral distances, the sharpness of the main peak in the spectrum is
rather moderate so as to give rise to a time-domain response of rather short duration. Increasing the epicentral distance from 400 to 1000 m
substantially increases the relative amplitude and sharpness of the satellite peaks thus giving rise (by interference with the principal resonance
peak) to marked beating in the time history. As the epicentral distance is further increased the amplitude and sharpness of the satellite peaks
decrease, thus decreasing the beating effect, while the sharpness of the principal resonance peak increases, thus increasing the duration of the
time-domain response. This latter effect seems to slow down at a certain large epicentral distance, beyond which the sharpness of the main
spectral peak as well as the duration of the time-domain response remain quasi-constant.

C 2005 RAS, GJI, 163, 192224
11
10
2|TF|
5
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
(Hz)
10
6
2|TF|
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
(Hz)
method, for xs = (0 m, 100 m), x = (6000 m, 0 m) (top panel) and xs = (0 m, 100 m), x = (7000 m, 0 m) (bottom panel), and excitation by a pseudo-Ricker
pulse with 0 = 0.5 Hz.
5 R E G I O N A L PAT H E F F E C T S
5.1 Choice of site, source and observation point
We now examine the manner in which a wave radiated from a source located underneath, but close to, the lower crustal boundary propagates
over long (i.e. regional) horizontal paths.
An actual, field-recorded, example of such motion, relative to the 2002 March 11 (shallow) seismic Denali (Alaska) event, at a free-field
ground location (i.e. 1.5 km from the buildings of the city of Anchorage) at an epicentral distance of more than 275 km, is given in Fig. 10
of Celebi (2004) (reproduced in Fig. 3 herein). Other examples of such regional path responses are found in Fig. 2 of Furumura & Kennett
(1998) (reproduced in Fig. 2 herein) and Savage (2004). A remarkable feature of the recorded ground motion in Celebi (2004) is its long
duration of over 125 s; it is of comparable duration in Furumura & Kennett (1998) at a site located just before the incident wave hits Mexico
City after having travelled a few hundred kilometers from the source location. The purpose of the lines which follow is to show that regional
path waves with such long duration can be obtained with our simple model.
We constructed a hopefully plausible crustal model starting with the parameters of a thin layer, modified-Nice configuration (for which
0 = 2000 kg m3 , 1 = 1300 kg m3 , c0 = 600 m s1 , c1 = 200 m s1 , Q 0 = , Q 1 = 30, h = 80 m), and by assuming conservation of such
quantities as k 1 h, 0 / 1 , etc. in going to a much thicker layer. Let us suppose that we have two configurations, one of which is thin-layered
and known (configuration with subscript 1), and the other is thick-layered and unknown (configuration with subscript 2). The layer in the
known configuration is relatively soft and lossy, whereas it is relatively hard (although always softer than the substratum) in the unknown
configuration. Since harder media are usually less lossy, we assume rather arbitrarily that the Q of the layer in the unknown configuration is
20 times larger than the Q in the known configuration, while the Qs of the substratum remain infinite in both configurations. Thus, we have:

C 2005 RAS, GJI, 163, 192224
800
300
700
200
600
500 100
2|u (xg,)| (mm)
u1(x ,t) (mm)

400
0
g
1
300
100
200
200
100
300
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 20 40 60 80 100 120 140 160 180 200
(Hz) t (s)
700
200
600 150
100
500
50
2|u (xg,)| (mm)
400
u1(x ,t) (mm)

0
g
1
300
50
200
100
100
150
0 200
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 20 40 60 80 100 120 140 160 180 200
(Hz) t (s)
Figure 25. 2 times the the modulus of the Fourier spectrum of ground displacement response (left panels) and corresponding modulus of the time history of
ground displacement response relative to the MC site with h = 50 m, as computed by the NQ method, for xs = (0 m, 100 m), x = (0 m, 0 m) (top panels) and
xs = (0 m, 100 m), x = (200 m, 0 m) (bottom panels). Excitation by a pseudo-Ricker pulse with 0 = 0.5 Hz.
Q 02 = Q 01 = and Q 12 = 20, Q 11 = 600. Conservation of k 1 h means k 11 h 1 = k 12 h 2 , whence c12 = c11 2 h 2 / 1 h 1 , or, if we choose 1 = 1 Hz,
2 = 0.08 Hz, and h 2 = 10 km, then c 2 = 2000 m s1 . Conservation of 0 / 1 means 01 / 11 = 02 / 12 , so that if we choose 02 = 2600 kg m3
(close to the density of granite), then 12 = 1690 kg m3 . We would also like to conserve wave speed proportions, that is, c01 /c11 = c02 /c12 , but
this turns out to give wave speeds in M 0 that are much larger than those for granite (3200 m s1 ) for the choice c12 = 2000 m s1 , so that we
arbitrarily chose c02 = 3000 m s1 (i.e. close to the wave speed in granite). Finally we chose to conserve the relative distance of the source to
the lower boundary of the layer, that is (x s2 h)/x s2 , which gives x s 2 = 12 km.
5.2 Spectra and time histories of ground response

The results for this new (crustal) configuration excited by the usual pseudo-Ricker pulse line sources are given in Fig. 29 wherein it can be
seen that the ground response far from the epicentred (300 km): (1) is dominated by the excitation of Love modes (notably the fundamental)
and (2) takes the form of a pulse which has a shape and duration (here approximately 100 s) not very different from that in Fig. 10 of Celebi
(2004), Furumura & Kennett (1998) and Savage (2004).
Thus, our theoretical model shows that it is quite possible for a source underneath, and relatively close to, a fairly thick, fairly hard crust
overlying a very hard substratum to give rise to a rather long-duration pulse even at large epicentral distances. What becomes of this pulse
when a city is located at this large lateral distance from the source constitutes an important, and as yet not fully elucidated, question (meaning
that although studies such as Faeh et al. (1994) are designed to take into account all that occurs between the distant source and the observation
point in the basin, the results that are offered are entirely of numerical nature and, therefore, do not provide an explanation of the underlying
physical processes).
6 DISCUSSION
We shall now address the questions raised in Section 1.3.

The answers to the first three questions are provided in the companion paper (Groby & Wirgin 2005) and are verified numerically in the
present contribution. The third question is treated from a different angle in Appendix B herein.
The fourth question was: how does the epicentral distance affect the response? This question has at least two aspects, concerning (i)
relatively short paths between the source and observer and (ii) regional paths.
If the epicentral distance is relatively small, the source is active (as in the majority of the field examples offered in Lee et al. 1980), and
the focal depth is not too large, our theory indicates that the time histories of the motion at the observation point on the ground result from a

C 2005 RAS, GJI, 163, 192224
550
500
100
450
400
50
350
2|u (xg,)| (mm)
u1(x ,t) (mm)

300
g
250
1
200
50
150
100
100
50
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 20 40 60 80 100 120 140 160 180 200
(Hz) t (s)
350
80
300
60
250 40
20
2|u (xg,)| (mm)
200
u (x ,t) (mm)
0
g
1
150
1
20
100
40
60
50
80
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 20 40 60 80 100 120 140 160 180 200
(Hz) t (s)
Figure 26. 2 times the the modulus of the Fourier spectrum of ground displacement response (left panels) and corresponding modulus of the time history
of ground displacement response relative to the MC site with h = 50 m, as computed by the NQ method, for xs = (0 m, 100 m), x = (400 m, 0 m) (top panels)
and xs = (0 m, 100 m), x = (1000 m, 0 m) (bottom panels). Excitation by a pseudo-Ricker pulse with 0 = 0.5 Hz.
complex interplay of SBW1 and SBW2, or alternatively of pole and branch cut contributions, with unpredictable features (such as irregular
beating, which, nevertheless, are revealed by the numerical analysis herein). Due to the formal identity of active and induced sources, the same
comment applies to induced sources located in the vicinity of the lower boundary of the layer. Furthermore, we find, for relatively moderate
crustal thicknesses, epicentral distances, and source depths, that the duration first increases, and then seems to saturate, with increasing
epicentral distance. We also find that the peak motion decreases with epicentral distance (more generally: with the distance of the source to
the observation point) when the layer losses are rather high (see, e.g. Figs 13 and 14). This is in contrast to what occurs for regional paths over
low loss layers, as has been observed by Furumura & Kennett (1998).
Consider the situation for regional paths. The commonly offered explanation of the lengthening of the duration with path length is
dispersion and scattering in the crustal layer, presumably due to unevennes of the boundaries and volumic heterogeneities of the layer
(Furumura & Kennett 1998). It will be recalled that our model has flat horizontal boundaries and we assumed both the layer and half-space to
be homogeneous, so as to rule out scattering in the above-mentioned sense. Nevertheless, our model, and the theory (notably the polebranch
cut version) with which it is associated, implicitly contains an explanation of large durations at large epicentral distances. We found that at
such locations the field behaves like a Love mode (it can be a mixture of these modes if the frequency is sufficiently high), the body waves
(contained in the branch cut integral) contributing little (asymptotically speaking) at such large distances because their amplitudes decay
spatially and those of the Love mode amplitudes less so (or not at all when the layer is free of dissipation). As the resonance associated with
the excitation of a Love mode is sharply peaked (in the frequency domain; and, in any case, more sharply peaked than the maximum associated
with constructive interference of standing body waves), the duration of the signal (in the time domain) associated with the Love mode, which,
as pointed out above, dominates the field at large epicentral distances, is relatively large (compared to the so-called normal duration due to
non-resonant standing body waves in the layer) at these large distances.
This is in overall agreement with the observations in field records (Novikova & Trifunac 1993, 1995; Furumura & Kennett 1998) whereby
the duration appears to be an increasing function of epicentral distance for regional path lengths. However, Furumura & Kennett (1998) find
that the peak response can also increase with epicentral distance, a feature that we have been unable to reproduce with our simple model, due
perhaps to the fact that this model does not take into account irregularities of the boundaries and composition of the layer. This point certainly
merits further investigation.
The fifth question was: how does the contrast of mechanical properties between the layer and the half-space affect the ground response?
There does not appear to be a clear-cut answer to this question (see, however, Shoji et al. 2004) in which it appears rather systematically
that softer layers lead to longer durations and lower peak response for a given epicentral distance), since the dependence on the mechanical
parameters is very much intermingled with that of the values of the geometrical and source parameters.

C 2005 RAS, GJI, 163, 192224
250 60
40
200
20
150
2|u (xg,)| (mm)
u (xg,t) (mm)
0
1
1
100
20
50
40
0 60
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 20 40 60 80 100 120 140 160 180 200
(Hz) t (s)
200 50
180 40
160 30
140 20
120
2|u (xg,)| (mm)
10
u (xg,t) (mm)
100 0
1
1
80 10
60 20
40 30
20 40
0 50
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 20 40 60 80 100 120 140 160 180 200
(Hz) t (s)
Figure 27. 2 times the the modulus of the Fourier spectrum of ground displacement response (left panels) and corresponding modulus of the time history
of ground displacement response relative to the MC site with h = 50 m, as computed by the NQ method, for xs = (0 m, 100 m), x = (2000 m, 0 m) (top panels)
On the whole, most of the answers to the first three questions in the companion paper (Groby & Wirgin 2005), and the fourth question
herein, hold in a qualitative sense whatever the contrast of mechanical properties (see, e.g. the results herein for the Nice and Mexico sites),
although there are some quantitative differences (e.g. the contrast influences considerably the intensity of the SBW1 contribution). Naturally,
the most remarkable features of the ground response of our simple configuration, which are due to the excitation of the fundamental Love
mode, can be observed only if the layer is softer than the substratum in the sense of the conditions of existence of Love modes given in the
companion paper. These conditions are so broad and widespread (for city-like sites) as to render the anomalous response described herein
a quite universal phenomenon.
The sixth question was: how does the thickness of the layer affect the response? In Fig. 13 it was found that increasing the layer thickness
increases the number of peaks, as well as the sharpness of each of the latter, in a given range of frequencies of the transfer function. This has the
effect of lowering the frequency of occurrence of the first peak so that the time-domain response will be largely conditioned by the spectrum
of the input pulse, assuming the latter to be centreed at a relatively high frequency. Thus, a low-frequency pulse can produce substantially the
same type of response for a thick layer as a relatively high-frequency pulse in a thin layer. This point is important in connection with the topic
of regional path effects mentioned in Section 1.
The seventh question was: how do the spectral characteristics of the incident pulse affect the response? The answer to this question
can be found by comparing the three subfigures in any one of Figs 1619. Obviously, the spectrum of the incident pulse is a key factor (see
section 8), which: (a) if it overlaps either a constructive interference peak or Love mode peak, gives rise to attenuated, quasi monochromatic
response, often of long duration (see Figs 1619 in which an example is given of a pulse having a duration of 4 s that gives rise to substantial
ground response of 200 s duration), (b) if it overlaps both a constructive interference peak and Love mode peak, gives rise to attenuated,
quasi monochromatic response with more or less regular beatings and (c) if it does not overlap significantly either a constructive interference
of Love mode peak, gives rise to a time-domain response that can be qualitatively quite similar to the input signal (see, e.g. left panel of
Fig. 17). When the sources are induced, their spectra will be modified with respect to that of the spectrum of the primary active source due
to diffraction and dispersion, so that an a priori unfavourable situation for anomalous responses from the point of view of the primary active
source may turn out to be favourable from the point of view of the induced sources.
7 C O N C LU S I O N A N D P E R S P E C T I V E S
This work originated in the observation that no satisfactory physical and/or mathematical explanation has been given until now of anomalous
seismic response in urban environments with soft layers or basins overlying a hard substratum. The principal reason for this knowledge gap

C 2005 RAS, GJI, 163, 192224
150 30
20
100 10
2|u (xg,)| (mm)
u (x ,t) (mm)
0
g
1
1
50 10
20
0 30
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 50 100 150 200
(Hz) t (s)
150 30
20
100 10
2|u (xg,)| (mm)
u (x ,t) (mm)
0
g
1
1
50 10
20
0 30
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 50 100 150 200
(Hz) t (s)
Figure 28. 2 times the the modulus of the Fourier spectrum of ground displacementresponse (left panels) and corresponding modulus of the time history of
ground displacement response relative to the MC site with h = 50 m, as computed by the NQ method, for xs = (0 m, 100 m), x = (6000 m, 0 m) (top panels)
probably lies in the complexity of the sites examined in previous (essentially numerical) studies: (1) a homogeneous or multilayered basin of
complicated form not including buildings (e.g. Semblat et al. 2000; Faeh et al. 1994) and (2) a homogeneous layer overlain by a periodic or
non-periodic set of blocks or buildings (e.g. Tsogka & Wirgin 2003; Groby et al. 2004). The choice was therefore made herein to simplify
as much as possible the characteristics of the site and its excitation, while retaining as many as possible of their essential features. Thus, it
was thought that: (i) the problem had to be treated at least as a 2-D one, (ii) the solicitation should not be a plane wave (for which coupling to
Love modes is impossible in the chosen configuration) but rather the wave radiated by a source which could be as simple as a line source (this
source eventually being able to mimic induced sources in more complicated configurations) and (iii) the soft component of the site could be a
layer (rather than a basin) with flat, horizontal boundaries (i.e. flat rather than irregular ground, as rendered by the presence of buildings, flat
interface with the substratum, rather than curved or irregular as for a basin or irregular layer).
The theoretical analysis in the companion paper (Groby & Wirgin 2005) reached a sort of limit when the horizontal wavenumber
integration was attempted. Thus, the integrals appearing in the frequency-domain response were carried out numerically and a parametric
study was made of the cumulative contributions of the SBW1 and SBW2. It was shown, as expected, that the SBW1 give the preponderant
contribution for remote sources, while both the SBW1 and SBW2 cumulative contributions can be significant for nearby sources. The
interference nature of the amplitudes of the individual SBW1 was shown to be maintained in the frequency-domain cumulative response of
these waves. The resonant nature of the amplitudes of the individual SBW2 was shown to be maintained in the frequency-domain cumulative
response of these waves.
It was also shown that it possible for a source, underneath, and relatively close to, a fairly thick (10 km), fairly hard crust overlying a
very hard substratum, to give rise to a rather long-duration pulse even at large (e.g. 300 km) epicentral distances, and that this finding is in
agreement with what has been observed in connection with real earthquakes (see, e.g. Celebi 2004; Furumura & Kennett 1998). We did not
carry out an extensive analysis of this finding, nor address the issue of what becomes of this pulse when it enters an urban centred located at
large lateral distances from the source (as was done numerically in works such as Faeh & Panza 1994; Faeh et al. 1994; Panza et al. 2001).
A question that naturally arises is whether the type of analysis carried out herein can be extended to more realistic configurations in
which induced sources are likely to play a major role. Our feeling is that this can be done provided some clever approximations are made in
the expressions for the response of these configurations.
Another question (alluded to in one of the previous paragraphs) is that of regional path effects on global response in cities such as Mexico
subject to earthquakes arising from laterally remote sources. This very important theoretical issue will have to be treated in depth, first in the
manner of the present contribution, to examine how the wave radiated by the source reaches the city site, why the peak response can increase
with path length (as observed by Furumura & Kennett 1998), what the nature of the waves is when the latter arrive in the city, and how these

C 2005 RAS, GJI, 163, 192224
5
x 10
3 3
2.5 2
2 1
a (x,t) (mm.s )
2
2 |TF|
1.5 0
1
1 1
0.5
2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 3
0 50 100 150 200
(Hz) t (s)
3
x 10
5
0.15
0.1 3
0.05
1
a (x,t) (mm.s )
2
a (xg,t) (mm.s )
2
0
0
1
1
1
0.05
2
3
0.1
0 50 100 150 200 5

0 50 100 150 200
t (s) t (s)
Figure 29. Ground motion at large epicentral distance (x 1 s1 x

= 300 km) in response to a pseudo-Ricker pulse line source underneath, and close to (focal
depth x s2 = 12 km), the lower boundary of a thick (h = 10 km), fairly hard, crust overlying a granite-like substratum ( 0 = 2600 kg m3 , 1 = 1690 kg m3 ,
c0 = 3000 m s1 , c1 = 2000 m s1 , Q 0 = , Q 1 = 600, h = 80 m) (parameters of the R site). Displacement transfer function, with same notations as in
Fig. 9 (upper left panel) and acceleration time histories for: a 0 = 0.05 Hz input pulse (upper right panel), a 0 = 0.1 Hz input pulse (lower left panel), a 0
= 0.2 Hz input pulse (lower right panel).
waves are converted within the city into the form they have been observed to take (quasi-Love or Rayleigh waves giving rise to high intensity,
extremely long (even longer than what was found herein) duration ground motion, accompanied by beatings).
Most of the extensions of the present work will have to be carried out first in the 2-D, shear horizontal wave context in order to discern
the essential issues. The extensions to the 2-D- P/SV (as in e.g. Faeh et al. (1994) and 3-D (as in e.g. Olsen 2000) cases with more general
types of sources Faeh et al. (1994) and Faeh & Panza (1994) are, of course, the requisites for a full understanding of what happens when a
seismic wave hits a realistic urban site.
AC K N OW L E D G M E N T S
This research was carried out partially within the framework of the Action Concertee Incitative Prevention des Catastrophes Naturelles
entitled Interaction site-ville et alea sismique en milieu urbain of the French Ministry of Research. We thank M. Celebi and and B.L.N.
Kennett for their permission to reproduce figures in their respective publications.
Carrier, G.F., Krook, M. & Pearson, C.E., 1983. Functions of a Complex

REFERENCES Variable, Hod Books, Ithaca.
Celebi, M., 2004. Responses of a 14-story (Anchorage, AK) building to far-
Abramowitz, M. & Stegun, A., 1986. Handbook of Mathematical Functions, distance (Ms=7.9) Denali (2002) and near distance earthquakes in 2002,
Dover, New York. Proc. 11th International Conference on Soil Dynamics & Earthquake En-
Alsop, L.E., 1970. The Leakymode period equationa plane-wave ap- gineering, pp. 895900, eds Doolin, D. et al., Stallion Press, Berkeley.
proach, Bull. seism. Soc. Am., 60, 19891998. Chavez-Garcia, F.J. & Bard, P.-Y., 1994. Site effects in Mexico City eight
Balendra, T. & Kong, K.H., 2004. Effects of ground motions in Singapore years after the September 1985 Michoacan earthquakes, Soil Dyn. Earth-
due to far-field earthquakes, Proc. 11th International Conference on Soil quake Engrg., 13, 229247.
Dynamics & Earthquake Engineering, pp. 255259, eds Doolin, D. et al., Chen, C.-H., Teng, T.-L. & Gung Y.-C., 1998. Ten-second Love-wave prop-
Stallion Press, Berkeley. agation and strong ground motions in Taiwan, J. geophys. Res., 103(B9),
Cardenas, M. & Chavez-Garcia, F.J., 2003. Regional path effects on seis- 21 25321 273.
mic wave propagation in central Mexico, Bull. seism. Soc. Am., 93, 973 Dennis, J.E. & Schnabel, R.B., 1983. Numerical Methods for Unconstrained
985. Optimization and Nonlinear Equations, Prentice-Hall, Englewood Cliffs.

C 2005 RAS, GJI, 163, 192224
Ewing, M., Jardetzky, W.S. & Press, F., 1957. Elastic Waves in Layered strong motion earthquake records, Technical Rept. California Institute of
Media, McGraw Hill, New York. Technology, EERL-80-01.
Faeh, D. & Panza, G.F., 1994. Realistic modeling of observed seismic motion Morse, P.M. & Feshbach, H., 1953. Methods of Theoretical Physics, McGraw
in complex sedimentary basins, Annal. Geofis., 37, 17711797. Hill, New York.
Faeh, D., Suhadolc, P., Mueller, S. & Panza, G.F., 1994. A hybrid method Novikova, E.I. & Trifunac, M.D., 1993. Duration of strong earthquake
for the estimation of ground motion in sedimentary basins: quantitative ground motion: physical basis and empirical equations, Technical Rept.
modeling of Mexico City, Bull. seism. Soc. Am., 84, 383399. University of Southern California, Dept. Civil Engineering, CE 93-02.
Furumura, T. & Kennett, B.L.N., 1998. On the nature of regional seismic Novikova, E.I. & Trifunac, M.D., 1995. Frequency dependent duration of
phases-III. The influence of crustal heterogenity on the wavefield for sub- strong earthquake ground motion: updated empirical equations, Techni-
duction earthquakes: the 1985 Michoacan and 1995 Copala, Guerrero, cal Rept. University of Southern California, Dept. Civil Engineering, CE
Mexico earthquakes, Geophys. J. Int., 135, 10601084. 95-01.
Groby, J.-P. & Tsogka, C., 2003. A time domain method for modelling Olsen, K.B., 2000. Site amplification in the Los Angeles basin from three-
wave propagation phenomena in viscoacoustic media, in Mathematical dimensional modeling of ground motion, Bull. seism. Soc. Am., 90, 77
and Numerical Aspects of Wave Propagation WAVES 2003, pp. 911915, 94.
eds Cohen, G.C. & Heikkola, E., Springer, Berlin. Panza, G.F., Vaccari, F. & Romanelli F., 2001. Realistic modelling of seismic
Groby, J.-P., Tsogka, C. & Wirgin, A., 2004. A time domain method input in urban areas: a UNESCO- IUGS-IGCP project, PAGEOPH, 158,
for modeling viscoelastic SH wave propagation in a city-like environ- 23892406.
ment, Proc. 11th International Conference on Soil Dynamics & Earth- Pekeris, C.L., 1955. The seismic surface pulse, Proc. Natl. Acad. Sci., 41,
quake Engineering, pp. 887894, eds Doolin, D. et al., Stallion Press, 469475.
Berkeley. Savage, B.K., 2004. Regional seismic wavefield propagation, PhD thesis,
Groby, J.-P. & Wirgin, A., 2005. 2-D ground motion at a soft viscoelastic California Institute of Technology, Pasadena.
layer/hard substratum site in response to SH cylindrical seismic waves Semblat, J.-F., Duval, A.-M. & Dangla, P., 2000. Numerical analysis of seis-
radiated by deep and shallow line sourcesI. Theory, Geophys. J. Int., mic wave amplification in Nice (France) and comparisons with experi-
doi:10111/j.1365-246X.2005.02712.x. ments, Soil Dynam. Earthquake Engrg., 19, 347362.
Hisada, Y., Yamamoto, S. & Tani S., 1988. Analysis of strong ground mo- Shoji, Y., Tanii, K. & Kamiyama, M., 2004. A study on the duration and
tion of plain and basin being composed of soft soil by fault model and amplitude characteristics of earthquake ground motions, in Proc. 11th
boundary element method, in Proc. Ninth World Conference on Earth- International Conference on Soil Dynamics & Earthquake Engineering,
quake Engineering, Paper 88, Japan Association for Earthquake Disaster pp. 157164, eds Doolin, D. et al., Stallion Press, Berkeley.
Prevention, Tokyo. Singh, S.K. & Ordaz, M., 1993. On the origin of long coda observed in the
Jennings, P.C., 1962. Velocity spectra of the Mexican earthquakes of 11 lake-bed strong-motion records of Mexico City, Bull. seism. Soc. Am., 83,
May and 19 May 1962, Technical Rept. California Institute of Technol- 12981306.
ogy, EERL.1962.002. Tsogka, C. & Wirgin, A., 2003. Simulation of seismic response in an ideal-
Jensen, F.B., Kuperman, W.A., Porter, M.B. & Schmidt, H., 1994. Compu- ized city, Soil. Dynam. Earthquake Engrg., 23, 391402.
tational Ocean Acoustics, AIP Press, New York, pp. 269276. Whittaker, E.T. & Watson, G.N., 1922. Modern Analysis, Chapters 5, 6,
Lee, D.M., Jennings, P.C. & Housner, G.W., 1980. A selection of important Cambridge University Press, Cambridge.
A P P E N D I X A : C O M P U T AT I O N O F U 1 ( X G , ) B Y T H E N U M E R I C A L
Q UA D R AT U R E M E T H O D
We show here how to obtain u 1 (x g , ), by direct numerical computation of the integral over the horizontal wavenumber variable k 1 .
We had:
u 1 (xg , ) = S()
+ s

i0 k20 () cos k1 x1 x1s ei [k2 ()(hx2 )]
0
i
dk1
0 k20 () i0 k20 () cos k21 ()h + 1 ()k21 () cos k21 ()h
+
= S() F(k1 , ) dk1 . (A1)
0
F(k 1 ) is exponentially decreasing for k 1 + so that:

+ K
u 1 (xg , ) = S() F(k1 , )dk1 S() F(k1 , ) dk1 , (A2)
0 0
with K a suitably chosen large number.

The integrand F(k 1 ) changes character at k 1 = k 0 . This suggests making the change of variables:
(1) in [0, k 0 ], k 1 ( ) = k 0 sin ( ), which leads to k 02 ( , ) = k 0 cos (), and to the integral:
i
I11 (xg , ) = S()

s

i0 k20 (, ) cos k1 ( ) x1 x1s ei [k2 (,)(hx2 )]
0
2
d,
0 i0 k20 (, ) cos k21 (, )h + 1 ()k21 (, ) cos k21 (, )h
(A3)
(2) in [k 0 , K ], k 1 (
) = k 0 cosh (
), which leads to k 02 (
, ) = ik 0 sinh (
), and to the integral:

C 2005 RAS, GJI, 163, 192224

I21 (xg , ) = S() F21 (xg , ,
)d
, (A4)
0
wherein := acosh( kK0 ) and

0 s

1 i0 k20 (
, ) cos k1 (
) x1 x1s e[k2 (
,)(hx2 )]
F2 (xg , ,
) :=
1
, (A5)
i0 k20 (
, ) cos k21 (
, )h + 1 ()k21 (
, ) cos k21 (
, )h
so that
u 1 (xg , ) = I11 (xg , ) + I21 (xg , ). (A6)
Special attention must be paid to the numerical evaluation of I 12 because its integrand is an oscillating function and because singularities
(whose existence are characterized by large peaks in the integrand, and which correspond to the existence of Love modes) are encountered in
the interval [0, ]. Thus, we isolate the locations
m (m = 1, 2, . . , M) of the zeros of the real and imaginary parts of the denominator of
F 12 (x g , ,
) so as to obtain:

1 1
m+1 1m+1
M1
1
I2 (xg , ) = S()
1
F2 (xg , ,
)d
+ S()
1
F21 (xg , ,
) d
0 m=1
m +2m
M

m +2m
+ F21 (xg , ,
) d
+ F21 (xg , ,
) d
. (A7)
m=1
m 1m
M +2M
Each quadrature is subsequently carried out by the trapezoidal method, with the following rules:
(1) for I 11 : 1500 sampling points

(2) for I 12 : K = k 0 cosh (9), with ( m1 + m2 ) a constant equal to K /10 000 and,
1 1
m+1 1m+1
(3) if poles are encountered: 3000 sampling points are employed in the integrals 0
1
and
m +2m (m = 1, 2, ... , M), 10 000 sampling

m +m
points for
M +m , and 5 sampling points for
m m2 (m = 1, 2, ... , M)
2 1
(4) if no poles are encountered, 10 000 sampling points are employed.
A P P E N D I X B : N U M E R I C A L A S P E C T S O F T H E C AU C H Y T H E O R E M ( C T ) M E T H O D
F O R T H E C O M P U T AT I O N O F U1 ( XG , )
B1 The basic formulae
The Fourier spectrum of the displacement field on the ground is:

+

0 exp ik1 x1 x1s + ik20 (x2s h)
u (xg , ) = S()
1
dk1 , (B1)
2 i0 k20 cos k21 h + k21 1 ()sin k21 h

wherein, k20 = (k 0 )2 k12 and k21 = (k 1 )2 k12 .
0 and 0 are real and 1 and 1 are generally complex frequency-dependent parameters. Thus, k 0 is a real number and k 1 a generally
complex function of .
To show clearly the influence of the Love modes on the ground response, we carry out the integration in the complex k 1 = k 1 + ik 1
plane via Cauchys theorem (Whittaker & Watson 1922; Carrier et al. 1983).
We shall attempt to evaluate the auxiliary integral

0 exp ik1 x1 x1s + ik20 x2s h
I (xg , ) = S() 1 1 dk1 :
C i k 2 cos k2 h + k 2 ()sin k2 h
2 0 0 1 1
(B2)
= F (k1 , xg , ) dk1 .
C
The contour C will presently be defined so as to enable u 1 (x g , ) to be computed from I (x g , ).
B2 Choice of path of integration in the complex k1 plane
For it to be possible to evaluate I (x g , ) by means of Cauchys theorem, the integrand F(k1 , xg , ) must be a single-valued function of k 1
along C.

C 2005 RAS, GJI, 163, 192224
Figure B1. The EJP integration contour C ABC D E F G H I J K A in the complex k 1 plane. The branch cuts do not depend on whether the layer is dissipative or
not, contrary to the location of the poles. Here we depict the location of the poles for the non-dissipative layer case. These poles shift upwards from the k 1 axis
when dissipation is present (see also Fig. B6).
All functions (i.e. cos (k 12 h) and k 12 sin (k 12 h)) involved in B.2 are even functions of the transverse wave number k 12 , so it doesnt matter
which sign of the square root is chosen for k 12 .
1 1
The integrand F(k1 , xg , ) is a multi-valued function of k 1 because of the two branches of the function k20 = (k 0 + k1 ) 2 (k 0 k1 ) 2 .
To make the integrand single-valued, one must introduce two cuts in the complex k 1 plane running from the branch points k 1 = k 0
(both located on the real k 1 axis due to our assumption that the substratum is non-dissipative) to infinity (or joining the two branch points
directly).
There exist a variety of choices of branch cuts. For instance, the so-called Pekeris cut (Pekeris 1955; Jensen et al. 1994) offers the advantage of
giving a very accurate representation of the ground response, particularly suitable for small epicentral distances. However, this choice is less
suitable for large epicentral distances and involves the difficult process of computation of the eigenvalues of the leaky modes (Alsop 1970;
Jensen et al. 1994).
The so-called EJP cuts K1 , K2 (Ewing et al. 1957; Jensen et al. 1994) constitute another and probably the simplest choice. Cut K1 runs
from k 1 = k 0 + i0 to k 1 = 0 + i0 and then from k 1 = 0 + i0 to k 1 = 0 + i. Cut K2 runs from k 1 = k 0 + i0 to k 1 = 0 + i0 and then from
k 1 = 0 + i0 to k 1 = 0 i. Since we assume that x s2 h is positive, and we choose observation points on the ground such that x 1 x s 1
0, then the appropriate (closed) path of integration in the complex k 1 plane is the one (C = C ABC D E F G H I J K A ) depicted in Fig. B1 which only
involves integration along the cut K1 (depicted by dark lines in Fig. B1).
Note that it is implicit that R . Furthermore, C AB , C BC and CC D are paths on the first Riemann sheet relative to K1 , CC D is a small
semicircular path of radius that disappears in the limit 0, C E F , C F G , CG H , C H I are paths on the second Riemann sheet relative to K1 ,
with C F G disappearing in the limit 0. In the limit 0, the path C I J becomes a path along the real k 1 axis from to 0, C J K disappears,
and C K A becomes a path along the real k 1 axis from 0 to .
This choice of integration contour is interesting only insofar as it leads to the disappearance of C AB and C H I in the limits R ,
0, 0. That this is so, is easily demonstrated.
B3 Application of Cauchys theorem
Cauchys theorem indicates that

F (k1 , xg , ) dk1 = 2i r j (xg , ) (B3)
C ABC D E F G H I J K A j
wherein rj is the residue of the jth pole located within C ABC D E F G H I J K A . It was shown in the companion paper (Groby & Wirgin 2005) that
when the layer is non-dissipative, the poles that belong to the first Riemann sheet all lie on the real axis in the complex k 1 plane and that there
are no poles on the second Riemann sheet for the EJP integration path (see also Jensen et al. 1994).
Decomposing the integration path into its parts gives

lim0 lim0 lim R F dk1 + F dk1 + Fdk1 + Fdk1 +
C AB C BC CC D CD E

F dk1 + F dk1 + F dk1 + F dk1 +
C C C CH I
EF FG GH
F dk1 + Fdk1 + F dk1 = 2i r j (xg , ), (B4)
CI J CJ K CK A j

C 2005 RAS, GJI, 163, 192224
12
10
2 |B|
6
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45
(Hz)
10
6
2 |B|
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45
(Hz)
Figure B2. Branch cut contributions at the MC site for a source, located at xs = (0 m, 100 m), radiating a 0 = 0.5 Hz pseudo-Ricker pulse. The dashed curves
designate 2 B versus frequency , the dotted curves 2 B+ versus , and the solid curves 2B versus . The top panel applies to x = (0 m, 1200 m),
the bottom panel to x = (0 m, 3000 m).
and, on account of previous remarks:

u 1 (xg , ) = u 1B (xg , ) + u 1R (xg , ), (B5)
wherein the branch cut contribution to u 1 (x g , ) is

u 1B (xg , ) := Fdk1 Fdk1 Fdk1 F dk1 , (B6)
C BC CC D CE F CG H
and the pole residue contribution to u 1 (x g , ) is

u 1R (xg , ) := 2i r j (xg , ). (B7)
j
B4 The branch cut contribution

On C BC we have k1 = ik1 , k20 = +k20+ := + (k 0 )2 + (k1 )2 , so that

0 s 0+ s
0 ek1 (x1 x1 ) eik2 (x2 h)
F dk1 = S() dk1 . (B8)
i k2 cos(k2 h) + k 2 sin(k 2 h)
2 0 0+ 1 1 1 1
C BC

On CC D we have k1 = k1 , k20 = +k20 := + (k 0 )2 (k1 )2 , so that

k0 s 0 s
0 eik1 (x1 x1 ) eik2 (x2 h)
Fdk1 = S() dk1 . (B9)
CC D 2 0 i0 k20 cos(k21 h) + k21 1 sin(k21 h)

C 2005 RAS, GJI, 163, 192224
On C E F we have k 1 = k 1 , k 02 = k 0
2 , so that

0 s 0 s
0
eik1 (x1 x1 ) eik2 (x2 h)
F dk1 = S() dk1 . (B10)
k 0 i k 2 cos(k2 h) + k 2 sin(k2 h)
2 0 0 1 1 1 1
CE F
On CG H we have k 1 = ik 1 , k 02 = k 0+
2 , so that

s 0+ s
0 ek1 (x1 x1 ) eik2 (x2 h)
F dk1 = S() dk1 . (B11)
CG H 2 0 i0 k20+ cos(k21 h) + k21 1 sin(k21 h)
Thus,

s
ek1 (x1 x1 )
1
u 1B (xg , ) = u 1B+ (xg , ) + u 1B (xg , ) = S()
0
k20+

cos k21 h cos k20+ (x2s h) + sin(k21 h) sin k20+ (x2s h)
dk1
cos2 k21 h + ( + )2 sin2 k21 h

k0 s
i eik1 (x1 x1 )
+S()
0 k20

cos(k21 h) cos k20 (x2s h) sin(k21 h) sin k20 (x2s h)
dk1 , (B12)
cos2 (k21 h) + ( )2 sin2 (k21 h)
wherein
k21 1
= 0 . (B13)
k2 0
The two integrals in (B.12) (whose integrands are well behaved due to the fact that there are no poles) are evaluated numerically in the manner
outlined in Appendix A.
To illustrate the relative contributions of the branch cut integrals as a function of epicentral distance, we depict, in Fig. B2, B+ :=
u 1B+ (xg , )/S(), B := u 1B (xg , )/S(), and B := (B+ + B )/S(). This is done for two ground locations (x = (1200 m,
0 m), x = (3000 m, 0 m)) at the MC site, for a source, located at xs = (0 m, 100 m), radiating a 0 = 0.5 Hz pseudo-Ricker pulse. One notices
that the branch cut contributions diminish as the epicentral distance increases.
B5 The pole residue contribution
To make the exposition as clear as possible, we only consider the case in which a single (Love mode) pole is included within the EJP contour
(this occurs for low frequencies). Furthermore, we assume that this pole is a first-order pole.
B5.1 Expression of the Residues

f (k1 ,)
The residue rj of C g(k 1 ,)
dk1 at a first-order pole k1 = k1 j ( j N) is (Whittaker & Watson 1922):
f (k1 j , )
r j () =
g(k1 ,)
. (B14)
k1 k1 =k1 j
Note that, at present, only one value of k 1 is assumed to lead to g(k 1 , ) = 0; we call this first-order pole k 1 . Consequently:

1 exp ik1 (x1 x1s ) + ik20 (x2s h)
r1 = S() 1 k =k . (B15)
2 i0 i0 hk20 1
k1 k20
1 h cos k21 h + k1 k1
k1
sin k 2 h 1 1
2 2
The next step is to find k 1 for a generally dissipative layer.
B5.2 Non-ambiguous determination of the first Love mode pole for a generally dissipative layer
Dissipation induces a perturbation of the non-dissipative case, which fact suggests looking for the zeros of the Love mode dispersion relation
in the dissipative case starting with the zeros of the Love mode dispersion relation relative to the non-dissipative case. The Love mode relation
dispersion takes the form (in both the non-dissipative and dissipative cases):

i0 k20 cos k21 h + 1 k21 sin k21 h = 0. (B16)

C 2005 RAS, GJI, 163, 192224
0.35
0.3
0.25
0.2

0.15
0.1
0.05
0
0 1 2 3 4 5 6 7 8
Figure B3. as a function of in the MC site with h = 50 m.
0.06
0.05
0.04
k1
0.03
0.02
0.01
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
(Hz]
Figure B4. The root of (B16) for the MC site with h = 50 m and Q1 taken to be infinite (i.e. the non-dissipative case). The dotted curve represents (k 0 ) as a
function of frequency , the dashed curve (k 1 ) as a function of , and the solid curve (k 1 ) as a function of .
In the non-dissipative case, 1 and c1 are real constants so that k 1 (as well as k 0 , due to the fact that we assumed that 0 and c0 are always real
constants) possess the attractive feature of being independent of frequency.
The general procedure for obtaining the zeros of (B16) in both cases was outlined in the companion paper (Groby & Wirgin 2005). Here,
we look into this matter in more detail.
We make the changes of variables:

k1 k1 2
= k h, = 0 , =
0
0
2 , = 2 1, (B17)
k k
so that (B16) takes the form:
0 cos () + 1 sin () = 0 (B18)
which is solved in exact manner for :

0
1
= Arctan , (B19)
1
the Arctan function becoming arctan (i.e. the principal value of Arctan) for the lowest-order Love mode. We depict as a function of in
Fig. B3. The difficulties related to this procedure are that:
(1) we are actually looking for complex root k 1 (or the related quantity ), which is (are) a function of , in the dissipative case. (resp.
) is simply related to (resp. k 1 ) through the relation c0 = h (resp. k 1 = /c0 ). Parameters and are both real. Thus, to get as a
function of we simply invert the file representing as a function of .
(2) the asymptotic behaviour of in the neighborhood of = 1 (see Fig. B3, shows that the frequency must be stepped in non-linear
manner in this neighborhood.
The result of this procedure in the non-dissipative case is depicted in Fig. B4.

C 2005 RAS, GJI, 163, 192224
Figure B5. MC site for h = 50 m (i.e. with dissipation). Top panel: (k 0

2 ) as a function of frequency . Bottom panel: (k 2 ) as a function of . Note the
0
difference in scales of the ordinates in the two panels.
We now turn to the dissipative case. A possibility for solving (B16) is to employ Mullers algorithm (a variant of the secant method (Dennis
& Schnabel 1983)) for finding a complex root of a complex function. We found that this does not work very well, even when associating the
initial guess with the real roots of the non-dissipative case.
The trick is to choose a more appropriate variable in (B16). Rather than search for k 1 as a function of frequency, we found it better to
search for k 02 as a function of so that the dispersion relation (B16) becomes:

0 2 0 2 0 2
i0 k20 cos h (k ) (k ) + k2
1 2 0 2 + (k ) (k ) + k2 sin h (k ) (k ) + k2
1 1 2 0 2 1 2 0 2 = 0, (B20)
by means of which we find the result depicted in Fig. B5. We see in this figure that dissipation in the layer (as in the MC site) leads (at all
except perhaps the lowest frequencies) to a Love mode associated predominantly with an evanescent wave in the substratum (i.e.|| (k 02 ||
>>||(k 02 ||). In the non-dissipative case, this wave becomes
a purely evanescent wave.

Once we have k 0 2 (), we compute k 1 () via k 1 = (k 0 )2 (k20 ())2 . This gives the result depicted in Fig. B6.
B5.3 Structure of the pole residue contribution to the displacement on the ground
Still assuming that the frequency is so low that only one (the fundamental) Love mode can be excited, the pole residue contribution to the
ground displacement becomes (see (B7))

exp ik1 (x1 x1s ) + ik20 (x2s h)
u 1R (xg , ) := 2ir1 (xg , ) = S() , (B21)
D(k1 , )
with
2
k20 := (k 0 )2 k1 (B22)
and

0 0
i i0 1 i hk2 1 1
D(k1 , ) = k 1 1
h cos k h + k 1 sin k h , (B23)
0 k20 2
k21 k21 2
k1 =k1

C 2005 RAS, GJI, 163, 192224
Figure B6. MC site (i.e. with dissipation) for h = 50 m. Top panel: the dotted curve represents (k 0 ) as a function of the frequency , the dashed curve (k 1 )
as a function of , and the solid curve: (k 1 ) as a function of . Bottom panel: the dashed curve represents (k 1 ) as a function of and the solid curve (k 1 )
as a function of . Note the differences in scales of the ordinates in the two panels.
k1 = k1 + ik1 , k20 = k20 + ik20 , (B24)

with
k1 0, k1 0, k20 0, k20 0. (B25)
Consequently

u 1R (xg , ) = A(xg , ) exp ik1 (x1 x1s ) + ik20 (x2s h) , (B26)
wherein
S()
A(xg , ) :=
exp k1 (x1 x1s ) k20 (x2s h) . (B27)
D(k1 , )
This result shows that the residue term in the ground response, associated with the fundamental Love mode, takes the form of an inhomogeneous
plane wave. The amplitude A and the oscillation period of this wave decrease as the epicentral distance x 1 x s1 increases. Furthermore A is
all the larger the closer the source is to the boundary between the layer and the substratum (i.e. the smaller is x s2 h). This behaviour is what
is generally observed in the main (Love mode) peak of the spectra (and transfer functions) of ground response.
B6 The combined branch cut and pole residue contributions

Although the pole residue contribution u 1R (xg , ) to u 1 (x g , ) has a relatively simple analytic (plane wave) form, the same cannot be said of
the branch cut contribution u 1B (xg , ) (see (B.12). This seems to make it difficult to account for the specific features of u 1 (x g , ) by means of
the expression one obtains from the Cauchy theorem (i.e. u 1 = u 1B + u 1R ).
Nevertheless, some typical features seem to emerge from the numerical computation of the terms u 1B , u 1R and of their sum. Consider
the transfer functions depicted in Fig. B7. It is seen in the top panel of this figure that, at the epicentral distance of 1200 m, the dominant

C 2005 RAS, GJI, 163, 192224
Figure B7. The various contributions to 2 times the modulus of the displacement transfer function, as computed by the Cauchy Theorem (CT) method,
relative to the MC site with h = 50 m, for xs = (0 m, 100 m) and excitation by a pseudo-Ricker pulse with 0 = 0.5 Hz. The thin continuous curves pertain
to the transfer function as computed by the CT method. The dashed curves pertain to the residue contribution to the transfer function. The dash-dot curves
pertain to the branch cut along [0, k0] contribution to the transfer function. The dotted curves pertain to the branch cut along [0, i] contribution to the transfer
function. Top panel: x = (1200 m, 0 m). Bottom panel: x = (3000 m, 0 m).
contribution to the total response, notably around the Love mode resonance peak, comes from the pole residue, whereas the oscillations beyond
(i.e. for higher frequencies) this peak arise from a complex interplay between the residue and branch cut integral terms. The same comments
apply to the larger epicentral distance of 3000 m (bottom panel), with, however, the sole residue term accounting more accurately for the real
height of the resonance peak, and the oscillations beyond this peak being of higher frequency than previously. This latter fact is probably due
to: i) the domination of the pole residue term and ii) the fact (pointed out previously) that the frequency of oscillation of this term (the latter

being of the form A exp[ik1 (x1 x1s ) + ik20 (x2s h)]) increases with increasing epicentral distance x 1 x s1 . The dominant nature of the
pole residue term and the fact that A is an exponentially decreasing function of the epicentral distance account for the fact that the moduli of
the pole residue and total transfer functions decrease with increasing epicentral distance x 1 x s1 .
This means, in particular, that the pole contribution associated with the Love mode dominates the total seismic response at large
epicentral distances. This point has important consequences as concerns the issue of regional earthquake response, and is further discussed in
Section 6.

C 2005 RAS, GJI, 163, 192224

ArticleGJI 2005 2

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

ArticleGJI 2005 2

Hochgeladen von

Copyright:

Verfügbare Formate

Geophys. J. Int. (2005) 163, 192224 doi: 10.1111/j.1365-246X.2005.02713.

Two-dimensional ground motion at a soft viscoelastic layer/hard

Jean-Philippe Groby1 and Armand Wirgin2

field examples taken from published literature.

1.1 Regional path effects

1.2 Effects of ground and underground irregularities

Table 1. Material and geometrical parameters of the site, source parame-

1.3 The questions addressed herein

2.1 Description of the generic bare urban site

2.2 Spacetime and spacefrequency domain representations of the impulsive force

2.3 Spectra and time histories of various input pulses

2.4 Site parameters

2.6 Method of computation of transfer functions

4.6 The influence of the epicentral distance on ground response

4.6.1 Transfer functions

5.1 Choice of site, source and observation point

2|u (xg,)| (mm)

u1(x ,t) (mm)

u1(x ,t) (mm)

5.2 Spectra and time histories of ground response

We shall now address the questions raised in Section 1.3.

2|u (xg,)| (mm)

u1(x ,t) (mm)

2|u (xg,)| (mm)

2|u (xg,)| (mm)

0 50 100 150 200 5

Figure 29. Ground motion at large epicentral distance (x 1 s1 x

Carrier, G.F., Krook, M. & Pearson, C.E., 1983. Functions of a Complex

F(k 1 ) is exponentially decreasing for k 1 + so that:

with K a suitably chosen large number.

wherein := acosh( kK0 ) and

u 1 (xg , ) = I11 (xg , ) + I21 (xg , ). (A6)

(1) for I 11 : 1500 sampling points

B1 The basic formulae

The Fourier spectrum of the displacement field on the ground is:

The contour C will presently be defined so as to enable u 1 (x g , ) to be computed from I (x g , ).

B2 Choice of path of integration in the complex k1 plane

B3 Application of Cauchys theorem

Cauchys theorem indicates that

and, on account of previous remarks:

and the pole residue contribution to u 1 (x g , ) is

B4 The branch cut contribution

B5 The pole residue contribution

B5.1 Expression of the Residues

The next step is to find k 1 for a generally dissipative layer.

Figure B3. as a function of in the MC site with h = 50 m.

Figure B5. MC site for h = 50 m (i.e. with dissipation). Top panel: (k 0

difference in scales of the ordinates in the two panels.

k1 = k1 + ik1 , k20 = k20 + ik20 , (B24)

B6 The combined branch cut and pole residue contributions

Das könnte Ihnen auch gefallen

The next step is to find k 1 for a generally dissipative layer.

Figure B5. MC site for h = 50 m (i.e. with dissipation). Top panel: (k 0

k1 = k1 + ik1 , k20 = k20 + ik20 , (B24)