J Geophys (1985) 58:153-174 Journal of

Geophysics

The reflectivity method: a tutorial

G. Miiller
Institute of Meteorology and Geophysics, University of Frankfurt, Feldbergstr. 47, 6000 Frankfurt,
Federal Republic of Germany

Abstract An extended reflectivity method is described Currently, however, the interest of seismologists and
by which complete seismograms for a point source in a theoreticians focuses strongly on methods which allow
layered half-space can be calculated. Starting with the the treatment of laterally heterogeneous media. This de-
differential equations and boundary conditions, the re- velopment has been prompted by many clear percep-
flection and transmission of plane waves at layered me- tions, collected in several disciplines of geosciences, that
dia is treated first, followed by the synthesis of point- the earth is a dynamic body with relatively rapid in-
source wave fields. The frequency-domain displace- ternal motions. These processes and related horizontal
ments of the half-space surface are expressed as slow- temperature differences may lead to pronounced lateral
ness integrals, and the most prominent parts of the in- variations in elastic and anelastic properties. Thus, for
tegrands are the reflectivities of the layers below and the seismologist, who is interested in depth ranges from
above the point source and a function which is closely the lower crust to the core, the earth becomes what it
related to the transmissivity of the layers above the has always been for those engaged in seismic prospect-
source. Reflectivities and transmissivities are calculated ing: a medium with truly three-dimensional inhomo-
by recursive methods which are numerically stable for geneities.
all frequencies and slownesses. Near- and far-field re- Methods for horizontally stratified earth models
sults are given for single-force and moment-tensor will, of course, continue to have their importance and
point sources. From the general results for the com- find applications in studies of structural properties and
plete medium response, partial responses can easily be sources of seismic waves. Hence, a good understanding
extracted, e.g. the original form of the reflectivity meth- of the relevant theory will always be a necessity. The
od which calculates only the response from the layers purpose of these lecture notes is to contribute to this
below the source. Thus, the extended reflectivity meth- understanding by presenting a self-contained theory of
od has a flexibility which is not available if propagator one of the methods for horizontally layered media, the
methods are used for the calculation of the integrands. reflectivity method, including all material that is neces-
Various other aspects of seismogram calculation are sary for the development of corresponding computer
addressed, such as extended sources, an earth-flattening programs for theoretical seismograms.
transformation and the inclusion of absorption for con- Before we start with the details, a few words about
stant and frequency-dependent Q. Theoretical seismo- the general scene of methods for horizontally stratified
grams are shown, first for body-wave propagation from media are in order. This scene is very vast now, and it
explosions in a crustal model and in a model which is practically impossible to mention all the different
came from seismic prospecting, and second for surface- methods that are in use. Much background material
wave propagation from a double-couple source. can be found in the text books and monographs by Pi-
Due to the tutorial nature of this paper the de- lant (1979), Aki and Richards (1980), Ben-Menahem
rivations are mostly rather detailed. It is hoped that and Singh (1981) and Kennett (1983). The most impor-
this will help interested newcomers to the field of tant theories and methods for wave propagation and
theoretical seismograms to get started. seismogram synthesis are the following (the references
given are only examples and far from complete):
Key words: Reflectivity method - Reflection and trans-
Generalized ray theory (Helmberger, 1968; Muller,
mission coefficients - Synthetic seismograms - Dissi-
1969; Ben-Menahem and Vered, 1973): the medium is
pation
approximated by homogeneous layers, and the wave
field is decomposed into elementary seismograms corre-
sponding to rays.
Full-wave theory (Cormier and Richards, 1977): a ray
1. Introduction
theory for inhomogeneous layers which takes account
The theory of seismic wave propagation in horizontally of frequency-dependent effects connected, e.g., with
stratified media, i.e. in media whose elastic and anelas- caustics and shadow zones.
tic properties depend only on depth, has been a main WKBJ theory (Chapman, 1978): a ray theory for in-
subject of research in seismology over the past decades. homogeneous layers which is more limited than full-
154

wave theory, as far as frequency-dependent effects are start with the displacement potentials for a point
concerned, but which allows very rapid computations. source in a homogeneous medium and their well-
Wavenumber or slowness integration methods (Kind, known representation by Sommerfeld integrals in the
1978; Cormier, 1980; Wang and Herrmann, 1980; wavenumber or slowness domain. Similar representa-
Ingate et al., 1983; Ha, 1984): representation of the tions are valid for the displacement components at
Fourier-transformed wave field of a layered medium by points with arbitrary location in a layered medium.
integrals over horizontal wavenumber or slowness; the The upgoing wave field in the layer with the source
reflectivity method belongs to this class of methods (see is synthesized as the sum of the direct wave from the
below). source and all possible interactions between the layer
Wavenumber summation methods (Alekseev and Mik- stack above the source and that below the source.
hailenko, 1980; Bouchon, 1981; Korn and Muller, The interaction terms contain the reflectivities of
1983; Spudich and Ascher, 1983; Olson et al., 1984; these zones: in the case of P-SV waves, they are
Campillo et al., 1984): both time- and frequency-do- 2 x 2 matrices whose elements are the P - P, SV- P,
main methods which are in principle very similar to P-SV, SV -SV reflection coefficients; in the case of SH
wavenumber integration methods, but the continuous waves, they are scalars. The upgoing wave field in the
distribution of wavenumbers is replaced by a discrete source layer is then transmitted through the layer stack
one. above the source with the aid of the transmissivity of
Modal summation method (Harvey, 1981; Panza, 1985): this zone, and the complete wave field in the half-space
representation of the wave field by normal modes of on top of the medium is obtained. An analytical limit-
Rayleigh and Love waves alone, either with the as- ing process which turns this half-space into vacuum
sumption of a perfect reflector at depth or without. gives, for points on the uppermost interface, the final
The first three of these methods are suitable for the expressions for the displacement components. The basic
calculation of body-wave contributions to seismograms, ideas of this wave field synthesis follow Kennett and
whereas the last three also allow surface waves to be Kerry (1979) and Kennett (1983), but some details are
in~luded, i.e. these methods are methods for complete different - mainly the treatment of transmission
se1smograms. through the layers above the source. The theory is pre-
The reflectivity method, which is described below in sented here for the simplest type of point source,
more detail, is a wavenumber or slowness integration radiating P, SV and SH waves, a single force of arbi-
method. The name stems from the fact that the func- trary orientation. Final results for a general moment-
tion which is integrated is mainly the reflection coef- tensor point source, including an explosion, are also
ficient or reflectivity of a layered medium. At first, this given.
is true only for a layered medium without a free surface In Section 5 we discuss the possibility of extracting,
and with a source on one side of those layers whose from the expressions for the complete wave field, sim-
reflections are sought; this is the geometry for which pler expressions which represent only partial responses
the reflectivity method was developed originally (Fuchs, of the layered medium. Moreover, compact far-field for-
1968; Fuchs and Muller, 1971). However, Kennett and mulas for extended sources consisting of several point
Kerry (1979) and Kennett (1983) have shown that in sources are compiled, and a few computational aspects
the case of a layered half-space, having a free surface of the reflectivity method are mentioned. Section 6 first
and a source at arbitrary depth, the integrand of the presents an earth-flattening transformation by which
wavenumber or slowness integrals can be expressed the reflectivity method can also be used for media with
mainly by the reflectivities of two partitions of the spherical geometry; it is adequate for many body and
medium - the layers above the source and those below. surface waves propagating in a spherical earth. Then
This form of the theory justifies the continued use of we describe how the theory of wave propagation for
the name reflectivity method. However, we emphasize purely elastic media is modified for dissipative media.
that there is no difference to the wavenumber or slow- Under the usual, experimentally confirmed assumption
ness integration methods quoted above in which the of a linear relation between stress and strain also in
integrand is calculated differently, namely by matrix or dissipative media at low strains, these media are com-
propagator techniques. That notations here indeed have pletely described by complex elastic moduli and hence
a certain degree of arbitrariness, is illustrated by the fact by complex wave velocities. A few possibilities of
that Luco and Apse! (1983) and Apse! and Luco (1983), choosing these velocities, according to non-causal or
whose treatment of the layered half-space comes closest causal absorption, are described. Finally, in Section 7
to the results presented below, call their method simply we present results of some synthetic-seismogram calcu-
a wavenumber-integration method. lations.
These lecture notes are organized as follows. After a
discussion of the differential equations for wave propa-
gation in horizontally stratified media and of the 2. Differential equations and boundary conditions
boundary conditions in Section 2, we derive in Section We assume that the medium consists of homogeneous
3 the reflection and transmission coefficients (or reflec- layers, separated by first-order discontinuities. If a me-
tivities and transmissivities) for plane waves, incident dium is continuously inhomogeneous (throughout or
on a plane interface or a stack of homogeneous layers piecewise), it is replaced by a sufficiently large number
separating two homogeneous half-spaces. The coef- of homogeneous layers; in smooth gradient zones it is
ficients for an interface are given analytically and those usually enough to choose roughly half the dominant
for a stack of layers are derived by a recursive algo- wavelength as layer thickness, whereas in transition
rithm first described by Kennett (1974). In Section 4 we zones with larger velocity gradients the layer thickness
 155

should be reduced further. The advantage of the homo- By inserting Eq. (5) into Eq. (1), again wave equations
geneous-layer approximation is that inside each layer are found, this time for all potentials:
the equation of motion takes a relatively simple form.
Its disadvantage is that boundary conditions have to be 2 1
V x=fJ2 X,w (7)
fulfilled at many interfaces. Analytical methods for in-
homogeneous layers (in contrast to numerical, e.g. fi-
nite-difference, methods) are not yet developed to a The second of these equations can be used to simplify
point where they really can compete with the methods the displacement component u 2 in Eq. (6):
for homogeneous layers.
The equation of motion of a homogeneous, isotro- (6')
pic elastic medium is

pu,tt =(A+ 2µ) grad div u- µ rot rot u, (1) The boundary conditions require continuity of the trac-
tion and displacement vectors across internal interfaces
where u is the displacement vector, p the density and ,1. in solid media. At a free surface the traction vanishes,
and µ the Lame parameters. Body forces due to gravity and the displacements are unspecified.
and seismic sources are not included in Eq. (1): it is If Cartesian coordinates are used, the traction com-
assumed that gravity has no other effect than to de- ponents, i.e. normal and tangential stresses, are:
termine, via self-compression, the (constant) values of p,
,1. and µ, and sources of seismic waves are included
through their known contributions to u (see Section 4).
Next we introduce displacement potentials, from Inserting Eq. (2) here, we obtain the first three bound-
which the displacements follow by spatial differen- ary conditions which require continuity of the quan-
tiation. For the first case of reflection and transmission tities
that we consider later, i.e. reflection and transmission of 1
plane waves (Section 3), we use Cartesian coordinates µ(2¢,xz-21/J.zz+ fJ21/J,tt), µuy,z'
(x, y, z), place the interfaces at constant values of z and
,1.
assume independence of the y coordinate. Then it is IX2 c/>,tt+2µ(¢.zz+I/J.xz) (8a)
most appropriate to derive the displacements ux and uz
from potentials ¢ and 1/J: at all interfaces, including a free surface where these
(2) quantities vanish. The conditions for displacements at
internal interfaces additionally require continuity of the
For the displacement uy, no potential is used. Inserting quantities
the corresponding representation of the displacement
vector (8b)

(3) In the case of cylindrical coordinates, the relevant

stresses are
(ey=unit vector in y-direction) into Eq. (1), one obtains
wave equations for ¢, 1/1 and uY: Pz,=µ(uz,,+u,), Pzq:,=µ (uq:,,z+~uz,q:,),

2 1 Pzz = ,1. div u+ 2 µ uz,z·

V 1/1= fJ2 I/J,i,, (4)
Inserting Eqs. (6) and (6'), one finds those quantities
which are continuous at interfaces, in analogy to Eqs.
Here, 17 2 = a 2 /ax 2 + a 2 /az 2 is the Laplace operator in (8a) and (8b). These quantities are not reproduced here.
two dimensions, ix= [(,1_ + 2 µ)/ pJ112 the velocity of P Closer inspection shows that they are continuous if the
waves and fJ=(µ/p) 112 the velocity of S waves. Equations simpler quantities
(4) imply decoupled propagation of P and S waves
within the layers.
In the second case considered below, reflection and
transmission of waves from a point source (Section 4),
µ ( c/>,z +2 P,zz - ;2 P,tt),

cylindrical coordinates (r, <p, z) are most appropriate. }, ( 1 ) (9a)

Then we use displacement potentials for all displace- µx,z, IX2 c/>,tt+2µ c/>,zz+ P,zzz- fJ2 P,ttz
ment components and obtain instead of Eqs. (2) and (3) and
(e,=unit vector in z-direction):
(9b)
u= grad¢+ rot rot(e, P) + rot(e, x) (5)

1 are continuous; these quantities contain no derivatives

u, =c/> , ,+ p , ,z+-x
r m ,y with respect to the horizontal coordinates r and <p. For
1 1 instance, if the first two functions in Eq. (9b) are con-
y r m+-r p mz-X
Um=-cp •Y •Y •
r (6) tinuous, so are the displacements u, and u"' in Eq. (6).
If the medium contains liquid layers, e.g. in cases
1 1 where wave propagation through the ocean or the
Uz = cp z- lJI rr - - lJI ,--2 lJI mm•
· · r ' r •YY earth's core is to be modelled, the displacement vector
156

is irrotational, u = grad </>, and the boundary conditions Case

are in principle different from those given above. How-
ever, computational experience shows that liquid layers 2 X
are modelled perfectly by solid layers whose S velocity
is of the order of 0.001 times the P velocity. Thus, algo- z
rithms for solid media are sufficient for all practical Fig. 1. Incidence of a plane wave at the interface z = 0 be-
purposes. tween two half-spaces 1 and 2. 9 is the angle of incidence
The continuity of the quantities in Eq. (8a and b) at
interfaces implies that interaction takes place between
P waves, derived from </>, and S waves, derived from i/1. ber k, in fulfilment of Snell's law. The relation between
The S waves, represented by uY and polarized horizon- k and the angle of incidence 9 (Fig. 1) is
tally, propagate independently. The S waves derived
from ijJ are called SV waves and those derived from u , k=~sin9. (11)
IX1
SH waves. Similar conclusions follow from (9a and b):
'P is the potential of the SV waves, and x the potential The vertical wavenumbers 11 _2 and l~, 2 are
of the SH waves. Note that, according to Eq. (6), SV
w2 )1;2 ( w2 )1;2
waves contribute also to the horizontal transverse com- 11,2 = (T-k2 ' l~.2 = ~/32 -k2 ' (12)
ponent u"' and SH waves to the horizontal radial com- 1, 2 1,2
ponent u,.. Both contributions are, however, near-field and the signs in front of them in Eq. (10) are negative
terms for point sources. In the far-field of point sources, (positive) for propagation in positive (negative) z-direc-
SV waves are polarized in rz-planes and SH waves, per- tion.
pendicular to r z-planes. The boundary conditions for z = 0 can be formulated
with the ¢- and ijJ-dependent quantities in Eq. (8a and
3. Plane waves in layered media b) and with Eq. (10). For instance, continuity of the
horizontal displacement ux yields:
3.1. Reflection and transmission at an interface <P1,x-1/11,z=<P2,x-1/12,z (z=0).
Here we consider the case of one interface, separating The following four equations for the reflection and
two homogeneous half-spaces, and determine the reflec- transmission coefficients are obtained in matrix form:
tion and transmission coefficients of this interface for
plane harmonic waves. These coefficients are essential
quantities for the treatment of layered media, as will be-
come clear in Section 3.2, but they are also of general (
~1k
l1W 2 -2µ1k 2
=~ /:
-2µ1kl~ 2µ2k 2 -p2W 2 -2µ2kl~
-kl~ )

interest to seismologists because they can sometimes be

2µ1kl1 P1W -2µ1k 2 2µ2kl2 2µ2k 2 -P2W 2
2
used for simple estimates of body-wave amplitudes. We
use Cartesian coordinates and the corresponding dis- (13)
placement components, equations and boundary con-
ditions compiled above. The interface is at z =0.
The two cases sketched in Fig. 1 have to be treated,
that of a downgoing incident wave (case I) and that of
an upgoing incident wave (case II). In case I the in-
cident wave travels in half-space 1, in case II in half- The apparent frequency dependence of these equations
space 2. We consider case I in more detail. can be removed if the wavenumbers are replaced by
If the incident wave is a P wave, the secondary slownesses:
waves produced at the interface are of P and SV type. k sin 9
Then we assume the following expressions for the dis- u =- = - - (horizontal slowness), (14)

j
W IX 1
placement potentials in both half-spaces (j = imaginary
unit): a 1,2-
- ~W- (- a -1,2
2 -u2)1;2
<Pi =ej(wt-kx-liz) (incident P wave), (vertical slownesses). (15)
+ Rdpp ej(wt-kx+l1z) (reflected P wave) b 1,2 =1i.2=(/3-2
W 1,2
-u2)112
,/, =Rd ej(wt-kx+f'tz)
'I' 1 ps (reflected SV wave), (10) Then, Eq. (13) tranforms into the equation
,-/, = Td ej(wt-kx-l2z) (transmitted P wave),
'+'2 pp
,/, = Td ej(wt-kx-l2z) (transmitted SV wave).
'I' 2 ps -u a2
The incident P wave has unit potential amplitude. -2µ1 ub1 2µ2 u2 -p2
Then, the amplitudes of the secondary waves are identi- P1 -2µ1 u2 2µ2 ua2
cal with the reflection and transmission coefficients for (16)
potential amplitudes. These coefficients have a super-
script d, indicating that they correspond to a down-
going incident wave. All terms in Eq. (10) have plane-
wave form and satisfy the wave equations (4). All waves
travel horizontally with the same horizontal wavenum-
 157

Table 1. Plane-wave reflection and transmission coefficients of than the P-SV case, since there is no conversion to P
a plane interface between two half-spaces (P velocities cx)i,z' S or SV waves. at the interface. The displacements are:
velocities /3 1 , 2 , densities Pi,z, horizontal slowness u). R and
Td are coefficients for potential amplitudes of P -SV waves, rd
and td coefficients for displacement amplitudes of SH waves. Uy! =ej(wt-kx-l',z) (incident SH wave),
The incident wave travels in the medium with index 1 (down-
ward propagation, case I of Fig. 1) +rdei<w<-kx+l\z) (reflected SH wave), (18)
(transmitted SH wave).
D~-Df
Df+D~
The reflection and transmission coefficients, which here
2ua 1 2 2 2 are coefficients for displacement amplitudes, follow from
--d--d {(cu -p 1+p 2)(cu + p 2)+c(cu -p 1 )a 2 b 2 }
D1+D2 the continuity conditions for uY and µ uy,z at z = 0 [ see
2plal 2 2
Eq. (8a and b)]. They are also given in Table 1.
-d--d {(cu +p 2)b 1 -(cu -p 1 )b 2 } P -SV reflection and transmission coefficients for
D1+D2
displacement amplitudes follow from the potential coef-
2p1Ua1 . 2 ficients in Table 1 by multiplication with factors which
--d--d {cu -p1+P2+ca2b1}
D1+D2 are quotients of the velocity of the incident wave and
D~ -Df-2p 1 p 2(a 1 b 2 -a 2 b 1) the velocity of the secondary wave.
Df +D~
The coefficients in Table 1 depend only on the ve-
locities and densities of the two half-spaces and on
2ub1 z z z slowness or angle of incidence, according to Eq. (14).
-d--d {(cu -p 1+p 2)(cu +p 2)+c(cu -p 1 )a 2 b 2 }
D1 +Dz They can become complex, if one or more of the verti-
2plbl 2 2
cal slownesses a 1 , 2 and b 1 , 2 are imaginary. In this case
-d--d {(cu + p 2)a 1 -(cu -p 1 )a 2 } some of the secondary waves are inhomogeneous waves
D1 +Dz
which propagate horizontally and whose amplitudes
2p1ub1 . z decay exponentially with increasing vertical distance
-d--d {cu -p1+P2+ca1b2}
D1 +Dz from the interface; this follows from Eqs. (10) or (18).
µ1b1-µzb2 In this case there is a slight frequency-dependence of
µ1b1+µ2b2
the reflection and transmission coefficients according to
the sign of w, as indicated in Table 1. For numerical cal-
2µ1 bl culations it is usually sufficient to consider w~O; then
the reflection and transmission coefficients of an inter-
face are frequency-independent.
D~ =(cu 2 -p 1 + p 2)2 u 2 + (cu 2 -p 1)2 a 2 b 2 + P1 p 2 a 2 b 1 The reflection and transmission coefficients in case
D~ = c 2 u 2 a 1 a 2 b 1 b 2 + (cu 2 + p 2)2 a 1 b 1 + p 1 P2 a 1 b 2 II of Fig. 1, i.e. for incidence at the interface z = 0 from
positive real or below, follow from those in Table 1 by two changes.
a1,2 =(cx;-,1-u2)112
negative imaginary for w > 0 or The first is an interchange of the parameters of both
b1, z =(/3;-,1-u2)112 positive imaginary for w < 0 media, the second a sign change for those P-SV coef-
ficients which imply wave conversion at the interface.
C= 2(µ1 - µ2), µ 1, 2 = p 1, 2 /3i, 2 This sign change is a consequence of the change, with
respect to case I, in propagation directions relative to
the z-axis. The coefficients are summarized in Table 2;
Application of Cramer's rule to this equation is cum- their superscript u indicates that the incident wave is
bersome but rewarding, since relatively compact ex- upgoing.
pressions for the reflection and transmission coefficients The coefficients in Tables 1 and 2 are the essential
result. They are given in Table 1. basis for the calculation of reflection and transmission
The case of an incident SV wave can be treated coefficients of layered media with an arbitrary number
along the same lines, starting with the appropriate po- of interfaces. The recursive calculation of these reflectiv-
tentials instead of Eq. (10). The reflection and trans- ities and transmissivities will be described in the next
mission coefficients follow from the equation section.
u
-u a2 3.2 Reflection and transmission at a layer stack
-2µ1 u b1 2µ2 u2 -pz
We consider a stack of n - l layers between two homo-
P1 -2µ1 u2 2µ2ua2 geneous half-spaces. The upper half-space has the index
0, the half-space at the bottom the index n, the parame-
ters of the medium i are IX;, /3;, P; and the thickness d;,
(17) and the layers i and i + 1 are separated by the interface
z = z;+ 1 . We treat the case of incident waves travelling
downwards in the upper half-space in some detail. The
results for the case of incident waves, travelling up-
and are included in Table 1. wards in the lower half-space, are summarized in Sec-
The case of an incident SH wave is much simpler tion 3.2.3.
158

Table 2. Plane-wave reflection and transmission coefficients as p SV

in Table!, but for an incident wave travelling in the medium
with index 2 (upward propagation, case II of Fig. 1)
Z=Zi+1
~/{ ~}1 Layer i
D~-D~
D~+D~ 0\fa2 \f2
Layeri+1

2 u a2 2 2 2
D"+D" {(cu -p 1 +p 2)(cu -p 1 )+c(cu +p 2 )a 1 b 1 } Fig. 2. Potential amplitudes of downgoing and upgoing waves
1 2 immediately above and below the interface z = z;+ 1
2p2a2 2 2
D"+D" {(cu +P2)b1 -(cu -p1)b 2}
1 2

2pzua2 2
- D" + D" {cu - p 1 + p 2 + ca 1 b 2 }
1 2

D~ -D~ -2p 1 p 2 (a 2 b 1 -a 1 b2 )
D~+D~
2ub
- D" +~" {(cuz -p1 + P2)(cu2 -pi)+ c(cu2 + Pz) a1 b1} MB.= (PPB;
1 2 (21)
I PSB;
2pzbz 2 2
D"+D" {(cu +P2la1 -(cu -p1)a2} The relation between the reflectivity matrices at the top
1 2
and the bottom of layer i is
2p2Ub2 2
D" + D" {cu - p 1 + p 2 + ca 2 b 1 }
1 2 (22)
µ2b2-µ1b1
r" with the phase matrix
µlb! +µ2b2
2µ2 b2
t" (23)

D~ =(c u2 -pt+ p 2)2 u2 + (cu2 + P2)2 a 1 b1 +pt Pz a1 b2 The relation between the reflectivity matrices at the
D~ =c 2 u 2 a 1 a 2 b 1 b 2 +(cu 2 -p 1 ) 2 a 2 b 2 + p 1 p 2 a 2 b 1 bottom of layer i, MBi, and at the top of layer i + 1,
MTi+ 1 , is more complicated; here the reflection and
Other quantities as in Table I
transmission coefficients of the interface z = zi+ 1 come
into play. We abbreviate the potential amplitudes of
downgoing and upgoing waves at the bottom of layer
3.2.1 P-SV case. In the P-SV case the displacement i, i.e. immediately above the interface, by a 1 , b 1 , c 1 , d 1
potentials in layer i are, in analogy to Eq. (10): (Fig. 2); similarly a 2 , b2 , c 2 , d 2 define the amplitudes
immediately below the interface (there should be no
confusion with slownesses and layer thicknesses). The
<pi= e-jkx [A; e-jl;(z-z,) + Bi e+ jl;(z-z,)]} . -
relation between these amplitudes is:
tf; i =e-jkx [ Ci e-jl:(z-z;) + D; e+jl:(z-z;)] 1-0, 1, ... ' n.
(19) a 2 =T:Pa 1 +R;Pb 2 +T,dPc 1 +R~Pd 2

The time factor eiwt has been suppressed. The first c2 = T:5 a 1 +R; 5 b 2 + 'f.~c 1 +R~5 d 2
(24)
terms in Eq. (19) represent all downgoing waves in b 1 =R!Pa 1 +T;Pb 2 +R~Pc 1 +'f.~d 2
layer i, the second terms all upgoing waves. Then we d 1 =R! 5 a 1 + TP",b 2 +R~5 C 1 + 'f.~d 2 .
define reflectivities at the top of layer i by the ampli-
tude ratios of upgoing and downgoing waves: The coefficients Rd,u and Td,u are the plane-wave re-
flection and transmission coefficients of the interface z
B. D. = zi+ 1 ; they follow from Tables 1 and 2 by replacing
PPI;=_!._, PSI;=_!._, the parameters a 1 , /3 1 , p 1 there by IX;, /3;, P; and a 2 , /3 2 ,
A; A;
Pz by IX;+ 1 , /3;+ 1 , Pi+ 1 . The left sides of Eq. (24) are the
B. D. amplitudes of the four waves travelling away from the
SPY;=_!._, SST;= c'·
C; interface; they are linear combinations of the ampli-
' tudes of all four incident waves, the weights being the
The first letter of each reflectivity denotes the type of appropriate reflection and transmission coefficients. We
the incident wave, the second letter denotes the type of combine these coefficients into matrices of interjace re-
the secondary wave and T stands for top. We combine flection and transmission coefficients
the reflectivities in the local reflectivity matrix
Rd,u _
(Rd,u
PP Rd,u)
sp
i+l - Rd,u Rd,u '
MT= (PPI; SPY;). (20) ps ss
' PSI; SST; yd,u)
Td,u _ cd,u
PP sp
yd,u (25)
Similarly, we have for the bottom of layer i:
i+l - yd,u '
ps ss
 159

and replace Eq. (24) by two vector equations: this is achieved by another application of Eq. (22) and
introduces only phase shifts.
(26) The great advantage of the recursive algorithm de-
scribed is that it is unconditionally stable for all
frequencies and slownesses. Other methods sometimes
suffer from overflow problems in the calculation of ex-
(27)
ponential functions for high frequencies and slownesses
(see e.g., Kennett, 1983, Sec. 6.2.2). Here, exponential
The amplitudes b 1 , d 1 , b 2 , d 2 of the upgoing waves can functions appear only in the phase matrix (23), and the
also be expressed with the aid of the reflectivities at the sign of their arguments guarantees that they are always
bottom of layer i and at the top of layer i + 1: bounded.
The transmission properties of the layered medium
b 1 =a 1 PPB;+ c 1 SPB;} (b 1) = MBi (a1) (28)
for incident waves coming from above can be found in
a similar way to the reflection properties. We define
d 1 =a 1 PSB;+c 1 SSB; d1 c1
transmissivities at the top of layer i by comparing the
amplitudes of the downgoing waves there and in the
(29) upper half-space [see Eq. (19)]:
~- A
Equation (26) is inserted into Eq. (29): PPT;=-',
Ao
- A
SPT; =-',
(!:)=MTi+l [T~. (::)+R~+l (!:)l Co

The solution of this equation is We combine them into a local transmissivity matrix

(dzb2) =[1-MTi+tRi+tJu -1 d
MTi+tTi+t
(al)
C1 (30)
-MTi= (PPT;
--
PST;
SP'f;)
-- .
SST;

where I is the 2 x 2 identity matrix. From Eqs. (27) and Similarly we have for the bottom of layer i:
(28) one obtains

~- (PPB SPB.) -
Inserting Eq. (30) here, we find the desired relation MBi = PSP' - ' =E-MT. (33)
!
SSB; I I

between MBi and MTi+ 1 :

with the phase matrix Ei from Eq. (23).
The relation between MBi and MTi+t follows from
a consideration of the interface z = Z;+ 1 , similar to that
Equations (31) and (22) together relate the reflectivity above (Fig. 2). We have:
matrices at the top of layers i + I and i.
Our final aim is to calculate the overall reflectivity (34)
matrix

(32) (35)

at the bottom of the upper half-space, since the ele- On the other hand, from Eqs. (26) and (29):
ments of this matrix are just the desired reflection coef-
ficients or reflectivities of the medium (for potential
amplitudes). Therefore, we apply Eqs. (31) and (22) re-
cursively. We start with i=n-1 and MTn=O, which is
(a2) =Ti+td (al) +Ri+t (b2)
Cz C1 d2
u

the logical condition for any level inside the lower half-
space. Then, the reflectivity matrix at the bottom of
layer n -1 is just the matrix of reflection coefficients of
=T1+1 e:) + R~+l MTi+l e:)-
the interface z = zn for downgoing incident waves: Hence:
MBn-t = R!. By successive applications of Eqs. (22) and
(31) one moves upwards through the medium, until i
= 0 is reached. The reflectivity matrix becomes increas-
ingly more complicated, since it now also represents
e:) = [I - R~+l MTi+1J- 1 T~. e:).
multiples and conversions produced at the interfaces. Inserting Eqs. (34) and (35), we obtain
For practical purposes it may be useful to relate the
reflectivity matrix to a level inside the upper half-space; (36)
160

The recursion, based on Eqs. (36) and (33), starts with i 3.2.3 Wave incidence from below. The layered medium is
= 0 and MB0 = I. The overall transmissivit y matrix is the same as before, but now the incident waves travel
upwards in the lower half-space. In this case the overall
reflectivity matrix RR" expresses, for a level at the top
(37) of the lower half-space, the potential-amplitude ratios
of downgoing reflected and upgoing incident waves.
and has as its elements the transmission coefficients or The overall transmissivity matrix TTu relates the ampli-
transmissivities of the medium (again for potential tudes of upgoing transmitted waves at the bottom of
amplitudes). the upper half-space and the amplitudes of the upgoing
It is, however, not advisable to perform the recur- incident waves at the top of the lower half-space. Simi-
sion in this form in practice because it would require lar definitions hold for the scalar SH reflectivity RR"
that the reflectivity matrices MT;+ 1 be calculated and and transmissivity TT".
stored prior to the calculation of TTd. Rather, one uses The recursion, leading to RR", is as follows:
the form

(38)
i = 0, 1, ... , n - I (44)
where

(39)

(E 0 = I, MTn= 0). If the matrix multiplication in Eq. (45)

(38) is performed from left to right, the calculation of
TTd can directly be combined with the calculation of
the reflectivity matrix RRd. The local reflectivity matrices NT; and NB; correspond
TTd is the transmissivity of the depth range between to the top and the bottom of layer i, respectively. They
the uppermost and the lowermost interface of the me- are different from the matrices MT; and MB; in Section
3.2.1, because they relate downgoing reflected and
dium. If it is desirable also to include parts of the half-
spaces, Eq. (38) can also be used with additional phase upgoing incident waves, whereas MT; and MB; relate
matrices (23) multiplied from right and left. upgoing reflected and downgoing incident waves.
The overall transmissivity matrix Tr is:
3.2.2 SH case. In this section we give only the results
for the scalar reflectivity RRd and the scalar transmis- (46)
sivity TTd, which here express the displacement-ampli-
tude ratio of the secondary (reflected or transmitted) (47)
and the incident SH wave. As in the P-SV case, RRd
refers to the interface z = z 1 between the upper half- The matrix multiplication in Eq. (46) is performed from
space and the first layer, and TTd to the lowermost in- left to right, since the calculation of TT" can then be
terface z = z" . It is clear that the principles of the deri- directly combined with that of RR".
vation of RRd and TTd are the same as in the P-SV The results for SH waves are:
case, but the derivation itself is much simpler; it is ac-
tually a good exercise for the interested reader. One ob-
tains:

(40) i = 0, 1, ... , n - 1 (48)

i = n - I, n - 2, ... , 0 (4 I ) (49)
n- 1
TT"= fl G; (50)
n- 1
t~+ I e- jt;d,
(42) (51)
1- r/+ 1 NB/

(43) 4. Waves from point sources in layered media

4.1 Elementary displacement potentials

In Eq . (41) one uses MT"= 0, and the factor F0 in Eq.
for cylindrical coordinates
(43) is obtained by setting d O = 0. The interface coef-
ficients rd·" and td·" follow from Tables I and 2. Com- Elementary solutions of the wave equations (7) for the
parison of the SH results with those for the P - SV P - SV potentials in layer i of the medium, which will
case, e.g. of Eq. (41) and Eqs. (31) plus (22), shows the be used later for the synthesis of point-source wave
great similarity in basic structure. fields, are the following:
<P· =
'

P= {
{c~s ((J}
c~s
l }
l
sml<p
J (kr) [A. e-jl,(z-z;) + B. e+ jl,(z-z;)J
' ' '

((J J(kr);.-[c.e-jl;<z-z;)+D-e+jl;<z-z,JJ
l
i=O, 1, ... ,n.
161

(52)

' sm l <p , Jk ' '

-----------------------
Ji(k r) is the Bessel function of integer order 1 dependent components F1 , F 2 , F3 . These components
= 0, 1, 2, ... , and the time factor ejw, has been omitted refer to a Cartesian coordinate system with, say, the x-
as in the corresponding plane-wave (Cartesian coor- axis pointing north, the y-axis pointing east and the z-
dinate) expressions, Eq. (19). axis pointing down. The displacement potentials of the
In the SV potential 'P; the term Uk)- 1 has been fac- single force have the following slowness-integral repre-
tored out since the equations for the coefficients A;, B;, sentations for an observer with cylindrical coordinates
C;, D;, which follow from the boundary conditions at (r, <p, z) in an infinite medium whose material properties
interfaces [see Eq. (9a and b)], then agree exactly with are those of layer m:
the equations for the coefficients in Eq. (19) which are
00
based on the continuity of the quantities (8a and b).
This has the important consequence that the plane-
4npm<l>s=81 ssign(z-zs)ulo(uwr)e-jwamlz-zsldu
0
wave reflectivity and transmissivity matrices that have
been derived in Section 3 can be used directly here. For +82
oo
J-.u2-J1(uwr)e-1wamlz-zsldu
.
(58)
instance, the amplitudes of the upgoing wave field at oJam
the uppermost interface z=z 1 , B 0 and D 0 , follow from u .
the amplitudes of the downgoing wave field, A 0 and 4n Pm If;= 81 J-.-b-Jo(U
00
w r) e-JWbmlz-zsl du
C 0 , with the aid of the reflectivity matrix RRd in Eq. o]W m

(32):

(53)
-1 .
x =r, J- - J (uwr)e-JWbmlz-zsldu
00
4np (60)
Similarly, if the amplitudes of the upgoing wave field at m s O /3;,,j bm 1

the lowermost interface z=zn, Bn and Dn, are given, B 0

and D 0 follow from am=(rx;;;2-u2)1;2 (positive real or
bm=(/3;;;2-u2)1/2 negative imaginary), (61)
(54)
81 =F3 , 82 =F1 cos<p+F2 sin<p,
where TT° is the transmissivity matrix in Eq. (46). r, = -F1 sin <p + F2 cos <p. (62)
The elementary SH potential in layer i is
These source-potential representations are very similar
X; = {c~sll ((J}
sm <p
J,(kr)
to the well-known Sommerfeld integral for an explosive
point source. They have been taken from Muller (1969)
and were corrected for a misprint in the second term of
. [E; e-jl;(z-z;) + F; e+ jl;(z-z,JJ, (55) the original form of Eq. (58).
and the scalar reflectivity and transm1ss1v1ty of the
plane-wave case can also be used here:
Layer
index
F0 =RRd E 0 (56) Receivers(z=OI
....r
0
, X X X X X X X X

with RRd from Eqs. (40) and (41), or 2

(57)
m- 1
Zm
with TTu according to Eqs. (50) and (51). m Point source- - - - - -- Zs
m+ l
Up till now, reflectivities and transmissivities have
been described for the complete layered medium. It is
Zi
clear that they can also be defined for partitions of the i <X; (Ji Yi o(;
medium by simply deleting those layers which are not
of interest. The reflectivities and transmissivities in this
Zn
form will be used later on. n
~z
Fig. 3. Layered medium, cons1stmg of n - 1 homogeneous
4.2 Displacement potentials for a single force layers between two homogeneous half-spaces. Layer parame-
ters are: a;= P velocity, /3; = S velocity, P; = density, d; = thick-
We consider now the layered medium with a single- ness, z; = depth to the top. On the left it is indicated to which
force point source at depth z5 in layer m (Fig. 3). The partition of the medium the reflectivities R ±, R ± and the
single force F is harmonic in time and has frequency- transmissivity T+, T+ correspond
162

The integrands in Eqs. (58H6O) have, in principle, Moreover, we write

the form of the elementary potentials (52) and (55) for l
= 0 and I= 1 if the wavenumbers there are replaced by (73)
slownesses,
for the SH source amplitudes, in order to have unified
notation.

with a; and b; defined in analogy to Eq. (61). Hence, it 4.3 Synthesis of the wave field at z = 0
is clear that the interaction of the source wave field
with the layer stacks above and below the source can The first step in the wave-field synthesis is to determine
also be represented by such slowness integrals. All that the complete upgoing wave field in layer m. This field is
has to be done is to extract from Eqs. (58H6O) source the sum of the direct upgoing waves, Eqs. (68), (69) and
amplitudes As, Bs, ... , Fs and then to describe the inter- (70), and all possible reflections and multiple reflections
action by equations similar to Eqs. (53) and (54) or Eqs. at the layer stacks above and below the level z = zm (in
(56) and (57). the sense of z;:;) in layer m. Each of these contributions,
The source amplitudes are different for downgoing and also their sum, can be written in a form similar to
and upgoing waves, and in the P-SV case they are Eq. (68), (69) or (70):
also different for the terms with different types of cp de- ro
pendence. For downgoing waves (z > zs) we obtain: 4n Pm q>u = s(el Bl lo+ ez Bz 11) e+jwam(Z-Zm) du
0
if)
1 .
4 TC Pm</>~ = J(e 1 As Jo+ ezAs211) e- jwam(z-zm) du (63) 4np IJIU=
m
J
00
- ( e D J +t: D J )e+JWbm(Z-Zm)du (74)
O ju W 1 1 0 2 2 1
0

4n Pm If'/ 4n Pm Xu=
00

srJ VJl e+ jwbm(Z-Zm) du.

1 .
= s-.-(81
if)

o]UW
csl lo+ ez Csz 11) e-JWbm(Z-Zm) du (64)
0

The contributions differ in the P-SV amplitude vectors

00

4npmx~= sr,EJ1 e-jwbm(Z-Zm)du (65)

0 V
1.2
= (B1,2)
D (75)
1, 2

(66) and in the scalar SH amplitude V, respectively, depend-

-1 ing on the type of interaction that is considered. In or-
Es=132 "b ep der to determine these amplitudes explicitly, we first
ml m define P-SV reflectivity matrices R- for the part z>zm
of the medium and R + for the part z ;£ zm and, likewise,
(67) scalar SH reflectivities R- and R + (see Fig. 3). R-, R-,
R+, R + are reflectivities of the type specified by Eq.
Similar expressions hold for upgoing waves (z < zs): (32), (40), (45) and (49), respectively. The practical com-
putation of these quantities should be clear from the
00
details that have been given in Section 3.
4np m'+'
,./..U=
s J
(e 1 B sI J O +e 2 B s2 J I )e+jwam(Z-Zm)du (68) In the P-SV case, the amplitude vectors V1 , 2 are:
0

for the direct wave field (/' ),

for the first multiple of the
(69) direct wave field (IV'),
ro for the second multiple (IV\/'),
4npmx~= sr,FJ1 e+jwbm(Z-Zm)du (70)
0
u2 for the reflected wave field ( V ),
Bs1=-ue;1, Bsz=-.-e;l for the first multiple of the
Jam reflected wave field ( \f'-.../' ),
u2
D sl=bep'
-1 D · -I
(71)
sz=-ruep'
m
The sum of all these wave fields has the amplitude vec-
tors

vl,2 =(l+R-R++R-R+R-R++ ... )(St2 +R-st2)

For later use, it is favourable to combine the P-SV =[I-R-R+J- 1 (S~.2+R-st2), (76)
source amplitudes into source amplitude vectors:
Their components [see Eq. (75)], when inserted into
sd1,2 = (Asl.
C 2) , S"1,2 = (Bs1,2).
D (72) Eq. (74), specify the complete upgoing P-SV wave field
sl,2 sl,2 in layer m.
 163

The corresponding SH potential is the expression The near-field displacements are:

for x" in Eq. (74) with
(77)

Note the structural similarity of Eqs. (76) and (77). (85)

The second step in the synthesis of the wave field for
z=0 is to determine, from the potentials (74), the po-
tentials ¢ 0 , P 0 and x0 in the upper half-space of the
layered medium and from these the displacement com- The total displacement vector is the sum of the near-
ponents at z = 0. The P -SV amplitude vectors in the field and far-field vectors. Closer inspection shows that
upper half-space, Vt 2 , follow from those in layer m, at the origin this vector (in Cartesian coordinates) is
V 1 , 2 in Eq. (76), by multiplication from the left with the finite and independent of cp, as it should be.
transmissivity matrix T+ of the layers between z = z;;; The results (83)---(85) reflect the well-known fact that,
and z=z 1 =0- [see Eq. (54) and Fig.3]: in the far field of a point source, P -SV waves are re-
stricted to the radial and the vertical component and
o
V 1,2 (B?,2) = T+V
= Do 1,2·
(78) SH waves to the transverse component. Only in the
1, 2 near field do P-SV waves appear also on the trans-
verse and SH waves also on the radial component.
Then the potentials ¢ 0 and P 0 are (z~0): Moreover, the two terms in Eq. (83) are easily identified
00
as the displacements due to a vertical force (i = 1) and a
4n Pm ¢ 0 = J(8 1 B? 10 +8 2 BP1 ) e+jwaoz du horizontal force (i = 2). A purely vertical force produces
0 no near-field displacements (8 2 = 11 = 0).
(79) The results (83-{85), supplemented by the factor
ejwi, are the displacements due to the harmonic single
force F = (F1 , F2 , F3 ) · ejwt_ As they stand, they can also
be considered as the Fourier transforms of the displace-
The SH potential is, accordingly, ments due to time-dependent force components, with F;
= F;(w) being the Fourier transform of the i-th com-
00
ponent. With this in mind, one obtains time-domain dis-
4npmxo= sr,T+ VJl e+jwbozdu (80) placement seismograms by inverse Fourier transfor-
0
mation of Eqs. (83)--(85).
with V from Eq. (77), and with the scalar transmissivity Equations (83)---(85) represent the complete displace-
y+ defined in the same way as T+. The calculation of ment field of the medium, including body and surface
T+ and y+ is performed with Eqs. (46) and (50), respec- waves, in spite of their derivation in terms of body-
tively; in these formulas the index n is replaced by m. wave notions such as reflection, transmission, upward
The displacement components follow from inserting and downward propagation and decomposition into
Eqs. (79) and (80) into Eq. (6) and (6'), respectively. multiple waves. Surface waves in the sense of normal
The results are given below for far-field and near-field modes are related to poles of the integrands on the
terms separately. They can be written in a compact positive real u-axis which follow from the dispersion
form, if the following definitions are used: equations:

det[J-R-R+J- 1 =0 (Rayleigh waves)

(81) (86)
l -R- R+ =0 (Love waves).

Further theoretical discussion of these equations and

how they are solved for the slowness-frequency relation
of Rayleigh and Love waves can be found in Kennett
(1983). Here, where the interest is concentrated on the
(82) calculation of theoretical seismograms, it is sufficient to
remark that the poles of the integrands are shifted
away from the real u-axis if absorption is introduced
The vector u should not be confused with the displace- via complex wave velocities (see Section 6.2). Hence,
ment vector. The far-field displacements at the level z straightforward numerical integration of Eqs. (83)-(85)
=0 are poses no problem in principle and both body- and sur-
face-wave contributions to theoretical seismograms are
obtained. It is, of course, possible to calculate only
body or surface waves by integrating only over the cor-
(83) responding slownesses.

00
4.4 Exact results for a free surface
4npmu{f = -wry J1 (uwr)uT+ Vdu
0
0 The displacements (83)-(85) were derived for the case
V=(l-R- R+)- 1 (S"+R-sd). (84) that the layered medium has a homogeneous upper half-
164

space with non-zero velocities cx 0 , /3 0 and density p 0 . Luco and Apse! (1983), whose approach starts directly
Realistic seismogram calculations, corresponding to a with the free-surface boundary conditions.
free surface z = 0, can probably be performed by choos-
ing cx 0 and p 0 approximately as the values of air, and
{3 0 ~ cx 0 . It is more elegant, however, to obtain exact 4.5 Results for a moment-tensor point source
free-surface conditions by the limiting process cx 0 --> 0,
/3 0 -->0, p 0 -->0. The parameters cx 0 , /3 0 , Po appear in the A generalized point source is represented by the sym-
matrix U, in the vector u [see Eq. (81)] and in the metric moment tensor
transmissivities T+ and y+. The transmissivities follow
from Eqs. (46) and (50), respectively, with the index n
replaced by the index m of the layer with the source
(see Fig. 3): (91)

T+=TrG, G=G1 G2···Gm-l (87)

y+ =t~ G, G=G 1 G 2 ... Gm_ 1. (88) This tensor represents a superpos1t10n of three single
couples without moment along the x-, y- and z-axes of
Tr is the matrix of P-SV transmission coefficients of the Cartesian coordinate system introduced earlier and
the interface z = z 1 = 0 for wave incidence from below, three double couples in the x y-, x z- and y z-planes.
as defined in Eq. (25), and t~ is the corresponding SH The tensor components may be frequency dependent.
transmission coefficient. These coefficients follow from The moment-tensor point source includes, as special ca-
Table 2 by changing the half-space index 1 to O and the ses, simpler point sources such as an explosion or a
index 2 to 1. Only in Tr and t~ do the parameters cx 0 , double couple of arbitrary orientation; it does not in-
/3 0 , Po appear; G and G are free of them. clude single forces.
We then introduce in the P -SV case the new quan- For instance, an explosion in layer m, having the
tities reduced displacement potential F(t), is represented by
Eq. (91) with vanishing mixed components and

and perform in them and in t~ the limiting process (92)

cx 0 --> 0 (hence a 0 --> oo ), {3 0 --> 0 (hence b 0 --> oo) and
p 0 --> 0. At first sight this does not look trivial in the P where F(w) is the Fourier transform of F(t) (Muller,
-SV case, but actually it is simple since only terms 1973). For a double couple of arbitrary orientation,
which contain the product a 0 b 0 have to be taken into specified by two orthogonal unit vectors f and n and by
account. We obtain: the moment function M(t), the moment tensor com-
ponents are

(93)

where M(w) is the Fourier transform of M(t). The vec-

( 2/3i ua 1 b 1 (1-2~iu 2 )b 1 ) (89) tors f and n are nodal-plane normals and have between
(1 - 2 /3i u 2 ) a 1 -2{3 1 ua 1 b 1
them a quadrant with P-wave motions towards the
hT = first row of H focus.
t~ =2. The wave field of each of the couples, combined in
Eq. (91 ), is found by spatial differentiation of the wave
Then the following substitutions are made in Eqs. (83}- fields of a single force with respect to the source coor-
(85): dinates. Details are omitted here; they can be found in
the text books mentioned in the introduction. In this
(90) way source displacement potentials similar to Eqs. (58)-
(60), representing a moment tensor in its general form
G and G are defined by Eqs. (87) and (88), respectively, (91), can be found. The subsequent treatment, leading
and calculated with the aid of Eqs. (47) and (51). to the displacements at the top of the layered medium,
Equations (83)--{85) together with Eq. (90) are the is exactly the same as in the case of the single force.
final results for the exact free-surface response of a In the following we give first the results for the far-
layered half-space to excitation by a single force. field displacements at z = 0 in a form similar to Eqs. (83)
In their essential structure our P - SV results are and (84):
simpler than the forms presented in Kennett [1983,
Eqs. (7.36) and (7.53)]. These forms contain a slightly
different transmissivity from our results and require the
calculation of a third reflectivity matrix in addition to
2npm (u~;)
uz
=w 2 ± Ji
i= 1
Ki
0
uT+VJu
(94)
R- and R+. The relative simplicity of the form obtained Vi= [I-R-R+J- 1 (sr+ R-st)
here will probably also imply some savings in comput- 2 00

ing time. We mention also that the above special treat- 2np m uff
<p
=w 2 "°' X J1·-uy+ Vdu
~ I l l

ment of the free surface z = 0 leads to results which ap- i= 1 0

pear to be identical in structure with the results of f!;=(l -R- R+)- 1 (s~+R- sf). (95)
 165

In these equations we have: 5. Discussion

K 1=½(M 11 cos 2 cp + M 22 sin 2 cp + M 12 sin2cp), 5.1 Partial responses
K 2 = ½M 3 3 , K 3 = M 13 cos cp + M 23 sin cp, In the following discussion reference is made only to
(96)
Jc 1 =½(M 11 -Mzz)sin2cp-M 12 cos2cp, the far-field P -SV displacements for a single force in
the form
Jc 2 = M 13 sin cp - M 23 cos cp

j1=h=J1,
j 3 =J 2 [J 1 , 2 from Eq. (82)] (97)
(102)
j 1 =11 (u wr), j 2 = -J0 (u wr).
The conclusions apply, of course, also to near-field dis-
The source amplitudes are [ea and ep from Eq. (67)]: placements, SH waves and moment-tensor sources.
Equation (102) yields, in the far-field approximation,

u
(J~ea
·u
am
3 -1
) the complete medium response, i.e. all possible interac-
tions and wave types are included. Various partial re-
S1 = j u2 e;; 1 ' sponses are immediately evident and they may be of
great practical importance. For instance, the original
form of the reflectivity method is obtained by setting Vi
= R-st, i.e. by considering only downgoing waves at
the source which are reflected once at the layers below,
(98) without any further reverberation between the layers
above and below the source. The matrix G
= G 1 G 2 ... Gm-t with Gi from Eq. (47) can be calculat-
ed either exactly, in cases where multiples and conver-
sions produced by the upgoing wave field in the layers
between the source and the free surface are of interest,
or with the approximation Gi = Ei Ti~ 1 which implies
only phase shifts in the layers and transmission losses
at the interfaces. Note that the reduced wave field con-
tains the P-P, SV-P, P-SV and SV-SV reflections
from the layers below the source which may still be
more than actually desired. In this case, one introduces
vanishing elements into R- and possibly also into Ti~t.
(99) Another important partial response is obtained by
choosing Vi=R-(R+sr+st). This is similar to the fore-
going case but, additionally, the reflection of the source
The remaining quantities in Eqs. (94) and (95) are
wave field at the layers above the source is included. In
identical with those in Eqs. (83) and (84). The near-field seismological terms this means, for instance, that be-
displacements are:
sides the mantle P phase the surface reflections pP and
s P at the source are also obtained. R+ is calculated
with the aid of Eqs. (44) and (45), where n is replaced
by m. Often the approximation in Eq. (44), NTi+t =Rf+ 1
+ Ti~ 1 NBi Ti~ 1 or even NTi+ 1 = Rr+ 1 + NBi, is sufficient.
These brief examples of partial responses illustrate
that the formalism presented here has great flexibility
(100) and that it is worthwhile to include a few such options
in a computer program together with the full calcu-
00 J lation. The flexibility of the reflectivity method is its
2np m u"f
z =j WK 4
0
J __l_vTT+v
ur
1
du main advantage in comparison with propagator meth-
0 ods which usually give the complete response of the
K4 = ½(M 22 - M 11 ) cos 2 cp - M 12 sin 2 cp. (101) medium. (An exception is Kind's (1985) treatment, by
propagator methods, of different source and receiver
In these expressions the Bessel functions 10 and 1 1 have structures which implies an incomplete response.)
been written without their argument u w r, the column
vector v has the components a 0 and - u, and all other
5.2 Extended sources
quantities are the same as in Eqs. (94) and (95); the
column vector u is defined in Eq. (81). Extended sources, modelled by a superposition of sev-
Exact results for a free surface follow, as in the case eral point sources, will often require the use of both
of a single force, by the substitutions (90) supplemented far- and near-field displacements. In this case it is ad-
by vTT+=2kT G, where kT is a row vector formed by visable to change from cylindrical to Cartesian coor-
the second row of matrix H in Eq. (89). Moreover, the dinates before the displacement fields of the individual
total displacement field is regular at the origin. point sources are superposed.
166

If the receiver distance from the source region is The SH displacement, corresponding to the P -SV
large compared with the source dimensions, and if all displacements ( 104), is:
point sources are located in the same layer m, simple
displacement formulas can be found which occasionally
may be useful.
4n Pm uff = -(l -j) (~)
"'
112
nr 0
f u 1I2 GVdu
(108)
We start with the generalization of Eq. (102) and V=(l -R- R+)- 1(SS"+R- SSa)
assume that the N point sources, which here are single N

forces, have the spectral representations F;, = (Flk, Fn, s su,d = L Y/k ssr·d (109)
k~ 1
F3k) and act at the hypocentres (xsk' Ysk' zsk) at the times
tk (so far tk = 0). Moreover, we assume the origin of the SStd =j Esr,d, (110)
Cartesian and cylindrical coordinate systems to be di- E from Eq. (107), sr,d from Eq. (73), Y/k from Eq. (62) for
rectly above the source region and the receiver coor- the k-th single force.
dinate r to be so large that the receiver azimuth, with Finally, we give the far-field results for an extend-
respect to each point source, is cp with good accuracy. ed source, consisting of several moment-tensor point
Then we have from Eq. (62) elk=F3k and e2k=Flkcoscp sources. The P - SV displacements are:
+ F2k sin cp, and not only can the vertical displacements
due to the different point sources be added, but also uff ) w3/2 co
the horizontal radial displacements; their sum is the 2npm ( 'ff = ( l - j ) - -11-2 Ju 112 HGVdu
-u 2 (nr) 0
displacement along the azimuth direction of the re- (111)
ceiver with good accuracy.
V = [I-R-R+J- 1(ss 0 + R-ssd)
3 N
In the far field the Bessel functions in the matrices
(82) can be replaced by the asymptotic form of the cor- SS0 'd= L L K;k ssr/ (112)
i~l k~l
responding Hankel functions of the second kind, which
implies that only waves propagating away from the
source are considered. This yields:
(113)
E u,d
SSu,d _J. _
3k - U
53k,
(103)
sr1/ fromEq. (98) with Zs replaced by zsk·
where b 1 = - j and b2 = I, and rk is the horizontal dis- The SH displacement is:
tance from the receiver to the k-th point source; rk in w3!2 oo

the denominator, but not in the exponential function, 2np uff= -(1-1·)--Ju 1I2 GVdu
m <P (n r)l/2 o
can be replaced by r. Finally, also the source amplitude (114)
vectors sr/ [see Eq. (72)] are different for different V = (1 - R - R +) - 1( s s" + R - s sa)
point sources: the source depth zs in Eq. (67) has to be 2 N

replaced by zsk· All other quantities in Eq. (102) are ss"·d= L L A;kssr/ (115)
independent of sources and receiver. i~l k~l

After some algebra, one obtains the following far- s s~·/ = E s~·ka, s s~·/ = j Es~/, (116)
field P-SV displacements for the extended source:
srkd from Eq. (99) with Zs replaced by zsk·
The factors K;k in Eq. (112) and Aikin Eq. (115) fol-
low from Eq. (96) for the k-th moment tensor, and E is
(104) given in Eq. (107).
The main applications of these results is in the
2 N modelling of extended earthquake sources. Such
ssu,d= L L 8ikssr/ (105) sources are represented by a sufficient number of shear
i~ I k~ 1 dislocations or double couples with moment tensors
E su,d according to Eq. (93), and these moment tensors are
ssu,d __ inserted above.
1k - U lk
(106)
E
ssu,d _ j-su,d 5.3 Computational aspects
2k - U 2k
Remarks on the numerical calculation of theoretical
(107) seismograms with the reflectivity method, i.e. on the
calculation of integrals like Eqs. (83) and (84) or Eqs.
ssr/ from Eqs. (72), (66), (67) and (71) with zs replaced (111) and (114), have been made by many authors (e.g.
by zsk· Fuchs and Muller, 1971; Kennett, 1979, 1980, 1983; Tem-
Equation (104) looks essentially like the point- me and Muller, 1982). So the following comments
source result (102) with new source amplitude vectors will be relatively brief.
SS and SSd. These vectors are now frequency and azi-
0 1) The calculation of the reflectivities and trans-
muth dependent, i.e. they reflect the directivity proper- missivities is usually the most time-consuming part in
ties of the composite source. Equation (104) is, of the computation, since it has to be done typically for
course, also useful for far-field calculations for only one several hundred slownesses and several ten to several
point source (N = 1). hundred frequencies. The frequency-dependence is in-
 167

troduced mainly through the phase matrices (23). The Wenzel compared the CDC CYBER 205, a vector com-
interface reflection and transmission coefficients (Ta- puter, and the SIEMENS 7880. Such gains in speed
bles 1 and 2) are frequency independent in the purely open up completely new possibilities in seismogram
elastic case, but they become slightly frequency depen- calculations for highly complicated layering and broad
dent if causal absorption is introduced, since then the frequency bands.
wave velocities are frequency dependent (see Section
6.2). We found it usually sufficient to calculate these
coefficients for the dominant frequency of the problem 6. Supplements
under study, such that they continue to depend only on
slowness. 6.1 Earth-flattening transformation
2) The slowness integration is performed with the Spherical earth models can be treated with the reflec-
trapezoidal rule and restricted to the slownesses of in- tivity method in the form described here after an earth-
terest. Numerical phases with the limiting slownesses flattening transformation, i.e. the spherical earth is re-
are often produced by this and they may occasionally placed by an equivalent or almost equivalent flat earth.
be a serious disturbance. Their amplitudes can be re- Various ray and wave theoretical aspects of this trans-
duced by application of cosine tapers at the ends of the formation have been discussed by Gerver and Marku-
slowness interval. shevich (1966), Biswas and Knopoff (1970), Chapman
3) Fast Fourier transformation is used to go from (1973) and Miiller (1977a); the following is a compi-
the time domain to the frequency domain and back lation of its main features.
and the usual rules for sampling in both domains are The depth and velocity transformation is
applied. The length of the seismograms which deter-
mines the frequency interval may be very long, in par- R R
ticular, when the complete-response integrals are used. z=Rln-, v1 (z) =-V 8 (r), (117)
r r
Then the number of frequencies may be very large. If a
partial-response integral is sufficient for the problem where v,(r) is the P or S velocity in the spherical earth
under study, this number may be reduced. at the radial distance r from the centre, R the earth's
4) In many cases it is favourable to use a reduced radius, z the depth in the flat earth and v1 (z) the trans-
time scale t-r/c -t 0 with suitably chosen values of re- formed velocity. Tracing of seismic rays through both
duction velocity c and minimum reduced time t 0 , since media shows that the spherical earth is mapped on a
then the calculation starts closer to the first arrival and cylindrical portion of the flat earth whose radius is the
the seismogram length is reduced. In the frequency do- epicentral distance n R of the antipode. The earth's cen-
main this means multiplication of the slowness integrals tre is mapped on a point with infinite depth and veloci-
by the factor ejw(r/c+taJ_ ties. Rays, leaving the source in both media under the
5) If, in spite of these possibilities, the seismogram same radiation angle with respect to the vertical, always
length has to be chosen shorter than the duration of form identical angles with the vertical at corresponding
the medium response, the resulting time-domain alias- depths. They also have identical travel times with the
ing, i.e. appearance of late energy early in the seismo- consequence that travel-time curves, e.g. at the surface,
gram, can be reduced or even avoided if one uses com- agree. These properties characterize Eq. (117) as a high-
plex frequencies w- j/, instead of w (Bouchon, 1979). frequency transformation.
This implies, as a consequence of the damping theorem The density transformation is not unique, but this is
of Fourier transformation, that instead of the desired no serious problem since the wave amplitudes are influ-
seismogram u(t) the damped version u(t) e- 11• is calcu- enced much less by the density structure than by the
lated. Depending on ,, this version actually has a shor- velocities except in the case of vertical wave propaga-
ter effective length and therefore is less disturbed by tion. In this case the wave amplitudes depend mainly
time-domain aliasing. Multiplication by e+t/r gives the on the impedance, i.e. on the velocity times density pro-
desired seismogram u(t). , is usually taken between duct, and therefore it is logical to make the impedances
20 % and 50 % of the chosen seismogram length. This in the spherical and the flat earth identical. This yields
method often gives satisfactory results, but sometimes it the density transformation
does just the opposite. This happens when there would
be energy in the seismogram prior to the time at which
(118)
the computation starts; the main source of such energy
are the numerical phases mentioned above. Then blown-
up amplitudes of this energy show up late in the Tests show that Eq. (118) is accurate enough also for
seismogram (after multiplication by e+ 11·) and they may non-vertical wave propagation.
exceed the amplitudes of the physical arrivals. Hence, The transformations ( 117) and (118) are applied prior
the suppression of time-domain aliasing via complex to the calculation of theoretical seismograms. This cal-
frequencies cannot be performed routinely but requires culation is performed under the assumption that the
special consideration in each case. same single force or moment tensor acts at the original
6) The reflectivity method can be programmed in and the transformed source, although the medium
such a way that efficient use can be made of modern properties there are usually different. The seismograms
vector computers (Sandmeier and Wenzel, 1985). Ac- so obtained do not yet correspond to the spherical
cording to these authors the speed of computation can earth. For this, they have to be multiplied, in the case
be increased by a factor of 20 to 30 relative to a of points at the surface, by the factor (see Miiller,
modern general-purpose computer; Sandmeier and 1977a)
168

K= (~)q (~)1;2,
r0 smLI
(119)
where 0 is the cubic dilatation and otherwise familiar
notation has been used.
The viscoelastic modulus in Eq. (120) is separated
where r0 is the radial distance of the source from the into real and imaginary parts, M = M 1 +j M 2 , or into
earth's centre, LI the epicentral distance, q = 1 for a sin- magnitude and phase, M = A ejq'. All these quantities in
gle force and q = 2 for a moment-tensor source. The principle have to be considered as frequency depen-
first factor in Eq. (119) serves as a correction for the dent. The quality factor Q is defined by
differences in velocity and density at the original and
(122)
the transformed source just mentioned. For deep-focus
earthquakes this factor can be 1.2-1.3 and, hence, it Increasing dissipation increases the phase shift cp be-
should not be neglected in calculations of absolute tween stress and strain and hence decreases Q. It can
amplitudes. The second factor in Eq. (119) is well- be shown that, if Q ~ 1, Q- 1 is proportional to the en-
known and takes account of the different ways in which ergy loss per period in a harmonic loading experi-
the wavefront expands in the spherical and the flat ment and therefore has a simple physical meaning. Q
earth. can be measured by different techniques, including
In spite of its high-frequency character, the earth- amplitude measurements of propagating waves, width
flattening transformation, Eqs. (117)--{119), has a broad measurements of spectral lines in spectra of free oscil-
range of applicability. It is useful for practically all lations and, of course, phase-shift measurements be-
body and surface waves travelling through the earth's tween stress and strain in forced oscillations.
mantle and also for most body waves traversing the An important point to note is that Eq. (120), and
core (e.g. Hage, 1983). Difficulties are manifest only for similarly Eq. (121), can be considered as a linear filter
waves propagating very close to the earth's centre equation. The filter, represented by the viscoelastic
where the transformation (117) breaks down and where modulus, must be causal, i.e. the filter output p(t) in the
the velocity structure in the spherical earth, corre- time domain must not start earlier than the filter input
sponding to the homogeneous-layer representation of s(t). This requirement imposes relations between M 1
the flat earth, oscillates strongly and hence deviates and M 2 , or A and cp, which are called dispersion or
from the true structure (Mi.iller, 1977b). In such cases, Kramers-Kronig relations. Those relating magnitude A
methods for synthetic seismograms, which work di- and phase cp [and hence Q, according to Eq. (122)] are
rectly in spherical geometry, are definitely superior (e.g. the most important in the present context:
Rial and Cormier, 1980).
1 + cp(w')
00

6.2 Dissipative media lnA(w)=B--P

n
J -,-dw'
-row -w
(123)
The following discussion of a way in which dissipation
of wave energy can be taken into account in seismo- 1 + In A (w')
00

gram calculations is tailored directly to the needs of

cp(w)=-P
n
J --dw'.
_ w' -w
00
(124)
seismology. The literature on anelastic and rheological
properties of earth materials is very vast. As a starting Here, only Eq. (123) is needed. For simple types of fre-
point for the interested reader, we mention only a book quency dependence of Q, the principal-value integral
by Christensen (1982) on viscoelasticity in general and can be calculated analytically and the constant B can
a review article by Minster (1980) which is geophysi- be determined either at high or at low frequencies. As a
cally oriented. consequence, the viscoelastic modulus is known for all
Dissipation or absorption of wave energy is often frequencies. If this procedure is followed for the rigidity
described by linear laws, i.e. it is assumed that stress µ(w) and the bulk modulus k(w), and if these complex
and strain are linearly related as in purely elastic me- moduli are used instead of the real moduli in the so-
dia. The difference to this case is that now phase shifts lution of an elastic wave-propagation problem, then the
occur between stress and strain. This implies that the frequency-domain solution of the corresponding visco-
elastic moduli are no longer real, but complex and pos- elastic problem is obtained. This is the correspondence
sibly frequency dependent. The simple one-dimensional principle of the linear theory of viscoelasticity. The
stress-strain relation is time-domain solution of the viscoelastic problem fol-
lows as usual by inverse Fourier transformation.
p(w) = M(w) s(w). (120) Working with µ(w) and k(w) is, however, not the
procedure that is normally used. Rather, one works
If £ is a shear strain, p a shear stress, then M is identi-
with wave velocities and hence replaces real velocities
cal with twice the complex rigidity µ. If £ is a volume by complex velocities. For P and S waves, we have the
strain or cubic dilatation, p a pressure (apart from the complex velocities
sign), then M is the complex bulk modulus k. As a
third example, if £ is the strain along a rod or wire, p M (w)] 112
the corresponding uniaxial stress, then M is the com- - [Ma(w)] 1/2 f3c(w)= [ _ p- (125)
a/w)- , '
plex Young's modulus. We will call M(w) the viscoelas- p p
tic modulus without specifying the mode of deformation.
The general three-dimensional viscoelastic stress-strain where p is the (real) density. The viscoelastic modulus
relation of an isotropic substance is for P waves is

(121) (126)
 169

0.98
_ 1_02

-0.1

l=====:=::aa~,-=-c-:.'------
-0.2
1.00
1.0•
g::;, . ,~,y Olwl = (-"-)y
Q(w,) a 200
Q\,..ir) Wr 1.02
0-.,1 , 200
1.00
I
-0.3 / 1.0,
/
/
1.02
Fig. 4. Relative d ispersion for phase velocity
1.00 c and group velocity U of body waves,
OA - - clwl/c lw,l
-0.•
- - c lwl/ c-.,1
corresponding to power laws for the quality
0.98 / factor Q with positive and negative
/
--- - - Ulwl / c lw, I - - - - Ulwl /c-.,1
0.96
exponents. The case y =O represents
10~-~, - ~
10~, - ~
10- , - ~
10~,- -,~o•- -1~0,~ ~10, ~.
\Q --10.....,_,~- I0
' -,'- - 10'-=
0 -wi---·fo-, - 103
frequency-independent Q
W/ Wr W/Wr

with the quality factor Qa following from absorption-band models have been investigated many
times in the literature (for a review, see Minster, 1980),
Qa- 1 = 4/32Q- I+ (1- 4 /32)Q - I. (127)
although often with unnecessary complications such as
3 (J.2 µ 3 (J.2 k cut-off frequencies introduced fo r mathematical con-
venience alone. Here we give the results for the case of
For S waves we have, accordingly: seismological interest, Q ~ 1. If Q is constant, this con-
dition applies for all frequencies, and in the case of the
Mp(w) = µ(w) (128) power law ( 131) we consider only frequencies for which
Qµ=Qµ. (129) Q ~ 1. In the constant- Q case one obtains the well-
known result
Qµ and Qk are the quality factors ofµ and k, and a and
/3 in Eq. (127) are rea l wave velocities taken for a typi- (132)
cal frequency. A familiar assumption is Qk ~ Qw i.e. that
there is much less dissipation in volume deformation
than in shear deformation. Then Qa depends o nly on and in the case of the power law (131)
Qµ, and Qa and Qp have the same frequency depen-
dence. In effect this implies a real, frequency-indepen-
dent bulk modulus, at least in the seismic frequency v0 (w) = v{ 1 +½ [Q(~,)- Q:w) ]
band. An often used relation is Qa = 2.25 Qp, correpond-
ing to a 2 = 3 {3 2 • yn j } ( 133)
·cot 2+ 2Q(w) .
The procedure to find the complex velocities ( 125) is
to make assumptions about Qa and Qp as functions of
In these expressions, v is a real velocity. The real part
frequency, to use Eqs. ( 122) and ( 123) for the determi-
of the complex velocity,
nation of Ma and MP and then to insert these moduli
into Eq. ( 125). In the following we will again disregard c(w) = Re vc(w), c(w,) = v, ( 134)
the distinction of P and S waves and work with M(w),
Q(w) and the complex velocity is the phase velocity of body-wave propagation. This
follows from the plane-wave expression
(130)

If Q is a frequency-independent constant, or if it follows

from the power law
~ exp Vw (t - x j )}

Q(w)=Q(w,) (:J ( 131)

c(w ) (1 + 2Q)
~exp { jw ( t - -X-) } exp { - WX
- -} , (135)
with the reference frequency w, and an exponent y be- c(w) 2c(w) Q
tween - 1 and + 1, the steps leading to the complex
velocity ( 130) are rather straightforward (Muller, 1983) which represents a wave with phase velocity c(w). The
and will not be repeated here. These Q laws and related imaginary part of the complex velocity is responsible
170

y in the case of the power law ( 13 1). The Preliminary

Reference Earth Model of Dziewonski and Anderson
(198 1) contains such a specification. Its reference fre-
quency is w, = 2 n s- 1 , and Qa and Qp are assumed con-
stant such that Eq. ( 132) applies and relates the veloci-
ties for d ifferent frequencies. O'Neill and Hill ( 1979)
have performed seismogram calculations with the re-
flectivity method and Eq. ( 132). They compared the re-
sults with seismograms calculated for the dispers ion-
free, frequency-independent complex velocity

5 6 1 8 ( 136)
VEL OC I TY I N KM / S

Fig. 5. P velocity-depth function of a crustal model for which

theoretical seismograms are shown in F ig. 6 wh ich leads to acausal body-wave arrivals. Therefore,
this simple velocity law is not well suited for body-
wave calculat ions, but it is often sufficient for surface
waves. Incorporation of the three velocity laws ( 132),
for absorption, since it leads to the exponential decay ( 133) and (136) in a computer program for theoretical
of the wave amplit udes in Eq. ( 135) with increasing seismograms offers enough possibilities for the modell-
propagation distance x. ing of absorption effects.
The frequency dependence of c(w) reflects the dis- An alternat ive to the use of complex velocit ies is
persion th at is connected with absorption. Dispersion is sometimes the use of dissipation operators which are
slight, of course, and bot h positive and negative ex- convolved wit h seismograms, computed for purely elas-
ponents y lead to an increase of phase velocity with tic med ia. D issipation operators follow from the plane-
freq uency (Fig. 4). Group velocity can formally be cal- wave delta-function response of a homogeneous ab-
culated from the phase velocity and has been included sorbing medium by appropriate averaging of the quali-
in Fig. 4. We have calculated synthetic seismograms for ty factor of an inhomogeneous medium along seismic
cases with relatively strong absorpt ion and fo und that rays. The use of such operators, normall y correspond-
first-arrival times agree quite well with travel times ing to frequency-independent Q, is very common in
computed from group velocity at the dominant fre- conjunctio n w ith seismogram calculations by general-
quency. Group velocity therefore appears to be a mea n- ized or asymptotic ray theory. Operators for the power
ingful velocity also in the case of weak dispersion. law (131) are given by Muller (1983). The use of dissi-
Synthetic-seismogram calculat io ns for dissipat ive pat ion operators, however, is restricted to body-wave
media with the reflectivity method require t he specifi- investigations with little or no interference of phases
cation of the reference frequency w,, the real layer ve- propagating along different rays. Interfering body
locity c(w ,) for P and S waves, the quality factor Q or waves and surface waves usually have to be treated
Q(w,) of each layer fo r P and S waves and the exponent with the aid of complex velocities.

=- 18 - 0

I I
'
J
fsM<
> 0,
:,- I 7 . 0
16 - 0
,,.. ' >
' ,.. -
15 - 0

' ~" --·--

;,-
- 14, 0
,.. ~ Fig.6. Record section of vertical-
- 13 - 0
component seismograms (e.g. for
"' /

st
- 12 . 0
c-,J - / displacement or particle velocity),
/ 11 . 0
/ calculated for the crustal model o f
/ I 0 -0
/
- F ig. 5 and an explosive point
- ./ ~ 9 -0 source at depth 30 m. Reduction
J ~P s ~s"1 1~ /
~ 8 -0 velocity is 6 km/s and true
✓
,/
✓
:,- 7 , 0 -
ui
amplitudes a re shown. Due to th e
Iyv ✓ =-
=-
6.0
5.0
uJ
:C
simplicity of the crustal mo del,
practically all seismic phases can
y =- 4 . 0 be identified. The seismogram
~ ~
~
.- ~
=-3.0
~ 2.0
sectio n also illustrates two
problems that can be connected

~ pt
f? I.0 with reflectivity calcula tio ns: the
,.. M :,- o. o appearance of numerical phases
Pg (indicated by dashed lines) and the

mr~
~ - I. 0

- I
L> ~ -2 . 0
-3-0
occurrence of time-domain aliasing
(at d istances beyond 140km)
0
0 0 0 0 0
'-: '-: 0 0
0 0 0 0
'-:
0
0
g
0
C>
0 g 0
~
0
0 - ~ ;;' 0
"'
0
"'
0
~
0
""
C,

"'
N ~
"'
DI ST ANCE (KM J
 171

7. Examples of theoretical seismograms

7./ Explosion-generated body waves
2
The first example of theoretical seismograms consists of ,: 3
reflection and refraction seismological records, generat- 0 4
0
ed by an explosive point source in a relatively simple - 5
model of the earth's crust. The P velocities of this mod- -z 6
el are shown in Fig. 5 and the theoretical seismograms
in Fig. 6. Reasonable assumptions about the S veloci-
ties and the densities were made which need not be de-
-
I
Cl.
w
0
8
7

9
tailed here. In the whole model, Q. is 500 and Qp is 10
250. The complex-velocity law ( 132) was applied, corre- 11
sponding to frequency-independent Q and causal ab-
so rption. The source is at depth 30 m and radiates a 2 3 4 5 6
VELOCITY IN KM /S
far-field pulse, e.g. for displacement or particle velocity,
with a dominant frequency of 5 Hz and an effective fre- Fig. 7. P velocity-depth function of a model relevant to seis-
quency band from 0 to 15 Hz; 307 frequencies were mic prospecting for coal. The synthetic seismograms are
shown in Fig. 8
used.
The complete P - SV response of the model was cal-
culated with far-field expressions similar to Eqs. ( I 04) sorge, I 983), produce even larger am plitude ratios. The
or ( 111 ). Therefore, reflections and multiple reflections, observational evidence from explosion-seismological ex-
produced by the earth's surface, are included. Due to periments is rather to the contrary, i.e. steep-angle
the nature o f the explosio n there is no direct radiation Moho reflections appear to be considerably weaker
of S waves. However, because o f the proximity of the than wide-angle reflections. This points to a Moho
explosion to the surface, S waves are effectively ra- transition similar in character to the transition assumed
diated in the form of th e surface reflection p S and the in Fig. 5, with partly gradual and partly step-like veloc-
non-geometrical wave S* (Hron and M ikha ilenko, ity and density increase wit ho ut pronounced la mi-
198 1). nation.
The phase-velocity range in th e calculation was Theoretical P - SV seismogra ms for a com plicated
3.54- 1, I 00 km/s, corresponding to the slowness window model (Fig. 7), taken from seismic prospect ing for coal,
0.0009- 0.2825 s/ km and including all body-wave veloci- are shown in Fig. 8 as a second example. The model
ties of the model; 1,300 equid ista nt slownesses were represents carboniferous rocks conta ining several
used. In spite of cosine taper ing fro m 3.54 to 3.56 km/s groups of coal seams and overla in by a complicated
a nd from 1,000 to 1, 100 km/s, the amplitudes of the overburden. The source is a n explosion a t depth 30 m,
numerical phases, mentioned in Section 5.3, are rather i.e. in the first layer. The frequency range is 0- 300 H z
strong at short distances. The Rayleigh wave tied to the wi th a dominant frequency of 100 H z; 307 frequencies
su rface of the model is suppressed, because its slowness have been used. T he phase-velocity range is I-
falls outside the slowness window. I, I 00 km/s, and 1,300 slownesses are distributed over
The theoretical seismograms in F ig. 6 show the di- this ra nge. Since the lowest S velocity of the model is
rect wave 1:, and the Moho reflectio ns PM P, PM S+ SM P 1.01 km/s, the seismograms include a ll body waves of
a nd SMS as the main phases. Multiple reflections are the model, but the Rayleigh waves, connected with the
very weak. As mentioned above, SM P and SM S actually waveguide formed by the first layer a nd represented by
do not leave th e source as SV waves but are produced the low frequencies at the end of the seismograms in
at the su rface by P-to-SV con versio n and S* exci- Fig. 8, may no t be modelled completely. Q. is 1,000 and
ta tion. The amplitude behaviour of the Moho reflec- Q11 is 500 th roughout the model. Time-d omain aliasing
tions is mainly determined by the refl ectivities of the has been successfull y suppressed by choosing the time r
Moho transition. Time-domain a liasing has not been (see Section 5.3) equal to 0.25 s; this is abo ut 25 % of
suppressed. As a consequence, SMS jumps from (co r- the seismogram length.
rect) late arrival times to (incorrect) ea rly arrival times The seismograms of Fig. 8 are t ru ly complicated re-
at a distance of about 140 km. T he reverse j ump is seen cords with much interference and only few phases
in the fast numerical phase at about 20 km. which can be rela ted to particular layers of th e model.
One purpose of the seismogram calculatio ns for the The band of strong amplitudes running across the re-
crustal mod el o f Fig. 5 and other models was to inves- cord section corresponds to waves in the top layers of
tigate quantitatively the amplitude ratio of steep-angle the overburden. The weaker energy prio r to this band
and wide-angle Moho reflections PM P. Steep-angle re- is due to waves that have travelled deeper th rough the
flections are st rongly influenced by the specia l fo rm of overburden. The weak phases in the time interval 0.4-
the velocity and density transition from the lower crust 0.55 s at sho rt distances are compressional reflections
to the uppermost ma ntle, whereas wide-angle reflec- from the groups of coal seams.
tio ns around the critical point have more similar ampli-
tudes. A first-ord er d iscontinuity is connected with an 7.2 Earthquake-generated swface waves
amplitude rat io of steep- to wide-a ngle reflections
around 0.5. La minated transitions, characterized by ve- The theoretical seismograms in Fig. 9 illustrate the
locity a nd density reversals, which have been suggested possibility of calculating surface waves with the reflec-
several times (e.g. F uchs, 1969; Deichmann and An- tivity method and of obtaining absolute amplitudes, e.g.
172

- o. 9

- 0,8

- 0,7

-
- 0,6

- o. 5
Cf)

-
- o. 4 w..J
:!: Fig. 8. Vertical-component
synthetic seismograms for the
- o. 3 model of Fig. 7. The source is an
explosion at depth 30 m. True
- o. 2 amplitudes are shown, but they are
partly clipped. No travel-time
reduction has been applied. A
- 0.1 weak numerical phase with
(practically) infinite phase velocity
is visible at short distances
C,
0 "'
0 :: "'- ~ "'
N
g ,..,
"' ~ "'
V ~ "' 0
"' ;: "'
"'0 "'0 "' .... 0
00 "'
00
0
"' "'"' -
C
C

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

DISTANCE lKM)

Tred (s)
,....
'
J>
C,
.,
C,
-,
~
C, C
..
'J
...
u
G (',
~,
...
C,
v>
G"' G
,.
'<
"'
.,
G
..
<J

C.
c, -, -:, -, -, -,

"
Rayleigh waves

0
P,,

~;~.l'(IIUtl'{,I ,. . ,.,/
s, ~

f-----~i~/•\,~)'0.,~:ii,~i{1WNW{Wi~W~,rv---- -
Airy
Rg
,~~if-- ---
phase
( fir s t (fundamental model

~ ~:: -';._-=--=--=--=--:::.--=--=--=--=--::--=--=--=--=--=..h.__i-=.
_. -g:::_
h-:::_
e-=,
r;.-m
:::__o___d-:::_
e-;:.)-:: _-:: _-=._-=._-:;._-=._-=._-=._-=._-=;_-=._-=._-=.-_=._-=,~-=--=--=--=--:=.-=--:..-:..-:..-;:.-:..-:..-:..-=:
257 µ/s Love waves
Fig. 9. Surface-wave seism ograms for
v1,:----------- particle velocity; the crusta l model and
the source are descri bed in the text.
1-----~~. ~ ~ - -- -
The Rayleigh-wave seismograms are for

1--------~NJv~- - - -- the vertical com ponent, the Love-wave

seismograms for the horizontal
transverse component. Times are
reduced with 8.1 km/s, and the
theoretical ,P. and Sn arrival times are
0 indicated in the seismograms for the
<O
~ ~~-~--~- -~ - ~ - -~ - ~ - - ~ - -~-~--~-~ distance 1,000 km

for displacement or particle velocity. A simple three- double couple is located at depth 10 km. Its moment
layer crust-mantle model has been assumed: the layer function, whose spectrum M(w) enters the double-cou-
thicknesses are 15/5/oo km, the P velocities ple moment tensor (93), corresponds to the build-up of
5.0/ 7.2/8.1 km/s, the S velocities 2.9/4.2/4.7 km/s, the the moment 10 2 5 dyne cm over a rise time of 10 s. The
densities 2.7/ 3.0/3.3 g/cm 3, the quality factors frequency range is 0-0.4 Hz a nd 400 frequencies were
700/700/ 225 for P waves a nd 300/300/ 100 for S waves. used. The simple complex-velocity law (136) was as-
Thicknesses, velocities and densities were slightly modi- sumed, the phase-velocity range is 2- 15 km/s and the
fied by the earth-flattening transformation. A strike-slip number of slownesses 500.
 173

The synthetic seismograms for particle velocity puter of the computing center, University of Frankfurt. I also
in Fig. 9 at epicentral distances of 1,000, 1,500 and thank Ingrid Hornchen for her efficient typing of the manu-
2,000 km were calculated with the aid of Eqs. (111) and script.
(114), respectively. The receivers have an azimuth of
30° with respect to one of the two vertical nodal planes
References
of the double couple. The main wave groups in Fig. 9
are surface waves, and the Love waves are much stron- Aki, K., Richards, P.G.: Quantitative seismology-theory and
ger than the Rayleigh waves. In both cases, the main methods, vol. I and II. San Francisco: Freeman and Co.
contributions come from the fundamental and the first 1980
higher mode. The Love waves show the typical surface- Alekseev, A.S., Mikhailenko, B.G.: The solution of dynamic
wave picture, i.e. regular dispersion in the first part of problems of elastic wave propagation in inhomogeneous
the seismograms and pronounced (fundamental-mode) media by a combination of partial separation of variables
Airy phases. The fundamental-mode Rayleigh wave and finite-difference methods. J. Geophys. 48, 161 - 172,
contributes the inversely dispersed Rg wave, and the 1980
Apse!, R.J., Luco, J.E.: On the Green's functions for a layered
first higher-mode Rayleigh wave is dominated by an half-space, Part II. Bull. Seismol. Soc. Am. 73, 931-951,
Airy phase with maximum group velocity. 1983
Seismogram calculations like those for Fig. 9 are Ben-Menahem, A., Vered, M.: Extension and interpretation of
useful, among others, for determinations of the moment the Cagniard-Pekeris method for dislocation sources. Bull.
and moment rise time of earthquakes from long-period Seismol. Soc. Am. 63, 1611-1636, 1973
seismograms. The rise time is a rough measure of the Ben-Menahem, A, Singh, S.J.: Seismic waves and sources.
rupture duration, and an estimate of rupture length fol- New York: Springer 1981
lows from it by multiplication with a reasonable value Biswas, N.N., Knopoff, L.: Exact earth-flattening calculation
for the rupture velocity. Modelling of Love waves at for Love waves. Bull. Seismol. Soc. Am. 60, 1123-1137,
1970
distances up to 20° can be very successful, even with Bouchon, M.: Discrete wave number representation of elastic
simple average models of the crust (Brustle and Muller, wave fields in three-space dimensions. J. Geophys. Res. 84,
1983). Rayleigh waves appear to be more strongly influ- 3609-3614, 1979
enced by details of the crust-mantle waveguide. Bouchon, M.: A simple method to calculate Green's functions
for elastic layered media. Bull. Seismol. Soc. Am. 71, 959-
971, 1981
8. Conclusions Brilstle, W., Millier, G.: Moment and duration of shallow
The extended form of the reflectivity method presented earthquakes from Love-wave modelling for regional
distances. Phys. Earth Planet. Inter. 32, 312-324, 1983
here will probably occupy an important place in the
Campillo, M., Bouchon, M., Massinon, B.: Theoretical study
spectrum of synthetic-seismogram methods for verti- of the excitation, spectral characteristics, and geometrical
cally inhomogeneous media. The main advantage in attenuation of regional seismic phases. Bull. Seismol. Soc.
comparison with propagator methods in their usual Am. 74, 79-90, 1984
form (e.g. Kind, 1978; Woodhouse, 1980; Kind and Chapman, C.H.: The earth. flattening transformation in body
Odom, 1983) is the possibility of calculating partial re- wave theory. Geophys. J.R. Astron. Soc. 35, 55-70, 1973
sponses of the medium. The main advantage in compari- Chapman, C.H.: A new method for computing synthetic seis-
son with wavenumber summation methods working in mograms. Geophys. J.R. Astron. Soc. 54, 481-518, 1978
the time domain, e.g. the Alekseev-Mikhailenko meth- Christensen, R.M.: Theory of viscoelasticity - an introduction
od (Alekseev and Mikhailenko, 1980), is that absorp- (Second ed.). New York: Academic Press 1982
Cormier, V.F.: The synthesis of complete seismograms in an
tion is easily modelled by complex wave velocities. earth model specified by radially inhomogeneous layers.
However, the reflectivity method cannot be considered Bull. Seismol. Soc. Am. 70, 691-716, 1980
as being the optimum method in all cases. For instance, Cormier, V.F., Richards, P.G.: Full wave theory applied to a
for models with many layers per wavelength the Alek- discontinuous velocity increase: the inner core boundary.
seev-Mikhailenko method will require less computing J. Geophys. 43, 3-31, 1977
time than the reflectivity method because the essential Deichmann, N., Ansorge, J.: Evidence for lamination in the
step, a finite-difference calculation in the depth-time lower continental crust beneath the Black Forest (South-
domain, is independent of the model complications; western Germany). J. Geophys. 52, 109-118, 1983
Dziewonski, A.M., Anderson, D.L.: Preliminary reference
whereas the corresponding step in the reflectivity meth-
earth model. Phys. Earth Planet. Inter. 25, 297-356, 1981
od requires more computing time for more complicated Fuchs, K.: The reflection of spherical waves from transition
models (Korn and Mi.iller, 1983). Moreover, the Alek- zones with arbitrary depth-dependent elastic moduli and
seev-Mikhailenko method can be formulated in the fre- density. J. Phys. Earth 16, Special Issue, 27-41, 1968
quency domain, such that absorption is easily incorpo- Fuchs, K.: On the properties of deep crustal reflectors. Z.
rated (Korn, 1985). Also, it remains to be seen whether Geophys. 35, 133-149, 1969
the reflectivity method in the form described here re- Fuchs, K., Millier, G.: Computation of synthetic seismograms
quires less or more computing time than propagator with the reflectivity method and comparison with obser-
methods. vations. Geophys. J.R. Astron. Soc. 23, 417-433, 1971
Gerver, M., Markushevich, V.: Determination of a seismic
wave velocity from the travel-time curve. Geophys. J.R.
Acknowledgements. I am grateful to Wolfgang Brilstle, Mi- Astron. Soc. 11, 165-173, 1966
chael Korn, Ji:irg Schlittenhardt and Paul Temme for many Ha, J.: Recurrence relations for computing complete P and
discussions on theoretical and practical aspects of the reflec- SV seismograms. Geophys. J.R. Astron. Soc. 79, 863-873,
tivity method. The seismograms shown in this paper were cal- 1984
culated by Wolfgang Brilstle and Paul Temme on the com- Hage, H.: Velocity constraints for the inner core inferred
174

from long-period PKP amplitudes. Phys. Earth Planet. In- sources of arbitrary orientation. Z. Geophys. 35, 347-371,
ter. 31, 171-185, 1983 1969
Harvey, D.J.: Seismogram synthesis using normal mode Millier, G.: Seismic moment and long-period radiation of
superposition: the locked mode approximation. Geophys. underground nuclear explosions. Bull. Seismal. Soc. Am.
J.R. Astron. Soc. 66, 37-69, 1981 63, 847-857, 1973
Helmberger, D.V.: The crust-mantle transition in the Bering Millier, G.: Earth-flattening approximation for body waves
Sea. Bull. Seismal. Soc. Am. 58, 179-214, 1968 derived from geometric ray theory-improvements, correc-
Hron, F., Mikhailenko, B.G.: Numerical modeling of non- tions and range of applicability. J. Geophys. 42, 429-436,
geometrical effects by the Alekseev-Mikhailenko method. 1977a
Bull. Seismal. Soc. Am. 71, 1011-1029, 1981 Millier, G.: Correction. J. Geophys. Res. 82, 2541-2542,
Ingate, S.F., Bock, G., Kind, R.: Synthesis of complete SH 1977b
seismograms. Geophys. J.R. Astron. Soc. 75, 261-274, Millier, G.: Rheological properties and velocity dispersion of
1983 a medium with power-law dependence of Q on frequency.
Kennett, B.L.N.: Reflections, rays and reverberations. Bull. J. Geophys. 54, 20-29, 1983
Seismal. Soc. Am. 65, 1643-1651, 1974 Olson, A.H., Orcutt, J.A., Frazier, G.A.: The discrete wave-
Kennett, B.L.N.: Theoretical reflection seismograms for an number/finite element method for synthetic seismograms.
elastic medium. Geophys. Prospecting 27, 301-321, 1979 Geophys. J.R. Astron. Soc. 77, 421-460, 1984
Kennett, B.L.N.: Seismic waves in a stratified medium II. O'Neil, M.E., Hill, D.P.: Causal absorption: its effect on syn-
theoretical seismograms. Geophys. J.R. Astron. Soc. 61, 1- thetic seismograms computed by the reflectivity method.
10, 1980 Bull. Seismal. Soc. Am. 69, 17-26, 1979
Kennett, B.L.N.: Seismic wave propagation in stratified me- Panza, G.F.: Synthetic seismograms: the Rayleigh waves mo-
dia. Cambridge: Cambridge University Press 1983 dal summation. J. Geophys. 58, 125-145, 1985
Kennett, B.L.N., Kerry, N.J.: Seismic waves in a stratified Pilant, W.L.: Elastic waves in the earth. Amsterdam: Elsevier
half-space. Geophys. J.R. Astron. Soc. 57, 557-583, 1979 1979
Kind, R.: The reflectivity method for a buried source. J. Geo- Rial, J.A., Cormier, V.F.: Seismic waves at the epicenter's
phys. 44, 603-612, 1978 antipode. J. Geophys. Res. 85, 2661-2668, 1980
Kind, R.: The reflectivity method for different source and re- Sandmeier, K.-J., Wenzel, F.: Synthetic seismograms for a
ceiver structures and comparison with GRF data. J. Geo- complex crustal model. Geophys. Res. Lett. 12, 1985 (in
phys. 58, 146-152, 1985 press)
Kind, R., Odom, R.I.: Improvements to layer matrix meth- Spudich, P., Ascher, U.: Calculation of complete theoretical
ods. J. Geophys. 53, 127-130, 1983 seismograms in vertically varying media using collocation
Korn, M.: Kombination von Integraltransformationen und methods. Geophys. J.R. Astron. Soc. 75, 101-124, 1983
Finiten-Differenzen-Methoden zur Berechnung der Wel- Temme, P., Millier, G.: Numerical simulation of vertical seis-
lenausbreitung in geschichteten Medien. Ph.D. thesis, Uni- mic profiling. J. Geophys. 50, 177-188, 1982
versity of Frankfurt, 1985 Wang, C.Y., Herrmann, R.B.: A numerical study of P-, SV-
Korn, M., Mi.iller, G.: Comparison of the Alekseev-Mik- and SH-wave generation in a plane layered medium. Bull.
hailenko method and the reflectivity method. Geophys. Seismol. Soc. Am. 70, 1015-1036, 1980
J.R. Astron. Soc. 72, 541-556, 1983 Woodhouse, J.H.: Efficient and stable methods for performing
Luco, J.E., Apse!, R.J.: On the Green's functions for a layered seismic calculations in stratified media. Proceedings of
half-space, Part I. Bull. Seismal. Soc. Am. 73, 909-929, Enrico Fermi International School of Physics, A. Dzie-
1983 wonski and E. Boschi, eds.: pp. 127-151. Amsterdam:
Minster, J.B.: Anelasticity and attenuation. Proceedings of North Holland 1980
Enrico Fermi International School of Physics, A. Dzie-
wonski and E. Boschi eds.: pp. 152-212. Amsterdam: No-
rth Holland 1980
Millier, G.: Theoretical seismograms for some types of point Received December 20, 1984; revised version June 12,
sources in layered media. Part III: Single force and dipole 1985/ Accepted July 3, 1985