P1: GQT/GUB P2: GLM Final Pages Qu: 00, 00, 00, 00

Encyclopedia of Physical Science and Technology EN014c-680 July 28, 2001 17:23

Seismology, Theoretical
Raul Madariaga
Ecole Normale Superieure

I. Introduction
II. Seismic Wave Propagation
III. Reflection and Refraction of Seismic Waves
IV. Seismic Waves in Heterogenous Media
V. Seismic Surface Waves
VI. Seismic Source Theory

GLOSSARY THEORETICAL SEISMOLOGY studies the propaga-

tion and generation of seismic waves in the earth. Because
Reflection and refraction Process by which seismic of heterogeneity of seismic wave velocities in the interior
waves are reflected and refracted by material discon- of our planet, seismic wave propagation is very complex:
tinuities according to Snells law; coupling between P seismic waves are refracted, reflected, and converted by
and S waves occurs at the interfaces. numerous internal discontinuities, and they are scattered
Seismic body waves Linear elastic waves propagating in and attenuated by inhomogeneities. They are also trapped
the interior of the earth are of two types: compressional as surface waves in the shallow, low-velocity layers of the
or P and shear or S waves, which are slower by a factor earth. The kinematic and dynamic properties of seismic
of about 1.73 with respect to P waves. waves are used by seismologists to study the interior of the
Seismic moment Torque or moment of one of the two earth and to determine the rupture process that generates
force couples in the usual model of an earthquake earthquakes.
source.
Seismic rays Trajectories described by high-frequency
P and S waves in heterogeneous media, where they I. INTRODUCTION
propagate according to the laws of geometric optics.
Seismic source Very fast slip on one or several discrete A. Brief Historical Review
segments of a fault surface causing earthquakes. Theoretical seismology originally developed as an appli-
Seismic surface waves Dispersive seismic waves trapped cation of classical elastodynamics to the propagation of
in the outer layers of the earth that propagate along the seismic waves in the earth. After seismologists recognized
earths surface. the main waves typesP, S, and surface wavesthey used

575
P1: GQT/GUB P2: GLM Final Pages
Encyclopedia of Physical Science and Technology EN014c-680 July 28, 2001 17:23

576 Seismology, Theoretical

their arrival times in seismograms to determine seismic considered to be linear and elastic. Seismic disturbances
wave velocities in the earth, and derived the first models are small incremental changes of elastic strains that in-
of the distribution of elastic constant in the earth. The teract very weakly with the large strains due to tectonic
main theoretical method was classical ray theory for a deformation of the earth. Thus for almost all practical pur-
radially stratified model of the earth. These models, de- poses, except the study of strong motion in the near field
veloped since the 1930s, permitted the accurate localiza- of an active fault, we can consider the earth to be a linear
tion of seismic events and the detailed study of seismicity. elastic but inhomogeneous body. Furthermore, it is fre-
Major improvements in the instrumentation between 1950 quently assumed that the earth is isotropic, although there
and 1965, extending the range of seismic observations to is clear evidence of small amounts of anisotropy in both
several decades in frequency, gave a major impulse to the- global and exploration seismology. Elastic waves are at-
oretical seismology. Matrix methods were developed for tenuated as they propagate mainly by scattering by small
the study of propagation in a stack of layers of constant but widely distributed heterogeneities in earths structure.
velocity. These techniques, still in use today, allowed the Attenuation is also consider as a secondary effect that can
calculation of realistic seismograms, including most of the be dealt with as a perturbation of wave propagation in an
features observed in actual records. A significant devel- elastic isotropic medium.
opment of this period was the use of the dispersion of sur-
face waves to invert for the velocity of shallow waveguide
of the earth. By 1960, the first observation of the earths II. SEISMIC WAVE PROPAGATION
normal modes had been made and a whole new area of
theoretical seismology was developed in order to analyze A. Elastodynamic Wave Equation
these new observations. At the same time, new techniques
were developed for the interpretation of reflection pro- We consider the deformation of this volume as a function
files in exploration seismology, where the large amount of time t with respect to a reference configurationfor
of data obtained require extensive computer processing. instance, the position r of the particles of this body at time
The widespread availability of computers has changed the t = 0. Let P be a point of coordinate r in the reference con-
emphasis in theoretical seismology from the study of sim- figuration, and P  its instantaneous position at time t. We
plified kinematical models of wave propagation to the define the displacement vector u(r, t) as the vector PP  .
simulation of complete seismograms that are then com- Separating the symmetric from the antisymmetric parts of
pared with observed records to invert for earth structure. the gradient of displacement tensor (u), we define the
Many different types of tomographic methods have been linear strain tensor,
 
developed for the study of the velocity structure of the 1 u i u j
earth. i j = + , (1)
2 x j xi
Frequently considered as a separate branch of theo-
retical seismology, source theory was developed in or- and the rotation tensor,
 
der to understand the faulting process that is at the origin 1 u i u j
of most earthquakes. One of the most significant results i j = , (2)
2 x j xi
was the observational and theoretical demonstration that
most earthquakes generate seismic waves with quadrupole where commas indicate derivatives with respect to the
symmetry. Later, in the period 19601990, major develop- space variables.
ments in the formulation of dislocation and crack models Deformation of an elastic body generates internal
of the earthquake source took place. These results are at stresses that resist strain. Since we assume that elastic
the basis of most current attempts to quantify earthquake waves produce only incremental changes in strain and
source parameters. stress, we can relate strain to stresses by linear elastic-
ity, so that the stress tensor associated with strain  is

B. Assumptions of Linearity and Elasticity i j = i j + i j , (3)

Although at very long time scales of the order of thousands where summation over repeated indices is assumed, =
of years to millions of years the earth deforms continu- u is the the dilatation or trace of the starin tensor (1).
ously, at the time scales of interest in seismology (from The and are the two Lames coefficients of linear
milliseconds to a few hours) the earth behaves like an isotropic elasticity. These two parameters define the main
elastic body. While in the immediate vicinity of a seismic elastic properties of the earth, and they vary both radially
source strains can be large, only a few kilometers away and laterally inside the earth. The determination of these
they drop to less than 105 so that deformation may be parameters is the main purpose of seismology.
P1: GQT/GUB P2: GLM Final Pages
Encyclopedia of Physical Science and Technology EN014c-680 July 28, 2001 17:23

Seismology, Theoretical 577

From the conservation of linear and angular momentum u = ( + 2) u, (7)

inside a closed elastic body, we derive the equation of
where double dots indicate double partial derivative with
elastodynamics,
respect to time t. This is a system of three second-order
2u partial differential equations with constant coefficients.
= div + f, (4)
t 2 There are two types of solutions of this equation that may
where is the density of mass of the elastic medium; be easily found using seismic potential defined by Lames
vector f(r, t) is a generalized force that represents sources theorem:
of elastic motion. In principle, we should also include in u = + , (8)
(4) the effect of gravity, but in most seismic applications
this can be neglected. Inserting the constitutive relation where is the scalar and the vector potential. Adding
(8) into the equation of motion (4), we obtain a system a gauge condition = 0, it can be shown that the any
of equations that has to be solved in order to compute vector displacement field u can be decomposed into a
elastic wave motions inside the earth. For a number of unique set of potential functions. Introducing the defi-
theoretical considerations, it is more appropriate to write nition (8) into the equation of motion (7), we find two
the equations of motion in the more symmetric form, independent equations for the potentials.
The first is the equation of dilatational, or P waves
vi = i j, j + f i , (5)
1
i j = vi,i + (vi, j + v j,i ) + m i j , (6) = 2 , (9)
2
where dots denote derivatives with respect to time while the second that of shear or S waves
the commas, partial derivation with respect to the space 1
component indicated after the comma, i.e., u i, j = u i / x j . = 2 , (10)
2
The vi (r, t) = u i (r, t) is the particle velocity at position r
and time t. The last term in the second equation includes where = ( + 2)/ is the speed of P waves and
a new source term, the so-called seismic moment distri- = / is that of S waves. In an elastic medium, the
bution. This term represent internal seismic sources such elastic constants and are always positive so that the
as explosions and faults. We will see later that this is the P-wave speed is always faster than that of S waves. This
main source of seismic energy in the earth. is at the origin of the notation P and S, because the first
Equations (5) are at the basis of all modern methods of seismologists called the first arrivals Primary (P) waves
simulation of seismic wave propagation inside the earth and later waves were called Secondary or S waves.
by finite differences. For any given distribution of den-
sity and elastic constants , inside the earth, we can in C. Plane Waves
principle solve numerically for velocities and stresses eav-
erywhere. Numerically, methods are currently being ex- The simplest possible solution of the wave equations
tensively used in the numerical simulation of elastic wave (9, 10) are plane waves; they have many general prop-
propagation in oil exploration and earthquake engineer- erties of seismic waves of more complex geometry. Plane
ing, and the first numerical solutions of wave propagation waves are difficult to generate in seismological practice;
in the entire earth are making their appearance in recent however, they provide an excellent approximation to local
publications. With increased computer power, especially seismic wave propagation in smoothly varying media, or
parallel machines, we envisage that in a few years most at far distances from the source. Let us define a Cartesian
seismological applications will use numerical simulation coordinate system in which the z axis points downward
of seismograms. into the earth, and x and y axes span the surface of the
For practical applications, and in order to understand earth. We consider a simple monochromatic (single fre-
the basic physics of elastic waves, we have to study simple quency) P wave propagating into the earth. Its potential
examples that can guide our intuition about wave propaga- can be written
tion in the earth. Most techniques for the study of earths (r, ) = () exp[i(t p r)], (11)
structure are derived from these simpler solutions.
where () is the complex-valued amplitude of the wave,
its circular frequency, and p the slowness vector. Equa-
B. Seismic Waves in a Homogeneous Medium
tion (11) is called a plane wave because the phase, defined
In homogeneous media the elastic constants and and as the argument of the exponential, is constant on planes
density are constant everywhere. In this case, the equa- p r t = const. When t, the travel time, increases, the
tion of motion simplifies significantly and we get constant phase plane moves in the direction defined by the
P1: GQT/GUB P2: GLM Final Pages
Encyclopedia of Physical Science and Technology EN014c-680 July 28, 2001 17:23

578 Seismology, Theoretical

vector p. The speed of wave propagation along the direc- 1

(R, ) = S() exp[i(t R/)], (14)
tion p is then c = 1/|p|. Inserting the plane wave solution R
(11) into the wave equation for the scalar potential (9) we where R = |r| is the distance from the source and S() is
find, carrying out the derivatives, that c = the P-wave the complex amplitudes or source spectrum. In the time
speed. From the definition of seismic potentials (8), one domain we write
finds that for P waves the displacement u P = ip so that  
1 R
plane P waves produce displacements that are longitudinal (R, t) = S t . (15)
(i.e., parallel to the direction of propagation p). R
Similarly, for S waves a plane wave of vector potential The most important new feature of spherical waves com-
can be written as pared to plane waves is the decrease in amplitude propor-
tional to 1/R as the wave moves away from the source.
(r, ) = () exp[i(t p r)]. (12)
This decrease in amplitude is usually called geometrical
In this case, inserting in (10), we find that the wave spreading by seismologists and dynamic correction in ap-
front moves at speed c = 1/|p| = , and displacement plied geophysics. With this correction, the time evolu-
u S = ip is perpendicular to the direction of prop- tion of a spherical wave is entirely contained in the phase
agation defined by the slowness vector p. Thus, plane S t = R/. Constant phase surfaces are spheres of radius
waves are transversely polarized. R = t. These concentric spheres are usually called wave
From the definition of plane P and S waves [(11) and fronts and are labeled by the travel time t.
(12)], we get the well-known relations between the period Cylindrical waves are more complex than spherical ones
of the seismic wave defined by T = 2/ and the wave- because in two dimensions waves develop a coda, a slowly
length defined by = 2/(|p|). Period and wavelength decreasing amplitude that trails the main wavefront. For a
are thus related by the classical relationship = cT , where point source, the simplest symmetric cylindrical wave is
c is the appropriate wave speed for P or S waves. i
So far we have studied single-frequency plane waves. (r, ) = S()H0(2) (k r ) exp(it), (16)
4
In practice, seismic waves are impulsive, not monochro-
where r is the radial distance from the line source, S()
matic. We can easily generate transient plane seismic sig-
is the source spectrum, H0(2) is a Hankel function of order
nals by the time Fourier transform:
2, and the cylindrical wavenumber k = /. In this case,

1 the time domain has a very simple form:
(r, t) = () exp[i(t p r)] d, (13)
2 1 H (t r/)
(r, t) = S(t)  . (17)
since seismic waves are always real signals, the fre- 4 t 2 r 2 / 2
quency domain amplitudes  must satisfy the condition The asterisk denotes time convolution. Cylindrical waves
() =  (), where the asterisk indicates complex are clearly more complex than its plane or spherical coun-
conjugation. In general, () contains information about terparts. This is due to the long tail in the amplitude of the
the spectral properties of the source like seismic moment, cylindrical wave that decreases very slowly, like an inverse
duration, corner frequencies, etc. square root, behind the wave front at t = r/. The tail is
The properties of plane waves that we have just dis- due to diffraction of the waves around the line source. A
cussed apply also to more general geometries under the particular type of S wave exists also in two dimensions:
condition that material properties change slowly. In that these are the so-called antiplane or SH waves. These are
case, we can use the ray theory and approximate wave waves that propagate in two dimensions but are polarized
propagation by high-frequency asymptotic as discussed in the direction perpendicular to the plane of propagation.
in a later section. These waves played a significant role in the development
of theoretical seismology because they are much simpler
to study than in-plane S waves (SV) that are always cou-
D. Spherical and Cylindrical Waves
pled to P waves.
Two simple and useful solutions of the wave equation for
P waves are spherical and cylindrical waves, generated by
E. Greens Function
a point source and a line source, respectively. Unfortu-
nately, there are no simple spherically symmetric S waves Plane, cylindrical, and spherical waves are frequently used
because vectors waves have lower order symmetries that in theoretical applications because they may be conve-
will be examined in next section for a point force. A simple niently superposed to find solutions to more complex prob-
spherical wave is lems like reflection and refraction of seismic waves. In any
P1: GQT/GUB P2: GLM Final Pages
Encyclopedia of Physical Science and Technology EN014c-680 July 28, 2001 17:23

Seismology, Theoretical 579

applications, however, we need to determine the radiation

from a certain distribution of body forces f(r, t). These
body forces appear as sources in the elastic wave Eq. (4).
The elastic waves radiated from an arbitrary distribution of
sources can be computed exactly with help of Green func-
tions, which are the fundamental solutions of the elastic
wave equation with a point source at the origin:
f(r, t) = e f (t)(r), (18)
where denotes the Dirac distribution, a generalized func-
tion that is zero everywhere except at the origin, and e f is
a unit vector of components ei that points in the direction
FIGURE 1 Radiation patterns for P and S waves for a single
of the applied force f. The determination of the solutions
vertical force.
of the elastodynamic wave Eq. (4) is the most fundamen-
tal problem of theoretical seismology. Its exact solution
was found by Love at the beginning of the century but it be discussed later in this chapter. The near field terms in
was well known in mechanics in the late nineteenth cen- (19) decay faster than those of the fast field, like R 3 ac-
tury. Using standard spherical coordinates (R , , ), we cording to this equation. Their effect is important only
can write the Greens function in a homogeneous elastic in the immediate vicinity of the source where they can
medium as dominate strong motion. The static field due a point force
  may be computed from the near field terms. These terms,
1 1 R
Gi j (r, t | 0, 0) = R P
(, ) t which appear in all elastodynamic solutions, are due to
4 2 R
  diffraction and coupling of P and S waves around the
1 1 R source.
+ R S
(, ) t (19)
4 2 R The Greens function may be used to determine the ra-
   diation from a generalized distribution of body forces by
1 1 R
+ R (, ) 3 H t
NF means of the representation theorem:
4 R  t
 
R u i (r, t) = G i j (r, t | r0 , t0 ) f j (r0 , t0 ) d V0 , (22)
H t , (20) 0 V0

where V0 denotes a volume (or support) of the region
where the three terms represent, respectively, P waves, S where body forces exist. Thus, knowing the Greens func-
waves, and the near-field or low-frequency terms. The far- tion, we can compute the radiation from very complex
field terms decay slowly like 1/R and have a phase that distributions of body forces like those that are equivalent
is identical to to that of the simple spherical waves we to seismic faulting in the earth.
studied in the last section. The main difference is in the
radiation patterns defined by
R P (, ) = cos , R S (, ) = sin , (21) III. REFLECTION AND REFRACTION
OF SEISMIC WAVES
where is the polar angle between the direction of the
point force and the direction of radiation from the source, Many of the most complex wave phenomena in seismol-
i.e., cos = e f r/R (see Fig. 1). The amplitude of the ogy are due to the interaction of seismic waves with the
seismic waves varies with angle of radiation in a typi- surface of the earth or internal boundaries. When a seis-
cal bipolar shape. For P waves, radiation is zero in the mic wave encounters a discontinuity in seismic speed, it
direction perpendicular to the point force, this defines a is refracted and partially or totally reflected by the in-
nodal plane along the direction = /2. For S waves, on terface. The sharpest discontinuities in the earth, like the
the other hand, the polar direction = 0 is a nodal line. Moho or the coremantle boundary, produce large ampli-
As illustrated in Fig. 1, the radiation is cylindrically sym- tude reflections and modify substantially the trajectories of
metric about the direction of the point force. This is a seismic body waves. The less important reflectors, inside
very important property, since observing the source from sedimentary basins, producer weaker reflections and mi-
different directions, one can determine the orientation of nor deflections of the seismic rays. These weak reflected
the point force. Seismic sources are generally more com- waves are used as a tool for the exploration of oil bear-
plex that the simple point force considered here, as will ing strata. Reflected and refracted waves have been and
P1: GQT/GUB P2: GLM Final Pages
Encyclopedia of Physical Science and Technology EN014c-680 July 28, 2001 17:23

580 Seismology, Theoretical



still are the main source of information about the internal R (r, ) = R exp i t px x pzP z ,
structure of the earth at all scales, from global to local 

structures. R (r, ) =  R exp i t px x pzS z . (25)

where pzS = 2 px2 . Inserting these potentials into the
definition of displacement (8) and stresses (3), we can
A. Seismic Reflections from the Surface
compute the stresses and, finally, impose the boundary
of the Earth
conditions (24). After solving for R and  R in terms of
The simplest problem that we can study is that of the the incident potential  I , we get the reflection coefficients:
reflection of seismic waves with the free surface of the
2
earth. Let a plane harmonic P wave, 4 px2 pzP pzS 2 2 px2
R PP =
2 ,

 4 px2 pzP pzS + 2 2 px2
I (r, ) =  I exp i t px x + pzP z , (23)

be incident on the free surface of an elastic medium. For 4 px pzP 2 2 px2

R PS =
2 ,
simplicity, we choose coordinate axes so that the slow- 4 px2 pzP pzS + 2 2 px2 (26)
ness vector p is located on the plane (x z). The x is the
coordinate along the surface and z the vertical coordinate where the denominator of these two expressions is
measured from the free surface. The slowness vector of the so-called Rayleigh determinant R( px ) = 4 px2 pzP pzS +
the incident wave is ( 2 px2 )2 . This function will play an important role in

the following.
p Ip = px , pzP , As indicated in Fig. 2, the kinematics of reflection from
p
where px and pz are the components of p IP along the x the free surface is better understood in terms of the angle
and z axes, respectively. As shown earlier in the section of incidence. Denoting by i s the angle of reflection of the
on plane waves, the slowness vector of plane P waves S wave from the free surface, we get
satisfy the relation p Ip  = 1/, so that we can express sin i p sin i s
px = = , (27)
the vertical
 component of slowness in the simple form
pzP = 2 px2 . Introducing the angle of incidence i of
so that, since < , the angle of reflection of the S wave,
the P wave, we can write px = sin i p / and pz = cos i p /.
i s is always less than that of the incident or reflected P
Thus the angle of incidenceand in fact the complete ge-
waves, i p .
ometry of the plane waveis determined by the horizontal
Similar results are obtained for an S wave incident on
slowness px .
the free surface (see Fig. 2b).
The reflected P wave and S wave are determined from

2
the condition that the surface be free of stress, i.e., 4 px2 pzP pzS 2 2 px2
R SS =
2 ,
x z = 0, zz = 0 (24) 4 p 2 pzP pzS + 2 2 px2

There are two conditions so that we need two reflected 4 px pzS 2 2 px2
waves, P and S waves reflected from the free surface. R SP =
2 . (28)
4 px2 pzP pzS + 2 2 px2
From simple considerations about direction of propaga-
tion (toward z positive), and wave speed we write these The Snell law (27) is the same, but there is an interesting
two waves as new situation in this case. If the angle of incidence of the

FIGURE 2 Geometry of reflection and refraction from a free surface for incident P (left) and S (right) waves.
P1: GQT/GUB P2: GLM Final Pages
Encyclopedia of Physical Science and Technology EN014c-680 July 28, 2001 17:23

Seismology, Theoretical 581

S wave is larger than a critical value sin i s > /, then the become increasingly complicated, but the basic kinematic
reflected P wave becomes an inhomogeneous wave. For properties of these waves derive from the SnellDescartes
this range of angles the vertical component of slowness principle pc = sin i c /c , where pc is the horizontal com-
becomes imaginary and the reflected P wave can be ponent of the slowness of the wave of type c (P or S)
written in layer , and i c is the corresponding angle of incidence

 
 or transmission with respect to the vertical axis. Here c
R (r, ) =  R exp i t px x exp pzP z , (29) is the appropriate wave velocity in layer . Depending on
p  the relative velocities of the layers and the type of incident
where pz = px2 2 . This is a wave that propagates
along the surface with an exponentially decreasing ampli- wave, one or more inhomogeneous waves may be gener-
tude with depth. This is a generalization of the concept ated at each interface. In fact, under certain restrictions for
of plane waves. These kind of boundary waves are very the values of the seismic velocities, it is possible to find
common in layered media and are generated whenever a eigensolutions called Stoneley waves, which propagate as
slow wave is converted into a faster one by reflection or inhomogeneous waves trapped near the interface. These
refraction by an interface. It is common in seismology to waves are important in studies of seismic wave propaga-
call these waves inhomogeneous, or head waves, or other tion in bore holes where they propagate at the interface
particular names depending on the mechanism of genera- between water and rock.
tion of the waves. Beyond the critical angle, the reflection
and refraction coefficients become complex, so that wave D. Attenuation
amplitudes contain a complex part that produces a phase
The earth is not a perfect elastic medium. Elastic waves in
shift during wave conversion.
the earth are attenuated as they propagate. This attenuation
is due to two effects: intrinsic dissipation and scattering.
Dissipation is the loss of energy of the wave due to internal
B. Rayleigh Waves at a Free Surface
friction or other mechanisms of conversion of seismic en-
Another important effect of the boundary conditions at ergy into heat. Scattering is the loss of energy of a seismic
the free surface is the presence of Rayleigh waves. These wave due to the presence of inhomogeneities along the
waves are free modes or eigensolutions of the wave equa- propagation path. These inhomogeneities radiate part of
tion for a homogeneous medium (4) together with the the energy in all directions so that the wavefront appears to
boundary conditions (24) and the condition of conver- lose energy. Energy flow in a seismic wave is proportional
gence of the wave field at infinity. Rayleigh waves cor- to the square of the complex amplitude A():
respond to a very particular combination of P and S 1
inhomogeneous waves. They are the roots of the denomi- Es = c|A()|2 (30)
2
nator R( px ) = 0 of the reflection coefficients (26) and (28).
These roots occur for a particular real value of the hori- where is the density and c the appropriate wave speed.
zontal slowness pxR , called the Rayleigh wave slowness. Attenuation is measured in terms of the quality factor
Its inverse c R = 1/ pxR is the Rayleigh wave speed, which Q(), which measures the rate at which energy is lost
is slower than both P and S waves. The Rayleigh wave from the wave:
speed is c = 0.91 for an elastic half-space with = , 1
=
c 1 d Es
. (31)
and vary around 0.9 for reasonable values of the ratio Q() ES d x
/. For shallow sources the Rayleigh waves dominate This equation has a simple solution, which written in terms
the elastic wave field observed on the free surface. In the of the amplitude gives
earth, seismic wave speeds increase with depth, so that the
free-surface effect is coupled to waveguide effects, mak- |A(x, )| = |A0 ()| exp(x/2cQ), (32)
ing the Rayleigh waves in the earth much more complex where A0 () is the amplitude of the wave at the source.
than those in an elastic half-space. The absolute value of the complex amplitude decreases ex-
ponentially at a rate controlled by Q(). This is, however,
not enough to determine the way a plane wave is af-
C. Reflection and Transmission
fected by attenuation because (31) does not permit us
at a Solid-Solid Boundary
to calculate the complex phase of the wave. If we as-
Internal discontinuities in seismic velocity or density will sume that the phase is zero and use |A| instead of A, it
generate reflected and transmitted waves and produce con- turns out that the wave will violate causality. The seis-
version between P and S waves. The corresponding ex- mic pulse calculated by Fourier inversion will broaden
pressions for the reflection and transmission coefficients up symmetrically around the arrival time of the P waves.
P1: GQT/GUB P2: GLM Final Pages
Encyclopedia of Physical Science and Technology EN014c-680 July 28, 2001 17:23

582 Seismology, Theoretical

Lower-frequency waves will appear to propagate faster the former c = , the P-wave velocity, while for the lat-
than higher-frequency ones. In order to respect causal- ter c = , the shear wave velocity. T (r, r0 ) is travel time
ity, a certain frequency-dependent phase correction has between the source point r0 and the observer at r. The
to be introduced. This phase correction is determined by 0 and c0 are the density and wave speed at the source,
the KramersKoenig relationships. Attenuation of seismic respectively. J (r, r0 ) is the ray Jacobian or geometrical
waves produces a certain amount of dispersion that has to spreading of the wavefronts; it will be defined later in the
be taken care of when comparing velocities determined paper. In many applications J may be negative or complex
by observation of seismic waves in different frequency so that the proper branch of the square root of J in (33)
ranges. should be chosen. The amplitude A(r, r0 ) is a complex
valued vector that contains information about the ampli-
tude and polarity of the waves. Finally, S() is the source
IV. SEISMIC WAVES IN HETEROGENOUS wavelet that contains information about the time variation
MEDIA of the source (or seismic source time function).
Equation (33) is an approximation to the wave equa-
The study of seismic wave propagation in simple homo- tion valid only at high frequencies. In this approximation
geneously layered elastic media is important for under- A0 , J , and T are assumed to be slowly varying functions
standing the basic features of seismic wave propagation of r; the only rapidly varying term in (33) being the expo-
and generation. The earth is, however, very heterogenous nential. This form of the solution simplifies the calculation
so that a proper understanding of propagation in inhomo- of seismograms in a substantial way. It is in fact simple to
geneous media is necessary to simulate seismograms and do the inverse Fourier transform of (33) in order to obtain
to invert the structure of the earth from observed seis- displacement in the time domain:
mic waves. Full numerical solutions to these problems   
0 c0
are becoming increasingly feasible thanks to the expo- u(r, t) =  A0 S[t T (r, r0 )] +
c J (r, r0 )
nential increase of computer power in recent years. How-
  
ever, even if we could compute numerical seismograms 0 c0
 A0 S [t T (r, r0 )], (34)
we need some practical approximations in order to un- c J (r, r0 )
derstand them and to identify the main seismic arrivals
(or phases) in those records. The most widely used of where S (t) is the Hilbert transform of s(t),

the approximate techniques is ray theory, which can be 1 S( )
S (t) = P.V. d,
simply described as an extension of spherical wave solu- t
tions to slowly varying spherical media. Conditions for the
validity of ray theory are difficult to establish, but they in- and P.V. denotes the principal value of the integral.
clude at least the following: the properties of the medium
must change very slowly on the scale of a wavelength B. Ray Tracing
and abrupt changes must be confined to well-defined in-
terfaces. Under these assumptions, propagation between The computation of synthetic seismograms by ray theory
material discontinuities can be computed with ray theory, consists in the computation of travel time T , geometrical
while the interaction of rays and waves with the interfaces spreading J , and the vector amplitude A. Travel times are
can be handled by the plane wave reflection and refraction obtained by ray tracing, the Jacobian J by ray perturba-
described in the previous section. tion theory (also called paraxial ray theory when it is based
on the Hamiltonian formulation). The vector amplitudes
are traced following an elementary coordinate system
A. Ray Theory for Body Waves along the ray propagation (see Fig. 3) and computing ap-
Ray theory is based on an ansatz or hypothesis about the propriate reflection and transmission coefficients along the
form of the elastic wave field, which is assumed to be of ray trajectory. Ray tracing equations can be determined
the form: by direct substitution of the ansatz (33) into the elastody-
 namic wave equation. We find
0 c0
u(r, ) = A0 (r, r0 ) S()ei T (r,r0 ) , (33) (T )2 = 2 , A0 (T ) = 0, (35)
c J (r, r0 )
where u(r, ) is the Fourier-transformed displacement at for P waves, and
point r in the elastic medium and is the circular fre-
(T )2 = 2 , A0 (T ) = 0, (36)
quency. The and c are the density and wave velocity at
point r. This expression applies both to P and S waves; for for S waves.
P1: GQT/GUB P2: GLM Final Pages
Encyclopedia of Physical Science and Technology EN014c-680 July 28, 2001 17:23

Seismology, Theoretical 583

dr
= cp, (37)
ds
dp 1
= ,
ds c
where c is either , the P-wave speed, or , that of S
waves. The second equation in (38) is closely related to
ray curvature given by
1
= cn , (38)
c
where n is the unit normal to the ray trajectory. Thus the
curvature of the ray is controlled by the gradient of slow-
FIGURE 3 Generalized ray coordinates. The s measures length ness. We can interpret it very simply: rays are deflected
along the ray, and 1 and 2 are coordinates on the wavefront that away of regions of high wave speed. This explains in a
label neighboring or paraxial rays to the central ray.
simple way a common observation for ray tracing in the
earth: rays are attracted by low velocity zones and are
rejected by regions of high speeds.
Let us define a wavefront as a surface of constant travel Solution of (38) requires the specification of initial or
time T (r) = const; then the vector p = T is the local boundary conditions. The simplest problem is to specify
slowness vector of the wave front. The p is perpendicu- the initial position r(s0 ) and slowness p(s0 ) for each ray
lar to the wave front and its modulus |p| is the slowness on some initial surface. For a point source, for instance,
at which the wave front moves locally. The two eikonal r(s0 ) is the same for all the rays, while p(s0 ) changes from
equations (35 and 36) simply state that the slowness for P ray to ray. Once the initial conditions are specified, the
waves is the inverse of the P-wave speed , and that the ray tracing system (38) can be integrated numerically, for
slowness of S is the inverse of its speed . The second instance, by the Runge Kutta method. Let us remark that
terms in (35 and 36) have a simple interpretation. The first the six ray tracing equations (38) are not really indepen-
means that the P-wave vector amplitude is parallel to the dent because |dr/ds| is always equal to 1, and |p| = c1 .
slowness vector |p|, while the second implies that the vec- Thus, actually only four of the equations are independent.
tor amplitude for S waves is perpendicular to the slowness In practice, when (38) is being solved numerically these
vector p. Thus ray tracing determines not just travel times relationships may be used as a consistency check. Once
but also the polarities of seismic waves. the rays have been traced, the travel time T (r, r0 ) may be
Given an initial wave front T0 (r), say, the successive calculated by direct integration of
positions of the front may be computed step by step using
(35 or 36). This is not a practical procedure, although it dT 1
=
has become popular in recent times for small scale ap- ds c
plications. The reason it is difficult to apply is that travel along each ray.
time is almost always a multivalued function that presents Solution of the initial value ray tracing problem is rela-
folds, cusps, and other singularities several that can be tively straightforward. In most seismological applications,
described very accurately with catastrophe theory. Direct however, the usual problem is to trace a ray that passes
integration of the eikonal can follow only one sheet of through two fixed points r0 and r1 . This is the so-called
the wave fronts, and fails at caustics and other singu- two-point ray tracing problem, which is very nonlinear and
larities. It is possible to evacuate these problems com- closely related to inverse problems. Most effective meth-
puting the normal to the wave front at every integra- ods for the solution of the two-point ray tracing problem
tion step and following the front along each of these are based on the iterative search of the initial slowness p1
vectors, but this technique has not yet been used in by continuation or ray perturbation methods.
practice. Given appropriate initial conditions, the set of rays and
The more frequent method for tracing rays is to derive wave fronts is uniquely determined in those regions of
from (35 or 36) a system of ray tracing equations. For space that are illuminated by the initial data. Because the
that purpose we introduce the ray coordinate s, as shown ray tracing system (38) is nonlinear, the ray field may
in Fig. 3, the curvilinear distance s along the ray. Re- present singularities. In order to understand these prob-
marking that the tangent to the ray t = dr/ds is parallel lems and to determine geometrical spreading, we remark
to the slowness vector p = T , we can easily derive the that the set of rays and wave fronts form a curvilinear coor-
system: dinate system. As shown in Fig. 3 we introduce orthogonal
P1: GQT/GUB P2: GLM Final Pages
Encyclopedia of Physical Science and Technology EN014c-680 July 28, 2001 17:23

584 Seismology, Theoretical

curvilinear coordinates 1 an 2 on the wave fronts in which may be any variable. In the particular case when
addition to the ray coordinate s. Each pair (1 , 2 ) defines s is the curvilinear coordinate, the rays that render (39)
a ray. The curvilinear coordinate set (s , 1 , 2 ) defined in stationary may be found by standard techniques of the
this form is usually called the ray coordinate system. Any calculus of variations. The Euler equations are
point P in the region illuminated by rays may be defined  
d 1 dr 1
by its ray coordinates. In this coordinate system the vol- = 0. (40)
ume element is d V = J (r, r0 ) ds d1 d 2 , where J is the ds c ds c
Jacobian of the transformation from Cartesian to ray co- which can be easily seen to be exactly equivalent to the
ordinates. Since ds is a curvilinear abscissa along the ray ray tracing equations (38) inserting the first equation of
the cross section (Fig. 2) of a beam of rays defined by the (38) into the second one. This is an alternative way to de-
four rays with coordinates 1 , 2 , 1 + d1 , and 2 + d2 rive the ray tracing equations without reference to the ray
is given by d S = J (r, r0 ) d1 d2 . Thus, J is a measure of ansatz, although the variational method does not permit
the variation of the cross section of this beam. J is usu- us to compute the amplitude variations in (33).
ally called geometrical spreading, because it measures the
spreading of the wavefront around the ray (1 , 2 ).
D. Rays in Vertically Heterogeneous Media
We can now explain the presence of J 1/2 in the expres-
sion for ray theoretical seismograms (3). Elastic energy If wave speeds vary only with depth, it is possible to find
flow across a wave-front element of cross section d S is a simple closed form solution of the ray tracing equations.
1 This is one of the first problems solved by seismologists
dE = c |u|2 J d1 d2 . and is the basis for numerous applications of ray theory
2
to seismic interpretation and inversion of travel times in
Since in the ray approximation energy flows along a beam
the earth. There is a good reason why this is a good ap-
of rays without lateral scattering, the energy flux across
proximation: wave speeds vary much more rapidly with
the cross section defined by d1 d2 must be conserved.
depth than they do laterally; for this reason lateral het-
Thus energy conservation along a ray tube implies that
erogeneities can be treated as small perturbations of a
amplitudes vary like (c J )1/2 as in (33). As mentioned
vertically stratified earth model. Initially, when the main
earlier, the transformation to ray coordinates may be sin-
objective of seismologists was to find the vertical vari-
gular. Near these singularities, J 0, so that the usual
ation of seismic wave speeds, the observed travel times
expressions of ray theory as given by (33) fail and other
were averaged in order to eliminate lateral variations. At
methods, like WKB or Gaussian beam summation, have
present, a preliminary reference earth model (PREM) has
to be used.
been proposed that serves as standard vertically stratified
distribution of seismic speeds. Most uses of ray theory
C. Variational Formulation
consist in computing travel times and synthetic seismo-
The ray tracing problem has been posed so far in its differ- grams for small perturbations of the PREM model. These
ential form. Alternatively, it may be posed in a variational models are also being used to locate earthquakes by itera-
form which may be used to develop alternative methods of tive inversion techniques. As data has improved in quality
solution of these equations, to introduce perturbation the- and precision earthquakes are increasingly being located
ory, to calculate wave fronts, etc. The most common use in laterally heterogenous media.
of the variational formulation in seismology is in travel In vertically heterogenous media, the second set of
time inversion. The starting point for this formulation is equations in (38) yields d px /ds = d p y /ds = 0 because the
Fermats principle, which may be stated in the following gradient of slowness is vertical. Therefore, the horizontal
form: among all trajectories joining two fixed points r0 components of slowness, px and p y , are conserved dur-
and r1 , a ray is the trajectory for which the travel time is ing ray propagation. This is completely equivalent to the
stationary. We write this condition in the form Snell law used in elementary ray tracing. The most obvious
 1 consequence of the conservation of px and p y is that rays
dr
T (r0 , r1 ) = u(r) ds = 0. (39) remain in the vertical plane defined by the initial point r0
0 ds and its initial slowness p0 . Another, perhaps less obvious,
where s is as before the curvilinear distance along the ray consequence is that the ray fronts have cylindrical sym-
and means variation with fixed end points. Let us note metry with respect to a vertical axis through the source.
that since the ray tracing problems are highly nonlinear Because of cylindrical symmetry we can choose the x axis
several rays may satisfy the variational condition (39). to coincide with the horizontal projection of vertical plane
The variational principle (39) looks for an extremal tra- that contains the ray. For simplicity seismologists use p
jectory without any constraint upon the ray coordinate s, instead of px for the horizontal slowness and call it ray
P1: GQT/GUB P2: GLM Final Pages
Encyclopedia of Physical Science and Technology EN014c-680 July 28, 2001 17:23

Seismology, Theoretical 585

ing the source, it will descend to a maximum depth z bS

from which it will return to the surface. There it will be
reflected as an S wave and converted into a reflected P
wave. The reflected S ray will again descend to depth z bS
and return back to the surface, where it will be reflected
and converted again. This will occur many times, but the
ray will always remain inside a zone of depth z bS from the
surface of the earth. The P waves generated by conversion
of the S waves on the earths surface will have the same
horizontal slowness p as the S waves, and they will also
FIGURE 4 Ray trajectory in a vertically heterogeneous medium. be constrained to a certain layer of depth z bP , which will be
shallower than z bS . Thus the whole set of converted P and
parameter. Then equation (38) can be integrated immedi- S waves will be trapped in a layer or wave-guide of depth
ately to z bS . In fact, since the depth of this waveguide is entirely
 z controlled by p, it is not really relevant whether the wave
p started as a P or an S wave of horizontal slowness p. The
x(z , x0 , z 0 , p) = x0 +  dz, (41)
z0 c 2 (z) p2 existence of this waveguide is entirely due to the increase
which gives the horizontal range of the ray as a function of of wave velocities with depth.
vertical position z and the initial conditions x0 , z 0 , and p. For a given frequency , all values of the apparent slow-
Likewise it is straightforward to integrate the travel time ness p will be associated with a family of multiply re-
along the trajectory (41): flected P and S waves. These multiply bouncing waves
 z will interfere, and depending on their frequency and the
1 velocity structure, constructive interference will occur for
T (z , x0 , z 0 , p) =  dz. (42)
2
z 0 c(z) c (z) p 2 a set of discrete values of p. A constructively interfering
set of P and S waves forms a surface wave, which will
Given a wave speed distribution c(z) for P or S waves,
propagate in the waveguide defined by the surface and the
(41) and (42) yield x and T as functions of depth z and ray
turning depth for that value of p. Surface waves are two-
parameter p. Since p is the horizontal component of slow-
dimensional waves so that their geometrical decay (R 1/2 )
ness, it can be expressed in terms of the incidence angle
is less rapid than that of noninterfering body waves (R 1 ).
i as p = sin i(z)/c(z) (see Fig. 4). Thus as c(z) increases
For this reason, at relatively large distances from the earth-
with depth, i also increases until a certain bottoming depth
quake, surface waves become the most prominent arrivals
z b at which i = /2. The ray is horizontal at this depth and
in seismograms. If we let the frequency change, con-
it returns toward the surface; thus for each ray parameter
structive interference will occur for p = pn (), where the
there is a maximum depth that the ray can attain. This is
index n [0, . . . , ] is the mode number. For each n
the basis for the inversion of seismic speeds from travel
there is a well-defined relation, called dispersion relation,
times and ray paths. In an earth model in which speed in-
between horizontal slowness and frequency. Dispersion is
creases monotonically with depth, rays of decreasing ray
frequently written in terms of the horizontal phase veloc-
parameter sample increasingly deeper parts of the earth.
ity (or phase velocity, for brief) c = 1/ pn () = cn (). The
This procedure is no longer possible when wave speeds
mode with the lowest value of cn is called the fundamental
present low velocity zones (LVZ) where the wave speed
mode and is labeled by n = 0. Higher values of n and cn
decreases with depth. In that case the speed within the
correspond to overtones of the surface waves.
low velocity zone can not be directly determined by travel
The most important characteristic of surface wave is
times of body waves. Surface wave dispersion and the fre-
dispersion. For body waves, the velocity of propagation
quencies of normal modes are a better approach to study
along the ray is constant and independent of frequency.
LVZs.
For surface waves, the horizontal phase velocity depends
on the frequency of the wave because the conditions for in-
V. SEISMIC SURFACE WAVES terference can only be satisfied by special values of phase
velocity and frequency. Dispersion relations are somewhat
As discussed in the preceding section, rays in a vertically difficult to obtain for realistic earth models; they have to be
heterogeneous model of the earths structure penetrate to computed numerically by a method that can be sketched
a maximum depth that depends on their horizontal slow- as follows. Let
ness or ray parameter p. Consider, for instance, an S-wave
ray of frequency and horizontal slowness p. After leav- u(x, z) = y(z) exp[i(t px)] (43)
P1: GQT/GUB P2: GLM Final Pages
Encyclopedia of Physical Science and Technology EN014c-680 July 28, 2001 17:23

586 Seismology, Theoretical

be the components of the displacement, where we have A simple way to do this is to do an inverse Fourier trans-
explicitly written that we look for waves propagating hor- form of the frequency domain expression (43). At suf-
izontally in the x direction with slowness p. The depth- ficiently high frequencies it is possible to calculate this
dependent functions y(z), which describe the vertical vari- Fourier transform asymptotically by using the stationary
ation of the amplitude of the surface wave, are usually phase technique. The main result of this analysis is that in
called the vertical eigenfunctions. Since velocity depends the time domain surface waves of predominant frequency
only on depth, inserting (43) into the equation of motion propagate with a group velocity U () = /k, where
will yield a system of three second-order ordinary differ- k = p is the horizontal wave number. In the earth, the
ential equations for the y(z) functions. These equations group velocity is always less than the phase velocity. De-
have to be solved together with the free-surface boundary pending on the velocity model, group velocity may vary
conditions i z = 0, for i [x, y, z] and the condition that very rapidly with frequency, especially for higher modes.
y(z) 0 when z 0. This set of equations and boundary It presents minima and maxima called Airy phases, which
conditions defines an eigenvalue problem for pn () with correspond to arrivals of relatively high amplitude.
associated eigenvectors yn (z) 0. Detailed examination
of the equations shows that they can be separated into two
A. Free Oscillations of the Earth
independent sets. The first and simpler one contains only
the component y y (z). These are waves that are polarized As the wavelength of the surface waves increases and it be-
horizontally in the transverse direction with respect to the comes comparable to the radius of the earth, a new quanti-
direction of propagation of the surface waves. These waves zation occurs and frequencies become discrete. The theory
are called Love waves. The other set of solutions are waves of normal modes is one of the most complex in seismology
polarized in the vertical plane, that is, y = [yx , 0, yz ], and because at low frequencies none of the usual simplifica-
are called Rayleigh waves. The eigenvalue problems for tions allowed by ray theory is applicable and the complete
Rayleigh and Love waves are solved independently by nu- equations of elastodynamics coupled with gravity pertur-
merical methods that yield both pn () [or cn ()] and the bations have to be solved. For the relatively simpler case
associated eigenvectors yn (z). The latter may eventually of earth models whose velocity and density depend only
be used to compute synthetic seismograms. on depth, there are standard programs to perform calcu-
Dispersion relations for Love and Rayleigh waves have lations of the frequencies of the normal modes. For more
some common properties that are useful in practical appli- realistic, laterally heterogeneous models, only perturba-
cations. The most important one is that dispersion curves tion techniques are currently used to compute the earths
for the different modes do not intersect each other in the eigenfrequencies.
,c plane. Also, c decreases monotonically from a max- In a vertically heterogeneous earth model, the elastody-
imum at the cutoff frequency down to a minimum con- namic equations can be separated in spherical coordinates,
trolled by the lowest shear wave velocity in the model. assuming that displacement is expanded in spherical har-
We can understand this by the following simple argument: monics. It turns out that just as with surface waves, the
As frequency decreases, wavelengths become longer, and equations separate into two sets. The first are the toroidal
for a given dispersion branch, the eigenfunctions pene- oscillations for which particle velocity is mainly tangen-
trate deeper into the earth. Hence, they propagate on a tial to the surface of the earth. At high frequencies the
deeper waveguide, and since the velocity increases with series of eigenfrequencies for toroidal modes merge con-
depth, they sample higher velocities. Since in most earth tinuously into the dispersion curves for Love waves. The
models the minimum velocity occurs on the surface, high- other type of free oscillation is spheroidal, with particle
frequency surface waves propagates with velocities very motions concentrated on vertical planes. At high frequen-
close to that of the shear wave velocity at the top of the cies, once the spectrum becomes continuous, spheroidal
crust. On the other hand, at longer periods the maximum oscillations merge into the dispersion curves for Rayleigh
phase velocity should approach the velocity in the deeper waves.
layers of the earth. At very long periods, however, the flat The surface pattern of displacement associated with a
model that we have considered so far becomes inappro- given mode of vibration of the earth is controlled by the
priate, and we have to take into account the sphericity of shape of the vector spherical harmonics. Spherical har-
the earth. monics are characterized by two numbers:  and m; 
So far we have considered monochromatic surface varies from 0 to , while  m . The number  mea-
waves of fixed frequency . Seismograms are recorded, sures the number of nodal circles that the spherical har-
of course, in the time domain, and although they may be monic has on the surface of the earth, while m measures the
processed by Fourier transformation, it is useful to under- number of these circles that intersect the earths equator. In
stand the behavior of surface waves in the time domain. a spherically symmetric earth model, the eigenfrequencies
P1: GQT/GUB P2: GLM Final Pages
Encyclopedia of Physical Science and Technology EN014c-680 July 28, 2001 17:23

Seismology, Theoretical 587

 are independent of the zonal number m; that is, these faulting that takes place at its source. Typically, for large
eigenfrequencies are degenerate. For each  there are thus earthquakes, a few meters of slip can take place in a few
(2 + 1) modes that share the same frequency. Every set of seconds generating strong seismic waves that we identify
such degenerate modes is called a multiplet. As mentioned as the main source of an earthquake.
earlier, the azimuthal number  determines the number of
nodal lines in the mode. It is therefore closely related to A. Seismic Moment Tensors
the horizontal wavelength of the mode on the surface of The most common theoretical approach to the study of
the earth. At high frequencies the wavelength is approxi- seismic radiation is to construct mechanical models equiv-
mately given by = 2a/( + 1/2) where a is the radius alent to faulting in the earth. Instead of describing the seis-
of the earth. Thus  controls the azimuthal variation of the mic event by the detailed distribution of slip on the fault,
free oscillation. Just as for surface waves, for every  there we represent it by a one, or a few, multipolar sources.
is a complete series of overtones labeled with the overtone Such an approach is deemed objective because it makes
number n, with n = 0 for the fundamental mode. no assumptions about the rupture process, not even the ex-
The spectrum of the earth is discrete because of its finite istence of a fault! Only later, once a particular geometry
size. It would, of course, be possible to construct synthetic has been chosen, can the multipolar sources be related to
seismograms at any frequency, even for body waves, by some specific geometry of the source.
summation of normal modes. This is indeed done for long Seismic sources are of internal origin so that they may
periods (down to about 60 sec), but it is a very inefficient not be due to point forces of the type we studied in
procedure at higher frequencies because of the numerous Section IIE. For a set of forces to be of internal origin,
overtone branches that interfere to produce the harmonics they must satisfy the following two conditions:
of surface waves. At even higher frequencies for periods  
less than 20 sec, say, modes of high overtone order inter- f(r0 ) d V0 = 0, r0 f(r0 ) d V0 = 0, (44)
fere to produce body waves. Numerical evaluation of in- V0 V0
terfering oscillatory sums is very inefficient in a computer, where V0 is any volume that surrounds the source. The
producing aliasing, beating, and other numerical artifacts. first condition states that the resultant of the body forces
For this reason, numerical methods based on modern spec- must be zero, so that there is no net force applied to the
tral finite element methods are being actively explored to earth by the earthquake; the second states that the total
generate synthetic seismograms. torque of these forces must also be zero. A distribution
that satisfies these two conditions is of internal origin and
is acceptable because it does not perturb the motion of the
VI. SEISMIC SOURCE THEORY earth in its orbit.
Any body force that satisfies the two conditions (44)
Many shallow earthquakes are accompanied by evidence necessarily derives from the divergence of a symmet-
of surface faulting. These faults have dimensions ranging ric tensor m that we call the moment tensor density
from a few meters to hundreds of kilometers, depending distribution:
on the size of the event. A substantial amount of informa-
tion demonstrates that practically all shallow earthquakes f(r, t) = m(r, t). (45)
are due to the fast propagation of rupture along one ore Moment tensors are well known in other areas of me-
more well-defined fault surfaces. Earthquakes occur in chanics where they are usually called spontaneous in-
the brittle, colder parts of the lithosphere in response to elastic stresses or transformational stresses following a
the slow but continuous accumulation of tectonic stresses nomenclature introduced by Eshelby. Backus introduced
due, in most cases, to the relative motion of the litho- the name stress glut because he interpreted m as an excess
spheric plates. Earthquakes are usually localized near the of stress with respect to elastic stresses. None of these
boundaries of the plates or, in their interiors, in zones other terms became popular in seismology.
that are mechanically weakened. It has been determined For a body force distribution that derives from a mo-
from geodetic measurements that the deeper parts of plate ment tensor distribution, the representation theorem (22)
boundaries creep continuously in response to plate motion. becomes
At the relatively colder temperatures that prevail at shal-  t
lower depths, the lithosphere cannot deform sufficiently u i (r, t) = G i j,k (r, t | r0 , t0 )m jk (r0 , t0 ) d V0 , (46)
fast to follow plate motions and stresses accumulate. When 0 V0

stresses reach a certain threshold, the fault becomes un- where the notation G i j,k means that the Greens function
stable and a fast process of stress release takes place. The for a point force G i j has been derived with respect to
main effect of an earthquake is the discontinuous slip or coordinate xk .
P1: GQT/GUB P2: GLM Final Pages
Encyclopedia of Physical Science and Technology EN014c-680 July 28, 2001 17:23

588 Seismology, Theoretical

In many applications, when the wavelengths of inter- fault. For a fault model all we would require is a scalar
est are much longer that the source dimensions, we can distribution of moment on the fault. It can be easily shown
replace the seismic moment by a concentrated source: that for a flat fault the moment density at any point on the
fault surface is simply the product of the elastic rigidity
m(r, t) = M (r r0 )(t t0 ), (47)
times the slip at that point D. The scalar seismic moment
where M is the point moment tensor equivalent to the in that case is
earthquake. Every element of M has units of mechanical 
moment (Nm). The moment tensor can be decomposed in M0 (t) = D(r, t) d S0 . (49)
S0
many ways into simpler fault models. The simplest of them
is a pure shear moment tensor component, m x y say. In this For the full description of an earthquake that took place
case the equivalent set of forces forms a double couple, on a fault of arbitrary geometry, we would have to specify
i.e., four forces in equilibrium with zero net moment. In a full moment tensor at each point of the fault and the
this case we write relation between the equivalent scalar moment and the
actual distribution is frequency dependent and complex.
m(r0 , t0 ) = M0 [ex e y + e y ex ](r0 ) (48) Such a model would require a complete description of
and we call M0 the scalar seismic moment of the source. the geometry of the fault surface and of the slip at every
This number has now replaced all previous methods to point of the fault. Although detailed models of this kind
quantify an earthquake including magnitude, intensity, have recently been inverted for a few very well-recorded
and energy. Since seismic moments span more than 10 earthquakes, for most events this is too ambitious at the
orders of magnitude, logarithmic scales are a more prac- present time. For this reason, fault models are simplified
tical way to estimate the size of an earthquake. In the by making geologically reasonable assumptions about the
1970s Kanamori introduced a new definition of mag- shape of the fault and the distribution of slip. The most
nitude that computes the so-called moment magnitude common assumptions is that the fault is planar, and that
log M0 = 1.5Mw + 9.3 directly from the moment. rupture starts at some predetermined hypocenter and then
A more general approach to the interpretation of the spreads over the fault surface at a fixed rupture speed.
point moment tensor M is to interpret in terms of couples Estimates of rupture speed for well-studied earthquakes
of sources. For that purpose we diagonalize the tensor always range between 60 and 80% of the shear wave speed.
M computing its three eigenvalues i and its associated As rupture develops starting from the hypocenter, the total
eigenvectors mi , where the index i varies from 1 to three. moment increases, and once rupture comes to a stop, it
Since the moment tensor is symmetric, all its eigenvalues eventually reaches a maximum (or static) value M0 .
and eigenvectors are real and orthogonal, so that the most
general point model of an earthquake is a set of three or- SEE ALSO THE FOLLOWING ARTICLES
thogonal linear dipoles of strengths 1 , 2 , 3 . A frequent
observation is that for many earthquakes the trace of the EARTHQUAKE MECHANISMS AND PLATE TECTONICS
moment tensor trace (M) is zero, so that 2 = (1 +3 ). EARTHQUAKE PREDICTION SEISMOLOGY, OBSERVA-
This conditions is generally interpreted as a lack of volume TIONAL WAVE PHENOMENA
change associated with earthquakes. This is the way they
are currently described in the daily reports of moment
tensors published by Harvard University, United States BIBLIOGRAPHY
Geological Survey, and several other institutions around
the world. Aki, K., and Richards, P. G. (1980). Quantitative Seismology, Freeman,
San Francisco.
Ben Menahem, A., and Singh, S. J. (1981). Seismic Waves and Sources,
B. Dislocation Model of Faulting Springer Verlag, Berlin and New York.
Dahlen, T., and Tromp, J. (1999) Theoretical Global Seismology,
Most earthquakes are due to faulting or slip along a well- Princeton University Press, Princeton, NJ.
defined fault surface. In the particular case in which the Kennett, B. L. N. (1985). Seismic Wave Propagation in Stratified
fault is perfectly flat, we can interpret the seismic moment Media, Cambridge University Press, London and New York.
Kostrov, B. V., and Das, S. (1989). Principles of Earthquake Source
tensor in a simpler way: one of the two couples of forces Mechanics, Cambridge University. Press, London and New York.
of the seismic moment would then extend along the fault Udas, A. (1999). Principles of Seismology, Cambridge University
plane; the other would be parallel to the normal to the Press, London and New York.