This action might not be possible to undo. Are you sure you want to continue?

BooksAudiobooksComicsSheet Music### Categories

### Categories

### Categories

Editors' Picks Books

Hand-picked favorites from

our editors

our editors

Editors' Picks Audiobooks

Hand-picked favorites from

our editors

our editors

Editors' Picks Comics

Hand-picked favorites from

our editors

our editors

Editors' Picks Sheet Music

Hand-picked favorites from

our editors

our editors

Top Books

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Audiobooks

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Comics

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Sheet Music

What's trending, bestsellers,

award-winners & more

award-winners & more

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

**Rep. Prog. Phys. 67 (2004) 1429–1496 PII: S0034-4885(04)25227-7
**

The physics of earthquakes

Hiroo Kanamori

1

and Emily E Brodsky

2

1

Seismological Laboratory, California Institute of Technology, Pasadena, CA 91125, USA

2

Department of Earth & Space Sciences, University of California, Los Angeles,

Los Angeles, CA 90095, USA

Received 21 January 2004, in ﬁnal form 20 April 2004

Published 12 July 2004

Online at stacks.iop.org/RoPP/67/1429

doi:10.1088/0034-4885/67/8/R03

Abstract

Earthquakes occur as a result of global plate motion. However, this simple picture is far from

complete. Some plate boundaries glide past each other smoothly, while others are punctuated

by catastrophic failures. Some earthquakes stop after only a few hundred metres while others

continue rupturing for a thousand kilometres. Earthquakes are sometimes triggered by other

large earthquakes thousands of kilometres away. We address these questions by dissecting the

observable phenomena and separating out the quantiﬁable features for comparison across

events. We begin with a discussion of stress in the crust followed by an overview of

earthquake phenomenology, focusing on the parameters that are readily measured by current

seismic techniques. We brieﬂy discuss how these parameters are related to the amplitude

and frequencies of the elastic waves measured by seismometers as well as direct geodetic

measurements of the Earth’s deformation. We then review the major processes thought to be

active during the rupture and discuss their relation to the observable parameters. We then take

a longer range view by discussing how earthquakes interact as a complex system. Finally, we

combine subjects to approach the key issue of earthquake initiation. This concluding discussion

will require using the processes introduced in the study of rupture as well as some novel

mechanisms. As our observational database improves, our computational ability accelerates

and our laboratories become more reﬁned, the next fewdecades promise to bring more insights

on earthquakes and perhaps some answers.

(Some ﬁgures in this article are in colour only in the electronic version)

0034-4885/04/081429+68$90.00 © 2004 IOP Publishing Ltd Printed in the UK 1429

1430 H Kanamori and E E Brodsky

Contents

Page

List of frequently used symbols 1432

1. Introduction 1433

2. Earthquakes and stress in the crust 1435

2.1. Plate motion and earthquake repeat times 1435

2.2. The state of stress in the crust 1436

Principal stresses and fault orientation 1438

Strength of the crust: laboratory and ﬁeld data 1439

Conﬂicting observations? 1440

Summary 1441

3. Quantifying earthquakes 1441

3.1. Earthquake source parameters and observables 1442

A formal description of the elastic problem 1442

3.1.1. Seismic source and displacement ﬁeld 1443

3.1.2. Seismic moment and magnitude 1445

3.1.3. Strain and stress drop 1446

3.1.4. Energy 1447

Radiated energy, E

R

1447

Potential energy 1448

3.1.5. Rupture mode, speed and directivity 1449

Directivity and source duration 1449

Rupture speed 1449

3.1.6. Earthquake rupture pattern 1450

3.2. Seismic scaling relations 1451

3.2.1. Scaling relations for static parameters 1451

3.2.2. Scaling relations for dynamic parameters 1454

4. Rupture processes 1455

4.1. Fracture mechanics 1455

4.1.1. An overview of the crack model 1455

4.1.2. Crack tip breakdown-zone 1457

4.1.3. Stability and growth of a crack 1457

Static crack 1459

Dynamic crack 1459

Rupture speed 1459

4.2. Frictional sliding 1460

4.2.1. Static and kinetic friction 1461

4.2.2. Rate- and state-dependent friction 1461

4.3. The link between the crack model and the friction model 1463

Direct determination of D

c

1463

4.4. Rupture energy budget 1463

4.5. Fault-zone processes: melting, ﬂuid pressurization and lubrication 1466

Melting 1466

The physics of earthquakes 1431

Thermal ﬂuid pressurization 1466

Lubrication 1468

4.6. Linking processes to the seismic data 1468

4.6.1. The interpretation of macroscopic seismological parameters 1468

Radiation efﬁciency 1468

The relation between radiation efﬁciency and rupture speed 1471

Summary and implications 1471

5. Earthquakes as a complex system 1473

The magnitude–frequency relationship

(the Gutenberg–Richter relation) 1473

Simple models 1474

6. Instability and triggering 1476

6.1. Instability 1476

6.1.1. Stick slip and instability 1476

Stiffness of the fault system 1478

6.1.2. Nucleation zone 1478

6.2. Triggering 1479

6.2.1. Observations 1479

6.2.2. Triggering with the rate- and state-dependent friction mechanism 1482

Spontaneous behaviour 1483

Loading at a uniform rate 1483

Stepwise change in loading 1483

6.2.3. Triggering with the stress corrosion mechanism 1484

6.2.4. Aftershocks and Omori’s Law 1486

State- and rate-dependent friction and Omori’s Law 1486

Stress corrosion model and Omori’s Law 1488

6.2.5. Hydrologic barrier removal 1490

7. Conclusions 1491

Acknowledgments 1492

References 1492

1432 H Kanamori and E E Brodsky

List of frequently used symbols

A constant in the rate- and state-friction law

a half-length of Mode III crack

α P-wave speed

B constant in the rate- and state-friction law

b slope of the earthquake magnitude–frequency relationship

β S-wave speed

χ probability of earthquake occurrence

D fault slip offset

¨

D average offset

D

0

critical slip of a crack

D

c

critical displacement in the slip-weakening models

δ slip on a frictional surface

˙

δ slip speed

E elastic modulus

E

R

radiated seismic energy

E

G

fracture energy of the earthquake

E

H

thermal energy (frictional energy loss) of the earthquake

˜ e scaled energy (the ratio of radiated seismic energy to seismic moment)

G dynamic energy release rate (dynamic crack extension force)

G

∗

static energy release rate (static crack extension force, speciﬁc

fracture energy)

G

∗

c

critical speciﬁc fracture energy

γ surface energy

η viscosity, seismic efﬁciency

η

R

radiation efﬁciency

K stress intensity factor

K

c

fracture toughness (critical stress intensity factor)

k stiffness of spring, permeability

k

f

stiffness of the fault

˜

L length scale of the fault

˜

L

n

nucleation length

l

0

crack breakdown length

M

0

seismic moment

M

w

earthquake magnitude (moment magnitude)

µ rigidity (shear modulus) or coefﬁcient of friction

µ

s

coefﬁcient of static friction

µ

k

coefﬁcient of kinetic friction

p pore pressure, power of the stress–corrosion relation, power of Omori’s Law

Q heat

R seismicity rate

r

0

background seismicity rate

ρ density

S fault area

σ

0

initial stress

The physics of earthquakes 1433

σ

1

ﬁnal stress (sections 3 to 6)

σ

f

frictional stress

σ

s

static stress drop (σ

0

−σ

1

)

σ

ij

stress tensor

(σ

1

, σ

2

, σ

3

) principal stresses (section 2)

σ

Y

yield stress

σ

n

normal stress

τ shear stress, source duration

¨ τ average source duration

˙ τ stress rate

θ state variable in rate- and state-dependent friction; angle between the

fault and the maximum compressional stress

u

i

displacement vector

V rupture speed

W

0

initial (before an earthquake) potential energy of the Earth

W

1

ﬁnal (after an earthquake) potential energy of the Earth

W change in the potential energy

W

0

change in the potential energy minus frictional energy

w width of the fault slip zone

1. Introduction

Why do earthquakes happen? This age-old question was solved at one level by the plate

tectonics revolution in the 1960s. Large, nearly rigid plates of the Earth slide past each other.

Earthquakes accommodate the motion (ﬁgure 1). However, this simple answer is far from

complete. Some plate boundaries glide past each other smoothly, while others are punctuated

by catastrophic failures. Why is so little motion accommodated by anything in between

these two extremes? Why do some earthquakes stop after only a few hundred metres while

others continue rupturing for a thousand kilometres? How do nearby earthquakes interact?

Why are earthquakes sometimes triggered by other large earthquakes thousands of kilometres

away?

Earthquake physicists have attempted to answer these questions by dissecting observable

phenomena and separating out the quantiﬁable features for comparison across events. We

begin this review with a discussion of stress in the crust followed by an overview of

earthquake phenomenology, focusing on the parameters that are readily measured by current

seismic techniques. We brieﬂy discuss how these parameters are related to the amplitude

and frequencies of the elastic waves measured by seismometers as well as direct geodetic

measurements of the Earth’s deformation. We then review the major processes thought to be

active during rupture and discuss their relationship to the observable parameters. We then take

a longer range view by discussing how earthquakes interact as a complex system. Finally,

we combine subjects to approach the key issue of earthquake initiation. This concluding

discussion will require using the processes introduced in the study of rupture, as well as some

novel mechanisms.

In this introductory review for non-specialists, we gloss over many exciting and important

advances in recent years ranging from the discovery of periodic slow slip events (Dragert

et al 2001) to the elucidation of fault structure revealed by new accurate location techniques

(Rubin et al 1999). Many of these recent advances are made possible by new technology

1434 H Kanamori and E E Brodsky

Figure 1. Global seismicity (from 1 January 1964 to 31 December 1995, magnitude range,

3.1–7.3, relocated data from the International Seismological Center catalogue) and plate motion.

Earthquakes occur at the boundaries between rigid plates of the Earth’s surface that move in different

directions (from Uyeda (1978)).

The physics of earthquakes 1435

such as satellite geodesy and high-power computation. In order to interpret the new

technological advances, we must return to and push the boundaries of classical mechanical

theories. The approach we take here is to emphasize the features of classical theory that

are directly applicable to current, cutting-edge topics. Where possible, we highlight modern

observations and laboratory results that conﬁrm, refute or extend elements of the classical

physics-based paradigm. Inevitably, our examples tend to be biased towards our own interests

and research. We hope that this review will equip the reader to be properly sceptical of our

results.

2. Earthquakes and stress in the crust

Earthquakes are a mechanism for accommodating large-scale motion of the Earth’s plates.

As the plates slide past each other, relative motion is sometimes accommodated by a relatively

constant gradual slip, at rates of the order of millimetres per year; while at other times, the

accumulated strain is released in earthquakes with slip rates of the order of metres per second.

Sometimes, slip is accommodated by slow earthquakes or creep events with velocities of the

order of centimetre per month between the two extreme cases. Current estimates are that about

80% of relative plate motion on continental boundaries is accommodated in rapid earthquakes

(Bird and Kagan 2004). With few exceptions, earthquakes do not generally occur at regular

intervals in time or space.

2.1. Plate motion and earthquake repeat times

The long-term loading of the Earth’s crust has been traditionally measured using geodetic

and geological methods. Geodesy is the branch of geophysics concerned with measuring the

size and shape of the Earth’s surface. The recent progress in space-based geodesy such as

the Global Positioning System (GPS) and satellite interferometry (InSAR) provides us with a

clear pattern of crustal movement and strain accumulation. Figure 2 shows the result of the

recent geodetic measurements in Southern California. The relative plate motion determined

from these data is about 2–7 cm per year which translates into a strain rate of approximately

3 10

−7

per year along plate boundaries. The strain also accumulates in plate interiors, but

at a much slower rate about 3 10

−8

per year or less, which is an order of magnitude smaller

than that at plate boundaries.

The shear strain change associated with large earthquakes (called coseismic strain drop)

has been estimated using geodetic and seismological methods. For large earthquakes, it is

of the order of 3 10

−5

–3 10

−4

(see sections 3.1.3 and 3.2.1). Since the rigidity of the

crustal rocks, µ, is about 3 10

4

MPa, this corresponds to a change in shear stress (i.e. static

stress drop) of about 1–10 MPa. This value is at least an order of magnitude smaller than that

associated with breaking intact rocks in laboratory, which is several hundred MPa.

Dividing the coseismic strain drop by the strain rate suggests that the repeat times of major

earthquakes at a given place are about 100–1000 years on plate boundaries, and 1000–10 000

years within plates. These values agree with what have been observed at many plate boundaries

and interiors. This is the basic long-term process that governs global earthquake activity.

Based on the above process, a simple sketch of the stresses generating earthquakes can be

drawn (ﬁgure 3(a)). Stress builds up on a fault plane until it reaches the breaking strength of

the rock. Then, an earthquake occurs, the stress is relaxed and a new cycle begins. Although

the basic process illustrated here is well understood and accurately measured, the details are

more complex. For example, the loading rate is not uniform in time. A large earthquake on a

segment of a fault changes the stress on the adjacent segments, either statically or dynamically,

1436 H Kanamori and E E Brodsky

Figure 2. Velocity vectors in Southern California determined by the GPS and other space-based

methods. Red lines (in the electronic version) indicate active faults. The ﬁgure is part of the

Southern California Earthquake Center’s web-site, http://www.scecdc.scec.org/group e/release.v2.

and accelerates or decelerates seismic activity depending on the fault geometry. The strength

of the crust is not constant in time either. Fluids may migrate in the Earth’s crust, thereby

weakening the crust signiﬁcantly and affecting the occurrence time of earthquakes. The

stress drop during earthquakes may also vary from event to event. Figure 3(b) illustrates

these complications schematically and their effect on the intervals between earthquakes.

Thus, although the overall long-term process is regular, considerable temporal ﬂuctuations

of seismicity are expected, which makes accurate prediction of earthquakes difﬁcult.

2.2. The state of stress in the crust

As outlined earlier, the simplest model for earthquake initiation is to assume that when the stress

accumulatedinthe plates exceeds some failure criterionona fault plane, anearthquake happens.

Evaluating this criterion requires both a measure of the resolved stress on the fault plane and a

quantiﬁable model for the failure threshold. Aﬁrst-order evaluation of the problemdates to the

groundbreaking work of Anderson (1905, 1951). He started with the fact that any stress ﬁeld

can be completely described by its principal stresses, which are given by the eigenvectors of the

stress tensor and are interpretable as the normal stresses in three orthogonal directions. He then

proposed that: (1) the stress state could be resolved by assuming that one principal stress is

vertical since the Earth’s surface is a free surface and (2) faulting occurs when the resolved

shear stress exceeds the internal friction on some plane in the medium. Internal friction is

deﬁned analogously with conventional sliding friction as a shear stress proportional to the

The physics of earthquakes 1437

Figure 3. Stress changes and earthquake sequence. (a) Regular sequence. (b) Irregular sequence

caused by the changes in loading rate and temporal variations in the strength of crust.

(a) (b) (c)

Figure 4. Schematic of the orientation of the principal stresses and the corresponding type of

faulting. The principal stresses are σ

1

> σ

2

> σ

3

. (a) Thrust faulting: the minimum principal

stress is vertical. (b) Strike-slip faulting: the intermediate principal stress is vertical. (c) Normal

faulting: the maximum principal stress is vertical (ﬁgure from Jaeger and Cook (1979) p 426).

normal stress on a plane. In this framework, faults are expected to accommodate horizontal

motion if the vertical axis is the intermediate principal stress and accommodate both vertical

and horizontal motion otherwise. Afault that has only horizontal motion is called ‘strike-slip’.

Combined vertical and horizontal extensional motion is called ‘normal’ faulting while vertical

and horizontal compressional motion is called ‘thrust’ faulting (ﬁgure 4). Each of these three

regimes corresponds to a particular orientation of the maximum principal stress.

Andersonian faulting theory has been remarkably successful in predicting and explaining

the occurrence and geometry of faults. However, as we show below, a few contradictory

observations cast doubt on enough parts of the paradigm that it is difﬁcult to apply to

1438 H Kanamori and E E Brodsky

Figure 5. Mohr circle diagram. Given principal stress magnitudes σ

1

and σ

3

, the locus of possible

combinations of shear and normal stresses resolved on a plane are given by (2.1) and (2.2) which

is plotted as the circle. The failure criterion (2.3) is the dashed line. The failure criterion in the

presence of pore ﬂuid is the solid line (2.7). Failure on a plane at an angle θ

opt

from the orientation

of σ

1

occurs if the circle intersects the failure line as it does at the ∗. Inset shows the deﬁnition

of θ.

earthquakes in a straightforward way. We have difﬁculty measuring the coefﬁcient of friction in

the crust and have reason to believe that it varies signiﬁcantly in time and space. The evidence

also suggests that high ﬂuid pressures are important in controlling frictional behaviour, yet the

precise values of the ever-changing ﬂuid pressures are also difﬁcult to measure deep within

the crust.

Principal stresses and fault orientation. Below we develop the formalism to quantitatively

evaluate the frictional failure criterion in terms of the principal stresses. We will use the

formalism to relate the observed geometry of faulting to the frictional strength of faults.

Denoting the principal stresses by σ

1

, σ

2

and σ

3

, where by deﬁnition σ

1

> σ

2

> σ

3

,

the relationships between the principal stresses and the resolved shear stress on a plane at an

angle θ to the maximum principal stress (σ

1

) can be written analytically and depicted with a

Mohr circle diagram (ﬁgure 5). The convention in rock mechanics is that positive values of

stresses are compressional. Since rocks are weak under tension, tensional strengths are usually

<20 MPa, i.e. <10% the compressional strengths (Lockner 1995), it is generally assumed that

all three principal stresses must be positive in the Earth.

The shear stress, τ, and the normal stress, σ, on the fault plane at an angle θ to σ

1

are

given, respectively, by

τ =

σ

1

−σ

3

2

sin 2θ (2.1)

and

σ = −

σ

1

−σ

3

2

cos 2θ +

σ

1

+ σ

3

2

. (2.2)

A Mohr circle diagram is a plot of these two resolved stresses. The normal stress is on the

x-axis and the shear stress is on the y-axis (Jaeger and Cook 1979). For a given set of principal

The physics of earthquakes 1439

stresses, the solutions to equations (2.1) and (2.2) fall on a circle (ﬁgure 5). Each point on the

circle represents a particular fault orientation. The angle

,

OO

/

Q in the diagram is 2θ.

In the 17th century, Guillaume Amonton ﬁrst established that the shear traction between

two surfaces is proportional to the load. Amonton’s Law for friction on a plane between two

surfaces is written in modern terms as

τ = µσ, (2.3)

where µis the coefﬁcient of friction. Amore complete description includes the cohesive stress

C in the shear stress, i.e. τ = µσ + C. However, the ratio of shear stress to normal stress,

τ/σ, is more straightforward to measure, therefore most studies use an effective coefﬁcient of

friction µ which includes the cohesive effects (Lockner and Beeler 2002).

The fault planes on which slip can occur with the minimum possible deviatoric stress

σ

1

−σ

3

, i.e. the minimum diameter of the Mohr’s circle, are the planes inclined at angles θ

opt

to σ

1

, such that (ﬁgure 5)

tan 2θ

opt

= ±

1

µ

. (2.4)

These two angles θ

opt

are known as the optimal angles because they are the angles at which

the rock will fracture in homogeneous, unﬂawed, intact rock.

Since real rocks are seldom intact, the more important criterion is the lock-up angle. If

a weak plane, such as a fault, exists in the crust, the slip can be constrained to occur on that

plane. In this case, for a given coefﬁcient of friction µ on the weak plane, slip can occur at

angles larger than the optimal angle θ

opt

. However, there is a maximum value of θ beyond

which slip cannot occur for any combination of positive stresses (Sibson 1985). The maximum

angle, θ

lu

, is known as the lock-up angle. From (2.1) to (2.3),

σ

1

σ

3

=

1 + µcot θ

1 −µtan θ

. (2.5)

The lock-up angle is the maximum value of θ that satisﬁes equation (2.5). For positive values

of σ

1

and σ

3

, the solution exists only if the denominator is positive, i.e. tan θ 1/µ. Therefore,

θ

lu

= tan

−1

1

µ

= 2θ

opt

. (2.6)

If a fault is observed to lie at an angle θ to the maximumprincipal stress when it is slipping,

then θ θ

lu

= tan

−1

(1/µ). Therefore, µ tan θ, and the observation gives a maximum

bound on the value of µ on the fault.

Strength of the crust: laboratory and ﬁeld data. Laboratory studies of rocks show that at

the depths typical of earthquakes µ = 0.6 to 0.85 for the majority of rocks (Byerlee 1978).

Therefore, equation (2.4) predicts that faults should form at angles of 25–30˚ to the maximum

principal stress σ

1

, if they are optimally oriented. Because σ

1

is horizontal and vertical

for thrust and normal faults, respectively (ﬁgure 4), the angles between the faults and the

horizontal surface (i.e. dip angles) should be about 25–30˚ for thrust and 60–65˚ for normal

faults if they are optimally oriented. Sibson and Xie (1998) check this criterion for the special

case of intraplate thrusts. They found that 40% of the faults fall into the optimal range and

none of their study sites violated the lock-up criterion. In general, only a handful of faults

anywhere have been found to exceed the lock-up criterion. We will return to these unusual cases

below.

The predictions of the Anderson–Byerlee mechanics have also been supported by ﬁeld

experiments. Boreholes are drilled and pumped full of high-pressure ﬂuid. The pressure at

which the wall of the borehole fractures and the orientation of the resulting fracture give a

1440 H Kanamori and E E Brodsky

measure of the magnitude and orientation of the least principal stress. More sophisticated

methods use the hoop stress to infer the maximum principal stress. When these experiments

are performed in an area prone to normal faulting, i.e. where the maximum principal stress is

vertical, the magnitude of the stresses resultant on the fracture plane and their orientation are

consistent with internal friction of 0.6 (Zoback and Healey 1984).

One complication to this simple picture was recognized early on. High ﬂuid pressures

can support part of the load across a fault and reduce the friction. In the presence of ﬂuids

equation (2.3) is modiﬁed to be

τ = µ(σ −p), (2.7)

where p is the pore pressure. Hubbert and Rubey (1959) ﬁrst recognized the importance of the

ﬂuid effect on fault friction. Fluid pressure at a certain depth should theoretically be determined

by the weight of the water column above. This state is called hydrostatic. In the course of

their work on oil exploration, Hubbert and Rubey observed that pressures in pockets of ﬂuids

in the crust commonly exceeded hydrostatic pressure. They connected this observation with

studies of faulting and proposed that the pore pressure p at a depth can approach the normal

stress σ on faults, resulting in low friction.

The most spectacular support for the importance of the Anderson–Byerlee paradigm

of failure as modiﬁed by Hubbert and Rubey came from the 1976 Rangeley experiment.

Earthquakes were induced by pumping water to increase the ﬂuid pressure at depth in an oil

ﬁeld with little surface indication of faulting (Raleigh et al 1976). Using equations (2.1), (2.2)

and (2.7), the observed fault orientation, the observed values of σ

1

and σ

3

fromin situ borehole

experiments and the measured value of µ on rock samples from the site, the researchers

successfully predicted the increase in pore pressure that is necessary to trigger earthquakes.

Conﬂicting observations? The most controversial aspect of the Anderson–Byerlee

formulation has been the applicability of the laboratory values of friction to natural settings.

A fault that fails according to equation (2.7) with µ = 0.6–0.85 and hydrostatic ﬂuid pressure

is called a strong fault. Three lines of evidence have complicated the Andersonian picture and

led researchers to question whether or not faults are strong before and during earthquakes.

The most often cited evidence against the strong fault hypothesis is based on heat ﬂow

data. If µ is high, the frictional stress on the fault should generate heat. This heat generation,

averaged over geological time should make a resolvably high level of heat ﬂow if the depth-

averaged shear stress is greater than 20 MPa. Lachenbruch and Sass (1980) showed that the

San Andreas fault generates no observable perturbation to the regional heat ﬂowpattern. Some

authors have suggested that regional-scale groundwater ﬂow may obscure such a signal, but

recent modelling has shown that the data are inconsistent with any known method of removing

the heat from the fault (Saffer et al 2003). Therefore, these difﬁcult heat ﬂow observations

stand as the best evidence that the San Andreas has a low resolved depth-averaged shear stress

(20 MPa). Since this stress is lower than that which can be achieved with hydrostatic pore

pressure and Byerlee friction, the fault is weak according to the deﬁnition at the beginning

of this section. If the pore pressure is hydrostatic, the upper limit of 20 MPa shear stress

corresponds to a maximum value of µ of 0.17. The heat ﬂow data is sensitive only to the

resolved shear stress, rather than the value of µ. Pore pressures that are more than 2.3 times

the hydrostatic values can also satisfy heat ﬂow constraint without requiring small µ. The heat

ﬂowobservations cannot distinguishbetweenhighpore pressure andlowintrinsic fault friction.

The second line of evidence comes from geological mapping. Low-angle normal faults

have nowbeen robustly documented in the geological record (e.g. Wernicke (1981)). Although

it is uncertain whether or not rapid slip occurred on these faults (as opposed to slow aseismic

The physics of earthquakes 1441

creep), it is clear that large-scale movement occurred on certain faults with dip angles of 20˚

and perhaps as low as 2˚ (Axen 2004). If the faulting occurred at the lock-up angle in the

more conservative case, the lock-up angle must be 70˚, which translates to µ = 0.4 from (2.6).

Therefore, µ 0.4 on the low-angle normal faults.

Note that high pore pressure does not affect the geological result, because combining (2.6)

with (2.1) and (2.2) yields

σ

1

−p

σ

3

−p

=

1 + µcot θ

1 −µtan θ

(2.8)

and the lock-up angle is still tan

−1

(1/µ) as long as σ

3

−p 0. The only alternative is that p

exceeds the minimum principal stress and the left-hand side (lhs) of (2.8) is negative.

Athirdline of evidence complicatingthe Anderson–Byerlee paradigmis that the maximum

principal stresses next to major strike-slip faults like the San Andreas in California and Nojima

in Japan are sometimes nearly normal to the fault (Zoback et al 1987, Ikeda 2001, Provost

and Houston 2003). On the creeping zone of the San Andreas in central California, Provost

and Houston ﬁnd that the angle θ between σ

1

and the fault is ∼80˚. Therefore, according to

equation (2.6) these areas must have µ < 0.2 in order to be able to support motion. Further

north on the fault, the angle θ varies from 40˚ to 70˚ implying a maximum value of µ varying

from 0.4 to 1.2 depending on location. In Southern California, Hardebeck and Hauksson

(2001) ﬁnd values of θ as low as 60˚. Once again, high pore pressures in the fault do not

remove the need for a low value of µ in the places with high θ, if these measurements of high

values of θ reﬂect the stress state directly on the fault. Both Byerlee (1992) and Rice (1992)

argue that the stress orientation observations may not reﬂect the state of stress within the core

of a pressurized, ﬂuid-ﬁlled fault. If it is true that the orientations are only measured outside

the fault core, then there is no constraint on the fault stress from this line of evidence.

Summary. The overall picture that is emerging is a good deal more complicated than the

Andersonian view. If the framework of equations (2.1), (2.2) and (2.7) is correct then in

areas with large, mature faults it appears that the µ applicable for initiation of slip must be

signiﬁcantly different from what is measured in the laboratory for intact rocks or immature

faults like Rangely. Moreover, the stress orientation data hint that these variables may vary in

time as well as space (Hardebeck and Hauksson 2001). Alternatively, pore pressure may be

so high that it exceeds the minimum principal stress. However, increasing the pore pressure

presents new problems as rocks can fail under tension with relatively low differential stresses.

An additional complication is that µ can depend on the slip rate and its history (Dieterich

1979). Clearly, our simple criterion for earthquakes proposed above is insufﬁcient to explain

this complexity of behaviour. In order to answer our question of why earthquakes begin, we

will have to dig deeper.

3. Quantifying earthquakes

In order to begin to answer these questions about earthquakes, we need to ﬁrst reviewthe major

observational facts and the parameters we use to quantify earthquakes. The most developed

method for measuring earthquakes is to measure the elastic wave-ﬁeld generated by the sudden

slip on a fault plane. Below, we discuss how the wave amplitude and frequencies are related to

the physical properties of the earthquake. We then list the most common earthquake parameters

derived from the wave-ﬁeld and discuss their dynamical signiﬁcance. Finally we explore the

scaling relationships between the observed parameters.

1442 H Kanamori and E E Brodsky

3.1. Earthquake source parameters and observables

Aformal description of the elastic problem. An earthquake is a failure process in Earth’s crust.

For a short-termprocess, we assume that the mediumis elastic. We imagine that an earthquake

perturbs the stress ﬁeld by relaxing the stress in a localized region S embedded in the elastic

medium. Prior to an earthquake, the crust is in equilibrium under some boundary conditions

with the initial displacement ¯ u

0

(¯ r) and the stress distribution σ

0

(¯ r), where ¯ r is the position

vector. The total potential energy (gravitational energy plus strain energy) of the system at

this stage is W

0

. In most seismological problems the displacement is assumed to be small and

linear elasticity theory is used. Then, at t = 0, i.e. the initiation time of an earthquake, a failure

occurs at a point in the medium called the earthquake hypocentre. Transient motion begins,

energy is radiated, and rupture propagates into a region, S, representing the earthquake rupture

zone. After the rupture propagation has stopped and the transient motion has subsided, the

displacement and stress become ¯ u

1

(¯ r) and σ

1

(¯ r). We denote the total potential energy of this

state by W

1

. (Note that in section 2.2 subscripts 1, 2 and 3 are used to indicate the principal

stresses; here, subscripts 0 and 1 are used to indicate the states before and after an earthquake,

respectively.)

The processes in the source region S are modelled by a localized inelastic process which

represents the result of the combination of brittle rupture and plastic yielding. The seismic

static displacement ﬁeld ¯ u(¯ r) is

¯ u(¯ r) = ¯ u

1

(¯ r) − ¯ u

0

(¯ r) (3.1)

and the stress drop is

σ(¯ r) = σ

0

(¯ r) −σ

1

(¯ r). (3.2)

The change in the potential energy is

W = W

0

−W

1

. (3.3)

During the failure process (i.e. coseismic process), some energy is radiated (radiated energy,

E

R

) and some energy is dissipated mechanically (fracture energy, E

G

) and thermally (thermal

energy, E

H

). Because some parts of the fracture energy eventually become thermal energy,

the distinction between E

G

and E

H

is model dependent.

To study an earthquake process, at least three approaches are possible.

(1) Spontaneous failure. In this case, the modelled failure growth is controlled by failure

criterion (or failure physics) at each point in the medium. Thus, the ﬁnal failure surface, or

volume, is determined by the failure process itself. This is the most physically desirable model,

but it requires the knowledge of every detail of the structure and properties of the medium.

Because it is difﬁcult to gain this information in the crust, this approach is seldom taken.

(2) Dynamic failure on a prescribed source region. In this approach, we ﬁx the geometry of the

source region. In most seismological problems, the source is a thin fault zone, and is modelled

as a planar failure surface. Then what controls the rupture is the friction law on the fault

plane (constitutive relation), and the elasto-dynamic equations are solved for a given friction

law (often parameterized) on the fault plane. The resulting displacement ﬁeld is compared

with the observed ﬁeld to determine the fault friction law. This approach has been taken in

recent years as more computer power is available. (A recent review on this subject is given by

Madariaga and Olsen (2002).)

(3) Kinematic model. In this approach, the wave-ﬁeld is computed for a prescribed slip motion

on the fault using the elastic dislocation theory. Then, the slip distribution on the fault is

determined by the inversion of observed seismic data. At this stage, no source physics is

The physics of earthquakes 1443

Figure 6. Representation of a dislocation (fault) seismic source. Left: a seismic fault represented

by a shear displacement offset D over a surface with an area S, embedded in a medium with

rigidity µ. Right: a force double couple equivalent to the dislocation model shown on left, in the

limit of point source (i.e. S →0, and D →∞while the product DS remains ﬁnite).

invoked. In this sense, this approach is called kinematic. However, once the slip is determined,

it can be used to compute the associated stress ﬁeld. The displacement and the stress ﬁeld

on the fault plane, together, can be used to infer the physical process involved in failure (i.e.

friction, etc). Since many methods for inversion of seismic data have been developed, this

approach is widely used. (A recent review on this subject is also given by Madariaga and

Olsen (2002).)

3.1.1. Seismic source and displacement ﬁeld. First, consider a very small fault (i.e. point

source) on which a displacement offset D (the difference between the displacements of the

two sides of a fault) occurs (ﬁgure 6, left).

We want to ﬁnd a set of forces that will generate a stress ﬁeld equivalent to the stress ﬁeld

generated by a given imposed displacement on the fault. Since the fault is entirely enclosed

by elastic crust and no work is done by external forces, both linear and angular momentum

must be conserved during faulting. It can be shown that the force system that respects these

conservation laws and produces a stress ﬁeld equivalent to the point dislocation source is

the combination of two perpendicular force couples (ﬁgure 6, right). This force system is

commonly called a double couple source. The moment of each force couple M

0

is given by

(Stekettee 1958, Maruyama 1964, Burridge and Knopoff 1964)

M

0

= µDS, (3.4)

where µ is the rigidity of the material surrounding the fault. (Note that in section 2.2, µ is

used for the coefﬁcient of friction, but in this section it is used to represent the rigidity. In the

later sections µ is used both for the rigidity and the coefﬁcient of friction. The distinction will

be clear from the text and context.) A ﬁnite fault model can be constructed by distributing

the point sources on a fault plane. The dimension of M

0

is [force] [length] = [energy]. In

seismology, it is common to use Nm for the unit of M

0

, rather than J (joule), because M

0

is

the moment of the equivalent force system, and does not directly represent any energy-related

quantity of the source.

For simplicity, we consider a homogeneous whole space with the density, ρ, the P-wave

(compressional wave) velocity, α and the S-wave (shear wave) velocity, β. In the absence

of any interfaces in the elastic medium, disturbances are propagated as simple elastic waves

known as body waves. If a point source with seismic moment, M

0

(t ), is placed at the origin,

the displacement in the far-ﬁeld is given in the polar coordinate, (r, θ, φ), as

¸

u

r

u

θ

u

φ

=

1

4πρrα

3

M

/

0

t −

r

α

¸

R

r

(θ, φ)

0

0

+

1

4πρrβ

3

M

/

0

t −

r

β

¸

0

R

θ

(θ, φ)

R

φ

(θ, φ)

, (3.5)

where the prime symbol denotes differentiation with respect to the argument.

1444 H Kanamori and E E Brodsky

Times (s)

Figure 7. Example of displacement from the M

w

= 7.9, 3 November 2002, Alaska

earthquake recorded by a broadband seismometer 3460 km away in Mammoth Lakes, California.

The components are radial (R), transverse (T) and vertical (Z). The radial and transverse components

are the two components on the horizontal plane. The early motion on these seismograms (between

400 and 800 s) shows P- and S-waves described by (3.5). Later motion (after 800 s) shows surface

waves produced by the interactions of the waves with boundaries in the earth and heterogeneous

structure.

The ﬁrst termis the P-wave and the second term, S-wave. R

r

(θ, φ), R

θ

(θ, φ) and R

φ

(θ, φ)

represent the radiationpatterns, whichdependonthe geometryof the source andthe observation

point. (For more details, see, e.g. Lay and Wallace (1995), Aki and Richards (2002).) These

displacement components are what are measured by seismometers (ﬁgure 7).

At short distances from the source, we have an additional term representing the near-ﬁeld

displacement. The primarycomponent of the near-ﬁelddisplacement is givenapproximatelyby

u ∝

1

4πµr

2

M

0

(t ). (3.6)

The near-ﬁeld displacement is important for the determination of detailed spatial and temporal

distribution of slip in the rupture zone. Far away from the fault, (3.6) is negligible as it falls

off much more quickly than the far-ﬁeld terms (1/r

2

as opposed to 1/r). The reason why the

near-ﬁeld and far-ﬁeld displacements are proportional to M

0

(t )/r

2

and M

/

0

(t )/r, respectively,

is that the near-ﬁeld is essentially determined by the motion on one side of the fault, while

the far-ﬁeld represents the contributions from both sides of the fault. This situation is similar

to that of an electric ﬁeld from a point charge and a dipole. (For more details, see Aki and

Richards (2002)).

The physics of earthquakes 1445

Figure 8. The near- and far-ﬁeld displacements from a point dislocation seismic source which

represents a fault slip motion given by a ramp function with duration τ.

If the fault motion is a linear ramp function, then M

0

(t ) is a ramp function, which, after

differentiated, produces a box-car far-ﬁeld wave form. In general, if the fault motion occurs

over a duration of τ, then the near-ﬁeld wave form is a ramp function and the far-ﬁeld wave

form is a pulse with a duration of τ (ﬁgure 8).

The time derivative of the seismic moment

˙

M

0

(t ) is called the moment-rate function or

the source time function, and its frequency spectrum is called the moment rate spectrum or the

source spectrum.

So far, we have discussed the seismic body waves that travel directly fromthe earthquake to

the seismometer. In the real situation, seismic body waves interact with the Earth’s free surface

and many internal structural boundaries to develop reﬂections as well as surface waves (i.e. the

wave trains in the later parts of ﬁgure 7). Observed surface waves are often very long period,

10–300 s, and carry the information of the seismic source at long period. When surface waves

propagate around the Earth many times, they can be interpreted as elastic oscillations of the

Earth. The long-period oscillations can be studied using the normal mode theory. The theories

of seismic surface waves and normal modes are well developed, and have been used effectively

to study earthquakes (Gilbert and Dziewonski 1975, Dahlen and Tromp 1998).

3.1.2. Seismic moment and magnitude. We now use the elastic theory developed above to

determine parameters of earthquakes that measure the size, energy and stress during rupture.

As shown by (3.5) and (3.6), the seismic moment can be determined from the integral

of the far-ﬁeld displacement, or from the amplitude of the near-ﬁeld displacement. In the

actual determination of the seismic moment, we need to include the effect of wave propagation

in a heterogeneous structure, geometry of the source and the ﬁniteness of the source. Many

seismological methods have been developed to handle these problems, and the seismic moment

can be determined accurately from seismic data (e.g. Lay and Wallace (1995)). For a ﬁnite

source with a fault area S on which the spatially averaged slip is

¨

D(offset), the seismic moment

M

0

is given byµ

¨

DS. Because M

0

depends on the two end states, before and after an earthquake,

it does not depend on the actual time history of faulting. In this sense, it is a static parameter.

If M

0

is determined by a seismic method, and if S is estimated by either a seismic or geodetic

method,

¨

D can be determined by using the relation

¨

D = M

0

/µS.

The seismic moment M

0

can be determined:

(1) From seismic data: the amplitude of long-period surface waves and normal-modes can

be used to determine M

0

most accurately, because long-period waves are least affected by

complex propagation effects. The amplitude and frequency spectra of seismic body waves can

be also used for smaller earthquakes. This is the most common method.

(2) From geodetic data: with the advent of space-based geodetic methods (e.g. GPS and

InSAR), this method is becoming more commonly used. The synthetic aperture radar (SAR)

interferometry was used for the 1992 Landers, California, earthquake (Massonnet et al 1993)

1446 H Kanamori and E E Brodsky

Table 1. Seismic moment determinations from different data sets.

Data M

0

(Nm) Reference

Hector Mine, California, Earthquake, 16 October 1999, M

w

= 7.1

Long-period surface waves 5.98 10

19

Harvard University

Seismic body waves 5.5 10

19

Earthquake Research Institute, Tokyo University

GPS and InSAR 6.7 10

19

Simons et al (2002)

to successfully map the regional static deformation ﬁeld associated with this earthquake.

To determine M

0

accurately, good spatial coverage around the source is required.

(3) Fromgeological data: the surface break of a fault can be used to estimate M

0

. However, the

distribution of slip where a fault meets the surface of the Earth does not necessarily represent

the slip at depths, and the resulting estimate of M

0

is inevitably inaccurate. However, for

historical events for which no instrumental data are available, this method is often used.

The redundant multiple methods allow us to verify that seismic moment is well-measured

by seismic methods to an accuracy unequalled by any other seismic parameters. Table 1 shows

the results for the 1999 Hector Mine, California, earthquake where the seismic moment was

independently measured by methods 1 and 2. The values determined by different methods

generally agree within 30%.

The following web-sites provide a catalogue of seismic moment of large earthquakes in

the world, compiled by the Seismology Group of Harvard University, Earthquake Information

Center of the Earthquake Research Institute of Tokyo University and the United States

Geological Survey, respectively.

• http://www.seismology.harvard.edu

• http://wwweic.eri.u-tokyo.ac.jp/EIC/EIC News/index-e.html

• http://neic.usgs.gov/neis/FM/previous mom.html

Seismic moments are the most modern and accurate quantiﬁcation of the size of an

earthquake; however, historically, magnitude scales were used for this purpose. Most

magnitude scales were deﬁned by the observed amplitude of seismic waves with some

corrections for attenuation with distance from the source, but these magnitudes are empirical

parameters and cannot be directly related to any speciﬁc physical parameter of the source.

Recently, the standard practice is to deﬁne the magnitude with the seismic moment. This

magnitude, M

w

, is deﬁned by the following relation:

M

w

=

log

10

M

0

1.5

−6.07 (M

0

in Nm). (3.7)

As mentioned above, M

0

is a static parameter and does not represent any dynamic properties

of the source. However, with the use of some scaling relations, it can be approximately related

to the total radiated energy, at least for large earthquakes (section 3.2.2). In this sense, M

0

or M

w

can be used as a useful quantiﬁcation parameter for an earthquake and its damaging

effects.

3.1.3. Strain and stress drop. As we discussed above, the stress drop caused by an earthquake

is σ(¯ r) = σ

0

(¯ r) −σ

1

(¯ r). We usually consider only the shear stress on the fault plane,

σ

s

= σ

0

−σ

1

(3.8)

The physics of earthquakes 1447

and call it the static stress drop associated with an earthquake. The strain drop ε

s

is given by

ε

s

= σ

s

/µ. In general, σ

s

varies spatially on the fault. The spatial average is given by

σ

s

=

1

S

S

σ

s

dS. (3.9)

Since the stress and strength distributions near a fault are non-uniform, the slip and stress

drop are, in general, complex functions of space. In most applications, we use the stress

drop averaged over the entire fault plane. The stress drop can be locally much higher than the

average. To be exact, the average stress drop is the spatial average of the stress drop, as given by

(3.9). However, the limited resolution of seismological methods often allows determinations

of only the average displacement over the fault plane, which in turn is used to compute the

average stress drop. With this approximation, we estimate σ

s

simply by

σ

s

≈ Cµ

¨

D

˜

L

, (3.10)

where,

¨

D is the average slip (offset),

˜

L is a characteristic rupture dimension, often deﬁned

by

√

S and C is a geometric constant of order unity. Unfortunately, given the limited spatial

resolution of seismic data, we cannot fully assess the validity of this approximation. However,

Madariaga (1977, 1979), Rudnicki and Kanamori (1981) and Das (1988) show that this is a

good approximation unless the variation of stress on the fault is extremely large.

We often use σ

s

to mean the average static stress drop in this sense. Some early

determinations of stress (strain) drops were made using D and

˜

L estimated from geodetic

data (e.g. 1927 Tango earthquake, Tsuboi (1933)).

More commonly, if the seismic moment is determined by either geodetic or seismological

methods, we use the following expression. Using M

0

= µ

¨

DS,

˜

L = S

1/2

and (3.10), we can

write

σ

s

= CM

0

˜

L

−3

= CM

0

S

−3/2

. (3.11)

If the length scale of the source is estimated from the geodetic data, aftershock area, tsunami

source area or other data, we can estimate the stress drop using (3.11) (e.g. Kanamori and

Anderson (1975), Abercrombie and Leary (1993)).

If the slip distribution on the fault plane can be determined from high-resolution seismic

data, it is possible to estimate the stress drop on the fault plane (Bouchon 1997).

Since σ

s

≈ CM

0

˜

L

−3

, an uncertainty in the length scale can cause a large uncertainty in

σ

s

: a factor of 2 uncertainty in

˜

Lresults in a factor of 8 uncertainty in σ

s

. Thus, an accurate

determination of earthquake source size, either S or

˜

L, is extremely important in determining

the stress drop.

3.1.4. Energy

Radiated energy, E

R

. The energy radiated by seismic waves, E

R

, is another important

physical parameter of an earthquake. In principle, if we can determine the wave-ﬁeld

completely, it is straightforward to estimate the radiated energy. For example, if the P-wave

displacement in a homogeneous medium is given by u

r

(r, t ), then the energy radiated in a

P-wave is given by

E

R,α

= ρα

S

0

+∞

−∞

˙ u

r

(r, t )

2

dt dS

0

, (3.12)

where S

0

is a spherical surface at a large distance surrounding the source. Similarly, the energy

radiated in an S-wave is given by

E

R,β

= ρβ

S

0

+∞

−∞

[ ˙ u

θ

(r, t )

2

+ ˙ u

φ

(r, t )

2

] dt dS

0

. (3.13)

1448 H Kanamori and E E Brodsky

Table 2. Determinations of radiated energy with different data sets and methods.

Data E

R

(J) Reference

Bhuji, India, Earthquake, 26 January 2001, M

w

= 7.6

Regional data 2.1 10

16

Singh et al (2004)

Teleseismic data 2.0 10

16

Venkataraman and Kanamori (2004)

Frequency-domain method 1.9 10

16

Singh et al (2004)

Hector Mine, California, Earthquake, 16 October 1999, M

w

= 7.1

Regional data 3.4 10

15

Boatwright et al (2002)

3 10

15

Venkataraman et al (2002)

Teleseismic data 3.2 10

15

Boatwright et al (2002)

2 10

15

Venkataraman et al (2002)

The total energy, E

R

, is the sum of E

R,α

and E

R,β

(e.g. Haskell (1964)). In practice, however,

the wave-ﬁeld in the Earth is extremely complex because of the complexity of the seismic

source, propagation effects, attenuation and scattering. Extensive efforts have been made in

recent years to accurately determine E

R

. For earthquakes for which high-quality seismic data

are available, E

R

can be estimated probably within a factor of 2–3 (McGarr and Fletcher 2002).

Some examples are shown in table 2.

Potential energy. The potential energy change in the crust due to an earthquake is

W =

1

2

(σ

0

+ σ

1

)DS, (3.14)

where the bar stands for the spatial average (Kostrov 1974, Dahlen 1977). Equation (3.14) can

be rewritten as,

W =

1

2

(σ

0

−σ

1

)DS + σ

1

DS =

1

2

σ

s

DS + σ

1

DS = W

0

+ σ

1

DS, (3.15)

where

W

0

=

1

2

σ

s

DS. (3.16)

Two difﬁculties are encountered. First, with seismological measurements alone, the absolute

value of the stresses, σ

0

and σ

1

cannot be determined. Only the difference σ

s

= σ

0

− σ

1

is

determined. Thus, we cannot compute W fromseismic data. As we discussed in section 2.2,

non-seismological methods give inconsistent results for background stress. Second, as we

discussed in section 3.1.3 for the stress drop, with the limited resolution of seismological

methods, the details of spatial variation of stress and displacement cannot be determined.

Thus we commonly use, instead of (3.16),

W

0

=

1

2

σ

s

DS. (3.17)

Unfortunately, it is not possible to accurately assess the errors associated with the

approximation of equation (3.17). It is a common practice to assume that the approximation

is sufﬁciently accurate if the spatial variation is not very rapid.

Although W cannot be determined by seismological methods, W

0

can be computed

from the seismologically determined parameters, σ

s

, D and S. In general σ

1

DS > 0, unless

a large scale overshoot occurs, and W

0

can be used as a lower bound of W. If the residual

stress σ

1

is small, W

0

is a good approximation ofW.

It is important to note that we can determine two kinds of energies, the radiated energy

(E

R

) and the lower bound of the potential energy change (W

0

), with seismological data

and methods. These two energies play an important role in understanding the physics of

earthquakes (section 4.4).

The physics of earthquakes 1449

3.1.5. Rupture mode, speed and directivity. Another observable feature of earthquakes is the

rupture pattern on the fault. Although the rupture pattern is not a parameter sensu stricto, since

it is not a single summaryquantity, it is another observable characterizationof the rupture. From

the rupture patterns we can deﬁne some secondary parameters describing rupture propagation

velocity, slip duration and directivity.

An earthquake occurs on a ﬁnite fault. It initiates from a point, called a hypocentre, and

propagates outward on the fault plane. From the gross rupture patterns, we classify the rupture

patterns into unilateral rupture, bilateral rupture and two-dimensional (approximately circular)

rupture patterns. In a unilateral rupture, the hypocentre is at the one end of the fault, and the

rupture propagates primarily in one direction toward the other end. One good example is the

recent Denali, Alaska, earthquake (M

w

= 7.9, 3 November 2002). In a bilateral rupture,

the rupture propagates in opposite directions from the hypocentre. Good examples include the

1989 Loma Prieta, California, earthquake (M

w

= 6.9), and the 1995 Kobe, Japan, earthquake

(M

w

= 6.9). Bilateral ruptures are not necessarily symmetric. The 1906 San Francisco,

California, earthquake is believed to have ruptured in both directions, but propagated further

to the north than the south. In the description of unilateral and the bilateral fault, the fault

geometry is assumed to be one-dimensional. In some earthquakes, the rupture propagates in

all directions on the fault plane. In these cases, a circular fault is often used to model the fault

plane.

Directivity and source duration. The source ﬁniteness and rupture propagation have an

important effect on seismic radiation. This effect, called directivity, is similar to the Doppler

effect.

As we discussed in section 3.1.1, the far-ﬁeld displacement is given by the time derivative

of the near-ﬁeld displacement and is pulse-like. However, because of the Doppler effect,

the observer located in the direction of the rupture will see a shorter pulse than the observer

in the direction away from the rupture direction. However, the area under the pulse-like

waveform (i.e. the displacement integrated over time) is constant, regardless of the azimuth

and is proportional to the seismic moment, M

0

. The duration of the pulse, ¨ τ, when averaged

over the azimuth, is proportional tothe lengthscale of the fault

˜

Ldividedbythe rupture speed, V

¨ τ =

˜

L

V

. (3.18)

For unilateral, bilateral and circular faults,

˜

L is commonly taken to represent the fault length,

half the fault length and the radius of the fault plane, in that order. The variation of the pulse

width as a function of azimuth due to rupture propagation has an important inﬂuence on ground

motions. As mentioned above, at a site towards the rupture propagation, the pulse becomes

larger and narrower, and produces stronger ground motions which often result in heavier shak-

ing damage, than at a site away from the rupture propagation. A good example is the 1995

Kobe, Japan, earthquake. The rupture of one of the bilateral segments propagated northeast

from the hypocentre toward Kobe and produced very strong ground motions in the city.

Rupture speed. As we will discuss later (section 4.6.1), the rupture speed is an important

parameter which reﬂects the dynamic characteristic of a fracture. In particular, the fraction of

the shear velocity that the shear crack rupture velocity achieves is related to the fracture energy.

Thus it is important to determine the rupture speed of earthquake faulting to understand the

nature of earthquake mechanics.

The rupture speed V has been determined for many large earthquakes. In general, for most

large shallow earthquakes, V is approximately 75–95% of the S-wave velocity at the depth

where the largest slip occurred. However, there are some exceptions. For some earthquakes

1450 H Kanamori and E E Brodsky

Figure 9. Complexity of earthquake rupture pattern. Rupture pattern for the 1992 Landers,

California, earthquake determined with inversion of seismic data (Wald and Heaton 1994).

(e.g. 2001 Kunlun, China, earthquake), super-shear rupture velocities, i.e. V > β, have been

reported (Bouchon and Vallee 2003). For some earthquakes (e.g. 1992 Nicaragua earthquake,

Kikuchi and Kanamori (1995)), a very slow rupture speed has been reported. For deep

earthquakes, an accurate determination of V is usually difﬁcult, because of the difﬁculty

in resolving the rupture pattern due to the lack of close-in observations. For the largest deep

earthquake, the 1994 Bolivian earthquake (M

w

= 8.3), the resolution of the seismic method

was good enough to determine V; a very low, (V/β) = 0.2 (e.g. Kikuchi and Kanamori

(1994)), rupture speed has been reported. For other smaller deep earthquakes, higher rupture

speeds have been reported (e.g. Tibi et al (2003)).

The relatively high rupture speeds observed for most shallow earthquakes is in striking

contrast with the rupture speeds observed in laboratory. Most of the laboratory data show

that the rupture speed for intact materials under tensile stress is at most 50% of the Rayleigh

wave speed. It is not possible to maintain a shear fault in intact materials, because the rupture

bifurcates and cannot produce a planar faulting. Higher rupture speeds have only been observed

for pre-cut samples. In a few pre-cut experiments the rupture velocities are even higher than

shear wave speed (Rosakis 2002).

The difference and similarity between earthquakes and laboratory fractures provide an

important clue to the mechanics of earthquake rupture, as we will discuss in section 4.6.1.

3.1.6. Earthquake rupture pattern. The slip distribution in real earthquakes is very complex.

With the advent of modern strong-motion seismographs and broad-band seismographs, it has

become possible to determine the actual slip distribution by inverting the observed seismic

waveforms. These studies demonstrate that the slip distribution on a fault plane is highly

heterogeneous in space and time, as shown for the 1992 Landers, California, earthquake

(ﬁgure 9, Wald and Heaton (1994)). However, in most modelling studies, short-period (usually

2 s or shorter) waves are ﬁltered out because of the difﬁculty in modelling such short-period

waves. At periods shorter than2 s (the correspondingwavelengthis about λ = 5 km), scattering

of waves and complexities of the source process produce wave forms too complex to be

explained with a simple model. Thus, these models should be regarded as long-wavelength

rupture patterns; the real slip distribution is probably far more complex with short wave-

length irregularities. Although the spatial resolution of these models are not always given,

it is probably of the order of λ/3. Short wavelength heterogeneity has been suggested by

The physics of earthquakes 1451

complex high-frequency wave forms seen on accelerograms recorded at short distances. Zeng

et al (1994) modelled an earthquake fault as a fractal distribution of patches. This complexity

suggests that the microscopic processes on a fault plane can be important in controlling the

rupture dynamics, as we will discuss in section 4.5.

3.2. Seismic scaling relations

Now that we have measured some average properties of rupture, we need to relate the

parameters to each other. The scaling relationships between the macroscopic source parameters

are useful for isolating general constraints on the microscopic forces and processes in the fault

zone during rupture. We will ﬁrst discuss a selection of the observed scalings with only a

cursory overview of the implications. A more detailed discussion of microscopic physics

follows in section 4.

3.2.1. Scaling relations for static parameters. The seismic moment, M

0

, is the source

parameter that can be determined most reliably. Thus, it is useful to investigate a scaling

relation between M

0

and another parameter that can be determined most directly from seismic

observations.

(1) M

0

versus source duration. We ﬁrst choose the duration of source process, τ. This

parameter can be determined from a seismogram, but it is not just the duration of a waveform

recorded on a seismogram. We must remove the propagation effects from the seismograms to

estimate the duration of rupture process at the source. τ is equal to the azimuthal average of

¨ τ discussed in section 3.1.5, equation (3.18). The existing data (Masayuki Kikuchi, written

communication (2001)) show a gross scaling relation

M

0

∝ τ

3

, (3.19)

as shown in ﬁgure 10.

(2) Moment versus fault area. It is not always easy to determine the fault area S (i.e. rupture

area), but by combining various kinds of data (e.g. aftershock area, surface rupture, geodetic

data, directivity and seismic inversion results), the rupture areas for large (M

w

> 6) earthquakes

have been estimated. Figure 11 shows the results, and suggests a scaling relation

M

0

∝ S

3/2

. (3.20)

The scaling relation given by (3.20) can be interpreted as follows. From equation (3.11),

the seismic moment M

0

, σ

s

, and the length scale S

1/2

are related by

M

0

=

1

C

σ

s

S

3/2

. (3.21)

Hereafter σ

s

is simply written as σ

s

for brevity. If σ

s

is constant, then M

0

∝ S

3/2

, which

is the scaling relation shown in ﬁgure 11. Thus, this scaling relation indicates that σ

s

is

roughly constant over a range of M

0

from 10

18

to 10

23

Nm. The level of the curve determines

the value of σ

s

. From ﬁgure 11 we can estimate that σ

s

is, on the average, approximately

3 MPa with a range 1–10 MPa. Because of the uncertainty in S, and the assumption for the

geometry of the fault plane, this estimate of σ

s

is not precise, but the approximate range

1–10 MPa is considered robust. There is some evidence that σ

s

varies for different types of

earthquakes, such as those on major plate boundaries and those in plate interiors, but the overall

difference is probably within this range. Figure 11 is the evidence that most earthquakes have

comparable stress drops in the range 1–10 MPa .

1452 H Kanamori and E E Brodsky

10

-2

10

-1

10

0

10

1

10

2

10

3

10

11

10

13

10

15

10

17

10

19

10

21

D

u

r

a

t

i

o

n

(

τ

)

,

s

M

0

, N m

M

w

8 6 4 2

Figure 10. The relation between the seismic moment M

0

(the corresponding M

w

is shown at the

top of the ﬁgure) and the source duration τ for shallow (depth <60 km) earthquakes. The data

are for events during the period from 1991 to 2001. The data for events larger than M

w

= 6.5

(M

0

= 710

18

Nm) are for the events worldwide. The data for smaller events in Japan are added.

(Masayuki Kikuchi, written communication (2001)). The horizontal and vertical alignments of the

data points are an artefact due to the rounding-off of the values used for M

0

and τ.

Comparison of this scaling relation, M

0

∝ S

3/2

, with the scaling relation, M

0

∝ τ

3

(3.19) suggests that τ ∝ S

1/2

. We deﬁne the length scale of the fault to be

˜

L ≡ S

1/2

. Since

˜

L ≈ Vτ (3.18) where V is the rupture speed, this means that V is constant for most shallow

earthquakes. As mentioned earlier, for some shallow large earthquakes, the rupture speed V is

directly determined to be 75–95%of the S-velocity. Thus, these results, taken together, suggest

that most large shallowearthquakes have σ

s

ranging from 1 to 10 MPa, and the rupture speed

V is roughly constant at 75–95% of the S-velocity. We note here that these are the general

scaling relations, and there are exceptions.

A similar analysis can be made for smaller earthquakes. However, it is difﬁcult to

determine the source dimension of small (e.g. M

w

< 3) earthquakes directly. In most cases,

the pulse width or the corner-frequency of the source spectrum is used to infer the source

dimension. (As discussed in section 3.1.5, the pulse width is, on the average, equal to L/V.

The physics of earthquakes 1453

6 7 8 9

10

1

10

2

10

3

10

4

10

5

10

6

10

-3

10

-2

10

-1

10

0

10

1

10

2

10

3

10

4

S

,

k

m

2

M

0

, 10

20

N m

M

w

Figure 11. The scaling relation between seismic moment M

0

(the corresponding M

w

is shown at

the top of the ﬁgure) and the fault area S for shallow earthquakes. The data are from Kanamori

and Anderson (1975) and Masayuki Kikuchi (written communication 2001). The unpublished data

provided by Kikuchi are for the events worldwide during the period 1991 to 2001 for which the

source dimension could be estimated.

If the rupture speed, V, is approximately equal to the S-velocity, the pulse width can be used to

estimate the fault length. The corner frequency of the spectrum of a pulse-like source function

is proportional to the reciprocal of the pulse width.) The general trend follows the M

0

∝ S

3/2

scaling, with σ

s

ranging from 0.1 to 100 MPa (Abercrombie 1995). This large range in σ

s

may be real, reﬂecting the heterogeneities of the crust on short length scales. It is also possible

that the large scatter is due to errors in determining the source dimension. At present, this

question is not resolved.

In the scaling relations discussed earlier, the length scale of the source is deﬁned by S

1/2

with the idea of representing the source dimension with just one parameter. However, different

faults have different aspect ratios (i.e. the ratio of fault length to width). For example, for long

crustal strike-slip earthquakes such as the 1906 San Francisco earthquake, the fault length

is about 350 km, but the depth-wise width of the fault is probably comparable to the upper

half of the crust, about 15 km. In contrast, large subduction-zone earthquakes like the 1964

1454 H Kanamori and E E Brodsky

Alaskan earthquake have a fault width as large as 200 km or more. In view of this variation

in aspect ratio, several investigators tried to investigate the scaling relation between M

0

and L

(e.g. Romanowicz and Rundle (1993), Scholz (1994), Romanowicz and Ruff (2002)). Several

different scaling relations, such as M

0

∝ L

3/2

and M

0

∝ L

2

, have been proposed for different

types of earthquakes and for different magnitude ranges.

3.2.2. Scaling relations for dynamic parameters. The radiated energy, E

R

, is another

macroscopic earthquake source parameter that can be determined by seismological methods

(see section 3.1.4). The ratio

˜ e =

E

R

M

0

=

1

µ

E

R

¨

DS

(3.22)

has long been used in seismology as a useful parameter that characterizes the dynamic

properties of an earthquake (Aki 1966, Wyss and Brune 1968). The ratio ˜ e multiplied by the

rigidity µis called the apparent stress. From(3.22), the ratio can be interpreted as proportional

to the energy radiated per unit fault area and per unit slip. In this sense, this scaling relation

represents a dynamic property of earthquakes. As we will discuss later (section 4.4), if the static

stress drop is constant, then ˜ e must be constant if small and large earthquakes are dynamically

similar. Seismologists are very concerned with whether or not earthquakes are dynamically

similar because of the implications of the observation for the predictability of the eventual size

of an earthquake. If small and large earthquakes are dynamically similar, then the initiation

process is scale-invariant and therefore the size of earthquakes is inherently unpredictable.

However, the converse statement is not true, so the observation of a lack of similarity cannot

prove the predictability of earthquake size.

In view of its importance for understanding the dynamic character of earthquakes, many

studies have been devoted to the determination of ˜ e. Unfortunately, it is difﬁcult to determine

E

R

accurately, because of the complex wave propagation effects in the Earth, especially for

small earthquakes, and the results were widely scattered.

Recent improvements in data quality and methodology have signiﬁcantly improved the

accuracy of E

R

determination for large earthquakes (e.g. M

w

> 6) (e.g. Boatwright and

Choy (1986), Boatwright et al (2002), Venkataraman et al (2002)). For small earthquakes, it is

still difﬁcult because the relatively high-frequency seismic waves excited by small earthquakes

are easily scattered and attenuated by the complex rock structures between the fault and a

seismic station. Nevertheless, using down-hole instruments, or with the careful removal of

path effects, large amounts of high-quality data for small earthquakes have been accumulated

(Abercrombie 1995, Mayeda and Walter 1996, Izutani and Kanamori 2001, Prejean and

Ellsworth 2001, Kinoshita and Ohike 2002). The ratio depends on many seismogenic

properties of the source region so that it varies signiﬁcantly for earthquakes in different

tectonic environments, suchas continental crust, subductionzone, deepseismic zone, etc (Choy

and Boatwright 1995, Perez-Campos and Beroza 2001, Venkataraman and Kanamori 2004).

However, the data for the same type of earthquakes exhibit an interesting trend. Figure 12

shows the results for crustal earthquakes in California and Japan.

Taken at face value, despite the large scatter, the average ratio ˜ e decreases as the magnitude,

M

w

, decreases. For large earthquakes (M

w

≈ 7), ˜ e is, on the average, approximately 510

−5

,

but it is approximately a factor of 10 smaller at M

w

≈ 3, and a factor of 100 smaller at

M

w

≈ 1. Results for even smaller earthquakes show even smaller values of ˜ e (e.g. Jost et al

(1998), Richardson and Jordan (2002)). Ide and Beroza (2001) suggested that many of the

published ˜ e versus M

w

relations could be biased to have decreased ˜ e for small events because of

inadequate corrections for path effects or the limited instrumental pass-band. These systematic

The physics of earthquakes 1455

10

-8

10

-7

10

-6

10

-5

10

-4

10

-3

0 1 2 3 4 5 6 7 8

Mayeda and Walter [1996]

Abercrombie [1995]

TERRAscope

Izutani and Kanamori [2001]

Large Eqs

E

R

/

M

0

M

w

Figure 12. The relation between ˜ e = E

R

/M

0

and M

w

(Abercrombie 1995, Mayeda and Walter

1996, Izutani and Kanamori 2001, Kanamori et al (1993), for TERRAscope data).

measurement errors could mean that the real ˜ e is scale-independent. At present, this question

remains unresolved. If future research ﬁnds that ˜ e varies as suggested by ﬁgure 12, then the

observation would imply that large and small earthquakes are dynamically different.

4. Rupture processes

We have now provided an overview of the stresses that generate earthquakes along with a

discussion of the measurable parameters and their interrelationships. The next step in our

inquiry into why earthquakes happen is to examine the rupture process itself.

4.1. Fracture mechanics

To interpret seismological data, crack models are often used in part because the theories on

cracks have been developed well. On the other hand, seismic faulting may be more intuitively

viewed as sliding on a frictional surface (fault) where the physics of friction, especially stick

slip, plays a key role. Seismic faulting in the Earth can be complex and we may require a

mixture of crack models and sliding models, or even other models to interpret it. Despite this

complexity, crack models and frictional sliding models provide a useful framework for the

interpretation of earthquake processes. Here, we limit our discussion to the very basic aspects

of these models.

4.1.1. An overview of the crack model. In crack mechanics, three types of crack geometries,

Mode I (tensile), Mode II (longitudinal shear) and Mode III (transverse shear), are used

1456 H Kanamori and E E Brodsky

z

x

y

Figure 13. Modes I, II and III cracks (from Lawn (1993)). The x, y and z axes indicate the

coordinate system used in ﬁgure 14.

Figure 14. Stress ﬁeld for a Mode III crack before and after crack formation. Dashed line is the

initial stress and solid line is the ﬁnal stress. The inset shows the Mode III crack geometry.

(ﬁgure 13). Although the difference between these models is important, for many problems

in crack mechanics, here we mainly use the Mode III crack for illustration purposes.

Figure 14 shows the stress distribution along the plane of a Mode III crack before (dashed)

and after (heavy curves) crack formation. The crack extends from z = −∞to +∞as shown

in the inset. After the crack is formed, the shear stress becomes inﬁnitely large just beyond

the crack tip, and drops to the frictional stress σ

f

on the crack surface.

For the coordinate system shown, the displacement w and the stress σ

zy

are

w =

σ

0

−σ

f

µ

(a

2

−x

2

)

1/2

x a (4.1)

and

σ

zy

= (σ

0

−σ

f

)

x

(x

2

−a

2

)

1/2

+ σ

f

x a (4.2)

(Knopoff 1958). At a small distance ε from the crack tip, x = a + ε, σ

zy

is proportional to

1/

√

ε. Speciﬁcally, the relationship is

σ

zy

=

K

√

2π

1

√

ε

+ σ

f

,

where K is the stress intensity factor deﬁned by

K ≡

√

πa(σ

0

−σ

f

). (4.3)

More detailed expressions for the stress intensity factors for Modes I, II and III cracks are

given in Rice (1980), Dmowska and Rice (1986) and Li (1987).

The strain energy release per unit length in z direction is (equations (3.14)–(3.16))

W =

(σ

0

+ σ

f

)DS

2

= W

0

+ σ

f

DS, (4.4)

The physics of earthquakes 1457

where

D = 2 ¨ w =

σ

0

−σ

f

µ

1

a

a

−a

(a

2

−x

2

)

1/2

dx =

πa

2

σ

0

−σ

f

µ

(4.5)

and

W

0

=

(σ

0

−σ

f

)DS

2

=

πa

2

(σ

0

−σ

f

)

2

2µ

(4.6)

and S = 2a and D is the average offset across the crack. In (4.4), the second term on the

right-hand side (rhs) is the frictional energy and the ﬁrst term, W

0

, is the portion of the strain

energy change that is not dissipated in the frictional process.

4.1.2. Crack tip breakdown-zone. The model discussed earlier is for the static case and it

provides the basic physics of dynamic crack propagation. If the stress just beyond the crack tip

becomes large enough to break the material, the crack grows. In the dynamic crack propagation

problem, the theory becomes complex because of the complex stress ﬁeld near the crack tip

and the strain energy ﬂux into the crack tip. Here, we discuss this problem using a simple

model. More rigorous and detailed discussions are by Freund (1989) and Lawn (1993).

In the simple model described in ﬁgure 14 (called the linear elastic fracture model

(LEFM)), the stress near the crack tip becomes indeﬁnitely large (solid curve in ﬁgure 14

(inset)). In the real material this does not occur. Instead, inelastic (e.g. plastic) yielding

occurs, and the stress becomes ﬁnite as shown by the broken curve in ﬁgure 15(a). The ﬁnite

stress at the crack tip, σ

Y

, is called the yield stress. Because of this breakdown process, the

stress just inside the crack does not drop to the constant frictional level σ

f

abruptly. Instead it

decreases gradually to σ

f

over a distance l

0

as shown by the broken curve in ﬁgure 15(a). Also,

slip, D, inside the crack increases gradually to the value, D

0

, expected for the case without

inelastic breakdown (i.e. LEFM), as shown in ﬁgure 15(b).

At a point just beyond the crack tip, the stress and slip vary as a function of time as shown

in ﬁgures 15(c) and (d), respectively. Figure 15(e) shows the shear stress σ

yz

at this point

as a function of slip D, as the crack tip passes by. The stress drops from σ

Y

to the constant

frictional stress σ

f

over a slip D

0

. This behaviour in which the stress on the fault plane

decreases as slip increases is called slip-weakening behaviour, and this model is often referred

to as the breakdown-zone slip-weakening model. (For the development of the concept, see

Dugdale (1960), Barenblatt (1962), Palmer and Rice (1973), Ida (1972), and for more detailed

discussions, see Rice (1980), Li (1987).)

4.1.3. Stability and growth of a crack. Now that we have an overview of crack physics, we

can consider the stability of a crack and its growth. The theory is based on Grifﬁth’s (1920)

concept which was initially developed for tensional cracks (Mode I). Here, we use the basic

concept, and apply it to seismological problems. More details are given by Lawn (1993).

Consider a Mode III crack with half-length, a, as discussed earlier. When a crack with

half-length a is inserted in a homogeneous medium under uniform shear stress σ

0

, the strain

energy is released. After subtracting the energy dissipated in friction, we obtain the energy

given by (4.6)

W

0

=

(σ

0

−σ

f

)DS

2

=

πa

2

(σ

0

−σ

f

)

2

2µ

, (4.7)

which is available for mechanical work for crack extension.

1458 H Kanamori and E E Brodsky

Figure 15. Breakdown-zone interpretation of slip-weakening process. (a) The stress ﬁeld near the

crack tip as a function of position. (b) The slip on the crack surface. (c) The temporal variation of

stress at a point which is initially just beyond the crack tip. The time is measured from the time

when the crack tip reached the point. (d) The temporal variation of the slip at a point which is

initially just beyond the crack tip. The time is measured fromthe instant when the crack tip reached

the point. (e) The shear stress at point which is initially just beyond the crack tip. As the crack tip

extends past this point, the stress drops from σ

Y

to σ

f

gradually. In all ﬁgures, the solid curves are

for the LEFM model. The broken curves indicate the deviation from the LEFM case when yielding

occurs.

Now consider a virtual extension of crack by δa. Then the strain energy that would be

released due to the virtual extension δa is, from (4.7)

δ(W

0

) =

∂W

0

∂a

δa =

πa(σ

0

−σ

f

)

2

µ

δa = 2G

∗

δa, (4.8)

where

G

∗

=

πa(σ

0

−σ

f

)

2

2µ

=

K

2

2µ

, (4.9)

where K is the stress intensity factor deﬁned by (4.3). G

∗

is called the static energy release

rate or crack extension force. The name is a little confusing because here ‘rate’ means per unit

area rather than unit time. The unit of G

∗

is energy per area. The factor 2 on the rhs of (4.8)

arises because the crack extends at both ends. Note that G

∗

is not a constant, but increases as

the crack size increases.

The physics of earthquakes 1459

Static crack. In case of a static (or quasi-static) crack, for the crack to be stable at half-

length a, this energy must be equal to twice the surface energy of the material near the crack

tip. That is,

G

∗

= G

∗

c

≡ 2γ, (4.10)

where γ is the surface energy per unit area which is necessary to create a new crack surface.

The factor 2 in (4.10) arises because surface energy is deﬁned for each side of the crack. If G

∗

given by (4.9) is larger than G

∗

c

, the crack will grow. G

∗

c

is called the critical speciﬁc fracture

energy. The stress intensity factor at this state, K

c

, is called the fracture toughness (or critical

stress intensity factor), which is related to G

∗

c

by equation (4.9), i.e.

G

∗

c

=

K

2

c

2µ

. (4.11)

Thus, the stability of a crack can be discussed either in term of the critical speciﬁc fracture

energy, G

∗

c

, or the critical stress intensity factor, K

c

.

In seismic faulting, we often generalize γ to include more surface area (e.g. damaged

zones) than just the normal area of crack extension, as is done in the Grifﬁth theory.

K and G

∗

are the important parameters in crack theory. The expressions for K are

independent of the mode of crack. The expression for G

∗

is the same for Modes I and II, but is

slightly different for Mode III, but considering the gross approximations used in seismological

applications, the differences are not important.

Dynamic crack. When G

∗

> G

∗

c

the crack propagates dynamically and some energy is

radiated out of the system as seismic waves. The total energy available for work from

equation (4.7) is divided between the virtual crack extension and the radiated energy. If

we denote the radiated energy by E

R

, the energy equation for virtual crack extension is no

longer given by (4.8). Instead,

δ(W

0

) −δ(E

R

) = 2Gδa, (4.12)

from which

G = G

∗

−

1

2

∂E

R

∂a

, (4.13)

where G is called the dynamic energy release rate. Then, the crack extension is governed by

G = 2γ (4.14)

instead of (4.10).

Rupture speed. The ratio of the dynamic energy release rate, G, to the static energy release

rate, G

∗

, is given by a function of rupture speed V = da/dt . Kostrov (1966), Eshelby (1969)

and Freund (1972) showed that the energy release rate, G, for a crack growing at a rupture

speed V is given approximately by

G = G

∗

g(V), (4.15)

where g(V) is a universal function of V.

For a Mode I (tensile) crack (Freund 1972)

g(V) = 1 −

V

c

R

, (4.16)

where c

R

is the Rayleigh-wave speed.

1460 H Kanamori and E E Brodsky

For a Mode II (longitudinal shear) crack (Fossum and Freund 1975)

g(V) =

1 −V/c

R

√

1 −V/β

, (4.17)

where β is the shear-wave speed.

For a Mode III (transverse shear) crack (Kostrov 1966, Eshelby 1969).

g(V) =

1 −(V/β)

1 + (V/β)

. (4.18)

The derivation of g(V) given above is actually complicated, but the classical Mott’s (1948)

theory is useful for understanding the basic physics. In Mott’s theory, the radiated energy E

R

is equated to the kinetic energy in the medium during rupture propagation and scales as

E

R

∝ a

2

˙ u

2

∝ a

2

∂u

∂a

2

˙ a

2

∝ a

2

∂u

∂a

2

V

2

,

where u is the displacement. Because the strain energy, W

0

, also scales as a

2

(4.7), E

R

can

be written as

E

R

=

1

B

2

V

β

2

W

0

, (4.19)

where B is a constant of the order of 1 (Lawn 1993, chapter 4; Mott 1948, Marder and

Fineberg 1996). Then, including the kinetic energy, the equation corresponding to (4.12) can

be written as,

δ(W

0

) −δE

R

=

¸

1 −

1

B

2

V

β

2

δ(W

0

) = 2Gδa (4.20)

from which

G = G

∗

g(V), (4.21)

where

g(V) = 1 −

V

2

(Bβ)

2

. (4.22)

Equation (4.22) has a similar form to relativistic contraction as used to calculate the

electromagnetic ﬁeld around a particle as it approaches the speed of light. Like the

electromagnetic case, acoustic waves also experience a relativistic effect because information

can only propagate through a ﬁnite distance in a ﬁnite time.

The equation for dynamic crack extension is given from (4.14) as

G

∗

g(V) = 2γ. (4.23)

In the limit of the rupture speed approaching the shear sound speed β, no energy is

dissipated mechanically for a Mode III crack and all the energy is radiated in elastic waves.

Modes I and II cracks display the same phenomenon at different limiting velocities. In the

limit of very small rupture speed, the relativistic contraction is irrelevant and g(V) approaches

unity.

4.2. Frictional sliding

Fault rupture can also be modelled as a frictional process. As two surfaces slide past each other

along a pre-existing fault, the dynamics can be dominated by the surface forces between the

two sides. Below, we quantify this frictional interaction in a way that is parallel to the crack

theory so that we can combine the two formulations in section 4.3.

The physics of earthquakes 1461

(a) (b)

Figure 16. Static and kinetic friction. (a) The coefﬁcient of friction. In the simple model, µ drops

fromµ

s

to µ

k

instantly, but in general, it drops to µ

k

after a slip D

c

. (b) The frictional stress. D

c

is

the critical slip. σ

f

is the frictional stress.

4.2.1. Static and kinetic friction. In the classical theory of friction, the coefﬁcient of static

friction µ

s

and the coefﬁcient of kinetic friction µ

k

are the most fundamental parameters

(note that µ is used for coefﬁcient of friction in this section rather than rigidity). If µ

k

< µ

s

,

instability can occur. In any physical system, static friction cannot drop to kinetic friction

instantly. A slip, D

c

, is required before static friction drops to kinetic friction, and steady

sliding begins (ﬁgure 16). The slip, D

c

, is called the critical slip, and is a key parameter in

frictional sliding models. In this case the friction σ

f

is a function of slip.

The hatched area in ﬁgure 16(b) indicates the extra energy per unit area expended in the

system compared with the case in which the friction instantly drops to the ﬁnal stress, σ

1

. D

c

in the frictional sliding model is often equated to D

0

of the critical slip of the slip-weakening

crack model (cf ﬁgure 15(e))

4.2.2. Rate- and state-dependent friction. The simple behaviour shown above can be

generalized by a rate- and state-dependent friction model. Dieterich and his collaborators

introduced the following friction law from experiments on many different materials (e.g.

Dieterich (1979), Scholz (2002)). According to this law, the coefﬁcient of friction µis given by

µ = µ

/

0

+ Aln

˙

δ + B ln θ, (4.24)

where

˙

δ is the sliding speed, θ is a state variable that accounts for the history of sliding and

µ

/

0

, A and B are constants. Speciﬁcally, θ is governed by the following differential equation:

˙

θ = 1 −

θ

˙

δ

D

c

. (4.25)

Although (4.24) is empirically derived, it still can be related in general terms to simple physics.

The secondtermAln

˙

δ represents a resistance similar toviscositygeneratedbydeformingsmall

irregularities on the sliding surface, or asperities. As they are deformed more quickly, they

have a greater resisting stress. The third term B ln θ describes the chemical adhesion between

surfaces that is assumed to increase with contact time. If

˙

δ = 0, then

˙

θ increases linearly with

time (4.25), i.e. the state variable is simply the time of contact.

1462 H Kanamori and E E Brodsky

Figure 17. Change in friction due to a sudden increase in sliding speed according to the rate- and

state-dependent friction.

Alternative forms of the evolution lawfor the state variable depend explicitly on slip rather

than hold time (e.g. Ruina (1983), Linker and Dieterich (1992)). Recent experimental works

favour the form in (4.25) (Beeler et al 1994).

If

˙

δ = constant, then, from (4.25), θ is given by

θ =

D

c

˙

δ

+

θ

0

−

D

c

˙

δ

exp

−

˙

δ(t −t

0

)

D

c

, (4.26)

where θ

0

is the value of θ at t = t

0

.

For illustration purposes, consider a case in which

˙

δ increases from

˙

δ

1

to

˙

δ

2

stepwise at

t = t

1

. Before t = t

1

we assume that sliding is in steady state at

˙

δ =

˙

δ

1

. Then θ = D

c

/

˙

δ

1

,

and, from (4.24), we obtain µ for t < t

1

as

µ

1

= µ

/

0

+ Aln

˙

δ

1

+ B ln

D

c

˙

δ

1

for t < t

1

. (4.27)

For t t

1

,

θ =

D

c

˙

δ

2

+

D

c

˙

δ

1

−

D

c

˙

δ

2

exp

¸

−

˙

δ

2

(t −t

1

)

D

c

¸

and

µ = µ

1

+ Aln

˙

δ

2

˙

δ

1

+ B ln

¸

˙

δ

1

˙

δ

2

+

1 −

˙

δ

1

˙

δ

2

exp

−

˙

δ

2

D

c

(t −t

1

)

¸

for t t

1

. (4.28)

As t →∞, µ = µ

2

where

µ

2

= µ

1

+ (A −B) ln

˙

δ

2

˙

δ

1

. (4.29)

Figure 17 shows µ as a function of slip.

The cases with (A − B) < 0 and (A − B) > 0 represent velocity weakening (generally

unstable) and velocity strengthening (generally stable), respectively. If (A − B) < 0, then

the friction initially increases, but eventually drops to µ

2

= µ

1

+ (A −B) ln(

˙

δ

2

/

˙

δ

1

) from µ

1

.

The constant D

c

is a scaling parameter which determines the amount of slip over which the

friction drops substantially. For example, let D

/

c

be the slip over which the friction drops by

(A −B) ln(

˙

δ

2

/

˙

δ

1

)/e. D

/

c

is proportional to D

c

but it also depends on the velocity ratio

˙

δ

2

/

˙

δ

1

,

and can be interpreted as the critical slip in the simple friction law shown in ﬁgure 16. For a

small ratio of

˙

δ

2

/

˙

δ

1

, D

/

c

≈ D

c

, but for a very large ratio of

˙

δ

2

/

˙

δ

1

, D

/

c

≈ (10–20)D

c

. Here, we

do not distinguish D

/

c

and D

c

, but if

˙

δ

2

/

˙

δ

1

is very large, D

c

and D

/

c

must be distinguished.

The behaviour shown in ﬁgure 17 is what was observed experimentally for many different

kinds of materials (Dieterich 1979), and is the basis of the rate- and state-dependent friction law.

The physics of earthquakes 1463

4.3. The link between the crack model and the friction model

Crack and frictional sliding models are frequently used in seismology to explain various aspects

of earthquake phenomena. Sometimes the terminology of crack mechanics is used and at other

times the terminology of friction mechanics is used. We need to link these two models to

understand the way these models are used in seismology.

The fracture energy, G

c

, in crack theory is the energy needed to create new crack surfaces

near the crack tip. Thus, the system must expend the threshold fracture energy G

c

before the

crack can extend. In contrast, in frictional sliding model, D

c

, is introduced as a critical slip

before rapid sliding begins at a constant friction. The ﬁnal value of the frictional stress σ

f

is equal to σ

1

. If the frictional stress varies more or less linearly as shown in ﬁgure 16, the

energy spent in the system before this happens can be approximately written as (hatched area

in ﬁgure 16(b))

1

2

(σ

0

−σ

1

)D

c

. (4.30)

Thus, if we are to link a crack model to a friction model, we can equate equation (4.30) to

G

c

, i.e.

G

c

=

1

2

(σ

0

−σ

1

)D

c

. (4.31)

Past the initial breakdown-zone, σ

f

is the same in the crack and friction models.

Direct determination of D

c

. With the recent availability of high-quality seismograms at short

distances, it is now possible to determine a bound on D

c

directly from seismograms (Ide and

Takeo 1997). With inversion of seismic data, the slip at a point on the fault plane, u(t ), can

be determined as a function of time. Then solving the equation of elastodynamics on the fault

plane, we can determine the shear stress, σ(t ) as a function of time. Eliminating t from u(t )

and σ(t ) leads to the slip dependence of stress σ(u) from which D

c

can be estimated. For the

1995 Kobe earthquake, Ide and Takeo (1997) found D

c

to be of the order of 0.5 min the deeper

part of the fault plane. However, in the process of inversion low-pass ﬁltering is applied to the

data, which tends to smooth the resulting σ(u) versus u relationship. Thus, the critical slip,

D

c

, thus determined is an upper bound. Mikumo et al (2003) developed a method to estimate

D

c

directly from slip-velocity records using elastodynamic modelling. With this method, D

c

for large earthquakes is also estimated to be of the order of 1 m.

The values of D

c

determined by laboratory friction experiments (equation (4.28)) are

approximately 5 orders of magnitude less than the upper bound derived from seismic studies.

Therefore, we conclude that either the seismically determined bound on D

c

is so extreme

that comparison with the laboratory values is not meaningful, or the slip-weakening process

at large slips is different from that of laboratory friction process. For example, Marone and

Kilgore (1933) suggest that D

c

is controlled by the thickness of fault gouge layers as well as the

surface roughness. Ohnaka and Shen (1999) proposed a scaling relation between D

c

and the

wavelength of the surface roughness. Alternative slip-weakening processes include thermal

pressurization, hydrodynamic lubrication (as discussed in section 4.5), plastic deformation and

micro-fracturing in the crust surrounding the fault.

4.4. Rupture energy budget

Since the crack and frictional processes are linked through the fracture energy G

c

, we can relate

the macroscopically observable energy budget to the microscopic processes in a surprisingly

general way. Any constraint on fracture energy obtained from the energy budget will provide

a strong bound on all microscopic rupture processes.

1464 H Kanamori and E E Brodsky

Figure 18. Illustration of simple stress release patterns during faulting. (a) ——: simple case of

immediate stress drop. - - - -: general case without slip-weakening. (b) Slip-weakening model:

hatched and cross-hatched areas indicate the fracture energy and frictional energy loss, respectively.

(c) The energy budget: hatched, cross-hatched and dotted areas indicate the fracture energy, thermal

(frictional) energy and radiated energy, in that order. All the ﬁgures are shown for unit area of the

fault plane.

An earthquake is viewed as a stress release process on a fault surface S . The solid lines

in ﬁgure 18(a) show the simplest case. At the initiation of an earthquake, the initial (before

an earthquake) shear stress on the fault plane σ

0

drops to a constant dynamic friction σ

f

, and

stays there, i.e. σ

f

= σ

1

. If the condition for instability is satisﬁed (Brace and Byerlee (1966),

Scholz (2002), also section 6.1.1), rapid fault slip motion begins and eventually stops. At the

end, the stress on the fault plane is σ

1

(ﬁnal stress) and the average slip (offset) is D. The

difference σ

s

= σ

0

− σ

1

is the static stress drop. During this process, the potential energy

(strain energy plus gravitational energy) of the system, W

0

, drops to W

1

= W

0

−W where

W is the strain energy drop, and the seismic wave is radiated carrying an energy E

R

. From

equation (3.14),

W = ¨ σDS, (4.32)

where ¨ σ = (σ

0

+ σ

1

)/2 is the average stress during faulting (section 3.1.4). Graphically, W

(for unit area) is given by the trapezoidal area shown in ﬁgure 18(c).

The variation of stress during faulting can be more complex than shown by the solid lines

in ﬁgure 18(a). For example, the stress may increase to the yield stress σ

Y

in the beginning of

the slip motion (curve (1) in ﬁgure 18(a)) because of loading caused by the advancing rupture

(ﬁgure 15(e)), or of a speciﬁc friction law such as the rate- and state-dependent friction law

(Dieterich 1979) (ﬁgure 17). In fact, some seismological inversion studies have shown this

The physics of earthquakes 1465

increase (Quin 1990, Miyatake 1992, Mikumo and Miyatake 1993, Beroza and Mikumo 1996,

Bouchon 1997, Ide and Takeo 1997). The stress difference, σ

Y

− σ

0

, is called the strength

excess. However, the amount of slip during this high stress stage is small so that little energy

is involved. Thus, we will not include it in our simple energy budget.

Also, the friction may not be constant during faulting. For instance, the friction may

drop drastically in the beginning and later resumes a somewhat larger value (curve (2) in

ﬁgure 18(a)), or it may decrease gradually to a constant level (ﬁgure 18(b)). As discussed

before, this behaviour in which the stress on the fault plane gradually decreases as slip

increases is often called slip-weakening (Rice 1980, Li 1987). Slip-weakening models have

been considered in seismological models by Brune (1970), Kikuchi and Fukao (1988), Heaton

(1990) and Kikuchi (1992).

If friction is not constant, the rupture dynamics is complicated, but for the energy budget

considered here, we formulate this problem referring to a simple case shown in ﬁgure 18(b).

The friction σ

f

gradually drops to a constant value σ

f0

until the slip becomes D

c

. (For simplicity,

here we assume that the ﬁnal stress σ

1

is equal to σ

f0

.) Then, we deﬁne the average friction

¨ σ

f

by

¨ σ

f

=

1

D

D

0

σ

f

(u) du, (4.33)

where u is the slip (offset) on the fault plane. Then, the total energy dissipation is given by

S

D

0

σ

f

(u) du = ¨ σ

f

DS. (4.34)

Figure 18(c) shows the partition of energy. The area under the trapezoid outlined by the heavy

lines represents the total potential energy change, W. The area under the curve labelled as

σ

f

is the total dissipated energy. Then, the radiated energy, E

R

, is the dotted area. Thus,

E

R

= W − ¨ σ

f

DS. (4.35)

As we discussed earlier, if we use the slip-weakening model, the hatched area in ﬁgure 18(c)

is the fracture energy, E

G

. Then, the total dissipated energy ¨ σ

f

DS can be divided into E

G

, and

the frictional energy, E

H

, represented by the cross-hatched area in ﬁgure 18(c). We should

note that this partition is model dependent; nevertheless it is based on the breakdown-zone

interpretation of the slip-weakening behaviour and is useful for interpretation of the energy

budget.

From ﬁgure 18(c), we obtain

E

R

=

σ

0

−σ

1

2

DS −E

G

=

σ

s

2

DS −E

G

=

σ

s

2µ

M

0

−E

G

, (4.36)

where M

0

= µDS is the seismic moment.

The ratio,

η

R

=

E

R

E

R

+ E

G

(4.37)

is called the radiation efﬁciency and is an important parameter which determines the dynamic

character of an earthquake (Husseini 1977). The radiation efﬁciency, η

R

, is different from

the seismic efﬁciency, η, which is given by E

R

/W. As discussed in section 3.1.4, W

cannot be determined directly by seismological methods and the seismic efﬁciency is difﬁcult

to determine. Because W E

R

+E

G

, η

R

η. If η

R

≈ 1, the breakdown zone is unimportant

and failure occurs primarily in the steady-state regime regardless of whether it is crack-like

or friction-dominated. On the other hand, if η

R

< 1, the microscopic breakdown process is

dominating the dynamics.

1466 H Kanamori and E E Brodsky

By combining (4.36) and (4.37), we obtain a relation between η

R

and observable

seismological parameters:

η

R

=

2µ

σ

s

˜ e, (4.38)

where

˜ e =

E

R

M

0

(4.39)

is the radiated energy scaled by the seismic moment, i.e. scaled energy. As shown in ﬁgure 12,

˜ e is always less than 10

−4

, and values of ˜ e for small earthquakes are often 1–2 orders of

magnitude less than those for large earthquakes. Whether the trend shown in ﬁgure 12 is real

or not is currently debated (Ide and Beroza 2001). If it is real, for typical values of static

stress drops, 1–10 MPa, and a shear modulus, 3 10

4

MPa, η

R

for small earthquakes must

be signiﬁcantly less than unity. The small values of η

R

for small earthquakes motivate us to

examine the micro-mechanisms of slip-weakening and breakdown.

4.5. Fault-zone processes: melting, ﬂuid pressurization and lubrication

Motivated by the importance of the non-elastic slip-weakening processes in the energy balance,

we now turn to the micromechanics of rupture beyond solid friction and crack models.

Melting. One of the ﬁrst such special mechanisms recognized was frictionally-induced

melting. As ﬁrst suggested by Jeffreys (1942), frictional dissipation may be high enough,

early in the rupture, to melt the wallrock. The silicate melt then reduces the friction for the

remainder of the earthquake. Studies by McKenzie and Brune (1972), Richards (1976) and

Cardwell et al (1978) quantitatively conﬁrmed the potential importance of frictional heating

during faulting.

Here, we consider a gross thermal budget during faulting under a frictional stress σ

f

. Let S

and D be the fault area and the displacement offset, respectively. Then the total heat generated

during faulting is Q = σ

f

DS. If we assume that the heat is distributed during seismic faulting

within a layer of thickness w around the rupture plane and there is negligible heat transport

away over the timescale of the earthquake, the average temperature rise T is given by

T =

Q

CρSw

=

σ

f

D

Cρw

, (4.40)

where C is the speciﬁc heat and ρ is the density. In general D increases with the earthquake

magnitude, M

w

. Using the scaling relations given in section 3.2.1, we can relate D to M

w

(with the static stress drop σ

s

as a parameter), and compute T as a function of magnitude

with three parameters, σ

f

, w and the static stress drop σ

s

. Figure 19 shows T for the case

with w = 1 cm as a function of M

w

.

If σ

f

is comparable to σ

s

about 10 MPa, the effect of shear heating is signiﬁcant. If the

thermal energy is contained within a few centimetres around the slip plane during seismic slip,

the temperature can easily rise by 100–1000˚C during a moderate-sized earthquake.

Thermal ﬂuid pressurization. Geological outcrops of faults suggest that many faults are ﬁlled

with aqueous ﬂuids or a viscous mixture of gouge and water when active. If this is true, then

another set of mechanisms are at work during rupture. One possibility is thermal pressurization

of the ﬂuid. This concept was introduced to seismology by Sibson (1973), and analysed in great

detail by Lachenbruch (1980), Mase and Smith (1985, 1987) and Andrews (2002). Under the

The physics of earthquakes 1467

10

-1

10

0

10

1

10

2

10

3

10

4

10

5

10

6

10

7

10

8

1 2 3 4 5 6 7 8 9 10

∆s

s

=10 MPa, w=1 cm

∆

T

,

d

e

g

r

e

e

s f

=

1

G

P

a

1

0

0

1

0

1

M

P

a

σ

s

σ

f

M

w

Figure 19. Predicted temperature rise, T , in a fault zone as a function of magnitude, M

w

, with

the frictional stress, σ

f

, as a parameter. The static stress drop σ

s

is assumed to be 10 MPa.

pressure–temperature conditions at the seismogenic depths, the thermal expansivity of water is

of the order of 10

−3

˚C

−1

, andsigniﬁcant increase inpore pressure withtemperature couldoccur.

If the ﬂuid does not escape (small permeability) and the surrounding rock is not compressible,

the pressure increase would be of the order of 1 MPa per ˚C(Lachenbruch 1980). In actual fault

zones, permeability and compressibility vary and the pressure increase may be less. The most

important parameter controlling the pressure change is the permeability. The analysis of

Lachenbruch (1980) and Mase and Smith (1985, 1987) suggests that if permeability is less

than 10

−18

m

2

, ﬂuid pressurization is most likely to occur with a temperature rise of less than

200˚C, and the friction will drop signiﬁcantly. Permeability in the crust varies over a very wide

range of more than a factor of 10

10

. Although the distribution of permeability can be complex,

these studies suggest that ﬂuid pressurization can play an important role, at least locally, in

reducing friction. Amodest T of 100–200˚Cwould likely increase the pore pressure enough

to signiﬁcantly reduce friction. Figure 19 shows that this moderate temperature increase can

occur even for intermediate-sized earthquakes (M

w

= 3–5).

The key question is: what is the thickness w of the fault slip zone? Geologists have

examined many old fault zones which were formed at depths and were brought to the surface

by long-term uplift (i.e. exhumed faults). Some fault zones have a very narrow (about 1 mm)

distinct slip zone where fault slips seem to have occurred repeatedly. According to Chester

and Chester (1998), the internal structure of the Punchbowl fault, California, implies that

earthquake ruptures were not only conﬁned to a layer of ﬁnely shattered rock, but also largely

localized to a thin prominent fracture surface. They suggest that mechanisms that are consistent

with the extreme localization of slip, such as thermal pressurization of pore ﬂuids, are most

compatible with their observations. In other cases, several narrow slip zones were found but

evidence shows that each slip zone represents a distinct slip event (i.e. an earthquake). Thus,

geological evidence suggests a narrow slip zone, at least for some faults, but this question

remains debatable (Sibson 2003).

1468 H Kanamori and E E Brodsky

Lubrication. If a fault zone is narrow and slightly rough, and if the material in the fault

zone behaves as a viscous ﬂuid, it is also possible that elastohydrodynamic lubrication plays

an important role in reducing friction for large events (Brodsky and Kanamori 2001). As in

a ball bearing, the transverse shear gradients in the ﬂuid are balanced by the longitudinal

pressure gradients and the pressure increases on the leading edge of irregularities in the fault

surface. An interesting consequence of this is that as the slip and slip velocity increase,

the hydrodynamic pressure within a narrow zone becomes large enough to smooth out the

irregularities on the fault surface by elastic deformation, thereby suppressing high-frequency

ground motion caused by the fault surfaces rubbing against each other. During the 1999

Chi-Chi, Taiwan, earthquake, the observed ground-motion near the northern end of the fault

was extremely large (>3.5 ms

−1

, the largest ever recorded), but short period acceleration was

not particularly strong and the shaking damage was not the worst (Ma et al 1999). This could

be a manifestation of the high-speed lubrication effects (Ma et al 2003). However, since this is

the only earthquake for which such large slip and slip velocity were instrumentally observed,

whether this is indeed a general behaviour or not is yet to be seen.

Any of these dynamic weakening mechanisms can explain the lack of elevated heat ﬂow

over seismogenic faults (section 2.2). There is some evidence that the heat ﬂow is slightly

elevated over the section of the San Andreas fault that slips gradually with no large earthquakes

(Sass et al 1997). Ahigh heat ﬂowover the aseismic section would be consistent with dynamic

weakening reducing the frictional dissipation where there are earthquakes.

Since a fault zone is probably complex and heterogeneous in stress, ﬂuid content,

permeability, porosity and compressibility, no single process is likely to dominate. In other

words, we do not necessarily expect a single continuous layer of melting and pressurization;

we envision, instead, a fault zone that consists of many regions where different mechanisms

are responsible for slip at different stress levels, producing complex rupture patterns as

observed.

In these discussions, the thickness of fault slip zones is the key parameter for understanding

fault dynamics. Of course, whether lubrication occurs or not depends on many factors such

as the effective permeability in the fault zone, compressibility of fault rocks and the viscosity

of melts; but in view of the large slip and slip velocity associated with seismic faulting, a

signiﬁcant drop in friction is likely to occur if the slip zone is narrow.

4.6. Linking processes to the seismic data

4.6.1. The interpretation of macroscopic seismological parameters

Radiation efﬁciency. As we discussed in section 4.4, in the breakdown-zone interpretation

of the slip-weakening model, the energy deﬁned by the cross-hatched area in ﬁgure 18(c) is

interpreted as frictional thermal energy, E

H

, and is subtracted fromthe total potential energy; it

does not directly control the dynamics of earthquake rupture. In contrast, the fracture energy,

E

G

, represents the mechanical energy loss during faulting, and controls the fault dynamics in

a fundamental way. Thus, the determination of fracture energy for earthquakes is critically

important for understanding the dynamics of faulting. Since the radiation efﬁciency, η

R

, is

directly related to E

G

by (4.37), ﬁrst we describe how we can determine η

R

from macroscopic

seismic parameters (Venkataraman and Kanamori 2004).

As shown by equation (4.38), we can estimate the radiation efﬁciency, η

R

, using the three

macroscopic seismological parameters, M

0

, E

R

and σ

s

. If η

R

= 1, no energy is dissipated

mechanically and the potential energy, after heat loss has been subtracted, is radiated as seismic

waves and the earthquake is considered a very ‘brittle’ event. In contrast, if η

R

= 0, the event

is quasi-static and no energy is radiated, even if the static stress drop is very large.

The physics of earthquakes 1469

Figure 20. Static stress drop plotted as a function of depth for the different types of earthquakes.

Deep: deep earthquakes; Intraplate: earthquakes which occur within the lithospheric plate; Crustal:

earthquakes which occur within continental crusts; Downdip and Interplate: earthquakes which

occur on the subduction-zone plate boundary; Tsunami: earthquake with a slow deformation at

the source which generates tsunamis disproportionately large for its magnitude (ﬁgure taken from

Venkataraman and Kanamori (2004)).

For many large earthquakes, the seismic moment, M

0

, and the radiated energy, E

R

have

been determined. The determination of the static stress drop, σ

s

, is a little more difﬁcult.

Although the scaling relation discussed in section 3.2.1 (ﬁgure 11) shows that most large

earthquakes have comparable stress drops in the range of 1–10 MPa, we need to determine

the stress drops for individual earthquakes for this purpose. Figure 20 shows the estimates for

large earthquakes.

For shallow earthquakes, σ

s

is in the range of 1–10 MPa, as discussed in section 3.2.

We can see a general trend of σ

s

increasing with depth. For the deepest earthquakes, σ

s

is

in the range of 20–200 MPa, and the average is roughly 10 times larger than that for shallow

earthquakes. This trend is roughly consistent with the result of Houston (2001) who found

that the source duration of deep earthquakes with comparable magnitudes is systematically

shorter than that of shallow earthquakes. This result, if interpreted using the assumption that

the rupture speed is, on the average, similar to the S-wave speed, suggests a trend similar to

that shown in ﬁgure 20.

Using the estimates of radiated energy, seismic moment and static stress drop, we can

determine the radiation efﬁciency for all these earthquakes using (4.38) (Venkaraman and

Kanamori 2004). Figure 21 shows the radiation efﬁciencies as a function of the magnitude,

M

w

. For a few earthquakes the computed η

R

is larger than 1. This is probably due to the errors

in the estimates of radiated energy and/or stress drops.

The radiationefﬁciencyof most earthquakes lies between0.25and1. Tsunami earthquakes

(earthquakes with slow deformation at the source which generate tsunamis disproportionately

1470 H Kanamori and E E Brodsky

Figure 21. Radiation efﬁciency η

R

= E

R

/(E

R

+ E

G

) as a function of M

w

. The different symbols

show different types of earthquakes as described in ﬁgure 20. Most earthquakes have radiation

efﬁciencies greater than 0.25, but tsunami earthquakes and two of the deep earthquakes (the Bolivia

earthquake and the Russia–China border earthquake) have small radiation efﬁciencies (ﬁgure taken

from Venkataraman and Kanamori (2004)).

large for its magnitude), however, have small radiation efﬁciencies (<0.25) and the two deep

earthquakes, the 1999 Russia–China border event and the 1994 deep Bolivia earthquake, have

small radiation efﬁciencies.

For the 1994 Bolivian earthquake (M

w

= 8.3, depth = 635 km), the largest deep

earthquake ever recorded, the source parameters couldbe determinedwell enoughtoinvestigate

the energy budget (Kanamori et al 1998). The result showed that W

0

= 1.4 10

18

J and

E

R

= 5 10

16

J, which is only 3% of W

0

, and the difference W

0

−E

R

= 1.35 10

18

J,

was not radiated, and must have been deposited near the focal region, probably in the form of

fracture energy in addition to the frictional energy. This energy 1.35 10

18

J is comparable

to the total thermal energy released during large volcanic eruptions such as the 1980 Mount

St Helens eruption. In other words, fracture and thermal energy at least comparable to that

released by a large volcanic eruption must have been released in a relatively small focal region,

about 50 50 km

2

, within a matter of about 1 min. The mechanical part of the process, i.e.

the earthquake observed as seismic waves, is only a small part of the whole process. Thus,

the Bolivia earthquake should be more appropriately viewed as a thermal process rather than

The physics of earthquakes 1471

a mechanical process. How much of the non-radiated energy goes to heat depends on the

details of the rupture process, which is unknown. However, it is possible that a substantial part

of the non-radiated energy was used to raise the temperature in the focal region signiﬁcantly.

The actual temperature rise, T , also depends on the thickness of the fault zone, which is not

known, but if it is of the order of a few centimetres, the temperature could have risen to above

10 000˚C (ﬁgure 19).

The relation between radiation efﬁciency and rupture speed. As discussed in section 4.1.3,

the energy release rate for a dynamic crack G is given as a function of the rupture speed V

(equation (4.15)). Since the fracture energy E

G

can be interpreted as the integral of Gover the

entire fault surface S, we can relate the radiation efﬁciency, η

R

, to rupture speed, V as follows.

In the simplest model discussed in section 4.1.3, we can use equations (4.7) and (4.9) to write

E

G

=

GdS = g(V)

G

∗

dS = g(V)W

0

(4.41)

and

η

R

=

E

R

E

R

+ E

G

=

E

R

W

0

, (4.42)

from which we obtain

η

R

= 1 −g(V). (4.43)

The average rupture speed is usually determined from the inversion of seismic waves and

the results can be non-unique, but for large earthquakes, the estimates of rupture speed are fairly

accurate. Most of these earthquakes have rupture speeds such that the ratio of rupture speed

to shear wave speed (V/β) is between 0.75 and 0.95. However, the 1994 Bolivia earthquake,

the 1999 Russia–China border event and the tsunami earthquakes, have small V/β, about 0.1

to 0.2. Figure 22 shows the upper and lower limit of radiation efﬁciencies that were determined

from the energy budget plotted against the upper and lower limit of V/β obtained from the

literature. The theoretical curves relating radiation efﬁciency to rupture speed for Modes I, II

and III cracks (equations (4.16)–(4.18)) are also plotted in the same ﬁgure. To the ﬁrst order,

the observed data follow the theoretical curves obtained from crack theory. Since rupture

speed is an independently determined quantity, this consistency of the observed relationship

between η

R

and V/β with the calculations from crack theory enhances the results shown in

ﬁgure 21.

Summary and implications. With the three seismologically observable macroscopic

parameters (seismic moment M

0

, radiated energy E

R

and the static stress drop σ

s

), we showed

that for most earthquakes, the radiation efﬁciency which is given by η

R

= E

R

/(E

R

+ E

G

) is

larger than 0.25, which means that the amount of energy mechanically dissipated during rupture

is comparable or smaller than the energy radiated as seismic waves. This conclusion seems to

be supported by the independent observations of the high rupture speed V. Note that this line

of reasoning poses no constraint on the energy E

H

dissipated directly as heat.

For tsunami earthquakes (slow seismic events) and some deep earthquakes, the radiation

efﬁciency is small, which means that the rupture process of these earthquakes involves more

dissipative processes than the average. One interpretation is that most tsunami earthquakes

involve rupture in soft deformable sediments, and a large amount of energy is used in

deformation. The mechanism of large deep earthquakes is not known well, but it is likely

that the rupture process in the pressure–temperature environment at large depths may involve

large amounts of plastic deformation with large amounts of energy dissipation.

1472 H Kanamori and E E Brodsky

V /

Figure 22. Radiation efﬁciencies determined from the radiated energy-to-moment ratios and

estimates of static stress drop (equations (4.38) and (4.39)) plotted against the estimates of the

ratio of rupture speed to shear wave speed obtained from literature. The symbols are the same as in

ﬁgure 20; for comparison, the theoretical curves relating radiation efﬁciency to rupture speed for

Modes I, II and III cracks have also been plotted (ﬁgure taken from Venkataraman and Kanamori

(2004)).

The relatively large radiation efﬁciency, i.e. relatively small critical fracture energy G

c

or

small fracture toughness K

c

(4.11), for most shallow earthquakes has an important implication

for rupture growth of earthquakes. As discussed in section 4.1.3, the rupture growth is

controlled by the balance between the dynamic stress intensity factor K and K

c

. As a rupture

grows, the length scale a increases and K increases (equation (4.3)). Thus, if K

c

is small,

on the average, the rupture is more likely to grow and develop into a runaway rupture. If

friction decreases as the slip increases, as discussed in section 4.5, the tendency for runaway

would increase because K is also proportional to (σ −σ

f

) (equation (4.3)). If this is the case,

once an earthquake is initiated, it will be difﬁcult to stop the rupture dynamically. To stop the

rupture, some external static features such as a strength barrier or irregular fault geometry may

be required. In terms of the friction model discussed in section 4.3, the small G

c

or K

c

means

small critical slip, D

c

(equation (4.31)).

The physics of earthquakes 1473

10

-1

10

0

10

1

10

2

10

3

10

4

10

5

3 4 5 6 7 8 9

N

(

M

)

Magnitude, M

Figure 23. Magnitude–frequency relationship for earthquakes in the world for the period 1904 to

1980. N(M) is the number of earthquakes per year with the magnitude M. The solid line shows

a slope of −1 on the semilog plot which corresponds to a b-value of 1. Note that, on the average,

approximately one earthquake with M 8 occurs every year. The data sources are as follows:

M 8, for the period 1904 to 1980 from Kanamori (1983); M = 5.5, 6.0, 6.5, 7.0 and 7.5, for

the period from 1976 to 2000 from Ekstrom (2000); M = 4 and 5, for the period January 1995 to

January 2000 from the catalogue of the Council of National Seismic System. For this range, the

catalogue may not be complete, and N may be slightly underestimated.

At present, the accuracy of the macroscopic source parameters, especially E

R

and σ

s

,

is not good enough to accurately estimate the fracture parameters G

c

, K

c

and D

c

, and to draw

more deﬁnitive conclusions on the rupture dynamics of earthquakes. Currently, extensive

efforts are being made to improve the accuracy of determinations of the macroscopic source

parameters.

5. Earthquakes as a complex system

Another possible approach to understanding why earthquakes happen is to take a broad

view beyond a single event. We can study earthquakes by dealing with large groups of

earthquakes statistically. The goal is to ﬁnd systems that robustly reproduce the general

patterns of seismicity regardless of the details of the rupture microphysics. This approach has

had considerable success characterizing the types of models that will reproduce the observed

magnitude–frequency relationship (i.e. Gutenberg–Richter relation) used in seismology.

The magnitude–frequency relationship (the Gutenberg–Richter relation). In general small

earthquakes are more frequent than large earthquakes. This is quantitatively stated by the

Gutenberg–Richter relation (Gutenberg and Richter (1941), a recent review is found in Utsu

(2002).) It describes the number of earthquakes expected of each size, or magnitude, in a given

area. In any area much larger than the rupture area of the largest earthquake considered, the

number of earthquakes, N(M), which have a magnitude greater than or equal to M is given

by the relation

log N(M) = a −bM, (5.1)

where a and b are constants. Figure 23 shows that the Gutenberg–Richter relationship

even applies to a seismicity catalogue encompassing the entire planet. Approximately

one earthquake with M 8 occurs every year somewhere in the Earth.

1474 H Kanamori and E E Brodsky

For most regions the value of b is nearly 1, as is the case in ﬁgure 23. This strikingly

consistent observation has motivated much of the study on fault networks and self-organized

criticality. (For more extensive discussions on this subject, see Main (1996), Turcotte (1997),

Rundle et al (2000) and Turcotte and Malamud (2002).) The primary conclusions of this

research are that over a wide (but ﬁnite) range of scales, fault networks are fractal. Cascades

of failure result when the faults are extremely close to failure prior to any late-stage triggering.

These cascades can be interpreted as an example of self-organized criticality.

The Gutenberg–Richter Law is a major tool in probabilistic hazard assessment. It allows

extrapolation fromthe rates of small earthquakes, which we observe easily, to the likelihood of

large events. Given the societal importance of the resulting hazard assessments, it is important

that the empirically-derived Gutenberg–Richter Law be put on ﬁrm physical ground.

Simple models. The starting point of complexity models is the assumption that the initiation,

growth and cessation of earthquake rupture are controlled by the complex interaction of fault-

bounded blocks on scales as small as individual cracks and as large as continents. Because of

the large number of elements involved and of the complexity of the interaction it may not be

determinable exactly how different parts of the crust interact with each other, well enough to

understand the earthquake process in a deterministic way. Nevertheless, some properties of

earthquakes such as the magnitude–frequency relationship can be understood as manifestations

of the general behaviour of complex systems.

The cruxof a successful earthquake complexitymodel is the recreationof critical behaviour

by setting up a system of elements with only local interactions being speciﬁed. A critical state

is when events of all sizes can occur and their frequency distribution follows a power law.

In the critical state, the local interactions can accumulate to generate long-range organization.

Self-organized criticality is when a system evolves to the critical state naturally without any

dependence on the initial conditions or tunable parameters (Bak et al 1988, Hergarten 2002).

In practice, systems are only critical within a certain range of scales determined by the overall

boundaries of the system. Because of the ubiquity of the Gutenberg–Richter distribution,

earthquakes are thought to be self-organized critical systems. Three speciﬁc models for

generating the critical state in earthquake processes are the mechanical slider-block system,

the percolation model and the sand pile model.

In the slider-block system (Burridge and Knopoff (1967), ﬁgure 24(a)), many blocks are

connected by a spring and the whole mass–spring system is dragged on a frictional surface.

The friction between the block and the surface is governed by a simple velocity-weakening

law. As the mass–spring system is dragged, some blocks slip intermittently. Most of the time,

a single block slips. This slip is interpreted as a small ‘earthquake’, and some potential energy

is released. However, occasionally slip of a block triggers slip of adjacent blocks causing a

larger earthquake with a larger amount of potential energy release. If more than one block

are triggered, we have a larger earthquake. After a series of events have occurred, we count

the number of events N(E) which released a potential energy larger than E. As shown in

ﬁgure 24(b), the relation between N(E) and E exhibits a power-law like behaviour, and when

plotted on a log–log diagram, the relation looks like the earthquake magnitude–frequency

relationship. In this slider-block model, the interaction between the blocks is described by a

differential equation that involves the spring stiffness, mass and friction.

In a percolation model (Otsuka 1971), a seismic fault is modelled by a distribution of many

small elements (patches). If one patch fails, then it can trigger failure of the adjacent patches

with some probability. This process continues until it stops spontaneously, and the whole

continuous failure corresponds to an earthquake rupture. If a patch can trigger nearby patches

at s sites with a transition probability, p, then e = ps is the expectancy of the number of patches

The physics of earthquakes 1475

(a)

(b)

Figure 24. (a) Mass–spring system sliding on a frictional surface (Burridge and Knopoff 1967).

(b) The relation between the number of events and released potential energy (Burridge and

Knopoff 1967).

to be triggered at each step. If e < 1, then the growth will eventually stop at a certain step

when a total of F patches have failed. This corresponds to a sub-critical state, i.e. the range of

interactions is limited. If this whole process is repeated many times, we ﬁnd a relation similar

to the magnitude–frequency relationship between log F and the number of cases, n, in which

at least F patches failed. Figure 25 shows the results of numerical simulations performed for

s = 10 and three cases, e = 0.8, 0.9 and 0.99. When e = 1, the system is critical. This

distribution corresponds to the magnitude–frequency distribution shown in ﬁgure 23.

In a sand pile model, a sand particle is added to a sand pile which has been maintained

in a critically stable state. Usually a small number of sand particles slide off the pile and into

adjacent areas as a result of the small additional stress. However, there is also a small but

ﬁnite probability that a large slide can happen as a result of simply adding one particle. Bak

et al (1988) used a set of simple rules that simulate the actual potentially complex physical

1476 H Kanamori and E E Brodsky

Figure 25. Magnitude–frequency relationships produced with a percolation model (ﬁgure taken

from Kanamori and Brodsky (2001)).

interaction between different cells. The result of the interaction yields a relationship like the

earthquake magnitude–frequency relationship (Bak and Tang 1989, Kadanoff 1991).

In all these models, what is essential is the interaction between the many elements which

make up the system. In the actual earthquake process, the interaction is between different parts

of a fault and between different faults. This interaction can be due to static as well as dynamic

processes. The robustness of the Gutenberg–Richter result tells us that this particular feature

of seismicity is insensitive to the microscopic physics controlling the failure and rupture. The

details of the microphysics are only important for addressing other questions, such as the

likelihood of a particular earthquake on a particular fault. The complexity models also show

that earthquake interactions have inherently chaotic elements in addition to the predictable

elements governed by the stress loading mechanismand fault structures. Unravelling the limits

and extent of the chaos is prerequisite for determining whether or not individual earthquakes are

predictable over societally useful timescales given the limited resolution of our observations

of the initial conditions.

In the above, we discussed only the magnitude–frequency relationship. Another

commonly observed seismicity pattern is the timing of aftershock decay which is known

as Omori Law. As we will discuss later (section 6.2.4), Omori’s Law can be explained with

several physical models.

6. Instability and triggering

We have now covered the major pieces of the earthquake puzzle: stress in the crust, observable

parameters, macroscopic observations, micromechanics and complex systems. We are now

ready to use these tools to broach the key problem of earthquake triggering and instability. We

will ﬁrst deal with frictional instability and then approach data that has bearing on other types

of initiation.

6.1. Instability

6.1.1. Stick slip and instability. Friction on a fault is not constant. Even in simple high-school

formulations, friction depends on slip velocity as dynamic friction is less than static friction.

The physics of earthquakes 1477

(a)

(b)

Figure 26. Stick slip model. (a) A spring-loaded slider block on a frictional surface. (b) Variation

of friction as a function of displacement (——) and the spring loading force (– – –).

In realistic fault conditions, the slip-velocity dependence can be more complex and friction

also varies as a function of time and slip distance. Therefore, sliding does not occur smoothly;

it occurs in a stop-and-go fashion. This behaviour is generally called stick slip (Brace and

Byerlee 1966).

Figure 26 shows a mechanismof stick slip (e.g. Rabinowicz (1995)) illustrated by a spring-

loaded slider-block model similar to that discussed in section 5. In this ﬁgure, k, δ, δ

0

, σ

n

and

τ are the spring constant, the displacement of the block, the displacement of the right-hand

end of the spring, the normal stress and the friction between the block and the surface, in that

order. Suppose we increase δ

0

by pulling the spring from the right. Then, the force balance is

given by

τ = k(δ

0

−δ) = −kδ + kδ

0

, (6.1)

where

τ = µσ

n

. (6.2)

The solid curve in ﬁgure 26(b) shows the variation of τ as a function of δ. The broken

line in ﬁgure 26(b) is the loading force exerted by the spring, given by the rhs of (6.1). The

intersection, B, between the broken line and the solid curve gives the equilibrium position

given by equation (6.1). As δ

0

increases, the broken line moves upward, and the intersection

moves to B

/

. Between points B and C, the block moves over the surface smoothly. At point C,

τ drops suddenly and the spring force exceeds τ, and the block moves abruptly along C–D

driven by the spring force. The area A

1

is approximately equal to A

2

. The block is stationary

at D until the spring force reaches point E with the increase of δ

0

, from where smooth motion

begins again.

More precisely, at point C,

τ

δ

= k (6.3)

1478 H Kanamori and E E Brodsky

Figure 27. A schematic ﬁgure showing the nucleation length,

˜

L, on a frictional fault plane. For a

rupture to grow,

˜

L must be larger than

˜

L

n

given by (6.7).

and the stick slip instability begins past this point where

τ

δ

> k. (6.4)

If the spring constant k is large, the slope of the broken line increases, and no instability occurs,

i.e. stable sliding occurs without stick-slip behaviour. Thus, the spring constant (stiffness of

the system) controls the stability for a given frictional property of the surface.

Stiffness of the fault system. In the models described above, the stiffness of the spring, k, plays

a key role in determining the stability. Then the question is: what is the stiffness of the crust?

Stiffness is deﬁned by the ratio of stress to displacement, e.g. k = τ/δ. Then, if we

consider a small crack with length scale

˜

L in a nucleation zone, the stress required to cause a

slip D is given by ED/

˜

L, where E is a relevant elastic modulus. Thus, the fault stiffness can

be deﬁned by

k

f

≈

E

˜

L

. (6.5)

6.1.2. Nucleation zone. Consider the nucleation of a slip on a frictional surface with the

normal stress σ

n

, the static friction coefﬁcient, µ

s

, kinetic friction coefﬁcient, µ

k

, and critical

slip, D

c

(ﬁgure 27).

Then from the stick-slip model (equation (6.3)), at the critical point,

k

f

D

c

= σ

n

(µ

s

−µ

k

). (6.6)

Then, combining this expression with the deﬁnition of k

f

, we deﬁne a critical fault length

scale

˜

L

n

:

˜

L

n

≡

˜

L ≈

ED

c

(µ

s

−µ

k

)σ

n

. (6.7)

According to the frictional instability model,

˜

L

n

in (6.7) is the nucleation length of this

frictional surface. If we assume that the laboratory measurements of µ

s

, µ

k

and D

c

on rocks

are appropriate for natural faults, at typical seismogenic depths where the normal stress is

∼200 MPa, D

c

is ∼10 µm, µ

s

− µ

k

∼ 0.1 and E is ∼5 10

10

Pa, then

˜

L

n

≈ 3 cm. For a

more sophisticated frictional model, the nucleation zone can be as large as 1 m (e.g. Lapusta

and Rice (2003)). For the latter values, the strain D

c

/

˜

L

n

in the nucleation zone prior to the

The physics of earthquakes 1479

earthquake is of the order 10

−5

. This relatively large strain could potentially be observed up

to ∼80 m away on modern strain metres. The major impediment to testing such a prediction

method is the very small size of the proposed nucleation zone. A geophysicist would have

to be very lucky to choose to put an instrument within 80 m of the right 1 m

2

patch of the

15 000 km

2

San Andreas fault!

The nucleation length model in (6.7) suggests that the smallest possible earthquakes have

magnitudes M

w

= −1 (from equation (3.7). M

w

= −1 corresponds to M

0

= 4 10

7

Nm).

Smaller earthquakes have been observed, although they are difﬁcult to detect. Dense networks

designed to capture extremely small earthquakes could help conﬁrm or refute the nucleation

length model by determining whether any lower bounds on earthquake size exists. If

the observation of very small earthquakes with M

w

< −1 is supported, then, either the

extrapolation of laboratory parameters from metre-scale samples to kilometre-scale faults is

problematic, or earthquake initiation involves other processes than simple frictional instability

as formulated here. Another strategy for studying earthquake initiation is to examine cases

where the immediate trigger of a real earthquake is known.

6.2. Triggering

During the past decade, seismologists have discovered that earthquakes commonly trigger

other earthquakes both in the near-ﬁeld and at distances approaching 4000 km(Hill et al 2002).

As one of the few cases in nature where the immediate cause of an earthquake is apparent,

triggering provides a fundamental clue into initiation. Observed cases of triggering are usually

separated into near-ﬁeld (<2–3 fault lengths) and far-ﬁeld, with a different set of mechanisms

operating in each regime. This distinction may be artiﬁcial, as the far-ﬁeld mechanism must

also operate in the near-ﬁeld, however, it is useful in order to separate plausible regimes for

certain physics. Therefore, we will retain the separation here with the above caveats.

6.2.1. Observations. Large earthquakes are followed by abundant smaller earthquakes called

aftershocks (section 6.2.4). Aftershocks are, therefore, the most commonly observed form of

earthquake interactions. Aftershocks form a cloud around the mainshock rupture plane that

can extend up to two fault lengths away. Beginning in the early 1990s, studies such as those

by King et al (1994) investigated the proposal that aftershocks are triggered by the static stress

changes due to the dislocation of the earthquake. As discussed in section 3.1, the deformation

of the crust by a slip on a fault plane generates an elastic strain ﬁeld surrounding the fault.

In some areas the strain is extensional, in others it is compressional. The pattern of dilatational

strain can most easily be seen for the simple example of a strike-slip fault (ﬁgure 28, right).

In addition to the dilatational strain, there is also a deviatoric stress component to the stress

ﬁeld. It is a combination of the shear and normal stress that will determine if a given fault

plane slips. Following the Anderson, Hubbert and Rubey failure criterion laid out in section 2,

we deﬁne the Coulomb stress change τ

c

on a fault by

τ

c

= −µ(σ

n

−p) + τ, (6.8)

where σ

n

and τ are the resolved normal and shear stress changes, respectively, on a given

fault orientation, p is the pore pressure change and µ is the coefﬁcient of friction. If τ

c

increases, then frictional fault slip is promoted (see equation (2.6)). The King et al strategy for

studying aftershocks is to map the calculated Coulomb stress change based on the observed slip

during an earthquake and compare the resulting ﬁeld with the observed aftershock distribution

(ﬁgure 28).

1480 H Kanamori and E E Brodsky

Figure 28. Change in Coulomb stress from the 1979 Homestead Valley mainshock and the

subsequent aftershocks. Red (positive) indicates that optimally oriented faults are stressed more

towards failure and purple (negative) indicates that failure is inhibited. White circles are observed

aftershocks (from King et al (1994)). Shown on the right is a schematic of a strike-slip fault with

slip in the directions shown by the arrows. The dilatational strain is compressional where there are

‘+’ signs and extensional where there are ‘−’ signs.

This method has had some success in predicting the location of aftershocks and even a

few large, nearby earthquakes (Stein 1999). Approximately 85% of the aftershocks of the

1992 M

w

= 7.3 Landers earthquake occurred where the Coulomb stress ﬁeld increased at

the time of the mainshock (Hardebeck et al 1998). A recent review can be found in Harris

(2002).

Equation (6.8) by itself does not fully describe the aftershock ﬁeld shown in ﬁgure 28.

There are some aftershocks in the areas where failure should have been inhibited by the

mainshock. This problem of a continual low aftershock rate in the destressed regions was

addressed by adding rate- and state-dependent friction (section 4.2.2) to the stress transfer

model (Stein et al 2003). Dieterich (1994) showed that if velocity and memory-dependence

are incorporated into the standard frictional coefﬁcient based on laboratory experiments, then

the rate of seismicity, rather than the absolute number of events, will be inﬂuenced by a stress

step. Therefore, we might expect some aftershocks to occur in all areas of the stress ﬁeld if

the background seismicity rate is fairly high, but the aftershock rate relative to the background

rate will vary systematically with the imposed Coulomb stress.

Since the pattern of static Coulomb stress increase is controlled by the mainshock fault

geometry, some other mechanisms couldpossiblyproduce similar predictions. Inparticular, the

dynamic stresses followa very similar pattern because the strong and weak areas of shaking are

also determined by the mainshock fault geometry. One major difference between the static and

dynamic stress is that only the dynamic ﬁeld is affected by rupture directivity (section 3.1.5).

The physics of earthquakes 1481

Figure 29. Triggering in Montana from the 2002 Denali, Alaska, earthquake (M

w

= 7.9). Top

panel is the seismic wave generated by the earthquake in Alaska as recorded by a seismometer in

Bozeman, Montana. The middle panel is the same record ﬁltered at high-frequencies to show local

earthquakes occurring in the vicinity of Montana during the passage of the seismic waves from the

Alaskan earthquake. Each green arrow is a local earthquake. The bottom panel is a magniﬁed view

of the record of one of these local earthquakes.

Kilb et al (2000) and Gomberg et al (2003) demonstrated that for earthquakes with strong

directivity, the aftershocks are better predicted by the dynamic than the static stress ﬁelds.

However, the exact mechanism for dynamic triggering is unclear. Furthermore, the dynamic

shaking cannot explain the rate decreases, or ‘stress shadows’, sometimes observed around

faults (Stein et al 2003). Static stress ﬁelds explain stress shadows by invoking negative

Coulombstress changes that move potential faults further fromfailure. The oscillatorydynamic

ﬁeld has no such negative effect. Areas are simply distinguished by stronger or weaker shaking.

Therefore, the current balance of evidence favours static stress as a primary mechanism for

generating aftershocks, but the debate is far from over.

Aftershocks can extend up to about 1–2 fault lengths from the original event. Past this

distance earthquakes were thought to have no effect until a very surprising observation in

1992. The magnitude 7.3 Landers earthquake in Southern California was followed by up to

10-fold increased seismicity in geothermal and volcanic areas up to 1500 km away, for days

after the mainshock (Hill et al 1993). Since that time, remote triggered seismicity has been

robustly documented for several large events and has now been seen up to 4000 km from the

mainshock (Brodsky et al 2000, Gomberg et al 2001, Eberhardt-Phillips 2003, Prejean et al

2004). Figure 29 shows a recent example. In all cases, the triggered sites are geothermal areas.

Increases in seismicity have been often observed within the surface wave trains indicating that

the seismic waves are the trigger. The triggered seismicity often persists for several days

indicating that the seismic waves have a sustained effect on the stress ﬁeld. The accompanying

deformation (Johnston et al 1995) also suggests a sustained stress in the triggered regions.

1482 H Kanamori and E E Brodsky

0

Figure 30. Slider-block sliding on a surface with rate- and state-dependent friction.

Artiﬁcially induced seismicity also gives a clue to the initiation process. In addition to the

well-controlled Rangely experiment discussed in section 2, humans have produced earthquakes

under less advantageous circumstances due to mining, reservoir ﬁlling and oil exploitation.

Many of these cases are consistent with pore pressure changes relieving the normal stress as

was observed for Rangely (Guha 2000). In mines, the excavation also directly removes the

load on faults producing the same effect (a recent review is found in McGarr et al (2002)).

To summarize, the observations of triggered and induced earthquakes imply that: (1) static

stress changes may be effective in the nearﬁeld triggering of aftershocks, (2) seismic waves

can trigger earthquakes at long-distances in geothermal areas and (3) pore pressure changes

can trigger seismicity. We now explore in detail some theoretical mechanisms for triggering

earthquakes that satisfy at least parts of these constraints. At present, no uniﬁed earthquake

model exists that satisﬁes all of them.

6.2.2. Triggering with the rate- and state-dependent friction mechanism. If friction on

a sliding surface is controlled by the rate- and state-dependent friction law discussed in

section 4.2.2, then a sudden change in loading causes a sudden increase in the sliding speed

which in turn results in accelerated seismic slip. This mechanism can be important in seismic

triggering, as shown by Dieterich (1994). Because this model is widely used in seismology,

we discuss this particular mechanism in greater detail than some others.

Consider a slider-block model shown in ﬁgure 30 in which friction is controlled by the

rate- and state-dependent friction given by (equation (4.24))

µ = µ

/

0

+ Aln

˙

δ + B ln θ. (6.9)

Consider the case where [θ

˙

δ/D

c

[ · 1 (e.g. large

˙

δ during coseismic slip). In this case,

from (4.25)

˙

θ = −

θ

˙

δ

D

c

, (6.10)

from which we obtain

θ = θ

0

exp

−

δ

D

c

(6.11)

and (6.9) becomes

µ = µ

/

0

+ Aln

˙

δ + B ln θ

0

−

B

D

c

δ. (6.12)

Then, (6.1) and (6.2) become

σ

n

µ

/

0

+ Aln

˙

δ + B ln θ

0

−

B

D

c

δ

= −kδ + kδ

0

. (6.13)

The physics of earthquakes 1483

Spontaneous behaviour. First, we examine the behaviour of this system under constant

loading, i.e. kδ

0

= τ

0

. Then, integrating (6.13) with the initial conditions, δ = 0 and

˙

δ =

˙

δ

0

at t = 0, we obtain

δ = −

A

H

ln

1 −

˙

δ

0

Ht

A

(6.14)

and

˙

δ =

¸

1

˙

δ

0

−

Ht

A

¸

−1

, (6.15)

where

H = −

k

σ

n

+

B

D

c

. (6.16)

For an unstable system, H > 0. Equations (6.15) shows that the sliding velocity spontaneously

increases with time, and at the time

t

f

=

A

H

1

˙

δ

0

(6.17)

˙

δ becomes inﬁnitely large; that is, an instability occurs, i.e. an earthquake occurs. The time t

f

is called the time-to-failure.

Loading at a uniform rate. Next, we add loading given by a linear function of time, i.e.

kδ

0

= τ(t ) = τ

0

+ ˙ τt, (6.18)

where ˙ τ is a constant loading rate. Then, from (6.13)

τ(t ) −kδ

σ

n

=

µ

/

0

+ Aln

˙

δ + B ln θ

0

−

B

D

c

δ

. (6.19)

We can integrate (6.19) to obtain,

δ = −

A

H

ln

¸

˙

δ

0

Hσ

n

˙ τ

¸

1 −exp

˙ τt

Aσ

n

¸

+ 1

¸

(6.20)

and

˙

δ =

¸¸

1

˙

δ

0

+

Hσ

n

˙ τ

¸

exp

−

˙ τt

Aσ

n

−

Hσ

n

˙ τ

¸

−1

. (6.21)

The functional forms of δ and

˙

δ given by (6.20) and (6.21), respectively, are a little complicated,

but both exhibit a monotonic behaviour in time that increases rapidly.

From equation (6.21), the time-to-failure, t

f

, is given by

t

f

=

Aσ

n

˙ τ

ln

˙ τ

Hσ

n

˙

δ

0

+ 1

. (6.22)

Stepwise change in loading. Next, let us consider the case where a sudden change in loading

occurs from τ

0

to τ

1

by τ at t = t

0

(i.e. τ

1

= τ

0

+ τ). This corresponds to the case when a

sudden change in the crustal stress occurs due to a large earthquake. Solving equation (6.13)

with kδ

0

= τ

0

for t < t

0

, and with kδ

0

= τ

0

+ τ for t > t

0

, and requiring the continuity of

slip at t = t

0

, i.e.

δ(t = t

0

−ε) = δ(t = t

0

+ ε),

1484 H Kanamori and E E Brodsky

0

1

2

3

4

5

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

1

10

100

1000

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

δ/(A/H)

0

/ δ δ

. .

0

/( / ) t A Hδ

.

0

/( / ) t A Hδ

.

Figure 31. Non-dimensional slip δ/(A/H) (left) and non-dimensional slip speed

˙

δ/

˙

δ

0

as a function

of non-dimensional time, t /(A/H

˙

δ

0

). A non-dimensional stepwise stress change τ/Aσ

n

= 1.25

is given at time t /(A/H

˙

δ

0

) = 0.5.

we can derive

˙

δ(t = t

0

+ ε) =

˙

δ(t = t

0

−ε) exp

τ

Aσ

n

. (6.23)

Figure 31 shows the non-dimensional slip, δ/(A/H), and slip speed,

˙

δ/

˙

δ

0

, as a function

of non-dimensional time t /(A/H

˙

δ

0

).

Thus, a step-wise change in loading by τ causes a step-wise increase in sliding velocity,

which in turn causes a step-wise decrease in the time-to-failure. This behaviour, a sudden

decrease in the time-to-failure due to a sudden loading, can be used to explain the triggering

of seismic activity, and aftershock behaviour (see section 6.2.4).

6.2.3. Triggering with the stress corrosion mechanism. Stress corrosion or sub-critical crack

growth is a process widely known in material science (Anderson and Grew 1977, Das and

Scholz 1981, Atkinson 1984, Main 1999, Gomberg 2001). Cracks in a purely brittle material

remain stable under the loading stress below the critical stress determined by the Grifﬁth

criterion. However, under certain environments, especially under high temperatures and with

ﬂuids, a crack can growspontaneously because of weakening near the crack tip due to chemical

‘corrosion’, even if the loading stress is belowthe critical level. In this case, a crack is growing

constantly, and eventually it will reach a critical state where it fails catastrophically. This

mechanism may be important for static triggering of earthquakes in Earth’s crust.

Large amounts of experimental data show that the growth rate of a crack with length x is

generally given by (Atkinson 1984),

dx

dt

= V

0

K

K

0

p

, (6.24)

where t is time, K is the stress intensity factor and p is a constant, usually 5 or larger.

Although most of the experimental data are obtained for tensile cracks, here we apply this

The physics of earthquakes 1485

model to seismic shear cracks. V

0

is the speed of crack growth at t = 0, when K = K

0

. In

the following, we assume that p > 2. K is given by equation (4.3)

K = Yx

1/2

σ, (6.25)

where σ is the loading stress, and Y is a constant determined by the geometry of the crack.

For a constant loading stress σ, (6.24) can be integrated as

x =

x

0

[1 −((p −2)/2)(V

0

/x

0

)t ]

(2/(p−2))

, (6.26)

where x

0

is the crack length at t = 0 (Main 1999). This can be rewritten as

x = x

0

1 +

t

mτ

m

, (6.27)

where τ = (x

0

/V

0

) and m = 2/(2 −p) < 0. Then,

˙ x = V

0

1 +

t

mτ

m−1

. (6.28)

From equations (6.27) and (6.28), the time-to-failure is given by

t

f

= −mτ = −m

x

0

V

0

. (6.29)

Now we consider the case in which the loading stress increases by τ at time t

1

(t

1

< t

f

).

At t = t

1

, the size and the growth rate of the crack are given by (6.27) and (6.28) as

x

1

= x

0

1 +

t

1

mτ

m

(6.30)

and

˙ x

1

= V

0

1 +

t

1

mτ

m−1

≡ V

1

(6.31)

and at this time, the growth rate suddenly increases with the step-wise increase of the load.

From equation (6.24) and (6.25), the relation between the speed just after t

1

, V

+

1

, and just

before t

1

, V

−

1

, is given by

V

+

1

=

¸

1 +

τ

σ

¸

p

V

−

1

. (6.32)

Thus, the time-to-failure measured from t

1

, is

t

/

f

= −

mx

1

V

+

1

(6.33)

and x and ˙ x after t

1

are given by

x = x

1

1 −

t −t

1

t

/

f

m

(6.34)

and

˙ x = V

+

1

1 −

t −t

1

t

/

f

m−1

. (6.35)

As given by equations (6.32) and (6.33), the speed increases suddenly and the time-to-failure

is shortened. Figure 32 shows a typical behaviour of x and ˙ x when the loading is increased

stepwise by τ at t = 0.2t

f

.

The stress corrosion behaviour is very similar to that found for the rate- and state-dependent

friction law(cf ﬁgures 31 and 32.) Asudden increase in loading (e.g. a static stress change due

to an earthquake) can accelerate the crack growth, shorten the time to failure, and contribute

to increase in the seismicity rate.

1486 H Kanamori and E E Brodsky

1

1.2

1.4

1.6

1.8

2

0.15 0.2 0.25

1

10

100

1000

10

4

0.15 0.2 0.25

x/x

0 0

/ x V

.

t / t

f

t / t

f

Figure 32. Non-dimensional crack length x/x

0

(left) and non-dimensional crack extension speed

˙ x/V

0

(right) as a function of non-dimensional time t /t

f

. Anon-dimensional stepwise stress change

τ/σ = 0.15 is imposed at time t /t

f

= 0.2.

6.2.4. Aftershocks and Omori’s Law. After a large earthquake (main shock), many smaller

earthquakes called aftershocks occur near the rupture zone of the earthquake. As ﬁrst

discovered by Omori (1894), the decay of aftershock activity follows a power law, usually

referred to as Omori’s Law. (For recent reviews, see Utsu et al (1995), Kisslinger (1996), Utsu

(2002).)

n(t ) =

K

t + c

, (6.36)

where n(t ) is the number of aftershocks larger than a given magnitude per unit time. Amodiﬁed

(or generalized) Omori’s Law is given by

n(t ) =

K

(t + c)

p

, (6.37)

where p is a constant, which is usually approximately equal to 1. Figure 33 shows two

examples. The ﬁrst one is for the 1891 M ≈ 8 Nobi, Japan, earthquake, for which Omori

found this relationship (Utsu et al 1995). It is already more than 100 years (36 500 days) since

the mainshock, and we can see that the relation (6.37) holds over a very long period of time

(p = 1 (constrained), c = 0.797 day). The second example is for the 1995 Kobe, Japan,

earthquake (Utsu 2002).

Why the aftershock decay follows a power lawgiven by (6.37) has attracted much attention

of many seismologists. Many different mechanisms have been proposed, e.g. post seismic

creep (e.g. Benioff (1951)), ﬂuid diffusion (Nur and Booker 1972), rate- and state-dependent

friction (Dieterich 1994), stress corrosion (Yamashita and Knopoff 1987, Gomberg 2001),

etc. Various mechanisms are reviewed in Utsu (1999). In the following, we summarize the

two recently developed models, the rate- and state-dependent friction model, and the stress

corrosion model.

State- and rate-dependent friction and Omori’s Law. Dieterich (1994) assumed that

seismicity rate is constant under the constant tectonic loading rate ˙ τ. This can be accomplished

The physics of earthquakes 1487

Time, days

Time, days

n(t) n(t)

1000

10

0.1

0.001

1

100

10000

Figure 33. The decay of aftershock activity following the 1891 Nobi, Japan, earthquake, and the

1995 Kobe, Japan earthquake (Utsu 2002).

if earthquake nuclei are distributed such that the time-to-failure, t

f0

(n), of the nth event is given

by nt . In this case, the constant seismicity rate is given by r

0

= (1/t ) (i.e. number of events

per unit time). The sliding speed,

˙

δ

n0

, of the nth nucleus with t

f0

(n) can be given by solving

equation (6.22) as,

˙

δ

n0

=

˙ τ

Hσ

1

exp( ˙ τnt /Aσ

n

) −1

. (6.38)

If the loading is increased by τ due to a mainshock, then, as we discussed in section 6.2.2,

the sliding speed increases step-wise by exp(τ/Aσ

n

), the time-to-failure changes, and

seismicity rate changes. The new time-to-failure, t

f

(n), for nucleus n can be given by

substituting the increased sliding velocity into (6.22). Thus,

t

f

(n) =

Aσ

n

˙ τ

ln

˙ τ

Hσ

n

˙

δ

n0

F

+ 1

, (6.39)

where F ≡ exp(τ/Aσ

n

). Substituting

˙

δ

n0

given by (6.38) in (6.39), and solving for n, we

obtain

n =

Aσ

n

˙ τt

ln

¸

1 + F

¸

exp

˙ τt

f

(n)

Aσ

n

−1

¸¸

. (6.40)

Here, n and t

f

(n) are discrete variables, but we can deﬁne the instantaneous seismicity rate

R by

R =

dn

dt

f

(n)

, (6.41)

taking n and t

f

(n) as continuous variables.

Thus,

R

r

0

=

1

1 −[(F −1)/F] exp(−˙ τt /Aσ

n

)

, (6.42)

where r

0

= 1/t is the background rate and t

f

(n) is now written as a continuous variable, t .

This relation is shown in ﬁgure 34.

1488 H Kanamori and E E Brodsky

10

-1

10

0

10

1

10

2

10

3

10

4

10

5

10

6

10

-8

10

-6

10

-4

10

-2

10

0

10

2

10

4

R

/

r

0

t /(A ) σ

n

/τ

.

Figure 34. Change in seismicity rate plotted as a function of non-dimensional time t /(Aσ

n

/˙ τ),

predicted by the rate- and state-dependent friction. The non-dimensional stress change is assumed

to be τ/(Aσ

n

) = 12.60.

For t → 0, (R/r

0

) = F ≡ exp(τ/Aσ

n

), which represents the sudden increase in

seismicity rate. For times comparable or larger than the timescale of the background stressing

Aσ

n

/˙ τ, the normalized rate (R/r

0

) approaches the steady state value, 1. Between these two

extremes, i.e. for (Aσ

n

/˙ τ)/F < t < (Aσ

n

/˙ τ).

R

r

0

≈

a

1

t + a

2

,

where a

1

= (F/(F −1))Aσ

n

/˙ τ and a

2

= (1/(F −1))Aσ

n

/˙ τ are constants. This is the form

of the Omori’s Law.

A test of this model would be whether the observed aftershock decay follows the trends

at very small and large t predicted by this model (ﬁgure 34). So far, these trends have not

been established observationally. The observational difﬁculties lie in detection thresholds.

Immediately after a large earthquake when t is small, many small earthquakes are missed

in a catalogue because the larger aftershocks mask their waveforms on seismograms. More

progress has been made with the large t limit although it is difﬁcult to measure the duration

of a sequence as the detectable measurement is dependent on the choice of spatial windows

(Gross and Kisslinger 1997).

Stress corrosion model and Omori’s Law. A similar 1/t trend can be predicted with the stress

corrosion model discussed in section 6.2.3. The results presented here are similar in parts

to Gomberg (2001) except that here the equations are formulated to be parallel to the above

derivation of Omori’s Law from rate- and state-dependent friction.

We assume a constant rate loading in the stress corrosion model,

σ = σ

0

+ ˙ τt.

Then substituting this in equations (6.24) and (6.25) which, after integration, leads to

x =

x

0

{1 −(σ

0

V

0

/2x

0

˙ τ)((p −2)/(1 + p))[(1 + ( ˙ τ/σ

0

)t )

p+1

−1]}

2/(p−2)

(6.43)

The physics of earthquakes 1489

and

˙ x =

V

0

(1 + ( ˙ τ/σ

0

)t )

p

{1 −(σ

0

V

0

/2x

0

˙ τ)((p −2)/(1 + p))[(1 + ( ˙ τ/σ

0

)t )

p+1

−1]}

2/(p−2)+1

. (6.44)

From (6.44), the time-to-failure t

f

can be determined as,

t

f

=

σ

0

˙ τ

¸

¸

2x

0

˙ τ(1 + p)

(p −2)σ

0

V

0

+ 1

¸

1(p+1)

−1

¸

. (6.45)

As discussed above, a constant seismicity rate r

0

= (1/t ) can be produced by distributing

earthquake nuclei such that the time-to-failure, t

f0

(n), of nth event is given by nt . The sliding

speed, V

n0

, of the nth nucleus with t

f0

(n) can be given by solving equation (6.45) as,

V

n0

=

2(1 + p)

p −2

x

0

˙ τ

σ

0

1

(( ˙ τ/σ

0

)nt + 1)

p+1

−1

. (6.46)

If the loading is increased by τ at t = 0 due to a mainshock, then, as we discussed

in section 6.2.3 with equation (6.23), the sliding speed increases step-wise by a factor of

F ≡ (1 + (τ/σ

0

))

p

, the time-to-failure changes, and the seismicity rate changes. The new

time-to-failure, t

f

(n), for a nucleus n can be given by substituting the increased sliding velocity,

FV

n0

, into (6.45). Thus,

t

f

(n) =

σ

0

˙ τ

¸

¸

2x

0

˙ τ(1 + p)

(p −2)σ

0

FV

n0

+ 1

¸

1/(p+1)

−1

¸

=

σ

0

˙ τ

¸

¸

1 +

(( ˙ τ/σ

0

)nt + 1)

p+1

−1

F

¸

1/(p+1)

−1

¸

. (6.47)

Solving this for n, we obtain

n =

σ

0

˙ τt

¸

¸

¸

F

¸

˙ τ

σ

0

t

f

(n) + 1

p+1

−1

+ 1

1/(p+1)

−1

¸

¸

¸

. (6.48)

Here, n and t

f

(n) are discrete variables, but we can deﬁne the instantaneous seismicity

rate R by

R =

dn

dt

f

(n)

(6.49)

taking n and t

f

(n) as continuous variables.

Thus,

R

r

0

= F

¸

F + (1 −F)

1 +

˙ τ

σ

0

t

−(p+1)

−p/(p+1)

, (6.50)

where t

f

(n) is now denoted simply by t . For t = 0, R/r

0

= F and for t · σ

0

/˙ τ,

R/r

0

= F

1/(p+1)

.

Because p is large, for t < σ

0

/˙ τ, R/r

0

has the form a

1

/((a

2

+ t )

p/(1+p)

) ≈

a

1

/(a

2

+ t ) which is the Omori’s law, where a

1

= (F/(F −1))(σ

0

/(p + 1) ˙ τ) and a

2

=

(1/(F −1))(σ

0

/(p + 1) ˙ τ).

An example of a solution with typical parameters is shown in ﬁgure 35. The similarity to

the rate and state model solution in ﬁgure 34 suggests that the form of the decay in seismicity

with time is controlled by the similar mathematical construction with both cases. In both cases,

the initial distribution of earthquake nuclei is designed to yield a constant rate under constant

stressing rates.

1490 H Kanamori and E E Brodsky

Figure 35. Change in seismicity rate plotted as a function of non-dimensional time t /(σ

0

/˙ τ),

predicted by the stress corrosion model. The parameters used are p = 40, and τ/σ = 0.5.

One difference between the stress corrosion and the rate- and state-dependent friction

model is that, in the former, the rate returns exactly to the background rate only for inﬁnite p

while, in the latter, it always returns to the background rate. If aftershocks are generated by

sub-critical crack growth, then we must rely on long-term relaxation processes such as viscous

ﬂow to prevent a continual ratcheting upwards of seismicity.

6.2.5. Hydrologic barrier removal. The above mechanisms emphasize the solid mechanics

of earthquake initiation. As discussed in section 2 and illustrated by the observations of

artiﬁcially induced seismicity, ﬂuid movement can reduce the strength on faults and initiate

earthquakes. Recent research has begun exploring quantitative models and new observational

techniques in order to constrain ﬂuid movements and their importance in natural faults. One

example of a ﬂuid-based triggering mechanism comes from recent work on the removal of

transient hydrologic barriers during ground shaking (Brodsky et al 2003). Seismic waves

can induce water ﬂow into faults as the differential stiffness of geological units generates a

hydraulic gradient when the seismic waves impose a long-wavelength, oscillating strain ﬁeld.

Even very small ﬂuid shear stresses ∼1 Pa are sufﬁcient to remove accumulations of sediment

or precipitate (Kessler 1993). The sediment or precipitate barriers blocked ﬂow prior to the

earthquake while maintaining a sharp pressure differential p which, according to the standard

formulation of ﬂow in porous media (Darcy’s Law), is of the order of

p =

U

d

hη

k

,

where U

d

is the average ﬂuid ﬂowvelocity (Darcy velocity), his the thickness of the barrier, η is

the viscosity of the water and k is the permeability of the rock (e.g. Freeze and Cherry (1979)).

When the earthquake occurs, the removal of the barrier redistributes this pressure difference

p in the fault zone. In places where the pressure rises, the frictional stress is reduced (see

equation (2.6)), and failure can occur. For realistic parameters, the pressure changes can be

The physics of earthquakes 1491

Figure 36. Energy release and casualties (number of death) per year during the 20th century. Five-

year running average is taken. The peak in the energy release in 1960 is mainly due to a sequence

of great earthquakes from 1952 to 1964 including the 1960 Chilean earthquake. The peak in the

casualty in the mid 1970s is due to the 1976 Tanshang, China, earthquake (updated from Kanamori

(1978)).

0.04 MPa (Brodsky et al 2003), which is sufﬁcient to trigger earthquakes based on static stress

studies of triggering thresholds in the nearﬁeld (Hardebeck et al 1998).

It is obvious that the ﬂuid-based models are not nearly as well developed quantitatively

as the solid models like rate- and state-dependent friction and stress corrosion. Much more

theoretical, observational and experimental work is necessary to develop the formalism to

compare both the ﬂuid and solid avenues for triggering. However, even at this stage the ﬂuid

approach is able to address some problems, such as sustained distant triggering, that elude

the solid models. Neither stress corrosion nor rate- and state-dependent friction can explain

sustained triggering fromseismic waves (Gomberg 2001). We look forward to the development

of this promising line of research.

7. Conclusions

We have presented in this paper an overview of earthquake physics with an emphasis on

initiation. There are many interesting and active areas of research that we have omitted because

of space (for recent reviews, see Lee et al (2002, 2003), National Research Council (2003)).

Subjects that we neglected include questions of how earthquakes stop, geological studies of

earthquake occurrence histories and methods of mitigating earthquake damage. The latter area

is particularly important for society. Figure 36 shows that casualties from earthquakes are not

correlated to magnitude or overall level of seismicity. They are more closely related to the

engineering infrastructures and preparedness of a region.

The above discussion should have made it clear that the quantitative prediction of

earthquake initiation is an extremely complicated and perhaps impossible task. Even in the best

case scenario of a predictable fault nucleation length, the nucleation length of 1 m requires

instruments to be too densely spaced to be practical. Perhaps one day we will be able to

accomplish accurate earthquake prediction, but the current state of the science implies that that

day is decades, if not centuries away. In the short term, it is more practical to save lives by using

the detailed knowledge we have about the propagation of seismic waves and strength of seismic

shaking to design buildings and infrastructure that will protect people during an earthquake.

Recent engineering advances such as active- and passive-controlled buildings and dense, high-

quality ground motion monitoring brings the goal of saving lives well within our grasp.

At the same time, we continue to build a basic scientiﬁc framework to learn why and

how earthquakes begin. Over the last 10–20 years we have unravelled parts of the puzzle

based on the state of stress in the crust, detailed slip inversions, laboratory friction models,

1492 H Kanamori and E E Brodsky

complex system modelling and triggering studies. A number of still unanswered questions

remain. Many of the questions highlighted during the course of this review will only be

addressed by improved instrumentation and observational techniques. Technical developments

like the recently deployed dense seismic networks of Japan and the Earthscope instrumentation

initiative in the United States may help us measure seismic parameters like stress drop and

radiated energy to sufﬁcient accuracy to ﬁnally address the dynamic similarity of earthquakes.

Another recent advance is the use of non-seismic data to study earthquakes. This review

emphasized elastic waves as the most developed method for characterizing earthquakes, but

insights from geodetic methods are becoming increasingly important. As the subject expands

beyond classic elasticity, we predict that input from geology and hydrogeology will play a

greater role in the decades to come. Combining disciplines might allowus to measure migrating

ﬂuids and if their pressures can equal or exceed the minimum principal stress as suggested

by normal faulting studies. We must also grapple with the thickness and physical properties

of the fault zone. Seismic and other observations will also have to address the heterogeneity

of stress and strength in the crust. Mechanisms for triggering can be differentiated by studies

targeting phenomena where the predictions diverge. The beginning of aftershock sequences,

the existence of very small earthquakes and the occurrence of long-range triggering are a few

areas with some resolving power.

Theorists and experimentalists also have their work cut out for them. Averages like

that taken in estimating the stress drop need to put on ﬁrmer theoretical ground. Theorists and

experimentalists will have to explore the relationship between initiation conditions and rupture

propagation. Is the same physics applicable or does a new set of processes come into play

once rapid slip has begun? What is the physical nature of the fracture energy termthat controls

dynamics ranging from slow quasi-static slip to brittle failure with high rupture speed and

efﬁcient energy radiation? What happens when the mechanisms such as thermal pressurization

and lubrication are combined in a single rupture model? The complexity community must try

to ascertain exactly how chaotic are earthquakes. Even chaotic systems, like weather, can be

predicted over short time horizons if the observations have sufﬁcient resolution. Quantiﬁcation

of the required resolution and the divergence rate from initial conditions would be a valuable

contribution.

As our observational database improves, our computational ability accelerates and our

laboratories become more reﬁned, the next few decades promise to bring more earthquake

insights and perhaps some answers.

Acknowledgments

We thank Allan Rubin and an anonymous reviewer for constructive comments.

References

Abercrombie R 1995 J. Geophys. Res. 100 24003–14

Abercrombie R and Leary P 1993 Geophys. Res. Lett. 20 1511–4

Aki K 1966 Bull. Earthquake Res. Inst. Tokyo Univ. 44 73–88

Aki K and Richards P G 2002 Quantitative Seismology (Sausalito: University Science Books) p 685

Anderson E M 1905 Trans. Edin. Geol. Soc. 8 393

Anderson E M 1951 The Dynamics of Faulting and Dyke Formation with Applications to Britain (Edinburgh: Oliver

and Boyd) p 206

Anderson O L and Grew P C 1977 Rev. Geophys. 15 77–104

Andrews D J 2002 J. Geophys. Res. 107 (B2) 2363

Atkinson B K 1984 J. Geophys. Res. 89 4077–114

The physics of earthquakes 1493

Axen G J 2004 Low-angle normal fault mechanics and crustal strength Rheology and Deformation in the Lithosphere

at Continental Margins ed G Karner et al (New York: Columbia University Press) pp 46–91

Bak P and Tang C 1989 J. Geophys. Res. 94 15635–7

Bak P et al 1988 Phys. Rev. A 38 364–74

Barenblatt G I 1962 The mathematical theory of equilibrium cracks in brittle fracture Advances in Applied Mechanics

vol 7 (London: Academic) pp 55–125

Beeler N M et al 1994 Geophys. Res. Lett. 21 1987–90

Benioff H 1951 Bull. Seismol. Soc. Am. 41 31–62

Beroza G C and Mikumo T 1996 J. Geophys. Res. 101 22449–60

Bird P and Kagan Y Y 2004 Bull. Seismol. Soc. Am. at press

Boatwright J and Choy G L 1986 J. Geophys. Res. 91 2095–112

Boatwright J et al 2002 Bull. Seismol. Soc. Am. 92 1241–55

Bouchon M 1997 J. Geophys. Res. 102 11731–44

Bouchon M and Vallee M 2003 Science 301 824–6

Brace W F and Byerlee J D 1966 Science 153 990–2

Brodsky E E and Kanamori H 2001 J. Geophys. Res. 106 16357–74

Brodsky E E et al 2000 Geophys. Res. Lett. 27 2741–4

Brodsky E E et al 2003 J. Geophys. Res. 108

Brune J N 1970 J. Geophys. Res. 75 4997–5009

Burridge R and Knopoff L 1964 Bull. Seismol. Soc. Am. 54 1875–88

Burridge R and Knopoff L 1967 Bull. Seismol. Soc. Am. 57 341–71

Byerlee J 1978 Pure Appl. Geophys. 116 615–26

Byerlee J 1992 Tectonophysics 211 295–303

Cardwell R K et al 1978 Geophys. J. R. Astron. Soc. 52 525–30

Chester F M and Chester J S 1998 Tectonophysics 295 199–221

Choy G L and Boatwright J L 1995 J. Geophys. Res. 100 18205–28

Dahlen F A 1977 Geophys. J. R. Astron. Soc. 48 239–61

Dahlen F A and Tromp J 1998 Theoretical Global Seismology (Princeton, NJ: Princeton University Press) p 1025

Das S 1988 Bull. Seismol. Soc. Am. 78 924–30

Das S and Scholz C H 1981 J. Geophys. Res. 86 6039–51

Dieterich J 1994 J. Geophys. Res. 99 2601–18

Dieterich J H 1979 J. Geophys. Res. 84 2161–8

Dmowska R and Rice J R 1986 Fracture theory and its seismological applications Theories in Solid Earth Physics

ed R Teisseyre (Warzawa: PWN-Polish Publishers) pp 187–255

Dragert H et al 2001 Science 292 1525–8

Dugdale D S 1960 J. Mech. Phys. Solids 8 100–4

Eberhart-Phillips D et al 2003 Science 300 1113–18

Ekstr¨ om E 2000 Global studies of earthquakes Problems in Geophysics for the New Millennium ed G E E Boschi and

A Morelli (Bologna: Editorice Compositori) pp 111–24

Eshelby J D 1969 J. Mech. Phys. Solids 17 177–99

Fossum A F and Freund L B 1975 J. Geophys. Res. 80 3343–7

Freeze R A and Cherry J A 1979 Groundwater (Englewood Cliffs, NJ: Prentice Hall)

Freund L B 1972 J. Elasticity 2 341–9

Freund L B 1989 Dynamic Fracture Mechanics (Cambridge: Cambridge University Press) p 563

Gilbert F and Dziewonski A M 1975 Phil. Trans. R. Soc. Lond. 278 187–269

Gomberg J 2001 J. Geophys. Res. 106 16253–63

Gomberg J et al 2001 Nature 411 462–6

Gomberg J et al 2003 Bull. Seismol. Soc. Am. 93 118–38

Grifﬁth A A 1920 Phil. Trans. R. Soc. Lond. A 221 169–98

Gross S and Kisslinger C 1997 J. Geophys. Res. Solid Earth 102 7603–12

Guha S K 2000 Induced Earthquakes (London: Kluwer) p 314

Gutenberg B and Richter C F 1941 Geol. Soc. Am. 34 1–131 (special paper)

Hardebeck J L and Hauksson E 2001 J. Geophys. Res. 106 21859–82

Hardebeck J L et al 1998 J. Geophys. Res. 103 24427–37

Harris R A 2002 Stress triggers, stress shadows, and seismic hazard International Handbook of Earthquake &

Engineering Seismology part B, ed H Kanamori et al (San Diego, CA: Academic) pp 1217–32

Haskell N 1964 Bull. Seismol. Soc. Am. 56 1811–42

Heaton T 1990 Phys. Earth Planet. Inter. 64 1–20

1494 H Kanamori and E E Brodsky

Hergaten S 2002 Self-Organized Criticality in Earth Systems (Berlin: Springer) p 272

Hill D P et al 2002 Phys. Today 55 41–7

Hill D P et al 1993 Science 260 1617–23

Houston H 2001 J. Geophys. Res. 106 11137–50

Hubbert M K and Rubey W W 1959 Geol. Soc. Am. 70 115–66 (Special paper)

Husseini M I 1977 Geophys. J. R. Astron. Soc. 49 699–714

Ida Y 1972 J. Geophys. Res. 77 3796–805

Ide S and Beroza G C 2001 Geophys. Res. Lett. 28 3349–52

Ide S and Takeo M 1997 J. Geophys. Res. 102 27379–91

Ikeda R et al 2001 Isl. Arc. 10 252–60

Izutani Y and Kanamori H 2001 Geophys. Res. Lett. 28 4007–10

Jaeger J C and Cook N G W 1979 Fundamentals of Rock Mechanics (London: Chapman and Hall) p 593

Jeffreys H 1942 Geol. Mag. 79 291–5

Johnston M J S et al 1995 Bull. Seismol. Soc. Am. 85 787–95

Jost M L et al 1998 Bull. Seismol. Soc. Am. 88 815–32

Kadanoff L P 1991 Phys. Today 44 9–10

Kanamori H 1978 Nature 271 411–14

Kanamori H 1983 Global seismicity Earthquakes: Observation, Theory and Interpretation ed H Kanamori and

E Boschi (New York: North-Holland) pp 596–608

Kanamori H and Anderson D L 1975 Bull. Seismol. Soc. Am. 65 1073–95

Kanamori H et al 1998 Science 279 839–42

Kanamori H and Brodsky E E 2001 Phys. Today 54 34–40

Kanamori H et al 1993 Bull. Seismol. Soc. Am. 83 330–46

Kessler J H 1993 Berkeley University of California, Berkeley

Kikuchi M 1992 Tectonophysics 211 107–13

Kikuchi M and Fukao Y 1988 Bull. Seismol. Soc. Am. 78 1707–24

Kikuchi M and Kanamori H 1994 Geophys. Res. Lett. 21 2341–4

Kikuchi M and Kanamori H 1995 Pure Appl. Geophys. 144 441–53

Kilb D et al 2000 Nature 408 570–4

King G C P et al 1994 Bull. Seismol. Soc. Am. 84 935–53

Kinoshita S and Ohike M 2002 Bull. Seismol. Soc. Am. 92 611–24

Kisslinger C 1996 Aftershocks and fault-zone properties Advances in Geophysics vol 38, ed R Dmowska and

B Saltzman (San Diego, CA: Academic) pp 1–36

Knopoff L 1958 Geophys. J. 1 44–52

Kostrov B V 1966 J. Appl. Math. Mech. 30 1241–8 (PMM) (Engl. tans.)

Kostrov B V 1974 Izv. Earth Phys. 1 23–40

Lachenbruch A H 1980 J. Geophys. Res. 85 6097–112

Lachenbruch A H and Sass J H 1980 J. Geophys. Res. 85 6185–222

Lapusta N and Rice J R 2003 J. Geophys. Res. 108 (B4) 2205

Lawn B 1993 Fracture of Brittle Solids 2nd edn (Cambridge: Cambridge University Press) p 378

Lay T and Wallace T C 1995 Modern Global Seismology (San Diego, CA: Academic) p 521

Lee W H K et al 2002 International Handbook of Earthquake & Engineering Seismology part A (San Diego, CA:

Academic) pp 1–933

Lee W H K et al 2003 International Handbook of Earthquake & Engineering Seismology part B (San Diego, CA:

Academic) pp 937–1945

Li V C 1987 Mechanics of shear rupture applied to earthquake zones Fracture Mechanics of Rock (London: Acdemic)

pp 351–428

Linker M and Dieterich J H 1992 J. Geophys. Res. Solid Earth 97 4923–40

Lockner DA1995Rockfailure inrockphysics &phase relations AHandbook of Physical Constants vol 3, edTJ Ahrens

(Washington, DC: American Geophysical Union) pp 127–47

Lockner DAandBeeler NM2002Rockfailure andearthquakes International Handbook of Earthquake &Engineering

Seismology part A, ed W H K Lee et al (San Diego, CA: Academic) pp 505–37

Ma K F et al 2003 Geophys. Res. Lett. 30 (5) 1244

Ma K-F et al 1999 EOS Trans.—AGU 80 605

Madariaga R 1977 Pure Appl. Geophys. 115 301–16

Madariaga R 1979 J. Geophys. Res. 84 2243–50

Madariaga R and Olsen K B 2002 Earthquake dynamics International Handbook of Earthquake & Engineering

Seismology part A, ed W H K Lee et al (San Diego, CA: Academic) pp 175–94

The physics of earthquakes 1495

Main I 1996 Rev. Geophys. 34 pp 433–62

Main I G 1999 Geophys. J. Int. 139 F1–6

Marder M and Fineberg J 1996 Phys. Today 49 24–9

Marone C and Kilgore B 1993 Nature 362 618–21

Maruyama T 1964 Bull. Earthquake Res. Inst. Tokyo Univ. 42 289–368

Mase C W and Smith L 1985 Pure Appl. Geophys. 122 583–607

Mase C W and Smith L 1987 J. Geophys. Res. 92 6249–72

Massonnet D et al 1993 Nature 364 138–42

Mayeda K and Walter W R 1996 J. Geophys. Res. 101 11195–208

McGarr A and Fletcher J B 2002 Bull. Seismol. Soc. Am. 92 1633–46

McGarr A et al 2002 Case histories of induced and triggered seismicity International Handbook of Earthquake &

Engineering Seismology part A, ed W H K Lee et al (San Diego, CA: Academic) pp 647–61

McKenzie D P and Brune J N 1972 Geophys. J. R. Astron. Soc. 29 65–78

Mikumo T and Miyatake T 1993 Geophys. J. Int. 112 481–96

Mikumo T et al 2003 Bull. Seismol. Soc. Am. 93 264–82

Miyatake T 1992 Geophys. Res. Lett. 19 1041–4

Mott N F 1948 Engineering 165 16

National Research Council 2003 Living on an Active Earth (Washington, DC: National Academies) p 418

Nur A and Booker J 1972 Science 175 885–7

Ohnaka M and Shen L-F 1999 J. Geophys. Res. 104 817–44

Omori F 1894 J. College of Science Imperial University of Tokyo 7 111–200

Otsuka M 1971 J. Seismol. Soc. Japan 24 215–27

Palmer A C and Rice J R 1973 Proc. R. Soc. Lond. Ser. A—Math. Phys. Eng. Sci. 332 527–48

Perez-Campos X and Beroza G C 2001 J. Geophys. Res. 106 11127–36

Prejean S G and Ellsworth W L 2001 Bull. Seismol. Soc. Am. 91 165–77

Prejean S G et al 2004 Bull. Seismol. Soc. Am. 94 at press

Provost A S and Houston H 2003 J. Geophys. Res. 108 (B3) 2175

Quin H 1990 Tectonophysics 175 93–117

Rabinowicz E 1995 Friction and Wear of Materials (New York: Wiley) p 315

Raleigh C B et al 1976 Science 191 1230–7

Rice J R 1980 The mechanics of earthquake rupture Physics of the Earth’s Interior ed A M Dziewonski and E Boschi

(Amsterdam: North-Holland) pp 555–649

Rice J R1992 Fault stress states, pore pressure distributions, and the weakness of the San Andreas fault Fault Mechanics

and Transport Properties of Rock ed B Evans and T-F Wong (San Diego, CA: Academic) pp 475–503

Richards P G 1976 Bull. Seismol. Soc. Am. 66 1–32

Richardson E and Jordan T H 2002 Bull. Seismol. Soc. Am. 92 1766–82

Romanowicz B and Ruff L J 2002 Geophys. Res. Lett. 29 (12) 1604

Romanowicz B and Rundle J B 1993 Bull. Seismol. Soc. Am. 83 1294–7

Rosakis A J 2002 Adv. Phys. 51 1189–257

Rubin A M et al 1999 Nature 400 635–41

Rudnicki J W and Kanamori H 1981 J. Geophys. Res. 86 1785–93

Ruina A L 1983 J. Geophys. Res. 88 10359–70

Rundle J B et al 2000 GeoComplexity and the Physics of Earth (Washington, DC: American Geophysical Union)

p 284

Saffer D M et al 2003 J. Geophys. Res. 108 (B5) 2274

Sass J H et al 1997 J. Geophys. Res. 102 27575–85

Scholz C H 1994 Bull. Seismol. Soc. Am. 84 215–18

Scholz C H 2002 The Mechanics of Earthquakes and Faulting (New York: Cambridge University Press) p 471

Sibson R H 1973 Nature 243 66–8

Sibson R H 1985 J. Struct. Geol. 7 751–4

Sibson R H 2003 Bull. Seismol. Soc. Am. 93 1169–78

Sibson R H and Xie G Y 1998 Bull. Seismol. Soc. Am. 88 1014–22

Simons M et al 2002 Bull. Seismol. Soc. Am. 92 1390–402

Singh S K et al 2004 Bull. Seismol. Soc. Am. 94 at press

Stein R S 1999 Nature 402 605–9

Stein R S et al 2003 Sci. Am. 288 72–9

Stekette J A 1958 Can. J. Phys. 36 1168–98

Tibi R et al 2003 J. Geophys. Res. 108 (B2) 2091

1496 H Kanamori and E E Brodsky

Tsuboi C 1933 Japan. J. Astron. Geophys. 10 93–248

Turcotte D L 1997 Fractals and Chaos in Geology and Geophysics (Cambridge, UK: Cambridge University Press)

p 398

Turcotte D L and Malamud B D 2002 Earthquakes as a complex system International Handbook of Earthquake &

Engineering Seismology part A, ed W H K Lee et al (San Diego, CA: Academic) pp 209–35

Utsu T 1999 Seismicity Studies: A Comprehensive Review (Tokyo: University of Tokyo Press) p 876 (in Japanese)

Utsu T 2002 Statistical features of seismicity International Handbook of Earthquake & Engineering Seismology

part A, ed W H K Lee (San Diego, CA: Academic) pp 719–32

Utsu T et al 1995 J. Phys. Earth 43 1–33

Uyeda S 1978 The New View of the Earth (San Francisco: W H Freeman and Company) p 217

Venkataraman A and Kanamori H 2004 J. Geophys. Res. 109 B05302, 10.1029/2003JB002549

Venkataraman A et al 2002 Bull. Seismol. Soc. Am. 92 1256–65

Wald D J and Heaton T H 1994 Bull. Seismol. Soc. Am. 84 668–91

Wernicke B 1981 Nature 291 645–8

Wyss M and Brune J N 1968 J. Geophys. Res. 73 4681–94

Yamashita T and Knopoff L 1987 Geophys. J. 91 13–26

Zeng Y H et al 1994 Geophys. Res. Lett. 21 725–8

Zoback M D and Healy J H 1984 Ann. Geophys. 2 689–98

Zoback M D et al 1987 Science 238 1105–11

1430

H Kanamori and E E Brodsky

Contents

1. 2.

3.

4.

List of frequently used symbols Introduction Earthquakes and stress in the crust 2.1. Plate motion and earthquake repeat times 2.2. The state of stress in the crust Principal stresses and fault orientation Strength of the crust: laboratory and ﬁeld data Conﬂicting observations? Summary Quantifying earthquakes 3.1. Earthquake source parameters and observables A formal description of the elastic problem 3.1.1. Seismic source and displacement ﬁeld 3.1.2. Seismic moment and magnitude 3.1.3. Strain and stress drop 3.1.4. Energy Radiated energy, ER Potential energy 3.1.5. Rupture mode, speed and directivity Directivity and source duration Rupture speed 3.1.6. Earthquake rupture pattern 3.2. Seismic scaling relations 3.2.1. Scaling relations for static parameters 3.2.2. Scaling relations for dynamic parameters Rupture processes 4.1. Fracture mechanics 4.1.1. An overview of the crack model 4.1.2. Crack tip breakdown-zone 4.1.3. Stability and growth of a crack Static crack Dynamic crack Rupture speed 4.2. Frictional sliding 4.2.1. Static and kinetic friction 4.2.2. Rate- and state-dependent friction 4.3. The link between the crack model and the friction model Direct determination of Dc 4.4. Rupture energy budget 4.5. Fault-zone processes: melting, ﬂuid pressurization and lubrication Melting

Page 1432 1433 1435 1435 1436 1438 1439 1440 1441 1441 1442 1442 1443 1445 1446 1447 1447 1448 1449 1449 1449 1450 1451 1451 1454 1455 1455 1455 1457 1457 1459 1459 1459 1460 1461 1461 1463 1463 1463 1466 1466

The physics of earthquakes

1431

Thermal ﬂuid pressurization Lubrication 4.6. Linking processes to the seismic data 4.6.1. The interpretation of macroscopic seismological parameters Radiation efﬁciency The relation between radiation efﬁciency and rupture speed Summary and implications 5. Earthquakes as a complex system The magnitude–frequency relationship (the Gutenberg–Richter relation) Simple models 6. Instability and triggering 6.1. Instability 6.1.1. Stick slip and instability Stiffness of the fault system 6.1.2. Nucleation zone 6.2. Triggering 6.2.1. Observations 6.2.2. Triggering with the rate- and state-dependent friction mechanism Spontaneous behaviour Loading at a uniform rate Stepwise change in loading 6.2.3. Triggering with the stress corrosion mechanism 6.2.4. Aftershocks and Omori’s Law State- and rate-dependent friction and Omori’s Law Stress corrosion model and Omori’s Law 6.2.5. Hydrologic barrier removal 7. Conclusions Acknowledgments References

1466 1468 1468 1468 1468 1471 1471 1473 1473 1474 1476 1476 1476 1478 1478 1479 1479 1482 1483 1483 1483 1484 1486 1486 1488 1490 1491 1492 1492

seismic efﬁciency radiation efﬁciency stress intensity factor fracture toughness (critical stress intensity factor) stiffness of spring. power of Omori’s Law heat seismicity rate background seismicity rate density fault area initial stress .and state-friction law half-length of Mode III crack P-wave speed constant in the rate.and state-friction law slope of the earthquake magnitude–frequency relationship S-wave speed probability of earthquake occurrence fault slip offset average offset critical slip of a crack critical displacement in the slip-weakening models slip on a frictional surface slip speed elastic modulus radiated seismic energy fracture energy of the earthquake thermal energy (frictional energy loss) of the earthquake scaled energy (the ratio of radiated seismic energy to seismic moment) dynamic energy release rate (dynamic crack extension force) static energy release rate (static crack extension force. permeability stiffness of the fault length scale of the fault nucleation length crack breakdown length seismic moment earthquake magnitude (moment magnitude) rigidity (shear modulus) or coefﬁcient of friction coefﬁcient of static friction coefﬁcient of kinetic friction pore pressure. power of the stress–corrosion relation. speciﬁc fracture energy) critical speciﬁc fracture energy surface energy viscosity.1432 H Kanamori and E E Brodsky List of frequently used symbols A a α B b β χ D ¯ D D0 Dc δ ˙ δ E ER EG EH e ˜ G G∗ G∗ c γ η ηR K Kc k kf ˜ L ˜ Ln l0 M0 Mw µ µs µk p Q R r0 ρ S σ0 constant in the rate.

and state-dependent friction. this simple answer is far from complete. σ2 . We begin this review with a discussion of stress in the crust followed by an overview of earthquake phenomenology. angle between the fault and the maximum compressional stress displacement vector rupture speed initial (before an earthquake) potential energy of the Earth ﬁnal (after an earthquake) potential energy of the Earth change in the potential energy change in the potential energy minus frictional energy width of the fault slip zone 1. We then review the major processes thought to be active during rupture and discuss their relationship to the observable parameters. source duration average source duration stress rate state variable in rate. We then take a longer range view by discussing how earthquakes interact as a complex system. we gloss over many exciting and important advances in recent years ranging from the discovery of periodic slow slip events (Dragert et al 2001) to the elucidation of fault structure revealed by new accurate location techniques (Rubin et al 1999). This concluding discussion will require using the processes introduced in the study of rupture. nearly rigid plates of the Earth slide past each other. we combine subjects to approach the key issue of earthquake initiation. We brieﬂy discuss how these parameters are related to the amplitude and frequencies of the elastic waves measured by seismometers as well as direct geodetic measurements of the Earth’s deformation. Earthquakes accommodate the motion (ﬁgure 1). However. Large. σ3 ) σY σn τ τ ¯ τ ˙ θ ui V W0 W1 W W0 w ﬁnal stress (sections 3 to 6) frictional stress static stress drop (σ0 − σ1 ) stress tensor principal stresses (section 2) yield stress normal stress shear stress. Why is so little motion accommodated by anything in between these two extremes? Why do some earthquakes stop after only a few hundred metres while others continue rupturing for a thousand kilometres? How do nearby earthquakes interact? Why are earthquakes sometimes triggered by other large earthquakes thousands of kilometres away? Earthquake physicists have attempted to answer these questions by dissecting observable phenomena and separating out the quantiﬁable features for comparison across events. In this introductory review for non-specialists. Many of these recent advances are made possible by new technology . Some plate boundaries glide past each other smoothly.The physics of earthquakes 1433 σ1 σf σs σij (σ1 . Finally. while others are punctuated by catastrophic failures. as well as some novel mechanisms. Introduction Why do earthquakes happen? This age-old question was solved at one level by the plate tectonics revolution in the 1960s. focusing on the parameters that are readily measured by current seismic techniques.

3. relocated data from the International Seismological Center catalogue) and plate motion.1–7.1434 H Kanamori and E E Brodsky Figure 1.3. Earthquakes occur at the boundaries between rigid plates of the Earth’s surface that move in different directions (from Uyeda (1978)). magnitude range. Global seismicity (from 1 January 1964 to 31 December 1995. .

2. Based on the above process. earthquakes do not generally occur at regular intervals in time or space. Inevitably. Where possible. A large earthquake on a segment of a fault changes the stress on the adjacent segments. while at other times. it is of the order of 3 × 10−5 –3 × 10−4 (see sections 3. Figure 2 shows the result of the recent geodetic measurements in Southern California. static stress drop) of about 1–10 MPa. either statically or dynamically. Since the rigidity of the crustal rocks. cutting-edge topics. As the plates slide past each other. µ. This is the basic long-term process that governs global earthquake activity. The approach we take here is to emphasize the features of classical theory that are directly applicable to current. refute or extend elements of the classical physics-based paradigm.3 and 3. The strain also accumulates in plate interiors. Although the basic process illustrated here is well understood and accurately measured. the details are more complex. These values agree with what have been observed at many plate boundaries and interiors.1. Stress builds up on a fault plane until it reaches the breaking strength of the rock. Dividing the coseismic strain drop by the strain rate suggests that the repeat times of major earthquakes at a given place are about 100–1000 years on plate boundaries. but at a much slower rate about 3 × 10−8 per year or less. For example.2. Then. Sometimes. a simple sketch of the stresses generating earthquakes can be drawn (ﬁgure 3(a)). This value is at least an order of magnitude smaller than that associated with breaking intact rocks in laboratory. 2. at rates of the order of millimetres per year. The relative plate motion determined from these data is about 2–7 cm per year which translates into a strain rate of approximately 3 × 10−7 per year along plate boundaries. Current estimates are that about 80% of relative plate motion on continental boundaries is accommodated in rapid earthquakes (Bird and Kagan 2004). our examples tend to be biased towards our own interests and research. Plate motion and earthquake repeat times The long-term loading of the Earth’s crust has been traditionally measured using geodetic and geological methods.e. relative motion is sometimes accommodated by a relatively constant gradual slip. the loading rate is not uniform in time. For large earthquakes. and 1000–10 000 years within plates.1. we highlight modern observations and laboratory results that conﬁrm. is about 3 × 104 MPa. the accumulated strain is released in earthquakes with slip rates of the order of metres per second. Earthquakes and stress in the crust Earthquakes are a mechanism for accommodating large-scale motion of the Earth’s plates.The physics of earthquakes 1435 such as satellite geodesy and high-power computation. the stress is relaxed and a new cycle begins. we must return to and push the boundaries of classical mechanical theories. which is an order of magnitude smaller than that at plate boundaries. The recent progress in space-based geodesy such as the Global Positioning System (GPS) and satellite interferometry (InSAR) provides us with a clear pattern of crustal movement and strain accumulation. The shear strain change associated with large earthquakes (called coseismic strain drop) has been estimated using geodetic and seismological methods. . which is several hundred MPa. slip is accommodated by slow earthquakes or creep events with velocities of the order of centimetre per month between the two extreme cases. an earthquake occurs. Geodesy is the branch of geophysics concerned with measuring the size and shape of the Earth’s surface. With few exceptions. We hope that this review will equip the reader to be properly sceptical of our results. this corresponds to a change in shear stress (i.1). In order to interpret the new technological advances.

the simplest model for earthquake initiation is to assume that when the stress accumulated in the plates exceeds some failure criterion on a fault plane. Evaluating this criterion requires both a measure of the resolved stress on the fault plane and a quantiﬁable model for the failure threshold. He started with the fact that any stress ﬁeld can be completely described by its principal stresses. The stress drop during earthquakes may also vary from event to event.v2. and accelerates or decelerates seismic activity depending on the fault geometry. 2. He then proposed that: (1) the stress state could be resolved by assuming that one principal stress is vertical since the Earth’s surface is a free surface and (2) faulting occurs when the resolved shear stress exceeds the internal friction on some plane in the medium. 1951). which are given by the eigenvectors of the stress tensor and are interpretable as the normal stresses in three orthogonal directions. http://www. A ﬁrst-order evaluation of the problem dates to the groundbreaking work of Anderson (1905. Fluids may migrate in the Earth’s crust.2. Thus. Red lines (in the electronic version) indicate active faults. The strength of the crust is not constant in time either.scec. The ﬁgure is part of the Southern California Earthquake Center’s web-site.1436 H Kanamori and E E Brodsky Figure 2.org/group e/release. although the overall long-term process is regular. The state of stress in the crust As outlined earlier. Velocity vectors in Southern California determined by the GPS and other space-based methods.scecdc. thereby weakening the crust signiﬁcantly and affecting the occurrence time of earthquakes. Internal friction is deﬁned analogously with conventional sliding friction as a shear stress proportional to the . Figure 3(b) illustrates these complications schematically and their effect on the intervals between earthquakes. considerable temporal ﬂuctuations of seismicity are expected. which makes accurate prediction of earthquakes difﬁcult. an earthquake happens.

normal stress on a plane. However. (a) Regular sequence. Each of these three regimes corresponds to a particular orientation of the maximum principal stress. A fault that has only horizontal motion is called ‘strike-slip’. as we show below. In this framework. (c) Normal faulting: the maximum principal stress is vertical (ﬁgure from Jaeger and Cook (1979) p 426). faults are expected to accommodate horizontal motion if the vertical axis is the intermediate principal stress and accommodate both vertical and horizontal motion otherwise. (a) (b) (c) Figure 4. (b) Irregular sequence caused by the changes in loading rate and temporal variations in the strength of crust. Andersonian faulting theory has been remarkably successful in predicting and explaining the occurrence and geometry of faults. (a) Thrust faulting: the minimum principal stress is vertical. (b) Strike-slip faulting: the intermediate principal stress is vertical.The physics of earthquakes 1437 Figure 3. Stress changes and earthquake sequence. The principal stresses are σ1 > σ2 > σ3 . a few contradictory observations cast doubt on enough parts of the paradigm that it is difﬁcult to apply to . Schematic of the orientation of the principal stresses and the corresponding type of faulting. Combined vertical and horizontal extensional motion is called ‘normal’ faulting while vertical and horizontal compressional motion is called ‘thrust’ faulting (ﬁgure 4).

We have difﬁculty measuring the coefﬁcient of friction in the crust and have reason to believe that it varies signiﬁcantly in time and space. The failure criterion in the presence of pore ﬂuid is the solid line (2. The convention in rock mechanics is that positive values of stresses are compressional. Given principal stress magnitudes σ1 and σ3 . respectively. τ . Since rocks are weak under tension. (2. Mohr circle diagram. by σ1 − σ3 sin 2θ (2. where by deﬁnition σ1 > σ2 > σ3 .1) and (2.2) which is plotted as the circle. Failure on a plane at an angle θopt from the orientation of σ1 occurs if the circle intersects the failure line as it does at the ∗. it is generally assumed that all three principal stresses must be positive in the Earth. σ . yet the precise values of the ever-changing ﬂuid pressures are also difﬁcult to measure deep within the crust. We will use the formalism to relate the observed geometry of faulting to the frictional strength of faults.2) cos 2θ + 2 2 A Mohr circle diagram is a plot of these two resolved stresses.3) is the dashed line. <10% the compressional strengths (Lockner 1995).1438 H Kanamori and E E Brodsky Figure 5.e. σ2 and σ3 .1) τ= 2 and σ1 − σ3 σ1 + σ3 σ =− .7). The normal stress is on the x-axis and the shear stress is on the y-axis (Jaeger and Cook 1979). and the normal stress. The failure criterion (2. For a given set of principal . The evidence also suggests that high ﬂuid pressures are important in controlling frictional behaviour. Denoting the principal stresses by σ1 . i. The shear stress. Principal stresses and fault orientation. earthquakes in a straightforward way. Inset shows the deﬁnition of θ . Below we develop the formalism to quantitatively evaluate the frictional failure criterion in terms of the principal stresses. the locus of possible combinations of shear and normal stresses resolved on a plane are given by (2. on the fault plane at an angle θ to σ1 are given. the relationships between the principal stresses and the resolved shear stress on a plane at an angle θ to the maximum principal stress (σ1 ) can be written analytically and depicted with a Mohr circle diagram (ﬁgure 5). tensional strengths are usually <20 MPa.

In this case. (2. is more straightforward to measure. (2.1) to (2. the slip can be constrained to occur on that plane.85 for the majority of rocks (Byerlee 1978). the minimum diameter of the Mohr’s circle. is known as the lock-up angle. dip angles) should be about 25–30˚ for thrust and 60–65˚ for normal faults if they are optimally oriented. the solutions to equations (2. The pressure at which the wall of the borehole fractures and the orientation of the resulting fracture give a . θlu . A more complete description includes the cohesive stress C in the shear stress. Since real rocks are seldom intact. equation (2.6) µ If a fault is observed to lie at an angle θ to the maximum principal stress when it is slipping. (2.e. i.e.3) where µ is the coefﬁcient of friction. We will return to these unusual cases below. such that (ﬁgure 5) 1 tan 2θopt = ± . Laboratory studies of rocks show that at the depths typical of earthquakes µ = 0. slip can occur at angles larger than the optimal angle θopt .The physics of earthquakes 1439 stresses. They found that 40% of the faults fall into the optimal range and none of their study sites violated the lock-up criterion. (2.2) fall on a circle (ﬁgure 5). Therefore. respectively (ﬁgure 4). θlu = tan−1 Strength of the crust: laboratory and ﬁeld data. intact rock.e. Guillaume Amonton ﬁrst established that the shear traction between two surfaces is proportional to the load. In general. µ bound on the value of µ on the fault. Each point on the circle represents a particular fault orientation. However. Amonton’s Law for friction on a plane between two surfaces is written in modern terms as τ = µσ. there is a maximum value of θ beyond which slip cannot occur for any combination of positive stresses (Sibson 1985). τ = µσ + C. i. In the 17th century. 1 + µ cot θ σ1 .4) predicts that faults should form at angles of 25–30˚ to the maximum principal stress σ1 . only a handful of faults anywhere have been found to exceed the lock-up criterion. therefore most studies use an effective coefﬁcient of friction µ which includes the cohesive effects (Lockner and Beeler 2002). Therefore.6 to 0. for a given coefﬁcient of friction µ on the weak plane. and the observation gives a maximum then θ θlu = tan−1 (1/µ). Boreholes are drilled and pumped full of high-pressure ﬂuid. the more important criterion is the lock-up angle.5) = σ3 1 − µ tan θ The lock-up angle is the maximum value of θ that satisﬁes equation (2. Sibson and Xie (1998) check this criterion for the special case of intraplate thrusts. unﬂawed. i. The angle OO Q in the diagram is 2θ.e.4) µ These two angles θopt are known as the optimal angles because they are the angles at which the rock will fracture in homogeneous. For positive values of σ1 and σ3 . the angles between the faults and the horizontal surface (i. Because σ1 is horizontal and vertical for thrust and normal faults. Therefore. The maximum angle. exists in the crust. The fault planes on which slip can occur with the minimum possible deviatoric stress σ1 − σ3 .1) and (2. are the planes inclined at angles θopt to σ1 . the ratio of shear stress to normal stress. If a weak plane. such as a fault. τ/σ . tan θ .3). However.5). the solution exists only if the denominator is positive. From (2. The predictions of the Anderson–Byerlee mechanics have also been supported by ﬁeld experiments. 1 = 2θopt . tan θ 1/µ. if they are optimally oriented.

Since this stress is lower than that which can be achieved with hydrostatic pore pressure and Byerlee friction. They connected this observation with studies of faulting and proposed that the pore pressure p at a depth can approach the normal stress σ on faults. Low-angle normal faults have now been robustly documented in the geological record (e. Fluid pressure at a certain depth should theoretically be determined by the weight of the water column above. the observed values of σ1 and σ3 from in situ borehole experiments and the measured value of µ on rock samples from the site.85 and hydrostatic ﬂuid pressure is called a strong fault. where the maximum principal stress is vertical. The most often cited evidence against the strong fault hypothesis is based on heat ﬂow data.7). i. In the presence of ﬂuids equation (2. This heat generation. A fault that fails according to equation (2.2) and (2.17. The most spectacular support for the importance of the Anderson–Byerlee paradigm of failure as modiﬁed by Hubbert and Rubey came from the 1976 Rangeley experiment. the frictional stress on the fault should generate heat. If the pore pressure is hydrostatic. Lachenbruch and Sass (1980) showed that the San Andreas fault generates no observable perturbation to the regional heat ﬂow pattern. (2. Therefore. these difﬁcult heat ﬂow observations stand as the best evidence that the San Andreas has a low resolved depth-averaged shear stress ( 20 MPa). The heat ﬂow observations can not distinguish between high pore pressure and low intrinsic fault friction. rather than the value of µ. If µ is high. averaged over geological time should make a resolvably high level of heat ﬂow if the depthaveraged shear stress is greater than 20 MPa. the researchers successfully predicted the increase in pore pressure that is necessary to trigger earthquakes.6 (Zoback and Healey 1984). the fault is weak according to the deﬁnition at the beginning of this section.g. but recent modelling has shown that the data are inconsistent with any known method of removing the heat from the fault (Saffer et al 2003). Some authors have suggested that regional-scale groundwater ﬂow may obscure such a signal.3 times the hydrostatic values can also satisfy heat ﬂow constraint without requiring small µ. resulting in low friction. Conﬂicting observations? The most controversial aspect of the Anderson–Byerlee formulation has been the applicability of the laboratory values of friction to natural settings.3) is modiﬁed to be τ = µ(σ − p). In the course of their work on oil exploration.6–0. Three lines of evidence have complicated the Andersonian picture and led researchers to question whether or not faults are strong before and during earthquakes. The second line of evidence comes from geological mapping. When these experiments are performed in an area prone to normal faulting. This state is called hydrostatic. One complication to this simple picture was recognized early on.1). (2.1440 H Kanamori and E E Brodsky measure of the magnitude and orientation of the least principal stress. the upper limit of 20 MPa shear stress corresponds to a maximum value of µ of 0. The heat ﬂow data is sensitive only to the resolved shear stress. More sophisticated methods use the hoop stress to infer the maximum principal stress.7) where p is the pore pressure. High ﬂuid pressures can support part of the load across a fault and reduce the friction. Wernicke (1981)). Hubbert and Rubey observed that pressures in pockets of ﬂuids in the crust commonly exceeded hydrostatic pressure.e. Although it is uncertain whether or not rapid slip occurred on these faults (as opposed to slow aseismic . Using equations (2.7) with µ = 0. Earthquakes were induced by pumping water to increase the ﬂuid pressure at depth in an oil ﬁeld with little surface indication of faulting (Raleigh et al 1976). Hubbert and Rubey (1959) ﬁrst recognized the importance of the ﬂuid effect on fault friction. the observed fault orientation. the magnitude of the stresses resultant on the fracture plane and their orientation are consistent with internal friction of 0. Pore pressures that are more than 2.

On the creeping zone of the San Andreas in central California.8) is negative. according to equation (2. Ikeda 2001.4 on the low-angle normal faults. However. Both Byerlee (1992) and Rice (1992) argue that the stress orientation observations may not reﬂect the state of stress within the core of a pressurized. Therefore. The overall picture that is emerging is a good deal more complicated than the Andersonian view. 3. If the faulting occurred at the lock-up angle in the more conservative case. Finally we explore the scaling relationships between the observed parameters. In Southern California. Alternatively. our simple criterion for earthquakes proposed above is insufﬁcient to explain this complexity of behaviour. The most developed method for measuring earthquakes is to measure the elastic wave-ﬁeld generated by the sudden slip on a fault plane. Summary. Below. mature faults it appears that the µ applicable for initiation of slip must be signiﬁcantly different from what is measured in the laboratory for intact rocks or immature faults like Rangely. Provost and Houston 2003). In order to answer our question of why earthquakes begin. Provost and Houston ﬁnd that the angle θ between σ1 and the fault is ∼80˚. which translates to µ = 0.4 to 1. then there is no constraint on the fault stress from this line of evidence. (2. if these measurements of high values of θ reﬂect the stress state directly on the fault. Hardebeck and Hauksson (2001) ﬁnd values of θ as low as 60˚. We then list the most common earthquake parameters derived from the wave-ﬁeld and discuss their dynamical signiﬁcance. pore pressure may be so high that it exceeds the minimum principal stress. . the angle θ varies from 40˚ to 70˚ implying a maximum value of µ varying from 0. Therefore. Note that high pore pressure does not affect the geological result.4 from (2.8) and the lock-up angle is still tan−1 (1/µ) as long as σ3 − p 0.6) these areas must have µ < 0. high pore pressures in the fault do not remove the need for a low value of µ in the places with high θ .6). The only alternative is that p exceeds the minimum principal stress and the left-hand side (lhs) of (2. µ 0. An additional complication is that µ can depend on the slip rate and its history (Dieterich 1979).1). If it is true that the orientations are only measured outside the fault core.1) and (2. the stress orientation data hint that these variables may vary in time as well as space (Hardebeck and Hauksson 2001).6) with (2. ﬂuid-ﬁlled fault. the lock-up angle must be 70˚. A third line of evidence complicating the Anderson–Byerlee paradigm is that the maximum principal stresses next to major strike-slip faults like the San Andreas in California and Nojima in Japan are sometimes nearly normal to the fault (Zoback et al 1987. we discuss how the wave amplitude and frequencies are related to the physical properties of the earthquake. we will have to dig deeper. because combining (2. If the framework of equations (2. Once again. Moreover.7) is correct then in areas with large. it is clear that large-scale movement occurred on certain faults with dip angles of 20˚ and perhaps as low as 2˚ (Axen 2004). increasing the pore pressure presents new problems as rocks can fail under tension with relatively low differential stresses. we need to ﬁrst review the major observational facts and the parameters we use to quantify earthquakes. Clearly. Quantifying earthquakes In order to begin to answer these questions about earthquakes.2 depending on location.2) yields σ1 − p 1 + µ cot θ = σ3 − p 1 − µ tan θ (2. Further north on the fault.2) and (2.The physics of earthquakes 1441 creep).2 in order to be able to support motion.

After the rupture propagation has stopped and the transient motion has subsided. The seismic static displacement ﬁeld u(r) is u(r) = u1 (r) − u0 (r) and the stress drop is σ (r) = σ 0 (r) − σ 1 (r). In this approach. To study an earthquake process. (3.e. We imagine that an earthquake perturbs the stress ﬁeld by relaxing the stress in a localized region S embedded in the elastic medium. In this approach. subscripts 0 and 1 are used to indicate the states before and after an earthquake. respectively. Transient motion begins. this approach is seldom taken. An earthquake is a failure process in Earth’s crust. we assume that the medium is elastic. (A recent review on this subject is given by Madariaga and Olsen (2002). S. During the failure process (i. the distinction between EG and EH is model dependent. EG ) and thermally (thermal energy. Because some parts of the fracture energy eventually become thermal energy.1442 H Kanamori and E E Brodsky 3. here. 2 and 3 are used to indicate the principal stresses. a failure occurs at a point in the medium called the earthquake hypocentre. For a short-term process. we ﬁx the geometry of the source region. and rupture propagates into a region. Then. no source physics is . the slip distribution on the fault is determined by the inversion of observed seismic data. Prior to an earthquake. is determined by the failure process itself.1. the source is a thin fault zone.) (3) Kinematic model.2) (3.) The processes in the source region S are modelled by a localized inelastic process which represents the result of the combination of brittle rupture and plastic yielding.2 subscripts 1. This is the most physically desirable model.3) The change in the potential energy is W = W0 − W 1 . Because it is difﬁcult to gain this information in the crust. or volume. (Note that in section 2. ER ) and some energy is dissipated mechanically (fracture energy. i. In most seismological problems. the modelled failure growth is controlled by failure criterion (or failure physics) at each point in the medium. where r is the position vector. Then what controls the rupture is the friction law on the fault plane (constitutive relation). We denote the total potential energy of this state by W1 . This approach has been taken in recent years as more computer power is available. The resulting displacement ﬁeld is compared with the observed ﬁeld to determine the fault friction law.1) (3. Thus. the crust is in equilibrium under some boundary conditions with the initial displacement u0 (r) and the stress distribution σ 0 (r). Earthquake source parameters and observables A formal description of the elastic problem. the displacement and stress become u1 (r) and σ 1 (r). some energy is radiated (radiated energy. EH ). representing the earthquake rupture zone. coseismic process). At this stage. at least three approaches are possible. and the elasto-dynamic equations are solved for a given friction law (often parameterized) on the fault plane. energy is radiated. In most seismological problems the displacement is assumed to be small and linear elasticity theory is used. the initiation time of an earthquake. but it requires the knowledge of every detail of the structure and properties of the medium.e. In this case. and is modelled as a planar failure surface. at t = 0. (2) Dynamic failure on a prescribed source region. the wave-ﬁeld is computed for a prescribed slip motion on the fault using the elastic dislocation theory. Then. The total potential energy (gravitational energy plus strain energy) of the system at this stage is W0 . the ﬁnal failure surface. (1) Spontaneous failure.

5) 4πρrα 3 0 α 4πρrβ 3 0 β 0 uφ Rφ (θ. etc). This force system is commonly called a double couple source. both linear and angular momentum must be conserved during faulting. as R (θ.e. the P-wave (compressional wave) velocity. embedded in a medium with rigidity µ. It can be shown that the force system that respects these conservation laws and produces a stress ﬁeld equivalent to the point dislocation source is the combination of two perpendicular force couples (ﬁgure 6. θ. is placed at the origin. Seismic source and displacement ﬁeld.e. If a point source with seismic moment. invoked. M0 (t). Representation of a dislocation (fault) seismic source. φ) 0 ur r r 1 1 r uθ = 0 + Rθ (θ. φ).2. but in this section it is used to represent the rigidity. First. this approach is widely used. In seismology. Since the fault is entirely enclosed by elastic crust and no work is done by external forces. together. The moment of each force couple M0 is given by (Stekettee 1958.1. The distinction will be clear from the text and context. In the later sections µ is used both for the rigidity and the coefﬁcient of friction.The physics of earthquakes 1443 Figure 6. M t− M t− (3. point source) on which a displacement offset D (the difference between the displacements of the two sides of a fault) occurs (ﬁgure 6. φ) where the prime symbol denotes differentiation with respect to the argument. Left: a seismic fault represented by a shear displacement offset D over a surface with an area S. Maruyama 1964. . can be used to infer the physical process involved in failure (i. the displacement in the far-ﬁeld is given in the polar coordinate. in the limit of point source (i.) A ﬁnite fault model can be constructed by distributing the point sources on a fault plane. The dimension of M0 is [force] × [length] = [energy]. (Note that in section 2. φ) . (r. this approach is called kinematic. rather than J (joule). Right: a force double couple equivalent to the dislocation model shown on left. and does not directly represent any energy-related quantity of the source.) 3.e. We want to ﬁnd a set of forces that will generate a stress ﬁeld equivalent to the stress ﬁeld generated by a given imposed displacement on the fault. In this sense. because M0 is the moment of the equivalent force system. (A recent review on this subject is also given by Madariaga and Olsen (2002). Burridge and Knopoff 1964) M0 = µDS. we consider a homogeneous whole space with the density. disturbances are propagated as simple elastic waves known as body waves. left). α and the S-wave (shear wave) velocity. β. S → 0. consider a very small fault (i.4) where µ is the rigidity of the material surrounding the fault. ρ. friction. However.1. µ is used for the coefﬁcient of friction. it can be used to compute the associated stress ﬁeld. it is common to use N m for the unit of M0 . right). and D → ∞ while the product DS remains ﬁnite). In the absence of any interfaces in the elastic medium. For simplicity. Since many methods for inversion of seismic data have been developed. The displacement and the stress ﬁeld on the fault plane. (3. once the slip is determined.

(For more details. Alaska earthquake recorded by a broadband seismometer 3460 km away in Mammoth Lakes. Aki and Richards (2002). 3 November 2002. φ). e.6) 4πµr 2 The near-ﬁeld displacement is important for the determination of detailed spatial and temporal distribution of slip in the rupture zone.9. (3. u∝ . Rr (θ. The reason why the near-ﬁeld and far-ﬁeld displacements are proportional to M0 (t)/r 2 and M0 (t)/r. which depend on the geometry of the source and the observation point. The early motion on these seismograms (between 400 and 800 s) shows P. Far away from the fault.6) is negligible as it falls off much more quickly than the far-ﬁeld terms (1/r 2 as opposed to 1/r). The components are radial (R).and S-waves described by (3. The primary component of the near-ﬁeld displacement is given approximately by 1 M0 (t).1444 H Kanamori and E E Brodsky Times (s) Figure 7. see Aki and Richards (2002)).g. φ) and Rφ (θ.5). see. California. we have an additional term representing the near-ﬁeld displacement.) These displacement components are what are measured by seismometers (ﬁgure 7). while the far-ﬁeld represents the contributions from both sides of the fault. Later motion (after 800 s) shows surface waves produced by the interactions of the waves with boundaries in the earth and heterogeneous structure. φ) represent the radiation patterns. Rθ (θ. S-wave. transverse (T) and vertical (Z). This situation is similar to that of an electric ﬁeld from a point charge and a dipole. Lay and Wallace (1995). (3. is that the near-ﬁeld is essentially determined by the motion on one side of the fault. respectively. (For more details. At short distances from the source. The radial and transverse components are the two components on the horizontal plane. The ﬁrst term is the P-wave and the second term. Example of displacement from the Mw = 7.

10–300 s. we have discussed the seismic body waves that travel directly from the earthquake to the seismometer. ˙ The time derivative of the seismic moment M0 (t) is called the moment-rate function or the source time function. Many seismological methods have been developed to handle these problems. after differentiated. and the seismic moment can be determined accurately from seismic data (e. If M0 is determined by a seismic method.The physics of earthquakes 1445 Figure 8. The near. In this sense. and its frequency spectrum is called the moment rate spectrum or the source spectrum. Observed surface waves are often very long period. the wave trains in the later parts of ﬁgure 7). seismic body waves interact with the Earth’s free surface and many internal structural boundaries to develop reﬂections as well as surface waves (i. In general. So far. If the fault motion is a linear ramp function. Lay and Wallace (1995)). The seismic moment M0 can be determined: (1) From seismic data: the amplitude of long-period surface waves and normal-modes can be used to determine M0 most accurately. As shown by (3. For a ﬁnite ¯ source with a fault area S on which the spatially averaged slip is D (offset). we need to include the effect of wave propagation in a heterogeneous structure. energy and stress during rupture. because long-period waves are least affected by complex propagation effects. and carry the information of the seismic source at long period. We now use the elastic theory developed above to determine parameters of earthquakes that measure the size. (2) From geodetic data: with the advent of space-based geodetic methods (e. California. they can be interpreted as elastic oscillations of the Earth. and have been used effectively to study earthquakes (Gilbert and Dziewonski 1975. or from the amplitude of the near-ﬁeld displacement. and if S is estimated by either a seismic or geodetic ¯ ¯ method.g. before and after an earthquake. geometry of the source and the ﬁniteness of the source. The theories of seismic surface waves and normal modes are well developed. The long-period oscillations can be studied using the normal mode theory.1.6). earthquake (Massonnet et al 1993) . then M0 (t) is a ramp function. Because M0 depends on the two end states. the seismic moment ¯ M0 is given byµDS. it is a static parameter. then the near-ﬁeld wave form is a ramp function and the far-ﬁeld wave form is a pulse with a duration of τ (ﬁgure 8). it does not depend on the actual time history of faulting. In the real situation.e.5) and (3. D can be determined by using the relation D = M0 /µS. This is the most common method. GPS and InSAR). Seismic moment and magnitude. 3. In the actual determination of the seismic moment. if the fault motion occurs over a duration of τ . produces a box-car far-ﬁeld wave form.g. The synthetic aperture radar (SAR) interferometry was used for the 1992 Landers. this method is becoming more commonly used.and far-ﬁeld displacements from a point dislocation seismic source which represents a fault slip motion given by a ramp function with duration τ . the seismic moment can be determined from the integral of the far-ﬁeld displacement. which. When surface waves propagate around the Earth many times.2. Dahlen and Tromp 1998). The amplitude and frequency spectra of seismic body waves can be also used for smaller earthquakes.

The following web-sites provide a catalogue of seismic moment of large earthquakes in the world. compiled by the Seismology Group of Harvard University. but these magnitudes are empirical parameters and cannot be directly related to any speciﬁc physical parameter of the source. Mw .5 (M0 in N m). is deﬁned by the following relation: Mw = log10 M0 − 6. As we discussed above. historically. the stress drop caused by an earthquake is σ (r) = σ 0 (r) − σ 1 (r). Earthquake. 3.2).eri. however. this method is often used. Recently.seismology.1 Long-period surface waves 5. Tokyo University Seismic body waves 5.jp/EIC/EIC News/index-e.8) . σs = σ0 − σ1 (3.98 × 1019 Harvard University Earthquake Research Institute. (3) From geological data: the surface break of a fault can be used to estimate M0 .7) As mentioned above. M0 or Mw can be used as a useful quantiﬁcation parameter for an earthquake and its damaging effects. at least for large earthquakes (section 3.edu • http://wwweic. Earthquake Information Center of the Earthquake Research Institute of Tokyo University and the United States Geological Survey.1. However. respectively. magnitude scales were used for this purpose. California. Seismic moment determinations from different data sets. Data M0 (N m) Reference H Kanamori and E E Brodsky Hector Mine.2. However.3. Mw = 7. Table 1 shows the results for the 1999 Hector Mine. Strain and stress drop. However. the standard practice is to deﬁne the magnitude with the seismic moment. The redundant multiple methods allow us to verify that seismic moment is well-measured by seismic methods to an accuracy unequalled by any other seismic parameters. and the resulting estimate of M0 is inevitably inaccurate. M0 is a static parameter and does not represent any dynamic properties of the source.gov/neis/FM/previous mom.7 × 1019 Simons et al (2002) to successfully map the regional static deformation ﬁeld associated with this earthquake. Most magnitude scales were deﬁned by the observed amplitude of seismic waves with some corrections for attenuation with distance from the source.html Seismic moments are the most modern and accurate quantiﬁcation of the size of an earthquake. • http://www. (3. California.harvard. for historical events for which no instrumental data are available. with the use of some scaling relations. the distribution of slip where a fault meets the surface of the Earth does not necessarily represent the slip at depths. 16 October 1999. We usually consider only the shear stress on the fault plane.html • http://neic. This magnitude. The values determined by different methods generally agree within 30%.u-tokyo. To determine M0 accurately. it can be approximately related to the total radiated energy.ac.07 1.usgs.5 × 1019 GPS and InSAR 6. good spatial coverage around the source is required. earthquake where the seismic moment was independently measured by methods 1 and 2.1446 Table 1. In this sense.

(3. However. More commonly. we can write ˜ σs = CM0 L−3 = CM0 S −3/2 . Madariaga (1977.β = ρβ +∞ S0 −∞ ˙ ˙ [uθ (r. In general. The stress drop can be locally much higher than the average. it is straightforward to estimate the radiated energy. is another important physical parameter of an earthquake. t)2 ] dt dS0 .11) (e. Rudnicki and Kanamori (1981) and Das (1988) show that this is a good approximation unless the variation of stress on the fault is extremely large. which in turn is used to compute the average stress drop. complex functions of space. σs varies spatially on the fault. (3. Tsuboi (1933)).11) If the length scale of the source is estimated from the geodetic data.α = ρα +∞ S0 −∞ ˙ ur (r.9) S S Since the stress and strength distributions near a fault are non-uniform. L = S 1/2 and (3. The energy radiated by seismic waves. For example.g. the limited resolution of seismological methods often allows determinations of only the average displacement over the fault plane. if the P-wave displacement in a homogeneous medium is given by ur (r. aftershock area. given the limited spatial resolution of seismic data. we cannot fully assess the validity of this approximation.9).The physics of earthquakes 1447 and call it the static stress drop associated with an earthquake. either S or L. If the slip distribution on the fault plane can be determined from high-resolution seismic data. Thus.g. ˜ Since σs ≈ CM0 L−3 . In most applications. L is a characteristic rupture dimension. Similarly. as given by (3. often deﬁned √ by S and C is a geometric constant of order unity. 3. (3. if we can determine the wave-ﬁeld completely. we use the following expression. is extremely important in determining the stress drop. In principle. we estimate σs simply by ¯ D (3. we use the stress drop averaged over the entire fault plane. The spatial average is given by 1 σs = σs dS. Some early ˜ determinations of stress (strain) drops were made using D and L estimated from geodetic data (e. then the energy radiated in a P-wave is given by ER. ˜ L ¯ ˜ where.13) . t). Energy Radiated energy. we can estimate the stress drop using (3. Abercrombie and Leary (1993)). 1927 Tango earthquake. ER . (3. an accurate ˜ determination of earthquake source size. the energy radiated in an S-wave is given by ER. With this approximation. Kanamori and Anderson (1975). tsunami source area or other data. However. ER . Unfortunately. if the seismic moment is determined by either geodetic or seismological ¯ ˜ methods.10) σs ≈ Cµ . The strain drop εs is given by εs = σs /µ. To be exact. Using M0 = µDS.4. We often use σs to mean the average static stress drop in this sense. it is possible to estimate the stress drop on the fault plane (Bouchon 1997).12) where S0 is a spherical surface at a large distance surrounding the source. 1979). an uncertainty in the length scale can cause a large uncertainty in ˜ σs : a factor of 2 uncertainty in L results in a factor of 8 uncertainty in σs . t)2 + uφ (r. the average stress drop is the spatial average of the stress drop. the slip and stress drop are. D is the average slip (offset).1. t)2 dt dS0 .10). in general.

California. (3. (3. attenuation and scattering. with the limited resolution of seismological methods.17). It is a common practice to assume that the approximation is sufﬁciently accurate if the spatial variation is not very rapid. Thus. ER can be estimated probably within a factor of 2–3 (McGarr and Fletcher 2002). ER . it is not possible to accurately assess the errors associated with the approximation of equation (3. Mw = 7. 1 W = 2 (σ0 − σ1 )DS + σ1 DS = 1 2 σs DS + σ1 DS = W0 + σ1 DS. propagation effects.α and ER. W0 can be computed from the seismologically determined parameters.3 for the stress drop.1 × 1016 Singh et al (2004) Teleseismic data 2. Although W cannot be determined by seismological methods. Equation (3. Second. Dahlen 1977). instead of (3. Haskell (1964)).2 × 1015 2 × 1015 Venkataraman et al (2002) The total energy. Earthquake.15) where W0 = 1 2 σs DS. σ0 and σ1 cannot be determined. the wave-ﬁeld in the Earth is extremely complex because of the complexity of the seismic source.16). Only the difference σs = σ0 − σ1 is determined. For earthquakes for which high-quality seismic data are available. the radiated energy (ER ) and the lower bound of the potential energy change ( W0 ). Earthquake.4 × 1015 Boatwright et al (2002) Venkataraman et al (2002) 3 × 1015 Boatwright et al (2002) Teleseismic data 3.14) can be rewritten as. with seismological data and methods. As we discussed in section 2. Extensive efforts have been made in recent years to accurately determine ER .16) Two difﬁculties are encountered.1. In practice. however.17) Unfortunately. as we discussed in section 3. The potential energy change in the crust due to an earthquake is 1 W = 2 (σ0 + σ1 )DS.β (e. Some examples are shown in table 2. Thus we commonly use. Mw = 7.6 Regional data 2. 26 January 2001.9 × 1016 Singh et al (2004) Hector Mine.4). If the residual stress σ1 is small. It is important to note that we can determine two kinds of energies. 16 October 1999. the details of spatial variation of stress and displacement cannot be determined. Determinations of radiated energy with different data sets and methods. (3.0 × 1016 Venkataraman and Kanamori (2004) Frequency-domain method 1.g. First.1 Regional data 3. W0 is a good approximation of W . In general σ1 DS > 0. Potential energy. Data ER (J) Reference Bhuji. unless a large scale overshoot occurs.14) where the bar stands for the spatial average (Kostrov 1974. W0 = 1 2 σs DS. non-seismological methods give inconsistent results for background stress. we cannot compute W from seismic data. the absolute value of the stresses. is the sum of ER. These two energies play an important role in understanding the physics of earthquakes (section 4.2. D and S. and W0 can be used as a lower bound of W .1448 H Kanamori and E E Brodsky Table 2. (3. with seismological measurements alone. . India. σs .

For some earthquakes . The rupture of one of the bilateral segments propagated northeast from the hypocentre toward Kobe and produced very strong ground motions in the city. regardless of the azimuth ¯ and is proportional to the seismic moment. Thus it is important to determine the rupture speed of earthquake faulting to understand the nature of earthquake mechanics. when averaged ˜ over the azimuth. This effect. As we discussed in section 3. the displacement integrated over time) is constant. half the fault length and the radius of the fault plane.The physics of earthquakes 1449 3. the pulse becomes larger and narrower. Another observable feature of earthquakes is the rupture pattern on the fault.1.9). V is approximately 75–95% of the S-wave velocity at the depth where the largest slip occurred. earthquake (Mw = 7. However. Rupture mode. than at a site away from the rupture propagation. is proportional to the length scale of the fault L divided by the rupture speed. the fraction of the shear velocity that the shear crack rupture velocity achieves is related to the fracture energy. for most large shallow earthquakes. earthquake is believed to have ruptured in both directions. L is commonly taken to represent the fault length.6. In a unilateral rupture. ¯ (3. The rupture speed V has been determined for many large earthquakes. earthquake (Mw = 6. the area under the pulse-like waveform (i. since it is not a single summary quantity. the observer located in the direction of the rupture will see a shorter pulse than the observer in the direction away from the rupture direction. because of the Doppler effect. there are some exceptions. bilateral and circular faults. earthquake (Mw = 6.18) V ˜ For unilateral. In general.9. In these cases. From the gross rupture patterns. called a hypocentre. However. bilateral rupture and two-dimensional (approximately circular) rupture patterns. California. and the 1995 Kobe.e. speed and directivity. the hypocentre is at the one end of the fault. Good examples include the 1989 Loma Prieta. earthquake. 3 November 2002). In some earthquakes. M0 . The variation of the pulse width as a function of azimuth due to rupture propagation has an important inﬂuence on ground motions. Rupture speed. As we will discuss later (section 4. the rupture speed is an important parameter which reﬂects the dynamic characteristic of a fracture. and the rupture propagates primarily in one direction toward the other end. V ˜ L τ= . it is another observable characterization of the rupture. The duration of the pulse. the far-ﬁeld displacement is given by the time derivative of the near-ﬁeld displacement and is pulse-like. and propagates outward on the fault plane. Directivity and source duration. we classify the rupture patterns into unilateral rupture. An earthquake occurs on a ﬁnite fault. and produces stronger ground motions which often result in heavier shaking damage. Japan. the fault geometry is assumed to be one-dimensional. A good example is the 1995 Kobe.9). is similar to the Doppler effect.1. One good example is the recent Denali. From the rupture patterns we can deﬁne some secondary parameters describing rupture propagation velocity. In the description of unilateral and the bilateral fault. In particular. but propagated further to the north than the south. slip duration and directivity. California.1. However. Japan. called directivity. It initiates from a point. The 1906 San Francisco. Bilateral ruptures are not necessarily symmetric. Alaska.5. in that order. In a bilateral rupture. a circular fault is often used to model the fault plane. The source ﬁniteness and rupture propagation have an important effect on seismic radiation. the rupture propagates in opposite directions from the hypocentre. τ .1). the rupture propagates in all directions on the fault plane. at a site towards the rupture propagation. Although the rupture pattern is not a parameter sensu stricto. As mentioned above.

Complexity of earthquake rupture pattern. Kikuchi and Kanamori (1994)). as we will discuss in section 4. scattering of waves and complexities of the source process produce wave forms too complex to be explained with a simple model. it is probably of the order of λ/3. rupture speed has been reported. Although the spatial resolution of these models are not always given.g. For some earthquakes (e. Tibi et al (2003)). the resolution of the seismic method was good enough to determine V . short-period (usually 2 s or shorter) waves are ﬁltered out because of the difﬁculty in modelling such short-period waves. California. Most of the laboratory data show that the rupture speed for intact materials under tensile stress is at most 50% of the Rayleigh wave speed. these models should be regarded as long-wavelength rupture patterns.1. i. have been reported (Bouchon and Vallee 2003). 2001 Kunlun.1. Thus. The slip distribution in real earthquakes is very complex. Earthquake rupture pattern.1450 H Kanamori and E E Brodsky Figure 9.g. as shown for the 1992 Landers. Rupture pattern for the 1992 Landers.g. With the advent of modern strong-motion seismographs and broad-band seismographs. because of the difﬁculty in resolving the rupture pattern due to the lack of close-in observations. At periods shorter than 2 s (the corresponding wavelength is about λ = 5 km). However. For the largest deep earthquake. California. (e.6. higher rupture speeds have been reported (e. 3. earthquake determined with inversion of seismic data (Wald and Heaton 1994). Short wavelength heterogeneity has been suggested by . the real slip distribution is probably far more complex with short wavelength irregularities.6. China.e. Wald and Heaton (1994)). it has become possible to determine the actual slip distribution by inverting the observed seismic waveforms. super-shear rupture velocities. Higher rupture speeds have only been observed for pre-cut samples. In a few pre-cut experiments the rupture velocities are even higher than shear wave speed (Rosakis 2002). The difference and similarity between earthquakes and laboratory fractures provide an important clue to the mechanics of earthquake rupture. earthquake). the 1994 Bolivian earthquake (Mw = 8. (V /β) = 0. in most modelling studies. a very low. an accurate determination of V is usually difﬁcult.2 (e.g. For deep earthquakes. earthquake (ﬁgure 9. These studies demonstrate that the slip distribution on a fault plane is highly heterogeneous in space and time. 1992 Nicaragua earthquake. For other smaller deep earthquakes. because the rupture bifurcates and cannot produce a planar faulting.3). a very slow rupture speed has been reported. V > β. It is not possible to maintain a shear fault in intact materials. Kikuchi and Kanamori (1995)). The relatively high rupture speeds observed for most shallow earthquakes is in striking contrast with the rupture speeds observed in laboratory.

11). Scaling relations for static parameters.20) can be interpreted as follows.5. The existing data (Masayuki Kikuchi. the rupture areas for large (Mw > 6) earthquakes have been estimated. This complexity suggests that the microscopic processes on a fault plane can be important in controlling the rupture dynamics. There is some evidence that σs varies for different types of earthquakes. This parameter can be determined from a seismogram.g. Figure 11 is the evidence that most earthquakes have comparable stress drops in the range 1–10 MPa . If σs is constant. it is useful to investigate a scaling relation between M0 and another parameter that can be determined most directly from seismic observations. but it is not just the duration of a waveform recorded on a seismogram. and suggests a scaling relation M0 ∝ S 3/2 . A more detailed discussion of microscopic physics follows in section 4. (2) Moment versus fault area. Thus. is the source parameter that can be determined most reliably. We will ﬁrst discuss a selection of the observed scalings with only a cursory overview of the implications. this estimate of σs is not precise.2. 3. rupture area). (1) M0 versus source duration. geodetic data. τ is equal to the azimuthal average of τ discussed in section 3.21) Hereafter σs is simply written as σs for brevity. We ﬁrst choose the duration of source process. τ . as shown in ﬁgure 10.5. this scaling relation indicates that σs is roughly constant over a range of M0 from 1018 to 1023 N m. The seismic moment. From ﬁgure 11 we can estimate that σs is. aftershock area. Figure 11 shows the results. directivity and seismic inversion results). It is not always easy to determine the fault area S (i. we need to relate the parameters to each other. surface rupture. on the average.2. (3. Because of the uncertainty in S. equation (3. Zeng et al (1994) modelled an earthquake fault as a fractal distribution of patches. C (3. σs . approximately 3 MPa with a range 1–10 MPa. Seismic scaling relations Now that we have measured some average properties of rupture. but the approximate range 1–10 MPa is considered robust. such as those on major plate boundaries and those in plate interiors. written ¯ communication (2001)) show a gross scaling relation M0 ∝ τ 3 .e. M0 . and the length scale S 1/2 are related by M0 = 1 σs S 3/2 . .19) The scaling relation given by (3. but by combining various kinds of data (e. From equation (3.18).The physics of earthquakes 1451 complex high-frequency wave forms seen on accelerograms recorded at short distances. The scaling relationships between the macroscopic source parameters are useful for isolating general constraints on the microscopic forces and processes in the fault zone during rupture. The level of the curve determines the value of σs . which is the scaling relation shown in ﬁgure 11. We must remove the propagation effects from the seismograms to estimate the duration of rupture process at the source. the seismic moment M0 . then M0 ∝ S 3/2 .1.20) (3. as we will discuss in section 4. Thus. 3. but the overall difference is probably within this range. and the assumption for the geometry of the fault plane.1.

We deﬁne the length scale of the fault to be L ≡ S 1/2 . The relation between the seismic moment M0 (the corresponding Mw is shown at the top of the ﬁgure) and the source duration τ for shallow (depth <60 km) earthquakes. Mw < 3) earthquakes directly. suggest that most large shallow earthquakes have σs ranging from 1 to 10 MPa. the pulse width or the corner-frequency of the source spectrum is used to infer the source dimension. equal to L/V . (Masayuki Kikuchi.5 (M0 = 7 × 1018 N m) are for the events worldwide. on the average. A similar analysis can be made for smaller earthquakes. Comparison of this scaling relation. and the rupture speed V is roughly constant at 75–95% of the S-velocity. M0 ∝ τ 3 ˜ (3.18) where V is the rupture speed.g. taken together. the pulse width is. Nm 0 17 10 19 10 21 Figure 10.1452 H Kanamori and E E Brodsky Mw 10 3 2 4 6 8 10 2 Duration (τ). .19) suggests that τ ∝ S 1/2 .1. with the scaling relation. M0 ∝ S 3/2 . However. (As discussed in section 3. for some shallow large earthquakes. The data for events larger than Mw = 6. and there are exceptions. it is difﬁcult to determine the source dimension of small (e. In most cases. Since ˜ ≈ V τ (3. Thus. As mentioned earlier. The data are for events during the period from 1991 to 2001. written communication (2001)). s 10 1 10 0 10 -1 10 -2 10 11 10 13 10 15 10 M. this means that V is constant for most shallow L earthquakes. The data for smaller events in Japan are added. We note here that these are the general scaling relations. the rupture speed V is directly determined to be 75–95% of the S-velocity. The horizontal and vertical alignments of the data points are an artefact due to the rounding-off of the values used for M0 and τ . these results.5.

If the rupture speed. large subduction-zone earthquakes like the 1964 . This large range in σs may be real.1 to 100 MPa (Abercrombie 1995). this question is not resolved. However. In the scaling relations discussed earlier. for long crustal strike-slip earthquakes such as the 1906 San Francisco earthquake. At present. with σs ranging from 0. 10 Nm Figure 11. different faults have different aspect ratios (i. V . The data are from Kanamori and Anderson (1975) and Masayuki Kikuchi (written communication 2001). about 15 km. The unpublished data provided by Kikuchi are for the events worldwide during the period 1991 to 2001 for which the source dimension could be estimated. the pulse width can be used to estimate the fault length. The scaling relation between seismic moment M0 (the corresponding Mw is shown at the top of the ﬁgure) and the fault area S for shallow earthquakes. The corner frequency of the spectrum of a pulse-like source function is proportional to the reciprocal of the pulse width. the fault length is about 350 km.The physics of earthquakes 1453 Mw 6 10 6 7 8 9 10 5 S. It is also possible that the large scatter is due to errors in determining the source dimension. In contrast.) The general trend follows the M0 ∝ S 3/2 scaling. the length scale of the source is deﬁned by S 1/2 with the idea of representing the source dimension with just one parameter.e. but the depth-wise width of the fault is probably comparable to the upper half of the crust. For example. reﬂecting the heterogeneities of the crust on short length scales. the ratio of fault length to width). km 10 3 2 10 4 10 2 10 1 10 -3 10 -2 10 -1 10 0 0 10 20 1 10 2 10 3 10 4 M . is approximately equal to the S-velocity.

Jost et al (1998). Kinoshita and Ohike 2002). because of the complex wave propagation effects in the Earth. decreases. it is still difﬁcult because the relatively high-frequency seismic waves excited by small earthquakes are easily scattered and attenuated by the complex rock structures between the fault and a seismic station. then e must be constant if small and large earthquakes are dynamically ˜ similar. Recent improvements in data quality and methodology have signiﬁcantly improved the accuracy of ER determination for large earthquakes (e. The ratio depends on many seismogenic properties of the source region so that it varies signiﬁcantly for earthquakes in different tectonic environments. and a factor of 100 smaller at ˜ Mw ≈ 1. have been proposed for different types of earthquakes and for different magnitude ranges. As we will discuss later (section 4. then the initiation process is scale-invariant and therefore the size of earthquakes is inherently unpredictable. Several different scaling relations. such as continental crust. However. the data for the same type of earthquakes exhibit an interesting trend. especially for small earthquakes. If small and large earthquakes are dynamically similar. several investigators tried to investigate the scaling relation between M0 and L (e.1454 H Kanamori and E E Brodsky Alaskan earthquake have a fault width as large as 200 km or more. the ratio can be interpreted as proportional to the energy radiated per unit fault area and per unit slip. Romanowicz and Ruff (2002)). Boatwright and Choy (1986). For small earthquakes. Richardson and Jordan (2002)). is another macroscopic earthquake source parameter that can be determined by seismological methods (see section 3.g. Seismologists are very concerned with whether or not earthquakes are dynamically similar because of the implications of the observation for the predictability of the eventual size of an earthquake. Romanowicz and Rundle (1993). Scholz (1994). For large earthquakes (Mw ≈ 7).g. the average ratio e decreases as the magnitude. on the average. ˜ but it is approximately a factor of 10 smaller at Mw ≈ 3. In view of its importance for understanding the dynamic character of earthquakes. In view of this variation in aspect ratio. so the observation of a lack of similarity cannot prove the predictability of earthquake size. From (3. Mw > 6) (e. Perez-Campos and Beroza 2001. if the static stress drop is constant. These systematic . The ratio e= ˜ 1 ER ER = ¯ M0 µ DS (3. Results for even smaller earthquakes show even smaller values of e (e.2. large amounts of high-quality data for small earthquakes have been accumulated (Abercrombie 1995. such as M0 ∝ L3/2 and M0 ∝ L2 . However. ER . Venkataraman and Kanamori 2004). The radiated energy. it is difﬁcult to determine ER accurately.g. Unfortunately.22). Taken at face value. ˜ Mw . this scaling relation represents a dynamic property of earthquakes. 3. Boatwright et al (2002).2. The ratio e multiplied by the ˜ rigidity µ is called the apparent stress. Prejean and Ellsworth 2001. despite the large scatter. Venkataraman et al (2002)). or with the careful removal of path effects. subduction zone. Wyss and Brune 1968). Izutani and Kanamori 2001. In this sense.4). Mayeda and Walter 1996. etc (Choy and Boatwright 1995. the converse statement is not true. deep seismic zone.1. using down-hole instruments.22) has long been used in seismology as a useful parameter that characterizes the dynamic properties of an earthquake (Aki 1966. many ˜ studies have been devoted to the determination of e. approximately 5 × 10−5 . Ide and Beroza (2001) suggested that many of the ˜ published e versus Mw relations could be biased to have decreased e for small events because of ˜ inadequate corrections for path effects or the limited instrumental pass-band.4). Nevertheless. e is. Scaling relations for dynamic parameters. Figure 12 shows the results for crustal earthquakes in California and Japan. and the results were widely scattered.g.

Mayeda and Walter ˜ 1996. seismic faulting may be more intuitively viewed as sliding on a frictional surface (fault) where the physics of friction. 4. Despite this complexity. On the other hand.1. measurement errors could mean that the real e is scale-independent. Mode I (tensile). Here. then the ˜ observation would imply that large and small earthquakes are dynamically different. The next step in our inquiry into why earthquakes happen is to examine the rupture process itself. Fracture mechanics To interpret seismological data. we limit our discussion to the very basic aspects of these models. are used . In crack mechanics. 4. Kanamori et al (1993).The physics of earthquakes Mayeda and Walter [1996] Abercrombie [1995] TERRAscope 1455 Izutani and Kanamori [2001] Large Eqs 10 10 10 10 10 10 -3 -4 -5 E /M R 0 -6 -7 -8 0 1 2 3 4 M 5 6 7 8 w Figure 12. especially stick slip. If future research ﬁnds that e varies as suggested by ﬁgure 12. 4. for TERRAscope data). An overview of the crack model. The relation between e = ER /M0 and Mw (Abercrombie 1995. three types of crack geometries. crack models are often used in part because the theories on cracks have been developed well. this question ˜ remains unresolved. Izutani and Kanamori 2001. plays a key role. Seismic faulting in the Earth can be complex and we may require a mixture of crack models and sliding models. Mode II (longitudinal shear) and Mode III (transverse shear). crack models and frictional sliding models provide a useful framework for the interpretation of earthquake processes.1. Rupture processes We have now provided an overview of the stresses that generate earthquakes along with a discussion of the measurable parameters and their interrelationships. or even other models to interpret it.1. At present.

16)) (σ0 + σf )DS = W0 + σf DS. The crack extends from z = −∞ to +∞ as shown in the inset.14)–(3. Stress ﬁeld for a Mode III crack before and after crack formation. Speciﬁcally. Figure 14.2) (x − a 2 )1/2 (Knopoff 1958). x = a + ε. Dashed line is the initial stress and solid line is the ﬁnal stress. Although the difference between these models is important. and drops to the frictional stress σf on the crack surface.1456 H Kanamori and E E Brodsky y x z Figure 13. (4. y and z axes indicate the coordinate system used in ﬁgure 14. II and III cracks are given in Rice (1980).4) W = 2 . The x. The strain energy release per unit length in z direction is (equations (3. II and III cracks (from Lawn (1993)).1) µ and x σzy = (σ0 − σf ) 2 + σf x a (4. Dmowska and Rice (1986) and Li (1987). The inset shows the Mode III crack geometry. Modes I. σzy is proportional to √ 1/ ε. the displacement w and the stress σzy are σ0 − σf w= (a 2 − x 2 )1/2 x a (4. the shear stress becomes inﬁnitely large just beyond the crack tip. the relationship is K 1 σzy = √ √ + σf . After the crack is formed. here we mainly use the Mode III crack for illustration purposes.3) More detailed expressions for the stress intensity factors for Modes I. (ﬁgure 13). At a small distance ε from the crack tip. 2π ε where K is the stress intensity factor deﬁned by √ K ≡ πa(σ0 − σf ). (4. For the coordinate system shown. for many problems in crack mechanics. Figure 14 shows the stress distribution along the plane of a Mode III crack before (dashed) and after (heavy curves) crack formation.

Also. In the dynamic crack propagation problem. When a crack with half-length a is inserted in a homogeneous medium under uniform shear stress σ0 . (For the development of the concept. slip. we discuss this problem using a simple model.6) σ0 − σf µ 1 a a −a (a 2 − x 2 )1/2 dx = πa 2 σ0 − σf µ (4.3. plastic) yielding occurs. Instead.1. we can consider the stability of a crack and its growth. as shown in ﬁgure 15(b). Here. Instead it decreases gradually to σf over a distance l0 as shown by the broken curve in ﬁgure 15(a). This behaviour in which the stress on the fault plane decreases as slip increases is called slip-weakening behaviour. LEFM). inelastic (e. inside the crack increases gradually to the value. as discussed earlier. and for more detailed discussions. D0 . Li (1987). we use the basic concept.1. as the crack tip passes by. a. The theory is based on Grifﬁth’s (1920) concept which was initially developed for tensional cracks (Mode I). 4. and this model is often referred to as the breakdown-zone slip-weakening model. and apply it to seismological problems. the crack grows. the stress near the crack tip becomes indeﬁnitely large (solid curve in ﬁgure 14 (inset)). we obtain the energy given by (4.4). If the stress just beyond the crack tip becomes large enough to break the material. In the real material this does not occur. the stress and slip vary as a function of time as shown in ﬁgures 15(c) and (d). The model discussed earlier is for the static case and it provides the basic physics of dynamic crack propagation.The physics of earthquakes 1457 where D = 2w = ¯ and W0 = π a 2 (σ0 − σf )2 (σ0 − σf )DS = 2 2µ (4. 2 2µ (4. After subtracting the energy dissipated in friction. Now that we have an overview of crack physics. Because of this breakdown process. the theory becomes complex because of the complex stress ﬁeld near the crack tip and the strain energy ﬂux into the crack tip. In (4. see Rice (1980). Here. D. Ida (1972). . The stress drops from σY to the constant frictional stress σf over a slip D0 . Stability and growth of a crack. expected for the case without inelastic breakdown (i. respectively.g. is called the yield stress.6) W0 = π a 2 (σ0 − σf )2 (σ0 − σf )DS = .7) which is available for mechanical work for crack extension. is the portion of the strain energy change that is not dissipated in the frictional process.2. see Dugdale (1960). At a point just beyond the crack tip. Crack tip breakdown-zone. the strain energy is released. and the stress becomes ﬁnite as shown by the broken curve in ﬁgure 15(a). More rigorous and detailed discussions are by Freund (1989) and Lawn (1993). Barenblatt (1962). the stress just inside the crack does not drop to the constant frictional level σf abruptly. σY . Figure 15(e) shows the shear stress σyz at this point as a function of slip D. More details are given by Lawn (1993).) 4. Palmer and Rice (1973). W0 . Consider a Mode III crack with half-length.5) and S = 2a and D is the average offset across the crack. In the simple model described in ﬁgure 14 (called the linear elastic fracture model (LEFM)). the second term on the right-hand side (rhs) is the frictional energy and the ﬁrst term. The ﬁnite stress at the crack tip.e.

∂a µ (4.3).8) arises because the crack extends at both ends. (a) The stress ﬁeld near the crack tip as a function of position. the stress drops from σY to σf gradually. As the crack tip extends past this point. Then the strain energy that would be released due to the virtual extension δa is. (d) The temporal variation of the slip at a point which is initially just beyond the crack tip. The unit of G∗ is energy per area.8) . (e) The shear stress at point which is initially just beyond the crack tip. (4. The broken curves indicate the deviation from the LEFM case when yielding occurs. G∗ = π a(σ0 − σf )2 ∂ W0 δa = δa = 2G∗ δa. (b) The slip on the crack surface.9) 2µ 2µ where K is the stress intensity factor deﬁned by (4. the solid curves are for the LEFM model. G∗ is called the static energy release rate or crack extension force. Note that G∗ is not a constant. The name is a little confusing because here ‘rate’ means per unit area rather than unit time.7) δ( W0 ) = where K2 πa(σ0 − σf )2 = . The factor 2 on the rhs of (4. but increases as the crack size increases. The time is measured from the time when the crack tip reached the point. Now consider a virtual extension of crack by δa. Breakdown-zone interpretation of slip-weakening process.1458 H Kanamori and E E Brodsky Figure 15. In all ﬁgures. (c) The temporal variation of stress at a point which is initially just beyond the crack tip. from (4. The time is measured from the instant when the crack tip reached the point.

The expressions for K are independent of the mode of crack. The expression for G∗ is the same for Modes I and II. The ratio of the dynamic energy release rate. When G∗ > G∗ the crack propagates dynamically and some energy is c radiated out of the system as seismic waves.11) Thus. this energy must be equal to twice the surface energy of the material near the crack tip. for a crack growing at a rupture speed V is given approximately by G = G∗ g(V ). . is called the fracture toughness (or critical stress intensity factor). Kc . Instead.10) arises because surface energy is deﬁned for each side of the crack. where g(V ) is a universal function of V . or the critical stress intensity factor.15) (4. K and G∗ are the important parameters in crack theory. the energy equation for virtual crack extension is no longer given by (4. cR (4. δ( W0 ) − δ(ER ) = 2Gδa. from which 1 ∂ER .10) where γ is the surface energy per unit area which is necessary to create a new crack surface. Rupture speed. For a Mode I (tensile) crack (Freund 1972) g(V ) = 1 − V . G.16) (4. In case of a static (or quasi-static) crack.14) (4.9). (4. Dynamic crack. Then. for the crack to be stable at halflength a. If G∗ given by (4. the stability of a crack can be discussed either in term of the critical speciﬁc fracture energy.10). Eshelby (1969) and Freund (1972) showed that the energy release rate.7) is divided between the virtual crack extension and the radiated energy. If we denote the radiated energy by ER . G. but considering the gross approximations used in seismological applications.9) is larger than G∗ . The stress intensity factor at this state.g. G∗ is called the critical speciﬁc fracture c c energy.The physics of earthquakes 1459 Static crack. which is related to G∗ by equation (4. That is.13) 2 ∂a where G is called the dynamic energy release rate. 2µ (4. to the static energy release rate. as is done in the Grifﬁth theory. the crack will grow. G∗ . but is slightly different for Mode III. the differences are not important. is given by a function of rupture speed V = da/dt. The total energy available for work from equation (4. G∗ . the crack extension is governed by G = G∗ − G = 2γ instead of (4. damaged zones) than just the normal area of crack extension.12) where cR is the Rayleigh-wave speed. i.e. Kc . c In seismic faulting. c (4. G∗ = G∗ ≡ 2γ . c G∗ = c 2 Kc . The factor 2 in (4. Kostrov (1966).8). we often generalize γ to include more surface area (e.

Marder and Fineberg 1996). δ( W0 ) − δER = 1 − from which G = G∗ g(V ). 4. For a Mode III (transverse shear) crack (Kostrov 1966. acoustic waves also experience a relativistic effect because information can only propagate through a ﬁnite distance in a ﬁnite time.21) 1 B2 V β 2 δ( W0 ) = 2Gδa (4. chapter 4. Like the electromagnetic case.2.19) B β where B is a constant of the order of 1 (Lawn 1993.17) (4. Then. In the limit of very small rupture speed.3. but the classical Mott’s (1948) theory is useful for understanding the basic physics. (4. where u is the displacement. g(V ) = 1 − (V /β) .22) has a similar form to relativistic contraction as used to calculate the electromagnetic ﬁeld around a particle as it approaches the speed of light. the dynamics can be dominated by the surface forces between the two sides. we quantify this frictional interaction in a way that is parallel to the crack theory so that we can combine the two formulations in section 4.22) (Bβ)2 Equation (4.1460 H Kanamori and E E Brodsky For a Mode II (longitudinal shear) crack (Fossum and Freund 1975) 1 − V /cR g(V ) = √ . Below. including the kinetic energy. the radiated energy ER is equated to the kinetic energy in the medium during rupture propagation and scales as ˙ ER ∝ a 2 u 2 ∝ a 2 ∂u ∂a 2 ˙ a2 ∝ a2 ∂u ∂a 2 V 2. 1 + (V /β) (4. As two surfaces slide past each other along a pre-existing fault. no energy is dissipated mechanically for a Mode III crack and all the energy is radiated in elastic waves. (4. W0 .20) . Frictional sliding Fault rupture can also be modelled as a frictional process.12) can be written as.23) In the limit of the rupture speed approaching the shear sound speed β. Eshelby 1969).18) The derivation of g(V ) given above is actually complicated. also scales as a 2 (4. ER can be written as 1 V 2 ER = 2 W0 . where V2 . (4.7).14) as g(V ) = 1 − G∗ g(V ) = 2γ . the relativistic contraction is irrelevant and g(V ) approaches unity. the equation corresponding to (4. Modes I and II cracks display the same phenomenon at different limiting velocities. 1 − V /β where β is the shear-wave speed. In Mott’s theory. The equation for dynamic crack extension is given from (4. Mott 1948. Because the strain energy. (4.

4. If δ = 0.and state-dependent friction model. According to this law. θ is governed by the following differential equation: ˙ θδ ˙ θ =1− . the coefﬁcient of static friction µs and the coefﬁcient of kinetic friction µk are the most fundamental parameters (note that µ is used for coefﬁcient of friction in this section rather than rigidity). If µk < µs . Scholz (2002)). it still can be related in general terms to simple physics. µ drops from µs to µk instantly.2. and is a key parameter in frictional sliding models. it drops to µk after a slip Dc . As they are deformed more quickly. Static and kinetic friction. is required before static friction drops to kinetic friction. or asperities. In this case the friction σf is a function of slip. θ is a state variable that accounts for the history of sliding and µ0 . Dc .g. and steady sliding begins (ﬁgure 16).e. The slip. is called the critical slip.and state-dependent friction. they have a greater resisting stress.1. (4.The physics of earthquakes 1461 (a) (b) Figure 16. Dc is the critical slip.24) is empirically derived. σ1 .25). then θ increases linearly with time (4. In the simple model. Speciﬁcally. The third term B ln θ describes the chemical adhesion between ˙ ˙ surfaces that is assumed to increase with contact time. static friction cannot drop to kinetic friction instantly.25) Although (4. (b) The frictional stress. Dieterich and his collaborators introduced the following friction law from experiments on many different materials (e. . A and B are constants. (a) The coefﬁcient of friction.24) ˙ where δ is the sliding speed.2. the state variable is simply the time of contact. Dc (4. the coefﬁcient of friction µ is given by ˙ µ = µ0 + A ln δ + B ln θ.2. Dc in the frictional sliding model is often equated to D0 of the critical slip of the slip-weakening crack model (cf ﬁgure 15(e)) 4. In the classical theory of friction. Dieterich (1979). A slip. but in general. ˙ The second term A ln δ represents a resistance similar to viscosity generated by deforming small irregularities on the sliding surface. In any physical system. instability can occur. The simple behaviour shown above can be generalized by a rate. Rate. The hatched area in ﬁgure 16(b) indicates the extra energy per unit area expended in the system compared with the case in which the friction instantly drops to the ﬁnal stress. Dc . Static and kinetic friction. i. σf is the frictional stress.

26) where θ0 is the value of θ at t = t0 . we obtain µ for t < t1 as Dc ˙ µ1 = µ0 + A ln δ1 + B ln ˙ δ1 For t t1 .and state-dependent friction law. and is the basis of the rate. and can be interpreted as the critical slip in the simple friction law shown in ﬁgure 16. Ruina (1983). Here.and state-dependent friction. then ˙ ˙ the friction initially increases. but eventually drops to µ2 = µ1 + (A − B) ln(δ2 /δ1 ) from µ1 . (4. let Dc be the slip over which the friction drops by ˙ ˙ ˙ ˙ (A − B) ln(δ2 /δ1 )/e. ˙ If δ = constant. ˙ ˙ ˙ For illustration purposes. Recent experimental works favour the form in (4. Dc and Dc must be distinguished. The cases with (A − B) < 0 and (A − B) > 0 represent velocity weakening (generally unstable) and velocity strengthening (generally stable). If (A − B) < 0.24). (4. then.25). Dc ≈ (10–20)Dc . For a ˙ ˙ ˙ ˙ small ratio of δ2 /δ1 . Dc (4. respectively. but for a very large ratio of δ2 /δ1 . we ˙2 /δ1 is very large. µ = µ2 where µ2 = µ1 + (A − B) ln . from (4.1462 H Kanamori and E E Brodsky Figure 17. θ is given by θ= Dc Dc + θ0 − ˙ ˙ δ δ exp − ˙ δ(t − t0 ) . ˙ ˙ t = t1 . from (4. For example.28) Dc + ˙ δ2 Dc Dc − ˙1 ˙ δ δ2 exp − ˙ δ2 (t − t1 ) Dc for t < t1 . Change in friction due to a sudden increase in sliding speed according to the rate. consider a case in which δ increases from δ1 to δ2 stepwise at ˙ = δ1 . Dc is proportional to Dc but it also depends on the velocity ratio δ2 /δ1 .g. θ= and µ = µ1 + A ln ˙ δ2 ˙ δ1 + B ln ˙ ˙ δ1 δ1 + 1− ˙ ˙ δ2 δ2 ˙ δ2 ˙ δ1 exp − ˙ δ2 (t − t1 ) Dc for t t1 . Then θ = Dc /δ1 . Dc ≈ Dc . Alternative forms of the evolution law for the state variable depend explicitly on slip rather than hold time (e. The constant Dc is a scaling parameter which determines the amount of slip over which the friction drops substantially. but if δ The behaviour shown in ﬁgure 17 is what was observed experimentally for many different kinds of materials (Dieterich 1979). Before t = t1 we assume that sliding is in steady state at δ and. Linker and Dieterich (1992)).25) (Beeler et al 1994). ˙ do not distinguish Dc and Dc . (4.27) As t → ∞.29) Figure 17 shows µ as a function of slip. .

Dc . The fracture energy. i. in crack theory is the energy needed to create new crack surfaces near the crack tip. With this method. Ohnaka and Shen (1999) proposed a scaling relation between Dc and the wavelength of the surface roughness.5). The ﬁnal value of the frictional stress σf is equal to σ1 . hydrodynamic lubrication (as discussed in section 4.4. u(t). Thus. (4. Then solving the equation of elastodynamics on the fault plane. it is now possible to determine a bound on Dc directly from seismograms (Ide and Takeo 1997).3. the critical slip.28)) are approximately 5 orders of magnitude less than the upper bound derived from seismic studies. is introduced as a critical slip before rapid sliding begins at a constant friction. the system must expend the threshold fracture energy Gc before the crack can extend. σ (t) as a function of time. Dc for large earthquakes is also estimated to be of the order of 1 m. Thus. With inversion of seismic data.The physics of earthquakes 1463 4.30) Thus. However. (4. Gc . Alternative slip-weakening processes include thermal pressurization. the energy spent in the system before this happens can be approximately written as (hatched area in ﬁgure 16(b)) 1 (σ 2 0 − σ1 )Dc .31) Past the initial breakdown-zone. σf is the same in the crack and friction models. 1 Gc = 2 (σ0 − σ1 )Dc . which tends to smooth the resulting σ (u) versus u relationship. Rupture energy budget Since the crack and frictional processes are linked through the fracture energy Gc . Any constraint on fracture energy obtained from the energy budget will provide a strong bound on all microscopic rupture processes. Ide and Takeo (1997) found Dc to be of the order of 0. Therefore. can be determined as a function of time. Dc .e. Marone and Kilgore (1933) suggest that Dc is controlled by the thickness of fault gouge layers as well as the surface roughness. we can relate the macroscopically observable energy budget to the microscopic processes in a surprisingly general way. For example. we can determine the shear stress. With the recent availability of high-quality seismograms at short distances. The values of Dc determined by laboratory friction experiments (equation (4. or the slip-weakening process at large slips is different from that of laboratory friction process. we can equate equation (4. .30) to Gc . we conclude that either the seismically determined bound on Dc is so extreme that comparison with the laboratory values is not meaningful. If the frictional stress varies more or less linearly as shown in ﬁgure 16. in frictional sliding model. if we are to link a crack model to a friction model. In contrast.5 m in the deeper part of the fault plane. plastic deformation and micro-fracturing in the crust surrounding the fault. in the process of inversion low-pass ﬁltering is applied to the data. Direct determination of Dc . For the 1995 Kobe earthquake. We need to link these two models to understand the way these models are used in seismology. the slip at a point on the fault plane. thus determined is an upper bound. Eliminating t from u(t) and σ (t) leads to the slip dependence of stress σ (u) from which Dc can be estimated. Mikumo et al (2003) developed a method to estimate Dc directly from slip-velocity records using elastodynamic modelling. Sometimes the terminology of crack mechanics is used and at other times the terminology of friction mechanics is used. 4. The link between the crack model and the friction model Crack and frictional sliding models are frequently used in seismology to explain various aspects of earthquake phenomena.

or of a speciﬁc friction law such as the rate. the stress on the fault plane is σ1 (ﬁnal stress) and the average slip (offset) is D. in that order. From equation (3. (b) Slip-weakening model: hatched and cross-hatched areas indicate the fracture energy and frictional energy loss.and state-dependent friction law (Dieterich 1979) (ﬁgure 17)..1. In fact. The solid lines in ﬁgure 18(a) show the simplest case. the potential energy (strain energy plus gravitational energy) of the system. some seismological inversion studies have shown this . An earthquake is viewed as a stress release process on a fault surface S . i. respectively.4). Graphically. All the ﬁgures are shown for unit area of the fault plane. and the seismic wave is radiated carrying an energy ER . (c) The energy budget: hatched.1). W = σ DS. The difference σs = σ0 − σ1 is the static stress drop. (a) ——: simple case of immediate stress drop.1464 H Kanamori and E E Brodsky Figure 18.14). ¯ (4. and stays there. Scholz (2002). also section 6. Illustration of simple stress release patterns during faulting. cross-hatched and dotted areas indicate the fracture energy. σf = σ1 . drops to W1 = W0 − W where W is the strain energy drop. . If the condition for instability is satisﬁed (Brace and Byerlee (1966). During this process. the initial (before an earthquake) shear stress on the fault plane σ0 drops to a constant dynamic friction σf . At the end. the stress may increase to the yield stress σY in the beginning of the slip motion (curve (1) in ﬁgure 18(a)) because of loading caused by the advancing rupture (ﬁgure 15(e)). rapid fault slip motion begins and eventually stops. For example. W ¯ (for unit area) is given by the trapezoidal area shown in ﬁgure 18(c).1.32) where σ = (σ0 + σ1 )/2 is the average stress during faulting (section 3. At the initiation of an earthquake. The variation of stress during faulting can be more complex than shown by the solid lines in ﬁgure 18(a).e.-: general case without slip-weakening.. W0 . thermal (frictional) energy and radiated energy.

The area under the trapezoid outlined by the heavy lines represents the total potential energy change. As discussed in section 3. nevertheless it is based on the breakdown-zone interpretation of the slip-weakening behaviour and is useful for interpretation of the energy budget. If friction is not constant. Thus.36) 2 2 2µ where M0 = µDS is the seismic moment. Because W ER +EG . (4. For instance. ER = W − σf DS. here we assume that the ﬁnal stress σ1 is equal to σf0 .35) As we discussed earlier. If ηR ≈ 1. ¯ (4. the microscopic breakdown process is or friction-dominated. ηR η. From ﬁgure 18(c). The radiation efﬁciency. which is given by ER / W . the total energy dissipation is given by σf = ¯ S 0 D σf (u) du = σf DS. the total dissipated energy σf DS can be divided into EG . EH . the radiated energy. ¯ (4. Bouchon 1997. the rupture dynamics is complicated. Then. and ¯ the frictional energy. Also. the friction may drop drastically in the beginning and later resumes a somewhat larger value (curve (2) in ﬁgure 18(a)).37) ER + EG is called the radiation efﬁciency and is an important parameter which determines the dynamic character of an earthquake (Husseini 1977). The ratio. We should note that this partition is model dependent. we obtain σ0 − σ 1 σs σs ER = DS − EG = DS − EG = M0 − EG . the amount of slip during this high stress stage is small so that little energy is involved.4. ER . However. Kikuchi and Fukao (1988).34) Figure 18(c) shows the partition of energy. On the other hand. we deﬁne the average friction σf by ¯ 1 D σf (u) du. η. EG . . if ηR dominating the dynamics. (4. represented by the cross-hatched area in ﬁgure 18(c).33) D 0 where u is the slip (offset) on the fault plane. (For simplicity. ER ηR = (4. ηR . Beroza and Mikumo 1996. is called the strength excess. or it may decrease gradually to a constant level (ﬁgure 18(b)). W . σY − σ0 . Heaton (1990) and Kikuchi (1992). the friction may not be constant during faulting. we formulate this problem referring to a simple case shown in ﬁgure 18(b). Ide and Takeo 1997). Mikumo and Miyatake 1993.The physics of earthquakes 1465 increase (Quin 1990. is the dotted area. this behaviour in which the stress on the fault plane gradually decreases as slip increases is often called slip-weakening (Rice 1980. Miyatake 1992. the breakdown zone is unimportant and failure occurs primarily in the steady-state regime regardless of whether it is crack-like 1. the hatched area in ﬁgure 18(c) is the fracture energy. As discussed before. we will not include it in our simple energy budget. Thus.1. Then. The stress difference. Li 1987). is different from the seismic efﬁciency.) Then. Slip-weakening models have been considered in seismological models by Brune (1970). The friction σf gradually drops to a constant value σf0 until the slip becomes Dc . W cannot be determined directly by seismological methods and the seismic efﬁciency is difﬁcult to determine. The area under the curve labelled as σf is the total dissipated energy. Then. if we use the slip-weakening model. but for the energy budget considered here.

and analysed in great detail by Lachenbruch (1980). and compute T as a function of magnitude with three parameters. the temperature can easily rise by 100–1000˚C during a moderate-sized earthquake. we consider a gross thermal budget during faulting under a frictional stress σf . One of the ﬁrst such special mechanisms recognized was frictionally-induced melting. Let S and D be the fault area and the displacement offset. 1987) and Andrews (2002). we now turn to the micromechanics of rupture beyond solid friction and crack models. ηR for small earthquakes must be signiﬁcantly less than unity.1. Mw . scaled energy.39) 2µ e. If σf is comparable to σs about 10 MPa. Richards (1976) and Cardwell et al (1978) quantitatively conﬁrmed the potential importance of frictional heating during faulting.2.e. 1–10 MPa.36) and (4. Studies by McKenzie and Brune (1972). Fault-zone processes: melting.40) where C is the speciﬁc heat and ρ is the density. the effect of shear heating is signiﬁcant. Mase and Smith (1985. Under the . Then the total heat generated during faulting is Q = σf DS. Geological outcrops of faults suggest that many faults are ﬁlled with aqueous ﬂuids or a viscous mixture of gouge and water when active. The silicate melt then reduces the friction for the remainder of the earthquake. One possibility is thermal pressurization of the ﬂuid. we obtain a relation between ηR and observable seismological parameters: ηR = where e= ˜ ER M0 (4. and values of e for small earthquakes are often 1–2 orders of ˜ ˜ magnitude less than those for large earthquakes. Melting. early in the rupture. i. Here.38) is the radiated energy scaled by the seismic moment. CρSw Cρw (4. 4.1466 H Kanamori and E E Brodsky By combining (4. If it is real. e is always less than 10−4 . and a shear modulus. then another set of mechanisms are at work during rupture. σf . ﬂuid pressurization and lubrication Motivated by the importance of the non-elastic slip-weakening processes in the energy balance. The small values of ηR for small earthquakes motivate us to examine the micro-mechanisms of slip-weakening and breakdown. w and the static stress drop σs . Whether the trend shown in ﬁgure 12 is real or not is currently debated (Ide and Beroza 2001). the average temperature rise T is given by T = σf D Q = . 3 × 104 MPa. This concept was introduced to seismology by Sibson (1973). In general D increases with the earthquake magnitude. If we assume that the heat is distributed during seismic faulting within a layer of thickness w around the rupture plane and there is negligible heat transport away over the timescale of the earthquake. respectively. If this is true. Using the scaling relations given in section 3. As shown in ﬁgure 12. Thermal ﬂuid pressurization. ˜ σs (4. we can relate D to Mw (with the static stress drop σs as a parameter). frictional dissipation may be high enough.5. As ﬁrst suggested by Jeffreys (1942). Figure 19 shows T for the case with w = 1 cm as a function of Mw .37). for typical values of static stress drops. If the thermal energy is contained within a few centimetres around the slip plane during seismic slip. to melt the wallrock.

The static stress drop σs is assumed to be 10 MPa. the thermal expansivity of water is of the order of 10−3 ˚C−1 . They suggest that mechanisms that are consistent with the extreme localization of slip. several narrow slip zones were found but evidence shows that each slip zone represents a distinct slip event (i. but this question remains debatable (Sibson 2003). The analysis of Lachenbruch (1980) and Mase and Smith (1985. in reducing friction. The most important parameter controlling the pressure change is the permeability. Some fault zones have a very narrow (about 1 mm) distinct slip zone where fault slips seem to have occurred repeatedly. 1987) suggests that if permeability is less than 10−18 m2 . exhumed faults). with the frictional stress. the pressure increase would be of the order of 1 MPa per ˚C (Lachenbruch 1980).The physics of earthquakes 10 10 10 8 1467 7 6 σs ∆s s=10 MPa. Mw . California. A modest T of 100–200˚C would likely increase the pore pressure enough to signiﬁcantly reduce friction. degree 10 10 10 10 10 10 10 5 4 σs f=1 G f Pa 3 2 1 100 10 Pa 1M 0 -1 1 2 3 4 5 6 7 8 9 10 Mw Figure 19. The key question is: what is the thickness w of the fault slip zone? Geologists have examined many old fault zones which were formed at depths and were brought to the surface by long-term uplift (i. such as thermal pressurization of pore ﬂuids. According to Chester and Chester (1998). Although the distribution of permeability can be complex. If the ﬂuid does not escape (small permeability) and the surrounding rock is not compressible.e. Figure 19 shows that this moderate temperature increase can occur even for intermediate-sized earthquakes (Mw = 3–5). and the friction will drop signiﬁcantly. w=1 cm ∆T. Permeability in the crust varies over a very wide range of more than a factor of 1010 . and signiﬁcant increase in pore pressure with temperature could occur. Thus. in a fault zone as a function of magnitude. T . at least locally. an earthquake). ﬂuid pressurization is most likely to occur with a temperature rise of less than 200˚C. the internal structure of the Punchbowl fault. geological evidence suggests a narrow slip zone. at least for some faults. permeability and compressibility vary and the pressure increase may be less. pressure–temperature conditions at the seismogenic depths. implies that earthquake ruptures were not only conﬁned to a layer of ﬁnely shattered rock. Predicted temperature rise. σf . . In other cases. are most compatible with their observations.e. as a parameter. but also largely localized to a thin prominent fracture surface. In actual fault zones. these studies suggest that ﬂuid pressurization can play an important role.

1468 H Kanamori and E E Brodsky Lubrication. is directly related to EG by (4. and is subtracted from the total potential energy.6. we do not necessarily expect a single continuous layer of melting and pressurization. no single process is likely to dominate. it is also possible that elastohydrodynamic lubrication plays an important role in reducing friction for large events (Brodsky and Kanamori 2001). the fracture energy. no energy is dissipated mechanically and the potential energy. In contrast. Taiwan. the thickness of fault slip zones is the key parameter for understanding fault dynamics. and if the material in the fault zone behaves as a viscous ﬂuid. If ηR = 1. the largest ever recorded). 4. EH . a fault zone that consists of many regions where different mechanisms are responsible for slip at different stress levels. but in view of the large slip and slip velocity associated with seismic faulting. represents the mechanical energy loss during faulting. we envision. In these discussions. permeability. During the 1999 Chi-Chi.4. producing complex rupture patterns as observed. a signiﬁcant drop in friction is likely to occur if the slip zone is narrow. whether this is indeed a general behaviour or not is yet to be seen. ER and σs .38).2).6. the hydrodynamic pressure within a narrow zone becomes large enough to smooth out the irregularities on the fault surface by elastic deformation. we can estimate the radiation efﬁciency. ﬁrst we describe how we can determine ηR from macroscopic seismic parameters (Venkataraman and Kanamori 2004). As shown by equation (4. if ηR = 0. the transverse shear gradients in the ﬂuid are balanced by the longitudinal pressure gradients and the pressure increases on the leading edge of irregularities in the fault surface. In other words. Since a fault zone is probably complex and heterogeneous in stress. but short period acceleration was not particularly strong and the shaking damage was not the worst (Ma et al 1999). ﬂuid content. after heat loss has been subtracted. A high heat ﬂow over the aseismic section would be consistent with dynamic weakening reducing the frictional dissipation where there are earthquakes. the event is quasi-static and no energy is radiated. the observed ground-motion near the northern end of the fault was extremely large (>3. The interpretation of macroscopic seismological parameters Radiation efﬁciency. since this is the only earthquake for which such large slip and slip velocity were instrumentally observed. Since the radiation efﬁciency. porosity and compressibility. An interesting consequence of this is that as the slip and slip velocity increase. However. it does not directly control the dynamics of earthquake rupture.5 m s−1 . thereby suppressing high-frequency ground motion caused by the fault surfaces rubbing against each other. There is some evidence that the heat ﬂow is slightly elevated over the section of the San Andreas fault that slips gradually with no large earthquakes (Sass et al 1997). Of course. in the breakdown-zone interpretation of the slip-weakening model. ηR . even if the static stress drop is very large. the energy deﬁned by the cross-hatched area in ﬁgure 18(c) is interpreted as frictional thermal energy. EG . In contrast. is radiated as seismic waves and the earthquake is considered a very ‘brittle’ event. As we discussed in section 4. Thus.37). M0 . Any of these dynamic weakening mechanisms can explain the lack of elevated heat ﬂow over seismogenic faults (section 2. the determination of fracture energy for earthquakes is critically important for understanding the dynamics of faulting. instead. . using the three macroscopic seismological parameters.1. As in a ball bearing. Linking processes to the seismic data 4. and controls the fault dynamics in a fundamental way. ηR . whether lubrication occurs or not depends on many factors such as the effective permeability in the fault zone. This could be a manifestation of the high-speed lubrication effects (Ma et al 2003). If a fault zone is narrow and slightly rough. compressibility of fault rocks and the viscosity of melts. earthquake.

This trend is roughly consistent with the result of Houston (2001) who found that the source duration of deep earthquakes with comparable magnitudes is systematically shorter than that of shallow earthquakes. the seismic moment. seismic moment and static stress drop. For shallow earthquakes. σs is in the range of 20–200 MPa. Tsunami earthquakes (earthquakes with slow deformation at the source which generate tsunamis disproportionately . we need to determine the stress drops for individual earthquakes for this purpose. is a little more difﬁcult. For the deepest earthquakes. We can see a general trend of σs increasing with depth.38) (Venkaraman and Kanamori 2004).2. suggests a trend similar to that shown in ﬁgure 20. Crustal: earthquakes which occur within continental crusts. Using the estimates of radiated energy.1 (ﬁgure 11) shows that most large earthquakes have comparable stress drops in the range of 1–10 MPa. For many large earthquakes. as discussed in section 3. similar to the S-wave speed.The physics of earthquakes 1469 Figure 20. Figure 20 shows the estimates for large earthquakes. σs is in the range of 1–10 MPa. For a few earthquakes the computed ηR is larger than 1. The radiation efﬁciency of most earthquakes lies between 0. and the average is roughly 10 times larger than that for shallow earthquakes. ER have been determined. Mw . and the radiated energy. The determination of the static stress drop.2. Deep: deep earthquakes. This result. we can determine the radiation efﬁciency for all these earthquakes using (4.25 and 1. Although the scaling relation discussed in section 3. σs . Figure 21 shows the radiation efﬁciencies as a function of the magnitude. on the average. Tsunami: earthquake with a slow deformation at the source which generates tsunamis disproportionately large for its magnitude (ﬁgure taken from Venkataraman and Kanamori (2004)). M0 . Static stress drop plotted as a function of depth for the different types of earthquakes. Downdip and Interplate: earthquakes which occur on the subduction-zone plate boundary. Intraplate: earthquakes which occur within the lithospheric plate. if interpreted using the assumption that the rupture speed is. This is probably due to the errors in the estimates of radiated energy and/or stress drops.

The result showed that W0 = 1.25) and the two deep earthquakes. This energy 1. the 1999 Russia–China border event and the 1994 deep Bolivia earthquake. which is only 3% of W0 . the earthquake observed as seismic waves.1470 H Kanamori and E E Brodsky Figure 21. The mechanical part of the process. but tsunami earthquakes and two of the deep earthquakes (the Bolivia earthquake and the Russia–China border earthquake) have small radiation efﬁciencies (ﬁgure taken from Venkataraman and Kanamori (2004)). however. Radiation efﬁciency ηR = ER /(ER + EG ) as a function of Mw . large for its magnitude). For the 1994 Bolivian earthquake (Mw = 8. and the difference W0 − ER = 1.3. have small radiation efﬁciencies (<0. fracture and thermal energy at least comparable to that released by a large volcanic eruption must have been released in a relatively small focal region.35 × 1018 J. the largest deep earthquake ever recorded.e.25. about 50 × 50 km2 . and must have been deposited near the focal region. probably in the form of fracture energy in addition to the frictional energy. i. the source parameters could be determined well enough to investigate the energy budget (Kanamori et al 1998). the Bolivia earthquake should be more appropriately viewed as a thermal process rather than .35 × 1018 J is comparable to the total thermal energy released during large volcanic eruptions such as the 1980 Mount St Helens eruption. was not radiated.4 × 1018 J and ER = 5 × 1016 J. The different symbols show different types of earthquakes as described in ﬁgure 20. is only a small part of the whole process. In other words. have small radiation efﬁciencies. depth = 635 km). within a matter of about 1 min. Most earthquakes have radiation efﬁciencies greater than 0. Thus.

the 1999 Russia–China border event and the tsunami earthquakes.The physics of earthquakes 1471 a mechanical process.18)) are also plotted in the same ﬁgure. also depends on the thickness of the fault zone. but for large earthquakes. but it is likely that the rupture process in the pressure–temperature environment at large depths may involve large amounts of plastic deformation with large amounts of energy dissipation.3. ER + EG W0 (4.41) from which we obtain ηR = 1 − g(V ). How much of the non-radiated energy goes to heat depends on the details of the rupture process. This conclusion seems to be supported by the independent observations of the high rupture speed V . As discussed in section 4. Figure 22 shows the upper and lower limit of radiation efﬁciencies that were determined from the energy budget plotted against the upper and lower limit of V /β obtained from the literature. but if it is of the order of a few centimetres. ηR . For tsunami earthquakes (slow seismic events) and some deep earthquakes. radiated energy ER and the static stress drop σs ). Since rupture speed is an independently determined quantity. to rupture speed. we showed that for most earthquakes.16)–(4.1 to 0. the observed data follow the theoretical curves obtained from crack theory. To the ﬁrst order. However.1. we can relate the radiation efﬁciency. Since the fracture energy EG can be interpreted as the integral of G over the entire fault surface S.75 and 0. T . it is possible that a substantial part of the non-radiated energy was used to raise the temperature in the focal region signiﬁcantly. Most of these earthquakes have rupture speeds such that the ratio of rupture speed to shear wave speed (V /β) is between 0. The mechanism of large deep earthquakes is not known well. (4. the energy release rate for a dynamic crack G is given as a function of the rupture speed V (equation (4.25. we can use equations (4. V as follows.1. this consistency of the observed relationship between ηR and V /β with the calculations from crack theory enhances the results shown in ﬁgure 21.2. and a large amount of energy is used in deformation. which means that the rupture process of these earthquakes involves more dissipative processes than the average. With the three seismologically observable macroscopic parameters (seismic moment M0 .9) to write EG = and ηR = ER ER = .7) and (4. The actual temperature rise. have small V /β. which is not known. II and III cracks (equations (4.15)). the radiation efﬁciency is small. .3. However. The theoretical curves relating radiation efﬁciency to rupture speed for Modes I. the 1994 Bolivia earthquake. Note that this line of reasoning poses no constraint on the energy EH dissipated directly as heat.42) G dS = g(V ) G∗ dS = g(V ) W0 (4. which is unknown. the estimates of rupture speed are fairly accurate. The relation between radiation efﬁciency and rupture speed.43) The average rupture speed is usually determined from the inversion of seismic waves and the results can be non-unique. which means that the amount of energy mechanically dissipated during rupture is comparable or smaller than the energy radiated as seismic waves. the radiation efﬁciency which is given by ηR = ER /(ER + EG ) is larger than 0. about 0. One interpretation is that most tsunami earthquakes involve rupture in soft deformable sediments. Summary and implications. the temperature could have risen to above 10 000˚C (ﬁgure 19). In the simplest model discussed in section 4.95.

relatively small critical fracture energy Gc or small fracture toughness Kc (4. for most shallow earthquakes has an important implication for rupture growth of earthquakes.1472 H Kanamori and E E Brodsky V/ Figure 22. it will be difﬁcult to stop the rupture dynamically. Dc (equation (4. if Kc is small. Radiation efﬁciencies determined from the radiated energy-to-moment ratios and estimates of static stress drop (equations (4.e. . If this is the case. i. the small Gc or Kc means small critical slip.3)). the length scale a increases and K increases (equation (4. II and III cracks have also been plotted (ﬁgure taken from Venkataraman and Kanamori (2004)). To stop the rupture.31)). The symbols are the same as in ﬁgure 20. once an earthquake is initiated.39)) plotted against the estimates of the ratio of rupture speed to shear wave speed obtained from literature. as discussed in section 4.11). In terms of the friction model discussed in section 4.1.3. the theoretical curves relating radiation efﬁciency to rupture speed for Modes I. As a rupture grows.3)). As discussed in section 4. on the average. the rupture growth is controlled by the balance between the dynamic stress intensity factor K and Kc . The relatively large radiation efﬁciency.3. the rupture is more likely to grow and develop into a runaway rupture. for comparison.5.38) and (4. Thus. the tendency for runaway would increase because K is also proportional to (σ − σf ) (equation (4. some external static features such as a strength barrier or irregular fault geometry may be required. If friction decreases as the slip increases.

which have a magnitude greater than or equal to M is given by the relation log N (M) = a − bM.5. Earthquakes as a complex system Another possible approach to understanding why earthquakes happen is to take a broad view beyond a single event. the accuracy of the macroscopic source parameters. This approach has had considerable success characterizing the types of models that will reproduce the observed magnitude–frequency relationship (i.5.e. a recent review is found in Utsu (2002). 6. on the average. . and to draw more deﬁnitive conclusions on the rupture dynamics of earthquakes. Note that. M = 5. in a given area. especially ER and σs . N (M) is the number of earthquakes per year with the magnitude M. The solid line shows a slope of −1 on the semilog plot which corresponds to a b-value of 1. and N may be slightly underestimated. The data sources are as follows: M 8. Magnitude–frequency relationship for earthquakes in the world for the period 1904 to 1980. In general small earthquakes are more frequent than large earthquakes. Approximately one earthquake with M 8 occurs every year somewhere in the Earth.The physics of earthquakes 1473 10 10 10 N(M) 5 4 3 10 10 10 2 1 0 10 -1 3 4 5 6 7 Magnitude. 6. is not good enough to accurately estimate the fracture parameters Gc . the catalogue may not be complete. 5.) It describes the number of earthquakes expected of each size. or magnitude. Currently.1) where a and b are constants. The goal is to ﬁnd systems that robustly reproduce the general patterns of seismicity regardless of the details of the rupture microphysics. We can study earthquakes by dealing with large groups of earthquakes statistically. M 8 9 Figure 23. 7. Kc and Dc . Gutenberg–Richter relation) used in seismology. At present. for the period from 1976 to 2000 from Ekstrom (2000). extensive efforts are being made to improve the accuracy of determinations of the macroscopic source parameters.0. for the period 1904 to 1980 from Kanamori (1983). the number of earthquakes.5. This is quantitatively stated by the Gutenberg–Richter relation (Gutenberg and Richter (1941). The magnitude–frequency relationship (the Gutenberg–Richter relation). N (M). M = 4 and 5. approximately one earthquake with M 8 occurs every year. Figure 23 shows that the Gutenberg–Richter relationship even applies to a seismicity catalogue encompassing the entire planet. In any area much larger than the rupture area of the largest earthquake considered.0 and 7. for the period January 1995 to January 2000 from the catalogue of the Council of National Seismic System. For this range. (5.

) The primary conclusions of this research are that over a wide (but ﬁnite) range of scales. As the mass–spring system is dragged. mass and friction. some blocks slip intermittently. Three speciﬁc models for generating the critical state in earthquake processes are the mechanical slider-block system. Because of the ubiquity of the Gutenberg–Richter distribution. it is important that the empirically-derived Gutenberg–Richter Law be put on ﬁrm physical ground. p. (For more extensive discussions on this subject. then e = ps is the expectancy of the number of patches . and when plotted on a log–log diagram. Hergarten 2002). In the slider-block system (Burridge and Knopoff (1967). The starting point of complexity models is the assumption that the initiation. and the whole continuous failure corresponds to an earthquake rupture. Because of the large number of elements involved and of the complexity of the interaction it may not be determinable exactly how different parts of the crust interact with each other. Rundle et al (2000) and Turcotte and Malamud (2002). Cascades of failure result when the faults are extremely close to failure prior to any late-stage triggering. then it can trigger failure of the adjacent patches with some probability. well enough to understand the earthquake process in a deterministic way. Simple models. and some potential energy is released. see Main (1996).1474 H Kanamori and E E Brodsky For most regions the value of b is nearly 1. The Gutenberg–Richter Law is a major tool in probabilistic hazard assessment. The friction between the block and the surface is governed by a simple velocity-weakening law. This process continues until it stops spontaneously. Turcotte (1997). This slip is interpreted as a small ‘earthquake’. occasionally slip of a block triggers slip of adjacent blocks causing a larger earthquake with a larger amount of potential energy release. to the likelihood of large events. ﬁgure 24(a)). In practice. Nevertheless. After a series of events have occurred. earthquakes are thought to be self-organized critical systems. systems are only critical within a certain range of scales determined by the overall boundaries of the system. These cascades can be interpreted as an example of self-organized criticality. In the critical state. the relation between N (E) and E exhibits a power-law like behaviour. If more than one block are triggered. Given the societal importance of the resulting hazard assessments. However. the interaction between the blocks is described by a differential equation that involves the spring stiffness. In a percolation model (Otsuka 1971). In this slider-block model. As shown in ﬁgure 24(b). Self-organized criticality is when a system evolves to the critical state naturally without any dependence on the initial conditions or tunable parameters (Bak et al 1988. we have a larger earthquake. as is the case in ﬁgure 23. The crux of a successful earthquake complexity model is the recreation of critical behaviour by setting up a system of elements with only local interactions being speciﬁed. some properties of earthquakes such as the magnitude–frequency relationship can be understood as manifestations of the general behaviour of complex systems. which we observe easily. a seismic fault is modelled by a distribution of many small elements (patches). the percolation model and the sand pile model. It allows extrapolation from the rates of small earthquakes. A critical state is when events of all sizes can occur and their frequency distribution follows a power law. If one patch fails. the local interactions can accumulate to generate long-range organization. Most of the time. many blocks are connected by a spring and the whole mass–spring system is dragged on a frictional surface. growth and cessation of earthquake rupture are controlled by the complex interaction of faultbounded blocks on scales as small as individual cracks and as large as continents. If a patch can trigger nearby patches at s sites with a transition probability. a single block slips. the relation looks like the earthquake magnitude–frequency relationship. we count the number of events N (E) which released a potential energy larger than E. fault networks are fractal. This strikingly consistent observation has motivated much of the study on fault networks and self-organized criticality.

Figure 25 shows the results of numerical simulations performed for s = 10 and three cases. there is also a small but ﬁnite probability that a large slide can happen as a result of simply adding one particle. in which at least F patches failed. If e < 1.8. When e = 1. However. This corresponds to a sub-critical state. Usually a small number of sand particles slide off the pile and into adjacent areas as a result of the small additional stress. we ﬁnd a relation similar to the magnitude–frequency relationship between log F and the number of cases. to be triggered at each step. If this whole process is repeated many times. then the growth will eventually stop at a certain step when a total of F patches have failed. Bak et al (1988) used a set of simple rules that simulate the actual potentially complex physical . (b) The relation between the number of events and released potential energy (Burridge and Knopoff 1967).9 and 0. (a) Mass–spring system sliding on a frictional surface (Burridge and Knopoff 1967). e = 0. This distribution corresponds to the magnitude–frequency distribution shown in ﬁgure 23. In a sand pile model. i. the system is critical.99.e. the range of interactions is limited. n. a sand particle is added to a sand pile which has been maintained in a critically stable state.The physics of earthquakes 1475 (a) (b) Figure 24. 0.

6. This interaction can be due to static as well as dynamic processes. In all these models. what is essential is the interaction between the many elements which make up the system. Even in simple high-school formulations. observable parameters. the interaction is between different parts of a fault and between different faults. We are now ready to use these tools to broach the key problem of earthquake triggering and instability. friction depends on slip velocity as dynamic friction is less than static friction. Another commonly observed seismicity pattern is the timing of aftershock decay which is known as Omori Law.1476 H Kanamori and E E Brodsky Figure 25. We will ﬁrst deal with frictional instability and then approach data that has bearing on other types of initiation. interaction between different cells.2.1. As we will discuss later (section 6. Kadanoff 1991). In the above.1. Omori’s Law can be explained with several physical models. The robustness of the Gutenberg–Richter result tells us that this particular feature of seismicity is insensitive to the microscopic physics controlling the failure and rupture. we discussed only the magnitude–frequency relationship. Instability and triggering We have now covered the major pieces of the earthquake puzzle: stress in the crust. Friction on a fault is not constant. Unravelling the limits and extent of the chaos is prerequisite for determining whether or not individual earthquakes are predictable over societally useful timescales given the limited resolution of our observations of the initial conditions. Instability 6. Stick slip and instability. macroscopic observations.1. In the actual earthquake process.4). The result of the interaction yields a relationship like the earthquake magnitude–frequency relationship (Bak and Tang 1989. Magnitude–frequency relationships produced with a percolation model (ﬁgure taken from Kanamori and Brodsky (2001)). such as the likelihood of a particular earthquake on a particular fault. 6. micromechanics and complex systems. . The details of the microphysics are only important for addressing other questions. The complexity models also show that earthquake interactions have inherently chaotic elements in addition to the predictable elements governed by the stress loading mechanism and fault structures.

More precisely. the broken line moves upward. in that order. it occurs in a stop-and-go fashion. the slip-velocity dependence can be more complex and friction also varies as a function of time and slip distance. (6. Suppose we increase δ0 by pulling the spring from the right. k. At point C. (a) A spring-loaded slider block on a frictional surface.2) (6. Rabinowicz (1995)) illustrated by a springloaded slider-block model similar to that discussed in section 5. The area A1 is approximately equal to A2 . (b) Variation of friction as a function of displacement (——) and the spring loading force (– – –). Therefore. Figure 26 shows a mechanism of stick slip (e. In this ﬁgure. Between points B and C. and the block moves abruptly along C–D driven by the spring force.The physics of earthquakes 1477 (a) (b) Figure 26. given by the rhs of (6.1) The solid curve in ﬁgure 26(b) shows the variation of τ as a function of δ. δ0 .1). the displacement of the right-hand end of the spring.1).g. This behaviour is generally called stick slip (Brace and Byerlee 1966). B. τ drops suddenly and the spring force exceeds τ . As δ0 increases. Stick slip model.3) . the block moves over the surface smoothly. where τ = µσn . σn and τ are the spring constant. and the intersection moves to B . τ =k δ (6. Then. The intersection. sliding does not occur smoothly. between the broken line and the solid curve gives the equilibrium position given by equation (6. from where smooth motion begins again. The broken line in ﬁgure 26(b) is the loading force exerted by the spring. the force balance is given by τ = k(δ0 − δ) = −kδ + kδ0 . the normal stress and the friction between the block and the surface. δ. The block is stationary at D until the spring force reaches point E with the increase of δ0 . In realistic fault conditions. the displacement of the block. at point C.

Stiffness of the fault system. we deﬁne a critical fault length ˜ scale Ln : EDc ˜ ˜ Ln ≡ L ≈ .1478 H Kanamori and E E Brodsky ˜ Figure 27. In the models described above. the stress required to cause a ˜ slip D is given by ED/L. If we assume that the laboratory measurements of µs . on a frictional fault plane.g. where E is a relevant elastic modulus. Consider the nucleation of a slip on a frictional surface with the normal stress σn . µs . at the critical point.7). For the latter values.3)). Dc (ﬁgure 27). µk . Then from the stick-slip model (equation (6.e. Thus. Then the question is: what is the stiffness of the crust? Stiffness is deﬁned by the ratio of stress to displacement. the spring constant (stiffness of the system) controls the stability for a given frictional property of the surface. Thus. µk and Dc on rocks are appropriate for natural faults. plays a key role in determining the stability. e. if we ˜ consider a small crack with length scale L in a nucleation zone. and critical slip.7) (µs − µk )σn ˜ According to the frictional instability model. Nucleation zone.6) Then. (6.g. For a ˜ ˜ rupture to grow. the static friction coefﬁcient.7) is the nucleation length of this frictional surface. the slope of the broken line increases. Then. k = τ/δ. (6.4) δ If the spring constant k is large. µs − µk ∼ 0. ˜ L 6.5) kf ≈ . and the stick slip instability begins past this point where τ > k. kf Dc = σn (µs − µk ). the strain Dc /Ln in the nucleation zone prior to the . combining this expression with the deﬁnition of kf . A schematic ﬁgure showing the nucleation length. and no instability occurs. the stiffness of the spring.1. k. then Ln ≈ 3 cm. stable sliding occurs without stick-slip behaviour. (6.2. For a more sophisticated frictional model. L must be larger than Ln given by (6. the fault stiffness can be deﬁned by E (6. i. Lapusta ˜ and Rice (2003)).1 and E is ∼5 × 1010 Pa. kinetic friction coefﬁcient. L. at typical seismogenic depths where the normal stress is ˜ ∼200 MPa. Dc is ∼10 µm. Ln in (6. the nucleation zone can be as large as 1 m (e.

in others it is compressional. Therefore. Large earthquakes are followed by abundant smaller earthquakes called aftershocks (section 6. .The physics of earthquakes 1479 earthquake is of the order 10−5 . The pattern of dilatational strain can most easily be seen for the simple example of a strike-slip fault (ﬁgure 28. As discussed in section 3. the deformation of the crust by a slip on a fault plane generates an elastic strain ﬁeld surrounding the fault. as the far-ﬁeld mechanism must also operate in the near-ﬁeld. If −1 is supported.2. right). As one of the few cases in nature where the immediate cause of an earthquake is apparent. we will retain the separation here with the above caveats. on a given fault orientation. Triggering During the past decade. Following the Anderson. we deﬁne the Coulomb stress change τc on a fault by τc = −µ( σn − p) + τ. therefore.8) where σn and τ are the resolved normal and shear stress changes. This relatively large strain could potentially be observed up to ∼80 m away on modern strain metres.2. 6. A geophysicist would have to be very lucky to choose to put an instrument within 80 m of the right 1 m2 patch of the 15 000 km2 San Andreas fault! The nucleation length model in (6. Aftershocks form a cloud around the mainshock rupture plane that can extend up to two fault lengths away.1. studies such as those by King et al (1994) investigated the proposal that aftershocks are triggered by the static stress changes due to the dislocation of the earthquake. 6.7) suggests that the smallest possible earthquakes have magnitudes Mw = −1 (from equation (3. Beginning in the early 1990s. Mw = −1 corresponds to M0 = 4 × 107 N m). This distinction may be artiﬁcial. Observations. with a different set of mechanisms operating in each regime. The major impediment to testing such a prediction method is the very small size of the proposed nucleation zone. It is a combination of the shear and normal stress that will determine if a given fault plane slips. In addition to the dilatational strain. (6. The King et al strategy for studying aftershocks is to map the calculated Coulomb stress change based on the observed slip during an earthquake and compare the resulting ﬁeld with the observed aftershock distribution (ﬁgure 28). although they are difﬁcult to detect. triggering provides a fundamental clue into initiation.4). the most commonly observed form of earthquake interactions. Aftershocks are.1. Hubbert and Rubey failure criterion laid out in section 2. Another strategy for studying earthquake initiation is to examine cases where the immediate trigger of a real earthquake is known. then frictional fault slip is promoted (see equation (2.6)). respectively. seismologists have discovered that earthquakes commonly trigger other earthquakes both in the near-ﬁeld and at distances approaching 4000 km (Hill et al 2002).2. If τc increases. either the the observation of very small earthquakes with Mw extrapolation of laboratory parameters from metre-scale samples to kilometre-scale faults is problematic. Smaller earthquakes have been observed. In some areas the strain is extensional. or earthquake initiation involves other processes than simple frictional instability as formulated here. it is useful in order to separate plausible regimes for certain physics.7). Observed cases of triggering are usually separated into near-ﬁeld (<2–3 fault lengths) and far-ﬁeld. Dense networks designed to capture extremely small earthquakes could help conﬁrm or refute the nucleation length model by determining whether any lower bounds on earthquake size exists. there is also a deviatoric stress component to the stress ﬁeld. however. p is the pore pressure change and µ is the coefﬁcient of friction. then.

This method has had some success in predicting the location of aftershocks and even a few large.8) by itself does not fully describe the aftershock ﬁeld shown in ﬁgure 28. will be inﬂuenced by a stress step.1480 H Kanamori and E E Brodsky Figure 28.2. the dynamic stresses follow a very similar pattern because the strong and weak areas of shaking are also determined by the mainshock fault geometry. A recent review can be found in Harris (2002). Therefore. Equation (6. Red (positive) indicates that optimally oriented faults are stressed more towards failure and purple (negative) indicates that failure is inhibited. but the aftershock rate relative to the background rate will vary systematically with the imposed Coulomb stress. nearby earthquakes (Stein 1999). This problem of a continual low aftershock rate in the destressed regions was addressed by adding rate.2) to the stress transfer model (Stein et al 2003). some other mechanisms could possibly produce similar predictions. Shown on the right is a schematic of a strike-slip fault with slip in the directions shown by the arrows. White circles are observed aftershocks (from King et al (1994)). In particular.1. One major difference between the static and dynamic stress is that only the dynamic ﬁeld is affected by rupture directivity (section 3. .3 Landers earthquake occurred where the Coulomb stress ﬁeld increased at the time of the mainshock (Hardebeck et al 1998). Change in Coulomb stress from the 1979 Homestead Valley mainshock and the subsequent aftershocks. Dieterich (1994) showed that if velocity and memory-dependence are incorporated into the standard frictional coefﬁcient based on laboratory experiments. rather than the absolute number of events. Since the pattern of static Coulomb stress increase is controlled by the mainshock fault geometry. we might expect some aftershocks to occur in all areas of the stress ﬁeld if the background seismicity rate is fairly high.5). There are some aftershocks in the areas where failure should have been inhibited by the mainshock. The dilatational strain is compressional where there are ‘+’ signs and extensional where there are ‘−’ signs. then the rate of seismicity. Approximately 85% of the aftershocks of the 1992 Mw = 7.and state-dependent friction (section 4.

However. earthquake (Mw = 7. Areas are simply distinguished by stronger or weaker shaking. Prejean et al 2004). Furthermore. . Figure 29 shows a recent example. In all cases. Montana. the triggered sites are geothermal areas. Increases in seismicity have been often observed within the surface wave trains indicating that the seismic waves are the trigger. The magnitude 7. Past this distance earthquakes were thought to have no effect until a very surprising observation in 1992. The triggered seismicity often persists for several days indicating that the seismic waves have a sustained effect on the stress ﬁeld. The oscillatory dynamic ﬁeld has no such negative effect. or ‘stress shadows’. Alaska. Gomberg et al 2001. remote triggered seismicity has been robustly documented for several large events and has now been seen up to 4000 km from the mainshock (Brodsky et al 2000. sometimes observed around faults (Stein et al 2003). The accompanying deformation (Johnston et al 1995) also suggests a sustained stress in the triggered regions. Each green arrow is a local earthquake. the dynamic shaking cannot explain the rate decreases. Therefore. Since that time. for days after the mainshock (Hill et al 1993). Triggering in Montana from the 2002 Denali. Aftershocks can extend up to about 1–2 fault lengths from the original event. Kilb et al (2000) and Gomberg et al (2003) demonstrated that for earthquakes with strong directivity. the current balance of evidence favours static stress as a primary mechanism for generating aftershocks.3 Landers earthquake in Southern California was followed by up to 10-fold increased seismicity in geothermal and volcanic areas up to 1500 km away.The physics of earthquakes 1481 Figure 29. The bottom panel is a magniﬁed view of the record of one of these local earthquakes. Eberhardt-Phillips 2003. the exact mechanism for dynamic triggering is unclear. Static stress ﬁelds explain stress shadows by invoking negative Coulomb stress changes that move potential faults further from failure.9). the aftershocks are better predicted by the dynamic than the static stress ﬁelds. Top panel is the seismic wave generated by the earthquake in Alaska as recorded by a seismometer in Bozeman. The middle panel is the same record ﬁltered at high-frequencies to show local earthquakes occurring in the vicinity of Montana during the passage of the seismic waves from the Alaskan earthquake. but the debate is far from over.

1482

H Kanamori and E E Brodsky

0

Figure 30. Slider-block sliding on a surface with rate- and state-dependent friction.

Artiﬁcially induced seismicity also gives a clue to the initiation process. In addition to the well-controlled Rangely experiment discussed in section 2, humans have produced earthquakes under less advantageous circumstances due to mining, reservoir ﬁlling and oil exploitation. Many of these cases are consistent with pore pressure changes relieving the normal stress as was observed for Rangely (Guha 2000). In mines, the excavation also directly removes the load on faults producing the same effect (a recent review is found in McGarr et al (2002)). To summarize, the observations of triggered and induced earthquakes imply that: (1) static stress changes may be effective in the nearﬁeld triggering of aftershocks, (2) seismic waves can trigger earthquakes at long-distances in geothermal areas and (3) pore pressure changes can trigger seismicity. We now explore in detail some theoretical mechanisms for triggering earthquakes that satisfy at least parts of these constraints. At present, no uniﬁed earthquake model exists that satisﬁes all of them. 6.2.2. Triggering with the rate- and state-dependent friction mechanism. If friction on a sliding surface is controlled by the rate- and state-dependent friction law discussed in section 4.2.2, then a sudden change in loading causes a sudden increase in the sliding speed which in turn results in accelerated seismic slip. This mechanism can be important in seismic triggering, as shown by Dieterich (1994). Because this model is widely used in seismology, we discuss this particular mechanism in greater detail than some others. Consider a slider-block model shown in ﬁgure 30 in which friction is controlled by the rate- and state-dependent friction given by (equation (4.24)) ˙ µ = µ0 + A ln δ + B ln θ. ˙ Consider the case where |θ δ/Dc | from (4.25) ˙ θδ ˙ θ =− , Dc from which we obtain θ = θ0 exp − and (6.9) becomes B ˙ δ. µ = µ0 + A ln δ + B ln θ0 − Dc Then, (6.1) and (6.2) become B ˙ δ = −kδ + kδ0 . σn µ0 + A ln δ + B ln θ0 − Dc (6.13) (6.12) δ Dc (6.11) (6.9) ˙ 1 (e.g. large δ during coseismic slip). In this case,

(6.10)

The physics of earthquakes

1483

Spontaneous behaviour. First, we examine the behaviour of this system under constant ˙ ˙ loading, i.e. kδ0 = τ0 . Then, integrating (6.13) with the initial conditions, δ = 0 and δ = δ0 at t = 0, we obtain ˙ δ0 H t A (6.14) δ = − ln 1 − H A and ˙ δ= where H =− k B + . σn Dc (6.16) 1 Ht − ˙0 A δ

−1

,

(6.15)

For an unstable system, H > 0. Equations (6.15) shows that the sliding velocity spontaneously increases with time, and at the time tf = A H 1 ˙0 δ (6.17)

˙ δ becomes inﬁnitely large; that is, an instability occurs, i.e. an earthquake occurs. The time tf is called the time-to-failure. Loading at a uniform rate. Next, we add loading given by a linear function of time, i.e. (6.18)

**kδ0 = τ (t) = τ0 + τ t, ˙ where τ is a constant loading rate. Then, from (6.13) ˙ τ (t) − kδ B ˙ = µ0 + A ln δ + B ln θ0 − δ . σn Dc We can integrate (6.19) to obtain, δ=− and ˙ δ= 1 H σn τt ˙ + exp − ˙0 τ ˙ Aσn δ − H σn τ ˙
**

−1

(6.19)

˙ A δ0 H σn τt ˙ ln 1 − exp H τ ˙ Aσn

+1

(6.20)

.

(6.21)

˙ The functional forms of δ and δ given by (6.20) and (6.21), respectively, are a little complicated, but both exhibit a monotonic behaviour in time that increases rapidly. From equation (6.21), the time-to-failure, tf , is given by tf = τ ˙ Aσn ln +1 . ˙ τ ˙ H σn δ0 (6.22)

Stepwise change in loading. Next, let us consider the case where a sudden change in loading occurs from τ0 to τ1 by τ at t = t0 (i.e. τ1 = τ0 + τ ). This corresponds to the case when a sudden change in the crustal stress occurs due to a large earthquake. Solving equation (6.13) with kδ0 = τ0 for t < t0 , and with kδ0 = τ0 + τ for t > t0 , and requiring the continuity of slip at t = t0 , i.e. δ(t = t0 − ε) = δ(t = t0 + ε),

1484

. . δ / δ0

H Kanamori and E E Brodsky

δ/(A/H)

5

1000

4

100

3

2

10

1

0

0

0.1

0.2

0.4 . t /( A / H δ 0 )

0.3

0.5

0.6

0.7

1

0

0.1

0.2

0.3 0.4 . t /( A / H δ 0 )

0.5

0.6

0.7

˙ ˙ Figure 31. Non-dimensional slip δ/(A/H ) (left) and non-dimensional slip speed δ/δ0 as a function ˙ of non-dimensional time, t/(A/H δ0 ). A non-dimensional stepwise stress change τ/Aσn = 1.25 ˙ is given at time t/(A/H δ0 ) = 0.5.

we can derive ˙ ˙ δ(t = t0 + ε) = δ(t = t0 − ε) exp τ Aσn . (6.23)

˙ ˙ Figure 31 shows the non-dimensional slip, δ/(A/H ), and slip speed, δ/δ0 , as a function ˙0 ). of non-dimensional time t/(A/H δ Thus, a step-wise change in loading by τ causes a step-wise increase in sliding velocity, which in turn causes a step-wise decrease in the time-to-failure. This behaviour, a sudden decrease in the time-to-failure due to a sudden loading, can be used to explain the triggering of seismic activity, and aftershock behaviour (see section 6.2.4). 6.2.3. Triggering with the stress corrosion mechanism. Stress corrosion or sub-critical crack growth is a process widely known in material science (Anderson and Grew 1977, Das and Scholz 1981, Atkinson 1984, Main 1999, Gomberg 2001). Cracks in a purely brittle material remain stable under the loading stress below the critical stress determined by the Grifﬁth criterion. However, under certain environments, especially under high temperatures and with ﬂuids, a crack can grow spontaneously because of weakening near the crack tip due to chemical ‘corrosion’, even if the loading stress is below the critical level. In this case, a crack is growing constantly, and eventually it will reach a critical state where it fails catastrophically. This mechanism may be important for static triggering of earthquakes in Earth’s crust. Large amounts of experimental data show that the growth rate of a crack with length x is generally given by (Atkinson 1984), dx = V0 dt K K0

p

,

(6.24)

where t is time, K is the stress intensity factor and p is a constant, usually 5 or larger. Although most of the experimental data are obtained for tensile cracks, here we apply this

K is given by equation (4. From equation (6. This can be rewritten as t m . (6.2tf . and contribute to increase in the seismicity rate.24) and (6.g.and state-dependent friction law (cf ﬁgures 31 and 32. The stress corrosion behaviour is very similar to that found for the rate.33).27) x = x0 1 + mτ where τ = (x0 /V0 ) and m = 2/(2 − p) < 0.34) x = x1 1 − tf and t − t1 m−1 ˙ .33) V1 ˙ and x and x after t1 are given by t − t1 m (6.27) and (6.3) K = Y x 1/2 σ.26) x= [1 − ((p − 2)/2)(V0 /x0 )t](2/(p−2)) where x0 is the crack length at t = 0 (Main 1999). the time-to-failure measured from t1 .27) and (6. (6. (6. (6.29) tf = −mτ = −m . is given by ˙ x = V0 1 + V1+ = 1 + τ p − V1 . (6. the growth rate suddenly increases with the step-wise increase of the load.32) and (6.35) x = V1+ 1 − tf As given by equations (6. the speed increases suddenly and the time-to-failure ˙ is shortened. when K = K0 .31) x1 = V0 1 + mτ and at this time. V1− . V0 Now we consider the case in which the loading stress increases by τ at time t1 (t1 < tf ). At t = t1 . For a constant loading stress σ .) A sudden increase in loading (e. .25) where σ is the loading stress. In the following. a static stress change due to an earthquake) can accelerate the crack growth. is mx1 tf = − + (6. Then. (6.32) σ Thus. the relation between the speed just after t1 . t m−1 . Figure 32 shows a typical behaviour of x and x when the loading is increased stepwise by τ at t = 0. V0 is the speed of crack growth at t = 0. the size and the growth rate of the crack are given by (6.28) as t1 m (6.28) mτ From equations (6.25). we assume that p > 2. and just before t1 . shorten the time to failure. (6. the time-to-failure is given by x0 (6. V1+ . and Y is a constant determined by the geometry of the crack.The physics of earthquakes 1485 model to seismic shear cracks.24) can be integrated as x0 .30) x1 = x0 1 + mτ and t1 m−1 ˙ ≡ V1 (6.28).

It is already more than 100 years (36 500 days) since the mainshock. The second example is for the 1995 Kobe. Why the aftershock decay follows a power law given by (6.36) n(t) = t +c where n(t) is the number of aftershocks larger than a given magnitude per unit time. and the stress corrosion model.6 100 1. Utsu (2002). 6. State. A modiﬁed (or generalized) Omori’s Law is given by K n(t) = .37) (t + c)p where p is a constant. earthquake. Japan. many smaller earthquakes called aftershocks occur near the rupture zone of the earthquake.g. post seismic creep (e. Dieterich (1994) assumed that seismicity rate is constant under the constant tectonic loading rate τ . Non-dimensional crack length x/x0 (left) and non-dimensional crack extension speed ˙ x/V0 (right) as a function of non-dimensional time t/tf . ﬂuid diffusion (Nur and Booker 1972).1486 H Kanamori and E E Brodsky .and rate-dependent friction and Omori’s Law. In the following. Figure 33 shows two examples. The ﬁrst one is for the 1891 M ≈ 8 Nobi.2 1 0. the decay of aftershock activity follows a power law. x / V0 x/x0 2 104 1.and state-dependent friction (Dieterich 1994). etc. stress corrosion (Yamashita and Knopoff 1987. A non-dimensional stepwise stress change τ/σ = 0. we summarize the two recently developed models.15 0. earthquake (Utsu 2002).37) has attracted much attention of many seismologists. Various mechanisms are reviewed in Utsu (1999). As ﬁrst discovered by Omori (1894).2. for which Omori found this relationship (Utsu et al 1995). which is usually approximately equal to 1.4.4 10 1. (6.25 1 0.2 0. Kisslinger (1996).2. c = 0. usually referred to as Omori’s Law.8 1000 1. Many different mechanisms have been proposed.15 is imposed at time t/tf = 0. e. This can be accomplished ˙ . Benioff (1951)).) K . Japan. rate.15 0.37) holds over a very long period of time (p = 1 (constrained).797 day). and we can see that the relation (6. (6. After a large earthquake (main shock). Aftershocks and Omori’s Law. Gomberg 2001). see Utsu et al (1995).25 t / tf t / tf Figure 32. (For recent reviews.and state-dependent friction model.g. the rate.2 0.

42) where r0 = 1/ t is the background rate and tf (n) is now written as a continuous variable. dtf (n) taking n and tf (n) as continuous variables.38) If the loading is increased by τ due to a mainshock.1 0.e.The physics of earthquakes n(t) 1000 n(t) 1487 10 0. t. n and tf (n) are discrete variables. days 10000 Time. and solving for n. The decay of aftershock activity following the 1891 Nobi. Substituting δn0 given by (6. if earthquake nuclei are distributed such that the time-to-failure. tf0 (n). of the nth event is given by n t. Japan. τ ˙ Aσn ln +1 .39) ˙ τ ˙ H σn δn0 F ˙ where F ≡ exp( τ /Aσn ). = r0 1 − [(F − 1)/F ] exp(−τ t/Aσn ) ˙ R= (6.22).2. Thus. Japan earthquake (Utsu 2002).38) in (6. (6. of the nth nucleus with tf0 (n) can be given by solving equation (6.2.39). days Figure 33. but we can deﬁne the instantaneous seismicity rate R by dn . (6. tf (n). Thus.001 1 100 Time.41) (6. as we discussed in section 6. The sliding speed. This relation is shown in ﬁgure 34.22) as. for nucleus n can be given by substituting the increased sliding velocity into (6. . The new time-to-failure. 1 R . and the 1995 Kobe. δn0 = H σ exp(τ n t/Aσn ) − 1 ˙ (6. then. number of events ˙ per unit time). the sliding speed increases step-wise by exp( τ /Aσn ). and seismicity rate changes. 1 τ ˙ ˙ .40) n= τ t ˙ Aσn tf (n) = Here. the constant seismicity rate is given by r0 = (1/ t) (i. δn0 . In this case. the time-to-failure changes. earthquake. we obtain τ tf (n) ˙ Aσn ln 1 + F exp −1 .

˙ predicted by the rate.2.1488 H Kanamori and E E Brodsky 10 10 10 10 R/r 0 6 5 4 3 10 10 10 10 2 1 0 -1 10 -8 10 -6 10 -4 .24) and (6.3. A similar 1/t trend can be predicted with the stress corrosion model discussed in section 6. leads to x0 x= (6. t /(Aσn /τ) 10 -2 10 0 10 2 10 4 Figure 34. We assume a constant rate loading in the stress corrosion model.43) {1 − (σ0 V0 /2x0 τ )((p − 2)/(1 + p))[(1 + (τ /σ0 )t)p+1 − 1]}2/(p−2) ˙ ˙ . More progress has been made with the large t limit although it is difﬁcult to measure the duration of a sequence as the detectable measurement is dependent on the choice of spatial windows (Gross and Kisslinger 1997). This is the form of the Omori’s Law. R a1 ≈ . The non-dimensional stress change is assumed to be τ/(Aσn ) = 12. 1. The observational difﬁculties lie in detection thresholds. for (Aσn /τ )/F < t < (Aσn /τ ). Change in seismicity rate plotted as a function of non-dimensional time t/(Aσn /τ ). For times comparable or larger than the timescale of the background stressing ˙ Aσn /τ .and state-dependent friction. ˙ σ = σ0 + τ t. So far. i. Stress corrosion model and Omori’s Law.and state-dependent friction. these trends have not been established observationally.25) which. after integration. Then substituting this in equations (6. The results presented here are similar in parts to Gomberg (2001) except that here the equations are formulated to be parallel to the above derivation of Omori’s Law from rate.60. many small earthquakes are missed in a catalogue because the larger aftershocks mask their waveforms on seismograms. which represents the sudden increase in seismicity rate. For t → 0. (R/r0 ) = F ≡ exp( τ /Aσn ). Immediately after a large earthquake when t is small. r0 t + a2 ˙ ˙ where a1 = (F /(F − 1))Aσn /τ and a2 = (1/(F − 1))Aσn /τ are constants. Between these two ˙ ˙ extremes. A test of this model would be whether the observed aftershock decay follows the trends at very small and large t predicted by this model (ﬁgure 34).e. the normalized rate (R/r0 ) approaches the steady state value.

tf = σ0 τ ˙ 2x0 τ (1 + p) ˙ +1 (p − 2)σ0 V0 1(p+1) (6. ˙ R/r0 = F 1/(p+1) . tf (n) = = σ0 τ ˙ σ0 τ ˙ 2x0 τ (1 + p) ˙ +1 (p − 2)σ0 F Vn0 1+ 1/(p+1) −1 1/(p+1) ((τ /σ0 )n t + 1)p+1 − 1 ˙ F −1 . a constant seismicity rate r0 = (1/ t) can be produced by distributing earthquake nuclei such that the time-to-failure. the initial distribution of earthquake nuclei is designed to yield a constant rate under constant stressing rates.48) Here. n and tf (n) are discrete variables.45) as. and the seismicity rate changes. (6. for a nucleus n can be given by substituting the increased sliding velocity. ˙ 1 2(1 + p) x0 τ .44).49) . (6.23). of the nth nucleus with tf0 (n) can be given by solving equation (6. Vn0 . For t = 0.The physics of earthquakes 1489 and ˙ x= V0 (1 + (τ /σ0 )t)p ˙ . but we can deﬁne the instantaneous seismicity rate R by dn dtf (n) taking n and tf (n) as continuous variables. tf0 (n). (6. R= R τ ˙ = F F + (1 − F ) 1 + t r0 σ0 −(p+1) −p/(p+1) (6. {1 − (σ0 V0 /2x0 τ )((p − 2)/(1 + p))[(1 + (τ /σ0 )t)p+1 − 1]}2/(p−2)+1 ˙ ˙ From (6. then. (6. The similarity to the rate and state model solution in ﬁgure 34 suggests that the form of the decay in seismicity with time is controlled by the similar mathematical construction with both cases.50) where tf (n) is now denoted simply by t. . The sliding speed. into (6. we obtain σ0 F n= τ t ˙ τ ˙ tf (n) + 1 σ0 p+1 1/(p+1) −1 +1 −1 .3 with equation (6.47) Solving this for n.45).46) Vn0 = p − 2 σ0 ((τ /σ0 )n t + 1)p+1 − 1 ˙ If the loading is increased by τ at t = 0 due to a mainshock. the time-to-failure changes. Thus. the sliding speed increases step-wise by a factor of F ≡ (1 + ( τ /σ0 ))p . the time-to-failure tf can be determined as. The new time-to-failure. An example of a solution with typical parameters is shown in ﬁgure 35. In both cases. (6. R/r0 = F and for t σ0 /τ . for t σ0 /τ . of nth event is given by n t.2. as we discussed in section 6. ˙ Because p is large.44) −1 . F Vn0 . R/r0 has the form a1 /((a2 + t)p/(1+p) ) ≈ ˙ a1 /(a2 + t) which is the Omori’s law.45) As discussed above. tf (n). where a1 = (F /(F − 1))(σ0 /(p + 1)τ ) and a2 = ˙ (1/(F − 1))(σ0 /(p + 1)τ ). Thus.

the rate returns exactly to the background rate only for inﬁnite p while. Seismic waves can induce water ﬂow into faults as the differential stiffness of geological units generates a hydraulic gradient when the seismic waves impose a long-wavelength. As discussed in section 2 and illustrated by the observations of artiﬁcially induced seismicity. in the former.6)).and state-dependent friction model is that. and τ /σ = 0. is of the order of p= Ud hη . the pressure changes can be . the removal of the barrier redistributes this pressure difference p in the fault zone. When the earthquake occurs. In places where the pressure rises. One example of a ﬂuid-based triggering mechanism comes from recent work on the removal of transient hydrologic barriers during ground shaking (Brodsky et al 2003). oscillating strain ﬁeld. One difference between the stress corrosion and the rate. it always returns to the background rate.1490 H Kanamori and E E Brodsky Figure 35. according to the standard formulation of ﬂow in porous media (Darcy’s Law). If aftershocks are generated by sub-critical crack growth. The parameters used are p = 40. Even very small ﬂuid shear stresses ∼1 Pa are sufﬁcient to remove accumulations of sediment or precipitate (Kessler 1993).5. the frictional stress is reduced (see equation (2. h is the thickness of the barrier. For realistic parameters. ﬂuid movement can reduce the strength on faults and initiate earthquakes. ˙ predicted by the stress corrosion model. and failure can occur. η is the viscosity of the water and k is the permeability of the rock (e. Recent research has begun exploring quantitative models and new observational techniques in order to constrain ﬂuid movements and their importance in natural faults. in the latter. 6. Hydrologic barrier removal.g.2. The sediment or precipitate barriers blocked ﬂow prior to the earthquake while maintaining a sharp pressure differential p which.5. Change in seismicity rate plotted as a function of non-dimensional time t/(σ0 /τ ). then we must rely on long-term relaxation processes such as viscous ﬂow to prevent a continual ratcheting upwards of seismicity. k where Ud is the average ﬂuid ﬂow velocity (Darcy velocity). Freeze and Cherry (1979)). The above mechanisms emphasize the solid mechanics of earthquake initiation.

detailed slip inversions. . highquality ground motion monitoring brings the goal of saving lives well within our grasp. laboratory friction models. In the short term. There are many interesting and active areas of research that we have omitted because of space (for recent reviews. Energy release and casualties (number of death) per year during the 20th century. We look forward to the development of this promising line of research. Subjects that we neglected include questions of how earthquakes stop.04 MPa (Brodsky et al 2003). that elude the solid models. if not centuries away. we continue to build a basic scientiﬁc framework to learn why and how earthquakes begin. Neither stress corrosion nor rate. observational and experimental work is necessary to develop the formalism to compare both the ﬂuid and solid avenues for triggering. They are more closely related to the engineering infrastructures and preparedness of a region. The peak in the casualty in the mid 1970s is due to the 1976 Tanshang.and state-dependent friction can explain sustained triggering from seismic waves (Gomberg 2001). China. Perhaps one day we will be able to accomplish accurate earthquake prediction.and passive-controlled buildings and dense.and state-dependent friction and stress corrosion. such as sustained distant triggering. The above discussion should have made it clear that the quantitative prediction of earthquake initiation is an extremely complicated and perhaps impossible task. Conclusions We have presented in this paper an overview of earthquake physics with an emphasis on initiation. it is more practical to save lives by using the detailed knowledge we have about the propagation of seismic waves and strength of seismic shaking to design buildings and infrastructure that will protect people during an earthquake. Much more theoretical. the nucleation length of 1 m requires instruments to be too densely spaced to be practical. but the current state of the science implies that that day is decades. National Research Council (2003)). 7. It is obvious that the ﬂuid-based models are not nearly as well developed quantitatively as the solid models like rate.The physics of earthquakes 1491 Figure 36. Recent engineering advances such as active. However. 2003). Over the last 10–20 years we have unravelled parts of the puzzle based on the state of stress in the crust. The latter area is particularly important for society. 0. Fiveyear running average is taken. even at this stage the ﬂuid approach is able to address some problems. which is sufﬁcient to trigger earthquakes based on static stress studies of triggering thresholds in the nearﬁeld (Hardebeck et al 1998). At the same time. earthquake (updated from Kanamori (1978)). see Lee et al (2002. Even in the best case scenario of a predictable fault nucleation length. The peak in the energy release in 1960 is mainly due to a sequence of great earthquakes from 1952 to 1964 including the 1960 Chilean earthquake. Figure 36 shows that casualties from earthquakes are not correlated to magnitude or overall level of seismicity. geological studies of earthquake occurrence histories and methods of mitigating earthquake damage.

Res. we predict that input from geology and hydrogeology will play a greater role in the decades to come. Theorists and experimentalists also have their work cut out for them. Soc. Combining disciplines might allow us to measure migrating ﬂuids and if their pressures can equal or exceed the minimum principal stress as suggested by normal faulting studies. Res. 89 4077–114 . 8 393 Anderson E M 1951 The Dynamics of Faulting and Dyke Formation with Applications to Britain (Edinburgh: Oliver and Boyd) p 206 Anderson O L and Grew P C 1977 Rev. Edin. Res. Geophys. the existence of very small earthquakes and the occurrence of long-range triggering are a few areas with some resolving power. but insights from geodetic methods are becoming increasingly important. Averages like that taken in estimating the stress drop need to put on ﬁrmer theoretical ground. Even chaotic systems. Inst. 44 73–88 Aki K and Richards P G 2002 Quantitative Seismology (Sausalito: University Science Books) p 685 Anderson E M 1905 Trans. Geophys. Tokyo Univ. Geophys. As our observational database improves. 100 24003–14 Abercrombie R and Leary P 1993 Geophys. the next few decades promise to bring more earthquake insights and perhaps some answers. Is the same physics applicable or does a new set of processes come into play once rapid slip has begun? What is the physical nature of the fracture energy term that controls dynamics ranging from slow quasi-static slip to brittle failure with high rupture speed and efﬁcient energy radiation? What happens when the mechanisms such as thermal pressurization and lubrication are combined in a single rupture model? The complexity community must try to ascertain exactly how chaotic are earthquakes. Many of the questions highlighted during the course of this review will only be addressed by improved instrumentation and observational techniques. The beginning of aftershock sequences. Geol. Mechanisms for triggering can be differentiated by studies targeting phenomena where the predictions diverge.1492 H Kanamori and E E Brodsky complex system modelling and triggering studies. Another recent advance is the use of non-seismic data to study earthquakes. Lett. Theorists and experimentalists will have to explore the relationship between initiation conditions and rupture propagation. A number of still unanswered questions remain. We must also grapple with the thickness and physical properties of the fault zone. can be predicted over short time horizons if the observations have sufﬁcient resolution. Seismic and other observations will also have to address the heterogeneity of stress and strength in the crust. our computational ability accelerates and our laboratories become more reﬁned. Quantiﬁcation of the required resolution and the divergence rate from initial conditions would be a valuable contribution. 20 1511–4 Aki K 1966 Bull. like weather. Geophys. Technical developments like the recently deployed dense seismic networks of Japan and the Earthscope instrumentation initiative in the United States may help us measure seismic parameters like stress drop and radiated energy to sufﬁcient accuracy to ﬁnally address the dynamic similarity of earthquakes. 107 (B2) 2363 Atkinson B K 1984 J. Acknowledgments We thank Allan Rubin and an anonymous reviewer for constructive comments. References Abercrombie R 1995 J. Res. Earthquake Res. As the subject expands beyond classic elasticity. 15 77–104 Andrews D J 2002 J. This review emphasized elastic waves as the most developed method for characterizing earthquakes.

Seismol. 100 18205–28 Dahlen F A 1977 Geophys. Soc. 75 4997–5009 Burridge R and Knopoff L 1964 Bull. Res. CA: Academic) pp 1217–32 Haskell N 1964 Bull. Soc. Astron. Am. Rev. 84 2161–8 Dmowska R and Rice J R 1986 Fracture theory and its seismological applications Theories in Solid Earth Physics ed R Teisseyre (Warzawa: PWN-Polish Publishers) pp 187–255 Dragert H et al 2001 Science 292 1525–8 Dugdale D S 1960 J. Soc. 99 2601–18 Dieterich J H 1979 J. 116 615–26 Byerlee J 1992 Tectonophysics 211 295–303 Cardwell R K et al 1978 Geophys. Solids 8 100–4 Eberhart-Phillips D et al 2003 Science 300 1113–18 Ekstr¨ m E 2000 Global studies of earthquakes Problems in Geophysics for the New Millennium ed G E E Boschi and o A Morelli (Bologna: Editorice Compositori) pp 111–24 Eshelby J D 1969 J. Am. Solids 17 177–99 Fossum A F and Freund L B 1975 J. 64 1–20 . R. Seismol. Geophys. 80 3343–7 Freeze R A and Cherry J A 1979 Groundwater (Englewood Cliffs. NJ: Prentice Hall) Freund L B 1972 J. Soc. Geophys. Res. 102 11731–44 Bouchon M and Vallee M 2003 Science 301 824–6 Brace W F and Byerlee J D 1966 Science 153 990–2 Brodsky E E and Kanamori H 2001 J. Soc. Lett. Geophys. Geophys. Soc. Astron. 57 341–71 Byerlee J 1978 Pure Appl. Res. Solid Earth 102 7603–12 Guha S K 2000 Induced Earthquakes (London: Kluwer) p 314 Gutenberg B and Richter C F 1941 Geol. Soc. Soc. Res. Geophys. 34 1–131 (special paper) Hardebeck J L and Hauksson E 2001 J. Phys. Lond. Am. 48 239–61 Dahlen F A and Tromp J 1998 Theoretical Global Seismology (Princeton. Geophys. and seismic hazard International Handbook of Earthquake & Engineering Seismology part B. 86 6039–51 Dieterich J 1994 J. NJ: Princeton University Press) p 1025 Das S 1988 Bull. Res. Res. Res. Res. Mech. 78 924–30 Das S and Scholz C H 1981 J. Earth Planet. Geophys. 101 22449–60 Bird P and Kagan Y Y 2004 Bull. 94 15635–7 Bak P et al 1988 Phys. Res. 106 16357–74 Brodsky E E et al 2000 Geophys. Am. Mech. Res. Geophys. Soc. stress shadows. R. Soc. Inter. Lond. Seismol. Seismol. Geophys. Trans. 27 2741–4 Brodsky E E et al 2003 J. A 221 169–98 Gross S and Kisslinger C 1997 J. J. Soc.The physics of earthquakes 1493 Axen G J 2004 Low-angle normal fault mechanics and crustal strength Rheology and Deformation in the Lithosphere at Continental Margins ed G Karner et al (New York: Columbia University Press) pp 46–91 Bak P and Tang C 1989 J. Seismol. Res. Am. Am. Res. Seismol. 21 1987–90 Benioff H 1951 Bull. R. Seismol. Geophys. A 38 364–74 Barenblatt G I 1962 The mathematical theory of equilibrium cracks in brittle fracture Advances in Applied Mechanics vol 7 (London: Academic) pp 55–125 Beeler N M et al 1994 Geophys. Res. Geophys. Geophys. 91 2095–112 Boatwright J et al 2002 Bull. 41 31–62 Beroza G C and Mikumo T 1996 J. Am. 278 187–269 Gomberg J 2001 J. Seismol. 54 1875–88 Burridge R and Knopoff L 1967 Bull. Res. 52 525–30 Chester F M and Chester J S 1998 Tectonophysics 295 199–221 Choy G L and Boatwright J L 1995 J. Geophys. 103 24427–37 Harris R A 2002 Stress triggers. Phys. J. ed H Kanamori et al (San Diego. Soc. Elasticity 2 341–9 Freund L B 1989 Dynamic Fracture Mechanics (Cambridge: Cambridge University Press) p 563 Gilbert F and Dziewonski A M 1975 Phil. 108 Brune J N 1970 J. Geophys. Am. Lett. Geophys. Soc. Am. Res. Res. Trans. Geophys. 56 1811–42 Heaton T 1990 Phys. Res. Geophys. 106 21859–82 Hardebeck J L et al 1998 J. Res. at press Boatwright J and Choy G L 1986 J. R. 92 1241–55 Bouchon M 1997 J. 106 16253–63 Gomberg J et al 2001 Nature 411 462–6 Gomberg J et al 2003 Bull. 93 118–38 Grifﬁth A A 1920 Phil.

Am. Mag. Seismol. 108 (B4) 2205 Lawn B 1993 Fracture of Brittle Solids 2nd edn (Cambridge: Cambridge University Press) p 378 Lay T and Wallace T C 1995 Modern Global Seismology (San Diego. Res. Seismol. Arc. 21 2341–4 Kikuchi M and Kanamori H 1995 Pure Appl. Geophys. Am. 1 44–52 Kostrov B V 1966 J. 28 4007–10 Jaeger J C and Cook N G W 1979 Fundamentals of Rock Mechanics (London: Chapman and Hall) p 593 Jeffreys H 1942 Geol. 92 611–24 Kisslinger C 1996 Aftershocks and fault-zone properties Advances in Geophysics vol 38. Seismol. Am. 10 252–60 Izutani Y and Kanamori H 2001 Geophys. Am. Geophys. Soc. 28 3349–52 Ide S and Takeo M 1997 J. Seismol. J. tans. Res. Soc. Geophys. 144 441–53 Kilb D et al 2000 Nature 408 570–4 King G C P et al 1994 Bull. 85 6185–222 Lapusta N and Rice J R 2003 J. Solid Earth 97 4923–40 Lockner D A 1995 Rock failure in rock physics & phase relations A Handbook of Physical Constants vol 3. Lett. Earth Phys. 1 23–40 Lachenbruch A H 1980 J. CA: Academic) pp 505–37 Ma K F et al 2003 Geophys. 102 27379–91 Ikeda R et al 2001 Isl. Soc. 65 1073–95 Kanamori H et al 1998 Science 279 839–42 Kanamori H and Brodsky E E 2001 Phys. 79 291–5 Johnston M J S et al 1995 Bull. 83 330–46 Kessler J H 1993 Berkeley University of California. 88 815–32 Kadanoff L P 1991 Phys. Res. CA: Academic) p 521 Lee W H K et al 2002 International Handbook of Earthquake & Engineering Seismology part A (San Diego. Lett. Res. CA: Academic) pp 937–1945 Li V C 1987 Mechanics of shear rupture applied to earthquake zones Fracture Mechanics of Rock (London: Acdemic) pp 351–428 Linker M and Dieterich J H 1992 J. 78 1707–24 Kikuchi M and Kanamori H 1994 Geophys. Res. ed R Dmowska and B Saltzman (San Diego. Res. Today 44 9–10 Kanamori H 1978 Nature 271 411–14 Kanamori H 1983 Global seismicity Earthquakes: Observation. Lett. 30 (5) 1244 Ma K-F et al 1999 EOS Trans. Mech. Res. Soc. ed W H K Lee et al (San Diego. Res. Seismol. CA: Academic) pp 1–36 Knopoff L 1958 Geophys. Geophys. J. Soc. Geophys. ed W H K Lee et al (San Diego. Geophys. 49 699–714 Ida Y 1972 J. 106 11137–50 Hubbert M K and Rubey W W 1959 Geol. Lett. 84 935–53 Kinoshita S and Ohike M 2002 Bull. Today 55 41–7 Hill D P et al 1993 Science 260 1617–23 Houston H 2001 J. Am. Am. Geophys. Geophys.) Kostrov B V 1974 Izv. Res. Soc. Appl. Soc. DC: American Geophysical Union) pp 127–47 Lockner D A and Beeler N M 2002 Rock failure and earthquakes International Handbook of Earthquake & Engineering Seismology part A. R. CA: Academic) pp 1–933 Lee W H K et al 2003 International Handbook of Earthquake & Engineering Seismology part B (San Diego. Seismol. Res. CA: Academic) pp 175–94 . Am. 77 3796–805 Ide S and Beroza G C 2001 Geophys. 70 115–66 (Special paper) Husseini M I 1977 Geophys. Today 54 34–40 Kanamori H et al 1993 Bull.1494 H Kanamori and E E Brodsky Hergaten S 2002 Self-Organized Criticality in Earth Systems (Berlin: Springer) p 272 Hill D P et al 2002 Phys. Am. Math. 30 1241–8 (PMM) (Engl. Astron. Seismol. 85 6097–112 Lachenbruch A H and Sass J H 1980 J. 85 787–95 Jost M L et al 1998 Bull. Berkeley Kikuchi M 1992 Tectonophysics 211 107–13 Kikuchi M and Fukao Y 1988 Bull. Geophys. 115 301–16 Madariaga R 1979 J. Soc. Res. Geophys. Res. Theory and Interpretation ed H Kanamori and E Boschi (New York: North-Holland) pp 596–608 Kanamori H and Anderson D L 1975 Bull. Soc.—AGU 80 605 Madariaga R 1977 Pure Appl. 84 2243–50 Madariaga R and Olsen K B 2002 Earthquake dynamics International Handbook of Earthquake & Engineering Seismology part A. ed T J Ahrens (Washington.

Soc. Soc. Res. Soc. Int. ed W H K Lee et al (San Diego. Res. 86 1785–93 Ruina A L 1983 J. 51 1189–257 Rubin A M et al 1999 Nature 400 635–41 Rudnicki J W and Kanamori H 1981 J. Eng. Res. CA: Academic) pp 647–61 McKenzie D P and Brune J N 1972 Geophys. Soc. Phys. Am. Soc. J. Am. Soc. Geophys. Res. 34 pp 433–62 Main I G 1999 Geophys. Geol. Seismol. J. Am. DC: National Academies) p 418 Nur A and Booker J 1972 Science 175 885–7 Ohnaka M and Shen L-F 1999 J. 122 583–607 Mase C W and Smith L 1987 J. 29 65–78 Mikumo T and Miyatake T 1993 Geophys. 94 at press Provost A S and Houston H 2003 J. Geophys. Geophys. 93 264–82 Miyatake T 1992 Geophys. Am. Soc. 66 1–32 Richardson E and Jordan T H 2002 Bull. J. Am. Seismol. 106 11127–36 Prejean S G and Ellsworth W L 2001 Bull. Sci. Seismol. Res. Am. Soc. Seismol. Seismol. and the weakness of the San Andreas fault Fault Mechanics and Transport Properties of Rock ed B Evans and T-F Wong (San Diego. 108 (B5) 2274 Sass J H et al 1997 J. 7 751–4 Sibson R H 2003 Bull. R. 83 1294–7 Rosakis A J 2002 Adv. 36 1168–98 Tibi R et al 2003 J. CA: Academic) pp 475–503 Richards P G 1976 Bull. Soc. Geophys. 84 215–18 Scholz C H 2002 The Mechanics of Earthquakes and Faulting (New York: Cambridge University Press) p 471 Sibson R H 1973 Nature 243 66–8 Sibson R H 1985 J. 104 817–44 Omori F 1894 J. Am. Geophys. Struct. 94 at press Stein R S 1999 Nature 402 605–9 Stein R S et al 2003 Sci. 101 11195–208 McGarr A and Fletcher J B 2002 Bull. 92 1766–82 Romanowicz B and Ruff L J 2002 Geophys. 102 27575–85 Scholz C H 1994 Bull. College of Science Imperial University of Tokyo 7 111–200 Otsuka M 1971 J. Am. pore pressure distributions. Am. Soc. Res. DC: American Geophysical Union) p 284 Saffer D M et al 2003 J. 29 (12) 1604 Romanowicz B and Rundle J B 1993 Bull. J. Res. Astron. Earthquake Res. Soc. Lond. 42 289–368 Mase C W and Smith L 1985 Pure Appl. 92 6249–72 Massonnet D et al 1993 Nature 364 138–42 Mayeda K and Walter W R 1996 J. Am. 108 (B2) 2091 . Res. Geophys. Today 49 24–9 Marone C and Kilgore B 1993 Nature 362 618–21 Maruyama T 1964 Bull. Geophys. 91 165–77 Prejean S G et al 2004 Bull. Geophys. 92 1633–46 McGarr A et al 2002 Case histories of induced and triggered seismicity International Handbook of Earthquake & Engineering Seismology part A. Geophys. Soc. 88 10359–70 Rundle J B et al 2000 GeoComplexity and the Physics of Earth (Washington. Ser. R. Res. 88 1014–22 Simons M et al 2002 Bull. Tokyo Univ. Seismol. Phys. 139 F1–6 Marder M and Fineberg J 1996 Phys. Lett. Soc. Inst. Res.The physics of earthquakes 1495 Main I 1996 Rev. 93 1169–78 Sibson R H and Xie G Y 1998 Bull. Int. Phys. Seismol. Seismol. 112 481–96 Mikumo T et al 2003 Bull. Japan 24 215–27 Palmer A C and Rice J R 1973 Proc. Seismol. Res. Am. Geophys. Soc. Res. 19 1041–4 Mott N F 1948 Engineering 165 16 National Research Council 2003 Living on an Active Earth (Washington. A—Math. Seismol. Seismol. Seismol. Soc. 92 1390–402 Singh S K et al 2004 Bull. Seismol. Lett. Geophys. 332 527–48 Perez-Campos X and Beroza G C 2001 J. Am. 288 72–9 Stekette J A 1958 Can. 108 (B3) 2175 Quin H 1990 Tectonophysics 175 93–117 Rabinowicz E 1995 Friction and Wear of Materials (New York: Wiley) p 315 Raleigh C B et al 1976 Science 191 1230–7 Rice J R 1980 The mechanics of earthquake rupture Physics of the Earth’s Interior ed A M Dziewonski and E Boschi (Amsterdam: North-Holland) pp 555–649 Rice J R 1992 Fault stress states. Am. Geophys.

UK: Cambridge University Press) p 398 Turcotte D L and Malamud B D 2002 Earthquakes as a complex system International Handbook of Earthquake & Engineering Seismology part A. 21 725–8 Zoback M D and Healy J H 1984 Ann. J. 10 93–248 Turcotte D L 1997 Fractals and Chaos in Geology and Geophysics (Cambridge. 91 13–26 Zeng Y H et al 1994 Geophys. Am. Am. Geophys.1496 H Kanamori and E E Brodsky Tsuboi C 1933 Japan. Geophys. 2 689–98 Zoback M D et al 1987 Science 238 1105–11 . 92 1256–65 Wald D J and Heaton T H 1994 Bull. J. 10. 109 B05302. Res. Geophys. Soc. 73 4681–94 Yamashita T and Knopoff L 1987 Geophys. Astron. Earth 43 1–33 Uyeda S 1978 The New View of the Earth (San Francisco: W H Freeman and Company) p 217 Venkataraman A and Kanamori H 2004 J. Seismol. Seismol. Soc. Res. Geophys. ed W H K Lee (San Diego. ed W H K Lee et al (San Diego. Phys. Lett. CA: Academic) pp 209–35 Utsu T 1999 Seismicity Studies: A Comprehensive Review (Tokyo: University of Tokyo Press) p 876 (in Japanese) Utsu T 2002 Statistical features of seismicity International Handbook of Earthquake & Engineering Seismology part A. Res.1029/2003JB002549 Venkataraman A et al 2002 Bull. 84 668–91 Wernicke B 1981 Nature 291 645–8 Wyss M and Brune J N 1968 J. CA: Academic) pp 719–32 Utsu T et al 1995 J.

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd