You are on page 1of 55

PAGEOPH, Vol. 121, 5/6 (1983) 0033-4553/83/050761-5551.50+0.

20/0
1983 Birkh/iuser Verlag, Basel

The Accommodation of Large Strains in the Upper


Lithosphere of the Earth and Other Solids by
Self-similar Fault Systems: the Geometrical Origin
of b-Value
GEOFFREY KING 1

Abstract - Plate tectonics has provided a method of visualizing the geometry o f the
deformation of the Earth's lithosphere on a large scale. The description is so concise that for
many purposes it provides an explanation of geological processes that overshadows the need to
understand the driving processes. The mechanics of the zones between the plates are less well
understood, particularly in continental regions where large areas are subject to deformation.
Both continuous and discontinuous models have been tried but both have obvious drawbacks.
In this paper concepts of geometrical self-similarity are adapted to provide a description of
the multiscale faulting that must occur in such environments. The fractal geometry of
Mandelbrot is applied to the problem of continental triple junctions and it is shown that certain
arrays of faults can "stabilize" a junction where three faults meet. The conditions required to do
this indicate that earthquakes of different sizes must occur in certain proportions. For simple
assumptions and conditions of triaxial deformation the proportion is that which is observed
globally for earthquakes. Thus, the b-value of unity found empirically by Gutenberg and Richter
and others can be regarded as a consequence of three-dimensional self-similar fault geometry.
The geometric description can be used to understand the way in which fault systems evolve.
Earthquakes initiate and terminate in regions where fault systems bend, because the bends
become zones subject to multiscale faulting. Movement on many faults in these regions
distributes the stress concentration of a propagating rupture front and terminates motion. The
multiple faults create offsets in the next fault to move. These offsets are the asperities that must
break before a new earthquake occurs.
The self-similar fault geometry requires that a substantial proportion of the deformation in a
fault system occur on minor faults and not on the main faults. The proportion of the deformation
taken up off the main fault depends on the form of the slip f u n c t i o n on the main fault.
The off-fault deformation produces aftershock sequences and forms background seismicity
and foreshocks. The geometric relation of aftershocks and foreshocks to the main faults suggests
that the former will tend to have b-values greater than unity and the latter b-values less than
unity.
The geometric description can be compared to the ideas of fracture mechanics, and it is
shown that for earthquake faulting and brittle deformation of the lithosphere in general, fracture
toughness, critical stress intensity factor and the Griffith Fracture Energy are not material
properties but properties of the geometry of fault systems.
Examined in terms of self-similar behaviour, concepts such as ductility can become ill defined.
What may be treated as ductile behaviour viewed at a large enough scale is seen to be brittle

' Builard Laboratories, Department of Earth Sciences, Cambridge University, Madingley


Rise, Madingley Road, Cambridge CB3 0EZ, England.
762 GeoffreyKing

when examined more closely. Clear-cut boundaries between the two phenomena do not
necessarily exist.
At very large scales and very small scales self-similar behaviour breaks down. Intermediate
scales occur but are not discussed at length. The upper scale limit is the Plate Tectonic scale,
which provides a very wide range of deformational boundary conditions, and the lower limit is
the scale of opening fissures. The latter process can be microscopic or macroscopic. At
whatever scale it occurs it is responsible for the confining pressure dependence of shear failure
that is responsible for frictional or dilatant behaviour of rocks. Although we do not discuss the
relations between fault motion and rotation, the large strains that can be accommodated by the
mechanisms we describe will, in general, be accompanied by rotations of similar magnitude.
This paper is concerned with the deformation of the brittle upper part of the Earth. Many
aspects of the descriptions, however, are also appropriate to the brittle deformation of any solid.

Key words: Earthquake; Faulting; Prediction; Fractal.

1. Introduction

The purpose of this paper is to discuss the mechanics of the Earth's


lithosphere and the geometry of faulting. The ideas presented originate in
a number of recent studies of earthquakes; but the paper addresses some
broader questions, in particular, the geometrical and mechanical
relations between large and smaU faults and the relation between
continuous and discontinuous deformation of rocks of the lithosphere.
To do so we shall consider scales that range from that of continents to
that of microfracture in rocks and seek to show that discovering rules
that relate deformation at one scale to that at another is crucial to
understanding how the lithosphere behaves mechanically.
The discussion applies the ideas of fractals developed by
MANDELBROT (1977) who suggests that geometrical self-similarity is a
widespread feature of the morphology of landscapes. He supports this
view with some synthetic landscapes generated using simple geometrical
rules. Two processes are obvious candidates for creating such fractal
geometry. One is the action of faulting, and the second is that of river
erosion.
While in general it is not possible to separate these two effects, under
some circumstances a separation is straightforward. We start this paper
with three examples of fault-determined geometry. They each indicate
that fault or fault-controlled geometries can be much less random than
the unfiltered Brownian form that Mandelbrot assumes and are chosen
to exclude the possibility that fluvial processes are associated with their
creation. This provides a justification for what may initially appear to be
an oversimplification in the theoretical faulting structures described here.
The simplicity of the faulting structures proposed is also why the
discussion differs from other considerations of fractal behaviour `of
faulting (e.g., ANDREWS, 1980a,b; KAGAN, 1982). The simple geometries
that are discussed can readily be made more complex by adding
Accommodationof Large Strains in the Upper Lithosphere of the Earth 763

randomness, but it does not appear that nature demands such


complexity.
Figure 1 shows a rock exposure. Although a range of rock face sizes
is visible, there are only a small number of orientations. Many exposures
of hard "rock that have not been greatly affected by weathering exhibit
this behaviour. Figure 2a is a lineament map of the surface of the planet
Mercury, and Figures 2b and c are directional analyses of the observed
features using Fourier transform methods (THOMAS and MASSON,
1983). The observed features presumably arise from faulting in the
brittle crust of the planet and any influence of rivers on Mercury can be
discounted. In this case three orientations are present. An examination of
lineament and joint orientation studies on Earth shows similar results
(e.g., TCHALENKO, 1970; HODGSON, 1961 ; OVERBEVand ROUGH, 1968;
BABCOCK, 1973). Patterns which show three orientations similar to
Figures 2b and c are most common. Figure 3 shows fault plane solutions
for the aftershocks of an earthquake in the Pyreenees (GAGNEPAIN-
BEVr~XX et al., 1982). The solutions are well controlled by a dense array
of stations and again the events fall into three groups. There is partial but
not complete spatial separatioh in the types of events so they cannot be
regarded as being on three discrete faults but must represent intermixed
faulting in different orientations, as the authors suggest. The faulting
mechanism we will discuss suggests that for two-dimensional (biaxial)
deformation faults should occur in sets of three, and for three-
dimensional (triaxial) deformation in sets of five. As we shall see later, a
horizontal section is likely to show three fault-strike directions; the
condition commonly observed.
Since earthquakes are a major manifestation of fault motion, certain
features of earthquakes underpin the following discussion. The most
striking is the self-similar behaviour of faults (TCHALENKO, 1970; KING,
1978) particularly as manifested in the earthquake frequency-magnitude
relation. A search for a geometrical means of understanding this
relationship has been a major impetus to this work.
In a discussion of Structural Geology, DE SXTTER (1956) stated that
there is no real difference between faulting and folding. To some extent it
is possible to show how this arises. Ductile behaviour, or what might be
described as pseudoductile behaviour, is not necessarily due to
microscopic crystalline mechanisms. (Here deformation due to the latter
is referred to as plasticity.) What appears to be ductile at one scale can
be entirely brittle when examined more closely. Thus whether a process
is described as brittle or ductile may depend on the scale at which it is
observed. Scale is very important. A concentration on a limited scale
range leads inevitably to confusion, which is a primary reason why a
marriage between laboratory studies of rock samples and the observed
field behaviour of rock has not been achieved.
764 Geoffrey King

Figure l(a). Shear planes in Jurassic recta volcanic rocks beside Route 132 near CoulterviUe,
California. The figure shows a general view of a roadside outcrop with planes of various sizes
visible. The outcrop has five directions of plane. These are identified in figure 1(b).

Under conditions where fractal fault behaviour operates, neither a


brittle nor a ductile description is satisfactory although either c a n be
applied with limited success. This explains the alternation o f the
geological theorists between block models (e.g., MCKENZIE, 1972; HILL,
1982) and continuum models (e.g., TAPPONmR and MOLNAR, 1976;
JACKSON and MCKENZIE, 1983) of continental deformation. In theory,
fractal systems are self similar at all scales, but in practice fractal
behaviour only operates over a range of scales between upper and
lower fractal limits. Defining what these limits are, and why they occur is
important to appreciating the physical significance of a natural fraetal.
Following a brief discussion of b-value, we consider large scale
deformation at Plate Boundaries. This scale provides the upper scale
limit of fractai fault behaviour. The behaviour of triple junctions is then
discussed, and we point out that it is the geometrical requirements of
deformation at triple junctions that requires that fractal fault behaviour
occur. A fractal geometry for faults at such a junction is proposed, and it
is noted that more faults can exist at such junctions than are required to
provide any specified amount of deformation. We then consider various
rules that determine why some possible faults are exploited rather than
Accommodation of Large Strains in the Upper Lithosphere of the Earth 765

Figure l(b). A detail of the outcrop in figure l(a) with the planes identified. Note the slicken-
siding visible on some planes.

others. These occupancy rules become complex when rupture pro-


pagation effects are considered, but a combination of an appropriate
three-dimensional fractal fault behaviour with a very simple rule
demanding continuous deformation using the minimum of fault area
predicts that the b-value of fault movement events is unity: the globally
observed average.
We then outline the application of these ideas to some model fault
systems and consider the inner limit of fractal fault behaviour. It is
shown that the inner fractal limit demands that solids subject to shear
deformation must expand and that this ultimately is the reason for
frictional laws and dilatancy prior to failure. Finally the problems of
initial fault formation are considered, although it is explained that the
766 Geoffrey King

a)
+ 2 5 /.,at.

4.~<>-. -'.. .... t.' ( Y~ - z'" |


''>"~'< ~."'.<'k'~ .....
~q\, +. ',', '( , , ,. \. / li, % I
"'+"'" ;," .-;,"-.',,"../,~,7" ,,k,!;',L~-., ",.' /
' ~ tt ~!"",
' ,.~
'/ ~l '''''>'
,.; , I ,' " \"' " ".";.4,
" ";'"'~
Iq 1; " ''~"+ 5(<" 13 " ~ i " i I/

o=
~n
M-~;..7.., ; ~ "t%~',; ~ , ''~" .~I'; :" z. ,~ / ~/,, ;;, h "~
o,
/..13j ~,,,~51 ,),1~3~\ ,. ~,,'r<., I'\t.~"~i, , ~ - ,'1 "~
v,~-,,).l,,~ , D';i~'[, ,~2 13'.~4.~,'U'. ~%..'" .% ,'j -
,.7"k"~ ~, ~ 5 ; , , ~ , ~ ''" ~ ~x I J , / ~ >.;,~" ,/ , ~i

V' ,.."r ,.i.~ "t ~-'.~,~.7; , 'z'~l

I
""..;, ; 'i,~.%, ,
" il ', - '.
'
,",~
,/#11 /
'
~,
I t
M#
' ""
/ ~%,,r';w
"~,
, , ' ~ . . , ' ,~'1 ,\7,~
I ~'-li#.4") ii 7 / ~ 4
,,;'~
--~
I

-25" I.~lt.

b) N c)
sun light

Figure 2. Alinement map of the surface of the planet Mercury taken from THOMAS and
MASSON (1983) is shown in (a). Diagrams of the orientation of linements are shown in (b) and
(c). The diagrams in (b) are for the four quadrants of the map, and (c) is for the map taken as a
whole.

discussions in the paper describe the relationship between complex fault


systems once they exist but do not apparently explain how they initiate.
This is analogous to plate tectonics describing the large-scale motion of
the Earth's lithosphere without describing the driving mechanism.
Accommodation of Large Strains in the Upper Lithosphere of the Earth 767
i : l i i i r I i

20 12 ! 3Q ~ 27

~ V

24
4T

3$

o .,,. 0"25' ' '= " ~ " "' "= 0 ~20 '

L L L . L 1

Figure 3. Aftershocks of the February 29, 1980 Pyrenean earthquake. The events range in
depth from 2 to 7 km. The upper hemisphere fault plane solutions shown were constructed using
upward going rays. The events fall into groups according to mechanism, but the groups are not
spatially separated, indicating that different fault mechanisms are intermixed,

The figures presented in the paper show how processes at different


scales relate. Drafting and printing problems, however, limit the range of
detail that we can present here. The reader is urged to imagine the effect
of including in some of the figures much greater detail than that shown.
It is only when this is done that the apparent oversimplifications will be
seen to disappear.

2. The magnitude-frequency relation o f e a r t h q u a k e s

The magnitude-frequency behaviour of earthquakes satisfies the


empirical relation
logN r = a - bm (1)
where N r is the total number of shocks of magnitude m or greater
(IsIJIMOTO and IOA, 1939; GUTENBERG and RICHTER, 1949). This holds
768 Geoffrey King

not only for earthquakes but also for microfracture in rock sample
studies (MoGI, 1962; SCHOLZ, 1968). It does not hold for volcanic
earthquakes (e.g., OKADA et al., 1981; TAKEO, 1983), a feature o f great
interest that will be discussed later.
The constant a is of little importance since it is determined only b y the
number of events in a sample. The value of b or "b-value" is of greater
importance. It is widely observed to be close to 1 and does not usually
fall below 0.7 or rise above 1.3. Various explanations of b-value have
been offered in terms of inhomogeneous stress distributions. These
demonstrate that the magnitude-frequency relation can be attributed to
suitable stress distributions, as indeed they must be, but they d o not
explain why such distributions exist. Such discussions transform the
question of "why do earthquakes obey the Gutenberg-Richter relation?"
to one of "Why do suitable stress distributions exist to cause
earthquakes to obey the Gutenberg-Richter relation?" We shall find here
that b-value is a direct consequence of long-term uniformity of finite
deformation under certain geometric constraints and is a necessary
consequence of the fractal nature of faulting. A consideration of stresses
is unnecessary and, in practice, is very confusing.
For the purposes of the later discussion, the magnitude-frequency
relation can be recast in more useful forms using the concept of seismic
moment (M0). This latter is the product of fault area (A), mean fault
displacement (d) and shear modulus (#) (M 0 =/~Ad). Since both fault
area and fault displacement are vectors, seismic moment is in fact a
t e n s o r (Mij). (For a recent description see MOLNAR, 1983). All o f the
discussion here concerns only the geometric attributes of the seismic
moment so we will use a quantity. ~0 = A d , which is again a tensor
( "~u) and we will refer to this as the moment.
The scalar moment (.//0) is empirically related to magnitude (m) by
log ~0 = cm + constants (2)
and thus
b
logNr = - - - log. ~0 + constants (3)

The empirically determined c varies over the magnitude range and


according to the definition of rn (magnitude) employed but m a y be
taken to be 1.5 (AI,:I, 1981; KANAMORIand ANDERSON, 1975).
The quantities A and d, which comprise the moment, are also related
empirically. The relation is normally noted by the statement that
earthquakes have relatively constant stress drops (e.g., SCHOLZ, 1982).
This is attributable geometrically to a constant ratio of fault displace-
ment to fault width ( L - t h e shortest dimension of the fault plane). Thus if
Accommodation of Large Strains in the Upper Lithosphere of the Earth 769

we restate constant stress drop as constant strain drop (s), we find that
d = sL. (4)
If we further assume that all the faults in a system have the same aspect
ratio, then
J" = sGL 3 (5)
where G is a geometric constant close to unity for faults with low aspects
ratios.
The result that..~" is proportional to L 3 is widely quoted and implies
three-dimensional self similarity (AKI, 1981). It is, however, worth
noting that the aspect ratio of faulting could vary systematically with
moment over a range of scales. We shall see that three-dimensional self
similarity implies triaxial deformation. Biaxial deformation occurs when
one fault dimension remains fixed for a wide range of fault areas. Self
similarity only describes a two-dimensional cross-section and moment is
proportional to L z under these circumstances.
By combining the foregoing results, we find for three-dimensional self
similarity that

3b
log N(L) = - - - - log L + constants (6)
c

and for two-dimensional self similarity


2b
log N(L) = - - - logL + constants (7)
c

In each case N ( L ) corresponds to the number of events with fault length


greater than L and is equivalent to N r in the tables.

3. Deformation at a continentat plate boundary

In contrast to typical oceanic plate boundaries, the deformation of


continental boundaries occurs over a broad zone. This can readily be
seen by comparing the seismicity of the western United States or the
Alpine-Himalayan belt with the seismicity of ocean basins. The reason
for such broad zones of deformation may be understood using the basic
geometrical ideas of plate tectonics.
In the oceanic environment three types of boundary are permissible:
transform faults, ridges, and trenches. The second two types do not
conserve area and generate and destroy plates. They exist because
oceanic lithosphere can be created from or returned to the mantle
reservoir. The narrow belts of seismicity associated with these features
770 Geoffrey King

result from the faulting necessary to change horizontal plate motion into
the vertical motion that carries material either into or out of the mantle.
This can be accomplished in a width not much greater than the plate
thickness. The greater apparent width of subduction zone seismicity is
due either to deep earthquakes in the subducting slabs or to deformation
of continental material associated with the subducting system.
The situation in a continental environment is very different.
Continental lithosphere cannot be derived from or returned to the
mantle, and thus the simple plate tectonic mechanisms for the creation
and destruction of plate area cannot operate. A normal fault can
accommodate some extension and a reverse fault some shortening, but
these features are limited to a total throw comparable to the crustal
thickness. In convergent or extensional regions greater motion can be
accommodated by the development of a series of parallel or subparallel
features that distribute crustal thinning or thickening over a wide area.
Other mechanisms such as overthrusting of one continental unit over
another or motion within the crust along detachment surfaces also
apparently occur to produce crustal thickening (e.g., WERNICKE and
BURCHFIELD, 1982; BURCHFIELD,1983; FROST and MARTIN,1982) and
major strike-slip faults form to transport material laterally to avoid
overthickening of the crust in particular places (TAPPONIER and
MOLNAR, 1976).
Whatever the mechanisms involved, it is clear that in the absence of
the mantle as a source or sink of material, deformation must spread over
a wide region. The ultimate cause of the difference between this
behaviour and the behaviour at ridges and trenches is the low density of
continental crust and not its rheological properties.
Some aspects of the nature of a continental plate boundary may be
understood by considering the motion in a zone between plates as if it
were purely ductile. Figure 4 shows the form of flow lines associated

const.
V

Figure 4. The compression of a hypothetical ductile crust over a ductile mantle. The flow lines
to produce the observed deformation must curve and vary in spacing. A system of faults in a
brittle medium required to accommodate the same overall deformation must be complex. The
largest faults can only approximate the deformation, and many smaller faults at different
orientations are needed to approach the continuous ductile deformation field.
Accommodation of Large Strains in the Upper Lithosphere of the Earth 771

with an idealized form of crustal thickening. Whatever the rheological


properties of the crust m a y be, the flow lines must bend. If we now
consider that the finite deformation is accommodated by faulting, then
the fault system must evolve with time and incorporate many faults of
different orientations. This is true whether the faults are straight or
curved. Only straight faults or faults which are arcs of circles can
accommodate motion without the creation of new structures, and the
type of deformation that can be accommodated by a small number of
such simple structures is very limited. Hence, in general, whatever system
of faults is actually created, that system must both envolve with time and
include faults in various directions.
A similar conclusion can be found by considering horizontal flow in a
hypothetical ductile region between two plates. This is shown in Figure
5. (Ductile models of this sort are discussed by MCKENZm, 1979; and

a} (V2)o
1/I//I
t . (Vt)o
~ i l l / i l l

/ /// / / / / / / / / /

fixed

b) > c) ,,,,-77T,,,,
_ /- (

/IIIIII[1" IIIIII/II

//*rl///ll/l / 1 1 1 1 1 1 1 [

Figure 5. Ductile deformation in a zone between rigid blocks modelling a continental plate
boundary. One block is taken to be fixed (lower) and the other can move as shown in (a). In (b),
the upper block displaces parallel to the block boundaries to produce a simple shear flow. This
can, in principle, be relieved by a simple shear fault (c). In practice it is not, for reasons
discussed later in the text. A component of motion perpendicular to the block boundaries results
in a much more complex flow. Under these boundary conditions a brittle material can only
respond with complex faulting.
772 GeoffreyKing

MCKENZIE and JACKSON, 1983.) One plate is taken to be fixed, and the
second moves in the plane of the paper. The velocity in the ductile band
is given by:
vl = allxl + al2x2 (8)
V2 = a 2 1 x I + a 2 2 x 2 (9)
In general (v:) 0 :~ 0 and (vl) 0 4= 0 and under these circumstances the av
are nonzero and are functions of position. A simple case occurs when
(v2) 0 = 0 (Figure 5b). The coefficient a12 is a constant and the other a o.
are zero. The deformation is simple shear as shown in Figure 5b. This
deformation can be accommodated, in principle, by a single simple
strike-slip fault (Figure 5c). In general the flow system is not simple.
Figure 5d shows one possible flow pattern. Figure 5e shows a possible
arrangement of faults accommodating it. It again follows that a system
of faults is needed to accommodate such a flow whether the faults are
straight or curved, and that they must evolve with time and include
active features in different directions.
The foregoing discussion has two purposes. One is to show that
multidirection faulting must occur under the conditions normally
found at continental plate boundaries; it is a consequence of finite
motion and the geometrical constraints placed on the system. Further-
more, the faulting will include more than one fault direction whether
viewed in plan or in section. Second, we will find that the fault systems,
required to accommodate finite deformation, form fractai sets, and that
the geometrical systems that we have described provide the upper fractal
limit to these sets and hence determine the upper bound on fractal
behaviour.
Before discussing fractal fault behaviour we note that, in general, finite
motion on two faults that meet at an angle requires the creation of a
third fault. This can be seen by referring to Figure 6. The third fault can
be in one of two positions, but the orientation and slip rate is specified by
the orientation and slip rate of the other two planes. The junction
between the three faults is a triple junction similar to those discussed in
connection with the evolution of oceanic plates and shall be discussed in
that context in the next section. It should be noted at this point and in
relation to Figure 6 that only motion in the plane of the paper is
important. This assumption is implicit in some later figures but wiU not
be restated.

4. C o n t i n e n t a l triple j u n c t i o n s

MCKENZtE and MORGAN (1969) discuss the stability of triple


junctions between plate boundaries that mix one or more of ridges,
Accommodation of Large Strains in the Upper Lithosphere of the Earth 773

(a)

-- c

A ~ A
\

(c) A

,/ t3 (d) l
E
\" "1
\ \\\
\

C A r B'
Figure 6. Two faults meeting must produce faulting in a third direction. Two possibilities are
shown (a) and (b), with their corresponding vector triangles (c) and (d). The faulting near the
junction of the faults (shaded) must be complex.

trenches and transform faults. They show that, provided certain


geometric rules are obeyed (demanded by the need for ridges to spread
at right angles and for trenches to consume on one side only), then all
junctions that include one trench or one ridge can occur without the
feature changing form. The inclusion of one or more structures, out of
three, that create or destroy material permits stable behaviour.
The only triple junction that is unconditionally unstable is that formed
by the junction of three strike-slip faults. We argued, however, in the last
section that such junctions are the only features that can occur on a
large scale in continental deformation; therefore how they behave is
critical.
First we note that the relative displacement on the three faults must
drop to zero at the actual point of the junction. This is shown
schematically in Figure 7. The faults have a slip function that reaches a
maximum at their centers and drops to zero at their ends. The faults are
consequently effective at producing deformation near to their centers,
but ineffective at their ends. The amount of deformation that must be
accommodated by other structures is indicated.
To understand how this takes place it is convenient to think in terms
774 Geoffrey King

B
i,i i r
/ I

.11- f ~

a)
\\
C

b)
J DISPLACEMENT
DISPLACEMENT
ON FAULT A
DISPLACEMENT
ON FAULT C

+
~ EQUIVALENT
c) | DISPLACEMENT
EQUIVALENT T ~ OFF FAULT B
DISPLACEMENT
OFF FAULT A
EQUIVALENT
I DISPLACEMENT
OFF FAULT C

d) /
~ b TOTAL
DISPLACEMENT
a" TOTAL DISPLACEMENT ~,
I ,~
~ AND EQUIVALENT
DISPLACEMENT
AND EQUIVALENT
DISPLACEMENT FOR
FAULT A
~ TOTAL DISPLACEMENT
~qJlll AND EQUIVALENT
"%,1 DISPLACEMENT FOR
FAULT

o
Figure 7. Processes where three faults meet. In (a) at the junction where the faults meet their
displacement is zero. The displacement on the faults increases away from the junction but will
come to zero again at the next junctions (not shown). The displacement as a function of position
along the faults is shown in (b). The area under the fault displacement curve is proportional to
the moment released by the fault (the figure is for a two-dimensional representation so that
moment is proportional to the product of fault length and displacement). The equivalent
displacement that must be accommodated by off-fault structures to permit main fault
displacement to come to zero at the barrier is shown in (c). This sums with the fault displacement
to give the equivalent total displacements shown in (d). Total displacements must sum to zero in
a vector fashion as shown.

of moment. To illustrate this the area that is hatched in Figure 7b is the


moment that can be accommodated by the main faults, and the hatching
in Figure 7c indicates the moment to be accommodated elsewhere such
that the total effective deformation is the same as if the faults actually
met with constant displacement (Figure 7d). At this point we shall
consider faults of infinite extent perpendicular to the direction of slip and
Accommodation of Large Strains in the Upper Lithosphere of the Earth 775

therefore regard the moment as proportional to the fault length squared


(biaxial deformation). The point of intersection of the faults we shall
term a "barrier" as defined by KING and YIELDING (1983) following
work of AKI (1979). We shall find later that the junction only creates the
conditions that produce a barrier and is not itself a barrier.
In order to proceed further we must make assumptions about the slip
function (i.e., the distribution of displacement on the fault plane). For
ease of visualization we take it to be triangular for the time being. Under
these circumstances 50 percent of the deformation (or moment release)
takes place on the main faults and 50 percent must be taken up
elsewhere. The choice of a triangular function and the 50:50 ratio is
simply for convenience of discussion in the absence of any better
alternative distribution. It might be argued that the slip distribution for a
crack in an infinite medium (similar to that shown in Figure 7b) is more
appropriate. However, later discussion will show that elastic approxi-
mations are not necessarily correct because the medium cannot be
regarded as a continuum at any scale. Thus the lack of clarity introduced
by considering a nontriangular slip function is not justified.
Before looking at an intersecting junction such as that shown in
Figure 7a we can visualize some of the features to be discussed later by
considering a single fault with a triangular slip function coming to zero
at two barriers (Figure 8). This slip function is shown by the shaded area
in Figure 8b. The rest of the deformation is taken up by an infinite
number of smaller faults. These are shown in Figure 8a but their position
is unconstrained. The smaller faults fall into a series of orders. Each
order has faults of half the length of the next highest order, and each
order is composed of twice the number of faults of the next lower order.
Order 1 is taken to be the main fault. These features are summarized in
Table 1. The moment of faults of each order is shown on the assumption
that the faults are constant in length in the third dimension (into the
paper) and hence moment is proportional to the square of fault length
(the fault of order 1 is assigned a length of 1).
Examining the right-hand column shows that the sum of the moment
of all the faults of order less than 1 equals the moment of the main fault
(2 = I + + + ~ + ~6 + ..., etc.). Using Table 1, a fault length
frequency relation can be plotted (Figure 9) and the slope compared with
relation 7. A b-value of c/2 is found for small events. It is important to
appreciate that this b-value does not define the relative amount of
deformation on and off the main fault. This can be seen if we consider a
system of faults with the behaviour shown in Table 2, which also gives a
b-value of c/2 for small events (log L < 0), but in this case the proportion
of deformation accommodated off the main fault is half the previous
case (1.5 = 1 + + ~6 + ~2 + .... etc.). (The slip function is no longer
triangular.)
776 Geoffrey King

a)
2
f.. e , , . . , . . . o ~
~A A

b)
Stip

Figure 8. Brittle deformation off the main fault plane. In (a) a single fault is shown with two
barriers at which displacement comes to zero. Schematically a set of smaller faults is indicated,
each ending in barriers. These faults accommodate deformation that the main fault cannot
accommodate. The way in which this takes place is shown in (b). The slip on the main fault is
indicated by the shaded triangle. The slip on the smaller faults is indicated by the smaller
triangles. The slip behaviour of all faults is required to be self similar (i.e., the slip functions are
similar triangles). The sum of the slip on the main fault and an infinite series of smaller faults is
required to give constant overall slip along the fault zone. This figure illustrates the principles to
be used to ensure self similarity and determine the relative proportion of off-fault and on-fault
deformation in later, more complex models. Notice that the triangular slip function requires that
50 percent of the deformation occur on the main fault and 50 percent off the main fault. This
ratio is chosen for much of the subsequent discussion because it makes it easy to visualize how
moment is distributed between different scales of faulting when the function is triangular.
However, the basic arguments do not depend on the slip function being triangular.

Table 1

Cumulative
number Total moment
including all Moment of of all faults
Order Length Number lower orders each fault of given order
(n) (L,,) (N.) (Nr) (.4".) (N..//.)

1 1 1 1 1 1
2 1/2 2 3 1/4 1/2
3 1/4 4 7 1/16 1/4
4 1/8 8 15 1/64 t/8
5 1/16 16 31 etc. 1/256 1/16

The origin of the difference can be seen by examining Figure 9 and


arises because true self-similar behaviour requires that fault size relations
be described by geometric series and not by continuous functions. Before
it is possible to determine how deformation is partitioned between larger
Accommodation of Large Strains in the Upper Lithosphere of the Earth 777

tog N

ble 1

0
-3 -2 -1 o tog L
Figure 9. The logarithm of the total number of faults with length greater than L plotted against
the logarithm of fault length L. The upper line is for Table 1, which corresponds to faulting with
50 percent of the deformation occurring on the fault and 50 percent off the fault (defined in
Section 9 as ~# = 1). The lower line is for Table 2 where the amount of off-fault deformation is
halved ( # = 0.5). Both lines have similar slopes and hence the b-value of both systems is the
same except near to log L = 0. Thus b-value alone does not define the ratio of on-fault to off-fault
deformation. Other geometrical information is needed.

Table 2

Cumulative
number Total moment
including all Moment of of all faults
Order Length Number lower orders each fault of given order
(n) (L) (N,,) (Nr) (. #',,) (Am,~,,)

1 1 1 1 1 1
2 1/4 4 5 1/16 1/4
3 1/16 16 21 1/256 1/16
4 1/64 64 85 1/4098 1/64
etc.

and smaller earthquakes, it is necessary to have information about where


secondary faults are located and what their dimensions are. It is for this
reason that the result based on b-value derived by WYss (1973)
suggesting that half the deformation in a region occurs on the largest
faults is not necessarily correct.
When a linear fault with barriers is considered, there are no
constraints placed on the locations of minor faults. However at a triple
junction the possible positions of secondary faulting become more tightly
constrained by the requirement that faults do not cross (i.e., where two
faults meet, one must have zero displacement) and the requirement that
the entire system must be self similar. Before this is discussed it is
necessary to consider fractal geometries.
778 GeoffreyKing

5. Self-similar geometry and fractals


In some recent books and articles Mandelbrot has drawn attention to
the wide distribution in nature of geometrical patterns that look the same
at whatever scale they are examined (i.e., they are self similar). These
self-similar figures he terms fractals and discusses their properties.
Natural fractals include trees, rivers, coastlines and the shapes of
mountains.
A basic property of fractals is illustrated by posing the apparently
innocuous question, " H o w long is the coast of Britain?" The question
has no simple answer, since the length increases indefinitely as the
interval at which it is measured decreases. It is, in fact, unreasonable to
pose the question unless a scale is specified at which further detail may
be ignored. It is possible, however, to ask how much the length increases
as the level of scrutiny increases. A plot of the logarithm of the length of
a coastline against the logarithms of the length of the yardstick used to
measure the coast yields a straight line. The slope of this line is 1 - D
where D is the fractal dimension of the coastline. The west coast of
Britain has a fractal dimension of 1.25 (i.e. length ~ (yardstick)-l/4).
The significance of fractal dimension that relates behaviour at one
scale to behaviour at another is most readily understood by considering
features that are less random than coastlines. Figure 10 shows an
example of the development of a fractal shape and the way in which it is
constructed. The shape is based on a triangle, which is termed the
initiator. This is then operated on at each successive scale by the
generator indicated at the bottom of the figure. The form of the generator
determines the fractal dimension D of the final figure

log N
D=~ or Nr = 1 (10)
log 1/r
where N is the number of equal segments in the generator and r is their
length as a fraction of the direct end-to-end length of the generator. The
example has N = 4 and r = ]. D is therefore log4/log 3 = 1.26.
The fractal dimension relates in an interesting way to Euclidean
dimension. A generator with a dimension of 1 that connects its two end
points will produce a straight line out of a straight line, and an initiator
operated on by such a generator will therefore remain unchanged. A
generator with a dimension of 2 will completely fill an area if no overlaps
occur, and a generator with a dimension of 3 will fill space. A generator
with a fractional dimension between 1 and 2 will partially fill an area,
and a generator with a dimension between 2 and 3 will partially fill a
volume. Not all generators are continuously connected and hence do not
produce figures that are continuously connected. Thus a generator with
D = 1 can turn a straight line into a dust. Higher dimension generators
Accommodation of Large Strains in the Upper Lithosphere of the Earth 779

Figure 10. An example of a simple fractal figure. The initiator is shown at the top of the page
and the generator is at the bottom. The intermediate figures are formed by successive operations
of the generator on the initiator. The generator has N -- 4, r = 3, D = log4/log3 ~_ 1.26 Note
that because the elements of the generator are non-parallel the fractal dimension of this figure will
change if it is distorted. This is not true for the faulting figures (11, 13).

can exhibit the same property including the generator that we adopt to
describe fault behaviour. The fractal dimension of figures produced by
such generators cannot be determined empirically in the same manner
used to determine the lengths of coastlines. Other methods must be
adopted, since a set of faults forms a disconnected rather than a
connected system.
It is important to appreciate that, while it is simple to produce a
generator of any chosen dimension, it is difficult to find generators that
780 GeoffreyKing

by not creating overlaps produce "real" figures. Once a generator is


found, the construction of a figure is very simple and results from
repeated application of a simple rule. Computer algorithms to d r a w such
figures are therefore very compact. Mandelbrot points out that if
understanding a phenomenon is no more than having a simple w a y to
describe that phenomenon, then fractals permit a large range of
phenomena to be understood. It is well worthwhile pondering the
significance of this statement.
Although a perfect fractal must exist at all scales, in practice there are
inner and outer limits at which fractal behaviour ceases (the inner and
outer fractal limits). It is clear that the fractal behaviour of the British
coast will cease at the atomic level if not at the crystalline level, and at
the upper scale limit, all coastal scales are limited by the circumference
of the Earth if not by smaller-scale tectonic processes.
Appreciating the nature of the fractal limits of a fractal process can be
the key to understanding why the system exists. This can be illustrated
by considering a tree. The upper fractal limit of a tree is the scale of the
trunk, which connects the tree to the root system below the ground. The
lower fractal limit is the scale at which twigs connect to the leaves. One
purpose of the fractal system is to carry fluid, which is best transported
in large conduits (the trunk and larger branches), and spread it over a
surface (the leaves). The fractal geometry of the branches connects from
the one-dimensional trunk to the two dimensions of the leaf c a n o p y in an
approximate way. The need to do this arises because photosynthesis
operates most effectively in approximately two-dimensional structures,
whereas fluid transmission is most readily accomplished in one-
dimensional structures. The root system operates in a slightly different
way, since fluid in the soil is collected from a volume and not from a
surface.
A fractal remains a fractal when it is incomplete. Trees grow by
adding new twigs at regular intervals. Most of these are lost as the tree
grows. Thus an algorithm to produce a tree must include rules to both
create and delete features. The fractal shape that we will use to describe
faulting includes more elements than are needed, and some o f the
simplicity of the complete figure is not apparent because only a part of it
ever comes into existence. Natural fractals in general may be expected to
behave in this way. Thus there is an important distinction between the
fractal that describes all the possible faults that could accommodate a
certain type of deformation and the actual fractal fault system that
occupies part of the theoretical figure. The two need not even have the
same fractal dimension.
A fundamental property of fractal shapes is that they cannot be
differentiated. The generator form appears when a small element is
examined, and a tangent is never the result of close inspection. Calculus
Accommodation of Large Strains in the Upper Lithosphere of the Earth 781

can be used at the inner fractal limit if Euclidean behaviour is restored,


or at any scale at which the features may be blurred such that a fractal
description is unimportant.
At any scale at which a fractal description is needed, and this applies
to some of the most important aspects of fault behaviour, calculus
cannot be employed. The earlier discussion concerning Figure 8
illustrates the type of problem to be encountered. Thus calculus-based
concepts such as stress and strain must be abandoned.

6. A fractal forrn for two-dimensional faulting

Provided that only a single-fault orientation is considered, all manner


of generators may be constructed to produce secondary faulting to
permit deformation around barriers. Where three faults intersect and no
faults are permitted to cross or duplicate, however, the fractal forms are
severely constrained. It is not clear to the author just how constrained.
One satisfactory structure is discussed but others may exist.
All the figures here are drawn for faults that meet at a 120 or 60
angle. This is not a limitation since any system that operates for these
angles will operate for any other angles. This may be appreciated by
imagining all the forthcoming figures to be drawn on highly deformable
rubber membranes.
Figure 11 shows part of a system of faults based on a three-fault
initiator system (which can extend to a hexagonal grip). The generator
produces four subfaults for each main fault, N = 4, r = 0.5. The fractal
dimension of this system is therefore 2. The fault system covers area
without any overlapping or fault crossings. Note that the figure as
presented is only part of the ideal figure for two reasons. First, some of
the faults that should result from the first application of the generator are
not shown. Second, only five applications of the generator are shown.
The deformation that such a system can accommodate is indicated in
Table 3. It is apparent from the table that, for this geometry, if all the
possible faults exist, then equal amounts of total moment are contributed
by all scales of faulting, and that the deformation provided by all the
smaller faults is greater than that provided by the largest fault. However,
the fault length that must be created to produce the same deformation
doubles with each increase in order. Thus, if the energy required to
deform the body depends only on the total length of faulting that must
move, then lower-order faulting will be produced in preference to
higher-order faults. This is the generally observed behaviour of faults.
If we adopt the triangular slip function used earlier, we can now
discuss how the possible fault positions can be partially occupied in a
self-similar fashion. The triangular slip function requires that all faulting
782 Geoffrey King

J
J

zjz b
J

0.5 05 ~ J

t-0 ~/

05 0.5

Figure 11. A fractal geometry for two-dimensional faulting corresponding to Figure 6a. The
initiator consists of the faults a, b, and c, and the generator is shown at the bottom of the figure.
Five orders of faulting are shown corresponding to four applications of the generator. The figure
has the property that no fault intersects any other fault. The generator has N = 4, r = , D = 2.
This remains true if the initiator is uniformly distorted, provided that the generator is subject to
the same deformation. The same is not true, in general, for fractal figures (e.g. figure 11).

at higher orders than a specified order accommodate the same


deformation as that order. Such a system is described by Table 4.
At each successive order both the proportion of subfaults utilized and
the moment relieved is halved. The total fault length at each scale is ~he
same, but the effectiveness of faulting to produce deformation reduces as
Accommodation of Large Strains in the Upper Lithosphere of the Earth 783
Table 3

Total fault Moment


Cumulative Proportion length at of each Total
Order Length Number number of positions given order fault moment
(n) (L) (n) (N r) used (N,L,) (,',) (N, .',)

1 1 I 1 1 1 1 1
2 1/2 4 5 1 (all) 2 1/4 1
3 1/4 16 21 I (all) 4 1/16 1
4 1/8 64 85 I (all) 8 1/64 1
etc.

Table 4

Total fault Moment


Cumulative Proportion length at of each Total
Order Length Number number of positions given order fault moment
(n) (L) (n) (N r) used (N~L,) (''n) (N~ If,)

1 1 1 1 I 1 1 1
2 1/2 2 3 1/2 1 1/4 1/2
3 1/4 4 7 1/4 1 1/16 1/4
4 1/8 8 15 1/8 1 1/64 1/8
etc.

the order decreases. This is a direct consequence of moment being


proportional to the square of fault length.
The arrangement utilizes only a small proportion of the fractal pattern
shown in Figure 11, and with N = 2 and r = the fractal dimension
becomes equal to 1.0. This arrangement can be considered to be the fault
system that will be produced in a completely rigid, brittle body in which
no deformation can be taken up in any other way. Any displacement on
the first-order fault must be accompanied instantly by motion at all
scales. If continuum deformation is permitted, then some fault motion
can occur without requiring further faulting at smaller scales. We shall
consider how this can operate in the next section.

7. Simple geometrical implications for fault behaviour


To allow for continuum deformation we shall examine the conse-
quences of another rule. We shall permit any fault to move one
displacement unit (inversely proportional to fault length) without motion
on any other faults. For the same fault to move a second time, however,
either two associated faults of the next highest order must move, or the
784 GeoffreyKing

equivalent deformation must occur on even higher-order faults. This can


be thought of in the following way. The deformation associated with one
unit of fault motion can be accommodated elastically in the surrounding
media, but this elastic strain must be relaxed by secondary faulting
before the fault can move again. The fractal system is only required to
extend down to a scale at which elasticity can accommodate the
remaining deformation.
The simplest sequence that obeys the above rule is shown in Figure
12. We move the main fault (order 1) three times, the first time without
the creation of secondary faulting (Figure 12a). Two second-order faults
then appear to allow it to move a second time (Figure 12b). Two more
second-order faults permit it to move a third time (Figure 12c). The
sequence is shown schematically in Figure 12d. It is helpful to appreciate
that each of the second two increments of faulting has a fractal
dimension of 1 (N = 2, r = ), but the resulting faulting has a fractal
dimension of 2 (N = 4, r = !) 2 and illustrates in a primitive way how the
fractal dimension of the process that can create a figure may differ from
that of the final figure.
An alternative arrangement to allow the main fault to move three
times is shown in Figure 13. The first two stages (Figures 13a and b) are
the same as before, but in the third stage, third-order faulting appears,
and the second-order faults move for a second time. In this case both the
process and the figure have fractal dimensions of 1.
Figures 12 and 13 are both so simple that they may appear
unrealistic. It is, however, straightforward in principle to construct fault
sequences in which the simplicity although present is much less obvious.
This is illustrated in Figure 14. All second-order faulting is absent, and
the compensating deformation at higher orders of faulting is distributed
randomly. In this example, the main fault has still moved only three
times. If we now imagine the effects of the other two main faults and of
more increments of movement, a very elaborate pattern of faulting can
result.
It should be pointed out that failure to fully exploit the lowest orders
of faulting requires that a greater total length of fault move or be created
than the minimum required to produce the necessary deformation.
Presumably, in terms of the energy of fault surface creation alone,
irregular systems are less efficient. In other words the creation of the
largest faults of the fractal pattern first (i.e. the scheme of Figure 12) is
optimal. However, other criteria also determine the overall energy
involved in faulting. We shall return to this question in a qualitative sense
when discussing examples of observed fault behaviour, but one
geometric figure deserves examination here.
This is shown in Figure 15. The figure is similar to Figure i 1 except
that fault c has been shifted to lie between a and b, and only faults that
Accommodation of Large Strains in the Upper Lithosphere of the Earth 785

2 2

b)
lr2

2 2

c) 1r2f3

3 3

Order

d)

Figure 12. Sequences of movement on a fault and subfaults. The faults should be regarded as
one orientation set from a system such as that shown in Figure 11. In (a) the main fault moves in
the first increment of deformation. This is indicated by the 1 marked beside it. The rest of the
deformation is accommodated elastically. In a second increment of deformation, (b), two
second-order faults move and the main fault moves again. This is shown by marking the
second-order faults and the main fault with the number 2. The final increment of motion is shown
in (c) and two more second-order faults (marked 3) move. The main fault, which has moved each
time, is marked 1, 2 and 3. The process is summarized in (d) which indicates a "family tree" of
fault orders and indicates which have moved and in which increment of deformation.

deform the region of the bend between faults a and b are included. We
can construct tables for the secondary faulting of this system noting that
a and b behave in the same way (Table 5) but differently from c (Table
6).
786 Geoffrey King

a) 1

2 2

b) 1,2

2,3 2,3
3 3 3
c) lf2,3

d) Order
1 1,2,3

2
3 33 33
Figure 13. A figure similar to Figure 12. The same deformation occurs for the first two
increments shown in (a) and (b). However, in (c) a third increment of motion occurs on
third-order faults and possible second-order faulting is not exploited. The "family tree" for the
sequence is shown in (d). Only faults marked with deformation increment numbers have moved.
The other lines correspond to possible but unexploited fault locations. The total fault length
created for three increments of motion on the main fault is the same as for Figure 12, but the
second-order faults must move twice.

With a system of the sort shown in Figure 15 it is apparent that the


system cannot repeat indefinitely (on a hexagonal grid) in the same way
as Figure i 1. The junction region can therefore be regarded as falling
under the influence of only half of each fault. Considered in this way, the
figures in brackets are appropriate for fault numbers and total m o m e n t
since the other numbers imply an unrealistic infinite repetition. However,
since the ratios remain the same whichever set is chosen, this does not
affect the following discussion. This sort of problem inevitably results at
Accommodation of Large Strains in the Upper Lithosphere of the Earth 787

a)

2 2 2
2 lj2 Z 2-
2- ,, b)
2, 2 2
2_..

2-

I
3 r/
1,3 z~ z3
--2 _L _L 2 2 3
If2~3
2 3 2 3 2 3 3

3 .2. 2--

z
3 3

Order
1
i 1,2,3
d)
2

3
2 33 23 32 23 3 2 32 3 323323

Figure 14. A more complex fault evolution than those shown in Figures 12 and 13. No
second-order faults move to permit a second increment of main fault motion. Instead, six
third-order and eight fourth-order faults move. The third increment of motion is accompanied by
motion on two further third-order faults and further motion on the six that moved in the second
increment of deformation. To permit this, motion also occurs on 12 fourth-order faults. The
"family tree" is shown in (d). Many possible fault positions are unoccupied, including all possible
second-order faults. The total length of faulting created is greater than for Figures 12 or 13.
78 8 Geoffrey King

b
Figure 15. A fractal form for faulting corresponding to Figure 6b. With deformation restricted
in the way shown, some faulting orders are only partially available or are absent.

the upper limit of fractal behaviour. The outer fractal limit boundary
conditions require that some parts of an idealized fractal figure cannot
exist.
It can be seen from Table 5 that a deformation system can proceed
that includes all orders of faulting for faults a and b, but for c, the
second-order faults generated by c are absent. There are also insufficient
third-order faults to permit repeated motion on the main fault. For
repeated motion, all faults to the fifth order must move for every
movement of the main fault, and only at higher orders can partial motion
occur. Thus the system requires more fault length to be created than the
system in Figure 11 for a given amount of total deformation, but the
deformation is more localized. Geometries of this type occur in the upper
Accommodation of Large Strains in the Upper Lithosphere of the Earth 789

Table 5

Moment
Order Length Number each fault Tota moment
(n) (L~) (N,) (~',) (N~ ,r,)
1 1 1 (1/2) 1 1 (1/2)
2 1/2 2 (1) 1/4 1/2 (1/4)
3 1/4 6 (3) 1/16 3/8 (3/16)
4 1/8 32 (16) 1/64 3/8 (3/16)
5 1/16 96 (48) 1/256 3/8 (3/16)
etc.

Table 6

Moment
Order Length Number each fault Total moment
(n) (Ln) (N~) (d"~) (N n ~n)

1 1 1 (1/2) I i (1/2)
2 I/2 0 (0) 1/4 0 (0)
3 1/4 4 (2) 1/16 1/4(1/8)
4 1/8 16 (4) 1/64 1/4 (1/8)
5 1/16 64 (32) 1/256 1/4 (I/8)
etc.

block of low angle dip-slip faults, and pairs of similar structures occur at
offsets in strike-slip faults.

8. Fraetal faulting in three dimensions

The discussion so far has concerned two-dimensional faulting. These


are systems in which all faults have one common dimension along which
no deformation occurs. The equivalent continuum behaviour is described
as plane strain. Such systems give a b-value of 0.5 c. Taking c to be 1.5,
this produces a b-value of 0.75, which is lower than that normally
observed. A more realistic b-value is obtained with three dimensional
deformation.
Drawing three-dimensional figures is a difficult, time-consuming
process. The nature of triaxial fractal deformation, however, can be
understood as a generalization of the previous discussion. We note first
of all that while three sets of planes are needed to permit biaxial
deformation, five are required to permit triaxial deformation. This
commonly quoted result is a consequence of von Mises' criterion (YON
MISES, 1928). It is the ability of metal crystals to permit dislocation
790 Geoffrey King

motion on five or more planes that allows metals to behave in a ductile


fashion. The rock deformation that we are discussing is, in effect, an
enforced ductile behaviour. Many rock faces show a rhomboidal
structure cut by two other planes. An example was shown in figure 1.
To discuss the geometry, we shall consider plane orientations that have
the angular relations of a square-based pyramid (Figure 16a) and a
generator of the form shown in Figure 16b. The figures produced by the
foregoing combination are very difficult to visualize. The generator must
be applied to the boundaries of the initiator planes in such a way that a
series of self-similar planes with the same angular relations is created. As
an aid to visualizing the three-dimensional geometry, the author
constructed a large number of pyramids similar to that shown in Figure
16a with sizes varying by factors of two.
It is immediately clear that however the five planes are chosen, they
cannot all be the same shape. Some can be rectangular and some
triangular. They cannot be either all triangular or all rectangular. The
implications of this are not considered.
For the new initiator we can construct a table (Table 7) that indicates
the deformation available at different scales in the same way that we
have done for two-dimensional systems. The moment is now propor-
tional to the cube of the length of a fault.
Again we find that more small faults exist than are required, if 50
percent of the deformation is to be accommodated by higher-order

a) b)

Figure 16. The initiator (a), and generator (b), for three-dimensional fractal faulting. N = 8, r =
, D = 3.
Accommodation of Large Strains in the Upper Lithosphere of the Earth 791

Table 7

Cumulative Moment of Fault Total


Order Length Number number each fault area moment
(n) (L,,) (N.) (Nr) (//.) (a.) (N. ~'.)

1 1 1 1 1 1 1
2 1/2 8 9 I/8 2 1
3 1/4 64 73 1/64 4 1
4 1/8 512 585 1/512 8 1
5 1/t6 4096 4681 1/4096 16 1
6 1/32 32768 37449 1/32768 32 1
etc.

Table 8

Cumulative Total Moment Total


Order Length Number number fault area each fault moment
(n) (L,,) (N,,) (Nr) (a,,) (//.) (N..//.)
1 1 I I I 1 1
2 1/2 4 5 1 1/8 I/2
3 1/4 16 21 1 1/64 1/4
4 1/8 64 85 1 1/512 1/8
5 1/16 256 341 1 1/4096 1/16
6 1/32 1024 1365 1 1/32769 1/32
etc,

faulting. Again, the fault area required to produce the same deformation
using higher fault orders is greater and increases rapidly as fault
dimensions are reduced. The three-dimensional system can therefore
behave in a similar way to the two-dimensional systems described
earlier. If we apply the 50 percent rule (or any other constant ratio rule)
and only partially occupy possible higher-order fault sites as shown in
Table 8, we find an interesting result. A plot of log N r against.~", has a
slope of 2/3. Taking c to have a value of 1.5, the resulting b-value is 1.0.
This is the value observed most generally for earthquakes. The
square-based pyramid of Figure 16a is deliberately placed on its side.
This is the orientation that conforms to the ANDERSON (1951) fault
criterion. The square vertical plane corresponds to strike-slip faulting
releasing simple shear. The other two vertical planes release pure shear
as conjugate planes, and the conjugate planes at an angle to the vertical
are normal or reverse faulting. Insofar as the Anderson criterion is
obeyed and the generator pyramid is in the orientation shown, horizontal
sections through any fault system will exhibit three fault-strike
orientations.
792 GeoffreyKing

9. Fractal dimension of active faults and the origin of b-value

Using the foregoing discussions as a basis, we can now generalize the


relation o f fractal dimension (D), b-value, and the p r o p o r t i o n o f
deformation taken up on the main fault and higher orders o f faulting.
First we note the following relations.
1
N=-- (11)

where N is the n u m b e r o f subsegments associated with a change o f scale


r, and D is the fractal dimension. The expression is a r e s t a t e m e n t of
equation (10).
At any specified order n the length of each fault L,, is
L, = r ~ (12)
and the corresponding m o m e n t
/ / . eL(L.) d (13)
where d relates fault dimension to moment. For a two-dimensional fault d
= 2, and for three-dimensional faulting d = 3.
The total n u m b e r of faults up to a given order n is given by
Nr=l +N I+N 2+...N n (14)
= 1 + N~ + (NO 2 + ... (N~)"
since N = N~
N~-I
Nr = (15)
N--1

Am
~_-- for N~>I (16)
N--I
F r o m 11, 12 and 13 Am =.//-~n/a so that the total n u m b e r o f faults with
m o m e n t greater than or equal t o . / / , is

N T ~- (17)

D
logN r = - - - - log. ~" + constants (18)
d
Thus b y c o m p a r i s o n with equation (3) we find
D b
-- = (19)
d c
Accommodation of Large Strains in the Upper Lithosphere of the Earth 793

This result agrees with that of AKI (I 981) provided d = 3. This fractal
dimension is however that which is associated with faults active at any
given time and must not be confused with the fractal dimension of the
fault system produced over a period of time, which has a higher
dimension (see Section 7).
Equation (19) suggests that the b-value is determined by the ratio of
fractal dimension to the parameter d which, in turn, is determined by
whether the deformation is biaxial or triaxial. This result must be treated
with caution since the empirical constant c could be similarly related.
The total moment ( # r ) released by the fault system can also be
calculated. L e t / / r = 1 + . d where the main fault has a moment of 1
and the ratio of off-main fault to on-main fault deformation is .~, then:
//r = 1 + ,~ (20)

= 1 + (ra-) + (r~-) 2 + ... + (r a-D) (21)


1
- - - (22)
1 - r a-

i
~ = - - (23)
r - a - 1
1
1
Nr a
We find that . d is a function of (l/r), N and D. The values that these
may take are constrained by the geometry of fractal faulting, and the
remainder of this section is concerned with exploring what these
constraints may be.
For the fault model we have outlined it appears that (l/r) must equal
2, and for triaxial deformation d = 3. We assumed that N = 4 but any
integral value between 1 and 7 is apparently possible. Table 9 shows .~,
D and b-value for varying values of N.
Although choosing values for N not equal to 4 is possible, alternatives
are less satisfactory in certain respects. Any odd value of N gives more
Table 9

N ., D b

1 ~ o o
2 } 1 0.5
3 ] 1.58 0.79
4 1 2 1.0
5 1} 2.32 1.16
6 3 2.58 1.29
7 7 2.8 1.4
794 GeoffreyKing

secondary faults at one end of a main fault than at the other. Thus strict
self similarity is not followed. From the possible even values, N = 2 gives
a very low b-value, and N = 6 a high b-value. Thus unless fractional
values of (I/r) are possible, the geometry with N = 4, (l/r) = 2, D = 2
and b = 1 is the most plausible. The slip function has the property t h a t as
much deformation occurs off the main fault as on the main fault ( ~ ' =
1). In the two-dimensional examples the slip function was triangular and
obviously unrealistic. For three dimensions the form of the slip function
can be visualised by referring to Figure 17. The figure caption describes
possible analytic functions to approximate the slip and notes t h a t one
with a sine squared form gives , ~ = 1. This is a physically plausible slip
function.
Despite the pleasing simplicity of the N = 4, (I/r) = 2 system we
should note that b-value is an average of many events and under such
circumstances it is possible to relax the requirement that (l/r) a n d N be
integers. The same effects as giving them fractional values can be created
by missing out a certain proportion of faults at all scale dimensions.
This can be illustrated by considering a two-dimensional system with
.;~ = 1 that has r = (N = 2) and gives
.z/r= I+++-~+~+ .... (25)
We now miss out alternate orders
/'r=l +0++0+~6 + .... (26)

a)
stip /m'y

b
sip fY

...~
Figure 17. The a p p r o x i m a t e form of the slip function with , ~ ~ 1 for a s q u a r e fault p l a n e : (a)
A circular or elliptical form that does not extend to the corners of the p l a n e gives a ~ ~ 0.9. (b) A
function e x t e n d i n g to the corners can be a p p r o x i m a t e d as u = sin x sin y, which gives , ~ ~ 1,5 or
u = sin 2 x sin 2 y, which gives. ~ = 1.0.
Accommodation of Large Strains in the Upper Lithosphere of the Earth 795

and find .9~-- 0.5 and an effective r = ( N = 4). Repeated faulting, however,
can utilize the intermediate scales. The rule that was proposed earlier
that any fault can only move a second time if two faults move at the next
order (i.e., N = 2 always), or the equivalent at higher orders, is relaxed.
The new condition is that N and hence r can take any value to give the
correct slip distribution. Fractional values of N are to be interpreted as
the result of orders being randomly or systematically missed. The
long-term average gives the appropriate N and r. A corollary is that the
slip function on the fault will not be identical for each event.
It is interesting to note that in general most fractional values of N will
result in apparently irregular temporal behaviour at high but not low
orders of faulting. Regularity depends on N being an integer. This has
implications for earthquake repeat time. At low fault orders (i.e., near to
the upper fractat limit), earthquakes will recur at regular intervals if the
plate boundary motion is constant. At higher orders the behaviour will
apparently be irregular because regularity appears only at periods that
are long, compared to the repeat time of the main events.

10. Some other considerations in the distribution of secondary faulting:


The relation to fracture mechanics
In the earlier discussion it was noted that not all of the possible fault
positions need to be occupied to accommodate the deformation that
must occur as secondary faulting. The discussion then proceeded as if
the only consideration were to minimize total fault area or to satisfy a
simple geometric rule such as confining deformation to one sector of a
fault junction.
Another consideration is likely to be important, and a discussion of it
indicates how the geometrical description presented so far can be related
to ideas of engineering fracture mechanics (e.g., KNOTT, 1973; BROECK,
1974; LAWN and WmSHAW, 1975). We proceed from the suggestion that
a fault can move once without the creation of secondary faulting. Once
an appropriate amount of secondary faulting has occurred, the fault can
move again. For this to occur the secondary faulting must, in some
approximate sense, relieve the strain produced by the first fault motion.
An alternative view is to say that the secondary faulting must
accommodate the off-fault moment produced by the first fault increment.
Previously we implied that this could be accomplished by any
suitably-sized fault or sequence of faults in any position in the fractal
distribution. N o w we suggest that there is a spatial constraint on where
these can be.
Near a fault tip the elastic strains have amplitudes of the form:
K~
e= (27)
X/'2 zra
796 GeoffreyKing

where K E is a strain equivalent of the stress intensity factor and a is the


distance from the fault tip. All the strain components obey this
expression but with different azimuthal distributions. For our purposes
we will simply note that strain is inversely proportional to the square
root of distance from the fault end.
The strain can be thought of as moment per unit area in two
dimensions (moment per unit volume in three dimensions). This is the
moment that must be released by secondary faulting. The moment ~', to
be released in any specified radius a0 is therefore
Ot 0

~ a f e d A a a3o/2. (26)
0

From Figure 18 we can calculate the moment that may be released by


the largest faults within any given a 0. This is

/ / I ct ao2. (27)

The two functions are plotted in Figure 19. If we equate a 0 with the
zone within which secondary faulting occurs, then we see that as the
zone expands the available fault moment to relieve the applied stress
increases rapidly. An equilibrium between the stress intensity factor and
the area of faulting can clearly be achieved with faulting preferably
occurring in the region where the main fault repeatedly produces most
stress. The balance, however, will be more subtle than Figure 19 suggests
because, for any given radius, several scales of subfaulting will operate
and not simply the largest. Furthermore, the distribution will be modified
by the azimuthal distribution of strain. We should also note that after
many fault movements the distribution of strain (moment) to be relieved
may no longer be similar to that for a single fault in an elastic medium.
Nonetheless, the description provides a way in which to understand
some further aspects of fault behaviour. Some aftershock distributions do
show maxima in regions where the main fault modelled as a simple crack
predicts further deformation (STEm and LISOWSKI, 1983). It should be
noted that faulting can only match the applied strain within a 0 (critical) if
each fault has a larger than usual number of subsidiary faults (i.e., D >
2). There are simply not enough positions for big faults near the crack tip
to accommodate all the deformation by low-order faulting. Outside the
region of a 0 (critical) the reverse is true, and fewer than the usual number
of large faults are needed (i.e., D < 2). Thus the b-value associated with
relieving crack-tip strains that are approximately elastic in form will have
a b-value greater than 1 near the end of the crack and less than 1 well
away from it. In the next section we suggest that the temporal change of
b-value associated with the earthquake cycle can be associated with
spatial changes in earthquake distribution in relation to the main faults.
Accommodation of Large Strains in the Upper Lithosphere of the Earth 797

f ~
/ /\.
a/\

b,

cl / <
. /
Figure 18. The faulting within a distance a 0 of a main fault. In (a) four third-order faults are
included; in (b) four fourth-order faults, and in (c) four fifth-order faults. The deformation that
can be accommodated by the fractal set is proportional to a~.

11. Rupture termination and initiation: The geometrical nature of


barriers and asperities

The relations described in the last section enable us now to discuss the
way in which a fault triple junction evolves and to discuss how it can act
both as a barrier to terminate rupture and also create the asperities that
rupture to initiate the next earthquake (see AKI, 1983). We can at the
same time relate the fractal description of faulting to the engineering
concepts of fracture toughness, critical stress intensity factor and
Griffith's energy release rate.
Figure 20a shows a sequence of earthquake events on a set of faults.
One event is shown to trigger the next event in a sequence that
798 Geoffrey King

ao 2

8J a0
criticat
Figure t9. The deformation that may be accommodated by fractal faulting (~ aoz) and required
by crack tip deformation (~ a03/2). A radius a 0 (critical) always exists at which more large faults
exist than are required to relieve the crack tip strains. This radius will scale with the length of the
main fault.

propagates from left to right. The junction that creates the barrier for the
first rupture is also associated with creating the asperity or asperities
associated with initiating the second rupture, and so forth. We assume
for the discussion that the events are separated by a time interval
sufficient for us to talk about aftershocks, periods of quiescence and
foreshocks. The events however, could proceed rapidly one after another
to form a multiple event.
Figure 20a indicates the first rupture propagating away from Asperity
1 towards the triple junction associated with Barrier 1. When it reaches
the triple junction, rupture commences on one or both of the other two
fault branches (shown) and on their subsidiaries (not shown). The result
is simultaneous motion of sets of interlocking faults with a wide range of
scale sizes. The motion distributes deformation over a zone and locks the
large faults. The dimensions of the zone spread to the point where
dynamic rupture terminates.
The multiple-scale faulting behaviour is analogous to ductile
behaviour at a crack tip in an engineering material. The plastic zone
inhibits the extension of the crack and contributes to the fracture
toughness of the material. In a general way the zone over which
subsidiary faulting spreads can be equated with a plastic zone size and
the relation can be used to discuss the fracture toughness of a triple
junction.
KING and YIELDING (1983) discussed this concept in relation to the E1
Asnam, Algeria earthquake and noted that for an ideal elastic-plastic
Accommodation of Large Strains in the Upper Lithosphere of the Earth 799

material and a mode III crack, the crack tip plastic zone radius is given
by:

a = - (28)
8

where K is the stress intensity factor and ay is the yield stress at the edge
of the plastic zone (e.g., RuI~NICm, 1980). If K is the critical stress
intensity factor, Kcrit, then acrit is the critical plastic zone (sometimes
called process zone) size that inhibits rupture.
In Figure 19b as rupture starts to extend from the first rupture plane
to other planes the effective plastic zone size grows until rupture is
suppressed. The process is independent of scale since a will grow to
provide an acrit that prevents further rupture, and this will scale with the
length of the first fault. Thus the square of the critical stress intensity
factor of a fault is proportional to its length. Noting that the Griffith's
energy release rate is proportional to the square of the critical stress
intensity factor, we find that the Griffith's energy is proportional to fault
length. This is a corollary of the often-quoted statement that earthquake
faulting is associated with constant stress drop (for example, Scr~oLz,
1982) and can now be seen to have a simple geometrical reason.
The final stages of an earthquake are associated with simultaneous
motion on a set of different planes, and it is therefore to be expected that
the last part of the energy release will be different from earlier parts. If
the deformation is triaxial then it will not have a double couple character,
and this can explain why the long period behaviour of some earthquakes
has a clear nondouble couple component (DzmwoNsKI and
WOODHOUSZ, 1983). When the main rupture ceases, the "plastic zone"
size is gradually extended by aftershocks that represent continued
motion on many different planes, but in a more gradual way.
In Section 10 it was noted that near to the triple junction the
deformation will be associated with a high b-value and that this
decreases with distance from the junction. With time and tectonic
loading, the deformation will spread and increase a o so that late-stage
aftershocks may be expected to have a lower b-value than earlier ones.
The process associated with the termination of rupture involves
interlocking faulting, and the process disrupts the continuity of the larger
fault planes. For rupture to continue in the second direction, asperities in
these planes formed by the termination process of the previous event
must be broken. Before the tectonic loading has reached a sufficient level
to break asperities, it is reasonable to expect a period of quiet,
particularly near the future epicenter. The number of asperities is
greatest nearest to the barrier of the last event.
When faults start to move again, those with fewest asperities, which
800 Geoffrey King

a)
~ , Rupture3 .,~

A j#Barrier 2 ~ Barrier3
%.~~ Asperify3

I Barrier 1
..~ ,,..
Asperity 1

b) /

v-- - d'rup ure


"-k

Rupturel

\ \/

Figure 20. Event sequence on a fault system with fractal subfaults. The overall sequence is
shown in (a). Rupture initiates at asperity 1 and is stopped by barrier 1 associated with the first
fault junction. This same junction creates the asperities to be broken before the second event that
then propagates to the barrier associated with the next fault junction. The third event initiates
from this junction. The stopping of the first rupture is shown in (b). Motion on the first fault
causes motion on one or both the faults beyond the barrier. This, in turn, causes fractal faulting
in the zone around the junction. For clarity only, one of the orientations of the fractal faulting,
that parallel to the first main fault, is shown. The faulting produces a "plastic" zone sufficiently
large to stop further rupture. The process of forming the fractal plastic zone creates asperities in
the plane where the second rupture will occur. Attention is drawn to Figure 2 in AKI (1983).
There will be more asperities near to the barrier than far from it. Thus, as the tectonic load
increases (c), the first faulting to move will be faults parallel to, but at a distance from the
nucleation site of the second rupture. When the second rupture initiates (d), it m a y be expected
to start close to, but not at the fault junction that stopped the first event.

are those furthest from the triple junction, will move first, giving rise to a
Mogi doughnut (e.g., MOGI, 1981). Because of their spatial position
these events may be expected to have a low b-value.
The second main rupture occurs when one of these events lies close
Accommodation of Large Strains in the Upper Lithosphere of the Earth 801

c)
/
/
/ J

/,
/ /

d)
H
~J ~ initiation
_.~/ rupture
of
/' ~" \Small
/ptastic
\
\
\
\

enough to the second fault plane to trigger motion on it. The reason for
low b-value for foreshocks can also be seen from a probabilistic
argument as well as a spatial one. Each of the new events may trigger a
larger or a much larger event, and this is the opposite behaviour to the
termination process where each event will trigger smaller events. A
situation in which an infinitesimal event triggers a finite event has a
b-value of zero. If the triggering is likely and not certain and the events
are both of finite size, the b-value will be low but not zero. (See also
KING and YIELDING, 1983). The final process of rupture initiation is
shown in Figure 20d with the main rupture being initiated a small
distance from the junction and rupturing back towards it as well as
towards the next barrier.
Figure 20 is two dimensional, and the processes in real earthquakes
802 Geoffrey King

are apparently three dimensional. Thus it should be appreciated that the


systems shown in the figures can be complicated by normal or reverse
faulting. In other words, they can include components of motion
perpendicular to the plane of the paper. It is also emphasized again that
for clarity only a small number of the total number of faults taking part
in each stage of the process is shown in Figure 19. The processes are all
self similar and involve motion on fault systems with three orientations
(five in three dimensions). When one direction predominates, it is a
partial rather than complete predominance and only temporary.
Figure 20a also indicates that all the faults have equal lengths. Thus a
unilateral propagation from left to right is plausible. However, this will
not be the general situation. Figure 21 shows a system with faults of
different lengths. Beside each fault a moment (.A'0) is shown indicating
the deformation that it can accommodate in one increment of faulting.
Thus it may be seen that the segments b and c must move twice for
each single motion on a and e, and segment d must move four times.
Short segments whose lengths are comparable to the plastic deformation
zone size of an adjacent segment will not move independently of the
main segment but will move with the larger segment in a multiple event.
In fact, the main segments may more usefully be considered to be
composed of systems of long and short segments that have this property
of moving collectively. The frictional behaviour is then related to the
angular distributions and lengths of these segments. The behaviour of
straight faults or, more correctly, nearly straight faults is considered in
the next section.

12. Nearly straight faults with barriers and asperities

In the previous sections we have described geometries with large


angles between the fault systems. It was pointed out, however, that these

c. \ ,v
a, 5zR= 1/l,
\
Figure 2 I. The length of fault segments need not be equal. Their repeat time (if they are
two-dimensional) may be expected to be inversely proportional to their length. Thus the single
propagation process indicated by Figure 19 will not occur. Small fault segments may be con-
strained to move when a larger adjacent segment moves. If the plastic zone for one segment is
comparable in size to a shorter fault segment, then the short segment will never move
independently. A large fault system can be regarded as a set of small fault systems that have this
property of moving collectively.
Accommodationof Large Strains in the Upper Lithosphereof the Earth 803

systems were topologically identical to any other system, and the reader
is again invited to imagine distorting the preceding figures as if they were
drawn on rubber sheets. When figures are distorted in this way, fault
angles can be made small and circular ductile zones become elliptical.
The behaviour, however, remains the same. A bend in a fault causes
fractal behaviour whether the bend is large or small.
Strike-slip fault systems such as the San Andreas can be described by
distorted versions of the preceding figures. Thus the commonly observed
en echelon offsets are describable by one or more pairs of triple
junctions (such as those in Figures 22a and b) interacting in various
ways (Figures 22c and d).
Two implications of this view are noteworthy. The first is that the
fault-plane solutions from events on all the faults will be similar and thus
the existence of different planes will not be very obvious from the data.
The second is that since the secondary faulting results from the
combined effect of a series of subparallel lozenge-shaped zones, it will lie
close to the mean direction of the fault zone. The narrowness of the
earthquake belts in California is presumably the origin of the belief that
aftershocks lie on a single fault plane. The data, however, are not
inconsistent with deformation over a zone of definite width and with slip
vectors that are not parallel to the fault strike. The observed zone of
shattering associated with faults such as the San Andreas is to be
expected from the mechanical processes caused by triple junctions
associated with fault off-sets. The offsets in a strike-slip fault system can
operate to initiate and terminate rupture in the same way as the single
junctions described earlier.
Figure 23a shows the form that a system may take at any particular
time. Bear in mind that each junction of three faults is the center of a
fracture system of the types shown in Figure 11 or 15 extending to the
smallest scale sizes. For comparison, Figure 23b shows the assumed
orientation of fault planes for aftershocks of the Coyote Lake
earthquake of 1979 (REASENBERG and ELLSWORTH, 1982). Note that all
the events are shown to have planes oriented in the a- and b-directions
(of Figure 24a), but some could quite plausibly be in the c-direction.

13. Dilatancy and the innerfractal limit

In the discussion of fault development in Section 7 it was suggested


that a fault of any given order could move once without the need for
higher order faults to move; the deformation associated with one
increment of displacement could be accommodated elastically. Under
these circumstances, the inner fractal limit is elastic behaviour and the
scale at which it occurs is determined by the total amount of
804 Geoffrey King

B D

A F E

,t
~
--i"-\
)
C
~ c F
"~, b

Ductile zone

Combined ductile zone

Figure 22. An offset in a strike-slip fault may be regarded as being composed of two or more
triple junctions, each with its own "ductile zone" of fractal faulting. These zones can be elongate,
and systems of paired junctions can create long zones that behave in an apparently ductile
fashion to control fault rupture. If the angle between the faults marked (a) and (b) is small, then
all of the faults will give similar focal mechanisms. Strike-slip faults produce elongated ductile
zones and all the fault plane solutions are similar. For this reason the geometric features of the
barriers on strike-slip faults are less readily visible than regions where volume deformation is
more obvious.

deformation. The more deformed a region becomes, the more smaller-


scale faults are created. At a small enough scale the region remains
unfractured.
It was pointed out in Section 4 that triple junctions where three faults
meet are unconditionally unstable in completely rigid material, because
none of the three elements permits a volume Change. For limited motion,
elasticity can perform the role of stabilising fault junctions by allowing
volume change. At low confining pressure, however, or in the presence of
fluid under pressure fault surfaces can separate over small areas. Thus at
Accommodation of Large Strains in the Upper Lithosphere of the Earth 805

a)

b)

Figure 23. An idealized fractal fault offset geometry, (a), to compare with the aftershock
locations of REASEr~BERG and ELLSWOR~ (1982), (b). The planes are assumed to lie in the b-
and a-directions of (a), but some could lie in the c-direction. Some planes suggest dip-slip motion,
which is required if the system is to have a component of triaxial deformation.

this small scale, if one of the fault directions permits opening, stable
triple junctions can form and repeated motion can occur without
fracturing more material. This is shown in Figure 24 where the b
806 GeoffreyKing

direction is permitted to accommodate some opening. The figure m u s t be


understood in a schematic way. In Figure 24a and b, the two possible
positions of large-scale shear faulting are shown, and Figure 24c and d
indicate that at some scale b-direction fissures can open. Figures 24e and
f emphasize that this occurs in a volume. The b-direction fissures can
have both conservative and nonconservative motion simultaneously.
The situation may reasonably be expected to be more complex than
that shown with opening occurring either simultaneously along
differently oriented planes, or more probably in some sequence
controlled by the evolution of the large-scale deformation process.
When fault surfaces may separate, motion can occur without the
creation of more fault surface, but the deformation must be accompanied
by a net expansion. There is consequently a trade-off between the energy
of creating more fault surface and the elastic energy associated with
opening fissures. The latter energy can clearly be modified dramatically
by the presence of fluids under pressure, and, as we shall see, fluids can
modify the scale of the lower fractal limit.
In circumstances in which a volume increase must occur to permit
deformation, even simple shear boundary conditions will cause complex
behaviour. Thus the simple situation depicted in Figure 5b will never
occur and all boundaries will behave to some extent like more complex
boundaries. The complex behaviour of strike-slip faults m a y be
attributed to this effect. Thus biaxial deformation will never occur in
practice. All deformation will have a triaxial component, and all systems
will have long-term b-values close to unity. In a more general sense,
friction itself can be thought of as owing its origin to dilatant behaviour at
the lower fractal limit. It has been noted by a number of authors that at
the ductile-brittle transition, an absence of dilatancy and an equality of
frictional strength and fracture strength occur at the same confining
pressure (e.g., BRACE, 1977). The foregoing discussion provides a means
of visualizing why this should occur. When dilatant fissuring is
suppressed, an applied simple shear can be relieved by simple shear
structures alone. Thus the multiple direction deformation that creates
triple junction barriers and asperities is removed, and friction and
seismicity consequently disappear.
Looking again at Figure 24, it might seem that of the two alternative
processes for accommodating motion associated with c, neither is more
favourable. The system in Figure 24c however allows a greater volume
to be dilatant than that in Figure 24d. There is some evidence that on
strike-slip systems, off-fault deformation favours the outside rather than
the inside of fault bends (Figure 25). The opposite is evidently true
however for listric faulting. The hanging-wall block is invariably more
distorted than the footwall block (e.g., WERNICKE and BURCHFIELD,
1982; FROST and MARTIN, 1982). In this case, the presence of the
Accommodation of Large Strains in the Upper Lithosphere of the Earth 807

stress-free surface can explain the greater susceptibility of the upper


block to dilatant, and hence shear, deformation.

a) b)
/
/
/
a

c) d)
c
c.,~/

f)

Figure 24. Dilatant structures at the lower fractal limit of faulting. Diagrams (a) and (b) show
the two basic generator systems. At the lower fractal limit dilatant structures exploit the
b-direction. This is shown in a schematic way in (c) and (d) where it is also indicated that the
region behaves as if the boundary conditions were changed to include some dilation. Figures (e)
and (f) indicate that the dilatant fault structures can exploit the outside or inside of the fault
bend.
808 Geoffrey King

37"30' , ,

CAO&

CSC~
AJST

AJSS JAL
\
JOB& L ~. HSP

37"OO' AJRGJTGA
JEC~AJE
JRR
A.GW
OF HP
&HSL

&JPL
/~! A"Fe
~ H HPF
CZ
&HCO
HCl

HA2 HKR HLTA

H0L ,HOR
MONTEREYI HJG,A HSF k ~sA

HFPA

es~ A .8EH
20 m 8EMA
J
KILOMETERS

BPC BVL

36"30' I I ] 1 I
122"00' 121"30' 121"00'

Figure 25. Figures suggesting that deformation occurs on the outside-of bends in strike-slip
fault systems (look along the faults to see the bends). The upper figure (a), is the aftershock zone
for the Coyote Lake, California, Earthquake of August 6, 1979 (REASENBERGand ELLSWORTH,
1982), and the figure (b), on the opposite page is the aftershock zone for the Bear Valley,
California, Earthquake sequence of February and March 1972 (ELeswoRxn, 1972).

14. The inner fractal limit and b-value

At the inner fractal limit the b-value must change. In tectonic


environments, this has not been observed. Nor has it been observed in
laboratory experiments (MOGI, 1962; SCHOLZ, 1968). This is presumably
because in both cases the scale of faulting at the fractal limit is too small
to be detected. Under volcanic conditions, this is not the case. Volcanic
Accommodation of Large Strains in the Upper Lithosphere of the Earth 809

",,, . .\ -. . .-" 36 50' _

",,, ,, .-:.L..: .... "--~, -.

"' , , i :':,

'~,'~ ""'~,~-L" ".: "~.-\\


..:.. ~%~...-.~-~.~..~......~.~ -

91111970 to 05:27 2122119"72 )~. .:." "


0
E55~EE~E3
5 KM
\ "\
j6 20' \,

Figure 25(b).

earthquakes behave in a completely different fashion and exhibit great


regularity of behaviour and characteristic scale lengths. In the volcanic
environment, large dilatant fissures in the form of magma or gas-filled
dikes are possible, and regular nonfractal geometries are possible.
The behaviour is discussed in papers by OKADA et al. (1981) and
TAKEO (1983), who report observations of two types of earthquake
families associated with the Usu volcano. One family has repeating
events of identical size and waveform, and the other has varying size and
identical waveform. The way in which events of this kind can come
810 Geoffrey King

about is shown in Figure 26. Shear faults on which the earthquakes


occur end at open fissures and not at fault junctions where fractal
smashing would occur to create smaller events. A system that produces
events of changing size is shown in Figure 26a and one that produces
events of the same size in Figure 26b. Many different geometries having
similar seismic properties to those illustrated are possible. A system
similar to Figure 24a is described from field observations by SEGALL and
POLLARD (1983) in an exhumed granite terrain in the Sierra Nevada.

a)

b)

,,g\~.x.~.

Figure 26. Repeating events on faults that terminate on fluid filled fissures. The upper diagram
(a) and (b), is for a fault ending on two short fissures. The fissures grow significantly with each
fault displacement resulting in a change in behaviour. The fault length remains unchanged; only
the amount of displacement varies with each event causing events of identical waveform but
different amplitude. The events are not self similar because the displacement to length ratio of
faulting changes with each event. The lower diagram (b), is for longer fissures. The change of
fissure length is unimportant and each event can be identical. Note that the foregoing are only
examples. The fluid fissures allow the faults to terminate without the fractal subfaulting that
produces self-similar behaviour. Many geometries can be stabilized to produce nonfractal
behaviour if dilatant structures are allowed. Similar behaviour may be expected from dip-slip
faults that cut through the brittle crust from the stress free ductile zone to the Earth's stress free
surface.
Accommodation of Large Strains in the Upper Lithosphere of the Earth 811

The features represented as dykes in Figure 25 could also be provided by


the Earth's stress free surface (e.g., TAKEO, 1983) and the base of the
brittle zone. Thus some variations in the fractal behaviour of
earthquakes and faulting may be expected at such scale dimensions. This
may explain the observations of AKI (1983) that earthquake spectra and
seismic scattering appear to be associated with structures having scales
of the order of the thickness of the brittle crust.

15. Problems with the formation offractal fault systems

Throughout our discussion it has been assumed that fault surfaces will
automatically form in the appropriate directions to accommodate the
applied finite deformation. Field observation does not contradict this
assumption. Examination of road cuts such as that shown in Figuire 1
commonly show simple planar surfaces except where human activity to
create the road cut has produced curved surfaces.
Our understanding of fracture mechanics, however, suggests that only
purely tensile fractures form in a planar fashion, and all other fractures
will curve (KNOTT, 1973; BROEK, 1974; LAWN and WmSHAW, 1975).
Man-made features in road cuts generally show the predicted curved
geometry. To explain planar features, SE~ALL and POLLARD (1983)
argue convincingly that the early stages in the development of simple
fault systems in the Sierra Nevada rely on the formation of new
structures initially in tension, with the structures only later being
exploited as shear planes. Some of their more complex zones, however,
appear to have geometries similar to those shown here (such as Figure
23), but they are unable to demonstrate how the more complex systems
formed. The discussion in this paper describes how deformation can
proceed but does not explain how features come into existence. Although
it seems likely that the faults in Segall and Pollard's simple systems
existed by chance formation in an earlier stress system, it is difficult to
avoid the conclusion that, to explain the more complex zones, some
process exists to form planar faults in an environment that is subject to
shear on a large scale.

16. Conclusions

The ideas presented here are, in effect, an extension of plate tectonics


to the general problem of the finite deformation of brittle elastic solids.
Finite motion is taken up on sets of planes with differing orientations.
Because these planes cannot intersect and have nonzero displacements,
intersections are associated with multiple faulting. This is described in
terms of a fractal geometry that allows faulting at one scale to be related
812 GeoffreyKing

to faulting at the next smaller scale. Two fundamental constraints are


placed on the fault systems. One is purely the geometric constraint that
faults cannot cross, and the second-is the constraint that sufficient
faults must exist to accommodate the applied deformation. The
latter constraint is more severe than the former. Fewer faults are
required at any phase of deformation than those that exist in a
theoretical pattern of nonintersecting faults, or than those that can come
into existence over a period of time. A system of active faults therefore
can have a lower fractal dimension than the system of fossil faults that a
period of activity will leave.
On the basis of some simple assumptions the fractal dimension of an
active fault system can be directly related to earthquake b-value. For a
triaxial system with a fractal dimension of 2 the b-value is 1.0. For
biaxial deformation with a fractal dimension of 1, the b-value is less than
this. Variations in b-value can therefore be attributed to variations in
deformation type, but it is argued that because of dilatancy at the lower
fractal limit all systems will behave in a triaxial fashion. Fault systems
evolve with time, and it is possible to show that this evolution will result
in b-value variations within a long-term average of unity. Within the
same context, other features of the observed seismic and geometric
behaviour o f fault systems can be explained and their relation to fracture
mechanical concepts indicated. The stopping processes involve motion
on planes with different orientations that spread deformation over a
volume. If this deformation is triaxial it will contribute a nondouble
couple component to the earthquake mechanism. The stopping processes
create the asperities that must break before the next main event initiates.
An important consequence of the behaviour described here is that the
fracture toughness of rocks subject to finite deformation is not a material
constant or even a quasi constant. At the most basic level this is because
for fracture toughness to be a natural constant requires that the material
involved have a built-in length scale. The self-similar behaviour of
earthquakes shows that the earth does not have this property. This result
will actually apply to the deformation of all solids, and similar
conclusions emerge from rock testing (WONG, 1982). The behaviour,
however, is more evident in the deformation of the Earth's crust where
the boundary conditions demand finite strain. In engineering situations
"large strains" result in "catastrophy", and instability conditions
associated with stress boundary conditions are more important.
The use of a Fractal description denies the use of calculus and hence
calculus-based concepts such as stress and strain. The significance of
this cannot be overemphasized. Calculus can be employed at the upper
and lower fractal limit but the "patching" together of fractal and
continuous deformation descriptions has to be an approximate process,
because at some scale level the assumptions of discontinuous and
Accommodation of Large Strains in the Upper Lithosphere of the Earth 813

continuous behaviour are contradictory. At the scale at which this


problem exists no alternative seems to be available except to examine
faults as they actually appear in the field and to explain their behaviour
on the basis of their observed geometry. The fractal description does
offer some support to the system of analysing Neotectonic deformation
adopted by some workers (e.g., ARMUOUet aL, 1982). The system uses a
large number of measurements of fault slickenside directions to calculate
the causal stress orientation. The self-similar behaviour of faulting
described here suggests that random sampling of faults over a wide
range of scales can be an acceptable process. The deformation behaviour
we describe, however, suggests that the quantities employed in
Neotectonic discussions should be redefined as strains rather than
stresses. Also the problems associated with approximating a fractal
system by a continuum demand that the procedure contain inherent
errors.
Throughout the paper we have not specifically considered rotations.
These will, in general, have amplitudes similar to the effective strains that
the mukiple fault systems relieve.

Acknowledgments

I would like to thank Keiiti Aki for inspiration and for supporting my
visit to M.I.T. I would also like to thank Bob Wallace who assisted me in
a visit to the U.S. Geological Survey where much of this paper was
written. John Beavan, Keiiti Aki and Robert MacDowell read the
manuscript, and I greatly appreciate their advice and comments. I
particularly appreciate careful reviews that have been helpful and
stimulating by Joe Andrews, Mike Ashby and Charlie Sammis. I have
enjoyed discussions with Earl Williams, and I thank Ann Rayner for
typing the manuscript.
The work has been supported by the Royal Society, the Natural
Environment Research Council (GR3/3904), the National Science
Foundation (CEE 8206456), and the United States Geological Survey.
This is Cambridge Earth Sciences Contribution number 446.
This paper is dedicated to the memory of Raffy Freund, a geologist of
great vision.

REFERENCES
AKI, K. (1979), Characterization of barriers on an earthquake fault, J. Geophys. Res. 84, B 11,
6140-6147.
AKI, K., A probabilistic synthesis of precursory phenomena, In Earthquake Prediction--An
International Review (Maurice Ewing Series 4, American Geophysical Union,"Washington,
D.C. 1981).
814 Geoffrey King

AKI, K. (1983), The use of a physical model of fault mechanics for earthquake prediction
(Preprint). Lecture delivered at U.S.G.S., Menlo Park, California.
ANDERSON, E. M., The Dynamics of Faulting (Oliver and Boyd, Ltd., London 1951).
ARMIaOU, R., CAaEY, E., and CISTERNAS, A. (1982), The inverse problem in microtectonics and
the separation of tectonic phases, Tectonophysics 82, 145-160.
ANDREWS, D. J. (1980a), A stochastic fault model 1. static case, J. Geophys. Res. 85, B7,
3867-3877.
ANDREWS, D. J. (1980b), A stochastic fault model 2. time-dependent case, J. Geophys. Res. 86,
B l l , 10821-10834.
BABCOCK, E. A. (1973), Regionaljointing in southern Alberta, Canada, J. Earth Sci. 10, 1769-
1781.
BRACE, W. F. (1977), Volume changes during fracture and friction: a review, Proceedings of
Conference II. Experimental Studies of Rock Friction. United States Geological Survey,
Office of Earthquake Studies, Menlo Pk., California 94025.
BROEK, D., Elementary Engineering Fracture Mechanics Martinus Nijhoff 1982, Hague.
BUaCHFIELD, B. C. (1983), The continental crust, Scientific American 249, 3, 130-142.
DE SITtEr, L. U., Structural Geology (McGraw-Hill, 1956).
DZIEWONSKI, A. M., and WOODHOUSE, J., (1983), An experiment in systematic study of global
seismicity: centroid-moment tensor solutions for 201 moderate and large earthquakes of
1981, J. Geophys. Res. 88, B4, 3247-3271.
ELLSWORTH, W. (1975), Bear Valley, California, earthquake sequence of February-March
1972, Bull. Seismol. Soc. Am. 65, 483-506.
FROST, G. F., and MARTIN, "D. L., Anderson-Hamilton Volume: Mesozoic-Cenozioc tectonic
evolution of the Colorado River region, California, Arizona, and Nevada (Cordilleran
Publishers. 6203 Lake Alturas Ave., San Diego, California 92119, 1982).
GAGNEPAIN-BEYNEIX, J., HAESLER, H., and MODIANO, T., (1982), The Pyrenean earthquake of
February 29, 1980: an example of complex faulting, Tectonophysics 85,273-290.
GUTENBURG, B., and RICHTER, C. F., Seismicity of the Earth and Associated Phenomena
(Princeton University Press, 1949).
HILL, D. P., (1982), Contemporary Block Tectonics: California and Nevada, J. Geophys. Res.
87, BT, 5433-5450.
HODGSON, R. A. (1961), Regional study of jointing in Comb Ridge-Navaja Mountain area,
Arizona and Utah, Bull. American Assoc. Petrol. Geologist 45. 1, 1-38.
ISHIMOTO, M., and IDA, K. (1939), Observations sur les s~isms enregistr& par le
microseismograph construit derni~rement (1), Bull. Earthq. Res. Inst. 17, 443-478.
JACKSON, J., and MCKENZIE, D. (1983), Active tectonics of the Alpine-Himalayan Belt between
Western Turkey and Pakistan, Geophys. J. R. Astra Soc. Lond. (submitted).
KANAMORI, H., and ANDERSON, D. (1975), Theoretical basis for some empirical relations in
seis,nology, Bull. Seism. Soc. Am. 65, 5, 1073-1095.
KAGAN, Y. Y. (1982), Stochastic model of earthquake fault geometry, Geophys. J.R. Astr. Soc.
Lond. 71,659-691.
KING, G. C. P., (I 978), Geological faults:fracture, creep and strabT, Phil. Trans. R. Soc. Lond.
A, 197-212.
KING, G. C. P., and YIELDING, G. (1983), The evolution of a thrust fault system: processes of
rupture initiation, propagation and termination in the 1980 El Asnam (Algeria) earthquake,
Geophys. J.R. Astr. Soc. 77, 915-933.
KNOTT, J. F.. Fundamentals of Fracture Mechanics (Wiley, New York. 1973).
LAWN, B. R-, and WlLSHAW, T. R., Fracture of Brittle Solids, (Cambridge University Press,
1975).
MANDELBROT, B. B., The Fractal Geometry of Nature (W. H. Freeman and Company, San
Francisco. 1977).
MCKENZIE. D, P., and MORGAN, W. J. (1969), Evolution of triple junctions, Nature 224, 5215,
125-133.
Accommodation of Large Strains in the Upper Lithosphere of the Earth 815

MCKENZlE, D., (1979), Finite deformation during fluid flow, Geophys, J.R. Astr. Soc., 58.
689-715.
MCKENZIE, D., and JACKSON, J. (1983), The relationship between strain rates, crustal
thickening, paleomagnetism, finite strain and fault movements within a deforming zone,
Earth Plant. Sci. Lett. 65, 182-202.
MCKENzm, D. (1972), Active tectonics of the Mediterranean region, Geophys. J.R. Astr. Soc.
30, 109-185.
MOGI, K., (1962), Study of the elastic shocks caused by the fracture of heterogeneous materials
and its relation to the earthquake phenomena, Bull. Earthq. Res. Inst. 40, 125-173.
MOGI, K., Seismicity in western Japan and long-term earthquake forecasting, In Earthquake
Prediction, Maurice Ewing Series 4. (ed. Simpson, D. W., and Richards, P. G.) (A.G.U.,
Washington, D.C., 1981)pp. 635-666.
MOLNAR, P. (1983), Average regional strain due to slip on numerous faults of different
orientations, .t. Geophys. Res. 88, B8, 6373-6394.
OKADA, H., WATANABE, H., YAMASHITA, H., and YOKOYAMA, I. (1981), Seismological
significance of the 1977-1978 eruptions and the magma intrusion process of Usu volcano,
Hokkaido, Volcanol. Geotherm. Res. 9, 311-334.
OVERBEY, W. K., and ROUGH, R. L. (1968), Surface studies predict orientation of induced
formation fractures, Producers Monthly 32, 16-19.
REASENBERG, P., and ELLSWORTH, W. L. (1982), Aftershocks of the Coyote Lake, California,
earthquake of August 6, 1979: a detailed stud),, J. Geophys. Res. 87, B 13 10,637-10,655.
RUDNICKI, J. W. (1980), Fracture mechanics applied to the Earth's crust, Ann. Rev. Earth
Plant. Sci. 8,489-525.
SCHOLZ, C. H. (1968), The frequency magnitude of microfracturing in rock and its relation to
earthquakes, Bull. Seism. Am. 58, 399-416.
SCHOLZ, C. H. (1982), Sealing laws for large earthquakes: consequences for physical models,
Bull, Seism. Soc. Am.-72, 1, 1-14.
SEGALL, P., and POLLARD, D. (1983), Nucleation and growth of strike-slip faults in granite, J.
Geophys. Res. 88,555-568.
STEIN, R., and LxSOWSKI, M. (1983), The 1979 Homestead Valley Earthquake Sequence,
California: Control of Aftershocks and Postseismic Deformation, J. Geophys. Res. 88, B8,
6477-6490.
TAKEO, M. (1983), Source mechanisms of Usu volcano, Japan, earthquakes and their tectonic
implications, Phys. Earth and Planet Int. 32, 241-264.
TAPPONNIER, P., and MOLNAR, P. (1976), Slip-line field theory and large-scale continental
tectonics, Nature 264, 319-324.
TCHALENKO, J. S. (1970), Similarities between shear zones of different magn#udes, Bull. Geol.
Soc. Am. 81, 1625-1640.
THOMAS, P. G., and MASSON, P. (1983), Tectonic evolution of Mercury; comparison with the
moon, Annales Geophysicae 1, 53-58.
VON MISES, L. Z. (1928), Mechanik der plastischen Formh'nderung yon Kristallen, Zeitschrift
ffir Angewandte Mathematik und Mechanik 8, 161-185.
WERNICKE, B., and BURCHFIELD, B. C. (1982), Modes of extensional tectonics, J. Struct. Geol.
4, 2, 105-115.
WONG, T. (1982), Shear fracture energy of Westerly Granite from post-failure behaviour, J.
Geophys. Res. 87, B2, 990-1000.
WYSs, M. (1973), Towards a physical understanding of the earthquake frequency distribution,
Geophys. J.R. Astr. Soc. 31,341-359.

(Received 15th November 1983, revised 1 lth April 1984, accepted 14th April 1984)

You might also like