You are on page 1of 20

WATER RESOURCES RESEARCH, VOL. 29, NO.

4, PAGES 775-794, APRIL 1993

PercolationTheory and Its Applicationto GroundwaterHydrology


BRIAN BERKOWITZ 1

HydrologicalService,Ministry.of Agriculture,Jerusalem,Lrael
ISAAC BALBERG

Racah Instituteof Physics,Hebrew Universi.ty,


Jerusalem,Israel

Thetheoryof percolation,
originally
proposed
over30yearsagoto describe
flowphenomena
in

porous
media,
hasundergone
enormous
development
inrecent
years,
primarily
inthefieldofphysics.
Theprincipal
advantage
ofpercolation
theory
isthatit provides
universal
lawswhich
determine
the
geometrical
andphysical
properties
ofthesystem.
Thissurvey
discusses
developments
andresults
in
percolation
theory
todate,
andidentifies
aspects
relevant
toproblems
ingroundwater
hydrology.
The
methods
of percolation
theoryarediscussed,
previous
applications
of thetheoryto hydrological
problems
arereviewed,
andfuturedirections
forstudyaresuggested.
1.

andpotential
applications
to groundwater
hydrology.
It is
hoped
thatthisreview
willproveto bea useful
contribution

INTRODUCTION

Thetheoryof percolation
wasfirstintroduced
sometime both as a serviceto researchersactive in the field of

hydrology,
andto theadvancement
of relevant
aspects
of
.ago
to describe
polymerization
[Flory,1941;Stockmayer,
percolation
theory.
1943]
andpenetration
of fluidsin porous
media[Broadbent
asfollows:in thenextsection,
andHammersley,
1957],and the subjecthas sincebeen Thepaperis organized
elements
of percolation
theoryareexplained,
and
intensively
studied,
primarily
in thefieldof physics
[e.g., essential
some
of
the
principal
results
pertaining
to
hydrology
and
Stauffer,
1985;
Balberg,
1987,andreferences
therein].
Howintegral
to
the
discussion
in
the
remainder
of
the
paper
are
ever,
littledirectuseofpercolation
theoryresults
hasbeen
Outstanding
problems
in groundwater
hydrolmadeto datein the field of hydrology[Thompson
et al., summarized.
ogy
are
then
briefly
surveyed,
followed
by
a
review
of
1987].
The theoryhasbeenextensively
developed
as a
existing
and
potential
applications
of
percolation
theory
to
branch
of statistical
physics,andhasfoundsuccessful
ap-

plications
ina diverse
range
ofproblems
including
design
of

problemsin hydrology.

electronic
andmagnetic
materials,
conceptualization
ofgeo-

metrical
andtopological
characteristics
ofporous
media,
and
understanding
of miscible
andimmiscible
displacements
in
disordered media.

2.

ELEMENTS
OF PERCOLATION
THEORY

2.1. What Is PercolationAll About?

Theprincipal
advantage
of percolation
theoryis thatit
Manyproperties
ofa macroscopic
system
areessentially
provides
universal
lawswhichdetermine
thegeometrical
determined
bytheconnectivity
ofthesystem
elements.
The
andphysical
properties
of a system.
These
lawsareinde- special
properties
ofasystem,
which
emerge
attheonset
of
pendent
ofthelocalgeometry
orconfiguration
ofthesystem. macroscopic
connectivity
withinit, areknown
aspercolaInparticular,
manytransport
processes
canbesuccessfully
tionphenomena.
Toillustrate
thephenomena,
andthebasic
understood
by considering
an idealized
transport
of an concepts
usedin percolation
theoryto characterize
the
abstract
fluidthrough
anabstract
medium.
If a fluidflows connectivity,
consider
thesimplest
modelof an irrigation
through
a medium
whichisitselfin some
sense
disordered
network of a "Texas farm."
(orrandom),
theflowthrough
thesystem
maybedescribed Let the "almostinfinite"irrigationsystemconsistof a
by a so-calledpercolationprocess.

square
network
ofpipes,
withthewater
supply
provided
by

Several
advances
in theapplication
ofpercolation
theory a riverlocated
atthenorthendof thefarm,asillustrated
in

fortheunderstanding
ofproblems
inporous
media
[HalperinFigure
!. When
some
ofthepipesegments
become
clogged,
etal., 1985]havebeenmadein thelastfewyears,with theintensity
of waterflowat thesouthern
endof thefarm
evidence
thatsomepercolation
models
canaccount
for decreases.
Thequestion
thenarisesastothepossibility
of
geometrical
properties
andtransport
phenomena
observed
in determining,
atthesouthern
endofthefarm,thenumber
of
porous
rocks[e.g.,Jerauld
et al., 1984a;
Balberg,
1986a; clogged
pipesegment
by monitoring
thewaterflow.In
Gueguen
andDienes,
1989;
Balberg
etal.,1991].
Inviewof particular,
it isofinterest
toknowthenumber
ofsegments
these
many
recent
developments,
it seems
prudent
toreviewthatmustbeclogged
(assuming,
of course,
thattheclogged
thestate
oftheartofpercolation
theory,
emphasizing
known segments
aredistributed
randomly)
inorderfornowaterto
arriveat the southern
end.The answerto the second
is givenbywhatis known
asthepercolation
IOnleave
atDepartment
ofGeological
Sciences,
University
of question
British
Columbia,
Vancouver,
BritishColumbia,
Canada.
threshold.
If thenumber
of unclogged
segments
isdenoted

byN, andtheir
number
atthethreshold
isNo,onecanshow
[Balberg,
1987;
Wong
etal.,1984]
thatthevolumetric
water
flow,Q, willbedetermined
bya power
lawoftheform

Copyright
1993bytheAmerican
Geophysical
Union.
Papernumber92WR02707.
(B43-1397/93/92WR-02707505.00
775

776

BERKOWITZ
AND]ALBERG:
PERCOLATION
THEORY

occupied
site,whileanintersection
closedto flowis repreI

rtinern

End of Farr

sentedby anunoccupied
site.Correspondingly,
two nearest
neighborsites are said to be "bound" if they are both
occupied.Any two sitesare called"connected"if thereisa
continuous
path of boundsitesbetweenthem. It is obvious

thata connected
pathallowstheflowof fluidor electricity
fromoneendof the systemto another.Hence the relationof
connectivityof the (so-calledmicroscopic)elementsof the
systemto the physicalpropertiesof the entire (or so-called
macroscopic)
system.In the followingsubsections,
thebasic

mathematicalconceptsof perco!ation theory are introduced

and quantifiedusing a rather abstract approach.At each


stageof the discussion,however, one can translate these
conceptsinto a concreteproblemof interest, suchastheone

Southern

illustrated in Figure 1.
End

of

[Zarm

2.2. FundamentalConceptsUsedin PercolationTheory


Fig. !. A possibleschematicillustrationof a "Texas irrigation
network" of pipes in which water flows (segments)andjunctions
(solid circles). The clogged pipes are represented here by the
missingsegments.Mathematically, one speaksof a squarelattice
with bonds (segments)and sites (solid circles).

Several simplegeometricaland statistical conceptsare


usedin percolationtheoryto quantitativelydescribesystems

suchas those introducedin the previous section (for a more


extended review, see Stauffer [1985]). To present these
concepts, consider the so-called square lattice shownin
Figure2. Eachpointon the latticerepresentsa sitethatmay
Qo(N - Nc) K
(1) be "occupied" or left "unoccupied." A basic assumption
of
percolationtheory is that the site occupawhere cis someexponentwhich can be found from theory (interaction-free)
and/orcomputersimulationand/orexperiment.Sucha sim- tion processis random, so that the occupation of one siteis
ple law holds for N relatively close to Nc (typically, for independentof occupation of its neighbors or of any other
N/Nc -< 2). For muchlarger N, otherconsiderations
canbe site. Of primary interest is characterization of the system
under a given fraction of occupied sites. For example,
applied which in many caseslead to other power laws.
The problemdescribedabove, and the simplelaw givenby assumethat there are No sitesin the two-dimensionalsquare
(1) that governsits solution,apply to many kindsof systems, lattice, with N of them occupied,as shown in Figure 2. Since
and as such, are much more generalthan one mightrealize at the occupationprocessis random, the probability of each
first glance. The above formulation can be used, for exam- site'sbeingoccupiedis simplyp = N/No. This "occupation
ple, to describeflow of water in porous rocks, flow of probability" is well defined in the No --> c limit. Unless
electricity in electrical networks and compositematerials, mentionedotherwise, it is assumedthroughout this paper
that the lattice is infinite.
penetration of termites in wood, and the spread of fire or
The second concept to be introduced is that of the
epidemics.In a percolation problem, one asks three basic
"cluster."
As mentionedabove, two occupiedsitesthatare
questions:(1) What is the geometricalor physicalproperty
that is relevant to the connectivity of the system under nearestneighborsare consideredbound, while two occupied
investigation(i.e., some property Q)?, (2) What is the sitesbetweenwhichthereis a continuouspath of boundsites
thresholdfor percolation(i.e., N c or an equivalentquanti- are consideredconnected.A group of connectedsitesis
ty)?, and (3) What is the exponentthat describesthe behav- called a cluster. Various clusters with one, two, and three
ior of Q near N c (i.e., K)?The interestand usefulness
of membersare illustratedin Figure 2. The questionthenarises
percolationtheory arise from the fact that many systems regardingthe number of clusters of each kind that canbe

have the same c.This rneansthat by findingcfor a very


simpleandsolvablemodel,onecanpredictthe valueof cfor
a very complicated system that one does not know how to
solve.This centralproperty of percolationtheoryis known

asuniversality,andis a resultof thefactthatthepercolation

O0 0 0 0 I 0
0

,,

transition (i.e., the transition between a connected and a

disconnected
network) is a second-orderphasetransition
(suchasthe well-studiedliquid-gasor magnetictransitions).
On the otherhand, the percolationthresholdmustbe determinedseparatelyfor each system,althoughsomegeneral
guidelinesfor its determination are available.

To becomefamiliar with and to quantifythe mostfundamentalconceptsofpercolationtheory,the networkshownin


FigureI is nowmodifiedto be a "latticesite" model,whereby rather than pipe segmentclogging,the systemcan be-

comecloggedonly at the intersections


of the pipes.Hence,
as shown in Figure 2, consider a mathematicallattice such

that an intersectionopen to flow is represented


by an

o o c o o
o

oooo

O0

Fig. 2. A square
lattice,in whichoccupied
sitesformone-site
clusters,
thetwokindsof two-site
clusters,
andthesixkinds
of
three-site clusters.

BERKOWITZ
ANDBALBERG:
PERCOLATION
THEORY

777

sites, the common definition of the average finite cluster


size, S, is the averageclustersize per occupiedsite,
C

S= E' nsS2/p
o

Since(as shownabove)ns is a polynomialfunctionof p, S


canbe expressedas a power seriesofp. For example,in the

squarelattice with the expressionsgiven above for ns (s =


1, 2, 3), one can easily (althoughwith lengthy calculation)

00I 00
C 0 I- 0 0

(3)

findthatS = I + 4p + 12p2 + 24p3 + 52p4+ 108p5+


224p6 - 412p7 + 844p8 + 1528p9 + -.-. Following
the

generalexpectation(see (1)) of a power law behavior, and


the particular findings in simple cases (e.g., the one-

Fig. 3. An illustrationof a finite latticewith varioussizeclusters.There is one cluster, the percolation cluster, that connectsthe
bottomand top boundaries,and the left and right boundaries,of the

finitesquarelattice.

dimensional
latticewherens = pS(1- p)2), oneexpects
that, nearPc, it is possibleto approximatea power series
suchas the one shown above for S by a power law of the
form

foundwhen the lattice occupationprobability is p. To have


a one-cluster,there must be an occupied site (probabilityp)

S:r(p c - p) -Y

(4)

where T is an exponent.Indeed, such an approximation

surrounded
byfourunoccupied
sites(probability
(1 - p)4). (e.g., usingthe Pad approximantsmethod[seeAlon et al.,
Hence,
thenumberof one-clusters
will be N0p(1 - p)4. 1990]) yields for y the same value as that obtained by
Similarly,for two-clusters,the number of such "left-right"

computersimulationtechniques.It is found that ? is inde-

clusters
isN0p2(1- p)6, andthereisthesamenumber
of pendent of the particular lattice under consideration,and
"up-down"clusters.For the limit No --> o, onemay define dependsonly on the dimensionalityof the system.In particthenumberof clustersper site, ns, wheres is the numberof ular, one finds that = 43/18 for two-dimensionalsystems,
skesin the cluster (the "size" of the cluster). Thus, it andthat ,/ 1.82for three-dimensionalsystems.In contrast,

follows
(seeFigure2) thatn1 -' P(1 - p)4, n2 = 2p2(1- the valueof p c dependson the particularlattice, but again
p)6,andn3= 2p3(1- p)8 +4p2(1_ p)7.Thecalculationthe Pc valuedeterminedby expressingthe series(e.g., (3))
ofnsfor largervaluesof s, for otherlattices,andfor higher with an asymptoticform for S {e.g., (4)) is the sameas that
for the samelattice(e.g., for
dimensions,
becomesincreasinglycomplicated,but the ba- foundby computersimulations
the squarelattice,p c = 0.59). Tables of values of p c and
sic"recipe" is apparentfrom the above example.
Followingthe above examples,it is also clear that with critical exponentsfor commontwo- and three-dimensional
increasing
p, the probability of finding larger clustersin- latticescan be found, for example, in the works by Stauffer
creases.In a finite lattice such as the one shown in Figure 3,

[ 1985]andAharony[ 1986].Note that for p > p c, the average

therewill be a p valuethatis sufficientlylargeto ensurethat cluster size is infinite, but the definitions given here are
atleastone cluster connectsthe "bottom" and "top" (or the

concernedwith finite clusters, and (4) describes their diver-

"left" and "right") ends of the lattice. This value of p, genceasp --->p c. It hasbeenfound, however,that the same
S for p > Pc, and thus(4) holdsin this
whichis well defined(say, in a computerexperiment)when /value describes
No-->m, is calledthe "critical occupation
probability,"and regime,exceptthat one has to replace(Pc - P) by (p -

isdenoted
pc. Thecorresponding
largestclusteris calledthe
percolation
cluster, and the value of Pc is known as the

So far, the statisticalpropertiesof the clustershave been

latticesitepercolationthreshold.The regime0 -< p < p c is considered.The next property to be consideredis the


extentof the cluster.Supposethat the sitesof a
calledtheregimebelowthe percolationthreshold,whilethe geometrical
regime
pc < P -< 1 is calledtheregimeabovethepercolation cluster of size s are located at r i, where i = 1, 2, .--, s.
threshold.
One can generallyarguethat thereis at mostone The centerof the clustercanbe definedby r0 = i ri/s, and
percolation
cluster[e.g., Stauffer,1985];this is obviousfor thus the so-calledradius of gyration of the cluster, Rs, is
theparticulartwo-dimensionalcase illustratedin Figure3. defined by
Abovethe percolationthreshold(i.e., for p > p c), onecan

alsodetermine
the probability
per site,P, thatan occupied
i
sitebelongs
to the percolation
cluster.SinceY.'snss,where
's isthesumoverall finitevaluesof s, istheprobability
per
The average(over all finite clusters)radiusof gyrationper
sitethatthesiteis occupied
andbelongs
to a finitecluster,it occupiedsite,, can thenbe definedby
followsthat

P=p-

Z!

nss

(2)

'=

Rjnss

nss~

(6)

Furthermore,
it can be shown[Stauffer,1985]that the Similar to S, the extent of the average cluster diverges as p
average
sizeof a finiteclusterpersite(aboveorbelowPc)is -'->Pc, sothat a behaviorof the form

'sns
s2, whichis the weighted
average
of clustersizes.
Since
oneisinterested
in theconnectivity
between
occupied

778

BERKOWITZ
ANDBALBERG:
PERCOLATION
THEORY

theabovepowerlaws)to identifytheEuclidean
dimension.
alityofthesystem,
d (d = 1, 2, or 3 in theaboveexamples}.
Thedefinition
of thedimensionality
of a system
canthen
be generalizedby

D = -log (M)/log(l/n)

{81

which is consistentwith the dimensiond knownfor the


aboveexamples.
Thus,for the Koch curve,the definition
0f
the lengthyieldsthe dimensionality
D = log (4)/log(3)

1.26,i.e., thiscurveisa self-similar


system
withnoninteger
dimensionality.Such a system can be called an "exact
fractal" with fractal dimensionD. Note that the fractal
dimensionis confinedwithin the two relevant Euclidean
dimensionsof the system.In the exampleused here,the
length of the Koch curve increases"faster" than that ofa

regularnontortuous
curvewith eachgeneration,
although
it
doesnot fill the plane. Hence, it is not surprisingthatforthe
Koch curve, 1 < D < 2. Similarly, for ramified structures
Fig. 4. The first five generationsof the Koch curve. The fully
developedKoch curve is obtainedby continuingthe rule depictedin
the present figure indefinitely.

enclosed
in a three-dimensional
space,it shouldbe expected
thatD

< 3.

Turning now to ramified percolation clusters, it is not

expectedthat successive
magnificationsof clusterportions
can be expected. It has been shownby variousapproaches
that v = 4/3 for two-dimensionalsystems,while v - 0.875 for
three-dimensionalsystems. Two sites of a cluster are consideredcorrelatedif they are connected,and thus e,which
characterizesthe average distanceover which two sitesare
connected,is called the "correlation length" of the system.
To obtain a better feeling of how large clusterslook, andto
characterizethe rather ramified ("Swiss cheese-like") struc-

will yield exactly the same picture as those seen without


magnifyingglasses.And yet, for any given picture, it isnot
possibleto tell at what magnification (i.e., scale) it was
taken. However, as for the Koch curve, there is a well.

definedrelationbetweenclustersizes andthecorresponding
clusterradiusR s, suchthat on average(over clustersofsize
s), one obtains

soc
Rs
D

(9)

ture of these clusters,some concepts,which can help in


quantifyingthe degreeof "ramification" of a cluster,canbe for all (large enough)s. In particular, any large but finite
by
used. The conceptsof self-similarity, fractals, and fractal portion of the percolationcluster can be characterized
this
D.
Hence
the
percolation
cluster
is
a
"statistical
fracdimensionare thereforeintroduced.As a simpleillustration,
consider the so-called Koch curve. This curve is constructed

tal," to be distinguishedfrom the "exact fractal" described

followinga specificrule, as shownin Figure4. In every above.Again, the dimensionalityof the Euclideanspaceisd
generationof thecurve,a segmentis dividedintothreeequal = 3, while for the percolationcluster, it is D < 3 whenp --,
thatPcr (p - p c)t, it hasbeenshown
segments,and the new segmentis the buildingblock of the Pc. In fact,noting
theoretically,and confirmedexperimentally and by computersimulations,that D = d - /3/. In the caseof clusters
in the two-dimensional
space,D = 91/48 and/3= 5/36,while
in thethree-dimensional
space,D = 2.52 and/3= 0.41.Note
ematical limit.
againthat since/3and , are universal,D is universal,andas
Considernow the use of magnifyingglassesto examine suchis uniquefor every d [e.g., Stauffer, 1985].
this curve. Suchglassesmagnifythe pictureby a factor of,
say,m, butalsocausea reductionin thefieldof visionby the
2.3. SomeFundamentalConceptsof PercolationTheory
samefactor. If suchglassesare appliedto the Koch curve,
for Fluid Flow
the pictureseenfor any m will be the same.In otherwords,
for eachscaleof length,the observedpictureisthe same.As
To furtherapproachthe problemof interestin thisreview.
next generation,suchthat the old segmentis replacedby a
structureof fournew segments.
Hencethe totallengthof the
curve in eachgenerationis 4/3 timeslongerthanthat of the
previousgeneration.This processis carriedout to its math-

such, the Koch curve is said to be self-similar. What is the

i.e., geometrical
and transportpropertiesin hydrological

lengthof the Koch curve9.It is obviouslyinfinite,unlikethe systems,


it is necessary
to reexaminethe systemdescribed
initial segmentfrom whichthe curvewasgenerated.In this in Figure1, andto returnto the conceptof "bonds"(pipe
sense,it is differentfrom the simpledivisionof a curvethat segments).Let p be the probability that a bond is open

is encountered
by simplychoosing
a differentunitlength(or
the yardstickusedfor its measurement).
For example,if a
segment
is dividedinton equalsegments
(ortakea yardstick
whichis n timessmaller),the numberof newsegments
is
M = n. Thedivisionof a square
by thedivision
of itsedges

("occupied"),
andpc bethecriticalvalueofp abovewhich
water will reach the southern end of the farm. Obviously.

flowproblems
areof interest
onlyforp > p c-Notethateven
in thisregime,manyof theopenpipesegments
will nothave
waterflowing
inthem,sinceeithertheydonotbelong
tothe

inton equalsegments
yieldsM = n2newsquares
(instead
of "percolating
cluster,"or theyform"dangling
branches"
the initialsinglesquare),andthe divisionof a cubeby (analogous
to dead-end
pores)which,althoughfilled,do
divisionof its edgesinto n equalsegments
yieldsM = n3 conductfluid.The ensembleof systemelements(pipesegcubes.Thelastthreeexamples
canbeused(byconsideringments)through
whichthefluidcanflowis calledthe"back-

BERKOWITZ
ANDBALBERG'
PERCOLATION
THEORY

779

andBalberg,1992].Theoverallpermeabilityof the systemis


then simplygivenaccordingto Darcy's law by
K = Q/AP

(1 I)

whereAft isthepressure
dropacrosstheentiresystem,
and
Q is the overallflow rate throughthe system.Since Q is

proportional
to /xff, the value of K dependsonly on the
connectivityof the system(i.e., on the bond occupationp)
and the geometryof the singlepipes.
The aboveapproachcan be generalizedto other typesof
networks,and is thus useful for computer simulationsof
finitemodelsystems.However, while providinga complete
solution,it has two major drawbacks. The first is that it is
necessaryto know all the k values in the system, and the
second is that, even if all of this information is available, it

will always be limited to finite (and thus relatively small)


systems. As such, the statistics used in the simulations may

yield very poor accuracy in the results, consideringthe


larger system that the model tries to simulate. While this
may be sufficientfor the Texas farmer, the hydrogeologist
L1
who considerspores in rocks requires an approach that
Fig. 5. An illustration of the backboneand the finite clusters circumventsthese drawbacks. This is where the universality
thatareisolatedwithin its net. Examinedhereis a sampleof edgeL
of percolationtheory becomeshelpful.
whichis a cut from an infinite network. The "squares" of the net
Consideragainthe simplebackboneillustrationshownin
havecomers("nodes"), edges("links"), and parallel connections
("blobs")with end-to-enddistanceof the order of the correlation Figure 5. From the definitionof k given in (10), it is clear that
the overall permeability decreasesas the length of the flow
lengthf. The lengthof a singlyconnectedbondin a link is L1.
path increases.Furthermore, since the correlation length se
divergesas p --> pc, it is expected that the permeability of
bone"of the system. Hence, with respect to consideration the "link," Kg, will dropsharplyasp --->Pc. Hence,the
ofsingle-phase
flow problems,it is sufficientto consideronly permeabilityexponent, ', is defined suchthat
the backbone.

What is the structure

of the backbone?

Recall that the

averagediameterof the finite cluster is se. Sincethe backbone


isa networkthat connectsthe ends of the entire system,the
averagediameter of the "holes" in the network must be of
orderf. Figure 5 illustrates the structure of the backbone
assuming
that a topological squarelikeapproximationis a
goodrepresentationof the backbone network. The network
consistsof "nodes" (or junctions) and "links" connecting

K = Ko(p- pc);

(12)

whereK0 isa constant


andK, isthepermeability
ofa "link"
(i.e., over the correlation length ). Turning to the permeability of the entire system, K, note that in a system of
dimensionalityd, if the length of the systemis L, then there
are, on average,L/se linksfrom one end of the sampleto the

other,and(L/Oa- parallel
pathscarrying
theflow.In this

"square" ("Texaslike farm") backbone model, there is no


flow in the "east-west" links, since all junctions along such
the"link" to the other is of the order of :(i.e., the average a link are at the samepressure. Hence, it follows that K =
radiusof possibleembeddedfinite clusters),but its actual K(L/O
a-/(L/se).
Forthree-dimensional
systems,
there
are

thenodes.As explainedabove,the distancefrom oneendof

detailed
(tortuous)lengthis, of course,longer.
L/s
elinks,but(L/)2 parallel
paths,
andthusK = K(L/).
Withthis understanding
of the flow networkgeometry,it In general, then
is now possibleto considerthe quantitativerelationship
K = K(L/)a- 2
(13)
between
the possibleflow and the probabilityof findingan
openbond.For a simplecircularpipeof crosssectioncrand and thus, considering(7), (12), and ,13),the critical behavior

length
i, the flow rate alongthe pipe, q, undera pressure
of the permeabilityK, which is definedby
dropof Ap is [e.g., Bear, !972]
Koc(p-Pc)"

q = cr2A/(8-r/l)mkAff

(10)

where/is the dynamicfluidviscosity.Note that herethe

local
permeability
isdefined
ask -- cr2/(8
r//).Theproblem

(14)

is determinedby the exponents ' and v accordingto

c= ,, + (d-

2)v

(15)

formulated
at the outset,i.e., to determinethe valueof the
overallsystempermeability,K, from the knownlocalper- The goodnewsis that r and v can be determinedin simple
meability,k, can be considerednow by usingpercolation s,stems,and thenusedto predict the behaviorof systems
for whichno, or very few (usually global)detailsare known.
theory.
If thepermeability
of eachpipe(i.e., corresponding
to the connectionsbetweenjunction pointsor "lattice The bad news is that in order to have an equals sign in (14),
sites")
is known,thenby useof KirchhoWs
lawfor each many more details are required, so that one must resort to
closed
loop(whichrequiresthatthe algebraic
sumof the proceduresof the kind outlinedby (!0)411). Thus, one of

fluxes
ateachjunction
equalzero),thecalculation
ofK ina the main contributionsof percolationtheory lies in its ability
finite
system
is straightforward
[Balberg,
1986b;Berkowitzto predictthe value of K. Since, as pointedout above, v can

780

BERKOWITZ
ANDBALBERG:
PERCOLATION
THEORY

Pc) of thebonds(i.e., trimming


a fraction(p - p)/pofall
occupied
bonds),thepercolation
thresholdis reached.However, this thresholdwill be reachedwhen, on average,one

singly
connected
bondperlinkis trimmed,
i.e., whenL(p
- Pc)/P = 1. Thus,

Ll(P-Pc) -1

(17)

Following
thefact thatthe totaltrue flow pathof thelink,
L2, is alwayslargerthan L1, regardlessof the particular
systemand its dimensionality,it must be concluded
from
(16) and (17) that 'K-> 1.
How differentis the true 'Kfrom 1? Computersimulations
Fig. 6. The first threegenerations
of the exact"permeability

and otherconsiderations
showthat the differenceis very
small, and that with increasingdimensionalityr 1.

fractal" (knownasthe hierarchicalmodel),whichis presentedhere Indeed the largestexistingdeviation, which is for d = 2, is


to show how a critical exponentcan be estimatedfrom a simple ' -1 - 0.3. For d = 3, thisdeviationis reducedto 0.1.To

illustratethat the difference' - 1 is of this order, consider

model..

the model illustrated in Figure 6, but now with the contribution of the "blobs." Assume that the permeability of the

be determined analytically (for two-dimensionalsystems)


usinga simplelatticemodel,the quantitythat stillremainsto
be found, which is closely related to the flow problem, is '.
In order to estimateK, it is necessaryto considerin more
detail the structureof the link, sinceit is the property of the
link that determinessr.As a simplemodelfor the detailed
structureof a link, consider the following exact fractal. The
fractal is generatedby increasingthe link three timesin each
generation, so that an additional new segment(resistor or
pipe) is added in parallel to the central segment. The
generationsequenceof this fractal is illustratedin Figure 6.

first-generationelement is k. Then the series and parallel


connectionin the secondgenerationyields a permeability
of
k/2.5. Thus when the length of the fractal, l, increasesbya
factor of 3, its permeability,K l, decreasesby a factorof 2.5.
According to relation (8), the fractal dimension of sucha

model,whichis definedby Kl ocl -D, is givenby


D = (log 2.5)/(log 3) 0.834

(18)

On the other hand, for links of length l = , which are of

interest
here,it follows
from(7)thatK o(p - pc)Dand

thus sr -- D v 1.11. It is apparentthen that even sucha


In the limit, the fractal consistsof "singly connectedbonds" crude (thoughelegant)model yields reasonablygoodvalues
(throughwhich all the flow shouldoccur); two in the second for sr.More preciseestimatesof , however, are obtained
generation, four in the third generation, etc. There are also by more elaborate calculationsor by computer simulations.
"blobs," i.e., bunchesof bonds or pipes that form parallel
flow paths within a link. Becauseof the highernumber of
paths, the blobs have a higher permeabilitythan the singly 2.4. Continuum Percolation
connected bonds, so that to a first approximationfor the
While one could be content with the achievements of
purpose of calculating the permeability, one can neglect
percolationtheorydescribedin sections2.1, 2.2, and2.3,it
them and assumethat the total length of the !ink is equal to
is apparentthat Texas farms in particular, and latticesysthe total (average) length of the singly connectedbonds in a
temsin general,are the exceptionrather than the rule among
link, L .
systemsin whichthe connectivitydeterminesthe behavior
The end-to-end distance of the link, , diverges as the
of a system. Most real systemspossessa much less 'orpercolationthresholdis approached,and thus L1 also didered" network,and a distributionof k values.This is,of
verges there. On the other hand, one cannot determine a

course,the casein real porousmedia,fracturedrocks,and

priori whetherL < or L > :. SinceL is tortuous,one


othergeologicaland hydrologicalsystems.As an example,
mightexpectthat L > :, while sincethe lengthof the blobs
Figure 7 illustratesa computer-generated
sampleof ranis neglected,one might expect that > L . The answeris in
fact determinedby the dimensionality.For examplefor d = domly alignedline segments.This can be envisionedasan
aerialphotographor a crosssectionof fracturesin a forma2, L < , whileforhigherdimensions,
L > . SinceK is tion, or a system of natural channels that serve the same
to be determined, and since

purposeas the artificiallyplacedpipesin the Texasfarm.

KL= k/L1

(16) The first questionthat arisespertainsto the quantitythat

shouldbe usedhere,insteadof p, so that a "proximity"

wherek is the permeabilityof eachbond(or pipe),andL is parametersuchas p - pc can be defined.


measuredin latticespaceunits(i.e., it is the numberof singly
To answerthisquestion,reconsider
nowthe sitesand
connected bonds in the link), a first-order determination of bondsin Figure2. In this lattice structure,each sitehas:
K (in the singlyconnectedbond model, i.e., assumingthat (=4 in this figure)possiblebonds.If a fractionp of these

is occupied,
theaverage
numberof bondspersiteis
Kf = KL) requires
estimation
of thedependence
of L on bonds
B = pz. Thepercolation
threshold,
orthecriticalnumber
of

P - Pc. The exactrigorousevaluationof the dependence


is
quite sophisticated,
and thus a simpleheuristicargumentis
used here to derive the result. Assumethat p bonds are
occupied,and that the systemis above the percolation
threshold.By trimmingrandomlythe excessfraction(p -

bondsper site, can thenbe definedby Bc = p cZ. Hence,

sincez is a constant,
B - Bc is as goodas proximity
parameter
asp - pc. Onecannowreturnto the continuum
modelof Figure7 andaskwhatthe averageB (i.e., number

BERKOWITZ
ANDBALBERG:
PERCOLATION
THEORY

781

1991].It alsoturnsoutthatthe excludedvolumeconcepthas


beenuseful[Batberg,1986a, 1987] for theunderstanding
of
theunexpected
very low or very highcriticalvolumeof the
conductingphaseat the onset of electrical or fluid flow in
compositematerialsand porousmedia (see discussionbelow).

Comparison
of FiguresI and7 revealsanothersignificant
difference
betweenlatticemodelsand real systems.While
thelengthof each"pipe" (or segment)is the samein Figure
1, the flowin the "equallengthfractures"in Figure7 takes
placealongsegmentsof different length.Thus, there is a
distribution
of flowsegments
and, as canfollowfrom (10), a
distributionof k values.The questionof interestis whether
this situationaffectsthe valuesof , that were derivedfor
models in which it was assumed that k is the same for all

Fig. 7. An filustration
of a continuum
model.Thelinesegments elementsthroughwhichfluid flow occurs.Sincethe percolationtheorypredictionsare accuratecloseto the threshold

maybe considered
to be fracturesor channelswhichcanconducta
ftaid.Thissystemis characterizedby the numberof segments
in its

(i.e., when:is very largein comparisonwith the detailsof

fixrite
version,
andby thenumberof bondspersegment
in itsinfinite the system),a "well-behaved"distributionof k values(e.g.,
version.The network shown here is just above the percolation

threshold.

Gaussian)
will manifestitselfby an average(K0) in (12), so
that Kis thesameasfor lattices.The situationappearsto be

ofintersectingsegmentsper line segment,or fracture)is, and

quite different, however, if the distribution of k values


divergesas k 0. Under such conditions,the decreasein K
asthe percolationthresholdis approachedfrom aboveis due
not only to the decreasingnumber of flow paths (as in the

thencomputeanyphysicalpropertyasa functionof B - Bc.


It hasbeenassumedthen, from the universalityapplicableto
lattice), but also to the fact that the participatingpaths
thevariouslattice systems,that all the resultspreviously consistof elementswith smaller and smaller k values (unlike

derivedfor lattice percolation are also applicableto such the lattice case). As will be seen below, such a distribution
continuumsystems, except that here the well-definedB may yield a K value that is different from that of the
Bc is usedinsteadof the lattice parameterp - p cnondivergingk value distributions(known as "lattice" modTo show that (1) describes the same critical behavior as

theother critical behaviorgiven above, i.e., that (up to a


proportionalityfactor) N - Nc is the same proximity
eter
as p - Pc and B - B c, the concept of the
excludedvolume is now introduced[Balberget al. 1984].
Theexcludedvolume (or area, in a two-dimensionalsystem)
isdefinedasthe volumearoundan objectin whichthe center
ofanotherobject must be in order for them to overlap.For
example,for spheresor cubes of volume V, the excluded
volume,Vex, is simplygivenby Vex = 8V. In caseswhere
theexcludedvolumeis differentfor two givenobjectsor two
givenintersections,the averageexcludedvolume,(Vex),
mustbe used. A simple example of sucha case is that of
"sticks," or fractures, as shown in Figure 7. If all the

els). The behavior associated with these different values is


known as the "nonuniversal"

behavior.

To demonstratethe concepts associated with the "nonuniversal" (nonlattice) behavior, consider now the case of a
porousmediumwhere the fluid flows throughthe overlapsof
the nearest pores. The rock material that surrounds the
pores is not permeable to the fluid. Such a system is
illustratedin Figure 8a, and the geometry of the intersection
of two poresis illustratedin Figure 8b. This systemis known
as the "inverted random void" (IRV) system. Following
(10), it is clear that the value of k is determined by the
narrowest region (smallest cross section) between two
spheres. This region is usually called the "neck" of the
bond, and is describedby the cylinder shownin Figure 8b.
fractures
areof length/0,andthereis anangieof Oi- Oj It is a goodapproximation[Feng et al., 1987]to assumethat
between
two of them, the corresponding
excluded"vol- k is determined essentially by this cylinder. Examination of
ume"
(area)
isA = Igsin[O/-Oj[.If 0i andOjarerandomlyFigure 8b showsthat the length of the cylinder, l, is of order

anduniformlydistributedbetween -,r/2 and ,r/2, the aver- 2(ae)2/2,withradiusof order(ae) 12,andthustheperme-

ageexcluded
"volume"is (2/z-)/g[Balberg
et at., 1984]. ability of the "neck" (see (10)) is

Nowthe numberof objectswhosecenterslie withinthe


ke= *r2(ae)2/{(8r)[2(ae)l/2]}--=
koe3/2 (20)
average
excludedvolumeis P(Vex),wherep isthedensityof
objects
in the system.But thisis alsothe averagenumberof
'Assume
nowthatthenormalized
distribution
of e values,
objects
thatoverlap,or arebound,to a givenobject(i.e.,B). h(e), is suchthat for somee0, h(e) = h0 for e < %. Note
Thus
that since e cannot exceed b (see Figure 8b), this distribution is normalizedfor any reasonableh(e) between e0and b.
(19)
It can be shown [Feng et al., 1987] that such a distribution
Forafinitesystem
(lattices
orcontinuum),
(N - Nc) o{O- does indeed exist for systems such as the one illustrated in
Pc),and (1) can be used.However, it is obviousthat for Figure 8. Sincethe quantity that determinesthe flow, and is
finitesystems
(forwhichpercolation
theory
gives
accuratelinearly additive,is the resistivity, or the inverseprobability,

results),
p - pc andB - Bc shouldbeused,respectively.k -I , theaverage,say,overa link, is foundby integrating
Inrecentyears,significant
progress
hasbeenmadein the over the singlyconnectedbondsin the link. For large enough
theory
ofcontinuum
percolation,
and,atpresent,
Bccanbe links, hi e) also appliesto the distribution in the link. On the
calculated
analytically
for manysystems
[Alonet al., 1990, other hand, the probability that the overlap yields a neck

782

BERKOWITZ
ANDBALBERG:
PERCOLATION
THEORY

f(s > 8) = 1- (Li/Lt)

h(e) de

(a)

Closeto the percolationthreshold,L1 --> oe,andthusit is


expectedthatsmallerandsmaller/valueswill be foundina
link. In particular,closeenoughto p c, for any e0thereisa
neckof "width" $ suchthat 8 < e0, and for this range

(b)

can be written as

f(s > 8) = exp [-Lh08]

(24}

This indicates(as is expected) [Feng et al., 1987] that the

typicalsmallest/in the link is givenby

8 = (Lh0)-
Fig. 8. An illustrationof (a) a sampleof "invertedvoids"
where a fluid is assumedto flow throughsphericalvoids, and the

(25)

In this model of singly connectedbonds, it follows then(see


(21)) that

background
is assumed
to be impermeable
(insulating),
and(b) the
geometryof two intersecting
permeablespheresof radiusa thatare
separated
by a distanceb. The cylinderrepresents
the"neck", i.e.,
regionwhich yieldsthe dominantresistanceto flow.

K[=L(k
-)
Lk
-f h(s)s-"
ds (26)
and thus

K-1 Lk-lho e-ude

associatedwith an overlap width between e and e + de is


h(e)d,, and thus the averaged ("weighted") reciprocal

(27)

permeability
(k-) in the singlyconnected
bondmodel wherethe upperboundof the integralis taken as m, simply
[Feng et al., 1987] is

toindicate
that$ issmall
andK-1isdetermined
essentially
by $ (seeabove).SinceLk-1 is the valueexpected
for

(k,

e-"h(e) de

(21)

lattice percolation(' was determined in the singlyconnectedbondmodelby L , see(17)), onecanwritein general


that

where u is an exponent (which is 3/2 in the exampleused


here, see {,20)).For u < 1, the integralfrom 0 to b converges

K'I= (K-l)lattice
h0 e-ude

andhassomefinitevalue;thus{k-) is a constant,
andK
will have the same behavior as in the lattice models ((13)(15)). This is since it can be assumed that the network is

composedof pipes of equal permeability,so that each pipe

hasthepermeability
(k;-) - . Foru > 1, onecannot
simply

For thecasethatyieldsthe nonuniversal


behavior,i.e., for
> 1, it follows from (25) and (28) that

consider the lower bound since with the above mentioned

h(e)distribution,
(k;-) willdiverge.
Rather,it isnecessary
to followthevalueof (k[) as e -->0, realizing
thatfor a
finite L there is always a neck of smallest in the link.
Furthermore, in this case, the upper bound of the integral is

unimportant,and for mathematicalconvenience,one can


replace b by o: in (21).
In order to find this neck, note first that the probability
that a given e is larger than a chosen 8 is

1-fh(e)
de

(22)

K- loc(K-l)lattice
Lu-

(29}

Then using the definition

reoc(p
- pc)''

(30)

wherep(orN, ina finitesystem;


seeabove)isthedensity
of
permeable
pores,
it follows
(considering
(14)and(17))thatg
= c+ u - 1. Hence,thereexistsa nonuniversal
behavior,
whereu andthusto'arenotonlya functionof d (dimension},
but alsoa functionof the particularsystemthat is charac-

terizedby u. Fora finitesystem,


it is obvious
thatthere
is,
by definition,
a smallest
& However,as longas this/is
smaller
than(Lh0)- , thebehavior
of thesystem
willbe

Consider now a percolating network of singly connected well describedby (30). Hence, in smallfinite samples,
a
to the universalbehavior
bonds. Blobs are not expected to be of significancein the "transition"fromthe nonuniversal
nonuniversalbehaviorof the entire system,dueto their high is expected
asp -->pc. Thequestion
is to beconsidered,
permeability, and may be consideredto be shorts.Of inter- when
onedeals
withfinitesystems,
pertains
totheproximity

estis the probabilitythatin a link in whichthereare onlyL i

ofthepercolation
threshold
andthecritical
exponents
t'ound

necks of "importance," 8 will be the e of the narrowest


neck. This probability is

for them to those calculated or measured in infinite s.stems

[Stauffer,1985].Thisissuehasfurtherconsequences..
since

BERKOWITZ
ANDBALBERG:
PERCOLATION
THEORY

783

evenif theh(e) distribution


is foundto peakat someera,as structureconsistingof cylindrical tubes of varying radius,

long
asp - pc is suchthatem< (Llhl)-l, wherehiis

and varyingdegreesof interconnection.In fact, analysesof

someaverageof h(e) between e0 and era, the critical pore structuresindicatethat the pore spaceis characterized
behavior
will appearto be nonuniversal.
by a wide distribution of channel sizes, and that random
As statedpreviously, the conspicuousfeaturesthat distin-

networks more closely represent actual systems of pores

guish
continuum
percolation
networks
fromlatticenetworks[e.g., Doyen, 1988]. But while other studiesbased on introare(1) the variablenumberof possibleconnecting
bondsper
site(or object),(2) the variationof the bondlength,and(3)
thelocalvariation in the bond direction. These aspectswill
beconsidered
further below.

ducingsomerandomnessinto lattice models [e.g., Ferrand


et al., !990] have yieldedinsightinto geometricalcharacteristicsandtransportbehaviorat the pore scale,there remains
considerableconcern over the validity of assumingthat the
Additional background reading on general percolation structuralnature of real porous media can be modeledby a
theorycanbe foundin the works by Broadbentand Ham- latticerepresentation.Percolationtheory, on the otherhand,
mersley
[1957],Frisch and Hammersley[1963],Shanteand hasshedlighton variousaspectsof network models,particKirkpatrick
[1971],Essam[1972],Stauffer[1985],andAha- ularly regardingthe nature and effects of network random-

tony[1986].Reviewsby Baudet et al. [1985]and Wong ness.


With respect to fractured formations, the problem of
[1988]
alsodiscussapplications
of statisticalphysicsto the
analysis
of propertiesand transportphenomenain porous characterizationof fracture network geometrycontinuesto
andfracturedrocks, including fractal conceptsand various

receive attention,particularly in terms of analysisof con-

percolation
theories(i.e., lattice,continuum,
andinvasion). nectivityand conditionsnecessaryto guaranteeor predict
the degreeof hydraulicconnectionwithin a network. While
someanalysesbasedon pure statisticalapproaches(involv3.
PROBLEMS IN HYDROLOGY
ing, for example,branching(Markov) processes[e.g., VereWhatdoesthe precedingexplan.
ationhave to do with Jones, 1977]) have yielded some insight into fracture nethydrology?
It is hoped that at this point, the connection work characteristics, percolation theory concepts have
betweenporousmedia and the "disorderedmedia" of per- proven especiallyuseful.
Historically, most relationshipsbetween hydraulic concolationis apparent, at least intuitively. By conceptualizing
a porespaceas a randommediumcomposedof conducting ductivityandporosityhavebeen developedtheoreticallyby
the pore spaceof a porousrock as a bundleof
"channels,"somepermittingflow and othersnot, the anal- representing
capillarytubes. The most frequently cited relationships
ogywith percolationnetworksseemsclear.
The history of the development of hydrology is well developedin thismannerare the Kozeny [1927]and Kozenyknown and will not be discussed here. Suffice it to say, Carmanequations[Carman, 1956], althoughmany modifihowever,that in recent years, there has been a growing cations have since been made to these equations. Other
relatehydraulicconductivity
to
realizationthat further significantadvancementof the sci- purelyempiricalexpressions
enceof hydrology requires a return to examinationand a mean(or "effective") diameter of the grainscomposingthe
development
of fundamentalaspectsof the subject[e.g., medium[e.g., Krurnbeinand Monk, 1943](as discussedby
KlerneL 1986; National Research Council, 199!]. While Bear [ 1972]),or to packingand sandshapefactors[e.g., Fair
somesignificantstrideshave been madein simulatingand and Hatch, 1933]. While such approaches are generally
for modelingmacroscopic
flow andunderstanding
predicting
propertiesof, and transportthrough,porousme- adequate
dia,numerous
questionsof fundamentalimportance
remain the consequencesof employing Darcy's law [e.g., Bear,
unanswered.
1972],they are not able to captureall aspectsof flow and
In fact, one can categorizecurrentfundamentalproblems transportbehavior, particularlythose significantlyinfluplaguing
hydrologists,both thoseinvolvedin researchand enced by processesoccurringat the microscopiclevel.

practicing
engineersdealingwith real-worldapplications, Characterization of flow in fracture networks has also been
primarilydueto uncertainties
regardingfracture
intofourgroups:(1) geometricalcharacterization
of porous hampered,
media;(2) characterization
and predictionof flow behavior networkconnectivity(as discussedabove). Many aspectsof
inporousmedia;(3) characterizationandpredictionof heat flow behavior can be successfullymodeled by percolation
propertiesareparticularlyuseful
andmasstransport
in porousmedia;and(4)characterizationtheory,andtheuniversality
andpredictionof multiphase(includingunsaturated)
flow in that someessentiallyunobtainabledetailedinformationis
behavior
in porousmedia.Understanding
of theseproblems unnecessary.
problempresentlyeluding
isfurthercomplicated
by the inherentheterogeneity
of such Probablythemostconspicuous
pertainsto the understanding
of, andabilityto
media,whichcreatesdifficultiesin obtainingandinterpreting hydrologists
bothin situand laboratorymeasurements.
Relatedto frac- model,solutetransportin porousmedia. Specifically,there
"difficulties"
turedmedia, these types of problemsbecomeeven more is a growingbodyof literaturedemonstrating

in applicationof the now classicaladvection-dispersion


in the meanEarlyconceptualizations
of porespacetopology
involved equation[e.g.,Bear, 1972],dueto uncertainty
of the dispersioncoefficient.Applicaconsidering
bundlesof capillarytubesof varyingradius(see, inganddetermination
forexample,
Bear [1972]for a review).Dueto the topolog- tion of the advection-dispersionequation assumesthat a
icalsimplicity,
essentially
allflowproperties
of suchsystemscontinuumapproachis applicable(i.e., thata representative
canbetreatedanalytically,andnumerous
flowandtransport elementaryvolume can be defined), and that dispersion
acute,and in some sense urgent.

as a (Fickian) diffusiveprocess.As
phenomena
can be simulatedsuccessfully.
However, such behavesmacroscopically
a
result
of
difficulties
applying
this equation, alternative
models
neglectthe fundamental
topological
randomness
of

arebeingconsidered,
andpercolation
theoryhas
porous
media.To betterapproximate
poreconnectivity,
Fatt approaches
[1956a,b, c] conceptualized
a porousmedium
asa lattice been shownto be useful in terms of describing"anomalous"

?84

BERKOWITZAND BALBERG:PERCOLATION
THEORY

(non-Fickian) diffusionand the scale dependencyof dispersion, in both porousmedia and fracture networks.
Finally, there exist even greatergeometricaland dynamic
complexitiesin multiphasetransportprocesses.Due to the
natureof the governingpartialdifferentialequations,andthe
various parameters and stability conditionsthat are involved, most work to date has been either experimentalin

forsystems
composed
of permeable
convex
objects
ofany
kind. The relevanceof the excludedvolumeconcept
to
practicalsystemsof interestin POROUS
mediaproblems
is
clearwhenconsidering
that (32) and (33) yield the critical
porosityapplicableto porousmedia. For example,these
equations,
throughtheir dependency
on V and Vex, allow

determination
of criticalporosities
for a varietyof systems,

nature,or basedon simplifiedequationspermittinganalyti- andcanaccountfor the fact that in certaintypesof rocks,


cal solution[e.g., Coskunerand Bentsen, 1990].Again, the flowis presentalthoughporosityis closeto zero (if poresare
usualtreatmenthas involved analysisof immiscibledisplace- consideredto be thin and sheetlike, i.e., there is a small
mentsin bundlesof capillary tubes.However, as above, the V/(Vex) ratio).
complexities of these phenomena have led researchersto
seek other approaches,includingpercolation theory.
In the following section, these aspectsare examined in
some detail. It is demonstratedhow percolationtheory can
addressthese issues, and successfultreatment of particular
problems associated with each of these four groups is
discussed.

In fact, (33) follows from the more fundamental relation-

ship betweencritical porosity and excludedvolume [Balberg, 1986a], which,for an infinitesystem,is givenby

qbc = 1 - exp (-p cV)

(34)

Relationshipsbetweenp and qbcan be derived from these


expressions.
In the caseof sphericalpores,for example,it is

well knownthat pVex = 2.8, VeffV = 8, and 4c 0.29.


4.

4.1.

APPLICATIONS OF PERCOLATION THEORY

Geometrical

Characteristics

Historically, one concept that has become a cornerstone


of continuumpercolation involves the relationshipbetween
the percolation threshold, Pc (i.e., some critical object
density)and porosity. For a systemof impermeablespheres,
wherein spherestouch each other at single points (and the

spheresare the "conductingpores"), it is clear that bc=


pcV, where bcis the critical volume fraction that is occupied by spheres,and V is the volume of eachsphere.This
quantity has been found to be dimensionallyinvariant (i.e.,
insensitiveto lattice structure).Generalizingthis conceptto
permeable spheres(i.e., pores), it was shown [Shante and
Kirkpatrick, 1971] that the average critical number of bonds
per site (which is sometimescalled the "average coordination number"), denoted B c, is related to the critical fractional pore volume bcby

bc = 1 - exp (-B c/8)

(31)

It hasbeenfurther shownthat B c is an invariantfor groups


of convexobjects[Balberget al., 1984].The valueof having
informationon the averagenumberof intersections
of a pore
with its neighborsis that it provides information on the
generalaveragetopologicalstructureof a porousmedium.
The conceptof the excludedvolume(discussed
above)has

The conceptsof percolation theory can also be usedfor


the studyof geometricalcharacteristicsof porousmedia.For
example, Yanukaet al. [1986] and Yanuka [1989a] consider
a geometricalmodelof a porousmedium, basedon a random
three-dimensionaljoint pore size distribution, and examine
the pore network structure.Their model, when compared
to
latticesin percolationtheory, is used to examinepercolation
probabilities and cluster formation, for different coordination numbers.It is found that the theory, numericalexperiments, and measurementson packings of glass beadsand
Berea sandstonegive similar results.
Similaranalysesof networkshave been appliedto fracture
systems.Chelidze [1982] suggestsa percolation modelfora
processof fracturing, which allows quantitative evaluation
of fractureclusterdevelopment.He considersbothfiniteand
infinite percolationlattices. The model requires knowledge
of several parameters: the total number of fractures, the
dimensionalityof the process,the coordinationnumberof
the network, and, for finite systemsof fractures, their length.
The idea is to describefracturing as a developingrandom
process.In particular, Chelidze [1982] developsa model
assumingthat the network is formed by a mergingsetof
smallfractureshaving no prescribedcenter of propagation
and which appearto be randomly distributedin space.He
successfully
correlatessomeexperimentalresultson fracturing processeswith the percolationfracture model. Furthermore,by assumingthat fracturedensityin a rock massunder

been useful in developing general empirical rules for the

nearlyconstantstressincreaseslinearly with time, hefinds


dependence
of the percolationthresholdPcon thegeometry thatmeasurements
of concentrations
of radontransported
in
of the objects(i.e., the pore shapes)makingup the system, groundwater,
takenprior to the earthquakein Tashkent
in
as well as on the macroscopicproperties (such as macro- 1966,can be explainedby the percolationmodel.
scopicanisotropy)of the system[Balberget al., 1984].More
The aboveconceptof the averagenumberof intersections
recently, exact analyticalcalculationshave been presented per fracture can be used as a measure of connectivityin
for the samepurpose[Drory et al., 1991].This concept, randomfracturegeometries[e.g., Robinson,1983,1984;
which accountsfor "interactions"(i.e., intersections)
be- Charlaix et al., 1984]. Robinson [1983] determinesby a
tween objects,is simplyconnectedto Bc (for permeable numericalprocedure
criticalpercolationdensities(i.e.,denobjects) by the relation
sity of fracturesrequiredto ensurea continuousfracture

connection
betweenopposing
facesof a rockmass)andthe
averagenumberof intersections
per "fracture"(line),for
systems
of finite-length
"fractures"(lines)uniformly
where(V,x) is the averageof the excludedvolumesof the various
domain.He alsoprovides
objects(seealso(19)).Balberg[ 1986a]thengeneralized
(3!) distributedin a two-dimensional
to
an estimateof the critical percolationdensitybasedon
lattice percolationprobability.For the varioussystems
(33)
4,c = 1 - exp [-(BcV/(V.x))]
considered,
at thepercolation
threshold,
theaverage
mareher
Bc = Pc(Vex)

(32)

BERKOWITZ
ANDBALBERG:
PERCOLATION
THEORY

of intersections
per line does not vary significantly.
He
shows
thatasthesizeof thedomainincreases,
thedensityof
fracturesmust increaseto ensureconnectivity.Robinson
[1984]
reexamines
hisprevious
resultsfor largersystems
of
fractures,and considersthree-dimensionaldomains.
Robinson [1983] also relates the threshold to fracture

length
anddensity,bythefindingthat(N(/)2)c- 1.54,for

785

systemsfor which networkrepresentations


can be applied
include,most clearly, fracturedmedia. Of course, such
modelsrely on varioussimplificationsof the pore space
morphology,
but whilethereis alwayssomequestionas to
their validityvis-h-visreal porousmedia,suchrepresentations seem unavoidable.

Rigorous
mappingof a porousmediumontoan equivalent

orthogonalsystems of fractures, where N is the actual networkof bondsleadsto a (variableaperture)networkwith


fracture
density(i.e., the numberof fracturesperunitarea), randomtopology.However, it has been shown [Jerauld et
t is the fracture length scale, and angle brackets denote al., 1984a, b] that a lattice network with coordination
average.This last result was extendedto three-dimensional number(i.e., the numberof bondsat eachjunction),z, equal
systems
of randomlyorienteddisk-shaped
fracturesof ra- to the averagenumber of bonds at each junction, B (disdiusr by Charlaix et al. [1984], who proposethat the cussedabove),of the disorderednetwork, possesses
transpercolation
threshold
canbe determined
from(Nr3)c port properties essentially identical to those of the disor0.15-0.3, dependingon fracture orientations.Again, these dered network. Jerauld et al. [1984a, b] examine
invariantsare essentiallythe B c invariantsdiscussedabove percolationpropertiesof two- and three-dimensionalrandom
[seeBalberget al., 1984].
networks.Monte Carlo realizationsof a particularrandom
Anotheraspect of interest, with regard to geometrical network, called the Voronoi construction, are generated,
characteristics,
is the relationshipbetweenthe percolation and geometricalpropertiesand conductivity of the systems
correlationlength, f, and the size of a representativeele- are comparedwith thoseof lattice networks. The comparimentaryvolume(REV). The generalphysicalsignificance
of son showsthat the parameters are qualitatively the same,
(whichcharacterizes
the averagedistanceoverwhichtwo and that the correlationlength, backbone, and conductivity
sitesare connected;see (7)) is that only for length scalesL exponents are almost identical to those of the lattice three>> f is the systemmacroscopicallyhomogeneous.Thus, the dimensionalnetwork. These results appearto supportunicorrelationlength is in a sense the "basic macroscopic versality arguments (see discussions above and below).
buildingblock" (see Figure 5) of the system. The REV of a Anotherdiscussion
of randomnetwork morphologyis given
porous
medium,associatedwith, for example,permeability, by Gawlinski and Redner [1983]. Thus, a lattice network
is defined by a scale at which permeability fluctuations structurethat is convenientfor computationalpurposescan
becomenegligible.As known from the densitydependence be considereda reasonableapproximationof at least some
of f (equations(7) and (19)), and shownby, for example, real porous media.
Robinson[1983] and Charlaix et al. [1987], for a fracture
network, the correlation length decreasesas the fracture
densityincreases; thus, the REV size also decreases.
As frequently noted in the literature [e.g., David et al.,
1990],systemswith nearly uniform distributionsof medium
properties(such as porosity or fracture aperture) can often
betreatedwith effective medium theories(in which a heterogeneous
system is represented by a homogeneousequiva-

lent),while systemswith broad distributionsof thesepropertiescanbe treatedwith variouscriticalpathanalysessuch


aspercolationtheory. The conceptof scaleis of fundamental
importance
in this regard,given that, at differentscalesof
consideration,the same medium can be consideredas either

homogeneous
or heterogeneous.
In caseswhereheterogene-

ityis presentat (relatively)smallscales,onecanintroduce


averagephysical and geometrical parametersto quantify
propertiesof the medium; the scale of averagingis defined

bytheREV. Belowthe REV, the mediumappearsstrongly


heterogeneous,
withhighvariabilityof properties;
abovethe
REV, the mediumappearsstatisticallyhomogeneous,
and
canbe modeledas an equivalenthomogeneous
medium.

4.2. Characterization
of FlowBehavior
Networkrepresentations
of porespaces
in porousmedia
havecontributed
significantly
to understanding
transport
of
fluidandmassin theporespace[e.g.,Fatt, 1956a,b, c;
Dullien,1979].At the microscopic
(pore)scale,theyhave
beenusefulin elucidating
effectsof poregeometry
and
topology
on flow phenomena
suchas, for example,immis-

cibledisplacement
[e.g.,ChatzisandDullien,1985;Wardlawetal., 1987;BluntandKing,!991;Ferrand
andCelia,
1992;
Heibaet al., 1992].On a macroscopic
scale,such

Percolation theory has been shown to be of considerable


use in predictingparameters(such as hydraulic conductivity
properties)that determinehydrological transport processes
in various porous media. Considerable effort has been devoted to analyzingelectrical resistivity of rocks, frequently
with reference to Archie's [1942] law; these results are

analogousto fluid flow in a porous medium, with Archie's


law beingreplacedby a Kozeny-Carman type equation.
Archie's law is an empirical result that simply relates the
electrical resistivity of a bulk rock to the product of the
resistivity of the conducting water contained in the pore
structure,and the porosityraised to some(fractional) power.
As for analogous(Kozeny-Carman) equationsfor fluid flow,
the exponent and coefficient of proportionality are determinedexperimentally.Various attempts to explain Archie's
law based on capillary tube models and effective medium
theorieshave met with only partial success.While measured
exponentsfor Archie's law are in the range of nonuniversal
critical conductivity exponents of percolation theory (see
above), the requirement of a percolation threshold very
close to zero (which is implied in Archie's law) would seem
to precludestraightforwardapplicationof percolationmodels to explain this law [Balberg, 1986a]. However, as
discussedbelow, a percolation threshold approachingzero
can, in fact, be explained quite easily.
Madden [1976] was among the first to consider random
networks as models of heterogeneousmedia, applied to the
study of electrical conductivity properties and Archie's law
behavior in porous media. He observes that for wide distributions of fracture and pore widths, the overall electrical
conductivityis not particularly sensitive to the topology of
the interconnections;this is confirmed by Jerauld et al.
[ 1984a, b] (discussedabove). Significantly, Madden [ 1976]

786

BERKOWITZAND BALBERGiP-ERCOLATION
THEORY

also concludesthat the geometricmean can be used to


accuratelypredictphysicalproperties(e.g., conductivity)of
heterogeneousmixtures of materials. This result, which
incorporatespercolationtheoryarguments,predatesnumerous later analyses of heterogeneousmedia, which also
concludethat the overall hydraulicconductivityis givenby
the geometricmean of the local hydraulicconductivitiesof
the individualelementsof the medium.Otherearlystudiesof
flow through"random maze" models,with brief reference
to percolationprocesses,are found in the works by ToreIll
and Scheidegger[197!] and Torelli [1972].
Percolationtheory considerations
are employedby Silliman and Wright [1988] in a stochasticanalysisaimed at

The modelconsiders
intersecting
conducting
permeable
"pores,"spheres
(in threedimensions)
andcircles(intwo
dimensions),
that are randomlydistributed
in space.The

local
hydraulic
conductivity
attheintersection
between
any

two permeable
spheres(circles)is eitherprescribed
or
determined
bythelocalgeometry,
andincorporates
features
illustrated
inFigure8 anddiscussed
in relation
tothisfigure.
Theresults,
whicharein excellent
agreement
withpredic.
tionsof continuum
percolation
theory,indicate
thatregardlessof the internalgeometryandconsidered
definitions
of

localhydraulicconductivity,
the overallhydraulicconductivity,K, in thetwo-dimensional
systemfollowsa universal
powerlaw relationship
givenby K o(N/No - 1).3.
identifyingpathsof high hydraulicconductivityin porous The resultsfor the three-dimensional
systemare alsoin
media. They representa three-dimensional
rock massby a agreementwith percolationtheory predictions;it is found
set of cubic grid elements,each possessing
a randomly that when the local conductivitiesfollow distributions
in
assigned
conductivity.This modelis analogous
to siteper- which some finite average value can be definedto be
colationon a cubiclattice(constructed
froma geometrically representative
ofthedistribution
of conductors
inthesystem
orderedcollectionof sites,andconnected
by bonds)wherein (suchas is the casewhenassuming
uniformor lognormal
all bonds are open (active), but individualsitesare active distributions),the overall (equivalent)hydraulicconductivonly with a fixed probability.The gridsgeneratedin their ity followsa universal
powerlaw relationship
givenbyK
Monte Carlo simulations were examined to locate connected

(N/No- 1)2.0.Furthermore,
it isfoundthata nonuniversal

pathsbetweeninletandoutletfacesof the cubicdomain,and behavior of the hydraulic conductivity arises when the
to identify"extremepaths"containing
the maximumvalue distribution
of localhydraulicconductivities
is diverging.
from amongthe minimumhydraulicconductivityvalues The nonuniversal
behaviorcan be predicted,since,asdisalongsuchpaths. Each "maximized minimum" was trans- cussedin relationto Figure 8, percolationtheoryalso
provides(analytically)conditionsunder which nonuniversal

formedto yieldthe "extremepathvalue,"which,for infinite


grids,is shownto be analogous
to 1 - p c- Calculation
of the
meanand varianceof the "extremepathvalue"for various
gridsizesindicates
that the asymptotic
"extremepathval-

behavior
maybeexpected
to arise[e.g.,Balberg,1987;
Feng

et al., 1987].Briefly,this theorycan alsobe explained


as
follows.Define7to be a transportpropertyof the medium
ue" is consistent with established estimates of the critical (suchas hydraulicconductivity),f(#) to be the distribution
percolationprobabilityfor site percolationon an infinite functionof 7 in the system,and h(/5) to be a distribution
cubic lattice.
function of a geometricalparameter, 8, relevant to the
In anotherstudy,Sillirnan[1990]usespercolation
theory systemandthetransport
property(e.g.,degreeof overlap
of
andMonte Carlo simulationsto analyzefinitecubiclattices intersecting
spheres,or, for a fracturenetwork,aperture;

of the type considered


by Sillimanand Wright[1988].He
demonstrates
that lattice geometryand discretization
can
significantly
influence
the meancriticalprobability
withina
discretizedrandomfield. In particular,it is shownthat

seeFigure8, where8 = e). Thenthe relationship


between
f(17)andh(/5)is givenby f(#) = h()(dlJ/d#)[Senet al.,
1985]. If h(/5) is such that f(17) oc17-, a nonuniversal
behaviormay be found[Kogutand Straley, 1979].The
refineddiscretizationof a lattice in the horizontaldirection nonuniversal
behaviorarisesif 0 < a < 1 (i.e., if thereis
cannotcompensate
for information
lost whenreducing
the sucha divergenceoff(!7) for # --> 0), with a criticalbehavior
verticaldiscretization
of thelattice,dueto thevariability
in characterized
by anexponent
x + a/(1 - a), wherex isthe
connectivity
(andthereforehydraulic
conductivity)
proper- universalexponent(suchas (). As mentionedabove,nontiesbetweentwo- and three-dimensional
systems.
Thisre- universality
arisesin caseswheref(17)diverges
as17-->0 due
sultisalsosupported,
for example,
by recalling
well-known to anaccelerated
dropin the sampleconductivity
beyond
the
percolation
theoryresultswhichshowthatthecritical(site) effectoffewerandfewerconductors
(i.e., asthepercolation
percolation
probability
fbr a three-dimensional
system
(Pc thresholdis approached
from above).
= 0.312) is considerablysmaller than that for a twoThe connection
betweenmodelsof this type andthe
dimensional
system(pc = 0.593). This observation
has dependence
of thehydraulic
conductivity
onporosity
ofreal
important
implications
in termsof numerical
modeling
of rockshasbeendiscussed
by severalauthors,including
for
hydrologicalsystems,where complexthree-dimensionalexample,Wonget al. [1984],Balberg[ 1986a],Thompson
et
flowdomains
are oftenreduced
in theirgeometrical
com- a!. [1987],andWong[1988].In fact,manyrealrocksexhibit
plexity.
nonzerohydraulic
conductivities
at extremelysmallporosiRenault [1991]calculatespercolationthresholds
for net- ties,whichin thesecases
indicates
a percolation
threshold
worksin whichthe conducting
capacities
of the bondsor approaching
zero. However,Balberg[1986a] showsthat
sitesare spatiallycorrelated.He considersseverallattice whilethe powerlaw dependence
of hydraulicconductivity
networks
intwoandthreedimensions,
andfindsthatperco- on porosityis validfor differentporosities,it is sensitive
to
lationthresholds
aredependent
uponthespatial
correlationthe microscopic
structure
of the pores.For example,
the
length.
criticalporosity(i.e., the porosityof the onsetof percolaBerkowitzand Ba!berg[1992]carriedout Monte Carlo tion)for spherical
poresis 0.29,whilefor poresshaped
as
computer
simulations
ona percolating
modelrepresentativethindisks,thecriticalporosity
approaches
zero.Thus,an
of someporous
media,andanalyzed
therelationship
be- apparent
difficulty
in applying
percolation
theoryto describe

tweenconductivity
andthenumber
of poresin thesystem. transportpropertiesof porousmediacan be resolved.

BERKOWITZ
ANDBALBERG:
PERCOLATiON
THEORY

787

conductive
fractures
withinthe percolating
cluster.By analogy to transportin semiconductors,they evaluate the ho-

mogenization
lengthand the permeabilityof the network.

Theypointoutthe existence
of two typesof heterogeneity
encounteredin porous media of low porosity: (1) poor
connectionof the pore space(or of the fracture network),
and (2) at a scalewell abovethe percolationthreshold,a

largedistribution
of pore(or fracture)aperturesandpermeabilities. Their results, which are valid when there exist
stronglyheterogeneous
local transportpropertiesin a wellconnectedfracturenetwork,constitutean extensionof perFig.9. The backbone(conducting
part of the percolating
clus- colationtheoryapplications,
which usuallydeal with netter)ofthesystemshownin Figure7. Notethattheapparently
dense works near the threshold. The main conclusion of this work
networkof Figure 7 is in fact not well connected.

is that as the distributionof aperturesbecomeswider, the


correlationlengthof the subnetworkresponsiblefor the bulk
of theflowincreases.
A similarapproachto that of Charlaix
A numberof studiesavailablein the literatureuseperco- et al. [1987],for the caseof a pipe network with a broad
lationtheory to analyze connectivities and conductivitiesof distribution
of radii, is givenby Katz and Thompson[ 1987].
fracturenetworks. Recall Figure 7, which illustrates a comThe importanceof scaleconsiderations
in heterogeneous
puter-generated
sampleof randomlyalignedline segments. porousmedia is discussedby Gueguen et al. [1991], who
Theline segmentsmay be considered to be fractures embed- illustratethe utility of percolationnetworksin investigating
dedin an impermeablerock. As discussedabove, sucha

transport properties within such media. They give two

system
easilylendsitselfto analysiswithpercolation
theory. examplesof applications,by showing (1) the pressuredeFurthermore,consideringsteadystateflow in sucha fracture
network,the "dead end" fracturescan be neglected,sothat
the network is reduced to a connected system of line
segmentsthat terminate at one another (known as the
"backbone").The backboneof the networkshownin Figure
7 is presentedin Figure 9; the dranmtic importanceof
connectivity
in such systemsis obvious.Conveniently,the
backbone
can also be analyzedwith percolationtheory.
Wilke et al. [1985] show that numerical simulations of

randomtwo-dimensionalplates ("fractures") embeddedin a


cubicdomaindisplaya criticalbehaviornear the percolation
threshold,which is characterizedby a correlationlength
similarto that of the site or bond percolationproblems.
Based
on thisbehavior,they suggestthat percolationtheory
results
for this classof systemscould be appliedto analysis
of flowproblemsin fractured formations.

pendenceof permeability in rocks, and (2) the density in

fracturevariationthroughthe Earth's crust.Theseexamples


alsoserveto indicatethat naturalfracturenetworksmay be
closeto the percolationthreshold (see discussionbelow). In

turn, this showsthat percolationtheory is helpful not only


for conceptualizing
porousmedia, but for modelingactual
natural systems.
Gueguen et al. [1986] and Gueguen and Dienes [1989]

analyze the hydraulicconductivity propertiesof fractured


media by combining percolation theory and a statistical
approach, together with a definition of the permeability
tensor of a fracture network developed by Dienes [1982].
Fractures are modeledas thin disks. Gueguen et al. [1986]

show that three geometrical characteristics of the medium


are important for conductivity determination: the mean
fracture spacing,the mean fracture radius, and the crack
Flowin a fracturenetworkis considered
by Englmanet al. shape factor (width/radius). The results are then used to
[1983],who analyze a random(continuum)distributionof discusstime evolution of conductivity, due to slow fracture
fractures
in a plane. They examinepercolationproperties growth or decay processes.These processescould be due to
(suchasoverallpermeability)of the network,asfunctionsof variationsin either tectonicstressesand/or hydrostaticpresthefracturearea densityand size of the medium(similarto sures, although Gueguen et al. [1986] suggest that the
Robinson
[1983, 1984]and Charlaix et al. [1984],discussed increase in fracture concentration through the percolation
previously),
and the modelis comparedto measurements
on threshold is stress controlled, while later evolution is conrealrock samples.A semiquantitative
expressionis then trolled by either of the two processes.It is also suggested
developed
that relatesfluid permeabilityto geometricalthat such a phenomenoncould result in large increases in
characteristics
of the fracture network. The expression conductivity without the development of larger-scale frac-

incorporates
a backbone
probabilityfactorthat measurestures.
howeffectively
a typicalfracturecontributes
to theoverall Gueguenand Dienes [ 1989]further extend theseresultsby
flow.Longand Witherspoon
[1985]developtwo-dimen-consideringtwo simplifiedmodels to describelow-porosity
sional
numericalmodelsof fracturenetworks,andexamine

rocks.

permeabilityas a function of fracture interconnection.They

dimensional tubes, and the other is based on interconnection

One model

is based

on interconnection

of one-

recognize
that theseresultscan be comparedto the perco- of two-dimensional(disk-shaped)fractures. As before, the

lation
resultsof Englmanet al. [1983],in termsof usingconductivity
isbased
onthreeparameters,
withthedegree
of
interconnection,f, deriving from percolationtheory. Their
percolation
theory to analyzenetworkconnectivity.
Charlaixet al. [1987]modela fracturedrockby a random analysisindicatesthat the percolation parameterf and the
array of plane random fracture networks with a broad porousmediumtortuosity parameter [e.g., Bear, 1972], r,

distribution
of apertures,
and showthat the equivalentare in fact related, since asf-->

0, r--> :. The authors use

Permeability
is determinedmainly by the critical (disor- thesemodelsto examinethe dependencyof conductivityon

dered)
subnetwork
necessary
for percolation.
Themethodporosity(givenin termsof the geometricalparameters),and
employs
analysisof "criticalbonds,"whichare the least comparethem with laboratory measurementsof rocks. The

788

BERKOWITZAND BALBERG:PERCOLATIONTHEORY

method, in contrast to those of Charlaix et al. [1987] and

old,aswellasvarious
otherproperties
(such
astheaverage

Katz and Thompson[1987], relies on additionalstatistical criticalobjectsize,e.g., the average"fracture"length)can


considerationsoutside of percolation theory, and is most be predicted
by use of excludedvolumeconsiderations.
appropriate for a narrow distribution of fracture apertures.
The results of Charlaix et al. [1987] and Katz and Thompson
[1987], on the other hand, are based on fracture networks
having a broad distribution of fracture apertures. Note also
that these results are more general than those of Englman et
al. [1983], who deal only with fracturesof constantaperture.
Furthermore, in contrast to theseother works, Gueg.enand
Dienes [1989] expressconductivity in terms of microscopic
geometricalparametersof the system.
The permeability of random two-dimensional fracture
networks is also examined in the flamework of percolation
theory by Hestir and Long [1990]. They first consider the
case for p = 1 (i.e., all fractures are present on an ordered
lattice, and a REV can be defined), and treat flow through
such a systemby an equivalent (or effective) medium approach (see, for example, David et al. [1990] and Harris

Followingsuchan approach[Balberg et al., 1984]for a

two-dimensional
system
of fractures,
Balberg
et al. [199I]
verifythe utility of expressions
suchas L c = 4.2/(,rN) m
whereL c is the criticalnormalizedfracturelength,andthe
coefficient4.2 is the B c of this system. A dimensional
invariancepropertyof excludedareasand volumeshasalso
been found, which is usually valid to within about 10%.

Thus,onecanestimate,
for example,the valueof Nc ina

two-dimensional
system
fromtherelationN c = Bc/(Aex),
where(Aex)denotes
theaverage
excluded
volumeperobject
andBc is knowna priori, say,by calculations
or computations. Note that this expressioncan be comparedto the
resultsof Robinson[1983, 1984],discussedin the previous
section.

Formationspossessingmultimodalpermeabilitydistributionscanalsobe analyzedin termsof percolationtheory.An

[1990]). They then examinenetwork parameters(bondpercolation probability p, coordinationnumber z, and connec-

attemptto estimatethe effectivepermeabilityof heterogeneous,two-component


(sandstone
andshale)formations,
by
tivity , definedas the averagenumberof intersections
per useof power-averaging
and percolationmodelapproaches,
fracture). It shouldbe noted that the connectivity,which is presented
by Deutsch[1989].The first approachsimply
these authors denote by ', is actually the parameter B, definesthe effectivepermeabilityas a power averageof the
definedabove,suchthat (see (19) and (32)) B =
permeabilityof each componentmultiplied by its relative
Theseparametersallow definitionof equivalentlattice per- volume. The percolationmodel employs a normalizedrelameability. Their connectivityvalue, ', is then modifiedto tionshipgiven by
allow the model to be adaptedto random systemsof fracke/kssCr(Vsh
c _ Vsh)t
(35)
tures. The analysisshowsquantitativelyhow permeability
decreases as shorter fractures are removed from the net-

where ke and kssdenoteeffectivepermeabilityand permework. They comparetheir estimatesfor percolationprobaabilityof the sandstone,
respectively,Vshcand Vshdenote
bility (P) and permeability(K) to thoseof Robinson[1983,
criticalvolumeand volume of the shale sequences,andt is
1984]and Englman et al. [1983], and find that while all three

the permeability exponent. Deutsch [1989] finds that both

modelsgive reasonableresultsnear the perco!ationthreshapproachesyield good effectivepermeabilityestimates,but


old and for fracturesof constantlength,only their modelis
concludesthat the power-averaging
approachis simplerasit
suitablefor predictingpermeabilityin systemswith random
requirescalibrationof fewer parameters.The appealof
length distribution of fractures.
percolationtheory for this class of problem is that the
Universal behaviorof hydraulicconductivityand electrieffective permeability of the formation decreasesas the
cal resistivity,similarto that discussed
abovefor porous relativevolumeof shaleincreases,and for a sufficientlyhigh
media(circularandsphericalpores),wasfoundby Balberg relative volumeof shale, the permeabilitybecomes,essenet al. [1991]for a two-dimensional
systemof fractures.They
tially, negligible.This behavioris similarto the approach
to
examinea computer-generated
analogmodel of a fracture

network,andobtaina universalpowerlaw relationship


for

the percolation threshold. The critical threshold for continu-

ity of one phaseor the other, assumingrandomdistribution


of the components,can be determinedfor simpleconfigurabutionof fractureapertures
in the networkon its transport tions. However, considerablework remains to extendperpropertiesis then presented, and basedon the limited field
colationtheory resultsto systemspossessing
multimodal
data available, it is suggestedthat the universalbehavior
hydraulicconductivity.An analysisof the effectof a distri-

predicted
by percolation
theoryis in generalto beexpected

permeability distributions.

for both the fluid hydraulicconductivityand the electrical

conductivity
in commonsystems
of fractures.In the analy- 4.3. Characterizationof Particle TransportPhenomena
sis,it is alsofoundthatlengthsof theconducting
elements
in
For particletransportby pure diffusion,it can be shown
the backbone(i.e., the conductingpart of the network)
[Gefen
et al., 1983]that on a percolationcluster,
followa powerlawdistribution
defined
byN l ocI-1.9, where

I is thelengthof theline segment


sections,
andNz is the
numberof elementsof lengthI in the backbone.This result

rrms
OC
t /Dw

(36)

is surprisingly
similarto distributions
reportedfrom mea- whererrm
s = (r2)1/2istheroot-mean-square
distance
travsurement and analysis of fractured formations in the field eledby a particlefromthe source(anglebracketsdenote
the
[e.g., Reches, !986]. It remainsto determinethe "universal- average
overa largepopulation
of particles),andt isthetime
ity" of theseresults,for differenttypesof theoreticaland requiredfor the particleto reacha distancer. The fractal
actual fracture networks.
dimension
of the randomwalk, D w, is generallygivenby
Balberg et al. [1991]alsofindthe thresholdfor the onsetof

D, = 2 + 0, where0 -> 0 [Gefenet al., 1983;Havlinand

electricalconduction
(Nc) to be in agreement
withtheoret- Ben-Avraham,1987],and 0 = 0 for a Fickian("normal")
icalvaluespredicted
frompercolation
theory.Thisthresh- diffusion. The random walk fractal dimension can alsobe

BERKOWITZ
ANDBALBERG:
PERCOLATION
THEORY

789

writtenin the form D w = D + R, whereD is the fmctal relationships


relevantto networkpropertiesnearthe percodimension
of the percolationcluster,and R is the so-called lationthreshold.
For example,in additionto (7),
electricalresistanceexponent [e.g., Aharony, 1986].This
XBor(p
- pc)I
K(p - pc)
(37)
lastquantity,whichhasbeenstudiedwhenanalyzingelec-

tricalresistanceproperties,is similarin meaningto srK,


whereX B is thefraction
of bonds(e.g.,fractures)
in the
whichwasalreadydiscussed
in regardto the permeability
clusterbackbone,andK is the conductivityof the network.
criticalbehavior.
The exponents
are dependentonly on the dimensionality
of
Thevaluesof theseexponentsfor purediffusionin square
the system(i.e., they are "universal"), and for a two-

two-dimensional
networks, at the percolationthreshold,are
dimensional
systemthey are given by [e.g., Stauffer,1985;
D = 91/48 and srR= 0.97. In contrast,for a fully connected

Sahimi,1987]/3s 0.53, and c 1.3. Theserelationships


can be usedindividually,or combined,to developscaling
result,in caseswhere D w > 2 (suchas near the percolation lawsfor otherexponentsof interest.For example,the mean
threshold),diffusion is generally called "anomalous." These flow velocity,v, can be related to the correlationlength,
values
varyaccordingto the dimensionof the network[e.g., as follows:
Aharony,1986;Stauffer, 1985]. Note that near the percolanetworkcorresponding
to p = 1, D = 2 and R = 0. As a

fionthreshold,particle displacementis slower than that for

vK/XS(P
- Pc) - Bf-

(38)

Fickian diffusion. A detailed set of Monte Carlo simulations


ontwo-dimensional honeycomb lattices, which confirms the

where 0s =
An intensiveand detailedstudy of dispersionprocessesin
valuesof these exponents, is given by Ahn et al. [1991].
The caseof pure diffusioncan be modeledby a simple porous media is presentedby Sahimi and Iredakin [1988],
randomwalk process, which allows particles to move ran- who incorporateseveral earlier sets of results (given as
references),and employ intensive Monte Carlo numerical
donflyin any direction within a percolation cluster. In
simulationand particletracking in random networks. Rangeneral,one can defineany numberof "rules" for directing
dom capillary networks are constructed (in both two and
the migration of particles on such a network. Diffusion in
threedimensions)by removingsegmentsfrom an orthogonal
disordered
media, modeledby random walks on percolation
lattice network, accordingto various criteria. As discussed
clusters,has been examined by so-called "ants in the
previously,while rigorousmappingof a porousmediumonto
labyrinth"routines. By definingdifferentrules which direct
an equivalentnetworkof bondsleads to a (variable aperture)
theresidencetime of ants (random walk particles) at each
network with random topology, it has been shown [Jerauld
site,onecan define "myopic" ants, "blind" ants, and "lazy et al., 1984a, b] (discussed above) that a lattice network
blind" ants, to name only three. Stauffer [1985] reviews structure of the type analyzed by these and other authors
anomalousdiffusion in this context, and also shows how the

abovescalingrelations (i.e., relations among critical exponents)between anomalous diffusion and the conductivity
exponentof the network can be derived. Monte Carlo
studiesof diffusion in percolating lattice networks have
verifiedthesepredictions.Wagner and Balberg [ 1987]study
omalous diffusion on continuum percolation networks,
and conclude that the critical behavior is consistent with that

ofothertransportpropertiesin the continuum.Furthermore,


scaling
exponentsare determinedby the connectivityof the
system;
changingthe residencetimes of the ants at eachsite
doesnot affectuniversality.Also, Sahimiet al. [1983]treat
diffusion
in percolationnetworkswith power law distributionsof conductances,similarto thosein continuumpercolation,and show how the random walk fractal dimension,
D,, may dependon the characteristicsof the distribution.

A detailedanalysisof dispersionphenomenain porous


media
usingpercolationtheoryis presentedby Sahimi[1987]
and Sahimi and Imdakm [1988, and references therein].
Sahimi[1987] derives a set of relationshipsrelevant to
geometrical
and hydrauliccharacteristicsof capillarytube
networks,andusesthem to establishvariousexpressions
for

would appear to be a reasonable approximation of at least


somereal porousmedia.
Sahimi and Iredakin [1988] show that as the percolation
threshold is approached from above, particle transport
(characterizedby mean square displacement)behavesin an
anomalousmanner, and therefore cannot be modeled by the
conventionaladvection-dispersionequation. The dispersion
coefficient is not constant. Rather, it is found to be scale

dependent,on eitherlengthor time, with exponentsvarying


for the flow and transportprocessesconsidered(i.e., proximity to percolation threshold; Peclet number for flow,
indicatingeither diffusiveor advective domination;consideration of diffusive transport into dead end and stagnant
regions). The methodsand analyses summarized in this

study may have considerableimpact on improvingunderstandingof diffusiveand dispersive transport processesin


porousmedia, and aid in developmentof models with a
realisticpredictivecapability.
Dispersionin two-dimensional
fracturenetworksnearthe
percolation
thresholdis examinedby BerkowitzandBraester
[ 1991b]. The analysisinvolvesa combinationof percolation
theory and randomwalk approaches,and is based on the

describing
dispersion
of contaminants
undervariouscondi- work of Sahimi [1987] and Sahimi and Iredakin [1988]. In a
tionsandassumptions,
nearthe percolation
threshold.
In similarprocedure,randomfracture networksare obtained
.particular,
the relationshipbetweenmean squareparticle by removingsegmentsfrom an orthogonallattice network
displacement
andtimeyieldsdifferentexponents,
dependinganddrawingthe apertureof eachsegmentrandomlyfrom a
ontheflow conditions(e.g., as characterizedby the Peclet lognormaldistribution.A randomwalk is then directedby
number;
see,for example,Bear [1972])and the transport the flow through the resulting network, and Monte Carlo
mechanisms
underconsideration
(e.g., advection,diffusion). simulationsare performedto track particlesthrough the
partiAdditional
discussion
of theserelationships
is givenby system.Neglectingparticlediffusionand considering
cle transportonly throughthe backbone,percolationtheory
Yanuka
[1992].
Sahimi[1987] summarizesa numberof usefulcritical considerations[Sahimi, 1987; Sahimi and !mdakm, 1988]

790

BERKOWITZAND BALBERG:PERCOLATION
THEORY

predicta dependenceof the meansquareparticledisplace- of a nonwetting


fluidusingpercolation
threshold
concepts.
ment on time raised to the power 1.27, rather than a power
of 1.0, which arisesfor the caseof regularFickian (diffusive)

Thisis appropriate
when"blobs"of the residual
phase,
whichformascontinuous
(percolation)
pathsforfluidflow,

by the wettingphase.
displacement.This predictionwas verified by numerical are eliminateddueto displacement
The sizedistributionof theseblobsdependson the connecexperiments.
Usingsimilrandom
networks,
Kopliketal. [1988]
con- tivity and topologyof the pore space.These authorsalso
sidera full distributionof particletransit times(i.e., times correlatethe residualnonwettingphasesaturationwiththe
requiredfor particlesto advancethrougha given lengthof capillarynumber.Similar analyseshave been usedto calcuthe system), and derive relationshipsfor higher-order mo- late two-phaserelative permeabilityrelationshipsfor both
ments (i.e., in addition to mean and variance) of the particle
displacements.The formulationtakesinto accountadvective
transport throughthe backboneand diffusivetransportinto
dead ends and stagnantregions. They develop a computational approachwhich allows determinationof transit time
distributionsof the particles, as a function of the macroscopic (mean) fluid velocity. Numerical calculationsare in
good agreementwith analytical predictions of percolation
theory.
The agreements between these various theoretical and
numerical results would seem to indicate that percolation
models can be of use in understandingdetailed transport
phenomenain porous and fractured formations. For example, the previous results illustrate the non-Fickian (anomalous) nature of dispersion in such systems, and can be
appliedto explainthe now well-knownscaledependencyof
the dispersioncoefficientin the advection-dispersion
equa-

drainageand imbibition.
Golden [1980] suggeststhe application of percolation
theory to describeflow throughunsaturatedporousmedia.
He observesthat percolation theory can provide a relationshipbetweenthe minimum and maximum saturationvalues
for a porous medium, based on analysis of percolation
thresholds.He further notes that hysteresis can be modeled

tion. It shouldalsobe pointedout that networksof the type


consideredin thesestudies(i.e., near the percolationthreshold) are fractal in nature. Thus, theseresults,giventhe fact
that they have now been derived for physicallybasedflow
systems,add considerabletheoreticaland quantitativesupport to expressionsdevelopedby, for example,Ross[1986],
Wheatcraft and Tyler [1988], and Cushman [!991], which
indicate the scale dependencyof dispersivityon fractal

mercury porosimetry and relative permeability curvesin


sandstone.They find that percolation theory for regular

paths.
As mentionedpreviously, one can arbitrarily define rules

with this approach,givingnew understanding


to the phenomenon.

More recently, Yanuka [1989a, b] has used pore size


distributionsfor a variety of porous media, together with
percolationtheory, to calculatecapillary pressurecurvesfor
both drainageand imbibition(i.e., hysteresisloops).These
predictionscomparedfavorablywith availableexperimental

data.ChatzisandDullien [1977]have alsomodeleddrainage


in two-dimensional systems. Later, Chatzis and Dullien
[1985] used two- and three-dimensional networks to examine

three-dimensional
networkscan be usedto modelcapillary
pressurecurves, as well as saturation and relative permeability curves, during drainage.
Ferrand et al. [1990] and Ferrand and Celia [1992]review
networkmodelingand percolationtheory, and their applica-

tion to porousmedia.They also discussthe use of percolation-basedmodelsin the simulationof fluid displacement
and
multiphaseflow phenomenain porousmedia. The emphasis

for directingmigrationof particleson a percolationcluster.


The physically based flow solutionsconsideredby, for
example,Koplik et al. [1988], Sahimi and Iredakin [1988],
and Berkowitzand Braester [1991b] prescribeparticular
setsof rules,andthusprovidea directlink betweenpercolation systemsand physicalprocesses
in porousand frac-

of their review, however, is on network models, rather than


percolationtheory; and it is important to differentiatebetween independentanalysisof network modelsand theiruse
in conjunctionwith actualpercolationtheory results.These
authorsindicatethat simplifiedlaboratory measurements
are
necessaryto determinethe constitutiverelationshipsthatare
tured media. Obviously, considerablework remains to fur- incorporatedin mathematicalmodelsof fluid flow in porous
ther establishand strengthenapplicabilityof percolation media.Physicallybasedmodelsof fluid-fluiddisplacement
at
theory to problemsof flow and transportin porousand the pore scale can be based on network models and percofractured media.
lationtheory.This enablesmeaningfulpredictionof constitutive relationships.
The first lattice models dealt almost exclusively with
4.4. Characterizationof Multiphaseand Unsaturated
Flow Phenomena

two-dimensionalnetwork representationsof inherently


three-dimensionalmedia. However, it has been shown[e.g.,

Configurations
adoptedby two immiscible
fluidspresent Chatzisand Dullien, 1977] that one shouldnot conclude
(or flowing) in a porous medium have been shown to be

critical percolationbehavior for three-dimensionalsystems

determinedby percolationmechanisms
[e.g., Chandleret
al., 1982].For example,asthepartialsaturation
of oneof the
fluidsdecreases
(duringimmiscible
displacement),
it may
loseits connectivity.Ultimately,if thesaturation
fallsbelow

fromtwo-dimensional
systems,sincethe criticalpercolation
threshold(pc = Bc/z) is approximately2/z for twodimensionalsystems,while it is reducedto about 1.5/z for
three-dimensional
systems,where z is the coordination

numberof the network. Other differencesbetweentwo- and


the percolation
threshold,
onlyisolated
(nonpercolating)
clustersof poresremain saturated.

three-dimensional
systems
for fluidflow are pointedoutby

Applicationsof percolationtheory to immisciblefluid Silliman [1990] (discussedabove).

displacement
havebeenstudied
fairlyextensively
(see,for
Whileearliermodels
ofimmiscible
displacement
consisted
example,
thebriefreviewsintheworksby KueperandFrind only of interconnectedtubes, more recent models [e.g.,
[1988]andKueperand McWhorter[1992]).For example, ChatzisandDullien,1985;Diazet al., 1987;Wardlawetal.,
Larsonet al. [ 1977, 1981a, b] calculateresidualsaturations 1987;Ferrand et al., 1990;Jerauld and Salter, 1990]con-

BERKOWITZ
ANDBALBERG:
PERCOLATION
THEORY

791

ceptualize
the pore spaceas networksof spherical
pores studieson fluid-solidreactionsin porousmediaas percolaconnected
(with variablecoordinationnumbersandgeome- tion processes
is givenby $ahimi et al. [1990].
tries)by cylindrical tubes of variable length and radius.

Maierand Laidlaw [1990]examinethe use of percolating


systems
with a dual networkof bondsand sites,andthe

5.

DISCUSSION AND CONCLUSIONS

associatedrelationship of fluid occupancy in connected


While this paperprovidesconsiderabledetail illustrating
bond-sitepairs. They show that symmetry between two the successful
applicationof percolationtheory and its
phases
displacing
oneanother
maynotalwaysexist.Yanuka conceptsto problemsrelevant to groundwater hydrology,
[1992]analyzessimilarmixedbond-sitenetworksto obtain one mustbe aware of the limitationsin its use for problems
with real porousmedia. For example, sincethe
relationsfor both capillarypressureand relative permeabil- associated
ityasfunctionsof the saturation
fractionof a givenphase. hydraulicconductivitycan vary by orders of magnitude
Characterization of pores accessible to each of the fluid among rocks of the same porosity (due to variability in
phases,
and the resultingdistributionof the fluidsin the microscopicpore structures),one might suggestthat the
of equalityin the power law relationshipsmaybe
system,have been achievedby a modifiedpercolation coefficients
theoryapproach,calledinvasionpercolation[e.g., Chandler of greaterpracticalsignificancethan the critical exponent.
et al., 1982; Wilkinson and Willemsen, 1983; Wilkinsonand Unfortunately, this coefficientis difficult to calculate, sinceit
Barsony,1984;Wilkinson,1986].Invasionpercolationusesa requiresthat detailsof the particular porousmediumunder

procedure
in which all sites of a network are assigned considerationbe taken into account. Percolation theory does
randomnumbersfrom a prescribed probability distribution
function,which is assumedto representresistanceto flow at
thesites.Fluid advancesover time, and at each time step,
the site with the smallest random number (i.e., least resistance)at the junction is occupied by the invading fluid. In
contrast to anomalous diffusion clusters, which advance

onlyif an open site is encountered,clustersof an invading


phasecontinueto grow at each time step.

not addressthis aspectof the problem. Furthermore, results


from percolation theory are based on systems near the
percolationthreshold,and the proximity of real porousrocks
to the thresholdand the validity of the critical relationships
away from the thresholdare matters of question.And finally,
whereas the statistical geometry of percolation clusters is
very well known, as are the critical exponentscharacterizing
geometricaland transportproperties, the randomnessof real
geologicalmedia (in both the laboratory and the field) is
generallypoorly understood,and their statisticalproperties
are only partially characterized.
In spite of these difficulties,percolation theory has been
shown here to provide useful predictions of the structural
parametersthat determinehydrologicaltransportprocesses,

Lenormand [1986] reviews mechanisms involved in immiscibledisplacement in two-dimensional lattice networks.


He showshow physical laws governing flow and meniscus
equilibrium
within capillarieslead to applicabilityof different
conceptualmodels of displacement, particularly invasion
percolationand diffusion-limited aggregation (DLA). The and both universal and nonuniversal behaviors have been
behaviorof immisciblefluid displacementin two dimensions shown to be predicted and identified. The virtue of the

is examinedand visualized experimentallyby Lenormand percolationtheory approach lies in its ability to handle
andZarcone [ 1985a], and a fractal dimension of the "clus- porousmediawith complex(microscopic)geometry and to
ter" that forms is determined. Such behavior is further
yield macroscopic
relationships.The practicalimportanceof
analyzedby Lenormand and Zarcone [1985b, 1989], who percolationtheory results(such as those discussedhere) is
showhowthe displacementmechanismscanbe describedby that, while not providingabsoluteanswers, they enable, for
invasionpercolationand DLA. Use of a modifiedmacro- example,predictionof the hydraulic conductivity behavior
scopic-scale
invasion percolationtheory has recentlybeen as a functionof porosity, based only on knowledgeof the
suggested
by Kueper and McWhorter [1992] to determine basicporestructure,while alsoprovidinginformationon the
large-scale
capillarypressuresaturationcurves.
generaltopologicalstructureof a porousmedium. A conIn contrast to "regular" (lattice) percolation theory, spicuousexampleof this is the explanationof Archie's law
whichhas been successfullyappliedto simulationof two- [Balberg,1986a]. Percolationtheory alsoyieldsinsightinto
phaseflowsdominatedby capillary pressures,DLA theory both miscible (dispersion) and immiscible displacement
hasprovenuseful in simulatingthe fingeringof a highly transportprocesses,and providesa meansto model essenmobilefluid into a very viscous fluid [e.g., Witten and tial features of many phenomena.
With regard to fracture systems, in cases where the
Sander,1983;Paterson, 1984;Chen and Wilkinson,1985;

DeGregoria,1985]. Hardy [1990] proposesa numerical fracture network is created due to stress, it can be postulated
modelbased on a modified diffusion-limited aggregation, [e.g., Gueguenet al., 1991]that conduction,and therefore
whichis ableto simulatesaturationprofiles(relativeperme- stress relief, occur the moment the threshold is reached.

abilitycurves)andproduction
histories(including
viscous Thus, one might reasonablyexpect that fracture networks
fingering
phenomena)
usefulfor oil reservoirengineering.created by such a processwill be near the percolation
Themodelcanalsosimulate
latticepercolation
andDLA as threshold.This suggestionis also made by Chelidze[1982],
specialcases.

who summarizesexperimental results indicating that the

Otherapplications
of percolation
theoryto multiphase
percolation
threshold
for propagating
andmerging
small
flowsystems
havebeenconsidered.
Forexample,
percola- fractures
isreached
simultaneously
withthedevelopment
of
tionhasbeenusedto examineclogging
in a porousmedium,
byrelating
the criticalthreshold
probability
to thetimeat

a "macrofracture."

Furthermore,it has been suggestedthat in real fractured

whichreactionwithinan exchangecolumnwill ceasedueto formations,the densityof fractures can increasethroughthe


plugging
of flowchannels
[SahimiandTsotsis,1985;Yortsos percolationthreshold.For example, Marsily [1985] and
andSharma,1986;!mdakmandSahimi,1991].A reviewof Wilke et al. [1985] report on boreholes drilled in various

792

BERKOWITZ
AND BALBERG:PERCOLATION
THEORY

G.E.,Theelectrical
resistivity
logasanaidindetermining
geologicallyand petrographically
similarformations(Britt- Archie,
somereservoir
characteristics,
Trans.Am.Inst.Min.Metall.Pet.
any andMassifCentral,France).Productionwellsdrilledin
Eng., I46, 54-62, 1942.
Brittany are successful,indicatinghigh fractureconnectivBalberg,
I., Excluded-volume
explanation
of Archie's
law,Phys.

ity, while wells in the Massif Central (which also intercept


Rev. B, 33(5), 3618-3620, 1986a.
fractures)have extremelylow yields.Theseexampleswould Balberg,I., Connectivity
andconductivityin 2-D and3-D fracture
systems,in Proceedingsof the InternationalConference
on
seemto indicatethat the questionof fractureconnectivityis
Fragmentation,
Form and Flow in FracturedMedia, Ann.Isr.
of prime importance.The lack of high productivityfrom
$oc.,vol. 8, editedby R. Englman
andZ. Jaeger,
pp.
wellsinterceptingfracturescanbeattributedto theexistence Phys.
89-101, Adam Hilger, Bristol, England, 1986b.

of finite(nonpercolating)
clusters.Oneimplicationof thisis Balberg,I., Recentdevelopments
in continuumpercolation,Philos.
that even in a domainthat appearsto be heavilyfractured,
Mag., 56(6), 991-1003, 1987.
there may not exist sufficientdensityto ensureconnectivity Balberg,I., C. H. Anderson,S. Alexander,and N. Wagner,
Excludedvolumeand its relationto the onsetof percolation,
(i.e., suchthat the systemis abovethe threshold;seeFigures
7 and 9). Thus, for fracture networks, it is of greatest

significanceto (1) predict theoretically the percolation


threshold,and (2) compareactualfracture densitywith this
threshold.

One shouldalso note the similarityof percolatingclusters


to channel systemsin a fracture "plane," which have been
intensively studied in recent years [e.g., Berkowitz and
Braester, 1991a]. Here too, percolationtheory may provide
a general means of analyzing flow and transport along a
fracture, especiallyin the absenceof detailedinformationon
the geometrical and hydraulic characteristicsof the system.
It alsoremainsto find physicalreasonsfor the existenceof
parameter distributions(such as fracture length and aperture) that have been assumed, or could be assumed, in
applicationof percolationtheory analysesto geologicaland
hydraulic properties of porous formations. Several cases
have been found that appearto indicatea closerelationship
betweendescriptionsof the featuresof field data and percolation models.However, while comparisonswith available
geologicaldata seemto supporttheseresults,additionalfield
data are required. One aim of this article is to provide an
impetusand useful reference for suggesting
the nature of
geologicaland geophysicaldata that shouldbe obtainedin
future field studies, in order to validate theoretical results
concerningstructuresand propertiesof real rocks.

Phys. Rev. B, 30(7), 3933-3943, 1984.


Balberg,I., B. Berkowitz, and G. E. Drachsler, Applicationof a

percolation
modelto flow in fracturedhardrocks,J. Geophys.
Res., 96(B6), 10,015-10,021, 1991.

Baudet, C., E. Charlaix, E. Clement, E. Guyon, J. Hulin, andC.


Leroy, Scalingconceptsin porousmedia, in Proceedings,NATO

Conference
on "Scalingin Disordered
Systems,"Geilo,Norway,
April 10-21,editedby R. Pynn,pp. 399-422,Plenum,New York,
1985.

Bear, J., Dynamicsof Fluids in PorousMedia, 764 pp., American


Elsevier, New York, 1972.

Berkowitz,B., andI. Balberg,Percolationapproachto the problem


of hydraulicconductivityin porous media, Transp. PorousMedia, 9(3), 275-286, 1992.

Berkowitz,B., and C. Braester,Solutetransportin fracturechannel


and parallel plate models, Geophys.Res. Lett., 8(2), 227-230,
1991a.

Berkowitz, B., and C. Braester, Dispersionin sub-representative


elementaryvolume fracture networks: Percolationtheoryand
randomwalk approaches,Water Resour. Res., 27(12), 3159-3164,
1991b.

Blunt, M., and P. King, Relative permeabilities from two- and


three-dimensionalpore-scalenetwork modelling, Transp.Porous
Media, 6, 407-433, 1991.

Broadbent,S. R., and J. M. HammersIcy, Percolationprocesses,


crystalsand mazes, Proc. Cambridge Philos. Soc., 53,629-641,
1957.

Carman, P. C., Flow of Gases Through Porous Media, Butterworths, London, 1956.

Chandler,R., J. Koplik, K. Lerman, and J. F. Wiilemsen,Capillary


Due to its generalapplicabilityand appeal, percolation displacementand percolation in porous media, J. Fluid Mech.,
119, 249-267, 1982.

theory has becomeubiquitousin many fields. However, Charlaix, E., E. Guyon, and N. Rivier, A criterion for percolation
muchof the work and resultsare only of secondaryapplithresholdin a random array of plates, Solid State Commun.,
50(11), 999-1002, 1984.
cabilityandperipheralvalueto groundwater
hydrology,and
have therefore not been mentioned in this review. For the

Charlaix, E., E. Guyon, and S. Roux, Permeability of a random

array of fracturesof widely varying apertures, Transp.Porous


samereasons,the many publicationsin related fields, such
Media, 2, 31-43, 1987.
aseffectivemediumtheoryandfractalanalysesof flow, have Chatzis,I., andF. A. L. Dullien,Modellingof porestructureby2-D
been excluded.

and 3-D networks with applicationto sandstones,J. Can. Pet.

Technol., 16, 97-108, 1977.


It is hopedthatthis overviewwill providethe readerwith
an understanding
of fundamentalconceptsin perolation Chatzis, I., and F. A. L. Dullien, The modeling of mercury

theoryas it pertainsto groundwater


hydrology,andthat it
will motivateadditionalresearchon relatedproblemsin
hydrologywithinthis promisingframework.

porosimetryand the relative permeabilityof mercuryin sandstonesusingpercolation


theory,Int. Chem.Eng.,25, 47-66,1985.

Chelidze, T. L., Percolation and fracture, Phys. Earth Planet.


Inter., 28, 93-10!, 1982.

Chen, J.-D., and D. Wilkinson, Pore-scaleviscousfingeringin


porous media, Phys. Rev. Lett., 55(18), 1892-1895, 1985.
REFERENCES

Coskuner,G., andR. G. Bentsen,An extendedtheoryto predictthe


onsetof viscousinstabilitiesfor miscibledisplacements
in porous

Aharony,
A., Percolation,
inDirections
in Condensed
MatterPhys- media, Transp. Porous Media, 5,473-490, 1990.
ics, editedby G. GrinsteinandG. Mazenko,pp. 1-50, Word Cushman,J. H., On diffusionin fractal porous media, Water
Scientific,Singapore,1986.

Resour. Res., 27(4), 643-644, 1991.

Ahn,J., Y. Furuhama,
Y. Li, andA. Suzuki,Analysis
of radionu- David, C., Y. Gueguen,and G. Pampoukis,Effectivemedium
clidetransport
through
fracture
networks
bypercolation
theory, theoryandnetwork
theoryapplied
to thetransport
properties
of
J. Nucl. Sci. Technol.,28(5), 433-446, 1991.

rock, J. Geophys.Res., 95(B5), 6993-7005, 1990.

Alon, U., A. Drory, and I. Balberg,Systematicderivationof DeGregoria,


A. J.,A predictive
MonteCarlosimulation
oftwo-fluid
percolation
thresholds
in continuum
systems,
Phys.Rev.A, 42(8),
flowthrough
porousmediaat finitemobilityratio,Phys.Fluids,
4634-4638, 1990.

28(10), 2933-2935, 1985.

Alon,U., I. Balberg,
andA. Drory,New,heuristic,
percolationDeutsch,
C., Calculating
effectiveabsolute
permeability
in sandcriterion
for continuum
systems,
Phys.Rev.Lett.,66(22),2879stone/shale
sequences,
SPEForm.Eval.,4(Sept.),343-348,989.
2882, 199!.
Diaz,C. E., I. Chatzis,andF. A. L., Dullien,Simulatkm
of

BERKOWITZ
ANDBALBERG:
PERCOLATION
THEORY

capillarypressurecurvesusingbondcorrelatedsitepercolation
on a simplecubicnetwork,Transp.PorousMedia, 2, 215-240,
1987.

Dienes,J. K., Permeability,percolationand statisticalcrack mechanics,in Issuesin Rock Mechanics, editedby R. E. Goodman

793

worksbasedon regularlattice percolationand equivalentmedia


theories,d. Geophys.Res., 95(B 13), 21,565-21,581, 1990.
Imdakm,A. O., and M. Sahimi, Computersimulationof particle
transport processesin flow through porous media, Chem. Eng.
Sci, 4618), 1977-1983, 1991.

and F. E. Heuze, pp. 86-94, AmericanInstitute of Mining, Jerauld,G. R., and S. J. Salter, The effectof pore-structure
on
Metallurgicaland Petroleum Engineers, New York, 1982.
hysteresisin relativepermeabilityand capillarypressure:PoreDoyen,P.M., Permeability,conductivity,and pore geometryof
level modeling,Transp.PorousMedia, 5, 103-151, 19.).
sandstone,J. Geophys. Res., 93(B7), 7729-7740, 1988.

Drory,A., I. Balberg,U. Alon, andB. Berkowitz,Analyticderivationof percolationthresholdsin anisotropicsystemsof permeable


objects,Phys.Rev. A, 43(12), 6604-6612,1991.
Dullien,F. A. L., Porous Media: Fluid Transportand Pore Structure, Academic, San Diego, Calif., 1979.

Englman,R., Y. Gur, and Z. Jaeger,Fluid flow througha crack


network in rocks, J. Appl. Mech., 50, 707-711, 1983.
Essam,J. W., Percolation and cluster size, in Phase Transitionsand

CriticalPhenomena,vol. 2, edited by C. Domb and M. S. Green,

pp. 197-270,Academic,SanDiego, Calif., 1972.


Fair, G. M., and L. P. Hatch, Fundamentalfactorsgoverningthe
streamlineflow of water through sand, J. Am. Water Works

Jerauld, G. R., J. C. Hatfield, L. E. Scriven, and H. T. Davis,

Percolationand conductionon Voronoi and triangularnetworks'


A casestudyin topologicaldisorder,J. Phys. C SolidState Phys.,
17(9), 1519-1529, 1984a.

Jerauld, G. R., L. E. Scriven, and H. T. Davis, Percolation and

conductionon the 3D Voronoi and regular networks' A second


case study in topologicaldisorder, J. Phys. C Solid State Phys.,
17(19), 3429-3439, 1984b.
Katz, A. J., and A. H. Thompson, Prediction of rock electrical
conductivityfrom mercury injection measurements,J. Geophys.
Res., 92(B1), 599-607, 1987.

Klemeg, V., Dilettantismin hydrology: Transition or destiny?,

Water Resour. Res., 22(9), !77S-188S, 1986.


Kogut, P.M., and J. Straley, Distribution-induced non-universality
Fatt, I., The network model of porousmedia,I, Capillarypressure
of the percolationconductivityexponents,J. Phys. C Solid State
characteristics,Trans. Am. Inst. Min. Metall. Pet. Eng., 207,
Phys., 12(11), 2151-2159, 1979.
144-159, 1956a.
Koplik, J., S. Redner,and D. Wilkinson, Transportand dispersion
Fart,I., The network model of porousmedia, II, Dynamicproperin random networks with percolation disorder, Phys. Rev. A,
ties of a single size tube network, Trans. Am. Inst. Min. Metall.
37(7), 2619-2636, 1988.
Pet. Eng., 207, 160-163, 1956b.
Kozeny,J., f2berkapillareLeitungdes Wassersim Boden,Sitzungsber. Akad. Wiss. Wien Math. Naturwiss. Kl., Abt. I, 136,
Fatt, I., The network model of porousmedia, III, Dynamicproper-

Assoc., 25, 1551-1565, 1933.

ties of networks with tube radius distribution, Trans. Am. Inst.

Min. Metall. Pet. Eng., 207, 164-177, 1956c.


Feng,S., B. I. Halperin, and P. N. Sen, Transport propertiesof
continuumsystemsnear the percolation threshold,Phys. Rev. B,
35(1), 197-214, 1987.
Ferrand,L. A., and M. A. Celia, The effect of heterogeneityon the
drainage capillary pressure-saturation relation, Water Resour.
Res., 28(3), 859-870, 1992.
Ferrand, L. A., M. A. Celia, and W. E. Soll, Percolation-based

modelsfor pore-to-lab scale calculations in multifluid porous


media, in Dynamics of Fluids in Hierarchical Porous Media,
editedby J. H. Cushman, pp. 463-483, Academic, San Diego,
Calif., 1990.

271-306, 1927.

Krumbein, W. C., and G. D. Monk, Permeability as a function of the


size parametersof unconsolidatedsands, Trans. Am. Inst. Min.
Metall. Eng., 151, 153-163, 1943.
Kueper, B. H., and E. O. Frind, An overview of immiscible
fingeringin porousmedia,J. Contain. ttydrol., 2, 95-!!0, 1988.
Kueper, B. H., and D. B. McWhorter, The use of macroscopic
percolation theory to construct large-scale capillary pressure
curves, Water Resour. Res. 28(9), ,t'5 'an 1992.
Larson, R. G., L. E. Scriven, and H. T. Davis, Percolation theor>'
of residualphasesin porousmedia, Nature, 268(4), 409-413, 1977.
Larson, R. G., L. E. Scriven, and H. T. Davis, Percolation theor>'
of two phaseflow in porousmedia, Chem. Eng. Sci., 36, 57-73,

Flory, P. J., Molecular size distribution in three dimensionalpoly1981a.


mers, J. Am. Chem. Soc., 63, 3083-3100, 1941.
Larson, R. G., H. T. Davis, and L. E. Scriven, Displacement of
Frisch,H. L., and J. M. HammersIcy, Percolationprocessesand
residualnonwettingfluidfrom porous media, Chem. Eng. Sci., 36,
relatedtopics, J. Soc. Ind. Appl. Math., II, 894-918, 1963.
75-85, 1981 b.
Gawlinski, E. T., and S. Redner, Monte-Carlo renormalisation Lenormand, R., Pattern growth and fluid displacementsthrough
groupfor continuum percolation with excluded-volumeinteracporousmedia,PhysicaA Amsterdam, 140, 114-123, 1986.
tions,J. Phys. A Math. Gen., 16(5), 1063-1071, 1983.
Lenormand, R., and C. Zarcone, Invasion percolation in an etched
Gefen,Y., A. Aharony, and S. Alexander, Anomalousdiffusionon
network: Measurement of a fractal dimension Phys. Re,'. ett.,
percolatingclusters,Phys. Rev. Lett., 50(1), 77-79, 1983.
54(20), 2226-2229, 1985a.
Golden,J. M., Percolationtheoryand modelsof unsaturated
porous Lenormand,R., and C. Zarcone, Two-phase flow experimentsin a
media, Water Resour. Res., 16(1), 201-209, 1980.

two-dimensional
permeablemedium,PCH PhysicoChem.HydroGueguen,Y., and J. Dienes, Transport propertiesof rocks from
dyn., 65/6), 497-506, 1985b.
statisticsand percolation,Math. Geol., 21(1), !-13, 1989.
Lenormand, R., and C. Zarcone, Capillary fingering: Percolation
Gueguen,Y., C. David, and M. Darot, Modelsand time constants
and fractal dimension, Transp. Porous Media, 4, 599-612, 1989.
for permeabilityevolution,Geophys.Res. Lett., 13(5),460-463,
Long, J. C. S., and P. A. Witherspoon,The relationshipof the
1986.
degreeof interconnection
to permeabilityin fracturenetworks,J.
Gueguen,Y., C. David, and P. Gavrilenko, Percolationnetworks
Geophys.Res., 90(B41,3087-3098, 1985.
and fluid transport in the crust, Geophys.Res. Lett., I8(5),
931-934, 1991.

Madden, T. R., Random networks and mixing laws, Geophysics,

41(6A), 1104-1125, 1976.


Halperin,
B. I., S. Feng,andP. N. Sen,Differences
betweenlattice
andcontinuum
percolationtransportexponents,
Phys.Rev. Lett., Maier, R., and W. G. Laidlaw, Fluid percolation in bond-site

54(22), 2391-2394, 1985.

size-correlatedthree-dimensionalnetworks, Transp. Porous Mediet,5,421-428, 1990.

Hardy,H. H., Jots---Amathematical


modelof microscopic
fluid
Marsily,G. de, Flow andtransportin fracturedrocks:Connectivity
flowin porousmedia,Transp.PorousMedia, 5, 27-47, 1990.
and scale effect, Mere. Int. Assoc. Hydrogeol., 1712}, 267-277,
Harris,
C. K., Application
ofgeneralised
effective-medium
theoryto
1985.
transport
in porousmedia,Transp.PorousMedia,5, 517-542,
1990.

National ResearchCouncil, Opportunities in the Hydrologic Sci-

ences,348 pp., National Academy Press, Washington,D.C.,


Havlin,S., and D. Ben-Avraham,Diffusionin disorderedmedia,
1991.
Adv.Phys., 36(6), 695-798, 1987.
Heiba,A. A., M. Sahimi,L. E. Scriven,andH. T. Davis,Perco- Paterson,L., Diffusion-limitedaggregationand two-fluid displacelationtheoryof two-phase
relativepermeability,
SPEReservoir mentsin porousmedia,Phys.Rev. Left., 5218), 1621-1624,1984.
Eng., 7(1), 123-132, 1992.

Hestir,K., and J. C. S. Long, Analyticalexpressions


for the
permeability
of randomtwo-dimensional
Poissonfracturenet-

Reches,Z., Networks or'shearfaults in the field and in experiments,


in Proceedingsof the International ConiC,
fence or1Fragmentation, Form and Flow in Frectured Media, Ann. !sr. Ptlys. Sot.,

794

BERKOWITZAND BALBERG.'PERCOLATIONTHEORY

vol. 8, edited by R. Englman and Z. Jaeger, pp. 42-51, Adam Wagner,


N., andi. Balberg,
Anomalous
diffusion
andcontinuum
percolation,J. Stat. Phys., 49(1/2), 369-382, 1987.
Hilger, Bristol, England, 1986.
Renault, P., The effectof spatiallycorrelatedblocking-upof some Wardlaw,
N. C., Y. Li, andD. Forbes,Pore-throat
sizecorrelation
bondsor nodesof a network on the percolationthreshold,Transp.
from capillarypressurecurves,Transp.PorousMedia, 2, 597.
Porous Media, 6, 451--468, 1991.

611, 1987.

Robinson,P. C., Connectivity of fracture systems--A percolation Wheatcraft,S. W., and S. W. Tyler, An explanationof scale.
theory approach,J. Phys. A Math. Gen., 16(3), 605-614, 1983.
dependent
dispersivity
inheterogeneous
aquifers
using
concepts
Robinson, P. C., Numerical calculationsof critical densitiesfor lines

and planes,J. Phys. A Math. Gen., 17(14), 2823-2830, 1984.


Ross, B., Dispersionin fractal fracture networks, Water Resour.
Res., 22(5), 823-827, 1986.

Sahimi, M., Hydrodynamicdispersionnear the percolationthreshold: Scalingand probability densities,J. Phys. A Math. Gen.,
20(18), L1293-LI298,

1987.

of fractalgeometry,WaterResour.Res., 24(4), 566-578, 1988.

Wilke,S.,E. Guyon,
andG. deMarsily,
Waterpenetration
through
fractured
rocks:Testof a tridimensional
percolation
description,
Math. Geol., 17(1), 17-27, 1985.

Wilkinson,
D., Percolation
effectsin immiscible
displacement,
Phys. Rev. A, 34(2), 1380-1391, 1986.

Wilkinson,D., and M. Barsony,Monte Carlo study of invasion

Sahimi, M., and A. O. Imdakm, The effect of morphological


percolation
clustersin two and three dimensions,
J. Phys.A
Math. Gen., 17(3), L129-L135, 1984.
disorder on hydrodynamicdispersionin flow through porous
media, J. Phys. A Math. Gen., 21(19), 3833-3870, 1988.
Wilkinson,D., and J. F. Willemsen, Invasion percolation:A new
Sahimi, M., and T. T. Tsotsis, A percolation model of catalyst
form of percolation
theory,J. Phys. A Math. Gen., 16(14},
3365-3376, 1983.
deactivation by site coverage and pore blockage, J. Catal., 96,
552-562, 1985.

Sahimi, M., B. D. Hughes, L. E. Scriven, and H. T. Davis,


Stochastic transport in disorderedsystems, J. Chern. Phys.,
78(11), 6849-6864, 1983.
Sahimi, M., G. R. Gavalas, and T. T. Tsotsis, Statistical and
continuummodelsof fluid-solidreactionsin porousmedia, Chem.
Eng. Sci., 45(6), 1443-1502, 1990.

Witten,T. A., and L. M. Sander,Diffusionlimitedaggregation,


Phys. Rex,. B, 27(9), 5686-5697, 1983.

Wong,P.-Z., The statistical


physicsof sedimentary
rock,Phys.
Today, 41(12), 24-32, 1988.

Wong, P.-Z., J. Koplik, and J.P. Tomanic, Conductivityand


permeabilityof rocks,Phys. Rev. B, 30(11), 6606-6614,1984.
Yanuka,M., Percolation
processes
andporousmedia,II, Computer
calculationsof percolationprobabilitiesand clusterformation,J.

Sen, P. N., J. N. Roberts,and B. I. Halperin, Nonuniversalcritical


exponentsfor transportin percolatingsystemswith a distribution
Colloid Interface Sci., 127(1), 35-47, 1989a.
of bond strengths,Phys. Rev. B, 32(5), 3306-3308, 1985.
Yanuka,M., Percolationprocesses
and porousmedia,III, PredicShante,V. K. S., andS. Kirkpatrick,An introduction
to percolation
tionof the capillaryhysteresis
loopfrom geometrical
andtopo.
theory, Adv. Phys., 20(85), 325-357, 1971.
logicalinformationof porespace,J. ColloidInterfaceSci., I27(1),

48-57, 1989b.
Silliman,S. E., The influenceof griddiscretization
on the percolation probabilitywithin discreterandomfields,J. Hydrol., 113, Yanuka,M., Percolation
theoryapproachto transportphenomena
177-191, 1990.

Silliman,S. E., andA. L. Wright,Stochastic


analysisof pathsof
highhydraulicconductivityin porousmedia,WaterResour.Res.,
24(11), 1901-1910, 1988.

Stauffer,D., Introduction
to PercoIationTheory,124pp., Taylor
and Francis, London, 1985.

Stockmayer,
W. H., Theoryof molecular
sizedistribution
andgel
formationin branched-chain
polymers,J. Chem.Phys.,11, 45-55,

in porousmedia, Transp. PorousMedia, 7, 265-282, 1992.


Yanuka, M., F. A. L. Dullien, and D. E. Elrick, Percolation
processesand porous media, I, Geometricaland topological

modelof porousmediausinga three-dimensional


joint poresize
distribution,J. ColloidInterface Sci., 112(1), 24-41, 1986.
Yortsos,Y. C., andM. Sharma,Applicationof percolation
theoryto
noncatalyticgas-solidreactions,Am. Inst Chem. Eng. J., 32,
46-55, 1986.

1943.

Thompson,
A. H., A. J. Katz,andC. E. Krohn,Themicrogeometry I. Balberg,RacahInstituteof Physics,HebrewUniversity,91904
andtransport
properties
of sedimentary
rock,Adv.Phys.,36(5), Jerusalem, Israel.
625-694, 1987.

B. Berkowitz,Departmentof GeologicalSciences,Universityof

Torelli, L., Computersimulationof the dispersion


phenomena BritishColumbia,6339StoresRoad, Vancouver, British Columbia,
occurring
duringflowthroughporousmedia,usinga randommaze
model,Pure Appl. Geophys.,96(4), 75-88, 1972.
Torelli, L., and A. E. Scheidegger,Randommaze modelsof flow
throughporousmedia,PureAppl. Geophys.,89(6),32-44, 1971.

Vere-Jones,
D., Statisticaltheoriesof crackpropagation,
Math.
Geol., 9(5), 455-481, 1977.

Canada V6T

1Z4.

(Received April 10, 1992;


revised October 19, 1992;
accepted November 13, 1992.)

All in-text references underlined in blue are linked to publications on ResearchGate, letting you access and read them immediately.

You might also like