Percolation Theory and Network Modeling Application

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/226215505
Percolation Theory and Network Modeling Applications in Soil Physics
Article in Surveys in Geophysics · January 1998

DOI: 10.1023/A:1006590500229
CITATIONS READS
325 3,132
2 authors, including:
Brian Berkowitz
Weizmann Institute of Science
379 PUBLICATIONS 16,667 CITATIONS
SEE PROFILE
All content following this page was uploaded by Brian Berkowitz on 13 June 2016.
The user has requested enhancement of the downloaded file.

PERCOLATION THEORY AND NETWORK MODELING
APPLICATIONS IN SOIL PHYSICS
BRIAN BERKOWITZ1 and ROBERT P. EWING2

1
Department of Environmental Sciences and Energy Research, Weizmann Institute of Science,
76100 Rehovot, Israel
2
USDA-ARS National Soil Tilth Laboratory, Ames, Iowa, USA (Now at: Department of Agronomy,
Iowa State University, Ames, Iowa 50011, USA)
Abstract. The application of percolation theory to porous media is closely tied to network mod-
els. A network model is a detailed model of a porous medium, generally incorporating pore-scale
descriptions of the medium and the physics of pore-scale events. Network models and percolation
theory are complementary: while network models have yielded insight into behavior at the pore scale,
percolation theory has shed light, at the larger scale, on the nature and effects of randomness in
porous media. This review discusses some basic aspects of percolation theory and its applications,
and explores work that explicitly links percolation theory to porous media using network models.
We then examine assumptions behind percolation theory and discuss how network models can be
adapted to capture the physics of water, air and solute movement in soils. Finally, we look at some
current work relating percolation theory and network models to soils.
Key words: Percolation theory, invasion percolation, network models, porous media, soil physics
1. Introduction
Percolation theory is a branch of probability theory dealing with properties of

random media. Originally conceived as dealing with crystals, mazes and random
media in general (Broadbent and Hammersley, 1957), it now appears in such
fields as petroleum engineering, hydrology, fractal mathematics, and the physics
of magnetic induction and phase transitions. As explained by the originators of
percolation theory (Broadbent and Hammersley, 1957), percolation and diffusion,
for example, can both be seen as dealing with fluids in a medium (in a very general
sense), but they differ in their emphasis: in diffusion, the randomness “belongs
to” the fluid, while in percolation theory, the emphasis is on the randomness of
the medium. Several reviews of percolation theory in general (e.g., Shante and
Kirkpatrick, 1971; Bunde and Havlin, 1991; Havlin and Bunde, 1991; Stauffer
and Aharony, 1992) and in relation to porous media in particular (e.g., Berkowitz
and Balberg, 1993) have been published recently. Percolation theory also figures
prominently in books on porous media by Dullien (1992), Adler (1992) and Sahimi
(1995).
The history of percolation theory as applied to porous media is closely tied
to network models. A network model is a detailed model of a porous medium,
generally incorporating pore-scale descriptions of the medium and the physics of
Surveys in Geophysics 19: 23–72, 1998.

c 1998 Kluwer Academic Publishers. Printed in the Netherlands.
24 BRIAN BERKOWITZ AND ROBERT P. EWING
pore-scale events. Network models and percolation theory are complementary:

while network models have yielded insight into behavior at the pore scale, percola-
tion theory has shed light on the larger scale behavior, accounting for randomness
in porous medium geometry, fluid properties, and their interplay. In each case,
behaviors are seen at the large scale (e.g., an assemblage of pores) that are not nec-
essarily predictable at the scale of the individual pore. Network models and their
application to porous media were recently reviewed by Ferrand et al. (1990) and
Ferrand and Celia (1992). However, the emphasis of their reviews is on network
models rather than percolation theory, and it is important to differentiate between
the use of pore-scale network models alone, and their use in conjunction with
percolation theory.
The objectives of this review are, first, to provide a useful summary and reference
work for soil scientists wanting to learn more about percolation theory and network
models and their application to soils, and second, to expose some of the many
remaining issues and challenges in this field. This review will first briefly sketch
some basics of percolation theory, and then explore some mathematical aspects.
We then discuss work that explicitly links percolation theory to porous media, often
using network models, and consider some applications in more depth. We examine
the assumptions behind percolation theory and discuss how network models can be
adapted to capture the physics of water, air, and solute movement in soils. Finally,
we look at some current work relating percolation theory and network models to
soils.
Many basic ideas in percolation can be illustrated quite easily. Suppose we have
a large array of squares (Figure 1a). Let us call the line intersections “sites”, and the
segments connecting them “bonds”. In a square lattice one bond is connected to six
nearest neighbor bonds, while a site has only four nearest neighbor sites. Suppose
that each site exists in only two possible states, ‘empty’ or ‘open’ (or ‘allowed’;
there is no universal terminology to describe these states, and one could just as
easily say ‘on’ or ‘off’), with open sites being denoted by the presence of a large
dot on the intersection, and suppose further that whether a site is open or empty is
random and independent of its neighbors. A bond is assumed to exist between each
pair of nearest neighbor sites on the lattice. If half the sites are open (Figure 1b),
we see that open sites tend to group into clusters of many shapes and sizes. We can
refer to these clusters by their size, i.e., a single open site with no immediate open
neighbors is a 1-cluster; two adjacent open sites with no open neighbors form a
2-cluster, and so on.
If the probability p of a site being open increases to 2/3 (Figure 1c), several
things happen. Most important, at some probability between 1/2 and 2/3, many
of the sites become joined into one giant cluster that spans the entire array both
vertically and horizontally. The probability at which this happens (approximately
0.593 for the square lattice sites) is called the critical probability pc , also known as
the percolation threshold. If we imagine that fluid can flow only through bonds that
connect open sites (drawn more thickly in Figure 1b,c), then below the threshold the
PERCOLATION THEORY AND NETWORK MODELING APPLICATIONS IN SOIL PHYSICS 25
lattice will have zero conductivity, while above the threshold conductivity will rise
as the probability increases. Hence, there is a strong relation between connectivity
of the (so-called microscopic) elements of the system and the physical properties of
the entire (or so-called macroscopic) system. Second, as the proportion of open sites
increases, the proportion of empty sites that have open neighbors also increases.
Third, once p > pc , if we run fluid across the lattice through bonds connecting
open sites, we will find that some of the bonds (the backbone) have fluid flowing
through them, while others are simply isolated dead-ends or dangling branches; the
proportion of these branches varies as a function of p. The proportion of open sites
belonging to the infinite cluster - in other words, the proportion of open sites that
would be penetrated by a fluid at the lattice boundaries – is called the accessibility
function. Fourth, the clusters grow larger (and merge) with an increase in p. And
finally, the reverse is happening to the empty sites: they are being divided and
squeezed into smaller clusters as the probability of open sites increases.
The above example varies only the proportion of open sites, so it is called site
percolation. An analogous procedure, called bond percolation, varies the proportion
of open bonds. The two are not interchangeable: there is no simple formula that
will predict bond percolation from site percolation (although bond problems can
be “mapped onto” site problems), and, in fact, pc for site percolation is always
greater than pc for bond percolation (Fisher, 1961). We will encounter a physical
significance to the distinction between bond and site percolation in Section 3 of
this review.
The ragged (“ramified”) edge of a cluster is reminiscent of fractals, and indeed,
clusters near the percolation threshold are known to be fractal (Stauffer and
Aharony, 1992). In fact, the fractal dimension of a percolating cluster is the same
(D 1.896) whether the percolation takes place on square, triangular, honeycomb,
Voronoi, or some other kinds of 2D lattices (Figures 2a–d) which differ widely in
their connectivity (which can be quantified by the coordination number z , defined
as the mean number of bonds per site). The fractal dimension does, however, change
with the dimensionality or embedding dimension d of the lattice: percolating clus-
ters on 2D lattices have different fractal dimensions than those on 3D lattices (such
as cubic lattices, Figure 2e). We will find other connections between percolation
theory and fractals in Sections 2 and 3.
The mathematics of percolation theory can be traced back to Flory (1941) and
Stockmayer (1943), who were trying to explain the process of gelation, in which
small molecules adhere to larger and larger ones to eventually form a gel. Because
these molecules do not loop back on themselves, but rather branch indefinitely, they
form a special family of lattices known as Cayley trees (Figure 2f), also called Bethe
lattices. These trees have only one possible path connecting any two sites, making
them much more amenable to mathematical treatment. As a consequence, much of
the mathematics of percolation theory was originally developed on Cayley trees.
However, these trees differ topologically from lattices which are multiply connected
Figure 1. (a): square lattice, (b): square lattice with 50% open sites, (c): square lattice with 67% open
sites. In (c), the backbone for flow from left to right is shown with dark bonds, while isolated clusters
and dangling branches are shown with light bonds.
hi
Figure 2. Examples of 2D lattices, with circles representing sites, and dark lines denoting bonds. (a):
square (z = 4), (b): triangular (z = 6), (c): honeycomb (z = 3), (d): Voronoi ( z = 6), (e): cubic (z =
6), and (f): Cayley tree with z = 3.
(having loops), like the pore space in a soil, so our discussion will focus on the
multiply connected lattices.
The form of percolation of interest to us was developed by Broadbent and
Hammersley (1957), who thought of it as a fluid (saturating what we called, above,
open sites or bonds) flowing into a maze (the medium). At the time, Broadbent was
involved in designing gas masks for use in coal mines (Hammersley and Welsh,
1980). These masks were to use porous carbon granules which would filter the
air, with cleaning occurring via surface adsorption. If the pore space within the
granules were sufficiently interconnected, gas could penetrate through them and
have access to the entire surface area; if not, the granules would not be sufficiently
permeable and so the mask would not work. Hammersley was meanwhile trying
to improve Monte Carlo methods for nuclear reactor design, and had organized
a symposium (Royal Statistical Society, 1954) to discuss this topic. He became
interested in Broadbent’s problem, and, according to Sahimi (1984), named it
percolation because he thought the fluid flowing into the maze resembled coffee
flowing through the grounds in a percolator.
2. The Mathematics of Percolation
Percolation theory and its many variants can be considered as part of a general
framework of statistical theories that deal with structural and transport properties
in porous media. The literature devoted to percolation theory is enormous, and
spans the fields of mathematics and statistical physics, as well as a variety of
engineering and other applied disciplines. As such, we shall introduce below only
some of the basic features of percolation theory; many other aspects of percolation
will be introduced in subsequent sections in the course of discussing its application.
2.1. BASICS AND DEFINITIONS
Many properties of a macroscopic system are essentially determined by the con-

nectivity of the system elements. The properties of a system which emerge at the
onset of macroscopic connectivity within it are known as percolation properties.
To illustrate the concept of connectivity, consider the square lattice of Figure 1c,
in terms of bond percolation. If, for example, the (connected) network of bonds
is fully saturated and conducts a fluid, and bonds are randomly removed from the
network, the intensity of flow between opposing sides of the network decreases.
The question then arises as to the possibility of determining, at one side of the
domain, the number of removed bonds by monitoring the fluid flow. In particular,
it is of interest to know the number of bonds that must be removed (randomly) in
order for no fluid to arrive at the side of the domain. The answer to the second
question is given by what is known as the percolation threshold. If the number
of bonds is denoted by N , and their number at the threshold is Nc (the “critical”
number), one can show (Balberg, 1987; Wong, 1984) that the volumetric fluid flow,
Q, will be determined by a power law of the form
Q / (N Nc) (1)
where is some (“critical”) exponent which can be found from theory and/or
computer simulation and/or experiment. Such a simple law holds for N relatively
close to Nc (typically, for N=Nc 2). For much larger N , other considerations can
be applied which in many cases lead to other power laws.
The problem described above, and the simple law given by Equation (1) that
governs its solution, apply to many kinds of systems, and as such, are much
more general than one might realize at first glance. The above formulation can be
used, for example, to describe flow of water in porous rocks, flow of electricity
in electrical networks and composite materials, penetration of termites in wood
and the spread of fire or epidemics. In a percolation problem, one asks three basic
questions: (a) What is the geometrical or physical property (e.g., Q) that is relevant
to the connectivity of the system under investigation?, (b) What is the threshold for
percolation (i.e., Nc or an equivalent quantity)?, and (c) What is the exponent that
describes the behavior of Q near Nc (i.e., )? One of the most interesting and useful
aspects of percolation theory is that many systems have the same . This means
that by finding for a simple and solvable model, one can predict the value of
for very complicated systems. This central property of percolation theory is known
as “universality”. On the other hand, the percolation threshold must be determined
separately for each system, although some general guidelines for its determination
are available.
Several simple geometrical and statistical concepts are used to quantitatively
describe percolation-type systems. To present these concepts, consider the so-called
square lattice shown in Figure 1b,c. Each point on the lattice represents a site that
is empty or open. A basic assumption of percolation theory is that the process
of assigning open sites is random, so that the probability of a site being open is
independent of the open/empty assignment of its neighbors or of any other site.
Of primary interest is characterization of the system under a given fraction of
open sites. For example, assume that there are N0 sites in the two-dimensional
square lattice, with N of them open, as shown in Figure 1b,c. Since the assignment
=
process is random, the probability p of each site being open is simply p N=N0 ;
this probability is well defined in the N0 ! 1 limit.
2.2. CLUSTER PROPERTIES
Another concept to be introduced is that of the “cluster”, which is simply a group of

connected sites. An important question that arises relates to the size distribution of
clusters that can be found when the lattice occupation probability is p; cluster size
distributions are important in the characterization of interconnected (conducting)
groupings of pores or fractures. Cluster sizes may vary, for example, as pores
or fractures become disconnected through mechanical or geochemical processes.
On a square lattice, to have a 1-cluster, there must be an open site (probability
p) surrounded by four empty sites (probability (1 p)4). Hence, the number of
( )
1-clusters will be N0 p 1 p 4 . Similarly, for 2-clusters, the number of such “left-
( )
right” clusters is N0 p2 1 p 6, and there is the same number of “up-down” clusters.
For the limit N0 ! 1, one may define the number of s-clusters per site, ns , where
s is the number of sites in the cluster (the “size” of the cluster). Thus, it follows
that n1 = ( ) = ( ) = (
p 1 p 4, n2 2p2 1 p 6, and n3 2p3 1 p 8 4p2 1 p 7 . ) + ( )
The calculation of ns for larger values of s, for other lattices, and for higher
dimensions, becomes increasingly complicated, but the basic “recipe” is apparent

from the above examples.
Following these examples, it is also clear that with increasing p, the probability
of finding larger clusters increases. In a finite lattice such as the one shown in
Figure 1c, there will be a p value that is sufficiently large to ensure that at least
one cluster connects the “bottom” and “top” (or the “left” and “right”) ends of the
lattice. This value of p, which is well-defined (say, in a computer experiment) when
N0 ! 1, is called the critical probability, and is denoted pc. The corresponding
largest cluster is called the percolation cluster, and the value of pc is known as the
lattice site percolation threshold. The regimes 0 p < pc and pc < p 1 are
called the regimes below and above the percolation threshold, respectively. One
can generally argue that there is at most one percolation cluster (e.g., Stauffer and
Aharony, 1992); this is obvious for the particular two-dimensional case illustrated
in Figure 1c. Above the percolation threshold (i.e., for p > pc ), one can also
determine the probability per site, P , that an open site belongs to the percolation

cluster. Since 0ns s (where 0 is the sum over all finite values of s) is the probability
per site that the site is open and belongs to a finite cluster, it follows that
P = p 0nss: (2)
Furthermore, it can be shown (Stauffer and Aharony, 1992) that the average

size of a finite cluster per site (above or below pc ) is 0 ns s2, which is the weighted
average of cluster sizes. Since one is interested in the connectivity between open
sites, the common definition of the average finite cluster size, S , is the average
cluster size per open site,
S = 0nss2=p: (3)
Since (as shown above) ns is a polynomial function of p, S can be expressed as

a power series of p. For example, in the square lattice with the expressions given
=
above for ns (s 1; 2; 3), one can readily (although with lengthy calculation) find
that S = 1 + 4p + 12p2 + 24p3 + 52p4 + 108p5 + 224p6 + 412p7 + 844p8 + 1528p9
+ . Following the general expectation (see Equation 1) of a power law behavior,
and the particular findings in simple cases (e.g., the one-dimensional lattice where
= ( ))
ns ps 1 p 2 , one expects that, near pc, it is possible to approximate a power
series such as the one shown above for S by a power law of the form
S / (pc p) (4)
where is an exponent. Indeed, an approximation such as Equation (4) (e.g., Alon

et al., 1990) yields for the same value as that obtained by computer simulation
techniques.
It is found that is independent of the particular lattice under consideration,
and depends only on the dimensionality of the system. In particular, one finds
that = 43/18 for two-dimensional systems, and 1.80 for three-dimensional
systems. (A convention in the percolation theory literature is that exact, theoretically
derived values are reported as fractions, while approximate values from computer
simulations are reported in decimal form). In contrast, the value of pc depends on
the particular lattice, but again the pc value determined by expressing the series
(e.g., Equation 3) with an asymptotic form for S (e.g., Equation 4) is the same
as that found by computer simulations for the same lattice (e.g., for the square
lattice, pc 0.593). Tables of values of pc and critical exponents for common
two- and three-dimensional lattices can be found, for example, in Stauffer and
Aharony (1992). Note that for p > pc , the average cluster size is infinite, but the
definitions given here are concerned with finite clusters, and Equation (4) describes
their divergence (i.e., approaching infinity) as p ! pc . It has been found, however,
that the same value describes S for p > pc , and thus Equation (4) holds in this
regime, except that one has to replace (pc p) by (p pc ).
So far, we have considered the statistical properties of the clusters. The next
property we discuss is the geometrical extent of the cluster. Suppose that the sites
=
of a cluster of size s are located at ri , where i 1; 2; : : :; s. The center of mass of
the cluster can be defined by r0 = ri =s, and thus the so-called radius of gyration
of the cluster, Rs , is defined by
Rs2 = (jri r0 j2 )=s: (5)
In other words, Rs can be related to the average distance between any two sites
within the same cluster. The average radius of gyration (over all finite clusters) per
open site, , can then be defined by
2 = 20Rs2nss2=0nss2: (6)
Thus, represents the average distance between any two cluster sites, averaged over
all finite clusters. The parameter is thus also called the correlation, or connectivity,
length. Similar to S , the extent of the average cluster diverges as p ! pc , so that a
behavior of the form
/ jp pcj (7)
can be expected. It has been shown by various approaches that = 4/3 for two-
dimensional systems, while 0.875 for three-dimensional systems. Two sites
are considered correlated if they are connected (i.e., if they belong to the same
cluster), and thus , which characterizes the average distance over which two sites
are connected, is called the “correlation length” of the system.
To characterize the rather ramified or ragged structure of these clusters, the
concepts of self-similarity, fractals and fractal dimension can be used. Objects such
as the Koch star and the Sierpinski gasket, which are defined by repetition of simple
construction rules, and are exactly self-similar over successive magnifications of
portions of the objects, can be called “exact fractals”. Non-exact fractals such as
percolation clusters are of course not expected to yield exactly the same picture
under successive magnifications. And yet, for any given picture, it is not possible
to tell at what magnification (i.e., scale) it was taken. However, as with the Koch
curve, there is a well-defined relation between cluster size s and the corresponding
cluster radius Rs , such that on average (over clusters of size s), one obtains
s / RsD (8)
for all (large enough) s. In particular, any large but finite portion of the percolation
cluster can be characterized by this D . Hence the percolation cluster is a “statistical
fractal”, to be distinguished from the “exact fractal” described above. The approach
is similar in 3D: the dimensionality of the Euclidean space is d = 3, while for the
( )
percolation cluster, it is D < 3 when p ! pc . In fact, noting that P / p pc , it has
been shown from purely theoretical arguments using power laws, and confirmed
experimentally by computer simulations, that D d = = . In the case of clusters
in two-dimensional space, D = 91/48 and = 5/36, while in three-dimensional
space, D 2.52 and 0.41. Note again that since and are universal, D is
universal, and as such is unique for every d (e.g., Stauffer and Aharony, 1992).
Thus, the use of percolation theory allows the quantitative introduction of scaling
approaches into the analysis of properties of heterogeneous porous and fractured
media. Due to the statistically self-similar (fractal) nature of percolation clusters,
which can be used to approximate heterogeneous media, properties arise that are
independent of scale, and can in principle be applied equally at the pore, laboratory
and field levels.
2.3. FLOW THROUGH PERCOLATION NETWORKS
We now proceed to the issue of conduction through the bonds. Obviously, flow
problems are of interest only for p > pc . Note that even in this regime, many of the
connected bonds will not have fluid flowing in them, since either they do not belong
to the percolating cluster, or they form “dangling branches” (analogous to dead-end
pores) which, although filled, do not conduct fluid. The ensemble of bonds through
which the fluid can flow is called the “backbone” of the system; when examining
single-phase flow problems, it is sufficient to consider only the backbone.
What is the structure of the backbone? Recall that the average diameter of the
finite clusters is . Since the backbone is a network that connects the ends of the
entire system, the average diameter of the “holes” (i.e., the finite clusters) in the
network must be of order . The network then consists of “nodes” (or junctions)
and "links" connecting the nodes (e.g., Stanley, 1977). As explained above, the
distance from one end of the “link” to the other is of the order of (i.e., the average
radius of possible embedded finite clusters), but its actual detailed (tortuous) length
is of course longer. With this understanding of the flow network geometry, it is now
possible to consider the quantitative relationship between the possible flow and the
probability of finding an open bond.
If the permeability of each bond in a network is known, then by use of Kirch-
hoff’s law for each closed loop (which requires that the algebraic sum of the fluxes
at each junction equal zero), the calculation of the overall system permeability,
K , in a finite system is straightforward (Balberg, 1986b; Berkowitz and Balberg,
1992). The value of K depends only on the connectivity of the system (i.e., on the
bond occupation p), the radii of the individual bonds, and the specific randomization
of the given case. This approach can be generalized to other types of networks,
and is thus useful for computer simulations of finite model systems. However,
while providing a complete solution, it has two major drawbacks. The first is that
it is necessary to know all the details of 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, considering the larger system that the model tries
to simulate.
Analysis of pores in rocks and soils requires an approach that circumvents these
drawbacks; this is where the universality of percolation theory becomes helpful.
Similar to power laws given by (4) and (7), scaling arguments indicate that the
critical behavior of the permeability K of an infinite system is defined by
K / (p pc) (9)
which is determined by the exponents and according to
= + (d 2): (10)
Conveniently, and can be determined in simple systems, and then used to

predict the behavior of systems for which no, or very few (usually global) details are
known. However, in order to have an equal sign in Equation (9), many more details
are required, so that one must resort to finite procedures such as those outlined
above using Kirchhoff’s law. Thus one of the main contributions of percolation
theory lies in its ability to predict the value of . Since, as pointed out above, can
be determined analytically (for two-dimensional systems) using a simple lattice
model, the quantity that still remains to be found, which is closely related to the
flow problem, is . Techniques for estimating , based on considerations of the
link structure, are discussed by Berkowitz and Balberg (1993).
2.4. CONTINUUM PERCOLATION
It is apparent that lattice systems, in general, are the exception rather than the rule
among systems in which the connectivity determines the behavior of a system. Most
real systems possess a much less “ordered” network, and a distribution of local bond
permeability values. This is of course the case in many geological porous media
Figure 3. An illustration of a continuum model. The line segments may be considered to be fractures
or channels which can conduct a fluid. This system is characterized by the number of segments in
its finite version, and by the number of bonds per segment in its infinite version. The network shown
here is just above the percolation threshold (after Berkowitz and Balberg, 1993).
such as fractured rocks. As an example, Figure 3 illustrates a computer-generated

sample of randomly-aligned line segments. This can be envisioned as an aerial
photograph or a cross-section of fractures or cracks in a formation, or a system
of natural channels. Percolation in such systems is called continuum percolation
(e.g., Halperin et al., 1985; Balberg, 1987), and the mathematics used to describe it
are similar, but not identical to, what we have presented thus far. The conspicuous
features that distinguish continuum percolation networks from lattice networks are
(i) the variable number of possible connecting bonds per site (or object), (ii) the
variation of the bond length, and (iii) the local variation in the bond direction. We
consider these aspects below.
The first question we ask pertains to the quantity that should be used here,
instead of p, so that a “proximity” parameter such as p pc can be defined. To
answer this question, reconsider now the sites and bonds in Figure 1c. In this lattice
structure, each site has z (= 4 in this figure) possible bonds. If a fraction p of these
bonds is open, the average number of bonds per site is B = pz. The percolation
=
threshold, or the critical number of bonds per site, can then be defined by Bc pc z .
Hence, since z is a constant, B Bc is as good a proximity parameter as p pc .
We can now return to the continuum percolation model of Figure 3 and ask what
the average B (i.e., number of intersecting segments per line segment, or fracture)
is, and then compute any physical property as a function of B Bc . It has been
assumed then, from the universality applicable to the various lattice systems, that
all the results previously derived for lattice percolation are also applicable to such
continuum systems, except that here the well-defined B Bc is used instead of the
lattice parameter p pc .
To show that Equation (1) describes the same critical behavior as the other
critical behaviors given above, i.e., that (up to a proportionality factor) N Nc is
the same proximity parameter as p pc and B Bc , the concept of the excluded
volume is now introduced (Balberg et al., 1984). The excluded volume (or area,
in a two-dimensional system) is defined as the volume around an object in which
the center of another object must be in order for them to overlap. For example, for
spheres or cubes of volume V , the excluded volume, Vex , is simply given by Vex =
8V . In cases where the excluded volume is different for two given objects or two
given intersections, the average excluded volume, hVex i, must be used. A simple
example of such a case is that of “sticks”, or fractures, as shown in Figure 3. If
all the fractures are of length l0 , and there is an angle of i j between two of
=
them, the corresponding excluded “volume” (area) is A l02 sin ji j j. If i and
j are randomly and uniformly distributed between =2 and =2, the average
( )
excluded “volume” is 2= l02 (Balberg et al., 1984). Now the number of objects
whose centers lie within the average excluded volume is hVex i, where is the
density of objects in the system. But this is also the average number of objects that
overlap, or are bound, to a given object (i.e., B ). Thus
B Bc = ( c)hVex i: (11)
( ) ( )
For a finite system (lattices or continuum), N Nc / c , and Equation
(1) can be used. However, for infinite systems (for which percolation theory gives
accurate results), p pc and B Bc should be used, respectively.
Comparison of Figures 1c and 3 reveals another significant difference between
lattice models and real systems. While the length of each bond is the same in
Figure 1, the flow in the “equal length fractures” in Figure 3 takes place along
segments of different length. Thus, there is a distribution of flow segments, and a
distribution of local hydraulic conductivity (k ) values. The question of interest is
whether this situation affects the values of that were derived for models in which
it was assumed that k is the same for all elements through which fluid flow occurs.
Since the percolation theory predictions are accurate close to the threshold (i.e.,
when is very large in comparison with the details of the system), a “well-behaved”
distribution of k values (e.g., Gaussian) will manifest itself by average parameters,
so that is the same as for lattices. The situation appears to be quite different,
however, if the distribution of k values diverges as k ! 0. Under such conditions,
the decrease in K as the percolation threshold is approached from above is due not
only to the decreasing number of flow paths, as in the lattice, but also to the fact
that the participating paths consist of elements with smaller and smaller k values
(unlike the lattice case). Such a distribution may yield a value that is different
from that of the non-diverging k value distributions. The behavior associated with
these different values is known as the “non-universal” behavior.
Historically, one of the fundamental results in continuum percolation involves
the relationship between the percolation threshold, c (i.e., some critical object
density) and porosity. For a system of non-overlapping spheres, wherein spheres
touch each other at single points, and the spheres are the “conducting pores”, it
is clear that c = cV , where c is the critical volume fraction that is occupied
by spheres, and V is the volume of each sphere. This quantity has been found to
be dimensionally invariant (i.e., insensitive to lattice structure). Generalizing this
concept to permeable spheres (i.e., pores), it was shown (Shante and Kirkpatrick,
1971) that the average critical number of bonds per site (which is sometimes called
the “average coordination number”), denoted Bc , is related to the critical fractional
pore volume c by
c = 1 exp ( Bc=8): (12)
It has been further shown that Bc is an invariant for groups of convex objects
(Balberg et al., 1984). The value of having information on the average number
of intersections of a pore with its neighbors is that it provides information on the
general average topological structure of a porous medium.
The concept of the excluded volume has been useful in developing general
empirical rules for the dependence of the percolation threshold c on the geometry
of the objects (i.e., pore shapes) making up the system, as well as on the macroscopic
properties, such as anisotropy, of the system (Balberg et al., 1984). More recently,
exact analytical calculations have been presented for the same purpose (Drory et al.,
1991). This concept, which accounts for “interactions” (i.e., intersections) between
objects is simply connected to Bc , for permeable objects, by the relation
Bc = chVex i (13)
where hVex i is the average of the excluded volumes of the objects (see also Equation
11). Balberg (1986a) then generalized Equation (12) to
c = 1 exp [ (BcV=hVex i)] (14)

for systems comprised of permeable convex objects of any kind. The relevance
of the excluded volume concept to practical systems of interest in porous media
problems is clear when considering that Equations (13) and (14) yield the critical
porosity applicable to porous media. For example, these equations, through their
dependency on V and Vex , allow determination of critical porosities for a variety of
systems, and can account for the fact that, in certain types of rocks, flow is present
although porosity is close to zero (if pores are considered to be thin and sheet-like,
i.e., there is a small V=hVex i ratio).
In fact, Equation (14) follows from the more fundamental relationship between
critical porosity and excluded volume (Balberg, 1986a), which, for an infinite
system, is given by
c = 1 exp ( c V ): (15)
Relationships between c and c can be derived from these expressions. In the case
of spherical pores, for example, it is well-known (Balberg, 1987) that c Vex = 2.8,
=
Vex =V 8 and c 0.29.
Another aspect of interest in random networks is the relationship between the
percolation correlation length, , and the size of a REV (representative elementary
volume). The general physical significance of (which characterizes the average
distance over which two sites are connected; see Equation 7) is that only for length
scales L is the system macroscopically homogeneous. Thus, the correlation
length is in a sense the basic macroscopic building block of the system. The REV of
a porous medium, associated with, for example, permeability, is defined by a scale
at which permeability fluctuations become negligible. As known from the density
dependence of (Equation 7), and shown by, for example, Robinson (1983) and
Charlaix et al. (1987), for a fracture network, the correlation length decreases as
the fracture density increases; thus, the REV size also decreases.
A natural question to ask is whether various porous media, and networks of
fractures and soil cracks, have a fractal structure. Fractal properties have been
identified for various porous materials, such as sandstone (e.g., Katz and Thompson,
1985), as well as for patterns of fracture and crack networks (e.g., Barton, 1995).
Study of fractal (and multifractal) properties of soils is also now receiving attention,
as is the problem of developing numerical methods for generation of realistic fractal
porous media.
This necessarily brief review has presented a few of the highlights of what
percolation theory has to offer to a mathematical treatment of porous media. The
theory offers a perspective on disordered systems that more conventional methods
clearly cannot match. Moreover, as will be seen below, this framework has provided
the basis for development of other, complementary, variants of percolation theory.
While the proximity of real porous rocks and soils to percolation thresholds is a
matter of question, the theory is flexible, being suited to working with discrete
pore and fracture system geometries, as well as a variety of flow and transport
phenomena.
3. Percolation Theory and Network Models of Porous Media
In this section, we give a brief history of the network model, trace the merging of
network models with percolation theory, and give some of the highlights of their
co-evolution.
3.1. EARLY NETWORK MODELS
Early conceptualizations of pore space topology involved considering bundles of

capillary tubes of varying radius (Figure 4a; see, e.g., Bear, 1972, for a review).
Due to the topological simplicity, essentially all flow properties of such systems
can be treated analytically, and numerous flow and transport phenomena can be
simulated successfully. However, such models neglect the fundamental topological
randomness of porous media. Analyses of pore structures indicate that the pore
space is characterized by a wide distribution of channel sizes, and that random
networks more closely represent actual systems of pores (e.g., Doyen, 1988). The
network model of porous media was developed by Fatt (1956a,b,c), who found
inadequate the existing conceptual models of his time, the bundle of parallel tubes
and the sphere pack. He argued that the sphere pack was still too complex for
a complete analysis, and the equations developed from it (such as the Kozeny-
Carmen equation) still had empirical coefficients. Meanwhile, the parallel tube
model was conceptually inadequate, structurally anisotropic, and it also needed to
be adjusted via empirical coefficients. Fatt (1956a) proposed a regular 2D lattice of
tubes with randomly assigned radii (Figure 4b), arguing that it was simple enough
to be analyzed, yet displayed enough of the complex behavior of porous media to
be useful.
Fatt (1956a) developed a drainage algorithm that starts with the network tubes
completely filled with the wetting fluid (typically water in the case of unsaturated
soil). As the capillary pressure is increased, air surrounding the network moves into
the largest tubes along the interface, which becomes larger and more complex as
more tubes drain. The pressure can be plotted against the proportion of the network
that is still water-filled to yield a drainage curve. Computer facilities at the time
were limited, so the drainage algorithm used paper and pencil accounting. Fatt
(1956a) compared drainage curves generated by draining different lattices (square,
triangular, honeycomb, double-hexagonal and bundle-of-tubes) with various pore
size distributions, and noted that differences between lattices were minor com-
pared to differences between pore size distributions. He found that the drainage
curves gave better predictions of the pore size distribution as the lattice connec-
tivity z increased. Finally, he derived a method for correcting the measured pore
size distribution to account for network effects such as the percolation threshold.
In subsequent papers, Fatt (1956b,c) measured relative permeability (unsaturated
hydraulic conductivity) of physical networks built of electrical resistors as they
were being “desorbed” following the algorithm developed in the first paper.
Figure 4. Evolution of pore space models. (a): parallel tubes, (b): tube network of Fatt (1956), (c):
ball-and-stick network of Chandler et al. (1982) and Koplik (1982), (d): 2D version of Toledo et al.’s
(1989) biconical pore network.
Several aspects of Fatt’s (1956a) work are related to percolation theory, and
many are still important today. While recognizing that real porous media are 3D,
he chose to work in 2D because it was simpler. The differences between 2D and
3D networks were not well understood until at least 1977 (Chatzis and Dullien),
and even today much important work is still done on 2D networks. Based on
observations of thin sections of sandstones, Fatt (1956a) made the assumption that
the length of a tube is inversely proportional to its radius. Interestingly, the opposite
assumption (length is proportional to radius), mistakenly attributed to Fatt (1956a)
by Mualem (1976), is incorporated into the widely used Mualem-van Genuchten
formulation (van Genuchten, 1980) for unsaturated flow. Fatt’s (1956a) drainage
algorithm, in a different guise, is virtually identical to a computational algorithm
suggested by Hammersley (1963), and the invasion percolation algorithm proposed
by Wilkinson and Willemsen in 1983 (see discussion below). Finally, Fatt (1956a)
assumed that, since the contact angle is 0 (perfect wetting), water can drain from
“trapped” pores via thin films. This assumption was controversial at the time (Rose
et al., 1956) and is still under discussion today (e.g., Blunt et al., 1992; Blunt and
Scher, 1995): depending on how trapping is handled, a variety of behaviors can be
seen (see Section 3.3).
Advances in network modeling were slow from Fatt (1956a,b,c) through the
1970s, at least partly because computer resources were limited. Most of the mod-
eling was done in the field of petroleum engineering, both because this was its field
of origin, and because petroleum engineers had easier access to computers than, for
example, agricultural scientists. The first computer-based network modeling was
apparently done on Illiac by Rose et al. (1956), who eventually studied 3D lattices
with at least 1728 nodes (Rose, 1957). Ksenzhek (1963) examined the effect of the
size of a 3D lattice on the drainage curve, and noted that lattices show a break-
through or bubbling pressure similar to real porous media. Nicholson (1968) used
a physical 3D network to study sorption and desorption isotherms, and remarked
that the presence of a network enhances hysteresis. Torelli and Scheidegger (1971)
referred to concepts from percolation theory and developed a network of uniform
diameter tubes for modeling dispersion; Torelli (1972) extended their work for lat-
tices with unequal tube diameters. Simon and Kelsey (1971, 1972) used networks
to simulate a waterflood (water displacing oil in an oil reservoir to enhance oil
recovery) and calculated its efficiency. Networks were also used to interpret mer-
cury porosimetry experiments (Androutsopoulos and Mann, 1979) and to explain,
at a pore scale, low oil recovery following a waterflood (Mann et al., 1981).
Broadbent and Hammersley’s (1957) seminal paper on percolation theory specif-
ically referred to porous media, but percolation theory came slowly to the attention
of hydrologists and soil scientists. In 1977, Levine et al. regretted that “[percolation]
theory has dealt mainly with abstract objects, such as bonds, sites and clusters, and
few direct applications have been made to flow through a porous medium.” The first
general application of percolation theory (without network models) in a journal read
by hydrologists and soil physicists may be Golden’s 1980 article, which proposed
using percolation mathematics both to predict unsaturated hydraulic conductivity
and to provide a more powerful language for discussing hysteresis via domain
concepts (e.g., Topp, 1971; Mualem and Dagan, 1975, and references therein).
Some of the earliest work connecting percolation theory and porous media
involves the phenomenon of oil blob creation and mobilization (Larson et al.,
1977). While this is not ordinarily of interest to soil scientists, we include it for
its applicability to soil pollution remediation as well as its historical interest. Oil
is often extracted using a waterflood, in which water is pumped into a series of
injection wells and the oil driven to and withdrawn from extraction wells. In the
process, blobs of oil are surrounded by water and so disconnected from the rest
of the oil (Chatzis et al., 1983). These blobs tend to stay in place, decreasing the
amount of oil that is recovered. Since large blobs are more readily swept along
by flowing water than small ones (Ng and Payatakes, 1980; Dias and Payatakes,
1986), the expected blob size distribution is of some interest. It turns out that the
size distribution, which depends on the connectivity of the pore space, can be
predicted using cluster size distributions given by percolation theory (Larson et al.,
1977).
The year 1977 saw, in journals occasionally read by porous media scientists,
the first solid links between percolation theory and network models. Chatzis and
Dullien (1977) stated that “penetration of the non-wetting phase into a network
of pores is closely analogous to the ‘bond’ problem of percolation theory,” while
Levine et al. (1977) noted that Fatt’s (1956a) network model was similar to a com-
putational method described by Hammersley (1963), and so it could, if sufficiently
large, be used to calculate percolation probabilities. Chatzis and Dullien (1977)
explicitly opened discussion on some of the assumptions of Fatt’s (1956a) model,
and implicitly made the connection that pore throats or constrictions are like the
bonds in percolation theory, while pore intersections or pore bodies are like sites.
The analogy between percolation on the one hand, and porous media processes on
the other, has since proven to be a very fruitful one, and yet, from the perspective
of 1977, it is not obvious that it should be so. There are many kinds of percolation
and porous medium processes, and the percolation process most analogous to the
drainage studied by Chatzis and Dullien (invasion percolation) had not even been
invented in 1977. We will not formally justify the analogy in this review, but will
simply point out that it is a natural one to make, given that each is based on discrete
units (pore, sites) and events (emptying, draining) that display on the aggregate
scale complex behaviors that are not apparent at the scale of the individual units.
Chatzis and Dullien’s (1977) article set the standard for much of the network
modeling that followed, so it is worth examining in more detail. They pointed out,
for example, the until then universal assumption that the pore intersections do not
have any volume of their own: in fact, most researchers nowadays attribute the
majority of the porosity to the pore bodies. They recalled the topological principle
that bicontinua (for our purposes, two continuous phases such as water and air)
cannot exist in two dimensions, so 3D simulations are necessary to reproduce
phenomena accurately. In addition, they pointed out that for a given coordination
number, both the percolation threshold and the pore accessibility function are
different in 2D and 3D. (Studies on the transition from 2D to 3D were published
by Silliman (1990), and will be discussed in Section 4.) Chatzis and Dullien
(1977) noted that the assumption that pore sizes are randomly distributed within
the network had never been justified; we will revisit this assumption in Section 4 of
this review. They showed that neither constant tube length, tube length proportional
to tube radius, nor tube length inversely proportional to tube radius gave good
agreement with experiment, and proposed that networks use tubes with bulges in
the middle. They demonstrated graphically that a minimum size network, of the
order of 40 40 in 2D and 20 20 20 in 3D, was required to give results
that reasonably approximated an infinite lattice. Finally, they noted that lower
coordination, z , gave later breakthrough, and coordination around 20 gave results
close to the bundle of tubes model; from this, they raised the important question,
how can one determine z for a given porous medium?
In a later article, Chatzis and Dullien (1981) pointed out that drainage and
imbibition were inherently different processes, in that drainage was controlled by
the pore throat radii, while imbibition was controlled by the pore body radii. This has
interesting implications in terms of percolation theory, as it suggests that drainage
is a bond percolation process, while imbibition is site percolation (see, e.g., Lane
et al., 1986). Upon further consideration, however, it is clear that a necessary, but
not sufficient, condition for this conclusion to be true is that the bonds and sites
must each be independently and randomly distributed throughout the medium. In
contrast, measurements of pore structure indicate that a correlation exists between
the size of a pore body and the adjacent pore throat radii; at the very least, the
radius of a pore throat must be smaller than the smallest of the two adjacent pore
bodies. As a consequence, Dullien and coworkers (e.g., Diaz et al., 1987; Kantzas
and Chatzis, 1988; Ioannidis and Chatzis, 1993) have since incorporated such a
correlation between the size of neighboring pore bonds and sites along with a
random distribution of site sizes across the network, resulting in a ’bond-correlated
site percolation’. Drainage can then be treated as a site percolation problem, due to
the assumption of randomly distributed site sizes, but the actual drainage process
is controlled by the size of the bonds connecting neighboring sites. This modeling
approach contributed substantially to the good agreement obtained between model
predictions and experimental measurements.
While some researchers continued to use Fatt’s convention of intersections
having no volume, after Chatzis and Dullien (1977) most started giving pore bodies
both size and volume. This practice was “formalized” by Koplik’s (1982) paper
on flow through networks, in which he justified the assumption that most of the
resistance to flow is in the narrow bonds, with the wider pore bodies contributing
relatively little resistance. A companion paper by Chandler et al. (1982) formalized
the convention for drainage; the two together had the effect of popularizing the
“ball-and-stick” model (Figure 4c) which is still widely used today.
The 1980s and early 1990s have seen an enormous increase in research using net-
work models linked with percolation theory. Following the lead of Fatt (1956a,b,c),
research focused mainly on relative permeability (also called unsaturated hydraulic
conductivity), and drainage and rewetting (porosimetry, immiscible displacement).
A third active area of research, growing out of the discussion between Fatt
(1956a,b,c) and Rose et al. (1956), was on the nature of the actual pore-scale
events, while a fourth area examined the structure of the porous medium itself. A
final area of research concerns random walks and dispersion.
3.2. PERMEABILITY
The calculation of the relative permeability of a porous medium solely from its
pore size distribution has been elusive, with statistical (e.g., Marshall, 1958) and
network (e.g., Farrell and Larson, 1972) approaches meeting only limited success.
An analogous problem exists in percolation theory: how does the conductivity
of a lattice change as bonds or sites are eliminated? As with Fatt (1956b), the
problem is commonly statistically mapped onto an electrical resistance analog; the
conductivity is then solved for, either with real resistance models or via Monte
Carlo simulations (see Section 2). Kirkpatrick (1971, 1973) derived an effective
medium theory (EMT) that gave good unsaturated conductivity estimates, using as
inputs the distribution function of conductors (derivable from pore size) and the
coordination number of the medium; this approach enjoys widespread use. Larson
et al. (1981) used Bethe lattices to derive expressions for two phase flow. Heiba
et al. (1982, 1984) presented an elegant combination of statistical and percolation
approaches that gave relative permeability for two- and three-phase flow. Because
of the extensive computer time and detailed knowledge of the pore structure that
are required, reliable estimates of saturated and relative permeability of individual
porous media are rare.
3.3. POROSIMETRY
Porosimetry, the characterization of a pore size distribution via drainage and imbibi-
tion pressure-saturation relationships, is a natural research area for both percolation
theory and network models. In percolation theory, the concept of invasion perco-
lation was developed (Lenormand and Bories, 1980; Wilkinson and Willemsen,
1983) to model the processes of drainage and imbibition. Invasion percolation is
applicable when considering how sites and bonds become filled, or occupied, by
a fluid continuum that penetrates (i.e., invades) the network from a well defined
source. In these instances, it is important to distinguish between a site or bond
which is ‘allowed’ (‘open’), and one that is actually ‘occupied’ by the invading
fluid. For example, in the case of a nonwetting fluid invading a pore network (i.e.,
drainage), pores that allow penetration may not actually become occupied, if the
paths leading to these pores contain pores that are too small to be penetrated at the
prevailing capillary pressure. Hence, this class of problems requires that we distin-
guish between allowed and occupied sites and bonds, and modify our percolation
theory framework accordingly.
Invasion and ordinary percolation are similar in that probabilities are assigned
to the sites or bonds randomly and independently. However, in invasion percola-
tion, unlike ordinary percolation, a (for example) unoccupied site (where we now
distinguish between ‘open’ and ‘occupied’) with no occupied neighbors cannot
become occupied; rather, unoccupied sites are “invaded” by neighboring occupied
sites (Figure 5). To put it another way, all occupied sites are ultimately connected
to each other through the occupation source, typically a side (of a 2D lattice) or
face (of a 3D lattice) from which the "invasion” spreads, like air moving into soil
pores through the soil surface during drainage. Each individual invasion changes
the interface; after each invasion, the new interface is scanned for the site or bond
with the lowest probability, which will be invaded next. This makes occupation
(of sites or bonds) subject to two rules: first, the probability must be correct (as in
ordinary percolation), and second, there must be a continuous pathway of occupied
sites or bonds, from the invasion source to the site or bond under consideration.
Wilkinson and Willemsen (1983) explicitly linked invasion percolation to slow
capillary processes in porous media, and clearly the analogy with porous media is
strengthened by this new variant. Following some initial doubts, it became clear
(Wilkinson and Barsony, 1984) that invasion percolation is in the same universality
class as ordinary percolation: most of the mathematical reasoning and results devel-
oped with ordinary percolation, such as the fractal dimension of the percolating
cluster, therefore apply to invasion percolation as well.
With invasion percolation came the concept of trapping (Lenormand and Bories,
1980; Wilkinson and Willemsen, 1983): once a cluster of unoccupied sites is
surrounded by occupied sites, it is “trapped” and the sites cannot be invaded. This
procedure, illustrated in Figure 5, is very sensitive to the order in which events
occur: one invasion can trap a large cluster of otherwise allowed sites. To extend
the analogy with porous media, trapping means that the fluids are incompressible
and no thin film flow is allowed, so once a cluster of water-filled pores (for example)
is surrounded by air during drainage, the water cannot escape and so air cannot
displace it. This is in contrast to Fatt’s (1956a) network model. It turns out that
trapping assumes greater importance in 2D than in 3D models (Dias and Wilkinson,
1986), and so it can often be ignored in 3D work.
An extensive body of work exists on wettability of the medium and its effect on
fluid migration and distribution. In the context of invasion processes, in particular,
it is important to recognize the key role of the wetting phase (e.g., Blunt et al., 1992;
Blunt and Scher, 1995; see also Section 3.4). For example, experimental evidence
demonstrates the mobility of the wetting phase, especially along crevices and pore
edges (e.g., Dullien et al., 1989; Dong and Chatzis, 1995). As a consequence, for
example, in a controlled infiltration process which proceeds sufficiently slowly
(e.g., using pressure or pump controls), individual pores are occupied (invaded)
independently of the presence of other pores in the path of the fluid. Cieplak and
Robbins (1990) and Koiller et al. (1992) noticed that, during drainage, there is a
transition from a fractal front to a relatively smooth front at some critical contact
Figure 5. A comparison of ordinary percolation, invasion percolation, and invasion percolation with
trapping. Invasion is from the top, and occupied sites are shaded. Numbers in the boxes at the top
denote the invasion probabilities of the individual sites, while numbers down the side refer to the
current occupation probability of the system. Sites marked with an “i” are invadable (sufficiently low
probability) but inaccessible. Sites marked with a “t” are invadable and accessible, but trapped. As the
occupation probability p increases, differences between ordinary and invasion percolation decrease,
while differences between invasion percolation with and without trapping increase.
angle whose value is dependent on the porosity. Sutanto (1991) has done some
elegant network modeling in conjunction with cryo-SEM examinations of fluid
distributions to examine the effects on slow displacement when the medium is
mixed-wet (randomly distributed oil-wet and water-wet surfaces, as often happens
in oil reservoirs); this work may be useful to investigators working with hydropho-
bic soils or organic pollutants in soils rich in organic matter. In general, a variety
of fluid migration and distribution behaviors can be seen, depending on how pore
geometry and topology, fluid and matrix properties, and boundary conditions (such
as invasion rate) affect trapping processes.
In terms of network modeling, invasion percolation had been used before it
was so named (e.g., Fatt, 1956a; Chandler et al., 1982), and it continued to be the
primary conceptual model for drainage and imbibition. Chatzis and Dullien (1985)
used 2D and 3D networks to model mercury intrusion from a sandstone sample and
to estimate the relative permeability curve. They obtained good agreement between
measured and simulated values when using cubic (z = 6) and tetrahedral (z = 4) 3D
lattices. More recently, effort has focused on modeling displacement in structured
rather than random media, and in the presence of other forces besides capillarity.
These will be discussed in Section 4 of this review.
The interpretation of mercury porosimetry data in light of percolation concepts
has been an active field of research, going back to Fatt’s (1956a) attempts to
reconstruct the pore size distribution from his simulated drainage curves. More
modern research along this line is presented by Yanuka (1989a,b) and Zhou and
Stenby (1993). Yanuka (1989a,b) used pore size distributions for a variety of porous
media, together with percolation theory, to calculate capillary pressure curves for
both drainage and imbibition (i.e., hysteresis loops). These predictions compared
favorably with available experimental data. A difficulty in this field of research
is that many factors are involved: the sample size, the coordination number, and
the degree of non-randomness (among other factors) all affect the retention curve
obtained from a given pore size distribution.
Mercury porosimetry is usually performed using injection to predetermined
pressures, followed by measurement of the volume of mercury injection. Such a
system yields discrete pressure versus saturation points which are often translated
into a smooth pore size distribution (Washburn, 1921). However, slow continuous
mercury injection, with a constant rate pressure increase and continuous monitor-
ing, reveals (Figure 6a) (Thompson et al., 1987) a "devil’s staircase” (Mandelbrot,
1983) of electrical resistance through the mercury, rather than a smooth curve, indi-
cating that the processes at work are not as smooth or continuous as our presentation
of the data.
When slow, constant rate volume injection is used, with a continuous recording
of the resulting pressure, the response (Figure 6b) (Yuan and Swanson, 1986) is
not so much a curve or staircase, as a jumble. The pressure alternately rises, as
the advancing mercury menisci are forced into narrow pore throats, then falls,
as the mercury enters a wider pore body, retracting mercury that was advancing
into tighter throats, as a new equilibrium is found. In other words, this sort of
porosimetry can be used to evaluate sizes of individual pores, and to distinguish
pore throat volume from pore body volume. Toledo et al. (1989) performed careful
network simulations that reproduced this more complex “devil’s jumble,” showing
how detailed information on the sample’s pore structure could be extracted from
Figure 6. (a): A “devil’s staircase” of electrical resistance through the mercury during pressure-
controlled mercury porosimetry (after Thompson et al., 1987), (b): A “devil’s jumble” of pressure
during volume-controlled porosimetry (after Toledo et al., 1989).
this new porosimetry method. Toledo et al.’s (1989) model is also important because
it shows a clear advantage of biconical pore throats (Figure 4d) over the earlier
ball-and-stick (Figure 4c) network model. Biconical throats (two cone sections
sharing the same center axis and overlapping near their points) were used both to
capture the converging-diverging nature of real pores, and because the cylindrical
tubes in the ball-and-stick model wrongly imply the presence of a stable interface
between connected pore bodies.
3.4. PORE-SCALE EVENTS
Attempts to model macroscopic behavior based on pore-scale events must necessar-

ily have a clear understanding of those events. While Fatt (1956a,b) had essentially
three events – complete replacement of one fluid by another (e.g., water by air),
drainage via thin film flow, and steady-state flow – the list has continued to expand.
Mohanty et al. (1980) modeled the phenomenon of choke-off (Figure 7), in which
the wetting fluid builds up at the narrow point of a pore throat, then spontaneously
coalesces to divide the non-wetting phase (oil in their case) in two. This is similar
to ordinary (non-invasion) bond percolation, in that it is the bonds that are cut,
and since the water is supplied by thin film flow, the cut bonds need not be part
of a visible or coherent front. Lenormand (1987) demonstrated a phenomenon he
called compact cluster growth, in which a choke-off appears to trigger neighboring
choke-offs, forming a saturated cluster ahead of the “wetting front”. While this
occurs primarily in micromodels with square tube-like pores, it may also apply
to porous media in which small pores are clustered together. Yuan and Swanson’s
(1986) and Toledo et al.’s (1989) porosimetry work, cited above, nicely illustrates
the existence of two distinct types of pore-scale displacements: reversible, as in
mercury advancing into a converging throat, and irreversible, such as a large pore
body suddenly being filled with mercury. This latter event is often referred to as
a Haines jump, after the soil scientist who first noticed that drainage can occur as
discrete events (Haines, 1930).
Pore level events are still under discussion, and new ones are added for both
theoretical reasons and in response to poor agreement between model and reality.
Blunt et al. (1992) show schematically how the trapping rules first proposed by
Rose et al. (1956) could potentially result in near-zero permeability, and thereby
propose a distinction between trapping in pore throats, and trapping in pore bodies.
Prat (1993), modeling drying of a porous medium via evaporation, found that the
drying patterns produced by invasion percolation with trapping did not reflect the
fact that drying was both driven and limited by diffusion. The trapping rule would
not dry water-filled clusters once they were surrounded, but Prat (1993) recognized
that the physics of the situation required that wetting fluid be transported through
the non-wetting phase. Transport was also limited by diffusion, necessitating the
calculation of the moisture gradient at each time step to account for different
evaporation rates at different distances from the “outside edge” of the medium.
More recently, Blunt and Scher (1995) provide a detailed investigation of wetting
behavior at the pore scale, accounting for competing flow in the centers of the pore
spaces and along crevices, and compare predictions of residual saturations based
on percolation theory.
Figure 7. “Choke-off” of a pore throat by the wetting fluid (after Mohanty et al., 1980). As the film
of wetting fluid thickens (II), the portion of the film at the throat becomes unstable and collapses (III),
blocking the pore to passage of non-wetting fluid.
3.5. PORE STRUCTURE
All of the network modeling discussed above was done on regular lattices with a
fairly limited range of coordination numbers, and the critical reader may wonder
whether this is applicable to natural porous media, with their chaotic structure and
microscopic pores whose connections defy study. Mohanty et al. (1980) pointed out
that, in principle, a porous medium could be completely mapped onto a network,
i.e. reconstructed symbolically in terms of connections, body radii and throat radii.
However, there are two somewhat arbitrary limits to the mapping (Mohanty et al.,
1980): the size of the smallest pore one was willing to map, and the definition
one used to define individual “pores”. They argued that statistical mapping, i.e.
determination of the pore size distribution and other functions, then “decorating”
a lattice of the correct coordination number by assigning values drawn at random
from the distribution, was therefore more practical than direct mapping. Some
assumptions and problems involved in this move from direct to statistical mapping
will be discussed in Section 4.
Direct mapping will yield an irregular lattice, whereas statistical mapping,
for the sake of convenience, is done on regular lattices. Winterfeld et al. (1981)
performed extensive Monte Carlo experiments and found that percolation and con-
duction properties of 2D Voronoi tesselations (Figure 2d) were essentially identical
to properties of hexagonal lattices. Their work was expanded and extended to 3D
by Jerauld et al. (1984a,b), who examined backbones and percolation exponents
of Voronoi and regular 2D and 3D systems. They found that regular and irregular
lattices behave essentially identically as long as they have identical coordination
numbers. The equivalence of regular and disordered systems formally justifies
application of work on regular lattices to real porous media. With respect to coor-
dination, work by Lin and Cohen (1982), Chatzis et al., (1983) and Yanuka et al.
(1984) has shown that sandstones have coordination numbers in the range of 4–8.
The coordination number applicable to soil is a more complex matter, however,
and discussion is deferred to Section 4.
In a related study, Gladden et al. (1995) demonstrated how spatial hetero-
geneities in pore size distributions can be characterized by use of nuclear magnetic
resonance imaging, and used percolation theory to quantify pore cluster sizes and
proximity to the percolation threshold. Analysis of "homogeneous” porous catalyst
support pellets indicated wide variability in these properties. Detailed, quantitative
understanding of pore structure can also be achieved by reconstruction of 3D pore
spaces from 2D serial section measurements of porous materials (e.g., Kwiecien
et al., 1990 and references therein; Adler, 1994). In addition, extremely detailed
images of the microstructure of porous rocks have been obtained using high resolu-
tion microtomography (e.g., Auzerais et al., 1996). While these methods are useful
complements to percolation theory and network modeling in the study of soils and
other porous media, discussion of them is beyond the scope of this review.
3.6. DISPERSION
A detailed analysis of dispersion phenomena in porous media using network mod-

els and percolation theory is presented by Sahimi (1984, 1987) and Sahimi and
Imdakm (1988). Sahimi (1987) derives a set of relationships relevant to geometrical
and hydraulic characteristics of capillary tube networks and uses them to establish
various expressions for describing dispersion of contaminants under various condi-
tions and assumptions, near the percolation threshold. In particular, the relationship
between mean square particle displacement and time yields different exponents,
depending on the flow conditions (e.g., as characterized by the Péclet number Pé
vl=D, where v is mean fluid velocity, l is the characteristic length (roughly, the
mean pore size), and D is the diffusion coefficient. This dimensionless number
gives the ratio of advective to diffusive forces.). Additional discussion of these
relationships is given by Berkowitz and Braester (1991) and Yanuka (1992).
A detailed study of dispersion processes in porous media was then presented
by Sahimi and Imdakm (1988), who incorporated several earlier sets of results
(given as references), and employed intensive Monte Carlo numerical simulation
and particle (solute) tracking in random capillary networks. Random networks
were constructed (in both two and three dimensions) by removing segments from
an orthogonal lattice network, according to various criteria. They showed that, as
the percolation threshold is approached from above, particle (solute) transport, as
characterized by mean square displacement, behaves in an anomalous manner. The
dispersion coefficient is not constant, so the conventional advection-dispersion
equation (e.g., Bear, 1972) cannot be applied. Rather, solute transport is found
to be scale-dependent, on either length or time, with exponents varying for the
flow and transport processes considered (i.e., proximity to percolation threshold;
Péclet number indicating either diffusive or advective domination; consideration of
diffusive transport into dead-end and stagnant regions). This framework provides
an important perspective for understanding anomalous dispersion, which is usually
only attributed to long-range spatial correlations in the permeability field.
Using similar random networks, Koplik et al. (1988) considered a full distrib-
ution of particle transit times (i.e., times required for particles to advance through
a given length of the system), and derived relationships for higher order moments
(i.e., in addition to mean and variance) of the particle displacements. Their formu-
lation takes into account advective transport through the backbone and diffusive
transport into dead ends and stagnant regions. They developed a computational
approach which allows determination of transit time distributions of the particles,
as a function of the macroscopic (mean) fluid velocity. Numerical calculations were
in good agreement with analytical predictions of percolation theory.
The agreements between these theoretical and numerical results indicate that
percolation models can be of use in understanding detailed transport phenome-
na in porous and fractured formations. For example, the results illustrating the
non-Fickian (anomalous) nature of dispersion in such systems can be applied to
explain the now well-known scale-dependency of the dispersion coefficient in
the advection-dispersion equation. It should also be pointed out that networks
of the type considered in these studies (i.e., near the percolation threshold) are
fractal in nature (see Section 2). Thus these results, which have been derived for
physically-based flow systems, add considerable theoretical and quantitative sup-
port to expressions developed by, for example, Ross (1986), Wheatcraft and Tyler
(1988), and Cushman (1991), which attribute the scale-dependency of the disper-
sion coefficient to fractal paths.
4. Percolation Theory and Soils: Assumptions, Problems and Variations
Throughout much of the preceding discussion of percolation theory and network

models, we have maintained several assumptions which now require further exam-
ination. These assumptions fall into two classes, namely medium effects and fluid
effects. They will be discussed in turn.
4.1. MEDIUM EFFECTS: LARGE-SCALE STRUCTURE
Percolation theory assumes that the medium is infinite, stationary (roughly, the
statistical properties do not vary with location), and random (in the sense that occu-
pation probabilities or pore sizes are assigned randomly and independently). Soils
and other geological porous media do not strictly meet any of these assumptions.
The deviation from infinity is obvious, although at the pore scale many geological
formations are large enough that extension of results to infinity seems reason-
able. Since many investigations of percolation theory are based on Monte Carlo
experimentation, the effect of the size of the problem domain has been examined.
As mentioned in Section 3, Chatzis and Dullien (1977) showed how the apparent
percolation threshold changes as a function of sample depth. Their results are rep-
resentative of the well-established (Stauffer and Aharony, 1992) finite-size scaling
effects of percolation. Chandler et al. (1982) found that, in two dimensions, the
residual (trapped) saturation increases as the lattice width increases, although it is
invariant with lattice length provided the length exceeds the width. Wilkinson and
Willemsen (1983) noted that residual saturation is invariant with lattice size on a
cubic lattice in three dimensions.
A less obvious effect of non-infinite media is the effect of the shape of the
medium, that is, whether it is wide or narrow with respect to the direction of per-
colation. A soil horizon or geological formation, for example, might extend for
kilometers in two dimensions, but only be several centimeters thick in the third.
Ewing and Gupta (1993a) examined this issue using site invasion percolation on a
cubic lattice, and examined effects of the lattice size and shape on the percolation
threshold, the degree of saturation at breakthrough and the degree of residual satu-
ration. Lattices that were wider than they were long in the direction of the invasion
under-estimated the percolation threshold, while lattices that were longer in the
invasion direction than they were wide, over-estimated the threshold (Figure 8).
Interestingly, the percolation threshold was constant for lattices that were of equal
size in all three dimensions. Wilkinson and Willemsen’s (1983) assertion of invari-
ant residual saturation in three dimensions was shown to be a simplification: where
the lattice width exceeds the length, residual saturation is invariant with width but
increases asymptotically with length. This behavior has also been seen by Hirsch
and Thompson (1994).
In a related study, Silliman (1990) used percolation theory and Monte Carlo
simulations to analyze finite cubic domain-type lattices. He demonstrated that
Figure 8. Site percolation thresholds on a cubic lattice as a function of lattice size, both in the
direction of the invasion and normal to the direction of invasion. The threshold is essentially constant
for cubes (adapted from Ewing and Gupta, 1993a).
lattice geometry and domain discretization can significantly influence the mean
critical probability within a discretized random field. In particular, it is shown that
refined discretization of a lattice in the horizontal direction cannot compensate
for information lost when reducing the vertical discretization of the lattice, due
to the variability in connectivity (and therefore hydraulic conductivity) properties
between two- and three-dimensional systems. This result is also supported by
recalling well-known percolation results which show, for example, that the critical
(site) percolation probability for a three-dimensional cubic system (pc = 0.312) is
considerably smaller than that for a two-dimensional square system (pc = 0.593).
This observation has important implications in terms of numerical modeling of
hydrological systems, where complex three-dimensional flow domains are often
reduced in their dimensionality for the sake of simplicity.
Another aspect of structure at a large scale is layering, which may be considered
a special case of anisotropy. Soils and geological media in general tend to be lay-
ered, while percolation theory in its basic form assumes that the medium is equally
random at all points. However, several investigators have performed network mod-
eling of layered media. Ferrand and Celia (1992) simulated drainage in lattices
composed of both coarse and fine media. The proportion of fine material was var-
ied from 29% to 59%, and at each proportion layered media were compared to
random media. In general, as the proportion of fine material increased, they saw an
increased difference between the drainage curves from layered and random media.
Effects of the number of layers, and the relative heterogeneity within and between
layers, were not considered in this study. Hansen et al. (1993) studied conductivity
both parallel and perpendicular to layering, assuming each layer to be internally
uniform and therefore treatable as a site with a known probability p. They found
that conductivity parallel to the layering could be predicted by effective medium
techniques (Kirkpatrick, 1971, 1973) with an error of 10% or less. However, con-
ductivity perpendicular to the layering was more complex (albeit one-dimensional),
and could only be explained by percolation theory.
While layering is the most obvious form of non-stationarity encountered in
soils, other forms may also appear. For example, rather than having discrete layers,
the medium may become continuously finer or less porous with depth, or the cor-
relation length may increase with depth. These effects have not been investigated,
although we note here that fining downward has some effects similar to those of
gravity, discussed below. A recent study by Friedman and Seaton (1996) considers
anisotropic lattices, and finds that permeability and diffusive properties depend
strongly on anisotropy induced by directionally different coordination numbers,
and by anisotropic pore size distributions.
4.2. MEDIUM EFFECTS: PORE-SCALE CORRELATIONS
Soils and geological porous media tend to be non-random at the scale of individual
pores or local pore domains. The most commonly studied non-randomness at this
level is correlation in the sizes of neighboring pores. In percolation theory, it is
assumed that each site is assigned a state independently of its neighbors (Section
2), although some aspects of anisotropy have been considered (see the discussion
preceding Equation (15), and Balberg and Binenbaum (1985)). This assumption
was implicitly included in the statistical mapping concept of Mohanty et al. (1980).
Some researchers working in percolation in porous media (e.g., Chatzis and Dullien,
1982; Wardlaw et al., 1987; Bryant et al., 1993) have noted that this assumption
does not apply to their materials: pore throats are often correlated with the size of
the pore bodies to which they connect. This observation has led to the development
of models that use some form of correlation between bonds and/or sites (e.g., Diaz
et al., 1987; Kantzas and Chatzis, 1988; Ioannidis and Chatzis, 1993; as discussed
in Section 3.1).
As pointed out by Jerauld and Salter (1990), there are several possible forms of
correlation, such as correlation between pore bodies and the throats that connect
them, correlation between adjacent bodies only, and correlation between neighbor-
ing throats only. Wardlaw, Laidlaw and coworkers tried several variations: Li et al.
(1986) and Wardlaw et al. (1987) used throats whose diameter was a function of
the two pore bodies it connected, while Maier and Laidlaw (1990) had the throat
diameter equal to the diameter of the smallest connecting site. Yanuka et al. (1986)
and Diaz et al. (1987) employed similar approaches. Correlation tends to decrease
both the percolation threshold and the amount of residual non-wetting phase fol-
lowing imbibition. In general, correlated networks are better than uncorrelated
networks in yielding drainage curves that resemble curves from real porous media
(usually sandstone). The same results might be obtained by using networks with
higher coordination, but since sandstone probably has coordination in the range of
4-8 (Lin and Cohen, 1982; Chatzis et al., 1983; Yanuka et al., 1984), increasing
the coordination of the lattice beyond this range is not a realistic alternative. At
present, however, it is not clear to what extent correlated networks still obey the
scaling laws followed by random networks.
Correlation between throat sizes and/or pore body sizes in soils is likely both
longer range and more complex than the simple methods used above. One approach
to dealing with this long range corrrelation is to use pore domains (groups of
interconnected pores that are considered as single units) rather than individual
pores as the units operated upon (Kueper and McWhorter, 1992; Ewing and Gupta,
1993a; Yortsos et al., 1993), which allows network approaches to be scaled up
from the pore level. A difficulty with this method, however, is that coordination
numbers are even more difficult to measure experimentally for domains than for
individual pores. Some insight may be provided by an interesting study in long
range correlation conducted by Renault (1991), who used random fields to assign
probabilities to sites and bonds, rather than assigning them independently. As
the correlation length was increased from 0 to 5 node spacings, the percolation
threshold fell, consistent with the studies cited above. What was remarkable was
that the thresholds for cubic lattices with site percolation, cubic lattices with bond
percolation, and tetrahedral lattices with bond percolation all converged to the
same range – approximately 0.15. Renault (1991) explained this by introducing
the concept of blob percolation, since percolation now occurred through blobs of
related sites or bonds. A follow-up study of accessibility characteristics of random
field lattices was published by Ioannidis et al. (1993).
During blob percolation in a random field correlated lattice, it is the coordination
of the blobs that determines the percolation threshold, so changing how the blobs are
constructed should change the threshold. Ewing and Gupta (1993b) point out that
Renault’s (1991) percolation threshold of approximately 0.15 is near the expected
value for site percolation on a 3D Voronoi lattice (0.145, cf. Jerauld et al., 1984b),
which has a coordination number of approximately 15.54 (Meijering, 1953), and
suggest that the blobs have the same connectivity structure as Voronoi polyhedra.
A percolation threshold of approximately 0.31 would presumably be obtained
using Ferrand and Celia’s (1992) “random heterogeneity” method, which simply
generates cubic blobs. If soil properties are structured like random fields, perhaps
they can be better modeled using domains structured as Voronoi polyhedra or, rather
more conveniently, tetrakaidecahedra (the union of a cube and an octahedron, with

z = 14) (Jerauld et al., 1984b).
Correlation affects permeability as well as drainage and imbibition. In an elab-
orate simulation study, Jerauld and Salter (1990) found that, as the degree of corre-
lation between nearby throats increased, wetting phase saturated permeability and
non-wetting phase relative permeability increased. An explanation for this increase
in saturated permeability with increased correlation can be found in Ambegaokar et
al.’s (1971) model of hopping conductivity, which holds that flow through a random
medium is dominated by a few pathways of high conductivity. Correlation appears
to increase the probability and/or the conductance of these individual pathways
(Jerauld and Salter, 1990). Interestingly, correlation in size between nearby pores
in sandstones has also been observed to affect miscible dispersion (Bretz et al.,
1988). This kind of small-scale property affecting larger scale behavior is typical
in percolation-type studies, and suggests that even pore-scale structure might be
worth examining in otherwise “structureless” soils.
A different type of fine scale non-randomness was mentioned by Ewing and
Gupta (1993a). They suggested that, in a random medium with a wide pore size
distribution, the coordination number could be a function of pore size, with larger
pores having higher coordination. This makes pore-scale percolation studies more
difficult: how can percolation scaling laws be used when z is a function of p?
They argued that, in structured soils, coordination numbers in the thousands are
possible, because of the presence of (for example) clay skins and macropores.
They also noted that in soils, the aspect ratio of pores could be linked to pore size,
since pores in clay tend to be flatter than pores between sand grains. This line of
reasoning can be extended further: for example, clay regions in soils may be more
prevalent than silt or sand regions, suggesting that the degree of clustering of pores
of similar size, and even the shape of the clusters, may also be functions of pore
size. Furthermore, contact angle may be different in pores whose walls are coated
with organic matter; if these pores tend to be clustered spatially or by pore size,
this would further affect fluid occupancy and thus the percolation threshold. Such
size-based characteristics represent a second major drawback to Mohanty et al.’s
(1980) statistical mapping. Detailed examinations of structured soils are needed to
help determine to what extent statistical mapping is valid in soils.
Moving beyond the blob-type correlation of Renault (1991), Ewing and Gupta
(1993b) considered different types of pore clustering. They clustered pore domains
with similar sized pores into different shapes such as line and plane segments, and
examined the effects of the size of these clusters, the proportion of the medium
occupied by the clusters, and the internal homogeneity of the clusters. Increased
cluster length, fraction and homogeneity lowered both the percolation threshold and
the residual saturation. Of the various cluster shapes examined, linear-type clusters
were the most efficient at lowering the percolation threshold or increasing perme-
ability. This agrees with the experience of soil physicists over the last decade, that
linking soil morphology to hydraulic properties might most profitably concentrate
on long structures (macropores) rather than on blob-type clusters (fabric). Related
current work in this direction involves disk- or sphere-shaped objects considered
as pores rather than assemblages of pore domains (e.g., Berkowitz and Balberg,
1992). These “inverted Swiss cheese” (or “inverted random void”) constructions
are an extension of the continuum “Swiss cheese” (or “random void”) models (e.g.,
Halperin et al., 1985), using the spheres, disks, or other shapes as the permeable
rather than the impermeable phase.
In summary, soils are structured in many different ways, making them different
from the infinite, random medium postulated by percolation theory. Some of these
differences (e.g., bond-correlated sites) have been recognized and studied for years,
though not always with the intention of applying percolation laws to them. Other
differences, such as large-scale correlations, anisotropies and pore-size dependence
of other properties, are relatively new in the literature. Here, as with most areas
of soil structure research, even the basics have not been universally agreed upon,
let alone their effects on the percolation properties of the resulting medium. This
suggests that there is considerable opportunity for further research.
4.3. FLUID EFFECTS: VISCOSITY
The structural considerations mentioned above all stem from properties of the medi-
um; we will now consider fluid-based effects. Factors such as viscosity and gravity
can complicate an already sufficiently complex capillary-driven flow process. We
will discuss these factors in turn.
Wilkinson and Willemsen (1983) described invasion percolation as applicable to
imbibition that proceeded at a constant flow rate, rather than at a constant pressure.
In addition, the process must be slow, allowing capillary forces to dominate over
viscous forces. Drainage and imbibition are closest to being invasion percolation-
like processes at low capillary numbers (Chatzis and Morrow, 1981; Lenormand
et al., 1988), where the capillary number Ca v= , for viscosity , velocity v
and interfacial tension . When we are dealing with the combination of immiscible
fluids at constant densities, high capillary numbers (for example, Ca > 10 4 ), and
(
viscosity ratios M < 1:0 M i =d , where subscripts i and d refer to the invader
and defender phases, respectively), viscous fingering (Figure 9a) can result.
A full description of the physics of viscous fingering is beyond the scope of this
review; readers are referred to reviews by Saffman (1986) and Homsy (1987) for
details. Briefly, the fingering comes about because, when viscous forces are non-
trivial, they affect “decisions” at the pore scale. To illustrate, suppose we have a one
meter square, horizontal sand layer through which oil is flowing. Ignoring diffusion,
the oil flows steadily along streamlines, with the local speed determined by the local
geometry. Assuming that the sand is relatively homogeneous, a front of a different
colored oil with otherwise identical properties would move relatively stably, with
distinct fingers (or ‘tongues’) advancing at different rates according to local flow.
Now suppose we start injecting air (negligible viscosity) instead of oil. A tongue
Figure 9. (a): Viscous fingering. Earlier times are shown with darker fluid. (b): Diffusion-limited
aggregation (DLA), with the seed being the left side. In both cases, the interface moves from left to
right.
of air that advances ahead of the front due to some local perturbation will then
have less resistance between it and the far end of the sand layer, so it will advance
further. At each instant, decisions at each point along the interface are determined
by the total resistance between that point and the end of the streamline. This results
in many individual tongues advancing rather than a single front, although where the
medium is narrow normal to the flow direction, a single tongue of air will dominate
(Figure 9a).
Paterson (1984) made the interesting observation that viscous fingering resem-
bles DLA or diffusion-limited aggregation patterns (Figure 9b) (Witten and Sander,
1983), which have fractal dimension of approximately 1.70 in 2D and 2.50 in 3D.
DLA can be simulated by making a “seed”, which can be a point, a line, or some
other shape. “Sticky” particles are then released nearby and diffuse freely until
they touch the seed, whereupon they stick to it and become part of it. The resulting
shape is dendritic (Figure 9b). Paterson’s (1984) article connected a real-world
phenomenon to a particular fractal type, and called the attention of physicists to
a problem that previously had occupied only a few specialists in porous media,
fluid dynamics, and petroleum engineering. The opposite process, anti-DLA or
diffusion-limited annihilation (Meakin and Deutch, 1986), has anti-particles dif-
fuse into a domain fully occupied by particles. Each anti-particle diffuses freely
until it reaches an occupied site, wherein both particle and anti-particle are removed
from the domain. Anti-DLA results in a stably expanding front similar to the Eden
growth model (Eden, 1961).
If we increase the interfacial tension in the example above, we see that it tends
to decrease the surface area of the interface, and so rounds and diminishes the
advancing finger. Thus, there could be three flow regimes in our example above:
capillary fingering, viscous fingering and stable displacement. Such a classification
was developed in an elegant combination of simulations and experiments by Lenor-
mand et al. (1988), who developed a “phase diagram" showing which locations on
the M , Ca plane correspond to which flow regime. These regimes also correspond
to the statistical models of invasion percolation, DLA and anti-DLA. In addition
to documenting and locating these regimes in the phase diagram, Lenormand et al.
(1988) also showed that the boundaries are not sudden: one regime changes contin-
uously into the next, in both form and (apparently) fractal dimension. The original
research was conducted using only drainage; subsequent work (Lenormand, 1990)
extends the approach to imbibition, and considers the influence of the aspect ratio
and the width of the pore size distribution.
Lenormand and coworkers’ (1988, 1990) experiments and simulations used
regular lattices in 2D, and correlated or structured media were never considered. In
addition, inertial and buoyancy effects were avoided. Nonetheless, the experiments
provide a useful framework for future work, and show that interesting deviations
from pure invasion percolation commonly arise in porous media. It remains to be
seen whether accessibility, scaling exponents, and other concepts from percolation
theory can be applied to flow that is not strictly percolation-like. Initial work on
continuously blending the different statistical models, in two dimensions, has been
published by Sahimi and Yortsos (1985), Leclerc and Neale (1988) and Kiriakidis
et al. (1990, 1991).
Viscous fingering represents an extreme that does not normally occur in soils,
since air, the low-viscosity fluid in soils, does not displace water at a rate sufficiently
high to put the flow into the viscous fingering regime. More relevant in soils is the
consideration that not all flow takes place at very low capillary numbers, so viscosity
can affect flow even when viscous fingering does not occur. The effect of viscosity
during infiltration in soils is actually to stabilize the front: due to the viscosity of
water, medium-sized pores near the soil surface may imbibe water sooner than
smaller pores deeper down. In Lenormand et al.’s (1988) phase diagram, this is the
anti-DLA (stable interface) regime. This is one of the reasons why percolation-like
water-air interfaces are not seen in soils; a second reason is discussed below.
4.4. FLUID EFFECTS: BUOYANCY
Another factor that can complicate immiscible fluid displacement is the effect of
gravity. More strictly, this effect can be called a buoyancy effect, as it depends on
the relative densities of the fluids involved. Simulations that have modeled drainage
or imbibition as a percolation process have, as a rule, ignored gravity (e.g., Fatt,
1956a,b,c; Chatzis and Dullien, 1985), meaning they resemble slow imbibition or
drainage in a zero gravity environment. These systems correspond exactly to the
original model of invasion percolation. If, on the other hand, gravity is allowed, the
pore radius at which a given pore can be invaded is a function of the pore’s location
in the gravity field.
The related subset of percolation theory is gradient percolation (e.g., Rosso et
al., 1986). The situation is analogous to the microscopic wetting pattern above
a static water table: as one moves farther above the water table, the radius of
the largest water-filled pore becomes smaller. Clearly, in the case of soil, buoyancy
forces will tend to stabilize a downward moving drying front (since air is less dense
than water) and destabilize a downward moving wetting front. Wilkinson (1986)
examined analytically how such a case would impact percolation mathematics in
3D, with particular emphasis on effects on the correlation length. Ioannidis et al.
(1996) examined the interacting effects of buoyancy forces and medium structure,
using a macroscopic or domain percolation approach. They noted that the buoyancy
forces become non-negligible within the individual domains as a function of both
the fluid density difference, and the size of the domains. Birovljev et al. (1991) used
both simulations and experiments to examine the 2D case, varying the effect of
gravity by tilting their 2D micromodel. More recent experiments and simulations
were performed by Frette et al. (1992), Prat (1993) and Hirsch and Thompson
(1995).
Under certain circumstances, a different form of fingering known as gravity
fingering can emerge (Glass et al., 1989a,b; Hillel, 1993). This fingering is rem-
iniscent of viscous fingering, but differs in that branching and rejoining rarely
happen once the fingers are formed. Stability analyses (Saffman and Taylor, 1958;
Raats, 1973; Parlange and Hill, 1976; Glass et al., 1989a) indicate that finger for-
mation is favored when gravity effects are greater than capillary effects, when the
medium has saturation and/or conductivity increasing with depth, and under vari-
ous other criteria. The fingers do not appear percolation-like, yet both this process
and its “inverse” (a low density fluid migrating upward through a higher density
fluid) can be simulated using an invasion percolation algorithm that includes grav-
ity (Frette et al., 1992). This suggests that stochastic growth models might readily
be adapted to handle gravity fingering as well as the other phenomena mentioned
earlier. Moreover, Lenormand et al.’s (1988) phase diagram might be extended
Figure 10. Proposed extension of the phase diagram of Lenormand et al. (1988) to account for
gravity effects. The shaded regions indicate conditions of applicability of invasion percolation, DLA
and anti-DLA (stable growth) models. By convention, positive values of Bo stabilize the front, and
negative values destabilize it.
from a 2D model to a 3D phase space encompassing Ca, M and Bo (Figure 10).

The Bond number, Bo, gives the ratio of gravity to capillary forces, and is defined

by Bo gl2 = , where g is gravitational acceleration, l is the characteristic

pore size, is the fluid density difference and is the interfacial tension. This
third dimension would make the diagram more applicable to situations commonly
encountered in soils.
5. Some Recent Work Relating Percolation Theory and Network Models to

Soils
Several recent works have appeared that concern percolation theory and soils,
involving pore continuity, porous media basics and issues of flow and transport.
Some of these studies used percolation theory directly, while others used pore-scale
network modeling without explicit reference to percolation theory. For example,
at least four articles have appeared in the soils literature that use network models
but have not connected with relevant aspects of percolation theory: Cox (1983),
studying drainage, Ewing and Gupta (1994), modeling clogging of pores during
surface sealing, and Steele and Nieber (1994a,b), examining diffusion of gases at
different water contents.
A study of pores in cracking clay soils by Scott et al. (1988a,b) analyzed the
numerical and topological density of cracks by making serial sections through
a resin-impregnated clay and following individual cracks as they branched and
merged. The two soils studied had numerical densities (number of discrete crack
networks) of approximately 34 cm 3 and 75 cm 3 , and topological densities (num-
ber of closed loops in the networks) of approximately 300 cm 3 and 195 cm 3 .
However, this study left unanswered the question of most interest to workers in
flow and transport: was there at least one crack network that spanned the entire
sample? A familiarity with percolation theory might have suggested this question,
along with the related issue of network size distribution. Also, using continuum
percolation, it would have been possible to estimate the percolation threshold for
the crack types represented; analysis of cluster sizes might also have shown whether
cracks were randomly placed or preferentially connected.
Fracture networks in geological formations have been analyzed extensively
(e.g., Berkowitz, 1994), and while such fracture networks are different in nature
than cracks in soil structures, the same conceptual approaches can be applicable.
Various geometrical and connectivity characteristics, as well as flow and transport
properties, of fracture networks, have been analyzed in terms of percolation theory.
The concept of the average number of intersections per fracture, and the critical
percolation density, have been used as a measure of connectivity in random fracture
geometries (e.g., Robinson, 1983, 1984; Charlaix et al., 1984). Moreover, Balberg
(1986b) and Berkowitz (1995) have shown that general connectivity and power
law relationships (of the form of Equation 1) characterize the density of fractures
and average number of intersections per fracture necessary to ensure network
connectivity, the likelihood of a fractured formation being hydraulically connected,
and the size and extent of fracture clusters.
A related issue is the accessibility, to microorganisms of different sizes, of pores
inside aggregates. Glasbey et al. (1991) approached this problem by synthesizing
3D images using an overlapping spheres model: spheres corresponding to a prede-
termined size distribution (representing soil solids) were synthesized inside a large
3D domain. Depending on the final porosity, the remaining pore space was more
or less connected and continuous. This scenario corresponds to the continuum or
“Swiss-cheese” model of percolation (Halperin et al., 1985; Feng et al., 1987), and
has also been studied in sintering experiments (see, e.g., Pathak et al., 1982 and
references therein). Exact relationships are available, at least for uniform sphere
sizes. In addition, the percolation threshold and the point at which the pore space
first becomes disconnected are useful percolation concepts in this context.
An innovative use of percolation theory was recently proposed by Luxmoore and
Ferrand (1993), who had observed in several previous studies that macropore flow
during rain events was composed of a mixture of “old” water already present in the
soil matrix, and “new,” just infiltrated water. Considering the soil to be composed
of macropores, mesopores and micropores, they hypothesized that chemicals held
in the mesopores between rain events were somehow moving into the macropores.
Mass flow seemed an unlikely explanation, leaving diffusion as the only available
mechanism. Using percolation theory, they estimated the surface area connecting
the mesopore and macropore domains, and found that diffusion during macropore
flow was not sufficient to explain the mixing. Consideration of percolation theory
led them to a third hypothesis, that mesopore regions were connected by macropore
links, like a blob and link model of a percolation backbone (Stanley, 1977). Between
rains, the soil dried to the point that the macropore links desaturated, and mesopore
water equilibrated with the soil solution. When rain fell again, the macropores
reconnected with the mesopore blobs, carrying with them the mesopore water,
and thus mixing “old” and “new" water. This work is continuing, emphasizing
characterization of soil heterogeneity across many scales with an eye to building
more realistic percolation models (Homer et al., 1994).
The analogy between drainage and invasion percolation suggests a further anal-
ogy between bubbling pressure (defined as the pressure at which the soil drains to
the point that air can pass through the entire sample) and the percolation threshold.
Katz and Thompson (1986) and Thompson et al. (1987) examined this analogy both
empirically (using mercury porosimetry) and theoretically, and derived an equa-
tion for saturated permeability which applies to a wide variety of porous rocks.
Their equation, k =( ( ))
r2 =0 =226, relates conductivity to the pore radius r that
corresponds to the bubbling pressure. Structural effects are handled in the =0 or
formation factor term, given by the ratio of the electrical conductivity of the medi-
um when saturated with an electrically conducting fluid to the conductivity of the
fluid itself. Interestingly, Katz and Thompson’s (1986) work relates the percolation
threshold to the inflection point on the drainage curve; this has implications for any
attempt to derive bubbling pressure from drainage data. A recent application of the
Katz and Thompson (1986) approach to clay-bearing sedimentary rock (Korvin,
1992) implies that the 3D percolation conductivity exponent t changes continuous-
ly from the lattice value 2.0 to a continuum value of 4.4 (Halperin et al., 1985) as
the texture moves from coarse sand to shale.
The relationship between bubbling pressure and conductivity shows up else-
where as well (e.g., Wang and Narasimhan, 1992), but has not always been recog-
nized as such. For example, it appears in the data of Laliberte and Brooks (1967),
but was only recently recognized as such (Ewing and Gupta, 1993b). Baver (1938)
proposed an equation for predicting saturated conductivity based on the inflection
point in the water characteristic curve. The equation for relative air permeability
derived by Brooks and Corey (1964) uses a proximity parameter similar to the
standard percolation theory form (p pc ) (see Equation 9), and air (being the
non-wetting fluid) corresponds well to the invading fluid in invasion percolation.
The unsaturated conductivity (relative permeability) functions of Brooks and
Corey (1964), Bouwer (1966), Campbell (1974) and van Genuchten (1980) also
use bubbling pressure as an input parameter. However, these relationships use a
proximity parameter quite different from the (p pc ) form typical of percolation

theory. This may be because the percolation equations are most accurate close
to the percolation threshold, while the water retention and conductivity function
cover a much wider range. At the dry end, for example, thin films (see Toledo et al.,
1990) dominate conductivity; Dullien et al. (1989), working with SEM and residual
saturations as low as 1%, argue that residual saturation is due to surface roughness.
On the other hand, water forms a continuous phase through virtually the entire
range of the retention and conductivity functions, in contrast to the assumptions of
percolation theory, indicating that percolation theory may not be appropriate for
these functions.
The analogy between bubbling pressure and the percolation threshold is most
useful in unstructured media or samples above the REV, and where buoyancy effects
are minimal. Unfortunately these two considerations clash in most agricultural
soils: the REV in a structured soil may be large, at which scale gravity is likely to
be non-negligible (Blunt et al., 1992).
6. Concluding Remarks
This review has been written with two principle objectives. The first was to provide
a useful summary and reference work for soil scientists wanting to learn more about
percolation theory and network models, and their application to soils. The second
was to expose some of the many remaining issues and challenges in this field. These
challenges are discussed throughout the review, and include the development of
critical relations for media that in some ways resemble both lattices and continua
(Section 2), relevant pore scale events (Section 3), the statistical properties of
structured soils, and the applicability of various statistical and network models to
different flow regimes (Sections 4 and 5).
Percolation may be considered as representing one of several statistical the-
ories that is applicable for calculating macroscopic, or ‘large-scale’ properties.
Depending on the properties of the one or more fluid (and/or gas) phases present
in the medium, the geometrical randomness of the medium itself, and the cou-
pling between fluid and medium properties, any of several statistical theories may
be applicable, including, in different ranges, percolation, DLA, anti-DLA and
gradient percolation, as well as effective medium theories. In terms of medium
geometry, standard percolation theory can be reasonably applied to poorly con-
nected and cracked soils and fractured rock; invasion percolation and its variants
can be usefully applied to model a range of multiphase displacement mechanisms.
From the many examples discussed above, it should be clear the challenge is not
to force non-percolation processes into the percolation mold, but rather to develop
statistical models that can represent the entire continuum, and to clarify the linkage
between statistical and physical models. In addition, since the statistical models
appear to be linked, it remains to determine what percolation theory can contribute
to the ensemble (e.g., the concept of universality), as well as what other statistical
models can contribute to percolation.
Acknowledgements
B. B. thanks his colleagues, Isaac Balberg and Harvey Scher, for fruitful discussions
and collaboration. B. B. holds the Barecha Foundation Career Development Chair.
R. P. E. thanks the Minnesota group for their interest in applying percolation theory
to soils, and the USDA-ARS for support during the research and writing of this
paper.
References
Adler, P.M.: 1992, Porous Media: Geometry and Transports, Butterworth-Heinemann, Boston.
Adler, P.M.: 1994, ‘The method of reconstructed porous media’, Current Topics in the Physics of
Fluids 1, 277–306.
Alon, U., Drory, A., and Balberg, I.: 1990, ‘Systematic derivation of percolation thresholds in
continuum systems’, Physical Review A 42, 4634–4638.
Ambegaokar, V., Halperin, B.I., and Langer, J.S.: 1971, ‘Hopping conductivity in disordered systems’,
Physical Review B 4, 2612–2620.
Androutsopoulos, G.P. and Mann, R.: 1979, ‘Evaluation of mercury porosimeter experiments using
a network pore structure model’, Chemical Engineering Science 34, 1203–1212.
Auzerais, F.M., Dunsmuir, J., Ferréol, B.B., Martys, N., Olson, J., Ramakrishnan, T.S., Rothman,
D.H., and Schwartz, L.M.: 1996, ‘Transport in sandstone: A study based on three dimensional
microtomography’, Geophysical Research Letters 23, 705–708.
Balberg, I.: 1986a, ‘Excluded-volume explanation of Archie’s law’, Physical Review B 33, 3618–
3620.
Balberg, I.: 1986b, ‘Connectivity and conductivity in 2-D and 3-D fracture systems’, in R. Englman
and Z. Jaeger (eds.), Proceedings of the International Conference on Fragmentation, Form and
Flow in Fractured Media, Annals of the Israel Physical Society 8, Adam Hilger, Bristol, pp.
89–101.
Balberg, I.: 1987, ‘Recent developments in continuum percolation’, Philosophical Magazine 56,
991–1003.
Balberg, I., Anderson, C.H., Alexander, S., and Wagner, N.: 1984, ‘Excluded volume and its relation
to the onset of percolation’, Physical Review B 30, 3933–3943.
Balberg, I. and Binenbaum, N.: 1985, ‘Cluster structure and conductivity of three-dimensional con-
tinuum systems’, Physical Review A 31, 1222–1225.
Barton, C.C.: 1995, ‘Fractal analysis of scaling and spatial clustering of fractures’, in C.C. Barton
and P.R. La Pointe (eds.), Fractals in the Earth Sciences, Plenum Press, New York, pp. 141–178.
Baver, L.D.: 1938, ‘Soil permeability in relation to non-capillary porosity’, Soil Science Society of
America Proceedings 3, 52–56.
Bear, J.: 1972, Dynamics of Fluids in Porous Media, American Elsevier, New York.
Berkowitz, B.: 1994, ‘Modelling flow and contaminant transport in fractured media’, in Y. Corapcioglu
(ed.), Advances in Porous Media, Vol. 2, Elsevier Science Publishers, New York, pp. 395–449.
Berkowitz, B.: 1995, ‘Analysis of fracture network connectivity using percolation theory’, Mathe-
matical Geology 27(4), 467–483.
Berkowitz, B. and Balberg, I.: 1992, ‘Percolation approach to the problem of hydraulic conductivity
in porous media’, Transport in Porous Media 9, 275–286.
Berkowitz, B. and Balberg, I.: 1993, ‘Percolation theory and its application to groundwater hydrology’,
Water Resources Research 29, 775–794.
Berkowitz, B. and Braester, C.: 1991, ‘Dispersion in sub-representative elementary volume fracture
networks: Percolation theory and random walk approaches’, Water Resources Research 27, 3159–
3164.
Birovljev, A., Furuberg, L., Feder, J., Jssang, T., Måly, K.J., and Aharony, A.: 1991, ‘Gravity invasion
percolation in two dimensions: experiments and simulations’, Physical Review Letters 67, 584–
587.
Blunt, M., King, M.J., and Scher, H.: 1992, ‘Simulation and theory of two-phase flow in porous
media’, Physical Review A 46, 7680–7699.
Blunt, M. and Scher, H.: 1995, ‘Pore-level modeling of wetting’, Physical Review E 52, 6387–6403.
Bouwer, H.: 1966, ‘Rapid field measurement of air entry value and hydraulic conductivity of soil as
significant parameters in flow system analysis’, Water Resources Research 2, 729–738.
Bretz, R.E., Specter, R.M., and Orr, F.M. Jr.: 1988, ‘Effect of pore structure on miscible displacement
in laboratory cores’, Society of Petroleum Engineers Reservoir Engineering 3, 857–866.
Broadbent, S.R. and Hammersley, J.M.: 1957, ‘Percolation processes. I. Crystals and Mazes’, Pro-
ceedings of the Cambridge Philosophical Society 53, 629–641.
Brooks, R.H. and Corey, A.T.: 1964, ‘Hydraulic properties of porous media’, Hydrological Paper 3,
Colorado State University, Fort Collins.
Bryant, S.L., King, P.R., and Mellor, D.W.: 1993, ‘Network model evaluation of permeability and
spatial correlation in a real random sphere packing’, Transport in Porous Media 11, 53–70.
Bunde, A. and Havlin, S.: 1991, ‘Percolation I’, in A. Bunde and S. Havlin (eds.), Fractals and
Disordered Systems, Springer-Verlag, Berlin, pp. 51–96.
Campbell, G. S.: 1974, ‘A simple method for determining unsaturated conductivity from moisture
retention data’, Soil Science 117, 311–314.
Chandler, R., Koplik, J., Lerman, K., and Willemsen, J.F.: 1982, ‘Capillary displacement and perco-
lation in porous media’, Journal of Fluid Mechanics 119, 249–267.
Charlaix, E., Guyon, E., and Rivier, N.: 1984, ‘A criterion for percolation threshold in a random array
of plates’, Solid State Communications 50, 999–1002.
Charlaix, E., Guyon, E., and Roux, S.: 1987, ‘Permeability of a random array of fractures of widely
varying apertures’, Transport in Porous Media 2, 31–43.
Chatzis, I. and Dullien, F.A.L.: 1977, ‘Modelling pore structures by 2-D and 3-D networks with
application to sandstones’, Canadian Journal of Petroleum Technology Jan.–Mar., 97–108.
Chatzis, I. and Dullien, F.A.L.: 1981, ‘Mercury porosimetry curves of sandstones, Mechanisms of
mercury penetration and withdrawal’, Powder Technology 29, 117–125.
Chatzis, I. and Dullien, F.A.L.: 1985, ‘The modeling of mercury porosimetry and the relative per-
meability of mercury in sandstones using percolation theory’, Int. Chemical Engineering 25,
47–66.
Chatzis, I. and Morrow, N.R.: 1981, ‘Correlation of capillary number relationships for sandstones’,
SPE paper 10114, 56th Ann. Fall Tech. Conf. and Exhib., San Antonio, TX, Oct. 5–7.
Chatzis, I., Morrow, N.R., and Lim, H.T.: 1983, ‘Magnitude and detailed structure of residual oil
saturation’, Society of Petroleum Engineers Journal 23, 311–326.
Cieplak, M. and Robbins, M.O.: 1990, ‘Influence of contact angle on quasistatic fluid invasion of
porous media’, Physical Review B 41, 508–521.
Cox, M.A.A.: 1983, ‘Analysis of a moisture characteristic curve employing a crystalline lattice of
capillary pores’, Journal of Soil Science 34, 649–658.
Cushman, J.H.: 1991, ‘On diffusion in fractal porous media’, Water Resources Research 27, 643–644.
Dias, M.M. and Payatakes, A.C.: 1986, ‘Network models for two-phase flow in porous media, Part
2. Motion of oil ganglia’, Journal of Fluid Mechanics 164, 337–358.
Dias, M.M. and Wilkinson, D.: 1986, ‘?’, Journal of Physics A 19, 3131-?.
Diaz, C.E., Chatzis, I., and Dullien, F.A.L.: 1987, ‘Simulation of capillary pressure curves using bond
correlated site percolation on a simple cubic network’, Transport in Porous Media 2, 215–240.
Dong, M. and Chatzis, I.: 1995, ‘The imbibition and flow of a wetting liquid along the corners of a
square capillary tube’, Journal of Colloid and Interface Science 172, 278–288.
Doyen, P.M.: 1988, ‘Permeability, conductivity, and pore geometry of sandstone’, Journal of Geo-
physical Research 93(B7), 7729–7740.
Drory, A., Balberg, I., Alon, U., and Berkowitz, B.: 1991, ‘Analytic derivation of percolation thresh-
olds in anisotropic systems of permeable objects’, Physical Review A 43, 6604–6612.
Dullien, F.A.L.: 1992, Porous Media: Fluid Transport and Pore Structure (2nd ed.), Academic Press,
San Diego.
Dullien, F.A.L., Zarcone, C., Macdonald, I.F., Collins, A., and Bochard, R.D.E.: 1989, ‘The effects
of surface roughness on the capillary pressure curves and the heights of capillary rise in glass
bead packs’, Journal of Colloid and Interface Science 127, 362–372.
Eden, M.: 1961, ‘A two-dimensional growth model’, in J. Neyman (ed.), Proceedings of the Fourth
Berkeley Symposium on Mathematics, Statistics, and Probability, Vol. 4, pp. 223–239, University
of California Press, Berkeley.
Ewing, R.P. and Gupta, S.C.: 1993a, ‘Modeling percolation properties of random media using a
domain network’, Water Resources Research 29, 3169–3178.
Ewing, R.P. and Gupta, S.C.: 1993b, ‘Percolation and permeability in partially structured networks’,
Water Resources Research 29, 3179–3188.
Ewing, R.P. and Gupta, S.C.: 1994, ‘Pore-scale network modeling of compaction and filtration during
surface sealing’, Soil Science Society of America Journal 58, 712–720.
Farrell, D.A. and Larson, W.E.: 1972, ‘Modeling the pore structure of porous media’, Water Resources
Research 8, 699–706.
Fatt, I.: 1956a, ‘The network model of porous media, I. Capillary pressure characteristics’, Petroleum
Transactions AIME 207, 144–159.
Fatt, I.: 1956b, ‘The network model of porous media, II. Dynamic properties of a single size tube
network’, Petroleum Transactions AIME 207, 160–163.
Fatt, I.: 1956c, ‘The network model of porous media, III. Dynamic properties of networks with tube
radius distribution’, Petroleum Transactions AIME 207, 164–177.
Feng, S., Halperin, B.I., and Sen, P.N.: 1987, ‘Transport properties of continuum systems near the
percolation threshold’, Physical Review B 35, 197–214.
Ferrand, L.A. and Celia, M.A.: 1992, ‘The effect of heterogeneity on the drainage capillary pressure-
saturation relation’, Water Resources Research 28, 859–870.
Ferrand, L.A., Celia, M.A., and Soll, W.E.: 1990, ‘Percolation-based models for pore-to-lab scale cal-
culations in multifluid porous media’, in J.H. Cushman (ed.), Dynamics of Fluids in Hierarchical
Porous Media, 463-483, Academic Press, San Diego, California.
Fisher, M.E.: 1961, ‘Critical probabilities for cluster size and percolation problems’, Journal of
Mathematical Physics 2, 620–627.
Flory, P.J.: 1941, ‘Molecular size distribution in three-dimensional polymers’, Journal of the Ameri-
can Chemical Society 63, 3083–3100.
Frette, V., Feder, J., Jssang, T., and Meakin, P.: 1992, ‘Buoyancy-driven fluid migration in porous
media’, Physical Review Letters 68, 3164–3167.
Friedman, S.P. and Seaton, N.A.: 1996, ‘On the transport properties of anisotropic networks of
capillaries’, Water Resources Research 32, 339–347.
Gladden, L.F., Hollewand, M.P. and Alexander, P.: 1995, ‘Characterization of structural inhomo-
geneities in porous media’, American Institute of Chemical Engineers Journal 41, 894–906.
Glasbey, C.A., Horgan, G.W., and Darbyshire, J.F.: 1991, ‘Image analysis and three-dimensional
modelling of pores in soil aggregates’, Journal of Soil Science 42, 479–486.
Glass, R.J., Parlange, J.-Y., and Steenhuis, T.S.: 1989a, ‘Wetting front instability I: Theoretical study
and dimensional analysis’, Water Resources Research 25, 1187–1194.
Glass, R.J., Steenhuis, T.S., and Parlange, J.-Y.: 1989b, ‘Wetting front instability II: Experimental
determination of relationships between system parameters and two dimensional unstable flow
field behavior in initially dry porous media’, Water Resources Research 25, 1195–1207.
Golden, J.M.: 1980, ‘Percolation theory and models of unsaturated porous media’, Water Resources
Research 16, 201–209.
Haines, W.B.: 1930, ‘Studies in the physical properties of soil, Part V’, Journal of Agricultural
Science 20, 97–116.
Halperin, B.I., Feng, S., and Sen, P.N.: 1985, ‘Differences between lattice and continuum percolation
exponents’, Physical Review Letters 54, 2391–2394.
Hammersley, J.M.: 1963, ‘A Monte Carlo solution of percolation in the cubic crystal’, in B. Adler, S.
Fernbach and M. Rotenberg (eds.), Methods in Computations Physics, Vol. 1: Statistical Physics,
Academic Press, New York.
Hammersley, J.M. and Welsh, D.J.A.: 1980, ‘Percolation theory and its ramifications’, Contemporary
Physics 21, 593–605.
Hansen, A., Hinrichsen, E.L., and Stauffer, D.: 1993, ‘Percolation in layered media – a conductivity
approach’, Transport in Porous Media 11, 45–52.
Havlin, S. and Bunde, A.: 1991, ‘Percolation II’, in A. Bunde and S. Havlin (eds.), Fractals and
Disordered Systems, pp. 97–150, Springer-Verlag, Berlin.
Heiba, A.A., Sahimi, M., Scriven, L.E., and Davis, H.T.: 1982, ‘Percolation theory of two-phase
relative permeability’, Paper SPE 11015, 57th Ann. Fall Tech. Conf. Exhib. of the Soc. Petr.
Eng. of the AIME, Sept. 26–29, New Orleans. (also in Society of Petroleum Engineers Reservoir
Engineering 7, 123–132, 1992).
Heiba, A.A., Davis, H.T., and Scriven, L.E.: 1984, ‘Statistical network theory of three-phase relative
permeabilities’, Paper SPE/DOE 12690, SPE/DOE 4th Symp. Enhanced Oil Recovery, Apr.
15–18, Tulsa.
Hillel, D.: 1993, ‘Unstable flow: a potentially significant mechanism of water and solute transport to
groundwater’, in D. Russo and G. Dagan (eds.), Water Flow and Solute Transport in Soils, pp.
123–135, Springer-Verlag, Berlin.
Hirsch, L.M. and Thompson, A.H.: 1994, ‘Size-dependent scaling of capillary invasion including
buoyancy and pore size distribution effects’, Physical Review E 50, 2069–2086.
Hirsch, L.M. and Thompson, A.H.: 1995, ‘Minimum saturations and buoyancy in secondary migra-
tion’, American Association of Petroleum Geologists 79, 696–710.
Homer, V.J., Jardine, P.M., Luxmoore, R.J., and Jenssen, P.: 1994, ‘Mapping soil pore heterogeneities
for percolation fluid/solute transport models’, Abstract H12A-03 (page 147), April 19th EOS
supplement: 1994 Spring Meeting of the American Geophysical Union, Baltimore.
Homsy, G.M.: 1987, ‘Viscous fingering in porous media’, Annual Reviews Fluid Mechanics 19,
271–311.
Ioannidis, M.A. and Chatzis, I.: 1993, ‘Network modelling of pore structure and transport properties
of porous media’, Chemical Engineering Science 48, 951–972.
Ioannidis, M.A., Chatzis, I., and Sudicky, E.A.: 1993, ‘The effect of spatial correlations on the acces-
sibility characteristics of three-dimensional cubic networks as related to drainage displacements
in porous media’, Water Resources Research 29, 1777–1785.
Ioannidis, M.A., Chatzis, I., and Dullien, F.A.L.: 1996, ‘Macroscopic percolation model of immiscible
displacement: Effects of buoyancy and spatial structure’, Water Resources Research 32, 3297–
3310.
Jerauld, G.R., Hatfield, J.C., Scriven, L.E., and Davis, H.T.: 1984a, ‘Percolation and conduction on
Voronoi and triangular networks: a case study in topological disorder’, Journal of Physics C 17,
1519–1529.
Jerauld, G.R., Scriven, L.E., and Davis, H.T.: 1984b, ‘Percolation and conduction on the 3D Voronoi
and regular networks: a second case study in topological disorder’, Journal of Physics C 17,
3429–3439.
Jerauld, G.R. and Salter, S.J.: 1990, ‘The effect of pore-structure on hysteresis in relative permeability
and capillary pressure: pore-level modeling’, Transport in Porous Media 5, 103–151.
Kantzas, A. and Chatzis, I.: 1988, ‘Network simulations of relative permeability curves using a bond
correlated-site percolation model of pore structure’, Chemical Engineering Communications 69,
191–214.
Katz, A.J. and Thompson, A.H.: 1985, ‘Fractal sandstone pores: Implications for conductivity and
pore formation’, Physical Review Letters 54, 1325–1328.
Katz, A.J. and Thompson, A.H.: 1986, ‘Quantitative prediction of permeability in porous rock’,
Physical Review B 34, 8179–8181.
Kiriakidis, D.G., Neale, G.H. and Mitsoulis, E.: 1990, ‘Numerical simulations of radial displacement
of a wetting fluid by a non-wetting fluid in a porous medium’, Journal Physics A.: Math. Gen.
23, 5089–5094.
Kiriakidis, D.G., Mitsoulis, E., and Neale, G.H.: 1991, ‘Linear displacement of a wetting fluid by an
immiscible non-wetting fluid in a porous medium: a predictive algorithm’, Canadian Journal of
Chemical Engineering 69, 557–563.
Kirkpatrick, S.: 1971, ‘Classical transport in disordered media: scaling and effective-medium theo-
ries’, Physical Review Letters 27, 1722–1725.
Kirkpatrick, S.: 1973, ‘Percolation and conduction’, Reviews of Modern Physics 45, 574–588.
Koiller, B., Ji, H., and Robbins, M.O.: 1992, ‘Fluid wetting properties and the invasion of square
networks’, Physical Review B 45, 7762–7767.
Koplik, J.: 1982, ‘Creeping flow in two-dimensional networks’, Journal of Fluid Mechanics 119,
219–247.
Koplik, J., Redner, S., and Wilkinson, D.: 1988, ‘Transport and dispersion in random networks with
percolation disorder’, Physical Review A: Math. Gen. 37, 2619–2636.
Korvin, G.: 1992, ‘A percolation model for the permeability of kaolinite-bearing sandstones’, Geo-
physical Transactions 37, 177–209.
Ksenzhek, O.S.: 1963, ‘Capillary equilibrium in porous media with intersecting pores’, Russian
Journal of Physical Chemistry 37, 691–694.
Kueper, B.H. and McWhorter, D.B.: 1992, ‘The use of macroscopic percolation theory to construct
large-scale capillary pressure curves’, Water Resources Research 28, 2425–2436.
Kwiecien, M.J., Macdonald, I.F., and Dullien, F.A.L.: 1990, ‘Three-dimensional reconstruction of
porous media from serial section data’, Journal of Microscopy 159, 343–359.
Laliberte, G.E. and Brooks, R.H.: 1967, ‘Hydraulic properties of disturbed soil materials affected by
porosity’, Soil Science Society of America Proceedings 31, 451–454.
Lane, A., Shah, N., and Conner, Jr., W.C.: 1986, ‘Measurement of the morphology of high-surface-
area solids: Porosimetry as a percolation process’, Journal of Colloid and Interface Science 109,
235–242.
Larson, R.G., Scriven, L.E., and Davis, H.T.: 1977, ‘Percolation theory of residual phases in porous
media’, Nature 268, 409–413.
Larson, R.G., Scriven, L.E., and Davis, H.T.: 1981, ‘Percolation theory of two phase flow in porous
media’, Chemical Engineering Science 36, 57–73.
Leclerc, D.F. and Neale, G.H.: 1988, ‘Monte Carlo simulations of radial displacement of oil from a
wetted porous medium: fractals, viscous fingering and invasion percolation’, Journal of Physics
A: Math. Gen. 21, 2979–2994.
Lenormand, R.: 1987, ‘Statistical physics and immiscible displacements through porous media’, in
J.R. Banavar, J. Koplik and K.W. Winkler (eds.), Physics and Chemistry of Porous Media II. AIP
Conference Proceedings no. 154, AIP, New York.
Lenormand, R.: 1990, ‘Liquids in porous media’, Journal of Physics: Condensed Matter 2, SA79–88.
Lenormand, R. and Bories: 1980, ‘Mécanique des fluides’, C. R. Acad. Sc. Paris 291B, 279–282.
Lenormand, R., Touboul, E. and Zarcone, C.: 1988, ‘Numerical models and experiments on immis-
cible displacements in porous media’, Journal of Fluid Mechanics 189, 165–187.
Levine, S., Reed, P., Shutts, G., and Neale, G.: 1977, ‘Some aspects of wetting/dewetting of a porous
medium’, Powder Technology 17, 163–181.
Li, Y., Laidlaw, W.G., and Wardlaw, N.C.: 1986, ‘Sensitivity of drainage and imbibition to pore
structures as revealed by computer simulation of displacement process’, Advances in Colloid
Interface Science 26, 1–68.
Lin, C. and Cohenm, M.H.: 1982, ‘Quantitative methods for microgeometric modeling’, Journal of
Applied Physics 53, 4152–4165.
Luxmoore, R.J. and Ferrand, L.A.: 1993, ‘Towards pore-scale analysis of preferential flow and
chemical transport’, in D. Russo and G. Dagan (eds.), Water Flow and Solute Transport in Soils,
pp. 45–60. Springer-Verlag, Berlin.
Maier, R. and Laidlaw, W.G.: 1990, ‘Fluid percolation in bond-site size-correlated three-dimensional
networks’, Transport in Porous Media 5, 421–428.
Mandelbrot, B.B.: 1983, The fractal geometry of nature, W.H. Freeman and Co., New York.
Mann, R., Androutsopoulos, G.P., and Golshan, H.: 1981, ‘Application of a stochastic network pore
model to oil-bearing rock with observations relevant to oil recovery’, Chemical Engineering
Science 36, 337–346.
Marshall, T.J.: 1958, ‘A relation between permeability and size distribution of pores’, Journal of Soil
Science 9, 1–8.
Meakin, P. and Deutch, J.M.: 1986, ‘The formation of surfaces by diffusion limited annihilation’,
Journal of Chemical Physics 85, 2320–2325.
Meijering, J.L.: 1953, ‘Interface area, edge length, and number of vertices in crystal aggregates with
random nucleation’, Philips Research Report 8, 270–290 (Report 218).
Mohanty, K.K, Davis, H.T., and Scriven, L.E.: 1980, ‘Physics of oil entrapment in water-wet rock’,
Paper SPE 9406, 55th Ann. Tech. Conf. and Exhib. Soc. Petr. Eng., Dallas, Texas, Sept. pp. 21–24.
Mualem, Y.: 1976, ‘A new model for predicting the hydraulic conductivity of unsaturated porous
media’, Water Resources Research 12, 513–522.
Mualem, Y. and Dagan, G.: 1975, ‘A dependent domain model of capillary hysteresis’, Water
Resources Research 11, 452–460.
Ng, K.M. and Payatakes, A.C.: 1980, ‘Stochastic simulation of the motion, break-up and stranding
of oil ganglia in water-wet granular porous media during immiscible displacement’, American
Institute of Chemical Engineers Journal 26, 419–429.
Nicholson, D.: 1968, ‘Capillary models for porous media, Part 2. Sorption desorption hysteresis in
three dimensional networks’, Transactions of the Faraday Society 64, 3416–3424.
Parlange, J.-Y. and Hill, D.E.: 1976, ‘Theoretical analysis of wetting front instability in soils’, Soil
Science 122, 236–239.
Paterson, L.: 1984, ‘Diffusion-limited aggregation and two-fluid displacements in porous media’,
Physical Review Letters 52, 1621–1624.
Pathak, P., Davis, H.T., and Scriven, L.E.: 1982, ‘Dependence of residual nonwetting liquid on pore
topology’, Paper SPE 11016, presented at the 57th Ann. Tech. Conf. and Exhib. of the Soc. Petr.
Eng. of AIME, New Orleans, Sept. 26–29.
Prat, M.: 1993, ‘Percolation model of drying under isothermal conditions in porous media’, Interna-
tional Journal of Multiphase Flow 19, 691–704.
Raats, P.A.C.: 1973, ‘Unstable wetting fronts in uniform and nonuniform soils’, Soil Science Society
of America Proceedings 37, 681–685.
Renault, P.: 1991, ‘The effect of spatially correlated blocking-up of some bonds or nodes of a network
on the percolation threshold’, Transport in Porous Media 6, 451–468.
Robinson, P.C.: 1983, ‘Connectivity of fracture systems – a percolation theory approach’, Journal of
Physics A: Math. Gen. 16, 605–614.
Robinson, P.C.: 1984, ‘Numerical calculations of critical densities for lines and planes’, Journal of
Physics A.: Math. Gen. 17, 2823–2830.
Rose, W.D.: 1957, ‘Studies of waterflood performance, III. Use of network models’, Illinois State
Geol. Survey Circ. 237.
Rose, W.D., Carpenter, D.W., and Witherspoon, P.A.: 1956, ‘Discussion [of Fatt’s papers]’, Petroleum
Transactions AIME 207, 179–181.
Ross, B.: 1986, ‘Dispersion in fractal fracture networks’, Water Resources Research 22, 823–827.
Rosso, M., Gouyet, J.F., and Sapoval, B.: 1986, ‘Gradient percolation in three dimensions and relation
to diffusion fronts’, Physical Review Letters 57, 3195–3198.
Royal Statistical Society: 1954, ‘Symposium on Monte Carlo Methods’, Journal of the Royal Statis-
tical Society B 16, 23–75.
Saffman, P.G.: 1986, ‘Viscous fingering in Hele-Shaw cells’, Journal of Fluid Mechanics 173, 73–94.
Saffman, P.G. and Taylor, G.I.: 1958, ‘The penetration of a fluid into a porous medium or Hele-
Shaw cell containing a more viscous liquid’, Proceedings of the Royal Society of London A245,
312–331.
Sahimi, M.: 1984, Transport and dispersion in porous media and related aspects of petroleum
recovery, Unpublished PhD thesis, University of Minnesota, Minneapolis, Minnesota.
Sahimi, M.: 1995, Flow and Transport in Porous Media and Fractured Rock, VCH, Weinheim.
Sahimi, M. and Yortsos, Y.C.: 1985, ‘Pattern formation in viscous fingering: a diffusion-limited
aggregation approach’, Physical Review A 32, 3762–3764.
Sahimi, M.: 1987, ‘Hydrodynamic dispersion near the percolation threshold: scaling and probability
densities’, Journal of Physics A: Math. Gen. 20, L1293–L1298.
Sahimi, M. and Imdakm, A.O.: 1988, ‘The effect of morphological disorder on hydrodynamic dis-
persion in flow through porous media’, Journal of Physics A: Math. Gen. 21, 3833–3870.
Scott, G.J.T., Webster, R., and Nortcliff, S.: 1988a, ‘The topology of pore structure in crack clay soil
I. The estimation of numerical density’, Journal of Soil Science 39, 303–314.
Scott, G.J.T., Webster, R., and Nortcliff, S.: 1988b, ‘The topology of pore structure in crack clay soil
II. Connectivity density and its estimation’, Journal of Soil Science 39, 315–326.
Shante, V. and Kirkpatrick, S.: 1971, ‘An introduction to percolation theory’, Advances in Physics
20, 325–357.
Silliman, S.E.: 1990, ‘The influence of grid discretization on the percolation probability within
discrete random fields’, Journal of Hydrology 113, 177–191.
Simon, R. and Kelsey, F.J.: 1971, ‘The use of capillary tube networks in reservoir performance studies,
I. Equal viscosity miscible displacements’, Society of Petroleum Engineers Journal 11, 99–112.
Simon, R. and Kelsey, F.J.: 1972, ‘The use of capillary tube networks in reservoir performance
studies, II. Effect of heterogeneity and mobility on miscible displacement efficiency’, Society of
Petroleum Engineers Journal 12, 345–351.
Stanley, H.E.: 1977, ‘Cluster shapes at the percolation threshold: an effective cluster dimensionality
and its connection with critical-point exponents’, Journal of Physics A: Math. Gen. 10, L211–
L220.
Stauffer, D. and Aharony, A.: 1992, Introduction to Percolation Theory (2nd ed.), Taylor & Francis,
London.
Steele, D.D. and Nieber, J.L.: 1994a, ‘Network modeling of diffusion coefficients for porous media,
I. Theory and model development’, Soil Science Society of America Journal 58, 1337–1345.
Steele, D.D. and Nieber, J.L.: 1994b, ‘Network modeling of diffusion coefficients for porous media,
II. Simulations, Soil Science Society of America Journal 58, 1346–1354.
Stockmayer, W.H.: 1943, ‘Theory of molecular size distribution and gel formation in branched-chain
polymers’, Journal of Chemical Physics 11, 45–55.
Sutanto, E.: 1991, Liquid distributions in porous media, PhD thesis, University of Minnesota, Min-
neapolis, Minnesota, 283 pp.
Thompson, A.H., Katz, A.J., and Krohn, C.E.: 1987, ‘The microgeometry and transport properties of
sedimentary rock’, Advances in Physics 36, 625–694.
Toledo, P.G., Scriven, L.E., and Davis, H.T.: 1989, ‘Pore-space statistics and capillary pressure curves
from volume-controlled porosimetry’, Paper SPE 19618, 64th Ann. Tech. Conf. and Exhib. of the
SPE, Oct. 8–11, San Antonio, Texas.
Toledo, P.G., Novy, R.A., Davis, H.T., and Scriven, L.E.: 1990, ‘Hydraulic conductivity of porous
media at low water content’, Soil Science Society of America Journal 54, 673–679.
Topp, G.C.: 1971, ‘Soil-water hysteresis - the domain theory extended to pore interaction conditions’,
Soil Science Society of America Proceedings 35, 219–225.
Torelli, L.: 1972, ‘Computer simulation of the dispersion phenomena occurring during flow through
porous media, using a random maze model’, Pure and Applied Geophysics 96, 75–88.
Torelli, L. and Scheidegger, A.E.: 1971, ‘Random maze models of flow through porous media’, Pure
and Applied Geophysics 89, 32–44.
van Genuchten, M.Th.: 1980, ‘A closed-form equation for predicting the hydraulic conductivity of
unsaturated soils’, Soil Science Society of America Journal 44, 892–898.
Wang, J.S.Y. and Narasimhan, T.N.: 1992, ‘Distributions and correlations of hydrologic parameters
of rocks and soils’, in M.Th. van Genuchten, F.J. Leij and L.J. Lund (eds.), Indirect Methods for
Estimating the Hydraulic Properties of Unsaturated Soils, pp. 169–176. University of California,
Riverside.
Wardlaw, N.C., Li, Y., and Forbes, D.: 1987, ‘Pore-throat size correlation from capillary pressure
curves’, Transport in Porous Media 2, 597–614.
Washburn, E.W.: 1921, ‘Note on a method of determining the distribution of pore sizes in a porous
material’, Proceedings of the National Academy of Science 7, 115–116.
Wheatcraft, S.W. and Tyler, S.W.: 1988, ‘An explanation of scale-dependent dispersivity in hetero-
geneous aquifers using concepts of fractal geometry’, Water Resources Research 24, 566–578.
Wilkinson, D.: 1986, ‘Percolation effects in immiscible displacement’, Physical Review A 34, 1380–
1391.
Wilkinson, D. and Barsony, M.: 1984, ‘Monte Carlo study of invasion percolation clusters in two and
three dimensions’, Journal of Physics A: Math. Gen. 17, L129–L135.
Wilkinson, D. and Willemsen, J.F.: 1983, ‘Invasion percolation: a new form of percolation theory’,
Journal of Physics A: Math. Gen. 16, 3365–3376.
Winterfeld, P.H., Scriven, L.E., and Davis, H.T.: 1981, ‘Percolation and conductivity of random
two-dimensional composites’, Journal of Physics C 14, 2361–2376.
Witten, T.A. and Sander, L.M.: 1983, ‘Diffusion-limited aggregation’, Physical Review B 27, 5686–
5697.
Wong, P.-Z.: 1984, ‘The statistical physics of sedimentary rock’, Physics Today 41, 24–32.
Yanuka, M., Dullien, F.A.L., and Elrick, D.E.: 1984, ‘Serial sectioning digitization of porous media for
two- and three-dimensional analysis and reconstruction’, Journal of Microscopy 135, 159–168.
Yanuka, M., Dullien, F.A.L., and Elrick, D.E.: 1986, ‘Percolation processes and porous media, I.
Geometrical and topological model of porous media using a three-dimensional joint pore size
distribution’, Journal of Colloid and Interface Science 112, 24–41.
Yanuka, M.: 1989a, ‘Percolation processes and porous media, II. Computer calculations of percolation
probabilities and cluster formation’, Journal of Colloid and Interface Science 127, 35–47.
Yanuka, M.: 1989b, ‘Percolation processes and porous media, III. Prediction of the capillary hysteresis
loop from geometrical and topological information of pore space’, Journal of Colloid and
Interface Science 127, 48–57.
Yanuka, M.: 1992, ‘Percolation theory approach to transport phenomena in porous media’, Transport
in Porous Media 7, 265–282.
Yortsos, Y.C., Satik, C., Bacri, J.-C., and Salin, D.: 1993, ‘Large-scale percolation theory of drainage’,
Transport in Porous Media 10, 171–1195.
Yuan, H.H. and Swanson, B.F.: 1986, ‘Resolving pore space characteristics by rate-controlled
porosimetry’, SPE/DOE paper 14892, Fifth SPE/DOE Symp. Enhanced Oil Recovery, Tulsa,
Oklahoma, April 20–23.
Zhou, D. and Stenby, E.H.: 1993, ‘Interpretation of capillary pressure curves using invasion percola-
tion theory’, Transport in Porous Media 11, 17–33.
View publication stats

Percolation Theory and Network Modeling Application

Uploaded by

Copyright:

Available Formats

Percolation Theory and Network Modeling Application

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Percolation Theory and Network Modeling Application

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Percolation Theory and Network Modeling Applications in Soil Physics

Article in Surveys in Geophysics · January 1998

The user has requested enhancement of the downloaded file.

BRIAN BERKOWITZ1 and ROBERT P. EWING2

Percolation theory is a branch of probability theory dealing with properties of

Surveys in Geophysics 19: 23–72, 1998.

pore-scale events. Network models and percolation theory are complementary:

2. The Mathematics of Percolation

2.1. BASICS AND DEFINITIONS

Many properties of a macroscopic system are essentially determined by the con-

2.2. CLUSTER PROPERTIES

Another concept to be introduced is that of the “cluster”, which is simply a group of

dimensions, becomes increasingly complicated, but the basic “recipe” is apparent

Since (as shown above) ns is a polynomial function of p, S can be expressed as

where is an exponent. Indeed, an approximation such as Equation (4) (e.g., Alon

Rs2 = (jri r0 j2 )=s: (5)

2.3. FLOW THROUGH PERCOLATION NETWORKS

which is determined by the exponents  and  according to

Conveniently,  and  can be determined in simple systems, and then used to

2.4. CONTINUUM PERCOLATION

such as fractured rocks. As an example, Figure 3 illustrates a computer-generated

c = 1 exp ( Bc=8): (12)

c = 1 exp [ (BcV=hVex i)] (14)

3. Percolation Theory and Network Models of Porous Media

3.1. EARLY NETWORK MODELS

Early conceptualizations of pore space topology involved considering bundles of

3.4. PORE-SCALE EVENTS

Attempts to model macroscopic behavior based on pore-scale events must necessar-

3.5. PORE STRUCTURE

A detailed analysis of dispersion phenomena in porous media using network mod-

4. Percolation Theory and Soils: Assumptions, Problems and Variations

Throughout much of the preceding discussion of percolation theory and network

4.1. MEDIUM EFFECTS: LARGE-SCALE STRUCTURE

4.2. MEDIUM EFFECTS: PORE-SCALE CORRELATIONS

more conveniently, tetrakaidecahedra (the union of a cube and an octahedron, with

4.3. FLUID EFFECTS: VISCOSITY

4.4. FLUID EFFECTS: BUOYANCY

from a 2D model to a 3D phase space encompassing Ca, M and Bo (Figure 10).

5. Some Recent Work Relating Percolation Theory and Network Models to

proximity parameter quite different from the (p pc ) form typical of percolation

View publication stats

You might also like

Rs2 = (jri r0 j2 )=s: (5)

which is determined by the exponents and according to

Conveniently, and can be determined in simple systems, and then used to

c = 1 exp ( Bc=8): (12)

c = 1 exp [ (BcV=hVex i)] (14)