Journal

of
Hydrology
ELSEVIER Journal of Hydrology187 (1996)195-213

Transports in fractal porous media

P.M. Adler
IPGP, 2, Place Jussien, 75252 Paris Cedex 05, France
Received 10 January 1994; accepted 19 October 1995

Abstract

A general overview of continuous deterministic fractals is given. The method of construction, the
fractal dimension and the major features of transport are summarized. Then the three major single
phase transports are briefly addressed.

I. Introduction

It is certainly well known and admitted now that the concept of fractals is relevant to the
study of real porous media, since the introduction of this concept by Mandelbrot (1982).
The following presentation is mostly based on our own work and surveys all our con-
tributions on deterministic fractals. This presentation is organized as follows. In Section 2,
the most classical continuous fractals are presented. The construction scheme due to
Hutchinson (1981) is recalled, as well as a short introduction to the concept and to the
determination of the fractal dimension.
Section 3 gives a unified approach to the description of porous media and to the analysis
of transports. Porous media are viewed as materials with multiple scales. Depending upon
the respective orders of magnitude of these scales, three basic situations are distinguished.
The most important one occurs when the material is homogeneous at the large scale, but
contains a fractal microstructure. In the second part of this section, the general properties
of the transports are summarized for each basic situation.
The rest of this paper deals with three transports of general interest: conduction, con-
vection, Taylor dispersion. This is always done in media which are fractal at the micro-
scale and spatially periodic at the large scale. Our attention is focused on the presentation
of the transport itself, on some numerical results and on some analytical arguments.
Conduction is addressed in Section 4 for two- and three-dimensional structures derived
from the Cantor set. It is quite interesting that most macroscopic conductivities follow the
0022-1694/96/$15.00 © 1996 - Elsevier Science B.V. All fights reserved
PII S0022-1694(96)03096-X
196 P.M. Adler~Journal of Hydrology 187 (1996) 195-213

classical Arehie's law, i.e. a power law as a function of porosity. Simple renormalization
arguments compare well with the data.
Convection is addressed in Section 5 for the same structures. The most interesting case
from a theoretical viewpoint is one-dimensional flow in Sierpinski carpets since numerical
data, approximate renormalization arguments and exact derivation coincide. It is also the
only case where permeability can be expressed as a function of porosity and fractal
dimension. However, one should not forget that the situation is much worse for two-
and three-dimensional flows.
Convection-diffusion processes are addressed in Section 6 by two different means, i.e.
by random walks and the method of moments. Some examples are given, but no
renormalization argument could be derived.
Some general remarks end up this paper in Section 7.

2. Serf-similar sets

Sets which are exactly self-similar provide a good introduction to fractals. As a tentative
definition (Mandelbrot, 1982), a body, a shape or a mathematical entity is self-similar
when each of its pieces is geometrically similar to the whole.
A general mode of construction was devised by Hutchinson (1981) through the action of
a set of similitudes on geometric sets. A similitude S is a geometric transformation which
is defined as the composition of a rotation, a homothety (or a contraction with a ratio r < 1)
and a translation by a vector t.
Several similitudes can be combined in order to generate a fractal. Suppose that one has
a set of similitudes {Si, i - 1. . . . . M} and a compact set K0; then successive sets K1, K2. . . .
can be obtained by application of the recursion relationship
M
K• - io.1Si(K, _ 1) (1)
This formula means that all the similitudes are applied to the set K~,t and that K~ is equal
to the union of all'these M 'reduced' sets. The index N is called the generation number.
For instance, the Sierpinski carpet (denoted by SO) can be obtained by M,- 8 similitudes
with a contraction ratio r - 1/3. The eight transformed images of the square which is the
initial compact Set K0, are rearranged into a new square as indicated in :Fig. l(a); this
process can be iterated. The iteration of the recursion process is shown in the right side of
Fig. 1. The result of an infinite number of iterations is called the Sierpinski carpet; this set
is invariant by successive applications of the elementary similitudes.
This invariance provides us with an operational definition of the fractal set constructed
by means of the Hutchinson's process; the fractal set K is the only set invariant by the set
of similitudes {Si, i - 1 . . . . . M}, i.e.
M
K- u S (K) (2)
i-1

The invariant and the unique (irrespective of the starting set K0) characters of K can be
rigorously proved (cf. Hutchinson, 1981).
 P.M. Adler~Journal of HydrolosY 187 (1996) 195-213 197

!if!~~ ~iii . . . . . . . !'!i

L
J i .....................

l
Fig. 1. Two-dimensional deterministic fractals at stage N - 1 and N + I. The porous medium is supposed to be
composed of unit cells identical to these ones. (a) SC; Co) BAH; (c) BAH.
198 P.M. Adler~Journal of Hydrology 187 (1996) 195-213

Another classical example, the Ben Avraham and Havlin carpet is displayed in Fig. l(b)
and (c). The contraction ratio is equal to 1/5 and M = 16. Three-dimensional examples are
displayed in Fig. 2. The fractal foam (denoted by FF) is a three-dimensional version of the
Sierpinski carpet; the parameters for this foam are equal to M - 26, r = 1/3. The Menger
sponge MS1 can be constructed as the fractal foam (cf. Fig. 2(b)); its parameters are
MSI: r=1/3, M-20 (3)
Usually, the concept of dimension is defined by the number of independent vectors which
can be found in a given space. A very different approach is the one corresponding to the
similarity dimension D~. In order to give an intuitive feeling for this dimension, one can
start with standard objects such as segments, squares and cubes. Such objects can be
covered by N(r) reduced versions of themselves where r is the contraction ratio. For
these simple objects, N(r) is obviously equal to
N(r) ffi (1/r) d (4)
where d is the usual spatial dimension. This formula can be solved for d
d - Log N(r)
Log 1/r (5)
More generally, the similarity dimension of self-similar objects constructed from a set of

I L .' e

<. h ~, 'f'~y -..< h ,

X

a. b.
Fig. 2. Three-dimensional deterministic fractals at stage N - 2. The porous medium is supposed to be composed of
unit cells identical to these ones. (a) FF, the inner cubes correspond to the solid phase; Co) both phases arc
continuous. MS1 is obtained when the liquid phase corresponds to L1, and MS2 in the opposite case.
 P.M. Adler~Journal of Hydrology 187 (1996) 195-213 199

M similitudes with the same contraction ratio r is defined as

Log M
D~i- Log 1/------~ (6)

since
N(r)-M (7)

This definition can be illustrated by the previous examples. The Sierpinski carpet consists
of M - 8 versions of itself reduced in the ratio r - 1/3. Hence, the similarity dimension of
the Sierpinski carpet is equal to

Dsi - Log 8.1.893 (8)

Log 3

Another Menger sponge can be introduced under the name of MS2, as illustrated in Fig. 2;
the liquid and solid phases are interchanged with respect to MS1. Note that MS2 is not a
genuine fractal and that its dimension is equal to 3.

3. Transports in fractals

We can now systematize both our approach to the description of porous media and our
analysis of the major transports which may occur in porous media (cf. also Adler, 1989,
1990, 1992).

3.1. Porous media with multiple structures

Historically, porous media were first modelled as structures which were statistically
homogeneous (in contrast with fractals which possess a dilational symmetry); this is the
reason why most of the anterior interest on porous media was focused on the former
structures which are adequately schematized by spatially periodic structures. Now the
fractal character of geological porous media is well established and we wish to give a
few examples of 'theoretical' media which will clearly display several symmetries (cf.
Adler, 1992).
A porous medium can be characterized by its two extreme scales I and L (Fig. 3). I is a
representative length scale of the elementary particles which compose the solid phase; L
corresponds to the external boundaries of the medium. It is assumed that
I <<L (9)
Let us assume that there is a fractal structure between two length scales 11 and/2; note that
in contrast with the mathematical fractals, real fractals are characterized by the fact that
these two scales are finite. However, we shall also assume that
l I <:<: l 2 (10)
Typically, these two scales should differ by at least one order of magnitude.
2011 P.M. Adler~Journal of Hydrology 187 (1996) 195-213

V 5V
solid

t
< -L > ,
liquid
a b
Fig. 3. Porousmediumwith its variouslengthscalesL and I. (b) is an enlargedview of (a).

Many different physical situations may be constructed by comparing the fractal lengths
11 and 12 to the lengths l and L. Roughly, three cases can be distinguished; all these
situations were illustrated by real examples in Adler (1992). The simplest is the case where
11~1; 12-L (11)
i.e. the whole porous medium is spanned by a fractal from the microscopic to the macro-
scopic scale. Such a situation is shown in Fig. 4(a).
The second interesting situation is the case where
l~ -- t << 12 << L (12)

i.e. a fractal structure existsbelow a certain length scale and a homogeneous structure
exists at large scale. A theoretical example isgiven in Fig. 4Co) with a spatially periodic
array whose unit cell is a Sierpinski gasket.
More sophisticated situations may be imagined. Let us just consider one of them, where
two different fracta! structures coexist in the same material; they are characterized by the
length scales (11, ]2) a~d (l:' 1, l'~). It.may b e assumed that
11 --~ l; 12 << 1'1; ' l'2 = L : . . . . (13)
This is tentatively illustrated by the network displayed in Fig. 4(c). An homogeneous
structure exists between 12 and l' 1. A real example of this situation could be a fractured
massif made o f limestone; at the small scales (i,e. for length scales smaller than/2), the
pore space is fractal, then homogeneous above 12; at the large scale, i.e. for scalesranging
from l' 1 to L, the fracture network itself is fractal.
Other structures could be easily generated, but it is more important to examine the
consequences of these structures on the analysis of transport processes.
 P.M.Adler~Journalof Hydrology 187 (1996) 195-213 201

< L •
a b

A
1,

Fig. 4. Typicalphysicalsituations.(a) the whole mediumis spannedby a fractal.Co) mediumhomogeneousat the

large scale and fractal at the microscale. (c) medium with two fractal substructures; the shadowed areas are
identical to that on the right,which consistsof a homogeneousnetworkof fractal subnetworks,which may be for
instance Pascal-Sierpinskigaskets.

3.2. Transports in fractal porous media

A standard problem with flow in porous media may be stated as follows. Let us consider
a solid porous material which occupies a macroscopic volume V; at the external boundary
0Vof this volume, various conditions are applied, such as pressure, on the fluid which fills
the voids of the porous medium as is shown in Fig. 4. The interface between the solid and
liquid phases is denoted by Sp. If it is assumed that the fluid is Newtonian and that the
Reynolds number is low (the pore dimensions are generally small), flow is locally
governed by the Stokes equations
V p - pV2v, V-v - 0 (14)
where v, p and/~ denote velocity, pressure and viscosity, respectively. These equations
must be supplemented by the no-slip condition of the fluid at the surface Sp of the solid
v - 0 on Sp (15a)
Overall boundary conditions should be added to aV; usually some parts of 0V1 are
impermeable and known external pressures arc imposed on some other parts 0V2: this
m a y b e summed up by
v ffiO on OV1, p=O on OV2 : (15b)
202 P.M.Adler~Journalof Hydrology 187 (1996) 195-213
Three basic situations can be encountered in porous media whose structure is fractal at the
microscale.
The first one is depicted in Fig. 4(a), where flow occurs through the fractal structure, i.e.
a source or a sink is assumed to exist inside the fractal. Such a situation can be charac-
terized by the relation between the flow rate J which goes out of the source and a measure
P of the pressure difference between the source and the outer boundary. The most frequent
situation corresponds to a source located at one of the external vertices, it is generally
expected that
J/e l, N.-.oo (16)
where N is recalled to be the generation number of the deterministic fractal. This formula
can be put under an equivalent form with the help of the ratio ldl2 (cf. F.~I. (11)); if r
denotes the contraction ratio, then ll/12 is equal to a; hence

J (11~, /~--.*0 (17)

It is generally believed that the proportionality factor in this expression depends upon the
detailed features of the medium while the exponent is independent of those details and
depends only on the gross features. This is the so-called universality property which will
be more precisely stated later.
Another important property is the speed with which these asymptotic laws are obtained,
i.e. the value of the ratio l~/12 (or equivalently of the generation number) below which a
power law is representative of the numerical results. Only one order of magnitude is
generally sufficient to reach the asymptotic regime; this is an amazingly good situation.
Were this not the case, fractals as models of real porous media would not be useful as they
are.
The second basic situation is the one depicted in Fig. 40)) where flow occurs in a porous
medium which is homogeneous at the large scale and fractal at the small one. Because of
homogeneity, a permeability K can be defined and introduced through Darcy law; the
seepage velocity ~ is linearly related to the macroscopic pressure gradient Vp

- - ~ (18)

The porous medium has been supposedto be isotropic. It is important to emphasize that K
only depends upon the geometry. In general, K is expected to scale as J/P, i.e. with the
same exponent

K ~ \~2] (19)

The last basic situation corresponds to flow around fractal structures, such as the Cantor-
Taylor brush or random aggregates (Adler, 1987; Vignes-Adler et al., 1987), hut such
situations are not of interest in the present context.
All these developments are valid almost unchanged for other transports, such as heat
conduction or diffusion when one is concerned by the analysis of the steady state. One has
to use the macroscopic heat conduction tensor instead of permeability.
 P.M. Adler~Journal of Hydrology 187 (1996) 195-213 203

The analysis of an unsteady transport process such as diffusion offers some new
features. Consider a solute which is released inside a porous medium at an arbitrary origin
O at time t - 0. The fluid is motionless and the solute diffuses through it. The average
position vector R is equal to zero whatever the structure, while the mean square displace-
ment dyadic R 2 is a time dependent tensor T(t)
S 2 I, T(t) (20)
In spatially periodic structures, it is already known that in the limit of very long times
R 2 - 2D't, t " * o~ (21)
where D* is the usual macroscopic diffusion tensor. Note that we use here the method of
moments specialized to the case__where there is no fluid motion. The specific feature of an
homogeneous medium is that R 2 is a linear function of time.
Now consider a Brownian particle released in a fractal porous medium such as the one
depicted by Eq. (11); it may be the Sierpinski gasket of Fig. 4(a). Here the solute as it
diffuses through the liquid phase, encounters larger and larger structures which possess
different scales, but which are otherwise self-similar. By a natural analogy with Eq. (21), it
is expected that R 2 reaches an asymptotic regime after a long time when the solute particle
has crossed many different scales. Eq. (21) is extended as
R-f - 2 ~ t 2/d" (22)
which is characterized by the introduction of the exponent dw; the homogeneous medium
is retrieved by dw - 2. This formula can be commented in precisely the same way as Eq.
(21). The 'diffusion tensor' D* and dw depend upon the fractal dimension, but D* is
believed to depend on the details of the fractal structure such as the precise shape of
the inclusions, while dw should not depend on them.
The two exponents ~ and dw correspond to the steady and the unsteady aspects of the
same phenomenon and they are related one to another (see e.g. Adler, 1992).
Another transport of general interest is Taylor dispersion where convective and diffu-
sive effects are mixed. Again in order to facilitate our understanding, let us recall the major
features of Taylor dispersion in an homogeneous (or spatially periodic) medium. The
mean displacement vector k is a linear function of time
R ~ ~'t, t---, oo (23)
where ~" is the interstitial velocity. The mean-square displacement is a simple general-
ization of Eq. (21)
( R - ~ ) 2 ___2/~t, t--* oo (24)
where D* is the dispersivity tensor.
These results may be generalized in the same way as the diffusive laws in fractal
structures. The example of Fig. 4(a) can be used again. A fluid is flowing through this
fractal medium and a Brownian particle is introduced into the fluid at point A. It is
qualitatively expected that the two first moments will scale as power laws
oc t2/din' ' (R _~)2 octt/dTv2 (25)
204 P.M. Adler~Journal of Hydrology 187 (1996) 195-213

where :the two exponents dTD1and dTD2are expected to depend upon the fractal structure
and upon the Peclet number. :
It might be useful to end up this general presentation by two remarks. Only power laws
were given here, but it might happen that they degenerate. For instance, they may
degenerate into constants, which is easily taken into account by exponents equal to
zero. But they can also degenerate in a more serious manner, namely into logarithmic
laws. Hence, this possible source of difficulty should be kept in mind when a n e w fractal
structure is,addressed. .-

Another general presentation of the previous properties could have been given since it
would be well suited to homogeneous porous media with fractal microstructures. Since
several macroscopic properties denoted by the generic symbol K are only functions of the
geometry of the medium (cf. Adler, 1992), the fractal dimension which is a geometric
parameter may appear in the expression of K. Hence, one would have
g l g ( B , Of .... ) (26)
Such a relation would have been practically" very useful, but it is not often verified in
theory or in practice.

X Z

.°.

I. . . . . . . . . . . _ _ J-- -.._ .J
Fig. 5. A porousmediumbasedon a Sierpinskicarpet(b - 3, 1- 1). The generationnumberis equalto 2; several
cells are displayedfor clarity.
 P.M. AdlerJJournalof Hydrology 187 (1994) 195-213 205

3.3. The studiedstructures

The structures which are going to be studied are two: or three-dimensional. They are’all
supposed to be spatially periodic at the large scale, ‘i.e. the space can be partitioned into
identical cells of size h, which contain solid inclusions.
The simplest sort of deterministic fractals is the usual version of the two-dimensional
Sierpinski carpets, denoted by SC which is illustrated in Fig. 5. A second two-dimensional
structure has been devised by Ben Avraham and Havlin (1983) and is denoted hereafter by
BAH; a picture similar to Fig. 5 could be produced for this medium.
Two sorts of three-dimensional fractals were studied. The fractal foam (denoted by FFJ
is a three-dimensional version of the Sierpinski carpet which is displayed in Fig. 2(a).
The Menger sponge can be constructed in the same way as the fractal foam, but it should
be noticed that both phases of the Menger sponge are connected in contrast with the three
previous cases where only one phase was connected; therefore, the two possible Menger
sponges will be examined and they are denoted by MS1 and MS2 (Fig. 2(b)).

4. Conduction in continuous deterministic fiactals

4.1. General

Because of the large-scale periodicity of&e media under consideration, the thermal
problem can be solved by considering a single unit cell, with periodic boundary conditions
(Thovert et al., 1990). The heat flow is governed by the Laplace equation

V2T=Q (27)
where T is the local temperature, together with the no:flux boundary condition
n-VT-Q on S, (28)
where S, is the fluid-solid interface, with unit normal vector n. In addition, it is assumed
that either the macroscopic temperature gradient m or the average heat flux 4

4-1 &q (29)

70 J ar,
is specified. Here, 7. is the volume of the unit cell. These two quantities are related by the
symmetric positive definite conductivity tensor

(30)
which depends only on the geometry of the medium; thus, it can be simplified according to
its symmetry. For example, all the fractal structures presented above possess a cubic
symmetry; hence, u is a spherical tensor
o-al (31)
where I is the unit tensor.
206 P.M.Adler~Journalof Hydrology 187 (1996) 195-213
The Neuman problem (Eq. (27), Eq. (28)) is solved via a second-order finite-difference
formulation and a conjugate - gradient iterative scheme.
Arc represents the ratio between the size of the unit cell and of the smallest inclusion or
void. For a Sie~inski carpet at stage N (cf. Eq. (1)), it is equal to 3 N. The mesh spacing A is
the same in the three directions; it is a fraction 1/n of the smallest length which exists in the
medium under consideration. Thus, for a fractal foam at stage N, it is given by
A -(h/3N)/n (32)

4.2. Results and discussion

The macroscopic conductivity of the deterministic two- and three-dimensional fractals

presented above was systematically calculated. It might be useful to note that the study of
'one-dimensional' configurations does not present any interest. Consider for instance the
porous medium represented in Fig. 5; assume the liquid phase conducting and the solid
phase insulating; when the macroscopic temperature gradient is along the z-axis, the
temperature is obviously constant in any xy-section, and the macroscopic conductivity
is simply proportional to the cross section.
The complete data relative to two and three dimensions are displayed in Fig. 6.
Generally speaking, the macroscopic conductivity of fractals is expected to scale as a
power law of the construction stage N, in the limit of large N (cf. Eq. (16))
a ~ S N, N---.oo (33)
where S is a constant which only depends upon the gross features of the fractal, but not
upon its details; this is the so-caUed universality property which is verified in most cases.
Let us consider for instance a Sierpinski carpet where the removed squares can be replaced
by disks whose diameters are equal to the sides of the corresponding squares; then, S
should be the same for both cases while the proportionality constant is different in Eq.
(33), except may be for very particular situations.

1.0.
.-....
05 • "~. ..s,

o-

\x.,
0.1. • .s,

• 8AH

OO5
o i ~ ~ L
N
Fig. 6. The macroscopicconductivityo for the deterministicfractalsas a functionof the generationnumberN.
 P.M.Adler~Journalof Hydrology 187 (1996) 195-213 207
Thanks to the universality property mentioned above, poor discretizations of the
geometry can be used to calculate large generation numbers N, without any influence
on S.
The data relative to MS2 behave in a particular way because porosity increases with N
as smaller and smaller holes are added. Hence, it is physically expected that the macro-
scopic conductivity tends towards one, i.e.
SMS2" 1 (34)

It is clear from the numerical results that the asymptotic regime is not reached yet. The
other configurations, whether they are two- or three-dimensional, quickly reach the
asymptotic power law Eq. (33) with an excellent precision. For all these configurations,
the porosity e behaves as
e=e v (35)
N can be eliminated between a and ~; one obtains the classical Arehie's law
o"~ 8= (36)
with an exponent m given by
Log S
m - ~ (37)
Log e
The values of the exponent m which is also called the Archie's exponent, are the following
for the various kinds of fractals: 1.64 (FF), 1.89 (SC), 1.59 (MS1), 2.05 (BAH). It is truly
remarkable to observe that the values of m belong to the experimental range 1.3-2.5. The
Archie's law is in a sense very robust since it is extremely well verified for the fractals
studied here.
Finally, approximate derivations of the factor S can be obtained (ef. Thovert et al.,
1990). Two kinds of approximations were used. The first one applies to 'external flows'
when the solid phase can be considered as a suspension, i.e. for the Sierpinskl carpet, the
fractal foam and the Menger sponge type I. The second one applies to 'capillary flows'
when the conductive phase can be grossly viewed as a capillary network, i.e. for two other
types of fractals. In most cases, these derivations give good estimates of S.
This discussion can be ended up by a few general remarks on these results~ There is no
obvious relationship like Eq. (26). It should be emphasized that the Archie's law works
quite well in many cases. Finally, the recent computational developments render possible
the calculation of complex configurations of practical or of theoretical interest and the
check of conjectures.

5. Convection in continuous deterministic fraetals

5.1. General

The Stokes flow problem can be first stated in general terms.(cf. Adler, 1986, 1988;
LcmaRre and Adler, 1990), so that all the following subsections can be easily presented as
208 P.M.Adler~Journalof Hydrology 187 (1996)195-213
particular cases. We are only interested in the sort of double structure symbolized by Eq.
(12), where the porous medium is spatially periodic at the large scale and fractal at the
small scale. Hence, let us consider a unit cell, which is usually square or cubic, and which
is characterized b y a transversal dimension LT (cf. Fig. 1 and Fig. 2) and a volume z0
(generally equal to L 2 or L3). The external surface of the unit cellis denoted by 0~'0.
Inside the unit cell, there is a deterministic fractal structure characterized by its
generation number N. The solid surface inside the unit cell is denoted by SN,r. Since the
structure of the porous medium is spatially periodic at the large scale, the general
formalism presented by Adler (1992) applies and is briefly recalled here without proof.
The low Reynolds number flow of a Newtonian fluid is governed by the usual Stokes
equations .,
Vp-/zV2v, V'v- 0 (38)
where v, p and/~ are the velocity, pressure and viscosity of the fluid, respectively. In
general, v satisfies the conditions
v - 0 on SN,v (39a)

v is spatially periodic (39b)

where SN,v denotes the surface of the solid bundles inside the unit cell.
This system of equations and conditions applies locally at each point R of the interstitial
fluid. In addition, it is supposed that either the seepage velocity vector ~ is specified, i.e.

~. 1 [Rds.v-a prescribed constant vector (40a)

• 0 J0r0
or else that the macroscopic pressure gradient Vp is specified
V p - a prescribed constant vector (40b)
As is well known, these two quantities are related by the permeability tensor K such that
p . . K_.~---~ (41)
/z
which is the generalization of Eq. (18) to anisotropic media. Let us now solve these
equations in the following particular cases which depend upon the dimension.

5.2. One-dimensional flow

This is a very interesting situation from the theoretical standpoint since the numerical
results can be justified either by approximate renormalization arguments or by rigorous
proofs based on standard properties of harmonic functions.
The class of porous media which is considered here, is derived from a particular variety
of fractals known as Sierpinski carpets (Mandelbrot, 1982). These carpets can be con-
structed in various ways (cf. Gefen et al., 1980, 1983). Let us start with squares of side a;
each ~uare is divided into b 2 subsquares and 12 of these subsquares are removed from the
center of the initial :squares. Here l and b are integers; this division and removal process
 P.M.Adler~Journal of Hydrology 187 (1996) 195-213 209

can be continued and the construction stage is denoted by N. A three-dimensional porous

medium is finally obtained as illustrated in Fig. 5; it is composed of unit cells of side a, and
the removed subsquares are prolongated in solid square bundles. Thus, at a large scale, the
porous medium is spatially periodic, while it is fractal inside the unit cell.
Let the z-axis be parallel to the axis of the solid bundles (cf, Fig. 5) and the x- and y-axes
be parallel to the sides of the squares. It is a simple matter to realize that K may be written
as

K- K± 0 (42)
o r//
? ;:

The Stokes equations are greatly simplified when the flow is parallel to the z-axis. The
component w of the velocity field along the z-axis verifies
a2w a2w lap
ax2 + ~ " " p. az (43)

where dpldz is a constant that does not depend on the transversal coordinates x and y. Eq.
(43) was dealt with in the three different ways mentioned above. These three methods
show that

However, we ~ n go further and the construction stage N can be eliminated between the
porosity 8 and the permeability; hence
K oc e [L°g(~- t2)/b']/tL°s(t~- t2)/b2l (45)

On the other hand, the carpets N + I are the union of b 2-12 carpets N reduced in the ratio b;
a direct application of Eq. (6) yields
Log(b2 _ 12)
Dsi " Log b (46)

Dsl can be used to simplify Eq. (52) as

K oc e ( 4 - o ' O / ( 2 - o ~ ) (47)

a relation which appears very attractive on theoretical grounds (el. Jacquin and Adler,
1987). It is one of the few examples where the macroscopic property can be directly
related to the fractal dimension as envisioned in Eq. (26).

5.3. Two- and three-dimensional flows

Two- and three-dimensional flows in structures such as the Sierpiuski carpets, the
fraetal foam, the two types of Menger sponges were thoroughly studied both numerically
and analytically. However, the situation is not as satisfactory as it was for conduction.
210 P.M.Adler~Journalof Hydrology187 (1996)195-213
First, it is found that the Carman equation is much less reliable than the Archie equation
which works for most regular fractals. Second, the status of simple scaling arguments is
also quite uncertain; in any case, it requires a check by other means such as the numerical
one in order to make sure that it is correct. It should be noticed that a renormalization of the
viscosity does not yield a correct value of the exponent.
This observation provides a transition to the next point; all these discrepancies are
certainly due to the fact that in contrast with conduction, permeability depends upon
the square of a length; hence, it is thought to be much more sensitive to all the details
of the geometry and a small error in the estimation of the scaling argument may imply a
large imprecision in the determination of the exponent. Moreover, the numerical calcula-
tions are much more limited for convection than for conduction since the number of
unknowns is multiplied by 4. Hence, the hope that the fractal dimension appears explicitly
in the expression of permeability (cf. Eq. (26)) is not fulfilled except for one-dimensional
flows.

6. Taylor dispersion in continuous deterministic fractals

6.1. General

The medium is again characterized by a double structure, i.e. spatially periodic at the
large scale and fractal at the small scale. The flow field v is assumed to be known inside the
cell as a result of the previous studies.
A neutrally buoyant, spherical Brownian particle is injected at some arbitrary interstitial
position R' at time t - 0; this particle is convected by the interstitial fluid and simulta-
neously undergoes Brownian motion characterized by the diffusion coefficient D. In the
limit of long times, the two first moments of the probability distribution M1 and M2 verify
Eq. (23) and Eq. (24) because the medium is spatially periodic at the large scale. The mean
interstitial fluid velocity ~" is usually defined as

v"~. l lr vd3R (48)

L

The macroscopic dispersion tensor D* can be calculated in two different ways. The first
one (Brenner, 1980) can be summarized as follows. The general expression of D* is

°I, VBt'VBd3R
D-~" "~L
L
(49)

B is a vector field satisfying

B(R)-B(R) -R " (50a)

where h is a spatially periodic vector which is solution of

v - ~* - V . ( v ~ ) - D V 2 b (50b)
n . V k - n on Sp (SOc)
 P.M. Adler~Journal of Hydrology 187 (1996) 195-213 211
The second classical manner to determine D* is to perform a Monte Carlo calculation by
simulating the displacement of a large number of particles through the fluid. Let Rj be the
position of particle j at time t. During the time step At, this particle is convected by the
fluid and undergoes a displacement v(Rj).At; because of the Brownian motion, it will also
undergo a random displacement 6. The new position of particle j at time t + At is
Rj + v(Rj).At + 6j (51)
This calculation can be repeated a large number of times on a large number of particles.
The two first moments can be evaluated with the precise location of particles and D* is
immediately derived from Eq. (25); the time derivative of the first moment can be used as a
useful check since it yields a statistical estimation of ~*. Note that the value of the
equivalent diffusion coefficient is proportional to i6jl2/At; the proportionality coefficient
depends upon the precise random rule (cf. Salles et al., 1993).

6.2. Results

Our progress in this field are mostly summarized by Salles et al. (1993). One-, two- and
three-dimensional structures could be addressed. Considerable progress could be made in
one-dimensional dispersion because the system is quite simpler in this case.
The general shape of the numerical results is always the same as it can be seen in Fig. 7.
The dispersion coefficient is roughly constant up to a Peclet number of 1. After a short
transition regime,/)" follows a power law
D*
- ~ oc p e = (52)

Typical values of this exponent are ranging between 1 and 2.

No dramatic effect occurs when the construction stage is increased. This was observed
in all the configurations which were studied by Salles et al. (1993), i.e. in Sierpinski

D;/D
106

10'

lO 2

1 A

P
10 103 10s 10~

Fig. 7. The dimensionlesslongitudinaldispersion~///D as a functionof the PecletnumberPe' - h3/itD V-p for

Mengcr sponges. Data are for: N - 1 (A); 2 (I); 3 (-).
212 P.M. Adler~Journal of Hydrology 187 (1996) 195-213
carpets and Menger sponges. No really convincing renormalization arguments could be
derived for the exponent ¢xin Eq. (52), even in the simple case of dispersion in Sierpinski
carpets when the flow is parallel to the bundles.
We shall only present the results obtained for Menger sponges (cf. Fig. 7). When .the
data are presented with a Peclet number based on the interstitial velocity, it seems that the
dispersivity at high Peclet numbers converges toward a limit when the construction stage
N is increased. However, the data for MS2 can be presented in a different way. Instead of
comparing I~//D for various construction stages at a fixed Peclet number, we can compare
the dispersivity in sponges when the size of the unit cell h, the molecular diffusivity D and
the average pressure gradient ~ are kept constant. The result is shown in Fig. 7; 1)~//D is
plotted versus

p . ~_~-ff. 3e
K--~--2Pe (53)

where K is the permeability of the medium. It turns out that the dispersion does not depend
upon N any more for Pe > 1. Since the ratio K/h 2 is almost constant for any N (el. Lemaitre
and Adler, i990), this means that 1)~//D depends only upon a modified Peclet number

Pe' ,.e.Pe.. -- (54)

D
where the interstitial velocity ~" is replaced by the seepage velocity ~.

7. Concluding remarks

The fractal approach to real porous media has been proved to be useful in theqast 10
years and this structure cannot be ignored any more. Many data have been accumulated
either experimentally or numerically. Many of them can be reasonably justified by
approximate renonnalization arguments similar to the ones presented here. A significant
effort should be devoted to strengthen the theoretical aspect of these arguments which are
very often derived on an ad hoc basis. The continuous deterministic fractals offer a good
opportunity to the development of this theoretical analysis.
Other physical situations such as deposition, can be fruitfully investigated in fractal
structures. This was recently performed in Menger sponges. At low Peclet numbers,
deposition is limited by diffusion and takes place preferentially at the walls which limit
the largest pores, which play the role of reservoirs of particles. The narrower parts are
preserved until the final steps of the process and permeability diminishes relatively slowly.
At high Peclet numbers, deposition occurs where convection is able to provide fresh brine.
In the Menger sponge, this occurs in large channels. In any case, it occurs along the
preferential paths of the convecting fluid. Thus, permeability is significantly diminished
by a small amount of deposited matter. It would be quite useful to rationalize all these
observations.
Many other routes are also opened to further investigations, such as the analysis of
random structures.
 P.M. Adler~Journal of Hydrology 187 (1996) 195-213 213

References

Adler, P.M., 1986. Transport processes in fractals. VI, Stokes flow through Sierpinski carpets. Phys. Fluids, 29:
15-22.
Adler, P.M., 1987. Hydrodynamic properties of fractal floes. Faraday Discussion Chem. Soc., 83: 145-153.
Adler, P.M., 1988. Fractal porous media, Ill. Transversal Stokes flow through random and Sierpinski carpets.
Trsnsp. Porous Media, 3: 185-198.
Adler, P.M., 1989. Flow in porous media. In: D. Avnir (Editor), The Fractal Approach to Heterogeneous
Chemistry. John Wiley, pp. 341-359.
Adler, P.M., 1990. Fractal porous media. In: J. Bear and M.Y. Corapcioglu (Editors), Transports Processes in
Porous Media. Nato ASI Series E, Applied Sci., Vol. 202, Kluwer, pp. 723-743.
Adler, P.M., 1992. Porous Media: Geometry and Transports. Butterworth/Heinemann.
Ben Avraham, D. and Havlin, S., 1983. Exact fractals with adjustable fractal and fracton dimensionality. J. Phys.,
AI6, L559-L563.
Brenner, H., 1980. Dispersion resulting from flow through spatially periodic porous media. Phil. Trans. R. Soc.,
London, A297: 81-133.
Gefen, Y., Mandelbort, B.B. and Aharony, A., 1980. Critical phenomena on fractal lattices. Phys. Rev. Left., 45:
855-858.
Gefen, Y., Meir, Y., Mandelbrot, B.B. and Aharony, A., 1983. Geometric implementation of hypercubic lattices
with non-integer dimensionality by use of low lacunarity fractal lattices. Phys. Rev. Lett., 50: 145-148.
Hu~hiuson, J.E., 1981. Fmctals and self-similarity. Indiana Univ. Math. J., 30: 713-747.
Jacquin, C.G. and Adler, P.M., 1987. Fractal porous media II, Geometry of porous geological structures. Trans.
Porous Media, 2: 571-596.
Lemaitre, R. and Adler, P.M., 1990. Fractai porous media, IV, Three dimensional Stokes flow through random
media and regular fractals. Transp. Porous Media, 5: 325-340.
Mandelbrot, B.B., 1982. The Fractal Geometry of Nature. Freeman, San Francisco.
Sallies, J., Pr~rors, L., Delannay, R., Thorert, J.-F., Auriault, J.L. and Adler, P.M., 1993. Taylor dispersion in
random media and regular fractals. Phys. Fluids, AS: 2348-2376.
Thovert, J.F., Wary, F. and Adler, P.M., 1990. Thermal conductivity of random media and regular fractals. J.
Appl. Phys., 68: 3872-3883.
Vigues-Adler, M., Adler, P.M. and Gougat, P., 1987. Transport process along fractals. The Cantor-Taylor brush.
Phys. Chem. Hydrodyn., 8: 401--422.