Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

ARTICLE IN PRESS

Physica A 374 (2007) 1–14


www.elsevier.com/locate/physa

A fractional-order Darcy’s law


J. Alberto Ochoa-Tapia, Francisco J. Valdes-Parada, Jose Alvarez-Ramirez
División de Ciencias Basicas e Ingenieria, Universidad Autonoma Metropolitana-Iztapalapa, Apartado Postal 55-534,
Mexico D.F. 09340, Mexico
Received 16 May 2006; received in revised form 13 June 2006
Available online 14 August 2006

Abstract

By using spatial averaging methods, in this work we derive a Darcy’s-type law from a fractional Newton’s law of
viscosity, which is intended to describe shear stress phenomena in non-homogeneous porous media. As a prerequisite
towards this end, we derive an extension of the spatial averaging theorem for fractional-order gradients. The usage of this
tool for averaging continuity and momentum equations yields a Darcy’s law with three contributions: (i) similar to the
classical Darcy’s law, a term depending on macroscopic pressure gradients and gravitational forces; (ii) a fractional
convective term induced by spatial porosity gradients; and (iii) a fractional Brinkman-type correction. In the three cases,
the corresponding permeability tensors should be computed from a fractional boundary-value problem within a
representative cell. Consistency of the resulting Darcy’s-type law is demonstrated by showing that it is reduced to the
classical one in the case of integer-order velocity gradients and homogeneous porous media.
r 2006 Published by Elsevier B.V.

Keywords: Darcy’s law; Fractional constitutive equation; Volume averaging

1. Introduction

In many practical situations, flow of fluids in porous media is described by the Darcy’s law, which relates
macroscopic pressure gradients to the flow velocity vector in the fluid phase:
1
vD;b ¼  Kb .ðrPm;b  rb gÞ, (1)
mb

where b indicates the fluid phase, vD;b is the Darcy velocity vector, Kb is the so-called permeability tensor, Pm;b
is the macroscopic pressure, rb is the mass density, and g is the gravity vector. Although the Darcy’s law was
proposed from empirical arguments, several formalizations have been reported. Hubbert [1] showed that an
equation similar to Darcy’s law can be obtained from averaging equations of flow for fluids on the microscale.
Whitaker [2,3] reported a more rigorous approach that includes the use of spatial averaging theorems. This
allowed to describe the Darcy’s law with all the variables defined directly in terms of averages of microscale or
local quantities. Quintard and Whitaker [4] have pointed out that results from Whitaker [3] are limited to
Corresponding author. Fax: +52 22 28044900.
E-mail address: jjar@xanum.uam.mx (J. Alvarez-Ramirez).

0378-4371/$ - see front matter r 2006 Published by Elsevier B.V.


doi:10.1016/j.physa.2006.07.033
ARTICLE IN PRESS
2 J.A. Ochoa-Tapia et al. / Physica A 374 (2007) 1–14

porous media with small porosity gradient. To overcome this problem, Quintard and Whitaker [5] introduced
a generalized average, which is equivalent to smoothening the problem before averaging. By doing this, the
effects of porosity heterogeneities (e.g., fractured media) are aggregated into the permeability tensor. As can
be noted, the pre-smoothening procedure yields essentially a homogeneous media for which spatial averaging
procedures can be readily applied. The drawback of a pre-smoothing approach is that important information
on the porous media structure, such as fractures, may not be reflected in the computed permeability
properties.
Recently, fractional calculus has been used to model particle transport in porous media [6]. The idea is to
use fractional-order density gradients to recover, at least at a phenomenological level, the inhomogeneities of
the porous media. Given the structural analogy between the Newton’s law of viscosity and the Fick’s law of
particles transport, the usage of fractional Newton’s law for description of momentum transport in non-
homogeneous porous media could be considered. The usage of fractional derivatives to describe fluid motion
in porous media is not new. He [7] studied seepage flow in porous media and observed that fractional
derivatives can be used to describe deviations from Darcy’s law behavior. Fractional viscoelastic Newton’s
equations have been considered to describe diffusion of fluids in porous media with memory [8,9]. The
rationale behind this modeling approach is that viscous relaxation processes display non-local dependency on
fluid motion, so that relatively distant fluid points are correlated by wave-like transport. Ad hoc combinations
of time-fractional derivatives and spatial-integer gradients have been proposed to describe anomalous
dispersion introduced by fluid motion stacked around porous media inhomogeneities (e.g., fracture). This
phenomena is reflected as non-Gaussian particle dispersion with heavy tails [10]. Experiments demonstrating
anomalous dispersion and its connection to non-local transport modeled with fractional derivatives, have been
also reported [11]. From a different perspective, van Tonder et al. [12] proposed fractional geometry to model
and explain deviations of the Darcy’s velocity due to fractured rocks in aquifers. In this way, an integer-order
transport equation is used to describe the local fluid motion in a fractional geometry. In a subsequent paper,
Riemann et al. [13] used this approach to estimate the Darcy’s velocity on an experimental test site located in
South Africa. Experiments showed that the estimated natural-flow velocity obtained by using fractional
dimension geometry is about two-times higher than the velocity estimated by using the standard two-
dimensional method. Although this approach has been applied with reasonable success in the analysis of
hydraulic test [13], it introduces parameters for which no sound physical interpretation exits in the case of non-
integer geometries. Recently, Cloot and Botha [14] generalized the classical Darcy’s law by regarding fluid flow
as a function of a non-integer order (spatial) derivative of the piezometric head. This constitutive equation,
together with the law of conservation of mass are then used to derive a new equation for groundwater flow.
The underlying basic assumption is that the fluid flow at a given point of the porous medium is governed not
only by the properties of the piezometric field at the specific position but also depends on the global spatial
distribution of that field in soil matrix. Numerical comparisons of the resulting fractional Darcy’s law with a
fractional Barker’s model [12] showed the superiority of the former approach to explain deviations from the
classical Darcy’s law.
Motivated by the results described above, in particular those by Cloot and Botha [14], this work explores
some implications of using fractional laws in the structure of Darcy’s-type laws. By using spatial averaging
methods, we derive a Darcy’s-type law from a fractional Newton’s law of viscosity, which is intended to
describe shear stress phenomena in non-homogeneous (e.g., fractured) porous media. As a prerequisite
towards this end, we derive an extension of the spatial averaging theorem for fractional-order gradients,
which, to the best of our knowledge, has not been reported previously. The usage of this tool for averaging
continuity and momentum equations yields a Darcy’s law with three contributions: (i) similar to the classical
Darcy’s-type law, a term depending on macroscopic pressure gradients and gravitational forces; (ii) a
fractional convective term induced by spatial porosity gradients; and (iii) a fractional Brinkman-type
correction. In the three cases, the corresponding permeability tensors should be computed from a fractional
boundary-value problem within a representative cell. Consistency of the resulting Darcy’s-type law is
demonstrated by showing that it is reduced to the classical one in the case of integer-order velocity gradients
and homogeneous porous media.
The paper is organized as follows. Section 2 provides some phenomenological arguments for using
fractional Newton’s law. Section 3 presents and discusses the main results, including the structure of the
ARTICLE IN PRESS
J.A. Ochoa-Tapia et al. / Physica A 374 (2007) 1–14 3

resulting Darcy’s law. Section 4 presents some conclusions. An Appendix describes the derivation of a
fractional-order spatial averaging theorem.

2. A fractional Newton’s law of viscosity

The Newton’s law of viscosity is based on a linear relationship between the viscous stress and the rate of
strain. This law for a simple one-dimensional flow is represented in the form of a single component of the
stress tensor being equal to the viscosity ðmÞ times a velocity gradient,
qvy
txy ¼ m . (2)
qx
Eq. (2) can be derived from simple mechanical arguments as follows. Consider two parallel plates of fluid
located at positions x and x þ Dx, and moving at velocities vy ðxÞ and vy ðx þ DxÞ, respectively (see Fig. 1).
From basic mechanics, the friction force (relative to the position x) between the plates is proportional to the
momentum difference. That is,
F xy ¼ K v ðrADxÞðvy ðxÞ  vy ðx þ DxÞÞ, (3)
where A is the interplate area, r is the density, and K v is a viscous friction coefficient. Notice that rADx is the
mass of the plate. Since the Taylor series approximation for x is
qvy
vy ðx þ DxÞ ¼ vy ðxÞ þ ðxÞDx þ OðDxÞ,
qx
where
q2 v y Dx2 q3 vy Dx3
OðDxÞ ¼ 2
ðxÞ þ 3 ðxÞ þ .
qx 2! qx 3!
Eq. (3) becomes
 
qvy
F xy ¼ K v ðrADxÞ ðxÞDx þ OðDxÞ .
qx
In order to recover the Newton’s law when Dx ! 0 is taken, one requires the following conditions:

 K v rDx2 ! m, and
 K v rOðDxÞDx ! 0.

Since the viscosity m is constant and Dx ! 0, it follows that K v ! 1 (i.e.), K v must be size-dependent). Thus,
Dx2 must decrease at the same rate that K v . Notice that tv ¼ K 1
v can be seen as a time-constant for viscous
1=2
force transfer. The above arguments imply that Dx must decrease like tv if m is to remain constant. Viscous
effects that occur during a given time tv are constrained to a spatial scale Dx. This means that there cannot be

Fig. 1. Schematic diagram of two fluid plates moving in the same direction.
ARTICLE IN PRESS
4 J.A. Ochoa-Tapia et al. / Physica A 374 (2007) 1–14

large momentum transfer deviations from the average one. Hence, the entire fluid flow must have constant
hydraulic properties on scales larger than some very small representative elementary volume in order for
Newton’s law to hold with a constant viscosity. Finally, Eq. (2) is obtained by taking the normalization
txy ¼ F xy =A. It should be stressed that the above arguments are not a rigorous (e.g., statistical) derivation of
the Newton’s law; rather, such arguments will be used below to motivate the usage of fractional-order laws of
viscosity.
Large velocity variations at the pore scale in, e.g., heterogeneous aquifers, are caused by the motion of fluid
through a disordered porous medium. Here, the hydraulic conductivity at different locations can vary over
many orders of magnitude. Hence, viscous effects that occur during a given time tv are not necessarily
constrained to a spatial scale Dx. In this case, the viscosity effects at the position x can be the result of viscous
interactions at scales beyond Dx. While the Newton’s law only involves local effects (represented by the
gradient qvy =qx), a generalization should incorporate non-local effects. That is, the viscous force F xy at the
position x arises as a consequence of interactions between fluid plates at positions Dx; 2Dx; . . . . A naive
interpretation is that the viscous force F xy is the (weighted) contribution of gradients at scales
Dx; 2Dx; . . . . It is, however, expected that the information relative to the direct vicinity of the specific
point under consideration will have a greater influence on the fluid flow than the information dealing with
events taking place far away from that point [14]. Fractional calculus, an extension of integer-order calculus,
can be used to describe non-local velocity gradients [15]. Fractional calculus is now being used in many
scientific and engineering fields, including electrical networks, electromagnetic theory, probability and
statistics [15–17].
Integer-order derivatives depend only on the local behavior of a function, meaning the slope of the function
at an infinitesimally small interval. Fractional derivatives, however, are non-local functions that can be
expressed as a linear combination of a qth order left and a right sided derivative:

Dqd ¼ ½12 ð1  dÞDqþ þ ½12 ð1 þ dÞDq , (4)


where d 2 ½1; þ1. The operator Dqþ ,
known as the Riemann–Liouville operator, is the derivative of a
function from 1 to x, while Dq , known as the Weyl operator, is the derivative from x to þ1 [17]. While the
Riemann–Liouville derivative has memory over the function from 1 to x, the Weyl operator induces
memory effects over the function from x to þ1. Both the Riemann–Liouville and the Weyl fractional
derivatives are the limit of a weighted average of the values of the function. For instance, the series definition
of the Riemann–Liouville operator is [16]

1 1 X1
Gðk  qÞ
Dqþ f ðxÞ ¼ lim q
f ðx  kDxÞ, (5)
Dx!0 GðqÞ Dx k¼0 Gðk þ 1Þ
where GðrÞ is the gamma function of r. In order to derive a fractional analogous to the Newton’s law of
viscosity, one should consider the generalized Taylor series [18]:
X
1
Dxnþq
vy ðx þ DxÞ ¼ Dnþq
d ðvy ðxÞÞ
k¼1
Gðn þ q þ 1Þ
Dxq
¼ Dqd ðvy ðxÞÞ þ OðDxq Þ, ð6Þ
Gðq þ 1Þ
where q is a rational number and Dqd , the qth derivative, is valid for both integer and fractional-order
derivatives. In Eq. (6), the residual is
X
1
Dxnþq Dxq
OðDxq Þ ¼ Dnþq
d ðvy ðxÞÞ  Dqd ðvy ðxÞÞ .
k¼1
Gðn þ q þ 1Þ Gðq þ 1Þ
Using Eq. (6) in Eq. (3), one has
 
Dxq
F xy ¼ K v ðrADxÞ Dqd ðvy ðxÞÞ þ OðDxq Þ . (7)
Gðq þ 1Þ
ARTICLE IN PRESS
J.A. Ochoa-Tapia et al. / Physica A 374 (2007) 1–14 5

Taking the limit as Dx ! 0 yields a fractional Newton’s law of viscosity:


txy ¼ mq Dqd ðvy ðxÞÞ, (8)
where mq is a fractional viscosity parameter. Note that the influence of the point x þ kDx on the viscous
transport is accounted for in the model by weighting the information with a factor W k ðqÞ ¼ Gðk  qÞ=
Gðk þ 1Þ. The fractional Newton’s law (8) is valid as Dx ! 0 when
qþ1
 K vGðqþ1Þ
rDx
! mq , and
 K v rOðDxq ÞDx ! 0.

In contrast to the non-fractional case, when a limit is taken as Dx ! 0, it must be hold that K v ! 1 at the
same rate as Dxqþ1 ! 0 in order for mq to be constant. Thus, Dx must decrease at the same rate as tqþ1 v .
Finally, a (symmetrized) multidimensional fractional Newton’s law of viscosity is obtained from Eq. (8) as
s ¼ mq ½rq vðxÞ þ rq vðxÞT ,
where rq is a qth order gradient.

3. A Darcy’s-type law from spatial averaging

Some implications of having a fractional Newton’s law on the structure of a Darcy’s-type law will be studied
in this section. To this end, consider a single-phase flow in a rigid porous medium (see Fig. 2), where b is used
to denote the fluid phase and s is used to denote the solid phase. The boundary value problem describing the
flow in the microscopic scale is
 rpb þ rb g þ mq rqþ1 vb ¼ 0; in the bphase,
=.vb ¼ 0; in the bphase,
vb ¼ 0; at Abs , ð9Þ
where, for simplicity in notation, we have used the expression
sb ¼ mq rq vðxÞ (10)
as the fractional Newton’s law. That is, we have assumed, without lost of generality, that rq vðxÞ is actually
symmetric. Here Abs represents the b2s interface contained within the macroscopic region illustrated in
Fig. 2.

Fig. 2. Porous medium and averaging volume with characteristic lengths.


ARTICLE IN PRESS
6 J.A. Ochoa-Tapia et al. / Physica A 374 (2007) 1–14

Consider a representative averaging volume V of the porous medium (see Fig. 2). Given some quantity cb
associated to the b-phase, its superficial average hcb i is defined as
Z
1
hcb i ¼ c dV , (11)
V Vb b

where V b is the b-phase volume contained into V. Note that b ¼ V b =V is the b-phase volume fraction
(porosity) of the representative volume V. The intrinsic average is defined by
Z
1
hcb ib ¼ c dV ,
V Vb b

so that

hcb i ¼ b hcb ib . (12)

In the sequel, the latter expression will be useful to reveal spatial porosity gradient effects on observable
macroscopic velocities.

3.1. Averaging of the continuity equation

Applying the averaging operator (11) and the spatial averaging theorem (SAT) (see Appendix) to the
continuity equation =.vb ¼ 0, one obtains
Z
1
.
= hvb i þ nbs .vb dS ¼ 0,
V b Abs
where nbs is the outer normal vector directed away from the b-phase. On the basis of the non-slip condition
vb ¼ 0, at Abs , the above equation is simplified to
=.hvb i ¼ 0.
Since hvb i ¼ b hvb ib , one obtains an alternate form of the continuity equation given by
=.hvb ib ¼ 1 . b
b rb hvb i . (13)

3.2. Averaging of the momentum equation

Similarly, applying the averaging operator to the fractional Stoke’s equation rpb þ rb g þ mq rqþ1 vb ¼ 0,
in the b-phase, gives hrpb i þ hrb gi þ hmq =.rq vb i ¼ 0. Variations of density and viscosity within the
averaging volume can be ignored, allowing to write this equation as

hrpb i þ b rb g þ mq h=.rq vb i ¼ 0. (14)

One can use the SAT (see Appendix) to represent the average of pressure gradient as
Z
1
hrpb i ¼ rhpb i þ nbs pb dS.
V Abs

While the superficial average hpb i is generally preferred in the analysis of flow in porous media, the intrinsic
average hpb ib is more convenient because it more closely corresponds to the measured value that one could
impose as a boundary condition. By using hpb i ¼ b hpb ib , one obtains
Z
b b 1
hrpb i ¼ b rhpb i þ hpb i rb þ nbs pb dS.
V Abs
ARTICLE IN PRESS
J.A. Ochoa-Tapia et al. / Physica A 374 (2007) 1–14 7

As usual in spatial averaging methods [19], it is convenient to decompose the point values of the pressure
according to pb ¼ hpb ib þ peb , where p~ b denotes local pressure fluctuations. This leads to the non-local form for
the average of the pressure gradient
Z Z
1 1
hrpb i ¼ b rhpb ib þ hpb ib rb þ nbs hpb ib dS þ nbs p~ b dS. (15)
V Abs V Abs

Since hpb ib is constant into the averaging volume, when the usual length-scale constraints are satisfied (i.e.,
l b 5ro 5L with reference to Fig. 2) the third term in the right hand side of the above equation can be developed
as follows [19]:
Z Z !
1 b 1
nbs hpb i dS ¼ nbs dS hpb ib
V Abs V Abs
¼  ðrb Þhpb ib ,
R
where the equality ð1=VÞ A nbs dS ¼ ðrb Þ was used [19]. Using this equation in Eq. (15) provides the
bs
following expression for the average of the gradient:
Z
1
hrpb i ¼ b rhpb i þ nbs p~ b dS. (16)
V Abs

Substitution of Eq. (16) into Eq. (14) provides


Z
1
b
b rhpb i  nbs p~ b dS þ b rb g þ mq h=.rq vb i ¼ 0. (17)
V Abs

Now, let us consider the averaging of the viscous term mq h=.rq vb i. Application of the SAT yields
Z
q q 1
. .
h= r vb i ¼ = hr vb i þ nbs .rq vb dS.
V Abs

By considering the spatial decomposition vb ¼ hvb ib þ v~ b , one can repeat the procedure used with the pressure
gradient to obtain
Z
1
h=.rq vb i ¼ =.hrq vb i  rb .rq hvb ib þ nbs .rq v~ b dS.
V Abs

An application of the fractional SAT (fSAT) (see Appendix) to the term hrq vb i leads to
" Z # Z
q 1 qþ1 1 1
h=.r vb i ¼ r hvb i þ =. nbs J ð1qÞ ðvb Þ dS  rb .rq hvb ib þ nbs .rq v~ b dS,
q V Abs V Abs

where J ð1qÞ ðvb Þ is a ð1  qÞth order fractional integral of the velocity field vb [11]. The second term in the RHS
of the above equation can be decomposed as
" Z # " Z ! # " Z #
1 1 b 1
r. nbs J ð1qÞ ðvb Þ dS ¼ r. nbs J ð1qÞ ð1Þ dS hvb i þ r. nbs J ð1qÞ ð~vb Þ dS .
V Abs V Abs V Abs

From the fSAT (see Appendix) it is not hard to show that


Z
1 1
nbs J ð1qÞ ð1Þ dS ¼  =q b ,
V Abs q

so that
" Z # " Z #
1 1 qþ1 b 1 q b 1
r. nbs J ð1qÞ ðvb Þ dS ¼  ðr b Þhvb i  = b .=hvb i þ r. nbs J ð1qÞ ð~vb Þ dS .
V Abs q q V Abs
ARTICLE IN PRESS
8 J.A. Ochoa-Tapia et al. / Physica A 374 (2007) 1–14

Consider the following order of magnitude estimates


Z !
1 q v~ b
nbs .= v~ b dS ¼ O qþ1
V Abs lb

and
" Z # !
1 v~ b
r. nbs J ð1qÞ ð~vb Þ dS ¼ O ,
V Abs Ll qb

where l b is the porous characteristic length, and we have made use of the fact that Abs =V is on the order of
l 1
b . So that if the length-scale constraint l b 5L is satisfied, the viscous term is simplified to
Z
1 1 1 1
hr.rq vb i ¼ rqþ1 hvb i  ðrqþ1 b Þhvb ib  =q b .=hvb ib  =b .=q hvb ib þ nbs .=q v~ b dS.
q q q V Abs

Using the relation given in Eq. (12), the above equation is modified to
Z
b 1 1
hr.rq vb i ¼ rqþ1 hvb ib  =q b .=hvb ib  =b .=q hvb ib þ nbs .=q v~ b dS. (18)
q q V Abs

It is observed that the second and third terms in the RHS of the above equation retain information (via the
gradient of the porosity) on the non-homogeneous nature of the porous medium within the averaging volume
V. For homogeneous porous medium, one has that the spatial variations of the porosity are negligible within
V, and the above mentioned terms can be ignored.

3.3. Closure problem

In order to develop the closed form of Eq. (18), we need to develop the governing differential equations and
boundary conditions for the spatial fluctuations of pressure and velocity p~ b and v~ b . This will lead us to a local
closure problem in terms of closure variables, which provides a method for establishing the structure of the
corresponding Darcy’s-type law. By substituting Eq. (18) in Eq. (17), one obtains
mq mq 1 q
 =hpb ib þ rb g þ rqþ1 hvb ib  1 . q b
b mq =b = hvb i   = b .=hvb ib
q q b
Z
1
þ nbs .ðIp~ b þ mq =q v~ b Þ dS ¼ 0. ð19Þ
V b Abs
On the other hand, a continuity equation for v~ b is obtained by subtracting Eq. (13) from =.vb ¼ 0 to obtain
b
=.v~ b ¼ 1 .
b =b hvb i ; in the b-phase. (20)
|ffl{zffl}
source

As in Whitaker [19], the terminology ‘‘source’’ is used here to denote driving terms for the boundary-value
problems. In the same way, by subtracting Eq. (19) from the fractional Stokes equation yields the spatial
velocity fluctuations equation
Z
1
~ qþ1
~
 = pb þ m q r vb  nbs .ðIp~ b þ mq =q v~ b Þ dS
V b Abs
 
1 m
1 q
¼ mq 1  rqþ1 hvb ib 1 . q
b mq =b = hvb i þb
b
=hvb ib .=q b , ð21Þ
q |fflfflfflfflfflffl{zfflfflfflfflfflffl} |fflfflfflffl{zfflfflfflffl} q |fflfflffl{zfflfflffl}
source source source

which holds for the b-phase. Finally, the non-slip boundary condition becomes
v~ b ¼  hvb ib ; at Abs . (22)
|ffl{zffl}
source
ARTICLE IN PRESS
J.A. Ochoa-Tapia et al. / Physica A 374 (2007) 1–14 9

Fig. 3. Periodic unit cell for a representative region of the porous medium.

Obviously, we do not want to solve Eqs. (20)–(22) in the macroscopic region. Instead, as recommended by
Whitaker [19], we wish to solve the closure problem in some representative region (see Fig. 3). To do this, we
treat the representative region as a unit cell in a spatially periodic model of a porous medium so that the
closure problem is given by Eqs. (20)–(22) together with the periodicity conditions:
p~ b ðr þ li Þ ¼ p~ b ðrÞ; i ¼ 1; 2; 3 (23)
and
v~ b ðr þ li Þ ¼ v~ b ðrÞ; i ¼ 1; 2; 3. (24)
Given the constant sources hvb ib , =hvb ib , =q hvb ib and rqþ1 hvb ib in the boundary value problem for p~ b and
v~ b , solutions of the form
v~ b ¼ Db .hvb ib þ Hb : =hvb ib þ Fb : =q hvb ib þ Cb .rqþ1 hvb ib (25)
and
p~ b ¼ mq db .hvb ib þ mq hb : =hvb ib þ mq f b : =q hvb ib þ mq cb .rqþ1 hvb ib (26)
are proposed. Substitution of the proposed solutions in Eqs. (20)–(24), give the following boundary value
problems for the closure variables:

 Problem I:
=.Db ¼ 1
b =b ; in the b-phase,
Z
1
=db þ rqþ1 Db  nbs .ðIdb þ =q Db Þ dS ¼ 0; in the b-phase,
Vb Abs

Db ¼ I; at Abs ,

Db ðr þ li Þ ¼ Db ðrÞ; db ðr þ li Þ ¼ db ðrÞ.

 Problem II:
=.Hb ¼ 0; in the b-phase,
ARTICLE IN PRESS
10 J.A. Ochoa-Tapia et al. / Physica A 374 (2007) 1–14
Z
1 1
b
=hb þ rqþ1 Hb  nbs .ðIhb þ =q Hb Þ dS ¼ =q b I; in the b-phase,
Vb Abs q

Hb ¼ 0; at Abs ,

Hb ðr þ li Þ ¼ Hb ðrÞ; hb ðr þ li Þ ¼ hb ðrÞ.

 Problem III:
=.Fb ¼ 0; in the b-phase,
Z
1  
qþ1
=f b þ r Fb  nbs . If b þ =q Fb dS ¼ 1
b =b I; in the b-phase,
V b Abs

Fb ¼ 0; at Abs ,

Fb ðr þ li Þ ¼ Fb ðrÞ; f b ðr þ li Þ ¼ f b ðrÞ.

 Problem IV:
=.Cb ¼ 0; in the b-phase,
Z  
qþ1 1 q 1
=cb þ r Cb  nbs .ðIcb þ = Cb Þ dS ¼  1  I; in the b-phase,
V b Abs q

Cb ¼ 0; at Abs ,

Cb ðr þ li Þ ¼ Cb ðrÞ; cb ðr þ li Þ ¼ cb ðrÞ.

In order to determine the necessity of solving the above boundary value problems, it is convenient to write
the closed form of the effective medium equation. This is done by substituting Eqs. (25) and (26) into Eq. (19):
q 1 . qþ1
=hpb ib þ rb g  b mq K1 . b 1 b
q;b hvb i þ mq K1q;b : = hvb i þ mq K2q;b r hvb ib þ mq K1 b
3q;b : =hvb i ¼ 0, (27)

where we have used the following definitions:


Z
1
b K1
q;b ¼ nbs .ðIdb þ =q Db Þ dS, (28)
V b Abs

Z
1
K1
1q;b ¼ ð1
b =b ÞI þ nbs .ðIhb þ =q Hb Þ dS, (29)
Vb Abs

Z
1
K1 1
2q;b ¼ q I þ nbs .ðIcb þ =q Cb Þ dS, (30)
Vb Abs

Z
1 1 q 1
K1
3q;b ¼ q b = b I þ nbs .ðIf b þ =q Fb Þ dS. (31)
Vb Abs

Eq. (27) can be re-arranged to give the following Darcy’s-type equation:


Kq;b
hvb i ¼  .ð=hpb ib  rb gÞ þ B1q;b : =q hvb ib þ B2q;b .rqþ1 hvb ib þ B3q;b : =hvb ib , (32)
mq
ARTICLE IN PRESS
J.A. Ochoa-Tapia et al. / Physica A 374 (2007) 1–14 11

where
B1q;b ¼ Kq;b .K1
1q;b , (33)
and
B2q;b ¼ Kq;b .K1
2q;b , (34)
and
B3q;b ¼ Kq;b .K1
3q;b . (35)
Eq. (32) describes the average (macroscopic) velocity in terms of observable variables rhpb ib , =q hvb ib ,
rqþ1 hvb ib , and =hvb ib . Some comments regarding the results obtained above, are in order:

(i) The term ðKq;b =mq Þ.ð=hpb ib  rb gÞ has the structure of a Darcy’s law. In contrast to the classical case,
here the permeability tensor Kq;b must be computed from a fractional boundary-value problem.
(ii) The term B1;q;b : =q hvb ib can be interpreted as a fractional convective correction induced by non-
homogeneities. In fact, if =q b ¼ 0 (i.e., homogeneous porous medium) Problem II has not a driving term
and the trivial solution (i.e., f b ¼ 0 and Fb ¼ 0) is the only one. As a consequence, K11;q;b  0 (see Eq. (29))
and B1;q;b ¼ 0.
(iii) The third term in Eq. (32) can be written as (see Eq. (30))
( " Z # )
1 q
B2;q;b .rqþ1 hvb ib ¼ q1 I.rqþ1 hvb ib þ nbs .ðIcb þ = Cb Þ dS .rqþ1 hvb ib .Kq;b .
V b Abs
While the first term in the RHS can be seen as a fractional Brinkman correction (i.e., fractional dispersive
correction), the second one arises from the fractional nature of the viscous effects. In fact, if q ¼ 1
ðð1  ð1=qÞÞI ¼ 0Þ, Problem III has only the trivial solution cb ¼ 0 and Cb ¼ 0, implying that K1 2;q;b ¼ I.
In such case, the last term in Eq. (31) reduces to the classical Brinkman correction Kq;b .r2 hvb ib .
(iv) The above three results demonstrate the consistency of the derived Darcy’s-type law since, after removing
fractional and non-homogeneous assumptions, it reduces to the classical Darcy’s law.

4. Conclusions

Using spatial averaging methods, in this paper we have derived an average equation for viscous flow in a
porous medium where a fractional Newton’s law of viscosity is satisfied. The rationale behind fractional
Newton’s law is that it can be used to describe correlation effects due to inhomogeneities in the porous
media structure. The result is a Darcy’s-type law where the permeability tensor should be computed from a
fractional boundary value problem. Contrary to classical Darcy’s law, here the Brinkman correction
corresponds to a fractional operator, and the corresponding permeability tensor is different from the Darcy’s
permeability tensor. This shows that fractional Newton’s law of viscosity yields a generalization of classical
Darcy’s law.

Appendix A. Fractional spatial averaging theorem

When applying spatial averaging methods, one requires to interchange differentiation and integration.
Given a closed and bounded manifold O with boundary qO, and some quantity F ðwÞ associated with O, the
spatial averaging theorem is given by [19]
Z
1
hrF ðwÞi ¼ rhF ðwÞi þ F ðwÞ.nqO dS,
V AqO
where nqO is the outer normal vector directed away from the boundary qO, and V is the volume of the
manifold O. Similarly, a fSAT is required to interchange averaging integration and fractional differentiation.
The following concepts and results were taken from Meerschaert et al. [20]. For a function F ðwÞ, its fractional
ARTICLE IN PRESS
12 J.A. Ochoa-Tapia et al. / Physica A 374 (2007) 1–14

divergence is given by =q F ðwÞ ¼ =J ð1qÞ ðF ðwÞÞ, where J ð1qÞ is the fractional integration operator, so that the
fractional divergence is just the classical divergence of the fractionally integrated function. Given a closed and
bounded manifold O with boundary qO, an application of the classical divergence theorem to the divergence
=J ð1qÞ F ðwÞ gives
Z Z
=q F ðwÞ dV ¼ J ð1qÞ ðF ðwÞÞnqO dS. (A.1)
O qO

Now, consider a two-phase heterogeneous domain V with a simply connected b-phase subdomain Vb . Let
Abs be the area of the impermeable bs-interface. Let Aio be the permeable boundary (i.e., the transport input/
output boundary). Hence, one has that: (i) Abs \ Aio ¼ ;, and (ii) Abs [ Aio ¼ qVb . For O ¼ Vb , these
properties allow to write Eq. (A.1) as follows:
Z Z Z
q
= F ðwÞ dV ¼ J ð1qÞ F ðwÞnio dS þ J ð1qÞ ðF ðwÞÞnbs dS, (A.2)
Vb Aio Abs

where nio and nbs are the unit normal vectors corresponding to the bs-interface and the transport input/output
boundary.R To obtain a fSAT useful for porous media averaging, let us derive an expression of the surface
integral A J ð1qÞ F ðwÞ.nio dS as a volume integral. To this end, consider the two-phase configuration shown
io
in Fig. 4. The idea is to characterize volume integral perturbations in a direction approaching the bs-interface.
Along the direction defined by the unit vector k, one has that
Z Z
dq
k.rq F ðwÞ dV ¼ q F ðwðsÞÞ dV , (A.3)
Vb ds Vb

where s is the parametrizing variable along k. For a given scalar gðsÞ, the following Grunwald formula for the
qth derivative [21] will be used in the sequel:
dq gðsÞ Xn
q
¼ lim h wj ðqÞgðs  jhÞ, (A.4)
dsq h!0
n!1 j¼0

where the weights wj ðqÞ are given by the binomial coefficients


Gðq þ 1Þ
wj ðqÞ ¼ ð1Þj .
Gðq  j þ 1Þj!
Using this definition in Eq. (A.3), one has that
Z Xn Z
dq q
F ðwðsÞÞ dV ¼ lim h wj ðqÞ F ðwðsÞÞ dV . (A.5)
dsq Vb h!0
n!1 j¼0 Vb ðsjhÞ

Fig. 4. Schematic diagram of volume perturbations.


ARTICLE IN PRESS
J.A. Ochoa-Tapia et al. / Physica A 374 (2007) 1–14 13

P
The property nj¼0 wj ðqÞ ¼ 0 implies that the intersection of the integrals will cancel in Eq. (A.5), so we obtain
Z Xn Z
dq q
F ðwðsÞÞ dV ¼ lim h w j ðqÞ F ðwðsÞÞ dV .
dsq Vb h!0
n!1 j¼0 DVb ðsjhÞ

Since the volumes DVb ðs  jhÞ, j ¼ 1; 2; . . ., tend to zero as h ! 0, we can use the simple geometrical
arguments presented in the derivation of the standard SAT in order to evaluate the integrals in the right hand
side of the latter equation. Each point on the surface of the domain V is translated a distance jh in the negative
of the k-direction in order to construct the volumes DVb ðs  jhÞ, j ¼ 1; 2; . . . [22]. The next step in this analysis
requires the representation of the volume integrals in terms of surface integrals. Following Howes and
Whitaker [23], for the jth volume DVb ðs  jhÞ, one has that the differential volume element can be represented
as dV ¼ jhk.nio dS þ Oðh2 Þ, where nio is the unit normal vector outward Vb ðs  jhÞ. The use of this relation
in Eq. (A.5) leads to the following expression:
Z Xn Z
dq q
F ðwðsÞÞ dV ¼ lim h ½jw j ðqÞ hk.nio F ðwðsÞÞ dS. (A.6)
dsq Vb h!0
n!1 j¼1 Aio ðsjhÞ

Note that summation carries out from j ¼ 1. Since h and k are independent of position, we can rewrite Eq.
(A.6) as follows:
Z " Z #
X
n
q qþ1
k.r F ðwðsÞÞ dV ¼ k. lim h ½jwj ðqÞ nio F ðwðsÞÞ dS , (A.7)
Vb h!0 Aio ðsjhÞ
n!1 j¼1

where we have used the relationship k.rq ¼ dq =dsq . For h sufficiently small, one has that
Aio ðs  jhÞ ’ Aio ; j ¼ 1; 2; . . . ,
which allows us to write the approximation
Z Z
nio F ðwðsÞÞ dS ’ nio F ðwðs  jhÞÞ dS.
Aio ðsjhÞ Aio

In this form, Eq. (7) can be expressed as


Z "Z " # #
X
n
q qþ1
k.r F ðwðsÞÞ dV ¼ k. lim h ½jwj ðqÞnio F ðwðs  jhÞÞ dS . (A.8)
Vb h!0
Aio n!1 j¼1

By introducing the shifting k ¼ j  1, and considering that Gðq þ 1Þ ¼ qGðqÞ, the following relationship can be
obtained from the expression for the binomial coefficients wj ðqÞ:
qGðqÞ
jwj ðqÞ ¼  ð1Þj
Gðq  j þ 1Þðj  1Þ!
qGðqÞ
¼ ð1Þj1
Gðq  j þ 1Þðj  1Þ!
GðqÞ
¼ qð1Þk
Gðq  kÞk!
¼ qwk ðq  1Þ; k ¼ 0; 1; 2; . . . .
This leads us to express Eq. (A.8) in the following form:
Z " Z " # #
X
n
k.rq F ðwðsÞÞ dV ¼ k. q lim h qþ1
wk ðq  1Þnio F ðwðs  khÞÞ dS . (A.9)
Vb h!0
Aio n!1 k¼0

For q 2 ð0; 1, 1  qo0. As a consequence, according to the definition (A.4), the term
X
n
lim hqþ1 wk ðq  1Þnio F ðwðs  khÞÞ,
h!0
n!1 k¼0
ARTICLE IN PRESS
14 J.A. Ochoa-Tapia et al. / Physica A 374 (2007) 1–14

corresponds to a fractional integration; i.e.,


Z " Z #
k.rq F ðwðsÞÞ dV ¼ k. q J ð1qÞ ðF ðwðsÞÞÞnio dS . (A.10)
Vb Aio
R R
Equivalently, =q V F ðwðsÞÞ dV ¼ q A J ð1qÞ ðF ðwÞÞnio dS. Consequently,
b io
Z Z Z
1 q
=F ðwÞ dV ¼ q = F ðwðsÞÞ dV þ J ð1qÞ ðF ðwÞÞ.nbs dS. (A.11)
Vb Vb Abs

Since
Z
1
hcðwÞi ¼ cðwÞ dV
V Vb

is the spatial averaging operator, equality (A.11) leads to the following fSAT:
Z
1
h=q F ðwÞi ¼ q1 =q hF ðwÞi þ J ð1qÞ ðF ðwÞÞnbs dS. (A.12)
V Abs
R
For q ¼ 1, the fSAT reduces to the classical SAT given by h=F ðwÞi ¼ =hF ðwÞi þ ð1=VÞ Abs F ðwÞnbs dS [23].

References

[1] M.K. Hubbert, The theory of groundwater motion, J. Geology 48 (1940) 785–793.
[2] S. Whitaker, Diffusion and dispersion in porous media, Ind. Eng. Chem. 61 (1969) 14–28.
[3] S. Whitaker, Flow in porous media I: a theoretical derivation of Darcy’s law, Transp. Porous Media 1 (1986) 3–25.
[4] M. Quintard, S. Whitaker, Transport in ordered and disordered porous media I: the cellular average and the use of weighting
functions, Transp. Porous Media 14 (1994) 163–177.
[5] M. Quintard, S. Whitaker, Transport in ordered and disordered porous media II: generalized volume averaging, Transp. Porous
Media 14 (1994) 179–206.
[6] R. Schumer, D.A. Benson, M.M. Meerschaert, S.W. Wheatcraft, Eulerian derivation of the fractional advection–dispersion equation,
J. Contaminant Hydrol. 48 (2001) 69–88.
[7] J.-H. He, Approximate analytical solution for seepage flow with fractional derivatives in porous media, Comput. Meth. Appl. Mech.
Eng. 167 (1998) 57–68.
[8] M. Caputo, Diffusion of fluids in porous media with memory, Geothermics 28 (1999) 113–130.
[9] M. Caputo, W. Plastino, Diffusion in porous layers with memory, Geophys. J. 10 (2002) 357–367.
[10] J.H. Cushman, M. Moroni, Statistical mechanics with three-dimensional particle tracking velocimetry experiments in the study of
anomalous dispersion I: theory, Phys. Fluids 13 (2001) 75–80.
[11] J.H. Cushman, M. Moroni, Statistical mechanics with three-dimensional particle tracking velocimetry experiments in the study of
anomalous dispersion II: experiments, Phys. Fluids 13 (2001) 81–91.
[12] G. van Tonder, K. Riemann, I. Dennis, Interpretation of single-well tracer tests using fractional-flow dimensions. Part 1: theory and
mathematical models, Hydrogeol. J. 10 (2002) 351–356.
[13] K. Riemann, G. van Tonder, P. Dzanga, Interpretation of single-well tracer tests using fractional-flow dimensions. Part 2: a case of
study, Hydrogeol. J. 10 (2002) 357–367.
[14] A. Cloot, J.F. Botha, A generalised groundwater flow equation using the concept of non-integer order derivatives, Water SA 32
(2006) 1–7.
[15] K.B. Oldham, J. Spanier, The Fractional Calculus, Academic Press, New York, 1974.
[16] K.S. Miller, B. Ross, An Introduction to the Fractional Calculus and Fractional Differential Equations, Wiley, New York, 1993.
[17] R. Gorenflo, F. Mainardi, Random walk models for space-fractional diffusion processes, Fractional Calculus Appl. Anal. 1 (1998)
167–191.
[18] T.J. Osler, Taylor’s series generalized for fractional derivatives and applications, SIAM J. Math. Anal. 2 (1971) 37–47.
[19] S. Whitaker, The Method of Volume Averaging, Kluwer Academic Publishers, Netherlands, 1999.
[20] M.M. Meerschaert, J. Mortensen, S.W. Wheatcraft, Fractional vector calculus for fractional advection–dispersion, Physica A 367
(2006) 181–190.
[21] S. Samko, A. Kilbas, O. Marichev, Fractional Integrals and Derivatives: Theory and Applications, Gordon and Breach, London,
1993.
[22] S. Whitaker, A simple geometrical derivation of the spatial averaging theorem, Chem. Eng. Education 19 (1985) 18.
[23] F.A. Howes, S. Whitaker, The spatial averaging theorem revisited, Chem. Eng. Sci. 40 (1985) 1387–1392.

You might also like