You are on page 1of 18

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING

Int. J. Numer. Meth. Engng. 46, 20432060 (1999)

MODELLING ION DIFFUSION MECHANISMS


IN POROUS MEDIA
E. SAMSON1; 2 , J. MARCHAND1; 2; , J.-L. ROBERT1 AND J.-P. BOURNAZEL3
1 Universit
e

2 SIMCO
3 Laboratoire

Laval; Quebec; Canada G1K 7P4


Technologies Inc.; 1400; boul. du Parc Technologique; Quebec; Canada G1P 4R7

dEtudes et de Recherches sur les Materiaux 23; rue de la Madeleine-BP 136 13631 Arles Cedex; France

SUMMARY
The main features of a numerical model aiming at predicting the drift of ions in electrolytic solutions are
presented. The mechanisms of ionic di usion are described by solving the NernstPlanck system of equations.
The electrical coupling between the various ionic uxes is accounted for by the Poisson equation. Two
algorithms using the nite element method for spatial discretization are compared for simple test cases. One
is based on the Picard iteration method while the other is based on the NewtonRaphson scheme. Test results
clearly indicate that the range of application is broader for the algorithm based on the NewtonRaphson
method. Selected examples of the application of the algorithm to more complex 1-D and 2-D cases are
given. Copyright ? 1999 John Wiley & Sons, Ltd.
KEY WORDS:

ionic di usion; NernstPlanck; Poisson; electrical coupling

1. INTRODUCTION
In many engineering problems, the behaviour of porous materials is directly a ected by the transport of ions under a concentration gradient. For instance, it has been shown that the swelling
of clays is predominantly controlled by the penetration of ions by di usion in their interlayer
spaces [1]. Given their in uence on various phenomena such as the ltration by ion exchange
membranes and the transport of pollutants in soils, the mechanisms of ionic di usion in porous
media has also received a great deal of attention from chemical and geological engineers [2; 3].
The process of ionic di usion remains of primary importance in many civil engineering problems
since the long-term durability of many building materials, such as concrete, is directly a ected by
the transport of chemical species [4].
Over the years, it has been established that the mechanisms of ionic di usion can be adequately
modelled by the NernstPlanck= Poisson set of equations [2; 4]. These equations take into account
the electrical coupling between the di erent ions present in an ideal solution (i.e. no chemical
activity e ects are considered). According to this model, the drift of an ionic species strongly
in uences that of all other ions dissolved in the electrolytic solution.

Correspondence to: J. Marchand, CRIB-Department of Civil Engineering, Laval University, Quebec, Canada G1K 7P4

CCC 0029-5981/99/36204318$17.50
Copyright ? 1999 John Wiley & Sons, Ltd.

Received 26 January 1999


Revised 6 April 1999

2044

E. SAMSON ET AL.

Although the electrical coupling between the various ionic uxes is well known to electrochemists and engineers, most existing models aiming at describing the mechanisms of ionic
di usion tend to neglect this phenomenon [3; 5]. Furthermore, a comprehensive bibliographical
review has recently shown that the proposed analytical or numerical ionic transport models are
unsatisfactory, all of them being limited to unidimensional and steady-state cases. A summary of
this literature survey is given in the following section.
In order to extend the application of the NernstPlanck= Poisson set of equations, two algorithms
were tested. The rst one is based on the Picard iteration method. The second algorithm uses a
NewtonRaphson scheme. Both rely on the nite element method for spatial discretization.
After this comparison, selected examples of calculations are presented to illustrate the application of the second algorithm to the treatment of steady-state and transient problems involving an
important number of multivalent ionic species. An example of application of the algorithm to the
resolution of 2-D cases (axisymmetrical geometries) is also given.

2. MATHEMATICAL MODEL
The NernstPlanck model, which describes the ux of an ionic species i in solution, is given by


zi F
ci grad (V )
(1)
ji = [Di ] grad (ci ) +
RT
where ji stands for the ux of the species i, [Di ] is the di usion coecient tensor of the species, ci
is the ionic concentration of the species, zi is the valence number of the species, F is the Faraday
constant, R is the perfect gas constant, T is the temperature, and V is the electrical potential that
is locally induced in the electrolytic solution by the movement of all ionic species.
It should be emphasized that the presence of this electrical potential is probably the most
important feature that distinguishes ionic di usion from molecular di usion. In an ionic solution,
the local electroneutrality shall be preserved at any point. The conservation of electroneutrality
requires that the transport of all di using species should be coupled. During the di usion process,
all ions are not drifting at the same speed. Some ions tend to di use at a higher rate. However,
any excess charge transferred by the faster ions builds up a local electric eld (Valso called the
di usion potential) which slows down the faster ions, and reciprocally accelerates the slower ionic
particles. The di usion potential has to be accounted for even in cases where an external electrical
eld is applied to the system. In that case, the di usion potential is superimposed to the external
eld.
It should also be underlined that equation (1) does not consider any chemical activity e ects
or the transport of ions by convection of the liquid phase in the pore system. The in uence of
activity phenomena on the mathematical treatment of the di usion problem is discussed elsewhere
[6]. A detailed discussion on the limits of equation (1) is given by Hel erich [2].
For each of the ionic species present in solution, the mass conservation law is given by
@ci
+ div( ji ) = 0
@t

(2)

This equation does not account for any chemical or physical interactions that can develop between
the solid and the various ionic species in solution. A comprehensive discussion of the in uence
Copyright ? 1999 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 46, 20432060 (1999)

MODELLING ION DIFFUSION MECHANISMS IN POROUS MEDIA

2045

of chemical reactions and physical interaction phenomena on the mathematical treatment of ionic
di usion problems is given in [7; 8].
By replacing equation (1) in equation (2), one nds the complete NernstPlanck equation:



zi F
@ci
div [Di ] grad (ci ) +
ci grad (V )
(3)
@t
RT
In order to complete this system of equations, one has to de ne one last relationship that will
couple the transport of all ionic species to one another. Over the past decades, numerous authors
have chosen to simplify the problem by assuming that the electroneutrality of the solution is
preserved at any points:
N
P
i=1

zi ci + w = 0

(4)

where N is the number of species and w is a xed charge density over the domain. For most
porous materials, the xed charge density is not a relevant parameter. It is, however, often used
in the description of biological systems like thin membranes.
In many cases, it is also assumed that the global ow of all ions across the membrane yields a
nil current:
N
P
i=1

zi ji = 0

(5)

However, as will be discussed in the following section, these assumptions are not always valid.
A more rigorous way to treat the problem [2] is to de ne the variation of the electric potential
according to the spatial distribution of the electric charges. This relationship is given by the Poisson
equation
2 V +


=0


(6)

where  is the electrical charge density and  is the dielectric constant of the surrounding medium.
The electrical charge density is a function of the concentration of the various ions in solution and
can be calculated using the following equation:
P
(7)
 = F( zi ci + w)

3. NUMERICAL MODELS FOR SOLVING THE NERNSTPLANCK= POISSON


SYSTEM OF EQUATIONS
3.1. Bibliographical review
Over the past decades, numerous authors have tried to develop solutions for the NernstPlanck
equation. The rst solution was obtained in 1890 by Planck himself [9]. He studied the case of
steady-state unidimensional ionic di usion of two monovalent species (+1 and 1) through a
membrane. In the problem considered by Planck, xed concentration and potential were imposed
Copyright ? 1999 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 46, 20432060 (1999)

2046

E. SAMSON ET AL.

on both sides of the domain. The solution was obtained using the electroneutrality and nil current
assumptions. The geometry and boundary conditions of this problem, although simple and very
idealized, still remain about the only ones for which an analytical solution is possible.
Later on, Schlogl [10], Hel erich [2] and Teorell [11] developed other solutions for the treatment
of unidimensional steady-state problem. Schlogl [10] derived an analytical solution using the same
boundary conditions as Planck. Schlogls solution was developed in such a way that it could
account for any number of ions, whatever their valence number. Once again, the electroneutrality
and nil current assumptions were at the basis of the development of his solution.
The application of all previous analytical solutions was limited to steady-state cases. Always
using the electroneutrality and nil current assumptions, Conti and Eisenman [12] proposed a different solution that could be applied to the treatment of unsteady-state problems. They developed
an expression for the variation of the electrical potential through a membrane but could not obtain
neither the concentration nor the electrical potential pro les. Furthermore, the validity of their
solution was solely restricted to electrolytic solutions made of monovalent ions.
It should also be emphasized that all the previous analytical solutions were developed on the
basis of the electroneutrality and nil current assumptions. In 1943, Goldman [13] presented a
solution for the steady-state case assuming a constant electric eld across the membrane (i.e. a
linear variation of the electrical potential across the system). This simpli cation of the problem
allowed the author to integrate directly the NernstPlanck equation.
The considerable diculty of developing analytical solutions for this system of equations has
led researchers to use numerical methods. In 1965, Cohen and Cooley [14] presented an algorithm
that allows solving the NernstPlanck equation for transient cases. Their solution was obtained
using a predictor-corrector scheme. But the predictor step uses Plancks analytical solution, thus
limiting the application of the algorithm to very simple cases.
More recently, Hwang and Hel erich [15] developed an algorithm that can be used to solve the
NernstPlanck equation for any number of ions, for any valence number, and for transient problems,
with a nite-di erence discretization. The discretization is not directly performed on equation (3).
The system of equations rst has to be transformed according to the electroneutrality and nil current
hypotheses. This transformation introduces two new terms that are used as iteration coecients
in the algorithm. However, after being transformed, the equations become very complex, which
makes the conversion in two or three dimensions extremely dicult. Moreover, the transformation
of the equations complicates the treatment of the boundary conditions. Patzay [16] modi ed the
approach by using three iteration coecients instead of two. In this case, no modi cation in the
treatment of the boundary conditions is required.
Harden and Viovy [17], who worked on membranes subjected to a current of variable intensity, have directly discretized the conservation equation (3) using the nite-di erence method. An
explicit Euler scheme is used for time discretization. For a case with N ionic species, the concentration pro les of N 1 ions are calculated using the concentration and potential calculated at
the preceding time step. For the remaining species, the concentration is calculated using the electroneutrality condition. The new electric eld is determined knowing that the current introduced
in the membrane must be equal to the internal current. The advantage of this method, compared
to all those previously described, is that it can be easily transposed in two or three dimensions.
This solution, however, still relies on the electroneutrality assumption.
All the previous numerical models are based on the assumption that the coupling of the Nernst
Planck equation with either the electroneutrality condition or the constant eld relationship yields
a reliable description of the ionic di usion mechanisms. Even if these hypotheses may constitute
Copyright ? 1999 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 46, 20432060 (1999)

MODELLING ION DIFFUSION MECHANISMS IN POROUS MEDIA

2047

a good choice in some practical cases, Hel erich [2] mentions that the Poisson equation should be
used in a more rigorous approach of the problem. MacGillivray and Hare [18; 19] have demonstrated that the electroneutrality and constant eld hypotheses are, in fact, nothing but particular
applications of the Poisson equation. The electroneutrality assumption is applicable only when the
concentrations are high while the constant eld hypothesis is rather valid for low concentrations.
Some researchers have tried to couple the NernstPlanck equation to the Poisson equation.
For instance, James et al. [20] coupled both equations to the Stokes equation to model the ow
of a liquid containing charged particles in a cylinder. Their analysis was limited to steady-state
cases. The conservation equations were discretized using the nite element method and a Galerkin
residual weighting. The algorithm consists in solving the equations one after the other starting
from an initial concentration pro le (Picard iterations), the values obtained being used as starting
points for the following iterations, until convergence is reached.
Kato [20] has proposed a numerical method to solve the NernstPlanck= Poisson system of
equations for unidimensional steady-state cases. Knowing that, once the steady state is reached,
the uxes are constant, a rst analytical integration of the NernstPlanck equation can be performed. The solution obtained has to be discretized afterwards by the nite-di erence method. The
numerical scheme used is similar to that of James et al. [20], the main di erence being that the
starting point is a potential pro le, which is subsequently used to calculate the concentrations.
As can be seen, the development of numerical solutions clearly appears to be the most promising
approach for the treatment of the NernstPlanck= Poisson system of equations. It should however
be emphasized that transient problems with any number of species for 1-D or 2-D cases with a
nite element discretization have never been investigated. As previously discussed, all attempts to
develop numerical solutions for this set of equations were limited to simpler cases. The object of
this paper is to compare the ability of two numerical algorithms to treat complex ionic di usion
problems using the NernstPlanck= Poisson equations.
3.2. First algorithm: uncoupled equations
The rst algorithm proposed is based on the Picard iteration technique, also known as the successive substitution technique [21]. It is a rst-order method, thus having a low rate of convergence.
The algorithm is similar to the numerical resolution proposed by James et al. [22] and Kato [20],
in which the equations are solved one after another.
The equations are discretized separately following the standard nite element method. The results
are thus presented without much detail. The weighted residual form of the NernstPlanck equation
(3) over the domain
is given by




Z
zi F
@ci
div [Di ] grad (ci ) +
ci grad (V )
d
= 0
(8)
W=
@t
RT

where is the weighting function. From that point on, an axisymmetrical case with an orthotropic
material is considered. Performing an integration by parts on W yields the following weak form:

 
 
Z
Z
@ci
zi F
c
V; r
r dr dz + h ; r ; z i [Di ] i; r +
ci
r dr dz
(9)
W=
c
V
@t
RT
i;
z
;z

The boundary terms are omitted, since all the simulations will be performed considering Dirichlet
conditions. Equation (9) is discretized according to the Galerkin method. The unknown ci is
Copyright ? 1999 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 46, 20432060 (1999)

2048

E. SAMSON ET AL.

interpolated at the nodes, as is

, according to
ci = hN i{cin }

(10)

= hN i{ n }

(11)

where hN i are the shape functions and the subscript n indicates node values. The elementary
matrices are written as
Z
[B]T [Di ] [B] + [B]T [Ei ] [N N ]r dr dz
(12)
[Kie ] =
Z
[Mie ] =

{N }hN ir dr dz

(13)

where [Ei ] is the matrix coupling the concentration of each


is written as

Dri zi F @V
0
RT @r
[Ei ] =
Dzi zi F @V
0
RT @z
The matrices [B] and [N N ] are de ned as

species to the electrical potential. It

@
@r hN i

[B] =

@
hN i
@z


hN i
[N N ] =
hN i

(14)

(15)

(16)

The various integrals are calculated using a Gaussian quadrature method. The terms @V=@r and
@V=@z in the matrix [Ei ] are evaluated at the integration points.
The Poisson equation is discretized using the same technique. The elementary matrices are
Z
[B]T [B]r dr dz
(17)
[K e ] =

e
P
Z
zi ci
r dr dz
(18)
{N }
{F e } =

e

Z
w
e
{Fs } =
(19)
{N } r dr dz

e

where {Fse } is the solicitation vector coming from the xed charge density w. The concentrations
in equation (18) are calculated at the integration points.
The resolution steps for a steady-state problem are presented. For a transient problem, a standard
 method [21] could be used.
1. An initial concentration is assumed for each ionic species.
2. The Poisson equation is solved using the concentration of the previous iteration level.
3. A loop is performed on all the ionic species:
(a) The gradient potential is calculated on the integration points from the numerical solution
of the Poisson equation. The results are used to build the matrix [Ei ].
Copyright ? 1999 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 46, 20432060 (1999)

MODELLING ION DIFFUSION MECHANISMS IN POROUS MEDIA

2049

(b) The solution of the NernstPlanck equation is obtained for the species i using the elementary matrices (12) and (13).
4. The L2 norm of the vector u = hc1 c2 : : : V ik hc1 c2 : : : V ik1 is calculated where k stands
for the iteration level.
5. If the norm is higher than a tolerance threshold , go back to step 2 by using the concentration
just calculated. The loops are performed until convergence is reached.
The main advantage of this algorithm is that the same calculation code can be used, whatever
the number of ionic species. Only the number of loops (step 3) is di erent.
3.3. Second algorithm: coupled equations
In the second algorithm, the NernstPlanck and Poisson equations are coupled and solved simultaneously. A classical NewtonRaphson method [21; 23] is used to solve the non-linear set of
equations. As it is a second-order scheme, convergence is expected to be faster than what is
obtained for Picard iteration method. Once again, the discretization follows the standard nite
element procedure.
The weighted residual form is written as [23]

Z
R1

h1 2 : : :i R2 d
= 0
(20)
W=

..

.
where the Ri s are the residuals associated to each of the equations and the i s are the corresponding weighting functions. For each of the ionic species, the residual is



@ci
zi F
Ri =
div [Di ] grad (ci ) +
ci grad (V )
(21)
@t
RT
and for the Poisson equation,
F P
( zi ci + w)
(22)

For a case limited to ionic species, the integration by parts leads to the following weak form:

Z
c1
W = hc1 c2 V i c2 r dr dz

Dr1 z1 F

c
c
+
V
D

r1 1; r
1 ;r

RT

D
z
F

z1
1
D c +

c1 V; z
z1 1; z

RT

Dr2 z2 F
c
D
c
+
V
r2
2;
r
2
;
r
r dr dz
+ hc1; r c1; z c2; r c2; z V; r V; z i
RT

D
z
F

z2 2

c2 V; z
Dz2 c2; z +

RT

;r

V; z
RV = 2 V +

Copyright ? 1999 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 46, 20432060 (1999)

2050

E. SAMSON ET AL.

Z
hc1 c2 V i

1
1


r dr dz
F

z 2 c2

0
r dr dz
(23)
+ hc1 c2 V i

F w

The weak form is discretized using a Galerkin weighting. The vector of the unknown variables
is written as

c1

c2

= [N ]{Un }
(24)
..

N2
NG
N1

N1
N2

NG

(25)
[N ] =

..
..
..

.
.
.
N2
NG
N1
+

hUn i = hc11 c21 : : : V1 c12 c22 : : : V2 c1G c2G : : : VG i

(26)

where G is the number of nodes in the element. The small dots (: : :) indicate the terms to add
when considering more ionic species, whereas the big dots ( ) stand for the missing shape
function terms. The subscripts i and j designate the species i and the node j. The elementary
matrices can thus be expressed as

T
[B] [D1 ][B] + [N ]T [D2 ] [N ] + [B]T [D3 ] [N N ]r dr dz
(27)
[K e ] =

{z
} |
{z
} |
{z
}
|

e


e
[KDe ]
[KPe ]
KNP
Z
[N ]T [D4 ][N ]r dr dz
(28)
[M e ] =

0
..
r dr dz
(29)
[N ]T
{F e } =
.

Fw


 
The matrix [K e ] is divided into three parts. The rst part, KDe , includes all the di usion terms. The
two other parts
 are related to the coupling between the concentration and the electrical potential.
The matrix KPe comes from the discretization of the Poisson equation (the third line in equation
e
(23)) and KNP
is related to the coupling between the concentration and the electrical potential
Copyright ? 1999 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 46, 20432060 (1999)

2051

MODELLING ION DIFFUSION MECHANISMS IN POROUS MEDIA

gradient in the NernstPlanck equation. The [N N ] matrix is given by

N2
NG
N1
N1
N
NG
2

N1
N2

NG

N
N

NG

1
2
[N N ] =

..
..

.
.

N2
N1
N1
N2
For an axisymmetrical case, the matrix [B] corresponds to

NG; x
N1; x
N1;y
NG;y

N1; x

NG; x

NG;y

1;y
[B] =

..

N1; x
N1;y
The matrices [D1 ][D4 ] are given by

Dr1

[D1 ] =

..

NG
NG

(30)

..

NG; x
NG;y

(31)

Dz1

Dr2

Dz2
..

(32)

[D2 ] =

Dr1 z1 F @V
RT @r

[D3 ] =

0
0
..
.
Fz1


Dz1 z1 F @V
RT @z

Fz2


(33)

Dr2 z2 F @V
RT @r

Dz2 z2 F @V
RT @z

..

.
0

(34)

0
Copyright ? 1999 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 46, 20432060 (1999)

2052

E. SAMSON ET AL.

[D4 ] =

1
..

(35)

0
The tangent matrix is obtained by calculating the rst variation of the weak form W (Equation
(20)) [23]:

R1
(36)
W = h1 2 : : :i R2 d
= 0

..
.
For a transient case solved with an implicit Euler scheme ( = 1), the elementary tangent matrix
is given by
Z
e
e
e
[BT ] [D5 ] [B] r dr dz
(37)
[KT ] = [M ] + t[K ] + t

where

[D5 ] =

:::

Dr1 z1 F
c1
RT
0
Dr2 z2 F
c2
RT
0
..
.
0
0

Dz1 z1 F

c1

RT

Dz2 z2 F
c2

RT

..

0
0

(38)

In contrast to the preceding numerical scheme, the characteristics of the various matrices vary
with the number of ionic species accounted for. Furthermore, larger matrices have to be stored.

4. COMPARISON OF THE TWO ALGORITHMS


In order to compare both algorithms, an example rst presented by Kato [20] will be used. It
is an unidimensional steady-state problem involving two ionic species. The development of the
elementary matrices for 1-D cases is not presented but is straightforward starting from the matrices
of the axisymmetrical case. The equations are rst rewritten in a non-dimensional form by stating
RT
V
F
x = Lx

V=

Copyright ? 1999 John Wiley & Sons, Ltd.

(39)
(40)
Int. J. Numer. Meth. Engng. 46, 20432060 (1999)

MODELLING ION DIFFUSION MECHANISMS IN POROUS MEDIA

Di = D i D0

2053
(41)

c0 D0
j
L i
ci = c0 ci
ji =

(42)
(43)

where the subscript 0 designates a reference value and L is the length of the domain. Equations (1) and (6) are rewritten as
 
 
Di d ci
Di d V

zi ci
(44)
ji =
D0 d x
D0 d x

d 2 V
+ 2 =0
2

d x

(45)

with
RT
c0 F 2 L2
P
 = zi ci + (w=c0 )

2 =

(46)
(47)

in the domain 06x61.

The value of the parameter 2 is determined mainly by the scale of


the problem under consideration. For a medium kept at 25 C (298 K) and saturated with an
ionic solution having a concentration of 1 mmol=l and a permittivity of 70832 1010 C=V m,
2 ' 2 1016 =L2 . For biological membranes that have thickness around 1 m, 2 ' 00002. For
porous construction materials like concrete, the characteristic length is of the order of about 1 cm
and thus 2 ' 2 1012 . For two problems having the same characteristic length L, a variation
in 2 corresponds to a variation in the level of concentration involved.
For this speci c example, it is assumed that two monovalent species are present in solution
(z1 = +1 and z2 = 1). For both species, D i = 1 and c0 = w = 1. It is also assumed that both
species have a dimensionless concentration of 10 at x = 0 and 15 at x = 1. The dimensionless
potential is set at 00 at x = 0 and 10 at x = 1.
To solve this problem, both algorithms were used with a linear two-node element.
4.1. Results obtained with the rst algorithm
To begin the calculations, initial concentrations were considered to vary linearly across the
system. Figure 1 shows the solution obtained with 2 = 05 and computed with 20 elements. Twelve
iterations were needed to perform the calculations. The di erence with the solution obtained by
Kato [20] does not exceed 2 per cent.
The next set of simulations consists in varying the value of 2 while keeping the same boundary
conditions. It appeared that under 2 = 025, no convergence could be obtained. Figure 2 clearly
shows the variation of the convergence rate as a function of 2 . All simulations were made with 20
elements. In order to rectify this situation, di erent solutions were tested: a variation in the numbers
of elements, an increase in the order of the interpolation polynomials, and a modi cation in the
initial solution introduced to the algorithm. Results revealed that all this work was done in vain.
A modi cation of the algorithm was also made, by updating the solution after a calculation step
Copyright ? 1999 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 46, 20432060 (1999)

2054

E. SAMSON ET AL.

Figure 1. First algorithm: solution for 2 = 05

Figure 2. First algorithm: norm vs. number of iterations for di erent values of 2

according to the following procedure:


{u}k1 =

{u}k1 + {u}k2
2

(48)

This modi cation allowed us to break the 025 limit, but only to nd the same problem for a
value of about 015. For such values of 2 , only thin membrane problems can be solved. This is
what justi ed the development of the coupled equation algorithm, in an attempt to broaden the
eld of application of the NernstPlanck= Poisson set of equations.
4.2. Results obtained with the second algorithm
The same tests were carried out for the second algorithm, and proved to provide much better
results. For the second algorithm, there is no evidence of a critical value of 2 for which divergence
occurs. The number of required iterations is lower and limited to approximately four. Furthermore,
this number varies very little according to 2 .
Copyright ? 1999 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 46, 20432060 (1999)

MODELLING ION DIFFUSION MECHANISMS IN POROUS MEDIA

2055

Figure 3. Second algorithm: solution for 2 = 00001

To illustrate the superior behaviour of the second algorithm, the same problem was solved,
but this time with 2 = 00001. The results are shown in Figure 3, and were obtained with 100
elements. More elements are needed to avoid oscillations near the boundaries, due to high gradient
values. Comparisons with the results of Kato [20] are impossible since the author did not perform
any tests for low values of 2 .
Following these results, it is clear that the second algorithm is much more robust. It was used
to obtain the results that will be presented afterwards.
Various other cases were considered to test the second algorithm. The rst example was inspired
from Hel erich [2]. It consists in a 1-D steady-state problem where four species are present:
+
+
Mg2+ , SO2
4 , Na and K . Dimensionless variables were used in the treatment of the problem.
2+
The di usion coecients were 10 for SO2
and Na+ , and 167 for K+ . At x = 0, the
4 , Mg
2
2+
boundary conditions were 10 for SO4 , 05 for Mg and zero for the other species. At x = 1,
2+
+
+
the conditions were 08345 for SO2
4 , 00 for Mg , 0169 for Na and 05 for K . The potential
was established at 00 at x = 0 and 50 at x = 1. The xed charge density in the membrane, w=c0 ,
was equal to 10. The only remaining unknown variable was the value of 2 .
The solution obtained analytically by Hel erich [2] for that same problem was developed on
the basis of the electroneutrality hypothesis. Considering that MacGillivray and Hare [18; 19]
have demonstrated that the validity of this assumption is restricted to very low values of 2 , the
calculations were performed with 2 = 1 1010 .
The solution, computed with 50 elements, is shown in Figure 4. The di erence with the analytical
solution presented by Hel erich [2] is about 3 per cent, which con rms the results of MacGillivray
and Hare [18; 19] concerning the value of 2 .
The other example concerns a transient di usion problem. It was selected to evaluate the in uence of the 2 parameter for a xed length domain. The calculations were performed using
an implicit Euler scheme. The dimensionless variables were once again used with the addition
of t= D0 t=L2 as a time variable. Two species were considered, with z1 = + 1 and D 1 = 3 for the
rst one, and z2 = 1, D 2 = 1 for the second. Initially, the concentrations were set equal to zero
over the entire domain, which consisted of a bar of length 1. At t= 0, a unit concentration was
imposed at x = 0 for both species. The concentrations were set at zero at x = 1.
Copyright ? 1999 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 46, 20432060 (1999)

2056

E. SAMSON ET AL.

Figure 4. Concentration pro les for the Hel erich problem

Figure 5. Concentration pro les at t= 001; 2 = 001

Two simulations were performed for values of 2 equal to 1 102 and 1 104 respectively.
This implies that, for a given domain length, the concentrations involved in the second case were
a hundred times higher than in the rst one. Results are given in Figures 5 and 6. These results
were obtained at t= 001 and reached after 200 steps of 5 105 . All calculations were performed
with 50 elements. As a comparative basis, the solutions computed without taking into account any
electrostatic e ect, thus derived on the sole basis of Ficks law, are also given in Figures 5 and 6.
In Figure 5, the in uence of the electrical coupling between the species is clearly noticeable.
On the one hand, results indicate that the progression of the fastest ion is slowed down by the
electrical potential. On the other, the slowest ion, with a di usion coecient being three times
lower, is accelerated. In the case of higher concentration (Figure 6), this phenomenon is even
more obvious. Both concentration pro les are so close that they became superposed.
These results con rm, once more, the conclusion of MacGillivray and Hare [18; 19], who suggested that the electroneutrality is gradually approached as the value of 2 decreases. Figure 7
shows that the electrical charge density  tends toward zero as 2 is reduced.
Copyright ? 1999 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 46, 20432060 (1999)

MODELLING ION DIFFUSION MECHANISMS IN POROUS MEDIA

2057

Figure 6. Concentration pro les at t = 001; 2 = 00001

Figure 7. Charge density for the two transient cases

The last example shows the use of this numerical scheme for a transient problem with an
axisymmetrical geometry. The simulations were performed to investigate the e ect of the presence
of electrical charges on the inner surface of a pore. In this case, ions are di using in and out of
the pore space. The pore, presented in Figure 8, has a radius of 1 m and a length of 10 m.

Five ions were considered in the simulations: OH , Na+ , K+ , SO2


4 , and Cl . Their respective
2
9
9
9
di usion coecients (in m =s) are: 5273 10 , 1334 10 , 1957 10 , 1065 109 , and
2032 109 .
The initial concentrations in the pore were xed at: 690 mmol= l of OH , 286 mmol= l of Na+ ,

500 mmol= l of K+ , 48 mmol= l of SO2


4 , and 0 mmol= l of Cl .
The boundary conditions, at the entrance of the pore, were xed at 500 mmol= l of Cl ,
800 mmol= l of Na+ , and 300 mmol= l of OH . For the electrical (di usion) potential, the initial
value was set at 0.
Two cases were investigated. In the rst case, it was assumed that no electrical charges were
present on the inner surface of the pore. In this case, the electrical potential was set at 0 at the
entrance of the pore, since a reference value has to be xed at some point. In the second case,
Copyright ? 1999 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 46, 20432060 (1999)

2058

E. SAMSON ET AL.

Figure 8. The axisymmetrical problem

Figure 9. Potential pro le in a pore

the electrical potential on the inner surface of the pore was set equal to 10 mV as shown in
Figure 8. The system was considered to remain in isothermal conditions (i.e. at 22 C ) over
the entire duration of the process. The dielectric constant of the system was xed at 70823
1010 C=V m. The two simulations were performed in ve time steps of 00002 s. For the spatial discretization, a regular mesh of 10 50 linear three-node triangles was used, as shown in
Figure 8.
According to the numerical simulations, the potential on the inner surface of the pore has no
signi cant e ect on the concentration pro les. Its in uence is limited to a small region at the
vicinity of the surface. Hence, the ux of ions in the pore is the same for both simulations.
For the electrical potential, the situation is however di erent. The potential on the inner surface
tends to lower the di erence of potential between both ends of the pore. For the case without any
electrical charges on the pore wall, this di erence has a value of 5 mV. In the other case, this
value drops to 31 mV.
Copyright ? 1999 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 46, 20432060 (1999)

MODELLING ION DIFFUSION MECHANISMS IN POROUS MEDIA

2059

5. CONCLUSION
The comparison between a numerical scheme based on the Picard iteration technique and another
scheme based on the NewtonRaphson method has clearly showed that the former, even though it
has already been used in some recent papers, cannot be used to solve the NernstPlanck= Poisson
system of equations for all cases.
By coupling all the equations and using the NewtonRaphson method, the NernstPlanck= Poisson
system of equations could be solved for problems with a high number of ionic species with di erent valence number, either for steady state or transient cases. Furthermore, the numerical scheme
works for 1-D geometry as well as 2-D and axisymmetrical ones, and could easily be extended
for 3-D cases.
This work opens the way for the treatment of complex multiionic species models of transport
in porous media involving chemical reactions between the species and the solid matrix, while
considering the electrical coupling.

REFERENCES
1. Dormieux L, Barboux P, Coussy O, Dangla P. A macroscopic model of the swelling phenomenon of a saturated clay.
European Journal of Mechanics A=Solids 1995; 14:9811004.
2. Hel erich F. Ion Exchange. McGraw-Hill: USA, 1962.
3. Zyvoloski GA, Robinson BA, Dash ZV, Trease LL. Models and methods summary for the FEHMN application,
FEHMN MMS, ECD-22, LA-UR-94-3787, Rev. 1, 1995.
4. Marchand J, Gerard B, Delagrave A. Ion transport mechanisms in cement-based materials. In Materials Science of
Concrete, vol. V, Skalny JP (ed.). American Ceramic Society, Ohio, 1998; 307400.
5. Bear J, Bachmat Y. Introduction to Modeling of Transport Phenomena in Porous Media. Kluwer Academic Publishers:
Netherlands, 1991.
6. Samson E, Marchand J. Numerical solution of the extended NernstPlanck model. Journal of Colloid and Interface
Science 1999; 215:18.
7. Reeves H, Kirkner DJ. Multicomponent mass transport with homogeneous and heterogeneous chemical reactions:
e ect of the chemistry on the choice of numerical algorithm1. Theory. Water Resources Research 1988; 24(10):
1719 1729.
8. Reeves H, Kirkner DJ. Multicomponent mass transport with homogeneous and heterogeneous chemical reactions:
e ect of the chemistry on the choice of numerical algorithm2. Numerical results. Water Resources Research 1988;
24(10):1730 1739.
9. Planck M. Ueber die potantialdi erenz zwischen zwei verdunnten losungen binarer electrolyte. Annals of Physics and
Chemistry 1890; 40:561570 (in German).
10. Schlogl R. Elektrodi usion in freier Losung und geladenen Membranen. Zeitschrift fur Physikalische Chemie, vol. 1.
1954; 305339 (in German).
11. Teorell T. Transport process and electrical phenomena in ionic membranes. Progress in Biophysics and Biophysical
Chemistry 1953; 3:305369.
12. Conti F, Eisenman G. The non-steady state membrane potential of ion exchangers with xed sites. Biophysical Journal
1965; 5:247 256.
13. Goldman DE. Journal of General Physiology 1943; 27:37 46.
14. Cohen H, Cooley JW. The numerical solution of the time-dependant NernstPlanck equations. Biophysical Journal
1965; 5:145162.
15. Hwang Y, Hel erich F. Generalized model for multispecies ion-exchange kinetics including fast reversible reactions.
Reactive Polymers 1987; 5:237 253.
16. Patzay G. A simpli ed numerical solution method for the NernstPlanck multicomponent ion exchange kinetics model.
Reactive and Functional Polymers 1995; 27:83 89.
17. Harden JL, Viovy JL. Numerical studies of pulsed iontophoresis through model membranes. Journal of Controlled
Released 1996; 38:129139.
18. MacGillivray AD. NernstPlanck equations and the electroneutrality and Donnan equilibrium assumptions. Journal of
Chemical Physics 1968; 48:2903 2907.
19. MacGillivray AD, Hare D. Applicability of Goldmans constant eld assumption to biological systems. Journal of
Thoretical Biology 1969; 25:113126.
Copyright ? 1999 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 46, 20432060 (1999)

2060

E. SAMSON ET AL.

20. Kato M. Numerical analysis of the NernstPlanckPoisson system. Journal of Thoretical Biology 1995; 177:
299304.
21. Reddy JN, Gartling DK. The Finite Element Method in Heat Transfer and Fluid Dynamics. CRC Press: USA, 1994.
22. James AE, Stillman JD, Williams DJA. Finite element solution of the equations governing the ow of electrolyte in
charged microporous membranes. International Journal for Numerical Methods in Fluids 1995; 20:1162 1178.
23. Zienkiewicz OC, Taylor RL. The Finite Element Method (4th edn). McGraw-Hill: USA, 1989.

Copyright ? 1999 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 46, 20432060 (1999)