Most models of compacted bentonite and claystone¹ assume that the material contains different domains with distinctly different properties. In particular, it is common to assume that the system contains both an aqueous domain (a bulk water solution) and one or several domains associated with the solid surfaces.² Models of this type is here called multiporosity models.

A “standard” procedure in multiporosity models is to assume that the diffusive flux in any domain can be fully expressed using quantities that relates only to that same domain, while a “total” flux is given by summing the contributions from each domain. In e.g. the case of having one bulk water domain and one surface domain, the diffusive flux is typically written something like

$$ \label{eq1} j = j_{bulk} + j_{surface} = -D_e\cdot\nabla c -D_s \cdot\nabla c_s \tag{1} $$

were \(D_e\) and \(c\) denote, respectively, the diffusion coefficient and concentration in the bulk water domain, and the corresponding parameters in the surface domain are denoted \(D_s\) and \(c_s\). It is important to keep in mind that what is referred to as a “total” flux in these descriptions is the experimentally accessible, macroscopic flux (here simply labelled \(j\)).

This approach is basically adopted by all proponents of multiporosity models, and has been so for a long time, see e.g.:

• van Schaik et al. 1966
• Rasmusen and Neretnieks 1983
• Jahnke and Radke 1985
• Muurinen et al. 1985
• Cheung and Gray 1988
• Oscarson 1994
• Eriksen et al. 1999
• Appelo and Wersin 2007
• Gimmi and Kosakowski 2011
• Shackelford and Moore 2013
• Bourg and Tournassat 2015
• Tournassat and Steefel 2015
• Glaus et al. 2015
• Tinnacher et al. 2016
• Alt-Epping et al. 2018
• Soler et al. 2019

There is, however, a big problem with this procedure:

Equation 1 does not make sense

The statement of eq. 1 — that the macroscopic, measurable flux has two independent contributions — is only true if the different domains are completely isolated from each other. This can e.g. be seen by making \(D_s\) vanishingly small in eq. 1; the flux is still non-zero due to the contribution from the completely unaffected \(j_{bulk}\)-term. In contrast, for non-isolated domains the expected behavior is that a macroscopic diffusive flux will vanish if the diffusivity of any domain is made to approach zero.

Consequently, a large part of the compacted bentonite research field seems to treat diffusion by assuming the different domains to be isolated “pipes”. But this is of course not how bentonite works! If you postulate multiple diffusive domains in the clay, you must certainly allow for diffusing species to be able to be transferred between them. And here is where it gets mysterious: the default choice is to also assume that the domains are non-isolated. In fact, the domains are typically assumed to be in equilibrium. In e.g. the case of having one bulk water and one surface domain, the equilibrium is usually expressed something like

$$
c_s = K\cdot c
\tag{2}
$$

where \(K\) is some sort of partition coefficient.

Thus, the “standard” treatment is to combine eq. 1 — even though it is only valid for isolated domains — with an equilibrium assumption (eq. 2) to arrive at expressions which typically look something like this

$$
j = \left (D_e + K\cdot D_s\right)\cdot \nabla c
\tag{3}
$$

Note how this equation has the appearance of an addition of diffusion coefficients, as the “total” flux now is related to a single concentration gradient (which in practice always is a bulk solution concentration).

Since eq. 3 results from combining two incompatible assumptions (isolation and equilibrium), it should come as no surprise that it is not a valid description for multiporosity models.

I cannot get my head around why this erroneous description has prevailed within the compacted clay research field for so long. The problem of deriving “effective” properties of heterogeneous systems appears in many research fields, and applies to many other quantities than diffusivity, e.g. heat conductivity, electrical conductance, elastic moduli, etc. It should therefore be clear also in the compacted clay world that obtaining the effective diffusivity for a system as a whole is considerably more involved than simply adding diffusivities from each contributing domain.

Some illustrations

To discuss the issue further, let’s make some illustrations. A molecule or ion in a porous system conducts random “jiggling” motion, exploring all regions to which it has access, as illustrated in this figure

This animation shows a simulated random walk at the microscopic scale, in a “made-up” generic porous system, consisting of solids (brown), a bulk water domain (blue), and a surface domain (pink). The green dots show the trace of the random walker; light green when the walker is located in the bulk water domain (with higher mobility), and darker green when it is located in the surface domain (with lower mobility). Note how the walker frequently switches between the domains — which is exactly what is required for the two domains to be in equilibrium.

As diffusivity is fundamentally related to the statistical properties of the random walk,³ it should be crystal clear that the macroscopic diffusion coefficient (reflecting mobility on length scales beyond what is shown in the animation) depends in a complicated way on the mobility of the involved domains, just as it depends on the detailed geometrical configuration. Oddly, the complex geometrical dependency is usually acknowledged within the bentonite research field (by means of e.g. formation and/or tortuosity factors), which makes it even more incomprehensible that the issue of multiple domain diffusivities is treated as it is.

In contrast, the figure below illustrates the microscopic behavior implied by eq. 1.

Here is shown one random walker in the bulk domain (green trace) and one in the surface domain (red trace). Since no exchange occurs between the domains, they are not in equilibrium (the concept does not apply). The lack of exchange between the domains is of course completely non-physical, but it is only under this assumption that the system is described by two independent fluxes (eq. 1). It may also be noted that the two types of walkers experience completely different geometrical configurations (i.e. formation factors), none of which corresponds to the geometrical configuration in the case of non-isolated domains.

A more reasonable formulation

If you insist on having a multiporosity model, the correct way to go about is to assume equilibrium within a so called representative elementary volume (REV), and to formulate the model on the macroscopic — averaged — level. On this level of description, it is legitimate to speak of both a bulk water and a surface concentration at any given point in the model (they represent REV averages), but there is of course only a single flux, as well as a single value of the chemical potential at any given point. The diffusive flux can be written quite generally as

\begin{equation} j = -\phi
D_\mathrm{macr.}\cdot\frac{\bar{c}}{RT}\nabla\mu
\end{equation}

where \(\bar{c}\) is the macroscopic, averaged, concentration, \(\phi\) is the porosity, and \(\mu\) is the (electro-) chemical potential. \(D_\mathrm{macr.}\) is the macroscopic diffusion coefficient, containing statistical information on how the diffusing substance traverses the porous system on length scales larger than the REV (in particular, it contains averaged information on the mobility in all involved domains, as well as their detailed geometrical configuration).

The value of \(\mu\) can be obtained from considering the microscopic scale. Since the system by assumption is in equilibrium within a REV, \(\mu\) has the same value everywhere within this volume. On the other hand, the (electro-) chemical potential may also be expressed e.g. as

\begin{equation}
\mu(\xi ) = \mu_0 + RT\cdot\ln a(\xi ) + zF\psi(\xi ) = \mathrm{const}
\end{equation}

where \(\xi\) is a microscopic variable denoting the position within the REV, \(\mu_0\) is a reference potential, \(a(\xi )\) is the chemical activity and \(\psi(\xi )\) the electrostatic potential. In the bulk water domain, the electrostatic potential is zero by definition (ignoring possible couplings between different types of charged species), and \(\mu\) can here be evaluated as (where the index “b” is used to denote bulk)

\begin{equation}
\mu_b = \mu_0 + RT\cdot\ln a_b
\tag{4}
\end{equation}

But this is by definition the value of the (electro-)chemical potential everywhere within the REV.

Thus, going back to the macroscopic scale and using eq. 4 for \(\mu\), we have for the flux

\begin{equation}
j = -\phi D_\mathrm{macr.}\cdot\frac{\bar{c}}{a_b}\nabla a_b
\end{equation}

Ignoring also possible gradients in activity coefficients, the flux can be simplified as

\begin{equation}
j = -\phi D_\mathrm{macr.}\cdot\frac{\bar{c}}{c}\nabla c
\end{equation}

Where, as before, \(c\) denotes a bulk solution concentration. For the case of having one bulk domain, and one surface domain, \(\bar{c}\) can be written

\begin{equation}
\bar{c} = \frac{\phi_b}{\phi}c + \frac{\phi_s}{\phi}c_s
\end{equation}

where \(\phi_b\) and \(\phi_s\) are the porosities for the bulk and surface domains, respectively (\(\phi = \phi_b + \phi_s\)). Using also eq. 2, the flux can finally be written

\begin{equation}
j = -D_\mathrm{macr.}\left (\phi_b+\phi_s\cdot K\right )\nabla c
\tag{5}
\end{equation}

This expression can be compared to the result from the “standard” procedure, eq. 3. Although eq. 3 is derived using incompatible, non-physical assumptions, it is similar to eq. 5 , but involves two (\(D_e\) and \(D_s\)) rather than a single diffusion parameter (\(D_\mathrm{macr.}\)). If these two parameters are used merely as fitting parameters (which they basically always are), eq. 3 can always be made to resemble eq. 5. In this sense, eq. 3 is not wrong. Rather, it is not even wrong.

---

Footnotes

[1] In the following I will only say bentonite, but I mean any type of clay system with significant swelling/ion exchange properties.

[2] I strongly oppose the way bulk water phases are included in such models, but this post is not a critique of that model choice. Here, I instead point out that virtually all published descriptions of multiporosity models handle the flux incorrectly.

[3] Specifically, the average of the square of the displacement of the walker is proportional to time, with the diffusion coefficient as proportionality constant (apart from a dimensional factor).