3.1 Bi-Velocity in One, Two, and Three Dimensions
In References
24 through
26, we have presented a basic formulation of the interdiffusion models in one, two, and three dimensions that is a generalization of the Darken bi-velocity method. Let an open and bounded set
Ω ⊂
Rn,
n = 1, 2, 3 with the smooth boundary ∂
Ω be given. Moreover, let
tp means the processing time. In two and three dimensions, a key assumption of the model is that there exists a scalar potential function,
Φ =
Φ(
t,
x), such that the vector drift velocity is expressed by the minus of its gradient:
$$ \upsilon^{D} = - {\text{grad}}\varPhi \;{\text{on }}\left[ {0,t_{p} } \right] \times \varOmega .$$
(4)
The existence of such potential is appropriate, even above the Tammann temperature (two-thirds melting temperature), because the viscosity of solids is very high and
\( {\text{rot}}{\kern 1pt} \,\upsilon^{D} = 0 \). In Reference
25, the model has been limited to the interdiffusion of the species having concentration-independent diffusivities. In Reference
26 the next step has been made and the system of the species having concentration-dependent diffusivities has been studied. The numerical results have been compared with the experiment. Here, we present a compendium about 1-D and multidimensional multicomponent interdiffusion based on our previous articles. The system under consideration has
s ≥ 2 components and it is represented by molar concentrations,
\( c_{1} ,\ldots,c_{s} {\kern 1pt} ,\;\,\,c_{i} = c_{i} \left( {t,x} \right) \) and molar ratios
\( N_{1} ,\ldots,N_{s} {\kern 1pt} ,\;\,\,N_{i} = N_{i} \left( {t,x} \right) = {{c_{i} \left( {t,x} \right)} \mathord{\left/ {\vphantom {{c_{i} \left( {t,x} \right)} {c\left( {t,x} \right)}}} \right. \kern-0pt} {c\left( {t,x} \right)}} \) of the components, where
c is the overall concentration
c = ∑
i=1
s
ci. We assume incompressible transport. In such case, the Euler theorem allows expressing the molar volume of a mixture of the components,
Ωm, having partial molar volumes,
Ωi =
const,
i = 1, ...,
s, as:
$$ \varOmega_{m} = \sum\nolimits_{i = 1}^{s} {\,\varOmega_{i} {\kern 1pt} N_{i} } , $$
(5)
which is accepted as Vegard’s rule and equivalently reads:
$$ \sum\nolimits_{i = 1}^{s} {\varOmega_{i} {\kern 1pt} c_{i} \,} = 1 .$$
(6)
The flux of the
ith species has two components, diffusion and drift, and we assume:
$$ J_{i} = - D_{i} {\kern 1pt} {\text{grad}}{\kern 1pt} c_{i} + c_{i} {\kern 1pt} \upsilon^{D}, $$
(7)
where the diffusivity is composition dependent,
Di =
Di(
c1, ...,
cs). Note that Eq. [
7] is a generalization of the Fick flux formula. For simplicity, we will apply in the following the diffusivities represented in terms of mole fractions,
Di =
Di(
Ni). The flux
Ji enters the continuity equation (Eq. [
3]). Multiplying Eq. [
3] by
Ωi, adding by its sides, and using the Vegard rule (Eq. [
6]), we obtain the volume continuity equation:
$$ {\text{div}}\left( {\sum\nolimits_{i = 1}^{s} {\varOmega_{i} \,J_{i} } } \right) = 0. $$
(8)
Define the initial condition on the concentrations:
$$ c_{i} (0,x) = c_{i,0} (x)\,\,\,{\text{on}}\,\,\,\,\varOmega ,\,\,\,i = 1,\ldots,s ,$$
(9)
and the boundary conditions for the fluxes:
$$ J_{i} \circ {\mathbf{n}} = j_{i} (t,x)\,\,{\text{on}}\,\,\,[0,t_{p} ]\, \times \partial \varOmega ,\,\,\,i = 1,\ldots,s, $$
(10)
where
n is the outside normal to the boundary ∂
Ω. Assume the Vegard rule on the initial concentrations:
$$ \sum\nolimits_{i = 1}^{s} {\varOmega_{i} \,c_{0,i} (x)} = 1\,\,{\text{on}}\,\,\varOmega. $$
(11)
We first formulate the 1-D model.[
27‐
29] Let
Ω = ( −
Λ,
Λ) ⊂
R. It follows from Eq. [
8] that:
$$ \sum\nolimits_{i = 1}^{s} {\varOmega_{i} \,J_{i} } = K(t), $$
(12)
where
K is an arbitrary function. By Eqs. [
10] and [
12], we get the unique:
$$ K(t) = \sum\nolimits_{i = 1}^{s} {\varOmega_{i} \,j_{i} (t,\varLambda )} = - \sum\nolimits_{i = 1}^{s} {\varOmega_{i} \,j_{i} (t, - \varLambda )} .$$
(13)
The second equality in Eq. [
13] can be treated also as an assumption on the boundary evolutions
ji. On the other hand, Eqs. [
6] and [
7] imply:
$$ \sum\nolimits_{i = 1}^{s} {\varOmega_{i} \,J_{i} } = - \sum\nolimits_{i = 1}^{s} {\varOmega_{i} \,D_{i} \frac{{\partial c_{i} }}{\partial x} + \upsilon^{D} }. $$
(14)
In consequence, from Eqs. [
9], [
13], and [
14], we obtain the formula on the drift:
$$ \upsilon^{D} = \sum\nolimits_{i = 1}^{s} {\varOmega_{i} \,D_{i} \frac{{\partial c_{i} }}{\partial x}} + K(t). $$
(15)
Finally, we get the strongly coupled nonlinear parabolic differential system:
$$ \frac{{\partial c_{i} }}{\partial t} + \frac{\partial }{\partial x}\left( { - D_{i} \frac{{\partial c_{i} }}{\partial x} + c_{i} \left( {\sum\nolimits_{k = 1}^{s} {\varOmega_{k} D_{k} \frac{{\partial c_{k} }}{\partial x} + } K(t)} \right)} \right) = 0 \, \,{\text{on}}\,\,[0,t_{p} ]\times \varOmega $$
(16)
and the nonlinear coupled boundary conditions:
$$ - D_{i} \frac{{\partial c_{i} }}{{\partial {\mathbf{n}}}} + c_{i} \left( {\sum\nolimits_{k = 1}^{s} {\varOmega_{k} D_{k} \frac{{\partial c_{k} }}{{\partial {\mathbf{n}}}}} + K(t){\mathbf{n}}} \right) = j_{i} (t,x) \, \,{\text{on}}\,\,[0,t_{p} ]\times \partial \varOmega , $$
(17)
$$ {\text{for}}\,\,\,i = 1,\,\ldots\,,\,\,s. $$
Now we formulate 2-D and 3-D models.[
24‐
26] Let
Ω ⊂
Rn,
n = 2, 3. In these cases, the volume continuity equation (Eq. [
8]) cannot be applied for finding the drift because it does not imply independency with respect to
x of the field under divergence. Assuming that the drift velocity can be written as minus of the gradient of the scalar field
Φ(
t,
x) (Eq. [
4]), we get the strongly coupled nonlinear parabolic-elliptic differential system:
$$ \left\{ \begin{aligned} &\frac{{\partial c_{i} }}{\partial t} + {\text{div}}{\kern 1pt} \left( { - D_{i} {\kern 1pt} {\text{grad}}{\kern 1pt} c_{i} - c_{i} {\text{grad}}{\kern 1pt} \varPhi } \right) = 0 \, \,{\text{on}}\,\,[0,t_{p} ]\times \varOmega , \hfill \\ &- \Delta {\kern 1pt} \varPhi = {\text{div}}\left( {\sum\nolimits_{k = 1}^{s} {\varOmega_{k} D_{k} {\text{grad}}{\kern 1pt} c_{k} } } \right) \, \,{\text{on}}\,\,[0,t_{p} ]\times \varOmega , \hfill \\ &\int\limits_{\varOmega } {\varPhi {\kern 1pt} {\kern 1pt} {\text{d}}x = 0} \, \,{\text{on}}\,\,[0,t_{p} ] ,\hfill \\ \end{aligned} \right. $$
(18)
for
i=1
, ..., s. The first of the preceding equations is the continuity equation (mass conservation) (Eq. [
3]). The second one is immediately generated by Eq. [
8] and it is an elliptic equation for potential
Φ. The third equation represents the uniqueness property for the second Poisson equation for
Φ together with the boundary condition defined later. For the system Eq. [
18], we want to find the boundary conditions. Multiplying Eq. [
7] by
Ωi and
n, adding by its sides, and using the Vegard rule (Eq. [
6]) and Eq. [
10], we have:
$$ \frac{\partial \varPhi }{{\partial {\mathbf{n}}}} = - \sum\nolimits_{k = 1}^{s} {\varOmega_{k} \left( {D_{k} \frac{{\partial c_{k} }}{{\partial {\mathbf{n}}}} + j_{k} (t,x)} \right)} \, \,{\text{on}}\,\,[0,t_{p} ]\times \partial \varOmega .$$
(19)
Hence, the nonlinear coupled boundary conditions are implied:
$$ \left\{ \begin{aligned} &- D_{i} \frac{{\partial c_{i} }}{{\partial {\mathbf{n}}}} - c_{i} \frac{\partial \varPhi }{{\partial {\mathbf{n}}}} = j_{i} \left( {t,x} \right)\;{\text{on }}[0,t_{p} ]\times \partial \varOmega , \hfill \\ &\frac{\partial \varPhi }{{\partial {\mathbf{n}}}} = - \sum\nolimits_{k = 1}^{s} {\varOmega_{k} \left( {D_{k} \frac{{\partial c_{k} }}{{\partial {\mathbf{n}}}} + j_{k} (t,x)} \right){\text{ on }}[0,t_{p} ]\times \partial \varOmega ,} \hfill \\ \end{aligned} \right. $$
(20)
for
i = 1, ...,
s. If an incompressible transport is assumed (constant overall volume), the compatibility condition on the boundary evolutions follows from the Gauss theorem:
$$ \int\limits_{\partial \varOmega } {\sum\nolimits_{i = 1}^{s} {\varOmega_{i} j_{i} \left( {t,x} \right)} {\kern 1pt} {\kern 1pt} dS = 0} \, \,{\text{on}}\,\,[0,t_{p} ]. $$
(21)
Let us stress that in 2-D and 3-D cases, the drift
υD has two or three coordinates, respectively. Hence, we cannot describe the diffusion model by the differential algebraic system (Eqs. [
3] and [
6]), because a number of unknowns is larger than a number of equations. In both cases, we have only (
s + 1) equations:
s continuity equations and one Poisson equation. By elimination of the vectorial drift-velocity component (Eq. [
4]), we have the same number of unknowns: the concentrations of
s components and the scalar potential. Obviously, in one dimension, this number is also the same. However, from a numerical and analytical point of view, it is better to study the equivalent differential system (Eq. [
16] in one dimension and Eq. [
18] in two and three dimensions). It is interesting that the parabolic-elliptic model Eqs. [
18], [
9], and [
20] can also be used in one dimension. Moreover, it leads, after integration of the elliptic equation for
Φ, to the parabolic model (Eqs. [
16], [
9], and [
17]).[
25]
3.2 Difference Scheme in Two Dimensions
To find numerical solutions for the model, we use the implicit finite difference method (FDM). Let us focus on the 2-D case,
Ω = (−
Λ,
Λ) × (−
Λ,
Λ) ⊂ R
2,
x = (
x1,
x2). It is assumed that the system is closed,
e.g.:
$$ j_{i} \left( {t,x_{1} ,x_{2} } \right) = j_{i} \left( {t,x_{1} ,x_{2} } \right) \equiv 0,(t,x_{1} ,x_{2} ) \in \left[ {0,t_{p} } \right] \times \partial \varOmega ,\;i = 1,\ldots,s $$
(22)
The initial concentrations are given (Eq. [
9]). The differential system (Eq. [
18]) and the boundary conditions (Eq. [
20]) can be written in the following form, respectively:
$$ \left\{ \begin{aligned} \frac{{\partial c_{i} }}{\partial t} = \frac{{\partial \left( {D_{i} {{\partial c_{i} } \mathord{\left/ {\vphantom {{\partial c_{i} } {\partial x_{1} }}} \right. \kern-0pt} {\partial x_{1} }}} \right)}}{{\partial x_{1} }} + \frac{{\partial \left( {D_{i} {{\partial c_{i} } \mathord{\left/ {\vphantom {{\partial c_{i} } {\partial x_{2} }}} \right. \kern-0pt} {\partial x_{2} }}} \right)}}{{\partial x_{2} }} + \frac{{\partial c_{i} }}{{\partial x_{1} }}\frac{\partial \varPhi }{{\partial x_{1} }} + \frac{{\partial c_{i} }}{{\partial x_{2} }}\frac{\partial \varPhi }{{\partial x_{2} }} + c_{i} \left( {\frac{{\partial^{2} \varPhi }}{{\partial x_{1}^{2} }} + \frac{{\partial^{2} \varPhi }}{{\partial x_{2}^{2} }}} \right){\text{ on }}\left[ {0,t_{p} } \right] \times \varOmega , \hfill \\ - \frac{{\partial^{2} \varPhi }}{{\partial x_{1}^{2} }} - \frac{{\partial^{2} \varPhi }}{{\partial x_{2}^{2} }} = \sum\limits_{k = 1}^{s - 1} {\varOmega_{k}
\left( \frac{\partial ((D_k-D_s)\partial c_k/\partial x_1)}{\partial x_1}+
\frac{\partial ((D_k-D_s)\partial c_k/\partial x_2)}{\partial x_2}\right) {\text{ on }}\left[ {0,t_{p} } \right] \times \varOmega ,} \hfill \\ \int_{\varOmega } {\varPhi \,dx = 0{\text{ on }}\left[ {0,t_{p} } \right]} , \hfill \\ \end{aligned} \right. $$
(23)
and
$$ \left\{ \begin{gathered} - D_{i} \frac{{\partial c_{i} }}{{\partial x_{j} }} + c_{i} \sum\nolimits_{{k = 1}}^{{s - 1}} {\Omega _{k} \left( {D_{k} - D_{s} } \right)} \frac{{\partial c_{k} }}{{\partial x_{j} }} = 0\,\,\,\,{\text{on }}\left[ {0,t_{p} } \right] \times \partial \Omega , \hfill \\ \frac{{\partial \Phi }}{{\partial x_{j} }} = - \sum\nolimits_{{k = 1}}^{{s - 1}} {\Omega _{k} \left( {D_{k} - D_{s} } \right)} \frac{{\partial c_{k} }}{{\partial x_{j} }}\,\,\,{\text{on}}\left[ {0,t_{p} } \right] \times \partial \Omega , \hfill \\ \end{gathered} \right. $$
(24)
for
j = 1, 2,
i = 1, ...,
s-1. The concentration of the
sth species,
cs, is calculated as:
$$ c_{s} = \frac{1}{{\varOmega_{s} }}\left( {1 - \sum\nolimits_{i = 1}^{s - 1} {\varOmega_{i} c_{i} } } \right) .$$
(25)
The modification with the use of Eq. [
25] allowed construction of the Vegard rule preserving the implicit finite difference method, generated by some linearization and splitting ideas used, to present the numerical solutions. Numerical methods that preserve some futures of a continuous model are very important and interesting from a physical and mathematical point of view. Because of the nonlinearity of the diffusion coefficients, we applied a middle points discretization that admits a special subtle approximation. In fact, two implicit linear difference schemes were defined for the system (Eqs. [
23] and [
24]), with the initial condition (Eq. [
9]):
1.
for the elliptic subsystem on the potential Φ;
2.
for the parabolic subsystem on the concentrations ci, i = 1, ..., s-1.
The Gauss elimination method has been applied to solve them. For details concerning the existence and uniqueness of the discrete solutions and other properties of the numerical method, we refer the readers to our previous articles.[
25,
26]