VARIATIONAL

TECHNIQUES IN
CHEMICAL
ENGINEERING
EXAMPLES ON
TRANSPORT
PROCESSES

Most processes in natural and

engineered systems inherently
involve flow and transport.
From atoms to galaxies, from
inorganic reactions to living
organisms, from oceans to
streambeds, flow and transport
processes occur across a wide
range of physical and chemical
applications.

Okon, John E
Akwa Ibom State
University (AKSU)
Advanced Chemical
Engineering Analysis
CPE 832
fgg
Variational Techniques in Chemical Engineering (Transport Processes)

1.0 Introduction
Most processes in natural and engineered systems inherently involve flow and
transport. From atoms to galaxies, from inorganic reactions to living organisms, from
oceans to streambeds, flow and transport processes occur across a wide range of
physical applications. Application areas include groundwater contamination, carbon
sequestration, air pollution, petroleum exploration and recovery, weather prediction,
drug delivery, material design, chemical separation processes, natural disaster
assessments, and others. As a result of its importance in a variety of problems of
interest, accurate and efficient numerical modeling of flow and transport is essential.

Interest in the numerical simulation of flow and transport mechanisms has increased
in recent years along with the rapid increase in computational power. Spectacular
advances in computer performance and emerging technologies of parallel distributed
grids and GPU computing have provided flow and transport modelers with tools to
simulate and study complex phenomena. However, efficient numerical algorithms, new
mathematical formulations, and advanced computational techniques are required if
researchers are to take full advantage of this progress and address new and existing
challenges in flow and transport modeling and simulation. One particular challenge
requires modelers to synthesize varied temporal and spatial scales at which flow and
transport mechanisms operate.

Variational methods are useful for finding numerical solutions of differential

equations, which are the corresponding Euler-Lagrange equations to the stationary
condition of the functional. Usually the functional is a maximum or a minimum with
respect to some function, but in some cases the functional is a saddle point. Fluid
dynamic problems are usually solved as a set of partial differential equations, given by
the momentum balance and the continuity equation (Bird et al, 2002). There are many
methods available, ranging from analytical solutions for simple problems to numerical
methods for more complicated ones, and the finite volume methods is one with
widespread use. Variational methods have been used only on a limited number of
cases in fluid dynamic problems, mostly due to the lack of a corresponding Variational
formulation where the stationary condition corresponds to the original set of
differential equations. In general, variational methods are useful for finding numerical

OKON JOHN|AKWA IBOM STATE UNIVERSITY 1

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

solutions of differential equations, which are the corresponding Euler-Lagrange

equations to the stationary condition of the functional. Usually the functional is a
maximum or a minimum with respect to some function, but in some cases the
functional is a saddle point.
In this paper we shall consider various techniques that have been developed over the
years and how they compare to the general conventional analytical and numerical
methods.

2.0 Saddle Point Variational Formulation

The proposed method is described as follows. Consider a function given by:

… (1)

This function has a saddle point at x1= 0 and x2 = 0. While there are many algorithms
to find a minimum point, there are not many algorithms for finding a saddle point.
Considering a transformation given by:

… (2)

… (3)

The problem can be rewritten as:

… (4)

Using an iterative procedure, fixing the value of y2, then the stationary value of y1 can
be found my minimizing the function given by Eq. (4). By a similar procedure, the
stationary value of y2 can be found.
Consider the function below:

… (5)

Subject to:

OKON JOHN|AKWA IBOM STATE UNIVERSITY 2

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

… (6a)

… (6b)

This function has stationary condition given by:

… (7a)

… (7b)

… (8a)

… (8b)

Also, consider that the boundary conditions for this problem are chosen so that the
set of equations (7a) – (8b) has the solution α (x 1, x2) = 0 and β (x1, x2) = 0, while u (x 1,
x2) = 0, and v (x1, x2) have nontrivial solutions.

Now, consider the following change of variables:

… (9)

… (10)

… (11)

… (12)

From equation (6) and (7) we can show that:

OKON JOHN|AKWA IBOM STATE UNIVERSITY 3

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

… (13)

… (14)

The functional given by equation (5) has a saddle point and it is known that in the
stationary condition α (x1, x2) = 0 and β (x1, x2) = 0. By applying equation (13) and (14)
in equation (5):

… (15)

then an iterative procedure can be used to find the stationary condition of the saddle
point:
 set the values of u2 and v2 at an initial given profile u2(0) and v2(0)
 keep u2(n) and v2(n) fixed, then minimize the functional in equation (18) with
respect to u1 and v1, satisfying the restriction given by equation (16), to get u 1(n)
and v1(n)
 make u(n+1) and v(n+1)
n
1 2 = v1
 repeat until [ u1 −u1 ] ≤ ε and [ v 1 −v 1 ] ≤ ε
(n+1) n (n+1 ) n

in this iterative procedure, the values of u2 and v2 are updated using the knowledge
that in the stationary point α = 0 and β = 0, which implies that u 1 = u2 and v1 = v2. In
order to have good convergence, a relaxation factor ω can be used as follows:

… (16)

… (17)

Finding the stationary value of Equation (18), satisfying restrictions (16) and (17), is
equivalent to solving the set of differential equations given by (7) - (9).

OKON JOHN|AKWA IBOM STATE UNIVERSITY 4

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

2.1 Mathematical Model

The fluid dynamic model considers a general heterogeneous system with two phases,
which could be gas-solid, liquid-solid, gas-liquid, or liquid-liquid (immiscible). The two
phases are designated by: i = 1,2. The equation for the steady state isothermal two
phase flow in cylindrical coordinates with axi-symmetry (r, z) is given by:

 Volumetric fraction balance:

ε1 + ε2 = 1

 Mass balance for i = 1,2

… (18)

If there is no mass transfer between the phases, ζ i = 0

 Momentum balance for i = 1,2

… (19)

… (20)

For a vertical pipe, gz = -g. for a horizontal pipe, both gz and gr may be neglected.
When the flow is in laminar regime, the velocity in the previous equations is the actual
velocity, while the viscosity is just the viscosity for a Newtonian fluid, µ. However,
when the flow is in turbulent regime, the velocity is an average over time fluctuations,
even when the model is for permanent flow, and the viscosity is an effective viscosity
given by:

… (21)

OKON JOHN|AKWA IBOM STATE UNIVERSITY 5

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

The turbulent viscosity, µ(t), can be calculated using different turbulence models, such
as k – ε (Hillmer et al, 1994). In this work, it is considered a zero order turbulence
model (Menzel et al, 1990;). For the sake of simplicity, since in this case µ (t) is a
function of local position and not on local velocity profiles. However, in zero order
models the turbulent velocity depends on the wall shear stress.
The diffusivity of one phase into another (D ij) affects the distribution of the volume
fractions of the phases inside the reactor volume (ε i), even if they are immiscible. It can
be shown that Dji = Dij and that these values may be variable. It can also be shown
that for turbulent flow the Schmidt number is equal to 1, so that:

… (22)

The force between the phases can be described by:

… (23)

… (24)

Where it can be seen that Fjir = - Fijr and Fjiz = - Fijr. The value CD may be considered
either a constant or a function of radial position only, and it can have different values
for different systems.
The transversal lift force is also known as the Magnus force. Here it is considered in
the most general case, where both phases can affect each other. In the case
considered in this work, where the velocities have only components on r and z
directions, it results in:

… (25)

… (26)

The boundary conditions for this model are the usual ones, both for the velocities at
the inlet, and for the centre and the walls.

OKON JOHN|AKWA IBOM STATE UNIVERSITY 6

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

2.2 Numerical Implementation

The proposed method was implemented in GAMS, solved with the CONOPT3 solver.
For a single-phase fluid, Equation (18) is used, while for two phases fluid flows a more
general functional is used that results in the model described previously. The
derivatives were calculated using second order finite differences, while the integrals
were calculated using the trapezoidal rule, which for cylindrical coordinates is given
by:

… (27)

Two case studies were solved using the proposed method. The first case considers a
two-dimensional laminar flow of a single fluid in a pipe and the results were compared
with an analytical solution. The second case considers a one-dimensional turbulent
laminar flow in a gas-liquid column and the results were compared with a numerical
solution using the finite volume method.

3.0 Variational Calculus to Wetting Phenomena

One of the most fundamental principles of physics requires that a system, when faced
with a choice between several possible states, select the energetically most favorable
option, the state with the lowest energy. An enormous variety of phenomena are
governed by this principle, and a range of mathematical methods have been developed
to determine the minimal possible energy values. Among these the variational
minimization technique is one of the most common.

The problem of determining the equilibrium shape of a wetting meniscus is proposed

as a rich example of application of the variational method. The problem describes a
common-place phenomenon that is conceptually simple and physically tangible for
undergraduate students. The proposal has a further appeal in that it illustrates how
more abstract variational boundary equations can be implemented. It also represents

OKON JOHN|AKWA IBOM STATE UNIVERSITY 7

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

a system that gives rise to both stable (minimum) and unstable (maximum) wetting
profiles, and by utilizing these it leads to a criterion earmarking eventual existence
limits of solutions of the variational equations.

3.1 Laplace and Young-Dupre Equations of Capillarity

one recognized way of determining the equilibrium shape of the interface between two
stationary fluids involved in the wetting of a solid surface Is by using a force balance
argument. A curved interface in equilibrium requires that the forces acting on a
surface element be in balance. In the case in point this force balance is expressed in
terms of the local mean (Gaussian) curvature and the pressure difference across the
ideal surface:

… (28)

Here, the pressure difference comes from buoyancy effects, g∆p = g (pL – pG). the two
principle radii of curvature, 1 and R2, are expressed differentially in terms of the local
shape function, z = z(r). by doing so the usual Laplace differential equation governing
the shape of the profile is obtained,

… (29)

A force balance argument, applied to the line of three phase contact on the solid
surface, is traditionally used to deduce a boundary condition to invoke when is
integrated: the vector sum of the three surface tensions acting on the contact line Is
zero. Applying the reasoning to the component of the net force in the direction parallel
to the solid (figure 1b), leads to the well-known force balance

… (30)

This relation is known as the young-dupre equation for the contact angle θ 0 = φ0 +
/2, defined as the angle between the fluid interface and the solid surface, measured
through the liquid region. φ0 denotes the angle the fluid interface makes with the

OKON JOHN|AKWA IBOM STATE UNIVERSITY 8

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

horizontal (figure 1). Relation (2) depends only on the different surface tension
coefficients and is independent of either the curvature of the solid, the curvature of the
contact line or the gravitational properties of the system. In the force balance
derivation of relation (2), one implicitly assumes the contact line has zero curvature.

Figure 1 Figure 2

(a) A schematic demonstrating the wetting Cross-sectional demonstrating the wetting of a

of constant circular cross-section. The semi-infinite paraboloid cylinder of shape, z s(rs) =
solid is shown in dark grey, while the zs(rs) = zp0 + λr2s, being a special case of an
liquid phase is shown in light grey. The axisymmetric solid of shape, zs(rs) = h(rs).
point (r0, z0) indicates the three-phase
contact point (line) (a) shows a hydrophilic solid for which zs > 0.

(b) shows a hydrophobic solid that is partially

(b) Close-up view of the contact region. immersed.
Arrows indicate the direction of surface
tensions considered as directed forces
per unit length.

OKON JOHN|AKWA IBOM STATE UNIVERSITY 9

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

Equation (2) can be rewritten in terms of the slope of the profile at the line of contact
with the solid to give the boundary condition, that the profile has to satisfy on the
surface

… (31)

The slope at the cylinder is opposite in sign to ∆γ, and since the profile is monotonic,
the height of the contact line, z 0, will have the same sign as ∆γ. Consequently, a
hydrophilic cylinder ∆γ > 0 will see a meniscus rise, while a hydrophobic cylinder, ∆γ <
0 will see a meniscus fall.

3.2 Energy Functional for Ideal Capillarity: Infinite Cylinder

An application of the variational functional begins with the definition of a functional,

normally an energy functional that is to be minimized. In this case the functional is
the energy change,

… (32)

Associated with the deformation of the fluid interface E and a preferential wetting of
the immersed solid. The first term represents the favorable increase in contact energy
resulting in a change in contact area with the solid, ∆A Ω. the second term represents
the unfavorable increase in surface energy, proportional to the increase in fluid
interfacial area ∆AE resulting from the induced deformation of E, propagating away
from the solid. The last term is also an unfavorable contribution whose origin is the
gravitational potential proportional to the mass of liquid displaced due to deformation.
At equilibrium these combine in such a way that the total energy change is minimal.

Although equation (32) applies generally to the wetting problem, we confine attention
here to a cylinder of infinite length and radius r 0, vertically immersed in a semi-infinite
bath of liquid, with a gaseous phase filling the upper half phase. An axisymmetric
meniscus is created with three phase circular contact line at height z 0.

OKON JOHN|AKWA IBOM STATE UNIVERSITY 10

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

Suppose an arbitrary, axisymmetric wetting profile denoted by z(r) with meniscus

height z0 is created. The change in the energy of the system measured with respect to
the undeformed interface can then be expressed as

… (33)

√
Where f:= γ(r 1+ z 2r −r ) + g∆prz2/2 is a sum of surface and gravitational contributions,
associated with the second and third terms of equation (32). The first term of (32), the
change in contact energy at the boundary, is here represented by ψ(z 0): =2∆γr0z0.

3.3 Energy Functional for Ideal Capillarity: Semi-Infinite Parabolic Cylinder

Considering a similar system wherein the infinite cylinder is replaced by a semi-

infinite, rotationally symmetric solid whose vertical cross-section is given by an
injection, monotonic increasing function of the form zs = h(r s); dh/drs =: hrs ≥ 0. As in
the previous section, a wetting meniscus is created when the solid is partially
immersed in the liquid bath. The circular, three phase contact line again takes on a
height, z0. The energy functional of the system is:

… (34)

Where,

… (35)

And

… (36)

Implementing the variational method involves adopting optimal profile functions, r ()
and z (), and expressing an arbitrary profile by the pair (ř, ž) where ř () = r () + dr ()
and ž () = z () +dz (). As before, the perturbations dr and dz will satisfy the
asymptotic conditions dr, dz → 0 as  → ∞. On the other hand, at the object-end of the
profile we set no requirements on the perturbation values, bar one. A natural inherent
constraint is that whatever form the surface profile eventually takes, it must reside on

OKON JOHN|AKWA IBOM STATE UNIVERSITY 11

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

or be connected to the solid surface, S (rs, zs) = h(rs) – zs = 0. Consequently, both

pairs (rs, zs) and (řs, žs) must adhere to this constraint, which then prevents the
perturbations (dr, dz) from varying completely independently. This additional
consideration can be included in a number of ways, most simply by creating a
modified functional

F = ∆F + µS, … (37)

Where µ is a Langrange multiplier. Variation of F can now be effected without

consideration of the constraint. The first-order approximation to the functional
change, df [dr, dz; r, z] = F [r + dr, z + dz] – F [r, z], found by expanding to first order in
the perturbations, dr and dz, which are now completely independent, possesses
several contributions. First there is the pair of Euler-Langrange differential equations.

… (38)

Arising from terms in the integral, ∆F, that are proportional to dr and dz, respectively.
Then there arise expressions proportional to the perturbations evaluated at the
boundary, dr ( = 0) and dz ( = 0), respectively,

… (39)

Eliminating the Langrange multiplier from between these two equations gives what
Courant and Hilbert refer to as a natural boundary condition

… (40)

4.0 Variational Principle for Mass Transport in Solids

Mass transport in solids is the fundamental kinetic process controlling both the
evolution of materials towards equilibrium and a variety of material properties.
Diffusion of atoms dictates everything from the stability of amorphous materials at
finite temperature, the design of Nano-scaled semiconductor devices, the processing of
structural metals including steels and super alloys, the performance of batteries and

OKON JOHN|AKWA IBOM STATE UNIVERSITY 12

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

fuel cells, to the degradation of materials due to corrosion or even irradiation. Since
Einstein, diffusion has been understood as mesoscale motion arising from many
individual atomic displacements, with significant effort over the last century to
experimentally measure and model theoretically. In the last forty years, computation
has played an increasingly important role, with different competing approximation
methods developing, combined with increasingly accurate methods to compute
transition state energies for atomic processes in transport. However, while we have
increasing accuracy in predicting atomic scale mechanisms, we lack a clear
methodology to compare accuracy of theoretical models that derive mesoscale
transport coefficients.
The modern macroscale description of mass transport comes from Onsager’s work on
non-equilibrium thermodynamics, where atomic fluxes J are linearly proportion to
small driving forces. A general driving force is the gradient of chemical potential of
species α. Then, the Onsager transport coefficients are second-rank tensors L(αβ) that
relate steady state fluxes in species α.

… (41)

are steady-state fluxes in response to perturbatively small driving forces in chemical

species ∆µβ. These transport coefficients can also be derived from a thermodynamic
extremal principle. for maximum entropy production, making the Onsager matrix
symmetric and positive semidefinite.
A brief, albeit incomplete list of methods to compute transport coefficients from atomic
mechanisms include stochastic methods like kinetic Monte Carlo, master-equation
methods like the self-consistent mean-field method and kinetic mean-field
approximations, path probability methods for irreversible thermodynamics, Green
function methods, and Ritz variational methods. The different approaches all have
different computational and theoretical complexity, rely on different approximations
which may or may not be controlled. However, the relationships between different
approximations is not always clear, and it is difficult to determine which of two
different calculations is more accurate, short of comparison to experimental results. In
what follows, we derive a general expression for the mass transport coefficients in a
solid system, and then cast this non-local form into an equivalent minimization

OKON JOHN|AKWA IBOM STATE UNIVERSITY 13

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

problem over thermodynamic averages of local quantities: a variational principle for

mass transport, with a simple physical interpretation. We show that different
computational approaches can be derived and compared with this principle, while also
providing a framework for the development of new types of approximations for
diffusion. We conclude with a quantitative comparison for a random alloy on a square
lattice.
Consider a system with chemical species α = A, B…, with discreet microstates {X}, and
transition states. For each state x and species α, Nαx of heat species are at positions {
α α α
X x :i=1 … N x }. Note that X x are a function of the state of X. if each state has an energy
i i

Ex then in the grand canonical ensemble, the equilibrium probability of occupying a

given microstate for chemical potentials µα at a temperature T is
… (42)

Where φ0 is a normalization constant – the grand potential – such that ∑ P 0x = 1. If the

x

chemical potentials were spatially inhomogeneous, then the term corresponding to the
sum over the chemistry would be
We assume that our system can achieve equilibrium through a Markovian process,
with transition rates W (x → x’) ≥ 0. then, by detailed balance,
0 0
P x W ( x → x ’ )=P x W ( x ' → x )If all non-zero rates conserve chemical species, then the
rates W (x → x’) are independent of the chemical potentials, and can only depend on
the initial and final states and temperature. The master equation for the evolution of a
time dependent probability P(t) is:

… (43)

And we introduce the matrix form:

… (44)

We identify steady state solutions of Eqn. (43) — which may not be equilibrium
solutions — as distributions where the right-hand side is zero for every x; we are
OKON JOHN|AKWA IBOM STATE UNIVERSITY 14
fgg
Variational Techniques in Chemical Engineering (Transport Processes)

interested in steady-state solutions that maintain infinitesimal gradients in chemical

potentials, for which we will compute fluxes. What follows is a generalization of results
derived previously for a lattice gas model; details are available in the supplemental
material. Consider a steady-state probability distribution Pssx : = Pssx (T, µA,…,∆µA,…,) in
the presence of infinitesimally small chemical potential gradient vectors ∆µ α. This
steady state probability distribution can have time independent fluxes Jα
corresponding to mass transport. For any (non-zero rate) transition ( x → x ’ ) we define
the mass transport

vector for each species α as δ X αXX ' : = ∑ X αx ' i - X αxi. This is the net change in position for
i

α α
all the atoms of species α, as N x = N xi when W( x → x ’ ) ≠ 0. Then the flux is:

… (45)

for total system volume V0. We make the ansatz that the steady-state probability
distribution for infinitesimal gradients
… (46)

up to first order in ∆µα, where δΦ0 is a change in the normalization relative to the
α
equilibrium distribution, and introducing the relaxation vectors η x that are to-be-
determined for each state x. These vectors are a generalization of the rate-dependent
relaxation in solute-vacancy exchange. If we substitute Eqn. (46) into Eqn. (43), set d
ss 0
P x ,/dt = 0 apply detailed balance, divide out by P xand require that it hold for arbitrary
∆µα we find

… (47)

We define the left-hand side as the velocity vector,

… (48)

So that equation (47) becomes

OKON JOHN|AKWA IBOM STATE UNIVERSITY 15

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

for the steady-state ansatz solution to be time invariant. Then, the transport
coefficients L(αβ) can be found by substituting the steady-state solution into Equation.
(45), while explicitly symmetrizing the summation which gives
… (49)

First, the separation of Equation. (49) into correlated and uncorrelated terms is
arbitrary. We introduce changes to the positions of atoms in a state while leaving the
α
rate matrix unchanged: Let y x be the sum of all displacements of atoms of species α in

state x. We can, without loss of generality, consider only cases where ∑ y αx Then, the
x

α
y x change the displacement, velocity, and relaxation vectors

α
as G is the pseudoinverse of W, and y x is orthogonal to the right null space of W.
Then, the Onsager coefficients are:

OKON JOHN|AKWA IBOM STATE UNIVERSITY 16

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

We can exploit this invariance by noting that, for α = β, the uncorrelated contribution
is positive semidefinite and the correlated contribution is negative semidefinite, as
Wxx’ and Gxx’ are negative semidefinite matrices. Thus, the maximum value of the
correlated contribution is zero, which corresponds with the minimal value of the
uncorrelated contribution, and so the equation for the transport coefficients can be
rewritten as:

… (50)

which is a variational principle for mass transport involving only thermodynamic

averages of local rate matrix. Here, the infimum of the tensor corresponds to the
α
tensor with the smallest trace. The values of y x that minimize Equation. (50) are found
by making the generalized force from the gradient of

… (51)

α α α
equal to zero; this is satisfied when y x = η x Moreover, the arguments y x that minimize
ℓα can then be used to compute the off-diagonal contributions,

… (52)

This variational principle for mass transport has multiple consequences. First, it
unifies multiple approaches for the computation of mass transport coefficients,
including kinetic Monte Carlo, Green function methods, and self-consistent mean-field
theory. Moreover, it provides a direct way to compare the accuracy of different

OKON JOHN|AKWA IBOM STATE UNIVERSITY 17

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

methods: outside of the convergence of stochastic sampling errors, once a mass

transport method is recast in a variational form, the minimal value of the diagonal
transport coefficients is necessarily closer to the true value. It also gives a simple
physical explanation for the correlation contributions in mass transport: the η αx values
are displacements that map a correlated random walk into an equivalent uncorrelated
random walk with identical transport coefficients. Finally, it provides a framework for
the construction of new algorithms for the computation of mass transport that
requires the minimization of a thermal average; as it is based on minimization,
α
different approximations for y x can be simultaneously introduced, while the process of
minimization finds the optimal solution.

OKON JOHN|AKWA IBOM STATE UNIVERSITY 18

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

References
Bird R.B., Stewart W.E., Lightfoot E.N., 2002, Transport Phenomena, John Wiley &
Sons Inc., 2nd Edition, p. 846-848.

Chen Z., Zheng C., Feng. Y., 1995, Modeling of three-phase fluidized beds based on
local bubble characteristics measurements, Chemical Engineering Science, 50, 231–
236.

Grienberger J., Hofmann H., 1992, Investigations and modelling of bubble columns,
Chemical Engineering Science, 47, 2215–2220.

Guirardello R., 2015, A variational formulation for steady two-dimensional laminar

flow, Far East Journal of Applied Mathematics, 93, 191–216.

Hillmer G., Weismantel L., Hofmann H. 1994, Investigations and modelling of slurry
bubble columns, Chemical Engineering Science, 49, 837–843.

Menzel T., Weide T., Staudacher O., Wein O., Onken U., 1990, Reynolds shear stress
for modeling of bubble column reactors, Ind. Eng. Chem. Res., 29, 988–994.

Torvik R., Svendsen H.F., 1990, Modelling of slurry reactors. A fundamental approach,
Chemical Engineering Science, 45, 2325–2332 .

R. W. Ballu_, S. M. Allen, and W. C. Carter, Kinetics of Materials (John Wiley & Sons,
Inc., 2005).

A. Einstein, Ann. d. Physik 322, 549 (1905).

A. R. Allnatt and A. B. Lidiard, “Atomic transport in solids,” (Cambridge University

Press, Cambridge, 1993) Chap. 5, pp. 202–203.

H. Mehrer, Dilution in Solids: Fundamentals, Methods, Materials, Di_usion-Controlled

Processes, Springer Series in Solid-State Sciences (Springer-Verlag Berlin Heidelberg,
2007).

A. Janotti, M. Krˇcmar, C. L. Fu, and R. C. Reed, Phys. Rev. Lett. 92, 085901 (2004).

OKON JOHN|AKWA IBOM STATE UNIVERSITY 19

fgg
Variational Techniques in Chemical Engineering (Transport Processes)

M. Mantina, Y. Wang, L. Q. Chen, Z. K. Liu, and C. Wolverton, Acta mater. 57, 4102
(2009).

T. Garnier, Z. Li, M. Nastar, P. Bellon, and D. R. Trinkle, Phys. Rev. B 90, 184301
(2014).

L. Onsager, Phys. Rev. 37, 405 (1931).

OKON JOHN|AKWA IBOM STATE UNIVERSITY 20