PORE-SCALE SIMULATION OF MULTIPHASE FLOW USING LATTICE BOLTZ.docx

PORE-SCALE SIMULATION OF MULTIPHASE FLOW
USING LATTICE BOLTZMANN
METHOD: DEVELOPMENTS AND APPLICATIONS
by
Jingwei Huang
Copyright by Jingwei Huang 2016
All Rights Reserved

ii
A thesis submitted to the Faculty and the Board of Trustees
of the Colorado School of Mines
in partial fulfillment of the requirements for the degree of
Master of Science (Petroleum
Engineering).
Golden, Colorado
Date ________________
Signed: _______________________
Jingwei Huang
Signed: _______________________
Dr. Xiaolong Yin
Thesis Advisor
Golden, Colorado
Date ________________

Signed: _____________________
Dr. Erdal Ozkan
Professor and Head
Department of Petroleum
iii
ABSTRACT
Multiphase flow is an important transport process that is
found in many natural phenomena
and engineering applications. In petroleum engineering, sub-
surface multiphase flow takes place
throughout the primary, secondary, and tertiary stages of
recovery. Advances in high-
performance computing and experimental techniques now allow
us to study multiphase flow at
the pore level to better understand pore-scale dynamic and its
continuum-level descriptions. In

this work, a parallelized multiphase flow simulator is developed
using the lattice Boltzmann
method with the color-gradient multiphase model. Several
boundary conditions for the
multiphase flow model are developed, and they allow the
simulator to handle a range of
practically relevant multiphase flows that are pressure-driven,
rate-driven, or gravity-driven. The
simulator is parallelized, which is critical for simulating flows
in computational domains of
meaningful sizes. The parallel performance is good in large
computation domains. Validation
tests show that the simulator is able to generate controlled
interfacial tension, wettability and can
accurately simulate dynamic problems. As an application, we
simulated pressure-driven
displacements in a porous medium made up by packed spheres
using fluids with different
viscosity ratios and wettability at different capillary numbers.
Our results are consistent with
published numerical and experimental results.
iv

TABLE OF CONTENTS
ABSTRACT
...............................................................................................
.................................... iii
LIST OF FIGURES
...............................................................................................
........................ vi
LIST OF TABLES
...............................................................................................
........................ viii
ACKNOWLEDGEMENTS
...............................................................................................
............ ix
CHAPTER 1 INTRODUCTION
...............................................................................................
..... 1
CHAPTER 2 LITERATURE REVIEW
......................................................................................... 5
2.1 Single-phase LB method and boundary condition
................................................................ 5
2.2 Two-phase LB method
...............................................................................................
........... 7
2.3 Boundary conditions for two-phase LB method
................................................................. 10

CHAPTER 3 SINGLE- AND MULTIPHASE LATTICE
BOLTZMANN METHODS AND
PARALLELIZATION
...............................................................................................
............... 12
3.1 Single-phase LB method and boundary conditions
............................................................. 12
3.1.1 Single-phase LB method
..............................................................................................
12
3.1.2 Pressure and Velocity boundary conditions for single-
phase LB method .................... 16
3.2 Multiphase LB model
...............................................................................................
........... 21
3.3 Parallelization of color-gradient LB code
........................................................................... 24
3.4 Test cases
...............................................................................................
.............................. 31
CHAPTER 4 BOUNDARY CONDITIONS FOR MULTIPHASE
LATTICE BOLTZMANN . 39
4.1 Evaluation of existing outflow boundary conditions for two-
phase LB method ................ 39
4.2 Modified periodic boundary condition for two-phase LB
method ..................................... 43

4.3 Pressure boundary condition for two-phase LB method
..................................................... 45
CHAPTER 5 SIMULATION OF IMMISCIBLE
DISPLACEMENTS IN A POROUS
MEDIUM
...............................................................................................
................................... 59
v
CHAPTER 6 CONCLUSIONS AND FUTURE WORK
............................................................. 74
6.1 Summary and Conclusions
...............................................................................................
... 74
6.2 Future work
...............................................................................................
.......................... 76
REFERENCES CITED
...............................................................................................
.................. 78

vi
LIST OF FIGURES
Figure 3.1 D3Q19 propagation scheme.
....................................................................................... 13
Figure 3.2 Velocity directions at boundary.
.................................................................................. 17
Figure 3.3 Scheme of decomposition: (a) Computation domain
before decomposition; (b)
Computation domains after decomposition.
................................................................ 26
Figure 3.4 Porous medium used for parallel performance test.
.................................................... 29
Figure 3.5 Speedup of multiphase lattice Boltzmann
code........................................................... 30
Figure 3.6 Parallel efficiency of multiphase lattice Boltzmann
code. .......................................... 31
Figure 3.7 Oscillation and relaxation of a square drop to
equilibrium spherical shape. ............... 33
Figure 3.8 Numerical measurement of Laplace law.
.................................................................... 35

Figure 3.9 Thickness of interface using color-gradient lattice
Boltzmann model. ....................... 36
Figure 3.10 Force balance at the contact line.
.............................................................................. 37
Figure 3.11 Contact angles achieved by setting � and � on
wall nodes. .............................. 38
Figure 3.12 Droplet deformation in Stokes flow.
......................................................................... 40
Figure 3.13 Deformation parameter as a function of ; at =0.1.
...................................... 41
Figure 4.1 Schematic of outlet boundary.
..................................................................................... 40
Figure 4.2 Phase distribution of a moving droplet in a channel
using the convective outflow
boundary condition.
...............................................................................................
..... 42
Figure 4.3 Phase distribution of two-phase displacement in a
channel using the convective
outflow boundary condition.
....................................................................................... 43
using the modified periodic
boundary condition.
...............................................................................................
..... 44
channel using modified periodic

boundary condition.
...............................................................................................
..... 45
Figure 4.6 Schematic of inlet and outlet
boundaries..................................................................... 46
Figure 4.7 Phase distribution (a) and corresponding
dimensionless pressure field (b) of a
moving droplet in a channel using the pressure
boundary condition. The
dimensionless pressure is defined as +⁄ .
.............................................. 51
Figure 4.8 Phase distribution (a) and corresponding pressure
field (b) of two-phase
displacement in a channel using the pressure
boundary condition. dimensionless
pressure is defined as +⁄ .
..................................................................... 53
Figure 4.9 Simulation setup for capillary filling. The capillary
tube has a length L and diameter
d. The red fluid is the invading fluid and the blue
fluid is the defending fluid. ......... 55
vii
Figure 4.10 The length of tube occupied by the filling fluid as a
function of time ∗ = / for
(a) drainage, M = 1 (b) imbibition M = 1 (c)
drainage M = 0.1 and (d) drainage
M = 10. The lines represent the analytical solution
based on Eqn. (4.22) and the

crosses mark the position (1, 1).
............................................................................... 56
Figure 4.11 Pressure distributions along the flow direction for
(a) drainage, M = 1 (b)
imbibition M = 1 (c) drainage M = 0.1 and (d)
drainage M = 10 at ∗ = / =
0.55, 0.55, 0.50, 0.50 respectively. P* is the
dimensionless pressure defined as
+⁄ .
...............................................................................................
....... 57
Figure 5.1 The cylindrical porous medium made up by packed
spheres. ..................................... 60
Figure 5.2 Two-phase distribution (a) and wetting-phase
pressure distribution (b) in the sphere
packing at different times (∗ = 0.12, 0.47, 0.71,
1.18), Case A. ................................. 63
1.18), Case B. ................................. 64
1.18), Case C. ................................. 65
1.18), Case D. ................................. 66
Figure 5.6 Saturation distributions at different times in Case A,
B, C, and D. Time interval ∆ ∗
between consecutive lines is 0.059. The red line
represents the saturation
distribution when breakthrough occurs.

...................................................................... 67
Figure 5.7 Phase pressure distribution of Case A, B, C, and D.
The open symbols represent
pressure profiles at times before breakthrough: ∗ =
0.89, 0.47, 0.30, 0.12
respectively. The lines represent the pressure
profiles after breakthrough: ∗ = 1.18
for Case C and D.
...............................................................................................
......... 68
Figure 5.8 Relative permeability vs. wetting phase saturation of
the sphere packing. Red and
green points denote breakthrough points. Time
interval ∆ ∗ between consecutive
points is 0.059.
...............................................................................................
.............. 69
viii
LIST OF TABLES
Table 3.1 Stationary bubble test. �� = � = , �� = � = , =
.......................................... 34
Table 3.2 Stationary bubble test. �� = � = , �� = � = , = .
....................................... 34
Table 4.1 Comparison of breakthrough times in the case of
capillary filling ............................... 58
Table 5.1 Parameters in different cases of immiscible

displacement ........................................... 61
ix
ACKNOWLEDGEMENTS
First of all, I would like to express my gratitude to my
advisor, Dr. Xiaolong Yin, for his
guidance, support and help for this work. I also want thank my
committee members, Dr. Hossein
Kazemi and Dr. Azra Tutuncu for their valuable comments. I
appreciate the help from Dr. Philip
Winterfeld for the help to use the EMG cluster. My appreciation
also goes to Dr. Timothy Kaiser,
Director of Research and High-Performance Computing at
Colorado School of Mines, for his
help on code testing on BlueM.
I thank my mother for her love, encouragement and support
all the time. I want to extend my

appreciations to all my friends at Colorado School of Mines.
Financial supports from the Unconventional Natural Gas and
Oil Institute, Coupled Integrated
Multiscale Measurement and Modeling consortium, and the
Department of Energy through the
Nuclear Energy University Program DE-NE0000719 are
gratefully acknowledged.
1
CHAPTER 1
INTRODUCTION
In petroleum engineering, multiphase flow through porous
media is a key process that
controls production throughout the primary, secondary, and
tertiary stages of a reservoir.
Understanding the dynamics of multiphase flow is essential for
one to predict and control
reservoir production. A large amount of theoretical,

experimental, and numerical studies have
already been dedicated to this area.
On the macroscopic level, Darcy’s law is used to describe
fluid flow through porous media. It
is known that Darcy’s law originated from averaging of the
Stokes flow (Whitaker 1998).
Description of multiphase flow on the continuum level, on the
other hand, usually uses a
multiphase extension of Darcy’s law, which is empirical and
does not have a strong theoretical
foundation (Hassanizadeh and Gray 1979). In this extension,
relative permeability is introduced
to account for the interaction of the two-phases in the pore
space. However, it is still a challenge
to model relative permeability as it is affected by various
factors including pore structure,
saturation, viscosity ratio, capillary number, wettability, as well
as the way that flow is generated
and maintained in porous medium. Traditionally, relative
permeability is determined by
experiment, which is time consuming and often difficult to
interpret. It is more difficult to
conduct multiphase flow experiments in tight porous media
because of their low permeability

(Karimi et al. 2015). Therefore, it is of great interest to study
multiphase flow from the level of
pores with numerical simulations so that one can attain better
understanding of pore-scale
dynamics and improve continuum-level descriptions.
2
Direct numerical simulation of multiphase flow through
porous media is an important
approach to study complex multiphase flow processes, such as
those in multiphase reactors and
heat/mass exchangers (Kreutzer et al. 2005; Dudukovic et al.
1999), microfluidics (Xu et al.
2014; Wu et al. 2012; Wu et al. 2013), oil and gas production
(Orr and Taber 1984; Cueto-
Felgueroso and Juanes 2013), contaminant transport and
remediation (Esposito and Thomson
1999; Mackay and Cherry 1989; Hunt et al. 1988), and
geological storage of greenhouse gases
(Szulczewski et al. 2012). Compared to continuum-scale
simulations and pore network models,

pore-scale direct numerical simulation directly solves the
dynamics of fluid motion in pore space,
and therefore is a more fundamental approach to understand the
driving and controlling
mechanisms of multiphase flow such as viscous coupling,
capillary pressure and the gravity
force (Joekar-Niasar and Hassanizadeh 2012).
Many approaches have been developed to directly simulate
multiphase flow in porous media.
These approaches include the phase field (PF) method (Cahn
and Hilliard 1958; Cahn and
Hilliard 1959; Badalassi et al. 2003; Lowengrub and
Truskinovsky 1998), the volume-of-fl uid
(VOF) method (Hirt and Nichols 1981; Rider and Kothe 1998;
Pilliod and Puckett 2004), the
level-set (LS) method (Osher and Sethian 1988; Sussman et al.
1998), the density functional
method (DFM) (Demianov et al. 2011) and the lattice
Boltzmann (LB) method (Frisch et al.
1987; Chen et al. 1991; d'Humières 1992; Gunstensen et al.
1991; Gunstensen and Rothman
1993; Grunau et al. 1993; Shan and Chen 1993; Shan and
Doolen 1995; Swift et al. 1995). In the
PF model, the interface is described as a thin transition layer

where unstable mixtures are
stabilized by non-local energy terms (Cahn and Hilliard 1959;
Lowengrub and Truskinovsky
1998). A very thin interface layer is required to improve the
accuracy of simulation which may
result in expensive computation cost (Badalassi et al. 2003).
The VOF algorithms consist of three
3
parts: interface reconstruction, advection and surface tension
(Gerlach et al. 2006). The VOF
method, as its name implies, tracks the volume of each fluid in
the cells that belong to the
interface (Gerlach et al. 2006). The LS method defines the
interface by a level-set function
(Osher and Sethian 1988) and has been widely used due to its
simplicity (Osher and Fedkiw
2001). Re-initialization techniques are employed to overcome
interface smearing and enforce
mass conservation (Sussman et al. 1994). However, VOF and LS
method may face numerical
instability when simulating flows in porous media where

capillary forces generally dominate
over viscous forces (Scardovelli and Zaleski 1999). In LB
methods, phase segregation is
modeled by inter-particle interactions and interface tracking is
not needed (He et al. 1999; Reis
and Phillips 2007). Similar to PF, interfaces are diffuse and
much thicker than real interfaces in
LB methods. However, for flows where interfacial thickness is
not crucial, owning to LB
methods’ simplicity and high computation efficiency (Liu et al.
2012), they have become
attractive numerical options for direct simulation of multiphase
flow (Kang et al. 2004; Kang et
al. 2005; Pan et al. 2004; Li et al. 2005; Aidun and Clausen
2010; Chen et al. 2014).
The objective of this study is to develop a multiphase lattice
Boltzmann model and a number
of flow driving conditions to facilitate simulations of
multiphase flow in porous media. A
parallel, color-gradient based multiphase lattice Boltzmann
model is built and several flow
driving conditions have been implemented and compared. With
the developed simulator, we
conducted simulations of immiscible displacement in a complex

geometry made up by packed
spheres.
The thesis is organized as follows: Chapter 1 introduces
motivations and objectives of this
study. Chapter 2 reviews previous work on single-phase lattice
Boltzmann, multiphase lattice
Boltzmann and relevant boundary conditions. Chapter 3 first
describes the lattice Boltzmann
4
model applied in this study and the validation tests. Then, the
parallel algorithm and parallel
performance of the simulator are presented. Chapter 4 studies
the performance of several
boundary conditions for multiphase lattice Boltzmann. In
Chapter 5, immiscible displacements
with different viscosity ratios, wettability, and capillary
numbers are simulated in a porous
medium made up by packed spheres. Conclusions and
recommended future work are presented
in Chapter 6.

5
CHAPTER 2
LITERATURE REVIEW
In this section, we summarize previous studies on single-
phase and multiphase lattice
Boltzmann models and applications. Different boundary
conditions for lattice Boltzmann are also
reviewed. Through this review, we show that developing proper
boundary conditions for
multiphase lattice Boltzmann method is critical for simulation
of many practically relevant flows.
2.1 Single-phase LB method and boundary condition
Traditional computation fluid dynamics (CFD) methods
directly solve the Navier-Stokes
equations that are derived from conservation of mass and
momentum on the continuum level.

Molecular dynamics, on the other hand, focuses on the motion
of individual molecules; they can
determine velocities and locations of molecules at every instant
of time. Lattice Boltzmann
method, as a mesoscopic method, deals with the velocity
distribution function of molecules. LB
method, compared with traditional CFD, has the advantage that
it does not require complex
discretization of partial differential operators; compared with
molecular dynamics, it does not
need to focus on the details of every molecule, which
significantly reduces computational cost.
With these said, it is also important to keep LB method’s
limitations in mind: due to its lattice
nature and connection to the kinetic theory of fluids, it is not as
flexible as traditional CFD when
it comes to complex fluids (e.g. non-Newtonian) and complex
boundary conditions. Because LB
method still relies on continuous descriptions of molecular
velocity distributions, it should not be
applied to flows with significant molecular granularity and
fluctuations.

6
Lattice Boltzmann method was originated from lattice gas
(Frisch et al. 1987) and gained a lot
of attention after it was invented. LB method models time
evolution of a discrete molecular
velocity distribution function through propagation and collision
steps. Propagation moves the
distribution function on a space-filling lattice to simulate the
streaming motion of molecules.
Collision, on the other hand, simulates the viscous relaxation of
the non-equilibrium part of the
velocity distribution. Many different collision schemes have
been developed. Single-Relaxation-
Time scheme (Chen et al. 1991; Qian et al. 1992), which is
based on the BGK collision operator
in Boltzmann equation (Bhatnagar et al. 1954), is very simple
but is more liable to numerical
instability in complex flows. Multiple-Relaxation-Time (MRT)
scheme, first proposed by
d'Humières (1992), applies different relaxation times to
different kinetic modes and is able to
achieve a better stability and accuracy with a moderate increase
in computational cost (Pan et al.

2006). Other collision models, for example, Two-Relaxation-
Time scheme (Ladd 1994) and
Generalized MRT collision scheme (Shan et al. 2007), have
their own characteristics but are
similar to MRT in terms of ideas and performance.
Single-phase lattice Boltzmann methods have been
successfully applied to simulate flow and
transport in porous media. As in all computational methods,
boundary conditions and flow
driving conditions are very important. In most LB applications,
a uniform external body force is
applied to drive fluid flow in a computational domain bounded
with the periodic boundary
condition. In such systems, the effects of a uniform body force
and a pressure gradient are
essentially equal. However, pressure-driven flows are not
always equal to body-force-driven
flows. It has been argued that they are equal only in flows with
a constant cross-section.
Moreover, the accuracy of body force-approach in a complex
geometry like porous media has
7

not been validated (Zhang and Kwok 2006). Hence, developing
pressure boundary condition or
velocity boundary condition is essential for extending
capabilities of LB method.
Boundary conditions for solid walls (no-slip boundary
condition) of LB method are related to
those from lattice gas automata, e.g., particle bounce-back
scheme. Chen et al. (1996) pointed
out that a simple extrapolation of the distributions on the
boundary can achieve second-order
accuracy. Maier et al. (1996) also proposed a similar
extrapolation method. However, these
methods need distribution’s orientation to the wall and require
special treatments for corner
nodes, which cannot be easily extended to general complex
geometries (Zou and He 1997). As to
pressure and velocity boundaries that often serve as inlet /
outlet conditions to drive fluid flow,
Zou and He (1997) built a pressure and velocity boundary
condition by applying a bounce-back
rule for the non-equilibrium distributions on the boundary
nodes. At inlet and outlet, zero
transverse velocities were assumed. Zhang and Kwok (2006)

proposed a modified periodic
boundary condition to implement a pressure differential. Their
model has superior performance
near inlet and outlet but requires periodicity. Kim and Pitsch
(2007) proposed yet another
boundary closure scheme for both compressible and
incompressible flows and their model shows
a higher accuracy than Zhang and Kwok (2006).
All of the above models have their respective advantages
and weaknesses. It is still an active
topic to design simple, steady and accurate boundary conditions
for lattice Boltzmann
simulations.
2.2 Two-phase LB method
A number of multiphase LB models have been developed for
simulation of multiphase flow,
including the color gradient model (Gunstensen et al. 1991;
Gunstensen and Rothman 1993;
8

Grunau et al. 1993; Liu et al. 2012), the pseudo-potential model
(Shan and Chen 1993; Shan and
Doolen 1995), the free energy model (Swift et al. 1995) and the
mean-field model (He et al.
1999). The first model, the color gradient model (Gunstensen et
al. 1991), was proposed based
on a two-component lattice gas model (Rothman and Keller
1988). Fluid is colored red or blue to
distinguish the two phases. A two-phase perturbation operator is
applied to generate interfacial
tension and a recoloring step forces the particle to move to the
fluids with the same color.
Variations of density and viscosity, though of limited ranges,
were implemented in a subsequent
study (Grunau et al. 1993). To reduce the computation cost of
the recoloring step, d’Ortona et al.
(1995) derived a simplified recoloring method. Latva-Kakko
and Rothman (2005) proposed a
new algorithm for the recoloring step to reduce lattice pinning
by allowing moderate phase
mixing at tangent of interface. In 2007, a new perturbation
operator that correctly recovers
Navier-Stokes equations for two-phase flows and can tackle
different viscosities and density

ratios was developed (Reis and Phillips 2007). In their work, an
explicit expression for the
interfacial tension was derived. Later, Latva-Kakko and
Rothman’s recoloring operator was
adopted to Reis and Phillips’ model, which led to reduced
spurious currents in two-dimensional
computations (Leclaire et al. 2011). TÖlke et al. (2002) built a
three-dimensional 19-velocity
model for immiscible binary fluids with different densities and
viscosity ratios. However, this 3D
model does not have an explicit expression for the interfacial
tension. Recently, Reis and Phillips’
two-dimensional model was extended to three-dimensional by
Liu et al. (2012). Latva-Kakko’s
recoloring operator was also implemented and an explicit
expression for the 3D interfacial
tension is available in Liu et al.’s work (2012). In the color
gradient model, the procedure to
separate different phases is time-consuming and high spurious
velocities can occur near the
interface (Chen and Doolen 1998). One advantage of the model
is that the interfacial tension,

9
contact angle, viscosity ratio and density ratio can all be
adjusted independently (Ahrenholz et al.
2008).
The second multiphase LBM model is the pseudopotential
model that directly describes
interactions between fluid particles using potentials (Shan and
Chen 1993). Shan and Doolen
(1995) then modified the basic model by redefining velocity and
significantly reduced errors in
recovering the macroscopic equation. However, the basic
pseudo-potential model suffers from
some limitations such as large spurious currents, limited range
of density and viscosity ratios,
lack of ability to adjust properties independently, and
thermodynamic inconsistency (Chen et al.
2014). A lot of work has been done to improve the
pseudopotential model. To name a few, it is
shown that interaction forces that have a higher order of
isotropy can suppress spurious currents
(Shan 2006; Sbragaglia et al. 2007). Incorporating a realistic
equation of state (EOS) can reduce
spurious currents and increase the range of density ratios (Yuan

and Schaefer 2006; Zhang and
Tian 2008). Proper forcing scheme also helps to suppress
spurious currents, improve
thermodynamic consistency and increase stability of pseudo-
potential models (Kupersthtokh et al.
2009; Guo et al. 2002). Recently, a critical review (Chen et al.
2014) was conducted, which
systematically listed the limitations and the improvements of
pseudopotential models.
The third model, the free energy model, was developed to
ensure thermodynamic consistency
by introducing a non-ideal pressure tensor directly into the
collision operator (Swift et al. 1995).
Although the original free energy model achieves mass and
momentum conservation, it lacks
Galilean invariance, despite that this problem can be partially
suppressed by adjusting terms in
the pressure tensor (Swift et al. 1996). In 2000, a Galilean
invariant free-energy model was
developed by Inamuro et al. (2000) using asymptotic analysis.
Kalarakis et al. (2002) also
proposed a two-dimensional free-energy model that recovers
Navier-Stokes equations and

10
preserves Galilean invariance. In the literature, comparisons of
the above three multiphase LB
models have been carried out and it was shown that the color
gradient model and the free energy
model have similar capabilities (Huang et al. 2011; Yang and
Boek 2013).
2.3 Boundary conditions for two-phase LB method
For multiphase LB methods, schemes to recover pressure or
velocity boundary conditions are
more difficult to develop than for single-phase LB. In most
publications of multiphase LB, a
periodic boundary condition is applied and flow is driven by a
body force. Such a set up cannot
simulate multiphase flow in experimental systems that are
driven by either pressure or a fixed
injection rate, with non-periodic boundaries. It is therefore
important to derive appropriate
boundary conditions and apply them to properly simulate
multiphase flow. Some studies have
been undertaken, for example, the stress-free boundary

condition (Liu et al. 2012). This method
can ensure mass conservation and improve numerical stability
for creeping flows (Liu et al.
2012). Neumann boundary condition has also been formulated
and tried in simulations of
immiscible displacement (Dong et al. 2010). A systemic study
on outflow boundary conditions
performed by Lou et al. (2013) showed that the convective
boundary condition behaved better
than Neumann and extrapolation boundary conditions. Recently,
Huang et al. (2014) conducted a
study of immiscible displacement using a velocity boundary
condition at inlet and a constant-
pressure boundary condition at outlet based on the non-
equilibrium bounce-back method. Later,
in a simulation of liquid CO2 displacing water, constant-flow-
rate and constant-pressure
boundary conditions were set at the inlet and outlet,
respectively (Liu et al. 2014). With these
developments, multiphase flows that are driven by the rate of
injection can now be properly
simulated.

11
The above brief review shows that currently there is no
established method to simulate
multiphase flow driven by controlled inlet and outlet pressure
differentials using LB model. It is
therefore important to develop a pressure boundary condition to
extend the capability of
multiphase LB methods. In most multiphase LB studies, a
uniform body force is used instead of
a pressure difference. However, a pressure differential cannot
always be replaced by a body force.
For single-phase flow, Zhang and Kwok (2006) showed that
flows driven by pressure differential
and body force are only equal in single-phase flow with
constant cross-sections. Kandhai et al.
(1999) also compared body force and pressure boundary
approaches in single-phase flow. Their
results show that for flows with small Reynolds numbers and
through simple geometries, body
force is equivalent to pressure boundary. However, for flows
with high Reynold numbers or
through complex geometries, a pressure boundary cannot always
be replaced by a body force. In

view of the importance of the pressure boundary and that few
studies have been undertaken to
implement pressure differentials for multiphase LB, in this
work, we will present a method to
implement pressure boundary conditions for multiphase lattice
Boltzmann simulations.
12
CHAPTER 3
SINGLE- AND MULTIPHASE LATTICE BOLTZMANN
METHODS AND
PARALLELIZATION
The section introduces lattice Boltzmann method for single-
phase flow and that for
multiphase flow based on the color gradient model. The parallel
performance of the multiphase
lattice Boltzmann code is tested on a super-computer. The
simulator is then verified by several

validation tests.
3.1 Single-phase LB method and boundary conditions
Single-phase lattice Boltzmann method has been widely
applied to simulate fluid flow and is
the fundamental for multiphase lattice Boltzmann. Here we
briefly introduce the single-phase LB
model as well as several boundary conditions.
3.1.1 Single-phase LB method
The Lattice Boltzmann equation can be regarded as a
discretization of the continuous
Boltzmann equation:
+ �.∇ � + ∇ � = Ω (3.1)
where is the velocity distribution function, � is particle
position. � is particle velocity, is
acceleration, Ω is the collision operator between particles. The
lattice Boltzmann method has two
processes, particle propagation and particle collision. Here,
particle is defined as a collection of
molecules on a certain lattice node. The governing transport
equation for LB method is
� + ∆t, + ∆t = �, + Ω �, (3.2)

13
where �, is the particle velocity distribution function, � is the
position vector of the fluid
node, defined on a cubic lattice, ∆t is the time step, is the
discretized velocity possessed by
molecules in , Ω represents the collision operator in LB
method, �, + Ω �, represents
post-collision populations that propagate to location � + ∆t in
next time step.
In this study, the D3Q19 scheme is used as shown in Figure
3.1. D3Q19 stands for three
dimensions and nineteen discretized velocities, as listed below
= { , , � =± , , , ,± , , , ,± � = ~± ,± , , ± , ,± , ,± ,± � = ~
(3.3)
Figure 3.1 D3Q19 propagation scheme.
Particle collision as denoted by Ω simulates the effect of
viscosity, and is very important for LB
method. Ω must satisfy mass and momentum conservation.
Several collision schemes have been
14

developed and here we apply the Multiple-Relaxation-Time
(MRT) (d'Humieres 2002) collision
method. MRT introduces multiple relaxation parameters to
control the rates of relaxation of
various moments of the particle velocity distribution. MRT has
better stability and accuracy
compared to the simpler BGK collision.
When applying the MRT scheme, a matrix is introduced into
the governing equation
� + ∆t, + ∆t = �, − �− ��[ �, − ] (3.4)
= �[ + .� + .� − � ] (3.5)
where � is a transformation matrix (19 × 19) for the D3Q19
scheme (d'Humières 2002). �− is
the inverse of �. � is the collision matrix in the moment space.
The diagonal collision matrix �
in the moment space is defined as (d'Humières 2002)
� ≡ dia� , , , , , , , , , , , , , , , , , , (3.6)
where are the relaxation parameters chosen as = = �⁄ , = − / − ,
== = �⁄ , = . represents the equilibrium distribution. The
difference between
and is the non-equilibrium distribution . is the lattice sound
speed (√⁄ ) and
is a weighting coefficient associated with discretized velocities:
= { / � =/ � = ~/ � = ~ (3.7)
Fluid density � and fluid velocity � can be obtained from the
first and second moments of the
distribution function

� = ∑ = ∑ (3.8)
�� = ∑ = ∑ (3.9)
15
The viscosity in lattice unit is correlated to the relaxation time
τ (one of the parameters in the
collision matrix �) by
= � τ − . (3.10)
where � is the static fluid density. The pressure is given as
= � (3.11)
After the collision step, the post-collision populations will
either stay at same node or
propagate to neighboring nodes based on Eqn. (3.2). If
neighboring node is solid, a boundary
condition must be applied. Here, we use the simple link-bounce-
back rule that recovers the no-
slip boundary condition (Ladd 1994). In this model, fluid-solid
interface is defined as a
collection of nodes whose positions are halfway along the link
that connects fluid and solid
nodes. The populations of fluids that propagate towards solid
nodes will be reflected back to the

direction where they came. For a moving interface, the link-
bounce-back rule is (Ladd 1994):
− �, + ∆t = �, − � .� � , (3.12)
where �, is the incoming population, − �, + ∆t is the bounce-
back population in the
opposite direction of �, , � = � + ∆t is the location of the
interface node, and � is the
velocity of the interface.
The mass, time and length units in LB simulation are related
to the static fluid density � and
the temporal and spatial lattice units that are designated as ∆t
and ∆x, respectively. A unit
conversion is required to relate simulation units to the actual
dimensions. For example, the
resolution of an image of porous medium is usually the real-
world equivalence of lattice
Boltzmann ∆x (always 1). By comparing the equilibrium density
of the lattice fluid to the actual
16
density, we can establish conversion in the mass unit. Then, we
can take the relation between the
lattice viscosity and the actual viscosity given by the equation
below to determine ∆
( ∆� ∆ ) = ( ∆� ∆ )�� (3.13)

After ∆t and ∆x are determined, other quantities in LB
simulation can be converted by using
dimensionless numbers. For example, lattice velocity, pressure,
and permeability can be
converted to real units by
( ∆∆ ) = ( ∆∆ )�� (3.14)
(� ∆∆ ) = (� ∆∆ ∆ )�� (3.15)
( �∆ ) = ( �∆ )�� (3.16)
where is the velocity, is the body force in simulation, ∆ ⁄ is
the actual pressure gradient, �
is the permeability.
3.1.2 Pressure and Velocity boundary conditions for single-
phase LB method
Several boundary conditions for single-phase LB have been
proposed. Here we choose to
present the pressure and velocity boundary condition derived by
Zou and He (1996), based on
which we developed a pressure boundary condition for
multiphase LB. In the original paper, the
method was presented for D2Q9 and D3Q15 LB models. Here,
we extend their derivations to the
D3Q19 model.
To implement the pressure boundary, non-equilibrium
distributions that are bounced back on

boundary nodes need to be modeled. Take the case of a channel
flow as shown in Figure 3.2 for
an example: flow is in the x direction and solid boundaries are
applied in the y and z directions.
17
At the inlet, after propagation, (i = 1, 3, 4, 5, 6, 7, 8, 9, 10,
11, 14, 15, 18, 19) are known;
(i = 2, 12, 13, 16, 17) are unknown distributions that need to be
determined. Assuming that == on the boundary nodes and the
density at boundary is specified as � , can be
determined from the two equations below:
Figure 3.2 Velocity directions at boundary.
∑ = �= (3.17)
� = + + + + − + + + + (3.18)
Then an explicit expression of is obtained
= � − [ + + + + + + + + + + + + + ]� (3.19)
By applying the bounce-back rule for the non-equilibriums of
the unknown (i = 2, 12, 13, 16,
17), we have

18
= + ( − ) (3.20)
where is the distributions in the opposite direction of . Using
Eqns. (3.5) and (3.20), all
unknown at the boundary nodes can be obtained. In order to
ensure that y- and z-momentum
are zero, (i = 2, 12, 13, 16, 17) are modified (Maier et al. 1996)
= + ( + ) i = 2, 12, 13, 16, 17 (3.21)
Here we choose:
α = / = − + − + − = − + − + −
(3.22)
This modification makes the x-momentum unchanged while y-
and z-momentum zero. Finally,
the unknown distributions at inlet are:
= + �
= + � − − + − + −
= + � + − + − + −
= + � − − + − + −
= + � + − + − + −

(3.23)
The same procedure can be applied to the outlet nodes. Density
at outlet boundary is specified as � . The unknown distributions
at the outlet are:
= − �
= − � − − + − + −
19
= − � + − + − + −
= − � − − + − + −
= − � + − + − + −
(3.24)
Similarly, we can get the unknown distributions when flow is in
y or z direction.
If flow is in the y direction, the unknown distributions at the
inlet are:
= + �
= + � − − + − + −
= + � + − + − + −

= + � − − + − + −
= + � + − + − + −
(3.25)
The unknown distributions at the outlet are:
= − �
= − � − − + − + −
= − � + − + − + −
= − � − − + − + −
= − � + − + − + −
(3.26)
If flow is in the z direction, the unknown distributions at the
inlet are:
20
= + �
= + � − − + − + −
= + � + − + − + −
= + � − − + − + −
= + � + − + − + −

(3.27)
The unknown distributions at the outlet are:
= − �
= − � − − + − + −
= − � + − + − + −
= − � − − + − + −
= − � + − + − + −
(3.28)
There are no special treatments for nodes at the corners.
The velocity boundary can also be derived similarly. Take
the flow in the x direction as an
example. As is known, from Eqns. (3.17) and (3.18), the
expression for density at the inlet is
� = + + + + + + + + + + + + +− (3.29)
Then, by following the same procedure, the unknown
distributions at the inlet can be obtained.

21
3.2 Multiphase LB model
In this work, we use a three-dimensional color gradient LB
model. As in all multiphase LB
models, hydrodynamics is recovered from the evolution of
discrete lattice velocity distribution
functions , where i denotes the index of the discrete velocity
distributions, and superscript k
represents fluid components. We assign the values of R and B to
k to represent the red and blue
components (� and ). In the color gradient model, for every
time step, each component
undergoes a collision operation and a propagation operation,
just like in single-phase LB model
� + ∆ , + ∆ = �, + Ω �, (3.30)
Here, �, is the velocity distribution function of component k at
position � and time t. is
the lattice velocity carried by �, , ∆ is time step. The lattice
velocity � is the D3Q19
model
= { , , � =± , , , ,± , , , ,± � = ~± ,± , , ± , ,± , ,± ,± � = ~
(3.31) Ω is the collision operator that includes three parts
(TÖlke et al. 2002)
Ω = (Ω ) [(Ω ) + (Ω ) ] (3.32)
(Ω ) is the single-component collision operator. When BGK
collision operator is used, it is:

(Ω ) = −� ( − ) (3.33)
where is the equilibrium distribution function of , � is the
dimensionless relaxation
time of component k. Mass and momentum are the first two
moments of the distribution
functions:
� = ∑ = ∑ (3.34)
22
�� = ∑∑ � =∑∑ � (3.35)
where � is the local fluid velocity, � is the density of
component k, � = �� + � is the total
density. In this study, both components are assumed to have
equal density. Then, the equilibrium
distribution functions are given by:
= � ( + [ �.� + �.� − � ]) (3.36)
where the weight coefficient are
= { / � =/ � = ~/ � = ~ (3.37)
The inter-component collision operator (Ω ) is a perturbation
operator defined as (Liu et al.
2012)
(Ω ) = |∇ � |[ .∇ �|∇ � | − ] (3.38)
where the phase field � is expressed by (Liu et al. 2012)

� �, = �� , − � �,�� , + � �, (3.39)
We choose = − / , − = / , − = / as in the work of Liu et al.
(2012). Using
these parameters, the theoretical expression of the interfacial
tension is (Liu et al. 2012)
� = � + � (3.40)
The interfacial tension is therefore controlled by three
parameters �, and relaxation time �.
In this study, we let � = = . The pressure can be obtained via
the equation of state and
given as
= � = � (3.41)
23
The relaxation time � is related to the kinematic viscosity of
the fluid
= (� − ) (3.42)
To accommodate large viscosity ratios, an interpolation scheme
is used to determine the
relaxation time � within an interface (Grunau 1993).
� = {
��, � > � , � > , � − � , � < − (3.43)
where is a free positive parameter that controls the zone of
application of the relaxation
time model. � = + + , = + + , and = =�� / �� + � , = �� − /
, = − / , = − � / , = / .

Because the multi-relaxation-time (MRT) scheme can
enhance numerical instability
(d'Humières 2002), suppress spurious velocities (Liu et al.
2014; Zhao and Fan 2010) and
minimize the viscosity-dependence of permeability (Li et al.
2005), we use MRT collision
operator instead of BGK in this study. As in the work of Liu et
al. (2014), the total distribution
function ( = � + ) is used in the single-component collision and
inter-component
collision steps. With MRT, (Ω ) and (Ω ) become
(Ω ) = − �− �� [ − ] (3.44)
(Ω ) = |∇ � |[ .∇ �|∇ � | − ] (3.45)
where � is an integer transformation tensor (d'Humières 2002).
The diagonal collision matrix �
in the moment space is defined as (d'Humières 2002)
� ≡ dia� , , , , , , , , , , , , , , , , , , (3.46)
24
where are the relaxation parameters chosen as = = �⁄ , = − / − ,
== = �⁄ , = .
The recoloring operator Ω� forces the separation of the two
components. A widely used
recoloring operator is the one proposed by Latva-Kokko and

Rothman (2005):
Ω� � = �� + �� , (3.47)
Ω� = �� − �� , (3.48) is a free parameter between
0 and 1 that controls interface thickness. � is the angle
between
the phase field gradient ∇ � and . To implement a contact
angle, it is assumed that the solid
wall is a mixture of two fluids with constant proportions, which
means that the solid wall has a
certain value of the phase field � . For instance, if the contact
angle of blue phase is , the
density of red and blue components on a solid wall are set as �
− cos and � + cos ,
respectively (Zhang et al. 2015).
3.3 Parallelization of color-gradient LB code
In practical applications, simulating three-dimensional flows
in a complex geometry often
needs millions to billions of lattices and very high
computational cost. A parallelized two-phase
code is therefore of significant value. Parallelization of the
color gradient model is based on the
existing single-phase code (Xiao and Yin 2015). Part of the
parallel schemes were inherited from
the single-phase code. Extensions to the parallel schemes were
built according to the procedures
of the color gradient LB model. In what follows, we present the

main steps of the parallelization
algorithm.
1. Parallelism of the single-phase LB code
25
(1) Build geometry
Read geometry from either experimentally obtained images
of porous media or images
generated by numerical reconstruction. Process the geometry
into a three-dimensional array that
contains (0, 1) (fluid = 1, solid = 0).
(2) Pre-processing
To parallelize the computation, the computation domain (the
three-dimensional array) is
decomposed into a number of subdomains based on the number
of processors to be used: 1, 2,
4, …, 2n, with n being the number of cores. Before flow
simulation, the following pre-processing
steps are needed.
a) Decompose the computation domain as shown in Figure

3.3 and construct a ghost layer
(green) for each subdomain. Ghost layer contains nodes that are
shared between cores and
they are used for communication between cores.
(a)
26
(b)
Figure 3.3 Scheme of decomposition: (a) Computation domain
before decomposition; (b)
Computation domains after decomposition.
b) Each core identifies the fluid nodes in its own ghost
layer.
c) Each core packs its ghost nodes into an array that contains
the coordinates of the nodes and
19 indices to ghost nodes’ neighbors.
d) Each core sends the list of ghost nodes to its neighbors.

e) With ghost layers built, every core constructs its own
local lattice Boltzmann node map.
(3) Single-phase LB flow simulation
a) Each core reads input parameters and the local node map.
b) Each core performs collision on every node in the local
node map.
c) Each core performs propagation. During a propagation
step, if the neighbor is local,
propagation is performed within the local subdomain. If the
neighbor is in the ghost layer,
communication is needed. The distribution to be propagation
therefore enters a specific
27
package based on its destination. MPI (Message Passing
Interface) is used to communicate
these packages when all of them are ready.
d) A master process collects the information from every core
to calculate the average
properties, such as saturation and superficial velocity.
e) Back to b) till all time steps are completed.

2. Parallelism of the color-gradient LB code
The parallel implementation of the two-phase code is an
extension of the parallel
implementation of the single-phase code. In the two-phase code,
there are two components. Each
component has its own velocity distribution for collision and
propagation.
(1) Build geometry
Same as the single-phase code.
(2) Pre-processing
Decompose the computational domain and identify fluid or
solid in the same way as the
single-phase code.
(3) Two-phase LB flow simulation
a) Each core reads input parameters and the local node map.
b) Each core calculates the local density of each component
and transfer the density of the
fluids nodes that on the sub-domain surface to a density
package for communication.
c) All cores exchange density packages by MPI and store
received packages to the local ghost

layer such that every fluid node in the local computational
domain can access all of its
neighbors’ densities.
d) Perform two-phase collision steps shown in Eqn. (3.38).
28
e) Propagate the distributions of each component like the
single-phase code.
f) A master process calculates the average properties of the
entire computation domain.
The performance of the parallelized two-phase LB simulator
was tested by simulating
immiscible displacements in a porous medium (Figure (3.4)).
Three different size of geometries
were used that contain 6403, 3203 and 1603 nodes respectively.
Figure 3.4 Porous medium used for parallel performance test.
Tests were performed on BlueM, which is a unique high-
performance computing system at

Colorado School of Mines (built by IBM) that contains two
partitions. One is named Mc2
29
(Energy) which runs on an IBM BlueGene Q and another is
named Aun (Golden) that uses the
iDataplex architecture. We only used Mc2 in the performance
test for consistency. Mc2 contains
512 nodes and has 8192 Gbytes of total memory. Each node has
16 cores (PowerPC A2 17 Core).
The performance of the two-phase code is evaluated according
to the following benchmarks:
(1) Speedup
Speedup SP is defined as:
SP = TTN (3.49)
where T is the simulation time using one core, TN is the
simulation time using N cores. The
speedup of different geometries using different number of cores
are shown in Figure 3.5.
Figure 3.5 Speedup of multiphase lattice Boltzmann code.

30
When comparing speedup in the different geometries, it can be
observed that speedup is higher
in the larger system. For larger systems, more computational
work is assigned to each core. The
fraction of communication overhead therefore is reduced,
leading to a higher speedup and a
higher parallel efficiency as we shall observe below. Though
not tested, we expect that the
speedup will continue to increase with increasing system size.
(2) Parallel Efficiency
Parallel efficiency is defined as the ratio of the speed up to
the number of cores:
EN = SpN (3.50)
From Figure 3.6, we can see that parallel efficiency of the two-
phase code decreases with
increasing number of cores used, owning to the increasing ratio
of computational load assigned
to each core versus the communication overhead.
Figure 3.6 Parallel efficiency of multiphase lattice Boltzmann

code.
31
As the size of geometry increases, the decreasing rate of
parallel efficiency slows down.
To summarize, a parallel LB simulator for two-phase flow in
porous media was built based on
the existing single-phase LB simulator. Parallel efficiency is
satisfactory. As the size of the
geometry increases, the speedup increases and the parallel
efficiency is higher. In these tests,
decomposition was done by consecutive bisections in three
directions. However, computational
cost in each subdomain is controlled by the actual number of
fluid nodes rather than the
subdomain size. Thus, load balance is not equal, which is an
issue that negatively affects the
parallel performance especially for flows through highly
heterogeneous porous media geometries.
3.4 Test cases
Several test cases are performed in this part to verify the

developed two-phase color gradient
LB simulator. When a drop of one fluid is immersed into
another, surface tension will make the
drop relax toward its equilibrium shape – a sphere or a circle.
As shown in Figure 3.7, an initial
square red drop ( × ) is centered in a × computation domain,
surrounded by the
blue fluid. The parameters in this test are �� = � = , �� = � =
, � = . × − . The
initial configuration is shown in the top left and different
snapshots are taken at different
simulation times. As the snapshots show, the square drop
deforms gradually under the influence
of the interfacial tension and becomes circular in the end.
Another test case characterizes the capillary pressure within
a stationary bubble. A spherical
red bubble is initialized at the center of a × × computation
domain, surrounded by the
blue fluid. According to the Young-Laplace equation, the
pressure difference ∆ across the
interface of a spherical bubble is related to the interfacial
tension by:
∆ = � (3.51)
32

where is the radius of bubble and � is the interfacial tension.
In our simulations, different
pressure differences can be obtained by changing the bubble
radius.
Figure 3.7 Oscillation and relaxation of a square drop to
equilibrium spherical shape.
As exemplified in Figure 3.8, the slope of ∆ vs. ⁄ is the
simulated interfacial tension.
Theoretically, the interfacial tension of the color model is given
as � = � + �. Without
33
losing generality, in this study, it is always assumed that � = =
. Table 3.1 and Table 3.2
give a comparison between simulated and theoretical results. It
can be seen that simulated
interfacial tensions agree well with theoretical results for both =
and = . cases. The
errors for the interfacial tension, calculated by = |� ℎ − � | � ℎ⁄
× %),
increase with A. = generates more accurate interfacial tensions
than = . . In many
multiphase LB models, spurious current, which refers to non-

zero, non-physical fluid velocities
in the vicinity of an interface, is an inevitable problem (Yuan
and Schaefer 2006). In this model,
a lower spurious current is obtained (|�| is the maximum
spurious current) when = . . At
the same time, however, affects the thickness of interface.
Figure 3.9 gives the density
distribution along a line that passes through the center of the
bubble. When = . , the
interfacial thickness is six lattice units. When = . , the thickness
decreases to five lattice
units. A lower therefore increases the thickness of the interface
and suppresses the spurious
current, however at the expense of the accuracy of the
interfacial tension.
Figure 3.8 Numerical measurement of Laplace law.
34
Table 3.1 Stationary bubble test. �� = � = , �� = � = , =
� � ℎ % |�|
1.00E-04 4.510E-05 4.44E-05 1.47% 2.79E-05
1.00E-03 4.515E-04 4.44E-04 1.59% 5.08E-05

1.00E-02 4.520E-03 4.44E-03 1.71% 2.02E-04
1.00E-01 4.537E-02 4.44E-02 2.08% 1.75E-03
Table 3.2 Stationary bubble test. �� = � = , �� = � = , = .
� � ℎ % |�|
1.00E-04 4.579E-05 4.44E-05 3.03% 2.27E-06
1.00E-03 4.579E-04 4.44E-04 3.03% 1.76E-05
1.00E-02 4.581E-03 4.44E-03 3.06% 1.00E-04
1.00E-01 4.597E-02 4.44E-02 3.43% 9.76E-04
Figure 3.9 Thickness of interface using color-gradient lattice
Boltzmann model.
35
One important ability of the color gradient LB method is
that it can produce controlled static
contact angle. As shown in Figure 3.10, the contact angle is
determined by the force balance at
the three-phase contact line.

Figure 3.10 Force balance at the contact line.
In the color gradient model, contact angle is achieved by setting
� and � on the wall nodes.
The contact angle can be analytically determined by (Latva-
Kakko and Rothman 2005):
= (� − �� ) (3.52)
where � is the initial density. Figure 3.11 shows the different
contact angles achieved by this
approach. A full range of contact angles can be achieved from
wetting to non-wetting, although
those at the extremes are difficult to simulate accurately,
owning to the finite thickness of the
fluid-fluid interface.
36
Figure 3.11 Contact angles achieved by setting � and � on
wall nodes.
To examine whether this multiphase LB model can properly
simulate the dynamics of two-

phase flow in the absence of solid walls, a Taylor deformation
test was performed. As shown in
Figure 3.12, when a spherical drop is placed in the middle of a
linear shear flow, viscous effect
tends to deform the drop, and interfacial tension acts to
maintain drop’s spherical shape. In this
test, two parallel plates move in the opposite direction to
generate a linear shear flow, in the
37
middle of which a spherical drop is placed. The deformation of
the drop is described as a
function of the Reynolds number and the capillary number. The
definitions of the Reynold
number and the capillary number are (Liu et al. 2012)
= �, = � (3.53)
where = ⁄ is shear flow rate. is the half of the distance between
two plates. is the
velocity of moving plate. R is the initial radius of droplet which
is 10 in lattice unit. = ρν is the
dynamic viscosity. For this case, the two fluid has same density
and viscosity. At the steady state,
the deformation of droplet is described using a deformation
parameter defined as

= −+ (3.54)
where a and b are the lengths of major and minor axes,
respectively. When flow is Stokes,
should obey the Taylor relation when is low (Taylor 1932)
= ⁄ (3.55)
Simulations were performed at different capillary numbers when
= . in a × ×
system. The effect of was also tested.
Figure 3.12 Droplet deformation in Stokes flow.
38
Figure 3.13 plots the deformation parameter as a function of
capillary number. It is can be
seen that the simulation results are in reasonable agreement
with the analytical solution. The
agreement is better when = . .
Figure 3.13 Deformation parameter as a function of ; at = 0.1.

39
CHAPTER 4
BOUNDARY CONDITIONS FOR MULTIPHASE LATTICE
BOLTZMANN
For lattice Boltzmann methods, it is not straightforward to
implement open boundary
conditions because, during propagation, distributions from the
outside of the computational
domain must be speculated. For multiphase LB method,
boundary conditions are not well
developed, because the distributions of both components, and
sometimes their interactions, must
be modeled at the inlet and outlet boundaries. In this section,
we study several boundary
conditions for multiphase LB method. First, we evaluate the
performance of an outflow
boundary and a modified periodic boundary for the color
gradient lattice Boltzmann model. Then,
a pressure boundary condition that can be applied to simulate
pressure-driven flows using

multiphase lattice Boltzmann is derived. As validations,
pressure-driven flows through a
capillary tube with different density ratios, viscosity ratios and
contact angles are tested.
4.1 Evaluation of existing outflow boundary conditions for two-
phase LB method
Several outflow boundary conditions have been tested
systematically by Lou et al. (2013)
using a mean-field multiphase LB model (He et al. 1999). Here,
we implement the convective
boundary condition, which behaves best among the three
outflow boundary conditions tested
(Lou et al. 2013), into the color-gradient multiphase LB model.
As shown in Figure 4.1, the
outlet boundary is located at = . Nodes at + and + are the ghost
nodes needed for
two-phase interactions at the boundary. Nodes at − and − are
the two nearest layers of
real fluid nodes before the outlet boundary.
40

Figure 4.1 Schematic of outlet boundary.
For the convective boundary condition, distributions at the
outlet boundary obey the
following equation
∂∂t + ∂ = (4.1)
where is the phase velocity normal to the outlet boundary (Lou
et al. 2013). Here, we choose
as the average velocity at layer − .
Using first-order finite difference, the convective boundary
condition can be written as
� , + ∆ − ,∆ + � , + ∆ − � − , + ∆∆ = , + ∆ − ,∆ + , + ∆ − − , +
∆∆ =
(4.2)
In lattice Boltzmann units, ∆ = , ∆ = , we therefore have
� , + = � , + � − , ++
, + = , + − , ++
(4.3)
41
Then the distribution of outlet boundary nodes , + is known.

For the two-phase
interaction at boundary, distributions at + are also set
according to the convective boundary
condition
� + , + = � + , + � , ++
+ , + = + , + , ++
(4.4)
To evaluate the performance of the convective boundary
condition, two tests were conducted
in a rectangular computational domain with length = , width �
= and height = .
Two pairs of solid walls are placed in the direction of and �,
outflow boundaries are applied
in the direction of . At the inlet, velocity boundary condition is
applied for both phases. In this
simulation, the parameters are �� = � = , �� = � = , � = . × −
, contact angle =°.
The first test is a moving drop in a channel flow. A red drop
initialized with radius =
was placed at ( , , = , , and surrounded by another blue fluid.
Density and
viscosity are the same for the two phases. The blue fluid is
injected continuously with a constant
rate. Performance of the outflow boundary condition is
evaluated by checking whether the drop
can leave the outlet boundary smoothly without significant
disturbance to the upstream flow nor

to the shape of the drop. Figure 4.2 shows the simulated exit of
the drop from the outlet using the
outflow boundary condition. When the red drop reaches the
outlet, it keeps its shape and move
out of the system freely without any unphysical phenomenon.
This result is consistent with the
report of Lou et al. (2013).
In the second test, we simulated the immiscible displacement
in the same computational
domain which is an important topic in many practical problems.
Initially, section is
42
occupied by the blue phase and the rest of the channel is
occupied by the red phase. Density and
viscosity are set as unity for both fluids. Also, the blue phase is
injected by a constant rate from
the inlet. Figure 4.3 shows the time evolution of the interface
during the immiscible displacement.
Similar to the moving drop case, the outflow boundary allows
the fluids move out of the domain
freely.

Based on the moving drop case and immiscible displacement
case, we can conclude that the
convective outflow boundary condition is implemented to the
color-gradient lattice Boltzmann
model successfully.
using the convective outflow
boundary condition.
43
channel using the convective
outflow boundary condition.
4.2 Modified periodic boundary condition for two-phase LB
method
In addition to the convective outflow boundary condition,
we also tested a modified periodic
boundary condition, designed to combine periodic and outflow
boundaries.

The idea of this modified periodic boundary condition is, at
the outlet, the convective
boundary condition is applied which has performed well in the
previous tests. At the inlet, to
keep the periodicity, fluid is allowed to re-enter from the outlet,
yet the color is changed to that
of the displacing fluid. In this way, the velocity field is forced
to be continuous between the inlet
and the outlet. Then, both fluids are driven by a uniform body
force applied to the entire
computation domain. In the simulations, the parameters are set
as �� = � = , �� = � = ,
44
� = . × − , contact angle = °. With this setup, we conducted the
same test of a drop
moving through a channel. The snapshots in Figure 4.4 show
that the drop can also move out of
the outlet with minimum disturbance to the upstream flow field
and to the shape of the drop. The
same test of immiscible displacement in a channel is also
conducted. As shown in Figure 4.5, the

two phases can also move out of the domain freely.
As a conclusion, the modified periodic boundary condition
works well in the color-gradient
lattice Boltzmann model. This boundary condition should be
very useful for simulations of
gravity-driven multiphase flow.
using the modified periodic
boundary condition.
45
channel using modified periodic
boundary condition.
4.3 Pressure boundary condition for two-phase LB method
At the end of this chapter, we present the derivation of
pressure inlet and outlet boundary
conditions for two-phase LB simulations, developed based on
the single-phase pressure

boundary condition proposed by Zou and He (1997). In their
work, the pressure boundary is
applied by letting the non-equilibrium part of distributions
bounce back from the boundary nodes.
In order to set up the pressure boundary condition for two-phase
flow, the non-equilibrium part
of the total distribution function is bounced back. Then, the
unknown distributions of each phase
are determined using its own density defined at the boundary.
46
Let us take the case of channel flow shown in Figure 4.6 as
an example. The boundary is
aligned with the x direction. Inlet and outlet boundary is located
at = (I for inlet) and =
(O for outlet) respectively. Nodes at = − and = + are the ghost
nodes that are
needed for two-phase interactions at the boundary. Nodes at x =
+ and x = − are the
nearest layers of real fluid nodes to the inlet and outlet. At the
inlet boundary = , a constant
density � is set.
Figure 4.6 Schematic of inlet and outlet boundaries.

At the inlet, � (i = 2, 12, 13, 16, 17), (i = 2, 12, 13, 16, 17)
are the unknown distributions
that need to be determined. Since we have
� = ∑∑ (4.5)
� = ∑( + + + + ) − ( + + + + ) (4.6)
the inlet velocity can be obtained as
47
� = � − ∑[ + + + + + + + +
+ ( + + + + )] (4.7)
Then, we use the bounce-back rule for the none-equilibrium part
of the total distribution function
(i = 2, 12, 13, 16, 17):
= + ( − ) (4.8)
where is the distribution function at the opposite direction of ,
= ∑ , is the total
equilibrium function. By using the definitions of the
equilibrium distribution functions in Eqn.
(3.36), for each component, we have
, − , = � ,� = (4.9)
, − , = � ,� = , , , (4.10)

Then by Eqns. (4.8) and (4.10), we can get
= + � ,� = (4.11)
= + � ,� = , , , (4.12)
In order to make sure that y- and z-momentum are zero, (i = 2,
12, 13, 16, 17) need to be
modified
= + ( + ),� = , , , (4.13)
This modification adds , to the y-, z-momenta and keep the x-
momentum unchanged. Here
we choose:
α = /
= ∑ − + − + − (4.14)
48
= ∑ − + − + −
Finally, the unknown total distributions at the inlet are:
= + �
= + � − ∑ − + − + −
= + � + ∑ − + − + −
= + � − ∑ − + − + −
= + � + ∑ − + − + −

(4.15)
Next, we need to divide to obtain the unknown distribution
function for each component.
Although the total density at the inlet � is defined, the
densities of component R and
component B (�� and � ) are still not known. To solve this
problem, we divide the total
distribution according to �� and � of the previous time
step:
� = �� + � (4.16)
= �� + � (4.17)
When calculating phase field gradient in the inter-component
collision and recoloring step at
boundary nodes, the information of ghost nodes is needed. We
assume the distribution of ghost
nodes at = − is the same as that of inlet boundary nodes: − = .
Similarly, at the outlet, by setting a constant density � at
outlet boundary nodes and
following the same procedure, the unknown distributions at the
outlet are:
49
= − �

= − � − ∑ − + − + −
= − � + ∑ − + − + −
= − � − ∑ − + − + −
= − � + ∑ − + − + −
(4.18)
These distributions are then divided to obtain � and according
to the densities of component
R and component B at the outlet (�� and � ) of the previous
time step.
� = �� + � (4.19)
= �� + � (4.20)
As at the inlet boundary, we assume the distribution of ghost
nodes at = + is the same as
that of outlet boundary nodes: + = .
To check the performance and accuracy of the pressure
boundary condition, three test cases
are carried out. The first two cases show the effect of the
pressure boundary condition on two-
phase flow. The third case is the simulation of capillary filling,
which is a problem with a
moving contact line that admits an analytical solution.
A. Effect of pressure boundary condition on two-phase flows

In this section, we test the pressure boundary condition to
see whether it generates two-phase
flow that are physically correct. The first test is performed in a
rectangular computational
50
domain with length = , width � = and height = . Two pairs of
solid walls are
placed in the direction of and �, pressure boundaries are
applied in the direction of . Initially,
the domain is filled with the blue phase, except that a red drop
with radius = is placed at
( , , = , , . By making � slightly above the equilibrium value of
unity and �
slightly below, a pressure difference is established between the
inlet and outlet, and fluid flow is
generated. The steady-state flow is characterized by the
Reynolds number Re and the capillary
number Ca that defined as
R� = ,Ca = � (4.21)
where is the velocity of blue phase along the flow direction. =
� is the dynamic viscosity
of the blue phase. � is the interfacial tension. In this
simulation, the parameters are �� = � = , �� = � = , � = . × −
, � = . , � = . , R� = . , Ca = . , contact
angle = °.
Figure 4.7(a) shows the transient motion of the two phases at

different simulation times. The
corresponding pressure fields are presented in Figure 4.7(b). As
the figures show, the fluids are
moved by the imposed pressure difference. When the drop
approaches the outlet, the interface
becomes perpendicular to the outlet boundary. Compared to
previously tested convective
boundary condition, the constant-pressure outlet does affect the
shape of the drop. However, this
deformation is reasonable because, based on the Laplace law,
pressure difference across a curved
interface is ∆ = � ⁄ + ⁄ where and are the local radii of
curvature. At the outlet
boundary, the pressure is constant, which means that the
pressure difference must be zero across
the interface. The radii of and , therefore, must be infinite. The
behavior of the drop is thus
consistent with the constraint imposed by the boundary. Plots of
the pressure fields in Figure
4.7(b) also demonstrate that a constant pressure difference is
maintained at all times. The
51

pressure inside the drop is higher than that outside because of
the interfacial tension. After the
drop moves out of the domain, the pressure field reverts to that
of the single-phase case.
Figure 4.7 Phase distribution (a) and corresponding
dimensionless pressure field (b) of a moving
droplet in a channel using the pressure boundary condition. The
dimensionless pressure is
defined as +⁄ .
52
The second test is an immiscible displacement, conducted in
the same computational domain
as that used in the moving drop test. Initially, the section is
occupied by the blue
phase; the rest of the channel is occupied by the red phase.
Fluid flow is driven by a constant
pressure difference. As in pressure-driven flows, varies with
time if viscosities of two phases
are not equal, we use the and when the computational domain
is fully saturated with the

defending fluid to compute R� and Ca . The parameters for the
simulation are: �� = � =
, �� = � = , � = . × − , � = . , � = . , R� = . , Ca = . ,
contact angle = .
Figure 4.8(a) shows the evolution of the interface in the two-
phase displacement test. Similar
to the moving drop case shown in Figure 4.7(a), as the red
phase leaves the computational
domain, the interface becomes perpendicular to the outlet
boundary. Pressure fields presented in
Figure 4.8(b) show that a constant pressure difference is also
maintained. Due to the curved
interface, the pressure in the domain also varies along the cross-
flow direction under the effect of
capillary pressure.
To further test the performance of the pressure boundary
condition, we varied the viscosity
ratio (ratio of kinematic viscosity between the defending and
invading fluids) and capillary
number Ca that are the two most important factors determining
the flow patterns (Lenormand et
al. 1988). Figure 4.8 also shows the simulation results with =
and Ca = . , while keeping
other parameters unchanged. In both cases, two-phase flows
generated are physically correct,

which demonstrates that the pressure boundary can handle a
range of viscosity ratios and
capillary numbers.
53
Figure 4.8 Phase distribution (a) and corresponding pressure
field (b) of two-phase displacement
in a channel using the pressure boundary condition. The
dimensionless pressure is defined as +⁄ .
54
B. Capillary filling
Displacement of a wetting fluid with a non-wetting fluid in a
capillary tube with a constant
pressure differential, as shown in Figure 4.9, is a problem with
a moving contact line that admits
an analytical solution. For a capillary tube that has a diameter ,
when one fluid (red fluid) is

injected to displace another fluid (blue fluid) under a constant
pressure differential, assuming that
the process is quasi-steady and there is no film drainage of the
wetting fluid, it is straightforward
to derive that the immiscible displacement can be described by
the analytical expression below:
� = − − + √ − − ∙ − + �− ∙
(4.22)
where � is the length occupied by the red phase, � and are,
respectively, the viscosity of
the red and blue phases, and are the inlet and outlet pressures,
� is the interfacial tension,
is the dynamic contact angle (Pooley et al. 2009). Based on this
equation, the breakthrough
time ( �⁄ = ) is
= ∙ � +− + � (4.23)
To compare the numerical results from simulation using
pressure boundary condition to the
analytical solution from Eqns. (4.22) and (4.23), a capillary
tube that has length = and
diameter = (lattice spacing) is set up. Initially, section . is
occupied by the
red phase, and the rest of the channel is occupied by the blue
phase. Displacement simulations
were run with the following parameters: �� = � = , = × − . To

study the effects of
viscosity ratio and wettability, four sets of simulations were
conducted: (a) = , � = . , � = . , = − . ; (b) = , � = . , � = . , = .
; (c) = . ,
55
� = . , � = . , = − . ; (d) = , � = . , � = . , = − . .
Figure 4.9 Simulation setup for capillary filling. The capillary
tube has a length L and diameter d.
The red fluid is the invading fluid and the blue fluid is the
defending fluid.
Figure 4.10 compares the simulation results and the
analytical solution from Eqn. (4.22). It
can be seen that, in the three drainage cases (case a, c, d), our
simulation results are consistent
with the analytical solutions for all the different viscosity ratios
/ pressure differences. However,
in the imbibition case, the interface in the simulation moves
slightly slower than the theoretical
result. The pressure distribution along the flow direction is
shown in Figure 4.11. In all four

cases, a constant pressure difference between the inlet and the
outlet is maintained and the
pressure gradient within each phase reflects different viscosities
correctly.
The breakthrough times from the simulations are compared
to the analytical solutions from
Eqn. (4.23) in Table 4.1. Again, in all the three drainage cases
(case a, c, d), the simulation
results are in good agreements with analytical results, while
there is about 8% error in the
imbibition case.
From the capillary filling tests, the accuracy of the pressure
boundary condition is verified
and the simulation results admit analytical solutions.
56
Figure 4.10 The length of tube occupied by the filling fluid as a

function of time ∗ = / for (a)
drainage, M = 1 (b) imbibition M = 1 (c) drainage M = 0.1 and
(d) drainage M = 10. The lines
represent the analytical solution based on Eqn. (4.22) and the
crosses mark the position (1, 1).
57
Figure 4.11 Pressure distributions along the flow direction for
(a) drainage, M = 1 (b) imbibition
M = 1 (c) drainage M = 0.1 and (d) drainage M = 10 at ∗ = / =
0.55, 0.55, 0.50, 0.50
respectively. P* is the dimensionless pressure defined as +⁄ .
58
Table 4.1 Comparison of breakthrough times in the case of
capillary filling
Type , ℎ , %

drainage 1 7.24E+5 7.2E+5 −0.55%
imbibition 1 7.20E+5 7.8E+5 8.3%
drainage 0.1 7.98E+5 7.7E+5 −3.5%
drainage 10 7.98E+5 7.9E+5 −1.0% % = ( , − , ℎ )/ , ℎ × %
59
CHAPTER 5
SIMULATION OF IMMISCIBLE DISPLACEMENTS IN A
POROUS MEDIUM
The objective of this chapter is to apply the developed
parallel multiphase LB simulator with
the color gradient model and the developed pressure boundary
condition for multiphase flow to
simulate a few meaningful cases of immiscible displacements in
a porous medium made up by
packed spheres.

The sphere packing is constructed by using a DEM (Discrete
Element Modeling) method to
fill spherical particles into a cm × cm × cm cubic domain
(Petunin et al. 2013). In the
sphere packing, the particle diameter obeys a Gaussian
distribution and varies from 1.21 mm to
5.97 mm; the number-averaged diameter is 3.34 mm. In
simulation, the sphere packing is
discretized into a × × voxel domain using a resolution of 0.2
mm. A column is
then cut out from the cubic sphere packing. The length of the
column L is 320, and the diameter
of the column d is 100. The slice of = is set as the inlet and the
slice of = is set as the
outlet. The column (Figure 5.1) is then sealed on the side using
solid no-slip walls to only allow
axial flow. To uniformly distribute the pressure and eliminate
the effect of solid phase at the
boundaries, we added ten layers of fluid nodes to the inlet and
the outlet. The porosity, �, of the
column is 22.8%.
Immiscible two-phase flows on the Darcy level are generally
described by the multiphase
extension of Darcy’s law
= �� , ∇ , � = , (5.1)

60
where � is the intrinsic permeability, � , is the relative
permeability of fluid i evaluated at a
particular wetting phase saturation , is the superficial velocity
of fluid i through the porous
medium, ∇ is the pressure gradient of fluid i and subscripts w
and nw denote wetting and non-
wetting phases, respectively. In this study, the relative
permeability is calculated by
� , = = ∇ =∇ , � = , (5.2)
With direct numerical simulation, the pressure gradients of both
phases can be obtained, which is
a distinctive advantage over experiments where phase pressures
are difficult, if not impossible, to
obtain.
Figure 5.1 The cylindrical porous medium made up by packed
spheres.
61

We first apply the pressure boundary to calculate the
intrinsic permeability � by setting the
sphere packs saturated with only one phase. We note that with
MRT, the absolute permeability
obtained, . × − m , shows little viscosity dependence.
To simulate immiscible displacement, initially, the entire
column (> ) is saturated by one
fluid (defending fluid) and the space before the porous column (
) is filled with the
other fluid (invading fluid). The invading fluid is then pushed
into the porous column by the
higher pressure to the inlet. At the outlet, a lower pressure is
applied. The densities of both
phases were set at 1. The kinematic viscosity of the defending
fluid was fixed at 0.167, and the
kinematic viscosity of the invading fluid was varied. By varying
the pressure difference relative
to the viscosity and the interfacial tension of the fluids, flows
with different capillary numbers
can be obtained. Results of four cases with different capillary
numbers, viscosity ratios, and
wettability, shown in Table 5.1, are reported in this thesis. Note
that in all cases, the density ratio
was kept at unity. As is evidenced in Eqns. (4.22) and (4.23),
when the Reynolds number is low,

the density ratio does not play a role in the dynamics of
displacement. We run all the cases until
a steady saturation is reached in the end. The simulation time
for each case is around 70 hours by
using 256 cores. The parallel performance of the code can be
found in Chapter 3.
Table 5.1 Parameters in different cases of immiscible
displacement
Case ∆P A cosθ Ca
A 0.0533 1E−4 −0.5 1 1.5E−1
B 0.0533 1E−4 −0.5 10 2.0E−1
C 0.0533 1E−1 −0.5 1 1.5E−4
D 0.0533 1E−1 0.5 1 1.5E−4
62
To clearly visualize the immiscible displacement inside the
porous medium, Figure 5.2
through Figure 5.5 show the distributions of the two phases and
corresponding pressure at the

slice cutting through the center of the column at different times
∗ for Case A, B, C, D
respectively. Here the dimensionless time ∗ is defined as the
ratio between the simulation time
and the residence time of the defending fluid in single-phase
flow:
∗ = ��∆ (5.3)
In Figure 5.2(a) through Figure 5.5(a), the red color represents
the invading fluid, and the yellow
color represents the defending fluid. The blue color represents
the solid phase. Figure 5.2(b)
through Figure 5.5(b) show the wetting phase pressure
distributions along the column, also at the
center slice at different times. It can be observed that the
defending fluid is displaced by the
invading fluid under a constant pressure difference between
inlet and outlet. The results
demonstrate that the pressure boundary condition can be applied
to simulate multiphase flow in
complex porous media.
In an attempt to quantify the displacement efficiency along
the axis of the column, we divide
the column equally into ten sections along the flow direction. In
every section, the saturations
and pressures of both phases are calculated. Figure 5.6 shows
the saturation distribution along

the column at different times for the four cases. Pressure
distributions along the flow direction
are presented in Figure 5.7. Figure 5.8 shows the relative
permeability curves using Eqn. (5.2)
for Case A through D.
By comparing the simulation results from different cases, we
can illustrate 1) the effect of
capillary number, 2) the effect of viscosity ratio, and 3) the
effect of wettability on two-phase
displacement.
63
packing at different times (∗ = 0.12, 0.47, 0.71, 1.18), Case A.
64

packing at different times (∗ = 0.12, 0.47, 0.71, 1.18), Case B.
65
packing at different times (∗ = 0.12, 0.47, 0.71, 1.18), Case C.
66

packing at different times (∗ = 0.12, 0.47, 0.71, 1.18), Case D.
67
Figure 5.6 Saturation distributions at different times in Case A,
B, C, and D. Time interval ∆ ∗
between consecutive lines is 0.059. The red line represents the
saturation distribution when
breakthrough occurs.
68
Figure 5.7 Phase pressure distribution of Case A, B, C, and D.
The open symbols represent
pressure profiles at times before breakthrough: ∗ = 0.89, 0.47,
0.30, 0.12 respectively. The lines

represent the pressure profiles after breakthrough: ∗ = 1.18 for
Case C and D.
69
Figure 5.8 Relative permeability vs. wetting phase saturation of
the sphere packing. Red and
green points denote breakthrough points. Time interval ∆ ∗
between consecutive points is 0.059.
1. Effect of capillary number
To evaluate the effect of the capillary number, Case A and
Case C were conducted with Ca =. and Ca = . × − ,
respectively. Case A represents situations with ultra-low
interfacial
tensions that are generally achieved with surfactants or with
nearly miscible fluids such as crude
oil and supercritical CO2 at high pressures. Case C represents
situations where the two fluids are
70
immiscible and the interfacial tension is high, e.g., a water-oil

displacement. The static contact
angle is °. Case A and Case C therefore represent typical
drainage processes.
Figure 5.6 shows that as time progresses, the saturation of
invading non-wetting phase
increases and a steady saturation profile is reached in the end
for both Case A and Case C. As the
porous medium in our simulation is homogeneous, the
displacement front takes a Buckley-
Leverett form (Buckley and Leverett 1942) and displacement
efficiency is high in both cases. An
obvious earlier breakthrough is observed in the high Ca case
(Case A, breakthrough when ∗ =. , = . ) compared to the low Ca
case (Case C, breakthrough when ∗ = . , =. ). The increased
breakthrough time for the low Ca case is the result of
significant adverse
capillary pressure. From Figure 5.7(a) and Figure 5.7(c), the
pressures of the two phase are
nearly identical when Ca = . , but are very different when Ca = .
× − . Another notable
observation is the difference between the pressure gradients of
the two phases. When Ca is low,
at the front of displacement, the non-wetting phase first
displaces the wetting phase in larger
pores where the capillary pressure is insignificant. The
pressures of the non-wetting phase and
the wetting phase are therefore nearly identical at the front of
displacement. In the region behind

the front, the non-wetting phase can enter smaller pores than at
the front due to flow pressure;
this leads to a higher difference between the pressures of the
non-wetting and wetting phases.
After breakthrough, the pressure gradients of wetting and non-
wetting phase become nearly
identical with a nearly constant capillary pressure difference.
The relative permeability curves of
the two phases are shown in Figure 5.8(a). We noticed that the
relative permeability of both
phases decrease with decreasing capillary number. When the
capillary number is small, flow is
blocked in some small pores, which decreases the relative
permeability of both invading and
71
defending fluids. Similar reductions have also been observed in
other numerical and
experimental studies (Li et al. 2005; Blunt and King 1991;
Avraam and Payatakes 1999).
2. Effect of viscosity ratio

By comparing Case A and Case B, we can observe the effect
of viscosity ratio on the
dynamics of multiphase flow. To eliminate the effect of the
capillary number, both Case A and
Case B were assigned with a high capillary number so that the
flow is dominated by viscous
forces. When ≠ , the interfacial tension can be obtained using
Laplace’s law by a bubble test.
In Case B, when = and = − , the interfacial tension expressed in
lattice unit is . ×
− .
According to the saturation profiles in Figure 5.6, in Case B,
when the viscosity of the
invading fluid is much lower than the viscosity of the defending
fluid, there is a much quicker
breakthrough (breakthrough when ∗ = . , = . ) than Case A
(breakthrough when ∗ =. , = . ). At the point of breakthrough,
significant amount of wetting fluid is left behind.
This phenomenon is known as viscous fingering (Saffman and
Taylor 1958; Homsy 1987). The
pressure profile in Figure 5.7(b) concaves up, which is another
indication that the less viscous
fluid has developed viscous fingers into the more viscous fluid.
In Figure 5.8(b), the shapes of
the relative permeability curves at high M also shows signs of
unstable and fingered

displacement. Compared with Case A, the less viscous non-
wetting fluid’s relative permeability
concaves down because it breaks into a more viscous fluid, and
therefore, has a slower speed
than expected. The more viscous wetting fluid’s relative
permeability concaves up, for the same
reason, as it is being penetrated by the front of the less viscous
fluid. This trend is in agreement
with results from pore network modeling by Blunt and King
(1991), though the flows there were
72
driven by a rate boundary applied to the inlet and a pressure
boundary applied to the outlet. In
some previous LB simulations (Li et al. 2005; Ghassemi and
Pak 2011; Huang and Lu 2009),
relative permeability is calculated using steady-state flows
through periodic domains, driven by a
body force. In those studies, is defined as the ratio of viscosity
between the non-wetting and
wetting fluids. An increased leads to significantly increased
non-wetting phase relative

permeability. The relative permeability of the wetting-phase,
however, is insensitive to the
change in (Li et al. 2005), which is very different from our
observations. This difference is a
clear evidence that relative permeability depends on the way in
which multiphase flows are
generated and maintained, as well as the steadiness of the flows.
3. Comparison between drainage and imbibition
Case C and Case D compare a drainage process and an
imbibition process in the sphere
packing. Saturation profiles are presented in Figure 5.6(c) and
Figure 5.6(d) respectively. A
much quicker breakthrough and higher residual saturations are
observed for the imbibition case
(breakthrough when ∗ = . , = . ) than the drainage case
(breakthrough when ∗ =. , = . ). The pressure distribution
during imbibition is shown in Figure 5.7(d).
Compared to drainage, at the front of imbibition, the wetting
phase first occupies the small pores.
This preferential occupation leads to a significant capillary
pressure at the front of displacement.
In the region behind the front, the wetting phase can gradually
enter larger pores; this leads to a

lower difference between the pressures of the non-wetting and
wetting phases. After
breakthrough, the pressure gradients of wetting and non-wetting
phase become similar. During
imbibition, the wetting phase tends to progress along the small
pores as well as the corners of the
pores, leaving the non-wetting phase trapped in large pores. In
later stage of imbibition, the non-
73
wetting phase’s distribution becomes more disconnected in the
imbibition process than in the
drainage process. The disconnected non-wetting phase
saturation reduces the relative
permeability. Note that saturations that reflect later stages of
imbibition correspond to earlier
stages of drainage. As shown in Figure 5.8(c), the relative
permeability of the non-wetting phase
in imbibition is initially higher than that in the drainage at high
non-wetting phase saturations;
when the non-wetting saturation is low, the relative
permeability of the non-wetting phase in

imbibition becomes similar to that in drainage. These relative
permeability curves are therefore
reasonable. Similar findings were reported in an experimental
study (Naar et al. 1962) and pore
network modeling work (Blunt 1997) for unconsolidated glass
beads. For the wetting phase, it is
always stated that there is little hysteresis; compared to
drainage, both higher and lower relative
permeability of the wetting phase during imbibition have been
reported (Jerauld and Salter 1990).
In this work, we observed that the wetting phase relative
permeability in imbibition is lower than
that in drainage.
74

CHAPTER 6
CONCLUSIONS AND FUTURE WORK
This thesis covers the development of a multiphase flow
simulator using the lattice
Boltzmann method, evaluation of several boundary conditions,
formulation of pressure boundary
condition, and simulations of pressure driven multiphase flow
through a complex porous
medium made up by packed spheres. In this chapter, the main
contributions of this work are
summarized, followed by several recommendations for future
work.
6.1 Summary and Conclusions
In this work, first we develop a multiphase lattice
Boltzmann simulator based on the color-
gradient model. This color gradient model can simulate fluids
with large density ratios, viscosity
ratios and produce controlled interfacial tension and contact
angle. Simulation of drop
deformation in a simple shear flow shows that this multiphase
model is able to accurately

simulate dynamic problems.
To simulate multiphase flows in a meaningful size that often
contains up to 109 lattice nodes,
the simulator is parallelized for distributed-memory clusters
using MPI (message passing
interface). The parallel performance of the code is tested on
BlueM, a high-performance
computing system at Colorado School of Mines. For the test
conducted in porous media that
contain 6403, 3203 and 1603 lattice nodes, a maximum speedup
of 208 is achieved using 256
cores, yielding a 0.81 parallel efficiency. These numbers are
reasonable upon considering that
the domain size and therefore the load of computation assigned
to each core is small relative to
75
the communication cost. As the computation loaded to each core
increases, higher speedup and
parallel efficiency are expected.
We note that boundary conditions are important for

simulation of multiphase flow through
porous media, and only a few work have been done to
implement appropriate boundary
conditions. In this study, we developed / tested several
boundary conditions for the color gradient
multiphase LB model, including the convective outflow
boundary condition, the modified
periodic boundary condition, and the pressure boundary
condition. The convective outflow
boundary condition allows the fluids to exit the computation
domain with minimum unphysical
disturbance. The modified periodic boundary also allows fluids
to exit freely and can be used to
simulate gravity-driven multiphase flows such as flows in a
centrifuge. The pressure boundary
condition, validated in a capillary filling test, enables us to
simulate pressure-driven multiphase
flow using the lattice Boltzmann method. The pressure
boundary condition, specifically, has not
been developed before this study and therefore is a new
contribution.
As an application, the developed simulator is used to study
pressure-driven immiscible
displacement in a porous medium made up by packed spheres.

By using direct numerical
simulation, pressures of each phase can be obtained, which is a
distinctive advantage over
experiments where phase pressures are difficult to measure.
Compared to previous LB
simulations, we observe similar effect from the capillary
number on multiphase flow. However,
significant difference in the effect of viscosity ratio is observed,
owning to the differences in the
way multiphase flow is generated. Using the pressure boundary
condition, drainage and
imbibition can be separately evaluated by simulating their
respective displacement processes;
these two processes cannot be distinguished in simulations with
periodic boundaries with body
76
force. The relative permeability curves obtained are consistent
with published pore network
modeling and experimental results.
In conclusion, we developed a capable and flexible lattice
Boltzmann simulator for

multiphase flow. Our simulator is validated and the results are
consistent with known
observations. Our parallel simulator allows us to simulate
multiphase flow in a cm × cm × cm cubic domain using a 0.2
mm resolution which requires 70 hours’ simulation time running
on 256 cores. The increasing parallel efficiency with increasing
computational load of every core
indicates that the simulator should be capable of carrying out
higher scale simulations on capable
clusters. This lattice Boltzmann simulator is a very useful tool
to solve multiphase flow through
porous media; however, it is highly desirable to compare lattice
Boltzmann results with other
methods.
6.2 Future work
Development of this multiphase flow simulator together with
flexible boundary conditions
opens up many research opportunities. Of the many applications
and theoretical studies that can
be conducted, we suggest a few directions that can be pursued
in the near future:
thesis, we can study the effect of
multiphase flow driving mechanisms on relative permeability

including the pressure-driven,
rate-driven and gravity-driven flows. This study is a direct
application of the methods
developed in this thesis, and the results should be beneficial for
improving the understanding
of relative permeability.
single- and multiphase flow in
porous media. Stress sensitivity of relative permeability can be
modeled by simulating
77
multiphase flow in a porous medium that has stress-induced
deformations. No similar studies
have been reported and therefore it is likely to generate a new
research direction.
flow and to study the effects of
pore geometry, network topology, and fluid properties. As
experimental work on
microfluidics is being carried out in our group, simulations of
microfluidic experiments can
be conducted. By comparison with experiments, we will have

opportunities for validation of
the simulation as well as the experimental results.
78
REFERENCES CITED
Aidun, C.K. and Clausen, J.R., 2010. “Lattice-Boltzmann
method for complex flows”. Annual
Review of Fluid Mechanics, 42, 439-472.
Ahrenholz, B., Tolke, T., Lehmann, P., Peters, A., Kaestner, A.,
Krafczyk, M. and Durner, W.,
2008. “Prediction of capillary hysteresis in a porous material
using lattice Boltzmann methods
and comparison to experimental data and a morphological
pore network model.” Advance in
Water Resources, 31, 1151-1173.
Avraam, D.G., Payatakes, A.C., 1999. “Flow mechanisms,
relative permeabilities, and coupling

effects in steady-state two-phase flow through porous media.
The case of strong wettability.”
Industrial & Engineering Chemistry Research, 38, 778-786.
Badalassi, V.E., Ceniceros, H.D. and Banerjee, S., 2003.
“Computation of multiphase systems
with phase field models.” Journal of Computational Physics,
190, 371-397.
Bhatnagar, P.L., Gross, E.P. and Krook, M. 1954. “A model for
collision processes in gases. I.
Small amplitude processes in charged and neutral one-
component systems.” Physical Review
94(3): 511.
Blunt, M, 1997. “Effects of heterogeneity and wetting on
relative permeability using pore level
modeling.” SPE Journal, 2, 70-87, SPE-36762-PA.
Blunt, M., King, P., 1991. “Relative permeabilities from two-
and three-dimensional pore-scale
network modeling.” Transport in Porous Media, 6, 407-433.
Buckley, S.E. and Leverett, M.C., 1942. “Mechanism of Fluid

Displacement in Sands.” SPE
Journal, 146, 107-116, SPE-942107-G.
79
Cahn, J.W. and Hilliard, J.E., 1958. “Free energy of a
nonuniform system. I. Interfacial free
energy.” The Journal of Chemical Physics, 28, 258.
Cahn, J.W. and Hilliard, J.E., 1959. “Free energy of a
nonuniform system III. Nucleation in a
two-component incompressible fluid.” The Journal of
Chemical Physics, 31, 688.
Chen, L., Kang, Q., Mu, Y., He, Y. and Tao, W., 2014. “A
critical review of the pseudopotential
multiphase lattice Boltzmann model: Methods and
applications.” International Journal of
Heat and Mass Transfer, 76, 210-236.
Chen, S., Chen, H., Martnez, D. and Matthaeus, W., 1991.
“Lattice Boltzmann model for

simulation of magnetohydrodynamics.” Physical Review
Letters, 67, 3776-3779.
Chen, S., Martínez, D. and Mei., R., 1996. “On boundary
conditions in lattice Boltzmann
methods." Physics of Fluids, 8, 2527.
Chen, S. and Doolen, G.D., 1998. “Lattice Boltzmann method
for fluid flows.” Annual Review
of Fluid Mechanics, 30, 329-364.
Cueto-Felgueroso, L. and Juanes, R., 2013. “Forecasting long-
term gas production from shale.”
Proceedings of the National Academy of Science of the
United States of America, 110,
19660-19661.
Demianov, A., Dinariev, O., Evseev, N., 2011. “Density
functional modelling in multiphase
compositional hydrodynamics.” The Canadian Journal of
Chemical Engineering, 89, 206-
226.
d'Humières, D. 1992. “Generalized lattice-Boltzmann

equations.” Rarefied Gas Dynamics:
Theory and Simulations, 450-458.
80
d'Humières, D., 2002. “Multiple-relaxation-time lattice
Boltzmann models in three dimensions.”
Philosophical Transactions of the Royal Society of London.
Series A: Mathematical, Physical
and Engineering Sciences, 360, 437-451.
d'Ortona, U., Salin, D., Cieplak, M., Rybka, R. B. and Banavar,
J. R., 1995. “Two-color
nonlinear Boltzmann cellular automata: Surface tension and
wetting.” Physical Review E,
51, 3718-3728.
Dong, B., Yan, Y.Y., Li, W. and Song, Y., 2010. “Lattice
Boltzmann simulation of viscous
fingering phenomenon of immiscible fluids displacement in
a channel.” Computers & Fluids,

39, 768-779.
Dudukovic, M.P., Larachi, F. and Mills, P.L., 1999.
“Multiphase reactors-revisited.” Chemical
Engineering Science, 54, 975-995.
Esposito, S.J. and Thomson, N.R., 1999. “Two-phase flow and
transport in a single fracture-
porous medium system.” Journal of Contaminant Hydrology,
37, 319-341.
Frisch, U., d'Humières, D., Hasslacher, B., Lallemand, P.,
Pomeau, Y. and Rivet, J.P., 1987.
“Lattice gas hydrodynamics in two and three dimensions.”
Complex System 1, 649-707.
Gerlach, D., Tomar, G., Biswas, G. and Durst, F., 2006.
“Comparison of volume-of-fluid
methods for surface tension-dominant two-phase flows.”
International Journal of Heat and
Mass Transfer, 49, 740-754.
Ghassemi, A., and Pak, A., 2011. “Numerical study of factors
influencing relative permeabilities

of two immiscible fluids through porous media using lattice
Boltzmann method.” Journal of
Petroleum Science and Engineering, 77, 135-145.
Grunau, D., Chen, S. and Egger, K., 1993. “A lattice Boltzmann
model for multi-phase fluid
flows.” Physics of Fluids A, 5, 2557.
81
Gunstensen, A.K., Rothman, D.H., Zaleski, S. and Zanetti, G.,
1991. “Lattice Boltzmann model
of immiscible fluids.” Physical Review A, 43, 4320.
Gunstensen, A.K. and Rothman, D.H., 1993. “Lattice-
Boltzmann studies of immiscible two-
phase flow through porous media.” Journal of Geophysical
Research: Solid Earth, 98,
6431-6441.
Guo, Z., Zheng, C. and Shi, B. 2002. “Discrete lattice effects on
the forcing term in the lattice
Boltzmann method.” Physical. Review E, 65, 046308.

Hassanizadeh, M. and Gray, W.G., 1979. “General conservation
equations for multi-phase
systems: 1. Averaging procedure.” Advance in Water
Resources, 2, 131-144.
He, X., Chen, S. and Zhang, R., 1999. “A lattice Boltzmann
scheme for incompressible
multiphase flow and its application in simulation of
Rayleigh-Taylor instability.” Journal of
Computation Physics, 152, 642-663.
Hirt, C.W. and Nichols, B.D., 1981. “Volume of fluid (VOF)
method for the dynamic of free
boundaries.” Journal of Computation Physics, 39, 201-225.
Homsy, G.M., 1987. “Viscous fingering in porous media.”
Annual Review of Fluid Mechanics,
19, 271-311.
Huang, H., Huang, J. and Lu, X., 2014. “Study of immiscible
displacements in porous media
using a color-gradient-based multiphase lattice Boltzmann
method.” Computers & Fluids, 93,

164-172.
Huang, H. and Lu, X., 2009. “Relative permeabilities and
coupling effects in steady-state gas-
liquid flow in porous media: A lattice Boltzmann study.”
Physics of Fluids, 21, 092104.
82
Huang, H., Wang, L. and Lu, X., 2011. “Evaluation of three
lattice Boltzmann models for
multiphase flows in porous media.” Computers and
Mathematics with Applications, 61, 3606-
3617.
Hunt, J.R., Sitar, N. and Udell, K.S., 1988. “Nonaqueous phase
liquid transport and cleanup 1.
analysis of mechanism.” Water Resource Research, 24,
1247-1258.
Inamuro, T., Konishi, N. and Ogino, F., 2000. “A Galilean
invariant model of the lattice

Boltzmann method for multiphase fluid flows using free
energy approach.” Computer Physics
Communications. 129, 32-45.
Jerauld, G.R. and Salter, S.J., 1990. “The effect of pore-
structure on hysteresis in relative
permeability and capillary pressure: pore-level modeling.”
Transport in Porous Media, 5,
103-151.
Joekar-Niasar, V. and Hassanizadeh, S.M., 2012. “Analysis of
fundamentals of two-phase flow
in porous media using dynamic pore-network models: A
review.” Critical Reviews in
Environmental Science and Technology, 42, 1895-1976.
Kalarkis, A.N., Burganos, V.N. and Payatakes, A.C., 2002.
“Galilean-invariant lattice Boltzmann
simulation of liquid-vapor interface dynamics.” Physical
Review E, 65, 056702.
Kandhai, D., Koponen, A., Hoekstra, A., Kataja, M., Timonen,
J. and Sloot, P.M.A., 1999.
“Implementation aspects of 3D lattice BGK: Boundaries,

accuracy, and a new fast relaxation
method.” Journal of Computation Physics, 150, 482-501.
Kang, Q., Zhang, D. and Chen, S., 2004. “Immiscible
displacement in a channel: simulation of
fingering in two dimensions.” Advance in Water Resources,
27, 13-22.
Kang, Q., Zhang, D. and Chen, S., 2005. “Displacement of a
three-dimensional immiscible
droplet in a duct.” Journal of Fluid Mechanics, 545, 41-66.
83
Karimi, S., Saidian, M., Prasad, M. and Kazemi, H. 2015.
“Reservoir rock characterization using
centrifuge and nuclear magnetic resonance: A laboratory
study of middle Bakken formation.”
SPE-175069-MS, presented at the SPE Annual Technical
Conference and Exhibition held in
Houston, Texas, USA.
Kim, S. H. and Pitsch, H., 2007. “A generalized periodic

boundary condition for lattice
Boltzmann method simulation of a pressure driven flow in a
periodic geometry.” Physics of
Fluids, 19, 108101.
Kreutzer, M.T., Kapteijn, F., Moulijn, J.A. and Heiszwolf, J.J.,
2005. “Multiphase monolith
reactors: chemical reaction engineering of segmented flow in
microchannels.” Chemical
Engineering Science, 60, 5895-5916.
Kupershtokh, A.L., Medvedev, D.A. and Karpov, D.I., 2009.
“On equations of state in a lattice
Boltzmann method.” Computers & Mathematics with.
Applications, 58, 965-974.
Ladd, A.J.C., 1994. “Numerical simulations of particulate
suspensions via a discretized
Boltzmann equation. Part 1. Theoretical foundation.” Journal
of Fluid Mechanics 271, 285.
Latva-Kokko, M. and Rothman, D.H., 2005. “Diffusion
properties of gradient-based lattice
Boltzmann models of immiscible fluids.” Physical Review E,
71, 056702.

Leclaire, S., Reggio, M. and Trepaníer, J., 2012. “Numerical
evaluation of the two recoloring
operators for an immiscible two-phase flow lattice
Boltzmann model.” Applied Mathematical
modeling, 39, 768.
Lenormand, R., Touboul, E. and Zarcone, C., 1988. “Numerical
models and experiments on
immiscible displacements in porous media.” Journal of Fluid
Mechanics, 189, 165-187.
Li, H., Pan, C. and Miller, C.T., 2005. “Pore scale investigation
of viscous coupling effects for
two-phase flow in porous media.” Physical Review E, 72,
026705.
84
Liu, H., Valocchi, A.J., and Kang, Q., 2012. “Three-
dimensional lattice Boltzmann model for
immiscible two-phase flow simulations.” Physical Review E,
85, 046309.

Liu, H., Valocchi, A.J., Zhang, Y. and Kang, Q., 2014. “Lattice
Boltzmann phase-field modeling
of thermocapillary flows in a confined microchannel.”
Journal of Computation Physics, 256,
334-356.
Liu, H., Valocchi, A.J., Werth, C., Kang, Q. and Oostrom M.,
2014. “Pore-scale simulation of
liquid CO2 displacement of water using a two-phase lattice
Boltzmann model.” Advance in
Water Resources, 73, 144-158.
Liu, H., Zhang, Y. and Valocchi, A.J., 2012. “Modeling and
simulation of thermocapillary flows
using lattice Boltzmann method.” Journal of Computation
Physics, 231, 4433-4453.
Lowengrub, J. and Truskinovsky, L., 1998. “Quasi-
incompressible Cahn-Hilliard fluids and
topological transitions.” Proceedings of the Royal Society
A: Mathematical, Physical and
Engineering Sciences, 454, 2617-2654.

Lou, Q., Guo, Z. and Shi, B., 2013. “Evaluation of outflow
boundary conditions for two-phase
lattice Boltzmann equation.” Physical Review E, 87, 063301.
Mackay, D.M. and Cherry, J.A., 1989. “Groundwater
contamination: pump-and-treat
remediation.” Environmental Science & Technology, 23,
630-636.
Maier, R. S., Bernard, R.S. and Grunau, D.W., 1996. “Boundary
conditions for the lattice
Boltzmann method.” Physics of Fluids, 8, 1788.
Naar, J., Wygal, R.J. and Henderson, J.H., 1962. “Imbibition
Relative permeability in
unconsolidated porous media.” SPE Journal, 2, 13-17, SPE-
213-PA.
85
Orr Jr, F.M. and Taber, J.J., 1984. “Use of carbon dioxide in
enhanced oil recovery.” Science,

224, 563-569.
Osher, S. and Fedkiw, R.P., 2001. “Level set methods: an
overview and some recent results.”
Journal of Computation Physics, 169, 463-502.
Osher, S. and Sethian, J.A., 1988. “Fronts propagating with
curvature-dependent speed:
Algorithms based on Hamilton-Jacobi formulations.” Journal
of Computation Physics, 79,
12-49.
Pan, C., Hilpert, M. and Miller, C.T., 2004. “Lattice-Boltzmann
simulation of two-phase flow in
porous media.” Water Resource Research, 40, W01501.
Pan, C., Luo, L.-S. and C. T. Miller, 2006. “An evaluation of
lattice Boltzmann schemes for
porous medium flow simulation.” Computer & Fluids 35,
898-909.
Petunin, V., Labra, C., Xiao, F., Tutuncu, A., Sun, J. and Yin,
X., 2013. “Porosity and
permeability change under stress and correlation to rock
texture.” Poromechanics, V, 1037-

1046.
Pilliod Jr, J.E. and Puckett, E.G., 2004. “Second-order accurate
volume-of-fluid algorithms for
tracking material interfaces.” Journal of Computation
Physics, 199, 465-502.
Pooley, C.M., Kusumaatmaja, H. and Yeomans, J.M., 2009.
“Modelling capillary filling
dynamics using lattice Boltzmann simulations.” European
Physical Journal-Special Topics,
171, 63-71.
Qian, Y. H., d'Humières, D. and Lallemand, P. 1992. “Lattice
BGK models for Navier-Stokes
equation.” Europhysics Letters, 17, 479-484.
86
Reis, T. and Phillips, T.N., 2007. “Lattice Boltzmann model for
simulating immiscible two-

phase flows.” Journal of Physics A, 40, 4033.
Rothman, D.H. and Keller, J.M., 1988. “Immiscible cellular-
automaton fluids.” Journal of
Statistical Physics ,52, 1119-1127.
Rider, W.J. and Kothe, D.B., 1998. “Reconstructing volume
tracking.” Journal of Computation
Physics, 141, 112-152.
Saffman, P.G. and Taylor, G., 1958. “The penetration of a fluid
into a porous medium or Hele-
Shaw cell containing a more viscous liquid.” Proceedings of
the Royal Society of London.
Series A, Mathematical and Physical Sciences, 245, 312-
329.
Sbragaglia, M., Benzi, R., Biferale, L., Succi, S., Sugiyama, K.
and Toschi, F., 2007.
“Generalized lattice Boltzmann method with multirange
pseudopotential.” Physical Review E,
75, 026702.
Scardovelli, R. and Zaleski, S., 1999. “Direct numerical

simulation of free-surface and interfacial
flow.” Annual Review of Fluid Mechanics, 31, 567-603.
Shan, X., 2006. “Analysis and reduction of the spurious current
in a class of multiphase lattice
Boltzmann models.” Physical Review E, 73, 047701.
Shan, X. and Chen, H., 2007. “A generalized multiple-
relaxation-time Boltzmann collision
model.” International Journal of Modern Physics C, 18, 635-
643.
Shan, X. and Chen, H., 1993. “Lattice Boltzmann model for
simulating flows with multiple
phases and components.” Physical Review E, 47, 1815.
Shan, X. and Doolen, G., 1995. “Multicomponent lattice-
Boltzmann model with interparticle
interaction” Journal of Statistical Physics, 81, 379-393.
http://www.worldscientific.com/worldscinet/ijmpc
87

Sussman, M., Fatemi, E., Smereka, P. and Osher, S., 1998. “An
improved level set method for
incompressible two-phase flows.” Computers & Fluids, 27,
663-680.
Sussman, M., Smereka, P. and Osher, S., 1994. “A level set
approach for computing solutions to
incompressible two-phase flow.” Journal of Computational
Physics, 114, 146-159.
Swift, M.R., Osborn, W.R. and Yeomans, J.M., 1995. “Lattice
Boltzmann simulation of nonideal
fluids.” Physical Review Letters, 75, 830.
Swift, M.R., Orlandini, E., Osborn, W.R. and Yeomans, J.M.
1996. “Lattice Boltzmann
simulation of liquid-gas and binary fluid system” Physical
Review E, 54, 5041.
Szulczewski, M.L., MacMinn, C.W., Herzog, H.J., and Juanes,
R., 2012. “Lifetime of carbon
capture and storage as a climate-change mitigation
technology.” Proceedings of the National
Academy of Science of the United States of America, 109,
5185-5189.

Taylor, G.I., 1932. “The viscosity of a fluid containing small
drops of another fluid.”
Proceedings of the Royal Society of London. Series A, 138,
41-48.
TÖlke, J., Krafczyk, M., Schulz, M. and Rank, E., 2002.
“Lattice Boltzmann simulations of
binary fluid flow through porous media.” Philosophical
Transactions of the Royal Society A,
360, 535.
Whitaker, S., 1998. “The method of volume averaging.”
Springer Science & Business Media.
Wu, Q., Ok, J.T., Sun, Y., Retterer, S.T., Neeves, K.B., Yin, X.,
Bai, B. and Ma, Y., 2013.
“Optic imaging of single and two-phase pressure-driven
flows in nano-scale channels.” Lab
on a Chip, 13, 1165-1171.
http://dx.doi.org/10.1039/1473-0189/2001
http://dx.doi.org/10.1039/1473-0189/2001

88
Wu, M., Xiao, F., Johnson-Paben, R.M., Retterer, S.T., Yin, X.
and Neeves, K.B., 2012. “Single-
and two-phase flow in microfluidic porous media analogs
based on Voronoi tessellation, Lab
on a Chip, 12, 253-261.
Xiao, F. and Yin, X., 2015. “Geometry models of porous media
based on Voronoi tessellations
and their porosity–permeability relations.” Computers &
Mathematics with Applications, In
press, available online 19 Oct 2015.
Xu, W., Ok, J.T., Xiao, F., Neeves, K.B. and Yin, X., 2014.
“Effect of pore geometry and
interfacial tension on water-oil displacement efficiency in
oil-wet microfluidic porous media
analogs.” Physics of Fluids, 26, 093102.
Yang, J. and Boek, E.S., 2013. “A comparison study of multi-
component Lattice Boltzmann
models for flow in porous media applications.” Computers

and Mathematics with
Applications, 65, 882-890.
Yuan, P. and Schaefer, L., 2006. “Equations of state in a lattice
Boltzmann model.” Physics of
Fluids, 18, 042101.
Zhang, J. and Kwok, D.Y., 2006. “Pressure boundary condition
of the lattice Boltzmann method
for fully developed periodic flows.” Physical Review E, 73,
047702.
Zhang, L., Kang, Q., Yao, J., Gao, Y., Sun, Z., Liu, H. and
Valocchi, A.J., 2015. “Pore scale
simulation of liquid and gas two phase flow based on digital
core technology.” Science China
Technological Science, 58, 1375-1384.
Zhang, J. and Tian, F., 2008. “A bottom-up approach to non-
ideal fluids in the lattice Boltzmann
Method.” Europhysics Letters, 81, 66005.
Zhao, Y. and Fan, L., 2010. “Multirelaxation-time interaction-
potential-based lattice Boltzmann

model for two-phase flow.” Physical Review E, 82, 046708.
http://www.sciencedirect.com/science/journal/08981221
https://www.researchgate.net/researcher/2001399492_Wei_Xu
https://www.researchgate.net/researcher/2006138318_Jeong_Ta
e_Ok
https://www.researchgate.net/researcher/59121949_Feng_Xiao
89
Zou, Q. and He, X., 1997. “On pressure and velocity boundary
conditions for the lattice
Boltzmann BGK model.” Physics of Fluids, 9, 1591.

PORE-SCALE SIMULATION OF MULTIPHASE FLOW USING LATTICE BOLTZ.docx

More Related Content

Similar to PORE-SCALE SIMULATION OF MULTIPHASE FLOW USING LATTICE BOLTZ.docx (20)

More from harrisonhoward80223 (20)

Recently uploaded (20)

PORE-SCALE SIMULATION OF MULTIPHASE FLOW USING LATTICE BOLTZ.docx