Chemical Engineering Science 65 (2010) 5615–5623

Contents lists available at ScienceDirect

Chemical Engineering Science

journal homepage: www.elsevier.com/locate/ces

A modified pseudopotential for a lattice Boltzmann simulation of bubbly flow

Malin Liu a,b,1, Zhao Yu b, Tiefeng Wang a,n, Jinfu Wang a, Liang-Shih Fan b
a
Department of Chemical Engineering, Tsinghua University, Beijing 100084, China
b
Department of Chemical and Biomolecular Engineering, The Ohio State University, Columbus, OH 43210, USA

a r t i c l e in fo abstract

Article history: The pseudopotential in Shan and Chen-type multiphase models was investigated and modified based on
Received 6 August 2009 a virial equation of state with newly proposed parameters. This modified pseudopotential was used in a
Received in revised form lattice Boltzmann model and shown to be suitable for simulating sufficiently large gas–liquid density
2 July 2010
ratios with good numerical stability and only small spurious velocities. The spurious velocity was
Accepted 10 August 2010
reduced by reducing the pseudo-sound speed by the use of suitable parameters. The multicomponent
Available online 13 August 2010
multiphase model based on this modified pseudopotential can be used in bubbly flow simulations.
Keywords: Bubble rise behavior was simulated using a 3D multicomponent and multiphase model with a high
Lattice Boltzmann model density ratio. The predicted terminal velocity and drag coefficient of a single bubble agreed well with
Pseudopotential
those calculated from empirical correlations. The drag coefficient of bubbles in the homogenous regime
Bubbly flow
decreased with increased gas holdup. A new relationship between the bubble drag coefficient and gas
Gas–liquid density ratio
Spurious velocity holdup in the homogenous regime was proposed.
Equation of state & 2010 Elsevier Ltd. All rights reserved.

1. Introduction
In recent years, the lattice Boltzmann method (LBM) has
become an alternative and attractive approach to simulate fluid
flow and transport phenomena, especially in complex and multi-
phase media. It is considered an effective simulation method on
the mesoscale because the microscopic forces can be directly
included in its description of interfacial phenomena of the
bubbles and droplets. Házi et al. (2002), in a review of lattice
Boltzmann models for gas–liquid simulations, divided them into
three types: color models (Gunstensen et al., 1991), pseudopo-
tential models (Shan and Chen, 1993; He et al., 1999) and free
energy models (Swift et al., 1996). These models distinguish two
different kinds of matter or state on the basis of a parameter.
Among these models, the pseudopotential model is the most
widely used due to its simplicity and high computational
efficiency (Sukop and Thorne, 2006).
However, most LBM models for bubble simulation are still
limited to conditions of low bubble Re, Mo, or Eo numbers due to
the limitation to a low density ratio, and the problems of large
spurious velocities and poor numerical stability. Many previous
works have suggested ways to improve the LBM for use with
higher gas–liquid density ratios for bubble simulations. Inamuro
et al. (2004) proposed a method for simulating two-phase flows
with a large density ratio of 1000 by solving the pressure Poisson
n
Corresponding author. Tel.: +86 10 62797490; fax: + 86 10 62772051.
E-mail address: wangtf@tsinghua.edu.cn (T. Wang).
1
Now working at Institute of Nuclear and New Energy Technology, Tsinghua
University.
equation with free energy. However, it was very time consuming
to solve the pressure Poisson equation. Lee and Lin (2005) used
the pressure evolution equation and simulated two-phase flows
with a density ratio of 1000. In their model (Lee and Lin, 2005),
the discretization of the collision step was different before and
after the streaming step, which also carried additional computa-
tional costs. Zheng et al. (2006) reported that high liquid–gas
density ratios can be realized by using an interface-capturing
scheme based on the Cahn–Hilliard equation. However, the
density ratio in their model (Zheng et al., 2006) was for the two
components, and not for the two phases. The two phases had the
same density in their model (Zheng et al., 2006). Most recently,
Lishchuk et al. (2008) reported a method for simulating gas–liquid
flow with high density ratios by introducing effective shear-
dependent forces acting in the region of the interface. All the
above methods introduced an additional treatment to the original
Shan and Chen-type model (Shan and Chen, 1993; He et al., 1999),
such as a pressure Poisson equation, different discretization steps,
interface capturing process and shear dependent forces. These
treatments increase the complexity and computational cost of the
simulation, which were at odds with the concise nature of
the LBM.
There have also been many previous works for reducing the
spurious velocity. Nourgaliev et al. (2002) found that the spurious
velocity could be reduced with a finite difference approach in the
streaming step. Wagner (2003) argued that that the spurious
velocity was caused by non-compatible discretization of the
driving forces, and reduced the maximum spurious velocity to
O(10  16) using the potential form of the surface tension.
However, a numerical viscosity correction term had to be added

0009-2509/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2010.08.014
 Edited by Foxit Reader
Copyright(C) by Foxit Software Company,2005-2007
5616 M. Liu et al. / Chemical Engineering Science 65 (2010) 5615–5623
For Evaluation Only.
due to numerical instability. Lishchuk et al. (2003) reported that in the external forces. Thus, the equilibrium velocity can be
the spurious velocity was due to the microscopic nature of the expressed as (Guo et al., 2002)
LBM in which the interface had a finite thickness. They " #
1 X
incorporated a sharp interface into the kinematics to reduce ueq ¼ fi ei þ tFðxÞ ð3Þ
the spurious velocity. Cristea and Sofonea (2003) found that the r i
directional derivative operator was the main reason for the
spurious velocity in the interfacial region. Lee and Fischer (2006) FðxÞ ¼ Fgravity ðxÞ þ Fint ðxÞ ð4Þ
reported that the spurious velocity could be eliminated by using
the potential form of the intermolecular force with a compact The gravity force was formulated as (Sankaranarayanan et al.,
isotropic discretization. Shan (2006) showed that the spurious 2002)
velocity was caused by insufficient isotropy of the discrete
Fgravity ðxÞ ¼ gU½rrðxÞ ð5Þ
gradient operators and could be reduced by using highly isotropic
gradient operators. All these previous works had to include some
additional treatments, e.g., a finite difference streaming step, The density and velocity were calculated as
numerical viscosity correction, sharp interface kinematics or X
r ¼ fi ð6  aÞ
highly isotropic gradient operators to reduce the spurious i
velocity.
X
The interaction potential, which determines the equation of ru ¼ fi ei þ 12FðxÞ ð6  bÞ
state of the fluid, has important effects on the density ratio range i
and spurious velocity of pseudopotential-based single component
multiphase LBMs. Yuan and Schaefer (2006) performed static The fluid momentum is defined as the average of the
simulations with a density ratio of 1000 by using an appropriate momentums before and after collision (Shan and Chen, 1994).
interaction potential. They did not need to use complicated There are also other approaches for incorporating the interaction
discretization or solve the Poisson equation, but only static cases force, such as direct body forcing. The interaction force Fint(x) was
without gravity were investigated. In fact, when the vapor density associated with the nonideal equation of state of the fluid,
was too low, the dynamic simulation was not easy. Yu and Fan and was obtained by the pseudopotential model. Fint(x) was
(2007) also proposed a new expression for the potential between an interaction force between nodes, which was the driving
species that allowed the use of the multicomponent multiphase force for phase separation. The macroscopic Navier–Stokes
model with high density ratios that reached steady state quickly. equations can be derived by Chapman–Enskog expansion
These works showed that an improvement of the interaction from Eqs. (1)–(6) with approximation to the second order (Guo
potential expression based on a modification of the equation of et al., 2002):
state would be helpful for improving the pseudopotential model
in bubbly flow simulation and to simulate bubble dynamics with @r
þ rru ¼ 0 ð7  aÞ
gravity. The objectives of the current study were to study the @t
influence of the equation of state on the LBMs for bubble @ h i
simulation, and to propose a new equation of state that can be ðruÞ þ rUðruuÞ ¼ rP þ nr rðru þ ðruÞÞT þF ð7  bÞ
@t
used for sufficiently large gas–liquid density ratios, which also
where the viscosity n ¼1/3(t  0.5) in the D2Q9 model. The
gave a small spurious velocity and good numerical stability. The
hydrodynamic pressure P in Eq. (7-b) satisfies the thermodynamic
paper is organized as follows. A brief description of the theoretical
equation of state from the detailed deducing process, so
basis of the LBM with a pseudopotential model is given first. Then
hydrodynamic pressure is equivalent to the thermodynamic
different pseudopotential models are compared and an improved
pressure in the bulk of each phase (He and Doolen, 2002; Shan,
pseudopotential model is proposed with an equation of state with
2008).
new parameters. Simulation results of bubbly flow are given and
verified by experimental results. Finally, a new relationship
between the bubble drag coefficient and gas holdup in the 2.2. Interaction potential in pseudopotential model
homogenous region is proposed based on the simulation results.
The pseudopotential-based LBM was first proposed by Shan
and Chen (1993, 1994). In physical chemistry, the intermolecular
2. Methodology potential is often approximated by the Lennard–Jones potential,
but this cannot be directly used in LBM because each LB particle
2.1. The pseudopotential LBM contains a large number of molecules. Shan and Chen (1993,
1994) used the nearest neighbor interaction potential to approx-
In a standard LBM without a force, the evolution of the particle imate the intermolecular potential. The interaction potential was
distribution is described by the equation modeled as
1  Vðx,x þei Þ ¼ Gwi cðxÞcðx þei Þ ð8Þ
fi ðx þ ei Dt,t þ DtÞfi ðx,tÞ ¼  fi ðx,tÞfieq ðx,tÞ ð1Þ
t and the interparticle force could then be calculated as
X
n
In the simplest implementation, with dx¼ dt¼ 1, the equili- Fint ðxÞ ¼ GcðxÞ wi cðx þei Þei ð9Þ
brium distribution function is expressed as i¼1
h i
fieq ðrðxÞ,ueq Þ ¼ wi rðxÞ 1 þ 3ei Uueq þ4:5ðei Uueq Þ2 1:5ðueq Þ2 ð2Þ where G is the strength of the interparticle potential, n is the
number of involved neighbors, c is a function of particle density
where wi is the weight factor, and ueq is the equilibrium velocity. and wi is the weight factors.
The equilibrium velocity is affected by the interacting force F that Using the mean field approximation for intermolecular attrac-
is manifested in a nonideal equation of state of the fluid and tion and the method of treating the exclusion volume effect
external forces. In this work, only the gravity force was considered proposed by Enskog, He et al. (1999) obtained the expression for
 M. Liu et al. / Chemical Engineering Science 65 (2010) 5615–5623 5617

the interparticle force from Eq. (9) as
2 the interparticle force can be calculated by Eq. (10). Since the
Fint ðxÞ ¼ rðr cs2 PÞ þ rkrr r
where k is a parameter for the surface tension. When it was set to
zero, the surface tension was only determined by the first item on
the right hand side. P is the thermodynamic pressure associated
with the equation of state (EOS). From Eq. (10), it can be seen that
the interparticle force will be different when a different EOS is
used. It was a straightforward idea to improve the pseudopoten-
tial model by changing the equation of state.
2.3. Maxwell construction
In thermodynamics, the densities of the liquid phase and vapor
phase are determined by the system equilibrium state. The
Maxwell construction is used to obtain the vapor and liquid
densities from a given P–V line. In a gas–liquid system, the molar
Gibbs free energy EG is
In the pseudopotential model, if the equation of state is given,
ð10Þ liquid phase and vapor phase are separated when the system
reaches the equilibrium state, the densities of the liquid phase
and vapor phase can be obtained. These simulation values can be
compared with the values obtained from the Maxwell construc-
tion. Good agreements were obtained (Yuan and Schaefer, 2006)
using different equations of state.
3. Model development
The calculation method of the interparticle force can be
deduced from Eq. (9) for different discretization schemes. In the
D2Q9 model, the interparticle force can be calculated using the
expression (Yuan and Schaefer, 2006)
c0
Fint ðxÞ ffi c0 cðxÞGrcðxÞ ¼  GrcðxÞ2 ¼ rGucðxÞ2 ð15Þ
2

dEG ¼ S dT þ V dP ð11Þ where

GucðxÞ2 ¼ Prcs2 ð16Þ

At equilibrium, the molar Gibbs free energy EG and the
temperature are equal in the two phases:
We can see that the interparticle force used by Yuan and
Z P2
Schaefer (2006) was the same as that used by He et al. (1999)
EG,2 EG,1 ¼ V dP ¼ 0 ð12Þ when k was set to zero, except that the coefficients were different.
P1
Cristea and Sofonea (2003) found that the spurious velocities
could be minimized when the gradient parameter k¼0 for a given
Eq. (12) was obtained from EOS. Yuan and Schaefer (2006) found that when k ¼0, the EOS had
Z P2 Z gas Z P2 a significant effect on the density ratio, spurious velocity and
V dP ¼ ½dðVPÞP dV ¼ P sat ðVg Vl Þ P dV ð13Þ numerical stability (Yuan and Schaefer, 2006). The interparticle
P1 liquid P1
potentials based on the van der Waals equation (VdW) and
Carnahan–Sterling equation (CS) used in the literature (Yuan and
It can be expressed as Schaefer, 2006) are summarized in Table 1. In the following
Z Vg sections, we will discuss the density ratio and spurious velocity of
P dV ¼ Psat ðVg Vl Þ ð14Þ different equations of state based on Eq. (16).
Vl
In Table 1, the model parameters have different values. The
parameters are a¼9/49, b¼2/21, R¼ 1 in the VdW equation,
Eq. (14) means that area A has to be equal to area B in Fig. 1. and a ¼1, b¼4, R¼ 1 in the CS equation. The VdW equation
From this, we can see that if the equation of state is known, the and CS equation are plotted as P–r curves in Fig. 2. The curves of
density of liquid and vapor phase can be obtained from the the equation of state for the ideal gas are also shown for
P–V line. comparison.
Fig. 2 shows that the liquid density in the VdW equation was
much higher than that in the CS equation. However, the density
ratio in the CS equation was higher than that in VdW equation
P P-V line because of a much smaller vapor density given by the CS equation
(Yuan and Schaefer, 2006). The parameters had a remarkable
influence on the simulation results; therefore, the CS equation and
VdW equation had to be compared with the same parameters. In
the following section, further comparison will be given and a new
equation of state will be proposed for a high density ratio by
increasing the liquid density rather than decreasing the vapor
B density.

A
Table 1
Mesoscopic interparticle potential and equation of state.

Equation of state Interparticle potentials (Yuan and

Schaefer, 2006)

V  
VdW P ¼ rRT ar2 GcðrÞ2 ¼ 2cr0 RT
rcs2
a
1br 1br
0 a¼ 9/49, b ¼2/21, c0 ¼6.0, c2s ¼1/3, R ¼1
 
Mole volume in liquid Mole volume in vapor CS rRT 2
GcðrÞ2 ¼ 2cr0 kCS 1b
P ¼ kCS 1b r ar
RT 2
r arcs
phase phase ð1brÞ½1 þ br=4 þ ðbr=4Þ2 ðbr=4Þ3  a¼ 1, b¼ 4, c0 ¼6.0, c2s ¼ 1/3, R ¼1
kCS ¼ ð1br=4Þ3
Fig. 1. The Maxwell construction.
5618 M. Liu et al. / Chemical Engineering Science 65 (2010) 5615–5623

0.2
2 T = 0.9 T = 0.9
T = 0.8 T = 0.7
T = 0.7 T = 0.4
Ideal Gas 0.1 Ideal Gas
1
P

P
0
0.0

-1
-0.1
0 3 6 9 0.0 0.1 0.2 0.3 0.4 0.5
 

Fig. 2. Pressure–density lines of different equations of state (a) VdW EOS (a¼ 9/49, b¼ 2/21, R¼ 1) and ideal gas EOS (c2s ¼ 1/3); (b) CS EOS (a¼ 1, b ¼4, R ¼1) and ideal gas
EOS (cs2 ¼ 1/3).

0.06 0.04
VdW, T = 0.8
0.05 VdW, T = 0.75 0.03
CS, T = 0.8
0.04
0.02
CS, T = 0.75
0.03
0.01
P

0.02
P

0.00
0.01
-0.01
0.00 VdW, T = 0.8
-0.02 CS, T = 0.8
-0.01 virial
-0.03
-0.02 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
0.0 0.1 0.2 0.3 0.4


Fig. 4. Pressure–density lines of VdW EOS (a¼1, b ¼4, R¼ 1), CS EOS (a ¼1, b¼ 4,
Fig. 3. Pressure–density lines of VdW EOS (a¼ 1, b¼ 4, R ¼1) and CS EOS (a ¼1,
R¼ 1) and virial EOS.
b¼ 4, R¼ 1).

3.1. Gas–liquid density ratio

The VdW equation and CS equation with the same parameters

are given in Fig. 3. It can be seen that the density ratio in the CS 0.015 Case 1
Maxwell Construction
equation was larger than that in the VdW equation, which agreed 0.010
0.02 Simualtion
with the simulation results. This was because the CS equation had 0.005
a much higher liquid density than the VdW equation at the same 0.000
P

temperature. It has been discussed that the simulation results -0.005

agreed with the Maxwell construction in the pressure–volume -0.010
0.01
plot using the CS EOS and VdW EOS (Yuan and Schaefer, 2006). -0.015
We can construct an equation of state for a higher density ratio by 1 10 100
P

V
increasing the liquid density in the pressure–density line with the
following virial-type equation:
0.00
P ¼ RTðr þ Br2 þC r3 Þ ð17Þ
where R¼1, B ¼  7.372, C¼10 and T¼0.1. The pressure–density Case 1
line of this new equation of state is given in Fig. 4. It can be seen Case 2
-0.01
that the virial EOS had a higher liquid density than the CS EOS; Case 3
thus, a higher gas–liquid ratio can be expected. It is noted here
when the temperature was changed, the simulation results using
0 5 10 15 20
virial equation (17) also agreed with the Maxwell construction in
the pressure–volume plot, as shown in Fig. 5. V
It should be clarified that the equation of state in the Fig. 5. Comparison of gas–liquid density ratio of virial equation of state obtained
simulation was only used to calculate the interparticle potential from simulations with theoretical values, Cases1–3 are the same in Table 3-b.
 M. Liu et al. / Chemical Engineering Science 65 (2010) 5615–5623 5619

and interparticle force, which was the driving force for the phase using multi-dimension Taylor series expansion (Shan, 2006). The
separation, and that it should not be used to determine principle can be summarized as follows. The interparticle force
thermodynamic behavior. In fact, the pseudopotential model is can be expressed as
not consistent with the thermodynamics theory except for X
n
c(r)pr (He and Doolen, 2002), when the equation of state Fint ðrÞ ¼  wa GcðrÞcðr þea Þea
would be 
a¼1

1 1
P ¼ RTðr þ Bur2 Þ ð18Þ ¼ Gc rcUE2 þ r3 cUE4 þ r5 cUE6 þ    ð19Þ
3! 5!

Eq. (18) only includes the first virial coefficient; thus, its
where
accuracy is limited in thermodynamics calculations. It can be
pointed out that when the LBM is used to investigate bubble X
n
EN ¼ wa ðea Þ1 ðea Þ2 . . .ðea ÞN ð20Þ
movement in a flow, the thermodynamic effect on the simulation a¼1
results is negligible. The equations of state, which were only used
to calculate the mesoscopic potential, can be looked upon as a
EN is the orientation tensor related to the discrete velocities.
function of pressure and density. Here we can construct relation-
In a given set of ea, it is impossible to make all EN isotropic. Then,
ships for the pressure and density and change the parameters to
there would be non-radial interparticle forces because a fully
have a higher density ratio and smaller spurious velocity.
isotropic medium cannot be obtained in the discrete scheme used.
The simulation results based on a single component multi-
The spurious velocities originate from the interparticle forces. For
phase model using the VdW EOS, CS EOS and Virial EOS are shown
example, in the D2Q9 scheme (Yuan and Schaefer, 2006), E4 was
in Table 2. The computation domain was 100  100 and the
isotropic, but E6 and higher order items were non-isotropic. Since
bubble radius was 20. It can be seen that the virial EOS can be
it was difficult to make E6 isotropic, it is a straightforward idea
used with a larger density ratio and increased gas and liquid
that the effect of the non-isotropic E6 can be reduced if r5c was
densities simultaneously. Although the CS EOS can also be used
smaller. However, this is a difficult task because of the higher
with a high density ratio of 103 by reducing the vapor density, it
order derivatives. However, the spurious velocity can also be
was not practical in the simulation of bubble flow because of the
reduced if we can reduce the interparticle force. The interparticle
numerical instability caused by the extremely small vapor
force can be expressed as
density. When the vapor density was too small, the velocity

changed rapidly even with a very small force. This should be @½GucðrÞ2  @ðPrcs2 Þ @P @r
avoided because the LBM is valid only in the incompressible limit Fint ðxÞ ffi ¼ ¼ cs2 ð21Þ
@x @x @r @x
(9u9/cs{1). The simulation can be stopped by rapid velocity
fluctuations. It was found that the algorithm was unstable when a
In general, the gradient of density cannot be obtained from the
rising bubble was simulated. In addition, the initialization could
EOS, but the pseudo-sound speed qP/qr is related to the EOS
not be easily performed in order to achieve a stable simulation.
directly. It can be assumed that a smaller spurious velocity is
We found that the virial EOS could be used with a density ratio as
obtained when the qP/qr was made smaller, which would require
high as 105.1, and it was very robust with different initializations.
a higher density at the same pressure. This can be obtained easily
Firstly, this was because qP/qr in the virial EOS was smaller than
by changing the parameter in the CS equation and virial equation.
that in the CS EOS near the equilibrium vapor and liquid density,
The specific approach was to reduce the density coefficient in the
so the interparticle force and the velocity change were smaller.
equation of state proportionally. We simulated six cases using the
Secondly, the vapor density from the virial equation was larger
CS equation and virial equation. The results are summarized in
than that from the CS equation at the same density ratio, so
Table 3 (bubble radius: 20; computation domain: 100  100). The
velocity fluctuations were smaller and algorithm stability was
above assumption was verified. It can be seen that the maximum
improved. Finally, the larger vapor density increased the numer-
spurious velocity was reduced by one order of magnitude when
ical accuracy because error accumulations from the many
qP/qr was made smaller to be 1/4 of the original value. In Fig. 6,
iterations in the simulation were reduced (Cristea, 2006). From
we compare the results in the literature from using a more
our different simulations under many conditions, we recommend
isotopic EN and our simulation results. It can be seen that there
this virial equation as quite suitable for bubble dynamics
was the same effect of reducing the spurious velocity when qP/qr
simulation.
was made smaller and when a more isotopic EN was used.
From the simulation results, we found that the density ratio
3.2. Spurious velocities was decreased slightly by increasing qP/qr using the CS equation,
while the density ratio was increased almost twice using the virial
The spurious velocity should be minimized in the simulations. equation. At the same spurious velocity of 0.0005, the density
It has been found that the spurious velocity in bubble simulations ratio with the virial equation was almost ten times larger than
was due to insufficient isotropy of the discrete gradient operator that with the CS equation. This means that the virial equation was
better than the CS equation for getting a higher density ratio and
Table 2 lower spurious velocity. It should be noted here that the width of
Density ratio on different equations of state.
the diffusive interface was also enlarged by reducing qP/qr as
rL rG Density ratio
shown in Fig. 7, so the density cannot be made too large.

VdW (T¼ 0.8) 0.161 0.0169 9.52

VdW (T¼ 0.75) 0.16996 0.01016 16.72
3.3. Multiple component pseudopotential model
VdW (T¼ 0.7) 0.178 0.005 35.60
CS (T¼0.8) 0.30717 0.01423 21.58 Two kinds of pseudopotential LBM are used in bubble
CS (T¼0.75) 0.3324 0.0043 77.30 simulations: the single component multiphase (SCMP) model
CS (T¼0.718) 0.3487 0.000454 768.1
Virial, Eq. (17) 0.5569 0.0053 105.1
(Yuan and Schaefer, 2006) and the multiple component multiphase
(MCMP) model (Shan and Doolen, 1995; Huang et al., 2007).
 Edited by Foxit Reader
Copyright(C) by Foxit Software Company,2005-2007
5620 M. Liu et al. / Chemical Engineering Science 65 (2010) 5615–5623
For Evaluation Only.
Table 3

Case qP/qr t rL rG Density ratio Vs,max

(a) The maximum spurious velocity (Vs,max) using CS equation (R¼ 2) under different parameters.
1: a¼ 4, b¼ 8, T¼ 0.8 s1a 0.8 0.158 0.0072 21.94 0.0065
2: a¼ 1, b¼ 4, T¼ 0.4 0.5s1 0.8 0.307 0.0142 21.62 0.0014
3: a¼ 0.25, b ¼2, T¼ 0.2 0.25s1 0.8 0.614 0.0285 21.54 0.0005
(b) The maximum spurious velocity using virial equation (R ¼1) under different parameters
a
1: B¼  7.372, C¼ 10, T¼0.1 s2 0.8 0.5569 0.0053 105.1 0.0064
2: B¼  3.686, C¼ 2.5, T¼ 0.05 0.5s2 0.8 1.109 0.0079 140.4 0.0018
3: B¼  1.843, C¼ 0.625, T¼ 0.025 0.25s2 0.8 2.206 0.0114 193.5 0.0005

a
s1 and s2 are the baseline for comparison.

0.007 multiphase model was further modified to the multiple component

multiphase model. To do this, the interaction force at each node of
a different species was given as follows to increase the rate to the
0.006
equilibrium state (Yu and Fan, 2007):
Maximum spurious velocity

X X
0.005 F1 ðxÞ ¼ G11 wi c1 ðxÞc1 ðx þ ei Þei G12 wi j1 ðxÞj2 ðx þ ei Þei
i¼1 i

0.004 ð22  aÞ
X X
0.003 F2 ðxÞ ¼ G22 wi c2 ðxÞc2 ðxþ ei Þei G12 wi j2 ðxÞj1 ðx þ ei Þei
i i
ð22  bÞ
0.002

0.001 Component 1 was the nonideal fluid with the virial equation,
SC EOS virial EOS, E4
and component 2 was assumed to be the ideal gas with G22 ¼0.
Shan, 2006 This work The interactions between the different species were recom-
0.000
mended as follows aimed at having high diffusivity (Yu and Fan,
E4 E6 E8 s 0.5s 0.25s
2007):
Rank N of the isotropic EN Sound velocity
j1 ðr2 ðrÞÞ ¼ 1expðr2 =r20 Þ ð23  aÞ
Fig. 6. The magnitude of the spurious velocity as a function of the pseudo-sound
speed qP/qr and the rank of isotropic tensor EN.
j2 ðr1 ðrÞÞ ¼ a0 expðr1 =r10 Þ ð23  bÞ

where 0 oa0 o1, r10 and r20 were parameters that depend on the
simulation conditions. In the following section, this multicompo-
Vs,max
nent multiphase model with a high density ratio was used to
0.0005
1 investigate the bubble rise behaviors.
0.0018
0.0064
4. Simulation results and discussion
Density

0.1 4.1. Static cases

In this section, a bubble without gravity was simulated in two

dimensions using the multiphase multicomponent LBM based on
the virial equation. The bubble surface tension was obtained as
0.01
the basis for the following bubble simulation. Bubble dimension-
less groups Mo, Eo and Re were used and were as follows
(Bhaga and Weber, 1981):
0 20 40 60 80 100 gðrl rg Þm4l rUd gðrl rg Þd2
Lattice units Mo ¼ , Re ¼ , Eo ¼ ð24Þ
r s 2
l
3 ml s
Fig. 7. Density profiles at different parameters using virial equation of state.

The surface tension was obtained from the simulation results

In Sections 3.1 and 3.2, the single component multiphase model
was used to develop the pseudopotential, for the sake of simplicity
and clear physical meaning. However, for practical bubble simula-
tion, the multicomponent multiphase model has been shown to be
more stable, and has been used more widely (Kurtoglu and Lin,
2006; Gupta and Kumar, 2008). One of the reasons indicated was
that the interparticle force in the SCMP model cannot have an
appropriate repulsive value. In order to simulate the repulsive
interaction existent in a real physical system, the single component
of different radius bubble with the Laplace equation, as shown in
Fig. 8. The good agreement between the simulation results and
the Laplace equation also verified that the multiphase pseudopo-
tential model can be used to simulate bubble behavior:
s
Pin Pout ¼ ð25Þ
r
Due to the diffusive interface in the LBM, the bubble edge was
set at the position where the density was 0.5(rL + rG) to determine
 M. Liu et al. / Chemical Engineering Science 65 (2010) 5615–5623 5621

0.0012 where Kb was a function of Mo given by

(
12, Kb r 12
Kb ¼ ð27Þ
0.0010 ð14:710:2ÞMo0:038 , Kb 4 12

0.0008 Joseph (2003) also obtained an expression for the bubble

terminal velocity using the theory of potential flow for a spherical
Pin-Pout

0.0006
0.0004
0.0002
Pin−Pout =
0.0000
0.00 0.02 0.04 0.06
1/r
Fig. 8. Comparison of the simulation results with the Laplace equation.
the bubble radius. The surface tension s was 0.0156 with the
virial equation with parameters R¼1, B¼  7.372, C¼ 10, T¼0.1.
4.2. Bubble rise simulation
There are extensive works on drag correlations for multiple
bubbles in the literature. The experimental results showed that
the drag for a bubble in a bubble swarm increases first and then
decreases with an increase in gas holdup (Simonnet et al., 2007).
The details of the interphase interaction are critical for a better
understanding and reliable simulation of gas–liquid multiphase
flow.
Recently, bubble dynamics have been investigated by the
lattice Boltzmann method in several previous studies, but the
density was limited to 2.66 (Gupta and Kumar, 2008) and 2.45
(Kurtoglu and Lin, 2006). Hua and Lou (2007) studied the effect of
the gas–liquid density ratio on the bubble behavior with a front
tracking method, and found that the dependence of the bubble
behavior on the density ratio became negligible only when the
density ratio was larger than 50. Therefore, it is necessary to
increase the density ratio for the bubble simulation, and many
investigators have tried to improve LBM for use with higher gas–
liquid density ratios. In the present study, the density ratio was
increased to 105 by only modifying the pseudopotential, not
adding any other additional treatments. Thus, the model in this
work remained the advantages of simplicity and high computa-
tional efficiency of the LBM. The density ratio was 105 in all
simulations of bubble dynamics in this work; thus correct gas
bubble behaviors (e.g., shape and terminal velocity) could be
obtained.
The 3D multicomponent multiphase model described in
Section 3.3 was used. The periodic boundary was used in all
directions; thus a regular array of rising bubbles was simulated.
The cubic simulation domain of length L, width W and height H
(L¼W¼H) was changed to obtain different gas holdups.
4.2.1. Bubble rise velocity
The isolated bubble rise velocity has been investigated
theoretically and experimentally in many studies. Fan and
Tsuchiya (1990) obtained the following expression for calculating
the isolated bubble rise velocity:
2 !0:8 !0:4 31:25
 0:25
sg 4 Eo 2:8 Eo0:5 5
ut ¼ þ þ ð26Þ
rl Kb Mo0:25 Eo0:5 2
cap bubble. These works were for an isolated bubble. The rise
velocity (ub,t) for a bubble in a bubble swarm, i.e., surrounded by
other bubbles, is influenced by the gas holdup. In our simulations,
the bubble radius was fixed at 15.2, and the simulation domain was
changed from 40 to 80 (40, 45, 50, 55, 60, 65, 70, 75, 80) to give
 different gas holdups. The bubble was first set at the center of the
simulation domain with zero velocity, and then the bubble was
r
accelerated by the buoyancy force to a final terminal velocity. When
the domain was large enough such that the bubble terminal velocity
0.08 did not change with gas holdup, this was the same as the isolated
bubble terminal velocity. From the simulation results, we found that
when the bubble center distance was larger than 4.605 (¼70/15.2)
times the bubble radius, bubble interaction could be omitted.
The bubble diameter and gravity were changed and the isolated
bubble terminal velocity was obtained in the same way with
different Eo and Mo numbers (s ¼0.0156, rl/rg ¼105, n ¼0.1667).
Fig. 9 compares the terminal velocities obtained from the simulation
results and the empirical correlation. It can be seen that the
simulated terminal velocities of an isolated bubble were in good
agreement with the empirical relationships. The bubble shape
simulated here are spherical or ellipsoidal. Bubbles of high
deformation (cap, skirt, etc.) may decrease the numerical stability.
This is not the intrinsic limitation of the model and can be avoided
by introducing improved numerical technologies such as adaptive
mesh refinement. Ellipsoidal cap and skirt bubbles with Eo ranging
from 15 to 99 have been successfully simulated by using adaptive
mesh refinement with the SC model, and the related results have
been reported in another paper (Yu and Fan, 2009). Although by
using Carnahan–Starling or Peng–Robinson equation of state, one
can reach density ratios higher than 1000 for simulations of static
bubbles, the simulation of a moving bubble with these models
becomes numerically unstable at high density ratio. Actually, the
density ratio was limited to 2.66 (Gupta and Kumar, 2008) and 2.45
(Kurtoglu and Lin, 2006) in the previous studies for bubble dynamics
simulations due to numerical instability caused by high density
ratio. With the model proposed in the present work, the simulation
of moving bubbles still had a good stability with density ratio 105.
For such a case, gas density can be neglected compared with liquid
density when considering the forces acting on the bubble.
4.2.2. Bubble drag coefficient
When the bubble rises in the fluid at the terminal velocity ut,
the drag force should be balanced by the buoyancy force. The drag
coefficient can be obtained from the 3D simulation results as
4gdb
Cd,iso ¼ ð28Þ
3u2t
The relationship of drag coefficient and the isolated bubble
Reynolds numbers with high Morton numbers (Mo 44  10–3) is
given as (Bhaga and Weber, 1981)
"  #1=0:9
16 0:9
Cd,iso ¼ 2:670:9 þ ð29Þ
Reb
The simulation results of the drag coefficient compared with
the empirical relationships are shown in Fig. 10. Good agreement
was obtained for Re o100.
5622 M. Liu et al. / Chemical Engineering Science 65 (2010) 5615–5623

L = 40 L = 50 L = 60 L = 70

0.050
Eq. (26) Simulation results
Simulation Results 0.09 g = 0.002, Mo = 0.0011
Bubble terminal velocity

0.045 g = 0.005, Mo = 0.0028

g = 0.008, Mo = 0.0045
0.040 0.07

ub,iso
0.035 0.05

0.030
0.03
0.025
0.01
35 40 45 50 55 60 65 70 75 80 1 4 7 10 13 16
L Eo

Fig. 9. The isolated bubble terminal velocity with different Eo and Mo numbers (gas liquid density ratio¼ 105): (a) the process of ascent for the isolated bubble rise
velocity; (b) bubble rise velocity versus simulation domain, the bubble radius (r ¼15.2) was fixed; (c) simulation results versus experimental empirical relationship.

1000
Bhaga & Weber, 1981
Simulation results

100
Cd,iso

10

1
0.01 0.1 1 10 100
Reb

Fig. 10. The isolated bubble drag coefficient versus Re number.

Fig. 11. Homogeneous regime and heterogeneous regime.

4.2.3. Bubble interaction in the homogeneous regime
The above study of an isolated bubble did not consider the
interaction between bubbles. In industrial reactors, the existence
of a large number of bubbles makes it such that there are many
other bubbles around each bubble. The bubble–bubble interaction
has a significant impact on bubble movement behavior under
different gas holdups. So the investigation of the bubble–bubble
interaction has much significance for the description of an actual
system.
It has been found that the drag coefficient of a bubble
surrounded by other bubbles is different from that of an isolated
bubble (Simonnet et al., 2007). It has been reported that the drag
coefficient increased with the gas holdup when the gas holdup is
smaller than a critical gas holdup (0.15 in air–pure water systems)
(Simonnet et al., 2007). This is the so-called bubble ‘‘hindered
interaction’’ in the homogeneous regime. When the gas holdup is
bigger than the critical gas holdup, the drag coefficient decreased
with the gas holdup because of the so-called bubble ‘‘cooperative
interaction’’ in the heterogeneous regime. The above changes are
shown in Fig. 11.
In our simulation, the regular array of bubbles showed it was a
homogeneous regime, so the bubble ‘‘hindered interaction’’ has to
be found. The simulations results are shown in Fig. 12, which is in
agreement with the above discussion. Based on our simulation
results, the empirical relationship in the homogeneous regime
was obtained as (Simonnet et al., 2007)
Cd
¼ 0:91ð1eg Þ2:264 ð30Þ
Cd,iso
It can be seen that the simulation results were in good
agreement with experimental results in the homogeneous regime.
It should be noted here the bubble was assigned in regular arrays
in the simulation, so only the similar homogeneous regime can be
simulated. The drag coefficient increased with the gas holdup in
 M. Liu et al. / Chemical Engineering Science 65 (2010) 5615–5623
1.8
Simulation (Homogeneous regime)
1.6 Experimental Results
Eq. (30)
1.4
1.2
Cd /Cd,iso
1.0
0.8
0.6
Homogeneous Hetergeneous
0.4
0.00 0.04 0.08 0.12 0.16
g
Fig. 12. Drag coefficient versus gas holdup: simulation and experimental results.
this case. The bubble shape and flow structure are different in the
homogeneous regime and heterogeneous regime, so the bubble
drag coefficient in bubble swarm is different. More detailed
information can be provided to explain this difference using more
simulation results in heterogeneous regime in the future.
5. Conclusions
The pseudopotential in Shan and Chen-type multiphase
models was investigated and modified based on a virial equation
of state with newly proposed parameters. This modified pseudo-
potential was used in a lattice Boltzmann model to simulate
bubble behavior. It was shown that this model was suitable for
simulating bubble dynamics of sufficiently large gas–liquid
density ratios, with good numerical stability and only small
spurious velocities. The conclusions were
 The CS equation can be used for a high density ratio only by
reducing the vapor density, which would cause numerical
instability. The proposed virial equation of state can be used
for a large density ratio by increasing the liquid density with a
constant vapor density. The proposed virial equation gave
much better numerical stability in the simulation.
 The spurious velocity can be decreased by reducing the
pseudo-sound speed qP/qr. This was done by choosing
appropriate parameters. The spurious velocity can be reduced
by an order of magnitude when the pseudo-sound speed was
decreased to 1/4 the original value. The virial equation can be
used with a much higher density ratio at the same spurious
velocity as the CS equation.
 The simulation of the isolated bubble rise was in good
agreement with the experimental empirical relationships for
the rise velocity and drag coefficient. A relationship between
the bubble drag coefficient and gas holdup was proposed
based on the simulation results. The drag coefficient of a
bubble in the homogenous regime increased with gas holdup,
which is in quantitative agreement with experimental results.
Acknowledgements
The authors gratefully acknowledge the financial supports by
Foundation for the Author of National Excellent Doctoral
5623
Dissertation of PR China (No. 200757) and National 973 Project
of China (No. 2007CB714302).
References
Bhaga, D., Weber, M.E., 1981. Bubbles in viscous liquids: shapes, wakes and
velocities. Journal of Fluid Mechanics 105, 61–85.
Cristea, A., 2006. Numerical effects in a finite difference lattice Boltzmann model for
liquid–vapor systems. International Journal of Modern Physics C 17, 1191–1201.
Cristea, A., Sofonea, V., 2003. Reduction of spurious velocity in finite difference
lattice Boltzmann models for liquid–vapor systems. International Journal of
Modern Physics C 14, 1251–1266.
Fan, L.S., Tsuchiya, K., 1990. Bubble Wake Dynamics in Liquids and Liquid–Solid
Suspensions. Butterworth-Heinemann, Boston.
Gunstensen, A.K., Rothman, D.H., Zaleski, S., Zanetti, G., 1991. Lattice Boltzmann
model of immiscible fluids. Physical Review A 43, 4320–4327.
Guo, Z.L., Zheng, C.G., Shi, B.C., 2002. Discrete lattice effects on the forcing term in
the lattice Boltzmann method. Physical Review E 65, 046308.
Gupta, A., Kumar, R., 2008. Lattice Boltzmann simulation to study multiple bubble
dynamics. International Journal of Heat and Mass Transfer 51, 5192–5203.
Hazi, G., Imre, A.R., Mayer, G., Farkas, I., 2002. Lattice Boltzmann methods for two-
0.20
phase flow modeling. Annals of Nuclear Energy 29, 1421–1453.
He, X.Y., Chen, S.Y., Zhang, R.Y., 1999. A lattice Boltzmann scheme for
incompressible multiphase flow and its application in simulation of Ray-
leigh–Taylor instability. Journal of Computational Physics 152, 642–663.
He, X.Y., Doolen, G.D., 2002. Thermodynamic foundations of kinetic theory and
lattice Boltzmann models for multiphase flows. Journal of Statistical Physics
107, 309–328.
Hua, J., Lou, J., 2007. Numerical simulation of bubble rising in viscous liquid.
Journal of Computational Physics 222, 769–795.
Huang, H., Thorne, D.T., Schaap, M.G., Sukop, M.C., 2007. Proposed approximation
for contact angles in Shan-and-Chen-type multicomponent multiphase lattice
Boltzmann models. Physical Review E 76, 066701.
Inamuro, T., Ogata, T., Tajima, S., Konishi, N., 2004. A lattice Boltzmann method for
incompressible two-phase flows with large density differences. Journal of
Computational Physics 198, 628–644.
Joseph, D.D., 2003. Rise velocity of a spherical cap bubble. Journal of Fluid
Mechanics 488, 213–223.
Kurtoglu, I.O., Lin, C.L., 2006. Lattice Boltzmann study of bubble dynamics.
Numerical Heat Transfer Part B—Fundamentals 50, 333–351.
Lee, T., Fischer, P.F., 2006. Eliminating parasitic currents in the lattice Boltzmann
equation method for nonideal gases. Physical Review E 74, 046709.
Lee, T., Lin, C.L., 2005. A stable discretization of the lattice Boltzmann equation for
simulation of incompressible two-phase flows at high density ratio. Journal of
Computational Physics 206, 16–47.
Lishchuk, S.V., Care, C.M., Halliday, I., 2003. Lattice Boltzmann algorithm for surface
tension with greatly reduced microcurrents. Physical Review E 67, 036701.
Lishchuk, S.V., Halliday, I., Care, C.M., 2008. Multicomponent lattice Boltzmann
method for fluids with a density contrast. Physical Review E 77, 036702.
Nourgaliev, R.R., Dinh, T.N., Sehgal, B.R., 2002. On lattice Boltzmann modeling of
phase transition in an isothermal non-ideal fluid. Nuclear Engineering and
Design 211, 153–171.
Sankaranarayanan, K., Shan, X., Kevrekidis, I.G., Sundaresan, S., 2002. Analysis of drag
and virtual mass forces in bubbly suspensions using an implicit formulation of
the lattice Boltzmann method. Journal of Fluid Mechanics 452, 61–96.
Shan, X.W., 2006. Analysis and reduction of the spurious current in a class of
multiphase lattice Boltzmann models. Physical Review E 73, 047701.
Shan, X.W., 2008. Pressure tensor calculation in a class of nonideal gas lattice
Boltzmann models. Physical Review E 77, 066702.
Shan, X.W., Chen, H.D., 1993. Lattice Boltzmann model for simulating flows with
multiple phases and components. Physical Review E 47, 1815–1819.
Shan, X.W., Chen, H.D., 1994. Simulation of nonideal gases and liquid–gas phase-
transitions by the lattice Boltzmann-equation. Physical Review E 49, 2941–2948.
Shan, X.W., Doolen, G., 1995. Multicomponent lattice-Boltzmann model with
interparticle interaction. Journal of Statistical Physics 81, 379–393.
Simonnet, M., Gentric, C., Olmos, E., Midoux, N., 2007. Experimental determination
of the drag coefficient in a swarm of bubbles. Chemical Engineering Science 62,
858–866.
Sukop, M.C., Thorne, D.T., 2006. Lattice Boltzmann Modeling: An Introduction for
Geoscientists and Engineers. Springer-Verlag, Berlin.
Swift, M.R., Orlandini, E., Osborn, W.R., Yeomans, J.M., 1996. Lattice Boltzmann
simulations of liquid–gas and binary fluid systems. Physical Review E 54,
5041–5052.
Wagner, A.J., 2003. The origin of spurious velocities in lattice Boltzmann.
International Journal of Modern Physics B 17, 193–196.
Yu, Z., Fan, L.S., 2007. An improved multi-component lattice Boltzmann method for
simulation of gas–liquid flows with high density ratio. AIChE Annual meeting.
Yu, Z., Fan, L.S., 2009. An interaction potential based lattice Boltzmann method
with adaptive mesh refinement (AMR) for two-phase flow simulation. Journal
of Computational Physics 228, 6456–6478.
Yuan, P., Schaefer, L., 2006. Equations of state in a lattice Boltzmann model. Physics
of Fluids 18, 042101.
Zheng, H.W., Shu, C., Chew, Y.T., 2006. A lattice Boltzmann model for multiphase
flows with large density ratio. Journal of Computational Physics 218, 353–371.