A versatile lattice Boltzmann model for

immiscible ternary fluid flows

Cite as: Phys. Fluids 31, 012108 (2019); https://doi.org/10.1063/1.5056765
Submitted: 13 September 2018 . Accepted: 20 December 2018 . Published Online: 15 January 2019

Yuan Yu (余愿), Haihu Liu (刘海湖) , Dong Liang (梁栋), and Yonghao Zhang (张勇豪)

COLLECTIONS

This paper was selected as an Editor’s Pick

ARTICLES YOU MAY BE INTERESTED IN

Computational analysis of self-similar capillary-driven thinning and pinch-off dynamics

during dripping using the volume-of-fluid method
Physics of Fluids 31, 021211 (2019); https://doi.org/10.1063/1.5061715

Coalescence dynamics of unequal sized drops

Physics of Fluids 31, 012105 (2019); https://doi.org/10.1063/1.5064516

Transient evolution of the heat transfer and the vapor film thickness at the drop impact in
the regime of film boiling
Physics of Fluids 30, 122109 (2018); https://doi.org/10.1063/1.5059388

Phys. Fluids 31, 012108 (2019); https://doi.org/10.1063/1.5056765 31, 012108

© 2019 Author(s).
 Physics of Fluids ARTICLE scitation.org/journal/phf

A versatile lattice Boltzmann model

for immiscible ternary fluid flows
Cite as: Phys. Fluids 31, 012108 (2019); doi: 10.1063/1.5056765
Submitted: 13 September 2018 • Accepted: 20 December 2018 •
Published Online: 15 January 2019

Yuan Yu (余愿),1 Haihu Liu (刘海湖),2,a) Dong Liang (梁栋),1,3 and Yonghao Zhang (张勇豪)4

AFFILIATIONS
1
School of Engineering, Sun Yat-Sen University, Guangzhou 510006, China
2
School of Energy and Power Engineering, Xi’an Jiaotong University, Xi’an 710049, China
3
Guangdong Provincial Key Laboratory of Fire Science and Technology, Guangzhou 51006, China
4
Department of Mechanical and Aerospace Engineering, University of Strathclyde, Glasgow G1 1XJ, United Kingdom

a)
Author to whom correspondence should be addressed: haihu.liu@mail.xjtu.edu.cn

ABSTRACT
We propose a lattice Boltzmann color-gradient model for immiscible ternary fluid flows, which is applicable to the fluids with a
full range of interfacial tensions, especially in near-critical and critical states. An interfacial force for N-phase systems is derived
and then introduced into the model using a body force scheme, which helps reduce spurious velocities. A generalized recoloring
algorithm is applied to produce phase segregation and ensure immiscibility of three different fluids, where an enhanced form of
segregation parameters is derived by considering the existence of Neumann’s triangle and the effect of the equilibrium contact
angle in a three-phase junction. The proposed model is first validated by two typical examples, namely, the Young-Laplace test
for a compound droplet and the spreading of a droplet between two stratified fluids. It is then used to study the structure
and stability of double droplets in a static matrix. Consistent with the theoretical stability diagram, seven possible equilibrium
morphologies are successfully reproduced by adjusting the interfacial tension ratio. By simulating near-critical and critical states
of double droplets where the outcomes are very sensitive to the model accuracy, we show that the present model is advantageous
to three-phase flow simulations and allows for accurate simulation of near-critical and critical states. Finally, we investigate the
influence of interfacial tension ratio on the behavior of a compound droplet in a three-dimensional shear flow, and four different
deformation and breakup modes are observed.
Published under license by AIP Publishing. https://doi.org/10.1063/1.5056765

I. INTRODUCTION applications in controlled release, separation, and encapsu-

An emulsion is a mixture of a dispersed phase as droplets
in another immiscible fluid that forms a continuous phase.
Two basic types of emulsions are the oil-in-water (O/W)
and water-in-oil (W/O) emulsions.1 Recently, more complex
systems referred to as double emulsions and Janus emul-
sions have received rapidly growing interest due to their
unique properties2–4 and potential applications.5–10 Double
emulsions, also known as “emulsion of emulsion” or “emul-
sion within emulsion,” are emulsions with smaller droplets
encapsulated in larger droplets. The shell fluid can serve as
a barrier between the core droplets and the outer environ-
ment, which makes double emulsions highly desirable for
lation.1–4 Janus emulsions, which are named after the two-
faced Roman god Janus, are highly structured fluids con-
sisting of emulsion droplets that have two distinct physical
properties.11 Because of their natural asymmetric ability in
the compositions and the shapes, Janus emulsions are often
used in the fields that need asymmetry in shape and mate-
rials. In the applications of emulsions, morphology is one of
the most important properties and closely related to other
emulsion properties such as rheology, droplet size, relative
stability, electrical conductivity, and zeta potential.2–4,12 A
number of theoretical and experimental studies have been
devoted to identifying different equilibrium morphologies
and their transformation. For example, Torza and Mason13

Phys. Fluids 31, 012108 (2019); doi: 10.1063/1.5056765 31, 012108-1

Published under license by AIP Publishing
 Physics of Fluids
studied the droplet morphology in terms of spreading coef-
ficients and obtained the theoretical relationship between the
droplet morphology and spreading coefficients. They exper-
imentally observed three equilibrium morphologies of dou-
ble droplets, i.e., complete engulfing, partial engulfing, and
non-engulfing, which correspond to three sets of spreading
coefficients. Beyond these three equilibrium states, Pannacci
et al.14 identified several new morphologies of double droplets
and found that the non-equilibrium morphologies can have
long lifetimes controlled by hydrodynamics, which facilitates
the use of double droplets to produce encapsulated particles
at early times and Janus particles at longer times. Guzowski
et al.15 presented a detailed theoretical analysis on the possi-
ble equilibrium morphologies of double droplets and designed
the structure of double emulsions by tuning the volumes of
the constituent segments experimentally. As a supplement to
theoretical and experimental studies, numerical modeling and
simulations are becoming increasingly popular in investigation
of the behavior of Janus/double emulsions, which are typical
of three-phase flow problems.
Traditionally, three-phase flows are simulated by solv-
ing the macroscopic Navier-Stokes equations together
with various approaches to capturing or tracking the
interfaces between fluids. Among these approaches, the
front-tracking,16 volume-of-fluid (VOF),17 level-set,18–20 and
phase-field21–26 methods are commonly used. However, the
front-tracking method is not suitable for simulating inter-
face breakup and coalescence; the VOF and level-set methods
require either sophisticated interface reconstruction algo-
rithms or unphysical re-initialization processes to represent
the interfaces; and the phase-field method yields an inter-
face thickness far greater than its actual value,27,28 which may
lead to unphysical dissolution of small droplets and mobility-
dependent numerical results.29 It still remains an open ques-
tion for the phase-field method to choose an optimal mobility,
even for a two-phase flow problem.30
In the past decades, the lattice Boltzmann (LB) method
has developed into a promising alternative to the tradi-
tional Navier-Stokes-based solvers, for simulating complex
flow problems.31–35 It is a pseudo-molecular method track-
ing evolution of the distribution function of an assembly of
molecules, built upon microscopic models and mesoscopic
kinetic equations.36 The LB method has several advantages
over the traditional Navier-Stokes-based solvers, e.g., the
algorithm simplicity and parallelizability, and the ease of han-
dling complex boundaries.37 In addition, its kinetic nature
allows a simple incorporation of microscopic physics with-
out suffering from the limitations in terms of length and time
scales typical of molecular dynamics simulations.38 Thus, the
LB method is particularly useful in the simulation of mul-
tiphase flows. The existing LB models for multiphase flows
can be generally classified into four categories: color-gradient
model,30,39,40 interparticle-potential model,41–45 phase-field-
based model,46–48 and mean-field theory model.49 These
models have shown great success as in dealing with two-
phase flow problems, and all of them except the mean-field
theory model have been extended to the modeling of
Phys. Fluids 31, 012108 (2019); doi: 10.1063/1.5056765
Published under license by AIP Publishing
ARTICLE scitation.org/journal/phf
immiscible ternary fluids; see, for example, Refs. 50–59.
However, the interparticle-potential model for three-phase
flows suffers from high spurious velocities, lack of flexibil-
ity in adjusting interfacial tensions and poor stability for
unequal fluid viscosities. The phase-field-based model can-
not conserve the mass for each fluid and produces mobility-
dependent numerical results, like those in the phase-field
method. These limitations can be overcome by the color-
gradient models,56,57 which are well suited to exploring the
dynamic processes occurring in ternary fluid systems as pre-
viously demonstrated by Fu et al.58 and Jiang and Tsuji.60 The
existing color-gradient models for three-phase flows, how-
ever, commonly suffer from a problem, i.e., three interfacial
tensions should satisfy a Neumann’s triangle. In industrial pro-
cesses, surfactants are often added to emulsions to stabilize
them against droplet coalescence. The presence of surfactants
could significantly modify the interfacial tensions so that the
interfacial tensions do not always yield a Neumann’s trian-
gle. To correctly predict the dynamical behavior of emulsions,
thereby allowing precise control over the droplet geometry
and composition, it is necessary for a numerical model to be
capable of simulating ternary fluids with a full range of inter-
facial tensions. On the other hand, it is challenging to simulate
the near-critical and critical states of a ternary fluid system
where the largest interfacial tension is close to the sum of
the other two as the outcomes are very sensitive to the model
accuracy.
In this paper, we develop a LB color-gradient model for
simulating immiscible ternary fluids with a full range of inter-
facial tensions. Based on the perturbation operator developed
by Leclaire et al.,57 an interfacial force formulation is derived
to describe the interactions among different fluids and is then
introduced into the model using a body force scheme, which
is found to effectively reduce spurious velocities. In addi-
tion, the recoloring algorithm proposed by Spencer et al.56 is
applied to maintain the interfaces and ensure immiscibility of
three different fluids, where an enhanced form of segregation
parameters is proposed by considering both the existence of
Neumann’s triangle and the effect of the equilibrium contact
angle in a three-phase junction. The capability and accuracy of
this model are first assessed by simulating the Young-Laplace
test for a compound droplet and the spreading of a droplet
between two stratified fluids. It is then used to study the
structure and stability of double droplets in a static matrix
fluid, where we emphasize the model’s capability for simu-
lating ternary fluid flows in near-critical and critical states.
Finally, the model is used to simulate a compound droplet in a
three-dimensional shear flow, and different droplet behaviors
are exhibited and explained.
II. NUMERICAL METHOD
The two-phase color-gradient LB model of Liu et al.30,61
is extended to the simulation of immiscible ternary fluids. The
ternary color-gradient model consists of three steps, i.e., the
collision step, the recoloring step, and the streaming step.
In the collision step, an interfacial force that describes the
31, 012108-2
 Physics of Fluids ARTICLE scitation.org/journal/phf

interactions among different fluids is derived from the per- D3Q19:

turbation operator presented in Leclaire et al.57 and is then
0 1 −1 0 0 0 0 1 −1 1 −1 1 −1 1 −1 0 0 0 0 
introduced by the body force scheme of Guo et al.62 In the  
recoloring step, an enhanced form of segregation parame- [e0 , e1 , . . . , e18 ] = 0 0 0 1 −1 0 0 1 1 −1 −1 0 0 0 0 1 −1 1 −1 ,
ters is proposed to ensure accurate phase segregation in a
0 0 0 0 0 1 −1 0 0 0 0 1 1 −1 −1 1 1 −1 −1
 
three-phase junction and to allow for the states where three (6)
interfacial tensions between the fluids cannot form a trian-
gle, known as Neumann’s triangle. The distribution functions
f i,r , f i,g , and f i,b are introduced to represent three immisci-


 1/3, i = 0;

wi =  i = 1, 2, 3, 4, 5, 6;

ble fluids, i.e., red fluid, green fluid, and blue fluid, where the  1/18, (7)
subscript i is the lattice velocity direction and ranges from  1/36, i = 7, 8, 9, . . . , 18.


0 to (n − 1) for a given m-dimensional DmQn lattice model.

In the present study, the D2Q9 and D3Q19 lattice models Using the concept of a continuum surface force to model
are used for two-dimensional and three-dimensional simula- the interfacial tension along with the constraints of mass con-
tions, respectively. The total distribution function is defined as servation and momentum conservation, Liu et al.30 derived
f i = k f i,k (k = r, g, or b), which undergoes a collision step as a generalized expression for the perturbation operator in
P
two-phase simulations. This perturbation operator was later
fi† (x, t) = fi (x, t) + Ωi (fi (x, t)) + Φi (x, t), (1) improved by Leclaire et al.57 to model the interfacial tensions
where fi (x, t) is the total distribution function in the ith veloc- between different fluids in three-phase simulations. Following
ity direction at the position x and the time t, fi† is the post- Leclaire et al.,57 the perturbation operator is given by
collision distribution function, Ωi is the Bhatnagar-Gross- X
Φi = Φi,k , (8)
Krook (BGK) collision operator, and Φi is the forcing term (also
k
known as perturbation operator), which contributes to the
XA C (ei · Gkl )2
 
mixed interfacial regions and creates the interfacial tensions Φi,k = kl kl
|Gkl | wi − B  , (9)
i
between different fluids. l,l,k
2  |Gkl | 2 
In the BGK collision operator, the total distribution func- ρ ρ ρ ρ
where Gkl = ρl ∇ ρk − ρk ∇ ρl is the color gradient57 and is intro-
tions are relaxed toward a local equilibrium with a single
relaxation time, duced to identify the location of the k − l interface, i.e., the
interface between the fluid k and the fluid l. Ckl is a concen-
1 f eq
g
tration factor that controls the activation of the interfacial
Ωi ( fi (x, t)) = − fi (x, t) − fi (x, t) , (2)
τf tension at the k − l interface and is given by57
eq
where τ f is the dimensionless relaxation time and fi is the ρ ρ
equilibrium distribution function of f i . The equilibrium distri- Ckl = min*106 0k 0l , 1+, (10)
ρ ρ -
k l
bution function is obtained by a second order Taylor expan- ,
sion of the Maxwell-Boltzmann distribution with respect to where ρk0 is the density of the pure fluid k and Akl is a param-
the local fluid velocity u, eter related to the interfacial tension between the fluids k and
2 l, i.e., σkl = 91 (Akl + Alk )τf . The generalized expression for Bi
e · u (ei · u) u · u 
 
eq
fi = wi ρ 1 + i 2 + − , (3) was given by Liu et al.,30 and it was, in particular, taken as
 cs 2c4s 2c2s  B0 = −4/27, B1–4 = 2/27, and B5–8 = 5/108 for the D2Q9
where ρ = k ρk is the total density and ρk is the density of the
P model57 and B0 = −1/3, B1–6 = 1/18, and B7–18 = 1/36 for the
√ D3Q19 model30. It is worth noting that Eq. (9) is not only lim-
fluid k; cs = c/ 3 is the speed of sound where c is related to
the lattice spacing δ x and time step δ t by c = δ x /δ t (for the ited to the case of ternary fluids but also applicable to N-phase
sake of simplicity, δ x = δ t = 1 is used hereafter); ei is the lattice (N > 3) systems.
velocity in the ith direction; and wi is the weighting factor. Using the Chapman-Enskog multiscale analysis, it is
For the D2Q9 and D3Q19 models, the lattice velocities and the shown that the perturbation operator, given by Eqs. (8) and
weight factors are given as follows: (9), can lead to the following interfacial force:
D2Q9:
σkl Ckl
X XX " #
0 1 0 −1 0 1 −1 −1 1  Fs = −∇ · *τf δt Φi ei ei + = ∇· |Gkl |(I − n n )
kl kl ,
2
[e0 , e1 , . . . , e8 ] =  , (4) , i - k l,l,k
0 0 1 0 −1 1 1 −1 −1 (11)
where nkl is the unit normal vector of the k − l interface and is
defined by nkl = Gkl /|Gkl |.
 4/9, i = 0;
Instead of using Eqs. (8) and (9), the effect of interfacial



wi =  i = 1, 2, 3, 4;

 1/9, (5) tensions is realized through the body force scheme of Guo
i = 5, 6, 7, 8.

 1/36,
et al.,62 which is able to reduce effectively spurious velocities



Phys. Fluids 31, 012108 (2019); doi: 10.1063/1.5056765 31, 012108-3

Published under license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

while keeping high numerical accuracy.61,63 According to Guo

et al.,62 the forcing term Φi in Eq. (1) is written as
! !
1 ei − u ei · u
Φi (x, t) = wi 1 − + e i · Fs (x, t)δt , (12)
2τf c2s c4s
where the local fluid velocity is defined by the averaged
momentum before and after the collision, i.e.,
X 1
ρu(x, t) = fi (x, t)ei + Fs (x, t)δt . (13)
2
i

For the sake of simplicity, the present model is limited to

FIG. 1. Neumann’s triangle relating the interfacial tensions (σ rg , σ rb , and σ gb )
ternary fluids with similar densities, which could be improved
and contact angles (ϕ rg , ϕ rb , and ϕ gb ) of three fluids coexisting in the static
by using the techniques recently developed by Ba et al.64 To equilibrium.
allow for unequal viscosities of the three fluids, we determine
the local kinematic viscosity ν by a harmonic mean
ρ X ρk as Neumann’s triangle, as shown in Fig. 1. When three inter-
= , (14)
ν νk facial tensions satisfy a Neumann’s triangle, the equilibrium
k
contact angle ϕ kl formed between the fluids in a three-phase
where ν k (k = r, g, or b) is the kinematic viscosity of the pure junction is related to the interfacial tensions by
fluid k. The local relaxation time τ f can be calculated from the
local viscosity using the following equation: σmk
2 + σ2 − σ2
ml kl
cos(ϕkl ) = . (18)
! 2σmk σml
1 2
ν = τf − c δt . (15) Spencer et al.56 theoretically showed that in a three-phase
2 s
junction, there should be a relationship between ϕ kl and the
The partial derivatives in the interfacial force Fs should (relative) interface thickness, which is controlled by the seg-
be evaluated through suitable difference schemes. To min- regation parameter β kl . Hence, it is of great importance to
imize the discretization errors, the fourth-order isotropic select a proper β kl in three-phase simulations. Several differ-
finite difference scheme ent forms of β kl have been provided in the literature. Spencer
1 X et al.56 proposed the first expression for the segregation
∂α ϕ(x, t) = 2 wi ϕ(x + ei δt , t)eiα (16) parameters, which is given by
cs i
βrg =β 0
is used to evaluate the derivatives of a variable ϕ.



27ρr ρg ρb 
 "

 #
0
β =β ϕ

1 + sin 1

Although the forcing term generates the interfacial ten-

 rb gb −
 ρ3 , (19)
sions, it does not guarantee the immiscibility of different flu- 

27ρr ρg ρb

 " #
ids. In order to minimize the mixing of the fluids, a recoloring

 βgb =β 0 1 + (sin ϕrb − 1)



step is applied. Based on the pioneering work of D’Ortona  ρ3
et al.,65 Latva-Kokko and Rothman66 developed a recoloring
where β 0 is the reference segregation parameter. Clearly, the
algorithm to demix two immiscible fluids, which can overcome
segregation parameters in Eq. (19) will degenerate into β kl = β 0
the lattice pinning problem and creates a symmetric distri-
at an interface where only two fluids are present. So it is sug-
bution of particles around the interface so that unphysical
gested to take β 0 = 0.7 to be consistent with the segregation
spurious velocities can be effectively reduced. This recoloring
parameter in the two-phase color-gradient model30. Leclaire
algorithm was later generalized by Spencer et al.56 to three-
et al.57 improved the segregation parameters of Spencer
phase fluid flows. Following Spencer et al.,56 the recolored
et al.56 by setting β kl = β 0 for the largest ϕ kl in Neumann’s
distribution functions of the fluid k (k = r, g, or b) are
triangle,
ρ X ρ ρ
‡
fi,k (x, t) = k fi† (x, t) + βkl wi k l nkl · ei , (17) 0
ρ ρ β

 kl with ϕmax
l,l,k βkl =  (20)
 β 0 + β 0 C [sin (π − ϕ
max − ϕkl ) − 1] otherwise,

‡ t
where is the recolored distribution functions of the fluid
fi,k

35ρr ρg ρb
 
k and β kl is a segregation parameter related to the thickness where ϕ max = max(ϕ kl ) and Ct = min , 1 . Leclaire
ρ 3
of the k − l interface. It should be noted that β kl = β lk in
et al.57 also mentioned to use β kl = β 0 when Neumann’s tri-
order to conserve mass and momentum during the recoloring
angle does not exist. Although Eqs. (19) and (20) work to some
process.
extent especially when Neumann’s triangle exists, they can-
In a three-phase junction, three interfacial tensions in the not accurately simulate the critical state where the largest
equilibrium form a triangle and this triangle of force is known interfacial tension equals the sum of the other two, which

Phys. Fluids 31, 012108 (2019); doi: 10.1063/1.5056765 31, 012108-4

Published under license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

will be shown later. Recently, Fu et al.58 seemed to have also domain is taken as 160 × 160, and it is filled with three different
noticed that Eqs. (19) and (20) do not always produce convinc- fluids, which are initialized as
ing results in three-phase simulations, so they simply selected
a constant β kl , i.e., 
 ρr = 1, ρg = ρb = 0 (x − 80)2 + (y − 80)2 ≤ R2r


βkl = β 0 .

(21) 

 ρg = 1, ρr = ρb = 0 R2r < (x − 80)2 + (y − 80)2 ≤ R2g

(27)
It is evident that the dependence of β kl on ϕ kl is not considered





in Eq. (21), and thus, incorrect results may be obtained, e.g., in  ρb = 1, ρr = ρg = 0

otherwise,

the critical state.
with Rg = 2Rr . This gives the initial condition that a com-
To overcome the aforementioned drawbacks associated
pound droplet is located in the center of the computational
with the existing β kl , an enhanced form of segregation param-
domain. The interfacial tensions and the fluid viscosities are
eters is proposed. First, we determine whether Neumann’s
taken as σ rg = σ rb = σ gb = 0.01 and ν r = ν g = ν b = 0.1, respec-
triangle exists by calculating
tively, and the periodic boundary conditions are used in both
σmk
2 + σ2 − σ2 x and y directions. According to Young-Laplace’s law, when
ml kl
Xkl = . (22) the system reaches the equilibrium state, the pressure differ-
2σmk σml
ence ∆p across an interface is related to the interfacial tension
It is easily seen from Eq. (18) that Neumann’s triangle will exist σ by
if |Xkl | < 1 for all kl. Then, the segregation parameter β kl is σ
∆p = , (28)
defined as a continuous function of Xkl , R
where R is the curvature radius of the interface. Equation (28)
35ρr ρg ρb
!
βkl = β 0 + β 0 min , 1 g(Xkl ), (23) allows us to quantify the modeling accuracy of interfacial
ρ3 tensions through the relative error
where 
∆pgb Rg + ∆prg Rr − σgb + σrg

= × 100%. (29)

 1 Xkl < −1 σgb + σrg



 1 − sin (arccos(Xkl )) −1 ≤ Xkl < 0


Table I shows the relative errors of interfacial tensions for


g(Xkl ) =  . (24)


 sin (arccos(Xkl )) − 1 0 ≤ Xkl ≤ 1 different values of Rr . All the relative errors  are below 1.5%,



 suggesting that our numerical results are in excellent agree-
1 < Xkl

 −1

ment with Young-Laplace’s law. In addition to the present
forcing method, i.e., Eqs. (11) and (12), the interfacial tension
It should be noted in the three-phase junction that
effects can also be realized by the forcing method of Leclaire
Eqs. (23) and (24) can be derived based on the following
et al.,57 i.e., Eqs. (8) and (9). It is of interest to compare the
relationship:
effect of these two different forcing methods on spurious
βrg βrb βgb velocities. Table II shows the maximum spurious velocities
= = , (25) (|u| max ) for various Rr , where the values of |u| max are magnified
sin(ϕrg ) sin(ϕrb ) sin(ϕgb )
by 105 times. It is seen that the maximum spurious veloci-
and the detailed process is shown in the Appendix. Clearly, the ties are almost independent of Rr for either forcing method
proposed β kl is consistent with the nature of diffuse interfaces, and that the present spurious velocities are always smaller
thus leading to more accurate results than using other forms than those obtained with the forcing method of Leclaire
of β kl . Moreover, it works well no matter if Neumann’s triangle et al.57
exists or not.
After the recoloring step, the red, green, and blue distri- TABLE I. The relative errors of interfacial tensions for various values of Rr .
bution functions propagate to the neighboring lattice nodes,
known as the propagation or streaming step, Rr 15 20 25 30

‡
fi,k (x + ei δt , t + δt ) = fi,k (x, t), k = {r, g, b}, (26)  (%) 1.3 0.95 0.83 0.57

with the post-propagation distribution functions used to com-

pute the densities of colored fluids by ρk = i f i,k .
P
TABLE II. The maximum spurious velocities (|u| max ) obtained with two different
forcing methods for various Rr .
III. NUMERICAL VALIDATIONS Rr 15 20 25 30
A. Young-Laplace test for a compound droplet Present forcing method 1.68 1.69 1.70 1.71
A compound droplet, which consists of a core and a shell |u | max × 105
Forcing method of
suspended in a third fluid, is simulated to assess whether the 3.15 3.28 3.27 3.29
Leclaire et al.
interfacial tensions are correctly modeled. The computational

Phys. Fluids 31, 012108 (2019); doi: 10.1063/1.5056765 31, 012108-5

Published under license by AIP Publishing
 Physics of Fluids
FIG. 2. The shape of a liquid lens at equilibrium.
B. Spreading of a droplet between two
stratified fluids
To assess the overall performance of the proposed model,
we simulate the spreading of a droplet between two other
immiscible fluids. The computational domain is set to be
160 × 160 lattices. Initially, a red circular droplet with the
radius R = 20 is placed in the center of the computational
domain, and the green and blue fluids are allocated to the
lower and upper halves of the computational domain outside
the droplet. Periodic boundary conditions are used in both
the x and y directions. Depending on the values of the inter-
facial tensions, two different spreading phenomena can be
observed, i.e., partial spreading and complete spreading.
We first consider the partial spreading, where three inter-
facial tensions yield a Neumann’s triangle. In a partial spread-
ing, the droplet can eventually reach a steady lens shape,
which is often characterized by the lens length D and the
heights h1 and h2 (see Fig. 2). The lens length and heights can
be analytically given as60,67
v
u and 4(d)]. Since three interfacial tensions do not satisfy a
u
u A
D = 2t
u
, (30a)
2  θ

1
sin θ sin θ − cos θi
i
P
i i
i=1
D 1 − cos θi
!
hi = with i = 1, 2,
2 sin θi
where A is the area of the red droplet; θ 1 = ϕ rg and
θ 2 = ϕ rb are the equilibrium contact angles that can be calcu-
lated from Eq. (18). Four groups of interfacial tensions are sim-
ulated with a constant σ gb of 0.01 but varying σ rb and σ rg , i.e.,
(a) σ rb = 0.01 and σ rg = 0.01, (b) σ rb = 0.0087 and σ rg = 0.005,
(c) σ rb = 0.0173 and σ rg = 0.02, and (d) σ rb = 0.0058 and
σ rg = 0.0115. The fluid viscosities are all kept at 0.1, and the
final fluid distributions are shown in Fig. 3. As expected, the
Phys. Fluids 31, 012108 (2019); doi: 10.1063/1.5056765
Published under license by AIP Publishing
ARTICLE scitation.org/journal/phf
droplet exhibits a lens shape in each of the cases considered,
and the geometrical sizes (D, h1 , and h2 ) of the lens are case
dependent. Based on the fluid distributions, we also quan-
tify the geometrical sizes of the lens and compare the sim-
ulated results with the analytical predictions from Eq. (30).
It is seen in Table III that the simulated results (denoted
by Ds , hs1 , and hs2 ) agree well with the analytical predictions
(denoted by Da , ha1 , and ha2 ) with the relative errors [defined
by E(χ) = | χ χ−χ
a s|
a × 100%, where χ = D, h1 , or h2 ] all around
1% except in the cases of small contact angles. The increased
errors at small contact angles are attributed to the low res-
olution in sharp corners, which were also found by Jiang and
Tsuji.60
We then consider the complete spreading, where three
interfacial tensions cannot yield a Neumann’s triangle. Two
different cases of complete spreading are simulated for
σ rg = 0.01 and σ rg = 0.015 at σ gb = σ rb = 0.005. Clearly,
σ rg = σ gb + σ rb in the first case, which corresponds to the
critical state, whereas σ rg > σ gb + σ rb in the second case,
which corresponds to the supercritical state. Figure 4 shows
the time evolution of the interface in both cases for a con-
stant fluid viscosity of 0.05, where the results obtained by
the model of Fu et al.58 are also plotted for comparison. The
simulation results from the present model show that in the
critical state, the red droplet sits exactly on the gb inter-
face in the end [Fig. 4(a)], while in the supercritical state, it
bounces off the gb interface and rises up to the blue fluid
[Fig. 4(b)]. By contrast, when the model of Fu et al.58 is
used, the red droplet keeps adhering to the gb interface and
three-phase regions are clearly seen in both states [Figs. 4(c)
Neumann’s triangle, a stable three-phase region cannot be
formed and the results shown in Figs. 4(c) and 4(d) are not
correct.
Finally, case (a) in Table III is chosen to examine whether
(30b) the grid resolution of R = 20 can produce grid-independent
results. Three different grid resolutions are considered, i.e.,
R = 10, 20, and 40, and the corresponding domain sizes are
80 × 80, 160 × 160, and 320 × 320, respectively. When all
simulations reach the steady states, we measure the geomet-
rical sizes of the lens and calculate the height-to-length ratios
n = h1D+h2
, which are shown in Table IV. It is seen that the rel-
ative error of the height-to-length ratio is as low as 1.5% at
R = 20, indicating that the grid resolution with R = 20 can
produce nearly grid-independent results.
FIG. 3. Final fluid distributions in the partial spread-
ing for (a) σ rb = 0.01, σ rg = 0.01; (b) σ rb = 0.0087,
σ rg = 0.005; (c) σ rb = 0.02, σ rg = 0.0173; and (d)
σ rb = 0.0115, σ rg = 0.0058. The third interfacial tension
is fixed at σ gb = 0.01.
31, 012108-6
 Physics of Fluids ARTICLE scitation.org/journal/phf

TABLE III. Comparison between the analytical predictions and simulated results for the geometrical sizes of the liquid lens.

Case Da ha1 ha2 Ds hs1 hs2 E(D) (%) E(h1 ) (%) E(h2 ) (%)

(a) 55.34 15.97 15.97 53.59 15.92 15.92 3.17 0.37 0.38
(b) 65.18 8.73 18.81 61.79 8.79 18.97 5.21 0.59 0.82
(c) 45.84 13.23 22.92 45.39 13.35 22.51 0.99 0.85 1.80
(d) 50.99 6.83 25.49 50.93 6.77 25.22 0.13 0.94 1.09

IV. STRUCTURE AND STABILITY OF DOUBLE
DROPLETS
Double emulsions have received considerable atten-
tion because of their potential applications in food science,
cosmetics, pharmaceuticals, and medical diagnostics. Since
emulsion properties and functions are related to the droplet
geometry and composition, it is of great importance, from a
numerical point of view, to accurately predict the topolog-
ical structure of double droplets when dispersed in another
immiscible fluid.
A. Stability diagram for double droplets
Consider a pair of equal-sized droplets, consisting of
red and green fluids and initially sitting next to each other,
immersed in the third fluid (blue fluid). Based on the theoret-
ical analysis, Guzowski et al.15 presented a stability diagram
that describes the possible topologies of double droplets and
their transitions in terms of two ratios of the interfacial ten-
sions (see the left panel of Fig. 5). In the stability diagram,
seven typical cases (represented by the solid points) are simu-
lated to examine if the present model is able to reproduce the

FIG. 4. Time evolution of the interface in the complete

spreading for [(a) and (c)] σ rg = 0.01 and [(b) and (d)]
σ rg = 0.015. The other two interfacial tensions are fixed at
σ gb = σ rb = 0.005. (a) and (b) are simulated by the present
model, while (c) and (d) are simulated by the model of Fu et
al.58 Note that the system has reached the steady state at
t = 50 000 in each case.

Phys. Fluids 31, 012108 (2019); doi: 10.1063/1.5056765 31, 012108-7

Published under license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

TABLE IV. The fluid properties in lattice unit, the height-to-length ratios n, and ξ = 1/(6kβ 0 ),68 where k is a geometric constant that is deter-
the relative errors E n on different grid resolutions. The relative error is defined as mined by69
n−n
En = n R=40 × 100%. Note that the fluid properties in lattice unit usually should
R=40 1 X wi ei ei
be changed on different grid resolutions in order to correspond to the same physical k= . (34)
system. 2 |ei |
i

R ρk νk σ kl n En (%) For the D2Q9 model, one can obtain k ≈ 0.1504 from Eq. (34),
and thus, ξ ≈ 1.5831 for β 0 = 0.7. The periodic boundary con-
80 × 80 10 1 0.05 0.005 0.613 4.8 ditions are used in both the x and y directions. All the fluids
160 × 160 20 1 0.10 0.010 0.594 1.5
are assumed to have equal viscosity of 0.1, and the interfacial
320 × 320 40 1 0.20 0.020 0.585 0
tension σ rg is fixed at 0.01. The simulations are run until an
equilibrium state is reached, and the equilibrium morpholo-
gies of double droplets for the seven cases are depicted in
correct morphologies of double droplets. These typical cases the right panel of Fig. 5. It is seen that seven different equilib-
σ σ σ σ
are (i) σgb
rg
= 1.7 and σrb
rg
= 0.5, (ii) σgb
rg
= 2 and σrb
rg
= 1, (iii) rium morphologies are exhibited, and they can be described as
σgb σrb σgb σrb σgb
= 0.4 and = 0.4, (iv) complete engulfing of the green fluid by the red fluid (i), crit-
σrg σrg σrg = 0.5 and σrg = 0.5, (v) σrg =1
σrb σgb σ σgb ical engulfing of the green fluid by the red fluid (ii), separate
and σrg = 1, (vi) σrg = 1 and σrb = 2, and (vii) σrg = 0.5 and
σrb
rg dispersion or non-engulfing (iii), kissing (iv), partial engulfing
σrg = 1.7, which cover all the possible morphologies identified (v), critical engulfing of the red fluid by the green fluid (vi),
by Guzowski et al.15 and complete engulfing of the red fluid by the green fluid
(vii). These simulation results are consistent with the theo-
The computational domain is taken to be [1, 120] × [1, 120],
retical predictions by Guzowski et al.15 By contrast, it still
and the initial fluid distributions are
remains a challenge to use the LB phase-field-based mod-
els for correctly reproducing all the equilibrium morphologies
q
2 2
 R − (x − 60.5) + (y − 60.5 − R) 
 
ρr (x, y) = 0.5 + 0.5 tanh  shown in Fig. 5. This is because when the largest interfacial
ξ  , (31)
  tension is greater than the other two, the free-energy den-
 sity becomes negative,52,53 which often leads to numerical
q instability.
R − (x − 60.5)2 + (y − 60.5 + R)2 
 
ρg (x, y) = 0.5 + 0.5 tanh   , (32) B. Near-critical and critical states
 ξ 
  For double droplets immersed in a static matrix, the criti-
cal state occurs when the largest interfacial tension equals the
ρb (x, y) = 1 − ρr (x, y) − ρg (x, y), (33) sum of the other two. As previously shown in Fig. 5, the criti-
cal state of double droplets can be subdivided into the kissing
where the droplet radius R = 20 lattices. The parameter ξ
state [(iv) in Fig. 5] and the critical engulfing state [(ii) or (vi)
is a measure of the interface thickness related to β 0 by
in Fig. 5]. For the convenience of description, we define the
near-critical state as the state where the largest interfacial
tension is close to the sum of the other two. It is challeng-
ing to accurately simulate the critical and near-critical states,
where a slight inaccuracy in modeling could lead to significant
simulation errors.
To highlight the strength of the present model for crit-
ical scenarios, we consider the kissing/near-kissing states
and the critical/near-critical engulfing states in a rectangu-
lar domain of [1, 300] × [1, 300]. The boundary conditions and
fluid viscosities are set the same as those in Sec. IV A. In the
kissing/near-kissing states, two equal-sized droplets with the
radii of Rr = Rg = 60 are initially placed with a distance of
drg , and they are symmetric with respect to the centerline
y = 150.5. The simulations are performed for a constant σ rg
of 0.01 but varying σ gb (=σ rb ), which is varied around the criti-
cal value of 0.005 with an increment of 2 × 10−4 . Note that the
initial distance drg depends on the value of σ gb and is given by
FIG. 5. Stability diagram representing possible morphologies of double droplets its analytical value in equilibrium as
(left panel) and equilibrium shapes of the droplets for the typical cases marked in
the stability diagram (right panel). The lines represent the critical morphologies or
the transitions between the regions of complete engulfing, partial engulfing, and  Rr σrg /σgb
 if σgb > 0.005;
drg =  (35)
non-engulfing.  R + R = 120 otherwise.
 r g

Phys. Fluids 31, 012108 (2019); doi: 10.1063/1.5056765 31, 012108-8

Published under license by AIP Publishing
 Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 6. (a) The interface length Lrg as

a function of σ gb in the kissing/near-
kissing states and (b) the interface
length Lrb as a function of σ rb in the
critical/near-critical engulfing states.

FIG. 7. The final fluid distributions obtained by (a) the

present model and (b) the model of Fu et al.58 for the kiss-
ing state and by (c) the present model and (d) the model of
Fu et al.58 for the critical engulfing state.

In the critical/near-critical engulfing states, we consider results as the one of Fu et al.,58 so the fluid distributions from
a green droplet with Rg = 80 entirely or partially engulf- it are not shown in the figure. Clearly, both critical states are
ing a red droplet with Rr = 60 for σ gb = σ rg = 0.01. σ rb is successfully reproduced by the present model but not by the
varied around the critical value of 0.02 with an increment model of Fu et al.58 These results indicate that the present
of 2 × 10−4 . With these parameters, we are able to analyti-
cally compute other geometric parameters, which are given
R sin θ
q
by drg = R2r + R2g − 2Rr Rg cos α, Rb = gsin θ g , and dgb = Rg cos θ g
b
− Rb cos θ b for σ rb < 0.02, and by drg = Rg − Rr = 20 for
σrb R2g +d2gr −R2r
σ rb ≥ 0.02. Herein, cos α = 2σgb , θg = arccos 2Rg dgr
, and
θ b = θ g + 2α. Again, we initialize the fluid distribution such that
it follows the analytical geometric parameters.
In addition to the present model, we also use the model
of Fu et al.58 and the model of Leclaire et al.57 for the simula-
tions. When the simulations reach the steady state, we quan-
tify the interface lengths Lrg in the kissing/near-kissing states
and Lrb in the critical/near-critical engulfing states. Figure 6
compares the simulated results from the present model with
those from the model of Fu et al.58 and the model of Leclaire
et al.,57 and the analytical solutions. It is seen that for either
kissing/near-kissing states or critical/near-critical engulfing
states, the simulated results from the present model are in
good agreement with the analytical solutions, while the sim-
ulated results from the other two models significantly deviate
from the analytical solutions. Figure 7 shows the final fluid dis-
tributions obtained by the present model and the model of
Fu et al.,58 in both kissing and critical engulfing states. Note FIG. 8. The dimensionless interface lengths Lrg /Rr and Lrb /Rr versus Rr in the
that the model of Leclaire et al.57 produces almost the same kissing state and the critical engulfing state.

Phys. Fluids 31, 012108 (2019); doi: 10.1063/1.5056765 31, 012108-9

Published under license by AIP Publishing
 Physics of Fluids
model is advantageous to simulate critical state in ternary
fluids.
Finally, it is interesting to investigate how the grid res-
olution influences the accuracy of simulating the kissing and
critical engulfing states for the present model and the model
of Fu et al.58 Six different resolutions are considered, i.e.,
Rr = 20, 40, 60, 80, 100, and 120. Note that the simulation
results with the grid resolution of Rr = 60 have been shown
in Fig. 6 (see only the critical state) and Fig. 7. For different
grid resolutions, the fluid properties in lattice unit and the
lattice domain should be adjusted accordingly to correspond
to the same physical system. Figure 8 shows the variation of
the dimensionless interface lengths Lrg /Rr and Lrb /Rr with
Rr for both models. It is seen that increasing grid resolu-
tion can improve the accuracy of simulation results, but the
improvements are negligibly small when Rr ≥ 60. Unlike the
present model, the model Fu et al.58 cannot produce the cor-
rect outcomes of both critical states even for very high grid
resolutions.
V. A COMPOUND DROPLET IN A
THREE-DIMENSIONAL SHEAR FLOW
In this section, we investigate the dynamical behavior of a
compound droplet when subject to a three-dimensional shear
flow. As illustrated in Fig. 9, a spherical compound droplet,
consisting of a red fluid core (Rr = 20) and a green fluid shell
(Rg = 30) immersed in a blue fluid, is initially placed in the cen-
ter of two parallel walls which move at an equal speed U but
in opposite directions. The distance between these two walls
is H = 120, and the resulting shear rate is γ̇ = 2U H . The size
FIG. 9. Illustration of a compound droplet in a three-dimensional shear flow. Due
to the symmetry, only a half of the flow domain in the y-direction is considered.
Phys. Fluids 31, 012108 (2019); doi: 10.1063/1.5056765
Published under license by AIP Publishing
ARTICLE scitation.org/journal/phf
of the entire flow domain is set as L × W × H = 400 × 150
× 120. The periodic boundary conditions are applied in the
x-direction, while the moving velocity boundary conditions
are imposed on the top and bottom walls using the scheme
proposed by Zou and He.70 Considering the symmetry, only a
half of the flow domain in the y-direction is simulated, with
the implementations following Liu et al.29
When a compound droplet is subject to a simple shear
flow, two primary forces, i.e., the viscous force and the cap-
illary force, are exerted on the droplet. The droplet behav-
ior can be characterized by several important dimensionless
numbers, including the Reynolds number (Re), the capillary
number (Ca), the viscosity ratio (λvis ), and the interfacial ten-
γ̇ ρb R2g
sion ratio (λint ), which are defined as follows: Re = µb ,
γ̇ µb Rg
Ca = σgb , λvis = ν r :ν g :ν b , and λint = σ gb :σ rb :σ rg . In this study,
the effect of interfacial tension ratio λint is investigated for
Re = 0.1, Ca = 0.45, and λvis = 1:1:1.
Seven different interfacial tension ratios are simulated,
i.e., (a) 1:0.4:0.4, (b) 1:0.5:0.5, (c) 1:1:3, (d) 1:1:2, (e) 1:1:1, (f) 1:2:1,
and (g) 1:3:1, which correspond to seven equilibrium morpholo-
gies shown previously in Fig. 5. All simulations are run until
the droplet reaches stable deformation or breaks up. The final
shapes of the compound droplet are shown in Fig. 10. Four
different modes are clearly seen for these interfacial tension
ratios considered. Cases (a) and (b) show the first breakup
mode, where the red fluid in the core comes out and wraps
around the green fluid like a thin film, and the droplet breaks
up into three parts. Case (c) shows the second breakup mode,
where the red fluid partly comes out of the green fluid and
splits into three parts, with one staying in the center with the
green fluid (which is disintegrated into two equal parts) and
the other two escaping from the center. Cases (d) and (e) show
the third breakup mode, where the red fluid comes out, splits
into two equal-sized parts, and separates and moves away
from the green droplet. In cases (f) and (g), the red fluid keeps
staying inside the green fluid, and the Janus droplet reaches a
stable deformation eventually.
The above four modes can be explained by the results in
Fig. 5 to some extent. For λint = 1:0.4:0.4 and 1:0.5:0.5, which
correspond to the morphologies (i) and (ii) in Fig. 5, the red
fluid favors to engulf the green fluid, whereas for λint = 1:2:1
and 1:3:1, which correspond to morphologies (vi) and (vii) in
Fig. 5, the green fluid favors to engulf the red fluid. For
λint = 1:1:1, the double droplets tend to form a Janus droplet
[see morphology (v) in Fig. 5], and the shear flow is strong
enough to overcome the adhesive force between the red and
green fluids. For λint = 1:1:2, the so-called kissing state occurs
[see morphology (iv) in Fig. 5], where there is no adhesive or
repulsive force between the red and green fluids, so the shear
flow can easily separate the red fluid from the green one. For
λint = 1:1:3, there exists a repulsive force between the red and
green fluids; such a repulsive force, on the one hand, can pro-
mote the separation between the red and green fluids and, on
the other hand, can balance with the shear force so that the
red fluid can steadily stay in the center with the green fluid.
31, 012108-10
 Physics of Fluids
All the above results indicate that the interfacial tension ratio
determines the interactions among different fluids, and thus
it has a great effect on the dynamical behavior of a compound
droplet.
VI. CONCLUSIONS
A LB color-gradient model is proposed to simulate immis-
cible ternary fluids with a full range of interfacial tensions.
An interfacial force formulation for N-phase (N ≥ 3) systems
is derived and then introduced into the model using a body
force scheme, which is found to effectively reduce spurious
velocities. A recoloring algorithm proposed by Spencer et al.56
is applied to produce the phase segregation and ensure the
immiscibility of three different fluids, where an enhanced form
of segregation parameters is proposed by considering the
existence of Neumann’s triangle and the effect of the equi-
librium contact angle in a three-phase junction. The model’s
capability in modeling interfacial tensions and reducing spu-
rious velocities is first validated by the Young-Laplace test for
a compound droplet, and its overall performance is assessed
by simulating the spreading of a droplet between two strat-
ified fluids, where both partial and complete spreadings are
predicted with satisfactory accuracy.
The present model is then used to study the stability
and structure of double droplets in a static matrix over a
wide range of interfacial tensions. By changing two ratios
of the interfacial tensions, seven possible equilibrium mor-
phologies are successfully reproduced, which are consistent
Phys. Fluids 31, 012108 (2019); doi: 10.1063/1.5056765
Published under license by AIP Publishing
ARTICLE scitation.org/journal/phf
FIG. 10. The final fluid distributions for a
compound droplet in a three-dimensional
shear flow at different interfacial ten-
sion ratios. The interfacial tension ratios
σ gb :σ rb :σ rg are (a) 1:0.4:0.4, (b)
1:0.5:0.5, (c) 1:1:3, (d) 1:1:2, (e) 1:1:1,
(f) 1:2:1, and (g) 1:3:1, respectively. In
each subfigure, the left half displays the
droplet surface, while the right half dis-
plays the sectional view at y = 0.5D.
with the theoretical stability diagram by Guzowski et al.15 We
then simulate the near-critical and critical states of double
droplets, which is challenging since the outcomes are very
sensitive to the model accuracy. It is found that the simulated
results from the present model agree well with the analyt-
ical solutions, while the simulated results from the existing
color-gradient models significantly deviate from the analytical
solutions, especially in critical states. Finally, we investigate
the influence of the interfacial tension ratio on a compound
droplet subject to a three-dimensional shear flow. Four differ-
ent deformation and breakup modes are observed. The inter-
facial tension ratio is identified as an important factor to affect
the dynamical behavior of a compound droplet. In summary,
the present work provides the first LB multiphase model that
allows for accurate simulation of ternary fluid flows with a full
range of interfacial tensions. Although this model is limited to
ternary fluids with similar densities, it is of great potential in
the study of droplet dynamics in microfluidic devices, where
the Bond number is so small that the effect of the density
ratio can be negligible compared to the capillary and viscous
effects.
ACKNOWLEDGMENTS
This work was supported by the National Natural Sci-
ence Foundation of China (Nos. 51506168 and 51711530130), the
National Key Research and Development Project of China (No.
2016YFB0200902), the China Postdoctoral Science Foundation
(No. 2016M590943), Guangdong Provincial Key Laboratory
31, 012108-11
 Physics of Fluids ARTICLE scitation.org/journal/phf

of Fire Science and Technology (No. 2010A060801010), and parameter β as56

Guangdong Provincial Scientific and Technological Project k
(No. 2011B090400518). Y. Yu was supported by the China ,d= (A4)
β
Scholarship Council for one year study at the University of where k is a proportionality constant. Combination of Eq. (A4)
Strathclyde, UK. H. Liu gratefully acknowledges the financial and Eqs. (A1)–(A3) leads to
support from the Thousand Youth Talents Program for Dis-
tinguished Young Scholars, the Young Talent Support Plan of βkl β
 = km
sin θm sin θl


Xi’an Jiaotong University. 



 βlk = βlm




(A5)
 sin θm sin θk
APPENDIX: DEVIATION OF AN ENHANCED FORM


β β


OF SEGREGATION PARAMETERS β kl

ml
= mk .



 sin θk sin θl

Consider a three-phase junction, as shown in Fig. 11. Ilk ,
Ilm , and Ikm are the sharp interfaces, and their relative posi- Given that β ij = β ji , we have
tions yield Neumann’s triangle. The fluid k could diffuse into
βkl β β
the fluid l and the fluid m, and the concentration of the fluid = km = lm ∼ β 0 . (A6)
k, xk , gradually decreases from 1 to 0 across the interfaces Ikl sin θm sin θl sin θk
and Ikm . Without loss of generality, we choose a diffuse thick- Following the choice of Spencer et al.,56 for θ m ≤ 90◦ , β kl can
ness dkl across the interface Ikl , where the concentration of be written as
the fluid k changes from xk 0 to x00k . Similarly, one can find the
βkl = β 0 [1 + Ct (sin θm − 1)]. (A7)
diffuse thickness dkm across the interface Ikm , and these two
diffuse interfaces intersect at the points A00 and A0 , which are Considering the directivity of diffusion process, for θ m > 90◦ ,
located at the line Ilm and its extension line, respectively. Obvi- we have
ously, the concentrations of the fluid k at the points A0 and A00 βkl = β 0 [1 + Ct (1 − sin θm )]. (A8)
are xk 0 and x00k , respectively. The simple geometric relationship In conclusion, an enhanced form of segregation parameters
in Fig. 11 gives β kl can be obtained, which is shown in Eqs. (23) and (24).
dkl dkm
= . (A1)
sin θl sin θm REFERENCES
In a similar manner, we have 1
A. S. Utada, E. Lorenceau, D. R. Link, P. D. Kaplan, H. A. Stone, and D.
dlk dlm A. Weitz, “Monodisperse double emulsions generated from a microcapillary
= , (A2) device,” Science 308, 537–541 (2005).
sin θk sin θm 2
N. Bhatia, S. Pandit, S. Agrawal, and D. Gupta, “A review on multiple
and emulsions,” Int. J. Pharm. Erudition 3, 22–30 (2013).
dml d 3
H. Lamba, K. Sathish, and L. Sabikhi, “Double emulsions: Emerging delivery
= mk . (A3)
sin θl sin θk system for plant bioactives,” Food Bioprocess Technol. 8, 709–728 (2015).
4
D. Chong, X. Liu, H. Ma, G. Huang, Y. L. Han, X. Cui, J. Yan, and F. Xu,
It is well known that in the color-gradient model, the thick-
“Advances in fabricating double-emulsion droplets and their biomedical
ness d of a diffuse interface is controlled by the segregation applications,” Microfluid. Nanofluid. 19, 1071–1090 (2015).
5
M. A. Augustin and Y. Hemar, “Nano- and micro-structured assemblies for
encapsulation of food ingredients,” Chem. Soc. Rev. 38, 902–912 (2009).
6
V. B. Patravale and S. D. Mandawgade, “Novel cosmetic delivery systems:
An application update,” Int. J. Cosmet. Sci. 30, 19–33 (2008).
7
E. E. Ekanem, S. A. Nabavi, G. T. Vladisavljević, and S. Gu, “Structured
biodegradable polymeric microparticles for drug delivery produced using
flow focusing glass microfluidic devices,” ACS Appl. Mater. Interfaces 7,
23132–23143 (2015).
8
B. Ahmed, D. A. Barrow, and T. Wirth, “Enhancement of reaction rates
by segmented fluid flow in capillary scale reactors,” Adv. Synth. Catal. 348,
1043–1048 (2006).
9
L. Chen, Y. Li, J. Fan, H. K. Bisoyi, D. A. Weitz, and Q. Li, “Microshells:
Photoresponsive monodisperse cholesteric liquid crystalline microshells
for tunable omnidirectional lasing enabled by a visible light-driven chiral
molecular switch,” Adv. Opt. Mater. 2, 904 (2014).
10
K. Funakoshi, H. Suzuki, and S. Takeuchi, “Lipid bilayer formation by con-
tacting monolayers in a microfluidic device for membrane protein analysis,”
Anal. Chem. 78, 8169–8174 (2006).
11
P. G. De Gennes, “Soft matter,” Rev. Mod. Phys. 64, 645–648 (1992).
12
G. Vladisavljević, R. A. Nuumani, and S. Nabavi, “Microfluidic production
of multiple emulsions,” Micromachines 8, 75 (2017).
13
S. Torza and S. Mason, “Coalescence of two immiscible liquid drops,”
FIG. 11. Diffuse interface description of a three-phase junction. Science 163, 813–814 (1969).

Phys. Fluids 31, 012108 (2019); doi: 10.1063/1.5056765 31, 012108-12

Published under license by AIP Publishing
 Physics of Fluids
14
N. Pannacci, H. Bruus, D. Bartolo, I. Etchart, T. Lockhart, Y. Hen-
nequin, H. Willaime, and P. Tabeling, “Equilibrium and nonequilibrium
states in microfluidic double emulsions,” Phys. Rev. Lett. 101, 164502
(2008).
15
J. Guzowski, P. M. Korczyk, S. Jakiela, and P. Garstecki, “The structure and
stability of multiple micro-droplets,” Soft Matter 8, 7269–7278 (2012).
16
M. Muradoglu and S. Tasoglu, “A front-tracking method for computa-
tional modeling of impact and spreading of viscous droplets on solid walls,”
Comput. Fluids 39, 615–625 (2010).
17
R. Bonhomme, J. Magnaudet, F. Duval, and B. Piar, “Inertial dynamics of
air bubbles crossing a horizontal fluid–fluid interface,” J. Fluid Mech. 707,
405–443 (2012).
18
H.-K. Zhao, T. Chan, B. Merriman, and S. Osher, “A variational level set
approach to multiphase motion,” J. Comput. Phys. 127, 179–195 (1996).
19
K. A. Smith, F. J. Solis, and D. Chopp, “A projection method for motion of
triple junctions by level sets,” Interfaces Free Boundaries 4, 263–276 (2002).
20
R. I. Saye and J. A. Sethian, “The Voronoi implicit interface method for
computing multiphase physics,” Proc. Natl. Acad. Sci. U. S. A. 108, 19498–
19503 (2011).
21
F. Boyer and C. Lapuerta, “Study of a three component Cahn-Hilliard flow
model,” ESAIM: Math. Modell. Numer. Anal. 40, 653–687 (2006).
22
F. Boyer, C. Lapuerta, S. Minjeaud, B. Piar, and M. Quintard, “Cahn-
Hilliard/Navier-Stokes model for the simulation of three-phase flows,”
Transp. Porous Media 82, 463–483 (2010).
23
J. Kim and J. S. Lowengrub, “Phase field modeling and simulation of three-
phase flows,” Interfaces Free Boundaries 7, 435–466 (2005).
24
J. Kim, “Phase field computations for ternary fluid flows,” Comput. Meth-
ods Appl. Mech. Eng. 196, 4779–4788 (2007).
25
J. Kim, “A generalized continuous surface tension force formulation for
phase-field models for multi-component immiscible fluid flows,” Comput.
Methods Appl. Mech. Eng. 198, 3105–3112 (2009).
26
J. Kim, “Phase-field models for multi-component fluid flows,” Commun.
Comput. Phys. 12, 613–661 (2012).
27
R. Folch and M. Plapp, “Quantitative phase-field modeling of two-phase
growth,” Phys. Rev. E 72, 011602 (2005).
28
R. Spatschek and B. Eidel, “Driving forces for interface kinetics and phase
field models,” Int. J. Solids Struct. 50, 2424–2436 (2013).
29
H. Liu, Y. Ba, L. Wu, Z. Li, G. Xi, and Y. Zhang, “A hybrid lattice Boltz-
mann and finite difference method for droplet dynamics with insoluble
surfactants,” J. Fluid Mech. 837, 381–412 (2018).
30
H. Liu, A. J. Valocchi, and Q. Kang, “Three-dimensional lattice Boltzmann
model for immiscible two-phase flow simulations,” Phys. Rev. E 85, 046309
(2012).
31
A. Mazloomi Moqaddam, S. S. Chikatamarla, and I. V. Karlin, “Simulation of
binary droplet collisions with the entropic lattice Boltzmann method,” Phys.
Fluids 28, 022106 (2016).
32
X.-D. Niu, Y. Li, Y.-R. Ma, M.-F. Chen, X. Li, and Q.-Z. Li, “A mass-
conserving multiphase lattice Boltzmann model for simulation of multi-
phase flows,” Phys. Fluids 30, 013302 (2018).
33
F. Qin, A. Mazloomi Moqaddam, Q. Kang, D. Derome, and J. Carmeliet,
“Entropic multiple-relaxation-time multirange pseudopotential lattice
Boltzmann model for two-phase flow,” Phys. Fluids 30, 032104
(2018).
34
E. Ezzatneshan, “Study of surface wettability effect on cavitation incep-
tion by implementation of the lattice Boltzmann method,” Phys. Fluids 29,
113304 (2017).
35
H. Yuan, Y. Wang, and C. Shu, “An adaptive mesh refinement-multiphase
lattice Boltzmann flux solver for simulation of complex binary fluid flows,”
Phys. Fluids 29, 123604 (2017).
36
X. He and L.-S. Luo, “A priori derivation of the lattice Boltzmann equation,”
Phys. Rev. E 55, R6333 (1997).
37
S. Chen and G. D. Doolen, “Lattice Boltzmann method for fluid flows,”
Annu. Rev. Fluid Mech. 30, 329–364 (1998).
38
C. K. Aidun and J. R. Clausen, “Lattice-Boltzmann method for complex
flows,” Annu. Rev. Fluid Mech. 42, 439–472 (2010).
Phys. Fluids 31, 012108 (2019); doi: 10.1063/1.5056765
Published under license by AIP Publishing
ARTICLE scitation.org/journal/phf
39
A. K. Gunstensen, D. H. Rothman, S. Zaleski, and G. Zanetti, “Lat-
tice Boltzmann model of immiscible fluids,” Phys. Rev. A 43, 4320–4327
(1991).
40
T. Reis and T. N. Phillips, “Lattice Boltzmann model for simulating
immiscible two-phase flows,” J. Phys. A: Math. Theor. 40, 4033–4053
(2007).
41
X. Shan and H. Chen, “Lattice Boltzmann model for simulating flows with
multiple phases and components,” Phys. Rev. E 47, 1815 (1993).
42
X. Shan and H. Chen, “Simulation of nonideal gases and liquid-gas phase
transitions by the lattice Boltzmann equation,” Phys. Rev. E 49, 2941–2948
(1994).
43
M. Sbragaglia, R. Benzi, L. Biferale, S. Succi, K. Sugiyama, and F. Toschi,
“Generalized lattice Boltzmann method with multirange pseudopotential,”
Phys. Rev. E 75, 026702 (2007).
44
M. Sbragaglia, R. Benzi, M. Bernaschi, and S. Succi, “The emergence of
supramolecular forces from lattice kinetic models of non-ideal fluids: Appli-
cations to the rheology of soft glassy materials,” Soft Matter 8, 10773–10782
(2012).
45
B. Dollet, A. Scagliarini, and M. Sbragaglia, “Two-dimensional plastic flow
of foams and emulsions in a channel: Experiments and lattice Boltzmann
simulations,” J. Fluid Mech. 766, 556–589 (2015).
46
M. R. Swift, E. Orlandini, W. Osborn, and J. Yeomans, “Lattice Boltzmann
simulations of liquid-gas and binary fluid systems,” Phys. Rev. E 54, 5041
(1996).
47
Y. Wang, C. Shu, H. Huang, and C. Teo, “Multiphase lattice Boltzmann
flux solver for incompressible multiphase flows with large density ratio,”
J. Comput. Phys. 280, 404–423 (2015).
48
Y. Wang, C. Shu, and L. M. Yang, “An improved multiphase lattice Boltz-
mann flux solver for three-dimensional flows with large density ratio and
high Reynolds number,” J. Comput. Phys. 302, 41–58 (2015).
49
X. He, S. Chen, and R. Zhang, “A lattice Boltzmann scheme for incompress-
ible multiphase flow and its application in simulation of Rayleigh-Taylor
instability,” J. Comput. Phys. 152, 642–663 (1999).
50
H. Chen, B. M. Boghosian, and P. V. Coveney, “A ternary lattice Boltz-
mann model for amphiphilic fluids,” Proc. R. Soc. A 456, 2043–2057
(2000).
51
O. Shardt, J. J. Derksen, and S. K. Mitra, “Simulations of Janus droplets at
equilibrium and in shear,” Phys. Fluids 26, 012104 (2014).
52
H. Liang, B. C. Shi, and Z. Chai, “Lattice Boltzmann modeling of three-
phase incompressible flows,” Phys. Rev. E 93, 013308 (2016).
53
C. Semprebon, T. Kruger, and H. Kusumaatmaja, “Ternary free energy lat-
tice Boltzmann model with tunable surface tensions and contact angles,”
Phys. Rev. E 93, 033305 (2016).
54
M. Wöhrwag, C. Semprebon, A. M. Moqaddam, I. Karlin, and H. Kusumaat-
maja, “Ternary free-energy entropic lattice Boltzmann model with a high
density ratio,” Phys. Rev. Lett. 120, 234501 (2018).
55
Y. Shi, G. Tang, and Y. Wang, “Simulation of three-component fluid flows
using the multiphase lattice Boltzmann flux solver,” J. Comput. Phys. 314,
228–243 (2016).
56
T. Spencer, I. Halliday, and C. M. Care, “Lattice Boltzmann equation
method for multiple immiscible continuum fluids,” Phys. Rev. E 82, 066701
(2010).
57
S. Leclaire, M. Reggio, and J. Trepanier, “Progress and investigation on
lattice Boltzmann modeling of multiple immiscible fluids or components
with variable density and viscosity ratios,” J. Comput. Phys. 246, 318–342
(2013).
58
Y. Fu, S. Zhao, L. Bai, Y. Jin, and Y. Cheng, “Numerical study of dou-
ble emulsion formation in microchannels by a ternary lattice Boltzmann
method,” Chem. Eng. Sci. 146, 126–134 (2016).
59
R. H. H. Abadi, M. H. Rahimian, and A. Fakhari, “Conservative phase-field
lattice-Boltzmann model for ternary fluids,” J. Comput. Phys. 374, 668–691
(2018).
60
F. Jiang and T. Tsuji, “Estimation of three-phase relative permeability by
simulating fluid dynamics directly on rock-microstructure images,” Water
Resour. Res. 53, 11–32, https://doi.org/10.1002/2016wr019098 (2017).
31, 012108-13
 Physics of Fluids ARTICLE scitation.org/journal/phf

61
H. Liu and Y. Zhang, “Modelling thermocapillary migration of a microflu-
idic droplet on a solid surface,” J. Comput. Phys. 280, 37–53 (2015).
62
Z. Guo, C. Zheng, and B. Shi, “Discrete lattice effects on the forcing term
in the lattice Boltzmann method.” Phys. Rev. E 65, 046308 (2002).
63
I. Halliday, C. M. Care, and A. P. Hollis, “Improved simulation of drop
dynamics in a shear flow at low Reynolds and capillary number,” Phys. Rev. E
73, 056708 (2006).
64
Y. Ba, H. Liu, Q. Li, Q. Kang, and J. Sun, “Multiple-relaxation-time color-
gradient lattice Boltzmann model for simulating two-phase flows with high
density ratio,” Phys. Rev. E 94, 023310 (2016).
65
U. D’Ortona, D. Salin, M. Cieplak, R. B. Rybka, and J. R. Banavar, “Two-color
nonlinear Boltzmann cellular automata: Surface tension and wetting,” Phys.
Rev. E 51, 3718 (1995).
66
M. Latva-Kokko and D. H. Rothman, “Diffusion properties of gradient-
based lattice Boltzmann models of immiscible fluids,” Phys. Rev. E 71, 056702
(2005).
67
J. S. Rowlinson and B. Widom, Molecular Theory of Capillarity (Courier
Corporation, 2013).
68
A. Riaud, S. Zhao, K. Wang, Y. Cheng, and G. Luo, “Lattice-Boltzmann
method for the simulation of multiphase mass transfer and reaction of dilute
species,” Phys. Rev. E 89, 053308 (2014).
69
H. Liu, L. Wu, Y. Ba, and G. Xi, “A lattice Boltzmann method for axisym-
metric thermocapillary flows,” Int. J. Heat Mass Transfer 104, 337–350
(2017).
70
Q. Zou and X. He, “On pressure and velocity boundary conditions for the
lattice Boltzmann BGK model,” Phys. Fluids 9, 1591–1598 (1997).

Phys. Fluids 31, 012108 (2019); doi: 10.1063/1.5056765 31, 012108-14

Published under license by AIP Publishing